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> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
434d53b0 | 70 | #include <linux/bootmem.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
d7876a08 | 99 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 100 | */ |
d6322faf | 101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 102 | |
6aa645ea IM |
103 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
104 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
105 | ||
1da177e4 LT |
106 | /* |
107 | * These are the 'tuning knobs' of the scheduler: | |
108 | * | |
a4ec24b4 | 109 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
110 | * Timeslices get refilled after they expire. |
111 | */ | |
1da177e4 | 112 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 113 | |
d0b27fa7 PZ |
114 | /* |
115 | * single value that denotes runtime == period, ie unlimited time. | |
116 | */ | |
117 | #define RUNTIME_INF ((u64)~0ULL) | |
118 | ||
5517d86b ED |
119 | #ifdef CONFIG_SMP |
120 | /* | |
121 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
122 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
123 | */ | |
124 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
125 | { | |
126 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
127 | } | |
128 | ||
129 | /* | |
130 | * Each time a sched group cpu_power is changed, | |
131 | * we must compute its reciprocal value | |
132 | */ | |
133 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
134 | { | |
135 | sg->__cpu_power += val; | |
136 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
137 | } | |
138 | #endif | |
139 | ||
e05606d3 IM |
140 | static inline int rt_policy(int policy) |
141 | { | |
3f33a7ce | 142 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
143 | return 1; |
144 | return 0; | |
145 | } | |
146 | ||
147 | static inline int task_has_rt_policy(struct task_struct *p) | |
148 | { | |
149 | return rt_policy(p->policy); | |
150 | } | |
151 | ||
1da177e4 | 152 | /* |
6aa645ea | 153 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 154 | */ |
6aa645ea IM |
155 | struct rt_prio_array { |
156 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
157 | struct list_head queue[MAX_RT_PRIO]; | |
158 | }; | |
159 | ||
d0b27fa7 | 160 | struct rt_bandwidth { |
ea736ed5 IM |
161 | /* nests inside the rq lock: */ |
162 | spinlock_t rt_runtime_lock; | |
163 | ktime_t rt_period; | |
164 | u64 rt_runtime; | |
165 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
166 | }; |
167 | ||
168 | static struct rt_bandwidth def_rt_bandwidth; | |
169 | ||
170 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
171 | ||
172 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
173 | { | |
174 | struct rt_bandwidth *rt_b = | |
175 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
176 | ktime_t now; | |
177 | int overrun; | |
178 | int idle = 0; | |
179 | ||
180 | for (;;) { | |
181 | now = hrtimer_cb_get_time(timer); | |
182 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
183 | ||
184 | if (!overrun) | |
185 | break; | |
186 | ||
187 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
188 | } | |
189 | ||
190 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
191 | } | |
192 | ||
193 | static | |
194 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
195 | { | |
196 | rt_b->rt_period = ns_to_ktime(period); | |
197 | rt_b->rt_runtime = runtime; | |
198 | ||
ac086bc2 PZ |
199 | spin_lock_init(&rt_b->rt_runtime_lock); |
200 | ||
d0b27fa7 PZ |
201 | hrtimer_init(&rt_b->rt_period_timer, |
202 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
203 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
204 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
205 | } | |
206 | ||
207 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
208 | { | |
209 | ktime_t now; | |
210 | ||
211 | if (rt_b->rt_runtime == RUNTIME_INF) | |
212 | return; | |
213 | ||
214 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
215 | return; | |
216 | ||
217 | spin_lock(&rt_b->rt_runtime_lock); | |
218 | for (;;) { | |
219 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
220 | break; | |
221 | ||
222 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
223 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
224 | hrtimer_start(&rt_b->rt_period_timer, | |
225 | rt_b->rt_period_timer.expires, | |
226 | HRTIMER_MODE_ABS); | |
227 | } | |
228 | spin_unlock(&rt_b->rt_runtime_lock); | |
229 | } | |
230 | ||
231 | #ifdef CONFIG_RT_GROUP_SCHED | |
232 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
233 | { | |
234 | hrtimer_cancel(&rt_b->rt_period_timer); | |
235 | } | |
236 | #endif | |
237 | ||
712555ee HC |
238 | /* |
239 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
240 | * detach_destroy_domains and partition_sched_domains. | |
241 | */ | |
242 | static DEFINE_MUTEX(sched_domains_mutex); | |
243 | ||
052f1dc7 | 244 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 245 | |
68318b8e SV |
246 | #include <linux/cgroup.h> |
247 | ||
29f59db3 SV |
248 | struct cfs_rq; |
249 | ||
6f505b16 PZ |
250 | static LIST_HEAD(task_groups); |
251 | ||
29f59db3 | 252 | /* task group related information */ |
4cf86d77 | 253 | struct task_group { |
052f1dc7 | 254 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
255 | struct cgroup_subsys_state css; |
256 | #endif | |
052f1dc7 PZ |
257 | |
258 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
259 | /* schedulable entities of this group on each cpu */ |
260 | struct sched_entity **se; | |
261 | /* runqueue "owned" by this group on each cpu */ | |
262 | struct cfs_rq **cfs_rq; | |
263 | unsigned long shares; | |
052f1dc7 PZ |
264 | #endif |
265 | ||
266 | #ifdef CONFIG_RT_GROUP_SCHED | |
267 | struct sched_rt_entity **rt_se; | |
268 | struct rt_rq **rt_rq; | |
269 | ||
d0b27fa7 | 270 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 271 | #endif |
6b2d7700 | 272 | |
ae8393e5 | 273 | struct rcu_head rcu; |
6f505b16 | 274 | struct list_head list; |
f473aa5e PZ |
275 | |
276 | struct task_group *parent; | |
277 | struct list_head siblings; | |
278 | struct list_head children; | |
29f59db3 SV |
279 | }; |
280 | ||
354d60c2 | 281 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
282 | |
283 | /* | |
284 | * Root task group. | |
285 | * Every UID task group (including init_task_group aka UID-0) will | |
286 | * be a child to this group. | |
287 | */ | |
288 | struct task_group root_task_group; | |
289 | ||
052f1dc7 | 290 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
291 | /* Default task group's sched entity on each cpu */ |
292 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
293 | /* Default task group's cfs_rq on each cpu */ | |
294 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 295 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
296 | |
297 | #ifdef CONFIG_RT_GROUP_SCHED | |
298 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
299 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad DG |
300 | #endif /* CONFIG_RT_GROUP_SCHED */ |
301 | #else /* !CONFIG_FAIR_GROUP_SCHED */ | |
eff766a6 | 302 | #define root_task_group init_task_group |
6d6bc0ad | 303 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6f505b16 | 304 | |
8ed36996 | 305 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
306 | * a task group's cpu shares. |
307 | */ | |
8ed36996 | 308 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 309 | |
052f1dc7 | 310 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
311 | #ifdef CONFIG_USER_SCHED |
312 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 313 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 314 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 315 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 316 | |
cb4ad1ff | 317 | /* |
2e084786 LJ |
318 | * A weight of 0 or 1 can cause arithmetics problems. |
319 | * A weight of a cfs_rq is the sum of weights of which entities | |
320 | * are queued on this cfs_rq, so a weight of a entity should not be | |
321 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
322 | * (The default weight is 1024 - so there's no practical |
323 | * limitation from this.) | |
324 | */ | |
18d95a28 | 325 | #define MIN_SHARES 2 |
2e084786 | 326 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 327 | |
052f1dc7 PZ |
328 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
329 | #endif | |
330 | ||
29f59db3 | 331 | /* Default task group. |
3a252015 | 332 | * Every task in system belong to this group at bootup. |
29f59db3 | 333 | */ |
434d53b0 | 334 | struct task_group init_task_group; |
29f59db3 SV |
335 | |
336 | /* return group to which a task belongs */ | |
4cf86d77 | 337 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 338 | { |
4cf86d77 | 339 | struct task_group *tg; |
9b5b7751 | 340 | |
052f1dc7 | 341 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 342 | tg = p->user->tg; |
052f1dc7 | 343 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
344 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
345 | struct task_group, css); | |
24e377a8 | 346 | #else |
41a2d6cf | 347 | tg = &init_task_group; |
24e377a8 | 348 | #endif |
9b5b7751 | 349 | return tg; |
29f59db3 SV |
350 | } |
351 | ||
352 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 353 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 354 | { |
052f1dc7 | 355 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
356 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
357 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 358 | #endif |
6f505b16 | 359 | |
052f1dc7 | 360 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
361 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
362 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 363 | #endif |
29f59db3 SV |
364 | } |
365 | ||
366 | #else | |
367 | ||
6f505b16 | 368 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
369 | static inline struct task_group *task_group(struct task_struct *p) |
370 | { | |
371 | return NULL; | |
372 | } | |
29f59db3 | 373 | |
052f1dc7 | 374 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 375 | |
6aa645ea IM |
376 | /* CFS-related fields in a runqueue */ |
377 | struct cfs_rq { | |
378 | struct load_weight load; | |
379 | unsigned long nr_running; | |
380 | ||
6aa645ea | 381 | u64 exec_clock; |
e9acbff6 | 382 | u64 min_vruntime; |
103638d9 | 383 | u64 pair_start; |
6aa645ea IM |
384 | |
385 | struct rb_root tasks_timeline; | |
386 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
387 | |
388 | struct list_head tasks; | |
389 | struct list_head *balance_iterator; | |
390 | ||
391 | /* | |
392 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
393 | * It is set to NULL otherwise (i.e when none are currently running). |
394 | */ | |
aa2ac252 | 395 | struct sched_entity *curr, *next; |
ddc97297 PZ |
396 | |
397 | unsigned long nr_spread_over; | |
398 | ||
62160e3f | 399 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
400 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
401 | ||
41a2d6cf IM |
402 | /* |
403 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
404 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
405 | * (like users, containers etc.) | |
406 | * | |
407 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
408 | * list is used during load balance. | |
409 | */ | |
41a2d6cf IM |
410 | struct list_head leaf_cfs_rq_list; |
411 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
412 | |
413 | #ifdef CONFIG_SMP | |
c09595f6 | 414 | /* |
c8cba857 | 415 | * the part of load.weight contributed by tasks |
c09595f6 | 416 | */ |
c8cba857 | 417 | unsigned long task_weight; |
c09595f6 | 418 | |
c8cba857 PZ |
419 | /* |
420 | * h_load = weight * f(tg) | |
421 | * | |
422 | * Where f(tg) is the recursive weight fraction assigned to | |
423 | * this group. | |
424 | */ | |
425 | unsigned long h_load; | |
c09595f6 | 426 | |
c8cba857 PZ |
427 | /* |
428 | * this cpu's part of tg->shares | |
429 | */ | |
430 | unsigned long shares; | |
f1d239f7 PZ |
431 | |
432 | /* | |
433 | * load.weight at the time we set shares | |
434 | */ | |
435 | unsigned long rq_weight; | |
c09595f6 | 436 | #endif |
6aa645ea IM |
437 | #endif |
438 | }; | |
1da177e4 | 439 | |
6aa645ea IM |
440 | /* Real-Time classes' related field in a runqueue: */ |
441 | struct rt_rq { | |
442 | struct rt_prio_array active; | |
63489e45 | 443 | unsigned long rt_nr_running; |
052f1dc7 | 444 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
445 | int highest_prio; /* highest queued rt task prio */ |
446 | #endif | |
fa85ae24 | 447 | #ifdef CONFIG_SMP |
73fe6aae | 448 | unsigned long rt_nr_migratory; |
a22d7fc1 | 449 | int overloaded; |
fa85ae24 | 450 | #endif |
6f505b16 | 451 | int rt_throttled; |
fa85ae24 | 452 | u64 rt_time; |
ac086bc2 | 453 | u64 rt_runtime; |
ea736ed5 | 454 | /* Nests inside the rq lock: */ |
ac086bc2 | 455 | spinlock_t rt_runtime_lock; |
6f505b16 | 456 | |
052f1dc7 | 457 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
458 | unsigned long rt_nr_boosted; |
459 | ||
6f505b16 PZ |
460 | struct rq *rq; |
461 | struct list_head leaf_rt_rq_list; | |
462 | struct task_group *tg; | |
463 | struct sched_rt_entity *rt_se; | |
464 | #endif | |
6aa645ea IM |
465 | }; |
466 | ||
57d885fe GH |
467 | #ifdef CONFIG_SMP |
468 | ||
469 | /* | |
470 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
471 | * variables. Each exclusive cpuset essentially defines an island domain by |
472 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
473 | * exclusive cpuset is created, we also create and attach a new root-domain |
474 | * object. | |
475 | * | |
57d885fe GH |
476 | */ |
477 | struct root_domain { | |
478 | atomic_t refcount; | |
479 | cpumask_t span; | |
480 | cpumask_t online; | |
637f5085 | 481 | |
0eab9146 | 482 | /* |
637f5085 GH |
483 | * The "RT overload" flag: it gets set if a CPU has more than |
484 | * one runnable RT task. | |
485 | */ | |
486 | cpumask_t rto_mask; | |
0eab9146 | 487 | atomic_t rto_count; |
6e0534f2 GH |
488 | #ifdef CONFIG_SMP |
489 | struct cpupri cpupri; | |
490 | #endif | |
57d885fe GH |
491 | }; |
492 | ||
dc938520 GH |
493 | /* |
494 | * By default the system creates a single root-domain with all cpus as | |
495 | * members (mimicking the global state we have today). | |
496 | */ | |
57d885fe GH |
497 | static struct root_domain def_root_domain; |
498 | ||
499 | #endif | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * This is the main, per-CPU runqueue data structure. | |
503 | * | |
504 | * Locking rule: those places that want to lock multiple runqueues | |
505 | * (such as the load balancing or the thread migration code), lock | |
506 | * acquire operations must be ordered by ascending &runqueue. | |
507 | */ | |
70b97a7f | 508 | struct rq { |
d8016491 IM |
509 | /* runqueue lock: */ |
510 | spinlock_t lock; | |
1da177e4 LT |
511 | |
512 | /* | |
513 | * nr_running and cpu_load should be in the same cacheline because | |
514 | * remote CPUs use both these fields when doing load calculation. | |
515 | */ | |
516 | unsigned long nr_running; | |
6aa645ea IM |
517 | #define CPU_LOAD_IDX_MAX 5 |
518 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 519 | unsigned char idle_at_tick; |
46cb4b7c | 520 | #ifdef CONFIG_NO_HZ |
15934a37 | 521 | unsigned long last_tick_seen; |
46cb4b7c SS |
522 | unsigned char in_nohz_recently; |
523 | #endif | |
d8016491 IM |
524 | /* capture load from *all* tasks on this cpu: */ |
525 | struct load_weight load; | |
6aa645ea IM |
526 | unsigned long nr_load_updates; |
527 | u64 nr_switches; | |
528 | ||
529 | struct cfs_rq cfs; | |
6f505b16 | 530 | struct rt_rq rt; |
6f505b16 | 531 | |
6aa645ea | 532 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
533 | /* list of leaf cfs_rq on this cpu: */ |
534 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
535 | #endif |
536 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 537 | struct list_head leaf_rt_rq_list; |
1da177e4 | 538 | #endif |
1da177e4 LT |
539 | |
540 | /* | |
541 | * This is part of a global counter where only the total sum | |
542 | * over all CPUs matters. A task can increase this counter on | |
543 | * one CPU and if it got migrated afterwards it may decrease | |
544 | * it on another CPU. Always updated under the runqueue lock: | |
545 | */ | |
546 | unsigned long nr_uninterruptible; | |
547 | ||
36c8b586 | 548 | struct task_struct *curr, *idle; |
c9819f45 | 549 | unsigned long next_balance; |
1da177e4 | 550 | struct mm_struct *prev_mm; |
6aa645ea | 551 | |
3e51f33f | 552 | u64 clock; |
6aa645ea | 553 | |
1da177e4 LT |
554 | atomic_t nr_iowait; |
555 | ||
556 | #ifdef CONFIG_SMP | |
0eab9146 | 557 | struct root_domain *rd; |
1da177e4 LT |
558 | struct sched_domain *sd; |
559 | ||
560 | /* For active balancing */ | |
561 | int active_balance; | |
562 | int push_cpu; | |
d8016491 IM |
563 | /* cpu of this runqueue: */ |
564 | int cpu; | |
1f11eb6a | 565 | int online; |
1da177e4 | 566 | |
a8a51d5e | 567 | unsigned long avg_load_per_task; |
1da177e4 | 568 | |
36c8b586 | 569 | struct task_struct *migration_thread; |
1da177e4 LT |
570 | struct list_head migration_queue; |
571 | #endif | |
572 | ||
8f4d37ec | 573 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
574 | #ifdef CONFIG_SMP |
575 | int hrtick_csd_pending; | |
576 | struct call_single_data hrtick_csd; | |
577 | #endif | |
8f4d37ec PZ |
578 | struct hrtimer hrtick_timer; |
579 | #endif | |
580 | ||
1da177e4 LT |
581 | #ifdef CONFIG_SCHEDSTATS |
582 | /* latency stats */ | |
583 | struct sched_info rq_sched_info; | |
584 | ||
585 | /* sys_sched_yield() stats */ | |
480b9434 KC |
586 | unsigned int yld_exp_empty; |
587 | unsigned int yld_act_empty; | |
588 | unsigned int yld_both_empty; | |
589 | unsigned int yld_count; | |
1da177e4 LT |
590 | |
591 | /* schedule() stats */ | |
480b9434 KC |
592 | unsigned int sched_switch; |
593 | unsigned int sched_count; | |
594 | unsigned int sched_goidle; | |
1da177e4 LT |
595 | |
596 | /* try_to_wake_up() stats */ | |
480b9434 KC |
597 | unsigned int ttwu_count; |
598 | unsigned int ttwu_local; | |
b8efb561 IM |
599 | |
600 | /* BKL stats */ | |
480b9434 | 601 | unsigned int bkl_count; |
1da177e4 LT |
602 | #endif |
603 | }; | |
604 | ||
f34e3b61 | 605 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 606 | |
dd41f596 IM |
607 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
608 | { | |
609 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
610 | } | |
611 | ||
0a2966b4 CL |
612 | static inline int cpu_of(struct rq *rq) |
613 | { | |
614 | #ifdef CONFIG_SMP | |
615 | return rq->cpu; | |
616 | #else | |
617 | return 0; | |
618 | #endif | |
619 | } | |
620 | ||
674311d5 NP |
621 | /* |
622 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 623 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
624 | * |
625 | * The domain tree of any CPU may only be accessed from within | |
626 | * preempt-disabled sections. | |
627 | */ | |
48f24c4d IM |
628 | #define for_each_domain(cpu, __sd) \ |
629 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
630 | |
631 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
632 | #define this_rq() (&__get_cpu_var(runqueues)) | |
633 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
634 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
635 | ||
3e51f33f PZ |
636 | static inline void update_rq_clock(struct rq *rq) |
637 | { | |
638 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
639 | } | |
640 | ||
bf5c91ba IM |
641 | /* |
642 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
643 | */ | |
644 | #ifdef CONFIG_SCHED_DEBUG | |
645 | # define const_debug __read_mostly | |
646 | #else | |
647 | # define const_debug static const | |
648 | #endif | |
649 | ||
017730c1 IM |
650 | /** |
651 | * runqueue_is_locked | |
652 | * | |
653 | * Returns true if the current cpu runqueue is locked. | |
654 | * This interface allows printk to be called with the runqueue lock | |
655 | * held and know whether or not it is OK to wake up the klogd. | |
656 | */ | |
657 | int runqueue_is_locked(void) | |
658 | { | |
659 | int cpu = get_cpu(); | |
660 | struct rq *rq = cpu_rq(cpu); | |
661 | int ret; | |
662 | ||
663 | ret = spin_is_locked(&rq->lock); | |
664 | put_cpu(); | |
665 | return ret; | |
666 | } | |
667 | ||
bf5c91ba IM |
668 | /* |
669 | * Debugging: various feature bits | |
670 | */ | |
f00b45c1 PZ |
671 | |
672 | #define SCHED_FEAT(name, enabled) \ | |
673 | __SCHED_FEAT_##name , | |
674 | ||
bf5c91ba | 675 | enum { |
f00b45c1 | 676 | #include "sched_features.h" |
bf5c91ba IM |
677 | }; |
678 | ||
f00b45c1 PZ |
679 | #undef SCHED_FEAT |
680 | ||
681 | #define SCHED_FEAT(name, enabled) \ | |
682 | (1UL << __SCHED_FEAT_##name) * enabled | | |
683 | ||
bf5c91ba | 684 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
685 | #include "sched_features.h" |
686 | 0; | |
687 | ||
688 | #undef SCHED_FEAT | |
689 | ||
690 | #ifdef CONFIG_SCHED_DEBUG | |
691 | #define SCHED_FEAT(name, enabled) \ | |
692 | #name , | |
693 | ||
983ed7a6 | 694 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
695 | #include "sched_features.h" |
696 | NULL | |
697 | }; | |
698 | ||
699 | #undef SCHED_FEAT | |
700 | ||
983ed7a6 | 701 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
702 | { |
703 | filp->private_data = inode->i_private; | |
704 | return 0; | |
705 | } | |
706 | ||
707 | static ssize_t | |
708 | sched_feat_read(struct file *filp, char __user *ubuf, | |
709 | size_t cnt, loff_t *ppos) | |
710 | { | |
711 | char *buf; | |
712 | int r = 0; | |
713 | int len = 0; | |
714 | int i; | |
715 | ||
716 | for (i = 0; sched_feat_names[i]; i++) { | |
717 | len += strlen(sched_feat_names[i]); | |
718 | len += 4; | |
719 | } | |
720 | ||
721 | buf = kmalloc(len + 2, GFP_KERNEL); | |
722 | if (!buf) | |
723 | return -ENOMEM; | |
724 | ||
725 | for (i = 0; sched_feat_names[i]; i++) { | |
726 | if (sysctl_sched_features & (1UL << i)) | |
727 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
728 | else | |
c24b7c52 | 729 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
730 | } |
731 | ||
732 | r += sprintf(buf + r, "\n"); | |
733 | WARN_ON(r >= len + 2); | |
734 | ||
735 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
736 | ||
737 | kfree(buf); | |
738 | ||
739 | return r; | |
740 | } | |
741 | ||
742 | static ssize_t | |
743 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
744 | size_t cnt, loff_t *ppos) | |
745 | { | |
746 | char buf[64]; | |
747 | char *cmp = buf; | |
748 | int neg = 0; | |
749 | int i; | |
750 | ||
751 | if (cnt > 63) | |
752 | cnt = 63; | |
753 | ||
754 | if (copy_from_user(&buf, ubuf, cnt)) | |
755 | return -EFAULT; | |
756 | ||
757 | buf[cnt] = 0; | |
758 | ||
c24b7c52 | 759 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
760 | neg = 1; |
761 | cmp += 3; | |
762 | } | |
763 | ||
764 | for (i = 0; sched_feat_names[i]; i++) { | |
765 | int len = strlen(sched_feat_names[i]); | |
766 | ||
767 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
768 | if (neg) | |
769 | sysctl_sched_features &= ~(1UL << i); | |
770 | else | |
771 | sysctl_sched_features |= (1UL << i); | |
772 | break; | |
773 | } | |
774 | } | |
775 | ||
776 | if (!sched_feat_names[i]) | |
777 | return -EINVAL; | |
778 | ||
779 | filp->f_pos += cnt; | |
780 | ||
781 | return cnt; | |
782 | } | |
783 | ||
784 | static struct file_operations sched_feat_fops = { | |
785 | .open = sched_feat_open, | |
786 | .read = sched_feat_read, | |
787 | .write = sched_feat_write, | |
788 | }; | |
789 | ||
790 | static __init int sched_init_debug(void) | |
791 | { | |
f00b45c1 PZ |
792 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
793 | &sched_feat_fops); | |
794 | ||
795 | return 0; | |
796 | } | |
797 | late_initcall(sched_init_debug); | |
798 | ||
799 | #endif | |
800 | ||
801 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 802 | |
b82d9fdd PZ |
803 | /* |
804 | * Number of tasks to iterate in a single balance run. | |
805 | * Limited because this is done with IRQs disabled. | |
806 | */ | |
807 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
808 | ||
2398f2c6 PZ |
809 | /* |
810 | * ratelimit for updating the group shares. | |
55cd5340 | 811 | * default: 0.25ms |
2398f2c6 | 812 | */ |
55cd5340 | 813 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 814 | |
fa85ae24 | 815 | /* |
9f0c1e56 | 816 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
817 | * default: 1s |
818 | */ | |
9f0c1e56 | 819 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 820 | |
6892b75e IM |
821 | static __read_mostly int scheduler_running; |
822 | ||
9f0c1e56 PZ |
823 | /* |
824 | * part of the period that we allow rt tasks to run in us. | |
825 | * default: 0.95s | |
826 | */ | |
827 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 828 | |
d0b27fa7 PZ |
829 | static inline u64 global_rt_period(void) |
830 | { | |
831 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
832 | } | |
833 | ||
834 | static inline u64 global_rt_runtime(void) | |
835 | { | |
e26873bb | 836 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
837 | return RUNTIME_INF; |
838 | ||
839 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
840 | } | |
fa85ae24 | 841 | |
1da177e4 | 842 | #ifndef prepare_arch_switch |
4866cde0 NP |
843 | # define prepare_arch_switch(next) do { } while (0) |
844 | #endif | |
845 | #ifndef finish_arch_switch | |
846 | # define finish_arch_switch(prev) do { } while (0) | |
847 | #endif | |
848 | ||
051a1d1a DA |
849 | static inline int task_current(struct rq *rq, struct task_struct *p) |
850 | { | |
851 | return rq->curr == p; | |
852 | } | |
853 | ||
4866cde0 | 854 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 855 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 856 | { |
051a1d1a | 857 | return task_current(rq, p); |
4866cde0 NP |
858 | } |
859 | ||
70b97a7f | 860 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
861 | { |
862 | } | |
863 | ||
70b97a7f | 864 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 865 | { |
da04c035 IM |
866 | #ifdef CONFIG_DEBUG_SPINLOCK |
867 | /* this is a valid case when another task releases the spinlock */ | |
868 | rq->lock.owner = current; | |
869 | #endif | |
8a25d5de IM |
870 | /* |
871 | * If we are tracking spinlock dependencies then we have to | |
872 | * fix up the runqueue lock - which gets 'carried over' from | |
873 | * prev into current: | |
874 | */ | |
875 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
876 | ||
4866cde0 NP |
877 | spin_unlock_irq(&rq->lock); |
878 | } | |
879 | ||
880 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 881 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
882 | { |
883 | #ifdef CONFIG_SMP | |
884 | return p->oncpu; | |
885 | #else | |
051a1d1a | 886 | return task_current(rq, p); |
4866cde0 NP |
887 | #endif |
888 | } | |
889 | ||
70b97a7f | 890 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
891 | { |
892 | #ifdef CONFIG_SMP | |
893 | /* | |
894 | * We can optimise this out completely for !SMP, because the | |
895 | * SMP rebalancing from interrupt is the only thing that cares | |
896 | * here. | |
897 | */ | |
898 | next->oncpu = 1; | |
899 | #endif | |
900 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
901 | spin_unlock_irq(&rq->lock); | |
902 | #else | |
903 | spin_unlock(&rq->lock); | |
904 | #endif | |
905 | } | |
906 | ||
70b97a7f | 907 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
908 | { |
909 | #ifdef CONFIG_SMP | |
910 | /* | |
911 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
912 | * We must ensure this doesn't happen until the switch is completely | |
913 | * finished. | |
914 | */ | |
915 | smp_wmb(); | |
916 | prev->oncpu = 0; | |
917 | #endif | |
918 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
919 | local_irq_enable(); | |
1da177e4 | 920 | #endif |
4866cde0 NP |
921 | } |
922 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 923 | |
b29739f9 IM |
924 | /* |
925 | * __task_rq_lock - lock the runqueue a given task resides on. | |
926 | * Must be called interrupts disabled. | |
927 | */ | |
70b97a7f | 928 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
929 | __acquires(rq->lock) |
930 | { | |
3a5c359a AK |
931 | for (;;) { |
932 | struct rq *rq = task_rq(p); | |
933 | spin_lock(&rq->lock); | |
934 | if (likely(rq == task_rq(p))) | |
935 | return rq; | |
b29739f9 | 936 | spin_unlock(&rq->lock); |
b29739f9 | 937 | } |
b29739f9 IM |
938 | } |
939 | ||
1da177e4 LT |
940 | /* |
941 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 942 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
943 | * explicitly disabling preemption. |
944 | */ | |
70b97a7f | 945 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
946 | __acquires(rq->lock) |
947 | { | |
70b97a7f | 948 | struct rq *rq; |
1da177e4 | 949 | |
3a5c359a AK |
950 | for (;;) { |
951 | local_irq_save(*flags); | |
952 | rq = task_rq(p); | |
953 | spin_lock(&rq->lock); | |
954 | if (likely(rq == task_rq(p))) | |
955 | return rq; | |
1da177e4 | 956 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 957 | } |
1da177e4 LT |
958 | } |
959 | ||
a9957449 | 960 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
961 | __releases(rq->lock) |
962 | { | |
963 | spin_unlock(&rq->lock); | |
964 | } | |
965 | ||
70b97a7f | 966 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
967 | __releases(rq->lock) |
968 | { | |
969 | spin_unlock_irqrestore(&rq->lock, *flags); | |
970 | } | |
971 | ||
1da177e4 | 972 | /* |
cc2a73b5 | 973 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 974 | */ |
a9957449 | 975 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
976 | __acquires(rq->lock) |
977 | { | |
70b97a7f | 978 | struct rq *rq; |
1da177e4 LT |
979 | |
980 | local_irq_disable(); | |
981 | rq = this_rq(); | |
982 | spin_lock(&rq->lock); | |
983 | ||
984 | return rq; | |
985 | } | |
986 | ||
8f4d37ec PZ |
987 | #ifdef CONFIG_SCHED_HRTICK |
988 | /* | |
989 | * Use HR-timers to deliver accurate preemption points. | |
990 | * | |
991 | * Its all a bit involved since we cannot program an hrt while holding the | |
992 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
993 | * reschedule event. | |
994 | * | |
995 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
996 | * rq->lock. | |
997 | */ | |
8f4d37ec PZ |
998 | |
999 | /* | |
1000 | * Use hrtick when: | |
1001 | * - enabled by features | |
1002 | * - hrtimer is actually high res | |
1003 | */ | |
1004 | static inline int hrtick_enabled(struct rq *rq) | |
1005 | { | |
1006 | if (!sched_feat(HRTICK)) | |
1007 | return 0; | |
ba42059f | 1008 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1009 | return 0; |
8f4d37ec PZ |
1010 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1011 | } | |
1012 | ||
8f4d37ec PZ |
1013 | static void hrtick_clear(struct rq *rq) |
1014 | { | |
1015 | if (hrtimer_active(&rq->hrtick_timer)) | |
1016 | hrtimer_cancel(&rq->hrtick_timer); | |
1017 | } | |
1018 | ||
8f4d37ec PZ |
1019 | /* |
1020 | * High-resolution timer tick. | |
1021 | * Runs from hardirq context with interrupts disabled. | |
1022 | */ | |
1023 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1024 | { | |
1025 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1026 | ||
1027 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1028 | ||
1029 | spin_lock(&rq->lock); | |
3e51f33f | 1030 | update_rq_clock(rq); |
8f4d37ec PZ |
1031 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1032 | spin_unlock(&rq->lock); | |
1033 | ||
1034 | return HRTIMER_NORESTART; | |
1035 | } | |
1036 | ||
95e904c7 | 1037 | #ifdef CONFIG_SMP |
31656519 PZ |
1038 | /* |
1039 | * called from hardirq (IPI) context | |
1040 | */ | |
1041 | static void __hrtick_start(void *arg) | |
b328ca18 | 1042 | { |
31656519 | 1043 | struct rq *rq = arg; |
b328ca18 | 1044 | |
31656519 PZ |
1045 | spin_lock(&rq->lock); |
1046 | hrtimer_restart(&rq->hrtick_timer); | |
1047 | rq->hrtick_csd_pending = 0; | |
1048 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1049 | } |
1050 | ||
31656519 PZ |
1051 | /* |
1052 | * Called to set the hrtick timer state. | |
1053 | * | |
1054 | * called with rq->lock held and irqs disabled | |
1055 | */ | |
1056 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1057 | { |
31656519 PZ |
1058 | struct hrtimer *timer = &rq->hrtick_timer; |
1059 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1060 | |
31656519 PZ |
1061 | timer->expires = time; |
1062 | ||
1063 | if (rq == this_rq()) { | |
1064 | hrtimer_restart(timer); | |
1065 | } else if (!rq->hrtick_csd_pending) { | |
1066 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1067 | rq->hrtick_csd_pending = 1; | |
1068 | } | |
b328ca18 PZ |
1069 | } |
1070 | ||
1071 | static int | |
1072 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1073 | { | |
1074 | int cpu = (int)(long)hcpu; | |
1075 | ||
1076 | switch (action) { | |
1077 | case CPU_UP_CANCELED: | |
1078 | case CPU_UP_CANCELED_FROZEN: | |
1079 | case CPU_DOWN_PREPARE: | |
1080 | case CPU_DOWN_PREPARE_FROZEN: | |
1081 | case CPU_DEAD: | |
1082 | case CPU_DEAD_FROZEN: | |
31656519 | 1083 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1084 | return NOTIFY_OK; |
1085 | } | |
1086 | ||
1087 | return NOTIFY_DONE; | |
1088 | } | |
1089 | ||
1090 | static void init_hrtick(void) | |
1091 | { | |
1092 | hotcpu_notifier(hotplug_hrtick, 0); | |
1093 | } | |
31656519 PZ |
1094 | #else |
1095 | /* | |
1096 | * Called to set the hrtick timer state. | |
1097 | * | |
1098 | * called with rq->lock held and irqs disabled | |
1099 | */ | |
1100 | static void hrtick_start(struct rq *rq, u64 delay) | |
1101 | { | |
1102 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1103 | } | |
b328ca18 | 1104 | |
31656519 | 1105 | static void init_hrtick(void) |
8f4d37ec | 1106 | { |
8f4d37ec | 1107 | } |
31656519 | 1108 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1109 | |
31656519 | 1110 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1111 | { |
31656519 PZ |
1112 | #ifdef CONFIG_SMP |
1113 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1114 | |
31656519 PZ |
1115 | rq->hrtick_csd.flags = 0; |
1116 | rq->hrtick_csd.func = __hrtick_start; | |
1117 | rq->hrtick_csd.info = rq; | |
1118 | #endif | |
8f4d37ec | 1119 | |
31656519 PZ |
1120 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1121 | rq->hrtick_timer.function = hrtick; | |
1122 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
8f4d37ec PZ |
1123 | } |
1124 | #else | |
1125 | static inline void hrtick_clear(struct rq *rq) | |
1126 | { | |
1127 | } | |
1128 | ||
8f4d37ec PZ |
1129 | static inline void init_rq_hrtick(struct rq *rq) |
1130 | { | |
1131 | } | |
1132 | ||
b328ca18 PZ |
1133 | static inline void init_hrtick(void) |
1134 | { | |
1135 | } | |
8f4d37ec PZ |
1136 | #endif |
1137 | ||
c24d20db IM |
1138 | /* |
1139 | * resched_task - mark a task 'to be rescheduled now'. | |
1140 | * | |
1141 | * On UP this means the setting of the need_resched flag, on SMP it | |
1142 | * might also involve a cross-CPU call to trigger the scheduler on | |
1143 | * the target CPU. | |
1144 | */ | |
1145 | #ifdef CONFIG_SMP | |
1146 | ||
1147 | #ifndef tsk_is_polling | |
1148 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1149 | #endif | |
1150 | ||
31656519 | 1151 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1152 | { |
1153 | int cpu; | |
1154 | ||
1155 | assert_spin_locked(&task_rq(p)->lock); | |
1156 | ||
31656519 | 1157 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1158 | return; |
1159 | ||
31656519 | 1160 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1161 | |
1162 | cpu = task_cpu(p); | |
1163 | if (cpu == smp_processor_id()) | |
1164 | return; | |
1165 | ||
1166 | /* NEED_RESCHED must be visible before we test polling */ | |
1167 | smp_mb(); | |
1168 | if (!tsk_is_polling(p)) | |
1169 | smp_send_reschedule(cpu); | |
1170 | } | |
1171 | ||
1172 | static void resched_cpu(int cpu) | |
1173 | { | |
1174 | struct rq *rq = cpu_rq(cpu); | |
1175 | unsigned long flags; | |
1176 | ||
1177 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1178 | return; | |
1179 | resched_task(cpu_curr(cpu)); | |
1180 | spin_unlock_irqrestore(&rq->lock, flags); | |
1181 | } | |
06d8308c TG |
1182 | |
1183 | #ifdef CONFIG_NO_HZ | |
1184 | /* | |
1185 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1186 | * idle CPU then this timer might expire before the next timer event | |
1187 | * which is scheduled to wake up that CPU. In case of a completely | |
1188 | * idle system the next event might even be infinite time into the | |
1189 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1190 | * leaves the inner idle loop so the newly added timer is taken into | |
1191 | * account when the CPU goes back to idle and evaluates the timer | |
1192 | * wheel for the next timer event. | |
1193 | */ | |
1194 | void wake_up_idle_cpu(int cpu) | |
1195 | { | |
1196 | struct rq *rq = cpu_rq(cpu); | |
1197 | ||
1198 | if (cpu == smp_processor_id()) | |
1199 | return; | |
1200 | ||
1201 | /* | |
1202 | * This is safe, as this function is called with the timer | |
1203 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1204 | * to idle and has not yet set rq->curr to idle then it will | |
1205 | * be serialized on the timer wheel base lock and take the new | |
1206 | * timer into account automatically. | |
1207 | */ | |
1208 | if (rq->curr != rq->idle) | |
1209 | return; | |
1210 | ||
1211 | /* | |
1212 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1213 | * lockless. The worst case is that the other CPU runs the | |
1214 | * idle task through an additional NOOP schedule() | |
1215 | */ | |
1216 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1217 | ||
1218 | /* NEED_RESCHED must be visible before we test polling */ | |
1219 | smp_mb(); | |
1220 | if (!tsk_is_polling(rq->idle)) | |
1221 | smp_send_reschedule(cpu); | |
1222 | } | |
6d6bc0ad | 1223 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1224 | |
6d6bc0ad | 1225 | #else /* !CONFIG_SMP */ |
31656519 | 1226 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1227 | { |
1228 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1229 | set_tsk_need_resched(p); |
c24d20db | 1230 | } |
6d6bc0ad | 1231 | #endif /* CONFIG_SMP */ |
c24d20db | 1232 | |
45bf76df IM |
1233 | #if BITS_PER_LONG == 32 |
1234 | # define WMULT_CONST (~0UL) | |
1235 | #else | |
1236 | # define WMULT_CONST (1UL << 32) | |
1237 | #endif | |
1238 | ||
1239 | #define WMULT_SHIFT 32 | |
1240 | ||
194081eb IM |
1241 | /* |
1242 | * Shift right and round: | |
1243 | */ | |
cf2ab469 | 1244 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1245 | |
a7be37ac PZ |
1246 | /* |
1247 | * delta *= weight / lw | |
1248 | */ | |
cb1c4fc9 | 1249 | static unsigned long |
45bf76df IM |
1250 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1251 | struct load_weight *lw) | |
1252 | { | |
1253 | u64 tmp; | |
1254 | ||
7a232e03 LJ |
1255 | if (!lw->inv_weight) { |
1256 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1257 | lw->inv_weight = 1; | |
1258 | else | |
1259 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1260 | / (lw->weight+1); | |
1261 | } | |
45bf76df IM |
1262 | |
1263 | tmp = (u64)delta_exec * weight; | |
1264 | /* | |
1265 | * Check whether we'd overflow the 64-bit multiplication: | |
1266 | */ | |
194081eb | 1267 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1268 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1269 | WMULT_SHIFT/2); |
1270 | else | |
cf2ab469 | 1271 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1272 | |
ecf691da | 1273 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1274 | } |
1275 | ||
1091985b | 1276 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1277 | { |
1278 | lw->weight += inc; | |
e89996ae | 1279 | lw->inv_weight = 0; |
45bf76df IM |
1280 | } |
1281 | ||
1091985b | 1282 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1283 | { |
1284 | lw->weight -= dec; | |
e89996ae | 1285 | lw->inv_weight = 0; |
45bf76df IM |
1286 | } |
1287 | ||
2dd73a4f PW |
1288 | /* |
1289 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1290 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1291 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1292 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1293 | * scaled version of the new time slice allocation that they receive on time |
1294 | * slice expiry etc. | |
1295 | */ | |
1296 | ||
dd41f596 IM |
1297 | #define WEIGHT_IDLEPRIO 2 |
1298 | #define WMULT_IDLEPRIO (1 << 31) | |
1299 | ||
1300 | /* | |
1301 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1302 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1303 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1304 | * that remained on nice 0. | |
1305 | * | |
1306 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1307 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1308 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1309 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1310 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1311 | */ |
1312 | static const int prio_to_weight[40] = { | |
254753dc IM |
1313 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1314 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1315 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1316 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1317 | /* 0 */ 1024, 820, 655, 526, 423, | |
1318 | /* 5 */ 335, 272, 215, 172, 137, | |
1319 | /* 10 */ 110, 87, 70, 56, 45, | |
1320 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1321 | }; |
1322 | ||
5714d2de IM |
1323 | /* |
1324 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1325 | * | |
1326 | * In cases where the weight does not change often, we can use the | |
1327 | * precalculated inverse to speed up arithmetics by turning divisions | |
1328 | * into multiplications: | |
1329 | */ | |
dd41f596 | 1330 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1331 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1332 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1333 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1334 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1335 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1336 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1337 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1338 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1339 | }; |
2dd73a4f | 1340 | |
dd41f596 IM |
1341 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1342 | ||
1343 | /* | |
1344 | * runqueue iterator, to support SMP load-balancing between different | |
1345 | * scheduling classes, without having to expose their internal data | |
1346 | * structures to the load-balancing proper: | |
1347 | */ | |
1348 | struct rq_iterator { | |
1349 | void *arg; | |
1350 | struct task_struct *(*start)(void *); | |
1351 | struct task_struct *(*next)(void *); | |
1352 | }; | |
1353 | ||
e1d1484f PW |
1354 | #ifdef CONFIG_SMP |
1355 | static unsigned long | |
1356 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1357 | unsigned long max_load_move, struct sched_domain *sd, | |
1358 | enum cpu_idle_type idle, int *all_pinned, | |
1359 | int *this_best_prio, struct rq_iterator *iterator); | |
1360 | ||
1361 | static int | |
1362 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1363 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1364 | struct rq_iterator *iterator); | |
e1d1484f | 1365 | #endif |
dd41f596 | 1366 | |
d842de87 SV |
1367 | #ifdef CONFIG_CGROUP_CPUACCT |
1368 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1369 | #else | |
1370 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1371 | #endif | |
1372 | ||
18d95a28 PZ |
1373 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1374 | { | |
1375 | update_load_add(&rq->load, load); | |
1376 | } | |
1377 | ||
1378 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1379 | { | |
1380 | update_load_sub(&rq->load, load); | |
1381 | } | |
1382 | ||
e7693a36 GH |
1383 | #ifdef CONFIG_SMP |
1384 | static unsigned long source_load(int cpu, int type); | |
1385 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1386 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1387 | |
a8a51d5e PZ |
1388 | static unsigned long cpu_avg_load_per_task(int cpu) |
1389 | { | |
1390 | struct rq *rq = cpu_rq(cpu); | |
1391 | ||
1392 | if (rq->nr_running) | |
1393 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1394 | ||
1395 | return rq->avg_load_per_task; | |
1396 | } | |
18d95a28 PZ |
1397 | |
1398 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1399 | |
c8cba857 | 1400 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1401 | |
1402 | /* | |
1403 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1404 | * leaving it for the final time. | |
1405 | */ | |
c8cba857 PZ |
1406 | static void |
1407 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1408 | { |
1409 | struct task_group *parent, *child; | |
1410 | ||
1411 | rcu_read_lock(); | |
1412 | parent = &root_task_group; | |
1413 | down: | |
b6a86c74 | 1414 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1415 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1416 | parent = child; | |
1417 | goto down; | |
1418 | ||
1419 | up: | |
1420 | continue; | |
1421 | } | |
b6a86c74 | 1422 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1423 | |
1424 | child = parent; | |
1425 | parent = parent->parent; | |
1426 | if (parent) | |
1427 | goto up; | |
1428 | rcu_read_unlock(); | |
1429 | } | |
1430 | ||
c09595f6 PZ |
1431 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1432 | ||
1433 | /* | |
1434 | * Calculate and set the cpu's group shares. | |
1435 | */ | |
1436 | static void | |
b6a86c74 | 1437 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1438 | unsigned long sd_shares, unsigned long sd_rq_weight) |
18d95a28 | 1439 | { |
c09595f6 PZ |
1440 | int boost = 0; |
1441 | unsigned long shares; | |
1442 | unsigned long rq_weight; | |
1443 | ||
c8cba857 | 1444 | if (!tg->se[cpu]) |
c09595f6 PZ |
1445 | return; |
1446 | ||
c8cba857 | 1447 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1448 | |
1449 | /* | |
1450 | * If there are currently no tasks on the cpu pretend there is one of | |
1451 | * average load so that when a new task gets to run here it will not | |
1452 | * get delayed by group starvation. | |
1453 | */ | |
1454 | if (!rq_weight) { | |
1455 | boost = 1; | |
1456 | rq_weight = NICE_0_LOAD; | |
1457 | } | |
1458 | ||
c8cba857 PZ |
1459 | if (unlikely(rq_weight > sd_rq_weight)) |
1460 | rq_weight = sd_rq_weight; | |
1461 | ||
c09595f6 PZ |
1462 | /* |
1463 | * \Sum shares * rq_weight | |
1464 | * shares = ----------------------- | |
1465 | * \Sum rq_weight | |
1466 | * | |
1467 | */ | |
c8cba857 | 1468 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1469 | |
1470 | /* | |
1471 | * record the actual number of shares, not the boosted amount. | |
1472 | */ | |
c8cba857 | 1473 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
f1d239f7 | 1474 | tg->cfs_rq[cpu]->rq_weight = rq_weight; |
c09595f6 PZ |
1475 | |
1476 | if (shares < MIN_SHARES) | |
1477 | shares = MIN_SHARES; | |
1478 | else if (shares > MAX_SHARES) | |
1479 | shares = MAX_SHARES; | |
1480 | ||
c8cba857 | 1481 | __set_se_shares(tg->se[cpu], shares); |
18d95a28 | 1482 | } |
c09595f6 PZ |
1483 | |
1484 | /* | |
c8cba857 PZ |
1485 | * Re-compute the task group their per cpu shares over the given domain. |
1486 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1487 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1488 | */ |
1489 | static void | |
c8cba857 | 1490 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1491 | { |
c8cba857 PZ |
1492 | unsigned long rq_weight = 0; |
1493 | unsigned long shares = 0; | |
1494 | int i; | |
c09595f6 | 1495 | |
c8cba857 PZ |
1496 | for_each_cpu_mask(i, sd->span) { |
1497 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1498 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1499 | } |
c09595f6 | 1500 | |
c8cba857 PZ |
1501 | if ((!shares && rq_weight) || shares > tg->shares) |
1502 | shares = tg->shares; | |
1503 | ||
1504 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1505 | shares = tg->shares; | |
c09595f6 | 1506 | |
cd80917e PZ |
1507 | if (!rq_weight) |
1508 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1509 | ||
c09595f6 PZ |
1510 | for_each_cpu_mask(i, sd->span) { |
1511 | struct rq *rq = cpu_rq(i); | |
1512 | unsigned long flags; | |
1513 | ||
1514 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1515 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1516 | spin_unlock_irqrestore(&rq->lock, flags); |
1517 | } | |
c09595f6 PZ |
1518 | } |
1519 | ||
1520 | /* | |
c8cba857 PZ |
1521 | * Compute the cpu's hierarchical load factor for each task group. |
1522 | * This needs to be done in a top-down fashion because the load of a child | |
1523 | * group is a fraction of its parents load. | |
c09595f6 | 1524 | */ |
b6a86c74 | 1525 | static void |
c8cba857 | 1526 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1527 | { |
c8cba857 | 1528 | unsigned long load; |
c09595f6 | 1529 | |
c8cba857 PZ |
1530 | if (!tg->parent) { |
1531 | load = cpu_rq(cpu)->load.weight; | |
1532 | } else { | |
1533 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1534 | load *= tg->cfs_rq[cpu]->shares; | |
1535 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1536 | } | |
c09595f6 | 1537 | |
c8cba857 | 1538 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1539 | } |
1540 | ||
c8cba857 PZ |
1541 | static void |
1542 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1543 | { |
c09595f6 PZ |
1544 | } |
1545 | ||
c8cba857 | 1546 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1547 | { |
2398f2c6 PZ |
1548 | u64 now = cpu_clock(raw_smp_processor_id()); |
1549 | s64 elapsed = now - sd->last_update; | |
1550 | ||
1551 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1552 | sd->last_update = now; | |
1553 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); | |
1554 | } | |
4d8d595d PZ |
1555 | } |
1556 | ||
3e5459b4 PZ |
1557 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1558 | { | |
1559 | spin_unlock(&rq->lock); | |
1560 | update_shares(sd); | |
1561 | spin_lock(&rq->lock); | |
1562 | } | |
1563 | ||
c8cba857 | 1564 | static void update_h_load(int cpu) |
c09595f6 | 1565 | { |
c8cba857 | 1566 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1567 | } |
1568 | ||
c09595f6 PZ |
1569 | #else |
1570 | ||
c8cba857 | 1571 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1572 | { |
1573 | } | |
1574 | ||
3e5459b4 PZ |
1575 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1576 | { | |
1577 | } | |
1578 | ||
18d95a28 PZ |
1579 | #endif |
1580 | ||
18d95a28 PZ |
1581 | #endif |
1582 | ||
30432094 | 1583 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1584 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1585 | { | |
30432094 | 1586 | #ifdef CONFIG_SMP |
34e83e85 IM |
1587 | cfs_rq->shares = shares; |
1588 | #endif | |
1589 | } | |
30432094 | 1590 | #endif |
e7693a36 | 1591 | |
dd41f596 | 1592 | #include "sched_stats.h" |
dd41f596 | 1593 | #include "sched_idletask.c" |
5522d5d5 IM |
1594 | #include "sched_fair.c" |
1595 | #include "sched_rt.c" | |
dd41f596 IM |
1596 | #ifdef CONFIG_SCHED_DEBUG |
1597 | # include "sched_debug.c" | |
1598 | #endif | |
1599 | ||
1600 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1601 | #define for_each_class(class) \ |
1602 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1603 | |
c09595f6 | 1604 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1605 | { |
1606 | rq->nr_running++; | |
9c217245 IM |
1607 | } |
1608 | ||
c09595f6 | 1609 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1610 | { |
1611 | rq->nr_running--; | |
9c217245 IM |
1612 | } |
1613 | ||
45bf76df IM |
1614 | static void set_load_weight(struct task_struct *p) |
1615 | { | |
1616 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1617 | p->se.load.weight = prio_to_weight[0] * 2; |
1618 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1619 | return; | |
1620 | } | |
45bf76df | 1621 | |
dd41f596 IM |
1622 | /* |
1623 | * SCHED_IDLE tasks get minimal weight: | |
1624 | */ | |
1625 | if (p->policy == SCHED_IDLE) { | |
1626 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1627 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1628 | return; | |
1629 | } | |
71f8bd46 | 1630 | |
dd41f596 IM |
1631 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1632 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1633 | } |
1634 | ||
2087a1ad GH |
1635 | static void update_avg(u64 *avg, u64 sample) |
1636 | { | |
1637 | s64 diff = sample - *avg; | |
1638 | *avg += diff >> 3; | |
1639 | } | |
1640 | ||
8159f87e | 1641 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1642 | { |
dd41f596 | 1643 | sched_info_queued(p); |
fd390f6a | 1644 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1645 | p->se.on_rq = 1; |
71f8bd46 IM |
1646 | } |
1647 | ||
69be72c1 | 1648 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1649 | { |
2087a1ad GH |
1650 | if (sleep && p->se.last_wakeup) { |
1651 | update_avg(&p->se.avg_overlap, | |
1652 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1653 | p->se.last_wakeup = 0; | |
1654 | } | |
1655 | ||
46ac22ba | 1656 | sched_info_dequeued(p); |
f02231e5 | 1657 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1658 | p->se.on_rq = 0; |
71f8bd46 IM |
1659 | } |
1660 | ||
14531189 | 1661 | /* |
dd41f596 | 1662 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1663 | */ |
14531189 IM |
1664 | static inline int __normal_prio(struct task_struct *p) |
1665 | { | |
dd41f596 | 1666 | return p->static_prio; |
14531189 IM |
1667 | } |
1668 | ||
b29739f9 IM |
1669 | /* |
1670 | * Calculate the expected normal priority: i.e. priority | |
1671 | * without taking RT-inheritance into account. Might be | |
1672 | * boosted by interactivity modifiers. Changes upon fork, | |
1673 | * setprio syscalls, and whenever the interactivity | |
1674 | * estimator recalculates. | |
1675 | */ | |
36c8b586 | 1676 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1677 | { |
1678 | int prio; | |
1679 | ||
e05606d3 | 1680 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1681 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1682 | else | |
1683 | prio = __normal_prio(p); | |
1684 | return prio; | |
1685 | } | |
1686 | ||
1687 | /* | |
1688 | * Calculate the current priority, i.e. the priority | |
1689 | * taken into account by the scheduler. This value might | |
1690 | * be boosted by RT tasks, or might be boosted by | |
1691 | * interactivity modifiers. Will be RT if the task got | |
1692 | * RT-boosted. If not then it returns p->normal_prio. | |
1693 | */ | |
36c8b586 | 1694 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1695 | { |
1696 | p->normal_prio = normal_prio(p); | |
1697 | /* | |
1698 | * If we are RT tasks or we were boosted to RT priority, | |
1699 | * keep the priority unchanged. Otherwise, update priority | |
1700 | * to the normal priority: | |
1701 | */ | |
1702 | if (!rt_prio(p->prio)) | |
1703 | return p->normal_prio; | |
1704 | return p->prio; | |
1705 | } | |
1706 | ||
1da177e4 | 1707 | /* |
dd41f596 | 1708 | * activate_task - move a task to the runqueue. |
1da177e4 | 1709 | */ |
dd41f596 | 1710 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1711 | { |
d9514f6c | 1712 | if (task_contributes_to_load(p)) |
dd41f596 | 1713 | rq->nr_uninterruptible--; |
1da177e4 | 1714 | |
8159f87e | 1715 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1716 | inc_nr_running(rq); |
1da177e4 LT |
1717 | } |
1718 | ||
1da177e4 LT |
1719 | /* |
1720 | * deactivate_task - remove a task from the runqueue. | |
1721 | */ | |
2e1cb74a | 1722 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1723 | { |
d9514f6c | 1724 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1725 | rq->nr_uninterruptible++; |
1726 | ||
69be72c1 | 1727 | dequeue_task(rq, p, sleep); |
c09595f6 | 1728 | dec_nr_running(rq); |
1da177e4 LT |
1729 | } |
1730 | ||
1da177e4 LT |
1731 | /** |
1732 | * task_curr - is this task currently executing on a CPU? | |
1733 | * @p: the task in question. | |
1734 | */ | |
36c8b586 | 1735 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1736 | { |
1737 | return cpu_curr(task_cpu(p)) == p; | |
1738 | } | |
1739 | ||
dd41f596 IM |
1740 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1741 | { | |
6f505b16 | 1742 | set_task_rq(p, cpu); |
dd41f596 | 1743 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1744 | /* |
1745 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1746 | * successfuly executed on another CPU. We must ensure that updates of | |
1747 | * per-task data have been completed by this moment. | |
1748 | */ | |
1749 | smp_wmb(); | |
dd41f596 | 1750 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1751 | #endif |
2dd73a4f PW |
1752 | } |
1753 | ||
cb469845 SR |
1754 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1755 | const struct sched_class *prev_class, | |
1756 | int oldprio, int running) | |
1757 | { | |
1758 | if (prev_class != p->sched_class) { | |
1759 | if (prev_class->switched_from) | |
1760 | prev_class->switched_from(rq, p, running); | |
1761 | p->sched_class->switched_to(rq, p, running); | |
1762 | } else | |
1763 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1764 | } | |
1765 | ||
1da177e4 | 1766 | #ifdef CONFIG_SMP |
c65cc870 | 1767 | |
e958b360 TG |
1768 | /* Used instead of source_load when we know the type == 0 */ |
1769 | static unsigned long weighted_cpuload(const int cpu) | |
1770 | { | |
1771 | return cpu_rq(cpu)->load.weight; | |
1772 | } | |
1773 | ||
cc367732 IM |
1774 | /* |
1775 | * Is this task likely cache-hot: | |
1776 | */ | |
e7693a36 | 1777 | static int |
cc367732 IM |
1778 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1779 | { | |
1780 | s64 delta; | |
1781 | ||
f540a608 IM |
1782 | /* |
1783 | * Buddy candidates are cache hot: | |
1784 | */ | |
d25ce4cd | 1785 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1786 | return 1; |
1787 | ||
cc367732 IM |
1788 | if (p->sched_class != &fair_sched_class) |
1789 | return 0; | |
1790 | ||
6bc1665b IM |
1791 | if (sysctl_sched_migration_cost == -1) |
1792 | return 1; | |
1793 | if (sysctl_sched_migration_cost == 0) | |
1794 | return 0; | |
1795 | ||
cc367732 IM |
1796 | delta = now - p->se.exec_start; |
1797 | ||
1798 | return delta < (s64)sysctl_sched_migration_cost; | |
1799 | } | |
1800 | ||
1801 | ||
dd41f596 | 1802 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1803 | { |
dd41f596 IM |
1804 | int old_cpu = task_cpu(p); |
1805 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1806 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1807 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1808 | u64 clock_offset; |
dd41f596 IM |
1809 | |
1810 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1811 | |
1812 | #ifdef CONFIG_SCHEDSTATS | |
1813 | if (p->se.wait_start) | |
1814 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1815 | if (p->se.sleep_start) |
1816 | p->se.sleep_start -= clock_offset; | |
1817 | if (p->se.block_start) | |
1818 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1819 | if (old_cpu != new_cpu) { |
1820 | schedstat_inc(p, se.nr_migrations); | |
1821 | if (task_hot(p, old_rq->clock, NULL)) | |
1822 | schedstat_inc(p, se.nr_forced2_migrations); | |
1823 | } | |
6cfb0d5d | 1824 | #endif |
2830cf8c SV |
1825 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1826 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1827 | |
1828 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1829 | } |
1830 | ||
70b97a7f | 1831 | struct migration_req { |
1da177e4 | 1832 | struct list_head list; |
1da177e4 | 1833 | |
36c8b586 | 1834 | struct task_struct *task; |
1da177e4 LT |
1835 | int dest_cpu; |
1836 | ||
1da177e4 | 1837 | struct completion done; |
70b97a7f | 1838 | }; |
1da177e4 LT |
1839 | |
1840 | /* | |
1841 | * The task's runqueue lock must be held. | |
1842 | * Returns true if you have to wait for migration thread. | |
1843 | */ | |
36c8b586 | 1844 | static int |
70b97a7f | 1845 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1846 | { |
70b97a7f | 1847 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1848 | |
1849 | /* | |
1850 | * If the task is not on a runqueue (and not running), then | |
1851 | * it is sufficient to simply update the task's cpu field. | |
1852 | */ | |
dd41f596 | 1853 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1854 | set_task_cpu(p, dest_cpu); |
1855 | return 0; | |
1856 | } | |
1857 | ||
1858 | init_completion(&req->done); | |
1da177e4 LT |
1859 | req->task = p; |
1860 | req->dest_cpu = dest_cpu; | |
1861 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1862 | |
1da177e4 LT |
1863 | return 1; |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * wait_task_inactive - wait for a thread to unschedule. | |
1868 | * | |
85ba2d86 RM |
1869 | * If @match_state is nonzero, it's the @p->state value just checked and |
1870 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1871 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1872 | * we return a positive number (its total switch count). If a second call | |
1873 | * a short while later returns the same number, the caller can be sure that | |
1874 | * @p has remained unscheduled the whole time. | |
1875 | * | |
1da177e4 LT |
1876 | * The caller must ensure that the task *will* unschedule sometime soon, |
1877 | * else this function might spin for a *long* time. This function can't | |
1878 | * be called with interrupts off, or it may introduce deadlock with | |
1879 | * smp_call_function() if an IPI is sent by the same process we are | |
1880 | * waiting to become inactive. | |
1881 | */ | |
85ba2d86 | 1882 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1883 | { |
1884 | unsigned long flags; | |
dd41f596 | 1885 | int running, on_rq; |
85ba2d86 | 1886 | unsigned long ncsw; |
70b97a7f | 1887 | struct rq *rq; |
1da177e4 | 1888 | |
3a5c359a AK |
1889 | for (;;) { |
1890 | /* | |
1891 | * We do the initial early heuristics without holding | |
1892 | * any task-queue locks at all. We'll only try to get | |
1893 | * the runqueue lock when things look like they will | |
1894 | * work out! | |
1895 | */ | |
1896 | rq = task_rq(p); | |
fa490cfd | 1897 | |
3a5c359a AK |
1898 | /* |
1899 | * If the task is actively running on another CPU | |
1900 | * still, just relax and busy-wait without holding | |
1901 | * any locks. | |
1902 | * | |
1903 | * NOTE! Since we don't hold any locks, it's not | |
1904 | * even sure that "rq" stays as the right runqueue! | |
1905 | * But we don't care, since "task_running()" will | |
1906 | * return false if the runqueue has changed and p | |
1907 | * is actually now running somewhere else! | |
1908 | */ | |
85ba2d86 RM |
1909 | while (task_running(rq, p)) { |
1910 | if (match_state && unlikely(p->state != match_state)) | |
1911 | return 0; | |
3a5c359a | 1912 | cpu_relax(); |
85ba2d86 | 1913 | } |
fa490cfd | 1914 | |
3a5c359a AK |
1915 | /* |
1916 | * Ok, time to look more closely! We need the rq | |
1917 | * lock now, to be *sure*. If we're wrong, we'll | |
1918 | * just go back and repeat. | |
1919 | */ | |
1920 | rq = task_rq_lock(p, &flags); | |
1921 | running = task_running(rq, p); | |
1922 | on_rq = p->se.on_rq; | |
85ba2d86 RM |
1923 | ncsw = 0; |
1924 | if (!match_state || p->state == match_state) { | |
1925 | ncsw = p->nivcsw + p->nvcsw; | |
1926 | if (unlikely(!ncsw)) | |
1927 | ncsw = 1; | |
1928 | } | |
3a5c359a | 1929 | task_rq_unlock(rq, &flags); |
fa490cfd | 1930 | |
85ba2d86 RM |
1931 | /* |
1932 | * If it changed from the expected state, bail out now. | |
1933 | */ | |
1934 | if (unlikely(!ncsw)) | |
1935 | break; | |
1936 | ||
3a5c359a AK |
1937 | /* |
1938 | * Was it really running after all now that we | |
1939 | * checked with the proper locks actually held? | |
1940 | * | |
1941 | * Oops. Go back and try again.. | |
1942 | */ | |
1943 | if (unlikely(running)) { | |
1944 | cpu_relax(); | |
1945 | continue; | |
1946 | } | |
fa490cfd | 1947 | |
3a5c359a AK |
1948 | /* |
1949 | * It's not enough that it's not actively running, | |
1950 | * it must be off the runqueue _entirely_, and not | |
1951 | * preempted! | |
1952 | * | |
1953 | * So if it wa still runnable (but just not actively | |
1954 | * running right now), it's preempted, and we should | |
1955 | * yield - it could be a while. | |
1956 | */ | |
1957 | if (unlikely(on_rq)) { | |
1958 | schedule_timeout_uninterruptible(1); | |
1959 | continue; | |
1960 | } | |
fa490cfd | 1961 | |
3a5c359a AK |
1962 | /* |
1963 | * Ahh, all good. It wasn't running, and it wasn't | |
1964 | * runnable, which means that it will never become | |
1965 | * running in the future either. We're all done! | |
1966 | */ | |
1967 | break; | |
1968 | } | |
85ba2d86 RM |
1969 | |
1970 | return ncsw; | |
1da177e4 LT |
1971 | } |
1972 | ||
1973 | /*** | |
1974 | * kick_process - kick a running thread to enter/exit the kernel | |
1975 | * @p: the to-be-kicked thread | |
1976 | * | |
1977 | * Cause a process which is running on another CPU to enter | |
1978 | * kernel-mode, without any delay. (to get signals handled.) | |
1979 | * | |
1980 | * NOTE: this function doesnt have to take the runqueue lock, | |
1981 | * because all it wants to ensure is that the remote task enters | |
1982 | * the kernel. If the IPI races and the task has been migrated | |
1983 | * to another CPU then no harm is done and the purpose has been | |
1984 | * achieved as well. | |
1985 | */ | |
36c8b586 | 1986 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1987 | { |
1988 | int cpu; | |
1989 | ||
1990 | preempt_disable(); | |
1991 | cpu = task_cpu(p); | |
1992 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1993 | smp_send_reschedule(cpu); | |
1994 | preempt_enable(); | |
1995 | } | |
1996 | ||
1997 | /* | |
2dd73a4f PW |
1998 | * Return a low guess at the load of a migration-source cpu weighted |
1999 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2000 | * |
2001 | * We want to under-estimate the load of migration sources, to | |
2002 | * balance conservatively. | |
2003 | */ | |
a9957449 | 2004 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2005 | { |
70b97a7f | 2006 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2007 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2008 | |
93b75217 | 2009 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2010 | return total; |
b910472d | 2011 | |
dd41f596 | 2012 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2013 | } |
2014 | ||
2015 | /* | |
2dd73a4f PW |
2016 | * Return a high guess at the load of a migration-target cpu weighted |
2017 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2018 | */ |
a9957449 | 2019 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2020 | { |
70b97a7f | 2021 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2022 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2023 | |
93b75217 | 2024 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2025 | return total; |
3b0bd9bc | 2026 | |
dd41f596 | 2027 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2028 | } |
2029 | ||
147cbb4b NP |
2030 | /* |
2031 | * find_idlest_group finds and returns the least busy CPU group within the | |
2032 | * domain. | |
2033 | */ | |
2034 | static struct sched_group * | |
2035 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2036 | { | |
2037 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2038 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2039 | int load_idx = sd->forkexec_idx; | |
2040 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2041 | ||
2042 | do { | |
2043 | unsigned long load, avg_load; | |
2044 | int local_group; | |
2045 | int i; | |
2046 | ||
da5a5522 BD |
2047 | /* Skip over this group if it has no CPUs allowed */ |
2048 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2049 | continue; |
da5a5522 | 2050 | |
147cbb4b | 2051 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2052 | |
2053 | /* Tally up the load of all CPUs in the group */ | |
2054 | avg_load = 0; | |
2055 | ||
363ab6f1 | 2056 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2057 | /* Bias balancing toward cpus of our domain */ |
2058 | if (local_group) | |
2059 | load = source_load(i, load_idx); | |
2060 | else | |
2061 | load = target_load(i, load_idx); | |
2062 | ||
2063 | avg_load += load; | |
2064 | } | |
2065 | ||
2066 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2067 | avg_load = sg_div_cpu_power(group, |
2068 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2069 | |
2070 | if (local_group) { | |
2071 | this_load = avg_load; | |
2072 | this = group; | |
2073 | } else if (avg_load < min_load) { | |
2074 | min_load = avg_load; | |
2075 | idlest = group; | |
2076 | } | |
3a5c359a | 2077 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2078 | |
2079 | if (!idlest || 100*this_load < imbalance*min_load) | |
2080 | return NULL; | |
2081 | return idlest; | |
2082 | } | |
2083 | ||
2084 | /* | |
0feaece9 | 2085 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2086 | */ |
95cdf3b7 | 2087 | static int |
7c16ec58 MT |
2088 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2089 | cpumask_t *tmp) | |
147cbb4b NP |
2090 | { |
2091 | unsigned long load, min_load = ULONG_MAX; | |
2092 | int idlest = -1; | |
2093 | int i; | |
2094 | ||
da5a5522 | 2095 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2096 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2097 | |
363ab6f1 | 2098 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2099 | load = weighted_cpuload(i); |
147cbb4b NP |
2100 | |
2101 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2102 | min_load = load; | |
2103 | idlest = i; | |
2104 | } | |
2105 | } | |
2106 | ||
2107 | return idlest; | |
2108 | } | |
2109 | ||
476d139c NP |
2110 | /* |
2111 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2112 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2113 | * SD_BALANCE_EXEC. | |
2114 | * | |
2115 | * Balance, ie. select the least loaded group. | |
2116 | * | |
2117 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2118 | * | |
2119 | * preempt must be disabled. | |
2120 | */ | |
2121 | static int sched_balance_self(int cpu, int flag) | |
2122 | { | |
2123 | struct task_struct *t = current; | |
2124 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2125 | |
c96d145e | 2126 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2127 | /* |
2128 | * If power savings logic is enabled for a domain, stop there. | |
2129 | */ | |
5c45bf27 SS |
2130 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2131 | break; | |
476d139c NP |
2132 | if (tmp->flags & flag) |
2133 | sd = tmp; | |
c96d145e | 2134 | } |
476d139c | 2135 | |
039a1c41 PZ |
2136 | if (sd) |
2137 | update_shares(sd); | |
2138 | ||
476d139c | 2139 | while (sd) { |
7c16ec58 | 2140 | cpumask_t span, tmpmask; |
476d139c | 2141 | struct sched_group *group; |
1a848870 SS |
2142 | int new_cpu, weight; |
2143 | ||
2144 | if (!(sd->flags & flag)) { | |
2145 | sd = sd->child; | |
2146 | continue; | |
2147 | } | |
476d139c NP |
2148 | |
2149 | span = sd->span; | |
2150 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2151 | if (!group) { |
2152 | sd = sd->child; | |
2153 | continue; | |
2154 | } | |
476d139c | 2155 | |
7c16ec58 | 2156 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2157 | if (new_cpu == -1 || new_cpu == cpu) { |
2158 | /* Now try balancing at a lower domain level of cpu */ | |
2159 | sd = sd->child; | |
2160 | continue; | |
2161 | } | |
476d139c | 2162 | |
1a848870 | 2163 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2164 | cpu = new_cpu; |
476d139c NP |
2165 | sd = NULL; |
2166 | weight = cpus_weight(span); | |
2167 | for_each_domain(cpu, tmp) { | |
2168 | if (weight <= cpus_weight(tmp->span)) | |
2169 | break; | |
2170 | if (tmp->flags & flag) | |
2171 | sd = tmp; | |
2172 | } | |
2173 | /* while loop will break here if sd == NULL */ | |
2174 | } | |
2175 | ||
2176 | return cpu; | |
2177 | } | |
2178 | ||
2179 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2180 | |
1da177e4 LT |
2181 | /*** |
2182 | * try_to_wake_up - wake up a thread | |
2183 | * @p: the to-be-woken-up thread | |
2184 | * @state: the mask of task states that can be woken | |
2185 | * @sync: do a synchronous wakeup? | |
2186 | * | |
2187 | * Put it on the run-queue if it's not already there. The "current" | |
2188 | * thread is always on the run-queue (except when the actual | |
2189 | * re-schedule is in progress), and as such you're allowed to do | |
2190 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2191 | * runnable without the overhead of this. | |
2192 | * | |
2193 | * returns failure only if the task is already active. | |
2194 | */ | |
36c8b586 | 2195 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2196 | { |
cc367732 | 2197 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2198 | unsigned long flags; |
2199 | long old_state; | |
70b97a7f | 2200 | struct rq *rq; |
1da177e4 | 2201 | |
b85d0667 IM |
2202 | if (!sched_feat(SYNC_WAKEUPS)) |
2203 | sync = 0; | |
2204 | ||
2398f2c6 PZ |
2205 | #ifdef CONFIG_SMP |
2206 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2207 | struct sched_domain *sd; | |
2208 | ||
2209 | this_cpu = raw_smp_processor_id(); | |
2210 | cpu = task_cpu(p); | |
2211 | ||
2212 | for_each_domain(this_cpu, sd) { | |
2213 | if (cpu_isset(cpu, sd->span)) { | |
2214 | update_shares(sd); | |
2215 | break; | |
2216 | } | |
2217 | } | |
2218 | } | |
2219 | #endif | |
2220 | ||
04e2f174 | 2221 | smp_wmb(); |
1da177e4 LT |
2222 | rq = task_rq_lock(p, &flags); |
2223 | old_state = p->state; | |
2224 | if (!(old_state & state)) | |
2225 | goto out; | |
2226 | ||
dd41f596 | 2227 | if (p->se.on_rq) |
1da177e4 LT |
2228 | goto out_running; |
2229 | ||
2230 | cpu = task_cpu(p); | |
cc367732 | 2231 | orig_cpu = cpu; |
1da177e4 LT |
2232 | this_cpu = smp_processor_id(); |
2233 | ||
2234 | #ifdef CONFIG_SMP | |
2235 | if (unlikely(task_running(rq, p))) | |
2236 | goto out_activate; | |
2237 | ||
5d2f5a61 DA |
2238 | cpu = p->sched_class->select_task_rq(p, sync); |
2239 | if (cpu != orig_cpu) { | |
2240 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2241 | task_rq_unlock(rq, &flags); |
2242 | /* might preempt at this point */ | |
2243 | rq = task_rq_lock(p, &flags); | |
2244 | old_state = p->state; | |
2245 | if (!(old_state & state)) | |
2246 | goto out; | |
dd41f596 | 2247 | if (p->se.on_rq) |
1da177e4 LT |
2248 | goto out_running; |
2249 | ||
2250 | this_cpu = smp_processor_id(); | |
2251 | cpu = task_cpu(p); | |
2252 | } | |
2253 | ||
e7693a36 GH |
2254 | #ifdef CONFIG_SCHEDSTATS |
2255 | schedstat_inc(rq, ttwu_count); | |
2256 | if (cpu == this_cpu) | |
2257 | schedstat_inc(rq, ttwu_local); | |
2258 | else { | |
2259 | struct sched_domain *sd; | |
2260 | for_each_domain(this_cpu, sd) { | |
2261 | if (cpu_isset(cpu, sd->span)) { | |
2262 | schedstat_inc(sd, ttwu_wake_remote); | |
2263 | break; | |
2264 | } | |
2265 | } | |
2266 | } | |
6d6bc0ad | 2267 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2268 | |
1da177e4 LT |
2269 | out_activate: |
2270 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2271 | schedstat_inc(p, se.nr_wakeups); |
2272 | if (sync) | |
2273 | schedstat_inc(p, se.nr_wakeups_sync); | |
2274 | if (orig_cpu != cpu) | |
2275 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2276 | if (cpu == this_cpu) | |
2277 | schedstat_inc(p, se.nr_wakeups_local); | |
2278 | else | |
2279 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2280 | update_rq_clock(rq); |
dd41f596 | 2281 | activate_task(rq, p, 1); |
1da177e4 LT |
2282 | success = 1; |
2283 | ||
2284 | out_running: | |
5b82a1b0 MD |
2285 | trace_mark(kernel_sched_wakeup, |
2286 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2287 | p->pid, p->state, rq, p, rq->curr); | |
4ae7d5ce IM |
2288 | check_preempt_curr(rq, p); |
2289 | ||
1da177e4 | 2290 | p->state = TASK_RUNNING; |
9a897c5a SR |
2291 | #ifdef CONFIG_SMP |
2292 | if (p->sched_class->task_wake_up) | |
2293 | p->sched_class->task_wake_up(rq, p); | |
2294 | #endif | |
1da177e4 | 2295 | out: |
2087a1ad GH |
2296 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2297 | ||
1da177e4 LT |
2298 | task_rq_unlock(rq, &flags); |
2299 | ||
2300 | return success; | |
2301 | } | |
2302 | ||
7ad5b3a5 | 2303 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2304 | { |
d9514f6c | 2305 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2306 | } |
1da177e4 LT |
2307 | EXPORT_SYMBOL(wake_up_process); |
2308 | ||
7ad5b3a5 | 2309 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2310 | { |
2311 | return try_to_wake_up(p, state, 0); | |
2312 | } | |
2313 | ||
1da177e4 LT |
2314 | /* |
2315 | * Perform scheduler related setup for a newly forked process p. | |
2316 | * p is forked by current. | |
dd41f596 IM |
2317 | * |
2318 | * __sched_fork() is basic setup used by init_idle() too: | |
2319 | */ | |
2320 | static void __sched_fork(struct task_struct *p) | |
2321 | { | |
dd41f596 IM |
2322 | p->se.exec_start = 0; |
2323 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2324 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2325 | p->se.last_wakeup = 0; |
2326 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2327 | |
2328 | #ifdef CONFIG_SCHEDSTATS | |
2329 | p->se.wait_start = 0; | |
dd41f596 IM |
2330 | p->se.sum_sleep_runtime = 0; |
2331 | p->se.sleep_start = 0; | |
dd41f596 IM |
2332 | p->se.block_start = 0; |
2333 | p->se.sleep_max = 0; | |
2334 | p->se.block_max = 0; | |
2335 | p->se.exec_max = 0; | |
eba1ed4b | 2336 | p->se.slice_max = 0; |
dd41f596 | 2337 | p->se.wait_max = 0; |
6cfb0d5d | 2338 | #endif |
476d139c | 2339 | |
fa717060 | 2340 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2341 | p->se.on_rq = 0; |
4a55bd5e | 2342 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2343 | |
e107be36 AK |
2344 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2345 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2346 | #endif | |
2347 | ||
1da177e4 LT |
2348 | /* |
2349 | * We mark the process as running here, but have not actually | |
2350 | * inserted it onto the runqueue yet. This guarantees that | |
2351 | * nobody will actually run it, and a signal or other external | |
2352 | * event cannot wake it up and insert it on the runqueue either. | |
2353 | */ | |
2354 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2355 | } |
2356 | ||
2357 | /* | |
2358 | * fork()/clone()-time setup: | |
2359 | */ | |
2360 | void sched_fork(struct task_struct *p, int clone_flags) | |
2361 | { | |
2362 | int cpu = get_cpu(); | |
2363 | ||
2364 | __sched_fork(p); | |
2365 | ||
2366 | #ifdef CONFIG_SMP | |
2367 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2368 | #endif | |
02e4bac2 | 2369 | set_task_cpu(p, cpu); |
b29739f9 IM |
2370 | |
2371 | /* | |
2372 | * Make sure we do not leak PI boosting priority to the child: | |
2373 | */ | |
2374 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2375 | if (!rt_prio(p->prio)) |
2376 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2377 | |
52f17b6c | 2378 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2379 | if (likely(sched_info_on())) |
52f17b6c | 2380 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2381 | #endif |
d6077cb8 | 2382 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2383 | p->oncpu = 0; |
2384 | #endif | |
1da177e4 | 2385 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2386 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2387 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2388 | #endif |
476d139c | 2389 | put_cpu(); |
1da177e4 LT |
2390 | } |
2391 | ||
2392 | /* | |
2393 | * wake_up_new_task - wake up a newly created task for the first time. | |
2394 | * | |
2395 | * This function will do some initial scheduler statistics housekeeping | |
2396 | * that must be done for every newly created context, then puts the task | |
2397 | * on the runqueue and wakes it. | |
2398 | */ | |
7ad5b3a5 | 2399 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2400 | { |
2401 | unsigned long flags; | |
dd41f596 | 2402 | struct rq *rq; |
1da177e4 LT |
2403 | |
2404 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2405 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2406 | update_rq_clock(rq); |
1da177e4 LT |
2407 | |
2408 | p->prio = effective_prio(p); | |
2409 | ||
b9dca1e0 | 2410 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2411 | activate_task(rq, p, 0); |
1da177e4 | 2412 | } else { |
1da177e4 | 2413 | /* |
dd41f596 IM |
2414 | * Let the scheduling class do new task startup |
2415 | * management (if any): | |
1da177e4 | 2416 | */ |
ee0827d8 | 2417 | p->sched_class->task_new(rq, p); |
c09595f6 | 2418 | inc_nr_running(rq); |
1da177e4 | 2419 | } |
5b82a1b0 MD |
2420 | trace_mark(kernel_sched_wakeup_new, |
2421 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2422 | p->pid, p->state, rq, p, rq->curr); | |
dd41f596 | 2423 | check_preempt_curr(rq, p); |
9a897c5a SR |
2424 | #ifdef CONFIG_SMP |
2425 | if (p->sched_class->task_wake_up) | |
2426 | p->sched_class->task_wake_up(rq, p); | |
2427 | #endif | |
dd41f596 | 2428 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2429 | } |
2430 | ||
e107be36 AK |
2431 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2432 | ||
2433 | /** | |
421cee29 RD |
2434 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2435 | * @notifier: notifier struct to register | |
e107be36 AK |
2436 | */ |
2437 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2438 | { | |
2439 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2440 | } | |
2441 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2442 | ||
2443 | /** | |
2444 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2445 | * @notifier: notifier struct to unregister |
e107be36 AK |
2446 | * |
2447 | * This is safe to call from within a preemption notifier. | |
2448 | */ | |
2449 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2450 | { | |
2451 | hlist_del(¬ifier->link); | |
2452 | } | |
2453 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2454 | ||
2455 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2456 | { | |
2457 | struct preempt_notifier *notifier; | |
2458 | struct hlist_node *node; | |
2459 | ||
2460 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2461 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2462 | } | |
2463 | ||
2464 | static void | |
2465 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2466 | struct task_struct *next) | |
2467 | { | |
2468 | struct preempt_notifier *notifier; | |
2469 | struct hlist_node *node; | |
2470 | ||
2471 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2472 | notifier->ops->sched_out(notifier, next); | |
2473 | } | |
2474 | ||
6d6bc0ad | 2475 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2476 | |
2477 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2478 | { | |
2479 | } | |
2480 | ||
2481 | static void | |
2482 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2483 | struct task_struct *next) | |
2484 | { | |
2485 | } | |
2486 | ||
6d6bc0ad | 2487 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2488 | |
4866cde0 NP |
2489 | /** |
2490 | * prepare_task_switch - prepare to switch tasks | |
2491 | * @rq: the runqueue preparing to switch | |
421cee29 | 2492 | * @prev: the current task that is being switched out |
4866cde0 NP |
2493 | * @next: the task we are going to switch to. |
2494 | * | |
2495 | * This is called with the rq lock held and interrupts off. It must | |
2496 | * be paired with a subsequent finish_task_switch after the context | |
2497 | * switch. | |
2498 | * | |
2499 | * prepare_task_switch sets up locking and calls architecture specific | |
2500 | * hooks. | |
2501 | */ | |
e107be36 AK |
2502 | static inline void |
2503 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2504 | struct task_struct *next) | |
4866cde0 | 2505 | { |
e107be36 | 2506 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2507 | prepare_lock_switch(rq, next); |
2508 | prepare_arch_switch(next); | |
2509 | } | |
2510 | ||
1da177e4 LT |
2511 | /** |
2512 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2513 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2514 | * @prev: the thread we just switched away from. |
2515 | * | |
4866cde0 NP |
2516 | * finish_task_switch must be called after the context switch, paired |
2517 | * with a prepare_task_switch call before the context switch. | |
2518 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2519 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2520 | * |
2521 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2522 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2523 | * with the lock held can cause deadlocks; see schedule() for |
2524 | * details.) | |
2525 | */ | |
a9957449 | 2526 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2527 | __releases(rq->lock) |
2528 | { | |
1da177e4 | 2529 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2530 | long prev_state; |
1da177e4 LT |
2531 | |
2532 | rq->prev_mm = NULL; | |
2533 | ||
2534 | /* | |
2535 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2536 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2537 | * schedule one last time. The schedule call will never return, and |
2538 | * the scheduled task must drop that reference. | |
c394cc9f | 2539 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2540 | * still held, otherwise prev could be scheduled on another cpu, die |
2541 | * there before we look at prev->state, and then the reference would | |
2542 | * be dropped twice. | |
2543 | * Manfred Spraul <manfred@colorfullife.com> | |
2544 | */ | |
55a101f8 | 2545 | prev_state = prev->state; |
4866cde0 NP |
2546 | finish_arch_switch(prev); |
2547 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2548 | #ifdef CONFIG_SMP |
2549 | if (current->sched_class->post_schedule) | |
2550 | current->sched_class->post_schedule(rq); | |
2551 | #endif | |
e8fa1362 | 2552 | |
e107be36 | 2553 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2554 | if (mm) |
2555 | mmdrop(mm); | |
c394cc9f | 2556 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2557 | /* |
2558 | * Remove function-return probe instances associated with this | |
2559 | * task and put them back on the free list. | |
9761eea8 | 2560 | */ |
c6fd91f0 | 2561 | kprobe_flush_task(prev); |
1da177e4 | 2562 | put_task_struct(prev); |
c6fd91f0 | 2563 | } |
1da177e4 LT |
2564 | } |
2565 | ||
2566 | /** | |
2567 | * schedule_tail - first thing a freshly forked thread must call. | |
2568 | * @prev: the thread we just switched away from. | |
2569 | */ | |
36c8b586 | 2570 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2571 | __releases(rq->lock) |
2572 | { | |
70b97a7f IM |
2573 | struct rq *rq = this_rq(); |
2574 | ||
4866cde0 NP |
2575 | finish_task_switch(rq, prev); |
2576 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2577 | /* In this case, finish_task_switch does not reenable preemption */ | |
2578 | preempt_enable(); | |
2579 | #endif | |
1da177e4 | 2580 | if (current->set_child_tid) |
b488893a | 2581 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2582 | } |
2583 | ||
2584 | /* | |
2585 | * context_switch - switch to the new MM and the new | |
2586 | * thread's register state. | |
2587 | */ | |
dd41f596 | 2588 | static inline void |
70b97a7f | 2589 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2590 | struct task_struct *next) |
1da177e4 | 2591 | { |
dd41f596 | 2592 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2593 | |
e107be36 | 2594 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2595 | trace_mark(kernel_sched_schedule, |
2596 | "prev_pid %d next_pid %d prev_state %ld " | |
2597 | "## rq %p prev %p next %p", | |
2598 | prev->pid, next->pid, prev->state, | |
2599 | rq, prev, next); | |
dd41f596 IM |
2600 | mm = next->mm; |
2601 | oldmm = prev->active_mm; | |
9226d125 ZA |
2602 | /* |
2603 | * For paravirt, this is coupled with an exit in switch_to to | |
2604 | * combine the page table reload and the switch backend into | |
2605 | * one hypercall. | |
2606 | */ | |
2607 | arch_enter_lazy_cpu_mode(); | |
2608 | ||
dd41f596 | 2609 | if (unlikely(!mm)) { |
1da177e4 LT |
2610 | next->active_mm = oldmm; |
2611 | atomic_inc(&oldmm->mm_count); | |
2612 | enter_lazy_tlb(oldmm, next); | |
2613 | } else | |
2614 | switch_mm(oldmm, mm, next); | |
2615 | ||
dd41f596 | 2616 | if (unlikely(!prev->mm)) { |
1da177e4 | 2617 | prev->active_mm = NULL; |
1da177e4 LT |
2618 | rq->prev_mm = oldmm; |
2619 | } | |
3a5f5e48 IM |
2620 | /* |
2621 | * Since the runqueue lock will be released by the next | |
2622 | * task (which is an invalid locking op but in the case | |
2623 | * of the scheduler it's an obvious special-case), so we | |
2624 | * do an early lockdep release here: | |
2625 | */ | |
2626 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2627 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2628 | #endif |
1da177e4 LT |
2629 | |
2630 | /* Here we just switch the register state and the stack. */ | |
2631 | switch_to(prev, next, prev); | |
2632 | ||
dd41f596 IM |
2633 | barrier(); |
2634 | /* | |
2635 | * this_rq must be evaluated again because prev may have moved | |
2636 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2637 | * frame will be invalid. | |
2638 | */ | |
2639 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2640 | } |
2641 | ||
2642 | /* | |
2643 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2644 | * | |
2645 | * externally visible scheduler statistics: current number of runnable | |
2646 | * threads, current number of uninterruptible-sleeping threads, total | |
2647 | * number of context switches performed since bootup. | |
2648 | */ | |
2649 | unsigned long nr_running(void) | |
2650 | { | |
2651 | unsigned long i, sum = 0; | |
2652 | ||
2653 | for_each_online_cpu(i) | |
2654 | sum += cpu_rq(i)->nr_running; | |
2655 | ||
2656 | return sum; | |
2657 | } | |
2658 | ||
2659 | unsigned long nr_uninterruptible(void) | |
2660 | { | |
2661 | unsigned long i, sum = 0; | |
2662 | ||
0a945022 | 2663 | for_each_possible_cpu(i) |
1da177e4 LT |
2664 | sum += cpu_rq(i)->nr_uninterruptible; |
2665 | ||
2666 | /* | |
2667 | * Since we read the counters lockless, it might be slightly | |
2668 | * inaccurate. Do not allow it to go below zero though: | |
2669 | */ | |
2670 | if (unlikely((long)sum < 0)) | |
2671 | sum = 0; | |
2672 | ||
2673 | return sum; | |
2674 | } | |
2675 | ||
2676 | unsigned long long nr_context_switches(void) | |
2677 | { | |
cc94abfc SR |
2678 | int i; |
2679 | unsigned long long sum = 0; | |
1da177e4 | 2680 | |
0a945022 | 2681 | for_each_possible_cpu(i) |
1da177e4 LT |
2682 | sum += cpu_rq(i)->nr_switches; |
2683 | ||
2684 | return sum; | |
2685 | } | |
2686 | ||
2687 | unsigned long nr_iowait(void) | |
2688 | { | |
2689 | unsigned long i, sum = 0; | |
2690 | ||
0a945022 | 2691 | for_each_possible_cpu(i) |
1da177e4 LT |
2692 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2693 | ||
2694 | return sum; | |
2695 | } | |
2696 | ||
db1b1fef JS |
2697 | unsigned long nr_active(void) |
2698 | { | |
2699 | unsigned long i, running = 0, uninterruptible = 0; | |
2700 | ||
2701 | for_each_online_cpu(i) { | |
2702 | running += cpu_rq(i)->nr_running; | |
2703 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2704 | } | |
2705 | ||
2706 | if (unlikely((long)uninterruptible < 0)) | |
2707 | uninterruptible = 0; | |
2708 | ||
2709 | return running + uninterruptible; | |
2710 | } | |
2711 | ||
48f24c4d | 2712 | /* |
dd41f596 IM |
2713 | * Update rq->cpu_load[] statistics. This function is usually called every |
2714 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2715 | */ |
dd41f596 | 2716 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2717 | { |
495eca49 | 2718 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2719 | int i, scale; |
2720 | ||
2721 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2722 | |
2723 | /* Update our load: */ | |
2724 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2725 | unsigned long old_load, new_load; | |
2726 | ||
2727 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2728 | ||
2729 | old_load = this_rq->cpu_load[i]; | |
2730 | new_load = this_load; | |
a25707f3 IM |
2731 | /* |
2732 | * Round up the averaging division if load is increasing. This | |
2733 | * prevents us from getting stuck on 9 if the load is 10, for | |
2734 | * example. | |
2735 | */ | |
2736 | if (new_load > old_load) | |
2737 | new_load += scale-1; | |
dd41f596 IM |
2738 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2739 | } | |
48f24c4d IM |
2740 | } |
2741 | ||
dd41f596 IM |
2742 | #ifdef CONFIG_SMP |
2743 | ||
1da177e4 LT |
2744 | /* |
2745 | * double_rq_lock - safely lock two runqueues | |
2746 | * | |
2747 | * Note this does not disable interrupts like task_rq_lock, | |
2748 | * you need to do so manually before calling. | |
2749 | */ | |
70b97a7f | 2750 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2751 | __acquires(rq1->lock) |
2752 | __acquires(rq2->lock) | |
2753 | { | |
054b9108 | 2754 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2755 | if (rq1 == rq2) { |
2756 | spin_lock(&rq1->lock); | |
2757 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2758 | } else { | |
c96d145e | 2759 | if (rq1 < rq2) { |
1da177e4 | 2760 | spin_lock(&rq1->lock); |
5e710e37 | 2761 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2762 | } else { |
2763 | spin_lock(&rq2->lock); | |
5e710e37 | 2764 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2765 | } |
2766 | } | |
6e82a3be IM |
2767 | update_rq_clock(rq1); |
2768 | update_rq_clock(rq2); | |
1da177e4 LT |
2769 | } |
2770 | ||
2771 | /* | |
2772 | * double_rq_unlock - safely unlock two runqueues | |
2773 | * | |
2774 | * Note this does not restore interrupts like task_rq_unlock, | |
2775 | * you need to do so manually after calling. | |
2776 | */ | |
70b97a7f | 2777 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2778 | __releases(rq1->lock) |
2779 | __releases(rq2->lock) | |
2780 | { | |
2781 | spin_unlock(&rq1->lock); | |
2782 | if (rq1 != rq2) | |
2783 | spin_unlock(&rq2->lock); | |
2784 | else | |
2785 | __release(rq2->lock); | |
2786 | } | |
2787 | ||
2788 | /* | |
2789 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2790 | */ | |
e8fa1362 | 2791 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2792 | __releases(this_rq->lock) |
2793 | __acquires(busiest->lock) | |
2794 | __acquires(this_rq->lock) | |
2795 | { | |
e8fa1362 SR |
2796 | int ret = 0; |
2797 | ||
054b9108 KK |
2798 | if (unlikely(!irqs_disabled())) { |
2799 | /* printk() doesn't work good under rq->lock */ | |
2800 | spin_unlock(&this_rq->lock); | |
2801 | BUG_ON(1); | |
2802 | } | |
1da177e4 | 2803 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2804 | if (busiest < this_rq) { |
1da177e4 LT |
2805 | spin_unlock(&this_rq->lock); |
2806 | spin_lock(&busiest->lock); | |
5e710e37 | 2807 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2808 | ret = 1; |
1da177e4 | 2809 | } else |
5e710e37 | 2810 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2811 | } |
e8fa1362 | 2812 | return ret; |
1da177e4 LT |
2813 | } |
2814 | ||
1b12bbc7 PZ |
2815 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2816 | __releases(busiest->lock) | |
2817 | { | |
2818 | spin_unlock(&busiest->lock); | |
2819 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2820 | } | |
2821 | ||
1da177e4 LT |
2822 | /* |
2823 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2824 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2825 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2826 | * the cpu_allowed mask is restored. |
2827 | */ | |
36c8b586 | 2828 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2829 | { |
70b97a7f | 2830 | struct migration_req req; |
1da177e4 | 2831 | unsigned long flags; |
70b97a7f | 2832 | struct rq *rq; |
1da177e4 LT |
2833 | |
2834 | rq = task_rq_lock(p, &flags); | |
2835 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2836 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2837 | goto out; |
2838 | ||
2839 | /* force the process onto the specified CPU */ | |
2840 | if (migrate_task(p, dest_cpu, &req)) { | |
2841 | /* Need to wait for migration thread (might exit: take ref). */ | |
2842 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2843 | |
1da177e4 LT |
2844 | get_task_struct(mt); |
2845 | task_rq_unlock(rq, &flags); | |
2846 | wake_up_process(mt); | |
2847 | put_task_struct(mt); | |
2848 | wait_for_completion(&req.done); | |
36c8b586 | 2849 | |
1da177e4 LT |
2850 | return; |
2851 | } | |
2852 | out: | |
2853 | task_rq_unlock(rq, &flags); | |
2854 | } | |
2855 | ||
2856 | /* | |
476d139c NP |
2857 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2858 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2859 | */ |
2860 | void sched_exec(void) | |
2861 | { | |
1da177e4 | 2862 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2863 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2864 | put_cpu(); |
476d139c NP |
2865 | if (new_cpu != this_cpu) |
2866 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2867 | } |
2868 | ||
2869 | /* | |
2870 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2871 | * Both runqueues must be locked. | |
2872 | */ | |
dd41f596 IM |
2873 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2874 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2875 | { |
2e1cb74a | 2876 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2877 | set_task_cpu(p, this_cpu); |
dd41f596 | 2878 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2879 | /* |
2880 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2881 | * to be always true for them. | |
2882 | */ | |
dd41f596 | 2883 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2884 | } |
2885 | ||
2886 | /* | |
2887 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2888 | */ | |
858119e1 | 2889 | static |
70b97a7f | 2890 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2891 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2892 | int *all_pinned) |
1da177e4 LT |
2893 | { |
2894 | /* | |
2895 | * We do not migrate tasks that are: | |
2896 | * 1) running (obviously), or | |
2897 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2898 | * 3) are cache-hot on their current CPU. | |
2899 | */ | |
cc367732 IM |
2900 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2901 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2902 | return 0; |
cc367732 | 2903 | } |
81026794 NP |
2904 | *all_pinned = 0; |
2905 | ||
cc367732 IM |
2906 | if (task_running(rq, p)) { |
2907 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2908 | return 0; |
cc367732 | 2909 | } |
1da177e4 | 2910 | |
da84d961 IM |
2911 | /* |
2912 | * Aggressive migration if: | |
2913 | * 1) task is cache cold, or | |
2914 | * 2) too many balance attempts have failed. | |
2915 | */ | |
2916 | ||
6bc1665b IM |
2917 | if (!task_hot(p, rq->clock, sd) || |
2918 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2919 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2920 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2921 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2922 | schedstat_inc(p, se.nr_forced_migrations); |
2923 | } | |
da84d961 IM |
2924 | #endif |
2925 | return 1; | |
2926 | } | |
2927 | ||
cc367732 IM |
2928 | if (task_hot(p, rq->clock, sd)) { |
2929 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2930 | return 0; |
cc367732 | 2931 | } |
1da177e4 LT |
2932 | return 1; |
2933 | } | |
2934 | ||
e1d1484f PW |
2935 | static unsigned long |
2936 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2937 | unsigned long max_load_move, struct sched_domain *sd, | |
2938 | enum cpu_idle_type idle, int *all_pinned, | |
2939 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2940 | { |
051c6764 | 2941 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2942 | struct task_struct *p; |
2943 | long rem_load_move = max_load_move; | |
1da177e4 | 2944 | |
e1d1484f | 2945 | if (max_load_move == 0) |
1da177e4 LT |
2946 | goto out; |
2947 | ||
81026794 NP |
2948 | pinned = 1; |
2949 | ||
1da177e4 | 2950 | /* |
dd41f596 | 2951 | * Start the load-balancing iterator: |
1da177e4 | 2952 | */ |
dd41f596 IM |
2953 | p = iterator->start(iterator->arg); |
2954 | next: | |
b82d9fdd | 2955 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2956 | goto out; |
051c6764 PZ |
2957 | |
2958 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2959 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2960 | p = iterator->next(iterator->arg); |
2961 | goto next; | |
1da177e4 LT |
2962 | } |
2963 | ||
dd41f596 | 2964 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2965 | pulled++; |
dd41f596 | 2966 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2967 | |
2dd73a4f | 2968 | /* |
b82d9fdd | 2969 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2970 | */ |
e1d1484f | 2971 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2972 | if (p->prio < *this_best_prio) |
2973 | *this_best_prio = p->prio; | |
dd41f596 IM |
2974 | p = iterator->next(iterator->arg); |
2975 | goto next; | |
1da177e4 LT |
2976 | } |
2977 | out: | |
2978 | /* | |
e1d1484f | 2979 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2980 | * so we can safely collect pull_task() stats here rather than |
2981 | * inside pull_task(). | |
2982 | */ | |
2983 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2984 | |
2985 | if (all_pinned) | |
2986 | *all_pinned = pinned; | |
e1d1484f PW |
2987 | |
2988 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2989 | } |
2990 | ||
dd41f596 | 2991 | /* |
43010659 PW |
2992 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2993 | * this_rq, as part of a balancing operation within domain "sd". | |
2994 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2995 | * |
2996 | * Called with both runqueues locked. | |
2997 | */ | |
2998 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2999 | unsigned long max_load_move, |
dd41f596 IM |
3000 | struct sched_domain *sd, enum cpu_idle_type idle, |
3001 | int *all_pinned) | |
3002 | { | |
5522d5d5 | 3003 | const struct sched_class *class = sched_class_highest; |
43010659 | 3004 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3005 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3006 | |
3007 | do { | |
43010659 PW |
3008 | total_load_moved += |
3009 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3010 | max_load_move - total_load_moved, |
a4ac01c3 | 3011 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3012 | class = class->next; |
c4acb2c0 GH |
3013 | |
3014 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3015 | break; | |
3016 | ||
43010659 | 3017 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3018 | |
43010659 PW |
3019 | return total_load_moved > 0; |
3020 | } | |
3021 | ||
e1d1484f PW |
3022 | static int |
3023 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3024 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3025 | struct rq_iterator *iterator) | |
3026 | { | |
3027 | struct task_struct *p = iterator->start(iterator->arg); | |
3028 | int pinned = 0; | |
3029 | ||
3030 | while (p) { | |
3031 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3032 | pull_task(busiest, p, this_rq, this_cpu); | |
3033 | /* | |
3034 | * Right now, this is only the second place pull_task() | |
3035 | * is called, so we can safely collect pull_task() | |
3036 | * stats here rather than inside pull_task(). | |
3037 | */ | |
3038 | schedstat_inc(sd, lb_gained[idle]); | |
3039 | ||
3040 | return 1; | |
3041 | } | |
3042 | p = iterator->next(iterator->arg); | |
3043 | } | |
3044 | ||
3045 | return 0; | |
3046 | } | |
3047 | ||
43010659 PW |
3048 | /* |
3049 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3050 | * part of active balancing operations within "domain". | |
3051 | * Returns 1 if successful and 0 otherwise. | |
3052 | * | |
3053 | * Called with both runqueues locked. | |
3054 | */ | |
3055 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3056 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3057 | { | |
5522d5d5 | 3058 | const struct sched_class *class; |
43010659 PW |
3059 | |
3060 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3061 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3062 | return 1; |
3063 | ||
3064 | return 0; | |
dd41f596 IM |
3065 | } |
3066 | ||
1da177e4 LT |
3067 | /* |
3068 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3069 | * domain. It calculates and returns the amount of weighted load which |
3070 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3071 | */ |
3072 | static struct sched_group * | |
3073 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3074 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3075 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3076 | { |
3077 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3078 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3079 | unsigned long max_pull; |
2dd73a4f PW |
3080 | unsigned long busiest_load_per_task, busiest_nr_running; |
3081 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3082 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3083 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3084 | int power_savings_balance = 1; | |
3085 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3086 | unsigned long min_nr_running = ULONG_MAX; | |
3087 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3088 | #endif | |
1da177e4 LT |
3089 | |
3090 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3091 | busiest_load_per_task = busiest_nr_running = 0; |
3092 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3093 | |
d15bcfdb | 3094 | if (idle == CPU_NOT_IDLE) |
7897986b | 3095 | load_idx = sd->busy_idx; |
d15bcfdb | 3096 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3097 | load_idx = sd->newidle_idx; |
3098 | else | |
3099 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3100 | |
3101 | do { | |
908a7c1b | 3102 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3103 | int local_group; |
3104 | int i; | |
908a7c1b | 3105 | int __group_imb = 0; |
783609c6 | 3106 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3107 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3108 | unsigned long sum_avg_load_per_task; |
3109 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3110 | |
3111 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3112 | ||
783609c6 SS |
3113 | if (local_group) |
3114 | balance_cpu = first_cpu(group->cpumask); | |
3115 | ||
1da177e4 | 3116 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3117 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3118 | sum_avg_load_per_task = avg_load_per_task = 0; |
3119 | ||
908a7c1b KC |
3120 | max_cpu_load = 0; |
3121 | min_cpu_load = ~0UL; | |
1da177e4 | 3122 | |
363ab6f1 | 3123 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3124 | struct rq *rq; |
3125 | ||
3126 | if (!cpu_isset(i, *cpus)) | |
3127 | continue; | |
3128 | ||
3129 | rq = cpu_rq(i); | |
2dd73a4f | 3130 | |
9439aab8 | 3131 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3132 | *sd_idle = 0; |
3133 | ||
1da177e4 | 3134 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3135 | if (local_group) { |
3136 | if (idle_cpu(i) && !first_idle_cpu) { | |
3137 | first_idle_cpu = 1; | |
3138 | balance_cpu = i; | |
3139 | } | |
3140 | ||
a2000572 | 3141 | load = target_load(i, load_idx); |
908a7c1b | 3142 | } else { |
a2000572 | 3143 | load = source_load(i, load_idx); |
908a7c1b KC |
3144 | if (load > max_cpu_load) |
3145 | max_cpu_load = load; | |
3146 | if (min_cpu_load > load) | |
3147 | min_cpu_load = load; | |
3148 | } | |
1da177e4 LT |
3149 | |
3150 | avg_load += load; | |
2dd73a4f | 3151 | sum_nr_running += rq->nr_running; |
dd41f596 | 3152 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3153 | |
3154 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3155 | } |
3156 | ||
783609c6 SS |
3157 | /* |
3158 | * First idle cpu or the first cpu(busiest) in this sched group | |
3159 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3160 | * domains. In the newly idle case, we will allow all the cpu's |
3161 | * to do the newly idle load balance. | |
783609c6 | 3162 | */ |
9439aab8 SS |
3163 | if (idle != CPU_NEWLY_IDLE && local_group && |
3164 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3165 | *balance = 0; |
3166 | goto ret; | |
3167 | } | |
3168 | ||
1da177e4 | 3169 | total_load += avg_load; |
5517d86b | 3170 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3171 | |
3172 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3173 | avg_load = sg_div_cpu_power(group, |
3174 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3175 | |
408ed066 PZ |
3176 | |
3177 | /* | |
3178 | * Consider the group unbalanced when the imbalance is larger | |
3179 | * than the average weight of two tasks. | |
3180 | * | |
3181 | * APZ: with cgroup the avg task weight can vary wildly and | |
3182 | * might not be a suitable number - should we keep a | |
3183 | * normalized nr_running number somewhere that negates | |
3184 | * the hierarchy? | |
3185 | */ | |
3186 | avg_load_per_task = sg_div_cpu_power(group, | |
3187 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3188 | ||
3189 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3190 | __group_imb = 1; |
3191 | ||
5517d86b | 3192 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3193 | |
1da177e4 LT |
3194 | if (local_group) { |
3195 | this_load = avg_load; | |
3196 | this = group; | |
2dd73a4f PW |
3197 | this_nr_running = sum_nr_running; |
3198 | this_load_per_task = sum_weighted_load; | |
3199 | } else if (avg_load > max_load && | |
908a7c1b | 3200 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3201 | max_load = avg_load; |
3202 | busiest = group; | |
2dd73a4f PW |
3203 | busiest_nr_running = sum_nr_running; |
3204 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3205 | group_imb = __group_imb; |
1da177e4 | 3206 | } |
5c45bf27 SS |
3207 | |
3208 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3209 | /* | |
3210 | * Busy processors will not participate in power savings | |
3211 | * balance. | |
3212 | */ | |
dd41f596 IM |
3213 | if (idle == CPU_NOT_IDLE || |
3214 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3215 | goto group_next; | |
5c45bf27 SS |
3216 | |
3217 | /* | |
3218 | * If the local group is idle or completely loaded | |
3219 | * no need to do power savings balance at this domain | |
3220 | */ | |
3221 | if (local_group && (this_nr_running >= group_capacity || | |
3222 | !this_nr_running)) | |
3223 | power_savings_balance = 0; | |
3224 | ||
dd41f596 | 3225 | /* |
5c45bf27 SS |
3226 | * If a group is already running at full capacity or idle, |
3227 | * don't include that group in power savings calculations | |
dd41f596 IM |
3228 | */ |
3229 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3230 | || !sum_nr_running) |
dd41f596 | 3231 | goto group_next; |
5c45bf27 | 3232 | |
dd41f596 | 3233 | /* |
5c45bf27 | 3234 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3235 | * This is the group from where we need to pick up the load |
3236 | * for saving power | |
3237 | */ | |
3238 | if ((sum_nr_running < min_nr_running) || | |
3239 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3240 | first_cpu(group->cpumask) < |
3241 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3242 | group_min = group; |
3243 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3244 | min_load_per_task = sum_weighted_load / |
3245 | sum_nr_running; | |
dd41f596 | 3246 | } |
5c45bf27 | 3247 | |
dd41f596 | 3248 | /* |
5c45bf27 | 3249 | * Calculate the group which is almost near its |
dd41f596 IM |
3250 | * capacity but still has some space to pick up some load |
3251 | * from other group and save more power | |
3252 | */ | |
3253 | if (sum_nr_running <= group_capacity - 1) { | |
3254 | if (sum_nr_running > leader_nr_running || | |
3255 | (sum_nr_running == leader_nr_running && | |
3256 | first_cpu(group->cpumask) > | |
3257 | first_cpu(group_leader->cpumask))) { | |
3258 | group_leader = group; | |
3259 | leader_nr_running = sum_nr_running; | |
3260 | } | |
48f24c4d | 3261 | } |
5c45bf27 SS |
3262 | group_next: |
3263 | #endif | |
1da177e4 LT |
3264 | group = group->next; |
3265 | } while (group != sd->groups); | |
3266 | ||
2dd73a4f | 3267 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3268 | goto out_balanced; |
3269 | ||
3270 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3271 | ||
3272 | if (this_load >= avg_load || | |
3273 | 100*max_load <= sd->imbalance_pct*this_load) | |
3274 | goto out_balanced; | |
3275 | ||
2dd73a4f | 3276 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3277 | if (group_imb) |
3278 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3279 | ||
1da177e4 LT |
3280 | /* |
3281 | * We're trying to get all the cpus to the average_load, so we don't | |
3282 | * want to push ourselves above the average load, nor do we wish to | |
3283 | * reduce the max loaded cpu below the average load, as either of these | |
3284 | * actions would just result in more rebalancing later, and ping-pong | |
3285 | * tasks around. Thus we look for the minimum possible imbalance. | |
3286 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3287 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3288 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3289 | * appear as very large values with unsigned longs. |
3290 | */ | |
2dd73a4f PW |
3291 | if (max_load <= busiest_load_per_task) |
3292 | goto out_balanced; | |
3293 | ||
3294 | /* | |
3295 | * In the presence of smp nice balancing, certain scenarios can have | |
3296 | * max load less than avg load(as we skip the groups at or below | |
3297 | * its cpu_power, while calculating max_load..) | |
3298 | */ | |
3299 | if (max_load < avg_load) { | |
3300 | *imbalance = 0; | |
3301 | goto small_imbalance; | |
3302 | } | |
0c117f1b SS |
3303 | |
3304 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3305 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3306 | |
1da177e4 | 3307 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3308 | *imbalance = min(max_pull * busiest->__cpu_power, |
3309 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3310 | / SCHED_LOAD_SCALE; |
3311 | ||
2dd73a4f PW |
3312 | /* |
3313 | * if *imbalance is less than the average load per runnable task | |
3314 | * there is no gaurantee that any tasks will be moved so we'll have | |
3315 | * a think about bumping its value to force at least one task to be | |
3316 | * moved | |
3317 | */ | |
7fd0d2dd | 3318 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3319 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3320 | unsigned int imbn; |
3321 | ||
3322 | small_imbalance: | |
3323 | pwr_move = pwr_now = 0; | |
3324 | imbn = 2; | |
3325 | if (this_nr_running) { | |
3326 | this_load_per_task /= this_nr_running; | |
3327 | if (busiest_load_per_task > this_load_per_task) | |
3328 | imbn = 1; | |
3329 | } else | |
408ed066 | 3330 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3331 | |
408ed066 | 3332 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3333 | busiest_load_per_task * imbn) { |
2dd73a4f | 3334 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3335 | return busiest; |
3336 | } | |
3337 | ||
3338 | /* | |
3339 | * OK, we don't have enough imbalance to justify moving tasks, | |
3340 | * however we may be able to increase total CPU power used by | |
3341 | * moving them. | |
3342 | */ | |
3343 | ||
5517d86b ED |
3344 | pwr_now += busiest->__cpu_power * |
3345 | min(busiest_load_per_task, max_load); | |
3346 | pwr_now += this->__cpu_power * | |
3347 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3348 | pwr_now /= SCHED_LOAD_SCALE; |
3349 | ||
3350 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3351 | tmp = sg_div_cpu_power(busiest, |
3352 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3353 | if (max_load > tmp) |
5517d86b | 3354 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3355 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3356 | |
3357 | /* Amount of load we'd add */ | |
5517d86b | 3358 | if (max_load * busiest->__cpu_power < |
33859f7f | 3359 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3360 | tmp = sg_div_cpu_power(this, |
3361 | max_load * busiest->__cpu_power); | |
1da177e4 | 3362 | else |
5517d86b ED |
3363 | tmp = sg_div_cpu_power(this, |
3364 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3365 | pwr_move += this->__cpu_power * | |
3366 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3367 | pwr_move /= SCHED_LOAD_SCALE; |
3368 | ||
3369 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3370 | if (pwr_move > pwr_now) |
3371 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3372 | } |
3373 | ||
1da177e4 LT |
3374 | return busiest; |
3375 | ||
3376 | out_balanced: | |
5c45bf27 | 3377 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3378 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3379 | goto ret; |
1da177e4 | 3380 | |
5c45bf27 SS |
3381 | if (this == group_leader && group_leader != group_min) { |
3382 | *imbalance = min_load_per_task; | |
3383 | return group_min; | |
3384 | } | |
5c45bf27 | 3385 | #endif |
783609c6 | 3386 | ret: |
1da177e4 LT |
3387 | *imbalance = 0; |
3388 | return NULL; | |
3389 | } | |
3390 | ||
3391 | /* | |
3392 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3393 | */ | |
70b97a7f | 3394 | static struct rq * |
d15bcfdb | 3395 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3396 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3397 | { |
70b97a7f | 3398 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3399 | unsigned long max_load = 0; |
1da177e4 LT |
3400 | int i; |
3401 | ||
363ab6f1 | 3402 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3403 | unsigned long wl; |
0a2966b4 CL |
3404 | |
3405 | if (!cpu_isset(i, *cpus)) | |
3406 | continue; | |
3407 | ||
48f24c4d | 3408 | rq = cpu_rq(i); |
dd41f596 | 3409 | wl = weighted_cpuload(i); |
2dd73a4f | 3410 | |
dd41f596 | 3411 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3412 | continue; |
1da177e4 | 3413 | |
dd41f596 IM |
3414 | if (wl > max_load) { |
3415 | max_load = wl; | |
48f24c4d | 3416 | busiest = rq; |
1da177e4 LT |
3417 | } |
3418 | } | |
3419 | ||
3420 | return busiest; | |
3421 | } | |
3422 | ||
77391d71 NP |
3423 | /* |
3424 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3425 | * so long as it is large enough. | |
3426 | */ | |
3427 | #define MAX_PINNED_INTERVAL 512 | |
3428 | ||
1da177e4 LT |
3429 | /* |
3430 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3431 | * tasks if there is an imbalance. | |
1da177e4 | 3432 | */ |
70b97a7f | 3433 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3434 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3435 | int *balance, cpumask_t *cpus) |
1da177e4 | 3436 | { |
43010659 | 3437 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3438 | struct sched_group *group; |
1da177e4 | 3439 | unsigned long imbalance; |
70b97a7f | 3440 | struct rq *busiest; |
fe2eea3f | 3441 | unsigned long flags; |
5969fe06 | 3442 | |
7c16ec58 MT |
3443 | cpus_setall(*cpus); |
3444 | ||
89c4710e SS |
3445 | /* |
3446 | * When power savings policy is enabled for the parent domain, idle | |
3447 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3448 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3449 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3450 | */ |
d15bcfdb | 3451 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3452 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3453 | sd_idle = 1; |
1da177e4 | 3454 | |
2d72376b | 3455 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3456 | |
0a2966b4 | 3457 | redo: |
c8cba857 | 3458 | update_shares(sd); |
0a2966b4 | 3459 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3460 | cpus, balance); |
783609c6 | 3461 | |
06066714 | 3462 | if (*balance == 0) |
783609c6 | 3463 | goto out_balanced; |
783609c6 | 3464 | |
1da177e4 LT |
3465 | if (!group) { |
3466 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3467 | goto out_balanced; | |
3468 | } | |
3469 | ||
7c16ec58 | 3470 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3471 | if (!busiest) { |
3472 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3473 | goto out_balanced; | |
3474 | } | |
3475 | ||
db935dbd | 3476 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3477 | |
3478 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3479 | ||
43010659 | 3480 | ld_moved = 0; |
1da177e4 LT |
3481 | if (busiest->nr_running > 1) { |
3482 | /* | |
3483 | * Attempt to move tasks. If find_busiest_group has found | |
3484 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3485 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3486 | * correctly treated as an imbalance. |
3487 | */ | |
fe2eea3f | 3488 | local_irq_save(flags); |
e17224bf | 3489 | double_rq_lock(this_rq, busiest); |
43010659 | 3490 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3491 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3492 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3493 | local_irq_restore(flags); |
81026794 | 3494 | |
46cb4b7c SS |
3495 | /* |
3496 | * some other cpu did the load balance for us. | |
3497 | */ | |
43010659 | 3498 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3499 | resched_cpu(this_cpu); |
3500 | ||
81026794 | 3501 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3502 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3503 | cpu_clear(cpu_of(busiest), *cpus); |
3504 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3505 | goto redo; |
81026794 | 3506 | goto out_balanced; |
0a2966b4 | 3507 | } |
1da177e4 | 3508 | } |
81026794 | 3509 | |
43010659 | 3510 | if (!ld_moved) { |
1da177e4 LT |
3511 | schedstat_inc(sd, lb_failed[idle]); |
3512 | sd->nr_balance_failed++; | |
3513 | ||
3514 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3515 | |
fe2eea3f | 3516 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3517 | |
3518 | /* don't kick the migration_thread, if the curr | |
3519 | * task on busiest cpu can't be moved to this_cpu | |
3520 | */ | |
3521 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3522 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3523 | all_pinned = 1; |
3524 | goto out_one_pinned; | |
3525 | } | |
3526 | ||
1da177e4 LT |
3527 | if (!busiest->active_balance) { |
3528 | busiest->active_balance = 1; | |
3529 | busiest->push_cpu = this_cpu; | |
81026794 | 3530 | active_balance = 1; |
1da177e4 | 3531 | } |
fe2eea3f | 3532 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3533 | if (active_balance) |
1da177e4 LT |
3534 | wake_up_process(busiest->migration_thread); |
3535 | ||
3536 | /* | |
3537 | * We've kicked active balancing, reset the failure | |
3538 | * counter. | |
3539 | */ | |
39507451 | 3540 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3541 | } |
81026794 | 3542 | } else |
1da177e4 LT |
3543 | sd->nr_balance_failed = 0; |
3544 | ||
81026794 | 3545 | if (likely(!active_balance)) { |
1da177e4 LT |
3546 | /* We were unbalanced, so reset the balancing interval */ |
3547 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3548 | } else { |
3549 | /* | |
3550 | * If we've begun active balancing, start to back off. This | |
3551 | * case may not be covered by the all_pinned logic if there | |
3552 | * is only 1 task on the busy runqueue (because we don't call | |
3553 | * move_tasks). | |
3554 | */ | |
3555 | if (sd->balance_interval < sd->max_interval) | |
3556 | sd->balance_interval *= 2; | |
1da177e4 LT |
3557 | } |
3558 | ||
43010659 | 3559 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3560 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3561 | ld_moved = -1; |
3562 | ||
3563 | goto out; | |
1da177e4 LT |
3564 | |
3565 | out_balanced: | |
1da177e4 LT |
3566 | schedstat_inc(sd, lb_balanced[idle]); |
3567 | ||
16cfb1c0 | 3568 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3569 | |
3570 | out_one_pinned: | |
1da177e4 | 3571 | /* tune up the balancing interval */ |
77391d71 NP |
3572 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3573 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3574 | sd->balance_interval *= 2; |
3575 | ||
48f24c4d | 3576 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3577 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3578 | ld_moved = -1; |
3579 | else | |
3580 | ld_moved = 0; | |
3581 | out: | |
c8cba857 PZ |
3582 | if (ld_moved) |
3583 | update_shares(sd); | |
c09595f6 | 3584 | return ld_moved; |
1da177e4 LT |
3585 | } |
3586 | ||
3587 | /* | |
3588 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3589 | * tasks if there is an imbalance. | |
3590 | * | |
d15bcfdb | 3591 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3592 | * this_rq is locked. |
3593 | */ | |
48f24c4d | 3594 | static int |
7c16ec58 MT |
3595 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3596 | cpumask_t *cpus) | |
1da177e4 LT |
3597 | { |
3598 | struct sched_group *group; | |
70b97a7f | 3599 | struct rq *busiest = NULL; |
1da177e4 | 3600 | unsigned long imbalance; |
43010659 | 3601 | int ld_moved = 0; |
5969fe06 | 3602 | int sd_idle = 0; |
969bb4e4 | 3603 | int all_pinned = 0; |
7c16ec58 MT |
3604 | |
3605 | cpus_setall(*cpus); | |
5969fe06 | 3606 | |
89c4710e SS |
3607 | /* |
3608 | * When power savings policy is enabled for the parent domain, idle | |
3609 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3610 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3611 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3612 | */ |
3613 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3614 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3615 | sd_idle = 1; |
1da177e4 | 3616 | |
2d72376b | 3617 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3618 | redo: |
3e5459b4 | 3619 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3620 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3621 | &sd_idle, cpus, NULL); |
1da177e4 | 3622 | if (!group) { |
d15bcfdb | 3623 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3624 | goto out_balanced; |
1da177e4 LT |
3625 | } |
3626 | ||
7c16ec58 | 3627 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3628 | if (!busiest) { |
d15bcfdb | 3629 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3630 | goto out_balanced; |
1da177e4 LT |
3631 | } |
3632 | ||
db935dbd NP |
3633 | BUG_ON(busiest == this_rq); |
3634 | ||
d15bcfdb | 3635 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3636 | |
43010659 | 3637 | ld_moved = 0; |
d6d5cfaf NP |
3638 | if (busiest->nr_running > 1) { |
3639 | /* Attempt to move tasks */ | |
3640 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3641 | /* this_rq->clock is already updated */ |
3642 | update_rq_clock(busiest); | |
43010659 | 3643 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3644 | imbalance, sd, CPU_NEWLY_IDLE, |
3645 | &all_pinned); | |
1b12bbc7 | 3646 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3647 | |
969bb4e4 | 3648 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3649 | cpu_clear(cpu_of(busiest), *cpus); |
3650 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3651 | goto redo; |
3652 | } | |
d6d5cfaf NP |
3653 | } |
3654 | ||
43010659 | 3655 | if (!ld_moved) { |
d15bcfdb | 3656 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3657 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3658 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3659 | return -1; |
3660 | } else | |
16cfb1c0 | 3661 | sd->nr_balance_failed = 0; |
1da177e4 | 3662 | |
3e5459b4 | 3663 | update_shares_locked(this_rq, sd); |
43010659 | 3664 | return ld_moved; |
16cfb1c0 NP |
3665 | |
3666 | out_balanced: | |
d15bcfdb | 3667 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3668 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3669 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3670 | return -1; |
16cfb1c0 | 3671 | sd->nr_balance_failed = 0; |
48f24c4d | 3672 | |
16cfb1c0 | 3673 | return 0; |
1da177e4 LT |
3674 | } |
3675 | ||
3676 | /* | |
3677 | * idle_balance is called by schedule() if this_cpu is about to become | |
3678 | * idle. Attempts to pull tasks from other CPUs. | |
3679 | */ | |
70b97a7f | 3680 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3681 | { |
3682 | struct sched_domain *sd; | |
dd41f596 IM |
3683 | int pulled_task = -1; |
3684 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3685 | cpumask_t tmpmask; |
1da177e4 LT |
3686 | |
3687 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3688 | unsigned long interval; |
3689 | ||
3690 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3691 | continue; | |
3692 | ||
3693 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3694 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3695 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3696 | sd, &tmpmask); | |
92c4ca5c CL |
3697 | |
3698 | interval = msecs_to_jiffies(sd->balance_interval); | |
3699 | if (time_after(next_balance, sd->last_balance + interval)) | |
3700 | next_balance = sd->last_balance + interval; | |
3701 | if (pulled_task) | |
3702 | break; | |
1da177e4 | 3703 | } |
dd41f596 | 3704 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3705 | /* |
3706 | * We are going idle. next_balance may be set based on | |
3707 | * a busy processor. So reset next_balance. | |
3708 | */ | |
3709 | this_rq->next_balance = next_balance; | |
dd41f596 | 3710 | } |
1da177e4 LT |
3711 | } |
3712 | ||
3713 | /* | |
3714 | * active_load_balance is run by migration threads. It pushes running tasks | |
3715 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3716 | * running on each physical CPU where possible, and avoids physical / | |
3717 | * logical imbalances. | |
3718 | * | |
3719 | * Called with busiest_rq locked. | |
3720 | */ | |
70b97a7f | 3721 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3722 | { |
39507451 | 3723 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3724 | struct sched_domain *sd; |
3725 | struct rq *target_rq; | |
39507451 | 3726 | |
48f24c4d | 3727 | /* Is there any task to move? */ |
39507451 | 3728 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3729 | return; |
3730 | ||
3731 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3732 | |
3733 | /* | |
39507451 | 3734 | * This condition is "impossible", if it occurs |
41a2d6cf | 3735 | * we need to fix it. Originally reported by |
39507451 | 3736 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3737 | */ |
39507451 | 3738 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3739 | |
39507451 NP |
3740 | /* move a task from busiest_rq to target_rq */ |
3741 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3742 | update_rq_clock(busiest_rq); |
3743 | update_rq_clock(target_rq); | |
39507451 NP |
3744 | |
3745 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3746 | for_each_domain(target_cpu, sd) { |
39507451 | 3747 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3748 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3749 | break; |
c96d145e | 3750 | } |
39507451 | 3751 | |
48f24c4d | 3752 | if (likely(sd)) { |
2d72376b | 3753 | schedstat_inc(sd, alb_count); |
39507451 | 3754 | |
43010659 PW |
3755 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3756 | sd, CPU_IDLE)) | |
48f24c4d IM |
3757 | schedstat_inc(sd, alb_pushed); |
3758 | else | |
3759 | schedstat_inc(sd, alb_failed); | |
3760 | } | |
1b12bbc7 | 3761 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3762 | } |
3763 | ||
46cb4b7c SS |
3764 | #ifdef CONFIG_NO_HZ |
3765 | static struct { | |
3766 | atomic_t load_balancer; | |
41a2d6cf | 3767 | cpumask_t cpu_mask; |
46cb4b7c SS |
3768 | } nohz ____cacheline_aligned = { |
3769 | .load_balancer = ATOMIC_INIT(-1), | |
3770 | .cpu_mask = CPU_MASK_NONE, | |
3771 | }; | |
3772 | ||
7835b98b | 3773 | /* |
46cb4b7c SS |
3774 | * This routine will try to nominate the ilb (idle load balancing) |
3775 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3776 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3777 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3778 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3779 | * arrives... | |
3780 | * | |
3781 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3782 | * for idle load balancing. ilb owner will still be part of | |
3783 | * nohz.cpu_mask.. | |
7835b98b | 3784 | * |
46cb4b7c SS |
3785 | * While stopping the tick, this cpu will become the ilb owner if there |
3786 | * is no other owner. And will be the owner till that cpu becomes busy | |
3787 | * or if all cpus in the system stop their ticks at which point | |
3788 | * there is no need for ilb owner. | |
3789 | * | |
3790 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3791 | * next busy scheduler_tick() | |
3792 | */ | |
3793 | int select_nohz_load_balancer(int stop_tick) | |
3794 | { | |
3795 | int cpu = smp_processor_id(); | |
3796 | ||
3797 | if (stop_tick) { | |
3798 | cpu_set(cpu, nohz.cpu_mask); | |
3799 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3800 | ||
3801 | /* | |
3802 | * If we are going offline and still the leader, give up! | |
3803 | */ | |
e761b772 | 3804 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3805 | atomic_read(&nohz.load_balancer) == cpu) { |
3806 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3807 | BUG(); | |
3808 | return 0; | |
3809 | } | |
3810 | ||
3811 | /* time for ilb owner also to sleep */ | |
3812 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3813 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3814 | atomic_set(&nohz.load_balancer, -1); | |
3815 | return 0; | |
3816 | } | |
3817 | ||
3818 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3819 | /* make me the ilb owner */ | |
3820 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3821 | return 1; | |
3822 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3823 | return 1; | |
3824 | } else { | |
3825 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3826 | return 0; | |
3827 | ||
3828 | cpu_clear(cpu, nohz.cpu_mask); | |
3829 | ||
3830 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3831 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3832 | BUG(); | |
3833 | } | |
3834 | return 0; | |
3835 | } | |
3836 | #endif | |
3837 | ||
3838 | static DEFINE_SPINLOCK(balancing); | |
3839 | ||
3840 | /* | |
7835b98b CL |
3841 | * It checks each scheduling domain to see if it is due to be balanced, |
3842 | * and initiates a balancing operation if so. | |
3843 | * | |
3844 | * Balancing parameters are set up in arch_init_sched_domains. | |
3845 | */ | |
a9957449 | 3846 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3847 | { |
46cb4b7c SS |
3848 | int balance = 1; |
3849 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3850 | unsigned long interval; |
3851 | struct sched_domain *sd; | |
46cb4b7c | 3852 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3853 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3854 | int update_next_balance = 0; |
d07355f5 | 3855 | int need_serialize; |
7c16ec58 | 3856 | cpumask_t tmp; |
1da177e4 | 3857 | |
46cb4b7c | 3858 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3859 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3860 | continue; | |
3861 | ||
3862 | interval = sd->balance_interval; | |
d15bcfdb | 3863 | if (idle != CPU_IDLE) |
1da177e4 LT |
3864 | interval *= sd->busy_factor; |
3865 | ||
3866 | /* scale ms to jiffies */ | |
3867 | interval = msecs_to_jiffies(interval); | |
3868 | if (unlikely(!interval)) | |
3869 | interval = 1; | |
dd41f596 IM |
3870 | if (interval > HZ*NR_CPUS/10) |
3871 | interval = HZ*NR_CPUS/10; | |
3872 | ||
d07355f5 | 3873 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3874 | |
d07355f5 | 3875 | if (need_serialize) { |
08c183f3 CL |
3876 | if (!spin_trylock(&balancing)) |
3877 | goto out; | |
3878 | } | |
3879 | ||
c9819f45 | 3880 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3881 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3882 | /* |
3883 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3884 | * longer idle, or one of our SMT siblings is |
3885 | * not idle. | |
3886 | */ | |
d15bcfdb | 3887 | idle = CPU_NOT_IDLE; |
1da177e4 | 3888 | } |
1bd77f2d | 3889 | sd->last_balance = jiffies; |
1da177e4 | 3890 | } |
d07355f5 | 3891 | if (need_serialize) |
08c183f3 CL |
3892 | spin_unlock(&balancing); |
3893 | out: | |
f549da84 | 3894 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3895 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3896 | update_next_balance = 1; |
3897 | } | |
783609c6 SS |
3898 | |
3899 | /* | |
3900 | * Stop the load balance at this level. There is another | |
3901 | * CPU in our sched group which is doing load balancing more | |
3902 | * actively. | |
3903 | */ | |
3904 | if (!balance) | |
3905 | break; | |
1da177e4 | 3906 | } |
f549da84 SS |
3907 | |
3908 | /* | |
3909 | * next_balance will be updated only when there is a need. | |
3910 | * When the cpu is attached to null domain for ex, it will not be | |
3911 | * updated. | |
3912 | */ | |
3913 | if (likely(update_next_balance)) | |
3914 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3915 | } |
3916 | ||
3917 | /* | |
3918 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3919 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3920 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3921 | */ | |
3922 | static void run_rebalance_domains(struct softirq_action *h) | |
3923 | { | |
dd41f596 IM |
3924 | int this_cpu = smp_processor_id(); |
3925 | struct rq *this_rq = cpu_rq(this_cpu); | |
3926 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3927 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3928 | |
dd41f596 | 3929 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3930 | |
3931 | #ifdef CONFIG_NO_HZ | |
3932 | /* | |
3933 | * If this cpu is the owner for idle load balancing, then do the | |
3934 | * balancing on behalf of the other idle cpus whose ticks are | |
3935 | * stopped. | |
3936 | */ | |
dd41f596 IM |
3937 | if (this_rq->idle_at_tick && |
3938 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3939 | cpumask_t cpus = nohz.cpu_mask; |
3940 | struct rq *rq; | |
3941 | int balance_cpu; | |
3942 | ||
dd41f596 | 3943 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3944 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3945 | /* |
3946 | * If this cpu gets work to do, stop the load balancing | |
3947 | * work being done for other cpus. Next load | |
3948 | * balancing owner will pick it up. | |
3949 | */ | |
3950 | if (need_resched()) | |
3951 | break; | |
3952 | ||
de0cf899 | 3953 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3954 | |
3955 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3956 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3957 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3958 | } |
3959 | } | |
3960 | #endif | |
3961 | } | |
3962 | ||
3963 | /* | |
3964 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3965 | * | |
3966 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3967 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3968 | * if the whole system is idle. | |
3969 | */ | |
dd41f596 | 3970 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3971 | { |
46cb4b7c SS |
3972 | #ifdef CONFIG_NO_HZ |
3973 | /* | |
3974 | * If we were in the nohz mode recently and busy at the current | |
3975 | * scheduler tick, then check if we need to nominate new idle | |
3976 | * load balancer. | |
3977 | */ | |
3978 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3979 | rq->in_nohz_recently = 0; | |
3980 | ||
3981 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3982 | cpu_clear(cpu, nohz.cpu_mask); | |
3983 | atomic_set(&nohz.load_balancer, -1); | |
3984 | } | |
3985 | ||
3986 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3987 | /* | |
3988 | * simple selection for now: Nominate the | |
3989 | * first cpu in the nohz list to be the next | |
3990 | * ilb owner. | |
3991 | * | |
3992 | * TBD: Traverse the sched domains and nominate | |
3993 | * the nearest cpu in the nohz.cpu_mask. | |
3994 | */ | |
3995 | int ilb = first_cpu(nohz.cpu_mask); | |
3996 | ||
434d53b0 | 3997 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3998 | resched_cpu(ilb); |
3999 | } | |
4000 | } | |
4001 | ||
4002 | /* | |
4003 | * If this cpu is idle and doing idle load balancing for all the | |
4004 | * cpus with ticks stopped, is it time for that to stop? | |
4005 | */ | |
4006 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4007 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4008 | resched_cpu(cpu); | |
4009 | return; | |
4010 | } | |
4011 | ||
4012 | /* | |
4013 | * If this cpu is idle and the idle load balancing is done by | |
4014 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4015 | */ | |
4016 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4017 | cpu_isset(cpu, nohz.cpu_mask)) | |
4018 | return; | |
4019 | #endif | |
4020 | if (time_after_eq(jiffies, rq->next_balance)) | |
4021 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4022 | } |
dd41f596 IM |
4023 | |
4024 | #else /* CONFIG_SMP */ | |
4025 | ||
1da177e4 LT |
4026 | /* |
4027 | * on UP we do not need to balance between CPUs: | |
4028 | */ | |
70b97a7f | 4029 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4030 | { |
4031 | } | |
dd41f596 | 4032 | |
1da177e4 LT |
4033 | #endif |
4034 | ||
1da177e4 LT |
4035 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4036 | ||
4037 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4038 | ||
f06febc9 FM |
4039 | /* |
4040 | * Return any ns on the sched_clock that have not yet been banked in | |
4041 | * @p in case that task is currently running. | |
4042 | * | |
4043 | * Called with task_rq_lock() held on @rq. | |
4044 | */ | |
4045 | static unsigned long long task_delta_exec(struct task_struct *p, struct rq *rq) | |
4046 | { | |
4047 | if (task_current(rq, p)) { | |
4048 | u64 delta_exec; | |
4049 | ||
4050 | update_rq_clock(rq); | |
4051 | delta_exec = rq->clock - p->se.exec_start; | |
4052 | if ((s64)delta_exec > 0) | |
4053 | return delta_exec; | |
4054 | } | |
4055 | return 0; | |
4056 | } | |
4057 | ||
1da177e4 | 4058 | /* |
41b86e9c IM |
4059 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4060 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4061 | */ |
41b86e9c | 4062 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4063 | { |
1da177e4 | 4064 | unsigned long flags; |
f06febc9 | 4065 | u64 ns; |
41b86e9c | 4066 | struct rq *rq; |
48f24c4d | 4067 | |
41b86e9c | 4068 | rq = task_rq_lock(p, &flags); |
f06febc9 FM |
4069 | ns = p->se.sum_exec_runtime + task_delta_exec(p, rq); |
4070 | task_rq_unlock(rq, &flags); | |
4071 | ||
4072 | return ns; | |
4073 | } | |
4074 | ||
4075 | /* | |
4076 | * Return sum_exec_runtime for the thread group plus any more ns on the | |
4077 | * sched_clock that have not yet been banked in case the task is currently | |
4078 | * running. | |
4079 | */ | |
4080 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4081 | { | |
4082 | unsigned long flags; | |
4083 | u64 ns; | |
4084 | struct rq *rq; | |
4085 | struct task_cputime totals; | |
4086 | ||
4087 | rq = task_rq_lock(p, &flags); | |
4088 | thread_group_cputime(p, &totals); | |
4089 | ns = totals.sum_exec_runtime + task_delta_exec(p, rq); | |
41b86e9c | 4090 | task_rq_unlock(rq, &flags); |
48f24c4d | 4091 | |
1da177e4 LT |
4092 | return ns; |
4093 | } | |
4094 | ||
1da177e4 LT |
4095 | /* |
4096 | * Account user cpu time to a process. | |
4097 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4098 | * @cputime: the cpu time spent in user space since the last update |
4099 | */ | |
4100 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4101 | { | |
4102 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4103 | cputime64_t tmp; | |
4104 | ||
4105 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4106 | account_group_user_time(p, cputime); |
1da177e4 LT |
4107 | |
4108 | /* Add user time to cpustat. */ | |
4109 | tmp = cputime_to_cputime64(cputime); | |
4110 | if (TASK_NICE(p) > 0) | |
4111 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4112 | else | |
4113 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4114 | /* Account for user time used */ |
4115 | acct_update_integrals(p); | |
1da177e4 LT |
4116 | } |
4117 | ||
94886b84 LV |
4118 | /* |
4119 | * Account guest cpu time to a process. | |
4120 | * @p: the process that the cpu time gets accounted to | |
4121 | * @cputime: the cpu time spent in virtual machine since the last update | |
4122 | */ | |
f7402e03 | 4123 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4124 | { |
4125 | cputime64_t tmp; | |
4126 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4127 | ||
4128 | tmp = cputime_to_cputime64(cputime); | |
4129 | ||
4130 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4131 | account_group_user_time(p, cputime); |
94886b84 LV |
4132 | p->gtime = cputime_add(p->gtime, cputime); |
4133 | ||
4134 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4135 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4136 | } | |
4137 | ||
c66f08be MN |
4138 | /* |
4139 | * Account scaled user cpu time to a process. | |
4140 | * @p: the process that the cpu time gets accounted to | |
4141 | * @cputime: the cpu time spent in user space since the last update | |
4142 | */ | |
4143 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4144 | { | |
4145 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4146 | } | |
4147 | ||
1da177e4 LT |
4148 | /* |
4149 | * Account system cpu time to a process. | |
4150 | * @p: the process that the cpu time gets accounted to | |
4151 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4152 | * @cputime: the cpu time spent in kernel space since the last update | |
4153 | */ | |
4154 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4155 | cputime_t cputime) | |
4156 | { | |
4157 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4158 | struct rq *rq = this_rq(); |
1da177e4 LT |
4159 | cputime64_t tmp; |
4160 | ||
983ed7a6 HH |
4161 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4162 | account_guest_time(p, cputime); | |
4163 | return; | |
4164 | } | |
94886b84 | 4165 | |
1da177e4 | 4166 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4167 | account_group_system_time(p, cputime); |
1da177e4 LT |
4168 | |
4169 | /* Add system time to cpustat. */ | |
4170 | tmp = cputime_to_cputime64(cputime); | |
4171 | if (hardirq_count() - hardirq_offset) | |
4172 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4173 | else if (softirq_count()) | |
4174 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4175 | else if (p != rq->idle) |
1da177e4 | 4176 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4177 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4178 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4179 | else | |
4180 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4181 | /* Account for system time used */ | |
4182 | acct_update_integrals(p); | |
1da177e4 LT |
4183 | } |
4184 | ||
c66f08be MN |
4185 | /* |
4186 | * Account scaled system cpu time to a process. | |
4187 | * @p: the process that the cpu time gets accounted to | |
4188 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4189 | * @cputime: the cpu time spent in kernel space since the last update | |
4190 | */ | |
4191 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4192 | { | |
4193 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4194 | } | |
4195 | ||
1da177e4 LT |
4196 | /* |
4197 | * Account for involuntary wait time. | |
4198 | * @p: the process from which the cpu time has been stolen | |
4199 | * @steal: the cpu time spent in involuntary wait | |
4200 | */ | |
4201 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4202 | { | |
4203 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4204 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4205 | struct rq *rq = this_rq(); |
1da177e4 LT |
4206 | |
4207 | if (p == rq->idle) { | |
4208 | p->stime = cputime_add(p->stime, steal); | |
f06febc9 | 4209 | account_group_system_time(p, steal); |
1da177e4 LT |
4210 | if (atomic_read(&rq->nr_iowait) > 0) |
4211 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4212 | else | |
4213 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4214 | } else |
1da177e4 LT |
4215 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4216 | } | |
4217 | ||
49048622 BS |
4218 | /* |
4219 | * Use precise platform statistics if available: | |
4220 | */ | |
4221 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4222 | cputime_t task_utime(struct task_struct *p) | |
4223 | { | |
4224 | return p->utime; | |
4225 | } | |
4226 | ||
4227 | cputime_t task_stime(struct task_struct *p) | |
4228 | { | |
4229 | return p->stime; | |
4230 | } | |
4231 | #else | |
4232 | cputime_t task_utime(struct task_struct *p) | |
4233 | { | |
4234 | clock_t utime = cputime_to_clock_t(p->utime), | |
4235 | total = utime + cputime_to_clock_t(p->stime); | |
4236 | u64 temp; | |
4237 | ||
4238 | /* | |
4239 | * Use CFS's precise accounting: | |
4240 | */ | |
4241 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4242 | ||
4243 | if (total) { | |
4244 | temp *= utime; | |
4245 | do_div(temp, total); | |
4246 | } | |
4247 | utime = (clock_t)temp; | |
4248 | ||
4249 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4250 | return p->prev_utime; | |
4251 | } | |
4252 | ||
4253 | cputime_t task_stime(struct task_struct *p) | |
4254 | { | |
4255 | clock_t stime; | |
4256 | ||
4257 | /* | |
4258 | * Use CFS's precise accounting. (we subtract utime from | |
4259 | * the total, to make sure the total observed by userspace | |
4260 | * grows monotonically - apps rely on that): | |
4261 | */ | |
4262 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4263 | cputime_to_clock_t(task_utime(p)); | |
4264 | ||
4265 | if (stime >= 0) | |
4266 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4267 | ||
4268 | return p->prev_stime; | |
4269 | } | |
4270 | #endif | |
4271 | ||
4272 | inline cputime_t task_gtime(struct task_struct *p) | |
4273 | { | |
4274 | return p->gtime; | |
4275 | } | |
4276 | ||
7835b98b CL |
4277 | /* |
4278 | * This function gets called by the timer code, with HZ frequency. | |
4279 | * We call it with interrupts disabled. | |
4280 | * | |
4281 | * It also gets called by the fork code, when changing the parent's | |
4282 | * timeslices. | |
4283 | */ | |
4284 | void scheduler_tick(void) | |
4285 | { | |
7835b98b CL |
4286 | int cpu = smp_processor_id(); |
4287 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4288 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4289 | |
4290 | sched_clock_tick(); | |
dd41f596 IM |
4291 | |
4292 | spin_lock(&rq->lock); | |
3e51f33f | 4293 | update_rq_clock(rq); |
f1a438d8 | 4294 | update_cpu_load(rq); |
fa85ae24 | 4295 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4296 | spin_unlock(&rq->lock); |
7835b98b | 4297 | |
e418e1c2 | 4298 | #ifdef CONFIG_SMP |
dd41f596 IM |
4299 | rq->idle_at_tick = idle_cpu(cpu); |
4300 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4301 | #endif |
1da177e4 LT |
4302 | } |
4303 | ||
6cd8a4bb SR |
4304 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4305 | defined(CONFIG_PREEMPT_TRACER)) | |
4306 | ||
4307 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4308 | { | |
4309 | if (in_lock_functions(addr)) { | |
4310 | addr = CALLER_ADDR2; | |
4311 | if (in_lock_functions(addr)) | |
4312 | addr = CALLER_ADDR3; | |
4313 | } | |
4314 | return addr; | |
4315 | } | |
1da177e4 | 4316 | |
43627582 | 4317 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4318 | { |
6cd8a4bb | 4319 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4320 | /* |
4321 | * Underflow? | |
4322 | */ | |
9a11b49a IM |
4323 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4324 | return; | |
6cd8a4bb | 4325 | #endif |
1da177e4 | 4326 | preempt_count() += val; |
6cd8a4bb | 4327 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4328 | /* |
4329 | * Spinlock count overflowing soon? | |
4330 | */ | |
33859f7f MOS |
4331 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4332 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4333 | #endif |
4334 | if (preempt_count() == val) | |
4335 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4336 | } |
4337 | EXPORT_SYMBOL(add_preempt_count); | |
4338 | ||
43627582 | 4339 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4340 | { |
6cd8a4bb | 4341 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4342 | /* |
4343 | * Underflow? | |
4344 | */ | |
9a11b49a IM |
4345 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4346 | return; | |
1da177e4 LT |
4347 | /* |
4348 | * Is the spinlock portion underflowing? | |
4349 | */ | |
9a11b49a IM |
4350 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4351 | !(preempt_count() & PREEMPT_MASK))) | |
4352 | return; | |
6cd8a4bb | 4353 | #endif |
9a11b49a | 4354 | |
6cd8a4bb SR |
4355 | if (preempt_count() == val) |
4356 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4357 | preempt_count() -= val; |
4358 | } | |
4359 | EXPORT_SYMBOL(sub_preempt_count); | |
4360 | ||
4361 | #endif | |
4362 | ||
4363 | /* | |
dd41f596 | 4364 | * Print scheduling while atomic bug: |
1da177e4 | 4365 | */ |
dd41f596 | 4366 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4367 | { |
838225b4 SS |
4368 | struct pt_regs *regs = get_irq_regs(); |
4369 | ||
4370 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4371 | prev->comm, prev->pid, preempt_count()); | |
4372 | ||
dd41f596 | 4373 | debug_show_held_locks(prev); |
e21f5b15 | 4374 | print_modules(); |
dd41f596 IM |
4375 | if (irqs_disabled()) |
4376 | print_irqtrace_events(prev); | |
838225b4 SS |
4377 | |
4378 | if (regs) | |
4379 | show_regs(regs); | |
4380 | else | |
4381 | dump_stack(); | |
dd41f596 | 4382 | } |
1da177e4 | 4383 | |
dd41f596 IM |
4384 | /* |
4385 | * Various schedule()-time debugging checks and statistics: | |
4386 | */ | |
4387 | static inline void schedule_debug(struct task_struct *prev) | |
4388 | { | |
1da177e4 | 4389 | /* |
41a2d6cf | 4390 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4391 | * schedule() atomically, we ignore that path for now. |
4392 | * Otherwise, whine if we are scheduling when we should not be. | |
4393 | */ | |
3f33a7ce | 4394 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4395 | __schedule_bug(prev); |
4396 | ||
1da177e4 LT |
4397 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4398 | ||
2d72376b | 4399 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4400 | #ifdef CONFIG_SCHEDSTATS |
4401 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4402 | schedstat_inc(this_rq(), bkl_count); |
4403 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4404 | } |
4405 | #endif | |
dd41f596 IM |
4406 | } |
4407 | ||
4408 | /* | |
4409 | * Pick up the highest-prio task: | |
4410 | */ | |
4411 | static inline struct task_struct * | |
ff95f3df | 4412 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4413 | { |
5522d5d5 | 4414 | const struct sched_class *class; |
dd41f596 | 4415 | struct task_struct *p; |
1da177e4 LT |
4416 | |
4417 | /* | |
dd41f596 IM |
4418 | * Optimization: we know that if all tasks are in |
4419 | * the fair class we can call that function directly: | |
1da177e4 | 4420 | */ |
dd41f596 | 4421 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4422 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4423 | if (likely(p)) |
4424 | return p; | |
1da177e4 LT |
4425 | } |
4426 | ||
dd41f596 IM |
4427 | class = sched_class_highest; |
4428 | for ( ; ; ) { | |
fb8d4724 | 4429 | p = class->pick_next_task(rq); |
dd41f596 IM |
4430 | if (p) |
4431 | return p; | |
4432 | /* | |
4433 | * Will never be NULL as the idle class always | |
4434 | * returns a non-NULL p: | |
4435 | */ | |
4436 | class = class->next; | |
4437 | } | |
4438 | } | |
1da177e4 | 4439 | |
dd41f596 IM |
4440 | /* |
4441 | * schedule() is the main scheduler function. | |
4442 | */ | |
4443 | asmlinkage void __sched schedule(void) | |
4444 | { | |
4445 | struct task_struct *prev, *next; | |
67ca7bde | 4446 | unsigned long *switch_count; |
dd41f596 | 4447 | struct rq *rq; |
31656519 | 4448 | int cpu; |
dd41f596 IM |
4449 | |
4450 | need_resched: | |
4451 | preempt_disable(); | |
4452 | cpu = smp_processor_id(); | |
4453 | rq = cpu_rq(cpu); | |
4454 | rcu_qsctr_inc(cpu); | |
4455 | prev = rq->curr; | |
4456 | switch_count = &prev->nivcsw; | |
4457 | ||
4458 | release_kernel_lock(prev); | |
4459 | need_resched_nonpreemptible: | |
4460 | ||
4461 | schedule_debug(prev); | |
1da177e4 | 4462 | |
31656519 | 4463 | if (sched_feat(HRTICK)) |
f333fdc9 | 4464 | hrtick_clear(rq); |
8f4d37ec | 4465 | |
1e819950 IM |
4466 | /* |
4467 | * Do the rq-clock update outside the rq lock: | |
4468 | */ | |
4469 | local_irq_disable(); | |
3e51f33f | 4470 | update_rq_clock(rq); |
1e819950 IM |
4471 | spin_lock(&rq->lock); |
4472 | clear_tsk_need_resched(prev); | |
1da177e4 | 4473 | |
1da177e4 | 4474 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4475 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4476 | prev->state = TASK_RUNNING; |
16882c1e | 4477 | else |
2e1cb74a | 4478 | deactivate_task(rq, prev, 1); |
dd41f596 | 4479 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4480 | } |
4481 | ||
9a897c5a SR |
4482 | #ifdef CONFIG_SMP |
4483 | if (prev->sched_class->pre_schedule) | |
4484 | prev->sched_class->pre_schedule(rq, prev); | |
4485 | #endif | |
f65eda4f | 4486 | |
dd41f596 | 4487 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4488 | idle_balance(cpu, rq); |
1da177e4 | 4489 | |
31ee529c | 4490 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4491 | next = pick_next_task(rq, prev); |
1da177e4 | 4492 | |
1da177e4 | 4493 | if (likely(prev != next)) { |
673a90a1 DS |
4494 | sched_info_switch(prev, next); |
4495 | ||
1da177e4 LT |
4496 | rq->nr_switches++; |
4497 | rq->curr = next; | |
4498 | ++*switch_count; | |
4499 | ||
dd41f596 | 4500 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4501 | /* |
4502 | * the context switch might have flipped the stack from under | |
4503 | * us, hence refresh the local variables. | |
4504 | */ | |
4505 | cpu = smp_processor_id(); | |
4506 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4507 | } else |
4508 | spin_unlock_irq(&rq->lock); | |
4509 | ||
8f4d37ec | 4510 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4511 | goto need_resched_nonpreemptible; |
8f4d37ec | 4512 | |
1da177e4 LT |
4513 | preempt_enable_no_resched(); |
4514 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4515 | goto need_resched; | |
4516 | } | |
1da177e4 LT |
4517 | EXPORT_SYMBOL(schedule); |
4518 | ||
4519 | #ifdef CONFIG_PREEMPT | |
4520 | /* | |
2ed6e34f | 4521 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4522 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4523 | * occur there and call schedule directly. |
4524 | */ | |
4525 | asmlinkage void __sched preempt_schedule(void) | |
4526 | { | |
4527 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4528 | |
1da177e4 LT |
4529 | /* |
4530 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4531 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4532 | */ |
beed33a8 | 4533 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4534 | return; |
4535 | ||
3a5c359a AK |
4536 | do { |
4537 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4538 | schedule(); |
3a5c359a | 4539 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4540 | |
3a5c359a AK |
4541 | /* |
4542 | * Check again in case we missed a preemption opportunity | |
4543 | * between schedule and now. | |
4544 | */ | |
4545 | barrier(); | |
4546 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4547 | } |
1da177e4 LT |
4548 | EXPORT_SYMBOL(preempt_schedule); |
4549 | ||
4550 | /* | |
2ed6e34f | 4551 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4552 | * off of irq context. |
4553 | * Note, that this is called and return with irqs disabled. This will | |
4554 | * protect us against recursive calling from irq. | |
4555 | */ | |
4556 | asmlinkage void __sched preempt_schedule_irq(void) | |
4557 | { | |
4558 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4559 | |
2ed6e34f | 4560 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4561 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4562 | ||
3a5c359a AK |
4563 | do { |
4564 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4565 | local_irq_enable(); |
4566 | schedule(); | |
4567 | local_irq_disable(); | |
3a5c359a | 4568 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4569 | |
3a5c359a AK |
4570 | /* |
4571 | * Check again in case we missed a preemption opportunity | |
4572 | * between schedule and now. | |
4573 | */ | |
4574 | barrier(); | |
4575 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4576 | } |
4577 | ||
4578 | #endif /* CONFIG_PREEMPT */ | |
4579 | ||
95cdf3b7 IM |
4580 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4581 | void *key) | |
1da177e4 | 4582 | { |
48f24c4d | 4583 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4584 | } |
1da177e4 LT |
4585 | EXPORT_SYMBOL(default_wake_function); |
4586 | ||
4587 | /* | |
41a2d6cf IM |
4588 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4589 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4590 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4591 | * | |
4592 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4593 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4594 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4595 | */ | |
4596 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4597 | int nr_exclusive, int sync, void *key) | |
4598 | { | |
2e45874c | 4599 | wait_queue_t *curr, *next; |
1da177e4 | 4600 | |
2e45874c | 4601 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4602 | unsigned flags = curr->flags; |
4603 | ||
1da177e4 | 4604 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4605 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4606 | break; |
4607 | } | |
4608 | } | |
4609 | ||
4610 | /** | |
4611 | * __wake_up - wake up threads blocked on a waitqueue. | |
4612 | * @q: the waitqueue | |
4613 | * @mode: which threads | |
4614 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4615 | * @key: is directly passed to the wakeup function |
1da177e4 | 4616 | */ |
7ad5b3a5 | 4617 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4618 | int nr_exclusive, void *key) |
1da177e4 LT |
4619 | { |
4620 | unsigned long flags; | |
4621 | ||
4622 | spin_lock_irqsave(&q->lock, flags); | |
4623 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4624 | spin_unlock_irqrestore(&q->lock, flags); | |
4625 | } | |
1da177e4 LT |
4626 | EXPORT_SYMBOL(__wake_up); |
4627 | ||
4628 | /* | |
4629 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4630 | */ | |
7ad5b3a5 | 4631 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4632 | { |
4633 | __wake_up_common(q, mode, 1, 0, NULL); | |
4634 | } | |
4635 | ||
4636 | /** | |
67be2dd1 | 4637 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4638 | * @q: the waitqueue |
4639 | * @mode: which threads | |
4640 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4641 | * | |
4642 | * The sync wakeup differs that the waker knows that it will schedule | |
4643 | * away soon, so while the target thread will be woken up, it will not | |
4644 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4645 | * with each other. This can prevent needless bouncing between CPUs. | |
4646 | * | |
4647 | * On UP it can prevent extra preemption. | |
4648 | */ | |
7ad5b3a5 | 4649 | void |
95cdf3b7 | 4650 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4651 | { |
4652 | unsigned long flags; | |
4653 | int sync = 1; | |
4654 | ||
4655 | if (unlikely(!q)) | |
4656 | return; | |
4657 | ||
4658 | if (unlikely(!nr_exclusive)) | |
4659 | sync = 0; | |
4660 | ||
4661 | spin_lock_irqsave(&q->lock, flags); | |
4662 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4663 | spin_unlock_irqrestore(&q->lock, flags); | |
4664 | } | |
4665 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4666 | ||
b15136e9 | 4667 | void complete(struct completion *x) |
1da177e4 LT |
4668 | { |
4669 | unsigned long flags; | |
4670 | ||
4671 | spin_lock_irqsave(&x->wait.lock, flags); | |
4672 | x->done++; | |
d9514f6c | 4673 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4674 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4675 | } | |
4676 | EXPORT_SYMBOL(complete); | |
4677 | ||
b15136e9 | 4678 | void complete_all(struct completion *x) |
1da177e4 LT |
4679 | { |
4680 | unsigned long flags; | |
4681 | ||
4682 | spin_lock_irqsave(&x->wait.lock, flags); | |
4683 | x->done += UINT_MAX/2; | |
d9514f6c | 4684 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4685 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4686 | } | |
4687 | EXPORT_SYMBOL(complete_all); | |
4688 | ||
8cbbe86d AK |
4689 | static inline long __sched |
4690 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4691 | { |
1da177e4 LT |
4692 | if (!x->done) { |
4693 | DECLARE_WAITQUEUE(wait, current); | |
4694 | ||
4695 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4696 | __add_wait_queue_tail(&x->wait, &wait); | |
4697 | do { | |
009e577e MW |
4698 | if ((state == TASK_INTERRUPTIBLE && |
4699 | signal_pending(current)) || | |
4700 | (state == TASK_KILLABLE && | |
4701 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4702 | timeout = -ERESTARTSYS; |
4703 | break; | |
8cbbe86d AK |
4704 | } |
4705 | __set_current_state(state); | |
1da177e4 LT |
4706 | spin_unlock_irq(&x->wait.lock); |
4707 | timeout = schedule_timeout(timeout); | |
4708 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4709 | } while (!x->done && timeout); |
1da177e4 | 4710 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4711 | if (!x->done) |
4712 | return timeout; | |
1da177e4 LT |
4713 | } |
4714 | x->done--; | |
ea71a546 | 4715 | return timeout ?: 1; |
1da177e4 | 4716 | } |
1da177e4 | 4717 | |
8cbbe86d AK |
4718 | static long __sched |
4719 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4720 | { |
1da177e4 LT |
4721 | might_sleep(); |
4722 | ||
4723 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4724 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4725 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4726 | return timeout; |
4727 | } | |
1da177e4 | 4728 | |
b15136e9 | 4729 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4730 | { |
4731 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4732 | } |
8cbbe86d | 4733 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4734 | |
b15136e9 | 4735 | unsigned long __sched |
8cbbe86d | 4736 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4737 | { |
8cbbe86d | 4738 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4739 | } |
8cbbe86d | 4740 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4741 | |
8cbbe86d | 4742 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4743 | { |
51e97990 AK |
4744 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4745 | if (t == -ERESTARTSYS) | |
4746 | return t; | |
4747 | return 0; | |
0fec171c | 4748 | } |
8cbbe86d | 4749 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4750 | |
b15136e9 | 4751 | unsigned long __sched |
8cbbe86d AK |
4752 | wait_for_completion_interruptible_timeout(struct completion *x, |
4753 | unsigned long timeout) | |
0fec171c | 4754 | { |
8cbbe86d | 4755 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4756 | } |
8cbbe86d | 4757 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4758 | |
009e577e MW |
4759 | int __sched wait_for_completion_killable(struct completion *x) |
4760 | { | |
4761 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4762 | if (t == -ERESTARTSYS) | |
4763 | return t; | |
4764 | return 0; | |
4765 | } | |
4766 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4767 | ||
be4de352 DC |
4768 | /** |
4769 | * try_wait_for_completion - try to decrement a completion without blocking | |
4770 | * @x: completion structure | |
4771 | * | |
4772 | * Returns: 0 if a decrement cannot be done without blocking | |
4773 | * 1 if a decrement succeeded. | |
4774 | * | |
4775 | * If a completion is being used as a counting completion, | |
4776 | * attempt to decrement the counter without blocking. This | |
4777 | * enables us to avoid waiting if the resource the completion | |
4778 | * is protecting is not available. | |
4779 | */ | |
4780 | bool try_wait_for_completion(struct completion *x) | |
4781 | { | |
4782 | int ret = 1; | |
4783 | ||
4784 | spin_lock_irq(&x->wait.lock); | |
4785 | if (!x->done) | |
4786 | ret = 0; | |
4787 | else | |
4788 | x->done--; | |
4789 | spin_unlock_irq(&x->wait.lock); | |
4790 | return ret; | |
4791 | } | |
4792 | EXPORT_SYMBOL(try_wait_for_completion); | |
4793 | ||
4794 | /** | |
4795 | * completion_done - Test to see if a completion has any waiters | |
4796 | * @x: completion structure | |
4797 | * | |
4798 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4799 | * 1 if there are no waiters. | |
4800 | * | |
4801 | */ | |
4802 | bool completion_done(struct completion *x) | |
4803 | { | |
4804 | int ret = 1; | |
4805 | ||
4806 | spin_lock_irq(&x->wait.lock); | |
4807 | if (!x->done) | |
4808 | ret = 0; | |
4809 | spin_unlock_irq(&x->wait.lock); | |
4810 | return ret; | |
4811 | } | |
4812 | EXPORT_SYMBOL(completion_done); | |
4813 | ||
8cbbe86d AK |
4814 | static long __sched |
4815 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4816 | { |
0fec171c IM |
4817 | unsigned long flags; |
4818 | wait_queue_t wait; | |
4819 | ||
4820 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4821 | |
8cbbe86d | 4822 | __set_current_state(state); |
1da177e4 | 4823 | |
8cbbe86d AK |
4824 | spin_lock_irqsave(&q->lock, flags); |
4825 | __add_wait_queue(q, &wait); | |
4826 | spin_unlock(&q->lock); | |
4827 | timeout = schedule_timeout(timeout); | |
4828 | spin_lock_irq(&q->lock); | |
4829 | __remove_wait_queue(q, &wait); | |
4830 | spin_unlock_irqrestore(&q->lock, flags); | |
4831 | ||
4832 | return timeout; | |
4833 | } | |
4834 | ||
4835 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4836 | { | |
4837 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4838 | } |
1da177e4 LT |
4839 | EXPORT_SYMBOL(interruptible_sleep_on); |
4840 | ||
0fec171c | 4841 | long __sched |
95cdf3b7 | 4842 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4843 | { |
8cbbe86d | 4844 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4845 | } |
1da177e4 LT |
4846 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4847 | ||
0fec171c | 4848 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4849 | { |
8cbbe86d | 4850 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4851 | } |
1da177e4 LT |
4852 | EXPORT_SYMBOL(sleep_on); |
4853 | ||
0fec171c | 4854 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4855 | { |
8cbbe86d | 4856 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4857 | } |
1da177e4 LT |
4858 | EXPORT_SYMBOL(sleep_on_timeout); |
4859 | ||
b29739f9 IM |
4860 | #ifdef CONFIG_RT_MUTEXES |
4861 | ||
4862 | /* | |
4863 | * rt_mutex_setprio - set the current priority of a task | |
4864 | * @p: task | |
4865 | * @prio: prio value (kernel-internal form) | |
4866 | * | |
4867 | * This function changes the 'effective' priority of a task. It does | |
4868 | * not touch ->normal_prio like __setscheduler(). | |
4869 | * | |
4870 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4871 | */ | |
36c8b586 | 4872 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4873 | { |
4874 | unsigned long flags; | |
83b699ed | 4875 | int oldprio, on_rq, running; |
70b97a7f | 4876 | struct rq *rq; |
cb469845 | 4877 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4878 | |
4879 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4880 | ||
4881 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4882 | update_rq_clock(rq); |
b29739f9 | 4883 | |
d5f9f942 | 4884 | oldprio = p->prio; |
dd41f596 | 4885 | on_rq = p->se.on_rq; |
051a1d1a | 4886 | running = task_current(rq, p); |
0e1f3483 | 4887 | if (on_rq) |
69be72c1 | 4888 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4889 | if (running) |
4890 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4891 | |
4892 | if (rt_prio(prio)) | |
4893 | p->sched_class = &rt_sched_class; | |
4894 | else | |
4895 | p->sched_class = &fair_sched_class; | |
4896 | ||
b29739f9 IM |
4897 | p->prio = prio; |
4898 | ||
0e1f3483 HS |
4899 | if (running) |
4900 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4901 | if (on_rq) { |
8159f87e | 4902 | enqueue_task(rq, p, 0); |
cb469845 SR |
4903 | |
4904 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4905 | } |
4906 | task_rq_unlock(rq, &flags); | |
4907 | } | |
4908 | ||
4909 | #endif | |
4910 | ||
36c8b586 | 4911 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4912 | { |
dd41f596 | 4913 | int old_prio, delta, on_rq; |
1da177e4 | 4914 | unsigned long flags; |
70b97a7f | 4915 | struct rq *rq; |
1da177e4 LT |
4916 | |
4917 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4918 | return; | |
4919 | /* | |
4920 | * We have to be careful, if called from sys_setpriority(), | |
4921 | * the task might be in the middle of scheduling on another CPU. | |
4922 | */ | |
4923 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4924 | update_rq_clock(rq); |
1da177e4 LT |
4925 | /* |
4926 | * The RT priorities are set via sched_setscheduler(), but we still | |
4927 | * allow the 'normal' nice value to be set - but as expected | |
4928 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4929 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4930 | */ |
e05606d3 | 4931 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4932 | p->static_prio = NICE_TO_PRIO(nice); |
4933 | goto out_unlock; | |
4934 | } | |
dd41f596 | 4935 | on_rq = p->se.on_rq; |
c09595f6 | 4936 | if (on_rq) |
69be72c1 | 4937 | dequeue_task(rq, p, 0); |
1da177e4 | 4938 | |
1da177e4 | 4939 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4940 | set_load_weight(p); |
b29739f9 IM |
4941 | old_prio = p->prio; |
4942 | p->prio = effective_prio(p); | |
4943 | delta = p->prio - old_prio; | |
1da177e4 | 4944 | |
dd41f596 | 4945 | if (on_rq) { |
8159f87e | 4946 | enqueue_task(rq, p, 0); |
1da177e4 | 4947 | /* |
d5f9f942 AM |
4948 | * If the task increased its priority or is running and |
4949 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4950 | */ |
d5f9f942 | 4951 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4952 | resched_task(rq->curr); |
4953 | } | |
4954 | out_unlock: | |
4955 | task_rq_unlock(rq, &flags); | |
4956 | } | |
1da177e4 LT |
4957 | EXPORT_SYMBOL(set_user_nice); |
4958 | ||
e43379f1 MM |
4959 | /* |
4960 | * can_nice - check if a task can reduce its nice value | |
4961 | * @p: task | |
4962 | * @nice: nice value | |
4963 | */ | |
36c8b586 | 4964 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4965 | { |
024f4747 MM |
4966 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4967 | int nice_rlim = 20 - nice; | |
48f24c4d | 4968 | |
e43379f1 MM |
4969 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4970 | capable(CAP_SYS_NICE)); | |
4971 | } | |
4972 | ||
1da177e4 LT |
4973 | #ifdef __ARCH_WANT_SYS_NICE |
4974 | ||
4975 | /* | |
4976 | * sys_nice - change the priority of the current process. | |
4977 | * @increment: priority increment | |
4978 | * | |
4979 | * sys_setpriority is a more generic, but much slower function that | |
4980 | * does similar things. | |
4981 | */ | |
4982 | asmlinkage long sys_nice(int increment) | |
4983 | { | |
48f24c4d | 4984 | long nice, retval; |
1da177e4 LT |
4985 | |
4986 | /* | |
4987 | * Setpriority might change our priority at the same moment. | |
4988 | * We don't have to worry. Conceptually one call occurs first | |
4989 | * and we have a single winner. | |
4990 | */ | |
e43379f1 MM |
4991 | if (increment < -40) |
4992 | increment = -40; | |
1da177e4 LT |
4993 | if (increment > 40) |
4994 | increment = 40; | |
4995 | ||
4996 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4997 | if (nice < -20) | |
4998 | nice = -20; | |
4999 | if (nice > 19) | |
5000 | nice = 19; | |
5001 | ||
e43379f1 MM |
5002 | if (increment < 0 && !can_nice(current, nice)) |
5003 | return -EPERM; | |
5004 | ||
1da177e4 LT |
5005 | retval = security_task_setnice(current, nice); |
5006 | if (retval) | |
5007 | return retval; | |
5008 | ||
5009 | set_user_nice(current, nice); | |
5010 | return 0; | |
5011 | } | |
5012 | ||
5013 | #endif | |
5014 | ||
5015 | /** | |
5016 | * task_prio - return the priority value of a given task. | |
5017 | * @p: the task in question. | |
5018 | * | |
5019 | * This is the priority value as seen by users in /proc. | |
5020 | * RT tasks are offset by -200. Normal tasks are centered | |
5021 | * around 0, value goes from -16 to +15. | |
5022 | */ | |
36c8b586 | 5023 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5024 | { |
5025 | return p->prio - MAX_RT_PRIO; | |
5026 | } | |
5027 | ||
5028 | /** | |
5029 | * task_nice - return the nice value of a given task. | |
5030 | * @p: the task in question. | |
5031 | */ | |
36c8b586 | 5032 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5033 | { |
5034 | return TASK_NICE(p); | |
5035 | } | |
150d8bed | 5036 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5037 | |
5038 | /** | |
5039 | * idle_cpu - is a given cpu idle currently? | |
5040 | * @cpu: the processor in question. | |
5041 | */ | |
5042 | int idle_cpu(int cpu) | |
5043 | { | |
5044 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5045 | } | |
5046 | ||
1da177e4 LT |
5047 | /** |
5048 | * idle_task - return the idle task for a given cpu. | |
5049 | * @cpu: the processor in question. | |
5050 | */ | |
36c8b586 | 5051 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5052 | { |
5053 | return cpu_rq(cpu)->idle; | |
5054 | } | |
5055 | ||
5056 | /** | |
5057 | * find_process_by_pid - find a process with a matching PID value. | |
5058 | * @pid: the pid in question. | |
5059 | */ | |
a9957449 | 5060 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5061 | { |
228ebcbe | 5062 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5063 | } |
5064 | ||
5065 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5066 | static void |
5067 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5068 | { |
dd41f596 | 5069 | BUG_ON(p->se.on_rq); |
48f24c4d | 5070 | |
1da177e4 | 5071 | p->policy = policy; |
dd41f596 IM |
5072 | switch (p->policy) { |
5073 | case SCHED_NORMAL: | |
5074 | case SCHED_BATCH: | |
5075 | case SCHED_IDLE: | |
5076 | p->sched_class = &fair_sched_class; | |
5077 | break; | |
5078 | case SCHED_FIFO: | |
5079 | case SCHED_RR: | |
5080 | p->sched_class = &rt_sched_class; | |
5081 | break; | |
5082 | } | |
5083 | ||
1da177e4 | 5084 | p->rt_priority = prio; |
b29739f9 IM |
5085 | p->normal_prio = normal_prio(p); |
5086 | /* we are holding p->pi_lock already */ | |
5087 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5088 | set_load_weight(p); |
1da177e4 LT |
5089 | } |
5090 | ||
961ccddd RR |
5091 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5092 | struct sched_param *param, bool user) | |
1da177e4 | 5093 | { |
83b699ed | 5094 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5095 | unsigned long flags; |
cb469845 | 5096 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5097 | struct rq *rq; |
1da177e4 | 5098 | |
66e5393a SR |
5099 | /* may grab non-irq protected spin_locks */ |
5100 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5101 | recheck: |
5102 | /* double check policy once rq lock held */ | |
5103 | if (policy < 0) | |
5104 | policy = oldpolicy = p->policy; | |
5105 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5106 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5107 | policy != SCHED_IDLE) | |
b0a9499c | 5108 | return -EINVAL; |
1da177e4 LT |
5109 | /* |
5110 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5111 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5112 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5113 | */ |
5114 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5115 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5116 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5117 | return -EINVAL; |
e05606d3 | 5118 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5119 | return -EINVAL; |
5120 | ||
37e4ab3f OC |
5121 | /* |
5122 | * Allow unprivileged RT tasks to decrease priority: | |
5123 | */ | |
961ccddd | 5124 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5125 | if (rt_policy(policy)) { |
8dc3e909 | 5126 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5127 | |
5128 | if (!lock_task_sighand(p, &flags)) | |
5129 | return -ESRCH; | |
5130 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5131 | unlock_task_sighand(p, &flags); | |
5132 | ||
5133 | /* can't set/change the rt policy */ | |
5134 | if (policy != p->policy && !rlim_rtprio) | |
5135 | return -EPERM; | |
5136 | ||
5137 | /* can't increase priority */ | |
5138 | if (param->sched_priority > p->rt_priority && | |
5139 | param->sched_priority > rlim_rtprio) | |
5140 | return -EPERM; | |
5141 | } | |
dd41f596 IM |
5142 | /* |
5143 | * Like positive nice levels, dont allow tasks to | |
5144 | * move out of SCHED_IDLE either: | |
5145 | */ | |
5146 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5147 | return -EPERM; | |
5fe1d75f | 5148 | |
37e4ab3f OC |
5149 | /* can't change other user's priorities */ |
5150 | if ((current->euid != p->euid) && | |
5151 | (current->euid != p->uid)) | |
5152 | return -EPERM; | |
5153 | } | |
1da177e4 | 5154 | |
725aad24 | 5155 | if (user) { |
b68aa230 | 5156 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5157 | /* |
5158 | * Do not allow realtime tasks into groups that have no runtime | |
5159 | * assigned. | |
5160 | */ | |
5161 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) | |
5162 | return -EPERM; | |
b68aa230 PZ |
5163 | #endif |
5164 | ||
725aad24 JF |
5165 | retval = security_task_setscheduler(p, policy, param); |
5166 | if (retval) | |
5167 | return retval; | |
5168 | } | |
5169 | ||
b29739f9 IM |
5170 | /* |
5171 | * make sure no PI-waiters arrive (or leave) while we are | |
5172 | * changing the priority of the task: | |
5173 | */ | |
5174 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5175 | /* |
5176 | * To be able to change p->policy safely, the apropriate | |
5177 | * runqueue lock must be held. | |
5178 | */ | |
b29739f9 | 5179 | rq = __task_rq_lock(p); |
1da177e4 LT |
5180 | /* recheck policy now with rq lock held */ |
5181 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5182 | policy = oldpolicy = -1; | |
b29739f9 IM |
5183 | __task_rq_unlock(rq); |
5184 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5185 | goto recheck; |
5186 | } | |
2daa3577 | 5187 | update_rq_clock(rq); |
dd41f596 | 5188 | on_rq = p->se.on_rq; |
051a1d1a | 5189 | running = task_current(rq, p); |
0e1f3483 | 5190 | if (on_rq) |
2e1cb74a | 5191 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5192 | if (running) |
5193 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5194 | |
1da177e4 | 5195 | oldprio = p->prio; |
dd41f596 | 5196 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5197 | |
0e1f3483 HS |
5198 | if (running) |
5199 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5200 | if (on_rq) { |
5201 | activate_task(rq, p, 0); | |
cb469845 SR |
5202 | |
5203 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5204 | } |
b29739f9 IM |
5205 | __task_rq_unlock(rq); |
5206 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5207 | ||
95e02ca9 TG |
5208 | rt_mutex_adjust_pi(p); |
5209 | ||
1da177e4 LT |
5210 | return 0; |
5211 | } | |
961ccddd RR |
5212 | |
5213 | /** | |
5214 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5215 | * @p: the task in question. | |
5216 | * @policy: new policy. | |
5217 | * @param: structure containing the new RT priority. | |
5218 | * | |
5219 | * NOTE that the task may be already dead. | |
5220 | */ | |
5221 | int sched_setscheduler(struct task_struct *p, int policy, | |
5222 | struct sched_param *param) | |
5223 | { | |
5224 | return __sched_setscheduler(p, policy, param, true); | |
5225 | } | |
1da177e4 LT |
5226 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5227 | ||
961ccddd RR |
5228 | /** |
5229 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5230 | * @p: the task in question. | |
5231 | * @policy: new policy. | |
5232 | * @param: structure containing the new RT priority. | |
5233 | * | |
5234 | * Just like sched_setscheduler, only don't bother checking if the | |
5235 | * current context has permission. For example, this is needed in | |
5236 | * stop_machine(): we create temporary high priority worker threads, | |
5237 | * but our caller might not have that capability. | |
5238 | */ | |
5239 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5240 | struct sched_param *param) | |
5241 | { | |
5242 | return __sched_setscheduler(p, policy, param, false); | |
5243 | } | |
5244 | ||
95cdf3b7 IM |
5245 | static int |
5246 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5247 | { |
1da177e4 LT |
5248 | struct sched_param lparam; |
5249 | struct task_struct *p; | |
36c8b586 | 5250 | int retval; |
1da177e4 LT |
5251 | |
5252 | if (!param || pid < 0) | |
5253 | return -EINVAL; | |
5254 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5255 | return -EFAULT; | |
5fe1d75f ON |
5256 | |
5257 | rcu_read_lock(); | |
5258 | retval = -ESRCH; | |
1da177e4 | 5259 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5260 | if (p != NULL) |
5261 | retval = sched_setscheduler(p, policy, &lparam); | |
5262 | rcu_read_unlock(); | |
36c8b586 | 5263 | |
1da177e4 LT |
5264 | return retval; |
5265 | } | |
5266 | ||
5267 | /** | |
5268 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5269 | * @pid: the pid in question. | |
5270 | * @policy: new policy. | |
5271 | * @param: structure containing the new RT priority. | |
5272 | */ | |
41a2d6cf IM |
5273 | asmlinkage long |
5274 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5275 | { |
c21761f1 JB |
5276 | /* negative values for policy are not valid */ |
5277 | if (policy < 0) | |
5278 | return -EINVAL; | |
5279 | ||
1da177e4 LT |
5280 | return do_sched_setscheduler(pid, policy, param); |
5281 | } | |
5282 | ||
5283 | /** | |
5284 | * sys_sched_setparam - set/change the RT priority of a thread | |
5285 | * @pid: the pid in question. | |
5286 | * @param: structure containing the new RT priority. | |
5287 | */ | |
5288 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5289 | { | |
5290 | return do_sched_setscheduler(pid, -1, param); | |
5291 | } | |
5292 | ||
5293 | /** | |
5294 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5295 | * @pid: the pid in question. | |
5296 | */ | |
5297 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5298 | { | |
36c8b586 | 5299 | struct task_struct *p; |
3a5c359a | 5300 | int retval; |
1da177e4 LT |
5301 | |
5302 | if (pid < 0) | |
3a5c359a | 5303 | return -EINVAL; |
1da177e4 LT |
5304 | |
5305 | retval = -ESRCH; | |
5306 | read_lock(&tasklist_lock); | |
5307 | p = find_process_by_pid(pid); | |
5308 | if (p) { | |
5309 | retval = security_task_getscheduler(p); | |
5310 | if (!retval) | |
5311 | retval = p->policy; | |
5312 | } | |
5313 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5314 | return retval; |
5315 | } | |
5316 | ||
5317 | /** | |
5318 | * sys_sched_getscheduler - get the RT priority of a thread | |
5319 | * @pid: the pid in question. | |
5320 | * @param: structure containing the RT priority. | |
5321 | */ | |
5322 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5323 | { | |
5324 | struct sched_param lp; | |
36c8b586 | 5325 | struct task_struct *p; |
3a5c359a | 5326 | int retval; |
1da177e4 LT |
5327 | |
5328 | if (!param || pid < 0) | |
3a5c359a | 5329 | return -EINVAL; |
1da177e4 LT |
5330 | |
5331 | read_lock(&tasklist_lock); | |
5332 | p = find_process_by_pid(pid); | |
5333 | retval = -ESRCH; | |
5334 | if (!p) | |
5335 | goto out_unlock; | |
5336 | ||
5337 | retval = security_task_getscheduler(p); | |
5338 | if (retval) | |
5339 | goto out_unlock; | |
5340 | ||
5341 | lp.sched_priority = p->rt_priority; | |
5342 | read_unlock(&tasklist_lock); | |
5343 | ||
5344 | /* | |
5345 | * This one might sleep, we cannot do it with a spinlock held ... | |
5346 | */ | |
5347 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5348 | ||
1da177e4 LT |
5349 | return retval; |
5350 | ||
5351 | out_unlock: | |
5352 | read_unlock(&tasklist_lock); | |
5353 | return retval; | |
5354 | } | |
5355 | ||
b53e921b | 5356 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5357 | { |
1da177e4 | 5358 | cpumask_t cpus_allowed; |
b53e921b | 5359 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5360 | struct task_struct *p; |
5361 | int retval; | |
1da177e4 | 5362 | |
95402b38 | 5363 | get_online_cpus(); |
1da177e4 LT |
5364 | read_lock(&tasklist_lock); |
5365 | ||
5366 | p = find_process_by_pid(pid); | |
5367 | if (!p) { | |
5368 | read_unlock(&tasklist_lock); | |
95402b38 | 5369 | put_online_cpus(); |
1da177e4 LT |
5370 | return -ESRCH; |
5371 | } | |
5372 | ||
5373 | /* | |
5374 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5375 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5376 | * usage count and then drop tasklist_lock. |
5377 | */ | |
5378 | get_task_struct(p); | |
5379 | read_unlock(&tasklist_lock); | |
5380 | ||
5381 | retval = -EPERM; | |
5382 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5383 | !capable(CAP_SYS_NICE)) | |
5384 | goto out_unlock; | |
5385 | ||
e7834f8f DQ |
5386 | retval = security_task_setscheduler(p, 0, NULL); |
5387 | if (retval) | |
5388 | goto out_unlock; | |
5389 | ||
f9a86fcb | 5390 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5391 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5392 | again: |
7c16ec58 | 5393 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5394 | |
8707d8b8 | 5395 | if (!retval) { |
f9a86fcb | 5396 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5397 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5398 | /* | |
5399 | * We must have raced with a concurrent cpuset | |
5400 | * update. Just reset the cpus_allowed to the | |
5401 | * cpuset's cpus_allowed | |
5402 | */ | |
5403 | new_mask = cpus_allowed; | |
5404 | goto again; | |
5405 | } | |
5406 | } | |
1da177e4 LT |
5407 | out_unlock: |
5408 | put_task_struct(p); | |
95402b38 | 5409 | put_online_cpus(); |
1da177e4 LT |
5410 | return retval; |
5411 | } | |
5412 | ||
5413 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5414 | cpumask_t *new_mask) | |
5415 | { | |
5416 | if (len < sizeof(cpumask_t)) { | |
5417 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5418 | } else if (len > sizeof(cpumask_t)) { | |
5419 | len = sizeof(cpumask_t); | |
5420 | } | |
5421 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5422 | } | |
5423 | ||
5424 | /** | |
5425 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5426 | * @pid: pid of the process | |
5427 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5428 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5429 | */ | |
5430 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5431 | unsigned long __user *user_mask_ptr) | |
5432 | { | |
5433 | cpumask_t new_mask; | |
5434 | int retval; | |
5435 | ||
5436 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5437 | if (retval) | |
5438 | return retval; | |
5439 | ||
b53e921b | 5440 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5441 | } |
5442 | ||
1da177e4 LT |
5443 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5444 | { | |
36c8b586 | 5445 | struct task_struct *p; |
1da177e4 | 5446 | int retval; |
1da177e4 | 5447 | |
95402b38 | 5448 | get_online_cpus(); |
1da177e4 LT |
5449 | read_lock(&tasklist_lock); |
5450 | ||
5451 | retval = -ESRCH; | |
5452 | p = find_process_by_pid(pid); | |
5453 | if (!p) | |
5454 | goto out_unlock; | |
5455 | ||
e7834f8f DQ |
5456 | retval = security_task_getscheduler(p); |
5457 | if (retval) | |
5458 | goto out_unlock; | |
5459 | ||
2f7016d9 | 5460 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5461 | |
5462 | out_unlock: | |
5463 | read_unlock(&tasklist_lock); | |
95402b38 | 5464 | put_online_cpus(); |
1da177e4 | 5465 | |
9531b62f | 5466 | return retval; |
1da177e4 LT |
5467 | } |
5468 | ||
5469 | /** | |
5470 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5471 | * @pid: pid of the process | |
5472 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5473 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5474 | */ | |
5475 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5476 | unsigned long __user *user_mask_ptr) | |
5477 | { | |
5478 | int ret; | |
5479 | cpumask_t mask; | |
5480 | ||
5481 | if (len < sizeof(cpumask_t)) | |
5482 | return -EINVAL; | |
5483 | ||
5484 | ret = sched_getaffinity(pid, &mask); | |
5485 | if (ret < 0) | |
5486 | return ret; | |
5487 | ||
5488 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5489 | return -EFAULT; | |
5490 | ||
5491 | return sizeof(cpumask_t); | |
5492 | } | |
5493 | ||
5494 | /** | |
5495 | * sys_sched_yield - yield the current processor to other threads. | |
5496 | * | |
dd41f596 IM |
5497 | * This function yields the current CPU to other tasks. If there are no |
5498 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5499 | */ |
5500 | asmlinkage long sys_sched_yield(void) | |
5501 | { | |
70b97a7f | 5502 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5503 | |
2d72376b | 5504 | schedstat_inc(rq, yld_count); |
4530d7ab | 5505 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5506 | |
5507 | /* | |
5508 | * Since we are going to call schedule() anyway, there's | |
5509 | * no need to preempt or enable interrupts: | |
5510 | */ | |
5511 | __release(rq->lock); | |
8a25d5de | 5512 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5513 | _raw_spin_unlock(&rq->lock); |
5514 | preempt_enable_no_resched(); | |
5515 | ||
5516 | schedule(); | |
5517 | ||
5518 | return 0; | |
5519 | } | |
5520 | ||
e7b38404 | 5521 | static void __cond_resched(void) |
1da177e4 | 5522 | { |
8e0a43d8 IM |
5523 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5524 | __might_sleep(__FILE__, __LINE__); | |
5525 | #endif | |
5bbcfd90 IM |
5526 | /* |
5527 | * The BKS might be reacquired before we have dropped | |
5528 | * PREEMPT_ACTIVE, which could trigger a second | |
5529 | * cond_resched() call. | |
5530 | */ | |
1da177e4 LT |
5531 | do { |
5532 | add_preempt_count(PREEMPT_ACTIVE); | |
5533 | schedule(); | |
5534 | sub_preempt_count(PREEMPT_ACTIVE); | |
5535 | } while (need_resched()); | |
5536 | } | |
5537 | ||
02b67cc3 | 5538 | int __sched _cond_resched(void) |
1da177e4 | 5539 | { |
9414232f IM |
5540 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5541 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5542 | __cond_resched(); |
5543 | return 1; | |
5544 | } | |
5545 | return 0; | |
5546 | } | |
02b67cc3 | 5547 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5548 | |
5549 | /* | |
5550 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5551 | * call schedule, and on return reacquire the lock. | |
5552 | * | |
41a2d6cf | 5553 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5554 | * operations here to prevent schedule() from being called twice (once via |
5555 | * spin_unlock(), once by hand). | |
5556 | */ | |
95cdf3b7 | 5557 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5558 | { |
95c354fe | 5559 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5560 | int ret = 0; |
5561 | ||
95c354fe | 5562 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5563 | spin_unlock(lock); |
95c354fe NP |
5564 | if (resched && need_resched()) |
5565 | __cond_resched(); | |
5566 | else | |
5567 | cpu_relax(); | |
6df3cecb | 5568 | ret = 1; |
1da177e4 | 5569 | spin_lock(lock); |
1da177e4 | 5570 | } |
6df3cecb | 5571 | return ret; |
1da177e4 | 5572 | } |
1da177e4 LT |
5573 | EXPORT_SYMBOL(cond_resched_lock); |
5574 | ||
5575 | int __sched cond_resched_softirq(void) | |
5576 | { | |
5577 | BUG_ON(!in_softirq()); | |
5578 | ||
9414232f | 5579 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5580 | local_bh_enable(); |
1da177e4 LT |
5581 | __cond_resched(); |
5582 | local_bh_disable(); | |
5583 | return 1; | |
5584 | } | |
5585 | return 0; | |
5586 | } | |
1da177e4 LT |
5587 | EXPORT_SYMBOL(cond_resched_softirq); |
5588 | ||
1da177e4 LT |
5589 | /** |
5590 | * yield - yield the current processor to other threads. | |
5591 | * | |
72fd4a35 | 5592 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5593 | * thread runnable and calls sys_sched_yield(). |
5594 | */ | |
5595 | void __sched yield(void) | |
5596 | { | |
5597 | set_current_state(TASK_RUNNING); | |
5598 | sys_sched_yield(); | |
5599 | } | |
1da177e4 LT |
5600 | EXPORT_SYMBOL(yield); |
5601 | ||
5602 | /* | |
41a2d6cf | 5603 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5604 | * that process accounting knows that this is a task in IO wait state. |
5605 | * | |
5606 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5607 | * has set its backing_dev_info: the queue against which it should throttle) | |
5608 | */ | |
5609 | void __sched io_schedule(void) | |
5610 | { | |
70b97a7f | 5611 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5612 | |
0ff92245 | 5613 | delayacct_blkio_start(); |
1da177e4 LT |
5614 | atomic_inc(&rq->nr_iowait); |
5615 | schedule(); | |
5616 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5617 | delayacct_blkio_end(); |
1da177e4 | 5618 | } |
1da177e4 LT |
5619 | EXPORT_SYMBOL(io_schedule); |
5620 | ||
5621 | long __sched io_schedule_timeout(long timeout) | |
5622 | { | |
70b97a7f | 5623 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5624 | long ret; |
5625 | ||
0ff92245 | 5626 | delayacct_blkio_start(); |
1da177e4 LT |
5627 | atomic_inc(&rq->nr_iowait); |
5628 | ret = schedule_timeout(timeout); | |
5629 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5630 | delayacct_blkio_end(); |
1da177e4 LT |
5631 | return ret; |
5632 | } | |
5633 | ||
5634 | /** | |
5635 | * sys_sched_get_priority_max - return maximum RT priority. | |
5636 | * @policy: scheduling class. | |
5637 | * | |
5638 | * this syscall returns the maximum rt_priority that can be used | |
5639 | * by a given scheduling class. | |
5640 | */ | |
5641 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5642 | { | |
5643 | int ret = -EINVAL; | |
5644 | ||
5645 | switch (policy) { | |
5646 | case SCHED_FIFO: | |
5647 | case SCHED_RR: | |
5648 | ret = MAX_USER_RT_PRIO-1; | |
5649 | break; | |
5650 | case SCHED_NORMAL: | |
b0a9499c | 5651 | case SCHED_BATCH: |
dd41f596 | 5652 | case SCHED_IDLE: |
1da177e4 LT |
5653 | ret = 0; |
5654 | break; | |
5655 | } | |
5656 | return ret; | |
5657 | } | |
5658 | ||
5659 | /** | |
5660 | * sys_sched_get_priority_min - return minimum RT priority. | |
5661 | * @policy: scheduling class. | |
5662 | * | |
5663 | * this syscall returns the minimum rt_priority that can be used | |
5664 | * by a given scheduling class. | |
5665 | */ | |
5666 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5667 | { | |
5668 | int ret = -EINVAL; | |
5669 | ||
5670 | switch (policy) { | |
5671 | case SCHED_FIFO: | |
5672 | case SCHED_RR: | |
5673 | ret = 1; | |
5674 | break; | |
5675 | case SCHED_NORMAL: | |
b0a9499c | 5676 | case SCHED_BATCH: |
dd41f596 | 5677 | case SCHED_IDLE: |
1da177e4 LT |
5678 | ret = 0; |
5679 | } | |
5680 | return ret; | |
5681 | } | |
5682 | ||
5683 | /** | |
5684 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5685 | * @pid: pid of the process. | |
5686 | * @interval: userspace pointer to the timeslice value. | |
5687 | * | |
5688 | * this syscall writes the default timeslice value of a given process | |
5689 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5690 | */ | |
5691 | asmlinkage | |
5692 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5693 | { | |
36c8b586 | 5694 | struct task_struct *p; |
a4ec24b4 | 5695 | unsigned int time_slice; |
3a5c359a | 5696 | int retval; |
1da177e4 | 5697 | struct timespec t; |
1da177e4 LT |
5698 | |
5699 | if (pid < 0) | |
3a5c359a | 5700 | return -EINVAL; |
1da177e4 LT |
5701 | |
5702 | retval = -ESRCH; | |
5703 | read_lock(&tasklist_lock); | |
5704 | p = find_process_by_pid(pid); | |
5705 | if (!p) | |
5706 | goto out_unlock; | |
5707 | ||
5708 | retval = security_task_getscheduler(p); | |
5709 | if (retval) | |
5710 | goto out_unlock; | |
5711 | ||
77034937 IM |
5712 | /* |
5713 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5714 | * tasks that are on an otherwise idle runqueue: | |
5715 | */ | |
5716 | time_slice = 0; | |
5717 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5718 | time_slice = DEF_TIMESLICE; |
1868f958 | 5719 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5720 | struct sched_entity *se = &p->se; |
5721 | unsigned long flags; | |
5722 | struct rq *rq; | |
5723 | ||
5724 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5725 | if (rq->cfs.load.weight) |
5726 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5727 | task_rq_unlock(rq, &flags); |
5728 | } | |
1da177e4 | 5729 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5730 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5731 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5732 | return retval; |
3a5c359a | 5733 | |
1da177e4 LT |
5734 | out_unlock: |
5735 | read_unlock(&tasklist_lock); | |
5736 | return retval; | |
5737 | } | |
5738 | ||
7c731e0a | 5739 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5740 | |
82a1fcb9 | 5741 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5742 | { |
1da177e4 | 5743 | unsigned long free = 0; |
36c8b586 | 5744 | unsigned state; |
1da177e4 | 5745 | |
1da177e4 | 5746 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5747 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5748 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5749 | #if BITS_PER_LONG == 32 |
1da177e4 | 5750 | if (state == TASK_RUNNING) |
cc4ea795 | 5751 | printk(KERN_CONT " running "); |
1da177e4 | 5752 | else |
cc4ea795 | 5753 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5754 | #else |
5755 | if (state == TASK_RUNNING) | |
cc4ea795 | 5756 | printk(KERN_CONT " running task "); |
1da177e4 | 5757 | else |
cc4ea795 | 5758 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5759 | #endif |
5760 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5761 | { | |
10ebffde | 5762 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5763 | while (!*n) |
5764 | n++; | |
10ebffde | 5765 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5766 | } |
5767 | #endif | |
ba25f9dc | 5768 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5769 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5770 | |
5fb5e6de | 5771 | show_stack(p, NULL); |
1da177e4 LT |
5772 | } |
5773 | ||
e59e2ae2 | 5774 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5775 | { |
36c8b586 | 5776 | struct task_struct *g, *p; |
1da177e4 | 5777 | |
4bd77321 IM |
5778 | #if BITS_PER_LONG == 32 |
5779 | printk(KERN_INFO | |
5780 | " task PC stack pid father\n"); | |
1da177e4 | 5781 | #else |
4bd77321 IM |
5782 | printk(KERN_INFO |
5783 | " task PC stack pid father\n"); | |
1da177e4 LT |
5784 | #endif |
5785 | read_lock(&tasklist_lock); | |
5786 | do_each_thread(g, p) { | |
5787 | /* | |
5788 | * reset the NMI-timeout, listing all files on a slow | |
5789 | * console might take alot of time: | |
5790 | */ | |
5791 | touch_nmi_watchdog(); | |
39bc89fd | 5792 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5793 | sched_show_task(p); |
1da177e4 LT |
5794 | } while_each_thread(g, p); |
5795 | ||
04c9167f JF |
5796 | touch_all_softlockup_watchdogs(); |
5797 | ||
dd41f596 IM |
5798 | #ifdef CONFIG_SCHED_DEBUG |
5799 | sysrq_sched_debug_show(); | |
5800 | #endif | |
1da177e4 | 5801 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5802 | /* |
5803 | * Only show locks if all tasks are dumped: | |
5804 | */ | |
5805 | if (state_filter == -1) | |
5806 | debug_show_all_locks(); | |
1da177e4 LT |
5807 | } |
5808 | ||
1df21055 IM |
5809 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5810 | { | |
dd41f596 | 5811 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5812 | } |
5813 | ||
f340c0d1 IM |
5814 | /** |
5815 | * init_idle - set up an idle thread for a given CPU | |
5816 | * @idle: task in question | |
5817 | * @cpu: cpu the idle task belongs to | |
5818 | * | |
5819 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5820 | * flag, to make booting more robust. | |
5821 | */ | |
5c1e1767 | 5822 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5823 | { |
70b97a7f | 5824 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5825 | unsigned long flags; |
5826 | ||
dd41f596 IM |
5827 | __sched_fork(idle); |
5828 | idle->se.exec_start = sched_clock(); | |
5829 | ||
b29739f9 | 5830 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5831 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5832 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5833 | |
5834 | spin_lock_irqsave(&rq->lock, flags); | |
5835 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5836 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5837 | idle->oncpu = 1; | |
5838 | #endif | |
1da177e4 LT |
5839 | spin_unlock_irqrestore(&rq->lock, flags); |
5840 | ||
5841 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5842 | #if defined(CONFIG_PREEMPT) |
5843 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5844 | #else | |
a1261f54 | 5845 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5846 | #endif |
dd41f596 IM |
5847 | /* |
5848 | * The idle tasks have their own, simple scheduling class: | |
5849 | */ | |
5850 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5851 | } |
5852 | ||
5853 | /* | |
5854 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5855 | * indicates which cpus entered this state. This is used | |
5856 | * in the rcu update to wait only for active cpus. For system | |
5857 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5858 | * always be CPU_MASK_NONE. | |
5859 | */ | |
5860 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5861 | ||
19978ca6 IM |
5862 | /* |
5863 | * Increase the granularity value when there are more CPUs, | |
5864 | * because with more CPUs the 'effective latency' as visible | |
5865 | * to users decreases. But the relationship is not linear, | |
5866 | * so pick a second-best guess by going with the log2 of the | |
5867 | * number of CPUs. | |
5868 | * | |
5869 | * This idea comes from the SD scheduler of Con Kolivas: | |
5870 | */ | |
5871 | static inline void sched_init_granularity(void) | |
5872 | { | |
5873 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5874 | const unsigned long limit = 200000000; | |
5875 | ||
5876 | sysctl_sched_min_granularity *= factor; | |
5877 | if (sysctl_sched_min_granularity > limit) | |
5878 | sysctl_sched_min_granularity = limit; | |
5879 | ||
5880 | sysctl_sched_latency *= factor; | |
5881 | if (sysctl_sched_latency > limit) | |
5882 | sysctl_sched_latency = limit; | |
5883 | ||
5884 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5885 | |
5886 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5887 | } |
5888 | ||
1da177e4 LT |
5889 | #ifdef CONFIG_SMP |
5890 | /* | |
5891 | * This is how migration works: | |
5892 | * | |
70b97a7f | 5893 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5894 | * runqueue and wake up that CPU's migration thread. |
5895 | * 2) we down() the locked semaphore => thread blocks. | |
5896 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5897 | * thread off the CPU) | |
5898 | * 4) it gets the migration request and checks whether the migrated | |
5899 | * task is still in the wrong runqueue. | |
5900 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5901 | * it and puts it into the right queue. | |
5902 | * 6) migration thread up()s the semaphore. | |
5903 | * 7) we wake up and the migration is done. | |
5904 | */ | |
5905 | ||
5906 | /* | |
5907 | * Change a given task's CPU affinity. Migrate the thread to a | |
5908 | * proper CPU and schedule it away if the CPU it's executing on | |
5909 | * is removed from the allowed bitmask. | |
5910 | * | |
5911 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5912 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5913 | * call is not atomic; no spinlocks may be held. |
5914 | */ | |
cd8ba7cd | 5915 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5916 | { |
70b97a7f | 5917 | struct migration_req req; |
1da177e4 | 5918 | unsigned long flags; |
70b97a7f | 5919 | struct rq *rq; |
48f24c4d | 5920 | int ret = 0; |
1da177e4 LT |
5921 | |
5922 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5923 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5924 | ret = -EINVAL; |
5925 | goto out; | |
5926 | } | |
5927 | ||
9985b0ba DR |
5928 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5929 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5930 | ret = -EINVAL; | |
5931 | goto out; | |
5932 | } | |
5933 | ||
73fe6aae | 5934 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5935 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5936 | else { |
cd8ba7cd MT |
5937 | p->cpus_allowed = *new_mask; |
5938 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5939 | } |
5940 | ||
1da177e4 | 5941 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5942 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5943 | goto out; |
5944 | ||
cd8ba7cd | 5945 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5946 | /* Need help from migration thread: drop lock and wait. */ |
5947 | task_rq_unlock(rq, &flags); | |
5948 | wake_up_process(rq->migration_thread); | |
5949 | wait_for_completion(&req.done); | |
5950 | tlb_migrate_finish(p->mm); | |
5951 | return 0; | |
5952 | } | |
5953 | out: | |
5954 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5955 | |
1da177e4 LT |
5956 | return ret; |
5957 | } | |
cd8ba7cd | 5958 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5959 | |
5960 | /* | |
41a2d6cf | 5961 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5962 | * this because either it can't run here any more (set_cpus_allowed() |
5963 | * away from this CPU, or CPU going down), or because we're | |
5964 | * attempting to rebalance this task on exec (sched_exec). | |
5965 | * | |
5966 | * So we race with normal scheduler movements, but that's OK, as long | |
5967 | * as the task is no longer on this CPU. | |
efc30814 KK |
5968 | * |
5969 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5970 | */ |
efc30814 | 5971 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5972 | { |
70b97a7f | 5973 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5974 | int ret = 0, on_rq; |
1da177e4 | 5975 | |
e761b772 | 5976 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5977 | return ret; |
1da177e4 LT |
5978 | |
5979 | rq_src = cpu_rq(src_cpu); | |
5980 | rq_dest = cpu_rq(dest_cpu); | |
5981 | ||
5982 | double_rq_lock(rq_src, rq_dest); | |
5983 | /* Already moved. */ | |
5984 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5985 | goto done; |
1da177e4 LT |
5986 | /* Affinity changed (again). */ |
5987 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 5988 | goto fail; |
1da177e4 | 5989 | |
dd41f596 | 5990 | on_rq = p->se.on_rq; |
6e82a3be | 5991 | if (on_rq) |
2e1cb74a | 5992 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5993 | |
1da177e4 | 5994 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5995 | if (on_rq) { |
5996 | activate_task(rq_dest, p, 0); | |
5997 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5998 | } |
b1e38734 | 5999 | done: |
efc30814 | 6000 | ret = 1; |
b1e38734 | 6001 | fail: |
1da177e4 | 6002 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6003 | return ret; |
1da177e4 LT |
6004 | } |
6005 | ||
6006 | /* | |
6007 | * migration_thread - this is a highprio system thread that performs | |
6008 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6009 | * another runqueue. | |
6010 | */ | |
95cdf3b7 | 6011 | static int migration_thread(void *data) |
1da177e4 | 6012 | { |
1da177e4 | 6013 | int cpu = (long)data; |
70b97a7f | 6014 | struct rq *rq; |
1da177e4 LT |
6015 | |
6016 | rq = cpu_rq(cpu); | |
6017 | BUG_ON(rq->migration_thread != current); | |
6018 | ||
6019 | set_current_state(TASK_INTERRUPTIBLE); | |
6020 | while (!kthread_should_stop()) { | |
70b97a7f | 6021 | struct migration_req *req; |
1da177e4 | 6022 | struct list_head *head; |
1da177e4 | 6023 | |
1da177e4 LT |
6024 | spin_lock_irq(&rq->lock); |
6025 | ||
6026 | if (cpu_is_offline(cpu)) { | |
6027 | spin_unlock_irq(&rq->lock); | |
6028 | goto wait_to_die; | |
6029 | } | |
6030 | ||
6031 | if (rq->active_balance) { | |
6032 | active_load_balance(rq, cpu); | |
6033 | rq->active_balance = 0; | |
6034 | } | |
6035 | ||
6036 | head = &rq->migration_queue; | |
6037 | ||
6038 | if (list_empty(head)) { | |
6039 | spin_unlock_irq(&rq->lock); | |
6040 | schedule(); | |
6041 | set_current_state(TASK_INTERRUPTIBLE); | |
6042 | continue; | |
6043 | } | |
70b97a7f | 6044 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6045 | list_del_init(head->next); |
6046 | ||
674311d5 NP |
6047 | spin_unlock(&rq->lock); |
6048 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6049 | local_irq_enable(); | |
1da177e4 LT |
6050 | |
6051 | complete(&req->done); | |
6052 | } | |
6053 | __set_current_state(TASK_RUNNING); | |
6054 | return 0; | |
6055 | ||
6056 | wait_to_die: | |
6057 | /* Wait for kthread_stop */ | |
6058 | set_current_state(TASK_INTERRUPTIBLE); | |
6059 | while (!kthread_should_stop()) { | |
6060 | schedule(); | |
6061 | set_current_state(TASK_INTERRUPTIBLE); | |
6062 | } | |
6063 | __set_current_state(TASK_RUNNING); | |
6064 | return 0; | |
6065 | } | |
6066 | ||
6067 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6068 | |
6069 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6070 | { | |
6071 | int ret; | |
6072 | ||
6073 | local_irq_disable(); | |
6074 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6075 | local_irq_enable(); | |
6076 | return ret; | |
6077 | } | |
6078 | ||
054b9108 | 6079 | /* |
3a4fa0a2 | 6080 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6081 | * NOTE: interrupts should be disabled by the caller |
6082 | */ | |
48f24c4d | 6083 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6084 | { |
efc30814 | 6085 | unsigned long flags; |
1da177e4 | 6086 | cpumask_t mask; |
70b97a7f IM |
6087 | struct rq *rq; |
6088 | int dest_cpu; | |
1da177e4 | 6089 | |
3a5c359a AK |
6090 | do { |
6091 | /* On same node? */ | |
6092 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6093 | cpus_and(mask, mask, p->cpus_allowed); | |
6094 | dest_cpu = any_online_cpu(mask); | |
6095 | ||
6096 | /* On any allowed CPU? */ | |
434d53b0 | 6097 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6098 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6099 | ||
6100 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6101 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6102 | cpumask_t cpus_allowed; |
6103 | ||
6104 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6105 | /* |
6106 | * Try to stay on the same cpuset, where the | |
6107 | * current cpuset may be a subset of all cpus. | |
6108 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6109 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6110 | * called within calls to cpuset_lock/cpuset_unlock. |
6111 | */ | |
3a5c359a | 6112 | rq = task_rq_lock(p, &flags); |
470fd646 | 6113 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6114 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6115 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6116 | |
3a5c359a AK |
6117 | /* |
6118 | * Don't tell them about moving exiting tasks or | |
6119 | * kernel threads (both mm NULL), since they never | |
6120 | * leave kernel. | |
6121 | */ | |
41a2d6cf | 6122 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6123 | printk(KERN_INFO "process %d (%s) no " |
6124 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6125 | task_pid_nr(p), p->comm, dead_cpu); |
6126 | } | |
3a5c359a | 6127 | } |
f7b4cddc | 6128 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6129 | } |
6130 | ||
6131 | /* | |
6132 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6133 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6134 | * for performance reasons the counter is not stricly tracking tasks to | |
6135 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6136 | * to keep the global sum constant after CPU-down: | |
6137 | */ | |
70b97a7f | 6138 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6139 | { |
7c16ec58 | 6140 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6141 | unsigned long flags; |
6142 | ||
6143 | local_irq_save(flags); | |
6144 | double_rq_lock(rq_src, rq_dest); | |
6145 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6146 | rq_src->nr_uninterruptible = 0; | |
6147 | double_rq_unlock(rq_src, rq_dest); | |
6148 | local_irq_restore(flags); | |
6149 | } | |
6150 | ||
6151 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6152 | static void migrate_live_tasks(int src_cpu) | |
6153 | { | |
48f24c4d | 6154 | struct task_struct *p, *t; |
1da177e4 | 6155 | |
f7b4cddc | 6156 | read_lock(&tasklist_lock); |
1da177e4 | 6157 | |
48f24c4d IM |
6158 | do_each_thread(t, p) { |
6159 | if (p == current) | |
1da177e4 LT |
6160 | continue; |
6161 | ||
48f24c4d IM |
6162 | if (task_cpu(p) == src_cpu) |
6163 | move_task_off_dead_cpu(src_cpu, p); | |
6164 | } while_each_thread(t, p); | |
1da177e4 | 6165 | |
f7b4cddc | 6166 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6167 | } |
6168 | ||
dd41f596 IM |
6169 | /* |
6170 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6171 | * It does so by boosting its priority to highest possible. |
6172 | * Used by CPU offline code. | |
1da177e4 LT |
6173 | */ |
6174 | void sched_idle_next(void) | |
6175 | { | |
48f24c4d | 6176 | int this_cpu = smp_processor_id(); |
70b97a7f | 6177 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6178 | struct task_struct *p = rq->idle; |
6179 | unsigned long flags; | |
6180 | ||
6181 | /* cpu has to be offline */ | |
48f24c4d | 6182 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6183 | |
48f24c4d IM |
6184 | /* |
6185 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6186 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6187 | */ |
6188 | spin_lock_irqsave(&rq->lock, flags); | |
6189 | ||
dd41f596 | 6190 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6191 | |
94bc9a7b DA |
6192 | update_rq_clock(rq); |
6193 | activate_task(rq, p, 0); | |
1da177e4 LT |
6194 | |
6195 | spin_unlock_irqrestore(&rq->lock, flags); | |
6196 | } | |
6197 | ||
48f24c4d IM |
6198 | /* |
6199 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6200 | * offline. |
6201 | */ | |
6202 | void idle_task_exit(void) | |
6203 | { | |
6204 | struct mm_struct *mm = current->active_mm; | |
6205 | ||
6206 | BUG_ON(cpu_online(smp_processor_id())); | |
6207 | ||
6208 | if (mm != &init_mm) | |
6209 | switch_mm(mm, &init_mm, current); | |
6210 | mmdrop(mm); | |
6211 | } | |
6212 | ||
054b9108 | 6213 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6214 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6215 | { |
70b97a7f | 6216 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6217 | |
6218 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6219 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6220 | |
6221 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6222 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6223 | |
48f24c4d | 6224 | get_task_struct(p); |
1da177e4 LT |
6225 | |
6226 | /* | |
6227 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6228 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6229 | * fine. |
6230 | */ | |
f7b4cddc | 6231 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6232 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6233 | spin_lock_irq(&rq->lock); |
1da177e4 | 6234 | |
48f24c4d | 6235 | put_task_struct(p); |
1da177e4 LT |
6236 | } |
6237 | ||
6238 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6239 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6240 | { | |
70b97a7f | 6241 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6242 | struct task_struct *next; |
48f24c4d | 6243 | |
dd41f596 IM |
6244 | for ( ; ; ) { |
6245 | if (!rq->nr_running) | |
6246 | break; | |
a8e504d2 | 6247 | update_rq_clock(rq); |
ff95f3df | 6248 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6249 | if (!next) |
6250 | break; | |
79c53799 | 6251 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6252 | migrate_dead(dead_cpu, next); |
e692ab53 | 6253 | |
1da177e4 LT |
6254 | } |
6255 | } | |
6256 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6257 | ||
e692ab53 NP |
6258 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6259 | ||
6260 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6261 | { |
6262 | .procname = "sched_domain", | |
c57baf1e | 6263 | .mode = 0555, |
e0361851 | 6264 | }, |
38605cae | 6265 | {0, }, |
e692ab53 NP |
6266 | }; |
6267 | ||
6268 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6269 | { |
c57baf1e | 6270 | .ctl_name = CTL_KERN, |
e0361851 | 6271 | .procname = "kernel", |
c57baf1e | 6272 | .mode = 0555, |
e0361851 AD |
6273 | .child = sd_ctl_dir, |
6274 | }, | |
38605cae | 6275 | {0, }, |
e692ab53 NP |
6276 | }; |
6277 | ||
6278 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6279 | { | |
6280 | struct ctl_table *entry = | |
5cf9f062 | 6281 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6282 | |
e692ab53 NP |
6283 | return entry; |
6284 | } | |
6285 | ||
6382bc90 MM |
6286 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6287 | { | |
cd790076 | 6288 | struct ctl_table *entry; |
6382bc90 | 6289 | |
cd790076 MM |
6290 | /* |
6291 | * In the intermediate directories, both the child directory and | |
6292 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6293 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6294 | * static strings and all have proc handlers. |
6295 | */ | |
6296 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6297 | if (entry->child) |
6298 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6299 | if (entry->proc_handler == NULL) |
6300 | kfree(entry->procname); | |
6301 | } | |
6382bc90 MM |
6302 | |
6303 | kfree(*tablep); | |
6304 | *tablep = NULL; | |
6305 | } | |
6306 | ||
e692ab53 | 6307 | static void |
e0361851 | 6308 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6309 | const char *procname, void *data, int maxlen, |
6310 | mode_t mode, proc_handler *proc_handler) | |
6311 | { | |
e692ab53 NP |
6312 | entry->procname = procname; |
6313 | entry->data = data; | |
6314 | entry->maxlen = maxlen; | |
6315 | entry->mode = mode; | |
6316 | entry->proc_handler = proc_handler; | |
6317 | } | |
6318 | ||
6319 | static struct ctl_table * | |
6320 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6321 | { | |
ace8b3d6 | 6322 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6323 | |
ad1cdc1d MM |
6324 | if (table == NULL) |
6325 | return NULL; | |
6326 | ||
e0361851 | 6327 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6328 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6329 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6330 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6331 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6332 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6333 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6334 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6335 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6336 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6337 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6338 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6339 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6340 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6341 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6342 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6343 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6344 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6345 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6346 | &sd->cache_nice_tries, |
6347 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6348 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6349 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6350 | /* &table[11] is terminator */ |
e692ab53 NP |
6351 | |
6352 | return table; | |
6353 | } | |
6354 | ||
9a4e7159 | 6355 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6356 | { |
6357 | struct ctl_table *entry, *table; | |
6358 | struct sched_domain *sd; | |
6359 | int domain_num = 0, i; | |
6360 | char buf[32]; | |
6361 | ||
6362 | for_each_domain(cpu, sd) | |
6363 | domain_num++; | |
6364 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6365 | if (table == NULL) |
6366 | return NULL; | |
e692ab53 NP |
6367 | |
6368 | i = 0; | |
6369 | for_each_domain(cpu, sd) { | |
6370 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6371 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6372 | entry->mode = 0555; |
e692ab53 NP |
6373 | entry->child = sd_alloc_ctl_domain_table(sd); |
6374 | entry++; | |
6375 | i++; | |
6376 | } | |
6377 | return table; | |
6378 | } | |
6379 | ||
6380 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6381 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6382 | { |
6383 | int i, cpu_num = num_online_cpus(); | |
6384 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6385 | char buf[32]; | |
6386 | ||
7378547f MM |
6387 | WARN_ON(sd_ctl_dir[0].child); |
6388 | sd_ctl_dir[0].child = entry; | |
6389 | ||
ad1cdc1d MM |
6390 | if (entry == NULL) |
6391 | return; | |
6392 | ||
97b6ea7b | 6393 | for_each_online_cpu(i) { |
e692ab53 | 6394 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6395 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6396 | entry->mode = 0555; |
e692ab53 | 6397 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6398 | entry++; |
e692ab53 | 6399 | } |
7378547f MM |
6400 | |
6401 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6402 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6403 | } | |
6382bc90 | 6404 | |
7378547f | 6405 | /* may be called multiple times per register */ |
6382bc90 MM |
6406 | static void unregister_sched_domain_sysctl(void) |
6407 | { | |
7378547f MM |
6408 | if (sd_sysctl_header) |
6409 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6410 | sd_sysctl_header = NULL; |
7378547f MM |
6411 | if (sd_ctl_dir[0].child) |
6412 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6413 | } |
e692ab53 | 6414 | #else |
6382bc90 MM |
6415 | static void register_sched_domain_sysctl(void) |
6416 | { | |
6417 | } | |
6418 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6419 | { |
6420 | } | |
6421 | #endif | |
6422 | ||
1f11eb6a GH |
6423 | static void set_rq_online(struct rq *rq) |
6424 | { | |
6425 | if (!rq->online) { | |
6426 | const struct sched_class *class; | |
6427 | ||
6428 | cpu_set(rq->cpu, rq->rd->online); | |
6429 | rq->online = 1; | |
6430 | ||
6431 | for_each_class(class) { | |
6432 | if (class->rq_online) | |
6433 | class->rq_online(rq); | |
6434 | } | |
6435 | } | |
6436 | } | |
6437 | ||
6438 | static void set_rq_offline(struct rq *rq) | |
6439 | { | |
6440 | if (rq->online) { | |
6441 | const struct sched_class *class; | |
6442 | ||
6443 | for_each_class(class) { | |
6444 | if (class->rq_offline) | |
6445 | class->rq_offline(rq); | |
6446 | } | |
6447 | ||
6448 | cpu_clear(rq->cpu, rq->rd->online); | |
6449 | rq->online = 0; | |
6450 | } | |
6451 | } | |
6452 | ||
1da177e4 LT |
6453 | /* |
6454 | * migration_call - callback that gets triggered when a CPU is added. | |
6455 | * Here we can start up the necessary migration thread for the new CPU. | |
6456 | */ | |
48f24c4d IM |
6457 | static int __cpuinit |
6458 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6459 | { |
1da177e4 | 6460 | struct task_struct *p; |
48f24c4d | 6461 | int cpu = (long)hcpu; |
1da177e4 | 6462 | unsigned long flags; |
70b97a7f | 6463 | struct rq *rq; |
1da177e4 LT |
6464 | |
6465 | switch (action) { | |
5be9361c | 6466 | |
1da177e4 | 6467 | case CPU_UP_PREPARE: |
8bb78442 | 6468 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6469 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6470 | if (IS_ERR(p)) |
6471 | return NOTIFY_BAD; | |
1da177e4 LT |
6472 | kthread_bind(p, cpu); |
6473 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6474 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6475 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6476 | task_rq_unlock(rq, &flags); |
6477 | cpu_rq(cpu)->migration_thread = p; | |
6478 | break; | |
48f24c4d | 6479 | |
1da177e4 | 6480 | case CPU_ONLINE: |
8bb78442 | 6481 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6482 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6483 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6484 | |
6485 | /* Update our root-domain */ | |
6486 | rq = cpu_rq(cpu); | |
6487 | spin_lock_irqsave(&rq->lock, flags); | |
6488 | if (rq->rd) { | |
6489 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6490 | |
6491 | set_rq_online(rq); | |
1f94ef59 GH |
6492 | } |
6493 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6494 | break; |
48f24c4d | 6495 | |
1da177e4 LT |
6496 | #ifdef CONFIG_HOTPLUG_CPU |
6497 | case CPU_UP_CANCELED: | |
8bb78442 | 6498 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6499 | if (!cpu_rq(cpu)->migration_thread) |
6500 | break; | |
41a2d6cf | 6501 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6502 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6503 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6504 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6505 | cpu_rq(cpu)->migration_thread = NULL; | |
6506 | break; | |
48f24c4d | 6507 | |
1da177e4 | 6508 | case CPU_DEAD: |
8bb78442 | 6509 | case CPU_DEAD_FROZEN: |
470fd646 | 6510 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6511 | migrate_live_tasks(cpu); |
6512 | rq = cpu_rq(cpu); | |
6513 | kthread_stop(rq->migration_thread); | |
6514 | rq->migration_thread = NULL; | |
6515 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6516 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6517 | update_rq_clock(rq); |
2e1cb74a | 6518 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6519 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6520 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6521 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6522 | migrate_dead_tasks(cpu); |
d2da272a | 6523 | spin_unlock_irq(&rq->lock); |
470fd646 | 6524 | cpuset_unlock(); |
1da177e4 LT |
6525 | migrate_nr_uninterruptible(rq); |
6526 | BUG_ON(rq->nr_running != 0); | |
6527 | ||
41a2d6cf IM |
6528 | /* |
6529 | * No need to migrate the tasks: it was best-effort if | |
6530 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6531 | * the requestors. | |
6532 | */ | |
1da177e4 LT |
6533 | spin_lock_irq(&rq->lock); |
6534 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6535 | struct migration_req *req; |
6536 | ||
1da177e4 | 6537 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6538 | struct migration_req, list); |
1da177e4 LT |
6539 | list_del_init(&req->list); |
6540 | complete(&req->done); | |
6541 | } | |
6542 | spin_unlock_irq(&rq->lock); | |
6543 | break; | |
57d885fe | 6544 | |
08f503b0 GH |
6545 | case CPU_DYING: |
6546 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6547 | /* Update our root-domain */ |
6548 | rq = cpu_rq(cpu); | |
6549 | spin_lock_irqsave(&rq->lock, flags); | |
6550 | if (rq->rd) { | |
6551 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6552 | set_rq_offline(rq); |
57d885fe GH |
6553 | } |
6554 | spin_unlock_irqrestore(&rq->lock, flags); | |
6555 | break; | |
1da177e4 LT |
6556 | #endif |
6557 | } | |
6558 | return NOTIFY_OK; | |
6559 | } | |
6560 | ||
6561 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6562 | * happens before everything else. | |
6563 | */ | |
26c2143b | 6564 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6565 | .notifier_call = migration_call, |
6566 | .priority = 10 | |
6567 | }; | |
6568 | ||
7babe8db | 6569 | static int __init migration_init(void) |
1da177e4 LT |
6570 | { |
6571 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6572 | int err; |
48f24c4d IM |
6573 | |
6574 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6575 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6576 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6577 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6578 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6579 | |
6580 | return err; | |
1da177e4 | 6581 | } |
7babe8db | 6582 | early_initcall(migration_init); |
1da177e4 LT |
6583 | #endif |
6584 | ||
6585 | #ifdef CONFIG_SMP | |
476f3534 | 6586 | |
3e9830dc | 6587 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6588 | |
099f98c8 GS |
6589 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6590 | { | |
6591 | switch (lvl) { | |
6592 | case SD_LV_NONE: | |
6593 | return "NONE"; | |
6594 | case SD_LV_SIBLING: | |
6595 | return "SIBLING"; | |
6596 | case SD_LV_MC: | |
6597 | return "MC"; | |
6598 | case SD_LV_CPU: | |
6599 | return "CPU"; | |
6600 | case SD_LV_NODE: | |
6601 | return "NODE"; | |
6602 | case SD_LV_ALLNODES: | |
6603 | return "ALLNODES"; | |
6604 | case SD_LV_MAX: | |
6605 | return "MAX"; | |
6606 | ||
6607 | } | |
6608 | return "MAX"; | |
6609 | } | |
6610 | ||
7c16ec58 MT |
6611 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6612 | cpumask_t *groupmask) | |
1da177e4 | 6613 | { |
4dcf6aff | 6614 | struct sched_group *group = sd->groups; |
434d53b0 | 6615 | char str[256]; |
1da177e4 | 6616 | |
434d53b0 | 6617 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6618 | cpus_clear(*groupmask); |
4dcf6aff IM |
6619 | |
6620 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6621 | ||
6622 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6623 | printk("does not load-balance\n"); | |
6624 | if (sd->parent) | |
6625 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6626 | " has parent"); | |
6627 | return -1; | |
41c7ce9a NP |
6628 | } |
6629 | ||
099f98c8 GS |
6630 | printk(KERN_CONT "span %s level %s\n", |
6631 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6632 | |
6633 | if (!cpu_isset(cpu, sd->span)) { | |
6634 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6635 | "CPU%d\n", cpu); | |
6636 | } | |
6637 | if (!cpu_isset(cpu, group->cpumask)) { | |
6638 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6639 | " CPU%d\n", cpu); | |
6640 | } | |
1da177e4 | 6641 | |
4dcf6aff | 6642 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6643 | do { |
4dcf6aff IM |
6644 | if (!group) { |
6645 | printk("\n"); | |
6646 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6647 | break; |
6648 | } | |
6649 | ||
4dcf6aff IM |
6650 | if (!group->__cpu_power) { |
6651 | printk(KERN_CONT "\n"); | |
6652 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6653 | "set\n"); | |
6654 | break; | |
6655 | } | |
1da177e4 | 6656 | |
4dcf6aff IM |
6657 | if (!cpus_weight(group->cpumask)) { |
6658 | printk(KERN_CONT "\n"); | |
6659 | printk(KERN_ERR "ERROR: empty group\n"); | |
6660 | break; | |
6661 | } | |
1da177e4 | 6662 | |
7c16ec58 | 6663 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6664 | printk(KERN_CONT "\n"); |
6665 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6666 | break; | |
6667 | } | |
1da177e4 | 6668 | |
7c16ec58 | 6669 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6670 | |
434d53b0 | 6671 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6672 | printk(KERN_CONT " %s", str); |
1da177e4 | 6673 | |
4dcf6aff IM |
6674 | group = group->next; |
6675 | } while (group != sd->groups); | |
6676 | printk(KERN_CONT "\n"); | |
1da177e4 | 6677 | |
7c16ec58 | 6678 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6679 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6680 | |
7c16ec58 | 6681 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6682 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6683 | "of domain->span\n"); | |
6684 | return 0; | |
6685 | } | |
1da177e4 | 6686 | |
4dcf6aff IM |
6687 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6688 | { | |
7c16ec58 | 6689 | cpumask_t *groupmask; |
4dcf6aff | 6690 | int level = 0; |
1da177e4 | 6691 | |
4dcf6aff IM |
6692 | if (!sd) { |
6693 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6694 | return; | |
6695 | } | |
1da177e4 | 6696 | |
4dcf6aff IM |
6697 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6698 | ||
7c16ec58 MT |
6699 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6700 | if (!groupmask) { | |
6701 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6702 | return; | |
6703 | } | |
6704 | ||
4dcf6aff | 6705 | for (;;) { |
7c16ec58 | 6706 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6707 | break; |
1da177e4 LT |
6708 | level++; |
6709 | sd = sd->parent; | |
33859f7f | 6710 | if (!sd) |
4dcf6aff IM |
6711 | break; |
6712 | } | |
7c16ec58 | 6713 | kfree(groupmask); |
1da177e4 | 6714 | } |
6d6bc0ad | 6715 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6716 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6717 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6718 | |
1a20ff27 | 6719 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6720 | { |
6721 | if (cpus_weight(sd->span) == 1) | |
6722 | return 1; | |
6723 | ||
6724 | /* Following flags need at least 2 groups */ | |
6725 | if (sd->flags & (SD_LOAD_BALANCE | | |
6726 | SD_BALANCE_NEWIDLE | | |
6727 | SD_BALANCE_FORK | | |
89c4710e SS |
6728 | SD_BALANCE_EXEC | |
6729 | SD_SHARE_CPUPOWER | | |
6730 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6731 | if (sd->groups != sd->groups->next) |
6732 | return 0; | |
6733 | } | |
6734 | ||
6735 | /* Following flags don't use groups */ | |
6736 | if (sd->flags & (SD_WAKE_IDLE | | |
6737 | SD_WAKE_AFFINE | | |
6738 | SD_WAKE_BALANCE)) | |
6739 | return 0; | |
6740 | ||
6741 | return 1; | |
6742 | } | |
6743 | ||
48f24c4d IM |
6744 | static int |
6745 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6746 | { |
6747 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6748 | ||
6749 | if (sd_degenerate(parent)) | |
6750 | return 1; | |
6751 | ||
6752 | if (!cpus_equal(sd->span, parent->span)) | |
6753 | return 0; | |
6754 | ||
6755 | /* Does parent contain flags not in child? */ | |
6756 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6757 | if (cflags & SD_WAKE_AFFINE) | |
6758 | pflags &= ~SD_WAKE_BALANCE; | |
6759 | /* Flags needing groups don't count if only 1 group in parent */ | |
6760 | if (parent->groups == parent->groups->next) { | |
6761 | pflags &= ~(SD_LOAD_BALANCE | | |
6762 | SD_BALANCE_NEWIDLE | | |
6763 | SD_BALANCE_FORK | | |
89c4710e SS |
6764 | SD_BALANCE_EXEC | |
6765 | SD_SHARE_CPUPOWER | | |
6766 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6767 | } |
6768 | if (~cflags & pflags) | |
6769 | return 0; | |
6770 | ||
6771 | return 1; | |
6772 | } | |
6773 | ||
57d885fe GH |
6774 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6775 | { | |
6776 | unsigned long flags; | |
57d885fe GH |
6777 | |
6778 | spin_lock_irqsave(&rq->lock, flags); | |
6779 | ||
6780 | if (rq->rd) { | |
6781 | struct root_domain *old_rd = rq->rd; | |
6782 | ||
1f11eb6a GH |
6783 | if (cpu_isset(rq->cpu, old_rd->online)) |
6784 | set_rq_offline(rq); | |
57d885fe | 6785 | |
dc938520 | 6786 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6787 | |
57d885fe GH |
6788 | if (atomic_dec_and_test(&old_rd->refcount)) |
6789 | kfree(old_rd); | |
6790 | } | |
6791 | ||
6792 | atomic_inc(&rd->refcount); | |
6793 | rq->rd = rd; | |
6794 | ||
dc938520 | 6795 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6796 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6797 | set_rq_online(rq); |
57d885fe GH |
6798 | |
6799 | spin_unlock_irqrestore(&rq->lock, flags); | |
6800 | } | |
6801 | ||
dc938520 | 6802 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6803 | { |
6804 | memset(rd, 0, sizeof(*rd)); | |
6805 | ||
dc938520 GH |
6806 | cpus_clear(rd->span); |
6807 | cpus_clear(rd->online); | |
6e0534f2 GH |
6808 | |
6809 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6810 | } |
6811 | ||
6812 | static void init_defrootdomain(void) | |
6813 | { | |
dc938520 | 6814 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6815 | atomic_set(&def_root_domain.refcount, 1); |
6816 | } | |
6817 | ||
dc938520 | 6818 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6819 | { |
6820 | struct root_domain *rd; | |
6821 | ||
6822 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6823 | if (!rd) | |
6824 | return NULL; | |
6825 | ||
dc938520 | 6826 | init_rootdomain(rd); |
57d885fe GH |
6827 | |
6828 | return rd; | |
6829 | } | |
6830 | ||
1da177e4 | 6831 | /* |
0eab9146 | 6832 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6833 | * hold the hotplug lock. |
6834 | */ | |
0eab9146 IM |
6835 | static void |
6836 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6837 | { |
70b97a7f | 6838 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6839 | struct sched_domain *tmp; |
6840 | ||
6841 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6842 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6843 | struct sched_domain *parent = tmp->parent; | |
6844 | if (!parent) | |
6845 | break; | |
1a848870 | 6846 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6847 | tmp->parent = parent->parent; |
1a848870 SS |
6848 | if (parent->parent) |
6849 | parent->parent->child = tmp; | |
6850 | } | |
245af2c7 SS |
6851 | } |
6852 | ||
1a848870 | 6853 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6854 | sd = sd->parent; |
1a848870 SS |
6855 | if (sd) |
6856 | sd->child = NULL; | |
6857 | } | |
1da177e4 LT |
6858 | |
6859 | sched_domain_debug(sd, cpu); | |
6860 | ||
57d885fe | 6861 | rq_attach_root(rq, rd); |
674311d5 | 6862 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6863 | } |
6864 | ||
6865 | /* cpus with isolated domains */ | |
67af63a6 | 6866 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6867 | |
6868 | /* Setup the mask of cpus configured for isolated domains */ | |
6869 | static int __init isolated_cpu_setup(char *str) | |
6870 | { | |
13b40c1e MT |
6871 | static int __initdata ints[NR_CPUS]; |
6872 | int i; | |
1da177e4 LT |
6873 | |
6874 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6875 | cpus_clear(cpu_isolated_map); | |
6876 | for (i = 1; i <= ints[0]; i++) | |
6877 | if (ints[i] < NR_CPUS) | |
6878 | cpu_set(ints[i], cpu_isolated_map); | |
6879 | return 1; | |
6880 | } | |
6881 | ||
8927f494 | 6882 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6883 | |
6884 | /* | |
6711cab4 SS |
6885 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6886 | * to a function which identifies what group(along with sched group) a CPU | |
6887 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6888 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6889 | * |
6890 | * init_sched_build_groups will build a circular linked list of the groups | |
6891 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6892 | * and ->cpu_power to 0. | |
6893 | */ | |
a616058b | 6894 | static void |
7c16ec58 | 6895 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6896 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6897 | struct sched_group **sg, |
6898 | cpumask_t *tmpmask), | |
6899 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6900 | { |
6901 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6902 | int i; |
6903 | ||
7c16ec58 MT |
6904 | cpus_clear(*covered); |
6905 | ||
363ab6f1 | 6906 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6907 | struct sched_group *sg; |
7c16ec58 | 6908 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6909 | int j; |
6910 | ||
7c16ec58 | 6911 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6912 | continue; |
6913 | ||
7c16ec58 | 6914 | cpus_clear(sg->cpumask); |
5517d86b | 6915 | sg->__cpu_power = 0; |
1da177e4 | 6916 | |
363ab6f1 | 6917 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6918 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6919 | continue; |
6920 | ||
7c16ec58 | 6921 | cpu_set(j, *covered); |
1da177e4 LT |
6922 | cpu_set(j, sg->cpumask); |
6923 | } | |
6924 | if (!first) | |
6925 | first = sg; | |
6926 | if (last) | |
6927 | last->next = sg; | |
6928 | last = sg; | |
6929 | } | |
6930 | last->next = first; | |
6931 | } | |
6932 | ||
9c1cfda2 | 6933 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6934 | |
9c1cfda2 | 6935 | #ifdef CONFIG_NUMA |
198e2f18 | 6936 | |
9c1cfda2 JH |
6937 | /** |
6938 | * find_next_best_node - find the next node to include in a sched_domain | |
6939 | * @node: node whose sched_domain we're building | |
6940 | * @used_nodes: nodes already in the sched_domain | |
6941 | * | |
41a2d6cf | 6942 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6943 | * finds the closest node not already in the @used_nodes map. |
6944 | * | |
6945 | * Should use nodemask_t. | |
6946 | */ | |
c5f59f08 | 6947 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6948 | { |
6949 | int i, n, val, min_val, best_node = 0; | |
6950 | ||
6951 | min_val = INT_MAX; | |
6952 | ||
076ac2af | 6953 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6954 | /* Start at @node */ |
076ac2af | 6955 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6956 | |
6957 | if (!nr_cpus_node(n)) | |
6958 | continue; | |
6959 | ||
6960 | /* Skip already used nodes */ | |
c5f59f08 | 6961 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6962 | continue; |
6963 | ||
6964 | /* Simple min distance search */ | |
6965 | val = node_distance(node, n); | |
6966 | ||
6967 | if (val < min_val) { | |
6968 | min_val = val; | |
6969 | best_node = n; | |
6970 | } | |
6971 | } | |
6972 | ||
c5f59f08 | 6973 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6974 | return best_node; |
6975 | } | |
6976 | ||
6977 | /** | |
6978 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6979 | * @node: node whose cpumask we're constructing | |
73486722 | 6980 | * @span: resulting cpumask |
9c1cfda2 | 6981 | * |
41a2d6cf | 6982 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6983 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6984 | * out optimally. | |
6985 | */ | |
4bdbaad3 | 6986 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6987 | { |
c5f59f08 | 6988 | nodemask_t used_nodes; |
c5f59f08 | 6989 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6990 | int i; |
9c1cfda2 | 6991 | |
4bdbaad3 | 6992 | cpus_clear(*span); |
c5f59f08 | 6993 | nodes_clear(used_nodes); |
9c1cfda2 | 6994 | |
4bdbaad3 | 6995 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6996 | node_set(node, used_nodes); |
9c1cfda2 JH |
6997 | |
6998 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6999 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7000 | |
c5f59f08 | 7001 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7002 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7003 | } |
9c1cfda2 | 7004 | } |
6d6bc0ad | 7005 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7006 | |
5c45bf27 | 7007 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7008 | |
9c1cfda2 | 7009 | /* |
48f24c4d | 7010 | * SMT sched-domains: |
9c1cfda2 | 7011 | */ |
1da177e4 LT |
7012 | #ifdef CONFIG_SCHED_SMT |
7013 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7014 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7015 | |
41a2d6cf | 7016 | static int |
7c16ec58 MT |
7017 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7018 | cpumask_t *unused) | |
1da177e4 | 7019 | { |
6711cab4 SS |
7020 | if (sg) |
7021 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7022 | return cpu; |
7023 | } | |
6d6bc0ad | 7024 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7025 | |
48f24c4d IM |
7026 | /* |
7027 | * multi-core sched-domains: | |
7028 | */ | |
1e9f28fa SS |
7029 | #ifdef CONFIG_SCHED_MC |
7030 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7031 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7032 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7033 | |
7034 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7035 | static int |
7c16ec58 MT |
7036 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7037 | cpumask_t *mask) | |
1e9f28fa | 7038 | { |
6711cab4 | 7039 | int group; |
7c16ec58 MT |
7040 | |
7041 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7042 | cpus_and(*mask, *mask, *cpu_map); | |
7043 | group = first_cpu(*mask); | |
6711cab4 SS |
7044 | if (sg) |
7045 | *sg = &per_cpu(sched_group_core, group); | |
7046 | return group; | |
1e9f28fa SS |
7047 | } |
7048 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7049 | static int |
7c16ec58 MT |
7050 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7051 | cpumask_t *unused) | |
1e9f28fa | 7052 | { |
6711cab4 SS |
7053 | if (sg) |
7054 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7055 | return cpu; |
7056 | } | |
7057 | #endif | |
7058 | ||
1da177e4 | 7059 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7060 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7061 | |
41a2d6cf | 7062 | static int |
7c16ec58 MT |
7063 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7064 | cpumask_t *mask) | |
1da177e4 | 7065 | { |
6711cab4 | 7066 | int group; |
48f24c4d | 7067 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7068 | *mask = cpu_coregroup_map(cpu); |
7069 | cpus_and(*mask, *mask, *cpu_map); | |
7070 | group = first_cpu(*mask); | |
1e9f28fa | 7071 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7072 | *mask = per_cpu(cpu_sibling_map, cpu); |
7073 | cpus_and(*mask, *mask, *cpu_map); | |
7074 | group = first_cpu(*mask); | |
1da177e4 | 7075 | #else |
6711cab4 | 7076 | group = cpu; |
1da177e4 | 7077 | #endif |
6711cab4 SS |
7078 | if (sg) |
7079 | *sg = &per_cpu(sched_group_phys, group); | |
7080 | return group; | |
1da177e4 LT |
7081 | } |
7082 | ||
7083 | #ifdef CONFIG_NUMA | |
1da177e4 | 7084 | /* |
9c1cfda2 JH |
7085 | * The init_sched_build_groups can't handle what we want to do with node |
7086 | * groups, so roll our own. Now each node has its own list of groups which | |
7087 | * gets dynamically allocated. | |
1da177e4 | 7088 | */ |
9c1cfda2 | 7089 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7090 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7091 | |
9c1cfda2 | 7092 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7093 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7094 | |
6711cab4 | 7095 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7096 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7097 | { |
6711cab4 SS |
7098 | int group; |
7099 | ||
7c16ec58 MT |
7100 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7101 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7102 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7103 | |
7104 | if (sg) | |
7105 | *sg = &per_cpu(sched_group_allnodes, group); | |
7106 | return group; | |
1da177e4 | 7107 | } |
6711cab4 | 7108 | |
08069033 SS |
7109 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7110 | { | |
7111 | struct sched_group *sg = group_head; | |
7112 | int j; | |
7113 | ||
7114 | if (!sg) | |
7115 | return; | |
3a5c359a | 7116 | do { |
363ab6f1 | 7117 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7118 | struct sched_domain *sd; |
08069033 | 7119 | |
3a5c359a AK |
7120 | sd = &per_cpu(phys_domains, j); |
7121 | if (j != first_cpu(sd->groups->cpumask)) { | |
7122 | /* | |
7123 | * Only add "power" once for each | |
7124 | * physical package. | |
7125 | */ | |
7126 | continue; | |
7127 | } | |
08069033 | 7128 | |
3a5c359a AK |
7129 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7130 | } | |
7131 | sg = sg->next; | |
7132 | } while (sg != group_head); | |
08069033 | 7133 | } |
6d6bc0ad | 7134 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7135 | |
a616058b | 7136 | #ifdef CONFIG_NUMA |
51888ca2 | 7137 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7138 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7139 | { |
a616058b | 7140 | int cpu, i; |
51888ca2 | 7141 | |
363ab6f1 | 7142 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7143 | struct sched_group **sched_group_nodes |
7144 | = sched_group_nodes_bycpu[cpu]; | |
7145 | ||
51888ca2 SV |
7146 | if (!sched_group_nodes) |
7147 | continue; | |
7148 | ||
076ac2af | 7149 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7150 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7151 | ||
7c16ec58 MT |
7152 | *nodemask = node_to_cpumask(i); |
7153 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7154 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7155 | continue; |
7156 | ||
7157 | if (sg == NULL) | |
7158 | continue; | |
7159 | sg = sg->next; | |
7160 | next_sg: | |
7161 | oldsg = sg; | |
7162 | sg = sg->next; | |
7163 | kfree(oldsg); | |
7164 | if (oldsg != sched_group_nodes[i]) | |
7165 | goto next_sg; | |
7166 | } | |
7167 | kfree(sched_group_nodes); | |
7168 | sched_group_nodes_bycpu[cpu] = NULL; | |
7169 | } | |
51888ca2 | 7170 | } |
6d6bc0ad | 7171 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7172 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7173 | { |
7174 | } | |
6d6bc0ad | 7175 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7176 | |
89c4710e SS |
7177 | /* |
7178 | * Initialize sched groups cpu_power. | |
7179 | * | |
7180 | * cpu_power indicates the capacity of sched group, which is used while | |
7181 | * distributing the load between different sched groups in a sched domain. | |
7182 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7183 | * there are asymmetries in the topology. If there are asymmetries, group | |
7184 | * having more cpu_power will pickup more load compared to the group having | |
7185 | * less cpu_power. | |
7186 | * | |
7187 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7188 | * the maximum number of tasks a group can handle in the presence of other idle | |
7189 | * or lightly loaded groups in the same sched domain. | |
7190 | */ | |
7191 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7192 | { | |
7193 | struct sched_domain *child; | |
7194 | struct sched_group *group; | |
7195 | ||
7196 | WARN_ON(!sd || !sd->groups); | |
7197 | ||
7198 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7199 | return; | |
7200 | ||
7201 | child = sd->child; | |
7202 | ||
5517d86b ED |
7203 | sd->groups->__cpu_power = 0; |
7204 | ||
89c4710e SS |
7205 | /* |
7206 | * For perf policy, if the groups in child domain share resources | |
7207 | * (for example cores sharing some portions of the cache hierarchy | |
7208 | * or SMT), then set this domain groups cpu_power such that each group | |
7209 | * can handle only one task, when there are other idle groups in the | |
7210 | * same sched domain. | |
7211 | */ | |
7212 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7213 | (child->flags & | |
7214 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7215 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7216 | return; |
7217 | } | |
7218 | ||
89c4710e SS |
7219 | /* |
7220 | * add cpu_power of each child group to this groups cpu_power | |
7221 | */ | |
7222 | group = child->groups; | |
7223 | do { | |
5517d86b | 7224 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7225 | group = group->next; |
7226 | } while (group != child->groups); | |
7227 | } | |
7228 | ||
7c16ec58 MT |
7229 | /* |
7230 | * Initializers for schedule domains | |
7231 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7232 | */ | |
7233 | ||
7234 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7235 | #define SD_INIT_FUNC(type) \ | |
7236 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7237 | { \ | |
7238 | memset(sd, 0, sizeof(*sd)); \ | |
7239 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7240 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7241 | } |
7242 | ||
7243 | SD_INIT_FUNC(CPU) | |
7244 | #ifdef CONFIG_NUMA | |
7245 | SD_INIT_FUNC(ALLNODES) | |
7246 | SD_INIT_FUNC(NODE) | |
7247 | #endif | |
7248 | #ifdef CONFIG_SCHED_SMT | |
7249 | SD_INIT_FUNC(SIBLING) | |
7250 | #endif | |
7251 | #ifdef CONFIG_SCHED_MC | |
7252 | SD_INIT_FUNC(MC) | |
7253 | #endif | |
7254 | ||
7255 | /* | |
7256 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7257 | * space as the usage in build_sched_domains() dictates. Used only | |
7258 | * if the amount of space is significant. | |
7259 | */ | |
7260 | struct allmasks { | |
7261 | cpumask_t tmpmask; /* make this one first */ | |
7262 | union { | |
7263 | cpumask_t nodemask; | |
7264 | cpumask_t this_sibling_map; | |
7265 | cpumask_t this_core_map; | |
7266 | }; | |
7267 | cpumask_t send_covered; | |
7268 | ||
7269 | #ifdef CONFIG_NUMA | |
7270 | cpumask_t domainspan; | |
7271 | cpumask_t covered; | |
7272 | cpumask_t notcovered; | |
7273 | #endif | |
7274 | }; | |
7275 | ||
7276 | #if NR_CPUS > 128 | |
7277 | #define SCHED_CPUMASK_ALLOC 1 | |
7278 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7279 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7280 | #else | |
7281 | #define SCHED_CPUMASK_ALLOC 0 | |
7282 | #define SCHED_CPUMASK_FREE(v) | |
7283 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7284 | #endif | |
7285 | ||
7286 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7287 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7288 | ||
1d3504fc HS |
7289 | static int default_relax_domain_level = -1; |
7290 | ||
7291 | static int __init setup_relax_domain_level(char *str) | |
7292 | { | |
30e0e178 LZ |
7293 | unsigned long val; |
7294 | ||
7295 | val = simple_strtoul(str, NULL, 0); | |
7296 | if (val < SD_LV_MAX) | |
7297 | default_relax_domain_level = val; | |
7298 | ||
1d3504fc HS |
7299 | return 1; |
7300 | } | |
7301 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7302 | ||
7303 | static void set_domain_attribute(struct sched_domain *sd, | |
7304 | struct sched_domain_attr *attr) | |
7305 | { | |
7306 | int request; | |
7307 | ||
7308 | if (!attr || attr->relax_domain_level < 0) { | |
7309 | if (default_relax_domain_level < 0) | |
7310 | return; | |
7311 | else | |
7312 | request = default_relax_domain_level; | |
7313 | } else | |
7314 | request = attr->relax_domain_level; | |
7315 | if (request < sd->level) { | |
7316 | /* turn off idle balance on this domain */ | |
7317 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7318 | } else { | |
7319 | /* turn on idle balance on this domain */ | |
7320 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7321 | } | |
7322 | } | |
7323 | ||
1da177e4 | 7324 | /* |
1a20ff27 DG |
7325 | * Build sched domains for a given set of cpus and attach the sched domains |
7326 | * to the individual cpus | |
1da177e4 | 7327 | */ |
1d3504fc HS |
7328 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7329 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7330 | { |
7331 | int i; | |
57d885fe | 7332 | struct root_domain *rd; |
7c16ec58 MT |
7333 | SCHED_CPUMASK_DECLARE(allmasks); |
7334 | cpumask_t *tmpmask; | |
d1b55138 JH |
7335 | #ifdef CONFIG_NUMA |
7336 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7337 | int sd_allnodes = 0; |
d1b55138 JH |
7338 | |
7339 | /* | |
7340 | * Allocate the per-node list of sched groups | |
7341 | */ | |
076ac2af | 7342 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7343 | GFP_KERNEL); |
d1b55138 JH |
7344 | if (!sched_group_nodes) { |
7345 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7346 | return -ENOMEM; |
d1b55138 | 7347 | } |
d1b55138 | 7348 | #endif |
1da177e4 | 7349 | |
dc938520 | 7350 | rd = alloc_rootdomain(); |
57d885fe GH |
7351 | if (!rd) { |
7352 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7353 | #ifdef CONFIG_NUMA |
7354 | kfree(sched_group_nodes); | |
7355 | #endif | |
57d885fe GH |
7356 | return -ENOMEM; |
7357 | } | |
7358 | ||
7c16ec58 MT |
7359 | #if SCHED_CPUMASK_ALLOC |
7360 | /* get space for all scratch cpumask variables */ | |
7361 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7362 | if (!allmasks) { | |
7363 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7364 | kfree(rd); | |
7365 | #ifdef CONFIG_NUMA | |
7366 | kfree(sched_group_nodes); | |
7367 | #endif | |
7368 | return -ENOMEM; | |
7369 | } | |
7370 | #endif | |
7371 | tmpmask = (cpumask_t *)allmasks; | |
7372 | ||
7373 | ||
7374 | #ifdef CONFIG_NUMA | |
7375 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7376 | #endif | |
7377 | ||
1da177e4 | 7378 | /* |
1a20ff27 | 7379 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7380 | */ |
363ab6f1 | 7381 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7382 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7383 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7384 | |
7c16ec58 MT |
7385 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7386 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7387 | |
7388 | #ifdef CONFIG_NUMA | |
dd41f596 | 7389 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7390 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7391 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7392 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7393 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7394 | sd->span = *cpu_map; |
7c16ec58 | 7395 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7396 | p = sd; |
6711cab4 | 7397 | sd_allnodes = 1; |
9c1cfda2 JH |
7398 | } else |
7399 | p = NULL; | |
7400 | ||
1da177e4 | 7401 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7402 | SD_INIT(sd, NODE); |
1d3504fc | 7403 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7404 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7405 | sd->parent = p; |
1a848870 SS |
7406 | if (p) |
7407 | p->child = sd; | |
9c1cfda2 | 7408 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7409 | #endif |
7410 | ||
7411 | p = sd; | |
7412 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7413 | SD_INIT(sd, CPU); |
1d3504fc | 7414 | set_domain_attribute(sd, attr); |
7c16ec58 | 7415 | sd->span = *nodemask; |
1da177e4 | 7416 | sd->parent = p; |
1a848870 SS |
7417 | if (p) |
7418 | p->child = sd; | |
7c16ec58 | 7419 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7420 | |
1e9f28fa SS |
7421 | #ifdef CONFIG_SCHED_MC |
7422 | p = sd; | |
7423 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7424 | SD_INIT(sd, MC); |
1d3504fc | 7425 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7426 | sd->span = cpu_coregroup_map(i); |
7427 | cpus_and(sd->span, sd->span, *cpu_map); | |
7428 | sd->parent = p; | |
1a848870 | 7429 | p->child = sd; |
7c16ec58 | 7430 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7431 | #endif |
7432 | ||
1da177e4 LT |
7433 | #ifdef CONFIG_SCHED_SMT |
7434 | p = sd; | |
7435 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7436 | SD_INIT(sd, SIBLING); |
1d3504fc | 7437 | set_domain_attribute(sd, attr); |
d5a7430d | 7438 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7439 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7440 | sd->parent = p; |
1a848870 | 7441 | p->child = sd; |
7c16ec58 | 7442 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7443 | #endif |
7444 | } | |
7445 | ||
7446 | #ifdef CONFIG_SCHED_SMT | |
7447 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7448 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7449 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7450 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7451 | ||
7452 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7453 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7454 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7455 | continue; |
7456 | ||
dd41f596 | 7457 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7458 | &cpu_to_cpu_group, |
7459 | send_covered, tmpmask); | |
1da177e4 LT |
7460 | } |
7461 | #endif | |
7462 | ||
1e9f28fa SS |
7463 | #ifdef CONFIG_SCHED_MC |
7464 | /* Set up multi-core groups */ | |
363ab6f1 | 7465 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7466 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7467 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7468 | ||
7469 | *this_core_map = cpu_coregroup_map(i); | |
7470 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7471 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7472 | continue; |
7c16ec58 | 7473 | |
dd41f596 | 7474 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7475 | &cpu_to_core_group, |
7476 | send_covered, tmpmask); | |
1e9f28fa SS |
7477 | } |
7478 | #endif | |
7479 | ||
1da177e4 | 7480 | /* Set up physical groups */ |
076ac2af | 7481 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7482 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7483 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7484 | |
7c16ec58 MT |
7485 | *nodemask = node_to_cpumask(i); |
7486 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7487 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7488 | continue; |
7489 | ||
7c16ec58 MT |
7490 | init_sched_build_groups(nodemask, cpu_map, |
7491 | &cpu_to_phys_group, | |
7492 | send_covered, tmpmask); | |
1da177e4 LT |
7493 | } |
7494 | ||
7495 | #ifdef CONFIG_NUMA | |
7496 | /* Set up node groups */ | |
7c16ec58 MT |
7497 | if (sd_allnodes) { |
7498 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7499 | ||
7500 | init_sched_build_groups(cpu_map, cpu_map, | |
7501 | &cpu_to_allnodes_group, | |
7502 | send_covered, tmpmask); | |
7503 | } | |
9c1cfda2 | 7504 | |
076ac2af | 7505 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7506 | /* Set up node groups */ |
7507 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7508 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7509 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7510 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7511 | int j; |
7512 | ||
7c16ec58 MT |
7513 | *nodemask = node_to_cpumask(i); |
7514 | cpus_clear(*covered); | |
7515 | ||
7516 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7517 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7518 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7519 | continue; |
d1b55138 | 7520 | } |
9c1cfda2 | 7521 | |
4bdbaad3 | 7522 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7523 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7524 | |
15f0b676 | 7525 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7526 | if (!sg) { |
7527 | printk(KERN_WARNING "Can not alloc domain group for " | |
7528 | "node %d\n", i); | |
7529 | goto error; | |
7530 | } | |
9c1cfda2 | 7531 | sched_group_nodes[i] = sg; |
363ab6f1 | 7532 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7533 | struct sched_domain *sd; |
9761eea8 | 7534 | |
9c1cfda2 JH |
7535 | sd = &per_cpu(node_domains, j); |
7536 | sd->groups = sg; | |
9c1cfda2 | 7537 | } |
5517d86b | 7538 | sg->__cpu_power = 0; |
7c16ec58 | 7539 | sg->cpumask = *nodemask; |
51888ca2 | 7540 | sg->next = sg; |
7c16ec58 | 7541 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7542 | prev = sg; |
7543 | ||
076ac2af | 7544 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7545 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7546 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7547 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7548 | |
7c16ec58 MT |
7549 | cpus_complement(*notcovered, *covered); |
7550 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7551 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7552 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7553 | break; |
7554 | ||
7c16ec58 MT |
7555 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7556 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7557 | continue; |
7558 | ||
15f0b676 SV |
7559 | sg = kmalloc_node(sizeof(struct sched_group), |
7560 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7561 | if (!sg) { |
7562 | printk(KERN_WARNING | |
7563 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7564 | goto error; |
9c1cfda2 | 7565 | } |
5517d86b | 7566 | sg->__cpu_power = 0; |
7c16ec58 | 7567 | sg->cpumask = *tmpmask; |
51888ca2 | 7568 | sg->next = prev->next; |
7c16ec58 | 7569 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7570 | prev->next = sg; |
7571 | prev = sg; | |
7572 | } | |
9c1cfda2 | 7573 | } |
1da177e4 LT |
7574 | #endif |
7575 | ||
7576 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7577 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7578 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7579 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7580 | ||
89c4710e | 7581 | init_sched_groups_power(i, sd); |
5c45bf27 | 7582 | } |
1da177e4 | 7583 | #endif |
1e9f28fa | 7584 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7585 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7586 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7587 | ||
89c4710e | 7588 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7589 | } |
7590 | #endif | |
1e9f28fa | 7591 | |
363ab6f1 | 7592 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7593 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7594 | ||
89c4710e | 7595 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7596 | } |
7597 | ||
9c1cfda2 | 7598 | #ifdef CONFIG_NUMA |
076ac2af | 7599 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7600 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7601 | |
6711cab4 SS |
7602 | if (sd_allnodes) { |
7603 | struct sched_group *sg; | |
f712c0c7 | 7604 | |
7c16ec58 MT |
7605 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7606 | tmpmask); | |
f712c0c7 SS |
7607 | init_numa_sched_groups_power(sg); |
7608 | } | |
9c1cfda2 JH |
7609 | #endif |
7610 | ||
1da177e4 | 7611 | /* Attach the domains */ |
363ab6f1 | 7612 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7613 | struct sched_domain *sd; |
7614 | #ifdef CONFIG_SCHED_SMT | |
7615 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7616 | #elif defined(CONFIG_SCHED_MC) |
7617 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7618 | #else |
7619 | sd = &per_cpu(phys_domains, i); | |
7620 | #endif | |
57d885fe | 7621 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7622 | } |
51888ca2 | 7623 | |
7c16ec58 | 7624 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7625 | return 0; |
7626 | ||
a616058b | 7627 | #ifdef CONFIG_NUMA |
51888ca2 | 7628 | error: |
7c16ec58 MT |
7629 | free_sched_groups(cpu_map, tmpmask); |
7630 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7631 | return -ENOMEM; |
a616058b | 7632 | #endif |
1da177e4 | 7633 | } |
029190c5 | 7634 | |
1d3504fc HS |
7635 | static int build_sched_domains(const cpumask_t *cpu_map) |
7636 | { | |
7637 | return __build_sched_domains(cpu_map, NULL); | |
7638 | } | |
7639 | ||
029190c5 PJ |
7640 | static cpumask_t *doms_cur; /* current sched domains */ |
7641 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7642 | static struct sched_domain_attr *dattr_cur; |
7643 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7644 | |
7645 | /* | |
7646 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7647 | * cpumask_t) fails, then fallback to a single sched domain, | |
7648 | * as determined by the single cpumask_t fallback_doms. | |
7649 | */ | |
7650 | static cpumask_t fallback_doms; | |
7651 | ||
22e52b07 HC |
7652 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7653 | { | |
7654 | } | |
7655 | ||
1a20ff27 | 7656 | /* |
41a2d6cf | 7657 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7658 | * For now this just excludes isolated cpus, but could be used to |
7659 | * exclude other special cases in the future. | |
1a20ff27 | 7660 | */ |
51888ca2 | 7661 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7662 | { |
7378547f MM |
7663 | int err; |
7664 | ||
22e52b07 | 7665 | arch_update_cpu_topology(); |
029190c5 PJ |
7666 | ndoms_cur = 1; |
7667 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7668 | if (!doms_cur) | |
7669 | doms_cur = &fallback_doms; | |
7670 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7671 | dattr_cur = NULL; |
7378547f | 7672 | err = build_sched_domains(doms_cur); |
6382bc90 | 7673 | register_sched_domain_sysctl(); |
7378547f MM |
7674 | |
7675 | return err; | |
1a20ff27 DG |
7676 | } |
7677 | ||
7c16ec58 MT |
7678 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7679 | cpumask_t *tmpmask) | |
1da177e4 | 7680 | { |
7c16ec58 | 7681 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7682 | } |
1da177e4 | 7683 | |
1a20ff27 DG |
7684 | /* |
7685 | * Detach sched domains from a group of cpus specified in cpu_map | |
7686 | * These cpus will now be attached to the NULL domain | |
7687 | */ | |
858119e1 | 7688 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7689 | { |
7c16ec58 | 7690 | cpumask_t tmpmask; |
1a20ff27 DG |
7691 | int i; |
7692 | ||
6382bc90 MM |
7693 | unregister_sched_domain_sysctl(); |
7694 | ||
363ab6f1 | 7695 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7696 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7697 | synchronize_sched(); |
7c16ec58 | 7698 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7699 | } |
7700 | ||
1d3504fc HS |
7701 | /* handle null as "default" */ |
7702 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7703 | struct sched_domain_attr *new, int idx_new) | |
7704 | { | |
7705 | struct sched_domain_attr tmp; | |
7706 | ||
7707 | /* fast path */ | |
7708 | if (!new && !cur) | |
7709 | return 1; | |
7710 | ||
7711 | tmp = SD_ATTR_INIT; | |
7712 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7713 | new ? (new + idx_new) : &tmp, | |
7714 | sizeof(struct sched_domain_attr)); | |
7715 | } | |
7716 | ||
029190c5 PJ |
7717 | /* |
7718 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7719 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7720 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7721 | * It destroys each deleted domain and builds each new domain. | |
7722 | * | |
7723 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7724 | * The masks don't intersect (don't overlap.) We should setup one |
7725 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7726 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7727 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7728 | * it as it is. | |
7729 | * | |
41a2d6cf IM |
7730 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7731 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7732 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7733 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7734 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 | 7735 | * |
dfb512ec MK |
7736 | * If doms_new==NULL it will be replaced with cpu_online_map. |
7737 | * ndoms_new==0 is a special case for destroying existing domains. | |
7738 | * It will not create the default domain. | |
7739 | * | |
029190c5 PJ |
7740 | * Call with hotplug lock held |
7741 | */ | |
1d3504fc HS |
7742 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7743 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7744 | { |
dfb512ec | 7745 | int i, j, n; |
029190c5 | 7746 | |
712555ee | 7747 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7748 | |
7378547f MM |
7749 | /* always unregister in case we don't destroy any domains */ |
7750 | unregister_sched_domain_sysctl(); | |
7751 | ||
dfb512ec | 7752 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7753 | |
7754 | /* Destroy deleted domains */ | |
7755 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7756 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7757 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7758 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7759 | goto match1; |
7760 | } | |
7761 | /* no match - a current sched domain not in new doms_new[] */ | |
7762 | detach_destroy_domains(doms_cur + i); | |
7763 | match1: | |
7764 | ; | |
7765 | } | |
7766 | ||
e761b772 MK |
7767 | if (doms_new == NULL) { |
7768 | ndoms_cur = 0; | |
e761b772 MK |
7769 | doms_new = &fallback_doms; |
7770 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7771 | dattr_new = NULL; | |
7772 | } | |
7773 | ||
029190c5 PJ |
7774 | /* Build new domains */ |
7775 | for (i = 0; i < ndoms_new; i++) { | |
7776 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7777 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7778 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7779 | goto match2; |
7780 | } | |
7781 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7782 | __build_sched_domains(doms_new + i, |
7783 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7784 | match2: |
7785 | ; | |
7786 | } | |
7787 | ||
7788 | /* Remember the new sched domains */ | |
7789 | if (doms_cur != &fallback_doms) | |
7790 | kfree(doms_cur); | |
1d3504fc | 7791 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7792 | doms_cur = doms_new; |
1d3504fc | 7793 | dattr_cur = dattr_new; |
029190c5 | 7794 | ndoms_cur = ndoms_new; |
7378547f MM |
7795 | |
7796 | register_sched_domain_sysctl(); | |
a1835615 | 7797 | |
712555ee | 7798 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7799 | } |
7800 | ||
5c45bf27 | 7801 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7802 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7803 | { |
95402b38 | 7804 | get_online_cpus(); |
dfb512ec MK |
7805 | |
7806 | /* Destroy domains first to force the rebuild */ | |
7807 | partition_sched_domains(0, NULL, NULL); | |
7808 | ||
e761b772 | 7809 | rebuild_sched_domains(); |
95402b38 | 7810 | put_online_cpus(); |
dfb512ec | 7811 | |
e761b772 | 7812 | return 0; |
5c45bf27 SS |
7813 | } |
7814 | ||
7815 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7816 | { | |
7817 | int ret; | |
7818 | ||
7819 | if (buf[0] != '0' && buf[0] != '1') | |
7820 | return -EINVAL; | |
7821 | ||
7822 | if (smt) | |
7823 | sched_smt_power_savings = (buf[0] == '1'); | |
7824 | else | |
7825 | sched_mc_power_savings = (buf[0] == '1'); | |
7826 | ||
7827 | ret = arch_reinit_sched_domains(); | |
7828 | ||
7829 | return ret ? ret : count; | |
7830 | } | |
7831 | ||
5c45bf27 | 7832 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7833 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7834 | char *page) | |
5c45bf27 SS |
7835 | { |
7836 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7837 | } | |
f718cd4a | 7838 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7839 | const char *buf, size_t count) |
5c45bf27 SS |
7840 | { |
7841 | return sched_power_savings_store(buf, count, 0); | |
7842 | } | |
f718cd4a AK |
7843 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7844 | sched_mc_power_savings_show, | |
7845 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7846 | #endif |
7847 | ||
7848 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7849 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7850 | char *page) | |
5c45bf27 SS |
7851 | { |
7852 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7853 | } | |
f718cd4a | 7854 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7855 | const char *buf, size_t count) |
5c45bf27 SS |
7856 | { |
7857 | return sched_power_savings_store(buf, count, 1); | |
7858 | } | |
f718cd4a AK |
7859 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7860 | sched_smt_power_savings_show, | |
6707de00 AB |
7861 | sched_smt_power_savings_store); |
7862 | #endif | |
7863 | ||
7864 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7865 | { | |
7866 | int err = 0; | |
7867 | ||
7868 | #ifdef CONFIG_SCHED_SMT | |
7869 | if (smt_capable()) | |
7870 | err = sysfs_create_file(&cls->kset.kobj, | |
7871 | &attr_sched_smt_power_savings.attr); | |
7872 | #endif | |
7873 | #ifdef CONFIG_SCHED_MC | |
7874 | if (!err && mc_capable()) | |
7875 | err = sysfs_create_file(&cls->kset.kobj, | |
7876 | &attr_sched_mc_power_savings.attr); | |
7877 | #endif | |
7878 | return err; | |
7879 | } | |
6d6bc0ad | 7880 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7881 | |
e761b772 | 7882 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7883 | /* |
e761b772 MK |
7884 | * Add online and remove offline CPUs from the scheduler domains. |
7885 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7886 | */ |
7887 | static int update_sched_domains(struct notifier_block *nfb, | |
7888 | unsigned long action, void *hcpu) | |
e761b772 MK |
7889 | { |
7890 | switch (action) { | |
7891 | case CPU_ONLINE: | |
7892 | case CPU_ONLINE_FROZEN: | |
7893 | case CPU_DEAD: | |
7894 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7895 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7896 | return NOTIFY_OK; |
7897 | ||
7898 | default: | |
7899 | return NOTIFY_DONE; | |
7900 | } | |
7901 | } | |
7902 | #endif | |
7903 | ||
7904 | static int update_runtime(struct notifier_block *nfb, | |
7905 | unsigned long action, void *hcpu) | |
1da177e4 | 7906 | { |
7def2be1 PZ |
7907 | int cpu = (int)(long)hcpu; |
7908 | ||
1da177e4 | 7909 | switch (action) { |
1da177e4 | 7910 | case CPU_DOWN_PREPARE: |
8bb78442 | 7911 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7912 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7913 | return NOTIFY_OK; |
7914 | ||
1da177e4 | 7915 | case CPU_DOWN_FAILED: |
8bb78442 | 7916 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7917 | case CPU_ONLINE: |
8bb78442 | 7918 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7919 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7920 | return NOTIFY_OK; |
7921 | ||
1da177e4 LT |
7922 | default: |
7923 | return NOTIFY_DONE; | |
7924 | } | |
1da177e4 | 7925 | } |
1da177e4 LT |
7926 | |
7927 | void __init sched_init_smp(void) | |
7928 | { | |
5c1e1767 NP |
7929 | cpumask_t non_isolated_cpus; |
7930 | ||
434d53b0 MT |
7931 | #if defined(CONFIG_NUMA) |
7932 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7933 | GFP_KERNEL); | |
7934 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7935 | #endif | |
95402b38 | 7936 | get_online_cpus(); |
712555ee | 7937 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7938 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7939 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7940 | if (cpus_empty(non_isolated_cpus)) |
7941 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7942 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7943 | put_online_cpus(); |
e761b772 MK |
7944 | |
7945 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7946 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7947 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7948 | #endif |
7949 | ||
7950 | /* RT runtime code needs to handle some hotplug events */ | |
7951 | hotcpu_notifier(update_runtime, 0); | |
7952 | ||
b328ca18 | 7953 | init_hrtick(); |
5c1e1767 NP |
7954 | |
7955 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7956 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7957 | BUG(); |
19978ca6 | 7958 | sched_init_granularity(); |
1da177e4 LT |
7959 | } |
7960 | #else | |
7961 | void __init sched_init_smp(void) | |
7962 | { | |
19978ca6 | 7963 | sched_init_granularity(); |
1da177e4 LT |
7964 | } |
7965 | #endif /* CONFIG_SMP */ | |
7966 | ||
7967 | int in_sched_functions(unsigned long addr) | |
7968 | { | |
1da177e4 LT |
7969 | return in_lock_functions(addr) || |
7970 | (addr >= (unsigned long)__sched_text_start | |
7971 | && addr < (unsigned long)__sched_text_end); | |
7972 | } | |
7973 | ||
a9957449 | 7974 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7975 | { |
7976 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7977 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7978 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7979 | cfs_rq->rq = rq; | |
7980 | #endif | |
67e9fb2a | 7981 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7982 | } |
7983 | ||
fa85ae24 PZ |
7984 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7985 | { | |
7986 | struct rt_prio_array *array; | |
7987 | int i; | |
7988 | ||
7989 | array = &rt_rq->active; | |
7990 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7991 | INIT_LIST_HEAD(array->queue + i); | |
7992 | __clear_bit(i, array->bitmap); | |
7993 | } | |
7994 | /* delimiter for bitsearch: */ | |
7995 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7996 | ||
052f1dc7 | 7997 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7998 | rt_rq->highest_prio = MAX_RT_PRIO; |
7999 | #endif | |
fa85ae24 PZ |
8000 | #ifdef CONFIG_SMP |
8001 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8002 | rt_rq->overloaded = 0; |
8003 | #endif | |
8004 | ||
8005 | rt_rq->rt_time = 0; | |
8006 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8007 | rt_rq->rt_runtime = 0; |
8008 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8009 | |
052f1dc7 | 8010 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8011 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8012 | rt_rq->rq = rq; |
8013 | #endif | |
fa85ae24 PZ |
8014 | } |
8015 | ||
6f505b16 | 8016 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8017 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8018 | struct sched_entity *se, int cpu, int add, | |
8019 | struct sched_entity *parent) | |
6f505b16 | 8020 | { |
ec7dc8ac | 8021 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8022 | tg->cfs_rq[cpu] = cfs_rq; |
8023 | init_cfs_rq(cfs_rq, rq); | |
8024 | cfs_rq->tg = tg; | |
8025 | if (add) | |
8026 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8027 | ||
8028 | tg->se[cpu] = se; | |
354d60c2 DG |
8029 | /* se could be NULL for init_task_group */ |
8030 | if (!se) | |
8031 | return; | |
8032 | ||
ec7dc8ac DG |
8033 | if (!parent) |
8034 | se->cfs_rq = &rq->cfs; | |
8035 | else | |
8036 | se->cfs_rq = parent->my_q; | |
8037 | ||
6f505b16 PZ |
8038 | se->my_q = cfs_rq; |
8039 | se->load.weight = tg->shares; | |
e05510d0 | 8040 | se->load.inv_weight = 0; |
ec7dc8ac | 8041 | se->parent = parent; |
6f505b16 | 8042 | } |
052f1dc7 | 8043 | #endif |
6f505b16 | 8044 | |
052f1dc7 | 8045 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8046 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8047 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8048 | struct sched_rt_entity *parent) | |
6f505b16 | 8049 | { |
ec7dc8ac DG |
8050 | struct rq *rq = cpu_rq(cpu); |
8051 | ||
6f505b16 PZ |
8052 | tg->rt_rq[cpu] = rt_rq; |
8053 | init_rt_rq(rt_rq, rq); | |
8054 | rt_rq->tg = tg; | |
8055 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8056 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8057 | if (add) |
8058 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8059 | ||
8060 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8061 | if (!rt_se) |
8062 | return; | |
8063 | ||
ec7dc8ac DG |
8064 | if (!parent) |
8065 | rt_se->rt_rq = &rq->rt; | |
8066 | else | |
8067 | rt_se->rt_rq = parent->my_q; | |
8068 | ||
6f505b16 | 8069 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8070 | rt_se->parent = parent; |
6f505b16 PZ |
8071 | INIT_LIST_HEAD(&rt_se->run_list); |
8072 | } | |
8073 | #endif | |
8074 | ||
1da177e4 LT |
8075 | void __init sched_init(void) |
8076 | { | |
dd41f596 | 8077 | int i, j; |
434d53b0 MT |
8078 | unsigned long alloc_size = 0, ptr; |
8079 | ||
8080 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8081 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8082 | #endif | |
8083 | #ifdef CONFIG_RT_GROUP_SCHED | |
8084 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8085 | #endif |
8086 | #ifdef CONFIG_USER_SCHED | |
8087 | alloc_size *= 2; | |
434d53b0 MT |
8088 | #endif |
8089 | /* | |
8090 | * As sched_init() is called before page_alloc is setup, | |
8091 | * we use alloc_bootmem(). | |
8092 | */ | |
8093 | if (alloc_size) { | |
5a9d3225 | 8094 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8095 | |
8096 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8097 | init_task_group.se = (struct sched_entity **)ptr; | |
8098 | ptr += nr_cpu_ids * sizeof(void **); | |
8099 | ||
8100 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8101 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8102 | |
8103 | #ifdef CONFIG_USER_SCHED | |
8104 | root_task_group.se = (struct sched_entity **)ptr; | |
8105 | ptr += nr_cpu_ids * sizeof(void **); | |
8106 | ||
8107 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8108 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8109 | #endif /* CONFIG_USER_SCHED */ |
8110 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8111 | #ifdef CONFIG_RT_GROUP_SCHED |
8112 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8113 | ptr += nr_cpu_ids * sizeof(void **); | |
8114 | ||
8115 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8116 | ptr += nr_cpu_ids * sizeof(void **); |
8117 | ||
8118 | #ifdef CONFIG_USER_SCHED | |
8119 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8120 | ptr += nr_cpu_ids * sizeof(void **); | |
8121 | ||
8122 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8123 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8124 | #endif /* CONFIG_USER_SCHED */ |
8125 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8126 | } |
dd41f596 | 8127 | |
57d885fe GH |
8128 | #ifdef CONFIG_SMP |
8129 | init_defrootdomain(); | |
8130 | #endif | |
8131 | ||
d0b27fa7 PZ |
8132 | init_rt_bandwidth(&def_rt_bandwidth, |
8133 | global_rt_period(), global_rt_runtime()); | |
8134 | ||
8135 | #ifdef CONFIG_RT_GROUP_SCHED | |
8136 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8137 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8138 | #ifdef CONFIG_USER_SCHED |
8139 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8140 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8141 | #endif /* CONFIG_USER_SCHED */ |
8142 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8143 | |
052f1dc7 | 8144 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8145 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8146 | INIT_LIST_HEAD(&init_task_group.children); |
8147 | ||
8148 | #ifdef CONFIG_USER_SCHED | |
8149 | INIT_LIST_HEAD(&root_task_group.children); | |
8150 | init_task_group.parent = &root_task_group; | |
8151 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8152 | #endif /* CONFIG_USER_SCHED */ |
8153 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8154 | |
0a945022 | 8155 | for_each_possible_cpu(i) { |
70b97a7f | 8156 | struct rq *rq; |
1da177e4 LT |
8157 | |
8158 | rq = cpu_rq(i); | |
8159 | spin_lock_init(&rq->lock); | |
7897986b | 8160 | rq->nr_running = 0; |
dd41f596 | 8161 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8162 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8163 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8164 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8165 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8166 | #ifdef CONFIG_CGROUP_SCHED |
8167 | /* | |
8168 | * How much cpu bandwidth does init_task_group get? | |
8169 | * | |
8170 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8171 | * gets 100% of the cpu resources in the system. This overall | |
8172 | * system cpu resource is divided among the tasks of | |
8173 | * init_task_group and its child task-groups in a fair manner, | |
8174 | * based on each entity's (task or task-group's) weight | |
8175 | * (se->load.weight). | |
8176 | * | |
8177 | * In other words, if init_task_group has 10 tasks of weight | |
8178 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8179 | * then A0's share of the cpu resource is: | |
8180 | * | |
8181 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8182 | * | |
8183 | * We achieve this by letting init_task_group's tasks sit | |
8184 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8185 | */ | |
ec7dc8ac | 8186 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8187 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8188 | root_task_group.shares = NICE_0_LOAD; |
8189 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8190 | /* |
8191 | * In case of task-groups formed thr' the user id of tasks, | |
8192 | * init_task_group represents tasks belonging to root user. | |
8193 | * Hence it forms a sibling of all subsequent groups formed. | |
8194 | * In this case, init_task_group gets only a fraction of overall | |
8195 | * system cpu resource, based on the weight assigned to root | |
8196 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8197 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8198 | * (init_cfs_rq) and having one entity represent this group of | |
8199 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8200 | */ | |
ec7dc8ac | 8201 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8202 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8203 | &per_cpu(init_sched_entity, i), i, 1, |
8204 | root_task_group.se[i]); | |
6f505b16 | 8205 | |
052f1dc7 | 8206 | #endif |
354d60c2 DG |
8207 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8208 | ||
8209 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8210 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8211 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8212 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8213 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8214 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8215 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8216 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8217 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8218 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8219 | root_task_group.rt_se[i]); | |
354d60c2 | 8220 | #endif |
dd41f596 | 8221 | #endif |
1da177e4 | 8222 | |
dd41f596 IM |
8223 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8224 | rq->cpu_load[j] = 0; | |
1da177e4 | 8225 | #ifdef CONFIG_SMP |
41c7ce9a | 8226 | rq->sd = NULL; |
57d885fe | 8227 | rq->rd = NULL; |
1da177e4 | 8228 | rq->active_balance = 0; |
dd41f596 | 8229 | rq->next_balance = jiffies; |
1da177e4 | 8230 | rq->push_cpu = 0; |
0a2966b4 | 8231 | rq->cpu = i; |
1f11eb6a | 8232 | rq->online = 0; |
1da177e4 LT |
8233 | rq->migration_thread = NULL; |
8234 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8235 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8236 | #endif |
8f4d37ec | 8237 | init_rq_hrtick(rq); |
1da177e4 | 8238 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8239 | } |
8240 | ||
2dd73a4f | 8241 | set_load_weight(&init_task); |
b50f60ce | 8242 | |
e107be36 AK |
8243 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8244 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8245 | #endif | |
8246 | ||
c9819f45 | 8247 | #ifdef CONFIG_SMP |
962cf36c | 8248 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8249 | #endif |
8250 | ||
b50f60ce HC |
8251 | #ifdef CONFIG_RT_MUTEXES |
8252 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8253 | #endif | |
8254 | ||
1da177e4 LT |
8255 | /* |
8256 | * The boot idle thread does lazy MMU switching as well: | |
8257 | */ | |
8258 | atomic_inc(&init_mm.mm_count); | |
8259 | enter_lazy_tlb(&init_mm, current); | |
8260 | ||
8261 | /* | |
8262 | * Make us the idle thread. Technically, schedule() should not be | |
8263 | * called from this thread, however somewhere below it might be, | |
8264 | * but because we are the idle thread, we just pick up running again | |
8265 | * when this runqueue becomes "idle". | |
8266 | */ | |
8267 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8268 | /* |
8269 | * During early bootup we pretend to be a normal task: | |
8270 | */ | |
8271 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8272 | |
8273 | scheduler_running = 1; | |
1da177e4 LT |
8274 | } |
8275 | ||
8276 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8277 | void __might_sleep(char *file, int line) | |
8278 | { | |
48f24c4d | 8279 | #ifdef in_atomic |
1da177e4 LT |
8280 | static unsigned long prev_jiffy; /* ratelimiting */ |
8281 | ||
8282 | if ((in_atomic() || irqs_disabled()) && | |
8283 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8284 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8285 | return; | |
8286 | prev_jiffy = jiffies; | |
91368d73 | 8287 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8288 | " context at %s:%d\n", file, line); |
8289 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8290 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8291 | debug_show_held_locks(current); |
3117df04 IM |
8292 | if (irqs_disabled()) |
8293 | print_irqtrace_events(current); | |
1da177e4 LT |
8294 | dump_stack(); |
8295 | } | |
8296 | #endif | |
8297 | } | |
8298 | EXPORT_SYMBOL(__might_sleep); | |
8299 | #endif | |
8300 | ||
8301 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8302 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8303 | { | |
8304 | int on_rq; | |
3e51f33f | 8305 | |
3a5e4dc1 AK |
8306 | update_rq_clock(rq); |
8307 | on_rq = p->se.on_rq; | |
8308 | if (on_rq) | |
8309 | deactivate_task(rq, p, 0); | |
8310 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8311 | if (on_rq) { | |
8312 | activate_task(rq, p, 0); | |
8313 | resched_task(rq->curr); | |
8314 | } | |
8315 | } | |
8316 | ||
1da177e4 LT |
8317 | void normalize_rt_tasks(void) |
8318 | { | |
a0f98a1c | 8319 | struct task_struct *g, *p; |
1da177e4 | 8320 | unsigned long flags; |
70b97a7f | 8321 | struct rq *rq; |
1da177e4 | 8322 | |
4cf5d77a | 8323 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8324 | do_each_thread(g, p) { |
178be793 IM |
8325 | /* |
8326 | * Only normalize user tasks: | |
8327 | */ | |
8328 | if (!p->mm) | |
8329 | continue; | |
8330 | ||
6cfb0d5d | 8331 | p->se.exec_start = 0; |
6cfb0d5d | 8332 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8333 | p->se.wait_start = 0; |
dd41f596 | 8334 | p->se.sleep_start = 0; |
dd41f596 | 8335 | p->se.block_start = 0; |
6cfb0d5d | 8336 | #endif |
dd41f596 IM |
8337 | |
8338 | if (!rt_task(p)) { | |
8339 | /* | |
8340 | * Renice negative nice level userspace | |
8341 | * tasks back to 0: | |
8342 | */ | |
8343 | if (TASK_NICE(p) < 0 && p->mm) | |
8344 | set_user_nice(p, 0); | |
1da177e4 | 8345 | continue; |
dd41f596 | 8346 | } |
1da177e4 | 8347 | |
4cf5d77a | 8348 | spin_lock(&p->pi_lock); |
b29739f9 | 8349 | rq = __task_rq_lock(p); |
1da177e4 | 8350 | |
178be793 | 8351 | normalize_task(rq, p); |
3a5e4dc1 | 8352 | |
b29739f9 | 8353 | __task_rq_unlock(rq); |
4cf5d77a | 8354 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8355 | } while_each_thread(g, p); |
8356 | ||
4cf5d77a | 8357 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8358 | } |
8359 | ||
8360 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8361 | |
8362 | #ifdef CONFIG_IA64 | |
8363 | /* | |
8364 | * These functions are only useful for the IA64 MCA handling. | |
8365 | * | |
8366 | * They can only be called when the whole system has been | |
8367 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8368 | * activity can take place. Using them for anything else would | |
8369 | * be a serious bug, and as a result, they aren't even visible | |
8370 | * under any other configuration. | |
8371 | */ | |
8372 | ||
8373 | /** | |
8374 | * curr_task - return the current task for a given cpu. | |
8375 | * @cpu: the processor in question. | |
8376 | * | |
8377 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8378 | */ | |
36c8b586 | 8379 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8380 | { |
8381 | return cpu_curr(cpu); | |
8382 | } | |
8383 | ||
8384 | /** | |
8385 | * set_curr_task - set the current task for a given cpu. | |
8386 | * @cpu: the processor in question. | |
8387 | * @p: the task pointer to set. | |
8388 | * | |
8389 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8390 | * are serviced on a separate stack. It allows the architecture to switch the |
8391 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8392 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8393 | * and caller must save the original value of the current task (see | |
8394 | * curr_task() above) and restore that value before reenabling interrupts and | |
8395 | * re-starting the system. | |
8396 | * | |
8397 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8398 | */ | |
36c8b586 | 8399 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8400 | { |
8401 | cpu_curr(cpu) = p; | |
8402 | } | |
8403 | ||
8404 | #endif | |
29f59db3 | 8405 | |
bccbe08a PZ |
8406 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8407 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8408 | { |
8409 | int i; | |
8410 | ||
8411 | for_each_possible_cpu(i) { | |
8412 | if (tg->cfs_rq) | |
8413 | kfree(tg->cfs_rq[i]); | |
8414 | if (tg->se) | |
8415 | kfree(tg->se[i]); | |
6f505b16 PZ |
8416 | } |
8417 | ||
8418 | kfree(tg->cfs_rq); | |
8419 | kfree(tg->se); | |
6f505b16 PZ |
8420 | } |
8421 | ||
ec7dc8ac DG |
8422 | static |
8423 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8424 | { |
29f59db3 | 8425 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8426 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8427 | struct rq *rq; |
29f59db3 SV |
8428 | int i; |
8429 | ||
434d53b0 | 8430 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8431 | if (!tg->cfs_rq) |
8432 | goto err; | |
434d53b0 | 8433 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8434 | if (!tg->se) |
8435 | goto err; | |
052f1dc7 PZ |
8436 | |
8437 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8438 | |
8439 | for_each_possible_cpu(i) { | |
9b5b7751 | 8440 | rq = cpu_rq(i); |
29f59db3 | 8441 | |
6f505b16 PZ |
8442 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8443 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8444 | if (!cfs_rq) |
8445 | goto err; | |
8446 | ||
6f505b16 PZ |
8447 | se = kmalloc_node(sizeof(struct sched_entity), |
8448 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8449 | if (!se) |
8450 | goto err; | |
8451 | ||
ec7dc8ac DG |
8452 | parent_se = parent ? parent->se[i] : NULL; |
8453 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8454 | } |
8455 | ||
8456 | return 1; | |
8457 | ||
8458 | err: | |
8459 | return 0; | |
8460 | } | |
8461 | ||
8462 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8463 | { | |
8464 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8465 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8466 | } | |
8467 | ||
8468 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8469 | { | |
8470 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8471 | } | |
6d6bc0ad | 8472 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8473 | static inline void free_fair_sched_group(struct task_group *tg) |
8474 | { | |
8475 | } | |
8476 | ||
ec7dc8ac DG |
8477 | static inline |
8478 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8479 | { |
8480 | return 1; | |
8481 | } | |
8482 | ||
8483 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8484 | { | |
8485 | } | |
8486 | ||
8487 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8488 | { | |
8489 | } | |
6d6bc0ad | 8490 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8491 | |
8492 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8493 | static void free_rt_sched_group(struct task_group *tg) |
8494 | { | |
8495 | int i; | |
8496 | ||
d0b27fa7 PZ |
8497 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8498 | ||
bccbe08a PZ |
8499 | for_each_possible_cpu(i) { |
8500 | if (tg->rt_rq) | |
8501 | kfree(tg->rt_rq[i]); | |
8502 | if (tg->rt_se) | |
8503 | kfree(tg->rt_se[i]); | |
8504 | } | |
8505 | ||
8506 | kfree(tg->rt_rq); | |
8507 | kfree(tg->rt_se); | |
8508 | } | |
8509 | ||
ec7dc8ac DG |
8510 | static |
8511 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8512 | { |
8513 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8514 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8515 | struct rq *rq; |
8516 | int i; | |
8517 | ||
434d53b0 | 8518 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8519 | if (!tg->rt_rq) |
8520 | goto err; | |
434d53b0 | 8521 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8522 | if (!tg->rt_se) |
8523 | goto err; | |
8524 | ||
d0b27fa7 PZ |
8525 | init_rt_bandwidth(&tg->rt_bandwidth, |
8526 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8527 | |
8528 | for_each_possible_cpu(i) { | |
8529 | rq = cpu_rq(i); | |
8530 | ||
6f505b16 PZ |
8531 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8532 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8533 | if (!rt_rq) | |
8534 | goto err; | |
29f59db3 | 8535 | |
6f505b16 PZ |
8536 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8537 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8538 | if (!rt_se) | |
8539 | goto err; | |
29f59db3 | 8540 | |
ec7dc8ac DG |
8541 | parent_se = parent ? parent->rt_se[i] : NULL; |
8542 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8543 | } |
8544 | ||
bccbe08a PZ |
8545 | return 1; |
8546 | ||
8547 | err: | |
8548 | return 0; | |
8549 | } | |
8550 | ||
8551 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8552 | { | |
8553 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8554 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8555 | } | |
8556 | ||
8557 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8558 | { | |
8559 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8560 | } | |
6d6bc0ad | 8561 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8562 | static inline void free_rt_sched_group(struct task_group *tg) |
8563 | { | |
8564 | } | |
8565 | ||
ec7dc8ac DG |
8566 | static inline |
8567 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8568 | { |
8569 | return 1; | |
8570 | } | |
8571 | ||
8572 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8573 | { | |
8574 | } | |
8575 | ||
8576 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8577 | { | |
8578 | } | |
6d6bc0ad | 8579 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8580 | |
d0b27fa7 | 8581 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8582 | static void free_sched_group(struct task_group *tg) |
8583 | { | |
8584 | free_fair_sched_group(tg); | |
8585 | free_rt_sched_group(tg); | |
8586 | kfree(tg); | |
8587 | } | |
8588 | ||
8589 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8590 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8591 | { |
8592 | struct task_group *tg; | |
8593 | unsigned long flags; | |
8594 | int i; | |
8595 | ||
8596 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8597 | if (!tg) | |
8598 | return ERR_PTR(-ENOMEM); | |
8599 | ||
ec7dc8ac | 8600 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8601 | goto err; |
8602 | ||
ec7dc8ac | 8603 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8604 | goto err; |
8605 | ||
8ed36996 | 8606 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8607 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8608 | register_fair_sched_group(tg, i); |
8609 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8610 | } |
6f505b16 | 8611 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8612 | |
8613 | WARN_ON(!parent); /* root should already exist */ | |
8614 | ||
8615 | tg->parent = parent; | |
f473aa5e | 8616 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8617 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8618 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8619 | |
9b5b7751 | 8620 | return tg; |
29f59db3 SV |
8621 | |
8622 | err: | |
6f505b16 | 8623 | free_sched_group(tg); |
29f59db3 SV |
8624 | return ERR_PTR(-ENOMEM); |
8625 | } | |
8626 | ||
9b5b7751 | 8627 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8628 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8629 | { |
29f59db3 | 8630 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8631 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8632 | } |
8633 | ||
9b5b7751 | 8634 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8635 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8636 | { |
8ed36996 | 8637 | unsigned long flags; |
9b5b7751 | 8638 | int i; |
29f59db3 | 8639 | |
8ed36996 | 8640 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8641 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8642 | unregister_fair_sched_group(tg, i); |
8643 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8644 | } |
6f505b16 | 8645 | list_del_rcu(&tg->list); |
f473aa5e | 8646 | list_del_rcu(&tg->siblings); |
8ed36996 | 8647 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8648 | |
9b5b7751 | 8649 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8650 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8651 | } |
8652 | ||
9b5b7751 | 8653 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8654 | * The caller of this function should have put the task in its new group |
8655 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8656 | * reflect its new group. | |
9b5b7751 SV |
8657 | */ |
8658 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8659 | { |
8660 | int on_rq, running; | |
8661 | unsigned long flags; | |
8662 | struct rq *rq; | |
8663 | ||
8664 | rq = task_rq_lock(tsk, &flags); | |
8665 | ||
29f59db3 SV |
8666 | update_rq_clock(rq); |
8667 | ||
051a1d1a | 8668 | running = task_current(rq, tsk); |
29f59db3 SV |
8669 | on_rq = tsk->se.on_rq; |
8670 | ||
0e1f3483 | 8671 | if (on_rq) |
29f59db3 | 8672 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8673 | if (unlikely(running)) |
8674 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8675 | |
6f505b16 | 8676 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8677 | |
810b3817 PZ |
8678 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8679 | if (tsk->sched_class->moved_group) | |
8680 | tsk->sched_class->moved_group(tsk); | |
8681 | #endif | |
8682 | ||
0e1f3483 HS |
8683 | if (unlikely(running)) |
8684 | tsk->sched_class->set_curr_task(rq); | |
8685 | if (on_rq) | |
7074badb | 8686 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8687 | |
29f59db3 SV |
8688 | task_rq_unlock(rq, &flags); |
8689 | } | |
6d6bc0ad | 8690 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8691 | |
052f1dc7 | 8692 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8693 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8694 | { |
8695 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8696 | int on_rq; |
8697 | ||
29f59db3 | 8698 | on_rq = se->on_rq; |
62fb1851 | 8699 | if (on_rq) |
29f59db3 SV |
8700 | dequeue_entity(cfs_rq, se, 0); |
8701 | ||
8702 | se->load.weight = shares; | |
e05510d0 | 8703 | se->load.inv_weight = 0; |
29f59db3 | 8704 | |
62fb1851 | 8705 | if (on_rq) |
29f59db3 | 8706 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8707 | } |
62fb1851 | 8708 | |
c09595f6 PZ |
8709 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8710 | { | |
8711 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8712 | struct rq *rq = cfs_rq->rq; | |
8713 | unsigned long flags; | |
8714 | ||
8715 | spin_lock_irqsave(&rq->lock, flags); | |
8716 | __set_se_shares(se, shares); | |
8717 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8718 | } |
8719 | ||
8ed36996 PZ |
8720 | static DEFINE_MUTEX(shares_mutex); |
8721 | ||
4cf86d77 | 8722 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8723 | { |
8724 | int i; | |
8ed36996 | 8725 | unsigned long flags; |
c61935fd | 8726 | |
ec7dc8ac DG |
8727 | /* |
8728 | * We can't change the weight of the root cgroup. | |
8729 | */ | |
8730 | if (!tg->se[0]) | |
8731 | return -EINVAL; | |
8732 | ||
18d95a28 PZ |
8733 | if (shares < MIN_SHARES) |
8734 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8735 | else if (shares > MAX_SHARES) |
8736 | shares = MAX_SHARES; | |
62fb1851 | 8737 | |
8ed36996 | 8738 | mutex_lock(&shares_mutex); |
9b5b7751 | 8739 | if (tg->shares == shares) |
5cb350ba | 8740 | goto done; |
29f59db3 | 8741 | |
8ed36996 | 8742 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8743 | for_each_possible_cpu(i) |
8744 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8745 | list_del_rcu(&tg->siblings); |
8ed36996 | 8746 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8747 | |
8748 | /* wait for any ongoing reference to this group to finish */ | |
8749 | synchronize_sched(); | |
8750 | ||
8751 | /* | |
8752 | * Now we are free to modify the group's share on each cpu | |
8753 | * w/o tripping rebalance_share or load_balance_fair. | |
8754 | */ | |
9b5b7751 | 8755 | tg->shares = shares; |
c09595f6 PZ |
8756 | for_each_possible_cpu(i) { |
8757 | /* | |
8758 | * force a rebalance | |
8759 | */ | |
8760 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8761 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8762 | } |
29f59db3 | 8763 | |
6b2d7700 SV |
8764 | /* |
8765 | * Enable load balance activity on this group, by inserting it back on | |
8766 | * each cpu's rq->leaf_cfs_rq_list. | |
8767 | */ | |
8ed36996 | 8768 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8769 | for_each_possible_cpu(i) |
8770 | register_fair_sched_group(tg, i); | |
f473aa5e | 8771 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8772 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8773 | done: |
8ed36996 | 8774 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8775 | return 0; |
29f59db3 SV |
8776 | } |
8777 | ||
5cb350ba DG |
8778 | unsigned long sched_group_shares(struct task_group *tg) |
8779 | { | |
8780 | return tg->shares; | |
8781 | } | |
052f1dc7 | 8782 | #endif |
5cb350ba | 8783 | |
052f1dc7 | 8784 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8785 | /* |
9f0c1e56 | 8786 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8787 | */ |
9f0c1e56 PZ |
8788 | static DEFINE_MUTEX(rt_constraints_mutex); |
8789 | ||
8790 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8791 | { | |
8792 | if (runtime == RUNTIME_INF) | |
8793 | return 1ULL << 16; | |
8794 | ||
6f6d6a1a | 8795 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8796 | } |
8797 | ||
b40b2e8e PZ |
8798 | #ifdef CONFIG_CGROUP_SCHED |
8799 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8800 | { | |
10b612f4 | 8801 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8802 | unsigned long total = 0; |
8803 | ||
8804 | if (!parent) { | |
8805 | if (global_rt_period() < period) | |
8806 | return 0; | |
8807 | ||
8808 | return to_ratio(period, runtime) < | |
8809 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8810 | } | |
8811 | ||
8812 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8813 | return 0; | |
8814 | ||
8815 | rcu_read_lock(); | |
8816 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8817 | if (tgi == tg) | |
8818 | continue; | |
8819 | ||
8820 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8821 | tgi->rt_bandwidth.rt_runtime); | |
8822 | } | |
8823 | rcu_read_unlock(); | |
8824 | ||
10b612f4 | 8825 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8826 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8827 | parent->rt_bandwidth.rt_runtime); | |
8828 | } | |
8829 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8830 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8831 | { |
8832 | struct task_group *tgi; | |
8833 | unsigned long total = 0; | |
9f0c1e56 | 8834 | unsigned long global_ratio = |
d0b27fa7 | 8835 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8836 | |
8837 | rcu_read_lock(); | |
9f0c1e56 PZ |
8838 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8839 | if (tgi == tg) | |
8840 | continue; | |
6f505b16 | 8841 | |
d0b27fa7 PZ |
8842 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8843 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8844 | } |
8845 | rcu_read_unlock(); | |
6f505b16 | 8846 | |
9f0c1e56 | 8847 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8848 | } |
b40b2e8e | 8849 | #endif |
6f505b16 | 8850 | |
521f1a24 DG |
8851 | /* Must be called with tasklist_lock held */ |
8852 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8853 | { | |
8854 | struct task_struct *g, *p; | |
8855 | do_each_thread(g, p) { | |
8856 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8857 | return 1; | |
8858 | } while_each_thread(g, p); | |
8859 | return 0; | |
8860 | } | |
8861 | ||
d0b27fa7 PZ |
8862 | static int tg_set_bandwidth(struct task_group *tg, |
8863 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8864 | { |
ac086bc2 | 8865 | int i, err = 0; |
9f0c1e56 | 8866 | |
9f0c1e56 | 8867 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8868 | read_lock(&tasklist_lock); |
ac086bc2 | 8869 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8870 | err = -EBUSY; |
8871 | goto unlock; | |
8872 | } | |
9f0c1e56 PZ |
8873 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8874 | err = -EINVAL; | |
8875 | goto unlock; | |
8876 | } | |
ac086bc2 PZ |
8877 | |
8878 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8879 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8880 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8881 | |
8882 | for_each_possible_cpu(i) { | |
8883 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8884 | ||
8885 | spin_lock(&rt_rq->rt_runtime_lock); | |
8886 | rt_rq->rt_runtime = rt_runtime; | |
8887 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8888 | } | |
8889 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8890 | unlock: |
521f1a24 | 8891 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8892 | mutex_unlock(&rt_constraints_mutex); |
8893 | ||
8894 | return err; | |
6f505b16 PZ |
8895 | } |
8896 | ||
d0b27fa7 PZ |
8897 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8898 | { | |
8899 | u64 rt_runtime, rt_period; | |
8900 | ||
8901 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8902 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8903 | if (rt_runtime_us < 0) | |
8904 | rt_runtime = RUNTIME_INF; | |
8905 | ||
8906 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8907 | } | |
8908 | ||
9f0c1e56 PZ |
8909 | long sched_group_rt_runtime(struct task_group *tg) |
8910 | { | |
8911 | u64 rt_runtime_us; | |
8912 | ||
d0b27fa7 | 8913 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8914 | return -1; |
8915 | ||
d0b27fa7 | 8916 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8917 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8918 | return rt_runtime_us; | |
8919 | } | |
d0b27fa7 PZ |
8920 | |
8921 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8922 | { | |
8923 | u64 rt_runtime, rt_period; | |
8924 | ||
8925 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8926 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8927 | ||
619b0488 R |
8928 | if (rt_period == 0) |
8929 | return -EINVAL; | |
8930 | ||
d0b27fa7 PZ |
8931 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8932 | } | |
8933 | ||
8934 | long sched_group_rt_period(struct task_group *tg) | |
8935 | { | |
8936 | u64 rt_period_us; | |
8937 | ||
8938 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8939 | do_div(rt_period_us, NSEC_PER_USEC); | |
8940 | return rt_period_us; | |
8941 | } | |
8942 | ||
8943 | static int sched_rt_global_constraints(void) | |
8944 | { | |
10b612f4 PZ |
8945 | struct task_group *tg = &root_task_group; |
8946 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8947 | int ret = 0; |
8948 | ||
10b612f4 PZ |
8949 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8950 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8951 | ||
d0b27fa7 | 8952 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8953 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8954 | ret = -EINVAL; |
8955 | mutex_unlock(&rt_constraints_mutex); | |
8956 | ||
8957 | return ret; | |
8958 | } | |
6d6bc0ad | 8959 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8960 | static int sched_rt_global_constraints(void) |
8961 | { | |
ac086bc2 PZ |
8962 | unsigned long flags; |
8963 | int i; | |
8964 | ||
8965 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8966 | for_each_possible_cpu(i) { | |
8967 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8968 | ||
8969 | spin_lock(&rt_rq->rt_runtime_lock); | |
8970 | rt_rq->rt_runtime = global_rt_runtime(); | |
8971 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8972 | } | |
8973 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8974 | ||
d0b27fa7 PZ |
8975 | return 0; |
8976 | } | |
6d6bc0ad | 8977 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8978 | |
8979 | int sched_rt_handler(struct ctl_table *table, int write, | |
8980 | struct file *filp, void __user *buffer, size_t *lenp, | |
8981 | loff_t *ppos) | |
8982 | { | |
8983 | int ret; | |
8984 | int old_period, old_runtime; | |
8985 | static DEFINE_MUTEX(mutex); | |
8986 | ||
8987 | mutex_lock(&mutex); | |
8988 | old_period = sysctl_sched_rt_period; | |
8989 | old_runtime = sysctl_sched_rt_runtime; | |
8990 | ||
8991 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8992 | ||
8993 | if (!ret && write) { | |
8994 | ret = sched_rt_global_constraints(); | |
8995 | if (ret) { | |
8996 | sysctl_sched_rt_period = old_period; | |
8997 | sysctl_sched_rt_runtime = old_runtime; | |
8998 | } else { | |
8999 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9000 | def_rt_bandwidth.rt_period = | |
9001 | ns_to_ktime(global_rt_period()); | |
9002 | } | |
9003 | } | |
9004 | mutex_unlock(&mutex); | |
9005 | ||
9006 | return ret; | |
9007 | } | |
68318b8e | 9008 | |
052f1dc7 | 9009 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9010 | |
9011 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9012 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9013 | { |
2b01dfe3 PM |
9014 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9015 | struct task_group, css); | |
68318b8e SV |
9016 | } |
9017 | ||
9018 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9019 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9020 | { |
ec7dc8ac | 9021 | struct task_group *tg, *parent; |
68318b8e | 9022 | |
2b01dfe3 | 9023 | if (!cgrp->parent) { |
68318b8e | 9024 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 9025 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
9026 | return &init_task_group.css; |
9027 | } | |
9028 | ||
ec7dc8ac DG |
9029 | parent = cgroup_tg(cgrp->parent); |
9030 | tg = sched_create_group(parent); | |
68318b8e SV |
9031 | if (IS_ERR(tg)) |
9032 | return ERR_PTR(-ENOMEM); | |
9033 | ||
9034 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 9035 | tg->css.cgroup = cgrp; |
68318b8e SV |
9036 | |
9037 | return &tg->css; | |
9038 | } | |
9039 | ||
41a2d6cf IM |
9040 | static void |
9041 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9042 | { |
2b01dfe3 | 9043 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9044 | |
9045 | sched_destroy_group(tg); | |
9046 | } | |
9047 | ||
41a2d6cf IM |
9048 | static int |
9049 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9050 | struct task_struct *tsk) | |
68318b8e | 9051 | { |
b68aa230 PZ |
9052 | #ifdef CONFIG_RT_GROUP_SCHED |
9053 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9054 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9055 | return -EINVAL; |
9056 | #else | |
68318b8e SV |
9057 | /* We don't support RT-tasks being in separate groups */ |
9058 | if (tsk->sched_class != &fair_sched_class) | |
9059 | return -EINVAL; | |
b68aa230 | 9060 | #endif |
68318b8e SV |
9061 | |
9062 | return 0; | |
9063 | } | |
9064 | ||
9065 | static void | |
2b01dfe3 | 9066 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9067 | struct cgroup *old_cont, struct task_struct *tsk) |
9068 | { | |
9069 | sched_move_task(tsk); | |
9070 | } | |
9071 | ||
052f1dc7 | 9072 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9073 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9074 | u64 shareval) |
68318b8e | 9075 | { |
2b01dfe3 | 9076 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9077 | } |
9078 | ||
f4c753b7 | 9079 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9080 | { |
2b01dfe3 | 9081 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9082 | |
9083 | return (u64) tg->shares; | |
9084 | } | |
6d6bc0ad | 9085 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9086 | |
052f1dc7 | 9087 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9088 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9089 | s64 val) |
6f505b16 | 9090 | { |
06ecb27c | 9091 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9092 | } |
9093 | ||
06ecb27c | 9094 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9095 | { |
06ecb27c | 9096 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9097 | } |
d0b27fa7 PZ |
9098 | |
9099 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9100 | u64 rt_period_us) | |
9101 | { | |
9102 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9103 | } | |
9104 | ||
9105 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9106 | { | |
9107 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9108 | } | |
6d6bc0ad | 9109 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9110 | |
fe5c7cc2 | 9111 | static struct cftype cpu_files[] = { |
052f1dc7 | 9112 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9113 | { |
9114 | .name = "shares", | |
f4c753b7 PM |
9115 | .read_u64 = cpu_shares_read_u64, |
9116 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9117 | }, |
052f1dc7 PZ |
9118 | #endif |
9119 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9120 | { |
9f0c1e56 | 9121 | .name = "rt_runtime_us", |
06ecb27c PM |
9122 | .read_s64 = cpu_rt_runtime_read, |
9123 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9124 | }, |
d0b27fa7 PZ |
9125 | { |
9126 | .name = "rt_period_us", | |
f4c753b7 PM |
9127 | .read_u64 = cpu_rt_period_read_uint, |
9128 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9129 | }, |
052f1dc7 | 9130 | #endif |
68318b8e SV |
9131 | }; |
9132 | ||
9133 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9134 | { | |
fe5c7cc2 | 9135 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9136 | } |
9137 | ||
9138 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9139 | .name = "cpu", |
9140 | .create = cpu_cgroup_create, | |
9141 | .destroy = cpu_cgroup_destroy, | |
9142 | .can_attach = cpu_cgroup_can_attach, | |
9143 | .attach = cpu_cgroup_attach, | |
9144 | .populate = cpu_cgroup_populate, | |
9145 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9146 | .early_init = 1, |
9147 | }; | |
9148 | ||
052f1dc7 | 9149 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9150 | |
9151 | #ifdef CONFIG_CGROUP_CPUACCT | |
9152 | ||
9153 | /* | |
9154 | * CPU accounting code for task groups. | |
9155 | * | |
9156 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9157 | * (balbir@in.ibm.com). | |
9158 | */ | |
9159 | ||
9160 | /* track cpu usage of a group of tasks */ | |
9161 | struct cpuacct { | |
9162 | struct cgroup_subsys_state css; | |
9163 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9164 | u64 *cpuusage; | |
9165 | }; | |
9166 | ||
9167 | struct cgroup_subsys cpuacct_subsys; | |
9168 | ||
9169 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9170 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9171 | { |
32cd756a | 9172 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9173 | struct cpuacct, css); |
9174 | } | |
9175 | ||
9176 | /* return cpu accounting group to which this task belongs */ | |
9177 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9178 | { | |
9179 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9180 | struct cpuacct, css); | |
9181 | } | |
9182 | ||
9183 | /* create a new cpu accounting group */ | |
9184 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9185 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9186 | { |
9187 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9188 | ||
9189 | if (!ca) | |
9190 | return ERR_PTR(-ENOMEM); | |
9191 | ||
9192 | ca->cpuusage = alloc_percpu(u64); | |
9193 | if (!ca->cpuusage) { | |
9194 | kfree(ca); | |
9195 | return ERR_PTR(-ENOMEM); | |
9196 | } | |
9197 | ||
9198 | return &ca->css; | |
9199 | } | |
9200 | ||
9201 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9202 | static void |
32cd756a | 9203 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9204 | { |
32cd756a | 9205 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9206 | |
9207 | free_percpu(ca->cpuusage); | |
9208 | kfree(ca); | |
9209 | } | |
9210 | ||
9211 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9212 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9213 | { |
32cd756a | 9214 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9215 | u64 totalcpuusage = 0; |
9216 | int i; | |
9217 | ||
9218 | for_each_possible_cpu(i) { | |
9219 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9220 | ||
9221 | /* | |
9222 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9223 | * platforms. | |
9224 | */ | |
9225 | spin_lock_irq(&cpu_rq(i)->lock); | |
9226 | totalcpuusage += *cpuusage; | |
9227 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9228 | } | |
9229 | ||
9230 | return totalcpuusage; | |
9231 | } | |
9232 | ||
0297b803 DG |
9233 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9234 | u64 reset) | |
9235 | { | |
9236 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9237 | int err = 0; | |
9238 | int i; | |
9239 | ||
9240 | if (reset) { | |
9241 | err = -EINVAL; | |
9242 | goto out; | |
9243 | } | |
9244 | ||
9245 | for_each_possible_cpu(i) { | |
9246 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9247 | ||
9248 | spin_lock_irq(&cpu_rq(i)->lock); | |
9249 | *cpuusage = 0; | |
9250 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9251 | } | |
9252 | out: | |
9253 | return err; | |
9254 | } | |
9255 | ||
d842de87 SV |
9256 | static struct cftype files[] = { |
9257 | { | |
9258 | .name = "usage", | |
f4c753b7 PM |
9259 | .read_u64 = cpuusage_read, |
9260 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9261 | }, |
9262 | }; | |
9263 | ||
32cd756a | 9264 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9265 | { |
32cd756a | 9266 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9267 | } |
9268 | ||
9269 | /* | |
9270 | * charge this task's execution time to its accounting group. | |
9271 | * | |
9272 | * called with rq->lock held. | |
9273 | */ | |
9274 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9275 | { | |
9276 | struct cpuacct *ca; | |
9277 | ||
9278 | if (!cpuacct_subsys.active) | |
9279 | return; | |
9280 | ||
9281 | ca = task_ca(tsk); | |
9282 | if (ca) { | |
9283 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9284 | ||
9285 | *cpuusage += cputime; | |
9286 | } | |
9287 | } | |
9288 | ||
9289 | struct cgroup_subsys cpuacct_subsys = { | |
9290 | .name = "cpuacct", | |
9291 | .create = cpuacct_create, | |
9292 | .destroy = cpuacct_destroy, | |
9293 | .populate = cpuacct_populate, | |
9294 | .subsys_id = cpuacct_subsys_id, | |
9295 | }; | |
9296 | #endif /* CONFIG_CGROUP_CPUACCT */ |