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 | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
5517d86b | 66 | #include <linux/reciprocal_div.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
434d53b0 | 71 | #include <linux/bootmem.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
0a16b607 | 75 | #include <trace/sched.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
d0b27fa7 PZ |
116 | /* |
117 | * single value that denotes runtime == period, ie unlimited time. | |
118 | */ | |
119 | #define RUNTIME_INF ((u64)~0ULL) | |
120 | ||
7e066fb8 MD |
121 | DEFINE_TRACE(sched_wait_task); |
122 | DEFINE_TRACE(sched_wakeup); | |
123 | DEFINE_TRACE(sched_wakeup_new); | |
124 | DEFINE_TRACE(sched_switch); | |
125 | DEFINE_TRACE(sched_migrate_task); | |
126 | ||
5517d86b ED |
127 | #ifdef CONFIG_SMP |
128 | /* | |
129 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
130 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
131 | */ | |
132 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
133 | { | |
134 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
135 | } | |
136 | ||
137 | /* | |
138 | * Each time a sched group cpu_power is changed, | |
139 | * we must compute its reciprocal value | |
140 | */ | |
141 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
142 | { | |
143 | sg->__cpu_power += val; | |
144 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
145 | } | |
146 | #endif | |
147 | ||
e05606d3 IM |
148 | static inline int rt_policy(int policy) |
149 | { | |
3f33a7ce | 150 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
151 | return 1; |
152 | return 0; | |
153 | } | |
154 | ||
155 | static inline int task_has_rt_policy(struct task_struct *p) | |
156 | { | |
157 | return rt_policy(p->policy); | |
158 | } | |
159 | ||
1da177e4 | 160 | /* |
6aa645ea | 161 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 162 | */ |
6aa645ea IM |
163 | struct rt_prio_array { |
164 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
165 | struct list_head queue[MAX_RT_PRIO]; | |
166 | }; | |
167 | ||
d0b27fa7 | 168 | struct rt_bandwidth { |
ea736ed5 IM |
169 | /* nests inside the rq lock: */ |
170 | spinlock_t rt_runtime_lock; | |
171 | ktime_t rt_period; | |
172 | u64 rt_runtime; | |
173 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
174 | }; |
175 | ||
176 | static struct rt_bandwidth def_rt_bandwidth; | |
177 | ||
178 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
179 | ||
180 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
181 | { | |
182 | struct rt_bandwidth *rt_b = | |
183 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
184 | ktime_t now; | |
185 | int overrun; | |
186 | int idle = 0; | |
187 | ||
188 | for (;;) { | |
189 | now = hrtimer_cb_get_time(timer); | |
190 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
191 | ||
192 | if (!overrun) | |
193 | break; | |
194 | ||
195 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
196 | } | |
197 | ||
198 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
199 | } | |
200 | ||
201 | static | |
202 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
203 | { | |
204 | rt_b->rt_period = ns_to_ktime(period); | |
205 | rt_b->rt_runtime = runtime; | |
206 | ||
ac086bc2 PZ |
207 | spin_lock_init(&rt_b->rt_runtime_lock); |
208 | ||
d0b27fa7 PZ |
209 | hrtimer_init(&rt_b->rt_period_timer, |
210 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
211 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
ccc7dadf | 212 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; |
d0b27fa7 PZ |
213 | } |
214 | ||
c8bfff6d KH |
215 | static inline int rt_bandwidth_enabled(void) |
216 | { | |
217 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
218 | } |
219 | ||
220 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
221 | { | |
222 | ktime_t now; | |
223 | ||
0b148fa0 | 224 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
225 | return; |
226 | ||
227 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
228 | return; | |
229 | ||
230 | spin_lock(&rt_b->rt_runtime_lock); | |
231 | for (;;) { | |
232 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
233 | break; | |
234 | ||
235 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
236 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
237 | hrtimer_start_expires(&rt_b->rt_period_timer, |
238 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
239 | } |
240 | spin_unlock(&rt_b->rt_runtime_lock); | |
241 | } | |
242 | ||
243 | #ifdef CONFIG_RT_GROUP_SCHED | |
244 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
245 | { | |
246 | hrtimer_cancel(&rt_b->rt_period_timer); | |
247 | } | |
248 | #endif | |
249 | ||
712555ee HC |
250 | /* |
251 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
252 | * detach_destroy_domains and partition_sched_domains. | |
253 | */ | |
254 | static DEFINE_MUTEX(sched_domains_mutex); | |
255 | ||
052f1dc7 | 256 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 257 | |
68318b8e SV |
258 | #include <linux/cgroup.h> |
259 | ||
29f59db3 SV |
260 | struct cfs_rq; |
261 | ||
6f505b16 PZ |
262 | static LIST_HEAD(task_groups); |
263 | ||
29f59db3 | 264 | /* task group related information */ |
4cf86d77 | 265 | struct task_group { |
052f1dc7 | 266 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
267 | struct cgroup_subsys_state css; |
268 | #endif | |
052f1dc7 PZ |
269 | |
270 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
271 | /* schedulable entities of this group on each cpu */ |
272 | struct sched_entity **se; | |
273 | /* runqueue "owned" by this group on each cpu */ | |
274 | struct cfs_rq **cfs_rq; | |
275 | unsigned long shares; | |
052f1dc7 PZ |
276 | #endif |
277 | ||
278 | #ifdef CONFIG_RT_GROUP_SCHED | |
279 | struct sched_rt_entity **rt_se; | |
280 | struct rt_rq **rt_rq; | |
281 | ||
d0b27fa7 | 282 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 283 | #endif |
6b2d7700 | 284 | |
ae8393e5 | 285 | struct rcu_head rcu; |
6f505b16 | 286 | struct list_head list; |
f473aa5e PZ |
287 | |
288 | struct task_group *parent; | |
289 | struct list_head siblings; | |
290 | struct list_head children; | |
29f59db3 SV |
291 | }; |
292 | ||
354d60c2 | 293 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
294 | |
295 | /* | |
296 | * Root task group. | |
297 | * Every UID task group (including init_task_group aka UID-0) will | |
298 | * be a child to this group. | |
299 | */ | |
300 | struct task_group root_task_group; | |
301 | ||
052f1dc7 | 302 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
303 | /* Default task group's sched entity on each cpu */ |
304 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
305 | /* Default task group's cfs_rq on each cpu */ | |
306 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 307 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
308 | |
309 | #ifdef CONFIG_RT_GROUP_SCHED | |
310 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
311 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 312 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 313 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 314 | #define root_task_group init_task_group |
9a7e0b18 | 315 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 316 | |
8ed36996 | 317 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
318 | * a task group's cpu shares. |
319 | */ | |
8ed36996 | 320 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 321 | |
052f1dc7 | 322 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
323 | #ifdef CONFIG_USER_SCHED |
324 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 325 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 326 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 327 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 328 | |
cb4ad1ff | 329 | /* |
2e084786 LJ |
330 | * A weight of 0 or 1 can cause arithmetics problems. |
331 | * A weight of a cfs_rq is the sum of weights of which entities | |
332 | * are queued on this cfs_rq, so a weight of a entity should not be | |
333 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
334 | * (The default weight is 1024 - so there's no practical |
335 | * limitation from this.) | |
336 | */ | |
18d95a28 | 337 | #define MIN_SHARES 2 |
2e084786 | 338 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 339 | |
052f1dc7 PZ |
340 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
341 | #endif | |
342 | ||
29f59db3 | 343 | /* Default task group. |
3a252015 | 344 | * Every task in system belong to this group at bootup. |
29f59db3 | 345 | */ |
434d53b0 | 346 | struct task_group init_task_group; |
29f59db3 SV |
347 | |
348 | /* return group to which a task belongs */ | |
4cf86d77 | 349 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 350 | { |
4cf86d77 | 351 | struct task_group *tg; |
9b5b7751 | 352 | |
052f1dc7 | 353 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 354 | tg = p->user->tg; |
052f1dc7 | 355 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
356 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
357 | struct task_group, css); | |
24e377a8 | 358 | #else |
41a2d6cf | 359 | tg = &init_task_group; |
24e377a8 | 360 | #endif |
9b5b7751 | 361 | return tg; |
29f59db3 SV |
362 | } |
363 | ||
364 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 365 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 366 | { |
052f1dc7 | 367 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
368 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
369 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 370 | #endif |
6f505b16 | 371 | |
052f1dc7 | 372 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
373 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
374 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 375 | #endif |
29f59db3 SV |
376 | } |
377 | ||
378 | #else | |
379 | ||
6f505b16 | 380 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
381 | static inline struct task_group *task_group(struct task_struct *p) |
382 | { | |
383 | return NULL; | |
384 | } | |
29f59db3 | 385 | |
052f1dc7 | 386 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 387 | |
6aa645ea IM |
388 | /* CFS-related fields in a runqueue */ |
389 | struct cfs_rq { | |
390 | struct load_weight load; | |
391 | unsigned long nr_running; | |
392 | ||
6aa645ea | 393 | u64 exec_clock; |
e9acbff6 | 394 | u64 min_vruntime; |
6aa645ea IM |
395 | |
396 | struct rb_root tasks_timeline; | |
397 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
398 | |
399 | struct list_head tasks; | |
400 | struct list_head *balance_iterator; | |
401 | ||
402 | /* | |
403 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
404 | * It is set to NULL otherwise (i.e when none are currently running). |
405 | */ | |
4793241b | 406 | struct sched_entity *curr, *next, *last; |
ddc97297 | 407 | |
5ac5c4d6 | 408 | unsigned int nr_spread_over; |
ddc97297 | 409 | |
62160e3f | 410 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
411 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
412 | ||
41a2d6cf IM |
413 | /* |
414 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
415 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
416 | * (like users, containers etc.) | |
417 | * | |
418 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
419 | * list is used during load balance. | |
420 | */ | |
41a2d6cf IM |
421 | struct list_head leaf_cfs_rq_list; |
422 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
423 | |
424 | #ifdef CONFIG_SMP | |
c09595f6 | 425 | /* |
c8cba857 | 426 | * the part of load.weight contributed by tasks |
c09595f6 | 427 | */ |
c8cba857 | 428 | unsigned long task_weight; |
c09595f6 | 429 | |
c8cba857 PZ |
430 | /* |
431 | * h_load = weight * f(tg) | |
432 | * | |
433 | * Where f(tg) is the recursive weight fraction assigned to | |
434 | * this group. | |
435 | */ | |
436 | unsigned long h_load; | |
c09595f6 | 437 | |
c8cba857 PZ |
438 | /* |
439 | * this cpu's part of tg->shares | |
440 | */ | |
441 | unsigned long shares; | |
f1d239f7 PZ |
442 | |
443 | /* | |
444 | * load.weight at the time we set shares | |
445 | */ | |
446 | unsigned long rq_weight; | |
c09595f6 | 447 | #endif |
6aa645ea IM |
448 | #endif |
449 | }; | |
1da177e4 | 450 | |
6aa645ea IM |
451 | /* Real-Time classes' related field in a runqueue: */ |
452 | struct rt_rq { | |
453 | struct rt_prio_array active; | |
63489e45 | 454 | unsigned long rt_nr_running; |
052f1dc7 | 455 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
456 | int highest_prio; /* highest queued rt task prio */ |
457 | #endif | |
fa85ae24 | 458 | #ifdef CONFIG_SMP |
73fe6aae | 459 | unsigned long rt_nr_migratory; |
a22d7fc1 | 460 | int overloaded; |
fa85ae24 | 461 | #endif |
6f505b16 | 462 | int rt_throttled; |
fa85ae24 | 463 | u64 rt_time; |
ac086bc2 | 464 | u64 rt_runtime; |
ea736ed5 | 465 | /* Nests inside the rq lock: */ |
ac086bc2 | 466 | spinlock_t rt_runtime_lock; |
6f505b16 | 467 | |
052f1dc7 | 468 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
469 | unsigned long rt_nr_boosted; |
470 | ||
6f505b16 PZ |
471 | struct rq *rq; |
472 | struct list_head leaf_rt_rq_list; | |
473 | struct task_group *tg; | |
474 | struct sched_rt_entity *rt_se; | |
475 | #endif | |
6aa645ea IM |
476 | }; |
477 | ||
57d885fe GH |
478 | #ifdef CONFIG_SMP |
479 | ||
480 | /* | |
481 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
482 | * variables. Each exclusive cpuset essentially defines an island domain by |
483 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
484 | * exclusive cpuset is created, we also create and attach a new root-domain |
485 | * object. | |
486 | * | |
57d885fe GH |
487 | */ |
488 | struct root_domain { | |
489 | atomic_t refcount; | |
490 | cpumask_t span; | |
491 | cpumask_t online; | |
637f5085 | 492 | |
0eab9146 | 493 | /* |
637f5085 GH |
494 | * The "RT overload" flag: it gets set if a CPU has more than |
495 | * one runnable RT task. | |
496 | */ | |
497 | cpumask_t rto_mask; | |
0eab9146 | 498 | atomic_t rto_count; |
6e0534f2 GH |
499 | #ifdef CONFIG_SMP |
500 | struct cpupri cpupri; | |
501 | #endif | |
57d885fe GH |
502 | }; |
503 | ||
dc938520 GH |
504 | /* |
505 | * By default the system creates a single root-domain with all cpus as | |
506 | * members (mimicking the global state we have today). | |
507 | */ | |
57d885fe GH |
508 | static struct root_domain def_root_domain; |
509 | ||
510 | #endif | |
511 | ||
1da177e4 LT |
512 | /* |
513 | * This is the main, per-CPU runqueue data structure. | |
514 | * | |
515 | * Locking rule: those places that want to lock multiple runqueues | |
516 | * (such as the load balancing or the thread migration code), lock | |
517 | * acquire operations must be ordered by ascending &runqueue. | |
518 | */ | |
70b97a7f | 519 | struct rq { |
d8016491 IM |
520 | /* runqueue lock: */ |
521 | spinlock_t lock; | |
1da177e4 LT |
522 | |
523 | /* | |
524 | * nr_running and cpu_load should be in the same cacheline because | |
525 | * remote CPUs use both these fields when doing load calculation. | |
526 | */ | |
527 | unsigned long nr_running; | |
6aa645ea IM |
528 | #define CPU_LOAD_IDX_MAX 5 |
529 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 530 | unsigned char idle_at_tick; |
46cb4b7c | 531 | #ifdef CONFIG_NO_HZ |
15934a37 | 532 | unsigned long last_tick_seen; |
46cb4b7c SS |
533 | unsigned char in_nohz_recently; |
534 | #endif | |
d8016491 IM |
535 | /* capture load from *all* tasks on this cpu: */ |
536 | struct load_weight load; | |
6aa645ea IM |
537 | unsigned long nr_load_updates; |
538 | u64 nr_switches; | |
539 | ||
540 | struct cfs_rq cfs; | |
6f505b16 | 541 | struct rt_rq rt; |
6f505b16 | 542 | |
6aa645ea | 543 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
544 | /* list of leaf cfs_rq on this cpu: */ |
545 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
546 | #endif |
547 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 548 | struct list_head leaf_rt_rq_list; |
1da177e4 | 549 | #endif |
1da177e4 LT |
550 | |
551 | /* | |
552 | * This is part of a global counter where only the total sum | |
553 | * over all CPUs matters. A task can increase this counter on | |
554 | * one CPU and if it got migrated afterwards it may decrease | |
555 | * it on another CPU. Always updated under the runqueue lock: | |
556 | */ | |
557 | unsigned long nr_uninterruptible; | |
558 | ||
36c8b586 | 559 | struct task_struct *curr, *idle; |
c9819f45 | 560 | unsigned long next_balance; |
1da177e4 | 561 | struct mm_struct *prev_mm; |
6aa645ea | 562 | |
3e51f33f | 563 | u64 clock; |
6aa645ea | 564 | |
1da177e4 LT |
565 | atomic_t nr_iowait; |
566 | ||
567 | #ifdef CONFIG_SMP | |
0eab9146 | 568 | struct root_domain *rd; |
1da177e4 LT |
569 | struct sched_domain *sd; |
570 | ||
571 | /* For active balancing */ | |
572 | int active_balance; | |
573 | int push_cpu; | |
d8016491 IM |
574 | /* cpu of this runqueue: */ |
575 | int cpu; | |
1f11eb6a | 576 | int online; |
1da177e4 | 577 | |
a8a51d5e | 578 | unsigned long avg_load_per_task; |
1da177e4 | 579 | |
36c8b586 | 580 | struct task_struct *migration_thread; |
1da177e4 LT |
581 | struct list_head migration_queue; |
582 | #endif | |
583 | ||
8f4d37ec | 584 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
585 | #ifdef CONFIG_SMP |
586 | int hrtick_csd_pending; | |
587 | struct call_single_data hrtick_csd; | |
588 | #endif | |
8f4d37ec PZ |
589 | struct hrtimer hrtick_timer; |
590 | #endif | |
591 | ||
1da177e4 LT |
592 | #ifdef CONFIG_SCHEDSTATS |
593 | /* latency stats */ | |
594 | struct sched_info rq_sched_info; | |
595 | ||
596 | /* sys_sched_yield() stats */ | |
480b9434 KC |
597 | unsigned int yld_exp_empty; |
598 | unsigned int yld_act_empty; | |
599 | unsigned int yld_both_empty; | |
600 | unsigned int yld_count; | |
1da177e4 LT |
601 | |
602 | /* schedule() stats */ | |
480b9434 KC |
603 | unsigned int sched_switch; |
604 | unsigned int sched_count; | |
605 | unsigned int sched_goidle; | |
1da177e4 LT |
606 | |
607 | /* try_to_wake_up() stats */ | |
480b9434 KC |
608 | unsigned int ttwu_count; |
609 | unsigned int ttwu_local; | |
b8efb561 IM |
610 | |
611 | /* BKL stats */ | |
480b9434 | 612 | unsigned int bkl_count; |
1da177e4 LT |
613 | #endif |
614 | }; | |
615 | ||
f34e3b61 | 616 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 617 | |
15afe09b | 618 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 619 | { |
15afe09b | 620 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
621 | } |
622 | ||
0a2966b4 CL |
623 | static inline int cpu_of(struct rq *rq) |
624 | { | |
625 | #ifdef CONFIG_SMP | |
626 | return rq->cpu; | |
627 | #else | |
628 | return 0; | |
629 | #endif | |
630 | } | |
631 | ||
674311d5 NP |
632 | /* |
633 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 634 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
635 | * |
636 | * The domain tree of any CPU may only be accessed from within | |
637 | * preempt-disabled sections. | |
638 | */ | |
48f24c4d IM |
639 | #define for_each_domain(cpu, __sd) \ |
640 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
641 | |
642 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
643 | #define this_rq() (&__get_cpu_var(runqueues)) | |
644 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
645 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
646 | ||
3e51f33f PZ |
647 | static inline void update_rq_clock(struct rq *rq) |
648 | { | |
649 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
650 | } | |
651 | ||
bf5c91ba IM |
652 | /* |
653 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
654 | */ | |
655 | #ifdef CONFIG_SCHED_DEBUG | |
656 | # define const_debug __read_mostly | |
657 | #else | |
658 | # define const_debug static const | |
659 | #endif | |
660 | ||
017730c1 IM |
661 | /** |
662 | * runqueue_is_locked | |
663 | * | |
664 | * Returns true if the current cpu runqueue is locked. | |
665 | * This interface allows printk to be called with the runqueue lock | |
666 | * held and know whether or not it is OK to wake up the klogd. | |
667 | */ | |
668 | int runqueue_is_locked(void) | |
669 | { | |
670 | int cpu = get_cpu(); | |
671 | struct rq *rq = cpu_rq(cpu); | |
672 | int ret; | |
673 | ||
674 | ret = spin_is_locked(&rq->lock); | |
675 | put_cpu(); | |
676 | return ret; | |
677 | } | |
678 | ||
bf5c91ba IM |
679 | /* |
680 | * Debugging: various feature bits | |
681 | */ | |
f00b45c1 PZ |
682 | |
683 | #define SCHED_FEAT(name, enabled) \ | |
684 | __SCHED_FEAT_##name , | |
685 | ||
bf5c91ba | 686 | enum { |
f00b45c1 | 687 | #include "sched_features.h" |
bf5c91ba IM |
688 | }; |
689 | ||
f00b45c1 PZ |
690 | #undef SCHED_FEAT |
691 | ||
692 | #define SCHED_FEAT(name, enabled) \ | |
693 | (1UL << __SCHED_FEAT_##name) * enabled | | |
694 | ||
bf5c91ba | 695 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
696 | #include "sched_features.h" |
697 | 0; | |
698 | ||
699 | #undef SCHED_FEAT | |
700 | ||
701 | #ifdef CONFIG_SCHED_DEBUG | |
702 | #define SCHED_FEAT(name, enabled) \ | |
703 | #name , | |
704 | ||
983ed7a6 | 705 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
706 | #include "sched_features.h" |
707 | NULL | |
708 | }; | |
709 | ||
710 | #undef SCHED_FEAT | |
711 | ||
983ed7a6 | 712 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
713 | { |
714 | filp->private_data = inode->i_private; | |
715 | return 0; | |
716 | } | |
717 | ||
718 | static ssize_t | |
719 | sched_feat_read(struct file *filp, char __user *ubuf, | |
720 | size_t cnt, loff_t *ppos) | |
721 | { | |
722 | char *buf; | |
723 | int r = 0; | |
724 | int len = 0; | |
725 | int i; | |
726 | ||
727 | for (i = 0; sched_feat_names[i]; i++) { | |
728 | len += strlen(sched_feat_names[i]); | |
729 | len += 4; | |
730 | } | |
731 | ||
732 | buf = kmalloc(len + 2, GFP_KERNEL); | |
733 | if (!buf) | |
734 | return -ENOMEM; | |
735 | ||
736 | for (i = 0; sched_feat_names[i]; i++) { | |
737 | if (sysctl_sched_features & (1UL << i)) | |
738 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
739 | else | |
c24b7c52 | 740 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
741 | } |
742 | ||
743 | r += sprintf(buf + r, "\n"); | |
744 | WARN_ON(r >= len + 2); | |
745 | ||
746 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
747 | ||
748 | kfree(buf); | |
749 | ||
750 | return r; | |
751 | } | |
752 | ||
753 | static ssize_t | |
754 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
755 | size_t cnt, loff_t *ppos) | |
756 | { | |
757 | char buf[64]; | |
758 | char *cmp = buf; | |
759 | int neg = 0; | |
760 | int i; | |
761 | ||
762 | if (cnt > 63) | |
763 | cnt = 63; | |
764 | ||
765 | if (copy_from_user(&buf, ubuf, cnt)) | |
766 | return -EFAULT; | |
767 | ||
768 | buf[cnt] = 0; | |
769 | ||
c24b7c52 | 770 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
771 | neg = 1; |
772 | cmp += 3; | |
773 | } | |
774 | ||
775 | for (i = 0; sched_feat_names[i]; i++) { | |
776 | int len = strlen(sched_feat_names[i]); | |
777 | ||
778 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
779 | if (neg) | |
780 | sysctl_sched_features &= ~(1UL << i); | |
781 | else | |
782 | sysctl_sched_features |= (1UL << i); | |
783 | break; | |
784 | } | |
785 | } | |
786 | ||
787 | if (!sched_feat_names[i]) | |
788 | return -EINVAL; | |
789 | ||
790 | filp->f_pos += cnt; | |
791 | ||
792 | return cnt; | |
793 | } | |
794 | ||
795 | static struct file_operations sched_feat_fops = { | |
796 | .open = sched_feat_open, | |
797 | .read = sched_feat_read, | |
798 | .write = sched_feat_write, | |
799 | }; | |
800 | ||
801 | static __init int sched_init_debug(void) | |
802 | { | |
f00b45c1 PZ |
803 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
804 | &sched_feat_fops); | |
805 | ||
806 | return 0; | |
807 | } | |
808 | late_initcall(sched_init_debug); | |
809 | ||
810 | #endif | |
811 | ||
812 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 813 | |
b82d9fdd PZ |
814 | /* |
815 | * Number of tasks to iterate in a single balance run. | |
816 | * Limited because this is done with IRQs disabled. | |
817 | */ | |
818 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
819 | ||
2398f2c6 PZ |
820 | /* |
821 | * ratelimit for updating the group shares. | |
55cd5340 | 822 | * default: 0.25ms |
2398f2c6 | 823 | */ |
55cd5340 | 824 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 825 | |
ffda12a1 PZ |
826 | /* |
827 | * Inject some fuzzyness into changing the per-cpu group shares | |
828 | * this avoids remote rq-locks at the expense of fairness. | |
829 | * default: 4 | |
830 | */ | |
831 | unsigned int sysctl_sched_shares_thresh = 4; | |
832 | ||
fa85ae24 | 833 | /* |
9f0c1e56 | 834 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
835 | * default: 1s |
836 | */ | |
9f0c1e56 | 837 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 838 | |
6892b75e IM |
839 | static __read_mostly int scheduler_running; |
840 | ||
9f0c1e56 PZ |
841 | /* |
842 | * part of the period that we allow rt tasks to run in us. | |
843 | * default: 0.95s | |
844 | */ | |
845 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 846 | |
d0b27fa7 PZ |
847 | static inline u64 global_rt_period(void) |
848 | { | |
849 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
850 | } | |
851 | ||
852 | static inline u64 global_rt_runtime(void) | |
853 | { | |
e26873bb | 854 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
855 | return RUNTIME_INF; |
856 | ||
857 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
858 | } | |
fa85ae24 | 859 | |
1da177e4 | 860 | #ifndef prepare_arch_switch |
4866cde0 NP |
861 | # define prepare_arch_switch(next) do { } while (0) |
862 | #endif | |
863 | #ifndef finish_arch_switch | |
864 | # define finish_arch_switch(prev) do { } while (0) | |
865 | #endif | |
866 | ||
051a1d1a DA |
867 | static inline int task_current(struct rq *rq, struct task_struct *p) |
868 | { | |
869 | return rq->curr == p; | |
870 | } | |
871 | ||
4866cde0 | 872 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 874 | { |
051a1d1a | 875 | return task_current(rq, p); |
4866cde0 NP |
876 | } |
877 | ||
70b97a7f | 878 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
879 | { |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 883 | { |
da04c035 IM |
884 | #ifdef CONFIG_DEBUG_SPINLOCK |
885 | /* this is a valid case when another task releases the spinlock */ | |
886 | rq->lock.owner = current; | |
887 | #endif | |
8a25d5de IM |
888 | /* |
889 | * If we are tracking spinlock dependencies then we have to | |
890 | * fix up the runqueue lock - which gets 'carried over' from | |
891 | * prev into current: | |
892 | */ | |
893 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
894 | ||
4866cde0 NP |
895 | spin_unlock_irq(&rq->lock); |
896 | } | |
897 | ||
898 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 899 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | return p->oncpu; | |
903 | #else | |
051a1d1a | 904 | return task_current(rq, p); |
4866cde0 NP |
905 | #endif |
906 | } | |
907 | ||
70b97a7f | 908 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
909 | { |
910 | #ifdef CONFIG_SMP | |
911 | /* | |
912 | * We can optimise this out completely for !SMP, because the | |
913 | * SMP rebalancing from interrupt is the only thing that cares | |
914 | * here. | |
915 | */ | |
916 | next->oncpu = 1; | |
917 | #endif | |
918 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
919 | spin_unlock_irq(&rq->lock); | |
920 | #else | |
921 | spin_unlock(&rq->lock); | |
922 | #endif | |
923 | } | |
924 | ||
70b97a7f | 925 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
926 | { |
927 | #ifdef CONFIG_SMP | |
928 | /* | |
929 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
930 | * We must ensure this doesn't happen until the switch is completely | |
931 | * finished. | |
932 | */ | |
933 | smp_wmb(); | |
934 | prev->oncpu = 0; | |
935 | #endif | |
936 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
937 | local_irq_enable(); | |
1da177e4 | 938 | #endif |
4866cde0 NP |
939 | } |
940 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 941 | |
b29739f9 IM |
942 | /* |
943 | * __task_rq_lock - lock the runqueue a given task resides on. | |
944 | * Must be called interrupts disabled. | |
945 | */ | |
70b97a7f | 946 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
947 | __acquires(rq->lock) |
948 | { | |
3a5c359a AK |
949 | for (;;) { |
950 | struct rq *rq = task_rq(p); | |
951 | spin_lock(&rq->lock); | |
952 | if (likely(rq == task_rq(p))) | |
953 | return rq; | |
b29739f9 | 954 | spin_unlock(&rq->lock); |
b29739f9 | 955 | } |
b29739f9 IM |
956 | } |
957 | ||
1da177e4 LT |
958 | /* |
959 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 960 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
961 | * explicitly disabling preemption. |
962 | */ | |
70b97a7f | 963 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
964 | __acquires(rq->lock) |
965 | { | |
70b97a7f | 966 | struct rq *rq; |
1da177e4 | 967 | |
3a5c359a AK |
968 | for (;;) { |
969 | local_irq_save(*flags); | |
970 | rq = task_rq(p); | |
971 | spin_lock(&rq->lock); | |
972 | if (likely(rq == task_rq(p))) | |
973 | return rq; | |
1da177e4 | 974 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 975 | } |
1da177e4 LT |
976 | } |
977 | ||
ad474cac ON |
978 | void task_rq_unlock_wait(struct task_struct *p) |
979 | { | |
980 | struct rq *rq = task_rq(p); | |
981 | ||
982 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
983 | spin_unlock_wait(&rq->lock); | |
984 | } | |
985 | ||
a9957449 | 986 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
987 | __releases(rq->lock) |
988 | { | |
989 | spin_unlock(&rq->lock); | |
990 | } | |
991 | ||
70b97a7f | 992 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
993 | __releases(rq->lock) |
994 | { | |
995 | spin_unlock_irqrestore(&rq->lock, *flags); | |
996 | } | |
997 | ||
1da177e4 | 998 | /* |
cc2a73b5 | 999 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1000 | */ |
a9957449 | 1001 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1002 | __acquires(rq->lock) |
1003 | { | |
70b97a7f | 1004 | struct rq *rq; |
1da177e4 LT |
1005 | |
1006 | local_irq_disable(); | |
1007 | rq = this_rq(); | |
1008 | spin_lock(&rq->lock); | |
1009 | ||
1010 | return rq; | |
1011 | } | |
1012 | ||
8f4d37ec PZ |
1013 | #ifdef CONFIG_SCHED_HRTICK |
1014 | /* | |
1015 | * Use HR-timers to deliver accurate preemption points. | |
1016 | * | |
1017 | * Its all a bit involved since we cannot program an hrt while holding the | |
1018 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1019 | * reschedule event. | |
1020 | * | |
1021 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1022 | * rq->lock. | |
1023 | */ | |
8f4d37ec PZ |
1024 | |
1025 | /* | |
1026 | * Use hrtick when: | |
1027 | * - enabled by features | |
1028 | * - hrtimer is actually high res | |
1029 | */ | |
1030 | static inline int hrtick_enabled(struct rq *rq) | |
1031 | { | |
1032 | if (!sched_feat(HRTICK)) | |
1033 | return 0; | |
ba42059f | 1034 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1035 | return 0; |
8f4d37ec PZ |
1036 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1037 | } | |
1038 | ||
8f4d37ec PZ |
1039 | static void hrtick_clear(struct rq *rq) |
1040 | { | |
1041 | if (hrtimer_active(&rq->hrtick_timer)) | |
1042 | hrtimer_cancel(&rq->hrtick_timer); | |
1043 | } | |
1044 | ||
8f4d37ec PZ |
1045 | /* |
1046 | * High-resolution timer tick. | |
1047 | * Runs from hardirq context with interrupts disabled. | |
1048 | */ | |
1049 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1050 | { | |
1051 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1052 | ||
1053 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1054 | ||
1055 | spin_lock(&rq->lock); | |
3e51f33f | 1056 | update_rq_clock(rq); |
8f4d37ec PZ |
1057 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1058 | spin_unlock(&rq->lock); | |
1059 | ||
1060 | return HRTIMER_NORESTART; | |
1061 | } | |
1062 | ||
95e904c7 | 1063 | #ifdef CONFIG_SMP |
31656519 PZ |
1064 | /* |
1065 | * called from hardirq (IPI) context | |
1066 | */ | |
1067 | static void __hrtick_start(void *arg) | |
b328ca18 | 1068 | { |
31656519 | 1069 | struct rq *rq = arg; |
b328ca18 | 1070 | |
31656519 PZ |
1071 | spin_lock(&rq->lock); |
1072 | hrtimer_restart(&rq->hrtick_timer); | |
1073 | rq->hrtick_csd_pending = 0; | |
1074 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1075 | } |
1076 | ||
31656519 PZ |
1077 | /* |
1078 | * Called to set the hrtick timer state. | |
1079 | * | |
1080 | * called with rq->lock held and irqs disabled | |
1081 | */ | |
1082 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1083 | { |
31656519 PZ |
1084 | struct hrtimer *timer = &rq->hrtick_timer; |
1085 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1086 | |
cc584b21 | 1087 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1088 | |
1089 | if (rq == this_rq()) { | |
1090 | hrtimer_restart(timer); | |
1091 | } else if (!rq->hrtick_csd_pending) { | |
1092 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1093 | rq->hrtick_csd_pending = 1; | |
1094 | } | |
b328ca18 PZ |
1095 | } |
1096 | ||
1097 | static int | |
1098 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1099 | { | |
1100 | int cpu = (int)(long)hcpu; | |
1101 | ||
1102 | switch (action) { | |
1103 | case CPU_UP_CANCELED: | |
1104 | case CPU_UP_CANCELED_FROZEN: | |
1105 | case CPU_DOWN_PREPARE: | |
1106 | case CPU_DOWN_PREPARE_FROZEN: | |
1107 | case CPU_DEAD: | |
1108 | case CPU_DEAD_FROZEN: | |
31656519 | 1109 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1110 | return NOTIFY_OK; |
1111 | } | |
1112 | ||
1113 | return NOTIFY_DONE; | |
1114 | } | |
1115 | ||
fa748203 | 1116 | static __init void init_hrtick(void) |
b328ca18 PZ |
1117 | { |
1118 | hotcpu_notifier(hotplug_hrtick, 0); | |
1119 | } | |
31656519 PZ |
1120 | #else |
1121 | /* | |
1122 | * Called to set the hrtick timer state. | |
1123 | * | |
1124 | * called with rq->lock held and irqs disabled | |
1125 | */ | |
1126 | static void hrtick_start(struct rq *rq, u64 delay) | |
1127 | { | |
1128 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1129 | } | |
b328ca18 | 1130 | |
006c75f1 | 1131 | static inline void init_hrtick(void) |
8f4d37ec | 1132 | { |
8f4d37ec | 1133 | } |
31656519 | 1134 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1135 | |
31656519 | 1136 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1137 | { |
31656519 PZ |
1138 | #ifdef CONFIG_SMP |
1139 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1140 | |
31656519 PZ |
1141 | rq->hrtick_csd.flags = 0; |
1142 | rq->hrtick_csd.func = __hrtick_start; | |
1143 | rq->hrtick_csd.info = rq; | |
1144 | #endif | |
8f4d37ec | 1145 | |
31656519 PZ |
1146 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1147 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1148 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1149 | } |
006c75f1 | 1150 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1151 | static inline void hrtick_clear(struct rq *rq) |
1152 | { | |
1153 | } | |
1154 | ||
8f4d37ec PZ |
1155 | static inline void init_rq_hrtick(struct rq *rq) |
1156 | { | |
1157 | } | |
1158 | ||
b328ca18 PZ |
1159 | static inline void init_hrtick(void) |
1160 | { | |
1161 | } | |
006c75f1 | 1162 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1163 | |
c24d20db IM |
1164 | /* |
1165 | * resched_task - mark a task 'to be rescheduled now'. | |
1166 | * | |
1167 | * On UP this means the setting of the need_resched flag, on SMP it | |
1168 | * might also involve a cross-CPU call to trigger the scheduler on | |
1169 | * the target CPU. | |
1170 | */ | |
1171 | #ifdef CONFIG_SMP | |
1172 | ||
1173 | #ifndef tsk_is_polling | |
1174 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1175 | #endif | |
1176 | ||
31656519 | 1177 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1178 | { |
1179 | int cpu; | |
1180 | ||
1181 | assert_spin_locked(&task_rq(p)->lock); | |
1182 | ||
31656519 | 1183 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1184 | return; |
1185 | ||
31656519 | 1186 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1187 | |
1188 | cpu = task_cpu(p); | |
1189 | if (cpu == smp_processor_id()) | |
1190 | return; | |
1191 | ||
1192 | /* NEED_RESCHED must be visible before we test polling */ | |
1193 | smp_mb(); | |
1194 | if (!tsk_is_polling(p)) | |
1195 | smp_send_reschedule(cpu); | |
1196 | } | |
1197 | ||
1198 | static void resched_cpu(int cpu) | |
1199 | { | |
1200 | struct rq *rq = cpu_rq(cpu); | |
1201 | unsigned long flags; | |
1202 | ||
1203 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1204 | return; | |
1205 | resched_task(cpu_curr(cpu)); | |
1206 | spin_unlock_irqrestore(&rq->lock, flags); | |
1207 | } | |
06d8308c TG |
1208 | |
1209 | #ifdef CONFIG_NO_HZ | |
1210 | /* | |
1211 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1212 | * idle CPU then this timer might expire before the next timer event | |
1213 | * which is scheduled to wake up that CPU. In case of a completely | |
1214 | * idle system the next event might even be infinite time into the | |
1215 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1216 | * leaves the inner idle loop so the newly added timer is taken into | |
1217 | * account when the CPU goes back to idle and evaluates the timer | |
1218 | * wheel for the next timer event. | |
1219 | */ | |
1220 | void wake_up_idle_cpu(int cpu) | |
1221 | { | |
1222 | struct rq *rq = cpu_rq(cpu); | |
1223 | ||
1224 | if (cpu == smp_processor_id()) | |
1225 | return; | |
1226 | ||
1227 | /* | |
1228 | * This is safe, as this function is called with the timer | |
1229 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1230 | * to idle and has not yet set rq->curr to idle then it will | |
1231 | * be serialized on the timer wheel base lock and take the new | |
1232 | * timer into account automatically. | |
1233 | */ | |
1234 | if (rq->curr != rq->idle) | |
1235 | return; | |
1236 | ||
1237 | /* | |
1238 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1239 | * lockless. The worst case is that the other CPU runs the | |
1240 | * idle task through an additional NOOP schedule() | |
1241 | */ | |
1242 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1243 | ||
1244 | /* NEED_RESCHED must be visible before we test polling */ | |
1245 | smp_mb(); | |
1246 | if (!tsk_is_polling(rq->idle)) | |
1247 | smp_send_reschedule(cpu); | |
1248 | } | |
6d6bc0ad | 1249 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1250 | |
6d6bc0ad | 1251 | #else /* !CONFIG_SMP */ |
31656519 | 1252 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1253 | { |
1254 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1255 | set_tsk_need_resched(p); |
c24d20db | 1256 | } |
6d6bc0ad | 1257 | #endif /* CONFIG_SMP */ |
c24d20db | 1258 | |
45bf76df IM |
1259 | #if BITS_PER_LONG == 32 |
1260 | # define WMULT_CONST (~0UL) | |
1261 | #else | |
1262 | # define WMULT_CONST (1UL << 32) | |
1263 | #endif | |
1264 | ||
1265 | #define WMULT_SHIFT 32 | |
1266 | ||
194081eb IM |
1267 | /* |
1268 | * Shift right and round: | |
1269 | */ | |
cf2ab469 | 1270 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1271 | |
a7be37ac PZ |
1272 | /* |
1273 | * delta *= weight / lw | |
1274 | */ | |
cb1c4fc9 | 1275 | static unsigned long |
45bf76df IM |
1276 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1277 | struct load_weight *lw) | |
1278 | { | |
1279 | u64 tmp; | |
1280 | ||
7a232e03 LJ |
1281 | if (!lw->inv_weight) { |
1282 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1283 | lw->inv_weight = 1; | |
1284 | else | |
1285 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1286 | / (lw->weight+1); | |
1287 | } | |
45bf76df IM |
1288 | |
1289 | tmp = (u64)delta_exec * weight; | |
1290 | /* | |
1291 | * Check whether we'd overflow the 64-bit multiplication: | |
1292 | */ | |
194081eb | 1293 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1294 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1295 | WMULT_SHIFT/2); |
1296 | else | |
cf2ab469 | 1297 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1298 | |
ecf691da | 1299 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1300 | } |
1301 | ||
1091985b | 1302 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1303 | { |
1304 | lw->weight += inc; | |
e89996ae | 1305 | lw->inv_weight = 0; |
45bf76df IM |
1306 | } |
1307 | ||
1091985b | 1308 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1309 | { |
1310 | lw->weight -= dec; | |
e89996ae | 1311 | lw->inv_weight = 0; |
45bf76df IM |
1312 | } |
1313 | ||
2dd73a4f PW |
1314 | /* |
1315 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1316 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1317 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1318 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1319 | * scaled version of the new time slice allocation that they receive on time |
1320 | * slice expiry etc. | |
1321 | */ | |
1322 | ||
dd41f596 IM |
1323 | #define WEIGHT_IDLEPRIO 2 |
1324 | #define WMULT_IDLEPRIO (1 << 31) | |
1325 | ||
1326 | /* | |
1327 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1328 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1329 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1330 | * that remained on nice 0. | |
1331 | * | |
1332 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1333 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1334 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1335 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1336 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1337 | */ |
1338 | static const int prio_to_weight[40] = { | |
254753dc IM |
1339 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1340 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1341 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1342 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1343 | /* 0 */ 1024, 820, 655, 526, 423, | |
1344 | /* 5 */ 335, 272, 215, 172, 137, | |
1345 | /* 10 */ 110, 87, 70, 56, 45, | |
1346 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1347 | }; |
1348 | ||
5714d2de IM |
1349 | /* |
1350 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1351 | * | |
1352 | * In cases where the weight does not change often, we can use the | |
1353 | * precalculated inverse to speed up arithmetics by turning divisions | |
1354 | * into multiplications: | |
1355 | */ | |
dd41f596 | 1356 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1357 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1358 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1359 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1360 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1361 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1362 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1363 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1364 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1365 | }; |
2dd73a4f | 1366 | |
dd41f596 IM |
1367 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1368 | ||
1369 | /* | |
1370 | * runqueue iterator, to support SMP load-balancing between different | |
1371 | * scheduling classes, without having to expose their internal data | |
1372 | * structures to the load-balancing proper: | |
1373 | */ | |
1374 | struct rq_iterator { | |
1375 | void *arg; | |
1376 | struct task_struct *(*start)(void *); | |
1377 | struct task_struct *(*next)(void *); | |
1378 | }; | |
1379 | ||
e1d1484f PW |
1380 | #ifdef CONFIG_SMP |
1381 | static unsigned long | |
1382 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1383 | unsigned long max_load_move, struct sched_domain *sd, | |
1384 | enum cpu_idle_type idle, int *all_pinned, | |
1385 | int *this_best_prio, struct rq_iterator *iterator); | |
1386 | ||
1387 | static int | |
1388 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1389 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1390 | struct rq_iterator *iterator); | |
e1d1484f | 1391 | #endif |
dd41f596 | 1392 | |
d842de87 SV |
1393 | #ifdef CONFIG_CGROUP_CPUACCT |
1394 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1395 | #else | |
1396 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1397 | #endif | |
1398 | ||
18d95a28 PZ |
1399 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1400 | { | |
1401 | update_load_add(&rq->load, load); | |
1402 | } | |
1403 | ||
1404 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1405 | { | |
1406 | update_load_sub(&rq->load, load); | |
1407 | } | |
1408 | ||
7940ca36 | 1409 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1410 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1411 | |
1412 | /* | |
1413 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1414 | * leaving it for the final time. | |
1415 | */ | |
eb755805 | 1416 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1417 | { |
1418 | struct task_group *parent, *child; | |
eb755805 | 1419 | int ret; |
c09595f6 PZ |
1420 | |
1421 | rcu_read_lock(); | |
1422 | parent = &root_task_group; | |
1423 | down: | |
eb755805 PZ |
1424 | ret = (*down)(parent, data); |
1425 | if (ret) | |
1426 | goto out_unlock; | |
c09595f6 PZ |
1427 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1428 | parent = child; | |
1429 | goto down; | |
1430 | ||
1431 | up: | |
1432 | continue; | |
1433 | } | |
eb755805 PZ |
1434 | ret = (*up)(parent, data); |
1435 | if (ret) | |
1436 | goto out_unlock; | |
c09595f6 PZ |
1437 | |
1438 | child = parent; | |
1439 | parent = parent->parent; | |
1440 | if (parent) | |
1441 | goto up; | |
eb755805 | 1442 | out_unlock: |
c09595f6 | 1443 | rcu_read_unlock(); |
eb755805 PZ |
1444 | |
1445 | return ret; | |
c09595f6 PZ |
1446 | } |
1447 | ||
eb755805 PZ |
1448 | static int tg_nop(struct task_group *tg, void *data) |
1449 | { | |
1450 | return 0; | |
c09595f6 | 1451 | } |
eb755805 PZ |
1452 | #endif |
1453 | ||
1454 | #ifdef CONFIG_SMP | |
1455 | static unsigned long source_load(int cpu, int type); | |
1456 | static unsigned long target_load(int cpu, int type); | |
1457 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1458 | ||
1459 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1460 | { | |
1461 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1462 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1463 | |
4cd42620 SR |
1464 | if (nr_running) |
1465 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1466 | else |
1467 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1468 | |
1469 | return rq->avg_load_per_task; | |
1470 | } | |
1471 | ||
1472 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1473 | |
c09595f6 PZ |
1474 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1475 | ||
1476 | /* | |
1477 | * Calculate and set the cpu's group shares. | |
1478 | */ | |
1479 | static void | |
ffda12a1 PZ |
1480 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1481 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1482 | { |
c09595f6 PZ |
1483 | int boost = 0; |
1484 | unsigned long shares; | |
1485 | unsigned long rq_weight; | |
1486 | ||
c8cba857 | 1487 | if (!tg->se[cpu]) |
c09595f6 PZ |
1488 | return; |
1489 | ||
c8cba857 | 1490 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1491 | |
1492 | /* | |
1493 | * If there are currently no tasks on the cpu pretend there is one of | |
1494 | * average load so that when a new task gets to run here it will not | |
1495 | * get delayed by group starvation. | |
1496 | */ | |
1497 | if (!rq_weight) { | |
1498 | boost = 1; | |
1499 | rq_weight = NICE_0_LOAD; | |
1500 | } | |
1501 | ||
c8cba857 PZ |
1502 | if (unlikely(rq_weight > sd_rq_weight)) |
1503 | rq_weight = sd_rq_weight; | |
1504 | ||
c09595f6 PZ |
1505 | /* |
1506 | * \Sum shares * rq_weight | |
1507 | * shares = ----------------------- | |
1508 | * \Sum rq_weight | |
1509 | * | |
1510 | */ | |
c8cba857 | 1511 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
ffda12a1 | 1512 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1513 | |
ffda12a1 PZ |
1514 | if (abs(shares - tg->se[cpu]->load.weight) > |
1515 | sysctl_sched_shares_thresh) { | |
1516 | struct rq *rq = cpu_rq(cpu); | |
1517 | unsigned long flags; | |
c09595f6 | 1518 | |
ffda12a1 PZ |
1519 | spin_lock_irqsave(&rq->lock, flags); |
1520 | /* | |
1521 | * record the actual number of shares, not the boosted amount. | |
1522 | */ | |
1523 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; | |
1524 | tg->cfs_rq[cpu]->rq_weight = rq_weight; | |
c09595f6 | 1525 | |
ffda12a1 PZ |
1526 | __set_se_shares(tg->se[cpu], shares); |
1527 | spin_unlock_irqrestore(&rq->lock, flags); | |
1528 | } | |
18d95a28 | 1529 | } |
c09595f6 PZ |
1530 | |
1531 | /* | |
c8cba857 PZ |
1532 | * Re-compute the task group their per cpu shares over the given domain. |
1533 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1534 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1535 | */ |
eb755805 | 1536 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1537 | { |
c8cba857 PZ |
1538 | unsigned long rq_weight = 0; |
1539 | unsigned long shares = 0; | |
eb755805 | 1540 | struct sched_domain *sd = data; |
c8cba857 | 1541 | int i; |
c09595f6 | 1542 | |
c8cba857 PZ |
1543 | for_each_cpu_mask(i, sd->span) { |
1544 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1545 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1546 | } |
c09595f6 | 1547 | |
c8cba857 PZ |
1548 | if ((!shares && rq_weight) || shares > tg->shares) |
1549 | shares = tg->shares; | |
1550 | ||
1551 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1552 | shares = tg->shares; | |
c09595f6 | 1553 | |
cd80917e PZ |
1554 | if (!rq_weight) |
1555 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1556 | ||
ffda12a1 PZ |
1557 | for_each_cpu_mask(i, sd->span) |
1558 | update_group_shares_cpu(tg, i, shares, rq_weight); | |
eb755805 PZ |
1559 | |
1560 | return 0; | |
c09595f6 PZ |
1561 | } |
1562 | ||
1563 | /* | |
c8cba857 PZ |
1564 | * Compute the cpu's hierarchical load factor for each task group. |
1565 | * This needs to be done in a top-down fashion because the load of a child | |
1566 | * group is a fraction of its parents load. | |
c09595f6 | 1567 | */ |
eb755805 | 1568 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1569 | { |
c8cba857 | 1570 | unsigned long load; |
eb755805 | 1571 | long cpu = (long)data; |
c09595f6 | 1572 | |
c8cba857 PZ |
1573 | if (!tg->parent) { |
1574 | load = cpu_rq(cpu)->load.weight; | |
1575 | } else { | |
1576 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1577 | load *= tg->cfs_rq[cpu]->shares; | |
1578 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1579 | } | |
c09595f6 | 1580 | |
c8cba857 | 1581 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1582 | |
eb755805 | 1583 | return 0; |
c09595f6 PZ |
1584 | } |
1585 | ||
c8cba857 | 1586 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1587 | { |
2398f2c6 PZ |
1588 | u64 now = cpu_clock(raw_smp_processor_id()); |
1589 | s64 elapsed = now - sd->last_update; | |
1590 | ||
1591 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1592 | sd->last_update = now; | |
eb755805 | 1593 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1594 | } |
4d8d595d PZ |
1595 | } |
1596 | ||
3e5459b4 PZ |
1597 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1598 | { | |
1599 | spin_unlock(&rq->lock); | |
1600 | update_shares(sd); | |
1601 | spin_lock(&rq->lock); | |
1602 | } | |
1603 | ||
eb755805 | 1604 | static void update_h_load(long cpu) |
c09595f6 | 1605 | { |
eb755805 | 1606 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1607 | } |
1608 | ||
c09595f6 PZ |
1609 | #else |
1610 | ||
c8cba857 | 1611 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1612 | { |
1613 | } | |
1614 | ||
3e5459b4 PZ |
1615 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1616 | { | |
1617 | } | |
1618 | ||
18d95a28 PZ |
1619 | #endif |
1620 | ||
18d95a28 PZ |
1621 | #endif |
1622 | ||
30432094 | 1623 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1624 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1625 | { | |
30432094 | 1626 | #ifdef CONFIG_SMP |
34e83e85 IM |
1627 | cfs_rq->shares = shares; |
1628 | #endif | |
1629 | } | |
30432094 | 1630 | #endif |
e7693a36 | 1631 | |
dd41f596 | 1632 | #include "sched_stats.h" |
dd41f596 | 1633 | #include "sched_idletask.c" |
5522d5d5 IM |
1634 | #include "sched_fair.c" |
1635 | #include "sched_rt.c" | |
dd41f596 IM |
1636 | #ifdef CONFIG_SCHED_DEBUG |
1637 | # include "sched_debug.c" | |
1638 | #endif | |
1639 | ||
1640 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1641 | #define for_each_class(class) \ |
1642 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1643 | |
c09595f6 | 1644 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1645 | { |
1646 | rq->nr_running++; | |
9c217245 IM |
1647 | } |
1648 | ||
c09595f6 | 1649 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1650 | { |
1651 | rq->nr_running--; | |
9c217245 IM |
1652 | } |
1653 | ||
45bf76df IM |
1654 | static void set_load_weight(struct task_struct *p) |
1655 | { | |
1656 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1657 | p->se.load.weight = prio_to_weight[0] * 2; |
1658 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1659 | return; | |
1660 | } | |
45bf76df | 1661 | |
dd41f596 IM |
1662 | /* |
1663 | * SCHED_IDLE tasks get minimal weight: | |
1664 | */ | |
1665 | if (p->policy == SCHED_IDLE) { | |
1666 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1667 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1668 | return; | |
1669 | } | |
71f8bd46 | 1670 | |
dd41f596 IM |
1671 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1672 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1673 | } |
1674 | ||
2087a1ad GH |
1675 | static void update_avg(u64 *avg, u64 sample) |
1676 | { | |
1677 | s64 diff = sample - *avg; | |
1678 | *avg += diff >> 3; | |
1679 | } | |
1680 | ||
8159f87e | 1681 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1682 | { |
dd41f596 | 1683 | sched_info_queued(p); |
fd390f6a | 1684 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1685 | p->se.on_rq = 1; |
71f8bd46 IM |
1686 | } |
1687 | ||
69be72c1 | 1688 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1689 | { |
2087a1ad GH |
1690 | if (sleep && p->se.last_wakeup) { |
1691 | update_avg(&p->se.avg_overlap, | |
1692 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1693 | p->se.last_wakeup = 0; | |
1694 | } | |
1695 | ||
46ac22ba | 1696 | sched_info_dequeued(p); |
f02231e5 | 1697 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1698 | p->se.on_rq = 0; |
71f8bd46 IM |
1699 | } |
1700 | ||
14531189 | 1701 | /* |
dd41f596 | 1702 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1703 | */ |
14531189 IM |
1704 | static inline int __normal_prio(struct task_struct *p) |
1705 | { | |
dd41f596 | 1706 | return p->static_prio; |
14531189 IM |
1707 | } |
1708 | ||
b29739f9 IM |
1709 | /* |
1710 | * Calculate the expected normal priority: i.e. priority | |
1711 | * without taking RT-inheritance into account. Might be | |
1712 | * boosted by interactivity modifiers. Changes upon fork, | |
1713 | * setprio syscalls, and whenever the interactivity | |
1714 | * estimator recalculates. | |
1715 | */ | |
36c8b586 | 1716 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1717 | { |
1718 | int prio; | |
1719 | ||
e05606d3 | 1720 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1721 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1722 | else | |
1723 | prio = __normal_prio(p); | |
1724 | return prio; | |
1725 | } | |
1726 | ||
1727 | /* | |
1728 | * Calculate the current priority, i.e. the priority | |
1729 | * taken into account by the scheduler. This value might | |
1730 | * be boosted by RT tasks, or might be boosted by | |
1731 | * interactivity modifiers. Will be RT if the task got | |
1732 | * RT-boosted. If not then it returns p->normal_prio. | |
1733 | */ | |
36c8b586 | 1734 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1735 | { |
1736 | p->normal_prio = normal_prio(p); | |
1737 | /* | |
1738 | * If we are RT tasks or we were boosted to RT priority, | |
1739 | * keep the priority unchanged. Otherwise, update priority | |
1740 | * to the normal priority: | |
1741 | */ | |
1742 | if (!rt_prio(p->prio)) | |
1743 | return p->normal_prio; | |
1744 | return p->prio; | |
1745 | } | |
1746 | ||
1da177e4 | 1747 | /* |
dd41f596 | 1748 | * activate_task - move a task to the runqueue. |
1da177e4 | 1749 | */ |
dd41f596 | 1750 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1751 | { |
d9514f6c | 1752 | if (task_contributes_to_load(p)) |
dd41f596 | 1753 | rq->nr_uninterruptible--; |
1da177e4 | 1754 | |
8159f87e | 1755 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1756 | inc_nr_running(rq); |
1da177e4 LT |
1757 | } |
1758 | ||
1da177e4 LT |
1759 | /* |
1760 | * deactivate_task - remove a task from the runqueue. | |
1761 | */ | |
2e1cb74a | 1762 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1763 | { |
d9514f6c | 1764 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1765 | rq->nr_uninterruptible++; |
1766 | ||
69be72c1 | 1767 | dequeue_task(rq, p, sleep); |
c09595f6 | 1768 | dec_nr_running(rq); |
1da177e4 LT |
1769 | } |
1770 | ||
1da177e4 LT |
1771 | /** |
1772 | * task_curr - is this task currently executing on a CPU? | |
1773 | * @p: the task in question. | |
1774 | */ | |
36c8b586 | 1775 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1776 | { |
1777 | return cpu_curr(task_cpu(p)) == p; | |
1778 | } | |
1779 | ||
dd41f596 IM |
1780 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1781 | { | |
6f505b16 | 1782 | set_task_rq(p, cpu); |
dd41f596 | 1783 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1784 | /* |
1785 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1786 | * successfuly executed on another CPU. We must ensure that updates of | |
1787 | * per-task data have been completed by this moment. | |
1788 | */ | |
1789 | smp_wmb(); | |
dd41f596 | 1790 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1791 | #endif |
2dd73a4f PW |
1792 | } |
1793 | ||
cb469845 SR |
1794 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1795 | const struct sched_class *prev_class, | |
1796 | int oldprio, int running) | |
1797 | { | |
1798 | if (prev_class != p->sched_class) { | |
1799 | if (prev_class->switched_from) | |
1800 | prev_class->switched_from(rq, p, running); | |
1801 | p->sched_class->switched_to(rq, p, running); | |
1802 | } else | |
1803 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1804 | } | |
1805 | ||
1da177e4 | 1806 | #ifdef CONFIG_SMP |
c65cc870 | 1807 | |
e958b360 TG |
1808 | /* Used instead of source_load when we know the type == 0 */ |
1809 | static unsigned long weighted_cpuload(const int cpu) | |
1810 | { | |
1811 | return cpu_rq(cpu)->load.weight; | |
1812 | } | |
1813 | ||
cc367732 IM |
1814 | /* |
1815 | * Is this task likely cache-hot: | |
1816 | */ | |
e7693a36 | 1817 | static int |
cc367732 IM |
1818 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1819 | { | |
1820 | s64 delta; | |
1821 | ||
f540a608 IM |
1822 | /* |
1823 | * Buddy candidates are cache hot: | |
1824 | */ | |
4793241b PZ |
1825 | if (sched_feat(CACHE_HOT_BUDDY) && |
1826 | (&p->se == cfs_rq_of(&p->se)->next || | |
1827 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1828 | return 1; |
1829 | ||
cc367732 IM |
1830 | if (p->sched_class != &fair_sched_class) |
1831 | return 0; | |
1832 | ||
6bc1665b IM |
1833 | if (sysctl_sched_migration_cost == -1) |
1834 | return 1; | |
1835 | if (sysctl_sched_migration_cost == 0) | |
1836 | return 0; | |
1837 | ||
cc367732 IM |
1838 | delta = now - p->se.exec_start; |
1839 | ||
1840 | return delta < (s64)sysctl_sched_migration_cost; | |
1841 | } | |
1842 | ||
1843 | ||
dd41f596 | 1844 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1845 | { |
dd41f596 IM |
1846 | int old_cpu = task_cpu(p); |
1847 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1848 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1849 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1850 | u64 clock_offset; |
dd41f596 IM |
1851 | |
1852 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1853 | |
1854 | #ifdef CONFIG_SCHEDSTATS | |
1855 | if (p->se.wait_start) | |
1856 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1857 | if (p->se.sleep_start) |
1858 | p->se.sleep_start -= clock_offset; | |
1859 | if (p->se.block_start) | |
1860 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1861 | if (old_cpu != new_cpu) { |
1862 | schedstat_inc(p, se.nr_migrations); | |
1863 | if (task_hot(p, old_rq->clock, NULL)) | |
1864 | schedstat_inc(p, se.nr_forced2_migrations); | |
1865 | } | |
6cfb0d5d | 1866 | #endif |
2830cf8c SV |
1867 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1868 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1869 | |
1870 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1871 | } |
1872 | ||
70b97a7f | 1873 | struct migration_req { |
1da177e4 | 1874 | struct list_head list; |
1da177e4 | 1875 | |
36c8b586 | 1876 | struct task_struct *task; |
1da177e4 LT |
1877 | int dest_cpu; |
1878 | ||
1da177e4 | 1879 | struct completion done; |
70b97a7f | 1880 | }; |
1da177e4 LT |
1881 | |
1882 | /* | |
1883 | * The task's runqueue lock must be held. | |
1884 | * Returns true if you have to wait for migration thread. | |
1885 | */ | |
36c8b586 | 1886 | static int |
70b97a7f | 1887 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1888 | { |
70b97a7f | 1889 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1890 | |
1891 | /* | |
1892 | * If the task is not on a runqueue (and not running), then | |
1893 | * it is sufficient to simply update the task's cpu field. | |
1894 | */ | |
dd41f596 | 1895 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1896 | set_task_cpu(p, dest_cpu); |
1897 | return 0; | |
1898 | } | |
1899 | ||
1900 | init_completion(&req->done); | |
1da177e4 LT |
1901 | req->task = p; |
1902 | req->dest_cpu = dest_cpu; | |
1903 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1904 | |
1da177e4 LT |
1905 | return 1; |
1906 | } | |
1907 | ||
1908 | /* | |
1909 | * wait_task_inactive - wait for a thread to unschedule. | |
1910 | * | |
85ba2d86 RM |
1911 | * If @match_state is nonzero, it's the @p->state value just checked and |
1912 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1913 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1914 | * we return a positive number (its total switch count). If a second call | |
1915 | * a short while later returns the same number, the caller can be sure that | |
1916 | * @p has remained unscheduled the whole time. | |
1917 | * | |
1da177e4 LT |
1918 | * The caller must ensure that the task *will* unschedule sometime soon, |
1919 | * else this function might spin for a *long* time. This function can't | |
1920 | * be called with interrupts off, or it may introduce deadlock with | |
1921 | * smp_call_function() if an IPI is sent by the same process we are | |
1922 | * waiting to become inactive. | |
1923 | */ | |
85ba2d86 | 1924 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1925 | { |
1926 | unsigned long flags; | |
dd41f596 | 1927 | int running, on_rq; |
85ba2d86 | 1928 | unsigned long ncsw; |
70b97a7f | 1929 | struct rq *rq; |
1da177e4 | 1930 | |
3a5c359a AK |
1931 | for (;;) { |
1932 | /* | |
1933 | * We do the initial early heuristics without holding | |
1934 | * any task-queue locks at all. We'll only try to get | |
1935 | * the runqueue lock when things look like they will | |
1936 | * work out! | |
1937 | */ | |
1938 | rq = task_rq(p); | |
fa490cfd | 1939 | |
3a5c359a AK |
1940 | /* |
1941 | * If the task is actively running on another CPU | |
1942 | * still, just relax and busy-wait without holding | |
1943 | * any locks. | |
1944 | * | |
1945 | * NOTE! Since we don't hold any locks, it's not | |
1946 | * even sure that "rq" stays as the right runqueue! | |
1947 | * But we don't care, since "task_running()" will | |
1948 | * return false if the runqueue has changed and p | |
1949 | * is actually now running somewhere else! | |
1950 | */ | |
85ba2d86 RM |
1951 | while (task_running(rq, p)) { |
1952 | if (match_state && unlikely(p->state != match_state)) | |
1953 | return 0; | |
3a5c359a | 1954 | cpu_relax(); |
85ba2d86 | 1955 | } |
fa490cfd | 1956 | |
3a5c359a AK |
1957 | /* |
1958 | * Ok, time to look more closely! We need the rq | |
1959 | * lock now, to be *sure*. If we're wrong, we'll | |
1960 | * just go back and repeat. | |
1961 | */ | |
1962 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 1963 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
1964 | running = task_running(rq, p); |
1965 | on_rq = p->se.on_rq; | |
85ba2d86 | 1966 | ncsw = 0; |
f31e11d8 | 1967 | if (!match_state || p->state == match_state) |
93dcf55f | 1968 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1969 | task_rq_unlock(rq, &flags); |
fa490cfd | 1970 | |
85ba2d86 RM |
1971 | /* |
1972 | * If it changed from the expected state, bail out now. | |
1973 | */ | |
1974 | if (unlikely(!ncsw)) | |
1975 | break; | |
1976 | ||
3a5c359a AK |
1977 | /* |
1978 | * Was it really running after all now that we | |
1979 | * checked with the proper locks actually held? | |
1980 | * | |
1981 | * Oops. Go back and try again.. | |
1982 | */ | |
1983 | if (unlikely(running)) { | |
1984 | cpu_relax(); | |
1985 | continue; | |
1986 | } | |
fa490cfd | 1987 | |
3a5c359a AK |
1988 | /* |
1989 | * It's not enough that it's not actively running, | |
1990 | * it must be off the runqueue _entirely_, and not | |
1991 | * preempted! | |
1992 | * | |
1993 | * So if it wa still runnable (but just not actively | |
1994 | * running right now), it's preempted, and we should | |
1995 | * yield - it could be a while. | |
1996 | */ | |
1997 | if (unlikely(on_rq)) { | |
1998 | schedule_timeout_uninterruptible(1); | |
1999 | continue; | |
2000 | } | |
fa490cfd | 2001 | |
3a5c359a AK |
2002 | /* |
2003 | * Ahh, all good. It wasn't running, and it wasn't | |
2004 | * runnable, which means that it will never become | |
2005 | * running in the future either. We're all done! | |
2006 | */ | |
2007 | break; | |
2008 | } | |
85ba2d86 RM |
2009 | |
2010 | return ncsw; | |
1da177e4 LT |
2011 | } |
2012 | ||
2013 | /*** | |
2014 | * kick_process - kick a running thread to enter/exit the kernel | |
2015 | * @p: the to-be-kicked thread | |
2016 | * | |
2017 | * Cause a process which is running on another CPU to enter | |
2018 | * kernel-mode, without any delay. (to get signals handled.) | |
2019 | * | |
2020 | * NOTE: this function doesnt have to take the runqueue lock, | |
2021 | * because all it wants to ensure is that the remote task enters | |
2022 | * the kernel. If the IPI races and the task has been migrated | |
2023 | * to another CPU then no harm is done and the purpose has been | |
2024 | * achieved as well. | |
2025 | */ | |
36c8b586 | 2026 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2027 | { |
2028 | int cpu; | |
2029 | ||
2030 | preempt_disable(); | |
2031 | cpu = task_cpu(p); | |
2032 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2033 | smp_send_reschedule(cpu); | |
2034 | preempt_enable(); | |
2035 | } | |
2036 | ||
2037 | /* | |
2dd73a4f PW |
2038 | * Return a low guess at the load of a migration-source cpu weighted |
2039 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2040 | * |
2041 | * We want to under-estimate the load of migration sources, to | |
2042 | * balance conservatively. | |
2043 | */ | |
a9957449 | 2044 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2045 | { |
70b97a7f | 2046 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2047 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2048 | |
93b75217 | 2049 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2050 | return total; |
b910472d | 2051 | |
dd41f596 | 2052 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2053 | } |
2054 | ||
2055 | /* | |
2dd73a4f PW |
2056 | * Return a high guess at the load of a migration-target cpu weighted |
2057 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2058 | */ |
a9957449 | 2059 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2060 | { |
70b97a7f | 2061 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2062 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2063 | |
93b75217 | 2064 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2065 | return total; |
3b0bd9bc | 2066 | |
dd41f596 | 2067 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2068 | } |
2069 | ||
147cbb4b NP |
2070 | /* |
2071 | * find_idlest_group finds and returns the least busy CPU group within the | |
2072 | * domain. | |
2073 | */ | |
2074 | static struct sched_group * | |
2075 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2076 | { | |
2077 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2078 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2079 | int load_idx = sd->forkexec_idx; | |
2080 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2081 | ||
2082 | do { | |
2083 | unsigned long load, avg_load; | |
2084 | int local_group; | |
2085 | int i; | |
2086 | ||
da5a5522 BD |
2087 | /* Skip over this group if it has no CPUs allowed */ |
2088 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2089 | continue; |
da5a5522 | 2090 | |
147cbb4b | 2091 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2092 | |
2093 | /* Tally up the load of all CPUs in the group */ | |
2094 | avg_load = 0; | |
2095 | ||
363ab6f1 | 2096 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2097 | /* Bias balancing toward cpus of our domain */ |
2098 | if (local_group) | |
2099 | load = source_load(i, load_idx); | |
2100 | else | |
2101 | load = target_load(i, load_idx); | |
2102 | ||
2103 | avg_load += load; | |
2104 | } | |
2105 | ||
2106 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2107 | avg_load = sg_div_cpu_power(group, |
2108 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2109 | |
2110 | if (local_group) { | |
2111 | this_load = avg_load; | |
2112 | this = group; | |
2113 | } else if (avg_load < min_load) { | |
2114 | min_load = avg_load; | |
2115 | idlest = group; | |
2116 | } | |
3a5c359a | 2117 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2118 | |
2119 | if (!idlest || 100*this_load < imbalance*min_load) | |
2120 | return NULL; | |
2121 | return idlest; | |
2122 | } | |
2123 | ||
2124 | /* | |
0feaece9 | 2125 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2126 | */ |
95cdf3b7 | 2127 | static int |
7c16ec58 MT |
2128 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2129 | cpumask_t *tmp) | |
147cbb4b NP |
2130 | { |
2131 | unsigned long load, min_load = ULONG_MAX; | |
2132 | int idlest = -1; | |
2133 | int i; | |
2134 | ||
da5a5522 | 2135 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2136 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2137 | |
363ab6f1 | 2138 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2139 | load = weighted_cpuload(i); |
147cbb4b NP |
2140 | |
2141 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2142 | min_load = load; | |
2143 | idlest = i; | |
2144 | } | |
2145 | } | |
2146 | ||
2147 | return idlest; | |
2148 | } | |
2149 | ||
476d139c NP |
2150 | /* |
2151 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2152 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2153 | * SD_BALANCE_EXEC. | |
2154 | * | |
2155 | * Balance, ie. select the least loaded group. | |
2156 | * | |
2157 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2158 | * | |
2159 | * preempt must be disabled. | |
2160 | */ | |
2161 | static int sched_balance_self(int cpu, int flag) | |
2162 | { | |
2163 | struct task_struct *t = current; | |
2164 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2165 | |
c96d145e | 2166 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2167 | /* |
2168 | * If power savings logic is enabled for a domain, stop there. | |
2169 | */ | |
5c45bf27 SS |
2170 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2171 | break; | |
476d139c NP |
2172 | if (tmp->flags & flag) |
2173 | sd = tmp; | |
c96d145e | 2174 | } |
476d139c | 2175 | |
039a1c41 PZ |
2176 | if (sd) |
2177 | update_shares(sd); | |
2178 | ||
476d139c | 2179 | while (sd) { |
7c16ec58 | 2180 | cpumask_t span, tmpmask; |
476d139c | 2181 | struct sched_group *group; |
1a848870 SS |
2182 | int new_cpu, weight; |
2183 | ||
2184 | if (!(sd->flags & flag)) { | |
2185 | sd = sd->child; | |
2186 | continue; | |
2187 | } | |
476d139c NP |
2188 | |
2189 | span = sd->span; | |
2190 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2191 | if (!group) { |
2192 | sd = sd->child; | |
2193 | continue; | |
2194 | } | |
476d139c | 2195 | |
7c16ec58 | 2196 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2197 | if (new_cpu == -1 || new_cpu == cpu) { |
2198 | /* Now try balancing at a lower domain level of cpu */ | |
2199 | sd = sd->child; | |
2200 | continue; | |
2201 | } | |
476d139c | 2202 | |
1a848870 | 2203 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2204 | cpu = new_cpu; |
476d139c NP |
2205 | sd = NULL; |
2206 | weight = cpus_weight(span); | |
2207 | for_each_domain(cpu, tmp) { | |
2208 | if (weight <= cpus_weight(tmp->span)) | |
2209 | break; | |
2210 | if (tmp->flags & flag) | |
2211 | sd = tmp; | |
2212 | } | |
2213 | /* while loop will break here if sd == NULL */ | |
2214 | } | |
2215 | ||
2216 | return cpu; | |
2217 | } | |
2218 | ||
2219 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2220 | |
1da177e4 LT |
2221 | /*** |
2222 | * try_to_wake_up - wake up a thread | |
2223 | * @p: the to-be-woken-up thread | |
2224 | * @state: the mask of task states that can be woken | |
2225 | * @sync: do a synchronous wakeup? | |
2226 | * | |
2227 | * Put it on the run-queue if it's not already there. The "current" | |
2228 | * thread is always on the run-queue (except when the actual | |
2229 | * re-schedule is in progress), and as such you're allowed to do | |
2230 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2231 | * runnable without the overhead of this. | |
2232 | * | |
2233 | * returns failure only if the task is already active. | |
2234 | */ | |
36c8b586 | 2235 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2236 | { |
cc367732 | 2237 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2238 | unsigned long flags; |
2239 | long old_state; | |
70b97a7f | 2240 | struct rq *rq; |
1da177e4 | 2241 | |
b85d0667 IM |
2242 | if (!sched_feat(SYNC_WAKEUPS)) |
2243 | sync = 0; | |
2244 | ||
2398f2c6 PZ |
2245 | #ifdef CONFIG_SMP |
2246 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2247 | struct sched_domain *sd; | |
2248 | ||
2249 | this_cpu = raw_smp_processor_id(); | |
2250 | cpu = task_cpu(p); | |
2251 | ||
2252 | for_each_domain(this_cpu, sd) { | |
2253 | if (cpu_isset(cpu, sd->span)) { | |
2254 | update_shares(sd); | |
2255 | break; | |
2256 | } | |
2257 | } | |
2258 | } | |
2259 | #endif | |
2260 | ||
04e2f174 | 2261 | smp_wmb(); |
1da177e4 LT |
2262 | rq = task_rq_lock(p, &flags); |
2263 | old_state = p->state; | |
2264 | if (!(old_state & state)) | |
2265 | goto out; | |
2266 | ||
dd41f596 | 2267 | if (p->se.on_rq) |
1da177e4 LT |
2268 | goto out_running; |
2269 | ||
2270 | cpu = task_cpu(p); | |
cc367732 | 2271 | orig_cpu = cpu; |
1da177e4 LT |
2272 | this_cpu = smp_processor_id(); |
2273 | ||
2274 | #ifdef CONFIG_SMP | |
2275 | if (unlikely(task_running(rq, p))) | |
2276 | goto out_activate; | |
2277 | ||
5d2f5a61 DA |
2278 | cpu = p->sched_class->select_task_rq(p, sync); |
2279 | if (cpu != orig_cpu) { | |
2280 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2281 | task_rq_unlock(rq, &flags); |
2282 | /* might preempt at this point */ | |
2283 | rq = task_rq_lock(p, &flags); | |
2284 | old_state = p->state; | |
2285 | if (!(old_state & state)) | |
2286 | goto out; | |
dd41f596 | 2287 | if (p->se.on_rq) |
1da177e4 LT |
2288 | goto out_running; |
2289 | ||
2290 | this_cpu = smp_processor_id(); | |
2291 | cpu = task_cpu(p); | |
2292 | } | |
2293 | ||
e7693a36 GH |
2294 | #ifdef CONFIG_SCHEDSTATS |
2295 | schedstat_inc(rq, ttwu_count); | |
2296 | if (cpu == this_cpu) | |
2297 | schedstat_inc(rq, ttwu_local); | |
2298 | else { | |
2299 | struct sched_domain *sd; | |
2300 | for_each_domain(this_cpu, sd) { | |
2301 | if (cpu_isset(cpu, sd->span)) { | |
2302 | schedstat_inc(sd, ttwu_wake_remote); | |
2303 | break; | |
2304 | } | |
2305 | } | |
2306 | } | |
6d6bc0ad | 2307 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2308 | |
1da177e4 LT |
2309 | out_activate: |
2310 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2311 | schedstat_inc(p, se.nr_wakeups); |
2312 | if (sync) | |
2313 | schedstat_inc(p, se.nr_wakeups_sync); | |
2314 | if (orig_cpu != cpu) | |
2315 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2316 | if (cpu == this_cpu) | |
2317 | schedstat_inc(p, se.nr_wakeups_local); | |
2318 | else | |
2319 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2320 | update_rq_clock(rq); |
dd41f596 | 2321 | activate_task(rq, p, 1); |
1da177e4 LT |
2322 | success = 1; |
2323 | ||
2324 | out_running: | |
0a16b607 | 2325 | trace_sched_wakeup(rq, p); |
15afe09b | 2326 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2327 | |
1da177e4 | 2328 | p->state = TASK_RUNNING; |
9a897c5a SR |
2329 | #ifdef CONFIG_SMP |
2330 | if (p->sched_class->task_wake_up) | |
2331 | p->sched_class->task_wake_up(rq, p); | |
2332 | #endif | |
1da177e4 | 2333 | out: |
2087a1ad GH |
2334 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2335 | ||
1da177e4 LT |
2336 | task_rq_unlock(rq, &flags); |
2337 | ||
2338 | return success; | |
2339 | } | |
2340 | ||
7ad5b3a5 | 2341 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2342 | { |
d9514f6c | 2343 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2344 | } |
1da177e4 LT |
2345 | EXPORT_SYMBOL(wake_up_process); |
2346 | ||
7ad5b3a5 | 2347 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2348 | { |
2349 | return try_to_wake_up(p, state, 0); | |
2350 | } | |
2351 | ||
1da177e4 LT |
2352 | /* |
2353 | * Perform scheduler related setup for a newly forked process p. | |
2354 | * p is forked by current. | |
dd41f596 IM |
2355 | * |
2356 | * __sched_fork() is basic setup used by init_idle() too: | |
2357 | */ | |
2358 | static void __sched_fork(struct task_struct *p) | |
2359 | { | |
dd41f596 IM |
2360 | p->se.exec_start = 0; |
2361 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2362 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2363 | p->se.last_wakeup = 0; |
2364 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2365 | |
2366 | #ifdef CONFIG_SCHEDSTATS | |
2367 | p->se.wait_start = 0; | |
dd41f596 IM |
2368 | p->se.sum_sleep_runtime = 0; |
2369 | p->se.sleep_start = 0; | |
dd41f596 IM |
2370 | p->se.block_start = 0; |
2371 | p->se.sleep_max = 0; | |
2372 | p->se.block_max = 0; | |
2373 | p->se.exec_max = 0; | |
eba1ed4b | 2374 | p->se.slice_max = 0; |
dd41f596 | 2375 | p->se.wait_max = 0; |
6cfb0d5d | 2376 | #endif |
476d139c | 2377 | |
fa717060 | 2378 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2379 | p->se.on_rq = 0; |
4a55bd5e | 2380 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2381 | |
e107be36 AK |
2382 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2383 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2384 | #endif | |
2385 | ||
1da177e4 LT |
2386 | /* |
2387 | * We mark the process as running here, but have not actually | |
2388 | * inserted it onto the runqueue yet. This guarantees that | |
2389 | * nobody will actually run it, and a signal or other external | |
2390 | * event cannot wake it up and insert it on the runqueue either. | |
2391 | */ | |
2392 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * fork()/clone()-time setup: | |
2397 | */ | |
2398 | void sched_fork(struct task_struct *p, int clone_flags) | |
2399 | { | |
2400 | int cpu = get_cpu(); | |
2401 | ||
2402 | __sched_fork(p); | |
2403 | ||
2404 | #ifdef CONFIG_SMP | |
2405 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2406 | #endif | |
02e4bac2 | 2407 | set_task_cpu(p, cpu); |
b29739f9 IM |
2408 | |
2409 | /* | |
2410 | * Make sure we do not leak PI boosting priority to the child: | |
2411 | */ | |
2412 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2413 | if (!rt_prio(p->prio)) |
2414 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2415 | |
52f17b6c | 2416 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2417 | if (likely(sched_info_on())) |
52f17b6c | 2418 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2419 | #endif |
d6077cb8 | 2420 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2421 | p->oncpu = 0; |
2422 | #endif | |
1da177e4 | 2423 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2424 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2425 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2426 | #endif |
476d139c | 2427 | put_cpu(); |
1da177e4 LT |
2428 | } |
2429 | ||
2430 | /* | |
2431 | * wake_up_new_task - wake up a newly created task for the first time. | |
2432 | * | |
2433 | * This function will do some initial scheduler statistics housekeeping | |
2434 | * that must be done for every newly created context, then puts the task | |
2435 | * on the runqueue and wakes it. | |
2436 | */ | |
7ad5b3a5 | 2437 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2438 | { |
2439 | unsigned long flags; | |
dd41f596 | 2440 | struct rq *rq; |
1da177e4 LT |
2441 | |
2442 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2443 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2444 | update_rq_clock(rq); |
1da177e4 LT |
2445 | |
2446 | p->prio = effective_prio(p); | |
2447 | ||
b9dca1e0 | 2448 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2449 | activate_task(rq, p, 0); |
1da177e4 | 2450 | } else { |
1da177e4 | 2451 | /* |
dd41f596 IM |
2452 | * Let the scheduling class do new task startup |
2453 | * management (if any): | |
1da177e4 | 2454 | */ |
ee0827d8 | 2455 | p->sched_class->task_new(rq, p); |
c09595f6 | 2456 | inc_nr_running(rq); |
1da177e4 | 2457 | } |
0a16b607 | 2458 | trace_sched_wakeup_new(rq, p); |
15afe09b | 2459 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2460 | #ifdef CONFIG_SMP |
2461 | if (p->sched_class->task_wake_up) | |
2462 | p->sched_class->task_wake_up(rq, p); | |
2463 | #endif | |
dd41f596 | 2464 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2465 | } |
2466 | ||
e107be36 AK |
2467 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2468 | ||
2469 | /** | |
421cee29 RD |
2470 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2471 | * @notifier: notifier struct to register | |
e107be36 AK |
2472 | */ |
2473 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2474 | { | |
2475 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2476 | } | |
2477 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2478 | ||
2479 | /** | |
2480 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2481 | * @notifier: notifier struct to unregister |
e107be36 AK |
2482 | * |
2483 | * This is safe to call from within a preemption notifier. | |
2484 | */ | |
2485 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2486 | { | |
2487 | hlist_del(¬ifier->link); | |
2488 | } | |
2489 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2490 | ||
2491 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2492 | { | |
2493 | struct preempt_notifier *notifier; | |
2494 | struct hlist_node *node; | |
2495 | ||
2496 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2497 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2498 | } | |
2499 | ||
2500 | static void | |
2501 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2502 | struct task_struct *next) | |
2503 | { | |
2504 | struct preempt_notifier *notifier; | |
2505 | struct hlist_node *node; | |
2506 | ||
2507 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2508 | notifier->ops->sched_out(notifier, next); | |
2509 | } | |
2510 | ||
6d6bc0ad | 2511 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2512 | |
2513 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2514 | { | |
2515 | } | |
2516 | ||
2517 | static void | |
2518 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2519 | struct task_struct *next) | |
2520 | { | |
2521 | } | |
2522 | ||
6d6bc0ad | 2523 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2524 | |
4866cde0 NP |
2525 | /** |
2526 | * prepare_task_switch - prepare to switch tasks | |
2527 | * @rq: the runqueue preparing to switch | |
421cee29 | 2528 | * @prev: the current task that is being switched out |
4866cde0 NP |
2529 | * @next: the task we are going to switch to. |
2530 | * | |
2531 | * This is called with the rq lock held and interrupts off. It must | |
2532 | * be paired with a subsequent finish_task_switch after the context | |
2533 | * switch. | |
2534 | * | |
2535 | * prepare_task_switch sets up locking and calls architecture specific | |
2536 | * hooks. | |
2537 | */ | |
e107be36 AK |
2538 | static inline void |
2539 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2540 | struct task_struct *next) | |
4866cde0 | 2541 | { |
e107be36 | 2542 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2543 | prepare_lock_switch(rq, next); |
2544 | prepare_arch_switch(next); | |
2545 | } | |
2546 | ||
1da177e4 LT |
2547 | /** |
2548 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2549 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2550 | * @prev: the thread we just switched away from. |
2551 | * | |
4866cde0 NP |
2552 | * finish_task_switch must be called after the context switch, paired |
2553 | * with a prepare_task_switch call before the context switch. | |
2554 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2555 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2556 | * |
2557 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2558 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2559 | * with the lock held can cause deadlocks; see schedule() for |
2560 | * details.) | |
2561 | */ | |
a9957449 | 2562 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2563 | __releases(rq->lock) |
2564 | { | |
1da177e4 | 2565 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2566 | long prev_state; |
1da177e4 LT |
2567 | |
2568 | rq->prev_mm = NULL; | |
2569 | ||
2570 | /* | |
2571 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2572 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2573 | * schedule one last time. The schedule call will never return, and |
2574 | * the scheduled task must drop that reference. | |
c394cc9f | 2575 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2576 | * still held, otherwise prev could be scheduled on another cpu, die |
2577 | * there before we look at prev->state, and then the reference would | |
2578 | * be dropped twice. | |
2579 | * Manfred Spraul <manfred@colorfullife.com> | |
2580 | */ | |
55a101f8 | 2581 | prev_state = prev->state; |
4866cde0 NP |
2582 | finish_arch_switch(prev); |
2583 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2584 | #ifdef CONFIG_SMP |
2585 | if (current->sched_class->post_schedule) | |
2586 | current->sched_class->post_schedule(rq); | |
2587 | #endif | |
e8fa1362 | 2588 | |
e107be36 | 2589 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2590 | if (mm) |
2591 | mmdrop(mm); | |
c394cc9f | 2592 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2593 | /* |
2594 | * Remove function-return probe instances associated with this | |
2595 | * task and put them back on the free list. | |
9761eea8 | 2596 | */ |
c6fd91f0 | 2597 | kprobe_flush_task(prev); |
1da177e4 | 2598 | put_task_struct(prev); |
c6fd91f0 | 2599 | } |
1da177e4 LT |
2600 | } |
2601 | ||
2602 | /** | |
2603 | * schedule_tail - first thing a freshly forked thread must call. | |
2604 | * @prev: the thread we just switched away from. | |
2605 | */ | |
36c8b586 | 2606 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2607 | __releases(rq->lock) |
2608 | { | |
70b97a7f IM |
2609 | struct rq *rq = this_rq(); |
2610 | ||
4866cde0 NP |
2611 | finish_task_switch(rq, prev); |
2612 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2613 | /* In this case, finish_task_switch does not reenable preemption */ | |
2614 | preempt_enable(); | |
2615 | #endif | |
1da177e4 | 2616 | if (current->set_child_tid) |
b488893a | 2617 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2618 | } |
2619 | ||
2620 | /* | |
2621 | * context_switch - switch to the new MM and the new | |
2622 | * thread's register state. | |
2623 | */ | |
dd41f596 | 2624 | static inline void |
70b97a7f | 2625 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2626 | struct task_struct *next) |
1da177e4 | 2627 | { |
dd41f596 | 2628 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2629 | |
e107be36 | 2630 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2631 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2632 | mm = next->mm; |
2633 | oldmm = prev->active_mm; | |
9226d125 ZA |
2634 | /* |
2635 | * For paravirt, this is coupled with an exit in switch_to to | |
2636 | * combine the page table reload and the switch backend into | |
2637 | * one hypercall. | |
2638 | */ | |
2639 | arch_enter_lazy_cpu_mode(); | |
2640 | ||
dd41f596 | 2641 | if (unlikely(!mm)) { |
1da177e4 LT |
2642 | next->active_mm = oldmm; |
2643 | atomic_inc(&oldmm->mm_count); | |
2644 | enter_lazy_tlb(oldmm, next); | |
2645 | } else | |
2646 | switch_mm(oldmm, mm, next); | |
2647 | ||
dd41f596 | 2648 | if (unlikely(!prev->mm)) { |
1da177e4 | 2649 | prev->active_mm = NULL; |
1da177e4 LT |
2650 | rq->prev_mm = oldmm; |
2651 | } | |
3a5f5e48 IM |
2652 | /* |
2653 | * Since the runqueue lock will be released by the next | |
2654 | * task (which is an invalid locking op but in the case | |
2655 | * of the scheduler it's an obvious special-case), so we | |
2656 | * do an early lockdep release here: | |
2657 | */ | |
2658 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2659 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2660 | #endif |
1da177e4 LT |
2661 | |
2662 | /* Here we just switch the register state and the stack. */ | |
2663 | switch_to(prev, next, prev); | |
2664 | ||
dd41f596 IM |
2665 | barrier(); |
2666 | /* | |
2667 | * this_rq must be evaluated again because prev may have moved | |
2668 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2669 | * frame will be invalid. | |
2670 | */ | |
2671 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2672 | } |
2673 | ||
2674 | /* | |
2675 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2676 | * | |
2677 | * externally visible scheduler statistics: current number of runnable | |
2678 | * threads, current number of uninterruptible-sleeping threads, total | |
2679 | * number of context switches performed since bootup. | |
2680 | */ | |
2681 | unsigned long nr_running(void) | |
2682 | { | |
2683 | unsigned long i, sum = 0; | |
2684 | ||
2685 | for_each_online_cpu(i) | |
2686 | sum += cpu_rq(i)->nr_running; | |
2687 | ||
2688 | return sum; | |
2689 | } | |
2690 | ||
2691 | unsigned long nr_uninterruptible(void) | |
2692 | { | |
2693 | unsigned long i, sum = 0; | |
2694 | ||
0a945022 | 2695 | for_each_possible_cpu(i) |
1da177e4 LT |
2696 | sum += cpu_rq(i)->nr_uninterruptible; |
2697 | ||
2698 | /* | |
2699 | * Since we read the counters lockless, it might be slightly | |
2700 | * inaccurate. Do not allow it to go below zero though: | |
2701 | */ | |
2702 | if (unlikely((long)sum < 0)) | |
2703 | sum = 0; | |
2704 | ||
2705 | return sum; | |
2706 | } | |
2707 | ||
2708 | unsigned long long nr_context_switches(void) | |
2709 | { | |
cc94abfc SR |
2710 | int i; |
2711 | unsigned long long sum = 0; | |
1da177e4 | 2712 | |
0a945022 | 2713 | for_each_possible_cpu(i) |
1da177e4 LT |
2714 | sum += cpu_rq(i)->nr_switches; |
2715 | ||
2716 | return sum; | |
2717 | } | |
2718 | ||
2719 | unsigned long nr_iowait(void) | |
2720 | { | |
2721 | unsigned long i, sum = 0; | |
2722 | ||
0a945022 | 2723 | for_each_possible_cpu(i) |
1da177e4 LT |
2724 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2725 | ||
2726 | return sum; | |
2727 | } | |
2728 | ||
db1b1fef JS |
2729 | unsigned long nr_active(void) |
2730 | { | |
2731 | unsigned long i, running = 0, uninterruptible = 0; | |
2732 | ||
2733 | for_each_online_cpu(i) { | |
2734 | running += cpu_rq(i)->nr_running; | |
2735 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2736 | } | |
2737 | ||
2738 | if (unlikely((long)uninterruptible < 0)) | |
2739 | uninterruptible = 0; | |
2740 | ||
2741 | return running + uninterruptible; | |
2742 | } | |
2743 | ||
48f24c4d | 2744 | /* |
dd41f596 IM |
2745 | * Update rq->cpu_load[] statistics. This function is usually called every |
2746 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2747 | */ |
dd41f596 | 2748 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2749 | { |
495eca49 | 2750 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2751 | int i, scale; |
2752 | ||
2753 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2754 | |
2755 | /* Update our load: */ | |
2756 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2757 | unsigned long old_load, new_load; | |
2758 | ||
2759 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2760 | ||
2761 | old_load = this_rq->cpu_load[i]; | |
2762 | new_load = this_load; | |
a25707f3 IM |
2763 | /* |
2764 | * Round up the averaging division if load is increasing. This | |
2765 | * prevents us from getting stuck on 9 if the load is 10, for | |
2766 | * example. | |
2767 | */ | |
2768 | if (new_load > old_load) | |
2769 | new_load += scale-1; | |
dd41f596 IM |
2770 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2771 | } | |
48f24c4d IM |
2772 | } |
2773 | ||
dd41f596 IM |
2774 | #ifdef CONFIG_SMP |
2775 | ||
1da177e4 LT |
2776 | /* |
2777 | * double_rq_lock - safely lock two runqueues | |
2778 | * | |
2779 | * Note this does not disable interrupts like task_rq_lock, | |
2780 | * you need to do so manually before calling. | |
2781 | */ | |
70b97a7f | 2782 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2783 | __acquires(rq1->lock) |
2784 | __acquires(rq2->lock) | |
2785 | { | |
054b9108 | 2786 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2787 | if (rq1 == rq2) { |
2788 | spin_lock(&rq1->lock); | |
2789 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2790 | } else { | |
c96d145e | 2791 | if (rq1 < rq2) { |
1da177e4 | 2792 | spin_lock(&rq1->lock); |
5e710e37 | 2793 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2794 | } else { |
2795 | spin_lock(&rq2->lock); | |
5e710e37 | 2796 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2797 | } |
2798 | } | |
6e82a3be IM |
2799 | update_rq_clock(rq1); |
2800 | update_rq_clock(rq2); | |
1da177e4 LT |
2801 | } |
2802 | ||
2803 | /* | |
2804 | * double_rq_unlock - safely unlock two runqueues | |
2805 | * | |
2806 | * Note this does not restore interrupts like task_rq_unlock, | |
2807 | * you need to do so manually after calling. | |
2808 | */ | |
70b97a7f | 2809 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2810 | __releases(rq1->lock) |
2811 | __releases(rq2->lock) | |
2812 | { | |
2813 | spin_unlock(&rq1->lock); | |
2814 | if (rq1 != rq2) | |
2815 | spin_unlock(&rq2->lock); | |
2816 | else | |
2817 | __release(rq2->lock); | |
2818 | } | |
2819 | ||
2820 | /* | |
2821 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2822 | */ | |
e8fa1362 | 2823 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2824 | __releases(this_rq->lock) |
2825 | __acquires(busiest->lock) | |
2826 | __acquires(this_rq->lock) | |
2827 | { | |
e8fa1362 SR |
2828 | int ret = 0; |
2829 | ||
054b9108 KK |
2830 | if (unlikely(!irqs_disabled())) { |
2831 | /* printk() doesn't work good under rq->lock */ | |
2832 | spin_unlock(&this_rq->lock); | |
2833 | BUG_ON(1); | |
2834 | } | |
1da177e4 | 2835 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2836 | if (busiest < this_rq) { |
1da177e4 LT |
2837 | spin_unlock(&this_rq->lock); |
2838 | spin_lock(&busiest->lock); | |
5e710e37 | 2839 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2840 | ret = 1; |
1da177e4 | 2841 | } else |
5e710e37 | 2842 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2843 | } |
e8fa1362 | 2844 | return ret; |
1da177e4 LT |
2845 | } |
2846 | ||
1b12bbc7 PZ |
2847 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2848 | __releases(busiest->lock) | |
2849 | { | |
2850 | spin_unlock(&busiest->lock); | |
2851 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2852 | } | |
2853 | ||
1da177e4 LT |
2854 | /* |
2855 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2856 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2857 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2858 | * the cpu_allowed mask is restored. |
2859 | */ | |
36c8b586 | 2860 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2861 | { |
70b97a7f | 2862 | struct migration_req req; |
1da177e4 | 2863 | unsigned long flags; |
70b97a7f | 2864 | struct rq *rq; |
1da177e4 LT |
2865 | |
2866 | rq = task_rq_lock(p, &flags); | |
2867 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2868 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2869 | goto out; |
2870 | ||
0a16b607 | 2871 | trace_sched_migrate_task(rq, p, dest_cpu); |
1da177e4 LT |
2872 | /* force the process onto the specified CPU */ |
2873 | if (migrate_task(p, dest_cpu, &req)) { | |
2874 | /* Need to wait for migration thread (might exit: take ref). */ | |
2875 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2876 | |
1da177e4 LT |
2877 | get_task_struct(mt); |
2878 | task_rq_unlock(rq, &flags); | |
2879 | wake_up_process(mt); | |
2880 | put_task_struct(mt); | |
2881 | wait_for_completion(&req.done); | |
36c8b586 | 2882 | |
1da177e4 LT |
2883 | return; |
2884 | } | |
2885 | out: | |
2886 | task_rq_unlock(rq, &flags); | |
2887 | } | |
2888 | ||
2889 | /* | |
476d139c NP |
2890 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2891 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2892 | */ |
2893 | void sched_exec(void) | |
2894 | { | |
1da177e4 | 2895 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2896 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2897 | put_cpu(); |
476d139c NP |
2898 | if (new_cpu != this_cpu) |
2899 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2900 | } |
2901 | ||
2902 | /* | |
2903 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2904 | * Both runqueues must be locked. | |
2905 | */ | |
dd41f596 IM |
2906 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2907 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2908 | { |
2e1cb74a | 2909 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2910 | set_task_cpu(p, this_cpu); |
dd41f596 | 2911 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2912 | /* |
2913 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2914 | * to be always true for them. | |
2915 | */ | |
15afe09b | 2916 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2917 | } |
2918 | ||
2919 | /* | |
2920 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2921 | */ | |
858119e1 | 2922 | static |
70b97a7f | 2923 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2924 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2925 | int *all_pinned) |
1da177e4 LT |
2926 | { |
2927 | /* | |
2928 | * We do not migrate tasks that are: | |
2929 | * 1) running (obviously), or | |
2930 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2931 | * 3) are cache-hot on their current CPU. | |
2932 | */ | |
cc367732 IM |
2933 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2934 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2935 | return 0; |
cc367732 | 2936 | } |
81026794 NP |
2937 | *all_pinned = 0; |
2938 | ||
cc367732 IM |
2939 | if (task_running(rq, p)) { |
2940 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2941 | return 0; |
cc367732 | 2942 | } |
1da177e4 | 2943 | |
da84d961 IM |
2944 | /* |
2945 | * Aggressive migration if: | |
2946 | * 1) task is cache cold, or | |
2947 | * 2) too many balance attempts have failed. | |
2948 | */ | |
2949 | ||
6bc1665b IM |
2950 | if (!task_hot(p, rq->clock, sd) || |
2951 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2952 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2953 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2954 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2955 | schedstat_inc(p, se.nr_forced_migrations); |
2956 | } | |
da84d961 IM |
2957 | #endif |
2958 | return 1; | |
2959 | } | |
2960 | ||
cc367732 IM |
2961 | if (task_hot(p, rq->clock, sd)) { |
2962 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2963 | return 0; |
cc367732 | 2964 | } |
1da177e4 LT |
2965 | return 1; |
2966 | } | |
2967 | ||
e1d1484f PW |
2968 | static unsigned long |
2969 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2970 | unsigned long max_load_move, struct sched_domain *sd, | |
2971 | enum cpu_idle_type idle, int *all_pinned, | |
2972 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2973 | { |
051c6764 | 2974 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2975 | struct task_struct *p; |
2976 | long rem_load_move = max_load_move; | |
1da177e4 | 2977 | |
e1d1484f | 2978 | if (max_load_move == 0) |
1da177e4 LT |
2979 | goto out; |
2980 | ||
81026794 NP |
2981 | pinned = 1; |
2982 | ||
1da177e4 | 2983 | /* |
dd41f596 | 2984 | * Start the load-balancing iterator: |
1da177e4 | 2985 | */ |
dd41f596 IM |
2986 | p = iterator->start(iterator->arg); |
2987 | next: | |
b82d9fdd | 2988 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2989 | goto out; |
051c6764 PZ |
2990 | |
2991 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2992 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2993 | p = iterator->next(iterator->arg); |
2994 | goto next; | |
1da177e4 LT |
2995 | } |
2996 | ||
dd41f596 | 2997 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2998 | pulled++; |
dd41f596 | 2999 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3000 | |
2dd73a4f | 3001 | /* |
b82d9fdd | 3002 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3003 | */ |
e1d1484f | 3004 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3005 | if (p->prio < *this_best_prio) |
3006 | *this_best_prio = p->prio; | |
dd41f596 IM |
3007 | p = iterator->next(iterator->arg); |
3008 | goto next; | |
1da177e4 LT |
3009 | } |
3010 | out: | |
3011 | /* | |
e1d1484f | 3012 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3013 | * so we can safely collect pull_task() stats here rather than |
3014 | * inside pull_task(). | |
3015 | */ | |
3016 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3017 | |
3018 | if (all_pinned) | |
3019 | *all_pinned = pinned; | |
e1d1484f PW |
3020 | |
3021 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3022 | } |
3023 | ||
dd41f596 | 3024 | /* |
43010659 PW |
3025 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3026 | * this_rq, as part of a balancing operation within domain "sd". | |
3027 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3028 | * |
3029 | * Called with both runqueues locked. | |
3030 | */ | |
3031 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3032 | unsigned long max_load_move, |
dd41f596 IM |
3033 | struct sched_domain *sd, enum cpu_idle_type idle, |
3034 | int *all_pinned) | |
3035 | { | |
5522d5d5 | 3036 | const struct sched_class *class = sched_class_highest; |
43010659 | 3037 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3038 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3039 | |
3040 | do { | |
43010659 PW |
3041 | total_load_moved += |
3042 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3043 | max_load_move - total_load_moved, |
a4ac01c3 | 3044 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3045 | class = class->next; |
c4acb2c0 GH |
3046 | |
3047 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3048 | break; | |
3049 | ||
43010659 | 3050 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3051 | |
43010659 PW |
3052 | return total_load_moved > 0; |
3053 | } | |
3054 | ||
e1d1484f PW |
3055 | static int |
3056 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3057 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3058 | struct rq_iterator *iterator) | |
3059 | { | |
3060 | struct task_struct *p = iterator->start(iterator->arg); | |
3061 | int pinned = 0; | |
3062 | ||
3063 | while (p) { | |
3064 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3065 | pull_task(busiest, p, this_rq, this_cpu); | |
3066 | /* | |
3067 | * Right now, this is only the second place pull_task() | |
3068 | * is called, so we can safely collect pull_task() | |
3069 | * stats here rather than inside pull_task(). | |
3070 | */ | |
3071 | schedstat_inc(sd, lb_gained[idle]); | |
3072 | ||
3073 | return 1; | |
3074 | } | |
3075 | p = iterator->next(iterator->arg); | |
3076 | } | |
3077 | ||
3078 | return 0; | |
3079 | } | |
3080 | ||
43010659 PW |
3081 | /* |
3082 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3083 | * part of active balancing operations within "domain". | |
3084 | * Returns 1 if successful and 0 otherwise. | |
3085 | * | |
3086 | * Called with both runqueues locked. | |
3087 | */ | |
3088 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3089 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3090 | { | |
5522d5d5 | 3091 | const struct sched_class *class; |
43010659 PW |
3092 | |
3093 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3094 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3095 | return 1; |
3096 | ||
3097 | return 0; | |
dd41f596 IM |
3098 | } |
3099 | ||
1da177e4 LT |
3100 | /* |
3101 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3102 | * domain. It calculates and returns the amount of weighted load which |
3103 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3104 | */ |
3105 | static struct sched_group * | |
3106 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3107 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3108 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3109 | { |
3110 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3111 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3112 | unsigned long max_pull; |
2dd73a4f PW |
3113 | unsigned long busiest_load_per_task, busiest_nr_running; |
3114 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3115 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3116 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3117 | int power_savings_balance = 1; | |
3118 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3119 | unsigned long min_nr_running = ULONG_MAX; | |
3120 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3121 | #endif | |
1da177e4 LT |
3122 | |
3123 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3124 | busiest_load_per_task = busiest_nr_running = 0; |
3125 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3126 | |
d15bcfdb | 3127 | if (idle == CPU_NOT_IDLE) |
7897986b | 3128 | load_idx = sd->busy_idx; |
d15bcfdb | 3129 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3130 | load_idx = sd->newidle_idx; |
3131 | else | |
3132 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3133 | |
3134 | do { | |
908a7c1b | 3135 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3136 | int local_group; |
3137 | int i; | |
908a7c1b | 3138 | int __group_imb = 0; |
783609c6 | 3139 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3140 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3141 | unsigned long sum_avg_load_per_task; |
3142 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3143 | |
3144 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3145 | ||
783609c6 SS |
3146 | if (local_group) |
3147 | balance_cpu = first_cpu(group->cpumask); | |
3148 | ||
1da177e4 | 3149 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3150 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3151 | sum_avg_load_per_task = avg_load_per_task = 0; |
3152 | ||
908a7c1b KC |
3153 | max_cpu_load = 0; |
3154 | min_cpu_load = ~0UL; | |
1da177e4 | 3155 | |
363ab6f1 | 3156 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3157 | struct rq *rq; |
3158 | ||
3159 | if (!cpu_isset(i, *cpus)) | |
3160 | continue; | |
3161 | ||
3162 | rq = cpu_rq(i); | |
2dd73a4f | 3163 | |
9439aab8 | 3164 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3165 | *sd_idle = 0; |
3166 | ||
1da177e4 | 3167 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3168 | if (local_group) { |
3169 | if (idle_cpu(i) && !first_idle_cpu) { | |
3170 | first_idle_cpu = 1; | |
3171 | balance_cpu = i; | |
3172 | } | |
3173 | ||
a2000572 | 3174 | load = target_load(i, load_idx); |
908a7c1b | 3175 | } else { |
a2000572 | 3176 | load = source_load(i, load_idx); |
908a7c1b KC |
3177 | if (load > max_cpu_load) |
3178 | max_cpu_load = load; | |
3179 | if (min_cpu_load > load) | |
3180 | min_cpu_load = load; | |
3181 | } | |
1da177e4 LT |
3182 | |
3183 | avg_load += load; | |
2dd73a4f | 3184 | sum_nr_running += rq->nr_running; |
dd41f596 | 3185 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3186 | |
3187 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3188 | } |
3189 | ||
783609c6 SS |
3190 | /* |
3191 | * First idle cpu or the first cpu(busiest) in this sched group | |
3192 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3193 | * domains. In the newly idle case, we will allow all the cpu's |
3194 | * to do the newly idle load balance. | |
783609c6 | 3195 | */ |
9439aab8 SS |
3196 | if (idle != CPU_NEWLY_IDLE && local_group && |
3197 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3198 | *balance = 0; |
3199 | goto ret; | |
3200 | } | |
3201 | ||
1da177e4 | 3202 | total_load += avg_load; |
5517d86b | 3203 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3204 | |
3205 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3206 | avg_load = sg_div_cpu_power(group, |
3207 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3208 | |
408ed066 PZ |
3209 | |
3210 | /* | |
3211 | * Consider the group unbalanced when the imbalance is larger | |
3212 | * than the average weight of two tasks. | |
3213 | * | |
3214 | * APZ: with cgroup the avg task weight can vary wildly and | |
3215 | * might not be a suitable number - should we keep a | |
3216 | * normalized nr_running number somewhere that negates | |
3217 | * the hierarchy? | |
3218 | */ | |
3219 | avg_load_per_task = sg_div_cpu_power(group, | |
3220 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3221 | ||
3222 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3223 | __group_imb = 1; |
3224 | ||
5517d86b | 3225 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3226 | |
1da177e4 LT |
3227 | if (local_group) { |
3228 | this_load = avg_load; | |
3229 | this = group; | |
2dd73a4f PW |
3230 | this_nr_running = sum_nr_running; |
3231 | this_load_per_task = sum_weighted_load; | |
3232 | } else if (avg_load > max_load && | |
908a7c1b | 3233 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3234 | max_load = avg_load; |
3235 | busiest = group; | |
2dd73a4f PW |
3236 | busiest_nr_running = sum_nr_running; |
3237 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3238 | group_imb = __group_imb; |
1da177e4 | 3239 | } |
5c45bf27 SS |
3240 | |
3241 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3242 | /* | |
3243 | * Busy processors will not participate in power savings | |
3244 | * balance. | |
3245 | */ | |
dd41f596 IM |
3246 | if (idle == CPU_NOT_IDLE || |
3247 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3248 | goto group_next; | |
5c45bf27 SS |
3249 | |
3250 | /* | |
3251 | * If the local group is idle or completely loaded | |
3252 | * no need to do power savings balance at this domain | |
3253 | */ | |
3254 | if (local_group && (this_nr_running >= group_capacity || | |
3255 | !this_nr_running)) | |
3256 | power_savings_balance = 0; | |
3257 | ||
dd41f596 | 3258 | /* |
5c45bf27 SS |
3259 | * If a group is already running at full capacity or idle, |
3260 | * don't include that group in power savings calculations | |
dd41f596 IM |
3261 | */ |
3262 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3263 | || !sum_nr_running) |
dd41f596 | 3264 | goto group_next; |
5c45bf27 | 3265 | |
dd41f596 | 3266 | /* |
5c45bf27 | 3267 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3268 | * This is the group from where we need to pick up the load |
3269 | * for saving power | |
3270 | */ | |
3271 | if ((sum_nr_running < min_nr_running) || | |
3272 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3273 | first_cpu(group->cpumask) < |
3274 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3275 | group_min = group; |
3276 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3277 | min_load_per_task = sum_weighted_load / |
3278 | sum_nr_running; | |
dd41f596 | 3279 | } |
5c45bf27 | 3280 | |
dd41f596 | 3281 | /* |
5c45bf27 | 3282 | * Calculate the group which is almost near its |
dd41f596 IM |
3283 | * capacity but still has some space to pick up some load |
3284 | * from other group and save more power | |
3285 | */ | |
3286 | if (sum_nr_running <= group_capacity - 1) { | |
3287 | if (sum_nr_running > leader_nr_running || | |
3288 | (sum_nr_running == leader_nr_running && | |
3289 | first_cpu(group->cpumask) > | |
3290 | first_cpu(group_leader->cpumask))) { | |
3291 | group_leader = group; | |
3292 | leader_nr_running = sum_nr_running; | |
3293 | } | |
48f24c4d | 3294 | } |
5c45bf27 SS |
3295 | group_next: |
3296 | #endif | |
1da177e4 LT |
3297 | group = group->next; |
3298 | } while (group != sd->groups); | |
3299 | ||
2dd73a4f | 3300 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3301 | goto out_balanced; |
3302 | ||
3303 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3304 | ||
3305 | if (this_load >= avg_load || | |
3306 | 100*max_load <= sd->imbalance_pct*this_load) | |
3307 | goto out_balanced; | |
3308 | ||
2dd73a4f | 3309 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3310 | if (group_imb) |
3311 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3312 | ||
1da177e4 LT |
3313 | /* |
3314 | * We're trying to get all the cpus to the average_load, so we don't | |
3315 | * want to push ourselves above the average load, nor do we wish to | |
3316 | * reduce the max loaded cpu below the average load, as either of these | |
3317 | * actions would just result in more rebalancing later, and ping-pong | |
3318 | * tasks around. Thus we look for the minimum possible imbalance. | |
3319 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3320 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3321 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3322 | * appear as very large values with unsigned longs. |
3323 | */ | |
2dd73a4f PW |
3324 | if (max_load <= busiest_load_per_task) |
3325 | goto out_balanced; | |
3326 | ||
3327 | /* | |
3328 | * In the presence of smp nice balancing, certain scenarios can have | |
3329 | * max load less than avg load(as we skip the groups at or below | |
3330 | * its cpu_power, while calculating max_load..) | |
3331 | */ | |
3332 | if (max_load < avg_load) { | |
3333 | *imbalance = 0; | |
3334 | goto small_imbalance; | |
3335 | } | |
0c117f1b SS |
3336 | |
3337 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3338 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3339 | |
1da177e4 | 3340 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3341 | *imbalance = min(max_pull * busiest->__cpu_power, |
3342 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3343 | / SCHED_LOAD_SCALE; |
3344 | ||
2dd73a4f PW |
3345 | /* |
3346 | * if *imbalance is less than the average load per runnable task | |
3347 | * there is no gaurantee that any tasks will be moved so we'll have | |
3348 | * a think about bumping its value to force at least one task to be | |
3349 | * moved | |
3350 | */ | |
7fd0d2dd | 3351 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3352 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3353 | unsigned int imbn; |
3354 | ||
3355 | small_imbalance: | |
3356 | pwr_move = pwr_now = 0; | |
3357 | imbn = 2; | |
3358 | if (this_nr_running) { | |
3359 | this_load_per_task /= this_nr_running; | |
3360 | if (busiest_load_per_task > this_load_per_task) | |
3361 | imbn = 1; | |
3362 | } else | |
408ed066 | 3363 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3364 | |
01c8c57d | 3365 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3366 | busiest_load_per_task * imbn) { |
2dd73a4f | 3367 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3368 | return busiest; |
3369 | } | |
3370 | ||
3371 | /* | |
3372 | * OK, we don't have enough imbalance to justify moving tasks, | |
3373 | * however we may be able to increase total CPU power used by | |
3374 | * moving them. | |
3375 | */ | |
3376 | ||
5517d86b ED |
3377 | pwr_now += busiest->__cpu_power * |
3378 | min(busiest_load_per_task, max_load); | |
3379 | pwr_now += this->__cpu_power * | |
3380 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3381 | pwr_now /= SCHED_LOAD_SCALE; |
3382 | ||
3383 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3384 | tmp = sg_div_cpu_power(busiest, |
3385 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3386 | if (max_load > tmp) |
5517d86b | 3387 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3388 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3389 | |
3390 | /* Amount of load we'd add */ | |
5517d86b | 3391 | if (max_load * busiest->__cpu_power < |
33859f7f | 3392 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3393 | tmp = sg_div_cpu_power(this, |
3394 | max_load * busiest->__cpu_power); | |
1da177e4 | 3395 | else |
5517d86b ED |
3396 | tmp = sg_div_cpu_power(this, |
3397 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3398 | pwr_move += this->__cpu_power * | |
3399 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3400 | pwr_move /= SCHED_LOAD_SCALE; |
3401 | ||
3402 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3403 | if (pwr_move > pwr_now) |
3404 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3405 | } |
3406 | ||
1da177e4 LT |
3407 | return busiest; |
3408 | ||
3409 | out_balanced: | |
5c45bf27 | 3410 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3411 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3412 | goto ret; |
1da177e4 | 3413 | |
5c45bf27 SS |
3414 | if (this == group_leader && group_leader != group_min) { |
3415 | *imbalance = min_load_per_task; | |
3416 | return group_min; | |
3417 | } | |
5c45bf27 | 3418 | #endif |
783609c6 | 3419 | ret: |
1da177e4 LT |
3420 | *imbalance = 0; |
3421 | return NULL; | |
3422 | } | |
3423 | ||
3424 | /* | |
3425 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3426 | */ | |
70b97a7f | 3427 | static struct rq * |
d15bcfdb | 3428 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3429 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3430 | { |
70b97a7f | 3431 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3432 | unsigned long max_load = 0; |
1da177e4 LT |
3433 | int i; |
3434 | ||
363ab6f1 | 3435 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3436 | unsigned long wl; |
0a2966b4 CL |
3437 | |
3438 | if (!cpu_isset(i, *cpus)) | |
3439 | continue; | |
3440 | ||
48f24c4d | 3441 | rq = cpu_rq(i); |
dd41f596 | 3442 | wl = weighted_cpuload(i); |
2dd73a4f | 3443 | |
dd41f596 | 3444 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3445 | continue; |
1da177e4 | 3446 | |
dd41f596 IM |
3447 | if (wl > max_load) { |
3448 | max_load = wl; | |
48f24c4d | 3449 | busiest = rq; |
1da177e4 LT |
3450 | } |
3451 | } | |
3452 | ||
3453 | return busiest; | |
3454 | } | |
3455 | ||
77391d71 NP |
3456 | /* |
3457 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3458 | * so long as it is large enough. | |
3459 | */ | |
3460 | #define MAX_PINNED_INTERVAL 512 | |
3461 | ||
1da177e4 LT |
3462 | /* |
3463 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3464 | * tasks if there is an imbalance. | |
1da177e4 | 3465 | */ |
70b97a7f | 3466 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3467 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3468 | int *balance, cpumask_t *cpus) |
1da177e4 | 3469 | { |
43010659 | 3470 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3471 | struct sched_group *group; |
1da177e4 | 3472 | unsigned long imbalance; |
70b97a7f | 3473 | struct rq *busiest; |
fe2eea3f | 3474 | unsigned long flags; |
5969fe06 | 3475 | |
7c16ec58 MT |
3476 | cpus_setall(*cpus); |
3477 | ||
89c4710e SS |
3478 | /* |
3479 | * When power savings policy is enabled for the parent domain, idle | |
3480 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3481 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3482 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3483 | */ |
d15bcfdb | 3484 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3485 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3486 | sd_idle = 1; |
1da177e4 | 3487 | |
2d72376b | 3488 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3489 | |
0a2966b4 | 3490 | redo: |
c8cba857 | 3491 | update_shares(sd); |
0a2966b4 | 3492 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3493 | cpus, balance); |
783609c6 | 3494 | |
06066714 | 3495 | if (*balance == 0) |
783609c6 | 3496 | goto out_balanced; |
783609c6 | 3497 | |
1da177e4 LT |
3498 | if (!group) { |
3499 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3500 | goto out_balanced; | |
3501 | } | |
3502 | ||
7c16ec58 | 3503 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3504 | if (!busiest) { |
3505 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3506 | goto out_balanced; | |
3507 | } | |
3508 | ||
db935dbd | 3509 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3510 | |
3511 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3512 | ||
43010659 | 3513 | ld_moved = 0; |
1da177e4 LT |
3514 | if (busiest->nr_running > 1) { |
3515 | /* | |
3516 | * Attempt to move tasks. If find_busiest_group has found | |
3517 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3518 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3519 | * correctly treated as an imbalance. |
3520 | */ | |
fe2eea3f | 3521 | local_irq_save(flags); |
e17224bf | 3522 | double_rq_lock(this_rq, busiest); |
43010659 | 3523 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3524 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3525 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3526 | local_irq_restore(flags); |
81026794 | 3527 | |
46cb4b7c SS |
3528 | /* |
3529 | * some other cpu did the load balance for us. | |
3530 | */ | |
43010659 | 3531 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3532 | resched_cpu(this_cpu); |
3533 | ||
81026794 | 3534 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3535 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3536 | cpu_clear(cpu_of(busiest), *cpus); |
3537 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3538 | goto redo; |
81026794 | 3539 | goto out_balanced; |
0a2966b4 | 3540 | } |
1da177e4 | 3541 | } |
81026794 | 3542 | |
43010659 | 3543 | if (!ld_moved) { |
1da177e4 LT |
3544 | schedstat_inc(sd, lb_failed[idle]); |
3545 | sd->nr_balance_failed++; | |
3546 | ||
3547 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3548 | |
fe2eea3f | 3549 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3550 | |
3551 | /* don't kick the migration_thread, if the curr | |
3552 | * task on busiest cpu can't be moved to this_cpu | |
3553 | */ | |
3554 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3555 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3556 | all_pinned = 1; |
3557 | goto out_one_pinned; | |
3558 | } | |
3559 | ||
1da177e4 LT |
3560 | if (!busiest->active_balance) { |
3561 | busiest->active_balance = 1; | |
3562 | busiest->push_cpu = this_cpu; | |
81026794 | 3563 | active_balance = 1; |
1da177e4 | 3564 | } |
fe2eea3f | 3565 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3566 | if (active_balance) |
1da177e4 LT |
3567 | wake_up_process(busiest->migration_thread); |
3568 | ||
3569 | /* | |
3570 | * We've kicked active balancing, reset the failure | |
3571 | * counter. | |
3572 | */ | |
39507451 | 3573 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3574 | } |
81026794 | 3575 | } else |
1da177e4 LT |
3576 | sd->nr_balance_failed = 0; |
3577 | ||
81026794 | 3578 | if (likely(!active_balance)) { |
1da177e4 LT |
3579 | /* We were unbalanced, so reset the balancing interval */ |
3580 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3581 | } else { |
3582 | /* | |
3583 | * If we've begun active balancing, start to back off. This | |
3584 | * case may not be covered by the all_pinned logic if there | |
3585 | * is only 1 task on the busy runqueue (because we don't call | |
3586 | * move_tasks). | |
3587 | */ | |
3588 | if (sd->balance_interval < sd->max_interval) | |
3589 | sd->balance_interval *= 2; | |
1da177e4 LT |
3590 | } |
3591 | ||
43010659 | 3592 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3593 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3594 | ld_moved = -1; |
3595 | ||
3596 | goto out; | |
1da177e4 LT |
3597 | |
3598 | out_balanced: | |
1da177e4 LT |
3599 | schedstat_inc(sd, lb_balanced[idle]); |
3600 | ||
16cfb1c0 | 3601 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3602 | |
3603 | out_one_pinned: | |
1da177e4 | 3604 | /* tune up the balancing interval */ |
77391d71 NP |
3605 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3606 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3607 | sd->balance_interval *= 2; |
3608 | ||
48f24c4d | 3609 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3610 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3611 | ld_moved = -1; |
3612 | else | |
3613 | ld_moved = 0; | |
3614 | out: | |
c8cba857 PZ |
3615 | if (ld_moved) |
3616 | update_shares(sd); | |
c09595f6 | 3617 | return ld_moved; |
1da177e4 LT |
3618 | } |
3619 | ||
3620 | /* | |
3621 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3622 | * tasks if there is an imbalance. | |
3623 | * | |
d15bcfdb | 3624 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3625 | * this_rq is locked. |
3626 | */ | |
48f24c4d | 3627 | static int |
7c16ec58 MT |
3628 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3629 | cpumask_t *cpus) | |
1da177e4 LT |
3630 | { |
3631 | struct sched_group *group; | |
70b97a7f | 3632 | struct rq *busiest = NULL; |
1da177e4 | 3633 | unsigned long imbalance; |
43010659 | 3634 | int ld_moved = 0; |
5969fe06 | 3635 | int sd_idle = 0; |
969bb4e4 | 3636 | int all_pinned = 0; |
7c16ec58 MT |
3637 | |
3638 | cpus_setall(*cpus); | |
5969fe06 | 3639 | |
89c4710e SS |
3640 | /* |
3641 | * When power savings policy is enabled for the parent domain, idle | |
3642 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3643 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3644 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3645 | */ |
3646 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3647 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3648 | sd_idle = 1; |
1da177e4 | 3649 | |
2d72376b | 3650 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3651 | redo: |
3e5459b4 | 3652 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3653 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3654 | &sd_idle, cpus, NULL); |
1da177e4 | 3655 | if (!group) { |
d15bcfdb | 3656 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3657 | goto out_balanced; |
1da177e4 LT |
3658 | } |
3659 | ||
7c16ec58 | 3660 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3661 | if (!busiest) { |
d15bcfdb | 3662 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3663 | goto out_balanced; |
1da177e4 LT |
3664 | } |
3665 | ||
db935dbd NP |
3666 | BUG_ON(busiest == this_rq); |
3667 | ||
d15bcfdb | 3668 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3669 | |
43010659 | 3670 | ld_moved = 0; |
d6d5cfaf NP |
3671 | if (busiest->nr_running > 1) { |
3672 | /* Attempt to move tasks */ | |
3673 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3674 | /* this_rq->clock is already updated */ |
3675 | update_rq_clock(busiest); | |
43010659 | 3676 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3677 | imbalance, sd, CPU_NEWLY_IDLE, |
3678 | &all_pinned); | |
1b12bbc7 | 3679 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3680 | |
969bb4e4 | 3681 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3682 | cpu_clear(cpu_of(busiest), *cpus); |
3683 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3684 | goto redo; |
3685 | } | |
d6d5cfaf NP |
3686 | } |
3687 | ||
43010659 | 3688 | if (!ld_moved) { |
d15bcfdb | 3689 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3690 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3691 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3692 | return -1; |
3693 | } else | |
16cfb1c0 | 3694 | sd->nr_balance_failed = 0; |
1da177e4 | 3695 | |
3e5459b4 | 3696 | update_shares_locked(this_rq, sd); |
43010659 | 3697 | return ld_moved; |
16cfb1c0 NP |
3698 | |
3699 | out_balanced: | |
d15bcfdb | 3700 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3701 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3702 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3703 | return -1; |
16cfb1c0 | 3704 | sd->nr_balance_failed = 0; |
48f24c4d | 3705 | |
16cfb1c0 | 3706 | return 0; |
1da177e4 LT |
3707 | } |
3708 | ||
3709 | /* | |
3710 | * idle_balance is called by schedule() if this_cpu is about to become | |
3711 | * idle. Attempts to pull tasks from other CPUs. | |
3712 | */ | |
70b97a7f | 3713 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3714 | { |
3715 | struct sched_domain *sd; | |
dd41f596 IM |
3716 | int pulled_task = -1; |
3717 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3718 | cpumask_t tmpmask; |
1da177e4 LT |
3719 | |
3720 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3721 | unsigned long interval; |
3722 | ||
3723 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3724 | continue; | |
3725 | ||
3726 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3727 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3728 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3729 | sd, &tmpmask); | |
92c4ca5c CL |
3730 | |
3731 | interval = msecs_to_jiffies(sd->balance_interval); | |
3732 | if (time_after(next_balance, sd->last_balance + interval)) | |
3733 | next_balance = sd->last_balance + interval; | |
3734 | if (pulled_task) | |
3735 | break; | |
1da177e4 | 3736 | } |
dd41f596 | 3737 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3738 | /* |
3739 | * We are going idle. next_balance may be set based on | |
3740 | * a busy processor. So reset next_balance. | |
3741 | */ | |
3742 | this_rq->next_balance = next_balance; | |
dd41f596 | 3743 | } |
1da177e4 LT |
3744 | } |
3745 | ||
3746 | /* | |
3747 | * active_load_balance is run by migration threads. It pushes running tasks | |
3748 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3749 | * running on each physical CPU where possible, and avoids physical / | |
3750 | * logical imbalances. | |
3751 | * | |
3752 | * Called with busiest_rq locked. | |
3753 | */ | |
70b97a7f | 3754 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3755 | { |
39507451 | 3756 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3757 | struct sched_domain *sd; |
3758 | struct rq *target_rq; | |
39507451 | 3759 | |
48f24c4d | 3760 | /* Is there any task to move? */ |
39507451 | 3761 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3762 | return; |
3763 | ||
3764 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3765 | |
3766 | /* | |
39507451 | 3767 | * This condition is "impossible", if it occurs |
41a2d6cf | 3768 | * we need to fix it. Originally reported by |
39507451 | 3769 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3770 | */ |
39507451 | 3771 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3772 | |
39507451 NP |
3773 | /* move a task from busiest_rq to target_rq */ |
3774 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3775 | update_rq_clock(busiest_rq); |
3776 | update_rq_clock(target_rq); | |
39507451 NP |
3777 | |
3778 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3779 | for_each_domain(target_cpu, sd) { |
39507451 | 3780 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3781 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3782 | break; |
c96d145e | 3783 | } |
39507451 | 3784 | |
48f24c4d | 3785 | if (likely(sd)) { |
2d72376b | 3786 | schedstat_inc(sd, alb_count); |
39507451 | 3787 | |
43010659 PW |
3788 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3789 | sd, CPU_IDLE)) | |
48f24c4d IM |
3790 | schedstat_inc(sd, alb_pushed); |
3791 | else | |
3792 | schedstat_inc(sd, alb_failed); | |
3793 | } | |
1b12bbc7 | 3794 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3795 | } |
3796 | ||
46cb4b7c SS |
3797 | #ifdef CONFIG_NO_HZ |
3798 | static struct { | |
3799 | atomic_t load_balancer; | |
41a2d6cf | 3800 | cpumask_t cpu_mask; |
46cb4b7c SS |
3801 | } nohz ____cacheline_aligned = { |
3802 | .load_balancer = ATOMIC_INIT(-1), | |
3803 | .cpu_mask = CPU_MASK_NONE, | |
3804 | }; | |
3805 | ||
7835b98b | 3806 | /* |
46cb4b7c SS |
3807 | * This routine will try to nominate the ilb (idle load balancing) |
3808 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3809 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3810 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3811 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3812 | * arrives... | |
3813 | * | |
3814 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3815 | * for idle load balancing. ilb owner will still be part of | |
3816 | * nohz.cpu_mask.. | |
7835b98b | 3817 | * |
46cb4b7c SS |
3818 | * While stopping the tick, this cpu will become the ilb owner if there |
3819 | * is no other owner. And will be the owner till that cpu becomes busy | |
3820 | * or if all cpus in the system stop their ticks at which point | |
3821 | * there is no need for ilb owner. | |
3822 | * | |
3823 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3824 | * next busy scheduler_tick() | |
3825 | */ | |
3826 | int select_nohz_load_balancer(int stop_tick) | |
3827 | { | |
3828 | int cpu = smp_processor_id(); | |
3829 | ||
3830 | if (stop_tick) { | |
3831 | cpu_set(cpu, nohz.cpu_mask); | |
3832 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3833 | ||
3834 | /* | |
3835 | * If we are going offline and still the leader, give up! | |
3836 | */ | |
e761b772 | 3837 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3838 | atomic_read(&nohz.load_balancer) == cpu) { |
3839 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3840 | BUG(); | |
3841 | return 0; | |
3842 | } | |
3843 | ||
3844 | /* time for ilb owner also to sleep */ | |
3845 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3846 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3847 | atomic_set(&nohz.load_balancer, -1); | |
3848 | return 0; | |
3849 | } | |
3850 | ||
3851 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3852 | /* make me the ilb owner */ | |
3853 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3854 | return 1; | |
3855 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3856 | return 1; | |
3857 | } else { | |
3858 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3859 | return 0; | |
3860 | ||
3861 | cpu_clear(cpu, nohz.cpu_mask); | |
3862 | ||
3863 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3864 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3865 | BUG(); | |
3866 | } | |
3867 | return 0; | |
3868 | } | |
3869 | #endif | |
3870 | ||
3871 | static DEFINE_SPINLOCK(balancing); | |
3872 | ||
3873 | /* | |
7835b98b CL |
3874 | * It checks each scheduling domain to see if it is due to be balanced, |
3875 | * and initiates a balancing operation if so. | |
3876 | * | |
3877 | * Balancing parameters are set up in arch_init_sched_domains. | |
3878 | */ | |
a9957449 | 3879 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3880 | { |
46cb4b7c SS |
3881 | int balance = 1; |
3882 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3883 | unsigned long interval; |
3884 | struct sched_domain *sd; | |
46cb4b7c | 3885 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3886 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3887 | int update_next_balance = 0; |
d07355f5 | 3888 | int need_serialize; |
7c16ec58 | 3889 | cpumask_t tmp; |
1da177e4 | 3890 | |
46cb4b7c | 3891 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3892 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3893 | continue; | |
3894 | ||
3895 | interval = sd->balance_interval; | |
d15bcfdb | 3896 | if (idle != CPU_IDLE) |
1da177e4 LT |
3897 | interval *= sd->busy_factor; |
3898 | ||
3899 | /* scale ms to jiffies */ | |
3900 | interval = msecs_to_jiffies(interval); | |
3901 | if (unlikely(!interval)) | |
3902 | interval = 1; | |
dd41f596 IM |
3903 | if (interval > HZ*NR_CPUS/10) |
3904 | interval = HZ*NR_CPUS/10; | |
3905 | ||
d07355f5 | 3906 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3907 | |
d07355f5 | 3908 | if (need_serialize) { |
08c183f3 CL |
3909 | if (!spin_trylock(&balancing)) |
3910 | goto out; | |
3911 | } | |
3912 | ||
c9819f45 | 3913 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3914 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3915 | /* |
3916 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3917 | * longer idle, or one of our SMT siblings is |
3918 | * not idle. | |
3919 | */ | |
d15bcfdb | 3920 | idle = CPU_NOT_IDLE; |
1da177e4 | 3921 | } |
1bd77f2d | 3922 | sd->last_balance = jiffies; |
1da177e4 | 3923 | } |
d07355f5 | 3924 | if (need_serialize) |
08c183f3 CL |
3925 | spin_unlock(&balancing); |
3926 | out: | |
f549da84 | 3927 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3928 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3929 | update_next_balance = 1; |
3930 | } | |
783609c6 SS |
3931 | |
3932 | /* | |
3933 | * Stop the load balance at this level. There is another | |
3934 | * CPU in our sched group which is doing load balancing more | |
3935 | * actively. | |
3936 | */ | |
3937 | if (!balance) | |
3938 | break; | |
1da177e4 | 3939 | } |
f549da84 SS |
3940 | |
3941 | /* | |
3942 | * next_balance will be updated only when there is a need. | |
3943 | * When the cpu is attached to null domain for ex, it will not be | |
3944 | * updated. | |
3945 | */ | |
3946 | if (likely(update_next_balance)) | |
3947 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3948 | } |
3949 | ||
3950 | /* | |
3951 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3952 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3953 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3954 | */ | |
3955 | static void run_rebalance_domains(struct softirq_action *h) | |
3956 | { | |
dd41f596 IM |
3957 | int this_cpu = smp_processor_id(); |
3958 | struct rq *this_rq = cpu_rq(this_cpu); | |
3959 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3960 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3961 | |
dd41f596 | 3962 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3963 | |
3964 | #ifdef CONFIG_NO_HZ | |
3965 | /* | |
3966 | * If this cpu is the owner for idle load balancing, then do the | |
3967 | * balancing on behalf of the other idle cpus whose ticks are | |
3968 | * stopped. | |
3969 | */ | |
dd41f596 IM |
3970 | if (this_rq->idle_at_tick && |
3971 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3972 | cpumask_t cpus = nohz.cpu_mask; |
3973 | struct rq *rq; | |
3974 | int balance_cpu; | |
3975 | ||
dd41f596 | 3976 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3977 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3978 | /* |
3979 | * If this cpu gets work to do, stop the load balancing | |
3980 | * work being done for other cpus. Next load | |
3981 | * balancing owner will pick it up. | |
3982 | */ | |
3983 | if (need_resched()) | |
3984 | break; | |
3985 | ||
de0cf899 | 3986 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3987 | |
3988 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3989 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3990 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3991 | } |
3992 | } | |
3993 | #endif | |
3994 | } | |
3995 | ||
3996 | /* | |
3997 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3998 | * | |
3999 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4000 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4001 | * if the whole system is idle. | |
4002 | */ | |
dd41f596 | 4003 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4004 | { |
46cb4b7c SS |
4005 | #ifdef CONFIG_NO_HZ |
4006 | /* | |
4007 | * If we were in the nohz mode recently and busy at the current | |
4008 | * scheduler tick, then check if we need to nominate new idle | |
4009 | * load balancer. | |
4010 | */ | |
4011 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4012 | rq->in_nohz_recently = 0; | |
4013 | ||
4014 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
4015 | cpu_clear(cpu, nohz.cpu_mask); | |
4016 | atomic_set(&nohz.load_balancer, -1); | |
4017 | } | |
4018 | ||
4019 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4020 | /* | |
4021 | * simple selection for now: Nominate the | |
4022 | * first cpu in the nohz list to be the next | |
4023 | * ilb owner. | |
4024 | * | |
4025 | * TBD: Traverse the sched domains and nominate | |
4026 | * the nearest cpu in the nohz.cpu_mask. | |
4027 | */ | |
4028 | int ilb = first_cpu(nohz.cpu_mask); | |
4029 | ||
434d53b0 | 4030 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4031 | resched_cpu(ilb); |
4032 | } | |
4033 | } | |
4034 | ||
4035 | /* | |
4036 | * If this cpu is idle and doing idle load balancing for all the | |
4037 | * cpus with ticks stopped, is it time for that to stop? | |
4038 | */ | |
4039 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4040 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4041 | resched_cpu(cpu); | |
4042 | return; | |
4043 | } | |
4044 | ||
4045 | /* | |
4046 | * If this cpu is idle and the idle load balancing is done by | |
4047 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4048 | */ | |
4049 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4050 | cpu_isset(cpu, nohz.cpu_mask)) | |
4051 | return; | |
4052 | #endif | |
4053 | if (time_after_eq(jiffies, rq->next_balance)) | |
4054 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4055 | } |
dd41f596 IM |
4056 | |
4057 | #else /* CONFIG_SMP */ | |
4058 | ||
1da177e4 LT |
4059 | /* |
4060 | * on UP we do not need to balance between CPUs: | |
4061 | */ | |
70b97a7f | 4062 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4063 | { |
4064 | } | |
dd41f596 | 4065 | |
1da177e4 LT |
4066 | #endif |
4067 | ||
1da177e4 LT |
4068 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4069 | ||
4070 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4071 | ||
4072 | /* | |
f06febc9 FM |
4073 | * Return any ns on the sched_clock that have not yet been banked in |
4074 | * @p in case that task is currently running. | |
1da177e4 | 4075 | */ |
bb34d92f | 4076 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4077 | { |
1da177e4 | 4078 | unsigned long flags; |
41b86e9c | 4079 | struct rq *rq; |
bb34d92f | 4080 | u64 ns = 0; |
48f24c4d | 4081 | |
41b86e9c | 4082 | rq = task_rq_lock(p, &flags); |
1508487e | 4083 | |
051a1d1a | 4084 | if (task_current(rq, p)) { |
f06febc9 FM |
4085 | u64 delta_exec; |
4086 | ||
a8e504d2 IM |
4087 | update_rq_clock(rq); |
4088 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4089 | if ((s64)delta_exec > 0) |
bb34d92f | 4090 | ns = delta_exec; |
41b86e9c | 4091 | } |
48f24c4d | 4092 | |
41b86e9c | 4093 | task_rq_unlock(rq, &flags); |
48f24c4d | 4094 | |
1da177e4 LT |
4095 | return ns; |
4096 | } | |
4097 | ||
1da177e4 LT |
4098 | /* |
4099 | * Account user cpu time to a process. | |
4100 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4101 | * @cputime: the cpu time spent in user space since the last update |
4102 | */ | |
4103 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4104 | { | |
4105 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4106 | cputime64_t tmp; | |
4107 | ||
4108 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4109 | account_group_user_time(p, cputime); |
1da177e4 LT |
4110 | |
4111 | /* Add user time to cpustat. */ | |
4112 | tmp = cputime_to_cputime64(cputime); | |
4113 | if (TASK_NICE(p) > 0) | |
4114 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4115 | else | |
4116 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4117 | /* Account for user time used */ |
4118 | acct_update_integrals(p); | |
1da177e4 LT |
4119 | } |
4120 | ||
94886b84 LV |
4121 | /* |
4122 | * Account guest cpu time to a process. | |
4123 | * @p: the process that the cpu time gets accounted to | |
4124 | * @cputime: the cpu time spent in virtual machine since the last update | |
4125 | */ | |
f7402e03 | 4126 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4127 | { |
4128 | cputime64_t tmp; | |
4129 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4130 | ||
4131 | tmp = cputime_to_cputime64(cputime); | |
4132 | ||
4133 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4134 | account_group_user_time(p, cputime); |
94886b84 LV |
4135 | p->gtime = cputime_add(p->gtime, cputime); |
4136 | ||
4137 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4138 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4139 | } | |
4140 | ||
c66f08be MN |
4141 | /* |
4142 | * Account scaled user cpu time to a process. | |
4143 | * @p: the process that the cpu time gets accounted to | |
4144 | * @cputime: the cpu time spent in user space since the last update | |
4145 | */ | |
4146 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4147 | { | |
4148 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4149 | } | |
4150 | ||
1da177e4 LT |
4151 | /* |
4152 | * Account system cpu time to a process. | |
4153 | * @p: the process that the cpu time gets accounted to | |
4154 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4155 | * @cputime: the cpu time spent in kernel space since the last update | |
4156 | */ | |
4157 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4158 | cputime_t cputime) | |
4159 | { | |
4160 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4161 | struct rq *rq = this_rq(); |
1da177e4 LT |
4162 | cputime64_t tmp; |
4163 | ||
983ed7a6 HH |
4164 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4165 | account_guest_time(p, cputime); | |
4166 | return; | |
4167 | } | |
94886b84 | 4168 | |
1da177e4 | 4169 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4170 | account_group_system_time(p, cputime); |
1da177e4 LT |
4171 | |
4172 | /* Add system time to cpustat. */ | |
4173 | tmp = cputime_to_cputime64(cputime); | |
4174 | if (hardirq_count() - hardirq_offset) | |
4175 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4176 | else if (softirq_count()) | |
4177 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4178 | else if (p != rq->idle) |
1da177e4 | 4179 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4180 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4181 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4182 | else | |
4183 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4184 | /* Account for system time used */ | |
4185 | acct_update_integrals(p); | |
1da177e4 LT |
4186 | } |
4187 | ||
c66f08be MN |
4188 | /* |
4189 | * Account scaled system cpu time to a process. | |
4190 | * @p: the process that the cpu time gets accounted to | |
4191 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4192 | * @cputime: the cpu time spent in kernel space since the last update | |
4193 | */ | |
4194 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4195 | { | |
4196 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4197 | } | |
4198 | ||
1da177e4 LT |
4199 | /* |
4200 | * Account for involuntary wait time. | |
4201 | * @p: the process from which the cpu time has been stolen | |
4202 | * @steal: the cpu time spent in involuntary wait | |
4203 | */ | |
4204 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4205 | { | |
4206 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4207 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4208 | struct rq *rq = this_rq(); |
1da177e4 LT |
4209 | |
4210 | if (p == rq->idle) { | |
4211 | p->stime = cputime_add(p->stime, steal); | |
f06febc9 | 4212 | account_group_system_time(p, steal); |
1da177e4 LT |
4213 | if (atomic_read(&rq->nr_iowait) > 0) |
4214 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4215 | else | |
4216 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4217 | } else |
1da177e4 LT |
4218 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4219 | } | |
4220 | ||
49048622 BS |
4221 | /* |
4222 | * Use precise platform statistics if available: | |
4223 | */ | |
4224 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4225 | cputime_t task_utime(struct task_struct *p) | |
4226 | { | |
4227 | return p->utime; | |
4228 | } | |
4229 | ||
4230 | cputime_t task_stime(struct task_struct *p) | |
4231 | { | |
4232 | return p->stime; | |
4233 | } | |
4234 | #else | |
4235 | cputime_t task_utime(struct task_struct *p) | |
4236 | { | |
4237 | clock_t utime = cputime_to_clock_t(p->utime), | |
4238 | total = utime + cputime_to_clock_t(p->stime); | |
4239 | u64 temp; | |
4240 | ||
4241 | /* | |
4242 | * Use CFS's precise accounting: | |
4243 | */ | |
4244 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4245 | ||
4246 | if (total) { | |
4247 | temp *= utime; | |
4248 | do_div(temp, total); | |
4249 | } | |
4250 | utime = (clock_t)temp; | |
4251 | ||
4252 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4253 | return p->prev_utime; | |
4254 | } | |
4255 | ||
4256 | cputime_t task_stime(struct task_struct *p) | |
4257 | { | |
4258 | clock_t stime; | |
4259 | ||
4260 | /* | |
4261 | * Use CFS's precise accounting. (we subtract utime from | |
4262 | * the total, to make sure the total observed by userspace | |
4263 | * grows monotonically - apps rely on that): | |
4264 | */ | |
4265 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4266 | cputime_to_clock_t(task_utime(p)); | |
4267 | ||
4268 | if (stime >= 0) | |
4269 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4270 | ||
4271 | return p->prev_stime; | |
4272 | } | |
4273 | #endif | |
4274 | ||
4275 | inline cputime_t task_gtime(struct task_struct *p) | |
4276 | { | |
4277 | return p->gtime; | |
4278 | } | |
4279 | ||
7835b98b CL |
4280 | /* |
4281 | * This function gets called by the timer code, with HZ frequency. | |
4282 | * We call it with interrupts disabled. | |
4283 | * | |
4284 | * It also gets called by the fork code, when changing the parent's | |
4285 | * timeslices. | |
4286 | */ | |
4287 | void scheduler_tick(void) | |
4288 | { | |
7835b98b CL |
4289 | int cpu = smp_processor_id(); |
4290 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4291 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4292 | |
4293 | sched_clock_tick(); | |
dd41f596 IM |
4294 | |
4295 | spin_lock(&rq->lock); | |
3e51f33f | 4296 | update_rq_clock(rq); |
f1a438d8 | 4297 | update_cpu_load(rq); |
fa85ae24 | 4298 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4299 | spin_unlock(&rq->lock); |
7835b98b | 4300 | |
e418e1c2 | 4301 | #ifdef CONFIG_SMP |
dd41f596 IM |
4302 | rq->idle_at_tick = idle_cpu(cpu); |
4303 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4304 | #endif |
1da177e4 LT |
4305 | } |
4306 | ||
6cd8a4bb SR |
4307 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4308 | defined(CONFIG_PREEMPT_TRACER)) | |
4309 | ||
4310 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4311 | { | |
4312 | if (in_lock_functions(addr)) { | |
4313 | addr = CALLER_ADDR2; | |
4314 | if (in_lock_functions(addr)) | |
4315 | addr = CALLER_ADDR3; | |
4316 | } | |
4317 | return addr; | |
4318 | } | |
1da177e4 | 4319 | |
43627582 | 4320 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4321 | { |
6cd8a4bb | 4322 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4323 | /* |
4324 | * Underflow? | |
4325 | */ | |
9a11b49a IM |
4326 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4327 | return; | |
6cd8a4bb | 4328 | #endif |
1da177e4 | 4329 | preempt_count() += val; |
6cd8a4bb | 4330 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4331 | /* |
4332 | * Spinlock count overflowing soon? | |
4333 | */ | |
33859f7f MOS |
4334 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4335 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4336 | #endif |
4337 | if (preempt_count() == val) | |
4338 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4339 | } |
4340 | EXPORT_SYMBOL(add_preempt_count); | |
4341 | ||
43627582 | 4342 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4343 | { |
6cd8a4bb | 4344 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4345 | /* |
4346 | * Underflow? | |
4347 | */ | |
9a11b49a IM |
4348 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4349 | return; | |
1da177e4 LT |
4350 | /* |
4351 | * Is the spinlock portion underflowing? | |
4352 | */ | |
9a11b49a IM |
4353 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4354 | !(preempt_count() & PREEMPT_MASK))) | |
4355 | return; | |
6cd8a4bb | 4356 | #endif |
9a11b49a | 4357 | |
6cd8a4bb SR |
4358 | if (preempt_count() == val) |
4359 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4360 | preempt_count() -= val; |
4361 | } | |
4362 | EXPORT_SYMBOL(sub_preempt_count); | |
4363 | ||
4364 | #endif | |
4365 | ||
4366 | /* | |
dd41f596 | 4367 | * Print scheduling while atomic bug: |
1da177e4 | 4368 | */ |
dd41f596 | 4369 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4370 | { |
838225b4 SS |
4371 | struct pt_regs *regs = get_irq_regs(); |
4372 | ||
4373 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4374 | prev->comm, prev->pid, preempt_count()); | |
4375 | ||
dd41f596 | 4376 | debug_show_held_locks(prev); |
e21f5b15 | 4377 | print_modules(); |
dd41f596 IM |
4378 | if (irqs_disabled()) |
4379 | print_irqtrace_events(prev); | |
838225b4 SS |
4380 | |
4381 | if (regs) | |
4382 | show_regs(regs); | |
4383 | else | |
4384 | dump_stack(); | |
dd41f596 | 4385 | } |
1da177e4 | 4386 | |
dd41f596 IM |
4387 | /* |
4388 | * Various schedule()-time debugging checks and statistics: | |
4389 | */ | |
4390 | static inline void schedule_debug(struct task_struct *prev) | |
4391 | { | |
1da177e4 | 4392 | /* |
41a2d6cf | 4393 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4394 | * schedule() atomically, we ignore that path for now. |
4395 | * Otherwise, whine if we are scheduling when we should not be. | |
4396 | */ | |
3f33a7ce | 4397 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4398 | __schedule_bug(prev); |
4399 | ||
1da177e4 LT |
4400 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4401 | ||
2d72376b | 4402 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4403 | #ifdef CONFIG_SCHEDSTATS |
4404 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4405 | schedstat_inc(this_rq(), bkl_count); |
4406 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4407 | } |
4408 | #endif | |
dd41f596 IM |
4409 | } |
4410 | ||
4411 | /* | |
4412 | * Pick up the highest-prio task: | |
4413 | */ | |
4414 | static inline struct task_struct * | |
ff95f3df | 4415 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4416 | { |
5522d5d5 | 4417 | const struct sched_class *class; |
dd41f596 | 4418 | struct task_struct *p; |
1da177e4 LT |
4419 | |
4420 | /* | |
dd41f596 IM |
4421 | * Optimization: we know that if all tasks are in |
4422 | * the fair class we can call that function directly: | |
1da177e4 | 4423 | */ |
dd41f596 | 4424 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4425 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4426 | if (likely(p)) |
4427 | return p; | |
1da177e4 LT |
4428 | } |
4429 | ||
dd41f596 IM |
4430 | class = sched_class_highest; |
4431 | for ( ; ; ) { | |
fb8d4724 | 4432 | p = class->pick_next_task(rq); |
dd41f596 IM |
4433 | if (p) |
4434 | return p; | |
4435 | /* | |
4436 | * Will never be NULL as the idle class always | |
4437 | * returns a non-NULL p: | |
4438 | */ | |
4439 | class = class->next; | |
4440 | } | |
4441 | } | |
1da177e4 | 4442 | |
dd41f596 IM |
4443 | /* |
4444 | * schedule() is the main scheduler function. | |
4445 | */ | |
4446 | asmlinkage void __sched schedule(void) | |
4447 | { | |
4448 | struct task_struct *prev, *next; | |
67ca7bde | 4449 | unsigned long *switch_count; |
dd41f596 | 4450 | struct rq *rq; |
31656519 | 4451 | int cpu; |
dd41f596 IM |
4452 | |
4453 | need_resched: | |
4454 | preempt_disable(); | |
4455 | cpu = smp_processor_id(); | |
4456 | rq = cpu_rq(cpu); | |
4457 | rcu_qsctr_inc(cpu); | |
4458 | prev = rq->curr; | |
4459 | switch_count = &prev->nivcsw; | |
4460 | ||
4461 | release_kernel_lock(prev); | |
4462 | need_resched_nonpreemptible: | |
4463 | ||
4464 | schedule_debug(prev); | |
1da177e4 | 4465 | |
31656519 | 4466 | if (sched_feat(HRTICK)) |
f333fdc9 | 4467 | hrtick_clear(rq); |
8f4d37ec | 4468 | |
8cd162ce | 4469 | spin_lock_irq(&rq->lock); |
3e51f33f | 4470 | update_rq_clock(rq); |
1e819950 | 4471 | clear_tsk_need_resched(prev); |
1da177e4 | 4472 | |
1da177e4 | 4473 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4474 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4475 | prev->state = TASK_RUNNING; |
16882c1e | 4476 | else |
2e1cb74a | 4477 | deactivate_task(rq, prev, 1); |
dd41f596 | 4478 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4479 | } |
4480 | ||
9a897c5a SR |
4481 | #ifdef CONFIG_SMP |
4482 | if (prev->sched_class->pre_schedule) | |
4483 | prev->sched_class->pre_schedule(rq, prev); | |
4484 | #endif | |
f65eda4f | 4485 | |
dd41f596 | 4486 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4487 | idle_balance(cpu, rq); |
1da177e4 | 4488 | |
31ee529c | 4489 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4490 | next = pick_next_task(rq, prev); |
1da177e4 | 4491 | |
1da177e4 | 4492 | if (likely(prev != next)) { |
673a90a1 DS |
4493 | sched_info_switch(prev, next); |
4494 | ||
1da177e4 LT |
4495 | rq->nr_switches++; |
4496 | rq->curr = next; | |
4497 | ++*switch_count; | |
4498 | ||
dd41f596 | 4499 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4500 | /* |
4501 | * the context switch might have flipped the stack from under | |
4502 | * us, hence refresh the local variables. | |
4503 | */ | |
4504 | cpu = smp_processor_id(); | |
4505 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4506 | } else |
4507 | spin_unlock_irq(&rq->lock); | |
4508 | ||
8f4d37ec | 4509 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4510 | goto need_resched_nonpreemptible; |
8f4d37ec | 4511 | |
1da177e4 LT |
4512 | preempt_enable_no_resched(); |
4513 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4514 | goto need_resched; | |
4515 | } | |
1da177e4 LT |
4516 | EXPORT_SYMBOL(schedule); |
4517 | ||
4518 | #ifdef CONFIG_PREEMPT | |
4519 | /* | |
2ed6e34f | 4520 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4521 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4522 | * occur there and call schedule directly. |
4523 | */ | |
4524 | asmlinkage void __sched preempt_schedule(void) | |
4525 | { | |
4526 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4527 | |
1da177e4 LT |
4528 | /* |
4529 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4530 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4531 | */ |
beed33a8 | 4532 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4533 | return; |
4534 | ||
3a5c359a AK |
4535 | do { |
4536 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4537 | schedule(); |
3a5c359a | 4538 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4539 | |
3a5c359a AK |
4540 | /* |
4541 | * Check again in case we missed a preemption opportunity | |
4542 | * between schedule and now. | |
4543 | */ | |
4544 | barrier(); | |
4545 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4546 | } |
1da177e4 LT |
4547 | EXPORT_SYMBOL(preempt_schedule); |
4548 | ||
4549 | /* | |
2ed6e34f | 4550 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4551 | * off of irq context. |
4552 | * Note, that this is called and return with irqs disabled. This will | |
4553 | * protect us against recursive calling from irq. | |
4554 | */ | |
4555 | asmlinkage void __sched preempt_schedule_irq(void) | |
4556 | { | |
4557 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4558 | |
2ed6e34f | 4559 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4560 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4561 | ||
3a5c359a AK |
4562 | do { |
4563 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4564 | local_irq_enable(); |
4565 | schedule(); | |
4566 | local_irq_disable(); | |
3a5c359a | 4567 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4568 | |
3a5c359a AK |
4569 | /* |
4570 | * Check again in case we missed a preemption opportunity | |
4571 | * between schedule and now. | |
4572 | */ | |
4573 | barrier(); | |
4574 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4575 | } |
4576 | ||
4577 | #endif /* CONFIG_PREEMPT */ | |
4578 | ||
95cdf3b7 IM |
4579 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4580 | void *key) | |
1da177e4 | 4581 | { |
48f24c4d | 4582 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4583 | } |
1da177e4 LT |
4584 | EXPORT_SYMBOL(default_wake_function); |
4585 | ||
4586 | /* | |
41a2d6cf IM |
4587 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4588 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4589 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4590 | * | |
4591 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4592 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4593 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4594 | */ | |
4595 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4596 | int nr_exclusive, int sync, void *key) | |
4597 | { | |
2e45874c | 4598 | wait_queue_t *curr, *next; |
1da177e4 | 4599 | |
2e45874c | 4600 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4601 | unsigned flags = curr->flags; |
4602 | ||
1da177e4 | 4603 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4604 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4605 | break; |
4606 | } | |
4607 | } | |
4608 | ||
4609 | /** | |
4610 | * __wake_up - wake up threads blocked on a waitqueue. | |
4611 | * @q: the waitqueue | |
4612 | * @mode: which threads | |
4613 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4614 | * @key: is directly passed to the wakeup function |
1da177e4 | 4615 | */ |
7ad5b3a5 | 4616 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4617 | int nr_exclusive, void *key) |
1da177e4 LT |
4618 | { |
4619 | unsigned long flags; | |
4620 | ||
4621 | spin_lock_irqsave(&q->lock, flags); | |
4622 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4623 | spin_unlock_irqrestore(&q->lock, flags); | |
4624 | } | |
1da177e4 LT |
4625 | EXPORT_SYMBOL(__wake_up); |
4626 | ||
4627 | /* | |
4628 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4629 | */ | |
7ad5b3a5 | 4630 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4631 | { |
4632 | __wake_up_common(q, mode, 1, 0, NULL); | |
4633 | } | |
4634 | ||
4635 | /** | |
67be2dd1 | 4636 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4637 | * @q: the waitqueue |
4638 | * @mode: which threads | |
4639 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4640 | * | |
4641 | * The sync wakeup differs that the waker knows that it will schedule | |
4642 | * away soon, so while the target thread will be woken up, it will not | |
4643 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4644 | * with each other. This can prevent needless bouncing between CPUs. | |
4645 | * | |
4646 | * On UP it can prevent extra preemption. | |
4647 | */ | |
7ad5b3a5 | 4648 | void |
95cdf3b7 | 4649 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4650 | { |
4651 | unsigned long flags; | |
4652 | int sync = 1; | |
4653 | ||
4654 | if (unlikely(!q)) | |
4655 | return; | |
4656 | ||
4657 | if (unlikely(!nr_exclusive)) | |
4658 | sync = 0; | |
4659 | ||
4660 | spin_lock_irqsave(&q->lock, flags); | |
4661 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4662 | spin_unlock_irqrestore(&q->lock, flags); | |
4663 | } | |
4664 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4665 | ||
65eb3dc6 KD |
4666 | /** |
4667 | * complete: - signals a single thread waiting on this completion | |
4668 | * @x: holds the state of this particular completion | |
4669 | * | |
4670 | * This will wake up a single thread waiting on this completion. Threads will be | |
4671 | * awakened in the same order in which they were queued. | |
4672 | * | |
4673 | * See also complete_all(), wait_for_completion() and related routines. | |
4674 | */ | |
b15136e9 | 4675 | void complete(struct completion *x) |
1da177e4 LT |
4676 | { |
4677 | unsigned long flags; | |
4678 | ||
4679 | spin_lock_irqsave(&x->wait.lock, flags); | |
4680 | x->done++; | |
d9514f6c | 4681 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4682 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4683 | } | |
4684 | EXPORT_SYMBOL(complete); | |
4685 | ||
65eb3dc6 KD |
4686 | /** |
4687 | * complete_all: - signals all threads waiting on this completion | |
4688 | * @x: holds the state of this particular completion | |
4689 | * | |
4690 | * This will wake up all threads waiting on this particular completion event. | |
4691 | */ | |
b15136e9 | 4692 | void complete_all(struct completion *x) |
1da177e4 LT |
4693 | { |
4694 | unsigned long flags; | |
4695 | ||
4696 | spin_lock_irqsave(&x->wait.lock, flags); | |
4697 | x->done += UINT_MAX/2; | |
d9514f6c | 4698 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4699 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4700 | } | |
4701 | EXPORT_SYMBOL(complete_all); | |
4702 | ||
8cbbe86d AK |
4703 | static inline long __sched |
4704 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4705 | { |
1da177e4 LT |
4706 | if (!x->done) { |
4707 | DECLARE_WAITQUEUE(wait, current); | |
4708 | ||
4709 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4710 | __add_wait_queue_tail(&x->wait, &wait); | |
4711 | do { | |
94d3d824 | 4712 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4713 | timeout = -ERESTARTSYS; |
4714 | break; | |
8cbbe86d AK |
4715 | } |
4716 | __set_current_state(state); | |
1da177e4 LT |
4717 | spin_unlock_irq(&x->wait.lock); |
4718 | timeout = schedule_timeout(timeout); | |
4719 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4720 | } while (!x->done && timeout); |
1da177e4 | 4721 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4722 | if (!x->done) |
4723 | return timeout; | |
1da177e4 LT |
4724 | } |
4725 | x->done--; | |
ea71a546 | 4726 | return timeout ?: 1; |
1da177e4 | 4727 | } |
1da177e4 | 4728 | |
8cbbe86d AK |
4729 | static long __sched |
4730 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4731 | { |
1da177e4 LT |
4732 | might_sleep(); |
4733 | ||
4734 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4735 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4736 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4737 | return timeout; |
4738 | } | |
1da177e4 | 4739 | |
65eb3dc6 KD |
4740 | /** |
4741 | * wait_for_completion: - waits for completion of a task | |
4742 | * @x: holds the state of this particular completion | |
4743 | * | |
4744 | * This waits to be signaled for completion of a specific task. It is NOT | |
4745 | * interruptible and there is no timeout. | |
4746 | * | |
4747 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4748 | * and interrupt capability. Also see complete(). | |
4749 | */ | |
b15136e9 | 4750 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4751 | { |
4752 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4753 | } |
8cbbe86d | 4754 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4755 | |
65eb3dc6 KD |
4756 | /** |
4757 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4758 | * @x: holds the state of this particular completion | |
4759 | * @timeout: timeout value in jiffies | |
4760 | * | |
4761 | * This waits for either a completion of a specific task to be signaled or for a | |
4762 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4763 | * interruptible. | |
4764 | */ | |
b15136e9 | 4765 | unsigned long __sched |
8cbbe86d | 4766 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4767 | { |
8cbbe86d | 4768 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4769 | } |
8cbbe86d | 4770 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4771 | |
65eb3dc6 KD |
4772 | /** |
4773 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4774 | * @x: holds the state of this particular completion | |
4775 | * | |
4776 | * This waits for completion of a specific task to be signaled. It is | |
4777 | * interruptible. | |
4778 | */ | |
8cbbe86d | 4779 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4780 | { |
51e97990 AK |
4781 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4782 | if (t == -ERESTARTSYS) | |
4783 | return t; | |
4784 | return 0; | |
0fec171c | 4785 | } |
8cbbe86d | 4786 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4787 | |
65eb3dc6 KD |
4788 | /** |
4789 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4790 | * @x: holds the state of this particular completion | |
4791 | * @timeout: timeout value in jiffies | |
4792 | * | |
4793 | * This waits for either a completion of a specific task to be signaled or for a | |
4794 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4795 | */ | |
b15136e9 | 4796 | unsigned long __sched |
8cbbe86d AK |
4797 | wait_for_completion_interruptible_timeout(struct completion *x, |
4798 | unsigned long timeout) | |
0fec171c | 4799 | { |
8cbbe86d | 4800 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4801 | } |
8cbbe86d | 4802 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4803 | |
65eb3dc6 KD |
4804 | /** |
4805 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4806 | * @x: holds the state of this particular completion | |
4807 | * | |
4808 | * This waits to be signaled for completion of a specific task. It can be | |
4809 | * interrupted by a kill signal. | |
4810 | */ | |
009e577e MW |
4811 | int __sched wait_for_completion_killable(struct completion *x) |
4812 | { | |
4813 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4814 | if (t == -ERESTARTSYS) | |
4815 | return t; | |
4816 | return 0; | |
4817 | } | |
4818 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4819 | ||
be4de352 DC |
4820 | /** |
4821 | * try_wait_for_completion - try to decrement a completion without blocking | |
4822 | * @x: completion structure | |
4823 | * | |
4824 | * Returns: 0 if a decrement cannot be done without blocking | |
4825 | * 1 if a decrement succeeded. | |
4826 | * | |
4827 | * If a completion is being used as a counting completion, | |
4828 | * attempt to decrement the counter without blocking. This | |
4829 | * enables us to avoid waiting if the resource the completion | |
4830 | * is protecting is not available. | |
4831 | */ | |
4832 | bool try_wait_for_completion(struct completion *x) | |
4833 | { | |
4834 | int ret = 1; | |
4835 | ||
4836 | spin_lock_irq(&x->wait.lock); | |
4837 | if (!x->done) | |
4838 | ret = 0; | |
4839 | else | |
4840 | x->done--; | |
4841 | spin_unlock_irq(&x->wait.lock); | |
4842 | return ret; | |
4843 | } | |
4844 | EXPORT_SYMBOL(try_wait_for_completion); | |
4845 | ||
4846 | /** | |
4847 | * completion_done - Test to see if a completion has any waiters | |
4848 | * @x: completion structure | |
4849 | * | |
4850 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4851 | * 1 if there are no waiters. | |
4852 | * | |
4853 | */ | |
4854 | bool completion_done(struct completion *x) | |
4855 | { | |
4856 | int ret = 1; | |
4857 | ||
4858 | spin_lock_irq(&x->wait.lock); | |
4859 | if (!x->done) | |
4860 | ret = 0; | |
4861 | spin_unlock_irq(&x->wait.lock); | |
4862 | return ret; | |
4863 | } | |
4864 | EXPORT_SYMBOL(completion_done); | |
4865 | ||
8cbbe86d AK |
4866 | static long __sched |
4867 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4868 | { |
0fec171c IM |
4869 | unsigned long flags; |
4870 | wait_queue_t wait; | |
4871 | ||
4872 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4873 | |
8cbbe86d | 4874 | __set_current_state(state); |
1da177e4 | 4875 | |
8cbbe86d AK |
4876 | spin_lock_irqsave(&q->lock, flags); |
4877 | __add_wait_queue(q, &wait); | |
4878 | spin_unlock(&q->lock); | |
4879 | timeout = schedule_timeout(timeout); | |
4880 | spin_lock_irq(&q->lock); | |
4881 | __remove_wait_queue(q, &wait); | |
4882 | spin_unlock_irqrestore(&q->lock, flags); | |
4883 | ||
4884 | return timeout; | |
4885 | } | |
4886 | ||
4887 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4888 | { | |
4889 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4890 | } |
1da177e4 LT |
4891 | EXPORT_SYMBOL(interruptible_sleep_on); |
4892 | ||
0fec171c | 4893 | long __sched |
95cdf3b7 | 4894 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4895 | { |
8cbbe86d | 4896 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4897 | } |
1da177e4 LT |
4898 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4899 | ||
0fec171c | 4900 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4901 | { |
8cbbe86d | 4902 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4903 | } |
1da177e4 LT |
4904 | EXPORT_SYMBOL(sleep_on); |
4905 | ||
0fec171c | 4906 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4907 | { |
8cbbe86d | 4908 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4909 | } |
1da177e4 LT |
4910 | EXPORT_SYMBOL(sleep_on_timeout); |
4911 | ||
b29739f9 IM |
4912 | #ifdef CONFIG_RT_MUTEXES |
4913 | ||
4914 | /* | |
4915 | * rt_mutex_setprio - set the current priority of a task | |
4916 | * @p: task | |
4917 | * @prio: prio value (kernel-internal form) | |
4918 | * | |
4919 | * This function changes the 'effective' priority of a task. It does | |
4920 | * not touch ->normal_prio like __setscheduler(). | |
4921 | * | |
4922 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4923 | */ | |
36c8b586 | 4924 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4925 | { |
4926 | unsigned long flags; | |
83b699ed | 4927 | int oldprio, on_rq, running; |
70b97a7f | 4928 | struct rq *rq; |
cb469845 | 4929 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4930 | |
4931 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4932 | ||
4933 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4934 | update_rq_clock(rq); |
b29739f9 | 4935 | |
d5f9f942 | 4936 | oldprio = p->prio; |
dd41f596 | 4937 | on_rq = p->se.on_rq; |
051a1d1a | 4938 | running = task_current(rq, p); |
0e1f3483 | 4939 | if (on_rq) |
69be72c1 | 4940 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4941 | if (running) |
4942 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4943 | |
4944 | if (rt_prio(prio)) | |
4945 | p->sched_class = &rt_sched_class; | |
4946 | else | |
4947 | p->sched_class = &fair_sched_class; | |
4948 | ||
b29739f9 IM |
4949 | p->prio = prio; |
4950 | ||
0e1f3483 HS |
4951 | if (running) |
4952 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4953 | if (on_rq) { |
8159f87e | 4954 | enqueue_task(rq, p, 0); |
cb469845 SR |
4955 | |
4956 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4957 | } |
4958 | task_rq_unlock(rq, &flags); | |
4959 | } | |
4960 | ||
4961 | #endif | |
4962 | ||
36c8b586 | 4963 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4964 | { |
dd41f596 | 4965 | int old_prio, delta, on_rq; |
1da177e4 | 4966 | unsigned long flags; |
70b97a7f | 4967 | struct rq *rq; |
1da177e4 LT |
4968 | |
4969 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4970 | return; | |
4971 | /* | |
4972 | * We have to be careful, if called from sys_setpriority(), | |
4973 | * the task might be in the middle of scheduling on another CPU. | |
4974 | */ | |
4975 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4976 | update_rq_clock(rq); |
1da177e4 LT |
4977 | /* |
4978 | * The RT priorities are set via sched_setscheduler(), but we still | |
4979 | * allow the 'normal' nice value to be set - but as expected | |
4980 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4981 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4982 | */ |
e05606d3 | 4983 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4984 | p->static_prio = NICE_TO_PRIO(nice); |
4985 | goto out_unlock; | |
4986 | } | |
dd41f596 | 4987 | on_rq = p->se.on_rq; |
c09595f6 | 4988 | if (on_rq) |
69be72c1 | 4989 | dequeue_task(rq, p, 0); |
1da177e4 | 4990 | |
1da177e4 | 4991 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4992 | set_load_weight(p); |
b29739f9 IM |
4993 | old_prio = p->prio; |
4994 | p->prio = effective_prio(p); | |
4995 | delta = p->prio - old_prio; | |
1da177e4 | 4996 | |
dd41f596 | 4997 | if (on_rq) { |
8159f87e | 4998 | enqueue_task(rq, p, 0); |
1da177e4 | 4999 | /* |
d5f9f942 AM |
5000 | * If the task increased its priority or is running and |
5001 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5002 | */ |
d5f9f942 | 5003 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5004 | resched_task(rq->curr); |
5005 | } | |
5006 | out_unlock: | |
5007 | task_rq_unlock(rq, &flags); | |
5008 | } | |
1da177e4 LT |
5009 | EXPORT_SYMBOL(set_user_nice); |
5010 | ||
e43379f1 MM |
5011 | /* |
5012 | * can_nice - check if a task can reduce its nice value | |
5013 | * @p: task | |
5014 | * @nice: nice value | |
5015 | */ | |
36c8b586 | 5016 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5017 | { |
024f4747 MM |
5018 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5019 | int nice_rlim = 20 - nice; | |
48f24c4d | 5020 | |
e43379f1 MM |
5021 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5022 | capable(CAP_SYS_NICE)); | |
5023 | } | |
5024 | ||
1da177e4 LT |
5025 | #ifdef __ARCH_WANT_SYS_NICE |
5026 | ||
5027 | /* | |
5028 | * sys_nice - change the priority of the current process. | |
5029 | * @increment: priority increment | |
5030 | * | |
5031 | * sys_setpriority is a more generic, but much slower function that | |
5032 | * does similar things. | |
5033 | */ | |
5034 | asmlinkage long sys_nice(int increment) | |
5035 | { | |
48f24c4d | 5036 | long nice, retval; |
1da177e4 LT |
5037 | |
5038 | /* | |
5039 | * Setpriority might change our priority at the same moment. | |
5040 | * We don't have to worry. Conceptually one call occurs first | |
5041 | * and we have a single winner. | |
5042 | */ | |
e43379f1 MM |
5043 | if (increment < -40) |
5044 | increment = -40; | |
1da177e4 LT |
5045 | if (increment > 40) |
5046 | increment = 40; | |
5047 | ||
5048 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5049 | if (nice < -20) | |
5050 | nice = -20; | |
5051 | if (nice > 19) | |
5052 | nice = 19; | |
5053 | ||
e43379f1 MM |
5054 | if (increment < 0 && !can_nice(current, nice)) |
5055 | return -EPERM; | |
5056 | ||
1da177e4 LT |
5057 | retval = security_task_setnice(current, nice); |
5058 | if (retval) | |
5059 | return retval; | |
5060 | ||
5061 | set_user_nice(current, nice); | |
5062 | return 0; | |
5063 | } | |
5064 | ||
5065 | #endif | |
5066 | ||
5067 | /** | |
5068 | * task_prio - return the priority value of a given task. | |
5069 | * @p: the task in question. | |
5070 | * | |
5071 | * This is the priority value as seen by users in /proc. | |
5072 | * RT tasks are offset by -200. Normal tasks are centered | |
5073 | * around 0, value goes from -16 to +15. | |
5074 | */ | |
36c8b586 | 5075 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5076 | { |
5077 | return p->prio - MAX_RT_PRIO; | |
5078 | } | |
5079 | ||
5080 | /** | |
5081 | * task_nice - return the nice value of a given task. | |
5082 | * @p: the task in question. | |
5083 | */ | |
36c8b586 | 5084 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5085 | { |
5086 | return TASK_NICE(p); | |
5087 | } | |
150d8bed | 5088 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5089 | |
5090 | /** | |
5091 | * idle_cpu - is a given cpu idle currently? | |
5092 | * @cpu: the processor in question. | |
5093 | */ | |
5094 | int idle_cpu(int cpu) | |
5095 | { | |
5096 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5097 | } | |
5098 | ||
1da177e4 LT |
5099 | /** |
5100 | * idle_task - return the idle task for a given cpu. | |
5101 | * @cpu: the processor in question. | |
5102 | */ | |
36c8b586 | 5103 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5104 | { |
5105 | return cpu_rq(cpu)->idle; | |
5106 | } | |
5107 | ||
5108 | /** | |
5109 | * find_process_by_pid - find a process with a matching PID value. | |
5110 | * @pid: the pid in question. | |
5111 | */ | |
a9957449 | 5112 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5113 | { |
228ebcbe | 5114 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5115 | } |
5116 | ||
5117 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5118 | static void |
5119 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5120 | { |
dd41f596 | 5121 | BUG_ON(p->se.on_rq); |
48f24c4d | 5122 | |
1da177e4 | 5123 | p->policy = policy; |
dd41f596 IM |
5124 | switch (p->policy) { |
5125 | case SCHED_NORMAL: | |
5126 | case SCHED_BATCH: | |
5127 | case SCHED_IDLE: | |
5128 | p->sched_class = &fair_sched_class; | |
5129 | break; | |
5130 | case SCHED_FIFO: | |
5131 | case SCHED_RR: | |
5132 | p->sched_class = &rt_sched_class; | |
5133 | break; | |
5134 | } | |
5135 | ||
1da177e4 | 5136 | p->rt_priority = prio; |
b29739f9 IM |
5137 | p->normal_prio = normal_prio(p); |
5138 | /* we are holding p->pi_lock already */ | |
5139 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5140 | set_load_weight(p); |
1da177e4 LT |
5141 | } |
5142 | ||
961ccddd RR |
5143 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5144 | struct sched_param *param, bool user) | |
1da177e4 | 5145 | { |
83b699ed | 5146 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5147 | unsigned long flags; |
cb469845 | 5148 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5149 | struct rq *rq; |
1da177e4 | 5150 | |
66e5393a SR |
5151 | /* may grab non-irq protected spin_locks */ |
5152 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5153 | recheck: |
5154 | /* double check policy once rq lock held */ | |
5155 | if (policy < 0) | |
5156 | policy = oldpolicy = p->policy; | |
5157 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5158 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5159 | policy != SCHED_IDLE) | |
b0a9499c | 5160 | return -EINVAL; |
1da177e4 LT |
5161 | /* |
5162 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5163 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5164 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5165 | */ |
5166 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5167 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5168 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5169 | return -EINVAL; |
e05606d3 | 5170 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5171 | return -EINVAL; |
5172 | ||
37e4ab3f OC |
5173 | /* |
5174 | * Allow unprivileged RT tasks to decrease priority: | |
5175 | */ | |
961ccddd | 5176 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5177 | if (rt_policy(policy)) { |
8dc3e909 | 5178 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5179 | |
5180 | if (!lock_task_sighand(p, &flags)) | |
5181 | return -ESRCH; | |
5182 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5183 | unlock_task_sighand(p, &flags); | |
5184 | ||
5185 | /* can't set/change the rt policy */ | |
5186 | if (policy != p->policy && !rlim_rtprio) | |
5187 | return -EPERM; | |
5188 | ||
5189 | /* can't increase priority */ | |
5190 | if (param->sched_priority > p->rt_priority && | |
5191 | param->sched_priority > rlim_rtprio) | |
5192 | return -EPERM; | |
5193 | } | |
dd41f596 IM |
5194 | /* |
5195 | * Like positive nice levels, dont allow tasks to | |
5196 | * move out of SCHED_IDLE either: | |
5197 | */ | |
5198 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5199 | return -EPERM; | |
5fe1d75f | 5200 | |
37e4ab3f OC |
5201 | /* can't change other user's priorities */ |
5202 | if ((current->euid != p->euid) && | |
5203 | (current->euid != p->uid)) | |
5204 | return -EPERM; | |
5205 | } | |
1da177e4 | 5206 | |
725aad24 | 5207 | if (user) { |
b68aa230 | 5208 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5209 | /* |
5210 | * Do not allow realtime tasks into groups that have no runtime | |
5211 | * assigned. | |
5212 | */ | |
9a7e0b18 PZ |
5213 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5214 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5215 | return -EPERM; |
b68aa230 PZ |
5216 | #endif |
5217 | ||
725aad24 JF |
5218 | retval = security_task_setscheduler(p, policy, param); |
5219 | if (retval) | |
5220 | return retval; | |
5221 | } | |
5222 | ||
b29739f9 IM |
5223 | /* |
5224 | * make sure no PI-waiters arrive (or leave) while we are | |
5225 | * changing the priority of the task: | |
5226 | */ | |
5227 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5228 | /* |
5229 | * To be able to change p->policy safely, the apropriate | |
5230 | * runqueue lock must be held. | |
5231 | */ | |
b29739f9 | 5232 | rq = __task_rq_lock(p); |
1da177e4 LT |
5233 | /* recheck policy now with rq lock held */ |
5234 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5235 | policy = oldpolicy = -1; | |
b29739f9 IM |
5236 | __task_rq_unlock(rq); |
5237 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5238 | goto recheck; |
5239 | } | |
2daa3577 | 5240 | update_rq_clock(rq); |
dd41f596 | 5241 | on_rq = p->se.on_rq; |
051a1d1a | 5242 | running = task_current(rq, p); |
0e1f3483 | 5243 | if (on_rq) |
2e1cb74a | 5244 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5245 | if (running) |
5246 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5247 | |
1da177e4 | 5248 | oldprio = p->prio; |
dd41f596 | 5249 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5250 | |
0e1f3483 HS |
5251 | if (running) |
5252 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5253 | if (on_rq) { |
5254 | activate_task(rq, p, 0); | |
cb469845 SR |
5255 | |
5256 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5257 | } |
b29739f9 IM |
5258 | __task_rq_unlock(rq); |
5259 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5260 | ||
95e02ca9 TG |
5261 | rt_mutex_adjust_pi(p); |
5262 | ||
1da177e4 LT |
5263 | return 0; |
5264 | } | |
961ccddd RR |
5265 | |
5266 | /** | |
5267 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5268 | * @p: the task in question. | |
5269 | * @policy: new policy. | |
5270 | * @param: structure containing the new RT priority. | |
5271 | * | |
5272 | * NOTE that the task may be already dead. | |
5273 | */ | |
5274 | int sched_setscheduler(struct task_struct *p, int policy, | |
5275 | struct sched_param *param) | |
5276 | { | |
5277 | return __sched_setscheduler(p, policy, param, true); | |
5278 | } | |
1da177e4 LT |
5279 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5280 | ||
961ccddd RR |
5281 | /** |
5282 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5283 | * @p: the task in question. | |
5284 | * @policy: new policy. | |
5285 | * @param: structure containing the new RT priority. | |
5286 | * | |
5287 | * Just like sched_setscheduler, only don't bother checking if the | |
5288 | * current context has permission. For example, this is needed in | |
5289 | * stop_machine(): we create temporary high priority worker threads, | |
5290 | * but our caller might not have that capability. | |
5291 | */ | |
5292 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5293 | struct sched_param *param) | |
5294 | { | |
5295 | return __sched_setscheduler(p, policy, param, false); | |
5296 | } | |
5297 | ||
95cdf3b7 IM |
5298 | static int |
5299 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5300 | { |
1da177e4 LT |
5301 | struct sched_param lparam; |
5302 | struct task_struct *p; | |
36c8b586 | 5303 | int retval; |
1da177e4 LT |
5304 | |
5305 | if (!param || pid < 0) | |
5306 | return -EINVAL; | |
5307 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5308 | return -EFAULT; | |
5fe1d75f ON |
5309 | |
5310 | rcu_read_lock(); | |
5311 | retval = -ESRCH; | |
1da177e4 | 5312 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5313 | if (p != NULL) |
5314 | retval = sched_setscheduler(p, policy, &lparam); | |
5315 | rcu_read_unlock(); | |
36c8b586 | 5316 | |
1da177e4 LT |
5317 | return retval; |
5318 | } | |
5319 | ||
5320 | /** | |
5321 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5322 | * @pid: the pid in question. | |
5323 | * @policy: new policy. | |
5324 | * @param: structure containing the new RT priority. | |
5325 | */ | |
41a2d6cf IM |
5326 | asmlinkage long |
5327 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5328 | { |
c21761f1 JB |
5329 | /* negative values for policy are not valid */ |
5330 | if (policy < 0) | |
5331 | return -EINVAL; | |
5332 | ||
1da177e4 LT |
5333 | return do_sched_setscheduler(pid, policy, param); |
5334 | } | |
5335 | ||
5336 | /** | |
5337 | * sys_sched_setparam - set/change the RT priority of a thread | |
5338 | * @pid: the pid in question. | |
5339 | * @param: structure containing the new RT priority. | |
5340 | */ | |
5341 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5342 | { | |
5343 | return do_sched_setscheduler(pid, -1, param); | |
5344 | } | |
5345 | ||
5346 | /** | |
5347 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5348 | * @pid: the pid in question. | |
5349 | */ | |
5350 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5351 | { | |
36c8b586 | 5352 | struct task_struct *p; |
3a5c359a | 5353 | int retval; |
1da177e4 LT |
5354 | |
5355 | if (pid < 0) | |
3a5c359a | 5356 | return -EINVAL; |
1da177e4 LT |
5357 | |
5358 | retval = -ESRCH; | |
5359 | read_lock(&tasklist_lock); | |
5360 | p = find_process_by_pid(pid); | |
5361 | if (p) { | |
5362 | retval = security_task_getscheduler(p); | |
5363 | if (!retval) | |
5364 | retval = p->policy; | |
5365 | } | |
5366 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5367 | return retval; |
5368 | } | |
5369 | ||
5370 | /** | |
5371 | * sys_sched_getscheduler - get the RT priority of a thread | |
5372 | * @pid: the pid in question. | |
5373 | * @param: structure containing the RT priority. | |
5374 | */ | |
5375 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5376 | { | |
5377 | struct sched_param lp; | |
36c8b586 | 5378 | struct task_struct *p; |
3a5c359a | 5379 | int retval; |
1da177e4 LT |
5380 | |
5381 | if (!param || pid < 0) | |
3a5c359a | 5382 | return -EINVAL; |
1da177e4 LT |
5383 | |
5384 | read_lock(&tasklist_lock); | |
5385 | p = find_process_by_pid(pid); | |
5386 | retval = -ESRCH; | |
5387 | if (!p) | |
5388 | goto out_unlock; | |
5389 | ||
5390 | retval = security_task_getscheduler(p); | |
5391 | if (retval) | |
5392 | goto out_unlock; | |
5393 | ||
5394 | lp.sched_priority = p->rt_priority; | |
5395 | read_unlock(&tasklist_lock); | |
5396 | ||
5397 | /* | |
5398 | * This one might sleep, we cannot do it with a spinlock held ... | |
5399 | */ | |
5400 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5401 | ||
1da177e4 LT |
5402 | return retval; |
5403 | ||
5404 | out_unlock: | |
5405 | read_unlock(&tasklist_lock); | |
5406 | return retval; | |
5407 | } | |
5408 | ||
b53e921b | 5409 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5410 | { |
1da177e4 | 5411 | cpumask_t cpus_allowed; |
b53e921b | 5412 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5413 | struct task_struct *p; |
5414 | int retval; | |
1da177e4 | 5415 | |
95402b38 | 5416 | get_online_cpus(); |
1da177e4 LT |
5417 | read_lock(&tasklist_lock); |
5418 | ||
5419 | p = find_process_by_pid(pid); | |
5420 | if (!p) { | |
5421 | read_unlock(&tasklist_lock); | |
95402b38 | 5422 | put_online_cpus(); |
1da177e4 LT |
5423 | return -ESRCH; |
5424 | } | |
5425 | ||
5426 | /* | |
5427 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5428 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5429 | * usage count and then drop tasklist_lock. |
5430 | */ | |
5431 | get_task_struct(p); | |
5432 | read_unlock(&tasklist_lock); | |
5433 | ||
5434 | retval = -EPERM; | |
5435 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5436 | !capable(CAP_SYS_NICE)) | |
5437 | goto out_unlock; | |
5438 | ||
e7834f8f DQ |
5439 | retval = security_task_setscheduler(p, 0, NULL); |
5440 | if (retval) | |
5441 | goto out_unlock; | |
5442 | ||
f9a86fcb | 5443 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5444 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5445 | again: |
7c16ec58 | 5446 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5447 | |
8707d8b8 | 5448 | if (!retval) { |
f9a86fcb | 5449 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5450 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5451 | /* | |
5452 | * We must have raced with a concurrent cpuset | |
5453 | * update. Just reset the cpus_allowed to the | |
5454 | * cpuset's cpus_allowed | |
5455 | */ | |
5456 | new_mask = cpus_allowed; | |
5457 | goto again; | |
5458 | } | |
5459 | } | |
1da177e4 LT |
5460 | out_unlock: |
5461 | put_task_struct(p); | |
95402b38 | 5462 | put_online_cpus(); |
1da177e4 LT |
5463 | return retval; |
5464 | } | |
5465 | ||
5466 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5467 | cpumask_t *new_mask) | |
5468 | { | |
5469 | if (len < sizeof(cpumask_t)) { | |
5470 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5471 | } else if (len > sizeof(cpumask_t)) { | |
5472 | len = sizeof(cpumask_t); | |
5473 | } | |
5474 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5475 | } | |
5476 | ||
5477 | /** | |
5478 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5479 | * @pid: pid of the process | |
5480 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5481 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5482 | */ | |
5483 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5484 | unsigned long __user *user_mask_ptr) | |
5485 | { | |
5486 | cpumask_t new_mask; | |
5487 | int retval; | |
5488 | ||
5489 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5490 | if (retval) | |
5491 | return retval; | |
5492 | ||
b53e921b | 5493 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5494 | } |
5495 | ||
1da177e4 LT |
5496 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5497 | { | |
36c8b586 | 5498 | struct task_struct *p; |
1da177e4 | 5499 | int retval; |
1da177e4 | 5500 | |
95402b38 | 5501 | get_online_cpus(); |
1da177e4 LT |
5502 | read_lock(&tasklist_lock); |
5503 | ||
5504 | retval = -ESRCH; | |
5505 | p = find_process_by_pid(pid); | |
5506 | if (!p) | |
5507 | goto out_unlock; | |
5508 | ||
e7834f8f DQ |
5509 | retval = security_task_getscheduler(p); |
5510 | if (retval) | |
5511 | goto out_unlock; | |
5512 | ||
2f7016d9 | 5513 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5514 | |
5515 | out_unlock: | |
5516 | read_unlock(&tasklist_lock); | |
95402b38 | 5517 | put_online_cpus(); |
1da177e4 | 5518 | |
9531b62f | 5519 | return retval; |
1da177e4 LT |
5520 | } |
5521 | ||
5522 | /** | |
5523 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5524 | * @pid: pid of the process | |
5525 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5526 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5527 | */ | |
5528 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5529 | unsigned long __user *user_mask_ptr) | |
5530 | { | |
5531 | int ret; | |
5532 | cpumask_t mask; | |
5533 | ||
5534 | if (len < sizeof(cpumask_t)) | |
5535 | return -EINVAL; | |
5536 | ||
5537 | ret = sched_getaffinity(pid, &mask); | |
5538 | if (ret < 0) | |
5539 | return ret; | |
5540 | ||
5541 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5542 | return -EFAULT; | |
5543 | ||
5544 | return sizeof(cpumask_t); | |
5545 | } | |
5546 | ||
5547 | /** | |
5548 | * sys_sched_yield - yield the current processor to other threads. | |
5549 | * | |
dd41f596 IM |
5550 | * This function yields the current CPU to other tasks. If there are no |
5551 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5552 | */ |
5553 | asmlinkage long sys_sched_yield(void) | |
5554 | { | |
70b97a7f | 5555 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5556 | |
2d72376b | 5557 | schedstat_inc(rq, yld_count); |
4530d7ab | 5558 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5559 | |
5560 | /* | |
5561 | * Since we are going to call schedule() anyway, there's | |
5562 | * no need to preempt or enable interrupts: | |
5563 | */ | |
5564 | __release(rq->lock); | |
8a25d5de | 5565 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5566 | _raw_spin_unlock(&rq->lock); |
5567 | preempt_enable_no_resched(); | |
5568 | ||
5569 | schedule(); | |
5570 | ||
5571 | return 0; | |
5572 | } | |
5573 | ||
e7b38404 | 5574 | static void __cond_resched(void) |
1da177e4 | 5575 | { |
8e0a43d8 IM |
5576 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5577 | __might_sleep(__FILE__, __LINE__); | |
5578 | #endif | |
5bbcfd90 IM |
5579 | /* |
5580 | * The BKS might be reacquired before we have dropped | |
5581 | * PREEMPT_ACTIVE, which could trigger a second | |
5582 | * cond_resched() call. | |
5583 | */ | |
1da177e4 LT |
5584 | do { |
5585 | add_preempt_count(PREEMPT_ACTIVE); | |
5586 | schedule(); | |
5587 | sub_preempt_count(PREEMPT_ACTIVE); | |
5588 | } while (need_resched()); | |
5589 | } | |
5590 | ||
02b67cc3 | 5591 | int __sched _cond_resched(void) |
1da177e4 | 5592 | { |
9414232f IM |
5593 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5594 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5595 | __cond_resched(); |
5596 | return 1; | |
5597 | } | |
5598 | return 0; | |
5599 | } | |
02b67cc3 | 5600 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5601 | |
5602 | /* | |
5603 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5604 | * call schedule, and on return reacquire the lock. | |
5605 | * | |
41a2d6cf | 5606 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5607 | * operations here to prevent schedule() from being called twice (once via |
5608 | * spin_unlock(), once by hand). | |
5609 | */ | |
95cdf3b7 | 5610 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5611 | { |
95c354fe | 5612 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5613 | int ret = 0; |
5614 | ||
95c354fe | 5615 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5616 | spin_unlock(lock); |
95c354fe NP |
5617 | if (resched && need_resched()) |
5618 | __cond_resched(); | |
5619 | else | |
5620 | cpu_relax(); | |
6df3cecb | 5621 | ret = 1; |
1da177e4 | 5622 | spin_lock(lock); |
1da177e4 | 5623 | } |
6df3cecb | 5624 | return ret; |
1da177e4 | 5625 | } |
1da177e4 LT |
5626 | EXPORT_SYMBOL(cond_resched_lock); |
5627 | ||
5628 | int __sched cond_resched_softirq(void) | |
5629 | { | |
5630 | BUG_ON(!in_softirq()); | |
5631 | ||
9414232f | 5632 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5633 | local_bh_enable(); |
1da177e4 LT |
5634 | __cond_resched(); |
5635 | local_bh_disable(); | |
5636 | return 1; | |
5637 | } | |
5638 | return 0; | |
5639 | } | |
1da177e4 LT |
5640 | EXPORT_SYMBOL(cond_resched_softirq); |
5641 | ||
1da177e4 LT |
5642 | /** |
5643 | * yield - yield the current processor to other threads. | |
5644 | * | |
72fd4a35 | 5645 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5646 | * thread runnable and calls sys_sched_yield(). |
5647 | */ | |
5648 | void __sched yield(void) | |
5649 | { | |
5650 | set_current_state(TASK_RUNNING); | |
5651 | sys_sched_yield(); | |
5652 | } | |
1da177e4 LT |
5653 | EXPORT_SYMBOL(yield); |
5654 | ||
5655 | /* | |
41a2d6cf | 5656 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5657 | * that process accounting knows that this is a task in IO wait state. |
5658 | * | |
5659 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5660 | * has set its backing_dev_info: the queue against which it should throttle) | |
5661 | */ | |
5662 | void __sched io_schedule(void) | |
5663 | { | |
70b97a7f | 5664 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5665 | |
0ff92245 | 5666 | delayacct_blkio_start(); |
1da177e4 LT |
5667 | atomic_inc(&rq->nr_iowait); |
5668 | schedule(); | |
5669 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5670 | delayacct_blkio_end(); |
1da177e4 | 5671 | } |
1da177e4 LT |
5672 | EXPORT_SYMBOL(io_schedule); |
5673 | ||
5674 | long __sched io_schedule_timeout(long timeout) | |
5675 | { | |
70b97a7f | 5676 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5677 | long ret; |
5678 | ||
0ff92245 | 5679 | delayacct_blkio_start(); |
1da177e4 LT |
5680 | atomic_inc(&rq->nr_iowait); |
5681 | ret = schedule_timeout(timeout); | |
5682 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5683 | delayacct_blkio_end(); |
1da177e4 LT |
5684 | return ret; |
5685 | } | |
5686 | ||
5687 | /** | |
5688 | * sys_sched_get_priority_max - return maximum RT priority. | |
5689 | * @policy: scheduling class. | |
5690 | * | |
5691 | * this syscall returns the maximum rt_priority that can be used | |
5692 | * by a given scheduling class. | |
5693 | */ | |
5694 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5695 | { | |
5696 | int ret = -EINVAL; | |
5697 | ||
5698 | switch (policy) { | |
5699 | case SCHED_FIFO: | |
5700 | case SCHED_RR: | |
5701 | ret = MAX_USER_RT_PRIO-1; | |
5702 | break; | |
5703 | case SCHED_NORMAL: | |
b0a9499c | 5704 | case SCHED_BATCH: |
dd41f596 | 5705 | case SCHED_IDLE: |
1da177e4 LT |
5706 | ret = 0; |
5707 | break; | |
5708 | } | |
5709 | return ret; | |
5710 | } | |
5711 | ||
5712 | /** | |
5713 | * sys_sched_get_priority_min - return minimum RT priority. | |
5714 | * @policy: scheduling class. | |
5715 | * | |
5716 | * this syscall returns the minimum rt_priority that can be used | |
5717 | * by a given scheduling class. | |
5718 | */ | |
5719 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5720 | { | |
5721 | int ret = -EINVAL; | |
5722 | ||
5723 | switch (policy) { | |
5724 | case SCHED_FIFO: | |
5725 | case SCHED_RR: | |
5726 | ret = 1; | |
5727 | break; | |
5728 | case SCHED_NORMAL: | |
b0a9499c | 5729 | case SCHED_BATCH: |
dd41f596 | 5730 | case SCHED_IDLE: |
1da177e4 LT |
5731 | ret = 0; |
5732 | } | |
5733 | return ret; | |
5734 | } | |
5735 | ||
5736 | /** | |
5737 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5738 | * @pid: pid of the process. | |
5739 | * @interval: userspace pointer to the timeslice value. | |
5740 | * | |
5741 | * this syscall writes the default timeslice value of a given process | |
5742 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5743 | */ | |
5744 | asmlinkage | |
5745 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5746 | { | |
36c8b586 | 5747 | struct task_struct *p; |
a4ec24b4 | 5748 | unsigned int time_slice; |
3a5c359a | 5749 | int retval; |
1da177e4 | 5750 | struct timespec t; |
1da177e4 LT |
5751 | |
5752 | if (pid < 0) | |
3a5c359a | 5753 | return -EINVAL; |
1da177e4 LT |
5754 | |
5755 | retval = -ESRCH; | |
5756 | read_lock(&tasklist_lock); | |
5757 | p = find_process_by_pid(pid); | |
5758 | if (!p) | |
5759 | goto out_unlock; | |
5760 | ||
5761 | retval = security_task_getscheduler(p); | |
5762 | if (retval) | |
5763 | goto out_unlock; | |
5764 | ||
77034937 IM |
5765 | /* |
5766 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5767 | * tasks that are on an otherwise idle runqueue: | |
5768 | */ | |
5769 | time_slice = 0; | |
5770 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5771 | time_slice = DEF_TIMESLICE; |
1868f958 | 5772 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5773 | struct sched_entity *se = &p->se; |
5774 | unsigned long flags; | |
5775 | struct rq *rq; | |
5776 | ||
5777 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5778 | if (rq->cfs.load.weight) |
5779 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5780 | task_rq_unlock(rq, &flags); |
5781 | } | |
1da177e4 | 5782 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5783 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5784 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5785 | return retval; |
3a5c359a | 5786 | |
1da177e4 LT |
5787 | out_unlock: |
5788 | read_unlock(&tasklist_lock); | |
5789 | return retval; | |
5790 | } | |
5791 | ||
7c731e0a | 5792 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5793 | |
82a1fcb9 | 5794 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5795 | { |
1da177e4 | 5796 | unsigned long free = 0; |
36c8b586 | 5797 | unsigned state; |
1da177e4 | 5798 | |
1da177e4 | 5799 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5800 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5801 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5802 | #if BITS_PER_LONG == 32 |
1da177e4 | 5803 | if (state == TASK_RUNNING) |
cc4ea795 | 5804 | printk(KERN_CONT " running "); |
1da177e4 | 5805 | else |
cc4ea795 | 5806 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5807 | #else |
5808 | if (state == TASK_RUNNING) | |
cc4ea795 | 5809 | printk(KERN_CONT " running task "); |
1da177e4 | 5810 | else |
cc4ea795 | 5811 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5812 | #endif |
5813 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5814 | { | |
10ebffde | 5815 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5816 | while (!*n) |
5817 | n++; | |
10ebffde | 5818 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5819 | } |
5820 | #endif | |
ba25f9dc | 5821 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5822 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5823 | |
5fb5e6de | 5824 | show_stack(p, NULL); |
1da177e4 LT |
5825 | } |
5826 | ||
e59e2ae2 | 5827 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5828 | { |
36c8b586 | 5829 | struct task_struct *g, *p; |
1da177e4 | 5830 | |
4bd77321 IM |
5831 | #if BITS_PER_LONG == 32 |
5832 | printk(KERN_INFO | |
5833 | " task PC stack pid father\n"); | |
1da177e4 | 5834 | #else |
4bd77321 IM |
5835 | printk(KERN_INFO |
5836 | " task PC stack pid father\n"); | |
1da177e4 LT |
5837 | #endif |
5838 | read_lock(&tasklist_lock); | |
5839 | do_each_thread(g, p) { | |
5840 | /* | |
5841 | * reset the NMI-timeout, listing all files on a slow | |
5842 | * console might take alot of time: | |
5843 | */ | |
5844 | touch_nmi_watchdog(); | |
39bc89fd | 5845 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5846 | sched_show_task(p); |
1da177e4 LT |
5847 | } while_each_thread(g, p); |
5848 | ||
04c9167f JF |
5849 | touch_all_softlockup_watchdogs(); |
5850 | ||
dd41f596 IM |
5851 | #ifdef CONFIG_SCHED_DEBUG |
5852 | sysrq_sched_debug_show(); | |
5853 | #endif | |
1da177e4 | 5854 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5855 | /* |
5856 | * Only show locks if all tasks are dumped: | |
5857 | */ | |
5858 | if (state_filter == -1) | |
5859 | debug_show_all_locks(); | |
1da177e4 LT |
5860 | } |
5861 | ||
1df21055 IM |
5862 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5863 | { | |
dd41f596 | 5864 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5865 | } |
5866 | ||
f340c0d1 IM |
5867 | /** |
5868 | * init_idle - set up an idle thread for a given CPU | |
5869 | * @idle: task in question | |
5870 | * @cpu: cpu the idle task belongs to | |
5871 | * | |
5872 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5873 | * flag, to make booting more robust. | |
5874 | */ | |
5c1e1767 | 5875 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5876 | { |
70b97a7f | 5877 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5878 | unsigned long flags; |
5879 | ||
5cbd54ef IM |
5880 | spin_lock_irqsave(&rq->lock, flags); |
5881 | ||
dd41f596 IM |
5882 | __sched_fork(idle); |
5883 | idle->se.exec_start = sched_clock(); | |
5884 | ||
b29739f9 | 5885 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5886 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5887 | __set_task_cpu(idle, cpu); |
1da177e4 | 5888 | |
1da177e4 | 5889 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5890 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5891 | idle->oncpu = 1; | |
5892 | #endif | |
1da177e4 LT |
5893 | spin_unlock_irqrestore(&rq->lock, flags); |
5894 | ||
5895 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5896 | #if defined(CONFIG_PREEMPT) |
5897 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5898 | #else | |
a1261f54 | 5899 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5900 | #endif |
dd41f596 IM |
5901 | /* |
5902 | * The idle tasks have their own, simple scheduling class: | |
5903 | */ | |
5904 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5905 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5906 | } |
5907 | ||
5908 | /* | |
5909 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5910 | * indicates which cpus entered this state. This is used | |
5911 | * in the rcu update to wait only for active cpus. For system | |
5912 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5913 | * always be CPU_MASK_NONE. | |
5914 | */ | |
5915 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5916 | ||
19978ca6 IM |
5917 | /* |
5918 | * Increase the granularity value when there are more CPUs, | |
5919 | * because with more CPUs the 'effective latency' as visible | |
5920 | * to users decreases. But the relationship is not linear, | |
5921 | * so pick a second-best guess by going with the log2 of the | |
5922 | * number of CPUs. | |
5923 | * | |
5924 | * This idea comes from the SD scheduler of Con Kolivas: | |
5925 | */ | |
5926 | static inline void sched_init_granularity(void) | |
5927 | { | |
5928 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5929 | const unsigned long limit = 200000000; | |
5930 | ||
5931 | sysctl_sched_min_granularity *= factor; | |
5932 | if (sysctl_sched_min_granularity > limit) | |
5933 | sysctl_sched_min_granularity = limit; | |
5934 | ||
5935 | sysctl_sched_latency *= factor; | |
5936 | if (sysctl_sched_latency > limit) | |
5937 | sysctl_sched_latency = limit; | |
5938 | ||
5939 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5940 | |
5941 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5942 | } |
5943 | ||
1da177e4 LT |
5944 | #ifdef CONFIG_SMP |
5945 | /* | |
5946 | * This is how migration works: | |
5947 | * | |
70b97a7f | 5948 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5949 | * runqueue and wake up that CPU's migration thread. |
5950 | * 2) we down() the locked semaphore => thread blocks. | |
5951 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5952 | * thread off the CPU) | |
5953 | * 4) it gets the migration request and checks whether the migrated | |
5954 | * task is still in the wrong runqueue. | |
5955 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5956 | * it and puts it into the right queue. | |
5957 | * 6) migration thread up()s the semaphore. | |
5958 | * 7) we wake up and the migration is done. | |
5959 | */ | |
5960 | ||
5961 | /* | |
5962 | * Change a given task's CPU affinity. Migrate the thread to a | |
5963 | * proper CPU and schedule it away if the CPU it's executing on | |
5964 | * is removed from the allowed bitmask. | |
5965 | * | |
5966 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5967 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5968 | * call is not atomic; no spinlocks may be held. |
5969 | */ | |
cd8ba7cd | 5970 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5971 | { |
70b97a7f | 5972 | struct migration_req req; |
1da177e4 | 5973 | unsigned long flags; |
70b97a7f | 5974 | struct rq *rq; |
48f24c4d | 5975 | int ret = 0; |
1da177e4 LT |
5976 | |
5977 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5978 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5979 | ret = -EINVAL; |
5980 | goto out; | |
5981 | } | |
5982 | ||
9985b0ba DR |
5983 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5984 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5985 | ret = -EINVAL; | |
5986 | goto out; | |
5987 | } | |
5988 | ||
73fe6aae | 5989 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5990 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5991 | else { |
cd8ba7cd MT |
5992 | p->cpus_allowed = *new_mask; |
5993 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5994 | } |
5995 | ||
1da177e4 | 5996 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5997 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5998 | goto out; |
5999 | ||
cd8ba7cd | 6000 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
6001 | /* Need help from migration thread: drop lock and wait. */ |
6002 | task_rq_unlock(rq, &flags); | |
6003 | wake_up_process(rq->migration_thread); | |
6004 | wait_for_completion(&req.done); | |
6005 | tlb_migrate_finish(p->mm); | |
6006 | return 0; | |
6007 | } | |
6008 | out: | |
6009 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6010 | |
1da177e4 LT |
6011 | return ret; |
6012 | } | |
cd8ba7cd | 6013 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6014 | |
6015 | /* | |
41a2d6cf | 6016 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6017 | * this because either it can't run here any more (set_cpus_allowed() |
6018 | * away from this CPU, or CPU going down), or because we're | |
6019 | * attempting to rebalance this task on exec (sched_exec). | |
6020 | * | |
6021 | * So we race with normal scheduler movements, but that's OK, as long | |
6022 | * as the task is no longer on this CPU. | |
efc30814 KK |
6023 | * |
6024 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6025 | */ |
efc30814 | 6026 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6027 | { |
70b97a7f | 6028 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6029 | int ret = 0, on_rq; |
1da177e4 | 6030 | |
e761b772 | 6031 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6032 | return ret; |
1da177e4 LT |
6033 | |
6034 | rq_src = cpu_rq(src_cpu); | |
6035 | rq_dest = cpu_rq(dest_cpu); | |
6036 | ||
6037 | double_rq_lock(rq_src, rq_dest); | |
6038 | /* Already moved. */ | |
6039 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6040 | goto done; |
1da177e4 LT |
6041 | /* Affinity changed (again). */ |
6042 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 6043 | goto fail; |
1da177e4 | 6044 | |
dd41f596 | 6045 | on_rq = p->se.on_rq; |
6e82a3be | 6046 | if (on_rq) |
2e1cb74a | 6047 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6048 | |
1da177e4 | 6049 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6050 | if (on_rq) { |
6051 | activate_task(rq_dest, p, 0); | |
15afe09b | 6052 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6053 | } |
b1e38734 | 6054 | done: |
efc30814 | 6055 | ret = 1; |
b1e38734 | 6056 | fail: |
1da177e4 | 6057 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6058 | return ret; |
1da177e4 LT |
6059 | } |
6060 | ||
6061 | /* | |
6062 | * migration_thread - this is a highprio system thread that performs | |
6063 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6064 | * another runqueue. | |
6065 | */ | |
95cdf3b7 | 6066 | static int migration_thread(void *data) |
1da177e4 | 6067 | { |
1da177e4 | 6068 | int cpu = (long)data; |
70b97a7f | 6069 | struct rq *rq; |
1da177e4 LT |
6070 | |
6071 | rq = cpu_rq(cpu); | |
6072 | BUG_ON(rq->migration_thread != current); | |
6073 | ||
6074 | set_current_state(TASK_INTERRUPTIBLE); | |
6075 | while (!kthread_should_stop()) { | |
70b97a7f | 6076 | struct migration_req *req; |
1da177e4 | 6077 | struct list_head *head; |
1da177e4 | 6078 | |
1da177e4 LT |
6079 | spin_lock_irq(&rq->lock); |
6080 | ||
6081 | if (cpu_is_offline(cpu)) { | |
6082 | spin_unlock_irq(&rq->lock); | |
6083 | goto wait_to_die; | |
6084 | } | |
6085 | ||
6086 | if (rq->active_balance) { | |
6087 | active_load_balance(rq, cpu); | |
6088 | rq->active_balance = 0; | |
6089 | } | |
6090 | ||
6091 | head = &rq->migration_queue; | |
6092 | ||
6093 | if (list_empty(head)) { | |
6094 | spin_unlock_irq(&rq->lock); | |
6095 | schedule(); | |
6096 | set_current_state(TASK_INTERRUPTIBLE); | |
6097 | continue; | |
6098 | } | |
70b97a7f | 6099 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6100 | list_del_init(head->next); |
6101 | ||
674311d5 NP |
6102 | spin_unlock(&rq->lock); |
6103 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6104 | local_irq_enable(); | |
1da177e4 LT |
6105 | |
6106 | complete(&req->done); | |
6107 | } | |
6108 | __set_current_state(TASK_RUNNING); | |
6109 | return 0; | |
6110 | ||
6111 | wait_to_die: | |
6112 | /* Wait for kthread_stop */ | |
6113 | set_current_state(TASK_INTERRUPTIBLE); | |
6114 | while (!kthread_should_stop()) { | |
6115 | schedule(); | |
6116 | set_current_state(TASK_INTERRUPTIBLE); | |
6117 | } | |
6118 | __set_current_state(TASK_RUNNING); | |
6119 | return 0; | |
6120 | } | |
6121 | ||
6122 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6123 | |
6124 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6125 | { | |
6126 | int ret; | |
6127 | ||
6128 | local_irq_disable(); | |
6129 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6130 | local_irq_enable(); | |
6131 | return ret; | |
6132 | } | |
6133 | ||
054b9108 | 6134 | /* |
3a4fa0a2 | 6135 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6136 | * NOTE: interrupts should be disabled by the caller |
6137 | */ | |
48f24c4d | 6138 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6139 | { |
efc30814 | 6140 | unsigned long flags; |
1da177e4 | 6141 | cpumask_t mask; |
70b97a7f IM |
6142 | struct rq *rq; |
6143 | int dest_cpu; | |
1da177e4 | 6144 | |
3a5c359a AK |
6145 | do { |
6146 | /* On same node? */ | |
6147 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6148 | cpus_and(mask, mask, p->cpus_allowed); | |
6149 | dest_cpu = any_online_cpu(mask); | |
6150 | ||
6151 | /* On any allowed CPU? */ | |
434d53b0 | 6152 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6153 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6154 | ||
6155 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6156 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6157 | cpumask_t cpus_allowed; |
6158 | ||
6159 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6160 | /* |
6161 | * Try to stay on the same cpuset, where the | |
6162 | * current cpuset may be a subset of all cpus. | |
6163 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6164 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6165 | * called within calls to cpuset_lock/cpuset_unlock. |
6166 | */ | |
3a5c359a | 6167 | rq = task_rq_lock(p, &flags); |
470fd646 | 6168 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6169 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6170 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6171 | |
3a5c359a AK |
6172 | /* |
6173 | * Don't tell them about moving exiting tasks or | |
6174 | * kernel threads (both mm NULL), since they never | |
6175 | * leave kernel. | |
6176 | */ | |
41a2d6cf | 6177 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6178 | printk(KERN_INFO "process %d (%s) no " |
6179 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6180 | task_pid_nr(p), p->comm, dead_cpu); |
6181 | } | |
3a5c359a | 6182 | } |
f7b4cddc | 6183 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6184 | } |
6185 | ||
6186 | /* | |
6187 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6188 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6189 | * for performance reasons the counter is not stricly tracking tasks to | |
6190 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6191 | * to keep the global sum constant after CPU-down: | |
6192 | */ | |
70b97a7f | 6193 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6194 | { |
7c16ec58 | 6195 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6196 | unsigned long flags; |
6197 | ||
6198 | local_irq_save(flags); | |
6199 | double_rq_lock(rq_src, rq_dest); | |
6200 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6201 | rq_src->nr_uninterruptible = 0; | |
6202 | double_rq_unlock(rq_src, rq_dest); | |
6203 | local_irq_restore(flags); | |
6204 | } | |
6205 | ||
6206 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6207 | static void migrate_live_tasks(int src_cpu) | |
6208 | { | |
48f24c4d | 6209 | struct task_struct *p, *t; |
1da177e4 | 6210 | |
f7b4cddc | 6211 | read_lock(&tasklist_lock); |
1da177e4 | 6212 | |
48f24c4d IM |
6213 | do_each_thread(t, p) { |
6214 | if (p == current) | |
1da177e4 LT |
6215 | continue; |
6216 | ||
48f24c4d IM |
6217 | if (task_cpu(p) == src_cpu) |
6218 | move_task_off_dead_cpu(src_cpu, p); | |
6219 | } while_each_thread(t, p); | |
1da177e4 | 6220 | |
f7b4cddc | 6221 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6222 | } |
6223 | ||
dd41f596 IM |
6224 | /* |
6225 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6226 | * It does so by boosting its priority to highest possible. |
6227 | * Used by CPU offline code. | |
1da177e4 LT |
6228 | */ |
6229 | void sched_idle_next(void) | |
6230 | { | |
48f24c4d | 6231 | int this_cpu = smp_processor_id(); |
70b97a7f | 6232 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6233 | struct task_struct *p = rq->idle; |
6234 | unsigned long flags; | |
6235 | ||
6236 | /* cpu has to be offline */ | |
48f24c4d | 6237 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6238 | |
48f24c4d IM |
6239 | /* |
6240 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6241 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6242 | */ |
6243 | spin_lock_irqsave(&rq->lock, flags); | |
6244 | ||
dd41f596 | 6245 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6246 | |
94bc9a7b DA |
6247 | update_rq_clock(rq); |
6248 | activate_task(rq, p, 0); | |
1da177e4 LT |
6249 | |
6250 | spin_unlock_irqrestore(&rq->lock, flags); | |
6251 | } | |
6252 | ||
48f24c4d IM |
6253 | /* |
6254 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6255 | * offline. |
6256 | */ | |
6257 | void idle_task_exit(void) | |
6258 | { | |
6259 | struct mm_struct *mm = current->active_mm; | |
6260 | ||
6261 | BUG_ON(cpu_online(smp_processor_id())); | |
6262 | ||
6263 | if (mm != &init_mm) | |
6264 | switch_mm(mm, &init_mm, current); | |
6265 | mmdrop(mm); | |
6266 | } | |
6267 | ||
054b9108 | 6268 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6269 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6270 | { |
70b97a7f | 6271 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6272 | |
6273 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6274 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6275 | |
6276 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6277 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6278 | |
48f24c4d | 6279 | get_task_struct(p); |
1da177e4 LT |
6280 | |
6281 | /* | |
6282 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6283 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6284 | * fine. |
6285 | */ | |
f7b4cddc | 6286 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6287 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6288 | spin_lock_irq(&rq->lock); |
1da177e4 | 6289 | |
48f24c4d | 6290 | put_task_struct(p); |
1da177e4 LT |
6291 | } |
6292 | ||
6293 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6294 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6295 | { | |
70b97a7f | 6296 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6297 | struct task_struct *next; |
48f24c4d | 6298 | |
dd41f596 IM |
6299 | for ( ; ; ) { |
6300 | if (!rq->nr_running) | |
6301 | break; | |
a8e504d2 | 6302 | update_rq_clock(rq); |
ff95f3df | 6303 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6304 | if (!next) |
6305 | break; | |
79c53799 | 6306 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6307 | migrate_dead(dead_cpu, next); |
e692ab53 | 6308 | |
1da177e4 LT |
6309 | } |
6310 | } | |
6311 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6312 | ||
e692ab53 NP |
6313 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6314 | ||
6315 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6316 | { |
6317 | .procname = "sched_domain", | |
c57baf1e | 6318 | .mode = 0555, |
e0361851 | 6319 | }, |
38605cae | 6320 | {0, }, |
e692ab53 NP |
6321 | }; |
6322 | ||
6323 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6324 | { |
c57baf1e | 6325 | .ctl_name = CTL_KERN, |
e0361851 | 6326 | .procname = "kernel", |
c57baf1e | 6327 | .mode = 0555, |
e0361851 AD |
6328 | .child = sd_ctl_dir, |
6329 | }, | |
38605cae | 6330 | {0, }, |
e692ab53 NP |
6331 | }; |
6332 | ||
6333 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6334 | { | |
6335 | struct ctl_table *entry = | |
5cf9f062 | 6336 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6337 | |
e692ab53 NP |
6338 | return entry; |
6339 | } | |
6340 | ||
6382bc90 MM |
6341 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6342 | { | |
cd790076 | 6343 | struct ctl_table *entry; |
6382bc90 | 6344 | |
cd790076 MM |
6345 | /* |
6346 | * In the intermediate directories, both the child directory and | |
6347 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6348 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6349 | * static strings and all have proc handlers. |
6350 | */ | |
6351 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6352 | if (entry->child) |
6353 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6354 | if (entry->proc_handler == NULL) |
6355 | kfree(entry->procname); | |
6356 | } | |
6382bc90 MM |
6357 | |
6358 | kfree(*tablep); | |
6359 | *tablep = NULL; | |
6360 | } | |
6361 | ||
e692ab53 | 6362 | static void |
e0361851 | 6363 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6364 | const char *procname, void *data, int maxlen, |
6365 | mode_t mode, proc_handler *proc_handler) | |
6366 | { | |
e692ab53 NP |
6367 | entry->procname = procname; |
6368 | entry->data = data; | |
6369 | entry->maxlen = maxlen; | |
6370 | entry->mode = mode; | |
6371 | entry->proc_handler = proc_handler; | |
6372 | } | |
6373 | ||
6374 | static struct ctl_table * | |
6375 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6376 | { | |
a5d8c348 | 6377 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6378 | |
ad1cdc1d MM |
6379 | if (table == NULL) |
6380 | return NULL; | |
6381 | ||
e0361851 | 6382 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6383 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6384 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6385 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6386 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6387 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6388 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6389 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6390 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6391 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6392 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6393 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6394 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6395 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6396 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6397 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6398 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6399 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6400 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6401 | &sd->cache_nice_tries, |
6402 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6403 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6404 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6405 | set_table_entry(&table[11], "name", sd->name, |
6406 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6407 | /* &table[12] is terminator */ | |
e692ab53 NP |
6408 | |
6409 | return table; | |
6410 | } | |
6411 | ||
9a4e7159 | 6412 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6413 | { |
6414 | struct ctl_table *entry, *table; | |
6415 | struct sched_domain *sd; | |
6416 | int domain_num = 0, i; | |
6417 | char buf[32]; | |
6418 | ||
6419 | for_each_domain(cpu, sd) | |
6420 | domain_num++; | |
6421 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6422 | if (table == NULL) |
6423 | return NULL; | |
e692ab53 NP |
6424 | |
6425 | i = 0; | |
6426 | for_each_domain(cpu, sd) { | |
6427 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6428 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6429 | entry->mode = 0555; |
e692ab53 NP |
6430 | entry->child = sd_alloc_ctl_domain_table(sd); |
6431 | entry++; | |
6432 | i++; | |
6433 | } | |
6434 | return table; | |
6435 | } | |
6436 | ||
6437 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6438 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6439 | { |
6440 | int i, cpu_num = num_online_cpus(); | |
6441 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6442 | char buf[32]; | |
6443 | ||
7378547f MM |
6444 | WARN_ON(sd_ctl_dir[0].child); |
6445 | sd_ctl_dir[0].child = entry; | |
6446 | ||
ad1cdc1d MM |
6447 | if (entry == NULL) |
6448 | return; | |
6449 | ||
97b6ea7b | 6450 | for_each_online_cpu(i) { |
e692ab53 | 6451 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6452 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6453 | entry->mode = 0555; |
e692ab53 | 6454 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6455 | entry++; |
e692ab53 | 6456 | } |
7378547f MM |
6457 | |
6458 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6459 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6460 | } | |
6382bc90 | 6461 | |
7378547f | 6462 | /* may be called multiple times per register */ |
6382bc90 MM |
6463 | static void unregister_sched_domain_sysctl(void) |
6464 | { | |
7378547f MM |
6465 | if (sd_sysctl_header) |
6466 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6467 | sd_sysctl_header = NULL; |
7378547f MM |
6468 | if (sd_ctl_dir[0].child) |
6469 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6470 | } |
e692ab53 | 6471 | #else |
6382bc90 MM |
6472 | static void register_sched_domain_sysctl(void) |
6473 | { | |
6474 | } | |
6475 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6476 | { |
6477 | } | |
6478 | #endif | |
6479 | ||
1f11eb6a GH |
6480 | static void set_rq_online(struct rq *rq) |
6481 | { | |
6482 | if (!rq->online) { | |
6483 | const struct sched_class *class; | |
6484 | ||
6485 | cpu_set(rq->cpu, rq->rd->online); | |
6486 | rq->online = 1; | |
6487 | ||
6488 | for_each_class(class) { | |
6489 | if (class->rq_online) | |
6490 | class->rq_online(rq); | |
6491 | } | |
6492 | } | |
6493 | } | |
6494 | ||
6495 | static void set_rq_offline(struct rq *rq) | |
6496 | { | |
6497 | if (rq->online) { | |
6498 | const struct sched_class *class; | |
6499 | ||
6500 | for_each_class(class) { | |
6501 | if (class->rq_offline) | |
6502 | class->rq_offline(rq); | |
6503 | } | |
6504 | ||
6505 | cpu_clear(rq->cpu, rq->rd->online); | |
6506 | rq->online = 0; | |
6507 | } | |
6508 | } | |
6509 | ||
1da177e4 LT |
6510 | /* |
6511 | * migration_call - callback that gets triggered when a CPU is added. | |
6512 | * Here we can start up the necessary migration thread for the new CPU. | |
6513 | */ | |
48f24c4d IM |
6514 | static int __cpuinit |
6515 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6516 | { |
1da177e4 | 6517 | struct task_struct *p; |
48f24c4d | 6518 | int cpu = (long)hcpu; |
1da177e4 | 6519 | unsigned long flags; |
70b97a7f | 6520 | struct rq *rq; |
1da177e4 LT |
6521 | |
6522 | switch (action) { | |
5be9361c | 6523 | |
1da177e4 | 6524 | case CPU_UP_PREPARE: |
8bb78442 | 6525 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6526 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6527 | if (IS_ERR(p)) |
6528 | return NOTIFY_BAD; | |
1da177e4 LT |
6529 | kthread_bind(p, cpu); |
6530 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6531 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6532 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6533 | task_rq_unlock(rq, &flags); |
6534 | cpu_rq(cpu)->migration_thread = p; | |
6535 | break; | |
48f24c4d | 6536 | |
1da177e4 | 6537 | case CPU_ONLINE: |
8bb78442 | 6538 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6539 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6540 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6541 | |
6542 | /* Update our root-domain */ | |
6543 | rq = cpu_rq(cpu); | |
6544 | spin_lock_irqsave(&rq->lock, flags); | |
6545 | if (rq->rd) { | |
6546 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6547 | |
6548 | set_rq_online(rq); | |
1f94ef59 GH |
6549 | } |
6550 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6551 | break; |
48f24c4d | 6552 | |
1da177e4 LT |
6553 | #ifdef CONFIG_HOTPLUG_CPU |
6554 | case CPU_UP_CANCELED: | |
8bb78442 | 6555 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6556 | if (!cpu_rq(cpu)->migration_thread) |
6557 | break; | |
41a2d6cf | 6558 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6559 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6560 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6561 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6562 | cpu_rq(cpu)->migration_thread = NULL; | |
6563 | break; | |
48f24c4d | 6564 | |
1da177e4 | 6565 | case CPU_DEAD: |
8bb78442 | 6566 | case CPU_DEAD_FROZEN: |
470fd646 | 6567 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6568 | migrate_live_tasks(cpu); |
6569 | rq = cpu_rq(cpu); | |
6570 | kthread_stop(rq->migration_thread); | |
6571 | rq->migration_thread = NULL; | |
6572 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6573 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6574 | update_rq_clock(rq); |
2e1cb74a | 6575 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6576 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6577 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6578 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6579 | migrate_dead_tasks(cpu); |
d2da272a | 6580 | spin_unlock_irq(&rq->lock); |
470fd646 | 6581 | cpuset_unlock(); |
1da177e4 LT |
6582 | migrate_nr_uninterruptible(rq); |
6583 | BUG_ON(rq->nr_running != 0); | |
6584 | ||
41a2d6cf IM |
6585 | /* |
6586 | * No need to migrate the tasks: it was best-effort if | |
6587 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6588 | * the requestors. | |
6589 | */ | |
1da177e4 LT |
6590 | spin_lock_irq(&rq->lock); |
6591 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6592 | struct migration_req *req; |
6593 | ||
1da177e4 | 6594 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6595 | struct migration_req, list); |
1da177e4 LT |
6596 | list_del_init(&req->list); |
6597 | complete(&req->done); | |
6598 | } | |
6599 | spin_unlock_irq(&rq->lock); | |
6600 | break; | |
57d885fe | 6601 | |
08f503b0 GH |
6602 | case CPU_DYING: |
6603 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6604 | /* Update our root-domain */ |
6605 | rq = cpu_rq(cpu); | |
6606 | spin_lock_irqsave(&rq->lock, flags); | |
6607 | if (rq->rd) { | |
6608 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6609 | set_rq_offline(rq); |
57d885fe GH |
6610 | } |
6611 | spin_unlock_irqrestore(&rq->lock, flags); | |
6612 | break; | |
1da177e4 LT |
6613 | #endif |
6614 | } | |
6615 | return NOTIFY_OK; | |
6616 | } | |
6617 | ||
6618 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6619 | * happens before everything else. | |
6620 | */ | |
26c2143b | 6621 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6622 | .notifier_call = migration_call, |
6623 | .priority = 10 | |
6624 | }; | |
6625 | ||
7babe8db | 6626 | static int __init migration_init(void) |
1da177e4 LT |
6627 | { |
6628 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6629 | int err; |
48f24c4d IM |
6630 | |
6631 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6632 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6633 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6634 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6635 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6636 | |
6637 | return err; | |
1da177e4 | 6638 | } |
7babe8db | 6639 | early_initcall(migration_init); |
1da177e4 LT |
6640 | #endif |
6641 | ||
6642 | #ifdef CONFIG_SMP | |
476f3534 | 6643 | |
3e9830dc | 6644 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6645 | |
099f98c8 GS |
6646 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6647 | { | |
6648 | switch (lvl) { | |
6649 | case SD_LV_NONE: | |
6650 | return "NONE"; | |
6651 | case SD_LV_SIBLING: | |
6652 | return "SIBLING"; | |
6653 | case SD_LV_MC: | |
6654 | return "MC"; | |
6655 | case SD_LV_CPU: | |
6656 | return "CPU"; | |
6657 | case SD_LV_NODE: | |
6658 | return "NODE"; | |
6659 | case SD_LV_ALLNODES: | |
6660 | return "ALLNODES"; | |
6661 | case SD_LV_MAX: | |
6662 | return "MAX"; | |
6663 | ||
6664 | } | |
6665 | return "MAX"; | |
6666 | } | |
6667 | ||
7c16ec58 MT |
6668 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6669 | cpumask_t *groupmask) | |
1da177e4 | 6670 | { |
4dcf6aff | 6671 | struct sched_group *group = sd->groups; |
434d53b0 | 6672 | char str[256]; |
1da177e4 | 6673 | |
434d53b0 | 6674 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6675 | cpus_clear(*groupmask); |
4dcf6aff IM |
6676 | |
6677 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6678 | ||
6679 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6680 | printk("does not load-balance\n"); | |
6681 | if (sd->parent) | |
6682 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6683 | " has parent"); | |
6684 | return -1; | |
41c7ce9a NP |
6685 | } |
6686 | ||
099f98c8 GS |
6687 | printk(KERN_CONT "span %s level %s\n", |
6688 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6689 | |
6690 | if (!cpu_isset(cpu, sd->span)) { | |
6691 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6692 | "CPU%d\n", cpu); | |
6693 | } | |
6694 | if (!cpu_isset(cpu, group->cpumask)) { | |
6695 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6696 | " CPU%d\n", cpu); | |
6697 | } | |
1da177e4 | 6698 | |
4dcf6aff | 6699 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6700 | do { |
4dcf6aff IM |
6701 | if (!group) { |
6702 | printk("\n"); | |
6703 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6704 | break; |
6705 | } | |
6706 | ||
4dcf6aff IM |
6707 | if (!group->__cpu_power) { |
6708 | printk(KERN_CONT "\n"); | |
6709 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6710 | "set\n"); | |
6711 | break; | |
6712 | } | |
1da177e4 | 6713 | |
4dcf6aff IM |
6714 | if (!cpus_weight(group->cpumask)) { |
6715 | printk(KERN_CONT "\n"); | |
6716 | printk(KERN_ERR "ERROR: empty group\n"); | |
6717 | break; | |
6718 | } | |
1da177e4 | 6719 | |
7c16ec58 | 6720 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6721 | printk(KERN_CONT "\n"); |
6722 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6723 | break; | |
6724 | } | |
1da177e4 | 6725 | |
7c16ec58 | 6726 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6727 | |
434d53b0 | 6728 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6729 | printk(KERN_CONT " %s", str); |
1da177e4 | 6730 | |
4dcf6aff IM |
6731 | group = group->next; |
6732 | } while (group != sd->groups); | |
6733 | printk(KERN_CONT "\n"); | |
1da177e4 | 6734 | |
7c16ec58 | 6735 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6736 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6737 | |
7c16ec58 | 6738 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6739 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6740 | "of domain->span\n"); | |
6741 | return 0; | |
6742 | } | |
1da177e4 | 6743 | |
4dcf6aff IM |
6744 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6745 | { | |
7c16ec58 | 6746 | cpumask_t *groupmask; |
4dcf6aff | 6747 | int level = 0; |
1da177e4 | 6748 | |
4dcf6aff IM |
6749 | if (!sd) { |
6750 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6751 | return; | |
6752 | } | |
1da177e4 | 6753 | |
4dcf6aff IM |
6754 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6755 | ||
7c16ec58 MT |
6756 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6757 | if (!groupmask) { | |
6758 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6759 | return; | |
6760 | } | |
6761 | ||
4dcf6aff | 6762 | for (;;) { |
7c16ec58 | 6763 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6764 | break; |
1da177e4 LT |
6765 | level++; |
6766 | sd = sd->parent; | |
33859f7f | 6767 | if (!sd) |
4dcf6aff IM |
6768 | break; |
6769 | } | |
7c16ec58 | 6770 | kfree(groupmask); |
1da177e4 | 6771 | } |
6d6bc0ad | 6772 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6773 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6774 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6775 | |
1a20ff27 | 6776 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6777 | { |
6778 | if (cpus_weight(sd->span) == 1) | |
6779 | return 1; | |
6780 | ||
6781 | /* Following flags need at least 2 groups */ | |
6782 | if (sd->flags & (SD_LOAD_BALANCE | | |
6783 | SD_BALANCE_NEWIDLE | | |
6784 | SD_BALANCE_FORK | | |
89c4710e SS |
6785 | SD_BALANCE_EXEC | |
6786 | SD_SHARE_CPUPOWER | | |
6787 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6788 | if (sd->groups != sd->groups->next) |
6789 | return 0; | |
6790 | } | |
6791 | ||
6792 | /* Following flags don't use groups */ | |
6793 | if (sd->flags & (SD_WAKE_IDLE | | |
6794 | SD_WAKE_AFFINE | | |
6795 | SD_WAKE_BALANCE)) | |
6796 | return 0; | |
6797 | ||
6798 | return 1; | |
6799 | } | |
6800 | ||
48f24c4d IM |
6801 | static int |
6802 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6803 | { |
6804 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6805 | ||
6806 | if (sd_degenerate(parent)) | |
6807 | return 1; | |
6808 | ||
6809 | if (!cpus_equal(sd->span, parent->span)) | |
6810 | return 0; | |
6811 | ||
6812 | /* Does parent contain flags not in child? */ | |
6813 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6814 | if (cflags & SD_WAKE_AFFINE) | |
6815 | pflags &= ~SD_WAKE_BALANCE; | |
6816 | /* Flags needing groups don't count if only 1 group in parent */ | |
6817 | if (parent->groups == parent->groups->next) { | |
6818 | pflags &= ~(SD_LOAD_BALANCE | | |
6819 | SD_BALANCE_NEWIDLE | | |
6820 | SD_BALANCE_FORK | | |
89c4710e SS |
6821 | SD_BALANCE_EXEC | |
6822 | SD_SHARE_CPUPOWER | | |
6823 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6824 | } |
6825 | if (~cflags & pflags) | |
6826 | return 0; | |
6827 | ||
6828 | return 1; | |
6829 | } | |
6830 | ||
57d885fe GH |
6831 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6832 | { | |
6833 | unsigned long flags; | |
57d885fe GH |
6834 | |
6835 | spin_lock_irqsave(&rq->lock, flags); | |
6836 | ||
6837 | if (rq->rd) { | |
6838 | struct root_domain *old_rd = rq->rd; | |
6839 | ||
1f11eb6a GH |
6840 | if (cpu_isset(rq->cpu, old_rd->online)) |
6841 | set_rq_offline(rq); | |
57d885fe | 6842 | |
dc938520 | 6843 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6844 | |
57d885fe GH |
6845 | if (atomic_dec_and_test(&old_rd->refcount)) |
6846 | kfree(old_rd); | |
6847 | } | |
6848 | ||
6849 | atomic_inc(&rd->refcount); | |
6850 | rq->rd = rd; | |
6851 | ||
dc938520 | 6852 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6853 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6854 | set_rq_online(rq); |
57d885fe GH |
6855 | |
6856 | spin_unlock_irqrestore(&rq->lock, flags); | |
6857 | } | |
6858 | ||
dc938520 | 6859 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6860 | { |
6861 | memset(rd, 0, sizeof(*rd)); | |
6862 | ||
dc938520 GH |
6863 | cpus_clear(rd->span); |
6864 | cpus_clear(rd->online); | |
6e0534f2 GH |
6865 | |
6866 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6867 | } |
6868 | ||
6869 | static void init_defrootdomain(void) | |
6870 | { | |
dc938520 | 6871 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6872 | atomic_set(&def_root_domain.refcount, 1); |
6873 | } | |
6874 | ||
dc938520 | 6875 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6876 | { |
6877 | struct root_domain *rd; | |
6878 | ||
6879 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6880 | if (!rd) | |
6881 | return NULL; | |
6882 | ||
dc938520 | 6883 | init_rootdomain(rd); |
57d885fe GH |
6884 | |
6885 | return rd; | |
6886 | } | |
6887 | ||
1da177e4 | 6888 | /* |
0eab9146 | 6889 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6890 | * hold the hotplug lock. |
6891 | */ | |
0eab9146 IM |
6892 | static void |
6893 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6894 | { |
70b97a7f | 6895 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6896 | struct sched_domain *tmp; |
6897 | ||
6898 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6899 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6900 | struct sched_domain *parent = tmp->parent; |
6901 | if (!parent) | |
6902 | break; | |
f29c9b1c | 6903 | |
1a848870 | 6904 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6905 | tmp->parent = parent->parent; |
1a848870 SS |
6906 | if (parent->parent) |
6907 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6908 | } else |
6909 | tmp = tmp->parent; | |
245af2c7 SS |
6910 | } |
6911 | ||
1a848870 | 6912 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6913 | sd = sd->parent; |
1a848870 SS |
6914 | if (sd) |
6915 | sd->child = NULL; | |
6916 | } | |
1da177e4 LT |
6917 | |
6918 | sched_domain_debug(sd, cpu); | |
6919 | ||
57d885fe | 6920 | rq_attach_root(rq, rd); |
674311d5 | 6921 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6922 | } |
6923 | ||
6924 | /* cpus with isolated domains */ | |
67af63a6 | 6925 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6926 | |
6927 | /* Setup the mask of cpus configured for isolated domains */ | |
6928 | static int __init isolated_cpu_setup(char *str) | |
6929 | { | |
13b40c1e MT |
6930 | static int __initdata ints[NR_CPUS]; |
6931 | int i; | |
1da177e4 LT |
6932 | |
6933 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6934 | cpus_clear(cpu_isolated_map); | |
6935 | for (i = 1; i <= ints[0]; i++) | |
6936 | if (ints[i] < NR_CPUS) | |
6937 | cpu_set(ints[i], cpu_isolated_map); | |
6938 | return 1; | |
6939 | } | |
6940 | ||
8927f494 | 6941 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6942 | |
6943 | /* | |
6711cab4 SS |
6944 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6945 | * to a function which identifies what group(along with sched group) a CPU | |
6946 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6947 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6948 | * |
6949 | * init_sched_build_groups will build a circular linked list of the groups | |
6950 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6951 | * and ->cpu_power to 0. | |
6952 | */ | |
a616058b | 6953 | static void |
7c16ec58 | 6954 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6955 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6956 | struct sched_group **sg, |
6957 | cpumask_t *tmpmask), | |
6958 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6959 | { |
6960 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6961 | int i; |
6962 | ||
7c16ec58 MT |
6963 | cpus_clear(*covered); |
6964 | ||
363ab6f1 | 6965 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6966 | struct sched_group *sg; |
7c16ec58 | 6967 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6968 | int j; |
6969 | ||
7c16ec58 | 6970 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6971 | continue; |
6972 | ||
7c16ec58 | 6973 | cpus_clear(sg->cpumask); |
5517d86b | 6974 | sg->__cpu_power = 0; |
1da177e4 | 6975 | |
363ab6f1 | 6976 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6977 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6978 | continue; |
6979 | ||
7c16ec58 | 6980 | cpu_set(j, *covered); |
1da177e4 LT |
6981 | cpu_set(j, sg->cpumask); |
6982 | } | |
6983 | if (!first) | |
6984 | first = sg; | |
6985 | if (last) | |
6986 | last->next = sg; | |
6987 | last = sg; | |
6988 | } | |
6989 | last->next = first; | |
6990 | } | |
6991 | ||
9c1cfda2 | 6992 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6993 | |
9c1cfda2 | 6994 | #ifdef CONFIG_NUMA |
198e2f18 | 6995 | |
9c1cfda2 JH |
6996 | /** |
6997 | * find_next_best_node - find the next node to include in a sched_domain | |
6998 | * @node: node whose sched_domain we're building | |
6999 | * @used_nodes: nodes already in the sched_domain | |
7000 | * | |
41a2d6cf | 7001 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7002 | * finds the closest node not already in the @used_nodes map. |
7003 | * | |
7004 | * Should use nodemask_t. | |
7005 | */ | |
c5f59f08 | 7006 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7007 | { |
7008 | int i, n, val, min_val, best_node = 0; | |
7009 | ||
7010 | min_val = INT_MAX; | |
7011 | ||
076ac2af | 7012 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7013 | /* Start at @node */ |
076ac2af | 7014 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7015 | |
7016 | if (!nr_cpus_node(n)) | |
7017 | continue; | |
7018 | ||
7019 | /* Skip already used nodes */ | |
c5f59f08 | 7020 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7021 | continue; |
7022 | ||
7023 | /* Simple min distance search */ | |
7024 | val = node_distance(node, n); | |
7025 | ||
7026 | if (val < min_val) { | |
7027 | min_val = val; | |
7028 | best_node = n; | |
7029 | } | |
7030 | } | |
7031 | ||
c5f59f08 | 7032 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7033 | return best_node; |
7034 | } | |
7035 | ||
7036 | /** | |
7037 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7038 | * @node: node whose cpumask we're constructing | |
73486722 | 7039 | * @span: resulting cpumask |
9c1cfda2 | 7040 | * |
41a2d6cf | 7041 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7042 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7043 | * out optimally. | |
7044 | */ | |
4bdbaad3 | 7045 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 7046 | { |
c5f59f08 | 7047 | nodemask_t used_nodes; |
c5f59f08 | 7048 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7049 | int i; |
9c1cfda2 | 7050 | |
4bdbaad3 | 7051 | cpus_clear(*span); |
c5f59f08 | 7052 | nodes_clear(used_nodes); |
9c1cfda2 | 7053 | |
4bdbaad3 | 7054 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7055 | node_set(node, used_nodes); |
9c1cfda2 JH |
7056 | |
7057 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7058 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7059 | |
c5f59f08 | 7060 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7061 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7062 | } |
9c1cfda2 | 7063 | } |
6d6bc0ad | 7064 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7065 | |
5c45bf27 | 7066 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7067 | |
9c1cfda2 | 7068 | /* |
48f24c4d | 7069 | * SMT sched-domains: |
9c1cfda2 | 7070 | */ |
1da177e4 LT |
7071 | #ifdef CONFIG_SCHED_SMT |
7072 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7073 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7074 | |
41a2d6cf | 7075 | static int |
7c16ec58 MT |
7076 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7077 | cpumask_t *unused) | |
1da177e4 | 7078 | { |
6711cab4 SS |
7079 | if (sg) |
7080 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7081 | return cpu; |
7082 | } | |
6d6bc0ad | 7083 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7084 | |
48f24c4d IM |
7085 | /* |
7086 | * multi-core sched-domains: | |
7087 | */ | |
1e9f28fa SS |
7088 | #ifdef CONFIG_SCHED_MC |
7089 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7090 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7091 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7092 | |
7093 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7094 | static int |
7c16ec58 MT |
7095 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7096 | cpumask_t *mask) | |
1e9f28fa | 7097 | { |
6711cab4 | 7098 | int group; |
7c16ec58 MT |
7099 | |
7100 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7101 | cpus_and(*mask, *mask, *cpu_map); | |
7102 | group = first_cpu(*mask); | |
6711cab4 SS |
7103 | if (sg) |
7104 | *sg = &per_cpu(sched_group_core, group); | |
7105 | return group; | |
1e9f28fa SS |
7106 | } |
7107 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7108 | static int |
7c16ec58 MT |
7109 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7110 | cpumask_t *unused) | |
1e9f28fa | 7111 | { |
6711cab4 SS |
7112 | if (sg) |
7113 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7114 | return cpu; |
7115 | } | |
7116 | #endif | |
7117 | ||
1da177e4 | 7118 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7119 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7120 | |
41a2d6cf | 7121 | static int |
7c16ec58 MT |
7122 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7123 | cpumask_t *mask) | |
1da177e4 | 7124 | { |
6711cab4 | 7125 | int group; |
48f24c4d | 7126 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7127 | *mask = cpu_coregroup_map(cpu); |
7128 | cpus_and(*mask, *mask, *cpu_map); | |
7129 | group = first_cpu(*mask); | |
1e9f28fa | 7130 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7131 | *mask = per_cpu(cpu_sibling_map, cpu); |
7132 | cpus_and(*mask, *mask, *cpu_map); | |
7133 | group = first_cpu(*mask); | |
1da177e4 | 7134 | #else |
6711cab4 | 7135 | group = cpu; |
1da177e4 | 7136 | #endif |
6711cab4 SS |
7137 | if (sg) |
7138 | *sg = &per_cpu(sched_group_phys, group); | |
7139 | return group; | |
1da177e4 LT |
7140 | } |
7141 | ||
7142 | #ifdef CONFIG_NUMA | |
1da177e4 | 7143 | /* |
9c1cfda2 JH |
7144 | * The init_sched_build_groups can't handle what we want to do with node |
7145 | * groups, so roll our own. Now each node has its own list of groups which | |
7146 | * gets dynamically allocated. | |
1da177e4 | 7147 | */ |
9c1cfda2 | 7148 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7149 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7150 | |
9c1cfda2 | 7151 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7152 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7153 | |
6711cab4 | 7154 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7155 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7156 | { |
6711cab4 SS |
7157 | int group; |
7158 | ||
7c16ec58 MT |
7159 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7160 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7161 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7162 | |
7163 | if (sg) | |
7164 | *sg = &per_cpu(sched_group_allnodes, group); | |
7165 | return group; | |
1da177e4 | 7166 | } |
6711cab4 | 7167 | |
08069033 SS |
7168 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7169 | { | |
7170 | struct sched_group *sg = group_head; | |
7171 | int j; | |
7172 | ||
7173 | if (!sg) | |
7174 | return; | |
3a5c359a | 7175 | do { |
363ab6f1 | 7176 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7177 | struct sched_domain *sd; |
08069033 | 7178 | |
3a5c359a AK |
7179 | sd = &per_cpu(phys_domains, j); |
7180 | if (j != first_cpu(sd->groups->cpumask)) { | |
7181 | /* | |
7182 | * Only add "power" once for each | |
7183 | * physical package. | |
7184 | */ | |
7185 | continue; | |
7186 | } | |
08069033 | 7187 | |
3a5c359a AK |
7188 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7189 | } | |
7190 | sg = sg->next; | |
7191 | } while (sg != group_head); | |
08069033 | 7192 | } |
6d6bc0ad | 7193 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7194 | |
a616058b | 7195 | #ifdef CONFIG_NUMA |
51888ca2 | 7196 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7197 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7198 | { |
a616058b | 7199 | int cpu, i; |
51888ca2 | 7200 | |
363ab6f1 | 7201 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7202 | struct sched_group **sched_group_nodes |
7203 | = sched_group_nodes_bycpu[cpu]; | |
7204 | ||
51888ca2 SV |
7205 | if (!sched_group_nodes) |
7206 | continue; | |
7207 | ||
076ac2af | 7208 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7209 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7210 | ||
7c16ec58 MT |
7211 | *nodemask = node_to_cpumask(i); |
7212 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7213 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7214 | continue; |
7215 | ||
7216 | if (sg == NULL) | |
7217 | continue; | |
7218 | sg = sg->next; | |
7219 | next_sg: | |
7220 | oldsg = sg; | |
7221 | sg = sg->next; | |
7222 | kfree(oldsg); | |
7223 | if (oldsg != sched_group_nodes[i]) | |
7224 | goto next_sg; | |
7225 | } | |
7226 | kfree(sched_group_nodes); | |
7227 | sched_group_nodes_bycpu[cpu] = NULL; | |
7228 | } | |
51888ca2 | 7229 | } |
6d6bc0ad | 7230 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7231 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7232 | { |
7233 | } | |
6d6bc0ad | 7234 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7235 | |
89c4710e SS |
7236 | /* |
7237 | * Initialize sched groups cpu_power. | |
7238 | * | |
7239 | * cpu_power indicates the capacity of sched group, which is used while | |
7240 | * distributing the load between different sched groups in a sched domain. | |
7241 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7242 | * there are asymmetries in the topology. If there are asymmetries, group | |
7243 | * having more cpu_power will pickup more load compared to the group having | |
7244 | * less cpu_power. | |
7245 | * | |
7246 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7247 | * the maximum number of tasks a group can handle in the presence of other idle | |
7248 | * or lightly loaded groups in the same sched domain. | |
7249 | */ | |
7250 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7251 | { | |
7252 | struct sched_domain *child; | |
7253 | struct sched_group *group; | |
7254 | ||
7255 | WARN_ON(!sd || !sd->groups); | |
7256 | ||
7257 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7258 | return; | |
7259 | ||
7260 | child = sd->child; | |
7261 | ||
5517d86b ED |
7262 | sd->groups->__cpu_power = 0; |
7263 | ||
89c4710e SS |
7264 | /* |
7265 | * For perf policy, if the groups in child domain share resources | |
7266 | * (for example cores sharing some portions of the cache hierarchy | |
7267 | * or SMT), then set this domain groups cpu_power such that each group | |
7268 | * can handle only one task, when there are other idle groups in the | |
7269 | * same sched domain. | |
7270 | */ | |
7271 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7272 | (child->flags & | |
7273 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7274 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7275 | return; |
7276 | } | |
7277 | ||
89c4710e SS |
7278 | /* |
7279 | * add cpu_power of each child group to this groups cpu_power | |
7280 | */ | |
7281 | group = child->groups; | |
7282 | do { | |
5517d86b | 7283 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7284 | group = group->next; |
7285 | } while (group != child->groups); | |
7286 | } | |
7287 | ||
7c16ec58 MT |
7288 | /* |
7289 | * Initializers for schedule domains | |
7290 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7291 | */ | |
7292 | ||
a5d8c348 IM |
7293 | #ifdef CONFIG_SCHED_DEBUG |
7294 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7295 | #else | |
7296 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7297 | #endif | |
7298 | ||
7c16ec58 | 7299 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7300 | |
7c16ec58 MT |
7301 | #define SD_INIT_FUNC(type) \ |
7302 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7303 | { \ | |
7304 | memset(sd, 0, sizeof(*sd)); \ | |
7305 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7306 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7307 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7308 | } |
7309 | ||
7310 | SD_INIT_FUNC(CPU) | |
7311 | #ifdef CONFIG_NUMA | |
7312 | SD_INIT_FUNC(ALLNODES) | |
7313 | SD_INIT_FUNC(NODE) | |
7314 | #endif | |
7315 | #ifdef CONFIG_SCHED_SMT | |
7316 | SD_INIT_FUNC(SIBLING) | |
7317 | #endif | |
7318 | #ifdef CONFIG_SCHED_MC | |
7319 | SD_INIT_FUNC(MC) | |
7320 | #endif | |
7321 | ||
7322 | /* | |
7323 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7324 | * space as the usage in build_sched_domains() dictates. Used only | |
7325 | * if the amount of space is significant. | |
7326 | */ | |
7327 | struct allmasks { | |
7328 | cpumask_t tmpmask; /* make this one first */ | |
7329 | union { | |
7330 | cpumask_t nodemask; | |
7331 | cpumask_t this_sibling_map; | |
7332 | cpumask_t this_core_map; | |
7333 | }; | |
7334 | cpumask_t send_covered; | |
7335 | ||
7336 | #ifdef CONFIG_NUMA | |
7337 | cpumask_t domainspan; | |
7338 | cpumask_t covered; | |
7339 | cpumask_t notcovered; | |
7340 | #endif | |
7341 | }; | |
7342 | ||
7343 | #if NR_CPUS > 128 | |
7344 | #define SCHED_CPUMASK_ALLOC 1 | |
7345 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7346 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7347 | #else | |
7348 | #define SCHED_CPUMASK_ALLOC 0 | |
7349 | #define SCHED_CPUMASK_FREE(v) | |
7350 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7351 | #endif | |
7352 | ||
7353 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7354 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7355 | ||
1d3504fc HS |
7356 | static int default_relax_domain_level = -1; |
7357 | ||
7358 | static int __init setup_relax_domain_level(char *str) | |
7359 | { | |
30e0e178 LZ |
7360 | unsigned long val; |
7361 | ||
7362 | val = simple_strtoul(str, NULL, 0); | |
7363 | if (val < SD_LV_MAX) | |
7364 | default_relax_domain_level = val; | |
7365 | ||
1d3504fc HS |
7366 | return 1; |
7367 | } | |
7368 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7369 | ||
7370 | static void set_domain_attribute(struct sched_domain *sd, | |
7371 | struct sched_domain_attr *attr) | |
7372 | { | |
7373 | int request; | |
7374 | ||
7375 | if (!attr || attr->relax_domain_level < 0) { | |
7376 | if (default_relax_domain_level < 0) | |
7377 | return; | |
7378 | else | |
7379 | request = default_relax_domain_level; | |
7380 | } else | |
7381 | request = attr->relax_domain_level; | |
7382 | if (request < sd->level) { | |
7383 | /* turn off idle balance on this domain */ | |
7384 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7385 | } else { | |
7386 | /* turn on idle balance on this domain */ | |
7387 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7388 | } | |
7389 | } | |
7390 | ||
1da177e4 | 7391 | /* |
1a20ff27 DG |
7392 | * Build sched domains for a given set of cpus and attach the sched domains |
7393 | * to the individual cpus | |
1da177e4 | 7394 | */ |
1d3504fc HS |
7395 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7396 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7397 | { |
7398 | int i; | |
57d885fe | 7399 | struct root_domain *rd; |
7c16ec58 MT |
7400 | SCHED_CPUMASK_DECLARE(allmasks); |
7401 | cpumask_t *tmpmask; | |
d1b55138 JH |
7402 | #ifdef CONFIG_NUMA |
7403 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7404 | int sd_allnodes = 0; |
d1b55138 JH |
7405 | |
7406 | /* | |
7407 | * Allocate the per-node list of sched groups | |
7408 | */ | |
076ac2af | 7409 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7410 | GFP_KERNEL); |
d1b55138 JH |
7411 | if (!sched_group_nodes) { |
7412 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7413 | return -ENOMEM; |
d1b55138 | 7414 | } |
d1b55138 | 7415 | #endif |
1da177e4 | 7416 | |
dc938520 | 7417 | rd = alloc_rootdomain(); |
57d885fe GH |
7418 | if (!rd) { |
7419 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7420 | #ifdef CONFIG_NUMA |
7421 | kfree(sched_group_nodes); | |
7422 | #endif | |
57d885fe GH |
7423 | return -ENOMEM; |
7424 | } | |
7425 | ||
7c16ec58 MT |
7426 | #if SCHED_CPUMASK_ALLOC |
7427 | /* get space for all scratch cpumask variables */ | |
7428 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7429 | if (!allmasks) { | |
7430 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7431 | kfree(rd); | |
7432 | #ifdef CONFIG_NUMA | |
7433 | kfree(sched_group_nodes); | |
7434 | #endif | |
7435 | return -ENOMEM; | |
7436 | } | |
7437 | #endif | |
7438 | tmpmask = (cpumask_t *)allmasks; | |
7439 | ||
7440 | ||
7441 | #ifdef CONFIG_NUMA | |
7442 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7443 | #endif | |
7444 | ||
1da177e4 | 7445 | /* |
1a20ff27 | 7446 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7447 | */ |
363ab6f1 | 7448 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7449 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7450 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7451 | |
7c16ec58 MT |
7452 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7453 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7454 | |
7455 | #ifdef CONFIG_NUMA | |
dd41f596 | 7456 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7457 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7458 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7459 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7460 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7461 | sd->span = *cpu_map; |
7c16ec58 | 7462 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7463 | p = sd; |
6711cab4 | 7464 | sd_allnodes = 1; |
9c1cfda2 JH |
7465 | } else |
7466 | p = NULL; | |
7467 | ||
1da177e4 | 7468 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7469 | SD_INIT(sd, NODE); |
1d3504fc | 7470 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7471 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7472 | sd->parent = p; |
1a848870 SS |
7473 | if (p) |
7474 | p->child = sd; | |
9c1cfda2 | 7475 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7476 | #endif |
7477 | ||
7478 | p = sd; | |
7479 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7480 | SD_INIT(sd, CPU); |
1d3504fc | 7481 | set_domain_attribute(sd, attr); |
7c16ec58 | 7482 | sd->span = *nodemask; |
1da177e4 | 7483 | sd->parent = p; |
1a848870 SS |
7484 | if (p) |
7485 | p->child = sd; | |
7c16ec58 | 7486 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7487 | |
1e9f28fa SS |
7488 | #ifdef CONFIG_SCHED_MC |
7489 | p = sd; | |
7490 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7491 | SD_INIT(sd, MC); |
1d3504fc | 7492 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7493 | sd->span = cpu_coregroup_map(i); |
7494 | cpus_and(sd->span, sd->span, *cpu_map); | |
7495 | sd->parent = p; | |
1a848870 | 7496 | p->child = sd; |
7c16ec58 | 7497 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7498 | #endif |
7499 | ||
1da177e4 LT |
7500 | #ifdef CONFIG_SCHED_SMT |
7501 | p = sd; | |
7502 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7503 | SD_INIT(sd, SIBLING); |
1d3504fc | 7504 | set_domain_attribute(sd, attr); |
d5a7430d | 7505 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7506 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7507 | sd->parent = p; |
1a848870 | 7508 | p->child = sd; |
7c16ec58 | 7509 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7510 | #endif |
7511 | } | |
7512 | ||
7513 | #ifdef CONFIG_SCHED_SMT | |
7514 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7515 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7516 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7517 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7518 | ||
7519 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7520 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7521 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7522 | continue; |
7523 | ||
dd41f596 | 7524 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7525 | &cpu_to_cpu_group, |
7526 | send_covered, tmpmask); | |
1da177e4 LT |
7527 | } |
7528 | #endif | |
7529 | ||
1e9f28fa SS |
7530 | #ifdef CONFIG_SCHED_MC |
7531 | /* Set up multi-core groups */ | |
363ab6f1 | 7532 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7533 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7534 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7535 | ||
7536 | *this_core_map = cpu_coregroup_map(i); | |
7537 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7538 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7539 | continue; |
7c16ec58 | 7540 | |
dd41f596 | 7541 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7542 | &cpu_to_core_group, |
7543 | send_covered, tmpmask); | |
1e9f28fa SS |
7544 | } |
7545 | #endif | |
7546 | ||
1da177e4 | 7547 | /* Set up physical groups */ |
076ac2af | 7548 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7549 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7550 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7551 | |
7c16ec58 MT |
7552 | *nodemask = node_to_cpumask(i); |
7553 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7554 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7555 | continue; |
7556 | ||
7c16ec58 MT |
7557 | init_sched_build_groups(nodemask, cpu_map, |
7558 | &cpu_to_phys_group, | |
7559 | send_covered, tmpmask); | |
1da177e4 LT |
7560 | } |
7561 | ||
7562 | #ifdef CONFIG_NUMA | |
7563 | /* Set up node groups */ | |
7c16ec58 MT |
7564 | if (sd_allnodes) { |
7565 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7566 | ||
7567 | init_sched_build_groups(cpu_map, cpu_map, | |
7568 | &cpu_to_allnodes_group, | |
7569 | send_covered, tmpmask); | |
7570 | } | |
9c1cfda2 | 7571 | |
076ac2af | 7572 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7573 | /* Set up node groups */ |
7574 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7575 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7576 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7577 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7578 | int j; |
7579 | ||
7c16ec58 MT |
7580 | *nodemask = node_to_cpumask(i); |
7581 | cpus_clear(*covered); | |
7582 | ||
7583 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7584 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7585 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7586 | continue; |
d1b55138 | 7587 | } |
9c1cfda2 | 7588 | |
4bdbaad3 | 7589 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7590 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7591 | |
15f0b676 | 7592 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7593 | if (!sg) { |
7594 | printk(KERN_WARNING "Can not alloc domain group for " | |
7595 | "node %d\n", i); | |
7596 | goto error; | |
7597 | } | |
9c1cfda2 | 7598 | sched_group_nodes[i] = sg; |
363ab6f1 | 7599 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7600 | struct sched_domain *sd; |
9761eea8 | 7601 | |
9c1cfda2 JH |
7602 | sd = &per_cpu(node_domains, j); |
7603 | sd->groups = sg; | |
9c1cfda2 | 7604 | } |
5517d86b | 7605 | sg->__cpu_power = 0; |
7c16ec58 | 7606 | sg->cpumask = *nodemask; |
51888ca2 | 7607 | sg->next = sg; |
7c16ec58 | 7608 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7609 | prev = sg; |
7610 | ||
076ac2af | 7611 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7612 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7613 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7614 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7615 | |
7c16ec58 MT |
7616 | cpus_complement(*notcovered, *covered); |
7617 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7618 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7619 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7620 | break; |
7621 | ||
7c16ec58 MT |
7622 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7623 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7624 | continue; |
7625 | ||
15f0b676 SV |
7626 | sg = kmalloc_node(sizeof(struct sched_group), |
7627 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7628 | if (!sg) { |
7629 | printk(KERN_WARNING | |
7630 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7631 | goto error; |
9c1cfda2 | 7632 | } |
5517d86b | 7633 | sg->__cpu_power = 0; |
7c16ec58 | 7634 | sg->cpumask = *tmpmask; |
51888ca2 | 7635 | sg->next = prev->next; |
7c16ec58 | 7636 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7637 | prev->next = sg; |
7638 | prev = sg; | |
7639 | } | |
9c1cfda2 | 7640 | } |
1da177e4 LT |
7641 | #endif |
7642 | ||
7643 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7644 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7645 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7646 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7647 | ||
89c4710e | 7648 | init_sched_groups_power(i, sd); |
5c45bf27 | 7649 | } |
1da177e4 | 7650 | #endif |
1e9f28fa | 7651 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7652 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7653 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7654 | ||
89c4710e | 7655 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7656 | } |
7657 | #endif | |
1e9f28fa | 7658 | |
363ab6f1 | 7659 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7660 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7661 | ||
89c4710e | 7662 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7663 | } |
7664 | ||
9c1cfda2 | 7665 | #ifdef CONFIG_NUMA |
076ac2af | 7666 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7667 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7668 | |
6711cab4 SS |
7669 | if (sd_allnodes) { |
7670 | struct sched_group *sg; | |
f712c0c7 | 7671 | |
7c16ec58 MT |
7672 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7673 | tmpmask); | |
f712c0c7 SS |
7674 | init_numa_sched_groups_power(sg); |
7675 | } | |
9c1cfda2 JH |
7676 | #endif |
7677 | ||
1da177e4 | 7678 | /* Attach the domains */ |
363ab6f1 | 7679 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7680 | struct sched_domain *sd; |
7681 | #ifdef CONFIG_SCHED_SMT | |
7682 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7683 | #elif defined(CONFIG_SCHED_MC) |
7684 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7685 | #else |
7686 | sd = &per_cpu(phys_domains, i); | |
7687 | #endif | |
57d885fe | 7688 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7689 | } |
51888ca2 | 7690 | |
7c16ec58 | 7691 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7692 | return 0; |
7693 | ||
a616058b | 7694 | #ifdef CONFIG_NUMA |
51888ca2 | 7695 | error: |
7c16ec58 MT |
7696 | free_sched_groups(cpu_map, tmpmask); |
7697 | SCHED_CPUMASK_FREE((void *)allmasks); | |
ca3273f9 | 7698 | kfree(rd); |
51888ca2 | 7699 | return -ENOMEM; |
a616058b | 7700 | #endif |
1da177e4 | 7701 | } |
029190c5 | 7702 | |
1d3504fc HS |
7703 | static int build_sched_domains(const cpumask_t *cpu_map) |
7704 | { | |
7705 | return __build_sched_domains(cpu_map, NULL); | |
7706 | } | |
7707 | ||
029190c5 PJ |
7708 | static cpumask_t *doms_cur; /* current sched domains */ |
7709 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7710 | static struct sched_domain_attr *dattr_cur; |
7711 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7712 | |
7713 | /* | |
7714 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7715 | * cpumask_t) fails, then fallback to a single sched domain, | |
7716 | * as determined by the single cpumask_t fallback_doms. | |
7717 | */ | |
7718 | static cpumask_t fallback_doms; | |
7719 | ||
22e52b07 HC |
7720 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7721 | { | |
7722 | } | |
7723 | ||
1a20ff27 | 7724 | /* |
41a2d6cf | 7725 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7726 | * For now this just excludes isolated cpus, but could be used to |
7727 | * exclude other special cases in the future. | |
1a20ff27 | 7728 | */ |
51888ca2 | 7729 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7730 | { |
7378547f MM |
7731 | int err; |
7732 | ||
22e52b07 | 7733 | arch_update_cpu_topology(); |
029190c5 PJ |
7734 | ndoms_cur = 1; |
7735 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7736 | if (!doms_cur) | |
7737 | doms_cur = &fallback_doms; | |
7738 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7739 | dattr_cur = NULL; |
7378547f | 7740 | err = build_sched_domains(doms_cur); |
6382bc90 | 7741 | register_sched_domain_sysctl(); |
7378547f MM |
7742 | |
7743 | return err; | |
1a20ff27 DG |
7744 | } |
7745 | ||
7c16ec58 MT |
7746 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7747 | cpumask_t *tmpmask) | |
1da177e4 | 7748 | { |
7c16ec58 | 7749 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7750 | } |
1da177e4 | 7751 | |
1a20ff27 DG |
7752 | /* |
7753 | * Detach sched domains from a group of cpus specified in cpu_map | |
7754 | * These cpus will now be attached to the NULL domain | |
7755 | */ | |
858119e1 | 7756 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7757 | { |
7c16ec58 | 7758 | cpumask_t tmpmask; |
1a20ff27 DG |
7759 | int i; |
7760 | ||
6382bc90 MM |
7761 | unregister_sched_domain_sysctl(); |
7762 | ||
363ab6f1 | 7763 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7764 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7765 | synchronize_sched(); |
7c16ec58 | 7766 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7767 | } |
7768 | ||
1d3504fc HS |
7769 | /* handle null as "default" */ |
7770 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7771 | struct sched_domain_attr *new, int idx_new) | |
7772 | { | |
7773 | struct sched_domain_attr tmp; | |
7774 | ||
7775 | /* fast path */ | |
7776 | if (!new && !cur) | |
7777 | return 1; | |
7778 | ||
7779 | tmp = SD_ATTR_INIT; | |
7780 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7781 | new ? (new + idx_new) : &tmp, | |
7782 | sizeof(struct sched_domain_attr)); | |
7783 | } | |
7784 | ||
029190c5 PJ |
7785 | /* |
7786 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7787 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7788 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7789 | * It destroys each deleted domain and builds each new domain. | |
7790 | * | |
7791 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7792 | * The masks don't intersect (don't overlap.) We should setup one |
7793 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7794 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7795 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7796 | * it as it is. | |
7797 | * | |
41a2d6cf IM |
7798 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7799 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
7800 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
7801 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
7802 | * the single partition 'fallback_doms', it also forces the domains | |
7803 | * to be rebuilt. | |
029190c5 | 7804 | * |
700018e0 LZ |
7805 | * If doms_new == NULL it will be replaced with cpu_online_map. |
7806 | * ndoms_new == 0 is a special case for destroying existing domains, | |
7807 | * and it will not create the default domain. | |
dfb512ec | 7808 | * |
029190c5 PJ |
7809 | * Call with hotplug lock held |
7810 | */ | |
1d3504fc HS |
7811 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7812 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7813 | { |
dfb512ec | 7814 | int i, j, n; |
029190c5 | 7815 | |
712555ee | 7816 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7817 | |
7378547f MM |
7818 | /* always unregister in case we don't destroy any domains */ |
7819 | unregister_sched_domain_sysctl(); | |
7820 | ||
dfb512ec | 7821 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7822 | |
7823 | /* Destroy deleted domains */ | |
7824 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7825 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7826 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7827 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7828 | goto match1; |
7829 | } | |
7830 | /* no match - a current sched domain not in new doms_new[] */ | |
7831 | detach_destroy_domains(doms_cur + i); | |
7832 | match1: | |
7833 | ; | |
7834 | } | |
7835 | ||
e761b772 MK |
7836 | if (doms_new == NULL) { |
7837 | ndoms_cur = 0; | |
e761b772 MK |
7838 | doms_new = &fallback_doms; |
7839 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7840 | dattr_new = NULL; | |
7841 | } | |
7842 | ||
029190c5 PJ |
7843 | /* Build new domains */ |
7844 | for (i = 0; i < ndoms_new; i++) { | |
7845 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7846 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7847 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7848 | goto match2; |
7849 | } | |
7850 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7851 | __build_sched_domains(doms_new + i, |
7852 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7853 | match2: |
7854 | ; | |
7855 | } | |
7856 | ||
7857 | /* Remember the new sched domains */ | |
7858 | if (doms_cur != &fallback_doms) | |
7859 | kfree(doms_cur); | |
1d3504fc | 7860 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7861 | doms_cur = doms_new; |
1d3504fc | 7862 | dattr_cur = dattr_new; |
029190c5 | 7863 | ndoms_cur = ndoms_new; |
7378547f MM |
7864 | |
7865 | register_sched_domain_sysctl(); | |
a1835615 | 7866 | |
712555ee | 7867 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7868 | } |
7869 | ||
5c45bf27 | 7870 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7871 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7872 | { |
95402b38 | 7873 | get_online_cpus(); |
dfb512ec MK |
7874 | |
7875 | /* Destroy domains first to force the rebuild */ | |
7876 | partition_sched_domains(0, NULL, NULL); | |
7877 | ||
e761b772 | 7878 | rebuild_sched_domains(); |
95402b38 | 7879 | put_online_cpus(); |
dfb512ec | 7880 | |
e761b772 | 7881 | return 0; |
5c45bf27 SS |
7882 | } |
7883 | ||
7884 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7885 | { | |
7886 | int ret; | |
7887 | ||
7888 | if (buf[0] != '0' && buf[0] != '1') | |
7889 | return -EINVAL; | |
7890 | ||
7891 | if (smt) | |
7892 | sched_smt_power_savings = (buf[0] == '1'); | |
7893 | else | |
7894 | sched_mc_power_savings = (buf[0] == '1'); | |
7895 | ||
7896 | ret = arch_reinit_sched_domains(); | |
7897 | ||
7898 | return ret ? ret : count; | |
7899 | } | |
7900 | ||
5c45bf27 | 7901 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7902 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7903 | char *page) | |
5c45bf27 SS |
7904 | { |
7905 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7906 | } | |
f718cd4a | 7907 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7908 | const char *buf, size_t count) |
5c45bf27 SS |
7909 | { |
7910 | return sched_power_savings_store(buf, count, 0); | |
7911 | } | |
f718cd4a AK |
7912 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7913 | sched_mc_power_savings_show, | |
7914 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7915 | #endif |
7916 | ||
7917 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7918 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7919 | char *page) | |
5c45bf27 SS |
7920 | { |
7921 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7922 | } | |
f718cd4a | 7923 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7924 | const char *buf, size_t count) |
5c45bf27 SS |
7925 | { |
7926 | return sched_power_savings_store(buf, count, 1); | |
7927 | } | |
f718cd4a AK |
7928 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7929 | sched_smt_power_savings_show, | |
6707de00 AB |
7930 | sched_smt_power_savings_store); |
7931 | #endif | |
7932 | ||
7933 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7934 | { | |
7935 | int err = 0; | |
7936 | ||
7937 | #ifdef CONFIG_SCHED_SMT | |
7938 | if (smt_capable()) | |
7939 | err = sysfs_create_file(&cls->kset.kobj, | |
7940 | &attr_sched_smt_power_savings.attr); | |
7941 | #endif | |
7942 | #ifdef CONFIG_SCHED_MC | |
7943 | if (!err && mc_capable()) | |
7944 | err = sysfs_create_file(&cls->kset.kobj, | |
7945 | &attr_sched_mc_power_savings.attr); | |
7946 | #endif | |
7947 | return err; | |
7948 | } | |
6d6bc0ad | 7949 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7950 | |
e761b772 | 7951 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7952 | /* |
e761b772 MK |
7953 | * Add online and remove offline CPUs from the scheduler domains. |
7954 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7955 | */ |
7956 | static int update_sched_domains(struct notifier_block *nfb, | |
7957 | unsigned long action, void *hcpu) | |
e761b772 MK |
7958 | { |
7959 | switch (action) { | |
7960 | case CPU_ONLINE: | |
7961 | case CPU_ONLINE_FROZEN: | |
7962 | case CPU_DEAD: | |
7963 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7964 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7965 | return NOTIFY_OK; |
7966 | ||
7967 | default: | |
7968 | return NOTIFY_DONE; | |
7969 | } | |
7970 | } | |
7971 | #endif | |
7972 | ||
7973 | static int update_runtime(struct notifier_block *nfb, | |
7974 | unsigned long action, void *hcpu) | |
1da177e4 | 7975 | { |
7def2be1 PZ |
7976 | int cpu = (int)(long)hcpu; |
7977 | ||
1da177e4 | 7978 | switch (action) { |
1da177e4 | 7979 | case CPU_DOWN_PREPARE: |
8bb78442 | 7980 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7981 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7982 | return NOTIFY_OK; |
7983 | ||
1da177e4 | 7984 | case CPU_DOWN_FAILED: |
8bb78442 | 7985 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7986 | case CPU_ONLINE: |
8bb78442 | 7987 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7988 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7989 | return NOTIFY_OK; |
7990 | ||
1da177e4 LT |
7991 | default: |
7992 | return NOTIFY_DONE; | |
7993 | } | |
1da177e4 | 7994 | } |
1da177e4 LT |
7995 | |
7996 | void __init sched_init_smp(void) | |
7997 | { | |
5c1e1767 NP |
7998 | cpumask_t non_isolated_cpus; |
7999 | ||
434d53b0 MT |
8000 | #if defined(CONFIG_NUMA) |
8001 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8002 | GFP_KERNEL); | |
8003 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8004 | #endif | |
95402b38 | 8005 | get_online_cpus(); |
712555ee | 8006 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 8007 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 8008 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
8009 | if (cpus_empty(non_isolated_cpus)) |
8010 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8011 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8012 | put_online_cpus(); |
e761b772 MK |
8013 | |
8014 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8015 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8016 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8017 | #endif |
8018 | ||
8019 | /* RT runtime code needs to handle some hotplug events */ | |
8020 | hotcpu_notifier(update_runtime, 0); | |
8021 | ||
b328ca18 | 8022 | init_hrtick(); |
5c1e1767 NP |
8023 | |
8024 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 8025 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 8026 | BUG(); |
19978ca6 | 8027 | sched_init_granularity(); |
1da177e4 LT |
8028 | } |
8029 | #else | |
8030 | void __init sched_init_smp(void) | |
8031 | { | |
19978ca6 | 8032 | sched_init_granularity(); |
1da177e4 LT |
8033 | } |
8034 | #endif /* CONFIG_SMP */ | |
8035 | ||
8036 | int in_sched_functions(unsigned long addr) | |
8037 | { | |
1da177e4 LT |
8038 | return in_lock_functions(addr) || |
8039 | (addr >= (unsigned long)__sched_text_start | |
8040 | && addr < (unsigned long)__sched_text_end); | |
8041 | } | |
8042 | ||
a9957449 | 8043 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8044 | { |
8045 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8046 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8047 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8048 | cfs_rq->rq = rq; | |
8049 | #endif | |
67e9fb2a | 8050 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8051 | } |
8052 | ||
fa85ae24 PZ |
8053 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8054 | { | |
8055 | struct rt_prio_array *array; | |
8056 | int i; | |
8057 | ||
8058 | array = &rt_rq->active; | |
8059 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8060 | INIT_LIST_HEAD(array->queue + i); | |
8061 | __clear_bit(i, array->bitmap); | |
8062 | } | |
8063 | /* delimiter for bitsearch: */ | |
8064 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8065 | ||
052f1dc7 | 8066 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8067 | rt_rq->highest_prio = MAX_RT_PRIO; |
8068 | #endif | |
fa85ae24 PZ |
8069 | #ifdef CONFIG_SMP |
8070 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8071 | rt_rq->overloaded = 0; |
8072 | #endif | |
8073 | ||
8074 | rt_rq->rt_time = 0; | |
8075 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8076 | rt_rq->rt_runtime = 0; |
8077 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8078 | |
052f1dc7 | 8079 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8080 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8081 | rt_rq->rq = rq; |
8082 | #endif | |
fa85ae24 PZ |
8083 | } |
8084 | ||
6f505b16 | 8085 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8086 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8087 | struct sched_entity *se, int cpu, int add, | |
8088 | struct sched_entity *parent) | |
6f505b16 | 8089 | { |
ec7dc8ac | 8090 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8091 | tg->cfs_rq[cpu] = cfs_rq; |
8092 | init_cfs_rq(cfs_rq, rq); | |
8093 | cfs_rq->tg = tg; | |
8094 | if (add) | |
8095 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8096 | ||
8097 | tg->se[cpu] = se; | |
354d60c2 DG |
8098 | /* se could be NULL for init_task_group */ |
8099 | if (!se) | |
8100 | return; | |
8101 | ||
ec7dc8ac DG |
8102 | if (!parent) |
8103 | se->cfs_rq = &rq->cfs; | |
8104 | else | |
8105 | se->cfs_rq = parent->my_q; | |
8106 | ||
6f505b16 PZ |
8107 | se->my_q = cfs_rq; |
8108 | se->load.weight = tg->shares; | |
e05510d0 | 8109 | se->load.inv_weight = 0; |
ec7dc8ac | 8110 | se->parent = parent; |
6f505b16 | 8111 | } |
052f1dc7 | 8112 | #endif |
6f505b16 | 8113 | |
052f1dc7 | 8114 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8115 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8116 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8117 | struct sched_rt_entity *parent) | |
6f505b16 | 8118 | { |
ec7dc8ac DG |
8119 | struct rq *rq = cpu_rq(cpu); |
8120 | ||
6f505b16 PZ |
8121 | tg->rt_rq[cpu] = rt_rq; |
8122 | init_rt_rq(rt_rq, rq); | |
8123 | rt_rq->tg = tg; | |
8124 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8125 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8126 | if (add) |
8127 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8128 | ||
8129 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8130 | if (!rt_se) |
8131 | return; | |
8132 | ||
ec7dc8ac DG |
8133 | if (!parent) |
8134 | rt_se->rt_rq = &rq->rt; | |
8135 | else | |
8136 | rt_se->rt_rq = parent->my_q; | |
8137 | ||
6f505b16 | 8138 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8139 | rt_se->parent = parent; |
6f505b16 PZ |
8140 | INIT_LIST_HEAD(&rt_se->run_list); |
8141 | } | |
8142 | #endif | |
8143 | ||
1da177e4 LT |
8144 | void __init sched_init(void) |
8145 | { | |
dd41f596 | 8146 | int i, j; |
434d53b0 MT |
8147 | unsigned long alloc_size = 0, ptr; |
8148 | ||
8149 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8150 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8151 | #endif | |
8152 | #ifdef CONFIG_RT_GROUP_SCHED | |
8153 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8154 | #endif |
8155 | #ifdef CONFIG_USER_SCHED | |
8156 | alloc_size *= 2; | |
434d53b0 MT |
8157 | #endif |
8158 | /* | |
8159 | * As sched_init() is called before page_alloc is setup, | |
8160 | * we use alloc_bootmem(). | |
8161 | */ | |
8162 | if (alloc_size) { | |
5a9d3225 | 8163 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8164 | |
8165 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8166 | init_task_group.se = (struct sched_entity **)ptr; | |
8167 | ptr += nr_cpu_ids * sizeof(void **); | |
8168 | ||
8169 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8170 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8171 | |
8172 | #ifdef CONFIG_USER_SCHED | |
8173 | root_task_group.se = (struct sched_entity **)ptr; | |
8174 | ptr += nr_cpu_ids * sizeof(void **); | |
8175 | ||
8176 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8177 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8178 | #endif /* CONFIG_USER_SCHED */ |
8179 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8180 | #ifdef CONFIG_RT_GROUP_SCHED |
8181 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8182 | ptr += nr_cpu_ids * sizeof(void **); | |
8183 | ||
8184 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8185 | ptr += nr_cpu_ids * sizeof(void **); |
8186 | ||
8187 | #ifdef CONFIG_USER_SCHED | |
8188 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8189 | ptr += nr_cpu_ids * sizeof(void **); | |
8190 | ||
8191 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8192 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8193 | #endif /* CONFIG_USER_SCHED */ |
8194 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8195 | } |
dd41f596 | 8196 | |
57d885fe GH |
8197 | #ifdef CONFIG_SMP |
8198 | init_defrootdomain(); | |
8199 | #endif | |
8200 | ||
d0b27fa7 PZ |
8201 | init_rt_bandwidth(&def_rt_bandwidth, |
8202 | global_rt_period(), global_rt_runtime()); | |
8203 | ||
8204 | #ifdef CONFIG_RT_GROUP_SCHED | |
8205 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8206 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8207 | #ifdef CONFIG_USER_SCHED |
8208 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8209 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8210 | #endif /* CONFIG_USER_SCHED */ |
8211 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8212 | |
052f1dc7 | 8213 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8214 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8215 | INIT_LIST_HEAD(&init_task_group.children); |
8216 | ||
8217 | #ifdef CONFIG_USER_SCHED | |
8218 | INIT_LIST_HEAD(&root_task_group.children); | |
8219 | init_task_group.parent = &root_task_group; | |
8220 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8221 | #endif /* CONFIG_USER_SCHED */ |
8222 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8223 | |
0a945022 | 8224 | for_each_possible_cpu(i) { |
70b97a7f | 8225 | struct rq *rq; |
1da177e4 LT |
8226 | |
8227 | rq = cpu_rq(i); | |
8228 | spin_lock_init(&rq->lock); | |
7897986b | 8229 | rq->nr_running = 0; |
dd41f596 | 8230 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8231 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8232 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8233 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8234 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8235 | #ifdef CONFIG_CGROUP_SCHED |
8236 | /* | |
8237 | * How much cpu bandwidth does init_task_group get? | |
8238 | * | |
8239 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8240 | * gets 100% of the cpu resources in the system. This overall | |
8241 | * system cpu resource is divided among the tasks of | |
8242 | * init_task_group and its child task-groups in a fair manner, | |
8243 | * based on each entity's (task or task-group's) weight | |
8244 | * (se->load.weight). | |
8245 | * | |
8246 | * In other words, if init_task_group has 10 tasks of weight | |
8247 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8248 | * then A0's share of the cpu resource is: | |
8249 | * | |
8250 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8251 | * | |
8252 | * We achieve this by letting init_task_group's tasks sit | |
8253 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8254 | */ | |
ec7dc8ac | 8255 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8256 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8257 | root_task_group.shares = NICE_0_LOAD; |
8258 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8259 | /* |
8260 | * In case of task-groups formed thr' the user id of tasks, | |
8261 | * init_task_group represents tasks belonging to root user. | |
8262 | * Hence it forms a sibling of all subsequent groups formed. | |
8263 | * In this case, init_task_group gets only a fraction of overall | |
8264 | * system cpu resource, based on the weight assigned to root | |
8265 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8266 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8267 | * (init_cfs_rq) and having one entity represent this group of | |
8268 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8269 | */ | |
ec7dc8ac | 8270 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8271 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8272 | &per_cpu(init_sched_entity, i), i, 1, |
8273 | root_task_group.se[i]); | |
6f505b16 | 8274 | |
052f1dc7 | 8275 | #endif |
354d60c2 DG |
8276 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8277 | ||
8278 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8279 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8280 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8281 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8282 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8283 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8284 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8285 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8286 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8287 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8288 | root_task_group.rt_se[i]); | |
354d60c2 | 8289 | #endif |
dd41f596 | 8290 | #endif |
1da177e4 | 8291 | |
dd41f596 IM |
8292 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8293 | rq->cpu_load[j] = 0; | |
1da177e4 | 8294 | #ifdef CONFIG_SMP |
41c7ce9a | 8295 | rq->sd = NULL; |
57d885fe | 8296 | rq->rd = NULL; |
1da177e4 | 8297 | rq->active_balance = 0; |
dd41f596 | 8298 | rq->next_balance = jiffies; |
1da177e4 | 8299 | rq->push_cpu = 0; |
0a2966b4 | 8300 | rq->cpu = i; |
1f11eb6a | 8301 | rq->online = 0; |
1da177e4 LT |
8302 | rq->migration_thread = NULL; |
8303 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8304 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8305 | #endif |
8f4d37ec | 8306 | init_rq_hrtick(rq); |
1da177e4 | 8307 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8308 | } |
8309 | ||
2dd73a4f | 8310 | set_load_weight(&init_task); |
b50f60ce | 8311 | |
e107be36 AK |
8312 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8313 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8314 | #endif | |
8315 | ||
c9819f45 | 8316 | #ifdef CONFIG_SMP |
962cf36c | 8317 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8318 | #endif |
8319 | ||
b50f60ce HC |
8320 | #ifdef CONFIG_RT_MUTEXES |
8321 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8322 | #endif | |
8323 | ||
1da177e4 LT |
8324 | /* |
8325 | * The boot idle thread does lazy MMU switching as well: | |
8326 | */ | |
8327 | atomic_inc(&init_mm.mm_count); | |
8328 | enter_lazy_tlb(&init_mm, current); | |
8329 | ||
8330 | /* | |
8331 | * Make us the idle thread. Technically, schedule() should not be | |
8332 | * called from this thread, however somewhere below it might be, | |
8333 | * but because we are the idle thread, we just pick up running again | |
8334 | * when this runqueue becomes "idle". | |
8335 | */ | |
8336 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8337 | /* |
8338 | * During early bootup we pretend to be a normal task: | |
8339 | */ | |
8340 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8341 | |
8342 | scheduler_running = 1; | |
1da177e4 LT |
8343 | } |
8344 | ||
8345 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8346 | void __might_sleep(char *file, int line) | |
8347 | { | |
48f24c4d | 8348 | #ifdef in_atomic |
1da177e4 LT |
8349 | static unsigned long prev_jiffy; /* ratelimiting */ |
8350 | ||
aef745fc IM |
8351 | if ((!in_atomic() && !irqs_disabled()) || |
8352 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8353 | return; | |
8354 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8355 | return; | |
8356 | prev_jiffy = jiffies; | |
8357 | ||
8358 | printk(KERN_ERR | |
8359 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8360 | file, line); | |
8361 | printk(KERN_ERR | |
8362 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8363 | in_atomic(), irqs_disabled(), | |
8364 | current->pid, current->comm); | |
8365 | ||
8366 | debug_show_held_locks(current); | |
8367 | if (irqs_disabled()) | |
8368 | print_irqtrace_events(current); | |
8369 | dump_stack(); | |
1da177e4 LT |
8370 | #endif |
8371 | } | |
8372 | EXPORT_SYMBOL(__might_sleep); | |
8373 | #endif | |
8374 | ||
8375 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8376 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8377 | { | |
8378 | int on_rq; | |
3e51f33f | 8379 | |
3a5e4dc1 AK |
8380 | update_rq_clock(rq); |
8381 | on_rq = p->se.on_rq; | |
8382 | if (on_rq) | |
8383 | deactivate_task(rq, p, 0); | |
8384 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8385 | if (on_rq) { | |
8386 | activate_task(rq, p, 0); | |
8387 | resched_task(rq->curr); | |
8388 | } | |
8389 | } | |
8390 | ||
1da177e4 LT |
8391 | void normalize_rt_tasks(void) |
8392 | { | |
a0f98a1c | 8393 | struct task_struct *g, *p; |
1da177e4 | 8394 | unsigned long flags; |
70b97a7f | 8395 | struct rq *rq; |
1da177e4 | 8396 | |
4cf5d77a | 8397 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8398 | do_each_thread(g, p) { |
178be793 IM |
8399 | /* |
8400 | * Only normalize user tasks: | |
8401 | */ | |
8402 | if (!p->mm) | |
8403 | continue; | |
8404 | ||
6cfb0d5d | 8405 | p->se.exec_start = 0; |
6cfb0d5d | 8406 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8407 | p->se.wait_start = 0; |
dd41f596 | 8408 | p->se.sleep_start = 0; |
dd41f596 | 8409 | p->se.block_start = 0; |
6cfb0d5d | 8410 | #endif |
dd41f596 IM |
8411 | |
8412 | if (!rt_task(p)) { | |
8413 | /* | |
8414 | * Renice negative nice level userspace | |
8415 | * tasks back to 0: | |
8416 | */ | |
8417 | if (TASK_NICE(p) < 0 && p->mm) | |
8418 | set_user_nice(p, 0); | |
1da177e4 | 8419 | continue; |
dd41f596 | 8420 | } |
1da177e4 | 8421 | |
4cf5d77a | 8422 | spin_lock(&p->pi_lock); |
b29739f9 | 8423 | rq = __task_rq_lock(p); |
1da177e4 | 8424 | |
178be793 | 8425 | normalize_task(rq, p); |
3a5e4dc1 | 8426 | |
b29739f9 | 8427 | __task_rq_unlock(rq); |
4cf5d77a | 8428 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8429 | } while_each_thread(g, p); |
8430 | ||
4cf5d77a | 8431 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8432 | } |
8433 | ||
8434 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8435 | |
8436 | #ifdef CONFIG_IA64 | |
8437 | /* | |
8438 | * These functions are only useful for the IA64 MCA handling. | |
8439 | * | |
8440 | * They can only be called when the whole system has been | |
8441 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8442 | * activity can take place. Using them for anything else would | |
8443 | * be a serious bug, and as a result, they aren't even visible | |
8444 | * under any other configuration. | |
8445 | */ | |
8446 | ||
8447 | /** | |
8448 | * curr_task - return the current task for a given cpu. | |
8449 | * @cpu: the processor in question. | |
8450 | * | |
8451 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8452 | */ | |
36c8b586 | 8453 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8454 | { |
8455 | return cpu_curr(cpu); | |
8456 | } | |
8457 | ||
8458 | /** | |
8459 | * set_curr_task - set the current task for a given cpu. | |
8460 | * @cpu: the processor in question. | |
8461 | * @p: the task pointer to set. | |
8462 | * | |
8463 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8464 | * are serviced on a separate stack. It allows the architecture to switch the |
8465 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8466 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8467 | * and caller must save the original value of the current task (see | |
8468 | * curr_task() above) and restore that value before reenabling interrupts and | |
8469 | * re-starting the system. | |
8470 | * | |
8471 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8472 | */ | |
36c8b586 | 8473 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8474 | { |
8475 | cpu_curr(cpu) = p; | |
8476 | } | |
8477 | ||
8478 | #endif | |
29f59db3 | 8479 | |
bccbe08a PZ |
8480 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8481 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8482 | { |
8483 | int i; | |
8484 | ||
8485 | for_each_possible_cpu(i) { | |
8486 | if (tg->cfs_rq) | |
8487 | kfree(tg->cfs_rq[i]); | |
8488 | if (tg->se) | |
8489 | kfree(tg->se[i]); | |
6f505b16 PZ |
8490 | } |
8491 | ||
8492 | kfree(tg->cfs_rq); | |
8493 | kfree(tg->se); | |
6f505b16 PZ |
8494 | } |
8495 | ||
ec7dc8ac DG |
8496 | static |
8497 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8498 | { |
29f59db3 | 8499 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8500 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8501 | struct rq *rq; |
29f59db3 SV |
8502 | int i; |
8503 | ||
434d53b0 | 8504 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8505 | if (!tg->cfs_rq) |
8506 | goto err; | |
434d53b0 | 8507 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8508 | if (!tg->se) |
8509 | goto err; | |
052f1dc7 PZ |
8510 | |
8511 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8512 | |
8513 | for_each_possible_cpu(i) { | |
9b5b7751 | 8514 | rq = cpu_rq(i); |
29f59db3 | 8515 | |
6f505b16 PZ |
8516 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8517 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8518 | if (!cfs_rq) |
8519 | goto err; | |
8520 | ||
6f505b16 PZ |
8521 | se = kmalloc_node(sizeof(struct sched_entity), |
8522 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8523 | if (!se) |
8524 | goto err; | |
8525 | ||
ec7dc8ac DG |
8526 | parent_se = parent ? parent->se[i] : NULL; |
8527 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8528 | } |
8529 | ||
8530 | return 1; | |
8531 | ||
8532 | err: | |
8533 | return 0; | |
8534 | } | |
8535 | ||
8536 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8537 | { | |
8538 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8539 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8540 | } | |
8541 | ||
8542 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8543 | { | |
8544 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8545 | } | |
6d6bc0ad | 8546 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8547 | static inline void free_fair_sched_group(struct task_group *tg) |
8548 | { | |
8549 | } | |
8550 | ||
ec7dc8ac DG |
8551 | static inline |
8552 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8553 | { |
8554 | return 1; | |
8555 | } | |
8556 | ||
8557 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8558 | { | |
8559 | } | |
8560 | ||
8561 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8562 | { | |
8563 | } | |
6d6bc0ad | 8564 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8565 | |
8566 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8567 | static void free_rt_sched_group(struct task_group *tg) |
8568 | { | |
8569 | int i; | |
8570 | ||
d0b27fa7 PZ |
8571 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8572 | ||
bccbe08a PZ |
8573 | for_each_possible_cpu(i) { |
8574 | if (tg->rt_rq) | |
8575 | kfree(tg->rt_rq[i]); | |
8576 | if (tg->rt_se) | |
8577 | kfree(tg->rt_se[i]); | |
8578 | } | |
8579 | ||
8580 | kfree(tg->rt_rq); | |
8581 | kfree(tg->rt_se); | |
8582 | } | |
8583 | ||
ec7dc8ac DG |
8584 | static |
8585 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8586 | { |
8587 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8588 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8589 | struct rq *rq; |
8590 | int i; | |
8591 | ||
434d53b0 | 8592 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8593 | if (!tg->rt_rq) |
8594 | goto err; | |
434d53b0 | 8595 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8596 | if (!tg->rt_se) |
8597 | goto err; | |
8598 | ||
d0b27fa7 PZ |
8599 | init_rt_bandwidth(&tg->rt_bandwidth, |
8600 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8601 | |
8602 | for_each_possible_cpu(i) { | |
8603 | rq = cpu_rq(i); | |
8604 | ||
6f505b16 PZ |
8605 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8606 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8607 | if (!rt_rq) | |
8608 | goto err; | |
29f59db3 | 8609 | |
6f505b16 PZ |
8610 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8611 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8612 | if (!rt_se) | |
8613 | goto err; | |
29f59db3 | 8614 | |
ec7dc8ac DG |
8615 | parent_se = parent ? parent->rt_se[i] : NULL; |
8616 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8617 | } |
8618 | ||
bccbe08a PZ |
8619 | return 1; |
8620 | ||
8621 | err: | |
8622 | return 0; | |
8623 | } | |
8624 | ||
8625 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8626 | { | |
8627 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8628 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8629 | } | |
8630 | ||
8631 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8632 | { | |
8633 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8634 | } | |
6d6bc0ad | 8635 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8636 | static inline void free_rt_sched_group(struct task_group *tg) |
8637 | { | |
8638 | } | |
8639 | ||
ec7dc8ac DG |
8640 | static inline |
8641 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8642 | { |
8643 | return 1; | |
8644 | } | |
8645 | ||
8646 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8647 | { | |
8648 | } | |
8649 | ||
8650 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8651 | { | |
8652 | } | |
6d6bc0ad | 8653 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8654 | |
d0b27fa7 | 8655 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8656 | static void free_sched_group(struct task_group *tg) |
8657 | { | |
8658 | free_fair_sched_group(tg); | |
8659 | free_rt_sched_group(tg); | |
8660 | kfree(tg); | |
8661 | } | |
8662 | ||
8663 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8664 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8665 | { |
8666 | struct task_group *tg; | |
8667 | unsigned long flags; | |
8668 | int i; | |
8669 | ||
8670 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8671 | if (!tg) | |
8672 | return ERR_PTR(-ENOMEM); | |
8673 | ||
ec7dc8ac | 8674 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8675 | goto err; |
8676 | ||
ec7dc8ac | 8677 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8678 | goto err; |
8679 | ||
8ed36996 | 8680 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8681 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8682 | register_fair_sched_group(tg, i); |
8683 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8684 | } |
6f505b16 | 8685 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8686 | |
8687 | WARN_ON(!parent); /* root should already exist */ | |
8688 | ||
8689 | tg->parent = parent; | |
f473aa5e | 8690 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8691 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8692 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8693 | |
9b5b7751 | 8694 | return tg; |
29f59db3 SV |
8695 | |
8696 | err: | |
6f505b16 | 8697 | free_sched_group(tg); |
29f59db3 SV |
8698 | return ERR_PTR(-ENOMEM); |
8699 | } | |
8700 | ||
9b5b7751 | 8701 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8702 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8703 | { |
29f59db3 | 8704 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8705 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8706 | } |
8707 | ||
9b5b7751 | 8708 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8709 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8710 | { |
8ed36996 | 8711 | unsigned long flags; |
9b5b7751 | 8712 | int i; |
29f59db3 | 8713 | |
8ed36996 | 8714 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8715 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8716 | unregister_fair_sched_group(tg, i); |
8717 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8718 | } |
6f505b16 | 8719 | list_del_rcu(&tg->list); |
f473aa5e | 8720 | list_del_rcu(&tg->siblings); |
8ed36996 | 8721 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8722 | |
9b5b7751 | 8723 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8724 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8725 | } |
8726 | ||
9b5b7751 | 8727 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8728 | * The caller of this function should have put the task in its new group |
8729 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8730 | * reflect its new group. | |
9b5b7751 SV |
8731 | */ |
8732 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8733 | { |
8734 | int on_rq, running; | |
8735 | unsigned long flags; | |
8736 | struct rq *rq; | |
8737 | ||
8738 | rq = task_rq_lock(tsk, &flags); | |
8739 | ||
29f59db3 SV |
8740 | update_rq_clock(rq); |
8741 | ||
051a1d1a | 8742 | running = task_current(rq, tsk); |
29f59db3 SV |
8743 | on_rq = tsk->se.on_rq; |
8744 | ||
0e1f3483 | 8745 | if (on_rq) |
29f59db3 | 8746 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8747 | if (unlikely(running)) |
8748 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8749 | |
6f505b16 | 8750 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8751 | |
810b3817 PZ |
8752 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8753 | if (tsk->sched_class->moved_group) | |
8754 | tsk->sched_class->moved_group(tsk); | |
8755 | #endif | |
8756 | ||
0e1f3483 HS |
8757 | if (unlikely(running)) |
8758 | tsk->sched_class->set_curr_task(rq); | |
8759 | if (on_rq) | |
7074badb | 8760 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8761 | |
29f59db3 SV |
8762 | task_rq_unlock(rq, &flags); |
8763 | } | |
6d6bc0ad | 8764 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8765 | |
052f1dc7 | 8766 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8767 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8768 | { |
8769 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8770 | int on_rq; |
8771 | ||
29f59db3 | 8772 | on_rq = se->on_rq; |
62fb1851 | 8773 | if (on_rq) |
29f59db3 SV |
8774 | dequeue_entity(cfs_rq, se, 0); |
8775 | ||
8776 | se->load.weight = shares; | |
e05510d0 | 8777 | se->load.inv_weight = 0; |
29f59db3 | 8778 | |
62fb1851 | 8779 | if (on_rq) |
29f59db3 | 8780 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8781 | } |
62fb1851 | 8782 | |
c09595f6 PZ |
8783 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8784 | { | |
8785 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8786 | struct rq *rq = cfs_rq->rq; | |
8787 | unsigned long flags; | |
8788 | ||
8789 | spin_lock_irqsave(&rq->lock, flags); | |
8790 | __set_se_shares(se, shares); | |
8791 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8792 | } |
8793 | ||
8ed36996 PZ |
8794 | static DEFINE_MUTEX(shares_mutex); |
8795 | ||
4cf86d77 | 8796 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8797 | { |
8798 | int i; | |
8ed36996 | 8799 | unsigned long flags; |
c61935fd | 8800 | |
ec7dc8ac DG |
8801 | /* |
8802 | * We can't change the weight of the root cgroup. | |
8803 | */ | |
8804 | if (!tg->se[0]) | |
8805 | return -EINVAL; | |
8806 | ||
18d95a28 PZ |
8807 | if (shares < MIN_SHARES) |
8808 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8809 | else if (shares > MAX_SHARES) |
8810 | shares = MAX_SHARES; | |
62fb1851 | 8811 | |
8ed36996 | 8812 | mutex_lock(&shares_mutex); |
9b5b7751 | 8813 | if (tg->shares == shares) |
5cb350ba | 8814 | goto done; |
29f59db3 | 8815 | |
8ed36996 | 8816 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8817 | for_each_possible_cpu(i) |
8818 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8819 | list_del_rcu(&tg->siblings); |
8ed36996 | 8820 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8821 | |
8822 | /* wait for any ongoing reference to this group to finish */ | |
8823 | synchronize_sched(); | |
8824 | ||
8825 | /* | |
8826 | * Now we are free to modify the group's share on each cpu | |
8827 | * w/o tripping rebalance_share or load_balance_fair. | |
8828 | */ | |
9b5b7751 | 8829 | tg->shares = shares; |
c09595f6 PZ |
8830 | for_each_possible_cpu(i) { |
8831 | /* | |
8832 | * force a rebalance | |
8833 | */ | |
8834 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8835 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8836 | } |
29f59db3 | 8837 | |
6b2d7700 SV |
8838 | /* |
8839 | * Enable load balance activity on this group, by inserting it back on | |
8840 | * each cpu's rq->leaf_cfs_rq_list. | |
8841 | */ | |
8ed36996 | 8842 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8843 | for_each_possible_cpu(i) |
8844 | register_fair_sched_group(tg, i); | |
f473aa5e | 8845 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8846 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8847 | done: |
8ed36996 | 8848 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8849 | return 0; |
29f59db3 SV |
8850 | } |
8851 | ||
5cb350ba DG |
8852 | unsigned long sched_group_shares(struct task_group *tg) |
8853 | { | |
8854 | return tg->shares; | |
8855 | } | |
052f1dc7 | 8856 | #endif |
5cb350ba | 8857 | |
052f1dc7 | 8858 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8859 | /* |
9f0c1e56 | 8860 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8861 | */ |
9f0c1e56 PZ |
8862 | static DEFINE_MUTEX(rt_constraints_mutex); |
8863 | ||
8864 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8865 | { | |
8866 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8867 | return 1ULL << 20; |
9f0c1e56 | 8868 | |
9a7e0b18 | 8869 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8870 | } |
8871 | ||
9a7e0b18 PZ |
8872 | /* Must be called with tasklist_lock held */ |
8873 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8874 | { |
9a7e0b18 | 8875 | struct task_struct *g, *p; |
b40b2e8e | 8876 | |
9a7e0b18 PZ |
8877 | do_each_thread(g, p) { |
8878 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8879 | return 1; | |
8880 | } while_each_thread(g, p); | |
b40b2e8e | 8881 | |
9a7e0b18 PZ |
8882 | return 0; |
8883 | } | |
b40b2e8e | 8884 | |
9a7e0b18 PZ |
8885 | struct rt_schedulable_data { |
8886 | struct task_group *tg; | |
8887 | u64 rt_period; | |
8888 | u64 rt_runtime; | |
8889 | }; | |
b40b2e8e | 8890 | |
9a7e0b18 PZ |
8891 | static int tg_schedulable(struct task_group *tg, void *data) |
8892 | { | |
8893 | struct rt_schedulable_data *d = data; | |
8894 | struct task_group *child; | |
8895 | unsigned long total, sum = 0; | |
8896 | u64 period, runtime; | |
b40b2e8e | 8897 | |
9a7e0b18 PZ |
8898 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8899 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8900 | |
9a7e0b18 PZ |
8901 | if (tg == d->tg) { |
8902 | period = d->rt_period; | |
8903 | runtime = d->rt_runtime; | |
b40b2e8e | 8904 | } |
b40b2e8e | 8905 | |
4653f803 PZ |
8906 | /* |
8907 | * Cannot have more runtime than the period. | |
8908 | */ | |
8909 | if (runtime > period && runtime != RUNTIME_INF) | |
8910 | return -EINVAL; | |
6f505b16 | 8911 | |
4653f803 PZ |
8912 | /* |
8913 | * Ensure we don't starve existing RT tasks. | |
8914 | */ | |
9a7e0b18 PZ |
8915 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8916 | return -EBUSY; | |
6f505b16 | 8917 | |
9a7e0b18 | 8918 | total = to_ratio(period, runtime); |
6f505b16 | 8919 | |
4653f803 PZ |
8920 | /* |
8921 | * Nobody can have more than the global setting allows. | |
8922 | */ | |
8923 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8924 | return -EINVAL; | |
6f505b16 | 8925 | |
4653f803 PZ |
8926 | /* |
8927 | * The sum of our children's runtime should not exceed our own. | |
8928 | */ | |
9a7e0b18 PZ |
8929 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8930 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8931 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8932 | |
9a7e0b18 PZ |
8933 | if (child == d->tg) { |
8934 | period = d->rt_period; | |
8935 | runtime = d->rt_runtime; | |
8936 | } | |
6f505b16 | 8937 | |
9a7e0b18 | 8938 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8939 | } |
6f505b16 | 8940 | |
9a7e0b18 PZ |
8941 | if (sum > total) |
8942 | return -EINVAL; | |
8943 | ||
8944 | return 0; | |
6f505b16 PZ |
8945 | } |
8946 | ||
9a7e0b18 | 8947 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8948 | { |
9a7e0b18 PZ |
8949 | struct rt_schedulable_data data = { |
8950 | .tg = tg, | |
8951 | .rt_period = period, | |
8952 | .rt_runtime = runtime, | |
8953 | }; | |
8954 | ||
8955 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8956 | } |
8957 | ||
d0b27fa7 PZ |
8958 | static int tg_set_bandwidth(struct task_group *tg, |
8959 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8960 | { |
ac086bc2 | 8961 | int i, err = 0; |
9f0c1e56 | 8962 | |
9f0c1e56 | 8963 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8964 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8965 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8966 | if (err) | |
9f0c1e56 | 8967 | goto unlock; |
ac086bc2 PZ |
8968 | |
8969 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8970 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8971 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8972 | |
8973 | for_each_possible_cpu(i) { | |
8974 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8975 | ||
8976 | spin_lock(&rt_rq->rt_runtime_lock); | |
8977 | rt_rq->rt_runtime = rt_runtime; | |
8978 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8979 | } | |
8980 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8981 | unlock: |
521f1a24 | 8982 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8983 | mutex_unlock(&rt_constraints_mutex); |
8984 | ||
8985 | return err; | |
6f505b16 PZ |
8986 | } |
8987 | ||
d0b27fa7 PZ |
8988 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8989 | { | |
8990 | u64 rt_runtime, rt_period; | |
8991 | ||
8992 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8993 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8994 | if (rt_runtime_us < 0) | |
8995 | rt_runtime = RUNTIME_INF; | |
8996 | ||
8997 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8998 | } | |
8999 | ||
9f0c1e56 PZ |
9000 | long sched_group_rt_runtime(struct task_group *tg) |
9001 | { | |
9002 | u64 rt_runtime_us; | |
9003 | ||
d0b27fa7 | 9004 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9005 | return -1; |
9006 | ||
d0b27fa7 | 9007 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9008 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9009 | return rt_runtime_us; | |
9010 | } | |
d0b27fa7 PZ |
9011 | |
9012 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9013 | { | |
9014 | u64 rt_runtime, rt_period; | |
9015 | ||
9016 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9017 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9018 | ||
619b0488 R |
9019 | if (rt_period == 0) |
9020 | return -EINVAL; | |
9021 | ||
d0b27fa7 PZ |
9022 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9023 | } | |
9024 | ||
9025 | long sched_group_rt_period(struct task_group *tg) | |
9026 | { | |
9027 | u64 rt_period_us; | |
9028 | ||
9029 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9030 | do_div(rt_period_us, NSEC_PER_USEC); | |
9031 | return rt_period_us; | |
9032 | } | |
9033 | ||
9034 | static int sched_rt_global_constraints(void) | |
9035 | { | |
4653f803 | 9036 | u64 runtime, period; |
d0b27fa7 PZ |
9037 | int ret = 0; |
9038 | ||
ec5d4989 HS |
9039 | if (sysctl_sched_rt_period <= 0) |
9040 | return -EINVAL; | |
9041 | ||
4653f803 PZ |
9042 | runtime = global_rt_runtime(); |
9043 | period = global_rt_period(); | |
9044 | ||
9045 | /* | |
9046 | * Sanity check on the sysctl variables. | |
9047 | */ | |
9048 | if (runtime > period && runtime != RUNTIME_INF) | |
9049 | return -EINVAL; | |
10b612f4 | 9050 | |
d0b27fa7 | 9051 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9052 | read_lock(&tasklist_lock); |
4653f803 | 9053 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9054 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9055 | mutex_unlock(&rt_constraints_mutex); |
9056 | ||
9057 | return ret; | |
9058 | } | |
6d6bc0ad | 9059 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9060 | static int sched_rt_global_constraints(void) |
9061 | { | |
ac086bc2 PZ |
9062 | unsigned long flags; |
9063 | int i; | |
9064 | ||
ec5d4989 HS |
9065 | if (sysctl_sched_rt_period <= 0) |
9066 | return -EINVAL; | |
9067 | ||
ac086bc2 PZ |
9068 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9069 | for_each_possible_cpu(i) { | |
9070 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9071 | ||
9072 | spin_lock(&rt_rq->rt_runtime_lock); | |
9073 | rt_rq->rt_runtime = global_rt_runtime(); | |
9074 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9075 | } | |
9076 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9077 | ||
d0b27fa7 PZ |
9078 | return 0; |
9079 | } | |
6d6bc0ad | 9080 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9081 | |
9082 | int sched_rt_handler(struct ctl_table *table, int write, | |
9083 | struct file *filp, void __user *buffer, size_t *lenp, | |
9084 | loff_t *ppos) | |
9085 | { | |
9086 | int ret; | |
9087 | int old_period, old_runtime; | |
9088 | static DEFINE_MUTEX(mutex); | |
9089 | ||
9090 | mutex_lock(&mutex); | |
9091 | old_period = sysctl_sched_rt_period; | |
9092 | old_runtime = sysctl_sched_rt_runtime; | |
9093 | ||
9094 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9095 | ||
9096 | if (!ret && write) { | |
9097 | ret = sched_rt_global_constraints(); | |
9098 | if (ret) { | |
9099 | sysctl_sched_rt_period = old_period; | |
9100 | sysctl_sched_rt_runtime = old_runtime; | |
9101 | } else { | |
9102 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9103 | def_rt_bandwidth.rt_period = | |
9104 | ns_to_ktime(global_rt_period()); | |
9105 | } | |
9106 | } | |
9107 | mutex_unlock(&mutex); | |
9108 | ||
9109 | return ret; | |
9110 | } | |
68318b8e | 9111 | |
052f1dc7 | 9112 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9113 | |
9114 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9115 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9116 | { |
2b01dfe3 PM |
9117 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9118 | struct task_group, css); | |
68318b8e SV |
9119 | } |
9120 | ||
9121 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9122 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9123 | { |
ec7dc8ac | 9124 | struct task_group *tg, *parent; |
68318b8e | 9125 | |
2b01dfe3 | 9126 | if (!cgrp->parent) { |
68318b8e | 9127 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9128 | return &init_task_group.css; |
9129 | } | |
9130 | ||
ec7dc8ac DG |
9131 | parent = cgroup_tg(cgrp->parent); |
9132 | tg = sched_create_group(parent); | |
68318b8e SV |
9133 | if (IS_ERR(tg)) |
9134 | return ERR_PTR(-ENOMEM); | |
9135 | ||
68318b8e SV |
9136 | return &tg->css; |
9137 | } | |
9138 | ||
41a2d6cf IM |
9139 | static void |
9140 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9141 | { |
2b01dfe3 | 9142 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9143 | |
9144 | sched_destroy_group(tg); | |
9145 | } | |
9146 | ||
41a2d6cf IM |
9147 | static int |
9148 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9149 | struct task_struct *tsk) | |
68318b8e | 9150 | { |
b68aa230 PZ |
9151 | #ifdef CONFIG_RT_GROUP_SCHED |
9152 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9153 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9154 | return -EINVAL; |
9155 | #else | |
68318b8e SV |
9156 | /* We don't support RT-tasks being in separate groups */ |
9157 | if (tsk->sched_class != &fair_sched_class) | |
9158 | return -EINVAL; | |
b68aa230 | 9159 | #endif |
68318b8e SV |
9160 | |
9161 | return 0; | |
9162 | } | |
9163 | ||
9164 | static void | |
2b01dfe3 | 9165 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9166 | struct cgroup *old_cont, struct task_struct *tsk) |
9167 | { | |
9168 | sched_move_task(tsk); | |
9169 | } | |
9170 | ||
052f1dc7 | 9171 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9172 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9173 | u64 shareval) |
68318b8e | 9174 | { |
2b01dfe3 | 9175 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9176 | } |
9177 | ||
f4c753b7 | 9178 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9179 | { |
2b01dfe3 | 9180 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9181 | |
9182 | return (u64) tg->shares; | |
9183 | } | |
6d6bc0ad | 9184 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9185 | |
052f1dc7 | 9186 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9187 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9188 | s64 val) |
6f505b16 | 9189 | { |
06ecb27c | 9190 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9191 | } |
9192 | ||
06ecb27c | 9193 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9194 | { |
06ecb27c | 9195 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9196 | } |
d0b27fa7 PZ |
9197 | |
9198 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9199 | u64 rt_period_us) | |
9200 | { | |
9201 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9202 | } | |
9203 | ||
9204 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9205 | { | |
9206 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9207 | } | |
6d6bc0ad | 9208 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9209 | |
fe5c7cc2 | 9210 | static struct cftype cpu_files[] = { |
052f1dc7 | 9211 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9212 | { |
9213 | .name = "shares", | |
f4c753b7 PM |
9214 | .read_u64 = cpu_shares_read_u64, |
9215 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9216 | }, |
052f1dc7 PZ |
9217 | #endif |
9218 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9219 | { |
9f0c1e56 | 9220 | .name = "rt_runtime_us", |
06ecb27c PM |
9221 | .read_s64 = cpu_rt_runtime_read, |
9222 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9223 | }, |
d0b27fa7 PZ |
9224 | { |
9225 | .name = "rt_period_us", | |
f4c753b7 PM |
9226 | .read_u64 = cpu_rt_period_read_uint, |
9227 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9228 | }, |
052f1dc7 | 9229 | #endif |
68318b8e SV |
9230 | }; |
9231 | ||
9232 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9233 | { | |
fe5c7cc2 | 9234 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9235 | } |
9236 | ||
9237 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9238 | .name = "cpu", |
9239 | .create = cpu_cgroup_create, | |
9240 | .destroy = cpu_cgroup_destroy, | |
9241 | .can_attach = cpu_cgroup_can_attach, | |
9242 | .attach = cpu_cgroup_attach, | |
9243 | .populate = cpu_cgroup_populate, | |
9244 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9245 | .early_init = 1, |
9246 | }; | |
9247 | ||
052f1dc7 | 9248 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9249 | |
9250 | #ifdef CONFIG_CGROUP_CPUACCT | |
9251 | ||
9252 | /* | |
9253 | * CPU accounting code for task groups. | |
9254 | * | |
9255 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9256 | * (balbir@in.ibm.com). | |
9257 | */ | |
9258 | ||
9259 | /* track cpu usage of a group of tasks */ | |
9260 | struct cpuacct { | |
9261 | struct cgroup_subsys_state css; | |
9262 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9263 | u64 *cpuusage; | |
9264 | }; | |
9265 | ||
9266 | struct cgroup_subsys cpuacct_subsys; | |
9267 | ||
9268 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9269 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9270 | { |
32cd756a | 9271 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9272 | struct cpuacct, css); |
9273 | } | |
9274 | ||
9275 | /* return cpu accounting group to which this task belongs */ | |
9276 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9277 | { | |
9278 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9279 | struct cpuacct, css); | |
9280 | } | |
9281 | ||
9282 | /* create a new cpu accounting group */ | |
9283 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9284 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9285 | { |
9286 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9287 | ||
9288 | if (!ca) | |
9289 | return ERR_PTR(-ENOMEM); | |
9290 | ||
9291 | ca->cpuusage = alloc_percpu(u64); | |
9292 | if (!ca->cpuusage) { | |
9293 | kfree(ca); | |
9294 | return ERR_PTR(-ENOMEM); | |
9295 | } | |
9296 | ||
9297 | return &ca->css; | |
9298 | } | |
9299 | ||
9300 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9301 | static void |
32cd756a | 9302 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9303 | { |
32cd756a | 9304 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9305 | |
9306 | free_percpu(ca->cpuusage); | |
9307 | kfree(ca); | |
9308 | } | |
9309 | ||
9310 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9311 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9312 | { |
32cd756a | 9313 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9314 | u64 totalcpuusage = 0; |
9315 | int i; | |
9316 | ||
9317 | for_each_possible_cpu(i) { | |
9318 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9319 | ||
9320 | /* | |
9321 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9322 | * platforms. | |
9323 | */ | |
9324 | spin_lock_irq(&cpu_rq(i)->lock); | |
9325 | totalcpuusage += *cpuusage; | |
9326 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9327 | } | |
9328 | ||
9329 | return totalcpuusage; | |
9330 | } | |
9331 | ||
0297b803 DG |
9332 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9333 | u64 reset) | |
9334 | { | |
9335 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9336 | int err = 0; | |
9337 | int i; | |
9338 | ||
9339 | if (reset) { | |
9340 | err = -EINVAL; | |
9341 | goto out; | |
9342 | } | |
9343 | ||
9344 | for_each_possible_cpu(i) { | |
9345 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9346 | ||
9347 | spin_lock_irq(&cpu_rq(i)->lock); | |
9348 | *cpuusage = 0; | |
9349 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9350 | } | |
9351 | out: | |
9352 | return err; | |
9353 | } | |
9354 | ||
d842de87 SV |
9355 | static struct cftype files[] = { |
9356 | { | |
9357 | .name = "usage", | |
f4c753b7 PM |
9358 | .read_u64 = cpuusage_read, |
9359 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9360 | }, |
9361 | }; | |
9362 | ||
32cd756a | 9363 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9364 | { |
32cd756a | 9365 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9366 | } |
9367 | ||
9368 | /* | |
9369 | * charge this task's execution time to its accounting group. | |
9370 | * | |
9371 | * called with rq->lock held. | |
9372 | */ | |
9373 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9374 | { | |
9375 | struct cpuacct *ca; | |
9376 | ||
9377 | if (!cpuacct_subsys.active) | |
9378 | return; | |
9379 | ||
9380 | ca = task_ca(tsk); | |
9381 | if (ca) { | |
9382 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9383 | ||
9384 | *cpuusage += cputime; | |
9385 | } | |
9386 | } | |
9387 | ||
9388 | struct cgroup_subsys cpuacct_subsys = { | |
9389 | .name = "cpuacct", | |
9390 | .create = cpuacct_create, | |
9391 | .destroy = cpuacct_destroy, | |
9392 | .populate = cpuacct_populate, | |
9393 | .subsys_id = cpuacct_subsys_id, | |
9394 | }; | |
9395 | #endif /* CONFIG_CGROUP_CPUACCT */ |