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> | |
1da177e4 | 73 | |
5517d86b | 74 | #include <asm/tlb.h> |
838225b4 | 75 | #include <asm/irq_regs.h> |
1da177e4 | 76 | |
6e0534f2 GH |
77 | #include "sched_cpupri.h" |
78 | ||
1da177e4 LT |
79 | /* |
80 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
81 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
82 | * and back. | |
83 | */ | |
84 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
85 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
86 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
87 | ||
88 | /* | |
89 | * 'User priority' is the nice value converted to something we | |
90 | * can work with better when scaling various scheduler parameters, | |
91 | * it's a [ 0 ... 39 ] range. | |
92 | */ | |
93 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
94 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
95 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
96 | ||
97 | /* | |
d7876a08 | 98 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 99 | */ |
d6322faf | 100 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 101 | |
6aa645ea IM |
102 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
103 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
104 | ||
1da177e4 LT |
105 | /* |
106 | * These are the 'tuning knobs' of the scheduler: | |
107 | * | |
a4ec24b4 | 108 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
109 | * Timeslices get refilled after they expire. |
110 | */ | |
1da177e4 | 111 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 112 | |
d0b27fa7 PZ |
113 | /* |
114 | * single value that denotes runtime == period, ie unlimited time. | |
115 | */ | |
116 | #define RUNTIME_INF ((u64)~0ULL) | |
117 | ||
5517d86b ED |
118 | #ifdef CONFIG_SMP |
119 | /* | |
120 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
121 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
122 | */ | |
123 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
124 | { | |
125 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
126 | } | |
127 | ||
128 | /* | |
129 | * Each time a sched group cpu_power is changed, | |
130 | * we must compute its reciprocal value | |
131 | */ | |
132 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
133 | { | |
134 | sg->__cpu_power += val; | |
135 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
136 | } | |
137 | #endif | |
138 | ||
e05606d3 IM |
139 | static inline int rt_policy(int policy) |
140 | { | |
3f33a7ce | 141 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
142 | return 1; |
143 | return 0; | |
144 | } | |
145 | ||
146 | static inline int task_has_rt_policy(struct task_struct *p) | |
147 | { | |
148 | return rt_policy(p->policy); | |
149 | } | |
150 | ||
1da177e4 | 151 | /* |
6aa645ea | 152 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 153 | */ |
6aa645ea IM |
154 | struct rt_prio_array { |
155 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
20b6331b | 156 | struct list_head queue[MAX_RT_PRIO]; |
6aa645ea IM |
157 | }; |
158 | ||
d0b27fa7 | 159 | struct rt_bandwidth { |
ea736ed5 IM |
160 | /* nests inside the rq lock: */ |
161 | spinlock_t rt_runtime_lock; | |
162 | ktime_t rt_period; | |
163 | u64 rt_runtime; | |
164 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
165 | }; |
166 | ||
167 | static struct rt_bandwidth def_rt_bandwidth; | |
168 | ||
169 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
170 | ||
171 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
172 | { | |
173 | struct rt_bandwidth *rt_b = | |
174 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
175 | ktime_t now; | |
176 | int overrun; | |
177 | int idle = 0; | |
178 | ||
179 | for (;;) { | |
180 | now = hrtimer_cb_get_time(timer); | |
181 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
182 | ||
183 | if (!overrun) | |
184 | break; | |
185 | ||
186 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
187 | } | |
188 | ||
189 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
190 | } | |
191 | ||
192 | static | |
193 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
194 | { | |
195 | rt_b->rt_period = ns_to_ktime(period); | |
196 | rt_b->rt_runtime = runtime; | |
197 | ||
ac086bc2 PZ |
198 | spin_lock_init(&rt_b->rt_runtime_lock); |
199 | ||
d0b27fa7 PZ |
200 | hrtimer_init(&rt_b->rt_period_timer, |
201 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
202 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
203 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
204 | } | |
205 | ||
206 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
207 | { | |
208 | ktime_t now; | |
209 | ||
210 | if (rt_b->rt_runtime == RUNTIME_INF) | |
211 | return; | |
212 | ||
213 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
214 | return; | |
215 | ||
216 | spin_lock(&rt_b->rt_runtime_lock); | |
217 | for (;;) { | |
218 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
219 | break; | |
220 | ||
221 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
222 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
223 | hrtimer_start(&rt_b->rt_period_timer, | |
224 | rt_b->rt_period_timer.expires, | |
225 | HRTIMER_MODE_ABS); | |
226 | } | |
227 | spin_unlock(&rt_b->rt_runtime_lock); | |
228 | } | |
229 | ||
230 | #ifdef CONFIG_RT_GROUP_SCHED | |
231 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
232 | { | |
233 | hrtimer_cancel(&rt_b->rt_period_timer); | |
234 | } | |
235 | #endif | |
236 | ||
712555ee HC |
237 | /* |
238 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
239 | * detach_destroy_domains and partition_sched_domains. | |
240 | */ | |
241 | static DEFINE_MUTEX(sched_domains_mutex); | |
242 | ||
052f1dc7 | 243 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 244 | |
68318b8e SV |
245 | #include <linux/cgroup.h> |
246 | ||
29f59db3 SV |
247 | struct cfs_rq; |
248 | ||
6f505b16 PZ |
249 | static LIST_HEAD(task_groups); |
250 | ||
29f59db3 | 251 | /* task group related information */ |
4cf86d77 | 252 | struct task_group { |
052f1dc7 | 253 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
254 | struct cgroup_subsys_state css; |
255 | #endif | |
052f1dc7 PZ |
256 | |
257 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
258 | /* schedulable entities of this group on each cpu */ |
259 | struct sched_entity **se; | |
260 | /* runqueue "owned" by this group on each cpu */ | |
261 | struct cfs_rq **cfs_rq; | |
262 | unsigned long shares; | |
052f1dc7 PZ |
263 | #endif |
264 | ||
265 | #ifdef CONFIG_RT_GROUP_SCHED | |
266 | struct sched_rt_entity **rt_se; | |
267 | struct rt_rq **rt_rq; | |
268 | ||
d0b27fa7 | 269 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 270 | #endif |
6b2d7700 | 271 | |
ae8393e5 | 272 | struct rcu_head rcu; |
6f505b16 | 273 | struct list_head list; |
f473aa5e PZ |
274 | |
275 | struct task_group *parent; | |
276 | struct list_head siblings; | |
277 | struct list_head children; | |
29f59db3 SV |
278 | }; |
279 | ||
354d60c2 | 280 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
281 | |
282 | /* | |
283 | * Root task group. | |
284 | * Every UID task group (including init_task_group aka UID-0) will | |
285 | * be a child to this group. | |
286 | */ | |
287 | struct task_group root_task_group; | |
288 | ||
052f1dc7 | 289 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
290 | /* Default task group's sched entity on each cpu */ |
291 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
292 | /* Default task group's cfs_rq on each cpu */ | |
293 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 294 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
295 | |
296 | #ifdef CONFIG_RT_GROUP_SCHED | |
297 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
298 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad DG |
299 | #endif /* CONFIG_RT_GROUP_SCHED */ |
300 | #else /* !CONFIG_FAIR_GROUP_SCHED */ | |
eff766a6 | 301 | #define root_task_group init_task_group |
6d6bc0ad | 302 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6f505b16 | 303 | |
8ed36996 | 304 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
305 | * a task group's cpu shares. |
306 | */ | |
8ed36996 | 307 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 308 | |
052f1dc7 | 309 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
310 | #ifdef CONFIG_USER_SCHED |
311 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 312 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 313 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 314 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 315 | |
cb4ad1ff | 316 | /* |
2e084786 LJ |
317 | * A weight of 0 or 1 can cause arithmetics problems. |
318 | * A weight of a cfs_rq is the sum of weights of which entities | |
319 | * are queued on this cfs_rq, so a weight of a entity should not be | |
320 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
321 | * (The default weight is 1024 - so there's no practical |
322 | * limitation from this.) | |
323 | */ | |
18d95a28 | 324 | #define MIN_SHARES 2 |
2e084786 | 325 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 326 | |
052f1dc7 PZ |
327 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
328 | #endif | |
329 | ||
29f59db3 | 330 | /* Default task group. |
3a252015 | 331 | * Every task in system belong to this group at bootup. |
29f59db3 | 332 | */ |
434d53b0 | 333 | struct task_group init_task_group; |
29f59db3 SV |
334 | |
335 | /* return group to which a task belongs */ | |
4cf86d77 | 336 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 337 | { |
4cf86d77 | 338 | struct task_group *tg; |
9b5b7751 | 339 | |
052f1dc7 | 340 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 341 | tg = p->user->tg; |
052f1dc7 | 342 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
343 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
344 | struct task_group, css); | |
24e377a8 | 345 | #else |
41a2d6cf | 346 | tg = &init_task_group; |
24e377a8 | 347 | #endif |
9b5b7751 | 348 | return tg; |
29f59db3 SV |
349 | } |
350 | ||
351 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 352 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 353 | { |
052f1dc7 | 354 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
355 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
356 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 357 | #endif |
6f505b16 | 358 | |
052f1dc7 | 359 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
360 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
361 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 362 | #endif |
29f59db3 SV |
363 | } |
364 | ||
365 | #else | |
366 | ||
6f505b16 | 367 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
368 | static inline struct task_group *task_group(struct task_struct *p) |
369 | { | |
370 | return NULL; | |
371 | } | |
29f59db3 | 372 | |
052f1dc7 | 373 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 374 | |
6aa645ea IM |
375 | /* CFS-related fields in a runqueue */ |
376 | struct cfs_rq { | |
377 | struct load_weight load; | |
378 | unsigned long nr_running; | |
379 | ||
6aa645ea | 380 | u64 exec_clock; |
e9acbff6 | 381 | u64 min_vruntime; |
103638d9 | 382 | u64 pair_start; |
6aa645ea IM |
383 | |
384 | struct rb_root tasks_timeline; | |
385 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
386 | |
387 | struct list_head tasks; | |
388 | struct list_head *balance_iterator; | |
389 | ||
390 | /* | |
391 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
392 | * It is set to NULL otherwise (i.e when none are currently running). |
393 | */ | |
aa2ac252 | 394 | struct sched_entity *curr, *next; |
ddc97297 PZ |
395 | |
396 | unsigned long nr_spread_over; | |
397 | ||
62160e3f | 398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
399 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
400 | ||
41a2d6cf IM |
401 | /* |
402 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
403 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
404 | * (like users, containers etc.) | |
405 | * | |
406 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
407 | * list is used during load balance. | |
408 | */ | |
41a2d6cf IM |
409 | struct list_head leaf_cfs_rq_list; |
410 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
411 | |
412 | #ifdef CONFIG_SMP | |
c09595f6 | 413 | /* |
c8cba857 | 414 | * the part of load.weight contributed by tasks |
c09595f6 | 415 | */ |
c8cba857 | 416 | unsigned long task_weight; |
c09595f6 | 417 | |
c8cba857 PZ |
418 | /* |
419 | * h_load = weight * f(tg) | |
420 | * | |
421 | * Where f(tg) is the recursive weight fraction assigned to | |
422 | * this group. | |
423 | */ | |
424 | unsigned long h_load; | |
c09595f6 | 425 | |
c8cba857 PZ |
426 | /* |
427 | * this cpu's part of tg->shares | |
428 | */ | |
429 | unsigned long shares; | |
c09595f6 | 430 | #endif |
6aa645ea IM |
431 | #endif |
432 | }; | |
1da177e4 | 433 | |
6aa645ea IM |
434 | /* Real-Time classes' related field in a runqueue: */ |
435 | struct rt_rq { | |
436 | struct rt_prio_array active; | |
63489e45 | 437 | unsigned long rt_nr_running; |
052f1dc7 | 438 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
439 | int highest_prio; /* highest queued rt task prio */ |
440 | #endif | |
fa85ae24 | 441 | #ifdef CONFIG_SMP |
73fe6aae | 442 | unsigned long rt_nr_migratory; |
a22d7fc1 | 443 | int overloaded; |
fa85ae24 | 444 | #endif |
6f505b16 | 445 | int rt_throttled; |
fa85ae24 | 446 | u64 rt_time; |
ac086bc2 | 447 | u64 rt_runtime; |
ea736ed5 | 448 | /* Nests inside the rq lock: */ |
ac086bc2 | 449 | spinlock_t rt_runtime_lock; |
6f505b16 | 450 | |
052f1dc7 | 451 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
452 | unsigned long rt_nr_boosted; |
453 | ||
6f505b16 PZ |
454 | struct rq *rq; |
455 | struct list_head leaf_rt_rq_list; | |
456 | struct task_group *tg; | |
457 | struct sched_rt_entity *rt_se; | |
458 | #endif | |
6aa645ea IM |
459 | }; |
460 | ||
57d885fe GH |
461 | #ifdef CONFIG_SMP |
462 | ||
463 | /* | |
464 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
465 | * variables. Each exclusive cpuset essentially defines an island domain by |
466 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
467 | * exclusive cpuset is created, we also create and attach a new root-domain |
468 | * object. | |
469 | * | |
57d885fe GH |
470 | */ |
471 | struct root_domain { | |
472 | atomic_t refcount; | |
473 | cpumask_t span; | |
474 | cpumask_t online; | |
637f5085 | 475 | |
0eab9146 | 476 | /* |
637f5085 GH |
477 | * The "RT overload" flag: it gets set if a CPU has more than |
478 | * one runnable RT task. | |
479 | */ | |
480 | cpumask_t rto_mask; | |
0eab9146 | 481 | atomic_t rto_count; |
6e0534f2 GH |
482 | #ifdef CONFIG_SMP |
483 | struct cpupri cpupri; | |
484 | #endif | |
57d885fe GH |
485 | }; |
486 | ||
dc938520 GH |
487 | /* |
488 | * By default the system creates a single root-domain with all cpus as | |
489 | * members (mimicking the global state we have today). | |
490 | */ | |
57d885fe GH |
491 | static struct root_domain def_root_domain; |
492 | ||
493 | #endif | |
494 | ||
1da177e4 LT |
495 | /* |
496 | * This is the main, per-CPU runqueue data structure. | |
497 | * | |
498 | * Locking rule: those places that want to lock multiple runqueues | |
499 | * (such as the load balancing or the thread migration code), lock | |
500 | * acquire operations must be ordered by ascending &runqueue. | |
501 | */ | |
70b97a7f | 502 | struct rq { |
d8016491 IM |
503 | /* runqueue lock: */ |
504 | spinlock_t lock; | |
1da177e4 LT |
505 | |
506 | /* | |
507 | * nr_running and cpu_load should be in the same cacheline because | |
508 | * remote CPUs use both these fields when doing load calculation. | |
509 | */ | |
510 | unsigned long nr_running; | |
6aa645ea IM |
511 | #define CPU_LOAD_IDX_MAX 5 |
512 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 513 | unsigned char idle_at_tick; |
46cb4b7c | 514 | #ifdef CONFIG_NO_HZ |
15934a37 | 515 | unsigned long last_tick_seen; |
46cb4b7c SS |
516 | unsigned char in_nohz_recently; |
517 | #endif | |
d8016491 IM |
518 | /* capture load from *all* tasks on this cpu: */ |
519 | struct load_weight load; | |
6aa645ea IM |
520 | unsigned long nr_load_updates; |
521 | u64 nr_switches; | |
522 | ||
523 | struct cfs_rq cfs; | |
6f505b16 | 524 | struct rt_rq rt; |
6f505b16 | 525 | |
6aa645ea | 526 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
527 | /* list of leaf cfs_rq on this cpu: */ |
528 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
529 | #endif |
530 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 531 | struct list_head leaf_rt_rq_list; |
1da177e4 | 532 | #endif |
1da177e4 LT |
533 | |
534 | /* | |
535 | * This is part of a global counter where only the total sum | |
536 | * over all CPUs matters. A task can increase this counter on | |
537 | * one CPU and if it got migrated afterwards it may decrease | |
538 | * it on another CPU. Always updated under the runqueue lock: | |
539 | */ | |
540 | unsigned long nr_uninterruptible; | |
541 | ||
36c8b586 | 542 | struct task_struct *curr, *idle; |
c9819f45 | 543 | unsigned long next_balance; |
1da177e4 | 544 | struct mm_struct *prev_mm; |
6aa645ea | 545 | |
3e51f33f | 546 | u64 clock; |
6aa645ea | 547 | |
1da177e4 LT |
548 | atomic_t nr_iowait; |
549 | ||
550 | #ifdef CONFIG_SMP | |
0eab9146 | 551 | struct root_domain *rd; |
1da177e4 LT |
552 | struct sched_domain *sd; |
553 | ||
554 | /* For active balancing */ | |
555 | int active_balance; | |
556 | int push_cpu; | |
d8016491 IM |
557 | /* cpu of this runqueue: */ |
558 | int cpu; | |
1f11eb6a | 559 | int online; |
1da177e4 | 560 | |
a8a51d5e PZ |
561 | unsigned long avg_load_per_task; |
562 | ||
36c8b586 | 563 | struct task_struct *migration_thread; |
1da177e4 LT |
564 | struct list_head migration_queue; |
565 | #endif | |
566 | ||
8f4d37ec PZ |
567 | #ifdef CONFIG_SCHED_HRTICK |
568 | unsigned long hrtick_flags; | |
569 | ktime_t hrtick_expire; | |
570 | struct hrtimer hrtick_timer; | |
571 | #endif | |
572 | ||
1da177e4 LT |
573 | #ifdef CONFIG_SCHEDSTATS |
574 | /* latency stats */ | |
575 | struct sched_info rq_sched_info; | |
576 | ||
577 | /* sys_sched_yield() stats */ | |
480b9434 KC |
578 | unsigned int yld_exp_empty; |
579 | unsigned int yld_act_empty; | |
580 | unsigned int yld_both_empty; | |
581 | unsigned int yld_count; | |
1da177e4 LT |
582 | |
583 | /* schedule() stats */ | |
480b9434 KC |
584 | unsigned int sched_switch; |
585 | unsigned int sched_count; | |
586 | unsigned int sched_goidle; | |
1da177e4 LT |
587 | |
588 | /* try_to_wake_up() stats */ | |
480b9434 KC |
589 | unsigned int ttwu_count; |
590 | unsigned int ttwu_local; | |
b8efb561 IM |
591 | |
592 | /* BKL stats */ | |
480b9434 | 593 | unsigned int bkl_count; |
1da177e4 | 594 | #endif |
fcb99371 | 595 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
596 | }; |
597 | ||
f34e3b61 | 598 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 599 | |
dd41f596 IM |
600 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
601 | { | |
602 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
603 | } | |
604 | ||
0a2966b4 CL |
605 | static inline int cpu_of(struct rq *rq) |
606 | { | |
607 | #ifdef CONFIG_SMP | |
608 | return rq->cpu; | |
609 | #else | |
610 | return 0; | |
611 | #endif | |
612 | } | |
613 | ||
674311d5 NP |
614 | /* |
615 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 616 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
617 | * |
618 | * The domain tree of any CPU may only be accessed from within | |
619 | * preempt-disabled sections. | |
620 | */ | |
48f24c4d IM |
621 | #define for_each_domain(cpu, __sd) \ |
622 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
623 | |
624 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
625 | #define this_rq() (&__get_cpu_var(runqueues)) | |
626 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
627 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
628 | ||
3e51f33f PZ |
629 | static inline void update_rq_clock(struct rq *rq) |
630 | { | |
631 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
632 | } | |
633 | ||
bf5c91ba IM |
634 | /* |
635 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
636 | */ | |
637 | #ifdef CONFIG_SCHED_DEBUG | |
638 | # define const_debug __read_mostly | |
639 | #else | |
640 | # define const_debug static const | |
641 | #endif | |
642 | ||
643 | /* | |
644 | * Debugging: various feature bits | |
645 | */ | |
f00b45c1 PZ |
646 | |
647 | #define SCHED_FEAT(name, enabled) \ | |
648 | __SCHED_FEAT_##name , | |
649 | ||
bf5c91ba | 650 | enum { |
f00b45c1 | 651 | #include "sched_features.h" |
bf5c91ba IM |
652 | }; |
653 | ||
f00b45c1 PZ |
654 | #undef SCHED_FEAT |
655 | ||
656 | #define SCHED_FEAT(name, enabled) \ | |
657 | (1UL << __SCHED_FEAT_##name) * enabled | | |
658 | ||
bf5c91ba | 659 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
660 | #include "sched_features.h" |
661 | 0; | |
662 | ||
663 | #undef SCHED_FEAT | |
664 | ||
665 | #ifdef CONFIG_SCHED_DEBUG | |
666 | #define SCHED_FEAT(name, enabled) \ | |
667 | #name , | |
668 | ||
983ed7a6 | 669 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
670 | #include "sched_features.h" |
671 | NULL | |
672 | }; | |
673 | ||
674 | #undef SCHED_FEAT | |
675 | ||
983ed7a6 | 676 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
677 | { |
678 | filp->private_data = inode->i_private; | |
679 | return 0; | |
680 | } | |
681 | ||
682 | static ssize_t | |
683 | sched_feat_read(struct file *filp, char __user *ubuf, | |
684 | size_t cnt, loff_t *ppos) | |
685 | { | |
686 | char *buf; | |
687 | int r = 0; | |
688 | int len = 0; | |
689 | int i; | |
690 | ||
691 | for (i = 0; sched_feat_names[i]; i++) { | |
692 | len += strlen(sched_feat_names[i]); | |
693 | len += 4; | |
694 | } | |
695 | ||
696 | buf = kmalloc(len + 2, GFP_KERNEL); | |
697 | if (!buf) | |
698 | return -ENOMEM; | |
699 | ||
700 | for (i = 0; sched_feat_names[i]; i++) { | |
701 | if (sysctl_sched_features & (1UL << i)) | |
702 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
703 | else | |
c24b7c52 | 704 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
705 | } |
706 | ||
707 | r += sprintf(buf + r, "\n"); | |
708 | WARN_ON(r >= len + 2); | |
709 | ||
710 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
711 | ||
712 | kfree(buf); | |
713 | ||
714 | return r; | |
715 | } | |
716 | ||
717 | static ssize_t | |
718 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
719 | size_t cnt, loff_t *ppos) | |
720 | { | |
721 | char buf[64]; | |
722 | char *cmp = buf; | |
723 | int neg = 0; | |
724 | int i; | |
725 | ||
726 | if (cnt > 63) | |
727 | cnt = 63; | |
728 | ||
729 | if (copy_from_user(&buf, ubuf, cnt)) | |
730 | return -EFAULT; | |
731 | ||
732 | buf[cnt] = 0; | |
733 | ||
c24b7c52 | 734 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
735 | neg = 1; |
736 | cmp += 3; | |
737 | } | |
738 | ||
739 | for (i = 0; sched_feat_names[i]; i++) { | |
740 | int len = strlen(sched_feat_names[i]); | |
741 | ||
742 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
743 | if (neg) | |
744 | sysctl_sched_features &= ~(1UL << i); | |
745 | else | |
746 | sysctl_sched_features |= (1UL << i); | |
747 | break; | |
748 | } | |
749 | } | |
750 | ||
751 | if (!sched_feat_names[i]) | |
752 | return -EINVAL; | |
753 | ||
754 | filp->f_pos += cnt; | |
755 | ||
756 | return cnt; | |
757 | } | |
758 | ||
759 | static struct file_operations sched_feat_fops = { | |
760 | .open = sched_feat_open, | |
761 | .read = sched_feat_read, | |
762 | .write = sched_feat_write, | |
763 | }; | |
764 | ||
765 | static __init int sched_init_debug(void) | |
766 | { | |
f00b45c1 PZ |
767 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
768 | &sched_feat_fops); | |
769 | ||
770 | return 0; | |
771 | } | |
772 | late_initcall(sched_init_debug); | |
773 | ||
774 | #endif | |
775 | ||
776 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 777 | |
b82d9fdd PZ |
778 | /* |
779 | * Number of tasks to iterate in a single balance run. | |
780 | * Limited because this is done with IRQs disabled. | |
781 | */ | |
782 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
783 | ||
2398f2c6 PZ |
784 | /* |
785 | * ratelimit for updating the group shares. | |
786 | * default: 0.5ms | |
787 | */ | |
788 | const_debug unsigned int sysctl_sched_shares_ratelimit = 500000; | |
789 | ||
fa85ae24 | 790 | /* |
9f0c1e56 | 791 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
792 | * default: 1s |
793 | */ | |
9f0c1e56 | 794 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 795 | |
6892b75e IM |
796 | static __read_mostly int scheduler_running; |
797 | ||
9f0c1e56 PZ |
798 | /* |
799 | * part of the period that we allow rt tasks to run in us. | |
800 | * default: 0.95s | |
801 | */ | |
802 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 803 | |
d0b27fa7 PZ |
804 | static inline u64 global_rt_period(void) |
805 | { | |
806 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
807 | } | |
808 | ||
809 | static inline u64 global_rt_runtime(void) | |
810 | { | |
811 | if (sysctl_sched_rt_period < 0) | |
812 | return RUNTIME_INF; | |
813 | ||
814 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
815 | } | |
fa85ae24 | 816 | |
1da177e4 | 817 | #ifndef prepare_arch_switch |
4866cde0 NP |
818 | # define prepare_arch_switch(next) do { } while (0) |
819 | #endif | |
820 | #ifndef finish_arch_switch | |
821 | # define finish_arch_switch(prev) do { } while (0) | |
822 | #endif | |
823 | ||
051a1d1a DA |
824 | static inline int task_current(struct rq *rq, struct task_struct *p) |
825 | { | |
826 | return rq->curr == p; | |
827 | } | |
828 | ||
4866cde0 | 829 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 830 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 831 | { |
051a1d1a | 832 | return task_current(rq, p); |
4866cde0 NP |
833 | } |
834 | ||
70b97a7f | 835 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
836 | { |
837 | } | |
838 | ||
70b97a7f | 839 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 840 | { |
da04c035 IM |
841 | #ifdef CONFIG_DEBUG_SPINLOCK |
842 | /* this is a valid case when another task releases the spinlock */ | |
843 | rq->lock.owner = current; | |
844 | #endif | |
8a25d5de IM |
845 | /* |
846 | * If we are tracking spinlock dependencies then we have to | |
847 | * fix up the runqueue lock - which gets 'carried over' from | |
848 | * prev into current: | |
849 | */ | |
850 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
851 | ||
4866cde0 NP |
852 | spin_unlock_irq(&rq->lock); |
853 | } | |
854 | ||
855 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 856 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
857 | { |
858 | #ifdef CONFIG_SMP | |
859 | return p->oncpu; | |
860 | #else | |
051a1d1a | 861 | return task_current(rq, p); |
4866cde0 NP |
862 | #endif |
863 | } | |
864 | ||
70b97a7f | 865 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
866 | { |
867 | #ifdef CONFIG_SMP | |
868 | /* | |
869 | * We can optimise this out completely for !SMP, because the | |
870 | * SMP rebalancing from interrupt is the only thing that cares | |
871 | * here. | |
872 | */ | |
873 | next->oncpu = 1; | |
874 | #endif | |
875 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
876 | spin_unlock_irq(&rq->lock); | |
877 | #else | |
878 | spin_unlock(&rq->lock); | |
879 | #endif | |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
887 | * We must ensure this doesn't happen until the switch is completely | |
888 | * finished. | |
889 | */ | |
890 | smp_wmb(); | |
891 | prev->oncpu = 0; | |
892 | #endif | |
893 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
894 | local_irq_enable(); | |
1da177e4 | 895 | #endif |
4866cde0 NP |
896 | } |
897 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 898 | |
b29739f9 IM |
899 | /* |
900 | * __task_rq_lock - lock the runqueue a given task resides on. | |
901 | * Must be called interrupts disabled. | |
902 | */ | |
70b97a7f | 903 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
904 | __acquires(rq->lock) |
905 | { | |
3a5c359a AK |
906 | for (;;) { |
907 | struct rq *rq = task_rq(p); | |
908 | spin_lock(&rq->lock); | |
909 | if (likely(rq == task_rq(p))) | |
910 | return rq; | |
b29739f9 | 911 | spin_unlock(&rq->lock); |
b29739f9 | 912 | } |
b29739f9 IM |
913 | } |
914 | ||
1da177e4 LT |
915 | /* |
916 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 917 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
918 | * explicitly disabling preemption. |
919 | */ | |
70b97a7f | 920 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
921 | __acquires(rq->lock) |
922 | { | |
70b97a7f | 923 | struct rq *rq; |
1da177e4 | 924 | |
3a5c359a AK |
925 | for (;;) { |
926 | local_irq_save(*flags); | |
927 | rq = task_rq(p); | |
928 | spin_lock(&rq->lock); | |
929 | if (likely(rq == task_rq(p))) | |
930 | return rq; | |
1da177e4 | 931 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 932 | } |
1da177e4 LT |
933 | } |
934 | ||
a9957449 | 935 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
936 | __releases(rq->lock) |
937 | { | |
938 | spin_unlock(&rq->lock); | |
939 | } | |
940 | ||
70b97a7f | 941 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
942 | __releases(rq->lock) |
943 | { | |
944 | spin_unlock_irqrestore(&rq->lock, *flags); | |
945 | } | |
946 | ||
1da177e4 | 947 | /* |
cc2a73b5 | 948 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 949 | */ |
a9957449 | 950 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
951 | __acquires(rq->lock) |
952 | { | |
70b97a7f | 953 | struct rq *rq; |
1da177e4 LT |
954 | |
955 | local_irq_disable(); | |
956 | rq = this_rq(); | |
957 | spin_lock(&rq->lock); | |
958 | ||
959 | return rq; | |
960 | } | |
961 | ||
8f4d37ec PZ |
962 | static void __resched_task(struct task_struct *p, int tif_bit); |
963 | ||
964 | static inline void resched_task(struct task_struct *p) | |
965 | { | |
966 | __resched_task(p, TIF_NEED_RESCHED); | |
967 | } | |
968 | ||
969 | #ifdef CONFIG_SCHED_HRTICK | |
970 | /* | |
971 | * Use HR-timers to deliver accurate preemption points. | |
972 | * | |
973 | * Its all a bit involved since we cannot program an hrt while holding the | |
974 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
975 | * reschedule event. | |
976 | * | |
977 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
978 | * rq->lock. | |
979 | */ | |
980 | static inline void resched_hrt(struct task_struct *p) | |
981 | { | |
982 | __resched_task(p, TIF_HRTICK_RESCHED); | |
983 | } | |
984 | ||
985 | static inline void resched_rq(struct rq *rq) | |
986 | { | |
987 | unsigned long flags; | |
988 | ||
989 | spin_lock_irqsave(&rq->lock, flags); | |
990 | resched_task(rq->curr); | |
991 | spin_unlock_irqrestore(&rq->lock, flags); | |
992 | } | |
993 | ||
994 | enum { | |
995 | HRTICK_SET, /* re-programm hrtick_timer */ | |
996 | HRTICK_RESET, /* not a new slice */ | |
b328ca18 | 997 | HRTICK_BLOCK, /* stop hrtick operations */ |
8f4d37ec PZ |
998 | }; |
999 | ||
1000 | /* | |
1001 | * Use hrtick when: | |
1002 | * - enabled by features | |
1003 | * - hrtimer is actually high res | |
1004 | */ | |
1005 | static inline int hrtick_enabled(struct rq *rq) | |
1006 | { | |
1007 | if (!sched_feat(HRTICK)) | |
1008 | return 0; | |
b328ca18 PZ |
1009 | if (unlikely(test_bit(HRTICK_BLOCK, &rq->hrtick_flags))) |
1010 | return 0; | |
8f4d37ec PZ |
1011 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Called to set the hrtick timer state. | |
1016 | * | |
1017 | * called with rq->lock held and irqs disabled | |
1018 | */ | |
1019 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1020 | { | |
1021 | assert_spin_locked(&rq->lock); | |
1022 | ||
1023 | /* | |
1024 | * preempt at: now + delay | |
1025 | */ | |
1026 | rq->hrtick_expire = | |
1027 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1028 | /* | |
1029 | * indicate we need to program the timer | |
1030 | */ | |
1031 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1032 | if (reset) | |
1033 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1034 | ||
1035 | /* | |
1036 | * New slices are called from the schedule path and don't need a | |
1037 | * forced reschedule. | |
1038 | */ | |
1039 | if (reset) | |
1040 | resched_hrt(rq->curr); | |
1041 | } | |
1042 | ||
1043 | static void hrtick_clear(struct rq *rq) | |
1044 | { | |
1045 | if (hrtimer_active(&rq->hrtick_timer)) | |
1046 | hrtimer_cancel(&rq->hrtick_timer); | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Update the timer from the possible pending state. | |
1051 | */ | |
1052 | static void hrtick_set(struct rq *rq) | |
1053 | { | |
1054 | ktime_t time; | |
1055 | int set, reset; | |
1056 | unsigned long flags; | |
1057 | ||
1058 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1059 | ||
1060 | spin_lock_irqsave(&rq->lock, flags); | |
1061 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1062 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1063 | time = rq->hrtick_expire; | |
1064 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1065 | spin_unlock_irqrestore(&rq->lock, flags); | |
1066 | ||
1067 | if (set) { | |
1068 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1069 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1070 | resched_rq(rq); | |
1071 | } else | |
1072 | hrtick_clear(rq); | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * High-resolution timer tick. | |
1077 | * Runs from hardirq context with interrupts disabled. | |
1078 | */ | |
1079 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1080 | { | |
1081 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1082 | ||
1083 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1084 | ||
1085 | spin_lock(&rq->lock); | |
3e51f33f | 1086 | update_rq_clock(rq); |
8f4d37ec PZ |
1087 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1088 | spin_unlock(&rq->lock); | |
1089 | ||
1090 | return HRTIMER_NORESTART; | |
1091 | } | |
1092 | ||
81d41d7e | 1093 | #ifdef CONFIG_SMP |
b328ca18 PZ |
1094 | static void hotplug_hrtick_disable(int cpu) |
1095 | { | |
1096 | struct rq *rq = cpu_rq(cpu); | |
1097 | unsigned long flags; | |
1098 | ||
1099 | spin_lock_irqsave(&rq->lock, flags); | |
1100 | rq->hrtick_flags = 0; | |
1101 | __set_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1102 | spin_unlock_irqrestore(&rq->lock, flags); | |
1103 | ||
1104 | hrtick_clear(rq); | |
1105 | } | |
1106 | ||
1107 | static void hotplug_hrtick_enable(int cpu) | |
1108 | { | |
1109 | struct rq *rq = cpu_rq(cpu); | |
1110 | unsigned long flags; | |
1111 | ||
1112 | spin_lock_irqsave(&rq->lock, flags); | |
1113 | __clear_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1114 | spin_unlock_irqrestore(&rq->lock, flags); | |
1115 | } | |
1116 | ||
1117 | static int | |
1118 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1119 | { | |
1120 | int cpu = (int)(long)hcpu; | |
1121 | ||
1122 | switch (action) { | |
1123 | case CPU_UP_CANCELED: | |
1124 | case CPU_UP_CANCELED_FROZEN: | |
1125 | case CPU_DOWN_PREPARE: | |
1126 | case CPU_DOWN_PREPARE_FROZEN: | |
1127 | case CPU_DEAD: | |
1128 | case CPU_DEAD_FROZEN: | |
1129 | hotplug_hrtick_disable(cpu); | |
1130 | return NOTIFY_OK; | |
1131 | ||
1132 | case CPU_UP_PREPARE: | |
1133 | case CPU_UP_PREPARE_FROZEN: | |
1134 | case CPU_DOWN_FAILED: | |
1135 | case CPU_DOWN_FAILED_FROZEN: | |
1136 | case CPU_ONLINE: | |
1137 | case CPU_ONLINE_FROZEN: | |
1138 | hotplug_hrtick_enable(cpu); | |
1139 | return NOTIFY_OK; | |
1140 | } | |
1141 | ||
1142 | return NOTIFY_DONE; | |
1143 | } | |
1144 | ||
1145 | static void init_hrtick(void) | |
1146 | { | |
1147 | hotcpu_notifier(hotplug_hrtick, 0); | |
1148 | } | |
81d41d7e | 1149 | #endif /* CONFIG_SMP */ |
b328ca18 PZ |
1150 | |
1151 | static void init_rq_hrtick(struct rq *rq) | |
8f4d37ec PZ |
1152 | { |
1153 | rq->hrtick_flags = 0; | |
1154 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1155 | rq->hrtick_timer.function = hrtick; | |
1156 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1157 | } | |
1158 | ||
1159 | void hrtick_resched(void) | |
1160 | { | |
1161 | struct rq *rq; | |
1162 | unsigned long flags; | |
1163 | ||
1164 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1165 | return; | |
1166 | ||
1167 | local_irq_save(flags); | |
1168 | rq = cpu_rq(smp_processor_id()); | |
1169 | hrtick_set(rq); | |
1170 | local_irq_restore(flags); | |
1171 | } | |
1172 | #else | |
1173 | static inline void hrtick_clear(struct rq *rq) | |
1174 | { | |
1175 | } | |
1176 | ||
1177 | static inline void hrtick_set(struct rq *rq) | |
1178 | { | |
1179 | } | |
1180 | ||
1181 | static inline void init_rq_hrtick(struct rq *rq) | |
1182 | { | |
1183 | } | |
1184 | ||
1185 | void hrtick_resched(void) | |
1186 | { | |
1187 | } | |
b328ca18 PZ |
1188 | |
1189 | static inline void init_hrtick(void) | |
1190 | { | |
1191 | } | |
8f4d37ec PZ |
1192 | #endif |
1193 | ||
c24d20db IM |
1194 | /* |
1195 | * resched_task - mark a task 'to be rescheduled now'. | |
1196 | * | |
1197 | * On UP this means the setting of the need_resched flag, on SMP it | |
1198 | * might also involve a cross-CPU call to trigger the scheduler on | |
1199 | * the target CPU. | |
1200 | */ | |
1201 | #ifdef CONFIG_SMP | |
1202 | ||
1203 | #ifndef tsk_is_polling | |
1204 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1205 | #endif | |
1206 | ||
8f4d37ec | 1207 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1208 | { |
1209 | int cpu; | |
1210 | ||
1211 | assert_spin_locked(&task_rq(p)->lock); | |
1212 | ||
8f4d37ec | 1213 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1214 | return; |
1215 | ||
8f4d37ec | 1216 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1217 | |
1218 | cpu = task_cpu(p); | |
1219 | if (cpu == smp_processor_id()) | |
1220 | return; | |
1221 | ||
1222 | /* NEED_RESCHED must be visible before we test polling */ | |
1223 | smp_mb(); | |
1224 | if (!tsk_is_polling(p)) | |
1225 | smp_send_reschedule(cpu); | |
1226 | } | |
1227 | ||
1228 | static void resched_cpu(int cpu) | |
1229 | { | |
1230 | struct rq *rq = cpu_rq(cpu); | |
1231 | unsigned long flags; | |
1232 | ||
1233 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1234 | return; | |
1235 | resched_task(cpu_curr(cpu)); | |
1236 | spin_unlock_irqrestore(&rq->lock, flags); | |
1237 | } | |
06d8308c TG |
1238 | |
1239 | #ifdef CONFIG_NO_HZ | |
1240 | /* | |
1241 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1242 | * idle CPU then this timer might expire before the next timer event | |
1243 | * which is scheduled to wake up that CPU. In case of a completely | |
1244 | * idle system the next event might even be infinite time into the | |
1245 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1246 | * leaves the inner idle loop so the newly added timer is taken into | |
1247 | * account when the CPU goes back to idle and evaluates the timer | |
1248 | * wheel for the next timer event. | |
1249 | */ | |
1250 | void wake_up_idle_cpu(int cpu) | |
1251 | { | |
1252 | struct rq *rq = cpu_rq(cpu); | |
1253 | ||
1254 | if (cpu == smp_processor_id()) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * This is safe, as this function is called with the timer | |
1259 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1260 | * to idle and has not yet set rq->curr to idle then it will | |
1261 | * be serialized on the timer wheel base lock and take the new | |
1262 | * timer into account automatically. | |
1263 | */ | |
1264 | if (rq->curr != rq->idle) | |
1265 | return; | |
1266 | ||
1267 | /* | |
1268 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1269 | * lockless. The worst case is that the other CPU runs the | |
1270 | * idle task through an additional NOOP schedule() | |
1271 | */ | |
1272 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1273 | ||
1274 | /* NEED_RESCHED must be visible before we test polling */ | |
1275 | smp_mb(); | |
1276 | if (!tsk_is_polling(rq->idle)) | |
1277 | smp_send_reschedule(cpu); | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1280 | |
6d6bc0ad | 1281 | #else /* !CONFIG_SMP */ |
8f4d37ec | 1282 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1283 | { |
1284 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1285 | set_tsk_thread_flag(p, tif_bit); |
c24d20db | 1286 | } |
6d6bc0ad | 1287 | #endif /* CONFIG_SMP */ |
c24d20db | 1288 | |
45bf76df IM |
1289 | #if BITS_PER_LONG == 32 |
1290 | # define WMULT_CONST (~0UL) | |
1291 | #else | |
1292 | # define WMULT_CONST (1UL << 32) | |
1293 | #endif | |
1294 | ||
1295 | #define WMULT_SHIFT 32 | |
1296 | ||
194081eb IM |
1297 | /* |
1298 | * Shift right and round: | |
1299 | */ | |
cf2ab469 | 1300 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1301 | |
a7be37ac PZ |
1302 | /* |
1303 | * delta *= weight / lw | |
1304 | */ | |
cb1c4fc9 | 1305 | static unsigned long |
45bf76df IM |
1306 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1307 | struct load_weight *lw) | |
1308 | { | |
1309 | u64 tmp; | |
1310 | ||
7a232e03 LJ |
1311 | if (!lw->inv_weight) { |
1312 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1313 | lw->inv_weight = 1; | |
1314 | else | |
1315 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1316 | / (lw->weight+1); | |
1317 | } | |
45bf76df IM |
1318 | |
1319 | tmp = (u64)delta_exec * weight; | |
1320 | /* | |
1321 | * Check whether we'd overflow the 64-bit multiplication: | |
1322 | */ | |
194081eb | 1323 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1324 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1325 | WMULT_SHIFT/2); |
1326 | else | |
cf2ab469 | 1327 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1328 | |
ecf691da | 1329 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1333 | { |
1334 | lw->weight += inc; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
1091985b | 1338 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1339 | { |
1340 | lw->weight -= dec; | |
e89996ae | 1341 | lw->inv_weight = 0; |
45bf76df IM |
1342 | } |
1343 | ||
2dd73a4f PW |
1344 | /* |
1345 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1346 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1347 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1348 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1349 | * scaled version of the new time slice allocation that they receive on time |
1350 | * slice expiry etc. | |
1351 | */ | |
1352 | ||
dd41f596 IM |
1353 | #define WEIGHT_IDLEPRIO 2 |
1354 | #define WMULT_IDLEPRIO (1 << 31) | |
1355 | ||
1356 | /* | |
1357 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1358 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1359 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1360 | * that remained on nice 0. | |
1361 | * | |
1362 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1363 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1364 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1365 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1366 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1367 | */ |
1368 | static const int prio_to_weight[40] = { | |
254753dc IM |
1369 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1370 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1371 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1372 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1373 | /* 0 */ 1024, 820, 655, 526, 423, | |
1374 | /* 5 */ 335, 272, 215, 172, 137, | |
1375 | /* 10 */ 110, 87, 70, 56, 45, | |
1376 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1377 | }; |
1378 | ||
5714d2de IM |
1379 | /* |
1380 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1381 | * | |
1382 | * In cases where the weight does not change often, we can use the | |
1383 | * precalculated inverse to speed up arithmetics by turning divisions | |
1384 | * into multiplications: | |
1385 | */ | |
dd41f596 | 1386 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1387 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1388 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1389 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1390 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1391 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1392 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1393 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1394 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1395 | }; |
2dd73a4f | 1396 | |
dd41f596 IM |
1397 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1398 | ||
1399 | /* | |
1400 | * runqueue iterator, to support SMP load-balancing between different | |
1401 | * scheduling classes, without having to expose their internal data | |
1402 | * structures to the load-balancing proper: | |
1403 | */ | |
1404 | struct rq_iterator { | |
1405 | void *arg; | |
1406 | struct task_struct *(*start)(void *); | |
1407 | struct task_struct *(*next)(void *); | |
1408 | }; | |
1409 | ||
e1d1484f PW |
1410 | #ifdef CONFIG_SMP |
1411 | static unsigned long | |
1412 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1413 | unsigned long max_load_move, struct sched_domain *sd, | |
1414 | enum cpu_idle_type idle, int *all_pinned, | |
1415 | int *this_best_prio, struct rq_iterator *iterator); | |
1416 | ||
1417 | static int | |
1418 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1419 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1420 | struct rq_iterator *iterator); | |
e1d1484f | 1421 | #endif |
dd41f596 | 1422 | |
d842de87 SV |
1423 | #ifdef CONFIG_CGROUP_CPUACCT |
1424 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1425 | #else | |
1426 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1427 | #endif | |
1428 | ||
18d95a28 PZ |
1429 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1430 | { | |
1431 | update_load_add(&rq->load, load); | |
1432 | } | |
1433 | ||
1434 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1435 | { | |
1436 | update_load_sub(&rq->load, load); | |
1437 | } | |
1438 | ||
e7693a36 GH |
1439 | #ifdef CONFIG_SMP |
1440 | static unsigned long source_load(int cpu, int type); | |
1441 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1442 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1443 | |
a8a51d5e PZ |
1444 | static unsigned long cpu_avg_load_per_task(int cpu) |
1445 | { | |
1446 | struct rq *rq = cpu_rq(cpu); | |
1447 | ||
1448 | if (rq->nr_running) | |
1449 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1450 | ||
1451 | return rq->avg_load_per_task; | |
1452 | } | |
1453 | ||
c09595f6 PZ |
1454 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1455 | ||
c8cba857 | 1456 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1457 | |
1458 | /* | |
1459 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1460 | * leaving it for the final time. | |
1461 | */ | |
c8cba857 PZ |
1462 | static void |
1463 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1464 | { |
1465 | struct task_group *parent, *child; | |
1466 | ||
1467 | rcu_read_lock(); | |
1468 | parent = &root_task_group; | |
1469 | down: | |
b6a86c74 | 1470 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1471 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1472 | parent = child; | |
1473 | goto down; | |
1474 | ||
1475 | up: | |
1476 | continue; | |
1477 | } | |
b6a86c74 | 1478 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1479 | |
1480 | child = parent; | |
1481 | parent = parent->parent; | |
1482 | if (parent) | |
1483 | goto up; | |
1484 | rcu_read_unlock(); | |
1485 | } | |
1486 | ||
c09595f6 PZ |
1487 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1488 | ||
1489 | /* | |
1490 | * Calculate and set the cpu's group shares. | |
1491 | */ | |
1492 | static void | |
b6a86c74 | 1493 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1494 | unsigned long sd_shares, unsigned long sd_rq_weight) |
c09595f6 PZ |
1495 | { |
1496 | int boost = 0; | |
1497 | unsigned long shares; | |
1498 | unsigned long rq_weight; | |
1499 | ||
c8cba857 | 1500 | if (!tg->se[cpu]) |
c09595f6 PZ |
1501 | return; |
1502 | ||
c8cba857 | 1503 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1504 | |
1505 | /* | |
1506 | * If there are currently no tasks on the cpu pretend there is one of | |
1507 | * average load so that when a new task gets to run here it will not | |
1508 | * get delayed by group starvation. | |
1509 | */ | |
1510 | if (!rq_weight) { | |
1511 | boost = 1; | |
1512 | rq_weight = NICE_0_LOAD; | |
1513 | } | |
1514 | ||
c8cba857 PZ |
1515 | if (unlikely(rq_weight > sd_rq_weight)) |
1516 | rq_weight = sd_rq_weight; | |
1517 | ||
c09595f6 PZ |
1518 | /* |
1519 | * \Sum shares * rq_weight | |
1520 | * shares = ----------------------- | |
1521 | * \Sum rq_weight | |
1522 | * | |
1523 | */ | |
c8cba857 | 1524 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1525 | |
1526 | /* | |
1527 | * record the actual number of shares, not the boosted amount. | |
1528 | */ | |
c8cba857 | 1529 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
c09595f6 PZ |
1530 | |
1531 | if (shares < MIN_SHARES) | |
1532 | shares = MIN_SHARES; | |
1533 | else if (shares > MAX_SHARES) | |
1534 | shares = MAX_SHARES; | |
1535 | ||
c8cba857 | 1536 | __set_se_shares(tg->se[cpu], shares); |
c09595f6 PZ |
1537 | } |
1538 | ||
1539 | /* | |
c8cba857 PZ |
1540 | * Re-compute the task group their per cpu shares over the given domain. |
1541 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1542 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1543 | */ |
1544 | static void | |
c8cba857 | 1545 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1546 | { |
c8cba857 PZ |
1547 | unsigned long rq_weight = 0; |
1548 | unsigned long shares = 0; | |
1549 | int i; | |
c09595f6 | 1550 | |
c8cba857 PZ |
1551 | for_each_cpu_mask(i, sd->span) { |
1552 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1553 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1554 | } |
c09595f6 | 1555 | |
c8cba857 PZ |
1556 | if ((!shares && rq_weight) || shares > tg->shares) |
1557 | shares = tg->shares; | |
1558 | ||
1559 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1560 | shares = tg->shares; | |
c09595f6 | 1561 | |
cd80917e PZ |
1562 | if (!rq_weight) |
1563 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1564 | ||
c09595f6 PZ |
1565 | for_each_cpu_mask(i, sd->span) { |
1566 | struct rq *rq = cpu_rq(i); | |
1567 | unsigned long flags; | |
1568 | ||
1569 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1570 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1571 | spin_unlock_irqrestore(&rq->lock, flags); |
1572 | } | |
c09595f6 PZ |
1573 | } |
1574 | ||
1575 | /* | |
c8cba857 PZ |
1576 | * Compute the cpu's hierarchical load factor for each task group. |
1577 | * This needs to be done in a top-down fashion because the load of a child | |
1578 | * group is a fraction of its parents load. | |
c09595f6 | 1579 | */ |
b6a86c74 | 1580 | static void |
c8cba857 | 1581 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1582 | { |
c8cba857 | 1583 | unsigned long load; |
c09595f6 | 1584 | |
c8cba857 PZ |
1585 | if (!tg->parent) { |
1586 | load = cpu_rq(cpu)->load.weight; | |
1587 | } else { | |
1588 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1589 | load *= tg->cfs_rq[cpu]->shares; | |
1590 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1591 | } | |
c09595f6 | 1592 | |
c8cba857 | 1593 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1594 | } |
1595 | ||
c8cba857 PZ |
1596 | static void |
1597 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1598 | { |
c09595f6 PZ |
1599 | } |
1600 | ||
c8cba857 | 1601 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1602 | { |
2398f2c6 PZ |
1603 | u64 now = cpu_clock(raw_smp_processor_id()); |
1604 | s64 elapsed = now - sd->last_update; | |
1605 | ||
1606 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1607 | sd->last_update = now; | |
1608 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); | |
1609 | } | |
4d8d595d PZ |
1610 | } |
1611 | ||
3e5459b4 PZ |
1612 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1613 | { | |
1614 | spin_unlock(&rq->lock); | |
1615 | update_shares(sd); | |
1616 | spin_lock(&rq->lock); | |
1617 | } | |
1618 | ||
c8cba857 | 1619 | static void update_h_load(int cpu) |
c09595f6 | 1620 | { |
c8cba857 | 1621 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1622 | } |
1623 | ||
1624 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) | |
1625 | { | |
1626 | cfs_rq->shares = shares; | |
1627 | } | |
1628 | ||
1629 | #else | |
1630 | ||
c8cba857 | 1631 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1632 | { |
1633 | } | |
1634 | ||
3e5459b4 PZ |
1635 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1636 | { | |
1637 | } | |
1638 | ||
c09595f6 PZ |
1639 | #endif |
1640 | ||
18d95a28 PZ |
1641 | #endif |
1642 | ||
dd41f596 | 1643 | #include "sched_stats.h" |
dd41f596 | 1644 | #include "sched_idletask.c" |
5522d5d5 IM |
1645 | #include "sched_fair.c" |
1646 | #include "sched_rt.c" | |
dd41f596 IM |
1647 | #ifdef CONFIG_SCHED_DEBUG |
1648 | # include "sched_debug.c" | |
1649 | #endif | |
1650 | ||
1651 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1652 | #define for_each_class(class) \ |
1653 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1654 | |
c09595f6 | 1655 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1656 | { |
1657 | rq->nr_running++; | |
9c217245 IM |
1658 | } |
1659 | ||
c09595f6 | 1660 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1661 | { |
1662 | rq->nr_running--; | |
9c217245 IM |
1663 | } |
1664 | ||
45bf76df IM |
1665 | static void set_load_weight(struct task_struct *p) |
1666 | { | |
1667 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1668 | p->se.load.weight = prio_to_weight[0] * 2; |
1669 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1670 | return; | |
1671 | } | |
45bf76df | 1672 | |
dd41f596 IM |
1673 | /* |
1674 | * SCHED_IDLE tasks get minimal weight: | |
1675 | */ | |
1676 | if (p->policy == SCHED_IDLE) { | |
1677 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1678 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1679 | return; | |
1680 | } | |
71f8bd46 | 1681 | |
dd41f596 IM |
1682 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1683 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1684 | } |
1685 | ||
8159f87e | 1686 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1687 | { |
dd41f596 | 1688 | sched_info_queued(p); |
fd390f6a | 1689 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1690 | p->se.on_rq = 1; |
71f8bd46 IM |
1691 | } |
1692 | ||
69be72c1 | 1693 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1694 | { |
f02231e5 | 1695 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1696 | p->se.on_rq = 0; |
71f8bd46 IM |
1697 | } |
1698 | ||
14531189 | 1699 | /* |
dd41f596 | 1700 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1701 | */ |
14531189 IM |
1702 | static inline int __normal_prio(struct task_struct *p) |
1703 | { | |
dd41f596 | 1704 | return p->static_prio; |
14531189 IM |
1705 | } |
1706 | ||
b29739f9 IM |
1707 | /* |
1708 | * Calculate the expected normal priority: i.e. priority | |
1709 | * without taking RT-inheritance into account. Might be | |
1710 | * boosted by interactivity modifiers. Changes upon fork, | |
1711 | * setprio syscalls, and whenever the interactivity | |
1712 | * estimator recalculates. | |
1713 | */ | |
36c8b586 | 1714 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1715 | { |
1716 | int prio; | |
1717 | ||
e05606d3 | 1718 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1719 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1720 | else | |
1721 | prio = __normal_prio(p); | |
1722 | return prio; | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * Calculate the current priority, i.e. the priority | |
1727 | * taken into account by the scheduler. This value might | |
1728 | * be boosted by RT tasks, or might be boosted by | |
1729 | * interactivity modifiers. Will be RT if the task got | |
1730 | * RT-boosted. If not then it returns p->normal_prio. | |
1731 | */ | |
36c8b586 | 1732 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1733 | { |
1734 | p->normal_prio = normal_prio(p); | |
1735 | /* | |
1736 | * If we are RT tasks or we were boosted to RT priority, | |
1737 | * keep the priority unchanged. Otherwise, update priority | |
1738 | * to the normal priority: | |
1739 | */ | |
1740 | if (!rt_prio(p->prio)) | |
1741 | return p->normal_prio; | |
1742 | return p->prio; | |
1743 | } | |
1744 | ||
1da177e4 | 1745 | /* |
dd41f596 | 1746 | * activate_task - move a task to the runqueue. |
1da177e4 | 1747 | */ |
dd41f596 | 1748 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1749 | { |
d9514f6c | 1750 | if (task_contributes_to_load(p)) |
dd41f596 | 1751 | rq->nr_uninterruptible--; |
1da177e4 | 1752 | |
8159f87e | 1753 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1754 | inc_nr_running(rq); |
1da177e4 LT |
1755 | } |
1756 | ||
1da177e4 LT |
1757 | /* |
1758 | * deactivate_task - remove a task from the runqueue. | |
1759 | */ | |
2e1cb74a | 1760 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1761 | { |
d9514f6c | 1762 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1763 | rq->nr_uninterruptible++; |
1764 | ||
69be72c1 | 1765 | dequeue_task(rq, p, sleep); |
c09595f6 | 1766 | dec_nr_running(rq); |
1da177e4 LT |
1767 | } |
1768 | ||
1da177e4 LT |
1769 | /** |
1770 | * task_curr - is this task currently executing on a CPU? | |
1771 | * @p: the task in question. | |
1772 | */ | |
36c8b586 | 1773 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1774 | { |
1775 | return cpu_curr(task_cpu(p)) == p; | |
1776 | } | |
1777 | ||
dd41f596 IM |
1778 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1779 | { | |
6f505b16 | 1780 | set_task_rq(p, cpu); |
dd41f596 | 1781 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1782 | /* |
1783 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1784 | * successfuly executed on another CPU. We must ensure that updates of | |
1785 | * per-task data have been completed by this moment. | |
1786 | */ | |
1787 | smp_wmb(); | |
dd41f596 | 1788 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1789 | #endif |
2dd73a4f PW |
1790 | } |
1791 | ||
cb469845 SR |
1792 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1793 | const struct sched_class *prev_class, | |
1794 | int oldprio, int running) | |
1795 | { | |
1796 | if (prev_class != p->sched_class) { | |
1797 | if (prev_class->switched_from) | |
1798 | prev_class->switched_from(rq, p, running); | |
1799 | p->sched_class->switched_to(rq, p, running); | |
1800 | } else | |
1801 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1802 | } | |
1803 | ||
1da177e4 | 1804 | #ifdef CONFIG_SMP |
c65cc870 | 1805 | |
e958b360 TG |
1806 | /* Used instead of source_load when we know the type == 0 */ |
1807 | static unsigned long weighted_cpuload(const int cpu) | |
1808 | { | |
1809 | return cpu_rq(cpu)->load.weight; | |
1810 | } | |
1811 | ||
cc367732 IM |
1812 | /* |
1813 | * Is this task likely cache-hot: | |
1814 | */ | |
e7693a36 | 1815 | static int |
cc367732 IM |
1816 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1817 | { | |
1818 | s64 delta; | |
1819 | ||
f540a608 IM |
1820 | /* |
1821 | * Buddy candidates are cache hot: | |
1822 | */ | |
d25ce4cd | 1823 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1824 | return 1; |
1825 | ||
cc367732 IM |
1826 | if (p->sched_class != &fair_sched_class) |
1827 | return 0; | |
1828 | ||
6bc1665b IM |
1829 | if (sysctl_sched_migration_cost == -1) |
1830 | return 1; | |
1831 | if (sysctl_sched_migration_cost == 0) | |
1832 | return 0; | |
1833 | ||
cc367732 IM |
1834 | delta = now - p->se.exec_start; |
1835 | ||
1836 | return delta < (s64)sysctl_sched_migration_cost; | |
1837 | } | |
1838 | ||
1839 | ||
dd41f596 | 1840 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1841 | { |
dd41f596 IM |
1842 | int old_cpu = task_cpu(p); |
1843 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1844 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1845 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1846 | u64 clock_offset; |
dd41f596 IM |
1847 | |
1848 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1849 | |
1850 | #ifdef CONFIG_SCHEDSTATS | |
1851 | if (p->se.wait_start) | |
1852 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1853 | if (p->se.sleep_start) |
1854 | p->se.sleep_start -= clock_offset; | |
1855 | if (p->se.block_start) | |
1856 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1857 | if (old_cpu != new_cpu) { |
1858 | schedstat_inc(p, se.nr_migrations); | |
1859 | if (task_hot(p, old_rq->clock, NULL)) | |
1860 | schedstat_inc(p, se.nr_forced2_migrations); | |
1861 | } | |
6cfb0d5d | 1862 | #endif |
2830cf8c SV |
1863 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1864 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1865 | |
1866 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1867 | } |
1868 | ||
70b97a7f | 1869 | struct migration_req { |
1da177e4 | 1870 | struct list_head list; |
1da177e4 | 1871 | |
36c8b586 | 1872 | struct task_struct *task; |
1da177e4 LT |
1873 | int dest_cpu; |
1874 | ||
1da177e4 | 1875 | struct completion done; |
70b97a7f | 1876 | }; |
1da177e4 LT |
1877 | |
1878 | /* | |
1879 | * The task's runqueue lock must be held. | |
1880 | * Returns true if you have to wait for migration thread. | |
1881 | */ | |
36c8b586 | 1882 | static int |
70b97a7f | 1883 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1884 | { |
70b97a7f | 1885 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1886 | |
1887 | /* | |
1888 | * If the task is not on a runqueue (and not running), then | |
1889 | * it is sufficient to simply update the task's cpu field. | |
1890 | */ | |
dd41f596 | 1891 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1892 | set_task_cpu(p, dest_cpu); |
1893 | return 0; | |
1894 | } | |
1895 | ||
1896 | init_completion(&req->done); | |
1da177e4 LT |
1897 | req->task = p; |
1898 | req->dest_cpu = dest_cpu; | |
1899 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1900 | |
1da177e4 LT |
1901 | return 1; |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * wait_task_inactive - wait for a thread to unschedule. | |
1906 | * | |
1907 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1908 | * else this function might spin for a *long* time. This function can't | |
1909 | * be called with interrupts off, or it may introduce deadlock with | |
1910 | * smp_call_function() if an IPI is sent by the same process we are | |
1911 | * waiting to become inactive. | |
1912 | */ | |
36c8b586 | 1913 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1914 | { |
1915 | unsigned long flags; | |
dd41f596 | 1916 | int running, on_rq; |
70b97a7f | 1917 | struct rq *rq; |
1da177e4 | 1918 | |
3a5c359a AK |
1919 | for (;;) { |
1920 | /* | |
1921 | * We do the initial early heuristics without holding | |
1922 | * any task-queue locks at all. We'll only try to get | |
1923 | * the runqueue lock when things look like they will | |
1924 | * work out! | |
1925 | */ | |
1926 | rq = task_rq(p); | |
fa490cfd | 1927 | |
3a5c359a AK |
1928 | /* |
1929 | * If the task is actively running on another CPU | |
1930 | * still, just relax and busy-wait without holding | |
1931 | * any locks. | |
1932 | * | |
1933 | * NOTE! Since we don't hold any locks, it's not | |
1934 | * even sure that "rq" stays as the right runqueue! | |
1935 | * But we don't care, since "task_running()" will | |
1936 | * return false if the runqueue has changed and p | |
1937 | * is actually now running somewhere else! | |
1938 | */ | |
1939 | while (task_running(rq, p)) | |
1940 | cpu_relax(); | |
fa490cfd | 1941 | |
3a5c359a AK |
1942 | /* |
1943 | * Ok, time to look more closely! We need the rq | |
1944 | * lock now, to be *sure*. If we're wrong, we'll | |
1945 | * just go back and repeat. | |
1946 | */ | |
1947 | rq = task_rq_lock(p, &flags); | |
1948 | running = task_running(rq, p); | |
1949 | on_rq = p->se.on_rq; | |
1950 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1951 | |
3a5c359a AK |
1952 | /* |
1953 | * Was it really running after all now that we | |
1954 | * checked with the proper locks actually held? | |
1955 | * | |
1956 | * Oops. Go back and try again.. | |
1957 | */ | |
1958 | if (unlikely(running)) { | |
1959 | cpu_relax(); | |
1960 | continue; | |
1961 | } | |
fa490cfd | 1962 | |
3a5c359a AK |
1963 | /* |
1964 | * It's not enough that it's not actively running, | |
1965 | * it must be off the runqueue _entirely_, and not | |
1966 | * preempted! | |
1967 | * | |
1968 | * So if it wa still runnable (but just not actively | |
1969 | * running right now), it's preempted, and we should | |
1970 | * yield - it could be a while. | |
1971 | */ | |
1972 | if (unlikely(on_rq)) { | |
1973 | schedule_timeout_uninterruptible(1); | |
1974 | continue; | |
1975 | } | |
fa490cfd | 1976 | |
3a5c359a AK |
1977 | /* |
1978 | * Ahh, all good. It wasn't running, and it wasn't | |
1979 | * runnable, which means that it will never become | |
1980 | * running in the future either. We're all done! | |
1981 | */ | |
1982 | break; | |
1983 | } | |
1da177e4 LT |
1984 | } |
1985 | ||
1986 | /*** | |
1987 | * kick_process - kick a running thread to enter/exit the kernel | |
1988 | * @p: the to-be-kicked thread | |
1989 | * | |
1990 | * Cause a process which is running on another CPU to enter | |
1991 | * kernel-mode, without any delay. (to get signals handled.) | |
1992 | * | |
1993 | * NOTE: this function doesnt have to take the runqueue lock, | |
1994 | * because all it wants to ensure is that the remote task enters | |
1995 | * the kernel. If the IPI races and the task has been migrated | |
1996 | * to another CPU then no harm is done and the purpose has been | |
1997 | * achieved as well. | |
1998 | */ | |
36c8b586 | 1999 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2000 | { |
2001 | int cpu; | |
2002 | ||
2003 | preempt_disable(); | |
2004 | cpu = task_cpu(p); | |
2005 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2006 | smp_send_reschedule(cpu); | |
2007 | preempt_enable(); | |
2008 | } | |
2009 | ||
2010 | /* | |
2dd73a4f PW |
2011 | * Return a low guess at the load of a migration-source cpu weighted |
2012 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2013 | * |
2014 | * We want to under-estimate the load of migration sources, to | |
2015 | * balance conservatively. | |
2016 | */ | |
a9957449 | 2017 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2018 | { |
70b97a7f | 2019 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2020 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2021 | |
93b75217 | 2022 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2023 | return total; |
b910472d | 2024 | |
dd41f596 | 2025 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2026 | } |
2027 | ||
2028 | /* | |
2dd73a4f PW |
2029 | * Return a high guess at the load of a migration-target cpu weighted |
2030 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2031 | */ |
a9957449 | 2032 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2033 | { |
70b97a7f | 2034 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2035 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2036 | |
93b75217 | 2037 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2038 | return total; |
3b0bd9bc | 2039 | |
dd41f596 | 2040 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2041 | } |
2042 | ||
147cbb4b NP |
2043 | /* |
2044 | * find_idlest_group finds and returns the least busy CPU group within the | |
2045 | * domain. | |
2046 | */ | |
2047 | static struct sched_group * | |
2048 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2049 | { | |
2050 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2051 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2052 | int load_idx = sd->forkexec_idx; | |
2053 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2054 | ||
2055 | do { | |
2056 | unsigned long load, avg_load; | |
2057 | int local_group; | |
2058 | int i; | |
2059 | ||
da5a5522 BD |
2060 | /* Skip over this group if it has no CPUs allowed */ |
2061 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2062 | continue; |
da5a5522 | 2063 | |
147cbb4b | 2064 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2065 | |
2066 | /* Tally up the load of all CPUs in the group */ | |
2067 | avg_load = 0; | |
2068 | ||
2069 | for_each_cpu_mask(i, group->cpumask) { | |
2070 | /* Bias balancing toward cpus of our domain */ | |
2071 | if (local_group) | |
2072 | load = source_load(i, load_idx); | |
2073 | else | |
2074 | load = target_load(i, load_idx); | |
2075 | ||
2076 | avg_load += load; | |
2077 | } | |
2078 | ||
2079 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2080 | avg_load = sg_div_cpu_power(group, |
2081 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2082 | |
2083 | if (local_group) { | |
2084 | this_load = avg_load; | |
2085 | this = group; | |
2086 | } else if (avg_load < min_load) { | |
2087 | min_load = avg_load; | |
2088 | idlest = group; | |
2089 | } | |
3a5c359a | 2090 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2091 | |
2092 | if (!idlest || 100*this_load < imbalance*min_load) | |
2093 | return NULL; | |
2094 | return idlest; | |
2095 | } | |
2096 | ||
2097 | /* | |
0feaece9 | 2098 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2099 | */ |
95cdf3b7 | 2100 | static int |
7c16ec58 MT |
2101 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2102 | cpumask_t *tmp) | |
147cbb4b NP |
2103 | { |
2104 | unsigned long load, min_load = ULONG_MAX; | |
2105 | int idlest = -1; | |
2106 | int i; | |
2107 | ||
da5a5522 | 2108 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2109 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2110 | |
7c16ec58 | 2111 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 2112 | load = weighted_cpuload(i); |
147cbb4b NP |
2113 | |
2114 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2115 | min_load = load; | |
2116 | idlest = i; | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | return idlest; | |
2121 | } | |
2122 | ||
476d139c NP |
2123 | /* |
2124 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2125 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2126 | * SD_BALANCE_EXEC. | |
2127 | * | |
2128 | * Balance, ie. select the least loaded group. | |
2129 | * | |
2130 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2131 | * | |
2132 | * preempt must be disabled. | |
2133 | */ | |
2134 | static int sched_balance_self(int cpu, int flag) | |
2135 | { | |
2136 | struct task_struct *t = current; | |
2137 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2138 | |
c96d145e | 2139 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2140 | /* |
2141 | * If power savings logic is enabled for a domain, stop there. | |
2142 | */ | |
5c45bf27 SS |
2143 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2144 | break; | |
476d139c NP |
2145 | if (tmp->flags & flag) |
2146 | sd = tmp; | |
c96d145e | 2147 | } |
476d139c | 2148 | |
039a1c41 PZ |
2149 | if (sd) |
2150 | update_shares(sd); | |
2151 | ||
476d139c | 2152 | while (sd) { |
7c16ec58 | 2153 | cpumask_t span, tmpmask; |
476d139c | 2154 | struct sched_group *group; |
1a848870 SS |
2155 | int new_cpu, weight; |
2156 | ||
2157 | if (!(sd->flags & flag)) { | |
2158 | sd = sd->child; | |
2159 | continue; | |
2160 | } | |
476d139c NP |
2161 | |
2162 | span = sd->span; | |
2163 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2164 | if (!group) { |
2165 | sd = sd->child; | |
2166 | continue; | |
2167 | } | |
476d139c | 2168 | |
7c16ec58 | 2169 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2170 | if (new_cpu == -1 || new_cpu == cpu) { |
2171 | /* Now try balancing at a lower domain level of cpu */ | |
2172 | sd = sd->child; | |
2173 | continue; | |
2174 | } | |
476d139c | 2175 | |
1a848870 | 2176 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2177 | cpu = new_cpu; |
476d139c NP |
2178 | sd = NULL; |
2179 | weight = cpus_weight(span); | |
2180 | for_each_domain(cpu, tmp) { | |
2181 | if (weight <= cpus_weight(tmp->span)) | |
2182 | break; | |
2183 | if (tmp->flags & flag) | |
2184 | sd = tmp; | |
2185 | } | |
2186 | /* while loop will break here if sd == NULL */ | |
2187 | } | |
2188 | ||
2189 | return cpu; | |
2190 | } | |
2191 | ||
2192 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2193 | |
1da177e4 LT |
2194 | /*** |
2195 | * try_to_wake_up - wake up a thread | |
2196 | * @p: the to-be-woken-up thread | |
2197 | * @state: the mask of task states that can be woken | |
2198 | * @sync: do a synchronous wakeup? | |
2199 | * | |
2200 | * Put it on the run-queue if it's not already there. The "current" | |
2201 | * thread is always on the run-queue (except when the actual | |
2202 | * re-schedule is in progress), and as such you're allowed to do | |
2203 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2204 | * runnable without the overhead of this. | |
2205 | * | |
2206 | * returns failure only if the task is already active. | |
2207 | */ | |
36c8b586 | 2208 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2209 | { |
cc367732 | 2210 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2211 | unsigned long flags; |
2212 | long old_state; | |
70b97a7f | 2213 | struct rq *rq; |
1da177e4 | 2214 | |
b85d0667 IM |
2215 | if (!sched_feat(SYNC_WAKEUPS)) |
2216 | sync = 0; | |
2217 | ||
2398f2c6 PZ |
2218 | #ifdef CONFIG_SMP |
2219 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2220 | struct sched_domain *sd; | |
2221 | ||
2222 | this_cpu = raw_smp_processor_id(); | |
2223 | cpu = task_cpu(p); | |
2224 | ||
2225 | for_each_domain(this_cpu, sd) { | |
2226 | if (cpu_isset(cpu, sd->span)) { | |
2227 | update_shares(sd); | |
2228 | break; | |
2229 | } | |
2230 | } | |
2231 | } | |
2232 | #endif | |
2233 | ||
04e2f174 | 2234 | smp_wmb(); |
1da177e4 LT |
2235 | rq = task_rq_lock(p, &flags); |
2236 | old_state = p->state; | |
2237 | if (!(old_state & state)) | |
2238 | goto out; | |
2239 | ||
dd41f596 | 2240 | if (p->se.on_rq) |
1da177e4 LT |
2241 | goto out_running; |
2242 | ||
2243 | cpu = task_cpu(p); | |
cc367732 | 2244 | orig_cpu = cpu; |
1da177e4 LT |
2245 | this_cpu = smp_processor_id(); |
2246 | ||
2247 | #ifdef CONFIG_SMP | |
2248 | if (unlikely(task_running(rq, p))) | |
2249 | goto out_activate; | |
2250 | ||
5d2f5a61 DA |
2251 | cpu = p->sched_class->select_task_rq(p, sync); |
2252 | if (cpu != orig_cpu) { | |
2253 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2254 | task_rq_unlock(rq, &flags); |
2255 | /* might preempt at this point */ | |
2256 | rq = task_rq_lock(p, &flags); | |
2257 | old_state = p->state; | |
2258 | if (!(old_state & state)) | |
2259 | goto out; | |
dd41f596 | 2260 | if (p->se.on_rq) |
1da177e4 LT |
2261 | goto out_running; |
2262 | ||
2263 | this_cpu = smp_processor_id(); | |
2264 | cpu = task_cpu(p); | |
2265 | } | |
2266 | ||
e7693a36 GH |
2267 | #ifdef CONFIG_SCHEDSTATS |
2268 | schedstat_inc(rq, ttwu_count); | |
2269 | if (cpu == this_cpu) | |
2270 | schedstat_inc(rq, ttwu_local); | |
2271 | else { | |
2272 | struct sched_domain *sd; | |
2273 | for_each_domain(this_cpu, sd) { | |
2274 | if (cpu_isset(cpu, sd->span)) { | |
2275 | schedstat_inc(sd, ttwu_wake_remote); | |
2276 | break; | |
2277 | } | |
2278 | } | |
2279 | } | |
6d6bc0ad | 2280 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2281 | |
1da177e4 LT |
2282 | out_activate: |
2283 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2284 | schedstat_inc(p, se.nr_wakeups); |
2285 | if (sync) | |
2286 | schedstat_inc(p, se.nr_wakeups_sync); | |
2287 | if (orig_cpu != cpu) | |
2288 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2289 | if (cpu == this_cpu) | |
2290 | schedstat_inc(p, se.nr_wakeups_local); | |
2291 | else | |
2292 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2293 | update_rq_clock(rq); |
dd41f596 | 2294 | activate_task(rq, p, 1); |
1da177e4 LT |
2295 | success = 1; |
2296 | ||
2297 | out_running: | |
4ae7d5ce IM |
2298 | check_preempt_curr(rq, p); |
2299 | ||
1da177e4 | 2300 | p->state = TASK_RUNNING; |
9a897c5a SR |
2301 | #ifdef CONFIG_SMP |
2302 | if (p->sched_class->task_wake_up) | |
2303 | p->sched_class->task_wake_up(rq, p); | |
2304 | #endif | |
1da177e4 LT |
2305 | out: |
2306 | task_rq_unlock(rq, &flags); | |
2307 | ||
2308 | return success; | |
2309 | } | |
2310 | ||
7ad5b3a5 | 2311 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2312 | { |
d9514f6c | 2313 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2314 | } |
1da177e4 LT |
2315 | EXPORT_SYMBOL(wake_up_process); |
2316 | ||
7ad5b3a5 | 2317 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2318 | { |
2319 | return try_to_wake_up(p, state, 0); | |
2320 | } | |
2321 | ||
1da177e4 LT |
2322 | /* |
2323 | * Perform scheduler related setup for a newly forked process p. | |
2324 | * p is forked by current. | |
dd41f596 IM |
2325 | * |
2326 | * __sched_fork() is basic setup used by init_idle() too: | |
2327 | */ | |
2328 | static void __sched_fork(struct task_struct *p) | |
2329 | { | |
dd41f596 IM |
2330 | p->se.exec_start = 0; |
2331 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2332 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2333 | p->se.last_wakeup = 0; |
2334 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2335 | |
2336 | #ifdef CONFIG_SCHEDSTATS | |
2337 | p->se.wait_start = 0; | |
dd41f596 IM |
2338 | p->se.sum_sleep_runtime = 0; |
2339 | p->se.sleep_start = 0; | |
dd41f596 IM |
2340 | p->se.block_start = 0; |
2341 | p->se.sleep_max = 0; | |
2342 | p->se.block_max = 0; | |
2343 | p->se.exec_max = 0; | |
eba1ed4b | 2344 | p->se.slice_max = 0; |
dd41f596 | 2345 | p->se.wait_max = 0; |
6cfb0d5d | 2346 | #endif |
476d139c | 2347 | |
fa717060 | 2348 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2349 | p->se.on_rq = 0; |
4a55bd5e | 2350 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2351 | |
e107be36 AK |
2352 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2353 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2354 | #endif | |
2355 | ||
1da177e4 LT |
2356 | /* |
2357 | * We mark the process as running here, but have not actually | |
2358 | * inserted it onto the runqueue yet. This guarantees that | |
2359 | * nobody will actually run it, and a signal or other external | |
2360 | * event cannot wake it up and insert it on the runqueue either. | |
2361 | */ | |
2362 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2363 | } |
2364 | ||
2365 | /* | |
2366 | * fork()/clone()-time setup: | |
2367 | */ | |
2368 | void sched_fork(struct task_struct *p, int clone_flags) | |
2369 | { | |
2370 | int cpu = get_cpu(); | |
2371 | ||
2372 | __sched_fork(p); | |
2373 | ||
2374 | #ifdef CONFIG_SMP | |
2375 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2376 | #endif | |
02e4bac2 | 2377 | set_task_cpu(p, cpu); |
b29739f9 IM |
2378 | |
2379 | /* | |
2380 | * Make sure we do not leak PI boosting priority to the child: | |
2381 | */ | |
2382 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2383 | if (!rt_prio(p->prio)) |
2384 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2385 | |
52f17b6c | 2386 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2387 | if (likely(sched_info_on())) |
52f17b6c | 2388 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2389 | #endif |
d6077cb8 | 2390 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2391 | p->oncpu = 0; |
2392 | #endif | |
1da177e4 | 2393 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2394 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2395 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2396 | #endif |
476d139c | 2397 | put_cpu(); |
1da177e4 LT |
2398 | } |
2399 | ||
2400 | /* | |
2401 | * wake_up_new_task - wake up a newly created task for the first time. | |
2402 | * | |
2403 | * This function will do some initial scheduler statistics housekeeping | |
2404 | * that must be done for every newly created context, then puts the task | |
2405 | * on the runqueue and wakes it. | |
2406 | */ | |
7ad5b3a5 | 2407 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2408 | { |
2409 | unsigned long flags; | |
dd41f596 | 2410 | struct rq *rq; |
1da177e4 LT |
2411 | |
2412 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2413 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2414 | update_rq_clock(rq); |
1da177e4 LT |
2415 | |
2416 | p->prio = effective_prio(p); | |
2417 | ||
b9dca1e0 | 2418 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2419 | activate_task(rq, p, 0); |
1da177e4 | 2420 | } else { |
1da177e4 | 2421 | /* |
dd41f596 IM |
2422 | * Let the scheduling class do new task startup |
2423 | * management (if any): | |
1da177e4 | 2424 | */ |
ee0827d8 | 2425 | p->sched_class->task_new(rq, p); |
c09595f6 | 2426 | inc_nr_running(rq); |
1da177e4 | 2427 | } |
dd41f596 | 2428 | check_preempt_curr(rq, p); |
9a897c5a SR |
2429 | #ifdef CONFIG_SMP |
2430 | if (p->sched_class->task_wake_up) | |
2431 | p->sched_class->task_wake_up(rq, p); | |
2432 | #endif | |
dd41f596 | 2433 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2434 | } |
2435 | ||
e107be36 AK |
2436 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2437 | ||
2438 | /** | |
421cee29 RD |
2439 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2440 | * @notifier: notifier struct to register | |
e107be36 AK |
2441 | */ |
2442 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2443 | { | |
2444 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2445 | } | |
2446 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2447 | ||
2448 | /** | |
2449 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2450 | * @notifier: notifier struct to unregister |
e107be36 AK |
2451 | * |
2452 | * This is safe to call from within a preemption notifier. | |
2453 | */ | |
2454 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2455 | { | |
2456 | hlist_del(¬ifier->link); | |
2457 | } | |
2458 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2459 | ||
2460 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2461 | { | |
2462 | struct preempt_notifier *notifier; | |
2463 | struct hlist_node *node; | |
2464 | ||
2465 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2466 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2467 | } | |
2468 | ||
2469 | static void | |
2470 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2471 | struct task_struct *next) | |
2472 | { | |
2473 | struct preempt_notifier *notifier; | |
2474 | struct hlist_node *node; | |
2475 | ||
2476 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2477 | notifier->ops->sched_out(notifier, next); | |
2478 | } | |
2479 | ||
6d6bc0ad | 2480 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2481 | |
2482 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2483 | { | |
2484 | } | |
2485 | ||
2486 | static void | |
2487 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2488 | struct task_struct *next) | |
2489 | { | |
2490 | } | |
2491 | ||
6d6bc0ad | 2492 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2493 | |
4866cde0 NP |
2494 | /** |
2495 | * prepare_task_switch - prepare to switch tasks | |
2496 | * @rq: the runqueue preparing to switch | |
421cee29 | 2497 | * @prev: the current task that is being switched out |
4866cde0 NP |
2498 | * @next: the task we are going to switch to. |
2499 | * | |
2500 | * This is called with the rq lock held and interrupts off. It must | |
2501 | * be paired with a subsequent finish_task_switch after the context | |
2502 | * switch. | |
2503 | * | |
2504 | * prepare_task_switch sets up locking and calls architecture specific | |
2505 | * hooks. | |
2506 | */ | |
e107be36 AK |
2507 | static inline void |
2508 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2509 | struct task_struct *next) | |
4866cde0 | 2510 | { |
e107be36 | 2511 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2512 | prepare_lock_switch(rq, next); |
2513 | prepare_arch_switch(next); | |
2514 | } | |
2515 | ||
1da177e4 LT |
2516 | /** |
2517 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2518 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2519 | * @prev: the thread we just switched away from. |
2520 | * | |
4866cde0 NP |
2521 | * finish_task_switch must be called after the context switch, paired |
2522 | * with a prepare_task_switch call before the context switch. | |
2523 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2524 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2525 | * |
2526 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2527 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2528 | * with the lock held can cause deadlocks; see schedule() for |
2529 | * details.) | |
2530 | */ | |
a9957449 | 2531 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2532 | __releases(rq->lock) |
2533 | { | |
1da177e4 | 2534 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2535 | long prev_state; |
1da177e4 LT |
2536 | |
2537 | rq->prev_mm = NULL; | |
2538 | ||
2539 | /* | |
2540 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2541 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2542 | * schedule one last time. The schedule call will never return, and |
2543 | * the scheduled task must drop that reference. | |
c394cc9f | 2544 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2545 | * still held, otherwise prev could be scheduled on another cpu, die |
2546 | * there before we look at prev->state, and then the reference would | |
2547 | * be dropped twice. | |
2548 | * Manfred Spraul <manfred@colorfullife.com> | |
2549 | */ | |
55a101f8 | 2550 | prev_state = prev->state; |
4866cde0 NP |
2551 | finish_arch_switch(prev); |
2552 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2553 | #ifdef CONFIG_SMP |
2554 | if (current->sched_class->post_schedule) | |
2555 | current->sched_class->post_schedule(rq); | |
2556 | #endif | |
e8fa1362 | 2557 | |
e107be36 | 2558 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2559 | if (mm) |
2560 | mmdrop(mm); | |
c394cc9f | 2561 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2562 | /* |
2563 | * Remove function-return probe instances associated with this | |
2564 | * task and put them back on the free list. | |
9761eea8 | 2565 | */ |
c6fd91f0 | 2566 | kprobe_flush_task(prev); |
1da177e4 | 2567 | put_task_struct(prev); |
c6fd91f0 | 2568 | } |
1da177e4 LT |
2569 | } |
2570 | ||
2571 | /** | |
2572 | * schedule_tail - first thing a freshly forked thread must call. | |
2573 | * @prev: the thread we just switched away from. | |
2574 | */ | |
36c8b586 | 2575 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2576 | __releases(rq->lock) |
2577 | { | |
70b97a7f IM |
2578 | struct rq *rq = this_rq(); |
2579 | ||
4866cde0 NP |
2580 | finish_task_switch(rq, prev); |
2581 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2582 | /* In this case, finish_task_switch does not reenable preemption */ | |
2583 | preempt_enable(); | |
2584 | #endif | |
1da177e4 | 2585 | if (current->set_child_tid) |
b488893a | 2586 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2587 | } |
2588 | ||
2589 | /* | |
2590 | * context_switch - switch to the new MM and the new | |
2591 | * thread's register state. | |
2592 | */ | |
dd41f596 | 2593 | static inline void |
70b97a7f | 2594 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2595 | struct task_struct *next) |
1da177e4 | 2596 | { |
dd41f596 | 2597 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2598 | |
e107be36 | 2599 | prepare_task_switch(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 LT |
2760 | spin_lock(&rq1->lock); |
2761 | spin_lock(&rq2->lock); | |
2762 | } else { | |
2763 | spin_lock(&rq2->lock); | |
2764 | spin_lock(&rq1->lock); | |
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); | |
2807 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2808 | ret = 1; |
1da177e4 LT |
2809 | } else |
2810 | spin_lock(&busiest->lock); | |
2811 | } | |
e8fa1362 | 2812 | return ret; |
1da177e4 LT |
2813 | } |
2814 | ||
1da177e4 LT |
2815 | /* |
2816 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2817 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2818 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2819 | * the cpu_allowed mask is restored. |
2820 | */ | |
36c8b586 | 2821 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2822 | { |
70b97a7f | 2823 | struct migration_req req; |
1da177e4 | 2824 | unsigned long flags; |
70b97a7f | 2825 | struct rq *rq; |
1da177e4 LT |
2826 | |
2827 | rq = task_rq_lock(p, &flags); | |
2828 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2829 | || unlikely(cpu_is_offline(dest_cpu))) | |
2830 | goto out; | |
2831 | ||
2832 | /* force the process onto the specified CPU */ | |
2833 | if (migrate_task(p, dest_cpu, &req)) { | |
2834 | /* Need to wait for migration thread (might exit: take ref). */ | |
2835 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2836 | |
1da177e4 LT |
2837 | get_task_struct(mt); |
2838 | task_rq_unlock(rq, &flags); | |
2839 | wake_up_process(mt); | |
2840 | put_task_struct(mt); | |
2841 | wait_for_completion(&req.done); | |
36c8b586 | 2842 | |
1da177e4 LT |
2843 | return; |
2844 | } | |
2845 | out: | |
2846 | task_rq_unlock(rq, &flags); | |
2847 | } | |
2848 | ||
2849 | /* | |
476d139c NP |
2850 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2851 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2852 | */ |
2853 | void sched_exec(void) | |
2854 | { | |
1da177e4 | 2855 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2856 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2857 | put_cpu(); |
476d139c NP |
2858 | if (new_cpu != this_cpu) |
2859 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2860 | } |
2861 | ||
2862 | /* | |
2863 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2864 | * Both runqueues must be locked. | |
2865 | */ | |
dd41f596 IM |
2866 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2867 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2868 | { |
2e1cb74a | 2869 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2870 | set_task_cpu(p, this_cpu); |
dd41f596 | 2871 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2872 | /* |
2873 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2874 | * to be always true for them. | |
2875 | */ | |
dd41f596 | 2876 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2877 | } |
2878 | ||
2879 | /* | |
2880 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2881 | */ | |
858119e1 | 2882 | static |
70b97a7f | 2883 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2884 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2885 | int *all_pinned) |
1da177e4 LT |
2886 | { |
2887 | /* | |
2888 | * We do not migrate tasks that are: | |
2889 | * 1) running (obviously), or | |
2890 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2891 | * 3) are cache-hot on their current CPU. | |
2892 | */ | |
cc367732 IM |
2893 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2894 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2895 | return 0; |
cc367732 | 2896 | } |
81026794 NP |
2897 | *all_pinned = 0; |
2898 | ||
cc367732 IM |
2899 | if (task_running(rq, p)) { |
2900 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2901 | return 0; |
cc367732 | 2902 | } |
1da177e4 | 2903 | |
da84d961 IM |
2904 | /* |
2905 | * Aggressive migration if: | |
2906 | * 1) task is cache cold, or | |
2907 | * 2) too many balance attempts have failed. | |
2908 | */ | |
2909 | ||
6bc1665b IM |
2910 | if (!task_hot(p, rq->clock, sd) || |
2911 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2912 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2913 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2914 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2915 | schedstat_inc(p, se.nr_forced_migrations); |
2916 | } | |
da84d961 IM |
2917 | #endif |
2918 | return 1; | |
2919 | } | |
2920 | ||
cc367732 IM |
2921 | if (task_hot(p, rq->clock, sd)) { |
2922 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2923 | return 0; |
cc367732 | 2924 | } |
1da177e4 LT |
2925 | return 1; |
2926 | } | |
2927 | ||
e1d1484f PW |
2928 | static unsigned long |
2929 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2930 | unsigned long max_load_move, struct sched_domain *sd, | |
2931 | enum cpu_idle_type idle, int *all_pinned, | |
2932 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2933 | { |
051c6764 | 2934 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2935 | struct task_struct *p; |
2936 | long rem_load_move = max_load_move; | |
1da177e4 | 2937 | |
e1d1484f | 2938 | if (max_load_move == 0) |
1da177e4 LT |
2939 | goto out; |
2940 | ||
81026794 NP |
2941 | pinned = 1; |
2942 | ||
1da177e4 | 2943 | /* |
dd41f596 | 2944 | * Start the load-balancing iterator: |
1da177e4 | 2945 | */ |
dd41f596 IM |
2946 | p = iterator->start(iterator->arg); |
2947 | next: | |
b82d9fdd | 2948 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2949 | goto out; |
051c6764 PZ |
2950 | |
2951 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2952 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2953 | p = iterator->next(iterator->arg); |
2954 | goto next; | |
1da177e4 LT |
2955 | } |
2956 | ||
dd41f596 | 2957 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2958 | pulled++; |
dd41f596 | 2959 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2960 | |
2dd73a4f | 2961 | /* |
b82d9fdd | 2962 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2963 | */ |
e1d1484f | 2964 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2965 | if (p->prio < *this_best_prio) |
2966 | *this_best_prio = p->prio; | |
dd41f596 IM |
2967 | p = iterator->next(iterator->arg); |
2968 | goto next; | |
1da177e4 LT |
2969 | } |
2970 | out: | |
2971 | /* | |
e1d1484f | 2972 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2973 | * so we can safely collect pull_task() stats here rather than |
2974 | * inside pull_task(). | |
2975 | */ | |
2976 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2977 | |
2978 | if (all_pinned) | |
2979 | *all_pinned = pinned; | |
e1d1484f PW |
2980 | |
2981 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2982 | } |
2983 | ||
dd41f596 | 2984 | /* |
43010659 PW |
2985 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2986 | * this_rq, as part of a balancing operation within domain "sd". | |
2987 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2988 | * |
2989 | * Called with both runqueues locked. | |
2990 | */ | |
2991 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2992 | unsigned long max_load_move, |
dd41f596 IM |
2993 | struct sched_domain *sd, enum cpu_idle_type idle, |
2994 | int *all_pinned) | |
2995 | { | |
5522d5d5 | 2996 | const struct sched_class *class = sched_class_highest; |
43010659 | 2997 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2998 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2999 | |
3000 | do { | |
43010659 PW |
3001 | total_load_moved += |
3002 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3003 | max_load_move - total_load_moved, |
a4ac01c3 | 3004 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3005 | class = class->next; |
43010659 | 3006 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3007 | |
43010659 PW |
3008 | return total_load_moved > 0; |
3009 | } | |
3010 | ||
e1d1484f PW |
3011 | static int |
3012 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3013 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3014 | struct rq_iterator *iterator) | |
3015 | { | |
3016 | struct task_struct *p = iterator->start(iterator->arg); | |
3017 | int pinned = 0; | |
3018 | ||
3019 | while (p) { | |
3020 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3021 | pull_task(busiest, p, this_rq, this_cpu); | |
3022 | /* | |
3023 | * Right now, this is only the second place pull_task() | |
3024 | * is called, so we can safely collect pull_task() | |
3025 | * stats here rather than inside pull_task(). | |
3026 | */ | |
3027 | schedstat_inc(sd, lb_gained[idle]); | |
3028 | ||
3029 | return 1; | |
3030 | } | |
3031 | p = iterator->next(iterator->arg); | |
3032 | } | |
3033 | ||
3034 | return 0; | |
3035 | } | |
3036 | ||
43010659 PW |
3037 | /* |
3038 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3039 | * part of active balancing operations within "domain". | |
3040 | * Returns 1 if successful and 0 otherwise. | |
3041 | * | |
3042 | * Called with both runqueues locked. | |
3043 | */ | |
3044 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3045 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3046 | { | |
5522d5d5 | 3047 | const struct sched_class *class; |
43010659 PW |
3048 | |
3049 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3050 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3051 | return 1; |
3052 | ||
3053 | return 0; | |
dd41f596 IM |
3054 | } |
3055 | ||
1da177e4 LT |
3056 | /* |
3057 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3058 | * domain. It calculates and returns the amount of weighted load which |
3059 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3060 | */ |
3061 | static struct sched_group * | |
3062 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3063 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3064 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3065 | { |
3066 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3067 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3068 | unsigned long max_pull; |
2dd73a4f PW |
3069 | unsigned long busiest_load_per_task, busiest_nr_running; |
3070 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3071 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3072 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3073 | int power_savings_balance = 1; | |
3074 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3075 | unsigned long min_nr_running = ULONG_MAX; | |
3076 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3077 | #endif | |
1da177e4 LT |
3078 | |
3079 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3080 | busiest_load_per_task = busiest_nr_running = 0; |
3081 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3082 | |
d15bcfdb | 3083 | if (idle == CPU_NOT_IDLE) |
7897986b | 3084 | load_idx = sd->busy_idx; |
d15bcfdb | 3085 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3086 | load_idx = sd->newidle_idx; |
3087 | else | |
3088 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3089 | |
3090 | do { | |
908a7c1b | 3091 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3092 | int local_group; |
3093 | int i; | |
908a7c1b | 3094 | int __group_imb = 0; |
783609c6 | 3095 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3096 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3097 | unsigned long sum_avg_load_per_task; |
3098 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3099 | |
3100 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3101 | ||
783609c6 SS |
3102 | if (local_group) |
3103 | balance_cpu = first_cpu(group->cpumask); | |
3104 | ||
1da177e4 | 3105 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3106 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3107 | sum_avg_load_per_task = avg_load_per_task = 0; |
3108 | ||
908a7c1b KC |
3109 | max_cpu_load = 0; |
3110 | min_cpu_load = ~0UL; | |
1da177e4 LT |
3111 | |
3112 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
3113 | struct rq *rq; |
3114 | ||
3115 | if (!cpu_isset(i, *cpus)) | |
3116 | continue; | |
3117 | ||
3118 | rq = cpu_rq(i); | |
2dd73a4f | 3119 | |
9439aab8 | 3120 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3121 | *sd_idle = 0; |
3122 | ||
1da177e4 | 3123 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3124 | if (local_group) { |
3125 | if (idle_cpu(i) && !first_idle_cpu) { | |
3126 | first_idle_cpu = 1; | |
3127 | balance_cpu = i; | |
3128 | } | |
3129 | ||
a2000572 | 3130 | load = target_load(i, load_idx); |
908a7c1b | 3131 | } else { |
a2000572 | 3132 | load = source_load(i, load_idx); |
908a7c1b KC |
3133 | if (load > max_cpu_load) |
3134 | max_cpu_load = load; | |
3135 | if (min_cpu_load > load) | |
3136 | min_cpu_load = load; | |
3137 | } | |
1da177e4 LT |
3138 | |
3139 | avg_load += load; | |
2dd73a4f | 3140 | sum_nr_running += rq->nr_running; |
dd41f596 | 3141 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3142 | |
3143 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3144 | } |
3145 | ||
783609c6 SS |
3146 | /* |
3147 | * First idle cpu or the first cpu(busiest) in this sched group | |
3148 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3149 | * domains. In the newly idle case, we will allow all the cpu's |
3150 | * to do the newly idle load balance. | |
783609c6 | 3151 | */ |
9439aab8 SS |
3152 | if (idle != CPU_NEWLY_IDLE && local_group && |
3153 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3154 | *balance = 0; |
3155 | goto ret; | |
3156 | } | |
3157 | ||
1da177e4 | 3158 | total_load += avg_load; |
5517d86b | 3159 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3160 | |
3161 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3162 | avg_load = sg_div_cpu_power(group, |
3163 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3164 | |
408ed066 PZ |
3165 | |
3166 | /* | |
3167 | * Consider the group unbalanced when the imbalance is larger | |
3168 | * than the average weight of two tasks. | |
3169 | * | |
3170 | * APZ: with cgroup the avg task weight can vary wildly and | |
3171 | * might not be a suitable number - should we keep a | |
3172 | * normalized nr_running number somewhere that negates | |
3173 | * the hierarchy? | |
3174 | */ | |
3175 | avg_load_per_task = sg_div_cpu_power(group, | |
3176 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3177 | ||
3178 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3179 | __group_imb = 1; |
3180 | ||
5517d86b | 3181 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3182 | |
1da177e4 LT |
3183 | if (local_group) { |
3184 | this_load = avg_load; | |
3185 | this = group; | |
2dd73a4f PW |
3186 | this_nr_running = sum_nr_running; |
3187 | this_load_per_task = sum_weighted_load; | |
3188 | } else if (avg_load > max_load && | |
908a7c1b | 3189 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3190 | max_load = avg_load; |
3191 | busiest = group; | |
2dd73a4f PW |
3192 | busiest_nr_running = sum_nr_running; |
3193 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3194 | group_imb = __group_imb; |
1da177e4 | 3195 | } |
5c45bf27 SS |
3196 | |
3197 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3198 | /* | |
3199 | * Busy processors will not participate in power savings | |
3200 | * balance. | |
3201 | */ | |
dd41f596 IM |
3202 | if (idle == CPU_NOT_IDLE || |
3203 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3204 | goto group_next; | |
5c45bf27 SS |
3205 | |
3206 | /* | |
3207 | * If the local group is idle or completely loaded | |
3208 | * no need to do power savings balance at this domain | |
3209 | */ | |
3210 | if (local_group && (this_nr_running >= group_capacity || | |
3211 | !this_nr_running)) | |
3212 | power_savings_balance = 0; | |
3213 | ||
dd41f596 | 3214 | /* |
5c45bf27 SS |
3215 | * If a group is already running at full capacity or idle, |
3216 | * don't include that group in power savings calculations | |
dd41f596 IM |
3217 | */ |
3218 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3219 | || !sum_nr_running) |
dd41f596 | 3220 | goto group_next; |
5c45bf27 | 3221 | |
dd41f596 | 3222 | /* |
5c45bf27 | 3223 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3224 | * This is the group from where we need to pick up the load |
3225 | * for saving power | |
3226 | */ | |
3227 | if ((sum_nr_running < min_nr_running) || | |
3228 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3229 | first_cpu(group->cpumask) < |
3230 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3231 | group_min = group; |
3232 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3233 | min_load_per_task = sum_weighted_load / |
3234 | sum_nr_running; | |
dd41f596 | 3235 | } |
5c45bf27 | 3236 | |
dd41f596 | 3237 | /* |
5c45bf27 | 3238 | * Calculate the group which is almost near its |
dd41f596 IM |
3239 | * capacity but still has some space to pick up some load |
3240 | * from other group and save more power | |
3241 | */ | |
3242 | if (sum_nr_running <= group_capacity - 1) { | |
3243 | if (sum_nr_running > leader_nr_running || | |
3244 | (sum_nr_running == leader_nr_running && | |
3245 | first_cpu(group->cpumask) > | |
3246 | first_cpu(group_leader->cpumask))) { | |
3247 | group_leader = group; | |
3248 | leader_nr_running = sum_nr_running; | |
3249 | } | |
48f24c4d | 3250 | } |
5c45bf27 SS |
3251 | group_next: |
3252 | #endif | |
1da177e4 LT |
3253 | group = group->next; |
3254 | } while (group != sd->groups); | |
3255 | ||
2dd73a4f | 3256 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3257 | goto out_balanced; |
3258 | ||
3259 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3260 | ||
3261 | if (this_load >= avg_load || | |
3262 | 100*max_load <= sd->imbalance_pct*this_load) | |
3263 | goto out_balanced; | |
3264 | ||
2dd73a4f | 3265 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3266 | if (group_imb) |
3267 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3268 | ||
1da177e4 LT |
3269 | /* |
3270 | * We're trying to get all the cpus to the average_load, so we don't | |
3271 | * want to push ourselves above the average load, nor do we wish to | |
3272 | * reduce the max loaded cpu below the average load, as either of these | |
3273 | * actions would just result in more rebalancing later, and ping-pong | |
3274 | * tasks around. Thus we look for the minimum possible imbalance. | |
3275 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3276 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3277 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3278 | * appear as very large values with unsigned longs. |
3279 | */ | |
2dd73a4f PW |
3280 | if (max_load <= busiest_load_per_task) |
3281 | goto out_balanced; | |
3282 | ||
3283 | /* | |
3284 | * In the presence of smp nice balancing, certain scenarios can have | |
3285 | * max load less than avg load(as we skip the groups at or below | |
3286 | * its cpu_power, while calculating max_load..) | |
3287 | */ | |
3288 | if (max_load < avg_load) { | |
3289 | *imbalance = 0; | |
3290 | goto small_imbalance; | |
3291 | } | |
0c117f1b SS |
3292 | |
3293 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3294 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3295 | |
1da177e4 | 3296 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3297 | *imbalance = min(max_pull * busiest->__cpu_power, |
3298 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3299 | / SCHED_LOAD_SCALE; |
3300 | ||
2dd73a4f PW |
3301 | /* |
3302 | * if *imbalance is less than the average load per runnable task | |
3303 | * there is no gaurantee that any tasks will be moved so we'll have | |
3304 | * a think about bumping its value to force at least one task to be | |
3305 | * moved | |
3306 | */ | |
7fd0d2dd | 3307 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3308 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3309 | unsigned int imbn; |
3310 | ||
3311 | small_imbalance: | |
3312 | pwr_move = pwr_now = 0; | |
3313 | imbn = 2; | |
3314 | if (this_nr_running) { | |
3315 | this_load_per_task /= this_nr_running; | |
3316 | if (busiest_load_per_task > this_load_per_task) | |
3317 | imbn = 1; | |
3318 | } else | |
408ed066 | 3319 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3320 | |
408ed066 | 3321 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3322 | busiest_load_per_task * imbn) { |
2dd73a4f | 3323 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3324 | return busiest; |
3325 | } | |
3326 | ||
3327 | /* | |
3328 | * OK, we don't have enough imbalance to justify moving tasks, | |
3329 | * however we may be able to increase total CPU power used by | |
3330 | * moving them. | |
3331 | */ | |
3332 | ||
5517d86b ED |
3333 | pwr_now += busiest->__cpu_power * |
3334 | min(busiest_load_per_task, max_load); | |
3335 | pwr_now += this->__cpu_power * | |
3336 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3337 | pwr_now /= SCHED_LOAD_SCALE; |
3338 | ||
3339 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3340 | tmp = sg_div_cpu_power(busiest, |
3341 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3342 | if (max_load > tmp) |
5517d86b | 3343 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3344 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3345 | |
3346 | /* Amount of load we'd add */ | |
5517d86b | 3347 | if (max_load * busiest->__cpu_power < |
33859f7f | 3348 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3349 | tmp = sg_div_cpu_power(this, |
3350 | max_load * busiest->__cpu_power); | |
1da177e4 | 3351 | else |
5517d86b ED |
3352 | tmp = sg_div_cpu_power(this, |
3353 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3354 | pwr_move += this->__cpu_power * | |
3355 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3356 | pwr_move /= SCHED_LOAD_SCALE; |
3357 | ||
3358 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3359 | if (pwr_move > pwr_now) |
3360 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3361 | } |
3362 | ||
1da177e4 LT |
3363 | return busiest; |
3364 | ||
3365 | out_balanced: | |
5c45bf27 | 3366 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3367 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3368 | goto ret; |
1da177e4 | 3369 | |
5c45bf27 SS |
3370 | if (this == group_leader && group_leader != group_min) { |
3371 | *imbalance = min_load_per_task; | |
3372 | return group_min; | |
3373 | } | |
5c45bf27 | 3374 | #endif |
783609c6 | 3375 | ret: |
1da177e4 LT |
3376 | *imbalance = 0; |
3377 | return NULL; | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3382 | */ | |
70b97a7f | 3383 | static struct rq * |
d15bcfdb | 3384 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3385 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3386 | { |
70b97a7f | 3387 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3388 | unsigned long max_load = 0; |
1da177e4 LT |
3389 | int i; |
3390 | ||
3391 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3392 | unsigned long wl; |
0a2966b4 CL |
3393 | |
3394 | if (!cpu_isset(i, *cpus)) | |
3395 | continue; | |
3396 | ||
48f24c4d | 3397 | rq = cpu_rq(i); |
dd41f596 | 3398 | wl = weighted_cpuload(i); |
2dd73a4f | 3399 | |
dd41f596 | 3400 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3401 | continue; |
1da177e4 | 3402 | |
dd41f596 IM |
3403 | if (wl > max_load) { |
3404 | max_load = wl; | |
48f24c4d | 3405 | busiest = rq; |
1da177e4 LT |
3406 | } |
3407 | } | |
3408 | ||
3409 | return busiest; | |
3410 | } | |
3411 | ||
77391d71 NP |
3412 | /* |
3413 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3414 | * so long as it is large enough. | |
3415 | */ | |
3416 | #define MAX_PINNED_INTERVAL 512 | |
3417 | ||
1da177e4 LT |
3418 | /* |
3419 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3420 | * tasks if there is an imbalance. | |
1da177e4 | 3421 | */ |
70b97a7f | 3422 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3423 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3424 | int *balance, cpumask_t *cpus) |
1da177e4 | 3425 | { |
43010659 | 3426 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3427 | struct sched_group *group; |
1da177e4 | 3428 | unsigned long imbalance; |
70b97a7f | 3429 | struct rq *busiest; |
fe2eea3f | 3430 | unsigned long flags; |
5969fe06 | 3431 | |
7c16ec58 MT |
3432 | cpus_setall(*cpus); |
3433 | ||
89c4710e SS |
3434 | /* |
3435 | * When power savings policy is enabled for the parent domain, idle | |
3436 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3437 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3438 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3439 | */ |
d15bcfdb | 3440 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3441 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3442 | sd_idle = 1; |
1da177e4 | 3443 | |
2d72376b | 3444 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3445 | |
0a2966b4 | 3446 | redo: |
c8cba857 | 3447 | update_shares(sd); |
0a2966b4 | 3448 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3449 | cpus, balance); |
783609c6 | 3450 | |
06066714 | 3451 | if (*balance == 0) |
783609c6 | 3452 | goto out_balanced; |
783609c6 | 3453 | |
1da177e4 LT |
3454 | if (!group) { |
3455 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3456 | goto out_balanced; | |
3457 | } | |
3458 | ||
7c16ec58 | 3459 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3460 | if (!busiest) { |
3461 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3462 | goto out_balanced; | |
3463 | } | |
3464 | ||
db935dbd | 3465 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3466 | |
3467 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3468 | ||
43010659 | 3469 | ld_moved = 0; |
1da177e4 LT |
3470 | if (busiest->nr_running > 1) { |
3471 | /* | |
3472 | * Attempt to move tasks. If find_busiest_group has found | |
3473 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3474 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3475 | * correctly treated as an imbalance. |
3476 | */ | |
fe2eea3f | 3477 | local_irq_save(flags); |
e17224bf | 3478 | double_rq_lock(this_rq, busiest); |
43010659 | 3479 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3480 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3481 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3482 | local_irq_restore(flags); |
81026794 | 3483 | |
46cb4b7c SS |
3484 | /* |
3485 | * some other cpu did the load balance for us. | |
3486 | */ | |
43010659 | 3487 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3488 | resched_cpu(this_cpu); |
3489 | ||
81026794 | 3490 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3491 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3492 | cpu_clear(cpu_of(busiest), *cpus); |
3493 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3494 | goto redo; |
81026794 | 3495 | goto out_balanced; |
0a2966b4 | 3496 | } |
1da177e4 | 3497 | } |
81026794 | 3498 | |
43010659 | 3499 | if (!ld_moved) { |
1da177e4 LT |
3500 | schedstat_inc(sd, lb_failed[idle]); |
3501 | sd->nr_balance_failed++; | |
3502 | ||
3503 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3504 | |
fe2eea3f | 3505 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3506 | |
3507 | /* don't kick the migration_thread, if the curr | |
3508 | * task on busiest cpu can't be moved to this_cpu | |
3509 | */ | |
3510 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3511 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3512 | all_pinned = 1; |
3513 | goto out_one_pinned; | |
3514 | } | |
3515 | ||
1da177e4 LT |
3516 | if (!busiest->active_balance) { |
3517 | busiest->active_balance = 1; | |
3518 | busiest->push_cpu = this_cpu; | |
81026794 | 3519 | active_balance = 1; |
1da177e4 | 3520 | } |
fe2eea3f | 3521 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3522 | if (active_balance) |
1da177e4 LT |
3523 | wake_up_process(busiest->migration_thread); |
3524 | ||
3525 | /* | |
3526 | * We've kicked active balancing, reset the failure | |
3527 | * counter. | |
3528 | */ | |
39507451 | 3529 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3530 | } |
81026794 | 3531 | } else |
1da177e4 LT |
3532 | sd->nr_balance_failed = 0; |
3533 | ||
81026794 | 3534 | if (likely(!active_balance)) { |
1da177e4 LT |
3535 | /* We were unbalanced, so reset the balancing interval */ |
3536 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3537 | } else { |
3538 | /* | |
3539 | * If we've begun active balancing, start to back off. This | |
3540 | * case may not be covered by the all_pinned logic if there | |
3541 | * is only 1 task on the busy runqueue (because we don't call | |
3542 | * move_tasks). | |
3543 | */ | |
3544 | if (sd->balance_interval < sd->max_interval) | |
3545 | sd->balance_interval *= 2; | |
1da177e4 LT |
3546 | } |
3547 | ||
43010659 | 3548 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3549 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3550 | ld_moved = -1; |
3551 | ||
3552 | goto out; | |
1da177e4 LT |
3553 | |
3554 | out_balanced: | |
1da177e4 LT |
3555 | schedstat_inc(sd, lb_balanced[idle]); |
3556 | ||
16cfb1c0 | 3557 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3558 | |
3559 | out_one_pinned: | |
1da177e4 | 3560 | /* tune up the balancing interval */ |
77391d71 NP |
3561 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3562 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3563 | sd->balance_interval *= 2; |
3564 | ||
48f24c4d | 3565 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3566 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3567 | ld_moved = -1; |
3568 | else | |
3569 | ld_moved = 0; | |
3570 | out: | |
c8cba857 PZ |
3571 | if (ld_moved) |
3572 | update_shares(sd); | |
c09595f6 | 3573 | return ld_moved; |
1da177e4 LT |
3574 | } |
3575 | ||
3576 | /* | |
3577 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3578 | * tasks if there is an imbalance. | |
3579 | * | |
d15bcfdb | 3580 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3581 | * this_rq is locked. |
3582 | */ | |
48f24c4d | 3583 | static int |
7c16ec58 MT |
3584 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3585 | cpumask_t *cpus) | |
1da177e4 LT |
3586 | { |
3587 | struct sched_group *group; | |
70b97a7f | 3588 | struct rq *busiest = NULL; |
1da177e4 | 3589 | unsigned long imbalance; |
43010659 | 3590 | int ld_moved = 0; |
5969fe06 | 3591 | int sd_idle = 0; |
969bb4e4 | 3592 | int all_pinned = 0; |
7c16ec58 MT |
3593 | |
3594 | cpus_setall(*cpus); | |
5969fe06 | 3595 | |
89c4710e SS |
3596 | /* |
3597 | * When power savings policy is enabled for the parent domain, idle | |
3598 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3599 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3600 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3601 | */ |
3602 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3603 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3604 | sd_idle = 1; |
1da177e4 | 3605 | |
2d72376b | 3606 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3607 | redo: |
3e5459b4 | 3608 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3609 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3610 | &sd_idle, cpus, NULL); |
1da177e4 | 3611 | if (!group) { |
d15bcfdb | 3612 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3613 | goto out_balanced; |
1da177e4 LT |
3614 | } |
3615 | ||
7c16ec58 | 3616 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3617 | if (!busiest) { |
d15bcfdb | 3618 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3619 | goto out_balanced; |
1da177e4 LT |
3620 | } |
3621 | ||
db935dbd NP |
3622 | BUG_ON(busiest == this_rq); |
3623 | ||
d15bcfdb | 3624 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3625 | |
43010659 | 3626 | ld_moved = 0; |
d6d5cfaf NP |
3627 | if (busiest->nr_running > 1) { |
3628 | /* Attempt to move tasks */ | |
3629 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3630 | /* this_rq->clock is already updated */ |
3631 | update_rq_clock(busiest); | |
43010659 | 3632 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3633 | imbalance, sd, CPU_NEWLY_IDLE, |
3634 | &all_pinned); | |
d6d5cfaf | 3635 | spin_unlock(&busiest->lock); |
0a2966b4 | 3636 | |
969bb4e4 | 3637 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3638 | cpu_clear(cpu_of(busiest), *cpus); |
3639 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3640 | goto redo; |
3641 | } | |
d6d5cfaf NP |
3642 | } |
3643 | ||
43010659 | 3644 | if (!ld_moved) { |
d15bcfdb | 3645 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3646 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3647 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3648 | return -1; |
3649 | } else | |
16cfb1c0 | 3650 | sd->nr_balance_failed = 0; |
1da177e4 | 3651 | |
3e5459b4 | 3652 | update_shares_locked(this_rq, sd); |
43010659 | 3653 | return ld_moved; |
16cfb1c0 NP |
3654 | |
3655 | out_balanced: | |
d15bcfdb | 3656 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3657 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3658 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3659 | return -1; |
16cfb1c0 | 3660 | sd->nr_balance_failed = 0; |
48f24c4d | 3661 | |
16cfb1c0 | 3662 | return 0; |
1da177e4 LT |
3663 | } |
3664 | ||
3665 | /* | |
3666 | * idle_balance is called by schedule() if this_cpu is about to become | |
3667 | * idle. Attempts to pull tasks from other CPUs. | |
3668 | */ | |
70b97a7f | 3669 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3670 | { |
3671 | struct sched_domain *sd; | |
dd41f596 IM |
3672 | int pulled_task = -1; |
3673 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3674 | cpumask_t tmpmask; |
1da177e4 LT |
3675 | |
3676 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3677 | unsigned long interval; |
3678 | ||
3679 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3680 | continue; | |
3681 | ||
3682 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3683 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3684 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3685 | sd, &tmpmask); | |
92c4ca5c CL |
3686 | |
3687 | interval = msecs_to_jiffies(sd->balance_interval); | |
3688 | if (time_after(next_balance, sd->last_balance + interval)) | |
3689 | next_balance = sd->last_balance + interval; | |
3690 | if (pulled_task) | |
3691 | break; | |
1da177e4 | 3692 | } |
dd41f596 | 3693 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3694 | /* |
3695 | * We are going idle. next_balance may be set based on | |
3696 | * a busy processor. So reset next_balance. | |
3697 | */ | |
3698 | this_rq->next_balance = next_balance; | |
dd41f596 | 3699 | } |
1da177e4 LT |
3700 | } |
3701 | ||
3702 | /* | |
3703 | * active_load_balance is run by migration threads. It pushes running tasks | |
3704 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3705 | * running on each physical CPU where possible, and avoids physical / | |
3706 | * logical imbalances. | |
3707 | * | |
3708 | * Called with busiest_rq locked. | |
3709 | */ | |
70b97a7f | 3710 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3711 | { |
39507451 | 3712 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3713 | struct sched_domain *sd; |
3714 | struct rq *target_rq; | |
39507451 | 3715 | |
48f24c4d | 3716 | /* Is there any task to move? */ |
39507451 | 3717 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3718 | return; |
3719 | ||
3720 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3721 | |
3722 | /* | |
39507451 | 3723 | * This condition is "impossible", if it occurs |
41a2d6cf | 3724 | * we need to fix it. Originally reported by |
39507451 | 3725 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3726 | */ |
39507451 | 3727 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3728 | |
39507451 NP |
3729 | /* move a task from busiest_rq to target_rq */ |
3730 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3731 | update_rq_clock(busiest_rq); |
3732 | update_rq_clock(target_rq); | |
39507451 NP |
3733 | |
3734 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3735 | for_each_domain(target_cpu, sd) { |
39507451 | 3736 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3737 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3738 | break; |
c96d145e | 3739 | } |
39507451 | 3740 | |
48f24c4d | 3741 | if (likely(sd)) { |
2d72376b | 3742 | schedstat_inc(sd, alb_count); |
39507451 | 3743 | |
43010659 PW |
3744 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3745 | sd, CPU_IDLE)) | |
48f24c4d IM |
3746 | schedstat_inc(sd, alb_pushed); |
3747 | else | |
3748 | schedstat_inc(sd, alb_failed); | |
3749 | } | |
39507451 | 3750 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3751 | } |
3752 | ||
46cb4b7c SS |
3753 | #ifdef CONFIG_NO_HZ |
3754 | static struct { | |
3755 | atomic_t load_balancer; | |
41a2d6cf | 3756 | cpumask_t cpu_mask; |
46cb4b7c SS |
3757 | } nohz ____cacheline_aligned = { |
3758 | .load_balancer = ATOMIC_INIT(-1), | |
3759 | .cpu_mask = CPU_MASK_NONE, | |
3760 | }; | |
3761 | ||
7835b98b | 3762 | /* |
46cb4b7c SS |
3763 | * This routine will try to nominate the ilb (idle load balancing) |
3764 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3765 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3766 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3767 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3768 | * arrives... | |
3769 | * | |
3770 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3771 | * for idle load balancing. ilb owner will still be part of | |
3772 | * nohz.cpu_mask.. | |
7835b98b | 3773 | * |
46cb4b7c SS |
3774 | * While stopping the tick, this cpu will become the ilb owner if there |
3775 | * is no other owner. And will be the owner till that cpu becomes busy | |
3776 | * or if all cpus in the system stop their ticks at which point | |
3777 | * there is no need for ilb owner. | |
3778 | * | |
3779 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3780 | * next busy scheduler_tick() | |
3781 | */ | |
3782 | int select_nohz_load_balancer(int stop_tick) | |
3783 | { | |
3784 | int cpu = smp_processor_id(); | |
3785 | ||
3786 | if (stop_tick) { | |
3787 | cpu_set(cpu, nohz.cpu_mask); | |
3788 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3789 | ||
3790 | /* | |
3791 | * If we are going offline and still the leader, give up! | |
3792 | */ | |
3793 | if (cpu_is_offline(cpu) && | |
3794 | atomic_read(&nohz.load_balancer) == cpu) { | |
3795 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3796 | BUG(); | |
3797 | return 0; | |
3798 | } | |
3799 | ||
3800 | /* time for ilb owner also to sleep */ | |
3801 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3802 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3803 | atomic_set(&nohz.load_balancer, -1); | |
3804 | return 0; | |
3805 | } | |
3806 | ||
3807 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3808 | /* make me the ilb owner */ | |
3809 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3810 | return 1; | |
3811 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3812 | return 1; | |
3813 | } else { | |
3814 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3815 | return 0; | |
3816 | ||
3817 | cpu_clear(cpu, nohz.cpu_mask); | |
3818 | ||
3819 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3820 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3821 | BUG(); | |
3822 | } | |
3823 | return 0; | |
3824 | } | |
3825 | #endif | |
3826 | ||
3827 | static DEFINE_SPINLOCK(balancing); | |
3828 | ||
3829 | /* | |
7835b98b CL |
3830 | * It checks each scheduling domain to see if it is due to be balanced, |
3831 | * and initiates a balancing operation if so. | |
3832 | * | |
3833 | * Balancing parameters are set up in arch_init_sched_domains. | |
3834 | */ | |
a9957449 | 3835 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3836 | { |
46cb4b7c SS |
3837 | int balance = 1; |
3838 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3839 | unsigned long interval; |
3840 | struct sched_domain *sd; | |
46cb4b7c | 3841 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3842 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3843 | int update_next_balance = 0; |
d07355f5 | 3844 | int need_serialize; |
7c16ec58 | 3845 | cpumask_t tmp; |
1da177e4 | 3846 | |
46cb4b7c | 3847 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3848 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3849 | continue; | |
3850 | ||
3851 | interval = sd->balance_interval; | |
d15bcfdb | 3852 | if (idle != CPU_IDLE) |
1da177e4 LT |
3853 | interval *= sd->busy_factor; |
3854 | ||
3855 | /* scale ms to jiffies */ | |
3856 | interval = msecs_to_jiffies(interval); | |
3857 | if (unlikely(!interval)) | |
3858 | interval = 1; | |
dd41f596 IM |
3859 | if (interval > HZ*NR_CPUS/10) |
3860 | interval = HZ*NR_CPUS/10; | |
3861 | ||
d07355f5 | 3862 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3863 | |
d07355f5 | 3864 | if (need_serialize) { |
08c183f3 CL |
3865 | if (!spin_trylock(&balancing)) |
3866 | goto out; | |
3867 | } | |
3868 | ||
c9819f45 | 3869 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3870 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3871 | /* |
3872 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3873 | * longer idle, or one of our SMT siblings is |
3874 | * not idle. | |
3875 | */ | |
d15bcfdb | 3876 | idle = CPU_NOT_IDLE; |
1da177e4 | 3877 | } |
1bd77f2d | 3878 | sd->last_balance = jiffies; |
1da177e4 | 3879 | } |
d07355f5 | 3880 | if (need_serialize) |
08c183f3 CL |
3881 | spin_unlock(&balancing); |
3882 | out: | |
f549da84 | 3883 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3884 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3885 | update_next_balance = 1; |
3886 | } | |
783609c6 SS |
3887 | |
3888 | /* | |
3889 | * Stop the load balance at this level. There is another | |
3890 | * CPU in our sched group which is doing load balancing more | |
3891 | * actively. | |
3892 | */ | |
3893 | if (!balance) | |
3894 | break; | |
1da177e4 | 3895 | } |
f549da84 SS |
3896 | |
3897 | /* | |
3898 | * next_balance will be updated only when there is a need. | |
3899 | * When the cpu is attached to null domain for ex, it will not be | |
3900 | * updated. | |
3901 | */ | |
3902 | if (likely(update_next_balance)) | |
3903 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3904 | } |
3905 | ||
3906 | /* | |
3907 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3908 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3909 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3910 | */ | |
3911 | static void run_rebalance_domains(struct softirq_action *h) | |
3912 | { | |
dd41f596 IM |
3913 | int this_cpu = smp_processor_id(); |
3914 | struct rq *this_rq = cpu_rq(this_cpu); | |
3915 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3916 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3917 | |
dd41f596 | 3918 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3919 | |
3920 | #ifdef CONFIG_NO_HZ | |
3921 | /* | |
3922 | * If this cpu is the owner for idle load balancing, then do the | |
3923 | * balancing on behalf of the other idle cpus whose ticks are | |
3924 | * stopped. | |
3925 | */ | |
dd41f596 IM |
3926 | if (this_rq->idle_at_tick && |
3927 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3928 | cpumask_t cpus = nohz.cpu_mask; |
3929 | struct rq *rq; | |
3930 | int balance_cpu; | |
3931 | ||
dd41f596 | 3932 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3933 | for_each_cpu_mask(balance_cpu, cpus) { |
3934 | /* | |
3935 | * If this cpu gets work to do, stop the load balancing | |
3936 | * work being done for other cpus. Next load | |
3937 | * balancing owner will pick it up. | |
3938 | */ | |
3939 | if (need_resched()) | |
3940 | break; | |
3941 | ||
de0cf899 | 3942 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3943 | |
3944 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3945 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3946 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3947 | } |
3948 | } | |
3949 | #endif | |
3950 | } | |
3951 | ||
3952 | /* | |
3953 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3954 | * | |
3955 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3956 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3957 | * if the whole system is idle. | |
3958 | */ | |
dd41f596 | 3959 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3960 | { |
46cb4b7c SS |
3961 | #ifdef CONFIG_NO_HZ |
3962 | /* | |
3963 | * If we were in the nohz mode recently and busy at the current | |
3964 | * scheduler tick, then check if we need to nominate new idle | |
3965 | * load balancer. | |
3966 | */ | |
3967 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3968 | rq->in_nohz_recently = 0; | |
3969 | ||
3970 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3971 | cpu_clear(cpu, nohz.cpu_mask); | |
3972 | atomic_set(&nohz.load_balancer, -1); | |
3973 | } | |
3974 | ||
3975 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3976 | /* | |
3977 | * simple selection for now: Nominate the | |
3978 | * first cpu in the nohz list to be the next | |
3979 | * ilb owner. | |
3980 | * | |
3981 | * TBD: Traverse the sched domains and nominate | |
3982 | * the nearest cpu in the nohz.cpu_mask. | |
3983 | */ | |
3984 | int ilb = first_cpu(nohz.cpu_mask); | |
3985 | ||
434d53b0 | 3986 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3987 | resched_cpu(ilb); |
3988 | } | |
3989 | } | |
3990 | ||
3991 | /* | |
3992 | * If this cpu is idle and doing idle load balancing for all the | |
3993 | * cpus with ticks stopped, is it time for that to stop? | |
3994 | */ | |
3995 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3996 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3997 | resched_cpu(cpu); | |
3998 | return; | |
3999 | } | |
4000 | ||
4001 | /* | |
4002 | * If this cpu is idle and the idle load balancing is done by | |
4003 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4004 | */ | |
4005 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4006 | cpu_isset(cpu, nohz.cpu_mask)) | |
4007 | return; | |
4008 | #endif | |
4009 | if (time_after_eq(jiffies, rq->next_balance)) | |
4010 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4011 | } |
dd41f596 IM |
4012 | |
4013 | #else /* CONFIG_SMP */ | |
4014 | ||
1da177e4 LT |
4015 | /* |
4016 | * on UP we do not need to balance between CPUs: | |
4017 | */ | |
70b97a7f | 4018 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4019 | { |
4020 | } | |
dd41f596 | 4021 | |
1da177e4 LT |
4022 | #endif |
4023 | ||
1da177e4 LT |
4024 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4025 | ||
4026 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4027 | ||
4028 | /* | |
41b86e9c IM |
4029 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4030 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4031 | */ |
41b86e9c | 4032 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4033 | { |
1da177e4 | 4034 | unsigned long flags; |
41b86e9c IM |
4035 | u64 ns, delta_exec; |
4036 | struct rq *rq; | |
48f24c4d | 4037 | |
41b86e9c IM |
4038 | rq = task_rq_lock(p, &flags); |
4039 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4040 | if (task_current(rq, p)) { |
a8e504d2 IM |
4041 | update_rq_clock(rq); |
4042 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4043 | if ((s64)delta_exec > 0) |
4044 | ns += delta_exec; | |
4045 | } | |
4046 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4047 | |
1da177e4 LT |
4048 | return ns; |
4049 | } | |
4050 | ||
1da177e4 LT |
4051 | /* |
4052 | * Account user cpu time to a process. | |
4053 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4054 | * @cputime: the cpu time spent in user space since the last update |
4055 | */ | |
4056 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4057 | { | |
4058 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4059 | cputime64_t tmp; | |
4060 | ||
4061 | p->utime = cputime_add(p->utime, cputime); | |
4062 | ||
4063 | /* Add user time to cpustat. */ | |
4064 | tmp = cputime_to_cputime64(cputime); | |
4065 | if (TASK_NICE(p) > 0) | |
4066 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4067 | else | |
4068 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4069 | } | |
4070 | ||
94886b84 LV |
4071 | /* |
4072 | * Account guest cpu time to a process. | |
4073 | * @p: the process that the cpu time gets accounted to | |
4074 | * @cputime: the cpu time spent in virtual machine since the last update | |
4075 | */ | |
f7402e03 | 4076 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4077 | { |
4078 | cputime64_t tmp; | |
4079 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4080 | ||
4081 | tmp = cputime_to_cputime64(cputime); | |
4082 | ||
4083 | p->utime = cputime_add(p->utime, cputime); | |
4084 | p->gtime = cputime_add(p->gtime, cputime); | |
4085 | ||
4086 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4087 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4088 | } | |
4089 | ||
c66f08be MN |
4090 | /* |
4091 | * Account scaled user cpu time to a process. | |
4092 | * @p: the process that the cpu time gets accounted to | |
4093 | * @cputime: the cpu time spent in user space since the last update | |
4094 | */ | |
4095 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4096 | { | |
4097 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4098 | } | |
4099 | ||
1da177e4 LT |
4100 | /* |
4101 | * Account system cpu time to a process. | |
4102 | * @p: the process that the cpu time gets accounted to | |
4103 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4104 | * @cputime: the cpu time spent in kernel space since the last update | |
4105 | */ | |
4106 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4107 | cputime_t cputime) | |
4108 | { | |
4109 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4110 | struct rq *rq = this_rq(); |
1da177e4 LT |
4111 | cputime64_t tmp; |
4112 | ||
983ed7a6 HH |
4113 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4114 | account_guest_time(p, cputime); | |
4115 | return; | |
4116 | } | |
94886b84 | 4117 | |
1da177e4 LT |
4118 | p->stime = cputime_add(p->stime, cputime); |
4119 | ||
4120 | /* Add system time to cpustat. */ | |
4121 | tmp = cputime_to_cputime64(cputime); | |
4122 | if (hardirq_count() - hardirq_offset) | |
4123 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4124 | else if (softirq_count()) | |
4125 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4126 | else if (p != rq->idle) |
1da177e4 | 4127 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4128 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4129 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4130 | else | |
4131 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4132 | /* Account for system time used */ | |
4133 | acct_update_integrals(p); | |
1da177e4 LT |
4134 | } |
4135 | ||
c66f08be MN |
4136 | /* |
4137 | * Account scaled system cpu time to a process. | |
4138 | * @p: the process that the cpu time gets accounted to | |
4139 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4140 | * @cputime: the cpu time spent in kernel space since the last update | |
4141 | */ | |
4142 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4143 | { | |
4144 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4145 | } | |
4146 | ||
1da177e4 LT |
4147 | /* |
4148 | * Account for involuntary wait time. | |
4149 | * @p: the process from which the cpu time has been stolen | |
4150 | * @steal: the cpu time spent in involuntary wait | |
4151 | */ | |
4152 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4153 | { | |
4154 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4155 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4156 | struct rq *rq = this_rq(); |
1da177e4 LT |
4157 | |
4158 | if (p == rq->idle) { | |
4159 | p->stime = cputime_add(p->stime, steal); | |
4160 | if (atomic_read(&rq->nr_iowait) > 0) | |
4161 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4162 | else | |
4163 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4164 | } else |
1da177e4 LT |
4165 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4166 | } | |
4167 | ||
7835b98b CL |
4168 | /* |
4169 | * This function gets called by the timer code, with HZ frequency. | |
4170 | * We call it with interrupts disabled. | |
4171 | * | |
4172 | * It also gets called by the fork code, when changing the parent's | |
4173 | * timeslices. | |
4174 | */ | |
4175 | void scheduler_tick(void) | |
4176 | { | |
7835b98b CL |
4177 | int cpu = smp_processor_id(); |
4178 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4179 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4180 | |
4181 | sched_clock_tick(); | |
dd41f596 IM |
4182 | |
4183 | spin_lock(&rq->lock); | |
3e51f33f | 4184 | update_rq_clock(rq); |
f1a438d8 | 4185 | update_cpu_load(rq); |
fa85ae24 | 4186 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4187 | spin_unlock(&rq->lock); |
7835b98b | 4188 | |
e418e1c2 | 4189 | #ifdef CONFIG_SMP |
dd41f596 IM |
4190 | rq->idle_at_tick = idle_cpu(cpu); |
4191 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4192 | #endif |
1da177e4 LT |
4193 | } |
4194 | ||
1da177e4 LT |
4195 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
4196 | ||
43627582 | 4197 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
4198 | { |
4199 | /* | |
4200 | * Underflow? | |
4201 | */ | |
9a11b49a IM |
4202 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4203 | return; | |
1da177e4 LT |
4204 | preempt_count() += val; |
4205 | /* | |
4206 | * Spinlock count overflowing soon? | |
4207 | */ | |
33859f7f MOS |
4208 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4209 | PREEMPT_MASK - 10); | |
1da177e4 LT |
4210 | } |
4211 | EXPORT_SYMBOL(add_preempt_count); | |
4212 | ||
43627582 | 4213 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
4214 | { |
4215 | /* | |
4216 | * Underflow? | |
4217 | */ | |
9a11b49a IM |
4218 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4219 | return; | |
1da177e4 LT |
4220 | /* |
4221 | * Is the spinlock portion underflowing? | |
4222 | */ | |
9a11b49a IM |
4223 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4224 | !(preempt_count() & PREEMPT_MASK))) | |
4225 | return; | |
4226 | ||
1da177e4 LT |
4227 | preempt_count() -= val; |
4228 | } | |
4229 | EXPORT_SYMBOL(sub_preempt_count); | |
4230 | ||
4231 | #endif | |
4232 | ||
4233 | /* | |
dd41f596 | 4234 | * Print scheduling while atomic bug: |
1da177e4 | 4235 | */ |
dd41f596 | 4236 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4237 | { |
838225b4 SS |
4238 | struct pt_regs *regs = get_irq_regs(); |
4239 | ||
4240 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4241 | prev->comm, prev->pid, preempt_count()); | |
4242 | ||
dd41f596 | 4243 | debug_show_held_locks(prev); |
e21f5b15 | 4244 | print_modules(); |
dd41f596 IM |
4245 | if (irqs_disabled()) |
4246 | print_irqtrace_events(prev); | |
838225b4 SS |
4247 | |
4248 | if (regs) | |
4249 | show_regs(regs); | |
4250 | else | |
4251 | dump_stack(); | |
dd41f596 | 4252 | } |
1da177e4 | 4253 | |
dd41f596 IM |
4254 | /* |
4255 | * Various schedule()-time debugging checks and statistics: | |
4256 | */ | |
4257 | static inline void schedule_debug(struct task_struct *prev) | |
4258 | { | |
1da177e4 | 4259 | /* |
41a2d6cf | 4260 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4261 | * schedule() atomically, we ignore that path for now. |
4262 | * Otherwise, whine if we are scheduling when we should not be. | |
4263 | */ | |
3f33a7ce | 4264 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4265 | __schedule_bug(prev); |
4266 | ||
1da177e4 LT |
4267 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4268 | ||
2d72376b | 4269 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4270 | #ifdef CONFIG_SCHEDSTATS |
4271 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4272 | schedstat_inc(this_rq(), bkl_count); |
4273 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4274 | } |
4275 | #endif | |
dd41f596 IM |
4276 | } |
4277 | ||
4278 | /* | |
4279 | * Pick up the highest-prio task: | |
4280 | */ | |
4281 | static inline struct task_struct * | |
ff95f3df | 4282 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4283 | { |
5522d5d5 | 4284 | const struct sched_class *class; |
dd41f596 | 4285 | struct task_struct *p; |
1da177e4 LT |
4286 | |
4287 | /* | |
dd41f596 IM |
4288 | * Optimization: we know that if all tasks are in |
4289 | * the fair class we can call that function directly: | |
1da177e4 | 4290 | */ |
dd41f596 | 4291 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4292 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4293 | if (likely(p)) |
4294 | return p; | |
1da177e4 LT |
4295 | } |
4296 | ||
dd41f596 IM |
4297 | class = sched_class_highest; |
4298 | for ( ; ; ) { | |
fb8d4724 | 4299 | p = class->pick_next_task(rq); |
dd41f596 IM |
4300 | if (p) |
4301 | return p; | |
4302 | /* | |
4303 | * Will never be NULL as the idle class always | |
4304 | * returns a non-NULL p: | |
4305 | */ | |
4306 | class = class->next; | |
4307 | } | |
4308 | } | |
1da177e4 | 4309 | |
dd41f596 IM |
4310 | /* |
4311 | * schedule() is the main scheduler function. | |
4312 | */ | |
4313 | asmlinkage void __sched schedule(void) | |
4314 | { | |
4315 | struct task_struct *prev, *next; | |
67ca7bde | 4316 | unsigned long *switch_count; |
dd41f596 | 4317 | struct rq *rq; |
f333fdc9 | 4318 | int cpu, hrtick = sched_feat(HRTICK); |
dd41f596 IM |
4319 | |
4320 | need_resched: | |
4321 | preempt_disable(); | |
4322 | cpu = smp_processor_id(); | |
4323 | rq = cpu_rq(cpu); | |
4324 | rcu_qsctr_inc(cpu); | |
4325 | prev = rq->curr; | |
4326 | switch_count = &prev->nivcsw; | |
4327 | ||
4328 | release_kernel_lock(prev); | |
4329 | need_resched_nonpreemptible: | |
4330 | ||
4331 | schedule_debug(prev); | |
1da177e4 | 4332 | |
f333fdc9 MG |
4333 | if (hrtick) |
4334 | hrtick_clear(rq); | |
8f4d37ec | 4335 | |
1e819950 IM |
4336 | /* |
4337 | * Do the rq-clock update outside the rq lock: | |
4338 | */ | |
4339 | local_irq_disable(); | |
3e51f33f | 4340 | update_rq_clock(rq); |
1e819950 IM |
4341 | spin_lock(&rq->lock); |
4342 | clear_tsk_need_resched(prev); | |
1da177e4 | 4343 | |
1da177e4 | 4344 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4345 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4346 | prev->state = TASK_RUNNING; |
16882c1e | 4347 | else |
2e1cb74a | 4348 | deactivate_task(rq, prev, 1); |
dd41f596 | 4349 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4350 | } |
4351 | ||
9a897c5a SR |
4352 | #ifdef CONFIG_SMP |
4353 | if (prev->sched_class->pre_schedule) | |
4354 | prev->sched_class->pre_schedule(rq, prev); | |
4355 | #endif | |
f65eda4f | 4356 | |
dd41f596 | 4357 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4358 | idle_balance(cpu, rq); |
1da177e4 | 4359 | |
31ee529c | 4360 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4361 | next = pick_next_task(rq, prev); |
1da177e4 | 4362 | |
1da177e4 | 4363 | if (likely(prev != next)) { |
673a90a1 DS |
4364 | sched_info_switch(prev, next); |
4365 | ||
1da177e4 LT |
4366 | rq->nr_switches++; |
4367 | rq->curr = next; | |
4368 | ++*switch_count; | |
4369 | ||
dd41f596 | 4370 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4371 | /* |
4372 | * the context switch might have flipped the stack from under | |
4373 | * us, hence refresh the local variables. | |
4374 | */ | |
4375 | cpu = smp_processor_id(); | |
4376 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4377 | } else |
4378 | spin_unlock_irq(&rq->lock); | |
4379 | ||
f333fdc9 MG |
4380 | if (hrtick) |
4381 | hrtick_set(rq); | |
8f4d37ec PZ |
4382 | |
4383 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4384 | goto need_resched_nonpreemptible; |
8f4d37ec | 4385 | |
1da177e4 LT |
4386 | preempt_enable_no_resched(); |
4387 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4388 | goto need_resched; | |
4389 | } | |
1da177e4 LT |
4390 | EXPORT_SYMBOL(schedule); |
4391 | ||
4392 | #ifdef CONFIG_PREEMPT | |
4393 | /* | |
2ed6e34f | 4394 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4395 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4396 | * occur there and call schedule directly. |
4397 | */ | |
4398 | asmlinkage void __sched preempt_schedule(void) | |
4399 | { | |
4400 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4401 | |
1da177e4 LT |
4402 | /* |
4403 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4404 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4405 | */ |
beed33a8 | 4406 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4407 | return; |
4408 | ||
3a5c359a AK |
4409 | do { |
4410 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4411 | schedule(); |
3a5c359a | 4412 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4413 | |
3a5c359a AK |
4414 | /* |
4415 | * Check again in case we missed a preemption opportunity | |
4416 | * between schedule and now. | |
4417 | */ | |
4418 | barrier(); | |
4419 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4420 | } |
1da177e4 LT |
4421 | EXPORT_SYMBOL(preempt_schedule); |
4422 | ||
4423 | /* | |
2ed6e34f | 4424 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4425 | * off of irq context. |
4426 | * Note, that this is called and return with irqs disabled. This will | |
4427 | * protect us against recursive calling from irq. | |
4428 | */ | |
4429 | asmlinkage void __sched preempt_schedule_irq(void) | |
4430 | { | |
4431 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4432 | |
2ed6e34f | 4433 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4434 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4435 | ||
3a5c359a AK |
4436 | do { |
4437 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4438 | local_irq_enable(); |
4439 | schedule(); | |
4440 | local_irq_disable(); | |
3a5c359a | 4441 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4442 | |
3a5c359a AK |
4443 | /* |
4444 | * Check again in case we missed a preemption opportunity | |
4445 | * between schedule and now. | |
4446 | */ | |
4447 | barrier(); | |
4448 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4449 | } |
4450 | ||
4451 | #endif /* CONFIG_PREEMPT */ | |
4452 | ||
95cdf3b7 IM |
4453 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4454 | void *key) | |
1da177e4 | 4455 | { |
48f24c4d | 4456 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4457 | } |
1da177e4 LT |
4458 | EXPORT_SYMBOL(default_wake_function); |
4459 | ||
4460 | /* | |
41a2d6cf IM |
4461 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4462 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4463 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4464 | * | |
4465 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4466 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4467 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4468 | */ | |
4469 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4470 | int nr_exclusive, int sync, void *key) | |
4471 | { | |
2e45874c | 4472 | wait_queue_t *curr, *next; |
1da177e4 | 4473 | |
2e45874c | 4474 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4475 | unsigned flags = curr->flags; |
4476 | ||
1da177e4 | 4477 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4478 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4479 | break; |
4480 | } | |
4481 | } | |
4482 | ||
4483 | /** | |
4484 | * __wake_up - wake up threads blocked on a waitqueue. | |
4485 | * @q: the waitqueue | |
4486 | * @mode: which threads | |
4487 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4488 | * @key: is directly passed to the wakeup function |
1da177e4 | 4489 | */ |
7ad5b3a5 | 4490 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4491 | int nr_exclusive, void *key) |
1da177e4 LT |
4492 | { |
4493 | unsigned long flags; | |
4494 | ||
4495 | spin_lock_irqsave(&q->lock, flags); | |
4496 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4497 | spin_unlock_irqrestore(&q->lock, flags); | |
4498 | } | |
1da177e4 LT |
4499 | EXPORT_SYMBOL(__wake_up); |
4500 | ||
4501 | /* | |
4502 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4503 | */ | |
7ad5b3a5 | 4504 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4505 | { |
4506 | __wake_up_common(q, mode, 1, 0, NULL); | |
4507 | } | |
4508 | ||
4509 | /** | |
67be2dd1 | 4510 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4511 | * @q: the waitqueue |
4512 | * @mode: which threads | |
4513 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4514 | * | |
4515 | * The sync wakeup differs that the waker knows that it will schedule | |
4516 | * away soon, so while the target thread will be woken up, it will not | |
4517 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4518 | * with each other. This can prevent needless bouncing between CPUs. | |
4519 | * | |
4520 | * On UP it can prevent extra preemption. | |
4521 | */ | |
7ad5b3a5 | 4522 | void |
95cdf3b7 | 4523 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4524 | { |
4525 | unsigned long flags; | |
4526 | int sync = 1; | |
4527 | ||
4528 | if (unlikely(!q)) | |
4529 | return; | |
4530 | ||
4531 | if (unlikely(!nr_exclusive)) | |
4532 | sync = 0; | |
4533 | ||
4534 | spin_lock_irqsave(&q->lock, flags); | |
4535 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4536 | spin_unlock_irqrestore(&q->lock, flags); | |
4537 | } | |
4538 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4539 | ||
b15136e9 | 4540 | void complete(struct completion *x) |
1da177e4 LT |
4541 | { |
4542 | unsigned long flags; | |
4543 | ||
4544 | spin_lock_irqsave(&x->wait.lock, flags); | |
4545 | x->done++; | |
d9514f6c | 4546 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4547 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4548 | } | |
4549 | EXPORT_SYMBOL(complete); | |
4550 | ||
b15136e9 | 4551 | void complete_all(struct completion *x) |
1da177e4 LT |
4552 | { |
4553 | unsigned long flags; | |
4554 | ||
4555 | spin_lock_irqsave(&x->wait.lock, flags); | |
4556 | x->done += UINT_MAX/2; | |
d9514f6c | 4557 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4558 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4559 | } | |
4560 | EXPORT_SYMBOL(complete_all); | |
4561 | ||
8cbbe86d AK |
4562 | static inline long __sched |
4563 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4564 | { |
1da177e4 LT |
4565 | if (!x->done) { |
4566 | DECLARE_WAITQUEUE(wait, current); | |
4567 | ||
4568 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4569 | __add_wait_queue_tail(&x->wait, &wait); | |
4570 | do { | |
009e577e MW |
4571 | if ((state == TASK_INTERRUPTIBLE && |
4572 | signal_pending(current)) || | |
4573 | (state == TASK_KILLABLE && | |
4574 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4575 | timeout = -ERESTARTSYS; |
4576 | break; | |
8cbbe86d AK |
4577 | } |
4578 | __set_current_state(state); | |
1da177e4 LT |
4579 | spin_unlock_irq(&x->wait.lock); |
4580 | timeout = schedule_timeout(timeout); | |
4581 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4582 | } while (!x->done && timeout); |
1da177e4 | 4583 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4584 | if (!x->done) |
4585 | return timeout; | |
1da177e4 LT |
4586 | } |
4587 | x->done--; | |
ea71a546 | 4588 | return timeout ?: 1; |
1da177e4 | 4589 | } |
1da177e4 | 4590 | |
8cbbe86d AK |
4591 | static long __sched |
4592 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4593 | { |
1da177e4 LT |
4594 | might_sleep(); |
4595 | ||
4596 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4597 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4598 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4599 | return timeout; |
4600 | } | |
1da177e4 | 4601 | |
b15136e9 | 4602 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4603 | { |
4604 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4605 | } |
8cbbe86d | 4606 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4607 | |
b15136e9 | 4608 | unsigned long __sched |
8cbbe86d | 4609 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4610 | { |
8cbbe86d | 4611 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4612 | } |
8cbbe86d | 4613 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4614 | |
8cbbe86d | 4615 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4616 | { |
51e97990 AK |
4617 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4618 | if (t == -ERESTARTSYS) | |
4619 | return t; | |
4620 | return 0; | |
0fec171c | 4621 | } |
8cbbe86d | 4622 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4623 | |
b15136e9 | 4624 | unsigned long __sched |
8cbbe86d AK |
4625 | wait_for_completion_interruptible_timeout(struct completion *x, |
4626 | unsigned long timeout) | |
0fec171c | 4627 | { |
8cbbe86d | 4628 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4629 | } |
8cbbe86d | 4630 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4631 | |
009e577e MW |
4632 | int __sched wait_for_completion_killable(struct completion *x) |
4633 | { | |
4634 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4635 | if (t == -ERESTARTSYS) | |
4636 | return t; | |
4637 | return 0; | |
4638 | } | |
4639 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4640 | ||
8cbbe86d AK |
4641 | static long __sched |
4642 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4643 | { |
0fec171c IM |
4644 | unsigned long flags; |
4645 | wait_queue_t wait; | |
4646 | ||
4647 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4648 | |
8cbbe86d | 4649 | __set_current_state(state); |
1da177e4 | 4650 | |
8cbbe86d AK |
4651 | spin_lock_irqsave(&q->lock, flags); |
4652 | __add_wait_queue(q, &wait); | |
4653 | spin_unlock(&q->lock); | |
4654 | timeout = schedule_timeout(timeout); | |
4655 | spin_lock_irq(&q->lock); | |
4656 | __remove_wait_queue(q, &wait); | |
4657 | spin_unlock_irqrestore(&q->lock, flags); | |
4658 | ||
4659 | return timeout; | |
4660 | } | |
4661 | ||
4662 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4663 | { | |
4664 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4665 | } |
1da177e4 LT |
4666 | EXPORT_SYMBOL(interruptible_sleep_on); |
4667 | ||
0fec171c | 4668 | long __sched |
95cdf3b7 | 4669 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4670 | { |
8cbbe86d | 4671 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4672 | } |
1da177e4 LT |
4673 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4674 | ||
0fec171c | 4675 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4676 | { |
8cbbe86d | 4677 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4678 | } |
1da177e4 LT |
4679 | EXPORT_SYMBOL(sleep_on); |
4680 | ||
0fec171c | 4681 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4682 | { |
8cbbe86d | 4683 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4684 | } |
1da177e4 LT |
4685 | EXPORT_SYMBOL(sleep_on_timeout); |
4686 | ||
b29739f9 IM |
4687 | #ifdef CONFIG_RT_MUTEXES |
4688 | ||
4689 | /* | |
4690 | * rt_mutex_setprio - set the current priority of a task | |
4691 | * @p: task | |
4692 | * @prio: prio value (kernel-internal form) | |
4693 | * | |
4694 | * This function changes the 'effective' priority of a task. It does | |
4695 | * not touch ->normal_prio like __setscheduler(). | |
4696 | * | |
4697 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4698 | */ | |
36c8b586 | 4699 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4700 | { |
4701 | unsigned long flags; | |
83b699ed | 4702 | int oldprio, on_rq, running; |
70b97a7f | 4703 | struct rq *rq; |
cb469845 | 4704 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4705 | |
4706 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4707 | ||
4708 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4709 | update_rq_clock(rq); |
b29739f9 | 4710 | |
d5f9f942 | 4711 | oldprio = p->prio; |
dd41f596 | 4712 | on_rq = p->se.on_rq; |
051a1d1a | 4713 | running = task_current(rq, p); |
0e1f3483 | 4714 | if (on_rq) |
69be72c1 | 4715 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4716 | if (running) |
4717 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4718 | |
4719 | if (rt_prio(prio)) | |
4720 | p->sched_class = &rt_sched_class; | |
4721 | else | |
4722 | p->sched_class = &fair_sched_class; | |
4723 | ||
b29739f9 IM |
4724 | p->prio = prio; |
4725 | ||
0e1f3483 HS |
4726 | if (running) |
4727 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4728 | if (on_rq) { |
8159f87e | 4729 | enqueue_task(rq, p, 0); |
cb469845 SR |
4730 | |
4731 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4732 | } |
4733 | task_rq_unlock(rq, &flags); | |
4734 | } | |
4735 | ||
4736 | #endif | |
4737 | ||
36c8b586 | 4738 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4739 | { |
dd41f596 | 4740 | int old_prio, delta, on_rq; |
1da177e4 | 4741 | unsigned long flags; |
70b97a7f | 4742 | struct rq *rq; |
1da177e4 LT |
4743 | |
4744 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4745 | return; | |
4746 | /* | |
4747 | * We have to be careful, if called from sys_setpriority(), | |
4748 | * the task might be in the middle of scheduling on another CPU. | |
4749 | */ | |
4750 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4751 | update_rq_clock(rq); |
1da177e4 LT |
4752 | /* |
4753 | * The RT priorities are set via sched_setscheduler(), but we still | |
4754 | * allow the 'normal' nice value to be set - but as expected | |
4755 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4756 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4757 | */ |
e05606d3 | 4758 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4759 | p->static_prio = NICE_TO_PRIO(nice); |
4760 | goto out_unlock; | |
4761 | } | |
dd41f596 | 4762 | on_rq = p->se.on_rq; |
c09595f6 | 4763 | if (on_rq) |
69be72c1 | 4764 | dequeue_task(rq, p, 0); |
1da177e4 | 4765 | |
1da177e4 | 4766 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4767 | set_load_weight(p); |
b29739f9 IM |
4768 | old_prio = p->prio; |
4769 | p->prio = effective_prio(p); | |
4770 | delta = p->prio - old_prio; | |
1da177e4 | 4771 | |
dd41f596 | 4772 | if (on_rq) { |
8159f87e | 4773 | enqueue_task(rq, p, 0); |
1da177e4 | 4774 | /* |
d5f9f942 AM |
4775 | * If the task increased its priority or is running and |
4776 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4777 | */ |
d5f9f942 | 4778 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4779 | resched_task(rq->curr); |
4780 | } | |
4781 | out_unlock: | |
4782 | task_rq_unlock(rq, &flags); | |
4783 | } | |
1da177e4 LT |
4784 | EXPORT_SYMBOL(set_user_nice); |
4785 | ||
e43379f1 MM |
4786 | /* |
4787 | * can_nice - check if a task can reduce its nice value | |
4788 | * @p: task | |
4789 | * @nice: nice value | |
4790 | */ | |
36c8b586 | 4791 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4792 | { |
024f4747 MM |
4793 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4794 | int nice_rlim = 20 - nice; | |
48f24c4d | 4795 | |
e43379f1 MM |
4796 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4797 | capable(CAP_SYS_NICE)); | |
4798 | } | |
4799 | ||
1da177e4 LT |
4800 | #ifdef __ARCH_WANT_SYS_NICE |
4801 | ||
4802 | /* | |
4803 | * sys_nice - change the priority of the current process. | |
4804 | * @increment: priority increment | |
4805 | * | |
4806 | * sys_setpriority is a more generic, but much slower function that | |
4807 | * does similar things. | |
4808 | */ | |
4809 | asmlinkage long sys_nice(int increment) | |
4810 | { | |
48f24c4d | 4811 | long nice, retval; |
1da177e4 LT |
4812 | |
4813 | /* | |
4814 | * Setpriority might change our priority at the same moment. | |
4815 | * We don't have to worry. Conceptually one call occurs first | |
4816 | * and we have a single winner. | |
4817 | */ | |
e43379f1 MM |
4818 | if (increment < -40) |
4819 | increment = -40; | |
1da177e4 LT |
4820 | if (increment > 40) |
4821 | increment = 40; | |
4822 | ||
4823 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4824 | if (nice < -20) | |
4825 | nice = -20; | |
4826 | if (nice > 19) | |
4827 | nice = 19; | |
4828 | ||
e43379f1 MM |
4829 | if (increment < 0 && !can_nice(current, nice)) |
4830 | return -EPERM; | |
4831 | ||
1da177e4 LT |
4832 | retval = security_task_setnice(current, nice); |
4833 | if (retval) | |
4834 | return retval; | |
4835 | ||
4836 | set_user_nice(current, nice); | |
4837 | return 0; | |
4838 | } | |
4839 | ||
4840 | #endif | |
4841 | ||
4842 | /** | |
4843 | * task_prio - return the priority value of a given task. | |
4844 | * @p: the task in question. | |
4845 | * | |
4846 | * This is the priority value as seen by users in /proc. | |
4847 | * RT tasks are offset by -200. Normal tasks are centered | |
4848 | * around 0, value goes from -16 to +15. | |
4849 | */ | |
36c8b586 | 4850 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4851 | { |
4852 | return p->prio - MAX_RT_PRIO; | |
4853 | } | |
4854 | ||
4855 | /** | |
4856 | * task_nice - return the nice value of a given task. | |
4857 | * @p: the task in question. | |
4858 | */ | |
36c8b586 | 4859 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4860 | { |
4861 | return TASK_NICE(p); | |
4862 | } | |
150d8bed | 4863 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4864 | |
4865 | /** | |
4866 | * idle_cpu - is a given cpu idle currently? | |
4867 | * @cpu: the processor in question. | |
4868 | */ | |
4869 | int idle_cpu(int cpu) | |
4870 | { | |
4871 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4872 | } | |
4873 | ||
1da177e4 LT |
4874 | /** |
4875 | * idle_task - return the idle task for a given cpu. | |
4876 | * @cpu: the processor in question. | |
4877 | */ | |
36c8b586 | 4878 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4879 | { |
4880 | return cpu_rq(cpu)->idle; | |
4881 | } | |
4882 | ||
4883 | /** | |
4884 | * find_process_by_pid - find a process with a matching PID value. | |
4885 | * @pid: the pid in question. | |
4886 | */ | |
a9957449 | 4887 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4888 | { |
228ebcbe | 4889 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4890 | } |
4891 | ||
4892 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4893 | static void |
4894 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4895 | { |
dd41f596 | 4896 | BUG_ON(p->se.on_rq); |
48f24c4d | 4897 | |
1da177e4 | 4898 | p->policy = policy; |
dd41f596 IM |
4899 | switch (p->policy) { |
4900 | case SCHED_NORMAL: | |
4901 | case SCHED_BATCH: | |
4902 | case SCHED_IDLE: | |
4903 | p->sched_class = &fair_sched_class; | |
4904 | break; | |
4905 | case SCHED_FIFO: | |
4906 | case SCHED_RR: | |
4907 | p->sched_class = &rt_sched_class; | |
4908 | break; | |
4909 | } | |
4910 | ||
1da177e4 | 4911 | p->rt_priority = prio; |
b29739f9 IM |
4912 | p->normal_prio = normal_prio(p); |
4913 | /* we are holding p->pi_lock already */ | |
4914 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4915 | set_load_weight(p); |
1da177e4 LT |
4916 | } |
4917 | ||
4918 | /** | |
72fd4a35 | 4919 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4920 | * @p: the task in question. |
4921 | * @policy: new policy. | |
4922 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4923 | * |
72fd4a35 | 4924 | * NOTE that the task may be already dead. |
1da177e4 | 4925 | */ |
95cdf3b7 IM |
4926 | int sched_setscheduler(struct task_struct *p, int policy, |
4927 | struct sched_param *param) | |
1da177e4 | 4928 | { |
83b699ed | 4929 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4930 | unsigned long flags; |
cb469845 | 4931 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4932 | struct rq *rq; |
1da177e4 | 4933 | |
66e5393a SR |
4934 | /* may grab non-irq protected spin_locks */ |
4935 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4936 | recheck: |
4937 | /* double check policy once rq lock held */ | |
4938 | if (policy < 0) | |
4939 | policy = oldpolicy = p->policy; | |
4940 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4941 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4942 | policy != SCHED_IDLE) | |
b0a9499c | 4943 | return -EINVAL; |
1da177e4 LT |
4944 | /* |
4945 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4946 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4947 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4948 | */ |
4949 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4950 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4951 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4952 | return -EINVAL; |
e05606d3 | 4953 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4954 | return -EINVAL; |
4955 | ||
37e4ab3f OC |
4956 | /* |
4957 | * Allow unprivileged RT tasks to decrease priority: | |
4958 | */ | |
4959 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4960 | if (rt_policy(policy)) { |
8dc3e909 | 4961 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4962 | |
4963 | if (!lock_task_sighand(p, &flags)) | |
4964 | return -ESRCH; | |
4965 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4966 | unlock_task_sighand(p, &flags); | |
4967 | ||
4968 | /* can't set/change the rt policy */ | |
4969 | if (policy != p->policy && !rlim_rtprio) | |
4970 | return -EPERM; | |
4971 | ||
4972 | /* can't increase priority */ | |
4973 | if (param->sched_priority > p->rt_priority && | |
4974 | param->sched_priority > rlim_rtprio) | |
4975 | return -EPERM; | |
4976 | } | |
dd41f596 IM |
4977 | /* |
4978 | * Like positive nice levels, dont allow tasks to | |
4979 | * move out of SCHED_IDLE either: | |
4980 | */ | |
4981 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4982 | return -EPERM; | |
5fe1d75f | 4983 | |
37e4ab3f OC |
4984 | /* can't change other user's priorities */ |
4985 | if ((current->euid != p->euid) && | |
4986 | (current->euid != p->uid)) | |
4987 | return -EPERM; | |
4988 | } | |
1da177e4 | 4989 | |
b68aa230 PZ |
4990 | #ifdef CONFIG_RT_GROUP_SCHED |
4991 | /* | |
4992 | * Do not allow realtime tasks into groups that have no runtime | |
4993 | * assigned. | |
4994 | */ | |
d0b27fa7 | 4995 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4996 | return -EPERM; |
4997 | #endif | |
4998 | ||
1da177e4 LT |
4999 | retval = security_task_setscheduler(p, policy, param); |
5000 | if (retval) | |
5001 | return retval; | |
b29739f9 IM |
5002 | /* |
5003 | * make sure no PI-waiters arrive (or leave) while we are | |
5004 | * changing the priority of the task: | |
5005 | */ | |
5006 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5007 | /* |
5008 | * To be able to change p->policy safely, the apropriate | |
5009 | * runqueue lock must be held. | |
5010 | */ | |
b29739f9 | 5011 | rq = __task_rq_lock(p); |
1da177e4 LT |
5012 | /* recheck policy now with rq lock held */ |
5013 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5014 | policy = oldpolicy = -1; | |
b29739f9 IM |
5015 | __task_rq_unlock(rq); |
5016 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5017 | goto recheck; |
5018 | } | |
2daa3577 | 5019 | update_rq_clock(rq); |
dd41f596 | 5020 | on_rq = p->se.on_rq; |
051a1d1a | 5021 | running = task_current(rq, p); |
0e1f3483 | 5022 | if (on_rq) |
2e1cb74a | 5023 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5024 | if (running) |
5025 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5026 | |
1da177e4 | 5027 | oldprio = p->prio; |
dd41f596 | 5028 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5029 | |
0e1f3483 HS |
5030 | if (running) |
5031 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5032 | if (on_rq) { |
5033 | activate_task(rq, p, 0); | |
cb469845 SR |
5034 | |
5035 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5036 | } |
b29739f9 IM |
5037 | __task_rq_unlock(rq); |
5038 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5039 | ||
95e02ca9 TG |
5040 | rt_mutex_adjust_pi(p); |
5041 | ||
1da177e4 LT |
5042 | return 0; |
5043 | } | |
5044 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
5045 | ||
95cdf3b7 IM |
5046 | static int |
5047 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5048 | { |
1da177e4 LT |
5049 | struct sched_param lparam; |
5050 | struct task_struct *p; | |
36c8b586 | 5051 | int retval; |
1da177e4 LT |
5052 | |
5053 | if (!param || pid < 0) | |
5054 | return -EINVAL; | |
5055 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5056 | return -EFAULT; | |
5fe1d75f ON |
5057 | |
5058 | rcu_read_lock(); | |
5059 | retval = -ESRCH; | |
1da177e4 | 5060 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5061 | if (p != NULL) |
5062 | retval = sched_setscheduler(p, policy, &lparam); | |
5063 | rcu_read_unlock(); | |
36c8b586 | 5064 | |
1da177e4 LT |
5065 | return retval; |
5066 | } | |
5067 | ||
5068 | /** | |
5069 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5070 | * @pid: the pid in question. | |
5071 | * @policy: new policy. | |
5072 | * @param: structure containing the new RT priority. | |
5073 | */ | |
41a2d6cf IM |
5074 | asmlinkage long |
5075 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5076 | { |
c21761f1 JB |
5077 | /* negative values for policy are not valid */ |
5078 | if (policy < 0) | |
5079 | return -EINVAL; | |
5080 | ||
1da177e4 LT |
5081 | return do_sched_setscheduler(pid, policy, param); |
5082 | } | |
5083 | ||
5084 | /** | |
5085 | * sys_sched_setparam - set/change the RT priority of a thread | |
5086 | * @pid: the pid in question. | |
5087 | * @param: structure containing the new RT priority. | |
5088 | */ | |
5089 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5090 | { | |
5091 | return do_sched_setscheduler(pid, -1, param); | |
5092 | } | |
5093 | ||
5094 | /** | |
5095 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5096 | * @pid: the pid in question. | |
5097 | */ | |
5098 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5099 | { | |
36c8b586 | 5100 | struct task_struct *p; |
3a5c359a | 5101 | int retval; |
1da177e4 LT |
5102 | |
5103 | if (pid < 0) | |
3a5c359a | 5104 | return -EINVAL; |
1da177e4 LT |
5105 | |
5106 | retval = -ESRCH; | |
5107 | read_lock(&tasklist_lock); | |
5108 | p = find_process_by_pid(pid); | |
5109 | if (p) { | |
5110 | retval = security_task_getscheduler(p); | |
5111 | if (!retval) | |
5112 | retval = p->policy; | |
5113 | } | |
5114 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5115 | return retval; |
5116 | } | |
5117 | ||
5118 | /** | |
5119 | * sys_sched_getscheduler - get the RT priority of a thread | |
5120 | * @pid: the pid in question. | |
5121 | * @param: structure containing the RT priority. | |
5122 | */ | |
5123 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5124 | { | |
5125 | struct sched_param lp; | |
36c8b586 | 5126 | struct task_struct *p; |
3a5c359a | 5127 | int retval; |
1da177e4 LT |
5128 | |
5129 | if (!param || pid < 0) | |
3a5c359a | 5130 | return -EINVAL; |
1da177e4 LT |
5131 | |
5132 | read_lock(&tasklist_lock); | |
5133 | p = find_process_by_pid(pid); | |
5134 | retval = -ESRCH; | |
5135 | if (!p) | |
5136 | goto out_unlock; | |
5137 | ||
5138 | retval = security_task_getscheduler(p); | |
5139 | if (retval) | |
5140 | goto out_unlock; | |
5141 | ||
5142 | lp.sched_priority = p->rt_priority; | |
5143 | read_unlock(&tasklist_lock); | |
5144 | ||
5145 | /* | |
5146 | * This one might sleep, we cannot do it with a spinlock held ... | |
5147 | */ | |
5148 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5149 | ||
1da177e4 LT |
5150 | return retval; |
5151 | ||
5152 | out_unlock: | |
5153 | read_unlock(&tasklist_lock); | |
5154 | return retval; | |
5155 | } | |
5156 | ||
b53e921b | 5157 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5158 | { |
1da177e4 | 5159 | cpumask_t cpus_allowed; |
b53e921b | 5160 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5161 | struct task_struct *p; |
5162 | int retval; | |
1da177e4 | 5163 | |
95402b38 | 5164 | get_online_cpus(); |
1da177e4 LT |
5165 | read_lock(&tasklist_lock); |
5166 | ||
5167 | p = find_process_by_pid(pid); | |
5168 | if (!p) { | |
5169 | read_unlock(&tasklist_lock); | |
95402b38 | 5170 | put_online_cpus(); |
1da177e4 LT |
5171 | return -ESRCH; |
5172 | } | |
5173 | ||
5174 | /* | |
5175 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5176 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5177 | * usage count and then drop tasklist_lock. |
5178 | */ | |
5179 | get_task_struct(p); | |
5180 | read_unlock(&tasklist_lock); | |
5181 | ||
5182 | retval = -EPERM; | |
5183 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5184 | !capable(CAP_SYS_NICE)) | |
5185 | goto out_unlock; | |
5186 | ||
e7834f8f DQ |
5187 | retval = security_task_setscheduler(p, 0, NULL); |
5188 | if (retval) | |
5189 | goto out_unlock; | |
5190 | ||
f9a86fcb | 5191 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5192 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5193 | again: |
7c16ec58 | 5194 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5195 | |
8707d8b8 | 5196 | if (!retval) { |
f9a86fcb | 5197 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5198 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5199 | /* | |
5200 | * We must have raced with a concurrent cpuset | |
5201 | * update. Just reset the cpus_allowed to the | |
5202 | * cpuset's cpus_allowed | |
5203 | */ | |
5204 | new_mask = cpus_allowed; | |
5205 | goto again; | |
5206 | } | |
5207 | } | |
1da177e4 LT |
5208 | out_unlock: |
5209 | put_task_struct(p); | |
95402b38 | 5210 | put_online_cpus(); |
1da177e4 LT |
5211 | return retval; |
5212 | } | |
5213 | ||
5214 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5215 | cpumask_t *new_mask) | |
5216 | { | |
5217 | if (len < sizeof(cpumask_t)) { | |
5218 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5219 | } else if (len > sizeof(cpumask_t)) { | |
5220 | len = sizeof(cpumask_t); | |
5221 | } | |
5222 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5223 | } | |
5224 | ||
5225 | /** | |
5226 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5227 | * @pid: pid of the process | |
5228 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5229 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5230 | */ | |
5231 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5232 | unsigned long __user *user_mask_ptr) | |
5233 | { | |
5234 | cpumask_t new_mask; | |
5235 | int retval; | |
5236 | ||
5237 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5238 | if (retval) | |
5239 | return retval; | |
5240 | ||
b53e921b | 5241 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5242 | } |
5243 | ||
1da177e4 LT |
5244 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5245 | { | |
36c8b586 | 5246 | struct task_struct *p; |
1da177e4 | 5247 | int retval; |
1da177e4 | 5248 | |
95402b38 | 5249 | get_online_cpus(); |
1da177e4 LT |
5250 | read_lock(&tasklist_lock); |
5251 | ||
5252 | retval = -ESRCH; | |
5253 | p = find_process_by_pid(pid); | |
5254 | if (!p) | |
5255 | goto out_unlock; | |
5256 | ||
e7834f8f DQ |
5257 | retval = security_task_getscheduler(p); |
5258 | if (retval) | |
5259 | goto out_unlock; | |
5260 | ||
2f7016d9 | 5261 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5262 | |
5263 | out_unlock: | |
5264 | read_unlock(&tasklist_lock); | |
95402b38 | 5265 | put_online_cpus(); |
1da177e4 | 5266 | |
9531b62f | 5267 | return retval; |
1da177e4 LT |
5268 | } |
5269 | ||
5270 | /** | |
5271 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5272 | * @pid: pid of the process | |
5273 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5274 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5275 | */ | |
5276 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5277 | unsigned long __user *user_mask_ptr) | |
5278 | { | |
5279 | int ret; | |
5280 | cpumask_t mask; | |
5281 | ||
5282 | if (len < sizeof(cpumask_t)) | |
5283 | return -EINVAL; | |
5284 | ||
5285 | ret = sched_getaffinity(pid, &mask); | |
5286 | if (ret < 0) | |
5287 | return ret; | |
5288 | ||
5289 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5290 | return -EFAULT; | |
5291 | ||
5292 | return sizeof(cpumask_t); | |
5293 | } | |
5294 | ||
5295 | /** | |
5296 | * sys_sched_yield - yield the current processor to other threads. | |
5297 | * | |
dd41f596 IM |
5298 | * This function yields the current CPU to other tasks. If there are no |
5299 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5300 | */ |
5301 | asmlinkage long sys_sched_yield(void) | |
5302 | { | |
70b97a7f | 5303 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5304 | |
2d72376b | 5305 | schedstat_inc(rq, yld_count); |
4530d7ab | 5306 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5307 | |
5308 | /* | |
5309 | * Since we are going to call schedule() anyway, there's | |
5310 | * no need to preempt or enable interrupts: | |
5311 | */ | |
5312 | __release(rq->lock); | |
8a25d5de | 5313 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5314 | _raw_spin_unlock(&rq->lock); |
5315 | preempt_enable_no_resched(); | |
5316 | ||
5317 | schedule(); | |
5318 | ||
5319 | return 0; | |
5320 | } | |
5321 | ||
e7b38404 | 5322 | static void __cond_resched(void) |
1da177e4 | 5323 | { |
8e0a43d8 IM |
5324 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5325 | __might_sleep(__FILE__, __LINE__); | |
5326 | #endif | |
5bbcfd90 IM |
5327 | /* |
5328 | * The BKS might be reacquired before we have dropped | |
5329 | * PREEMPT_ACTIVE, which could trigger a second | |
5330 | * cond_resched() call. | |
5331 | */ | |
1da177e4 LT |
5332 | do { |
5333 | add_preempt_count(PREEMPT_ACTIVE); | |
5334 | schedule(); | |
5335 | sub_preempt_count(PREEMPT_ACTIVE); | |
5336 | } while (need_resched()); | |
5337 | } | |
5338 | ||
02b67cc3 | 5339 | int __sched _cond_resched(void) |
1da177e4 | 5340 | { |
9414232f IM |
5341 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5342 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5343 | __cond_resched(); |
5344 | return 1; | |
5345 | } | |
5346 | return 0; | |
5347 | } | |
02b67cc3 | 5348 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5349 | |
5350 | /* | |
5351 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5352 | * call schedule, and on return reacquire the lock. | |
5353 | * | |
41a2d6cf | 5354 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5355 | * operations here to prevent schedule() from being called twice (once via |
5356 | * spin_unlock(), once by hand). | |
5357 | */ | |
95cdf3b7 | 5358 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5359 | { |
95c354fe | 5360 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5361 | int ret = 0; |
5362 | ||
95c354fe | 5363 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5364 | spin_unlock(lock); |
95c354fe NP |
5365 | if (resched && need_resched()) |
5366 | __cond_resched(); | |
5367 | else | |
5368 | cpu_relax(); | |
6df3cecb | 5369 | ret = 1; |
1da177e4 | 5370 | spin_lock(lock); |
1da177e4 | 5371 | } |
6df3cecb | 5372 | return ret; |
1da177e4 | 5373 | } |
1da177e4 LT |
5374 | EXPORT_SYMBOL(cond_resched_lock); |
5375 | ||
5376 | int __sched cond_resched_softirq(void) | |
5377 | { | |
5378 | BUG_ON(!in_softirq()); | |
5379 | ||
9414232f | 5380 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5381 | local_bh_enable(); |
1da177e4 LT |
5382 | __cond_resched(); |
5383 | local_bh_disable(); | |
5384 | return 1; | |
5385 | } | |
5386 | return 0; | |
5387 | } | |
1da177e4 LT |
5388 | EXPORT_SYMBOL(cond_resched_softirq); |
5389 | ||
1da177e4 LT |
5390 | /** |
5391 | * yield - yield the current processor to other threads. | |
5392 | * | |
72fd4a35 | 5393 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5394 | * thread runnable and calls sys_sched_yield(). |
5395 | */ | |
5396 | void __sched yield(void) | |
5397 | { | |
5398 | set_current_state(TASK_RUNNING); | |
5399 | sys_sched_yield(); | |
5400 | } | |
1da177e4 LT |
5401 | EXPORT_SYMBOL(yield); |
5402 | ||
5403 | /* | |
41a2d6cf | 5404 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5405 | * that process accounting knows that this is a task in IO wait state. |
5406 | * | |
5407 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5408 | * has set its backing_dev_info: the queue against which it should throttle) | |
5409 | */ | |
5410 | void __sched io_schedule(void) | |
5411 | { | |
70b97a7f | 5412 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5413 | |
0ff92245 | 5414 | delayacct_blkio_start(); |
1da177e4 LT |
5415 | atomic_inc(&rq->nr_iowait); |
5416 | schedule(); | |
5417 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5418 | delayacct_blkio_end(); |
1da177e4 | 5419 | } |
1da177e4 LT |
5420 | EXPORT_SYMBOL(io_schedule); |
5421 | ||
5422 | long __sched io_schedule_timeout(long timeout) | |
5423 | { | |
70b97a7f | 5424 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5425 | long ret; |
5426 | ||
0ff92245 | 5427 | delayacct_blkio_start(); |
1da177e4 LT |
5428 | atomic_inc(&rq->nr_iowait); |
5429 | ret = schedule_timeout(timeout); | |
5430 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5431 | delayacct_blkio_end(); |
1da177e4 LT |
5432 | return ret; |
5433 | } | |
5434 | ||
5435 | /** | |
5436 | * sys_sched_get_priority_max - return maximum RT priority. | |
5437 | * @policy: scheduling class. | |
5438 | * | |
5439 | * this syscall returns the maximum rt_priority that can be used | |
5440 | * by a given scheduling class. | |
5441 | */ | |
5442 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5443 | { | |
5444 | int ret = -EINVAL; | |
5445 | ||
5446 | switch (policy) { | |
5447 | case SCHED_FIFO: | |
5448 | case SCHED_RR: | |
5449 | ret = MAX_USER_RT_PRIO-1; | |
5450 | break; | |
5451 | case SCHED_NORMAL: | |
b0a9499c | 5452 | case SCHED_BATCH: |
dd41f596 | 5453 | case SCHED_IDLE: |
1da177e4 LT |
5454 | ret = 0; |
5455 | break; | |
5456 | } | |
5457 | return ret; | |
5458 | } | |
5459 | ||
5460 | /** | |
5461 | * sys_sched_get_priority_min - return minimum RT priority. | |
5462 | * @policy: scheduling class. | |
5463 | * | |
5464 | * this syscall returns the minimum rt_priority that can be used | |
5465 | * by a given scheduling class. | |
5466 | */ | |
5467 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5468 | { | |
5469 | int ret = -EINVAL; | |
5470 | ||
5471 | switch (policy) { | |
5472 | case SCHED_FIFO: | |
5473 | case SCHED_RR: | |
5474 | ret = 1; | |
5475 | break; | |
5476 | case SCHED_NORMAL: | |
b0a9499c | 5477 | case SCHED_BATCH: |
dd41f596 | 5478 | case SCHED_IDLE: |
1da177e4 LT |
5479 | ret = 0; |
5480 | } | |
5481 | return ret; | |
5482 | } | |
5483 | ||
5484 | /** | |
5485 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5486 | * @pid: pid of the process. | |
5487 | * @interval: userspace pointer to the timeslice value. | |
5488 | * | |
5489 | * this syscall writes the default timeslice value of a given process | |
5490 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5491 | */ | |
5492 | asmlinkage | |
5493 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5494 | { | |
36c8b586 | 5495 | struct task_struct *p; |
a4ec24b4 | 5496 | unsigned int time_slice; |
3a5c359a | 5497 | int retval; |
1da177e4 | 5498 | struct timespec t; |
1da177e4 LT |
5499 | |
5500 | if (pid < 0) | |
3a5c359a | 5501 | return -EINVAL; |
1da177e4 LT |
5502 | |
5503 | retval = -ESRCH; | |
5504 | read_lock(&tasklist_lock); | |
5505 | p = find_process_by_pid(pid); | |
5506 | if (!p) | |
5507 | goto out_unlock; | |
5508 | ||
5509 | retval = security_task_getscheduler(p); | |
5510 | if (retval) | |
5511 | goto out_unlock; | |
5512 | ||
77034937 IM |
5513 | /* |
5514 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5515 | * tasks that are on an otherwise idle runqueue: | |
5516 | */ | |
5517 | time_slice = 0; | |
5518 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5519 | time_slice = DEF_TIMESLICE; |
1868f958 | 5520 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5521 | struct sched_entity *se = &p->se; |
5522 | unsigned long flags; | |
5523 | struct rq *rq; | |
5524 | ||
5525 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5526 | if (rq->cfs.load.weight) |
5527 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5528 | task_rq_unlock(rq, &flags); |
5529 | } | |
1da177e4 | 5530 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5531 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5532 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5533 | return retval; |
3a5c359a | 5534 | |
1da177e4 LT |
5535 | out_unlock: |
5536 | read_unlock(&tasklist_lock); | |
5537 | return retval; | |
5538 | } | |
5539 | ||
2ed6e34f | 5540 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5541 | |
82a1fcb9 | 5542 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5543 | { |
1da177e4 | 5544 | unsigned long free = 0; |
36c8b586 | 5545 | unsigned state; |
1da177e4 | 5546 | |
1da177e4 | 5547 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5548 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5549 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5550 | #if BITS_PER_LONG == 32 |
1da177e4 | 5551 | if (state == TASK_RUNNING) |
cc4ea795 | 5552 | printk(KERN_CONT " running "); |
1da177e4 | 5553 | else |
cc4ea795 | 5554 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5555 | #else |
5556 | if (state == TASK_RUNNING) | |
cc4ea795 | 5557 | printk(KERN_CONT " running task "); |
1da177e4 | 5558 | else |
cc4ea795 | 5559 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5560 | #endif |
5561 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5562 | { | |
10ebffde | 5563 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5564 | while (!*n) |
5565 | n++; | |
10ebffde | 5566 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5567 | } |
5568 | #endif | |
ba25f9dc | 5569 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5570 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5571 | |
5fb5e6de | 5572 | show_stack(p, NULL); |
1da177e4 LT |
5573 | } |
5574 | ||
e59e2ae2 | 5575 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5576 | { |
36c8b586 | 5577 | struct task_struct *g, *p; |
1da177e4 | 5578 | |
4bd77321 IM |
5579 | #if BITS_PER_LONG == 32 |
5580 | printk(KERN_INFO | |
5581 | " task PC stack pid father\n"); | |
1da177e4 | 5582 | #else |
4bd77321 IM |
5583 | printk(KERN_INFO |
5584 | " task PC stack pid father\n"); | |
1da177e4 LT |
5585 | #endif |
5586 | read_lock(&tasklist_lock); | |
5587 | do_each_thread(g, p) { | |
5588 | /* | |
5589 | * reset the NMI-timeout, listing all files on a slow | |
5590 | * console might take alot of time: | |
5591 | */ | |
5592 | touch_nmi_watchdog(); | |
39bc89fd | 5593 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5594 | sched_show_task(p); |
1da177e4 LT |
5595 | } while_each_thread(g, p); |
5596 | ||
04c9167f JF |
5597 | touch_all_softlockup_watchdogs(); |
5598 | ||
dd41f596 IM |
5599 | #ifdef CONFIG_SCHED_DEBUG |
5600 | sysrq_sched_debug_show(); | |
5601 | #endif | |
1da177e4 | 5602 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5603 | /* |
5604 | * Only show locks if all tasks are dumped: | |
5605 | */ | |
5606 | if (state_filter == -1) | |
5607 | debug_show_all_locks(); | |
1da177e4 LT |
5608 | } |
5609 | ||
1df21055 IM |
5610 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5611 | { | |
dd41f596 | 5612 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5613 | } |
5614 | ||
f340c0d1 IM |
5615 | /** |
5616 | * init_idle - set up an idle thread for a given CPU | |
5617 | * @idle: task in question | |
5618 | * @cpu: cpu the idle task belongs to | |
5619 | * | |
5620 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5621 | * flag, to make booting more robust. | |
5622 | */ | |
5c1e1767 | 5623 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5624 | { |
70b97a7f | 5625 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5626 | unsigned long flags; |
5627 | ||
dd41f596 IM |
5628 | __sched_fork(idle); |
5629 | idle->se.exec_start = sched_clock(); | |
5630 | ||
b29739f9 | 5631 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5632 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5633 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5634 | |
5635 | spin_lock_irqsave(&rq->lock, flags); | |
5636 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5637 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5638 | idle->oncpu = 1; | |
5639 | #endif | |
1da177e4 LT |
5640 | spin_unlock_irqrestore(&rq->lock, flags); |
5641 | ||
5642 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5643 | #if defined(CONFIG_PREEMPT) |
5644 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5645 | #else | |
a1261f54 | 5646 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5647 | #endif |
dd41f596 IM |
5648 | /* |
5649 | * The idle tasks have their own, simple scheduling class: | |
5650 | */ | |
5651 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5652 | } |
5653 | ||
5654 | /* | |
5655 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5656 | * indicates which cpus entered this state. This is used | |
5657 | * in the rcu update to wait only for active cpus. For system | |
5658 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5659 | * always be CPU_MASK_NONE. | |
5660 | */ | |
5661 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5662 | ||
19978ca6 IM |
5663 | /* |
5664 | * Increase the granularity value when there are more CPUs, | |
5665 | * because with more CPUs the 'effective latency' as visible | |
5666 | * to users decreases. But the relationship is not linear, | |
5667 | * so pick a second-best guess by going with the log2 of the | |
5668 | * number of CPUs. | |
5669 | * | |
5670 | * This idea comes from the SD scheduler of Con Kolivas: | |
5671 | */ | |
5672 | static inline void sched_init_granularity(void) | |
5673 | { | |
5674 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5675 | const unsigned long limit = 200000000; | |
5676 | ||
5677 | sysctl_sched_min_granularity *= factor; | |
5678 | if (sysctl_sched_min_granularity > limit) | |
5679 | sysctl_sched_min_granularity = limit; | |
5680 | ||
5681 | sysctl_sched_latency *= factor; | |
5682 | if (sysctl_sched_latency > limit) | |
5683 | sysctl_sched_latency = limit; | |
5684 | ||
5685 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5686 | } |
5687 | ||
1da177e4 LT |
5688 | #ifdef CONFIG_SMP |
5689 | /* | |
5690 | * This is how migration works: | |
5691 | * | |
70b97a7f | 5692 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5693 | * runqueue and wake up that CPU's migration thread. |
5694 | * 2) we down() the locked semaphore => thread blocks. | |
5695 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5696 | * thread off the CPU) | |
5697 | * 4) it gets the migration request and checks whether the migrated | |
5698 | * task is still in the wrong runqueue. | |
5699 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5700 | * it and puts it into the right queue. | |
5701 | * 6) migration thread up()s the semaphore. | |
5702 | * 7) we wake up and the migration is done. | |
5703 | */ | |
5704 | ||
5705 | /* | |
5706 | * Change a given task's CPU affinity. Migrate the thread to a | |
5707 | * proper CPU and schedule it away if the CPU it's executing on | |
5708 | * is removed from the allowed bitmask. | |
5709 | * | |
5710 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5711 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5712 | * call is not atomic; no spinlocks may be held. |
5713 | */ | |
cd8ba7cd | 5714 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5715 | { |
70b97a7f | 5716 | struct migration_req req; |
1da177e4 | 5717 | unsigned long flags; |
70b97a7f | 5718 | struct rq *rq; |
48f24c4d | 5719 | int ret = 0; |
1da177e4 LT |
5720 | |
5721 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5722 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5723 | ret = -EINVAL; |
5724 | goto out; | |
5725 | } | |
5726 | ||
9985b0ba DR |
5727 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5728 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5729 | ret = -EINVAL; | |
5730 | goto out; | |
5731 | } | |
5732 | ||
73fe6aae | 5733 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5734 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5735 | else { |
cd8ba7cd MT |
5736 | p->cpus_allowed = *new_mask; |
5737 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5738 | } |
5739 | ||
1da177e4 | 5740 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5741 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5742 | goto out; |
5743 | ||
cd8ba7cd | 5744 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5745 | /* Need help from migration thread: drop lock and wait. */ |
5746 | task_rq_unlock(rq, &flags); | |
5747 | wake_up_process(rq->migration_thread); | |
5748 | wait_for_completion(&req.done); | |
5749 | tlb_migrate_finish(p->mm); | |
5750 | return 0; | |
5751 | } | |
5752 | out: | |
5753 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5754 | |
1da177e4 LT |
5755 | return ret; |
5756 | } | |
cd8ba7cd | 5757 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5758 | |
5759 | /* | |
41a2d6cf | 5760 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5761 | * this because either it can't run here any more (set_cpus_allowed() |
5762 | * away from this CPU, or CPU going down), or because we're | |
5763 | * attempting to rebalance this task on exec (sched_exec). | |
5764 | * | |
5765 | * So we race with normal scheduler movements, but that's OK, as long | |
5766 | * as the task is no longer on this CPU. | |
efc30814 KK |
5767 | * |
5768 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5769 | */ |
efc30814 | 5770 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5771 | { |
70b97a7f | 5772 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5773 | int ret = 0, on_rq; |
1da177e4 LT |
5774 | |
5775 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5776 | return ret; |
1da177e4 LT |
5777 | |
5778 | rq_src = cpu_rq(src_cpu); | |
5779 | rq_dest = cpu_rq(dest_cpu); | |
5780 | ||
5781 | double_rq_lock(rq_src, rq_dest); | |
5782 | /* Already moved. */ | |
5783 | if (task_cpu(p) != src_cpu) | |
5784 | goto out; | |
5785 | /* Affinity changed (again). */ | |
5786 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5787 | goto out; | |
5788 | ||
dd41f596 | 5789 | on_rq = p->se.on_rq; |
6e82a3be | 5790 | if (on_rq) |
2e1cb74a | 5791 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5792 | |
1da177e4 | 5793 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5794 | if (on_rq) { |
5795 | activate_task(rq_dest, p, 0); | |
5796 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5797 | } |
efc30814 | 5798 | ret = 1; |
1da177e4 LT |
5799 | out: |
5800 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5801 | return ret; |
1da177e4 LT |
5802 | } |
5803 | ||
5804 | /* | |
5805 | * migration_thread - this is a highprio system thread that performs | |
5806 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5807 | * another runqueue. | |
5808 | */ | |
95cdf3b7 | 5809 | static int migration_thread(void *data) |
1da177e4 | 5810 | { |
1da177e4 | 5811 | int cpu = (long)data; |
70b97a7f | 5812 | struct rq *rq; |
1da177e4 LT |
5813 | |
5814 | rq = cpu_rq(cpu); | |
5815 | BUG_ON(rq->migration_thread != current); | |
5816 | ||
5817 | set_current_state(TASK_INTERRUPTIBLE); | |
5818 | while (!kthread_should_stop()) { | |
70b97a7f | 5819 | struct migration_req *req; |
1da177e4 | 5820 | struct list_head *head; |
1da177e4 | 5821 | |
1da177e4 LT |
5822 | spin_lock_irq(&rq->lock); |
5823 | ||
5824 | if (cpu_is_offline(cpu)) { | |
5825 | spin_unlock_irq(&rq->lock); | |
5826 | goto wait_to_die; | |
5827 | } | |
5828 | ||
5829 | if (rq->active_balance) { | |
5830 | active_load_balance(rq, cpu); | |
5831 | rq->active_balance = 0; | |
5832 | } | |
5833 | ||
5834 | head = &rq->migration_queue; | |
5835 | ||
5836 | if (list_empty(head)) { | |
5837 | spin_unlock_irq(&rq->lock); | |
5838 | schedule(); | |
5839 | set_current_state(TASK_INTERRUPTIBLE); | |
5840 | continue; | |
5841 | } | |
70b97a7f | 5842 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5843 | list_del_init(head->next); |
5844 | ||
674311d5 NP |
5845 | spin_unlock(&rq->lock); |
5846 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5847 | local_irq_enable(); | |
1da177e4 LT |
5848 | |
5849 | complete(&req->done); | |
5850 | } | |
5851 | __set_current_state(TASK_RUNNING); | |
5852 | return 0; | |
5853 | ||
5854 | wait_to_die: | |
5855 | /* Wait for kthread_stop */ | |
5856 | set_current_state(TASK_INTERRUPTIBLE); | |
5857 | while (!kthread_should_stop()) { | |
5858 | schedule(); | |
5859 | set_current_state(TASK_INTERRUPTIBLE); | |
5860 | } | |
5861 | __set_current_state(TASK_RUNNING); | |
5862 | return 0; | |
5863 | } | |
5864 | ||
5865 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5866 | |
5867 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5868 | { | |
5869 | int ret; | |
5870 | ||
5871 | local_irq_disable(); | |
5872 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5873 | local_irq_enable(); | |
5874 | return ret; | |
5875 | } | |
5876 | ||
054b9108 | 5877 | /* |
3a4fa0a2 | 5878 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5879 | * NOTE: interrupts should be disabled by the caller |
5880 | */ | |
48f24c4d | 5881 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5882 | { |
efc30814 | 5883 | unsigned long flags; |
1da177e4 | 5884 | cpumask_t mask; |
70b97a7f IM |
5885 | struct rq *rq; |
5886 | int dest_cpu; | |
1da177e4 | 5887 | |
3a5c359a AK |
5888 | do { |
5889 | /* On same node? */ | |
5890 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5891 | cpus_and(mask, mask, p->cpus_allowed); | |
5892 | dest_cpu = any_online_cpu(mask); | |
5893 | ||
5894 | /* On any allowed CPU? */ | |
434d53b0 | 5895 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5896 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5897 | ||
5898 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5899 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5900 | cpumask_t cpus_allowed; |
5901 | ||
5902 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5903 | /* |
5904 | * Try to stay on the same cpuset, where the | |
5905 | * current cpuset may be a subset of all cpus. | |
5906 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5907 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5908 | * called within calls to cpuset_lock/cpuset_unlock. |
5909 | */ | |
3a5c359a | 5910 | rq = task_rq_lock(p, &flags); |
470fd646 | 5911 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5912 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5913 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5914 | |
3a5c359a AK |
5915 | /* |
5916 | * Don't tell them about moving exiting tasks or | |
5917 | * kernel threads (both mm NULL), since they never | |
5918 | * leave kernel. | |
5919 | */ | |
41a2d6cf | 5920 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5921 | printk(KERN_INFO "process %d (%s) no " |
5922 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5923 | task_pid_nr(p), p->comm, dead_cpu); |
5924 | } | |
3a5c359a | 5925 | } |
f7b4cddc | 5926 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5927 | } |
5928 | ||
5929 | /* | |
5930 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5931 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5932 | * for performance reasons the counter is not stricly tracking tasks to | |
5933 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5934 | * to keep the global sum constant after CPU-down: | |
5935 | */ | |
70b97a7f | 5936 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5937 | { |
7c16ec58 | 5938 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5939 | unsigned long flags; |
5940 | ||
5941 | local_irq_save(flags); | |
5942 | double_rq_lock(rq_src, rq_dest); | |
5943 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5944 | rq_src->nr_uninterruptible = 0; | |
5945 | double_rq_unlock(rq_src, rq_dest); | |
5946 | local_irq_restore(flags); | |
5947 | } | |
5948 | ||
5949 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5950 | static void migrate_live_tasks(int src_cpu) | |
5951 | { | |
48f24c4d | 5952 | struct task_struct *p, *t; |
1da177e4 | 5953 | |
f7b4cddc | 5954 | read_lock(&tasklist_lock); |
1da177e4 | 5955 | |
48f24c4d IM |
5956 | do_each_thread(t, p) { |
5957 | if (p == current) | |
1da177e4 LT |
5958 | continue; |
5959 | ||
48f24c4d IM |
5960 | if (task_cpu(p) == src_cpu) |
5961 | move_task_off_dead_cpu(src_cpu, p); | |
5962 | } while_each_thread(t, p); | |
1da177e4 | 5963 | |
f7b4cddc | 5964 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5965 | } |
5966 | ||
dd41f596 IM |
5967 | /* |
5968 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5969 | * It does so by boosting its priority to highest possible. |
5970 | * Used by CPU offline code. | |
1da177e4 LT |
5971 | */ |
5972 | void sched_idle_next(void) | |
5973 | { | |
48f24c4d | 5974 | int this_cpu = smp_processor_id(); |
70b97a7f | 5975 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5976 | struct task_struct *p = rq->idle; |
5977 | unsigned long flags; | |
5978 | ||
5979 | /* cpu has to be offline */ | |
48f24c4d | 5980 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5981 | |
48f24c4d IM |
5982 | /* |
5983 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5984 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5985 | */ |
5986 | spin_lock_irqsave(&rq->lock, flags); | |
5987 | ||
dd41f596 | 5988 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5989 | |
94bc9a7b DA |
5990 | update_rq_clock(rq); |
5991 | activate_task(rq, p, 0); | |
1da177e4 LT |
5992 | |
5993 | spin_unlock_irqrestore(&rq->lock, flags); | |
5994 | } | |
5995 | ||
48f24c4d IM |
5996 | /* |
5997 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5998 | * offline. |
5999 | */ | |
6000 | void idle_task_exit(void) | |
6001 | { | |
6002 | struct mm_struct *mm = current->active_mm; | |
6003 | ||
6004 | BUG_ON(cpu_online(smp_processor_id())); | |
6005 | ||
6006 | if (mm != &init_mm) | |
6007 | switch_mm(mm, &init_mm, current); | |
6008 | mmdrop(mm); | |
6009 | } | |
6010 | ||
054b9108 | 6011 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6012 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6013 | { |
70b97a7f | 6014 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6015 | |
6016 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6017 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6018 | |
6019 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6020 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6021 | |
48f24c4d | 6022 | get_task_struct(p); |
1da177e4 LT |
6023 | |
6024 | /* | |
6025 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6026 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6027 | * fine. |
6028 | */ | |
f7b4cddc | 6029 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6030 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6031 | spin_lock_irq(&rq->lock); |
1da177e4 | 6032 | |
48f24c4d | 6033 | put_task_struct(p); |
1da177e4 LT |
6034 | } |
6035 | ||
6036 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6037 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6038 | { | |
70b97a7f | 6039 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6040 | struct task_struct *next; |
48f24c4d | 6041 | |
dd41f596 IM |
6042 | for ( ; ; ) { |
6043 | if (!rq->nr_running) | |
6044 | break; | |
a8e504d2 | 6045 | update_rq_clock(rq); |
ff95f3df | 6046 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6047 | if (!next) |
6048 | break; | |
6049 | migrate_dead(dead_cpu, next); | |
e692ab53 | 6050 | |
1da177e4 LT |
6051 | } |
6052 | } | |
6053 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6054 | ||
e692ab53 NP |
6055 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6056 | ||
6057 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6058 | { |
6059 | .procname = "sched_domain", | |
c57baf1e | 6060 | .mode = 0555, |
e0361851 | 6061 | }, |
38605cae | 6062 | {0, }, |
e692ab53 NP |
6063 | }; |
6064 | ||
6065 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6066 | { |
c57baf1e | 6067 | .ctl_name = CTL_KERN, |
e0361851 | 6068 | .procname = "kernel", |
c57baf1e | 6069 | .mode = 0555, |
e0361851 AD |
6070 | .child = sd_ctl_dir, |
6071 | }, | |
38605cae | 6072 | {0, }, |
e692ab53 NP |
6073 | }; |
6074 | ||
6075 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6076 | { | |
6077 | struct ctl_table *entry = | |
5cf9f062 | 6078 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6079 | |
e692ab53 NP |
6080 | return entry; |
6081 | } | |
6082 | ||
6382bc90 MM |
6083 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6084 | { | |
cd790076 | 6085 | struct ctl_table *entry; |
6382bc90 | 6086 | |
cd790076 MM |
6087 | /* |
6088 | * In the intermediate directories, both the child directory and | |
6089 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6090 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6091 | * static strings and all have proc handlers. |
6092 | */ | |
6093 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6094 | if (entry->child) |
6095 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6096 | if (entry->proc_handler == NULL) |
6097 | kfree(entry->procname); | |
6098 | } | |
6382bc90 MM |
6099 | |
6100 | kfree(*tablep); | |
6101 | *tablep = NULL; | |
6102 | } | |
6103 | ||
e692ab53 | 6104 | static void |
e0361851 | 6105 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6106 | const char *procname, void *data, int maxlen, |
6107 | mode_t mode, proc_handler *proc_handler) | |
6108 | { | |
e692ab53 NP |
6109 | entry->procname = procname; |
6110 | entry->data = data; | |
6111 | entry->maxlen = maxlen; | |
6112 | entry->mode = mode; | |
6113 | entry->proc_handler = proc_handler; | |
6114 | } | |
6115 | ||
6116 | static struct ctl_table * | |
6117 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6118 | { | |
ace8b3d6 | 6119 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6120 | |
ad1cdc1d MM |
6121 | if (table == NULL) |
6122 | return NULL; | |
6123 | ||
e0361851 | 6124 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6125 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6126 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6127 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6128 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6129 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6130 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6131 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6132 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6133 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6134 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6135 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6136 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6137 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6138 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6139 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6140 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6141 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6142 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6143 | &sd->cache_nice_tries, |
6144 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6145 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6146 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6147 | /* &table[11] is terminator */ |
e692ab53 NP |
6148 | |
6149 | return table; | |
6150 | } | |
6151 | ||
9a4e7159 | 6152 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6153 | { |
6154 | struct ctl_table *entry, *table; | |
6155 | struct sched_domain *sd; | |
6156 | int domain_num = 0, i; | |
6157 | char buf[32]; | |
6158 | ||
6159 | for_each_domain(cpu, sd) | |
6160 | domain_num++; | |
6161 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6162 | if (table == NULL) |
6163 | return NULL; | |
e692ab53 NP |
6164 | |
6165 | i = 0; | |
6166 | for_each_domain(cpu, sd) { | |
6167 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6168 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6169 | entry->mode = 0555; |
e692ab53 NP |
6170 | entry->child = sd_alloc_ctl_domain_table(sd); |
6171 | entry++; | |
6172 | i++; | |
6173 | } | |
6174 | return table; | |
6175 | } | |
6176 | ||
6177 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6178 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6179 | { |
6180 | int i, cpu_num = num_online_cpus(); | |
6181 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6182 | char buf[32]; | |
6183 | ||
7378547f MM |
6184 | WARN_ON(sd_ctl_dir[0].child); |
6185 | sd_ctl_dir[0].child = entry; | |
6186 | ||
ad1cdc1d MM |
6187 | if (entry == NULL) |
6188 | return; | |
6189 | ||
97b6ea7b | 6190 | for_each_online_cpu(i) { |
e692ab53 | 6191 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6192 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6193 | entry->mode = 0555; |
e692ab53 | 6194 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6195 | entry++; |
e692ab53 | 6196 | } |
7378547f MM |
6197 | |
6198 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6199 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6200 | } | |
6382bc90 | 6201 | |
7378547f | 6202 | /* may be called multiple times per register */ |
6382bc90 MM |
6203 | static void unregister_sched_domain_sysctl(void) |
6204 | { | |
7378547f MM |
6205 | if (sd_sysctl_header) |
6206 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6207 | sd_sysctl_header = NULL; |
7378547f MM |
6208 | if (sd_ctl_dir[0].child) |
6209 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6210 | } |
e692ab53 | 6211 | #else |
6382bc90 MM |
6212 | static void register_sched_domain_sysctl(void) |
6213 | { | |
6214 | } | |
6215 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6216 | { |
6217 | } | |
6218 | #endif | |
6219 | ||
1f11eb6a GH |
6220 | static void set_rq_online(struct rq *rq) |
6221 | { | |
6222 | if (!rq->online) { | |
6223 | const struct sched_class *class; | |
6224 | ||
6225 | cpu_set(rq->cpu, rq->rd->online); | |
6226 | rq->online = 1; | |
6227 | ||
6228 | for_each_class(class) { | |
6229 | if (class->rq_online) | |
6230 | class->rq_online(rq); | |
6231 | } | |
6232 | } | |
6233 | } | |
6234 | ||
6235 | static void set_rq_offline(struct rq *rq) | |
6236 | { | |
6237 | if (rq->online) { | |
6238 | const struct sched_class *class; | |
6239 | ||
6240 | for_each_class(class) { | |
6241 | if (class->rq_offline) | |
6242 | class->rq_offline(rq); | |
6243 | } | |
6244 | ||
6245 | cpu_clear(rq->cpu, rq->rd->online); | |
6246 | rq->online = 0; | |
6247 | } | |
6248 | } | |
6249 | ||
1da177e4 LT |
6250 | /* |
6251 | * migration_call - callback that gets triggered when a CPU is added. | |
6252 | * Here we can start up the necessary migration thread for the new CPU. | |
6253 | */ | |
48f24c4d IM |
6254 | static int __cpuinit |
6255 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6256 | { |
1da177e4 | 6257 | struct task_struct *p; |
48f24c4d | 6258 | int cpu = (long)hcpu; |
1da177e4 | 6259 | unsigned long flags; |
70b97a7f | 6260 | struct rq *rq; |
1da177e4 LT |
6261 | |
6262 | switch (action) { | |
5be9361c | 6263 | |
1da177e4 | 6264 | case CPU_UP_PREPARE: |
8bb78442 | 6265 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6266 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6267 | if (IS_ERR(p)) |
6268 | return NOTIFY_BAD; | |
1da177e4 LT |
6269 | kthread_bind(p, cpu); |
6270 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6271 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6272 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6273 | task_rq_unlock(rq, &flags); |
6274 | cpu_rq(cpu)->migration_thread = p; | |
6275 | break; | |
48f24c4d | 6276 | |
1da177e4 | 6277 | case CPU_ONLINE: |
8bb78442 | 6278 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6279 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6280 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6281 | |
6282 | /* Update our root-domain */ | |
6283 | rq = cpu_rq(cpu); | |
6284 | spin_lock_irqsave(&rq->lock, flags); | |
6285 | if (rq->rd) { | |
6286 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6287 | |
6288 | set_rq_online(rq); | |
1f94ef59 GH |
6289 | } |
6290 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6291 | break; |
48f24c4d | 6292 | |
1da177e4 LT |
6293 | #ifdef CONFIG_HOTPLUG_CPU |
6294 | case CPU_UP_CANCELED: | |
8bb78442 | 6295 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6296 | if (!cpu_rq(cpu)->migration_thread) |
6297 | break; | |
41a2d6cf | 6298 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6299 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6300 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6301 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6302 | cpu_rq(cpu)->migration_thread = NULL; | |
6303 | break; | |
48f24c4d | 6304 | |
1da177e4 | 6305 | case CPU_DEAD: |
8bb78442 | 6306 | case CPU_DEAD_FROZEN: |
470fd646 | 6307 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6308 | migrate_live_tasks(cpu); |
6309 | rq = cpu_rq(cpu); | |
6310 | kthread_stop(rq->migration_thread); | |
6311 | rq->migration_thread = NULL; | |
6312 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6313 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6314 | update_rq_clock(rq); |
2e1cb74a | 6315 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6316 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6317 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6318 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6319 | migrate_dead_tasks(cpu); |
d2da272a | 6320 | spin_unlock_irq(&rq->lock); |
470fd646 | 6321 | cpuset_unlock(); |
1da177e4 LT |
6322 | migrate_nr_uninterruptible(rq); |
6323 | BUG_ON(rq->nr_running != 0); | |
6324 | ||
41a2d6cf IM |
6325 | /* |
6326 | * No need to migrate the tasks: it was best-effort if | |
6327 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6328 | * the requestors. | |
6329 | */ | |
1da177e4 LT |
6330 | spin_lock_irq(&rq->lock); |
6331 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6332 | struct migration_req *req; |
6333 | ||
1da177e4 | 6334 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6335 | struct migration_req, list); |
1da177e4 LT |
6336 | list_del_init(&req->list); |
6337 | complete(&req->done); | |
6338 | } | |
6339 | spin_unlock_irq(&rq->lock); | |
6340 | break; | |
57d885fe | 6341 | |
08f503b0 GH |
6342 | case CPU_DYING: |
6343 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6344 | /* Update our root-domain */ |
6345 | rq = cpu_rq(cpu); | |
6346 | spin_lock_irqsave(&rq->lock, flags); | |
6347 | if (rq->rd) { | |
6348 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6349 | set_rq_offline(rq); |
57d885fe GH |
6350 | } |
6351 | spin_unlock_irqrestore(&rq->lock, flags); | |
6352 | break; | |
1da177e4 LT |
6353 | #endif |
6354 | } | |
6355 | return NOTIFY_OK; | |
6356 | } | |
6357 | ||
6358 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6359 | * happens before everything else. | |
6360 | */ | |
26c2143b | 6361 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6362 | .notifier_call = migration_call, |
6363 | .priority = 10 | |
6364 | }; | |
6365 | ||
e6fe6649 | 6366 | void __init migration_init(void) |
1da177e4 LT |
6367 | { |
6368 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6369 | int err; |
48f24c4d IM |
6370 | |
6371 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6372 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6373 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6374 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6375 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6376 | } |
6377 | #endif | |
6378 | ||
6379 | #ifdef CONFIG_SMP | |
476f3534 | 6380 | |
3e9830dc | 6381 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6382 | |
099f98c8 GS |
6383 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6384 | { | |
6385 | switch (lvl) { | |
6386 | case SD_LV_NONE: | |
6387 | return "NONE"; | |
6388 | case SD_LV_SIBLING: | |
6389 | return "SIBLING"; | |
6390 | case SD_LV_MC: | |
6391 | return "MC"; | |
6392 | case SD_LV_CPU: | |
6393 | return "CPU"; | |
6394 | case SD_LV_NODE: | |
6395 | return "NODE"; | |
6396 | case SD_LV_ALLNODES: | |
6397 | return "ALLNODES"; | |
6398 | case SD_LV_MAX: | |
6399 | return "MAX"; | |
6400 | ||
6401 | } | |
6402 | return "MAX"; | |
6403 | } | |
6404 | ||
7c16ec58 MT |
6405 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6406 | cpumask_t *groupmask) | |
1da177e4 | 6407 | { |
4dcf6aff | 6408 | struct sched_group *group = sd->groups; |
434d53b0 | 6409 | char str[256]; |
1da177e4 | 6410 | |
434d53b0 | 6411 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6412 | cpus_clear(*groupmask); |
4dcf6aff IM |
6413 | |
6414 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6415 | ||
6416 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6417 | printk("does not load-balance\n"); | |
6418 | if (sd->parent) | |
6419 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6420 | " has parent"); | |
6421 | return -1; | |
41c7ce9a NP |
6422 | } |
6423 | ||
099f98c8 GS |
6424 | printk(KERN_CONT "span %s level %s\n", |
6425 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6426 | |
6427 | if (!cpu_isset(cpu, sd->span)) { | |
6428 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6429 | "CPU%d\n", cpu); | |
6430 | } | |
6431 | if (!cpu_isset(cpu, group->cpumask)) { | |
6432 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6433 | " CPU%d\n", cpu); | |
6434 | } | |
1da177e4 | 6435 | |
4dcf6aff | 6436 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6437 | do { |
4dcf6aff IM |
6438 | if (!group) { |
6439 | printk("\n"); | |
6440 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6441 | break; |
6442 | } | |
6443 | ||
4dcf6aff IM |
6444 | if (!group->__cpu_power) { |
6445 | printk(KERN_CONT "\n"); | |
6446 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6447 | "set\n"); | |
6448 | break; | |
6449 | } | |
1da177e4 | 6450 | |
4dcf6aff IM |
6451 | if (!cpus_weight(group->cpumask)) { |
6452 | printk(KERN_CONT "\n"); | |
6453 | printk(KERN_ERR "ERROR: empty group\n"); | |
6454 | break; | |
6455 | } | |
1da177e4 | 6456 | |
7c16ec58 | 6457 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6458 | printk(KERN_CONT "\n"); |
6459 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6460 | break; | |
6461 | } | |
1da177e4 | 6462 | |
7c16ec58 | 6463 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6464 | |
434d53b0 | 6465 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6466 | printk(KERN_CONT " %s", str); |
1da177e4 | 6467 | |
4dcf6aff IM |
6468 | group = group->next; |
6469 | } while (group != sd->groups); | |
6470 | printk(KERN_CONT "\n"); | |
1da177e4 | 6471 | |
7c16ec58 | 6472 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6473 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6474 | |
7c16ec58 | 6475 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6476 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6477 | "of domain->span\n"); | |
6478 | return 0; | |
6479 | } | |
1da177e4 | 6480 | |
4dcf6aff IM |
6481 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6482 | { | |
7c16ec58 | 6483 | cpumask_t *groupmask; |
4dcf6aff | 6484 | int level = 0; |
1da177e4 | 6485 | |
4dcf6aff IM |
6486 | if (!sd) { |
6487 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6488 | return; | |
6489 | } | |
1da177e4 | 6490 | |
4dcf6aff IM |
6491 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6492 | ||
7c16ec58 MT |
6493 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6494 | if (!groupmask) { | |
6495 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6496 | return; | |
6497 | } | |
6498 | ||
4dcf6aff | 6499 | for (;;) { |
7c16ec58 | 6500 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6501 | break; |
1da177e4 LT |
6502 | level++; |
6503 | sd = sd->parent; | |
33859f7f | 6504 | if (!sd) |
4dcf6aff IM |
6505 | break; |
6506 | } | |
7c16ec58 | 6507 | kfree(groupmask); |
1da177e4 | 6508 | } |
6d6bc0ad | 6509 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6510 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6511 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6512 | |
1a20ff27 | 6513 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6514 | { |
6515 | if (cpus_weight(sd->span) == 1) | |
6516 | return 1; | |
6517 | ||
6518 | /* Following flags need at least 2 groups */ | |
6519 | if (sd->flags & (SD_LOAD_BALANCE | | |
6520 | SD_BALANCE_NEWIDLE | | |
6521 | SD_BALANCE_FORK | | |
89c4710e SS |
6522 | SD_BALANCE_EXEC | |
6523 | SD_SHARE_CPUPOWER | | |
6524 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6525 | if (sd->groups != sd->groups->next) |
6526 | return 0; | |
6527 | } | |
6528 | ||
6529 | /* Following flags don't use groups */ | |
6530 | if (sd->flags & (SD_WAKE_IDLE | | |
6531 | SD_WAKE_AFFINE | | |
6532 | SD_WAKE_BALANCE)) | |
6533 | return 0; | |
6534 | ||
6535 | return 1; | |
6536 | } | |
6537 | ||
48f24c4d IM |
6538 | static int |
6539 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6540 | { |
6541 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6542 | ||
6543 | if (sd_degenerate(parent)) | |
6544 | return 1; | |
6545 | ||
6546 | if (!cpus_equal(sd->span, parent->span)) | |
6547 | return 0; | |
6548 | ||
6549 | /* Does parent contain flags not in child? */ | |
6550 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6551 | if (cflags & SD_WAKE_AFFINE) | |
6552 | pflags &= ~SD_WAKE_BALANCE; | |
6553 | /* Flags needing groups don't count if only 1 group in parent */ | |
6554 | if (parent->groups == parent->groups->next) { | |
6555 | pflags &= ~(SD_LOAD_BALANCE | | |
6556 | SD_BALANCE_NEWIDLE | | |
6557 | SD_BALANCE_FORK | | |
89c4710e SS |
6558 | SD_BALANCE_EXEC | |
6559 | SD_SHARE_CPUPOWER | | |
6560 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6561 | } |
6562 | if (~cflags & pflags) | |
6563 | return 0; | |
6564 | ||
6565 | return 1; | |
6566 | } | |
6567 | ||
57d885fe GH |
6568 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6569 | { | |
6570 | unsigned long flags; | |
57d885fe GH |
6571 | |
6572 | spin_lock_irqsave(&rq->lock, flags); | |
6573 | ||
6574 | if (rq->rd) { | |
6575 | struct root_domain *old_rd = rq->rd; | |
6576 | ||
1f11eb6a GH |
6577 | if (cpu_isset(rq->cpu, old_rd->online)) |
6578 | set_rq_offline(rq); | |
57d885fe | 6579 | |
dc938520 | 6580 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6581 | |
57d885fe GH |
6582 | if (atomic_dec_and_test(&old_rd->refcount)) |
6583 | kfree(old_rd); | |
6584 | } | |
6585 | ||
6586 | atomic_inc(&rd->refcount); | |
6587 | rq->rd = rd; | |
6588 | ||
dc938520 | 6589 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6590 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6591 | set_rq_online(rq); |
57d885fe GH |
6592 | |
6593 | spin_unlock_irqrestore(&rq->lock, flags); | |
6594 | } | |
6595 | ||
dc938520 | 6596 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6597 | { |
6598 | memset(rd, 0, sizeof(*rd)); | |
6599 | ||
dc938520 GH |
6600 | cpus_clear(rd->span); |
6601 | cpus_clear(rd->online); | |
6e0534f2 GH |
6602 | |
6603 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6604 | } |
6605 | ||
6606 | static void init_defrootdomain(void) | |
6607 | { | |
dc938520 | 6608 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6609 | atomic_set(&def_root_domain.refcount, 1); |
6610 | } | |
6611 | ||
dc938520 | 6612 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6613 | { |
6614 | struct root_domain *rd; | |
6615 | ||
6616 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6617 | if (!rd) | |
6618 | return NULL; | |
6619 | ||
dc938520 | 6620 | init_rootdomain(rd); |
57d885fe GH |
6621 | |
6622 | return rd; | |
6623 | } | |
6624 | ||
1da177e4 | 6625 | /* |
0eab9146 | 6626 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6627 | * hold the hotplug lock. |
6628 | */ | |
0eab9146 IM |
6629 | static void |
6630 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6631 | { |
70b97a7f | 6632 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6633 | struct sched_domain *tmp; |
6634 | ||
6635 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6636 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6637 | struct sched_domain *parent = tmp->parent; | |
6638 | if (!parent) | |
6639 | break; | |
1a848870 | 6640 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6641 | tmp->parent = parent->parent; |
1a848870 SS |
6642 | if (parent->parent) |
6643 | parent->parent->child = tmp; | |
6644 | } | |
245af2c7 SS |
6645 | } |
6646 | ||
1a848870 | 6647 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6648 | sd = sd->parent; |
1a848870 SS |
6649 | if (sd) |
6650 | sd->child = NULL; | |
6651 | } | |
1da177e4 LT |
6652 | |
6653 | sched_domain_debug(sd, cpu); | |
6654 | ||
57d885fe | 6655 | rq_attach_root(rq, rd); |
674311d5 | 6656 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6657 | } |
6658 | ||
6659 | /* cpus with isolated domains */ | |
67af63a6 | 6660 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6661 | |
6662 | /* Setup the mask of cpus configured for isolated domains */ | |
6663 | static int __init isolated_cpu_setup(char *str) | |
6664 | { | |
6665 | int ints[NR_CPUS], i; | |
6666 | ||
6667 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6668 | cpus_clear(cpu_isolated_map); | |
6669 | for (i = 1; i <= ints[0]; i++) | |
6670 | if (ints[i] < NR_CPUS) | |
6671 | cpu_set(ints[i], cpu_isolated_map); | |
6672 | return 1; | |
6673 | } | |
6674 | ||
8927f494 | 6675 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6676 | |
6677 | /* | |
6711cab4 SS |
6678 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6679 | * to a function which identifies what group(along with sched group) a CPU | |
6680 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6681 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6682 | * |
6683 | * init_sched_build_groups will build a circular linked list of the groups | |
6684 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6685 | * and ->cpu_power to 0. | |
6686 | */ | |
a616058b | 6687 | static void |
7c16ec58 | 6688 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6689 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6690 | struct sched_group **sg, |
6691 | cpumask_t *tmpmask), | |
6692 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6693 | { |
6694 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6695 | int i; |
6696 | ||
7c16ec58 MT |
6697 | cpus_clear(*covered); |
6698 | ||
6699 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6700 | struct sched_group *sg; |
7c16ec58 | 6701 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6702 | int j; |
6703 | ||
7c16ec58 | 6704 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6705 | continue; |
6706 | ||
7c16ec58 | 6707 | cpus_clear(sg->cpumask); |
5517d86b | 6708 | sg->__cpu_power = 0; |
1da177e4 | 6709 | |
7c16ec58 MT |
6710 | for_each_cpu_mask(j, *span) { |
6711 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6712 | continue; |
6713 | ||
7c16ec58 | 6714 | cpu_set(j, *covered); |
1da177e4 LT |
6715 | cpu_set(j, sg->cpumask); |
6716 | } | |
6717 | if (!first) | |
6718 | first = sg; | |
6719 | if (last) | |
6720 | last->next = sg; | |
6721 | last = sg; | |
6722 | } | |
6723 | last->next = first; | |
6724 | } | |
6725 | ||
9c1cfda2 | 6726 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6727 | |
9c1cfda2 | 6728 | #ifdef CONFIG_NUMA |
198e2f18 | 6729 | |
9c1cfda2 JH |
6730 | /** |
6731 | * find_next_best_node - find the next node to include in a sched_domain | |
6732 | * @node: node whose sched_domain we're building | |
6733 | * @used_nodes: nodes already in the sched_domain | |
6734 | * | |
41a2d6cf | 6735 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6736 | * finds the closest node not already in the @used_nodes map. |
6737 | * | |
6738 | * Should use nodemask_t. | |
6739 | */ | |
c5f59f08 | 6740 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6741 | { |
6742 | int i, n, val, min_val, best_node = 0; | |
6743 | ||
6744 | min_val = INT_MAX; | |
6745 | ||
6746 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6747 | /* Start at @node */ | |
6748 | n = (node + i) % MAX_NUMNODES; | |
6749 | ||
6750 | if (!nr_cpus_node(n)) | |
6751 | continue; | |
6752 | ||
6753 | /* Skip already used nodes */ | |
c5f59f08 | 6754 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6755 | continue; |
6756 | ||
6757 | /* Simple min distance search */ | |
6758 | val = node_distance(node, n); | |
6759 | ||
6760 | if (val < min_val) { | |
6761 | min_val = val; | |
6762 | best_node = n; | |
6763 | } | |
6764 | } | |
6765 | ||
c5f59f08 | 6766 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6767 | return best_node; |
6768 | } | |
6769 | ||
6770 | /** | |
6771 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6772 | * @node: node whose cpumask we're constructing | |
73486722 | 6773 | * @span: resulting cpumask |
9c1cfda2 | 6774 | * |
41a2d6cf | 6775 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6776 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6777 | * out optimally. | |
6778 | */ | |
4bdbaad3 | 6779 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6780 | { |
c5f59f08 | 6781 | nodemask_t used_nodes; |
c5f59f08 | 6782 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6783 | int i; |
9c1cfda2 | 6784 | |
4bdbaad3 | 6785 | cpus_clear(*span); |
c5f59f08 | 6786 | nodes_clear(used_nodes); |
9c1cfda2 | 6787 | |
4bdbaad3 | 6788 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6789 | node_set(node, used_nodes); |
9c1cfda2 JH |
6790 | |
6791 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6792 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6793 | |
c5f59f08 | 6794 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6795 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6796 | } |
9c1cfda2 | 6797 | } |
6d6bc0ad | 6798 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6799 | |
5c45bf27 | 6800 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6801 | |
9c1cfda2 | 6802 | /* |
48f24c4d | 6803 | * SMT sched-domains: |
9c1cfda2 | 6804 | */ |
1da177e4 LT |
6805 | #ifdef CONFIG_SCHED_SMT |
6806 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6807 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6808 | |
41a2d6cf | 6809 | static int |
7c16ec58 MT |
6810 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6811 | cpumask_t *unused) | |
1da177e4 | 6812 | { |
6711cab4 SS |
6813 | if (sg) |
6814 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6815 | return cpu; |
6816 | } | |
6d6bc0ad | 6817 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6818 | |
48f24c4d IM |
6819 | /* |
6820 | * multi-core sched-domains: | |
6821 | */ | |
1e9f28fa SS |
6822 | #ifdef CONFIG_SCHED_MC |
6823 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6824 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6825 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6826 | |
6827 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6828 | static int |
7c16ec58 MT |
6829 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6830 | cpumask_t *mask) | |
1e9f28fa | 6831 | { |
6711cab4 | 6832 | int group; |
7c16ec58 MT |
6833 | |
6834 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6835 | cpus_and(*mask, *mask, *cpu_map); | |
6836 | group = first_cpu(*mask); | |
6711cab4 SS |
6837 | if (sg) |
6838 | *sg = &per_cpu(sched_group_core, group); | |
6839 | return group; | |
1e9f28fa SS |
6840 | } |
6841 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6842 | static int |
7c16ec58 MT |
6843 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6844 | cpumask_t *unused) | |
1e9f28fa | 6845 | { |
6711cab4 SS |
6846 | if (sg) |
6847 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6848 | return cpu; |
6849 | } | |
6850 | #endif | |
6851 | ||
1da177e4 | 6852 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6853 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6854 | |
41a2d6cf | 6855 | static int |
7c16ec58 MT |
6856 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6857 | cpumask_t *mask) | |
1da177e4 | 6858 | { |
6711cab4 | 6859 | int group; |
48f24c4d | 6860 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6861 | *mask = cpu_coregroup_map(cpu); |
6862 | cpus_and(*mask, *mask, *cpu_map); | |
6863 | group = first_cpu(*mask); | |
1e9f28fa | 6864 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6865 | *mask = per_cpu(cpu_sibling_map, cpu); |
6866 | cpus_and(*mask, *mask, *cpu_map); | |
6867 | group = first_cpu(*mask); | |
1da177e4 | 6868 | #else |
6711cab4 | 6869 | group = cpu; |
1da177e4 | 6870 | #endif |
6711cab4 SS |
6871 | if (sg) |
6872 | *sg = &per_cpu(sched_group_phys, group); | |
6873 | return group; | |
1da177e4 LT |
6874 | } |
6875 | ||
6876 | #ifdef CONFIG_NUMA | |
1da177e4 | 6877 | /* |
9c1cfda2 JH |
6878 | * The init_sched_build_groups can't handle what we want to do with node |
6879 | * groups, so roll our own. Now each node has its own list of groups which | |
6880 | * gets dynamically allocated. | |
1da177e4 | 6881 | */ |
9c1cfda2 | 6882 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6883 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6884 | |
9c1cfda2 | 6885 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6886 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6887 | |
6711cab4 | 6888 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6889 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6890 | { |
6711cab4 SS |
6891 | int group; |
6892 | ||
7c16ec58 MT |
6893 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6894 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6895 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6896 | |
6897 | if (sg) | |
6898 | *sg = &per_cpu(sched_group_allnodes, group); | |
6899 | return group; | |
1da177e4 | 6900 | } |
6711cab4 | 6901 | |
08069033 SS |
6902 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6903 | { | |
6904 | struct sched_group *sg = group_head; | |
6905 | int j; | |
6906 | ||
6907 | if (!sg) | |
6908 | return; | |
3a5c359a AK |
6909 | do { |
6910 | for_each_cpu_mask(j, sg->cpumask) { | |
6911 | struct sched_domain *sd; | |
08069033 | 6912 | |
3a5c359a AK |
6913 | sd = &per_cpu(phys_domains, j); |
6914 | if (j != first_cpu(sd->groups->cpumask)) { | |
6915 | /* | |
6916 | * Only add "power" once for each | |
6917 | * physical package. | |
6918 | */ | |
6919 | continue; | |
6920 | } | |
08069033 | 6921 | |
3a5c359a AK |
6922 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6923 | } | |
6924 | sg = sg->next; | |
6925 | } while (sg != group_head); | |
08069033 | 6926 | } |
6d6bc0ad | 6927 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6928 | |
a616058b | 6929 | #ifdef CONFIG_NUMA |
51888ca2 | 6930 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6931 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6932 | { |
a616058b | 6933 | int cpu, i; |
51888ca2 SV |
6934 | |
6935 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6936 | struct sched_group **sched_group_nodes |
6937 | = sched_group_nodes_bycpu[cpu]; | |
6938 | ||
51888ca2 SV |
6939 | if (!sched_group_nodes) |
6940 | continue; | |
6941 | ||
6942 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6943 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6944 | ||
7c16ec58 MT |
6945 | *nodemask = node_to_cpumask(i); |
6946 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6947 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6948 | continue; |
6949 | ||
6950 | if (sg == NULL) | |
6951 | continue; | |
6952 | sg = sg->next; | |
6953 | next_sg: | |
6954 | oldsg = sg; | |
6955 | sg = sg->next; | |
6956 | kfree(oldsg); | |
6957 | if (oldsg != sched_group_nodes[i]) | |
6958 | goto next_sg; | |
6959 | } | |
6960 | kfree(sched_group_nodes); | |
6961 | sched_group_nodes_bycpu[cpu] = NULL; | |
6962 | } | |
51888ca2 | 6963 | } |
6d6bc0ad | 6964 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 6965 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6966 | { |
6967 | } | |
6d6bc0ad | 6968 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6969 | |
89c4710e SS |
6970 | /* |
6971 | * Initialize sched groups cpu_power. | |
6972 | * | |
6973 | * cpu_power indicates the capacity of sched group, which is used while | |
6974 | * distributing the load between different sched groups in a sched domain. | |
6975 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6976 | * there are asymmetries in the topology. If there are asymmetries, group | |
6977 | * having more cpu_power will pickup more load compared to the group having | |
6978 | * less cpu_power. | |
6979 | * | |
6980 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6981 | * the maximum number of tasks a group can handle in the presence of other idle | |
6982 | * or lightly loaded groups in the same sched domain. | |
6983 | */ | |
6984 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6985 | { | |
6986 | struct sched_domain *child; | |
6987 | struct sched_group *group; | |
6988 | ||
6989 | WARN_ON(!sd || !sd->groups); | |
6990 | ||
6991 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6992 | return; | |
6993 | ||
6994 | child = sd->child; | |
6995 | ||
5517d86b ED |
6996 | sd->groups->__cpu_power = 0; |
6997 | ||
89c4710e SS |
6998 | /* |
6999 | * For perf policy, if the groups in child domain share resources | |
7000 | * (for example cores sharing some portions of the cache hierarchy | |
7001 | * or SMT), then set this domain groups cpu_power such that each group | |
7002 | * can handle only one task, when there are other idle groups in the | |
7003 | * same sched domain. | |
7004 | */ | |
7005 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7006 | (child->flags & | |
7007 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7008 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7009 | return; |
7010 | } | |
7011 | ||
89c4710e SS |
7012 | /* |
7013 | * add cpu_power of each child group to this groups cpu_power | |
7014 | */ | |
7015 | group = child->groups; | |
7016 | do { | |
5517d86b | 7017 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7018 | group = group->next; |
7019 | } while (group != child->groups); | |
7020 | } | |
7021 | ||
7c16ec58 MT |
7022 | /* |
7023 | * Initializers for schedule domains | |
7024 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7025 | */ | |
7026 | ||
7027 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7028 | #define SD_INIT_FUNC(type) \ | |
7029 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7030 | { \ | |
7031 | memset(sd, 0, sizeof(*sd)); \ | |
7032 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7033 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7034 | } |
7035 | ||
7036 | SD_INIT_FUNC(CPU) | |
7037 | #ifdef CONFIG_NUMA | |
7038 | SD_INIT_FUNC(ALLNODES) | |
7039 | SD_INIT_FUNC(NODE) | |
7040 | #endif | |
7041 | #ifdef CONFIG_SCHED_SMT | |
7042 | SD_INIT_FUNC(SIBLING) | |
7043 | #endif | |
7044 | #ifdef CONFIG_SCHED_MC | |
7045 | SD_INIT_FUNC(MC) | |
7046 | #endif | |
7047 | ||
7048 | /* | |
7049 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7050 | * space as the usage in build_sched_domains() dictates. Used only | |
7051 | * if the amount of space is significant. | |
7052 | */ | |
7053 | struct allmasks { | |
7054 | cpumask_t tmpmask; /* make this one first */ | |
7055 | union { | |
7056 | cpumask_t nodemask; | |
7057 | cpumask_t this_sibling_map; | |
7058 | cpumask_t this_core_map; | |
7059 | }; | |
7060 | cpumask_t send_covered; | |
7061 | ||
7062 | #ifdef CONFIG_NUMA | |
7063 | cpumask_t domainspan; | |
7064 | cpumask_t covered; | |
7065 | cpumask_t notcovered; | |
7066 | #endif | |
7067 | }; | |
7068 | ||
7069 | #if NR_CPUS > 128 | |
7070 | #define SCHED_CPUMASK_ALLOC 1 | |
7071 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7072 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7073 | #else | |
7074 | #define SCHED_CPUMASK_ALLOC 0 | |
7075 | #define SCHED_CPUMASK_FREE(v) | |
7076 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7077 | #endif | |
7078 | ||
7079 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7080 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7081 | ||
1d3504fc HS |
7082 | static int default_relax_domain_level = -1; |
7083 | ||
7084 | static int __init setup_relax_domain_level(char *str) | |
7085 | { | |
30e0e178 LZ |
7086 | unsigned long val; |
7087 | ||
7088 | val = simple_strtoul(str, NULL, 0); | |
7089 | if (val < SD_LV_MAX) | |
7090 | default_relax_domain_level = val; | |
7091 | ||
1d3504fc HS |
7092 | return 1; |
7093 | } | |
7094 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7095 | ||
7096 | static void set_domain_attribute(struct sched_domain *sd, | |
7097 | struct sched_domain_attr *attr) | |
7098 | { | |
7099 | int request; | |
7100 | ||
7101 | if (!attr || attr->relax_domain_level < 0) { | |
7102 | if (default_relax_domain_level < 0) | |
7103 | return; | |
7104 | else | |
7105 | request = default_relax_domain_level; | |
7106 | } else | |
7107 | request = attr->relax_domain_level; | |
7108 | if (request < sd->level) { | |
7109 | /* turn off idle balance on this domain */ | |
7110 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7111 | } else { | |
7112 | /* turn on idle balance on this domain */ | |
7113 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7114 | } | |
7115 | } | |
7116 | ||
1da177e4 | 7117 | /* |
1a20ff27 DG |
7118 | * Build sched domains for a given set of cpus and attach the sched domains |
7119 | * to the individual cpus | |
1da177e4 | 7120 | */ |
1d3504fc HS |
7121 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7122 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7123 | { |
7124 | int i; | |
57d885fe | 7125 | struct root_domain *rd; |
7c16ec58 MT |
7126 | SCHED_CPUMASK_DECLARE(allmasks); |
7127 | cpumask_t *tmpmask; | |
d1b55138 JH |
7128 | #ifdef CONFIG_NUMA |
7129 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7130 | int sd_allnodes = 0; |
d1b55138 JH |
7131 | |
7132 | /* | |
7133 | * Allocate the per-node list of sched groups | |
7134 | */ | |
5cf9f062 | 7135 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 7136 | GFP_KERNEL); |
d1b55138 JH |
7137 | if (!sched_group_nodes) { |
7138 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7139 | return -ENOMEM; |
d1b55138 | 7140 | } |
d1b55138 | 7141 | #endif |
1da177e4 | 7142 | |
dc938520 | 7143 | rd = alloc_rootdomain(); |
57d885fe GH |
7144 | if (!rd) { |
7145 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7146 | #ifdef CONFIG_NUMA |
7147 | kfree(sched_group_nodes); | |
7148 | #endif | |
57d885fe GH |
7149 | return -ENOMEM; |
7150 | } | |
7151 | ||
7c16ec58 MT |
7152 | #if SCHED_CPUMASK_ALLOC |
7153 | /* get space for all scratch cpumask variables */ | |
7154 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7155 | if (!allmasks) { | |
7156 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7157 | kfree(rd); | |
7158 | #ifdef CONFIG_NUMA | |
7159 | kfree(sched_group_nodes); | |
7160 | #endif | |
7161 | return -ENOMEM; | |
7162 | } | |
7163 | #endif | |
7164 | tmpmask = (cpumask_t *)allmasks; | |
7165 | ||
7166 | ||
7167 | #ifdef CONFIG_NUMA | |
7168 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7169 | #endif | |
7170 | ||
1da177e4 | 7171 | /* |
1a20ff27 | 7172 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7173 | */ |
1a20ff27 | 7174 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 7175 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7176 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7177 | |
7c16ec58 MT |
7178 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7179 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7180 | |
7181 | #ifdef CONFIG_NUMA | |
dd41f596 | 7182 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7183 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7184 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7185 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7186 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7187 | sd->span = *cpu_map; |
7c16ec58 | 7188 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7189 | p = sd; |
6711cab4 | 7190 | sd_allnodes = 1; |
9c1cfda2 JH |
7191 | } else |
7192 | p = NULL; | |
7193 | ||
1da177e4 | 7194 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7195 | SD_INIT(sd, NODE); |
1d3504fc | 7196 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7197 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7198 | sd->parent = p; |
1a848870 SS |
7199 | if (p) |
7200 | p->child = sd; | |
9c1cfda2 | 7201 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7202 | #endif |
7203 | ||
7204 | p = sd; | |
7205 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7206 | SD_INIT(sd, CPU); |
1d3504fc | 7207 | set_domain_attribute(sd, attr); |
7c16ec58 | 7208 | sd->span = *nodemask; |
1da177e4 | 7209 | sd->parent = p; |
1a848870 SS |
7210 | if (p) |
7211 | p->child = sd; | |
7c16ec58 | 7212 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7213 | |
1e9f28fa SS |
7214 | #ifdef CONFIG_SCHED_MC |
7215 | p = sd; | |
7216 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7217 | SD_INIT(sd, MC); |
1d3504fc | 7218 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7219 | sd->span = cpu_coregroup_map(i); |
7220 | cpus_and(sd->span, sd->span, *cpu_map); | |
7221 | sd->parent = p; | |
1a848870 | 7222 | p->child = sd; |
7c16ec58 | 7223 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7224 | #endif |
7225 | ||
1da177e4 LT |
7226 | #ifdef CONFIG_SCHED_SMT |
7227 | p = sd; | |
7228 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7229 | SD_INIT(sd, SIBLING); |
1d3504fc | 7230 | set_domain_attribute(sd, attr); |
d5a7430d | 7231 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7232 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7233 | sd->parent = p; |
1a848870 | 7234 | p->child = sd; |
7c16ec58 | 7235 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7236 | #endif |
7237 | } | |
7238 | ||
7239 | #ifdef CONFIG_SCHED_SMT | |
7240 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 7241 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
7242 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7243 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7244 | ||
7245 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7246 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7247 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7248 | continue; |
7249 | ||
dd41f596 | 7250 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7251 | &cpu_to_cpu_group, |
7252 | send_covered, tmpmask); | |
1da177e4 LT |
7253 | } |
7254 | #endif | |
7255 | ||
1e9f28fa SS |
7256 | #ifdef CONFIG_SCHED_MC |
7257 | /* Set up multi-core groups */ | |
7258 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
7259 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7260 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7261 | ||
7262 | *this_core_map = cpu_coregroup_map(i); | |
7263 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7264 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7265 | continue; |
7c16ec58 | 7266 | |
dd41f596 | 7267 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7268 | &cpu_to_core_group, |
7269 | send_covered, tmpmask); | |
1e9f28fa SS |
7270 | } |
7271 | #endif | |
7272 | ||
1da177e4 LT |
7273 | /* Set up physical groups */ |
7274 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
7275 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7276 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7277 | |
7c16ec58 MT |
7278 | *nodemask = node_to_cpumask(i); |
7279 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7280 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7281 | continue; |
7282 | ||
7c16ec58 MT |
7283 | init_sched_build_groups(nodemask, cpu_map, |
7284 | &cpu_to_phys_group, | |
7285 | send_covered, tmpmask); | |
1da177e4 LT |
7286 | } |
7287 | ||
7288 | #ifdef CONFIG_NUMA | |
7289 | /* Set up node groups */ | |
7c16ec58 MT |
7290 | if (sd_allnodes) { |
7291 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7292 | ||
7293 | init_sched_build_groups(cpu_map, cpu_map, | |
7294 | &cpu_to_allnodes_group, | |
7295 | send_covered, tmpmask); | |
7296 | } | |
9c1cfda2 JH |
7297 | |
7298 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7299 | /* Set up node groups */ | |
7300 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7301 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7302 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7303 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7304 | int j; |
7305 | ||
7c16ec58 MT |
7306 | *nodemask = node_to_cpumask(i); |
7307 | cpus_clear(*covered); | |
7308 | ||
7309 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7310 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7311 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7312 | continue; |
d1b55138 | 7313 | } |
9c1cfda2 | 7314 | |
4bdbaad3 | 7315 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7316 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7317 | |
15f0b676 | 7318 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7319 | if (!sg) { |
7320 | printk(KERN_WARNING "Can not alloc domain group for " | |
7321 | "node %d\n", i); | |
7322 | goto error; | |
7323 | } | |
9c1cfda2 | 7324 | sched_group_nodes[i] = sg; |
7c16ec58 | 7325 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7326 | struct sched_domain *sd; |
9761eea8 | 7327 | |
9c1cfda2 JH |
7328 | sd = &per_cpu(node_domains, j); |
7329 | sd->groups = sg; | |
9c1cfda2 | 7330 | } |
5517d86b | 7331 | sg->__cpu_power = 0; |
7c16ec58 | 7332 | sg->cpumask = *nodemask; |
51888ca2 | 7333 | sg->next = sg; |
7c16ec58 | 7334 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7335 | prev = sg; |
7336 | ||
7337 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7338 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7339 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7340 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7341 | |
7c16ec58 MT |
7342 | cpus_complement(*notcovered, *covered); |
7343 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7344 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7345 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7346 | break; |
7347 | ||
7c16ec58 MT |
7348 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7349 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7350 | continue; |
7351 | ||
15f0b676 SV |
7352 | sg = kmalloc_node(sizeof(struct sched_group), |
7353 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7354 | if (!sg) { |
7355 | printk(KERN_WARNING | |
7356 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7357 | goto error; |
9c1cfda2 | 7358 | } |
5517d86b | 7359 | sg->__cpu_power = 0; |
7c16ec58 | 7360 | sg->cpumask = *tmpmask; |
51888ca2 | 7361 | sg->next = prev->next; |
7c16ec58 | 7362 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7363 | prev->next = sg; |
7364 | prev = sg; | |
7365 | } | |
9c1cfda2 | 7366 | } |
1da177e4 LT |
7367 | #endif |
7368 | ||
7369 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7370 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7371 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7372 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7373 | ||
89c4710e | 7374 | init_sched_groups_power(i, sd); |
5c45bf27 | 7375 | } |
1da177e4 | 7376 | #endif |
1e9f28fa | 7377 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7378 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7379 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7380 | ||
89c4710e | 7381 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7382 | } |
7383 | #endif | |
1e9f28fa | 7384 | |
5c45bf27 | 7385 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7386 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7387 | ||
89c4710e | 7388 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7389 | } |
7390 | ||
9c1cfda2 | 7391 | #ifdef CONFIG_NUMA |
08069033 SS |
7392 | for (i = 0; i < MAX_NUMNODES; i++) |
7393 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7394 | |
6711cab4 SS |
7395 | if (sd_allnodes) { |
7396 | struct sched_group *sg; | |
f712c0c7 | 7397 | |
7c16ec58 MT |
7398 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7399 | tmpmask); | |
f712c0c7 SS |
7400 | init_numa_sched_groups_power(sg); |
7401 | } | |
9c1cfda2 JH |
7402 | #endif |
7403 | ||
1da177e4 | 7404 | /* Attach the domains */ |
1a20ff27 | 7405 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7406 | struct sched_domain *sd; |
7407 | #ifdef CONFIG_SCHED_SMT | |
7408 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7409 | #elif defined(CONFIG_SCHED_MC) |
7410 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7411 | #else |
7412 | sd = &per_cpu(phys_domains, i); | |
7413 | #endif | |
57d885fe | 7414 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7415 | } |
51888ca2 | 7416 | |
7c16ec58 | 7417 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7418 | return 0; |
7419 | ||
a616058b | 7420 | #ifdef CONFIG_NUMA |
51888ca2 | 7421 | error: |
7c16ec58 MT |
7422 | free_sched_groups(cpu_map, tmpmask); |
7423 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7424 | return -ENOMEM; |
a616058b | 7425 | #endif |
1da177e4 | 7426 | } |
029190c5 | 7427 | |
1d3504fc HS |
7428 | static int build_sched_domains(const cpumask_t *cpu_map) |
7429 | { | |
7430 | return __build_sched_domains(cpu_map, NULL); | |
7431 | } | |
7432 | ||
029190c5 PJ |
7433 | static cpumask_t *doms_cur; /* current sched domains */ |
7434 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7435 | static struct sched_domain_attr *dattr_cur; |
7436 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7437 | |
7438 | /* | |
7439 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7440 | * cpumask_t) fails, then fallback to a single sched domain, | |
7441 | * as determined by the single cpumask_t fallback_doms. | |
7442 | */ | |
7443 | static cpumask_t fallback_doms; | |
7444 | ||
22e52b07 HC |
7445 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7446 | { | |
7447 | } | |
7448 | ||
5c8e1ed1 MK |
7449 | /* |
7450 | * Free current domain masks. | |
7451 | * Called after all cpus are attached to NULL domain. | |
7452 | */ | |
7453 | static void free_sched_domains(void) | |
7454 | { | |
7455 | ndoms_cur = 0; | |
7456 | if (doms_cur != &fallback_doms) | |
7457 | kfree(doms_cur); | |
7458 | doms_cur = &fallback_doms; | |
7459 | } | |
7460 | ||
1a20ff27 | 7461 | /* |
41a2d6cf | 7462 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7463 | * For now this just excludes isolated cpus, but could be used to |
7464 | * exclude other special cases in the future. | |
1a20ff27 | 7465 | */ |
51888ca2 | 7466 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7467 | { |
7378547f MM |
7468 | int err; |
7469 | ||
22e52b07 | 7470 | arch_update_cpu_topology(); |
029190c5 PJ |
7471 | ndoms_cur = 1; |
7472 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7473 | if (!doms_cur) | |
7474 | doms_cur = &fallback_doms; | |
7475 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7476 | dattr_cur = NULL; |
7378547f | 7477 | err = build_sched_domains(doms_cur); |
6382bc90 | 7478 | register_sched_domain_sysctl(); |
7378547f MM |
7479 | |
7480 | return err; | |
1a20ff27 DG |
7481 | } |
7482 | ||
7c16ec58 MT |
7483 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7484 | cpumask_t *tmpmask) | |
1da177e4 | 7485 | { |
7c16ec58 | 7486 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7487 | } |
1da177e4 | 7488 | |
1a20ff27 DG |
7489 | /* |
7490 | * Detach sched domains from a group of cpus specified in cpu_map | |
7491 | * These cpus will now be attached to the NULL domain | |
7492 | */ | |
858119e1 | 7493 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7494 | { |
7c16ec58 | 7495 | cpumask_t tmpmask; |
1a20ff27 DG |
7496 | int i; |
7497 | ||
6382bc90 MM |
7498 | unregister_sched_domain_sysctl(); |
7499 | ||
1a20ff27 | 7500 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7501 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7502 | synchronize_sched(); |
7c16ec58 | 7503 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7504 | } |
7505 | ||
1d3504fc HS |
7506 | /* handle null as "default" */ |
7507 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7508 | struct sched_domain_attr *new, int idx_new) | |
7509 | { | |
7510 | struct sched_domain_attr tmp; | |
7511 | ||
7512 | /* fast path */ | |
7513 | if (!new && !cur) | |
7514 | return 1; | |
7515 | ||
7516 | tmp = SD_ATTR_INIT; | |
7517 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7518 | new ? (new + idx_new) : &tmp, | |
7519 | sizeof(struct sched_domain_attr)); | |
7520 | } | |
7521 | ||
029190c5 PJ |
7522 | /* |
7523 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7524 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7525 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7526 | * It destroys each deleted domain and builds each new domain. | |
7527 | * | |
7528 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7529 | * The masks don't intersect (don't overlap.) We should setup one |
7530 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7531 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7532 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7533 | * it as it is. | |
7534 | * | |
41a2d6cf IM |
7535 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7536 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7537 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7538 | * and partition_sched_domains() will fallback to the single partition | |
7539 | * 'fallback_doms'. | |
7540 | * | |
7541 | * Call with hotplug lock held | |
7542 | */ | |
1d3504fc HS |
7543 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7544 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7545 | { |
7546 | int i, j; | |
7547 | ||
712555ee | 7548 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7549 | |
7378547f MM |
7550 | /* always unregister in case we don't destroy any domains */ |
7551 | unregister_sched_domain_sysctl(); | |
7552 | ||
029190c5 PJ |
7553 | if (doms_new == NULL) { |
7554 | ndoms_new = 1; | |
7555 | doms_new = &fallback_doms; | |
7556 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
1d3504fc | 7557 | dattr_new = NULL; |
029190c5 PJ |
7558 | } |
7559 | ||
7560 | /* Destroy deleted domains */ | |
7561 | for (i = 0; i < ndoms_cur; i++) { | |
7562 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7563 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7564 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7565 | goto match1; |
7566 | } | |
7567 | /* no match - a current sched domain not in new doms_new[] */ | |
7568 | detach_destroy_domains(doms_cur + i); | |
7569 | match1: | |
7570 | ; | |
7571 | } | |
7572 | ||
7573 | /* Build new domains */ | |
7574 | for (i = 0; i < ndoms_new; i++) { | |
7575 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7576 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7577 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7578 | goto match2; |
7579 | } | |
7580 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7581 | __build_sched_domains(doms_new + i, |
7582 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7583 | match2: |
7584 | ; | |
7585 | } | |
7586 | ||
7587 | /* Remember the new sched domains */ | |
7588 | if (doms_cur != &fallback_doms) | |
7589 | kfree(doms_cur); | |
1d3504fc | 7590 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7591 | doms_cur = doms_new; |
1d3504fc | 7592 | dattr_cur = dattr_new; |
029190c5 | 7593 | ndoms_cur = ndoms_new; |
7378547f MM |
7594 | |
7595 | register_sched_domain_sysctl(); | |
a1835615 | 7596 | |
712555ee | 7597 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7598 | } |
7599 | ||
5c45bf27 | 7600 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7601 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7602 | { |
7603 | int err; | |
7604 | ||
95402b38 | 7605 | get_online_cpus(); |
712555ee | 7606 | mutex_lock(&sched_domains_mutex); |
5c45bf27 | 7607 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7608 | free_sched_domains(); |
5c45bf27 | 7609 | err = arch_init_sched_domains(&cpu_online_map); |
712555ee | 7610 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7611 | put_online_cpus(); |
5c45bf27 SS |
7612 | |
7613 | return err; | |
7614 | } | |
7615 | ||
7616 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7617 | { | |
7618 | int ret; | |
7619 | ||
7620 | if (buf[0] != '0' && buf[0] != '1') | |
7621 | return -EINVAL; | |
7622 | ||
7623 | if (smt) | |
7624 | sched_smt_power_savings = (buf[0] == '1'); | |
7625 | else | |
7626 | sched_mc_power_savings = (buf[0] == '1'); | |
7627 | ||
7628 | ret = arch_reinit_sched_domains(); | |
7629 | ||
7630 | return ret ? ret : count; | |
7631 | } | |
7632 | ||
5c45bf27 SS |
7633 | #ifdef CONFIG_SCHED_MC |
7634 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7635 | { | |
7636 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7637 | } | |
48f24c4d IM |
7638 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7639 | const char *buf, size_t count) | |
5c45bf27 SS |
7640 | { |
7641 | return sched_power_savings_store(buf, count, 0); | |
7642 | } | |
6707de00 AB |
7643 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7644 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7645 | #endif |
7646 | ||
7647 | #ifdef CONFIG_SCHED_SMT | |
7648 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7649 | { | |
7650 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7651 | } | |
48f24c4d IM |
7652 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7653 | const char *buf, size_t count) | |
5c45bf27 SS |
7654 | { |
7655 | return sched_power_savings_store(buf, count, 1); | |
7656 | } | |
6707de00 AB |
7657 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7658 | sched_smt_power_savings_store); | |
7659 | #endif | |
7660 | ||
7661 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7662 | { | |
7663 | int err = 0; | |
7664 | ||
7665 | #ifdef CONFIG_SCHED_SMT | |
7666 | if (smt_capable()) | |
7667 | err = sysfs_create_file(&cls->kset.kobj, | |
7668 | &attr_sched_smt_power_savings.attr); | |
7669 | #endif | |
7670 | #ifdef CONFIG_SCHED_MC | |
7671 | if (!err && mc_capable()) | |
7672 | err = sysfs_create_file(&cls->kset.kobj, | |
7673 | &attr_sched_mc_power_savings.attr); | |
7674 | #endif | |
7675 | return err; | |
7676 | } | |
6d6bc0ad | 7677 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7678 | |
1da177e4 | 7679 | /* |
41a2d6cf | 7680 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7681 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7682 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7683 | * which will prevent rebalancing while the sched domains are recalculated. |
7684 | */ | |
7685 | static int update_sched_domains(struct notifier_block *nfb, | |
7686 | unsigned long action, void *hcpu) | |
7687 | { | |
7def2be1 PZ |
7688 | int cpu = (int)(long)hcpu; |
7689 | ||
1da177e4 | 7690 | switch (action) { |
1da177e4 | 7691 | case CPU_DOWN_PREPARE: |
8bb78442 | 7692 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 PZ |
7693 | disable_runtime(cpu_rq(cpu)); |
7694 | /* fall-through */ | |
7695 | case CPU_UP_PREPARE: | |
7696 | case CPU_UP_PREPARE_FROZEN: | |
1a20ff27 | 7697 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7698 | free_sched_domains(); |
1da177e4 LT |
7699 | return NOTIFY_OK; |
7700 | ||
7def2be1 | 7701 | |
1da177e4 | 7702 | case CPU_DOWN_FAILED: |
8bb78442 | 7703 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7704 | case CPU_ONLINE: |
8bb78442 | 7705 | case CPU_ONLINE_FROZEN: |
7def2be1 PZ |
7706 | enable_runtime(cpu_rq(cpu)); |
7707 | /* fall-through */ | |
7708 | case CPU_UP_CANCELED: | |
7709 | case CPU_UP_CANCELED_FROZEN: | |
1da177e4 | 7710 | case CPU_DEAD: |
8bb78442 | 7711 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7712 | /* |
7713 | * Fall through and re-initialise the domains. | |
7714 | */ | |
7715 | break; | |
7716 | default: | |
7717 | return NOTIFY_DONE; | |
7718 | } | |
7719 | ||
5c8e1ed1 MK |
7720 | #ifndef CONFIG_CPUSETS |
7721 | /* | |
7722 | * Create default domain partitioning if cpusets are disabled. | |
7723 | * Otherwise we let cpusets rebuild the domains based on the | |
7724 | * current setup. | |
7725 | */ | |
7726 | ||
1da177e4 | 7727 | /* The hotplug lock is already held by cpu_up/cpu_down */ |
1a20ff27 | 7728 | arch_init_sched_domains(&cpu_online_map); |
5c8e1ed1 | 7729 | #endif |
1da177e4 LT |
7730 | |
7731 | return NOTIFY_OK; | |
7732 | } | |
1da177e4 LT |
7733 | |
7734 | void __init sched_init_smp(void) | |
7735 | { | |
5c1e1767 NP |
7736 | cpumask_t non_isolated_cpus; |
7737 | ||
434d53b0 MT |
7738 | #if defined(CONFIG_NUMA) |
7739 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7740 | GFP_KERNEL); | |
7741 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7742 | #endif | |
95402b38 | 7743 | get_online_cpus(); |
712555ee | 7744 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7745 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7746 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7747 | if (cpus_empty(non_isolated_cpus)) |
7748 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7749 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7750 | put_online_cpus(); |
1da177e4 LT |
7751 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7752 | hotcpu_notifier(update_sched_domains, 0); | |
b328ca18 | 7753 | init_hrtick(); |
5c1e1767 NP |
7754 | |
7755 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7756 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7757 | BUG(); |
19978ca6 | 7758 | sched_init_granularity(); |
1da177e4 LT |
7759 | } |
7760 | #else | |
7761 | void __init sched_init_smp(void) | |
7762 | { | |
19978ca6 | 7763 | sched_init_granularity(); |
1da177e4 LT |
7764 | } |
7765 | #endif /* CONFIG_SMP */ | |
7766 | ||
7767 | int in_sched_functions(unsigned long addr) | |
7768 | { | |
1da177e4 LT |
7769 | return in_lock_functions(addr) || |
7770 | (addr >= (unsigned long)__sched_text_start | |
7771 | && addr < (unsigned long)__sched_text_end); | |
7772 | } | |
7773 | ||
a9957449 | 7774 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7775 | { |
7776 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7777 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7778 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7779 | cfs_rq->rq = rq; | |
7780 | #endif | |
67e9fb2a | 7781 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7782 | } |
7783 | ||
fa85ae24 PZ |
7784 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7785 | { | |
7786 | struct rt_prio_array *array; | |
7787 | int i; | |
7788 | ||
7789 | array = &rt_rq->active; | |
7790 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
20b6331b | 7791 | INIT_LIST_HEAD(array->queue + i); |
fa85ae24 PZ |
7792 | __clear_bit(i, array->bitmap); |
7793 | } | |
7794 | /* delimiter for bitsearch: */ | |
7795 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7796 | ||
052f1dc7 | 7797 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7798 | rt_rq->highest_prio = MAX_RT_PRIO; |
7799 | #endif | |
fa85ae24 PZ |
7800 | #ifdef CONFIG_SMP |
7801 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7802 | rt_rq->overloaded = 0; |
7803 | #endif | |
7804 | ||
7805 | rt_rq->rt_time = 0; | |
7806 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7807 | rt_rq->rt_runtime = 0; |
7808 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7809 | |
052f1dc7 | 7810 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7811 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7812 | rt_rq->rq = rq; |
7813 | #endif | |
fa85ae24 PZ |
7814 | } |
7815 | ||
6f505b16 | 7816 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7817 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7818 | struct sched_entity *se, int cpu, int add, | |
7819 | struct sched_entity *parent) | |
6f505b16 | 7820 | { |
ec7dc8ac | 7821 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7822 | tg->cfs_rq[cpu] = cfs_rq; |
7823 | init_cfs_rq(cfs_rq, rq); | |
7824 | cfs_rq->tg = tg; | |
7825 | if (add) | |
7826 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7827 | ||
7828 | tg->se[cpu] = se; | |
354d60c2 DG |
7829 | /* se could be NULL for init_task_group */ |
7830 | if (!se) | |
7831 | return; | |
7832 | ||
ec7dc8ac DG |
7833 | if (!parent) |
7834 | se->cfs_rq = &rq->cfs; | |
7835 | else | |
7836 | se->cfs_rq = parent->my_q; | |
7837 | ||
6f505b16 PZ |
7838 | se->my_q = cfs_rq; |
7839 | se->load.weight = tg->shares; | |
e05510d0 | 7840 | se->load.inv_weight = 0; |
ec7dc8ac | 7841 | se->parent = parent; |
6f505b16 | 7842 | } |
052f1dc7 | 7843 | #endif |
6f505b16 | 7844 | |
052f1dc7 | 7845 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7846 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7847 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7848 | struct sched_rt_entity *parent) | |
6f505b16 | 7849 | { |
ec7dc8ac DG |
7850 | struct rq *rq = cpu_rq(cpu); |
7851 | ||
6f505b16 PZ |
7852 | tg->rt_rq[cpu] = rt_rq; |
7853 | init_rt_rq(rt_rq, rq); | |
7854 | rt_rq->tg = tg; | |
7855 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7856 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7857 | if (add) |
7858 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7859 | ||
7860 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7861 | if (!rt_se) |
7862 | return; | |
7863 | ||
ec7dc8ac DG |
7864 | if (!parent) |
7865 | rt_se->rt_rq = &rq->rt; | |
7866 | else | |
7867 | rt_se->rt_rq = parent->my_q; | |
7868 | ||
6f505b16 | 7869 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7870 | rt_se->parent = parent; |
6f505b16 PZ |
7871 | INIT_LIST_HEAD(&rt_se->run_list); |
7872 | } | |
7873 | #endif | |
7874 | ||
1da177e4 LT |
7875 | void __init sched_init(void) |
7876 | { | |
dd41f596 | 7877 | int i, j; |
434d53b0 MT |
7878 | unsigned long alloc_size = 0, ptr; |
7879 | ||
7880 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7881 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7882 | #endif | |
7883 | #ifdef CONFIG_RT_GROUP_SCHED | |
7884 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7885 | #endif |
7886 | #ifdef CONFIG_USER_SCHED | |
7887 | alloc_size *= 2; | |
434d53b0 MT |
7888 | #endif |
7889 | /* | |
7890 | * As sched_init() is called before page_alloc is setup, | |
7891 | * we use alloc_bootmem(). | |
7892 | */ | |
7893 | if (alloc_size) { | |
5a9d3225 | 7894 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7895 | |
7896 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7897 | init_task_group.se = (struct sched_entity **)ptr; | |
7898 | ptr += nr_cpu_ids * sizeof(void **); | |
7899 | ||
7900 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7901 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7902 | |
7903 | #ifdef CONFIG_USER_SCHED | |
7904 | root_task_group.se = (struct sched_entity **)ptr; | |
7905 | ptr += nr_cpu_ids * sizeof(void **); | |
7906 | ||
7907 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7908 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7909 | #endif /* CONFIG_USER_SCHED */ |
7910 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7911 | #ifdef CONFIG_RT_GROUP_SCHED |
7912 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7913 | ptr += nr_cpu_ids * sizeof(void **); | |
7914 | ||
7915 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7916 | ptr += nr_cpu_ids * sizeof(void **); |
7917 | ||
7918 | #ifdef CONFIG_USER_SCHED | |
7919 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7920 | ptr += nr_cpu_ids * sizeof(void **); | |
7921 | ||
7922 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7923 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7924 | #endif /* CONFIG_USER_SCHED */ |
7925 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7926 | } |
dd41f596 | 7927 | |
57d885fe GH |
7928 | #ifdef CONFIG_SMP |
7929 | init_defrootdomain(); | |
7930 | #endif | |
7931 | ||
d0b27fa7 PZ |
7932 | init_rt_bandwidth(&def_rt_bandwidth, |
7933 | global_rt_period(), global_rt_runtime()); | |
7934 | ||
7935 | #ifdef CONFIG_RT_GROUP_SCHED | |
7936 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7937 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7938 | #ifdef CONFIG_USER_SCHED |
7939 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7940 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7941 | #endif /* CONFIG_USER_SCHED */ |
7942 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7943 | |
052f1dc7 | 7944 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7945 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7946 | INIT_LIST_HEAD(&init_task_group.children); |
7947 | ||
7948 | #ifdef CONFIG_USER_SCHED | |
7949 | INIT_LIST_HEAD(&root_task_group.children); | |
7950 | init_task_group.parent = &root_task_group; | |
7951 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7952 | #endif /* CONFIG_USER_SCHED */ |
7953 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7954 | |
0a945022 | 7955 | for_each_possible_cpu(i) { |
70b97a7f | 7956 | struct rq *rq; |
1da177e4 LT |
7957 | |
7958 | rq = cpu_rq(i); | |
7959 | spin_lock_init(&rq->lock); | |
fcb99371 | 7960 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7961 | rq->nr_running = 0; |
dd41f596 | 7962 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7963 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7964 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7965 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7966 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7967 | #ifdef CONFIG_CGROUP_SCHED |
7968 | /* | |
7969 | * How much cpu bandwidth does init_task_group get? | |
7970 | * | |
7971 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7972 | * gets 100% of the cpu resources in the system. This overall | |
7973 | * system cpu resource is divided among the tasks of | |
7974 | * init_task_group and its child task-groups in a fair manner, | |
7975 | * based on each entity's (task or task-group's) weight | |
7976 | * (se->load.weight). | |
7977 | * | |
7978 | * In other words, if init_task_group has 10 tasks of weight | |
7979 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7980 | * then A0's share of the cpu resource is: | |
7981 | * | |
7982 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7983 | * | |
7984 | * We achieve this by letting init_task_group's tasks sit | |
7985 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7986 | */ | |
ec7dc8ac | 7987 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7988 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
7989 | root_task_group.shares = NICE_0_LOAD; |
7990 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
7991 | /* |
7992 | * In case of task-groups formed thr' the user id of tasks, | |
7993 | * init_task_group represents tasks belonging to root user. | |
7994 | * Hence it forms a sibling of all subsequent groups formed. | |
7995 | * In this case, init_task_group gets only a fraction of overall | |
7996 | * system cpu resource, based on the weight assigned to root | |
7997 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
7998 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
7999 | * (init_cfs_rq) and having one entity represent this group of | |
8000 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8001 | */ | |
ec7dc8ac | 8002 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8003 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8004 | &per_cpu(init_sched_entity, i), i, 1, |
8005 | root_task_group.se[i]); | |
6f505b16 | 8006 | |
052f1dc7 | 8007 | #endif |
354d60c2 DG |
8008 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8009 | ||
8010 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8011 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8012 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8013 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8014 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8015 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8016 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8017 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8018 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8019 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8020 | root_task_group.rt_se[i]); | |
354d60c2 | 8021 | #endif |
dd41f596 | 8022 | #endif |
1da177e4 | 8023 | |
dd41f596 IM |
8024 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8025 | rq->cpu_load[j] = 0; | |
1da177e4 | 8026 | #ifdef CONFIG_SMP |
41c7ce9a | 8027 | rq->sd = NULL; |
57d885fe | 8028 | rq->rd = NULL; |
1da177e4 | 8029 | rq->active_balance = 0; |
dd41f596 | 8030 | rq->next_balance = jiffies; |
1da177e4 | 8031 | rq->push_cpu = 0; |
0a2966b4 | 8032 | rq->cpu = i; |
1f11eb6a | 8033 | rq->online = 0; |
1da177e4 LT |
8034 | rq->migration_thread = NULL; |
8035 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8036 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8037 | #endif |
8f4d37ec | 8038 | init_rq_hrtick(rq); |
1da177e4 | 8039 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8040 | } |
8041 | ||
2dd73a4f | 8042 | set_load_weight(&init_task); |
b50f60ce | 8043 | |
e107be36 AK |
8044 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8045 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8046 | #endif | |
8047 | ||
c9819f45 CL |
8048 | #ifdef CONFIG_SMP |
8049 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
8050 | #endif | |
8051 | ||
b50f60ce HC |
8052 | #ifdef CONFIG_RT_MUTEXES |
8053 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8054 | #endif | |
8055 | ||
1da177e4 LT |
8056 | /* |
8057 | * The boot idle thread does lazy MMU switching as well: | |
8058 | */ | |
8059 | atomic_inc(&init_mm.mm_count); | |
8060 | enter_lazy_tlb(&init_mm, current); | |
8061 | ||
8062 | /* | |
8063 | * Make us the idle thread. Technically, schedule() should not be | |
8064 | * called from this thread, however somewhere below it might be, | |
8065 | * but because we are the idle thread, we just pick up running again | |
8066 | * when this runqueue becomes "idle". | |
8067 | */ | |
8068 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8069 | /* |
8070 | * During early bootup we pretend to be a normal task: | |
8071 | */ | |
8072 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8073 | |
8074 | scheduler_running = 1; | |
1da177e4 LT |
8075 | } |
8076 | ||
8077 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8078 | void __might_sleep(char *file, int line) | |
8079 | { | |
48f24c4d | 8080 | #ifdef in_atomic |
1da177e4 LT |
8081 | static unsigned long prev_jiffy; /* ratelimiting */ |
8082 | ||
8083 | if ((in_atomic() || irqs_disabled()) && | |
8084 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8085 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8086 | return; | |
8087 | prev_jiffy = jiffies; | |
91368d73 | 8088 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8089 | " context at %s:%d\n", file, line); |
8090 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8091 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8092 | debug_show_held_locks(current); |
3117df04 IM |
8093 | if (irqs_disabled()) |
8094 | print_irqtrace_events(current); | |
1da177e4 LT |
8095 | dump_stack(); |
8096 | } | |
8097 | #endif | |
8098 | } | |
8099 | EXPORT_SYMBOL(__might_sleep); | |
8100 | #endif | |
8101 | ||
8102 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8103 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8104 | { | |
8105 | int on_rq; | |
3e51f33f | 8106 | |
3a5e4dc1 AK |
8107 | update_rq_clock(rq); |
8108 | on_rq = p->se.on_rq; | |
8109 | if (on_rq) | |
8110 | deactivate_task(rq, p, 0); | |
8111 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8112 | if (on_rq) { | |
8113 | activate_task(rq, p, 0); | |
8114 | resched_task(rq->curr); | |
8115 | } | |
8116 | } | |
8117 | ||
1da177e4 LT |
8118 | void normalize_rt_tasks(void) |
8119 | { | |
a0f98a1c | 8120 | struct task_struct *g, *p; |
1da177e4 | 8121 | unsigned long flags; |
70b97a7f | 8122 | struct rq *rq; |
1da177e4 | 8123 | |
4cf5d77a | 8124 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8125 | do_each_thread(g, p) { |
178be793 IM |
8126 | /* |
8127 | * Only normalize user tasks: | |
8128 | */ | |
8129 | if (!p->mm) | |
8130 | continue; | |
8131 | ||
6cfb0d5d | 8132 | p->se.exec_start = 0; |
6cfb0d5d | 8133 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8134 | p->se.wait_start = 0; |
dd41f596 | 8135 | p->se.sleep_start = 0; |
dd41f596 | 8136 | p->se.block_start = 0; |
6cfb0d5d | 8137 | #endif |
dd41f596 IM |
8138 | |
8139 | if (!rt_task(p)) { | |
8140 | /* | |
8141 | * Renice negative nice level userspace | |
8142 | * tasks back to 0: | |
8143 | */ | |
8144 | if (TASK_NICE(p) < 0 && p->mm) | |
8145 | set_user_nice(p, 0); | |
1da177e4 | 8146 | continue; |
dd41f596 | 8147 | } |
1da177e4 | 8148 | |
4cf5d77a | 8149 | spin_lock(&p->pi_lock); |
b29739f9 | 8150 | rq = __task_rq_lock(p); |
1da177e4 | 8151 | |
178be793 | 8152 | normalize_task(rq, p); |
3a5e4dc1 | 8153 | |
b29739f9 | 8154 | __task_rq_unlock(rq); |
4cf5d77a | 8155 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8156 | } while_each_thread(g, p); |
8157 | ||
4cf5d77a | 8158 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8159 | } |
8160 | ||
8161 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8162 | |
8163 | #ifdef CONFIG_IA64 | |
8164 | /* | |
8165 | * These functions are only useful for the IA64 MCA handling. | |
8166 | * | |
8167 | * They can only be called when the whole system has been | |
8168 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8169 | * activity can take place. Using them for anything else would | |
8170 | * be a serious bug, and as a result, they aren't even visible | |
8171 | * under any other configuration. | |
8172 | */ | |
8173 | ||
8174 | /** | |
8175 | * curr_task - return the current task for a given cpu. | |
8176 | * @cpu: the processor in question. | |
8177 | * | |
8178 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8179 | */ | |
36c8b586 | 8180 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8181 | { |
8182 | return cpu_curr(cpu); | |
8183 | } | |
8184 | ||
8185 | /** | |
8186 | * set_curr_task - set the current task for a given cpu. | |
8187 | * @cpu: the processor in question. | |
8188 | * @p: the task pointer to set. | |
8189 | * | |
8190 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8191 | * are serviced on a separate stack. It allows the architecture to switch the |
8192 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8193 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8194 | * and caller must save the original value of the current task (see | |
8195 | * curr_task() above) and restore that value before reenabling interrupts and | |
8196 | * re-starting the system. | |
8197 | * | |
8198 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8199 | */ | |
36c8b586 | 8200 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8201 | { |
8202 | cpu_curr(cpu) = p; | |
8203 | } | |
8204 | ||
8205 | #endif | |
29f59db3 | 8206 | |
bccbe08a PZ |
8207 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8208 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8209 | { |
8210 | int i; | |
8211 | ||
8212 | for_each_possible_cpu(i) { | |
8213 | if (tg->cfs_rq) | |
8214 | kfree(tg->cfs_rq[i]); | |
8215 | if (tg->se) | |
8216 | kfree(tg->se[i]); | |
6f505b16 PZ |
8217 | } |
8218 | ||
8219 | kfree(tg->cfs_rq); | |
8220 | kfree(tg->se); | |
6f505b16 PZ |
8221 | } |
8222 | ||
ec7dc8ac DG |
8223 | static |
8224 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8225 | { |
29f59db3 | 8226 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8227 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8228 | struct rq *rq; |
29f59db3 SV |
8229 | int i; |
8230 | ||
434d53b0 | 8231 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8232 | if (!tg->cfs_rq) |
8233 | goto err; | |
434d53b0 | 8234 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8235 | if (!tg->se) |
8236 | goto err; | |
052f1dc7 PZ |
8237 | |
8238 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8239 | |
8240 | for_each_possible_cpu(i) { | |
9b5b7751 | 8241 | rq = cpu_rq(i); |
29f59db3 | 8242 | |
6f505b16 PZ |
8243 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8244 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8245 | if (!cfs_rq) |
8246 | goto err; | |
8247 | ||
6f505b16 PZ |
8248 | se = kmalloc_node(sizeof(struct sched_entity), |
8249 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8250 | if (!se) |
8251 | goto err; | |
8252 | ||
ec7dc8ac DG |
8253 | parent_se = parent ? parent->se[i] : NULL; |
8254 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8255 | } |
8256 | ||
8257 | return 1; | |
8258 | ||
8259 | err: | |
8260 | return 0; | |
8261 | } | |
8262 | ||
8263 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8264 | { | |
8265 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8266 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8267 | } | |
8268 | ||
8269 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8270 | { | |
8271 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8272 | } | |
6d6bc0ad | 8273 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8274 | static inline void free_fair_sched_group(struct task_group *tg) |
8275 | { | |
8276 | } | |
8277 | ||
ec7dc8ac DG |
8278 | static inline |
8279 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8280 | { |
8281 | return 1; | |
8282 | } | |
8283 | ||
8284 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8285 | { | |
8286 | } | |
8287 | ||
8288 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8289 | { | |
8290 | } | |
6d6bc0ad | 8291 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8292 | |
8293 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8294 | static void free_rt_sched_group(struct task_group *tg) |
8295 | { | |
8296 | int i; | |
8297 | ||
d0b27fa7 PZ |
8298 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8299 | ||
bccbe08a PZ |
8300 | for_each_possible_cpu(i) { |
8301 | if (tg->rt_rq) | |
8302 | kfree(tg->rt_rq[i]); | |
8303 | if (tg->rt_se) | |
8304 | kfree(tg->rt_se[i]); | |
8305 | } | |
8306 | ||
8307 | kfree(tg->rt_rq); | |
8308 | kfree(tg->rt_se); | |
8309 | } | |
8310 | ||
ec7dc8ac DG |
8311 | static |
8312 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8313 | { |
8314 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8315 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8316 | struct rq *rq; |
8317 | int i; | |
8318 | ||
434d53b0 | 8319 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8320 | if (!tg->rt_rq) |
8321 | goto err; | |
434d53b0 | 8322 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8323 | if (!tg->rt_se) |
8324 | goto err; | |
8325 | ||
d0b27fa7 PZ |
8326 | init_rt_bandwidth(&tg->rt_bandwidth, |
8327 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8328 | |
8329 | for_each_possible_cpu(i) { | |
8330 | rq = cpu_rq(i); | |
8331 | ||
6f505b16 PZ |
8332 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8333 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8334 | if (!rt_rq) | |
8335 | goto err; | |
29f59db3 | 8336 | |
6f505b16 PZ |
8337 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8338 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8339 | if (!rt_se) | |
8340 | goto err; | |
29f59db3 | 8341 | |
ec7dc8ac DG |
8342 | parent_se = parent ? parent->rt_se[i] : NULL; |
8343 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8344 | } |
8345 | ||
bccbe08a PZ |
8346 | return 1; |
8347 | ||
8348 | err: | |
8349 | return 0; | |
8350 | } | |
8351 | ||
8352 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8353 | { | |
8354 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8355 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8356 | } | |
8357 | ||
8358 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8359 | { | |
8360 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8361 | } | |
6d6bc0ad | 8362 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8363 | static inline void free_rt_sched_group(struct task_group *tg) |
8364 | { | |
8365 | } | |
8366 | ||
ec7dc8ac DG |
8367 | static inline |
8368 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8369 | { |
8370 | return 1; | |
8371 | } | |
8372 | ||
8373 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8374 | { | |
8375 | } | |
8376 | ||
8377 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8378 | { | |
8379 | } | |
6d6bc0ad | 8380 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8381 | |
d0b27fa7 | 8382 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8383 | static void free_sched_group(struct task_group *tg) |
8384 | { | |
8385 | free_fair_sched_group(tg); | |
8386 | free_rt_sched_group(tg); | |
8387 | kfree(tg); | |
8388 | } | |
8389 | ||
8390 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8391 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8392 | { |
8393 | struct task_group *tg; | |
8394 | unsigned long flags; | |
8395 | int i; | |
8396 | ||
8397 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8398 | if (!tg) | |
8399 | return ERR_PTR(-ENOMEM); | |
8400 | ||
ec7dc8ac | 8401 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8402 | goto err; |
8403 | ||
ec7dc8ac | 8404 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8405 | goto err; |
8406 | ||
8ed36996 | 8407 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8408 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8409 | register_fair_sched_group(tg, i); |
8410 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8411 | } |
6f505b16 | 8412 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8413 | |
8414 | WARN_ON(!parent); /* root should already exist */ | |
8415 | ||
8416 | tg->parent = parent; | |
8417 | list_add_rcu(&tg->siblings, &parent->children); | |
8418 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8419 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8420 | |
9b5b7751 | 8421 | return tg; |
29f59db3 SV |
8422 | |
8423 | err: | |
6f505b16 | 8424 | free_sched_group(tg); |
29f59db3 SV |
8425 | return ERR_PTR(-ENOMEM); |
8426 | } | |
8427 | ||
9b5b7751 | 8428 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8429 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8430 | { |
29f59db3 | 8431 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8432 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8433 | } |
8434 | ||
9b5b7751 | 8435 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8436 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8437 | { |
8ed36996 | 8438 | unsigned long flags; |
9b5b7751 | 8439 | int i; |
29f59db3 | 8440 | |
8ed36996 | 8441 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8442 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8443 | unregister_fair_sched_group(tg, i); |
8444 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8445 | } |
6f505b16 | 8446 | list_del_rcu(&tg->list); |
f473aa5e | 8447 | list_del_rcu(&tg->siblings); |
8ed36996 | 8448 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8449 | |
9b5b7751 | 8450 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8451 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8452 | } |
8453 | ||
9b5b7751 | 8454 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8455 | * The caller of this function should have put the task in its new group |
8456 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8457 | * reflect its new group. | |
9b5b7751 SV |
8458 | */ |
8459 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8460 | { |
8461 | int on_rq, running; | |
8462 | unsigned long flags; | |
8463 | struct rq *rq; | |
8464 | ||
8465 | rq = task_rq_lock(tsk, &flags); | |
8466 | ||
29f59db3 SV |
8467 | update_rq_clock(rq); |
8468 | ||
051a1d1a | 8469 | running = task_current(rq, tsk); |
29f59db3 SV |
8470 | on_rq = tsk->se.on_rq; |
8471 | ||
0e1f3483 | 8472 | if (on_rq) |
29f59db3 | 8473 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8474 | if (unlikely(running)) |
8475 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8476 | |
6f505b16 | 8477 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8478 | |
810b3817 PZ |
8479 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8480 | if (tsk->sched_class->moved_group) | |
8481 | tsk->sched_class->moved_group(tsk); | |
8482 | #endif | |
8483 | ||
0e1f3483 HS |
8484 | if (unlikely(running)) |
8485 | tsk->sched_class->set_curr_task(rq); | |
8486 | if (on_rq) | |
7074badb | 8487 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8488 | |
29f59db3 SV |
8489 | task_rq_unlock(rq, &flags); |
8490 | } | |
6d6bc0ad | 8491 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8492 | |
052f1dc7 | 8493 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8494 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8495 | { |
8496 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8497 | int on_rq; |
8498 | ||
29f59db3 | 8499 | on_rq = se->on_rq; |
62fb1851 | 8500 | if (on_rq) |
29f59db3 SV |
8501 | dequeue_entity(cfs_rq, se, 0); |
8502 | ||
8503 | se->load.weight = shares; | |
e05510d0 | 8504 | se->load.inv_weight = 0; |
29f59db3 | 8505 | |
62fb1851 | 8506 | if (on_rq) |
29f59db3 | 8507 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8508 | } |
62fb1851 | 8509 | |
c09595f6 PZ |
8510 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8511 | { | |
8512 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8513 | struct rq *rq = cfs_rq->rq; | |
8514 | unsigned long flags; | |
8515 | ||
8516 | spin_lock_irqsave(&rq->lock, flags); | |
8517 | __set_se_shares(se, shares); | |
8518 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8519 | } |
8520 | ||
8ed36996 PZ |
8521 | static DEFINE_MUTEX(shares_mutex); |
8522 | ||
4cf86d77 | 8523 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8524 | { |
8525 | int i; | |
8ed36996 | 8526 | unsigned long flags; |
c61935fd | 8527 | |
ec7dc8ac DG |
8528 | /* |
8529 | * We can't change the weight of the root cgroup. | |
8530 | */ | |
8531 | if (!tg->se[0]) | |
8532 | return -EINVAL; | |
8533 | ||
18d95a28 PZ |
8534 | if (shares < MIN_SHARES) |
8535 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8536 | else if (shares > MAX_SHARES) |
8537 | shares = MAX_SHARES; | |
62fb1851 | 8538 | |
8ed36996 | 8539 | mutex_lock(&shares_mutex); |
9b5b7751 | 8540 | if (tg->shares == shares) |
5cb350ba | 8541 | goto done; |
29f59db3 | 8542 | |
8ed36996 | 8543 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8544 | for_each_possible_cpu(i) |
8545 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8546 | list_del_rcu(&tg->siblings); |
8ed36996 | 8547 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8548 | |
8549 | /* wait for any ongoing reference to this group to finish */ | |
8550 | synchronize_sched(); | |
8551 | ||
8552 | /* | |
8553 | * Now we are free to modify the group's share on each cpu | |
8554 | * w/o tripping rebalance_share or load_balance_fair. | |
8555 | */ | |
9b5b7751 | 8556 | tg->shares = shares; |
c09595f6 PZ |
8557 | for_each_possible_cpu(i) { |
8558 | /* | |
8559 | * force a rebalance | |
8560 | */ | |
8561 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8562 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8563 | } |
29f59db3 | 8564 | |
6b2d7700 SV |
8565 | /* |
8566 | * Enable load balance activity on this group, by inserting it back on | |
8567 | * each cpu's rq->leaf_cfs_rq_list. | |
8568 | */ | |
8ed36996 | 8569 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8570 | for_each_possible_cpu(i) |
8571 | register_fair_sched_group(tg, i); | |
f473aa5e | 8572 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8573 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8574 | done: |
8ed36996 | 8575 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8576 | return 0; |
29f59db3 SV |
8577 | } |
8578 | ||
5cb350ba DG |
8579 | unsigned long sched_group_shares(struct task_group *tg) |
8580 | { | |
8581 | return tg->shares; | |
8582 | } | |
052f1dc7 | 8583 | #endif |
5cb350ba | 8584 | |
052f1dc7 | 8585 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8586 | /* |
9f0c1e56 | 8587 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8588 | */ |
9f0c1e56 PZ |
8589 | static DEFINE_MUTEX(rt_constraints_mutex); |
8590 | ||
8591 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8592 | { | |
8593 | if (runtime == RUNTIME_INF) | |
8594 | return 1ULL << 16; | |
8595 | ||
6f6d6a1a | 8596 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8597 | } |
8598 | ||
b40b2e8e PZ |
8599 | #ifdef CONFIG_CGROUP_SCHED |
8600 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8601 | { | |
10b612f4 | 8602 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8603 | unsigned long total = 0; |
8604 | ||
8605 | if (!parent) { | |
8606 | if (global_rt_period() < period) | |
8607 | return 0; | |
8608 | ||
8609 | return to_ratio(period, runtime) < | |
8610 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8611 | } | |
8612 | ||
8613 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8614 | return 0; | |
8615 | ||
8616 | rcu_read_lock(); | |
8617 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8618 | if (tgi == tg) | |
8619 | continue; | |
8620 | ||
8621 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8622 | tgi->rt_bandwidth.rt_runtime); | |
8623 | } | |
8624 | rcu_read_unlock(); | |
8625 | ||
10b612f4 | 8626 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8627 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8628 | parent->rt_bandwidth.rt_runtime); | |
8629 | } | |
8630 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8631 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8632 | { |
8633 | struct task_group *tgi; | |
8634 | unsigned long total = 0; | |
9f0c1e56 | 8635 | unsigned long global_ratio = |
d0b27fa7 | 8636 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8637 | |
8638 | rcu_read_lock(); | |
9f0c1e56 PZ |
8639 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8640 | if (tgi == tg) | |
8641 | continue; | |
6f505b16 | 8642 | |
d0b27fa7 PZ |
8643 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8644 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8645 | } |
8646 | rcu_read_unlock(); | |
6f505b16 | 8647 | |
9f0c1e56 | 8648 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8649 | } |
b40b2e8e | 8650 | #endif |
6f505b16 | 8651 | |
521f1a24 DG |
8652 | /* Must be called with tasklist_lock held */ |
8653 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8654 | { | |
8655 | struct task_struct *g, *p; | |
8656 | do_each_thread(g, p) { | |
8657 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8658 | return 1; | |
8659 | } while_each_thread(g, p); | |
8660 | return 0; | |
8661 | } | |
8662 | ||
d0b27fa7 PZ |
8663 | static int tg_set_bandwidth(struct task_group *tg, |
8664 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8665 | { |
ac086bc2 | 8666 | int i, err = 0; |
9f0c1e56 | 8667 | |
9f0c1e56 | 8668 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8669 | read_lock(&tasklist_lock); |
ac086bc2 | 8670 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8671 | err = -EBUSY; |
8672 | goto unlock; | |
8673 | } | |
9f0c1e56 PZ |
8674 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8675 | err = -EINVAL; | |
8676 | goto unlock; | |
8677 | } | |
ac086bc2 PZ |
8678 | |
8679 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8680 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8681 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8682 | |
8683 | for_each_possible_cpu(i) { | |
8684 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8685 | ||
8686 | spin_lock(&rt_rq->rt_runtime_lock); | |
8687 | rt_rq->rt_runtime = rt_runtime; | |
8688 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8689 | } | |
8690 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8691 | unlock: |
521f1a24 | 8692 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8693 | mutex_unlock(&rt_constraints_mutex); |
8694 | ||
8695 | return err; | |
6f505b16 PZ |
8696 | } |
8697 | ||
d0b27fa7 PZ |
8698 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8699 | { | |
8700 | u64 rt_runtime, rt_period; | |
8701 | ||
8702 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8703 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8704 | if (rt_runtime_us < 0) | |
8705 | rt_runtime = RUNTIME_INF; | |
8706 | ||
8707 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8708 | } | |
8709 | ||
9f0c1e56 PZ |
8710 | long sched_group_rt_runtime(struct task_group *tg) |
8711 | { | |
8712 | u64 rt_runtime_us; | |
8713 | ||
d0b27fa7 | 8714 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8715 | return -1; |
8716 | ||
d0b27fa7 | 8717 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8718 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8719 | return rt_runtime_us; | |
8720 | } | |
d0b27fa7 PZ |
8721 | |
8722 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8723 | { | |
8724 | u64 rt_runtime, rt_period; | |
8725 | ||
8726 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8727 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8728 | ||
8729 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8730 | } | |
8731 | ||
8732 | long sched_group_rt_period(struct task_group *tg) | |
8733 | { | |
8734 | u64 rt_period_us; | |
8735 | ||
8736 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8737 | do_div(rt_period_us, NSEC_PER_USEC); | |
8738 | return rt_period_us; | |
8739 | } | |
8740 | ||
8741 | static int sched_rt_global_constraints(void) | |
8742 | { | |
10b612f4 PZ |
8743 | struct task_group *tg = &root_task_group; |
8744 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8745 | int ret = 0; |
8746 | ||
10b612f4 PZ |
8747 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8748 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8749 | ||
d0b27fa7 | 8750 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8751 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8752 | ret = -EINVAL; |
8753 | mutex_unlock(&rt_constraints_mutex); | |
8754 | ||
8755 | return ret; | |
8756 | } | |
6d6bc0ad | 8757 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8758 | static int sched_rt_global_constraints(void) |
8759 | { | |
ac086bc2 PZ |
8760 | unsigned long flags; |
8761 | int i; | |
8762 | ||
8763 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8764 | for_each_possible_cpu(i) { | |
8765 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8766 | ||
8767 | spin_lock(&rt_rq->rt_runtime_lock); | |
8768 | rt_rq->rt_runtime = global_rt_runtime(); | |
8769 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8770 | } | |
8771 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8772 | ||
d0b27fa7 PZ |
8773 | return 0; |
8774 | } | |
6d6bc0ad | 8775 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8776 | |
8777 | int sched_rt_handler(struct ctl_table *table, int write, | |
8778 | struct file *filp, void __user *buffer, size_t *lenp, | |
8779 | loff_t *ppos) | |
8780 | { | |
8781 | int ret; | |
8782 | int old_period, old_runtime; | |
8783 | static DEFINE_MUTEX(mutex); | |
8784 | ||
8785 | mutex_lock(&mutex); | |
8786 | old_period = sysctl_sched_rt_period; | |
8787 | old_runtime = sysctl_sched_rt_runtime; | |
8788 | ||
8789 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8790 | ||
8791 | if (!ret && write) { | |
8792 | ret = sched_rt_global_constraints(); | |
8793 | if (ret) { | |
8794 | sysctl_sched_rt_period = old_period; | |
8795 | sysctl_sched_rt_runtime = old_runtime; | |
8796 | } else { | |
8797 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8798 | def_rt_bandwidth.rt_period = | |
8799 | ns_to_ktime(global_rt_period()); | |
8800 | } | |
8801 | } | |
8802 | mutex_unlock(&mutex); | |
8803 | ||
8804 | return ret; | |
8805 | } | |
68318b8e | 8806 | |
052f1dc7 | 8807 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8808 | |
8809 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8810 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8811 | { |
2b01dfe3 PM |
8812 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8813 | struct task_group, css); | |
68318b8e SV |
8814 | } |
8815 | ||
8816 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8817 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8818 | { |
ec7dc8ac | 8819 | struct task_group *tg, *parent; |
68318b8e | 8820 | |
2b01dfe3 | 8821 | if (!cgrp->parent) { |
68318b8e | 8822 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8823 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8824 | return &init_task_group.css; |
8825 | } | |
8826 | ||
ec7dc8ac DG |
8827 | parent = cgroup_tg(cgrp->parent); |
8828 | tg = sched_create_group(parent); | |
68318b8e SV |
8829 | if (IS_ERR(tg)) |
8830 | return ERR_PTR(-ENOMEM); | |
8831 | ||
8832 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8833 | tg->css.cgroup = cgrp; |
68318b8e SV |
8834 | |
8835 | return &tg->css; | |
8836 | } | |
8837 | ||
41a2d6cf IM |
8838 | static void |
8839 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8840 | { |
2b01dfe3 | 8841 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8842 | |
8843 | sched_destroy_group(tg); | |
8844 | } | |
8845 | ||
41a2d6cf IM |
8846 | static int |
8847 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8848 | struct task_struct *tsk) | |
68318b8e | 8849 | { |
b68aa230 PZ |
8850 | #ifdef CONFIG_RT_GROUP_SCHED |
8851 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8852 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8853 | return -EINVAL; |
8854 | #else | |
68318b8e SV |
8855 | /* We don't support RT-tasks being in separate groups */ |
8856 | if (tsk->sched_class != &fair_sched_class) | |
8857 | return -EINVAL; | |
b68aa230 | 8858 | #endif |
68318b8e SV |
8859 | |
8860 | return 0; | |
8861 | } | |
8862 | ||
8863 | static void | |
2b01dfe3 | 8864 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8865 | struct cgroup *old_cont, struct task_struct *tsk) |
8866 | { | |
8867 | sched_move_task(tsk); | |
8868 | } | |
8869 | ||
052f1dc7 | 8870 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8871 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8872 | u64 shareval) |
68318b8e | 8873 | { |
2b01dfe3 | 8874 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8875 | } |
8876 | ||
f4c753b7 | 8877 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8878 | { |
2b01dfe3 | 8879 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8880 | |
8881 | return (u64) tg->shares; | |
8882 | } | |
6d6bc0ad | 8883 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8884 | |
052f1dc7 | 8885 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8886 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8887 | s64 val) |
6f505b16 | 8888 | { |
06ecb27c | 8889 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8890 | } |
8891 | ||
06ecb27c | 8892 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8893 | { |
06ecb27c | 8894 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8895 | } |
d0b27fa7 PZ |
8896 | |
8897 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8898 | u64 rt_period_us) | |
8899 | { | |
8900 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8901 | } | |
8902 | ||
8903 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8904 | { | |
8905 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8906 | } | |
6d6bc0ad | 8907 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8908 | |
fe5c7cc2 | 8909 | static struct cftype cpu_files[] = { |
052f1dc7 | 8910 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8911 | { |
8912 | .name = "shares", | |
f4c753b7 PM |
8913 | .read_u64 = cpu_shares_read_u64, |
8914 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8915 | }, |
052f1dc7 PZ |
8916 | #endif |
8917 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8918 | { |
9f0c1e56 | 8919 | .name = "rt_runtime_us", |
06ecb27c PM |
8920 | .read_s64 = cpu_rt_runtime_read, |
8921 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8922 | }, |
d0b27fa7 PZ |
8923 | { |
8924 | .name = "rt_period_us", | |
f4c753b7 PM |
8925 | .read_u64 = cpu_rt_period_read_uint, |
8926 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8927 | }, |
052f1dc7 | 8928 | #endif |
68318b8e SV |
8929 | }; |
8930 | ||
8931 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8932 | { | |
fe5c7cc2 | 8933 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8934 | } |
8935 | ||
8936 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8937 | .name = "cpu", |
8938 | .create = cpu_cgroup_create, | |
8939 | .destroy = cpu_cgroup_destroy, | |
8940 | .can_attach = cpu_cgroup_can_attach, | |
8941 | .attach = cpu_cgroup_attach, | |
8942 | .populate = cpu_cgroup_populate, | |
8943 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8944 | .early_init = 1, |
8945 | }; | |
8946 | ||
052f1dc7 | 8947 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8948 | |
8949 | #ifdef CONFIG_CGROUP_CPUACCT | |
8950 | ||
8951 | /* | |
8952 | * CPU accounting code for task groups. | |
8953 | * | |
8954 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8955 | * (balbir@in.ibm.com). | |
8956 | */ | |
8957 | ||
8958 | /* track cpu usage of a group of tasks */ | |
8959 | struct cpuacct { | |
8960 | struct cgroup_subsys_state css; | |
8961 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8962 | u64 *cpuusage; | |
8963 | }; | |
8964 | ||
8965 | struct cgroup_subsys cpuacct_subsys; | |
8966 | ||
8967 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8968 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8969 | { |
32cd756a | 8970 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8971 | struct cpuacct, css); |
8972 | } | |
8973 | ||
8974 | /* return cpu accounting group to which this task belongs */ | |
8975 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8976 | { | |
8977 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8978 | struct cpuacct, css); | |
8979 | } | |
8980 | ||
8981 | /* create a new cpu accounting group */ | |
8982 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8983 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8984 | { |
8985 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8986 | ||
8987 | if (!ca) | |
8988 | return ERR_PTR(-ENOMEM); | |
8989 | ||
8990 | ca->cpuusage = alloc_percpu(u64); | |
8991 | if (!ca->cpuusage) { | |
8992 | kfree(ca); | |
8993 | return ERR_PTR(-ENOMEM); | |
8994 | } | |
8995 | ||
8996 | return &ca->css; | |
8997 | } | |
8998 | ||
8999 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9000 | static void |
32cd756a | 9001 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9002 | { |
32cd756a | 9003 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9004 | |
9005 | free_percpu(ca->cpuusage); | |
9006 | kfree(ca); | |
9007 | } | |
9008 | ||
9009 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9010 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9011 | { |
32cd756a | 9012 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9013 | u64 totalcpuusage = 0; |
9014 | int i; | |
9015 | ||
9016 | for_each_possible_cpu(i) { | |
9017 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9018 | ||
9019 | /* | |
9020 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9021 | * platforms. | |
9022 | */ | |
9023 | spin_lock_irq(&cpu_rq(i)->lock); | |
9024 | totalcpuusage += *cpuusage; | |
9025 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9026 | } | |
9027 | ||
9028 | return totalcpuusage; | |
9029 | } | |
9030 | ||
0297b803 DG |
9031 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9032 | u64 reset) | |
9033 | { | |
9034 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9035 | int err = 0; | |
9036 | int i; | |
9037 | ||
9038 | if (reset) { | |
9039 | err = -EINVAL; | |
9040 | goto out; | |
9041 | } | |
9042 | ||
9043 | for_each_possible_cpu(i) { | |
9044 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9045 | ||
9046 | spin_lock_irq(&cpu_rq(i)->lock); | |
9047 | *cpuusage = 0; | |
9048 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9049 | } | |
9050 | out: | |
9051 | return err; | |
9052 | } | |
9053 | ||
d842de87 SV |
9054 | static struct cftype files[] = { |
9055 | { | |
9056 | .name = "usage", | |
f4c753b7 PM |
9057 | .read_u64 = cpuusage_read, |
9058 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9059 | }, |
9060 | }; | |
9061 | ||
32cd756a | 9062 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9063 | { |
32cd756a | 9064 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9065 | } |
9066 | ||
9067 | /* | |
9068 | * charge this task's execution time to its accounting group. | |
9069 | * | |
9070 | * called with rq->lock held. | |
9071 | */ | |
9072 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9073 | { | |
9074 | struct cpuacct *ca; | |
9075 | ||
9076 | if (!cpuacct_subsys.active) | |
9077 | return; | |
9078 | ||
9079 | ca = task_ca(tsk); | |
9080 | if (ca) { | |
9081 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9082 | ||
9083 | *cpuusage += cputime; | |
9084 | } | |
9085 | } | |
9086 | ||
9087 | struct cgroup_subsys cpuacct_subsys = { | |
9088 | .name = "cpuacct", | |
9089 | .create = cpuacct_create, | |
9090 | .destroy = cpuacct_destroy, | |
9091 | .populate = cpuacct_populate, | |
9092 | .subsys_id = cpuacct_subsys_id, | |
9093 | }; | |
9094 | #endif /* CONFIG_CGROUP_CPUACCT */ |