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) { } |
29f59db3 | 368 | |
052f1dc7 | 369 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 370 | |
6aa645ea IM |
371 | /* CFS-related fields in a runqueue */ |
372 | struct cfs_rq { | |
373 | struct load_weight load; | |
374 | unsigned long nr_running; | |
375 | ||
6aa645ea | 376 | u64 exec_clock; |
e9acbff6 | 377 | u64 min_vruntime; |
103638d9 | 378 | u64 pair_start; |
6aa645ea IM |
379 | |
380 | struct rb_root tasks_timeline; | |
381 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
382 | |
383 | struct list_head tasks; | |
384 | struct list_head *balance_iterator; | |
385 | ||
386 | /* | |
387 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
388 | * It is set to NULL otherwise (i.e when none are currently running). |
389 | */ | |
aa2ac252 | 390 | struct sched_entity *curr, *next; |
ddc97297 PZ |
391 | |
392 | unsigned long nr_spread_over; | |
393 | ||
62160e3f | 394 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
395 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
396 | ||
41a2d6cf IM |
397 | /* |
398 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
399 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
400 | * (like users, containers etc.) | |
401 | * | |
402 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
403 | * list is used during load balance. | |
404 | */ | |
41a2d6cf IM |
405 | struct list_head leaf_cfs_rq_list; |
406 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
407 | |
408 | #ifdef CONFIG_SMP | |
c09595f6 | 409 | /* |
c8cba857 | 410 | * the part of load.weight contributed by tasks |
c09595f6 | 411 | */ |
c8cba857 | 412 | unsigned long task_weight; |
c09595f6 | 413 | |
c8cba857 PZ |
414 | /* |
415 | * h_load = weight * f(tg) | |
416 | * | |
417 | * Where f(tg) is the recursive weight fraction assigned to | |
418 | * this group. | |
419 | */ | |
420 | unsigned long h_load; | |
c09595f6 | 421 | |
c8cba857 PZ |
422 | /* |
423 | * this cpu's part of tg->shares | |
424 | */ | |
425 | unsigned long shares; | |
c09595f6 | 426 | #endif |
6aa645ea IM |
427 | #endif |
428 | }; | |
1da177e4 | 429 | |
6aa645ea IM |
430 | /* Real-Time classes' related field in a runqueue: */ |
431 | struct rt_rq { | |
432 | struct rt_prio_array active; | |
63489e45 | 433 | unsigned long rt_nr_running; |
052f1dc7 | 434 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
435 | int highest_prio; /* highest queued rt task prio */ |
436 | #endif | |
fa85ae24 | 437 | #ifdef CONFIG_SMP |
73fe6aae | 438 | unsigned long rt_nr_migratory; |
a22d7fc1 | 439 | int overloaded; |
fa85ae24 | 440 | #endif |
6f505b16 | 441 | int rt_throttled; |
fa85ae24 | 442 | u64 rt_time; |
ac086bc2 | 443 | u64 rt_runtime; |
ea736ed5 | 444 | /* Nests inside the rq lock: */ |
ac086bc2 | 445 | spinlock_t rt_runtime_lock; |
6f505b16 | 446 | |
052f1dc7 | 447 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
448 | unsigned long rt_nr_boosted; |
449 | ||
6f505b16 PZ |
450 | struct rq *rq; |
451 | struct list_head leaf_rt_rq_list; | |
452 | struct task_group *tg; | |
453 | struct sched_rt_entity *rt_se; | |
454 | #endif | |
6aa645ea IM |
455 | }; |
456 | ||
57d885fe GH |
457 | #ifdef CONFIG_SMP |
458 | ||
459 | /* | |
460 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
461 | * variables. Each exclusive cpuset essentially defines an island domain by |
462 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
463 | * exclusive cpuset is created, we also create and attach a new root-domain |
464 | * object. | |
465 | * | |
57d885fe GH |
466 | */ |
467 | struct root_domain { | |
468 | atomic_t refcount; | |
469 | cpumask_t span; | |
470 | cpumask_t online; | |
637f5085 | 471 | |
0eab9146 | 472 | /* |
637f5085 GH |
473 | * The "RT overload" flag: it gets set if a CPU has more than |
474 | * one runnable RT task. | |
475 | */ | |
476 | cpumask_t rto_mask; | |
0eab9146 | 477 | atomic_t rto_count; |
6e0534f2 GH |
478 | #ifdef CONFIG_SMP |
479 | struct cpupri cpupri; | |
480 | #endif | |
57d885fe GH |
481 | }; |
482 | ||
dc938520 GH |
483 | /* |
484 | * By default the system creates a single root-domain with all cpus as | |
485 | * members (mimicking the global state we have today). | |
486 | */ | |
57d885fe GH |
487 | static struct root_domain def_root_domain; |
488 | ||
489 | #endif | |
490 | ||
1da177e4 LT |
491 | /* |
492 | * This is the main, per-CPU runqueue data structure. | |
493 | * | |
494 | * Locking rule: those places that want to lock multiple runqueues | |
495 | * (such as the load balancing or the thread migration code), lock | |
496 | * acquire operations must be ordered by ascending &runqueue. | |
497 | */ | |
70b97a7f | 498 | struct rq { |
d8016491 IM |
499 | /* runqueue lock: */ |
500 | spinlock_t lock; | |
1da177e4 LT |
501 | |
502 | /* | |
503 | * nr_running and cpu_load should be in the same cacheline because | |
504 | * remote CPUs use both these fields when doing load calculation. | |
505 | */ | |
506 | unsigned long nr_running; | |
6aa645ea IM |
507 | #define CPU_LOAD_IDX_MAX 5 |
508 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 509 | unsigned char idle_at_tick; |
46cb4b7c | 510 | #ifdef CONFIG_NO_HZ |
15934a37 | 511 | unsigned long last_tick_seen; |
46cb4b7c SS |
512 | unsigned char in_nohz_recently; |
513 | #endif | |
d8016491 IM |
514 | /* capture load from *all* tasks on this cpu: */ |
515 | struct load_weight load; | |
6aa645ea IM |
516 | unsigned long nr_load_updates; |
517 | u64 nr_switches; | |
518 | ||
519 | struct cfs_rq cfs; | |
6f505b16 | 520 | struct rt_rq rt; |
6f505b16 | 521 | |
6aa645ea | 522 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
523 | /* list of leaf cfs_rq on this cpu: */ |
524 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
525 | #endif |
526 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 527 | struct list_head leaf_rt_rq_list; |
1da177e4 | 528 | #endif |
1da177e4 LT |
529 | |
530 | /* | |
531 | * This is part of a global counter where only the total sum | |
532 | * over all CPUs matters. A task can increase this counter on | |
533 | * one CPU and if it got migrated afterwards it may decrease | |
534 | * it on another CPU. Always updated under the runqueue lock: | |
535 | */ | |
536 | unsigned long nr_uninterruptible; | |
537 | ||
36c8b586 | 538 | struct task_struct *curr, *idle; |
c9819f45 | 539 | unsigned long next_balance; |
1da177e4 | 540 | struct mm_struct *prev_mm; |
6aa645ea | 541 | |
3e51f33f | 542 | u64 clock; |
6aa645ea | 543 | |
1da177e4 LT |
544 | atomic_t nr_iowait; |
545 | ||
546 | #ifdef CONFIG_SMP | |
0eab9146 | 547 | struct root_domain *rd; |
1da177e4 LT |
548 | struct sched_domain *sd; |
549 | ||
550 | /* For active balancing */ | |
551 | int active_balance; | |
552 | int push_cpu; | |
d8016491 IM |
553 | /* cpu of this runqueue: */ |
554 | int cpu; | |
1f11eb6a | 555 | int online; |
1da177e4 | 556 | |
a8a51d5e PZ |
557 | unsigned long avg_load_per_task; |
558 | ||
36c8b586 | 559 | struct task_struct *migration_thread; |
1da177e4 LT |
560 | struct list_head migration_queue; |
561 | #endif | |
562 | ||
8f4d37ec PZ |
563 | #ifdef CONFIG_SCHED_HRTICK |
564 | unsigned long hrtick_flags; | |
565 | ktime_t hrtick_expire; | |
566 | struct hrtimer hrtick_timer; | |
567 | #endif | |
568 | ||
1da177e4 LT |
569 | #ifdef CONFIG_SCHEDSTATS |
570 | /* latency stats */ | |
571 | struct sched_info rq_sched_info; | |
572 | ||
573 | /* sys_sched_yield() stats */ | |
480b9434 KC |
574 | unsigned int yld_exp_empty; |
575 | unsigned int yld_act_empty; | |
576 | unsigned int yld_both_empty; | |
577 | unsigned int yld_count; | |
1da177e4 LT |
578 | |
579 | /* schedule() stats */ | |
480b9434 KC |
580 | unsigned int sched_switch; |
581 | unsigned int sched_count; | |
582 | unsigned int sched_goidle; | |
1da177e4 LT |
583 | |
584 | /* try_to_wake_up() stats */ | |
480b9434 KC |
585 | unsigned int ttwu_count; |
586 | unsigned int ttwu_local; | |
b8efb561 IM |
587 | |
588 | /* BKL stats */ | |
480b9434 | 589 | unsigned int bkl_count; |
1da177e4 | 590 | #endif |
fcb99371 | 591 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
592 | }; |
593 | ||
f34e3b61 | 594 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 595 | |
dd41f596 IM |
596 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
597 | { | |
598 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
599 | } | |
600 | ||
0a2966b4 CL |
601 | static inline int cpu_of(struct rq *rq) |
602 | { | |
603 | #ifdef CONFIG_SMP | |
604 | return rq->cpu; | |
605 | #else | |
606 | return 0; | |
607 | #endif | |
608 | } | |
609 | ||
674311d5 NP |
610 | /* |
611 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 612 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
613 | * |
614 | * The domain tree of any CPU may only be accessed from within | |
615 | * preempt-disabled sections. | |
616 | */ | |
48f24c4d IM |
617 | #define for_each_domain(cpu, __sd) \ |
618 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
619 | |
620 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
621 | #define this_rq() (&__get_cpu_var(runqueues)) | |
622 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
623 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
624 | ||
3e51f33f PZ |
625 | static inline void update_rq_clock(struct rq *rq) |
626 | { | |
627 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
628 | } | |
629 | ||
bf5c91ba IM |
630 | /* |
631 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
632 | */ | |
633 | #ifdef CONFIG_SCHED_DEBUG | |
634 | # define const_debug __read_mostly | |
635 | #else | |
636 | # define const_debug static const | |
637 | #endif | |
638 | ||
639 | /* | |
640 | * Debugging: various feature bits | |
641 | */ | |
f00b45c1 PZ |
642 | |
643 | #define SCHED_FEAT(name, enabled) \ | |
644 | __SCHED_FEAT_##name , | |
645 | ||
bf5c91ba | 646 | enum { |
f00b45c1 | 647 | #include "sched_features.h" |
bf5c91ba IM |
648 | }; |
649 | ||
f00b45c1 PZ |
650 | #undef SCHED_FEAT |
651 | ||
652 | #define SCHED_FEAT(name, enabled) \ | |
653 | (1UL << __SCHED_FEAT_##name) * enabled | | |
654 | ||
bf5c91ba | 655 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
656 | #include "sched_features.h" |
657 | 0; | |
658 | ||
659 | #undef SCHED_FEAT | |
660 | ||
661 | #ifdef CONFIG_SCHED_DEBUG | |
662 | #define SCHED_FEAT(name, enabled) \ | |
663 | #name , | |
664 | ||
983ed7a6 | 665 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
666 | #include "sched_features.h" |
667 | NULL | |
668 | }; | |
669 | ||
670 | #undef SCHED_FEAT | |
671 | ||
983ed7a6 | 672 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
673 | { |
674 | filp->private_data = inode->i_private; | |
675 | return 0; | |
676 | } | |
677 | ||
678 | static ssize_t | |
679 | sched_feat_read(struct file *filp, char __user *ubuf, | |
680 | size_t cnt, loff_t *ppos) | |
681 | { | |
682 | char *buf; | |
683 | int r = 0; | |
684 | int len = 0; | |
685 | int i; | |
686 | ||
687 | for (i = 0; sched_feat_names[i]; i++) { | |
688 | len += strlen(sched_feat_names[i]); | |
689 | len += 4; | |
690 | } | |
691 | ||
692 | buf = kmalloc(len + 2, GFP_KERNEL); | |
693 | if (!buf) | |
694 | return -ENOMEM; | |
695 | ||
696 | for (i = 0; sched_feat_names[i]; i++) { | |
697 | if (sysctl_sched_features & (1UL << i)) | |
698 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
699 | else | |
c24b7c52 | 700 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
701 | } |
702 | ||
703 | r += sprintf(buf + r, "\n"); | |
704 | WARN_ON(r >= len + 2); | |
705 | ||
706 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
707 | ||
708 | kfree(buf); | |
709 | ||
710 | return r; | |
711 | } | |
712 | ||
713 | static ssize_t | |
714 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
715 | size_t cnt, loff_t *ppos) | |
716 | { | |
717 | char buf[64]; | |
718 | char *cmp = buf; | |
719 | int neg = 0; | |
720 | int i; | |
721 | ||
722 | if (cnt > 63) | |
723 | cnt = 63; | |
724 | ||
725 | if (copy_from_user(&buf, ubuf, cnt)) | |
726 | return -EFAULT; | |
727 | ||
728 | buf[cnt] = 0; | |
729 | ||
c24b7c52 | 730 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
731 | neg = 1; |
732 | cmp += 3; | |
733 | } | |
734 | ||
735 | for (i = 0; sched_feat_names[i]; i++) { | |
736 | int len = strlen(sched_feat_names[i]); | |
737 | ||
738 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
739 | if (neg) | |
740 | sysctl_sched_features &= ~(1UL << i); | |
741 | else | |
742 | sysctl_sched_features |= (1UL << i); | |
743 | break; | |
744 | } | |
745 | } | |
746 | ||
747 | if (!sched_feat_names[i]) | |
748 | return -EINVAL; | |
749 | ||
750 | filp->f_pos += cnt; | |
751 | ||
752 | return cnt; | |
753 | } | |
754 | ||
755 | static struct file_operations sched_feat_fops = { | |
756 | .open = sched_feat_open, | |
757 | .read = sched_feat_read, | |
758 | .write = sched_feat_write, | |
759 | }; | |
760 | ||
761 | static __init int sched_init_debug(void) | |
762 | { | |
f00b45c1 PZ |
763 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
764 | &sched_feat_fops); | |
765 | ||
766 | return 0; | |
767 | } | |
768 | late_initcall(sched_init_debug); | |
769 | ||
770 | #endif | |
771 | ||
772 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 773 | |
b82d9fdd PZ |
774 | /* |
775 | * Number of tasks to iterate in a single balance run. | |
776 | * Limited because this is done with IRQs disabled. | |
777 | */ | |
778 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
779 | ||
2398f2c6 PZ |
780 | /* |
781 | * ratelimit for updating the group shares. | |
782 | * default: 0.5ms | |
783 | */ | |
784 | const_debug unsigned int sysctl_sched_shares_ratelimit = 500000; | |
785 | ||
fa85ae24 | 786 | /* |
9f0c1e56 | 787 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
788 | * default: 1s |
789 | */ | |
9f0c1e56 | 790 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 791 | |
6892b75e IM |
792 | static __read_mostly int scheduler_running; |
793 | ||
9f0c1e56 PZ |
794 | /* |
795 | * part of the period that we allow rt tasks to run in us. | |
796 | * default: 0.95s | |
797 | */ | |
798 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 799 | |
d0b27fa7 PZ |
800 | static inline u64 global_rt_period(void) |
801 | { | |
802 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
803 | } | |
804 | ||
805 | static inline u64 global_rt_runtime(void) | |
806 | { | |
807 | if (sysctl_sched_rt_period < 0) | |
808 | return RUNTIME_INF; | |
809 | ||
810 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
811 | } | |
fa85ae24 | 812 | |
1da177e4 | 813 | #ifndef prepare_arch_switch |
4866cde0 NP |
814 | # define prepare_arch_switch(next) do { } while (0) |
815 | #endif | |
816 | #ifndef finish_arch_switch | |
817 | # define finish_arch_switch(prev) do { } while (0) | |
818 | #endif | |
819 | ||
051a1d1a DA |
820 | static inline int task_current(struct rq *rq, struct task_struct *p) |
821 | { | |
822 | return rq->curr == p; | |
823 | } | |
824 | ||
4866cde0 | 825 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 826 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 827 | { |
051a1d1a | 828 | return task_current(rq, p); |
4866cde0 NP |
829 | } |
830 | ||
70b97a7f | 831 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
832 | { |
833 | } | |
834 | ||
70b97a7f | 835 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 836 | { |
da04c035 IM |
837 | #ifdef CONFIG_DEBUG_SPINLOCK |
838 | /* this is a valid case when another task releases the spinlock */ | |
839 | rq->lock.owner = current; | |
840 | #endif | |
8a25d5de IM |
841 | /* |
842 | * If we are tracking spinlock dependencies then we have to | |
843 | * fix up the runqueue lock - which gets 'carried over' from | |
844 | * prev into current: | |
845 | */ | |
846 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
847 | ||
4866cde0 NP |
848 | spin_unlock_irq(&rq->lock); |
849 | } | |
850 | ||
851 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 852 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
853 | { |
854 | #ifdef CONFIG_SMP | |
855 | return p->oncpu; | |
856 | #else | |
051a1d1a | 857 | return task_current(rq, p); |
4866cde0 NP |
858 | #endif |
859 | } | |
860 | ||
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
862 | { |
863 | #ifdef CONFIG_SMP | |
864 | /* | |
865 | * We can optimise this out completely for !SMP, because the | |
866 | * SMP rebalancing from interrupt is the only thing that cares | |
867 | * here. | |
868 | */ | |
869 | next->oncpu = 1; | |
870 | #endif | |
871 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
872 | spin_unlock_irq(&rq->lock); | |
873 | #else | |
874 | spin_unlock(&rq->lock); | |
875 | #endif | |
876 | } | |
877 | ||
70b97a7f | 878 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
879 | { |
880 | #ifdef CONFIG_SMP | |
881 | /* | |
882 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
883 | * We must ensure this doesn't happen until the switch is completely | |
884 | * finished. | |
885 | */ | |
886 | smp_wmb(); | |
887 | prev->oncpu = 0; | |
888 | #endif | |
889 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
890 | local_irq_enable(); | |
1da177e4 | 891 | #endif |
4866cde0 NP |
892 | } |
893 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 894 | |
b29739f9 IM |
895 | /* |
896 | * __task_rq_lock - lock the runqueue a given task resides on. | |
897 | * Must be called interrupts disabled. | |
898 | */ | |
70b97a7f | 899 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
900 | __acquires(rq->lock) |
901 | { | |
3a5c359a AK |
902 | for (;;) { |
903 | struct rq *rq = task_rq(p); | |
904 | spin_lock(&rq->lock); | |
905 | if (likely(rq == task_rq(p))) | |
906 | return rq; | |
b29739f9 | 907 | spin_unlock(&rq->lock); |
b29739f9 | 908 | } |
b29739f9 IM |
909 | } |
910 | ||
1da177e4 LT |
911 | /* |
912 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 913 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
914 | * explicitly disabling preemption. |
915 | */ | |
70b97a7f | 916 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
917 | __acquires(rq->lock) |
918 | { | |
70b97a7f | 919 | struct rq *rq; |
1da177e4 | 920 | |
3a5c359a AK |
921 | for (;;) { |
922 | local_irq_save(*flags); | |
923 | rq = task_rq(p); | |
924 | spin_lock(&rq->lock); | |
925 | if (likely(rq == task_rq(p))) | |
926 | return rq; | |
1da177e4 | 927 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 928 | } |
1da177e4 LT |
929 | } |
930 | ||
a9957449 | 931 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
932 | __releases(rq->lock) |
933 | { | |
934 | spin_unlock(&rq->lock); | |
935 | } | |
936 | ||
70b97a7f | 937 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
938 | __releases(rq->lock) |
939 | { | |
940 | spin_unlock_irqrestore(&rq->lock, *flags); | |
941 | } | |
942 | ||
1da177e4 | 943 | /* |
cc2a73b5 | 944 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 945 | */ |
a9957449 | 946 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
947 | __acquires(rq->lock) |
948 | { | |
70b97a7f | 949 | struct rq *rq; |
1da177e4 LT |
950 | |
951 | local_irq_disable(); | |
952 | rq = this_rq(); | |
953 | spin_lock(&rq->lock); | |
954 | ||
955 | return rq; | |
956 | } | |
957 | ||
8f4d37ec PZ |
958 | static void __resched_task(struct task_struct *p, int tif_bit); |
959 | ||
960 | static inline void resched_task(struct task_struct *p) | |
961 | { | |
962 | __resched_task(p, TIF_NEED_RESCHED); | |
963 | } | |
964 | ||
965 | #ifdef CONFIG_SCHED_HRTICK | |
966 | /* | |
967 | * Use HR-timers to deliver accurate preemption points. | |
968 | * | |
969 | * Its all a bit involved since we cannot program an hrt while holding the | |
970 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
971 | * reschedule event. | |
972 | * | |
973 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
974 | * rq->lock. | |
975 | */ | |
976 | static inline void resched_hrt(struct task_struct *p) | |
977 | { | |
978 | __resched_task(p, TIF_HRTICK_RESCHED); | |
979 | } | |
980 | ||
981 | static inline void resched_rq(struct rq *rq) | |
982 | { | |
983 | unsigned long flags; | |
984 | ||
985 | spin_lock_irqsave(&rq->lock, flags); | |
986 | resched_task(rq->curr); | |
987 | spin_unlock_irqrestore(&rq->lock, flags); | |
988 | } | |
989 | ||
990 | enum { | |
991 | HRTICK_SET, /* re-programm hrtick_timer */ | |
992 | HRTICK_RESET, /* not a new slice */ | |
b328ca18 | 993 | HRTICK_BLOCK, /* stop hrtick operations */ |
8f4d37ec PZ |
994 | }; |
995 | ||
996 | /* | |
997 | * Use hrtick when: | |
998 | * - enabled by features | |
999 | * - hrtimer is actually high res | |
1000 | */ | |
1001 | static inline int hrtick_enabled(struct rq *rq) | |
1002 | { | |
1003 | if (!sched_feat(HRTICK)) | |
1004 | return 0; | |
b328ca18 PZ |
1005 | if (unlikely(test_bit(HRTICK_BLOCK, &rq->hrtick_flags))) |
1006 | return 0; | |
8f4d37ec PZ |
1007 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * Called to set the hrtick timer state. | |
1012 | * | |
1013 | * called with rq->lock held and irqs disabled | |
1014 | */ | |
1015 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1016 | { | |
1017 | assert_spin_locked(&rq->lock); | |
1018 | ||
1019 | /* | |
1020 | * preempt at: now + delay | |
1021 | */ | |
1022 | rq->hrtick_expire = | |
1023 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1024 | /* | |
1025 | * indicate we need to program the timer | |
1026 | */ | |
1027 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1028 | if (reset) | |
1029 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1030 | ||
1031 | /* | |
1032 | * New slices are called from the schedule path and don't need a | |
1033 | * forced reschedule. | |
1034 | */ | |
1035 | if (reset) | |
1036 | resched_hrt(rq->curr); | |
1037 | } | |
1038 | ||
1039 | static void hrtick_clear(struct rq *rq) | |
1040 | { | |
1041 | if (hrtimer_active(&rq->hrtick_timer)) | |
1042 | hrtimer_cancel(&rq->hrtick_timer); | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Update the timer from the possible pending state. | |
1047 | */ | |
1048 | static void hrtick_set(struct rq *rq) | |
1049 | { | |
1050 | ktime_t time; | |
1051 | int set, reset; | |
1052 | unsigned long flags; | |
1053 | ||
1054 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1055 | ||
1056 | spin_lock_irqsave(&rq->lock, flags); | |
1057 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1058 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1059 | time = rq->hrtick_expire; | |
1060 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1061 | spin_unlock_irqrestore(&rq->lock, flags); | |
1062 | ||
1063 | if (set) { | |
1064 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1065 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1066 | resched_rq(rq); | |
1067 | } else | |
1068 | hrtick_clear(rq); | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * High-resolution timer tick. | |
1073 | * Runs from hardirq context with interrupts disabled. | |
1074 | */ | |
1075 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1076 | { | |
1077 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1078 | ||
1079 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1080 | ||
1081 | spin_lock(&rq->lock); | |
3e51f33f | 1082 | update_rq_clock(rq); |
8f4d37ec PZ |
1083 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1084 | spin_unlock(&rq->lock); | |
1085 | ||
1086 | return HRTIMER_NORESTART; | |
1087 | } | |
1088 | ||
81d41d7e | 1089 | #ifdef CONFIG_SMP |
b328ca18 PZ |
1090 | static void hotplug_hrtick_disable(int cpu) |
1091 | { | |
1092 | struct rq *rq = cpu_rq(cpu); | |
1093 | unsigned long flags; | |
1094 | ||
1095 | spin_lock_irqsave(&rq->lock, flags); | |
1096 | rq->hrtick_flags = 0; | |
1097 | __set_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1098 | spin_unlock_irqrestore(&rq->lock, flags); | |
1099 | ||
1100 | hrtick_clear(rq); | |
1101 | } | |
1102 | ||
1103 | static void hotplug_hrtick_enable(int cpu) | |
1104 | { | |
1105 | struct rq *rq = cpu_rq(cpu); | |
1106 | unsigned long flags; | |
1107 | ||
1108 | spin_lock_irqsave(&rq->lock, flags); | |
1109 | __clear_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1110 | spin_unlock_irqrestore(&rq->lock, flags); | |
1111 | } | |
1112 | ||
1113 | static int | |
1114 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1115 | { | |
1116 | int cpu = (int)(long)hcpu; | |
1117 | ||
1118 | switch (action) { | |
1119 | case CPU_UP_CANCELED: | |
1120 | case CPU_UP_CANCELED_FROZEN: | |
1121 | case CPU_DOWN_PREPARE: | |
1122 | case CPU_DOWN_PREPARE_FROZEN: | |
1123 | case CPU_DEAD: | |
1124 | case CPU_DEAD_FROZEN: | |
1125 | hotplug_hrtick_disable(cpu); | |
1126 | return NOTIFY_OK; | |
1127 | ||
1128 | case CPU_UP_PREPARE: | |
1129 | case CPU_UP_PREPARE_FROZEN: | |
1130 | case CPU_DOWN_FAILED: | |
1131 | case CPU_DOWN_FAILED_FROZEN: | |
1132 | case CPU_ONLINE: | |
1133 | case CPU_ONLINE_FROZEN: | |
1134 | hotplug_hrtick_enable(cpu); | |
1135 | return NOTIFY_OK; | |
1136 | } | |
1137 | ||
1138 | return NOTIFY_DONE; | |
1139 | } | |
1140 | ||
1141 | static void init_hrtick(void) | |
1142 | { | |
1143 | hotcpu_notifier(hotplug_hrtick, 0); | |
1144 | } | |
81d41d7e | 1145 | #endif /* CONFIG_SMP */ |
b328ca18 PZ |
1146 | |
1147 | static void init_rq_hrtick(struct rq *rq) | |
8f4d37ec PZ |
1148 | { |
1149 | rq->hrtick_flags = 0; | |
1150 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1151 | rq->hrtick_timer.function = hrtick; | |
1152 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1153 | } | |
1154 | ||
1155 | void hrtick_resched(void) | |
1156 | { | |
1157 | struct rq *rq; | |
1158 | unsigned long flags; | |
1159 | ||
1160 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1161 | return; | |
1162 | ||
1163 | local_irq_save(flags); | |
1164 | rq = cpu_rq(smp_processor_id()); | |
1165 | hrtick_set(rq); | |
1166 | local_irq_restore(flags); | |
1167 | } | |
1168 | #else | |
1169 | static inline void hrtick_clear(struct rq *rq) | |
1170 | { | |
1171 | } | |
1172 | ||
1173 | static inline void hrtick_set(struct rq *rq) | |
1174 | { | |
1175 | } | |
1176 | ||
1177 | static inline void init_rq_hrtick(struct rq *rq) | |
1178 | { | |
1179 | } | |
1180 | ||
1181 | void hrtick_resched(void) | |
1182 | { | |
1183 | } | |
b328ca18 PZ |
1184 | |
1185 | static inline void init_hrtick(void) | |
1186 | { | |
1187 | } | |
8f4d37ec PZ |
1188 | #endif |
1189 | ||
c24d20db IM |
1190 | /* |
1191 | * resched_task - mark a task 'to be rescheduled now'. | |
1192 | * | |
1193 | * On UP this means the setting of the need_resched flag, on SMP it | |
1194 | * might also involve a cross-CPU call to trigger the scheduler on | |
1195 | * the target CPU. | |
1196 | */ | |
1197 | #ifdef CONFIG_SMP | |
1198 | ||
1199 | #ifndef tsk_is_polling | |
1200 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1201 | #endif | |
1202 | ||
8f4d37ec | 1203 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1204 | { |
1205 | int cpu; | |
1206 | ||
1207 | assert_spin_locked(&task_rq(p)->lock); | |
1208 | ||
8f4d37ec | 1209 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1210 | return; |
1211 | ||
8f4d37ec | 1212 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1213 | |
1214 | cpu = task_cpu(p); | |
1215 | if (cpu == smp_processor_id()) | |
1216 | return; | |
1217 | ||
1218 | /* NEED_RESCHED must be visible before we test polling */ | |
1219 | smp_mb(); | |
1220 | if (!tsk_is_polling(p)) | |
1221 | smp_send_reschedule(cpu); | |
1222 | } | |
1223 | ||
1224 | static void resched_cpu(int cpu) | |
1225 | { | |
1226 | struct rq *rq = cpu_rq(cpu); | |
1227 | unsigned long flags; | |
1228 | ||
1229 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1230 | return; | |
1231 | resched_task(cpu_curr(cpu)); | |
1232 | spin_unlock_irqrestore(&rq->lock, flags); | |
1233 | } | |
06d8308c TG |
1234 | |
1235 | #ifdef CONFIG_NO_HZ | |
1236 | /* | |
1237 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1238 | * idle CPU then this timer might expire before the next timer event | |
1239 | * which is scheduled to wake up that CPU. In case of a completely | |
1240 | * idle system the next event might even be infinite time into the | |
1241 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1242 | * leaves the inner idle loop so the newly added timer is taken into | |
1243 | * account when the CPU goes back to idle and evaluates the timer | |
1244 | * wheel for the next timer event. | |
1245 | */ | |
1246 | void wake_up_idle_cpu(int cpu) | |
1247 | { | |
1248 | struct rq *rq = cpu_rq(cpu); | |
1249 | ||
1250 | if (cpu == smp_processor_id()) | |
1251 | return; | |
1252 | ||
1253 | /* | |
1254 | * This is safe, as this function is called with the timer | |
1255 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1256 | * to idle and has not yet set rq->curr to idle then it will | |
1257 | * be serialized on the timer wheel base lock and take the new | |
1258 | * timer into account automatically. | |
1259 | */ | |
1260 | if (rq->curr != rq->idle) | |
1261 | return; | |
1262 | ||
1263 | /* | |
1264 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1265 | * lockless. The worst case is that the other CPU runs the | |
1266 | * idle task through an additional NOOP schedule() | |
1267 | */ | |
1268 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1269 | ||
1270 | /* NEED_RESCHED must be visible before we test polling */ | |
1271 | smp_mb(); | |
1272 | if (!tsk_is_polling(rq->idle)) | |
1273 | smp_send_reschedule(cpu); | |
1274 | } | |
6d6bc0ad | 1275 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1276 | |
6d6bc0ad | 1277 | #else /* !CONFIG_SMP */ |
8f4d37ec | 1278 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1279 | { |
1280 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1281 | set_tsk_thread_flag(p, tif_bit); |
c24d20db | 1282 | } |
6d6bc0ad | 1283 | #endif /* CONFIG_SMP */ |
c24d20db | 1284 | |
45bf76df IM |
1285 | #if BITS_PER_LONG == 32 |
1286 | # define WMULT_CONST (~0UL) | |
1287 | #else | |
1288 | # define WMULT_CONST (1UL << 32) | |
1289 | #endif | |
1290 | ||
1291 | #define WMULT_SHIFT 32 | |
1292 | ||
194081eb IM |
1293 | /* |
1294 | * Shift right and round: | |
1295 | */ | |
cf2ab469 | 1296 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1297 | |
a7be37ac PZ |
1298 | /* |
1299 | * delta *= weight / lw | |
1300 | */ | |
cb1c4fc9 | 1301 | static unsigned long |
45bf76df IM |
1302 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1303 | struct load_weight *lw) | |
1304 | { | |
1305 | u64 tmp; | |
1306 | ||
7a232e03 LJ |
1307 | if (!lw->inv_weight) { |
1308 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1309 | lw->inv_weight = 1; | |
1310 | else | |
1311 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1312 | / (lw->weight+1); | |
1313 | } | |
45bf76df IM |
1314 | |
1315 | tmp = (u64)delta_exec * weight; | |
1316 | /* | |
1317 | * Check whether we'd overflow the 64-bit multiplication: | |
1318 | */ | |
194081eb | 1319 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1320 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1321 | WMULT_SHIFT/2); |
1322 | else | |
cf2ab469 | 1323 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1324 | |
ecf691da | 1325 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1326 | } |
1327 | ||
1091985b | 1328 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1329 | { |
1330 | lw->weight += inc; | |
e89996ae | 1331 | lw->inv_weight = 0; |
45bf76df IM |
1332 | } |
1333 | ||
1091985b | 1334 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1335 | { |
1336 | lw->weight -= dec; | |
e89996ae | 1337 | lw->inv_weight = 0; |
45bf76df IM |
1338 | } |
1339 | ||
2dd73a4f PW |
1340 | /* |
1341 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1342 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1343 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1344 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1345 | * scaled version of the new time slice allocation that they receive on time |
1346 | * slice expiry etc. | |
1347 | */ | |
1348 | ||
dd41f596 IM |
1349 | #define WEIGHT_IDLEPRIO 2 |
1350 | #define WMULT_IDLEPRIO (1 << 31) | |
1351 | ||
1352 | /* | |
1353 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1354 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1355 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1356 | * that remained on nice 0. | |
1357 | * | |
1358 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1359 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1360 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1361 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1362 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1363 | */ |
1364 | static const int prio_to_weight[40] = { | |
254753dc IM |
1365 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1366 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1367 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1368 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1369 | /* 0 */ 1024, 820, 655, 526, 423, | |
1370 | /* 5 */ 335, 272, 215, 172, 137, | |
1371 | /* 10 */ 110, 87, 70, 56, 45, | |
1372 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1373 | }; |
1374 | ||
5714d2de IM |
1375 | /* |
1376 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1377 | * | |
1378 | * In cases where the weight does not change often, we can use the | |
1379 | * precalculated inverse to speed up arithmetics by turning divisions | |
1380 | * into multiplications: | |
1381 | */ | |
dd41f596 | 1382 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1383 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1384 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1385 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1386 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1387 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1388 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1389 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1390 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1391 | }; |
2dd73a4f | 1392 | |
dd41f596 IM |
1393 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1394 | ||
1395 | /* | |
1396 | * runqueue iterator, to support SMP load-balancing between different | |
1397 | * scheduling classes, without having to expose their internal data | |
1398 | * structures to the load-balancing proper: | |
1399 | */ | |
1400 | struct rq_iterator { | |
1401 | void *arg; | |
1402 | struct task_struct *(*start)(void *); | |
1403 | struct task_struct *(*next)(void *); | |
1404 | }; | |
1405 | ||
e1d1484f PW |
1406 | #ifdef CONFIG_SMP |
1407 | static unsigned long | |
1408 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1409 | unsigned long max_load_move, struct sched_domain *sd, | |
1410 | enum cpu_idle_type idle, int *all_pinned, | |
1411 | int *this_best_prio, struct rq_iterator *iterator); | |
1412 | ||
1413 | static int | |
1414 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1415 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1416 | struct rq_iterator *iterator); | |
e1d1484f | 1417 | #endif |
dd41f596 | 1418 | |
d842de87 SV |
1419 | #ifdef CONFIG_CGROUP_CPUACCT |
1420 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1421 | #else | |
1422 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1423 | #endif | |
1424 | ||
18d95a28 PZ |
1425 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1426 | { | |
1427 | update_load_add(&rq->load, load); | |
1428 | } | |
1429 | ||
1430 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1431 | { | |
1432 | update_load_sub(&rq->load, load); | |
1433 | } | |
1434 | ||
e7693a36 GH |
1435 | #ifdef CONFIG_SMP |
1436 | static unsigned long source_load(int cpu, int type); | |
1437 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1438 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1439 | |
a8a51d5e PZ |
1440 | static unsigned long cpu_avg_load_per_task(int cpu) |
1441 | { | |
1442 | struct rq *rq = cpu_rq(cpu); | |
1443 | ||
1444 | if (rq->nr_running) | |
1445 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1446 | ||
1447 | return rq->avg_load_per_task; | |
1448 | } | |
1449 | ||
c09595f6 PZ |
1450 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1451 | ||
c8cba857 | 1452 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1453 | |
1454 | /* | |
1455 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1456 | * leaving it for the final time. | |
1457 | */ | |
c8cba857 PZ |
1458 | static void |
1459 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1460 | { |
1461 | struct task_group *parent, *child; | |
1462 | ||
1463 | rcu_read_lock(); | |
1464 | parent = &root_task_group; | |
1465 | down: | |
b6a86c74 | 1466 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1467 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1468 | parent = child; | |
1469 | goto down; | |
1470 | ||
1471 | up: | |
1472 | continue; | |
1473 | } | |
b6a86c74 | 1474 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1475 | |
1476 | child = parent; | |
1477 | parent = parent->parent; | |
1478 | if (parent) | |
1479 | goto up; | |
1480 | rcu_read_unlock(); | |
1481 | } | |
1482 | ||
c09595f6 PZ |
1483 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1484 | ||
1485 | /* | |
1486 | * Calculate and set the cpu's group shares. | |
1487 | */ | |
1488 | static void | |
b6a86c74 | 1489 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1490 | unsigned long sd_shares, unsigned long sd_rq_weight) |
c09595f6 PZ |
1491 | { |
1492 | int boost = 0; | |
1493 | unsigned long shares; | |
1494 | unsigned long rq_weight; | |
1495 | ||
c8cba857 | 1496 | if (!tg->se[cpu]) |
c09595f6 PZ |
1497 | return; |
1498 | ||
c8cba857 | 1499 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1500 | |
1501 | /* | |
1502 | * If there are currently no tasks on the cpu pretend there is one of | |
1503 | * average load so that when a new task gets to run here it will not | |
1504 | * get delayed by group starvation. | |
1505 | */ | |
1506 | if (!rq_weight) { | |
1507 | boost = 1; | |
1508 | rq_weight = NICE_0_LOAD; | |
1509 | } | |
1510 | ||
c8cba857 PZ |
1511 | if (unlikely(rq_weight > sd_rq_weight)) |
1512 | rq_weight = sd_rq_weight; | |
1513 | ||
c09595f6 PZ |
1514 | /* |
1515 | * \Sum shares * rq_weight | |
1516 | * shares = ----------------------- | |
1517 | * \Sum rq_weight | |
1518 | * | |
1519 | */ | |
c8cba857 | 1520 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1521 | |
1522 | /* | |
1523 | * record the actual number of shares, not the boosted amount. | |
1524 | */ | |
c8cba857 | 1525 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
c09595f6 PZ |
1526 | |
1527 | if (shares < MIN_SHARES) | |
1528 | shares = MIN_SHARES; | |
1529 | else if (shares > MAX_SHARES) | |
1530 | shares = MAX_SHARES; | |
1531 | ||
c8cba857 | 1532 | __set_se_shares(tg->se[cpu], shares); |
c09595f6 PZ |
1533 | } |
1534 | ||
1535 | /* | |
c8cba857 PZ |
1536 | * Re-compute the task group their per cpu shares over the given domain. |
1537 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1538 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1539 | */ |
1540 | static void | |
c8cba857 | 1541 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1542 | { |
c8cba857 PZ |
1543 | unsigned long rq_weight = 0; |
1544 | unsigned long shares = 0; | |
1545 | int i; | |
c09595f6 | 1546 | |
c8cba857 PZ |
1547 | for_each_cpu_mask(i, sd->span) { |
1548 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1549 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1550 | } |
c09595f6 | 1551 | |
c8cba857 PZ |
1552 | if ((!shares && rq_weight) || shares > tg->shares) |
1553 | shares = tg->shares; | |
1554 | ||
1555 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1556 | shares = tg->shares; | |
c09595f6 | 1557 | |
cd80917e PZ |
1558 | if (!rq_weight) |
1559 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1560 | ||
c09595f6 PZ |
1561 | for_each_cpu_mask(i, sd->span) { |
1562 | struct rq *rq = cpu_rq(i); | |
1563 | unsigned long flags; | |
1564 | ||
1565 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1566 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1567 | spin_unlock_irqrestore(&rq->lock, flags); |
1568 | } | |
c09595f6 PZ |
1569 | } |
1570 | ||
1571 | /* | |
c8cba857 PZ |
1572 | * Compute the cpu's hierarchical load factor for each task group. |
1573 | * This needs to be done in a top-down fashion because the load of a child | |
1574 | * group is a fraction of its parents load. | |
c09595f6 | 1575 | */ |
b6a86c74 | 1576 | static void |
c8cba857 | 1577 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1578 | { |
c8cba857 | 1579 | unsigned long load; |
c09595f6 | 1580 | |
c8cba857 PZ |
1581 | if (!tg->parent) { |
1582 | load = cpu_rq(cpu)->load.weight; | |
1583 | } else { | |
1584 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1585 | load *= tg->cfs_rq[cpu]->shares; | |
1586 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1587 | } | |
c09595f6 | 1588 | |
c8cba857 | 1589 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1590 | } |
1591 | ||
c8cba857 PZ |
1592 | static void |
1593 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1594 | { |
c09595f6 PZ |
1595 | } |
1596 | ||
c8cba857 | 1597 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1598 | { |
2398f2c6 PZ |
1599 | u64 now = cpu_clock(raw_smp_processor_id()); |
1600 | s64 elapsed = now - sd->last_update; | |
1601 | ||
1602 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1603 | sd->last_update = now; | |
1604 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); | |
1605 | } | |
4d8d595d PZ |
1606 | } |
1607 | ||
3e5459b4 PZ |
1608 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1609 | { | |
1610 | spin_unlock(&rq->lock); | |
1611 | update_shares(sd); | |
1612 | spin_lock(&rq->lock); | |
1613 | } | |
1614 | ||
c8cba857 | 1615 | static void update_h_load(int cpu) |
c09595f6 | 1616 | { |
c8cba857 | 1617 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1618 | } |
1619 | ||
1620 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) | |
1621 | { | |
1622 | cfs_rq->shares = shares; | |
1623 | } | |
1624 | ||
1625 | #else | |
1626 | ||
c8cba857 | 1627 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1628 | { |
1629 | } | |
1630 | ||
3e5459b4 PZ |
1631 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1632 | { | |
1633 | } | |
1634 | ||
c09595f6 PZ |
1635 | #endif |
1636 | ||
18d95a28 PZ |
1637 | #endif |
1638 | ||
dd41f596 | 1639 | #include "sched_stats.h" |
dd41f596 | 1640 | #include "sched_idletask.c" |
5522d5d5 IM |
1641 | #include "sched_fair.c" |
1642 | #include "sched_rt.c" | |
dd41f596 IM |
1643 | #ifdef CONFIG_SCHED_DEBUG |
1644 | # include "sched_debug.c" | |
1645 | #endif | |
1646 | ||
1647 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1648 | #define for_each_class(class) \ |
1649 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1650 | |
c09595f6 | 1651 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1652 | { |
1653 | rq->nr_running++; | |
9c217245 IM |
1654 | } |
1655 | ||
c09595f6 | 1656 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1657 | { |
1658 | rq->nr_running--; | |
9c217245 IM |
1659 | } |
1660 | ||
45bf76df IM |
1661 | static void set_load_weight(struct task_struct *p) |
1662 | { | |
1663 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1664 | p->se.load.weight = prio_to_weight[0] * 2; |
1665 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1666 | return; | |
1667 | } | |
45bf76df | 1668 | |
dd41f596 IM |
1669 | /* |
1670 | * SCHED_IDLE tasks get minimal weight: | |
1671 | */ | |
1672 | if (p->policy == SCHED_IDLE) { | |
1673 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1674 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1675 | return; | |
1676 | } | |
71f8bd46 | 1677 | |
dd41f596 IM |
1678 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1679 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1680 | } |
1681 | ||
8159f87e | 1682 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1683 | { |
dd41f596 | 1684 | sched_info_queued(p); |
fd390f6a | 1685 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1686 | p->se.on_rq = 1; |
71f8bd46 IM |
1687 | } |
1688 | ||
69be72c1 | 1689 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1690 | { |
f02231e5 | 1691 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1692 | p->se.on_rq = 0; |
71f8bd46 IM |
1693 | } |
1694 | ||
14531189 | 1695 | /* |
dd41f596 | 1696 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1697 | */ |
14531189 IM |
1698 | static inline int __normal_prio(struct task_struct *p) |
1699 | { | |
dd41f596 | 1700 | return p->static_prio; |
14531189 IM |
1701 | } |
1702 | ||
b29739f9 IM |
1703 | /* |
1704 | * Calculate the expected normal priority: i.e. priority | |
1705 | * without taking RT-inheritance into account. Might be | |
1706 | * boosted by interactivity modifiers. Changes upon fork, | |
1707 | * setprio syscalls, and whenever the interactivity | |
1708 | * estimator recalculates. | |
1709 | */ | |
36c8b586 | 1710 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1711 | { |
1712 | int prio; | |
1713 | ||
e05606d3 | 1714 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1715 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1716 | else | |
1717 | prio = __normal_prio(p); | |
1718 | return prio; | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * Calculate the current priority, i.e. the priority | |
1723 | * taken into account by the scheduler. This value might | |
1724 | * be boosted by RT tasks, or might be boosted by | |
1725 | * interactivity modifiers. Will be RT if the task got | |
1726 | * RT-boosted. If not then it returns p->normal_prio. | |
1727 | */ | |
36c8b586 | 1728 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1729 | { |
1730 | p->normal_prio = normal_prio(p); | |
1731 | /* | |
1732 | * If we are RT tasks or we were boosted to RT priority, | |
1733 | * keep the priority unchanged. Otherwise, update priority | |
1734 | * to the normal priority: | |
1735 | */ | |
1736 | if (!rt_prio(p->prio)) | |
1737 | return p->normal_prio; | |
1738 | return p->prio; | |
1739 | } | |
1740 | ||
1da177e4 | 1741 | /* |
dd41f596 | 1742 | * activate_task - move a task to the runqueue. |
1da177e4 | 1743 | */ |
dd41f596 | 1744 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1745 | { |
d9514f6c | 1746 | if (task_contributes_to_load(p)) |
dd41f596 | 1747 | rq->nr_uninterruptible--; |
1da177e4 | 1748 | |
8159f87e | 1749 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1750 | inc_nr_running(rq); |
1da177e4 LT |
1751 | } |
1752 | ||
1da177e4 LT |
1753 | /* |
1754 | * deactivate_task - remove a task from the runqueue. | |
1755 | */ | |
2e1cb74a | 1756 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1757 | { |
d9514f6c | 1758 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1759 | rq->nr_uninterruptible++; |
1760 | ||
69be72c1 | 1761 | dequeue_task(rq, p, sleep); |
c09595f6 | 1762 | dec_nr_running(rq); |
1da177e4 LT |
1763 | } |
1764 | ||
1da177e4 LT |
1765 | /** |
1766 | * task_curr - is this task currently executing on a CPU? | |
1767 | * @p: the task in question. | |
1768 | */ | |
36c8b586 | 1769 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1770 | { |
1771 | return cpu_curr(task_cpu(p)) == p; | |
1772 | } | |
1773 | ||
dd41f596 IM |
1774 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1775 | { | |
6f505b16 | 1776 | set_task_rq(p, cpu); |
dd41f596 | 1777 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1778 | /* |
1779 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1780 | * successfuly executed on another CPU. We must ensure that updates of | |
1781 | * per-task data have been completed by this moment. | |
1782 | */ | |
1783 | smp_wmb(); | |
dd41f596 | 1784 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1785 | #endif |
2dd73a4f PW |
1786 | } |
1787 | ||
cb469845 SR |
1788 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1789 | const struct sched_class *prev_class, | |
1790 | int oldprio, int running) | |
1791 | { | |
1792 | if (prev_class != p->sched_class) { | |
1793 | if (prev_class->switched_from) | |
1794 | prev_class->switched_from(rq, p, running); | |
1795 | p->sched_class->switched_to(rq, p, running); | |
1796 | } else | |
1797 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1798 | } | |
1799 | ||
1da177e4 | 1800 | #ifdef CONFIG_SMP |
c65cc870 | 1801 | |
e958b360 TG |
1802 | /* Used instead of source_load when we know the type == 0 */ |
1803 | static unsigned long weighted_cpuload(const int cpu) | |
1804 | { | |
1805 | return cpu_rq(cpu)->load.weight; | |
1806 | } | |
1807 | ||
cc367732 IM |
1808 | /* |
1809 | * Is this task likely cache-hot: | |
1810 | */ | |
e7693a36 | 1811 | static int |
cc367732 IM |
1812 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1813 | { | |
1814 | s64 delta; | |
1815 | ||
f540a608 IM |
1816 | /* |
1817 | * Buddy candidates are cache hot: | |
1818 | */ | |
d25ce4cd | 1819 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1820 | return 1; |
1821 | ||
cc367732 IM |
1822 | if (p->sched_class != &fair_sched_class) |
1823 | return 0; | |
1824 | ||
6bc1665b IM |
1825 | if (sysctl_sched_migration_cost == -1) |
1826 | return 1; | |
1827 | if (sysctl_sched_migration_cost == 0) | |
1828 | return 0; | |
1829 | ||
cc367732 IM |
1830 | delta = now - p->se.exec_start; |
1831 | ||
1832 | return delta < (s64)sysctl_sched_migration_cost; | |
1833 | } | |
1834 | ||
1835 | ||
dd41f596 | 1836 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1837 | { |
dd41f596 IM |
1838 | int old_cpu = task_cpu(p); |
1839 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1840 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1841 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1842 | u64 clock_offset; |
dd41f596 IM |
1843 | |
1844 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1845 | |
1846 | #ifdef CONFIG_SCHEDSTATS | |
1847 | if (p->se.wait_start) | |
1848 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1849 | if (p->se.sleep_start) |
1850 | p->se.sleep_start -= clock_offset; | |
1851 | if (p->se.block_start) | |
1852 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1853 | if (old_cpu != new_cpu) { |
1854 | schedstat_inc(p, se.nr_migrations); | |
1855 | if (task_hot(p, old_rq->clock, NULL)) | |
1856 | schedstat_inc(p, se.nr_forced2_migrations); | |
1857 | } | |
6cfb0d5d | 1858 | #endif |
2830cf8c SV |
1859 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1860 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1861 | |
1862 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1863 | } |
1864 | ||
70b97a7f | 1865 | struct migration_req { |
1da177e4 | 1866 | struct list_head list; |
1da177e4 | 1867 | |
36c8b586 | 1868 | struct task_struct *task; |
1da177e4 LT |
1869 | int dest_cpu; |
1870 | ||
1da177e4 | 1871 | struct completion done; |
70b97a7f | 1872 | }; |
1da177e4 LT |
1873 | |
1874 | /* | |
1875 | * The task's runqueue lock must be held. | |
1876 | * Returns true if you have to wait for migration thread. | |
1877 | */ | |
36c8b586 | 1878 | static int |
70b97a7f | 1879 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1880 | { |
70b97a7f | 1881 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1882 | |
1883 | /* | |
1884 | * If the task is not on a runqueue (and not running), then | |
1885 | * it is sufficient to simply update the task's cpu field. | |
1886 | */ | |
dd41f596 | 1887 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1888 | set_task_cpu(p, dest_cpu); |
1889 | return 0; | |
1890 | } | |
1891 | ||
1892 | init_completion(&req->done); | |
1da177e4 LT |
1893 | req->task = p; |
1894 | req->dest_cpu = dest_cpu; | |
1895 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1896 | |
1da177e4 LT |
1897 | return 1; |
1898 | } | |
1899 | ||
1900 | /* | |
1901 | * wait_task_inactive - wait for a thread to unschedule. | |
1902 | * | |
1903 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1904 | * else this function might spin for a *long* time. This function can't | |
1905 | * be called with interrupts off, or it may introduce deadlock with | |
1906 | * smp_call_function() if an IPI is sent by the same process we are | |
1907 | * waiting to become inactive. | |
1908 | */ | |
36c8b586 | 1909 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1910 | { |
1911 | unsigned long flags; | |
dd41f596 | 1912 | int running, on_rq; |
70b97a7f | 1913 | struct rq *rq; |
1da177e4 | 1914 | |
3a5c359a AK |
1915 | for (;;) { |
1916 | /* | |
1917 | * We do the initial early heuristics without holding | |
1918 | * any task-queue locks at all. We'll only try to get | |
1919 | * the runqueue lock when things look like they will | |
1920 | * work out! | |
1921 | */ | |
1922 | rq = task_rq(p); | |
fa490cfd | 1923 | |
3a5c359a AK |
1924 | /* |
1925 | * If the task is actively running on another CPU | |
1926 | * still, just relax and busy-wait without holding | |
1927 | * any locks. | |
1928 | * | |
1929 | * NOTE! Since we don't hold any locks, it's not | |
1930 | * even sure that "rq" stays as the right runqueue! | |
1931 | * But we don't care, since "task_running()" will | |
1932 | * return false if the runqueue has changed and p | |
1933 | * is actually now running somewhere else! | |
1934 | */ | |
1935 | while (task_running(rq, p)) | |
1936 | cpu_relax(); | |
fa490cfd | 1937 | |
3a5c359a AK |
1938 | /* |
1939 | * Ok, time to look more closely! We need the rq | |
1940 | * lock now, to be *sure*. If we're wrong, we'll | |
1941 | * just go back and repeat. | |
1942 | */ | |
1943 | rq = task_rq_lock(p, &flags); | |
1944 | running = task_running(rq, p); | |
1945 | on_rq = p->se.on_rq; | |
1946 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1947 | |
3a5c359a AK |
1948 | /* |
1949 | * Was it really running after all now that we | |
1950 | * checked with the proper locks actually held? | |
1951 | * | |
1952 | * Oops. Go back and try again.. | |
1953 | */ | |
1954 | if (unlikely(running)) { | |
1955 | cpu_relax(); | |
1956 | continue; | |
1957 | } | |
fa490cfd | 1958 | |
3a5c359a AK |
1959 | /* |
1960 | * It's not enough that it's not actively running, | |
1961 | * it must be off the runqueue _entirely_, and not | |
1962 | * preempted! | |
1963 | * | |
1964 | * So if it wa still runnable (but just not actively | |
1965 | * running right now), it's preempted, and we should | |
1966 | * yield - it could be a while. | |
1967 | */ | |
1968 | if (unlikely(on_rq)) { | |
1969 | schedule_timeout_uninterruptible(1); | |
1970 | continue; | |
1971 | } | |
fa490cfd | 1972 | |
3a5c359a AK |
1973 | /* |
1974 | * Ahh, all good. It wasn't running, and it wasn't | |
1975 | * runnable, which means that it will never become | |
1976 | * running in the future either. We're all done! | |
1977 | */ | |
1978 | break; | |
1979 | } | |
1da177e4 LT |
1980 | } |
1981 | ||
1982 | /*** | |
1983 | * kick_process - kick a running thread to enter/exit the kernel | |
1984 | * @p: the to-be-kicked thread | |
1985 | * | |
1986 | * Cause a process which is running on another CPU to enter | |
1987 | * kernel-mode, without any delay. (to get signals handled.) | |
1988 | * | |
1989 | * NOTE: this function doesnt have to take the runqueue lock, | |
1990 | * because all it wants to ensure is that the remote task enters | |
1991 | * the kernel. If the IPI races and the task has been migrated | |
1992 | * to another CPU then no harm is done and the purpose has been | |
1993 | * achieved as well. | |
1994 | */ | |
36c8b586 | 1995 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1996 | { |
1997 | int cpu; | |
1998 | ||
1999 | preempt_disable(); | |
2000 | cpu = task_cpu(p); | |
2001 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2002 | smp_send_reschedule(cpu); | |
2003 | preempt_enable(); | |
2004 | } | |
2005 | ||
2006 | /* | |
2dd73a4f PW |
2007 | * Return a low guess at the load of a migration-source cpu weighted |
2008 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2009 | * |
2010 | * We want to under-estimate the load of migration sources, to | |
2011 | * balance conservatively. | |
2012 | */ | |
a9957449 | 2013 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2014 | { |
70b97a7f | 2015 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2016 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2017 | |
93b75217 | 2018 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2019 | return total; |
b910472d | 2020 | |
dd41f596 | 2021 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2022 | } |
2023 | ||
2024 | /* | |
2dd73a4f PW |
2025 | * Return a high guess at the load of a migration-target cpu weighted |
2026 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2027 | */ |
a9957449 | 2028 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2029 | { |
70b97a7f | 2030 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2031 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2032 | |
93b75217 | 2033 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2034 | return total; |
3b0bd9bc | 2035 | |
dd41f596 | 2036 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2037 | } |
2038 | ||
147cbb4b NP |
2039 | /* |
2040 | * find_idlest_group finds and returns the least busy CPU group within the | |
2041 | * domain. | |
2042 | */ | |
2043 | static struct sched_group * | |
2044 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2045 | { | |
2046 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2047 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2048 | int load_idx = sd->forkexec_idx; | |
2049 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2050 | ||
2051 | do { | |
2052 | unsigned long load, avg_load; | |
2053 | int local_group; | |
2054 | int i; | |
2055 | ||
da5a5522 BD |
2056 | /* Skip over this group if it has no CPUs allowed */ |
2057 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2058 | continue; |
da5a5522 | 2059 | |
147cbb4b | 2060 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2061 | |
2062 | /* Tally up the load of all CPUs in the group */ | |
2063 | avg_load = 0; | |
2064 | ||
2065 | for_each_cpu_mask(i, group->cpumask) { | |
2066 | /* Bias balancing toward cpus of our domain */ | |
2067 | if (local_group) | |
2068 | load = source_load(i, load_idx); | |
2069 | else | |
2070 | load = target_load(i, load_idx); | |
2071 | ||
2072 | avg_load += load; | |
2073 | } | |
2074 | ||
2075 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2076 | avg_load = sg_div_cpu_power(group, |
2077 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2078 | |
2079 | if (local_group) { | |
2080 | this_load = avg_load; | |
2081 | this = group; | |
2082 | } else if (avg_load < min_load) { | |
2083 | min_load = avg_load; | |
2084 | idlest = group; | |
2085 | } | |
3a5c359a | 2086 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2087 | |
2088 | if (!idlest || 100*this_load < imbalance*min_load) | |
2089 | return NULL; | |
2090 | return idlest; | |
2091 | } | |
2092 | ||
2093 | /* | |
0feaece9 | 2094 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2095 | */ |
95cdf3b7 | 2096 | static int |
7c16ec58 MT |
2097 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2098 | cpumask_t *tmp) | |
147cbb4b NP |
2099 | { |
2100 | unsigned long load, min_load = ULONG_MAX; | |
2101 | int idlest = -1; | |
2102 | int i; | |
2103 | ||
da5a5522 | 2104 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2105 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2106 | |
7c16ec58 | 2107 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 2108 | load = weighted_cpuload(i); |
147cbb4b NP |
2109 | |
2110 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2111 | min_load = load; | |
2112 | idlest = i; | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | return idlest; | |
2117 | } | |
2118 | ||
476d139c NP |
2119 | /* |
2120 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2121 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2122 | * SD_BALANCE_EXEC. | |
2123 | * | |
2124 | * Balance, ie. select the least loaded group. | |
2125 | * | |
2126 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2127 | * | |
2128 | * preempt must be disabled. | |
2129 | */ | |
2130 | static int sched_balance_self(int cpu, int flag) | |
2131 | { | |
2132 | struct task_struct *t = current; | |
2133 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2134 | |
c96d145e | 2135 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2136 | /* |
2137 | * If power savings logic is enabled for a domain, stop there. | |
2138 | */ | |
5c45bf27 SS |
2139 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2140 | break; | |
476d139c NP |
2141 | if (tmp->flags & flag) |
2142 | sd = tmp; | |
c96d145e | 2143 | } |
476d139c | 2144 | |
039a1c41 PZ |
2145 | if (sd) |
2146 | update_shares(sd); | |
2147 | ||
476d139c | 2148 | while (sd) { |
7c16ec58 | 2149 | cpumask_t span, tmpmask; |
476d139c | 2150 | struct sched_group *group; |
1a848870 SS |
2151 | int new_cpu, weight; |
2152 | ||
2153 | if (!(sd->flags & flag)) { | |
2154 | sd = sd->child; | |
2155 | continue; | |
2156 | } | |
476d139c NP |
2157 | |
2158 | span = sd->span; | |
2159 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2160 | if (!group) { |
2161 | sd = sd->child; | |
2162 | continue; | |
2163 | } | |
476d139c | 2164 | |
7c16ec58 | 2165 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2166 | if (new_cpu == -1 || new_cpu == cpu) { |
2167 | /* Now try balancing at a lower domain level of cpu */ | |
2168 | sd = sd->child; | |
2169 | continue; | |
2170 | } | |
476d139c | 2171 | |
1a848870 | 2172 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2173 | cpu = new_cpu; |
476d139c NP |
2174 | sd = NULL; |
2175 | weight = cpus_weight(span); | |
2176 | for_each_domain(cpu, tmp) { | |
2177 | if (weight <= cpus_weight(tmp->span)) | |
2178 | break; | |
2179 | if (tmp->flags & flag) | |
2180 | sd = tmp; | |
2181 | } | |
2182 | /* while loop will break here if sd == NULL */ | |
2183 | } | |
2184 | ||
2185 | return cpu; | |
2186 | } | |
2187 | ||
2188 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2189 | |
1da177e4 LT |
2190 | /*** |
2191 | * try_to_wake_up - wake up a thread | |
2192 | * @p: the to-be-woken-up thread | |
2193 | * @state: the mask of task states that can be woken | |
2194 | * @sync: do a synchronous wakeup? | |
2195 | * | |
2196 | * Put it on the run-queue if it's not already there. The "current" | |
2197 | * thread is always on the run-queue (except when the actual | |
2198 | * re-schedule is in progress), and as such you're allowed to do | |
2199 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2200 | * runnable without the overhead of this. | |
2201 | * | |
2202 | * returns failure only if the task is already active. | |
2203 | */ | |
36c8b586 | 2204 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2205 | { |
cc367732 | 2206 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2207 | unsigned long flags; |
2208 | long old_state; | |
70b97a7f | 2209 | struct rq *rq; |
1da177e4 | 2210 | |
b85d0667 IM |
2211 | if (!sched_feat(SYNC_WAKEUPS)) |
2212 | sync = 0; | |
2213 | ||
2398f2c6 PZ |
2214 | #ifdef CONFIG_SMP |
2215 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2216 | struct sched_domain *sd; | |
2217 | ||
2218 | this_cpu = raw_smp_processor_id(); | |
2219 | cpu = task_cpu(p); | |
2220 | ||
2221 | for_each_domain(this_cpu, sd) { | |
2222 | if (cpu_isset(cpu, sd->span)) { | |
2223 | update_shares(sd); | |
2224 | break; | |
2225 | } | |
2226 | } | |
2227 | } | |
2228 | #endif | |
2229 | ||
04e2f174 | 2230 | smp_wmb(); |
1da177e4 LT |
2231 | rq = task_rq_lock(p, &flags); |
2232 | old_state = p->state; | |
2233 | if (!(old_state & state)) | |
2234 | goto out; | |
2235 | ||
dd41f596 | 2236 | if (p->se.on_rq) |
1da177e4 LT |
2237 | goto out_running; |
2238 | ||
2239 | cpu = task_cpu(p); | |
cc367732 | 2240 | orig_cpu = cpu; |
1da177e4 LT |
2241 | this_cpu = smp_processor_id(); |
2242 | ||
2243 | #ifdef CONFIG_SMP | |
2244 | if (unlikely(task_running(rq, p))) | |
2245 | goto out_activate; | |
2246 | ||
5d2f5a61 DA |
2247 | cpu = p->sched_class->select_task_rq(p, sync); |
2248 | if (cpu != orig_cpu) { | |
2249 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2250 | task_rq_unlock(rq, &flags); |
2251 | /* might preempt at this point */ | |
2252 | rq = task_rq_lock(p, &flags); | |
2253 | old_state = p->state; | |
2254 | if (!(old_state & state)) | |
2255 | goto out; | |
dd41f596 | 2256 | if (p->se.on_rq) |
1da177e4 LT |
2257 | goto out_running; |
2258 | ||
2259 | this_cpu = smp_processor_id(); | |
2260 | cpu = task_cpu(p); | |
2261 | } | |
2262 | ||
e7693a36 GH |
2263 | #ifdef CONFIG_SCHEDSTATS |
2264 | schedstat_inc(rq, ttwu_count); | |
2265 | if (cpu == this_cpu) | |
2266 | schedstat_inc(rq, ttwu_local); | |
2267 | else { | |
2268 | struct sched_domain *sd; | |
2269 | for_each_domain(this_cpu, sd) { | |
2270 | if (cpu_isset(cpu, sd->span)) { | |
2271 | schedstat_inc(sd, ttwu_wake_remote); | |
2272 | break; | |
2273 | } | |
2274 | } | |
2275 | } | |
6d6bc0ad | 2276 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2277 | |
1da177e4 LT |
2278 | out_activate: |
2279 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2280 | schedstat_inc(p, se.nr_wakeups); |
2281 | if (sync) | |
2282 | schedstat_inc(p, se.nr_wakeups_sync); | |
2283 | if (orig_cpu != cpu) | |
2284 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2285 | if (cpu == this_cpu) | |
2286 | schedstat_inc(p, se.nr_wakeups_local); | |
2287 | else | |
2288 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2289 | update_rq_clock(rq); |
dd41f596 | 2290 | activate_task(rq, p, 1); |
1da177e4 LT |
2291 | success = 1; |
2292 | ||
2293 | out_running: | |
4ae7d5ce IM |
2294 | check_preempt_curr(rq, p); |
2295 | ||
1da177e4 | 2296 | p->state = TASK_RUNNING; |
9a897c5a SR |
2297 | #ifdef CONFIG_SMP |
2298 | if (p->sched_class->task_wake_up) | |
2299 | p->sched_class->task_wake_up(rq, p); | |
2300 | #endif | |
1da177e4 LT |
2301 | out: |
2302 | task_rq_unlock(rq, &flags); | |
2303 | ||
2304 | return success; | |
2305 | } | |
2306 | ||
7ad5b3a5 | 2307 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2308 | { |
d9514f6c | 2309 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2310 | } |
1da177e4 LT |
2311 | EXPORT_SYMBOL(wake_up_process); |
2312 | ||
7ad5b3a5 | 2313 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2314 | { |
2315 | return try_to_wake_up(p, state, 0); | |
2316 | } | |
2317 | ||
1da177e4 LT |
2318 | /* |
2319 | * Perform scheduler related setup for a newly forked process p. | |
2320 | * p is forked by current. | |
dd41f596 IM |
2321 | * |
2322 | * __sched_fork() is basic setup used by init_idle() too: | |
2323 | */ | |
2324 | static void __sched_fork(struct task_struct *p) | |
2325 | { | |
dd41f596 IM |
2326 | p->se.exec_start = 0; |
2327 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2328 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2329 | p->se.last_wakeup = 0; |
2330 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2331 | |
2332 | #ifdef CONFIG_SCHEDSTATS | |
2333 | p->se.wait_start = 0; | |
dd41f596 IM |
2334 | p->se.sum_sleep_runtime = 0; |
2335 | p->se.sleep_start = 0; | |
dd41f596 IM |
2336 | p->se.block_start = 0; |
2337 | p->se.sleep_max = 0; | |
2338 | p->se.block_max = 0; | |
2339 | p->se.exec_max = 0; | |
eba1ed4b | 2340 | p->se.slice_max = 0; |
dd41f596 | 2341 | p->se.wait_max = 0; |
6cfb0d5d | 2342 | #endif |
476d139c | 2343 | |
fa717060 | 2344 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2345 | p->se.on_rq = 0; |
4a55bd5e | 2346 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2347 | |
e107be36 AK |
2348 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2349 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2350 | #endif | |
2351 | ||
1da177e4 LT |
2352 | /* |
2353 | * We mark the process as running here, but have not actually | |
2354 | * inserted it onto the runqueue yet. This guarantees that | |
2355 | * nobody will actually run it, and a signal or other external | |
2356 | * event cannot wake it up and insert it on the runqueue either. | |
2357 | */ | |
2358 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2359 | } |
2360 | ||
2361 | /* | |
2362 | * fork()/clone()-time setup: | |
2363 | */ | |
2364 | void sched_fork(struct task_struct *p, int clone_flags) | |
2365 | { | |
2366 | int cpu = get_cpu(); | |
2367 | ||
2368 | __sched_fork(p); | |
2369 | ||
2370 | #ifdef CONFIG_SMP | |
2371 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2372 | #endif | |
02e4bac2 | 2373 | set_task_cpu(p, cpu); |
b29739f9 IM |
2374 | |
2375 | /* | |
2376 | * Make sure we do not leak PI boosting priority to the child: | |
2377 | */ | |
2378 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2379 | if (!rt_prio(p->prio)) |
2380 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2381 | |
52f17b6c | 2382 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2383 | if (likely(sched_info_on())) |
52f17b6c | 2384 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2385 | #endif |
d6077cb8 | 2386 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2387 | p->oncpu = 0; |
2388 | #endif | |
1da177e4 | 2389 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2390 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2391 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2392 | #endif |
476d139c | 2393 | put_cpu(); |
1da177e4 LT |
2394 | } |
2395 | ||
2396 | /* | |
2397 | * wake_up_new_task - wake up a newly created task for the first time. | |
2398 | * | |
2399 | * This function will do some initial scheduler statistics housekeeping | |
2400 | * that must be done for every newly created context, then puts the task | |
2401 | * on the runqueue and wakes it. | |
2402 | */ | |
7ad5b3a5 | 2403 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2404 | { |
2405 | unsigned long flags; | |
dd41f596 | 2406 | struct rq *rq; |
1da177e4 LT |
2407 | |
2408 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2409 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2410 | update_rq_clock(rq); |
1da177e4 LT |
2411 | |
2412 | p->prio = effective_prio(p); | |
2413 | ||
b9dca1e0 | 2414 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2415 | activate_task(rq, p, 0); |
1da177e4 | 2416 | } else { |
1da177e4 | 2417 | /* |
dd41f596 IM |
2418 | * Let the scheduling class do new task startup |
2419 | * management (if any): | |
1da177e4 | 2420 | */ |
ee0827d8 | 2421 | p->sched_class->task_new(rq, p); |
c09595f6 | 2422 | inc_nr_running(rq); |
1da177e4 | 2423 | } |
dd41f596 | 2424 | check_preempt_curr(rq, p); |
9a897c5a SR |
2425 | #ifdef CONFIG_SMP |
2426 | if (p->sched_class->task_wake_up) | |
2427 | p->sched_class->task_wake_up(rq, p); | |
2428 | #endif | |
dd41f596 | 2429 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2430 | } |
2431 | ||
e107be36 AK |
2432 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2433 | ||
2434 | /** | |
421cee29 RD |
2435 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2436 | * @notifier: notifier struct to register | |
e107be36 AK |
2437 | */ |
2438 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2439 | { | |
2440 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2441 | } | |
2442 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2443 | ||
2444 | /** | |
2445 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2446 | * @notifier: notifier struct to unregister |
e107be36 AK |
2447 | * |
2448 | * This is safe to call from within a preemption notifier. | |
2449 | */ | |
2450 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2451 | { | |
2452 | hlist_del(¬ifier->link); | |
2453 | } | |
2454 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2455 | ||
2456 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2457 | { | |
2458 | struct preempt_notifier *notifier; | |
2459 | struct hlist_node *node; | |
2460 | ||
2461 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2462 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2463 | } | |
2464 | ||
2465 | static void | |
2466 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2467 | struct task_struct *next) | |
2468 | { | |
2469 | struct preempt_notifier *notifier; | |
2470 | struct hlist_node *node; | |
2471 | ||
2472 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2473 | notifier->ops->sched_out(notifier, next); | |
2474 | } | |
2475 | ||
6d6bc0ad | 2476 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2477 | |
2478 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2479 | { | |
2480 | } | |
2481 | ||
2482 | static void | |
2483 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2484 | struct task_struct *next) | |
2485 | { | |
2486 | } | |
2487 | ||
6d6bc0ad | 2488 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2489 | |
4866cde0 NP |
2490 | /** |
2491 | * prepare_task_switch - prepare to switch tasks | |
2492 | * @rq: the runqueue preparing to switch | |
421cee29 | 2493 | * @prev: the current task that is being switched out |
4866cde0 NP |
2494 | * @next: the task we are going to switch to. |
2495 | * | |
2496 | * This is called with the rq lock held and interrupts off. It must | |
2497 | * be paired with a subsequent finish_task_switch after the context | |
2498 | * switch. | |
2499 | * | |
2500 | * prepare_task_switch sets up locking and calls architecture specific | |
2501 | * hooks. | |
2502 | */ | |
e107be36 AK |
2503 | static inline void |
2504 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2505 | struct task_struct *next) | |
4866cde0 | 2506 | { |
e107be36 | 2507 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2508 | prepare_lock_switch(rq, next); |
2509 | prepare_arch_switch(next); | |
2510 | } | |
2511 | ||
1da177e4 LT |
2512 | /** |
2513 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2514 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2515 | * @prev: the thread we just switched away from. |
2516 | * | |
4866cde0 NP |
2517 | * finish_task_switch must be called after the context switch, paired |
2518 | * with a prepare_task_switch call before the context switch. | |
2519 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2520 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2521 | * |
2522 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2523 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2524 | * with the lock held can cause deadlocks; see schedule() for |
2525 | * details.) | |
2526 | */ | |
a9957449 | 2527 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2528 | __releases(rq->lock) |
2529 | { | |
1da177e4 | 2530 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2531 | long prev_state; |
1da177e4 LT |
2532 | |
2533 | rq->prev_mm = NULL; | |
2534 | ||
2535 | /* | |
2536 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2537 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2538 | * schedule one last time. The schedule call will never return, and |
2539 | * the scheduled task must drop that reference. | |
c394cc9f | 2540 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2541 | * still held, otherwise prev could be scheduled on another cpu, die |
2542 | * there before we look at prev->state, and then the reference would | |
2543 | * be dropped twice. | |
2544 | * Manfred Spraul <manfred@colorfullife.com> | |
2545 | */ | |
55a101f8 | 2546 | prev_state = prev->state; |
4866cde0 NP |
2547 | finish_arch_switch(prev); |
2548 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2549 | #ifdef CONFIG_SMP |
2550 | if (current->sched_class->post_schedule) | |
2551 | current->sched_class->post_schedule(rq); | |
2552 | #endif | |
e8fa1362 | 2553 | |
e107be36 | 2554 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2555 | if (mm) |
2556 | mmdrop(mm); | |
c394cc9f | 2557 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2558 | /* |
2559 | * Remove function-return probe instances associated with this | |
2560 | * task and put them back on the free list. | |
9761eea8 | 2561 | */ |
c6fd91f0 | 2562 | kprobe_flush_task(prev); |
1da177e4 | 2563 | put_task_struct(prev); |
c6fd91f0 | 2564 | } |
1da177e4 LT |
2565 | } |
2566 | ||
2567 | /** | |
2568 | * schedule_tail - first thing a freshly forked thread must call. | |
2569 | * @prev: the thread we just switched away from. | |
2570 | */ | |
36c8b586 | 2571 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2572 | __releases(rq->lock) |
2573 | { | |
70b97a7f IM |
2574 | struct rq *rq = this_rq(); |
2575 | ||
4866cde0 NP |
2576 | finish_task_switch(rq, prev); |
2577 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2578 | /* In this case, finish_task_switch does not reenable preemption */ | |
2579 | preempt_enable(); | |
2580 | #endif | |
1da177e4 | 2581 | if (current->set_child_tid) |
b488893a | 2582 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2583 | } |
2584 | ||
2585 | /* | |
2586 | * context_switch - switch to the new MM and the new | |
2587 | * thread's register state. | |
2588 | */ | |
dd41f596 | 2589 | static inline void |
70b97a7f | 2590 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2591 | struct task_struct *next) |
1da177e4 | 2592 | { |
dd41f596 | 2593 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2594 | |
e107be36 | 2595 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2596 | mm = next->mm; |
2597 | oldmm = prev->active_mm; | |
9226d125 ZA |
2598 | /* |
2599 | * For paravirt, this is coupled with an exit in switch_to to | |
2600 | * combine the page table reload and the switch backend into | |
2601 | * one hypercall. | |
2602 | */ | |
2603 | arch_enter_lazy_cpu_mode(); | |
2604 | ||
dd41f596 | 2605 | if (unlikely(!mm)) { |
1da177e4 LT |
2606 | next->active_mm = oldmm; |
2607 | atomic_inc(&oldmm->mm_count); | |
2608 | enter_lazy_tlb(oldmm, next); | |
2609 | } else | |
2610 | switch_mm(oldmm, mm, next); | |
2611 | ||
dd41f596 | 2612 | if (unlikely(!prev->mm)) { |
1da177e4 | 2613 | prev->active_mm = NULL; |
1da177e4 LT |
2614 | rq->prev_mm = oldmm; |
2615 | } | |
3a5f5e48 IM |
2616 | /* |
2617 | * Since the runqueue lock will be released by the next | |
2618 | * task (which is an invalid locking op but in the case | |
2619 | * of the scheduler it's an obvious special-case), so we | |
2620 | * do an early lockdep release here: | |
2621 | */ | |
2622 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2623 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2624 | #endif |
1da177e4 LT |
2625 | |
2626 | /* Here we just switch the register state and the stack. */ | |
2627 | switch_to(prev, next, prev); | |
2628 | ||
dd41f596 IM |
2629 | barrier(); |
2630 | /* | |
2631 | * this_rq must be evaluated again because prev may have moved | |
2632 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2633 | * frame will be invalid. | |
2634 | */ | |
2635 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2636 | } |
2637 | ||
2638 | /* | |
2639 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2640 | * | |
2641 | * externally visible scheduler statistics: current number of runnable | |
2642 | * threads, current number of uninterruptible-sleeping threads, total | |
2643 | * number of context switches performed since bootup. | |
2644 | */ | |
2645 | unsigned long nr_running(void) | |
2646 | { | |
2647 | unsigned long i, sum = 0; | |
2648 | ||
2649 | for_each_online_cpu(i) | |
2650 | sum += cpu_rq(i)->nr_running; | |
2651 | ||
2652 | return sum; | |
2653 | } | |
2654 | ||
2655 | unsigned long nr_uninterruptible(void) | |
2656 | { | |
2657 | unsigned long i, sum = 0; | |
2658 | ||
0a945022 | 2659 | for_each_possible_cpu(i) |
1da177e4 LT |
2660 | sum += cpu_rq(i)->nr_uninterruptible; |
2661 | ||
2662 | /* | |
2663 | * Since we read the counters lockless, it might be slightly | |
2664 | * inaccurate. Do not allow it to go below zero though: | |
2665 | */ | |
2666 | if (unlikely((long)sum < 0)) | |
2667 | sum = 0; | |
2668 | ||
2669 | return sum; | |
2670 | } | |
2671 | ||
2672 | unsigned long long nr_context_switches(void) | |
2673 | { | |
cc94abfc SR |
2674 | int i; |
2675 | unsigned long long sum = 0; | |
1da177e4 | 2676 | |
0a945022 | 2677 | for_each_possible_cpu(i) |
1da177e4 LT |
2678 | sum += cpu_rq(i)->nr_switches; |
2679 | ||
2680 | return sum; | |
2681 | } | |
2682 | ||
2683 | unsigned long nr_iowait(void) | |
2684 | { | |
2685 | unsigned long i, sum = 0; | |
2686 | ||
0a945022 | 2687 | for_each_possible_cpu(i) |
1da177e4 LT |
2688 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2689 | ||
2690 | return sum; | |
2691 | } | |
2692 | ||
db1b1fef JS |
2693 | unsigned long nr_active(void) |
2694 | { | |
2695 | unsigned long i, running = 0, uninterruptible = 0; | |
2696 | ||
2697 | for_each_online_cpu(i) { | |
2698 | running += cpu_rq(i)->nr_running; | |
2699 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2700 | } | |
2701 | ||
2702 | if (unlikely((long)uninterruptible < 0)) | |
2703 | uninterruptible = 0; | |
2704 | ||
2705 | return running + uninterruptible; | |
2706 | } | |
2707 | ||
48f24c4d | 2708 | /* |
dd41f596 IM |
2709 | * Update rq->cpu_load[] statistics. This function is usually called every |
2710 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2711 | */ |
dd41f596 | 2712 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2713 | { |
495eca49 | 2714 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2715 | int i, scale; |
2716 | ||
2717 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2718 | |
2719 | /* Update our load: */ | |
2720 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2721 | unsigned long old_load, new_load; | |
2722 | ||
2723 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2724 | ||
2725 | old_load = this_rq->cpu_load[i]; | |
2726 | new_load = this_load; | |
a25707f3 IM |
2727 | /* |
2728 | * Round up the averaging division if load is increasing. This | |
2729 | * prevents us from getting stuck on 9 if the load is 10, for | |
2730 | * example. | |
2731 | */ | |
2732 | if (new_load > old_load) | |
2733 | new_load += scale-1; | |
dd41f596 IM |
2734 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2735 | } | |
48f24c4d IM |
2736 | } |
2737 | ||
dd41f596 IM |
2738 | #ifdef CONFIG_SMP |
2739 | ||
1da177e4 LT |
2740 | /* |
2741 | * double_rq_lock - safely lock two runqueues | |
2742 | * | |
2743 | * Note this does not disable interrupts like task_rq_lock, | |
2744 | * you need to do so manually before calling. | |
2745 | */ | |
70b97a7f | 2746 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2747 | __acquires(rq1->lock) |
2748 | __acquires(rq2->lock) | |
2749 | { | |
054b9108 | 2750 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2751 | if (rq1 == rq2) { |
2752 | spin_lock(&rq1->lock); | |
2753 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2754 | } else { | |
c96d145e | 2755 | if (rq1 < rq2) { |
1da177e4 LT |
2756 | spin_lock(&rq1->lock); |
2757 | spin_lock(&rq2->lock); | |
2758 | } else { | |
2759 | spin_lock(&rq2->lock); | |
2760 | spin_lock(&rq1->lock); | |
2761 | } | |
2762 | } | |
6e82a3be IM |
2763 | update_rq_clock(rq1); |
2764 | update_rq_clock(rq2); | |
1da177e4 LT |
2765 | } |
2766 | ||
2767 | /* | |
2768 | * double_rq_unlock - safely unlock two runqueues | |
2769 | * | |
2770 | * Note this does not restore interrupts like task_rq_unlock, | |
2771 | * you need to do so manually after calling. | |
2772 | */ | |
70b97a7f | 2773 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2774 | __releases(rq1->lock) |
2775 | __releases(rq2->lock) | |
2776 | { | |
2777 | spin_unlock(&rq1->lock); | |
2778 | if (rq1 != rq2) | |
2779 | spin_unlock(&rq2->lock); | |
2780 | else | |
2781 | __release(rq2->lock); | |
2782 | } | |
2783 | ||
2784 | /* | |
2785 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2786 | */ | |
e8fa1362 | 2787 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2788 | __releases(this_rq->lock) |
2789 | __acquires(busiest->lock) | |
2790 | __acquires(this_rq->lock) | |
2791 | { | |
e8fa1362 SR |
2792 | int ret = 0; |
2793 | ||
054b9108 KK |
2794 | if (unlikely(!irqs_disabled())) { |
2795 | /* printk() doesn't work good under rq->lock */ | |
2796 | spin_unlock(&this_rq->lock); | |
2797 | BUG_ON(1); | |
2798 | } | |
1da177e4 | 2799 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2800 | if (busiest < this_rq) { |
1da177e4 LT |
2801 | spin_unlock(&this_rq->lock); |
2802 | spin_lock(&busiest->lock); | |
2803 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2804 | ret = 1; |
1da177e4 LT |
2805 | } else |
2806 | spin_lock(&busiest->lock); | |
2807 | } | |
e8fa1362 | 2808 | return ret; |
1da177e4 LT |
2809 | } |
2810 | ||
1da177e4 LT |
2811 | /* |
2812 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2813 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2814 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2815 | * the cpu_allowed mask is restored. |
2816 | */ | |
36c8b586 | 2817 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2818 | { |
70b97a7f | 2819 | struct migration_req req; |
1da177e4 | 2820 | unsigned long flags; |
70b97a7f | 2821 | struct rq *rq; |
1da177e4 LT |
2822 | |
2823 | rq = task_rq_lock(p, &flags); | |
2824 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2825 | || unlikely(cpu_is_offline(dest_cpu))) | |
2826 | goto out; | |
2827 | ||
2828 | /* force the process onto the specified CPU */ | |
2829 | if (migrate_task(p, dest_cpu, &req)) { | |
2830 | /* Need to wait for migration thread (might exit: take ref). */ | |
2831 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2832 | |
1da177e4 LT |
2833 | get_task_struct(mt); |
2834 | task_rq_unlock(rq, &flags); | |
2835 | wake_up_process(mt); | |
2836 | put_task_struct(mt); | |
2837 | wait_for_completion(&req.done); | |
36c8b586 | 2838 | |
1da177e4 LT |
2839 | return; |
2840 | } | |
2841 | out: | |
2842 | task_rq_unlock(rq, &flags); | |
2843 | } | |
2844 | ||
2845 | /* | |
476d139c NP |
2846 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2847 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2848 | */ |
2849 | void sched_exec(void) | |
2850 | { | |
1da177e4 | 2851 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2852 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2853 | put_cpu(); |
476d139c NP |
2854 | if (new_cpu != this_cpu) |
2855 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2856 | } |
2857 | ||
2858 | /* | |
2859 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2860 | * Both runqueues must be locked. | |
2861 | */ | |
dd41f596 IM |
2862 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2863 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2864 | { |
2e1cb74a | 2865 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2866 | set_task_cpu(p, this_cpu); |
dd41f596 | 2867 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2868 | /* |
2869 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2870 | * to be always true for them. | |
2871 | */ | |
dd41f596 | 2872 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2873 | } |
2874 | ||
2875 | /* | |
2876 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2877 | */ | |
858119e1 | 2878 | static |
70b97a7f | 2879 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2880 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2881 | int *all_pinned) |
1da177e4 LT |
2882 | { |
2883 | /* | |
2884 | * We do not migrate tasks that are: | |
2885 | * 1) running (obviously), or | |
2886 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2887 | * 3) are cache-hot on their current CPU. | |
2888 | */ | |
cc367732 IM |
2889 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2890 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2891 | return 0; |
cc367732 | 2892 | } |
81026794 NP |
2893 | *all_pinned = 0; |
2894 | ||
cc367732 IM |
2895 | if (task_running(rq, p)) { |
2896 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2897 | return 0; |
cc367732 | 2898 | } |
1da177e4 | 2899 | |
da84d961 IM |
2900 | /* |
2901 | * Aggressive migration if: | |
2902 | * 1) task is cache cold, or | |
2903 | * 2) too many balance attempts have failed. | |
2904 | */ | |
2905 | ||
6bc1665b IM |
2906 | if (!task_hot(p, rq->clock, sd) || |
2907 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2908 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2909 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2910 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2911 | schedstat_inc(p, se.nr_forced_migrations); |
2912 | } | |
da84d961 IM |
2913 | #endif |
2914 | return 1; | |
2915 | } | |
2916 | ||
cc367732 IM |
2917 | if (task_hot(p, rq->clock, sd)) { |
2918 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2919 | return 0; |
cc367732 | 2920 | } |
1da177e4 LT |
2921 | return 1; |
2922 | } | |
2923 | ||
e1d1484f PW |
2924 | static unsigned long |
2925 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2926 | unsigned long max_load_move, struct sched_domain *sd, | |
2927 | enum cpu_idle_type idle, int *all_pinned, | |
2928 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2929 | { |
051c6764 | 2930 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2931 | struct task_struct *p; |
2932 | long rem_load_move = max_load_move; | |
1da177e4 | 2933 | |
e1d1484f | 2934 | if (max_load_move == 0) |
1da177e4 LT |
2935 | goto out; |
2936 | ||
81026794 NP |
2937 | pinned = 1; |
2938 | ||
1da177e4 | 2939 | /* |
dd41f596 | 2940 | * Start the load-balancing iterator: |
1da177e4 | 2941 | */ |
dd41f596 IM |
2942 | p = iterator->start(iterator->arg); |
2943 | next: | |
b82d9fdd | 2944 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2945 | goto out; |
051c6764 PZ |
2946 | |
2947 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2948 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2949 | p = iterator->next(iterator->arg); |
2950 | goto next; | |
1da177e4 LT |
2951 | } |
2952 | ||
dd41f596 | 2953 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2954 | pulled++; |
dd41f596 | 2955 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2956 | |
2dd73a4f | 2957 | /* |
b82d9fdd | 2958 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2959 | */ |
e1d1484f | 2960 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2961 | if (p->prio < *this_best_prio) |
2962 | *this_best_prio = p->prio; | |
dd41f596 IM |
2963 | p = iterator->next(iterator->arg); |
2964 | goto next; | |
1da177e4 LT |
2965 | } |
2966 | out: | |
2967 | /* | |
e1d1484f | 2968 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2969 | * so we can safely collect pull_task() stats here rather than |
2970 | * inside pull_task(). | |
2971 | */ | |
2972 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2973 | |
2974 | if (all_pinned) | |
2975 | *all_pinned = pinned; | |
e1d1484f PW |
2976 | |
2977 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2978 | } |
2979 | ||
dd41f596 | 2980 | /* |
43010659 PW |
2981 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2982 | * this_rq, as part of a balancing operation within domain "sd". | |
2983 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2984 | * |
2985 | * Called with both runqueues locked. | |
2986 | */ | |
2987 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2988 | unsigned long max_load_move, |
dd41f596 IM |
2989 | struct sched_domain *sd, enum cpu_idle_type idle, |
2990 | int *all_pinned) | |
2991 | { | |
5522d5d5 | 2992 | const struct sched_class *class = sched_class_highest; |
43010659 | 2993 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2994 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2995 | |
2996 | do { | |
43010659 PW |
2997 | total_load_moved += |
2998 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2999 | max_load_move - total_load_moved, |
a4ac01c3 | 3000 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3001 | class = class->next; |
43010659 | 3002 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3003 | |
43010659 PW |
3004 | return total_load_moved > 0; |
3005 | } | |
3006 | ||
e1d1484f PW |
3007 | static int |
3008 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3009 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3010 | struct rq_iterator *iterator) | |
3011 | { | |
3012 | struct task_struct *p = iterator->start(iterator->arg); | |
3013 | int pinned = 0; | |
3014 | ||
3015 | while (p) { | |
3016 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3017 | pull_task(busiest, p, this_rq, this_cpu); | |
3018 | /* | |
3019 | * Right now, this is only the second place pull_task() | |
3020 | * is called, so we can safely collect pull_task() | |
3021 | * stats here rather than inside pull_task(). | |
3022 | */ | |
3023 | schedstat_inc(sd, lb_gained[idle]); | |
3024 | ||
3025 | return 1; | |
3026 | } | |
3027 | p = iterator->next(iterator->arg); | |
3028 | } | |
3029 | ||
3030 | return 0; | |
3031 | } | |
3032 | ||
43010659 PW |
3033 | /* |
3034 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3035 | * part of active balancing operations within "domain". | |
3036 | * Returns 1 if successful and 0 otherwise. | |
3037 | * | |
3038 | * Called with both runqueues locked. | |
3039 | */ | |
3040 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3041 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3042 | { | |
5522d5d5 | 3043 | const struct sched_class *class; |
43010659 PW |
3044 | |
3045 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3046 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3047 | return 1; |
3048 | ||
3049 | return 0; | |
dd41f596 IM |
3050 | } |
3051 | ||
1da177e4 LT |
3052 | /* |
3053 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3054 | * domain. It calculates and returns the amount of weighted load which |
3055 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3056 | */ |
3057 | static struct sched_group * | |
3058 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3059 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3060 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3061 | { |
3062 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3063 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3064 | unsigned long max_pull; |
2dd73a4f PW |
3065 | unsigned long busiest_load_per_task, busiest_nr_running; |
3066 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3067 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3068 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3069 | int power_savings_balance = 1; | |
3070 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3071 | unsigned long min_nr_running = ULONG_MAX; | |
3072 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3073 | #endif | |
1da177e4 LT |
3074 | |
3075 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3076 | busiest_load_per_task = busiest_nr_running = 0; |
3077 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3078 | |
d15bcfdb | 3079 | if (idle == CPU_NOT_IDLE) |
7897986b | 3080 | load_idx = sd->busy_idx; |
d15bcfdb | 3081 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3082 | load_idx = sd->newidle_idx; |
3083 | else | |
3084 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3085 | |
3086 | do { | |
908a7c1b | 3087 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3088 | int local_group; |
3089 | int i; | |
908a7c1b | 3090 | int __group_imb = 0; |
783609c6 | 3091 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3092 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3093 | unsigned long sum_avg_load_per_task; |
3094 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3095 | |
3096 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3097 | ||
783609c6 SS |
3098 | if (local_group) |
3099 | balance_cpu = first_cpu(group->cpumask); | |
3100 | ||
1da177e4 | 3101 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3102 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3103 | sum_avg_load_per_task = avg_load_per_task = 0; |
3104 | ||
908a7c1b KC |
3105 | max_cpu_load = 0; |
3106 | min_cpu_load = ~0UL; | |
1da177e4 LT |
3107 | |
3108 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
3109 | struct rq *rq; |
3110 | ||
3111 | if (!cpu_isset(i, *cpus)) | |
3112 | continue; | |
3113 | ||
3114 | rq = cpu_rq(i); | |
2dd73a4f | 3115 | |
9439aab8 | 3116 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3117 | *sd_idle = 0; |
3118 | ||
1da177e4 | 3119 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3120 | if (local_group) { |
3121 | if (idle_cpu(i) && !first_idle_cpu) { | |
3122 | first_idle_cpu = 1; | |
3123 | balance_cpu = i; | |
3124 | } | |
3125 | ||
a2000572 | 3126 | load = target_load(i, load_idx); |
908a7c1b | 3127 | } else { |
a2000572 | 3128 | load = source_load(i, load_idx); |
908a7c1b KC |
3129 | if (load > max_cpu_load) |
3130 | max_cpu_load = load; | |
3131 | if (min_cpu_load > load) | |
3132 | min_cpu_load = load; | |
3133 | } | |
1da177e4 LT |
3134 | |
3135 | avg_load += load; | |
2dd73a4f | 3136 | sum_nr_running += rq->nr_running; |
dd41f596 | 3137 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3138 | |
3139 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3140 | } |
3141 | ||
783609c6 SS |
3142 | /* |
3143 | * First idle cpu or the first cpu(busiest) in this sched group | |
3144 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3145 | * domains. In the newly idle case, we will allow all the cpu's |
3146 | * to do the newly idle load balance. | |
783609c6 | 3147 | */ |
9439aab8 SS |
3148 | if (idle != CPU_NEWLY_IDLE && local_group && |
3149 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3150 | *balance = 0; |
3151 | goto ret; | |
3152 | } | |
3153 | ||
1da177e4 | 3154 | total_load += avg_load; |
5517d86b | 3155 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3156 | |
3157 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3158 | avg_load = sg_div_cpu_power(group, |
3159 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3160 | |
408ed066 PZ |
3161 | |
3162 | /* | |
3163 | * Consider the group unbalanced when the imbalance is larger | |
3164 | * than the average weight of two tasks. | |
3165 | * | |
3166 | * APZ: with cgroup the avg task weight can vary wildly and | |
3167 | * might not be a suitable number - should we keep a | |
3168 | * normalized nr_running number somewhere that negates | |
3169 | * the hierarchy? | |
3170 | */ | |
3171 | avg_load_per_task = sg_div_cpu_power(group, | |
3172 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3173 | ||
3174 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3175 | __group_imb = 1; |
3176 | ||
5517d86b | 3177 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3178 | |
1da177e4 LT |
3179 | if (local_group) { |
3180 | this_load = avg_load; | |
3181 | this = group; | |
2dd73a4f PW |
3182 | this_nr_running = sum_nr_running; |
3183 | this_load_per_task = sum_weighted_load; | |
3184 | } else if (avg_load > max_load && | |
908a7c1b | 3185 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3186 | max_load = avg_load; |
3187 | busiest = group; | |
2dd73a4f PW |
3188 | busiest_nr_running = sum_nr_running; |
3189 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3190 | group_imb = __group_imb; |
1da177e4 | 3191 | } |
5c45bf27 SS |
3192 | |
3193 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3194 | /* | |
3195 | * Busy processors will not participate in power savings | |
3196 | * balance. | |
3197 | */ | |
dd41f596 IM |
3198 | if (idle == CPU_NOT_IDLE || |
3199 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3200 | goto group_next; | |
5c45bf27 SS |
3201 | |
3202 | /* | |
3203 | * If the local group is idle or completely loaded | |
3204 | * no need to do power savings balance at this domain | |
3205 | */ | |
3206 | if (local_group && (this_nr_running >= group_capacity || | |
3207 | !this_nr_running)) | |
3208 | power_savings_balance = 0; | |
3209 | ||
dd41f596 | 3210 | /* |
5c45bf27 SS |
3211 | * If a group is already running at full capacity or idle, |
3212 | * don't include that group in power savings calculations | |
dd41f596 IM |
3213 | */ |
3214 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3215 | || !sum_nr_running) |
dd41f596 | 3216 | goto group_next; |
5c45bf27 | 3217 | |
dd41f596 | 3218 | /* |
5c45bf27 | 3219 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3220 | * This is the group from where we need to pick up the load |
3221 | * for saving power | |
3222 | */ | |
3223 | if ((sum_nr_running < min_nr_running) || | |
3224 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3225 | first_cpu(group->cpumask) < |
3226 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3227 | group_min = group; |
3228 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3229 | min_load_per_task = sum_weighted_load / |
3230 | sum_nr_running; | |
dd41f596 | 3231 | } |
5c45bf27 | 3232 | |
dd41f596 | 3233 | /* |
5c45bf27 | 3234 | * Calculate the group which is almost near its |
dd41f596 IM |
3235 | * capacity but still has some space to pick up some load |
3236 | * from other group and save more power | |
3237 | */ | |
3238 | if (sum_nr_running <= group_capacity - 1) { | |
3239 | if (sum_nr_running > leader_nr_running || | |
3240 | (sum_nr_running == leader_nr_running && | |
3241 | first_cpu(group->cpumask) > | |
3242 | first_cpu(group_leader->cpumask))) { | |
3243 | group_leader = group; | |
3244 | leader_nr_running = sum_nr_running; | |
3245 | } | |
48f24c4d | 3246 | } |
5c45bf27 SS |
3247 | group_next: |
3248 | #endif | |
1da177e4 LT |
3249 | group = group->next; |
3250 | } while (group != sd->groups); | |
3251 | ||
2dd73a4f | 3252 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3253 | goto out_balanced; |
3254 | ||
3255 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3256 | ||
3257 | if (this_load >= avg_load || | |
3258 | 100*max_load <= sd->imbalance_pct*this_load) | |
3259 | goto out_balanced; | |
3260 | ||
2dd73a4f | 3261 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3262 | if (group_imb) |
3263 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3264 | ||
1da177e4 LT |
3265 | /* |
3266 | * We're trying to get all the cpus to the average_load, so we don't | |
3267 | * want to push ourselves above the average load, nor do we wish to | |
3268 | * reduce the max loaded cpu below the average load, as either of these | |
3269 | * actions would just result in more rebalancing later, and ping-pong | |
3270 | * tasks around. Thus we look for the minimum possible imbalance. | |
3271 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3272 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3273 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3274 | * appear as very large values with unsigned longs. |
3275 | */ | |
2dd73a4f PW |
3276 | if (max_load <= busiest_load_per_task) |
3277 | goto out_balanced; | |
3278 | ||
3279 | /* | |
3280 | * In the presence of smp nice balancing, certain scenarios can have | |
3281 | * max load less than avg load(as we skip the groups at or below | |
3282 | * its cpu_power, while calculating max_load..) | |
3283 | */ | |
3284 | if (max_load < avg_load) { | |
3285 | *imbalance = 0; | |
3286 | goto small_imbalance; | |
3287 | } | |
0c117f1b SS |
3288 | |
3289 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3290 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3291 | |
1da177e4 | 3292 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3293 | *imbalance = min(max_pull * busiest->__cpu_power, |
3294 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3295 | / SCHED_LOAD_SCALE; |
3296 | ||
2dd73a4f PW |
3297 | /* |
3298 | * if *imbalance is less than the average load per runnable task | |
3299 | * there is no gaurantee that any tasks will be moved so we'll have | |
3300 | * a think about bumping its value to force at least one task to be | |
3301 | * moved | |
3302 | */ | |
7fd0d2dd | 3303 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3304 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3305 | unsigned int imbn; |
3306 | ||
3307 | small_imbalance: | |
3308 | pwr_move = pwr_now = 0; | |
3309 | imbn = 2; | |
3310 | if (this_nr_running) { | |
3311 | this_load_per_task /= this_nr_running; | |
3312 | if (busiest_load_per_task > this_load_per_task) | |
3313 | imbn = 1; | |
3314 | } else | |
408ed066 | 3315 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3316 | |
408ed066 | 3317 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3318 | busiest_load_per_task * imbn) { |
2dd73a4f | 3319 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3320 | return busiest; |
3321 | } | |
3322 | ||
3323 | /* | |
3324 | * OK, we don't have enough imbalance to justify moving tasks, | |
3325 | * however we may be able to increase total CPU power used by | |
3326 | * moving them. | |
3327 | */ | |
3328 | ||
5517d86b ED |
3329 | pwr_now += busiest->__cpu_power * |
3330 | min(busiest_load_per_task, max_load); | |
3331 | pwr_now += this->__cpu_power * | |
3332 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3333 | pwr_now /= SCHED_LOAD_SCALE; |
3334 | ||
3335 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3336 | tmp = sg_div_cpu_power(busiest, |
3337 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3338 | if (max_load > tmp) |
5517d86b | 3339 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3340 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3341 | |
3342 | /* Amount of load we'd add */ | |
5517d86b | 3343 | if (max_load * busiest->__cpu_power < |
33859f7f | 3344 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3345 | tmp = sg_div_cpu_power(this, |
3346 | max_load * busiest->__cpu_power); | |
1da177e4 | 3347 | else |
5517d86b ED |
3348 | tmp = sg_div_cpu_power(this, |
3349 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3350 | pwr_move += this->__cpu_power * | |
3351 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3352 | pwr_move /= SCHED_LOAD_SCALE; |
3353 | ||
3354 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3355 | if (pwr_move > pwr_now) |
3356 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3357 | } |
3358 | ||
1da177e4 LT |
3359 | return busiest; |
3360 | ||
3361 | out_balanced: | |
5c45bf27 | 3362 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3363 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3364 | goto ret; |
1da177e4 | 3365 | |
5c45bf27 SS |
3366 | if (this == group_leader && group_leader != group_min) { |
3367 | *imbalance = min_load_per_task; | |
3368 | return group_min; | |
3369 | } | |
5c45bf27 | 3370 | #endif |
783609c6 | 3371 | ret: |
1da177e4 LT |
3372 | *imbalance = 0; |
3373 | return NULL; | |
3374 | } | |
3375 | ||
3376 | /* | |
3377 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3378 | */ | |
70b97a7f | 3379 | static struct rq * |
d15bcfdb | 3380 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3381 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3382 | { |
70b97a7f | 3383 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3384 | unsigned long max_load = 0; |
1da177e4 LT |
3385 | int i; |
3386 | ||
3387 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3388 | unsigned long wl; |
0a2966b4 CL |
3389 | |
3390 | if (!cpu_isset(i, *cpus)) | |
3391 | continue; | |
3392 | ||
48f24c4d | 3393 | rq = cpu_rq(i); |
dd41f596 | 3394 | wl = weighted_cpuload(i); |
2dd73a4f | 3395 | |
dd41f596 | 3396 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3397 | continue; |
1da177e4 | 3398 | |
dd41f596 IM |
3399 | if (wl > max_load) { |
3400 | max_load = wl; | |
48f24c4d | 3401 | busiest = rq; |
1da177e4 LT |
3402 | } |
3403 | } | |
3404 | ||
3405 | return busiest; | |
3406 | } | |
3407 | ||
77391d71 NP |
3408 | /* |
3409 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3410 | * so long as it is large enough. | |
3411 | */ | |
3412 | #define MAX_PINNED_INTERVAL 512 | |
3413 | ||
1da177e4 LT |
3414 | /* |
3415 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3416 | * tasks if there is an imbalance. | |
1da177e4 | 3417 | */ |
70b97a7f | 3418 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3419 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3420 | int *balance, cpumask_t *cpus) |
1da177e4 | 3421 | { |
43010659 | 3422 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3423 | struct sched_group *group; |
1da177e4 | 3424 | unsigned long imbalance; |
70b97a7f | 3425 | struct rq *busiest; |
fe2eea3f | 3426 | unsigned long flags; |
5969fe06 | 3427 | |
7c16ec58 MT |
3428 | cpus_setall(*cpus); |
3429 | ||
89c4710e SS |
3430 | /* |
3431 | * When power savings policy is enabled for the parent domain, idle | |
3432 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3433 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3434 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3435 | */ |
d15bcfdb | 3436 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3437 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3438 | sd_idle = 1; |
1da177e4 | 3439 | |
2d72376b | 3440 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3441 | |
0a2966b4 | 3442 | redo: |
c8cba857 | 3443 | update_shares(sd); |
0a2966b4 | 3444 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3445 | cpus, balance); |
783609c6 | 3446 | |
06066714 | 3447 | if (*balance == 0) |
783609c6 | 3448 | goto out_balanced; |
783609c6 | 3449 | |
1da177e4 LT |
3450 | if (!group) { |
3451 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3452 | goto out_balanced; | |
3453 | } | |
3454 | ||
7c16ec58 | 3455 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3456 | if (!busiest) { |
3457 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3458 | goto out_balanced; | |
3459 | } | |
3460 | ||
db935dbd | 3461 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3462 | |
3463 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3464 | ||
43010659 | 3465 | ld_moved = 0; |
1da177e4 LT |
3466 | if (busiest->nr_running > 1) { |
3467 | /* | |
3468 | * Attempt to move tasks. If find_busiest_group has found | |
3469 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3470 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3471 | * correctly treated as an imbalance. |
3472 | */ | |
fe2eea3f | 3473 | local_irq_save(flags); |
e17224bf | 3474 | double_rq_lock(this_rq, busiest); |
43010659 | 3475 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3476 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3477 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3478 | local_irq_restore(flags); |
81026794 | 3479 | |
46cb4b7c SS |
3480 | /* |
3481 | * some other cpu did the load balance for us. | |
3482 | */ | |
43010659 | 3483 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3484 | resched_cpu(this_cpu); |
3485 | ||
81026794 | 3486 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3487 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3488 | cpu_clear(cpu_of(busiest), *cpus); |
3489 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3490 | goto redo; |
81026794 | 3491 | goto out_balanced; |
0a2966b4 | 3492 | } |
1da177e4 | 3493 | } |
81026794 | 3494 | |
43010659 | 3495 | if (!ld_moved) { |
1da177e4 LT |
3496 | schedstat_inc(sd, lb_failed[idle]); |
3497 | sd->nr_balance_failed++; | |
3498 | ||
3499 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3500 | |
fe2eea3f | 3501 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3502 | |
3503 | /* don't kick the migration_thread, if the curr | |
3504 | * task on busiest cpu can't be moved to this_cpu | |
3505 | */ | |
3506 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3507 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3508 | all_pinned = 1; |
3509 | goto out_one_pinned; | |
3510 | } | |
3511 | ||
1da177e4 LT |
3512 | if (!busiest->active_balance) { |
3513 | busiest->active_balance = 1; | |
3514 | busiest->push_cpu = this_cpu; | |
81026794 | 3515 | active_balance = 1; |
1da177e4 | 3516 | } |
fe2eea3f | 3517 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3518 | if (active_balance) |
1da177e4 LT |
3519 | wake_up_process(busiest->migration_thread); |
3520 | ||
3521 | /* | |
3522 | * We've kicked active balancing, reset the failure | |
3523 | * counter. | |
3524 | */ | |
39507451 | 3525 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3526 | } |
81026794 | 3527 | } else |
1da177e4 LT |
3528 | sd->nr_balance_failed = 0; |
3529 | ||
81026794 | 3530 | if (likely(!active_balance)) { |
1da177e4 LT |
3531 | /* We were unbalanced, so reset the balancing interval */ |
3532 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3533 | } else { |
3534 | /* | |
3535 | * If we've begun active balancing, start to back off. This | |
3536 | * case may not be covered by the all_pinned logic if there | |
3537 | * is only 1 task on the busy runqueue (because we don't call | |
3538 | * move_tasks). | |
3539 | */ | |
3540 | if (sd->balance_interval < sd->max_interval) | |
3541 | sd->balance_interval *= 2; | |
1da177e4 LT |
3542 | } |
3543 | ||
43010659 | 3544 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3545 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3546 | ld_moved = -1; |
3547 | ||
3548 | goto out; | |
1da177e4 LT |
3549 | |
3550 | out_balanced: | |
1da177e4 LT |
3551 | schedstat_inc(sd, lb_balanced[idle]); |
3552 | ||
16cfb1c0 | 3553 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3554 | |
3555 | out_one_pinned: | |
1da177e4 | 3556 | /* tune up the balancing interval */ |
77391d71 NP |
3557 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3558 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3559 | sd->balance_interval *= 2; |
3560 | ||
48f24c4d | 3561 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3562 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3563 | ld_moved = -1; |
3564 | else | |
3565 | ld_moved = 0; | |
3566 | out: | |
c8cba857 PZ |
3567 | if (ld_moved) |
3568 | update_shares(sd); | |
c09595f6 | 3569 | return ld_moved; |
1da177e4 LT |
3570 | } |
3571 | ||
3572 | /* | |
3573 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3574 | * tasks if there is an imbalance. | |
3575 | * | |
d15bcfdb | 3576 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3577 | * this_rq is locked. |
3578 | */ | |
48f24c4d | 3579 | static int |
7c16ec58 MT |
3580 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3581 | cpumask_t *cpus) | |
1da177e4 LT |
3582 | { |
3583 | struct sched_group *group; | |
70b97a7f | 3584 | struct rq *busiest = NULL; |
1da177e4 | 3585 | unsigned long imbalance; |
43010659 | 3586 | int ld_moved = 0; |
5969fe06 | 3587 | int sd_idle = 0; |
969bb4e4 | 3588 | int all_pinned = 0; |
7c16ec58 MT |
3589 | |
3590 | cpus_setall(*cpus); | |
5969fe06 | 3591 | |
89c4710e SS |
3592 | /* |
3593 | * When power savings policy is enabled for the parent domain, idle | |
3594 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3595 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3596 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3597 | */ |
3598 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3599 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3600 | sd_idle = 1; |
1da177e4 | 3601 | |
2d72376b | 3602 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3603 | redo: |
3e5459b4 | 3604 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3605 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3606 | &sd_idle, cpus, NULL); |
1da177e4 | 3607 | if (!group) { |
d15bcfdb | 3608 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3609 | goto out_balanced; |
1da177e4 LT |
3610 | } |
3611 | ||
7c16ec58 | 3612 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3613 | if (!busiest) { |
d15bcfdb | 3614 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3615 | goto out_balanced; |
1da177e4 LT |
3616 | } |
3617 | ||
db935dbd NP |
3618 | BUG_ON(busiest == this_rq); |
3619 | ||
d15bcfdb | 3620 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3621 | |
43010659 | 3622 | ld_moved = 0; |
d6d5cfaf NP |
3623 | if (busiest->nr_running > 1) { |
3624 | /* Attempt to move tasks */ | |
3625 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3626 | /* this_rq->clock is already updated */ |
3627 | update_rq_clock(busiest); | |
43010659 | 3628 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3629 | imbalance, sd, CPU_NEWLY_IDLE, |
3630 | &all_pinned); | |
d6d5cfaf | 3631 | spin_unlock(&busiest->lock); |
0a2966b4 | 3632 | |
969bb4e4 | 3633 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3634 | cpu_clear(cpu_of(busiest), *cpus); |
3635 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3636 | goto redo; |
3637 | } | |
d6d5cfaf NP |
3638 | } |
3639 | ||
43010659 | 3640 | if (!ld_moved) { |
d15bcfdb | 3641 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3642 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3643 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3644 | return -1; |
3645 | } else | |
16cfb1c0 | 3646 | sd->nr_balance_failed = 0; |
1da177e4 | 3647 | |
3e5459b4 | 3648 | update_shares_locked(this_rq, sd); |
43010659 | 3649 | return ld_moved; |
16cfb1c0 NP |
3650 | |
3651 | out_balanced: | |
d15bcfdb | 3652 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3653 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3654 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3655 | return -1; |
16cfb1c0 | 3656 | sd->nr_balance_failed = 0; |
48f24c4d | 3657 | |
16cfb1c0 | 3658 | return 0; |
1da177e4 LT |
3659 | } |
3660 | ||
3661 | /* | |
3662 | * idle_balance is called by schedule() if this_cpu is about to become | |
3663 | * idle. Attempts to pull tasks from other CPUs. | |
3664 | */ | |
70b97a7f | 3665 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3666 | { |
3667 | struct sched_domain *sd; | |
dd41f596 IM |
3668 | int pulled_task = -1; |
3669 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3670 | cpumask_t tmpmask; |
1da177e4 LT |
3671 | |
3672 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3673 | unsigned long interval; |
3674 | ||
3675 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3676 | continue; | |
3677 | ||
3678 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3679 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3680 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3681 | sd, &tmpmask); | |
92c4ca5c CL |
3682 | |
3683 | interval = msecs_to_jiffies(sd->balance_interval); | |
3684 | if (time_after(next_balance, sd->last_balance + interval)) | |
3685 | next_balance = sd->last_balance + interval; | |
3686 | if (pulled_task) | |
3687 | break; | |
1da177e4 | 3688 | } |
dd41f596 | 3689 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3690 | /* |
3691 | * We are going idle. next_balance may be set based on | |
3692 | * a busy processor. So reset next_balance. | |
3693 | */ | |
3694 | this_rq->next_balance = next_balance; | |
dd41f596 | 3695 | } |
1da177e4 LT |
3696 | } |
3697 | ||
3698 | /* | |
3699 | * active_load_balance is run by migration threads. It pushes running tasks | |
3700 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3701 | * running on each physical CPU where possible, and avoids physical / | |
3702 | * logical imbalances. | |
3703 | * | |
3704 | * Called with busiest_rq locked. | |
3705 | */ | |
70b97a7f | 3706 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3707 | { |
39507451 | 3708 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3709 | struct sched_domain *sd; |
3710 | struct rq *target_rq; | |
39507451 | 3711 | |
48f24c4d | 3712 | /* Is there any task to move? */ |
39507451 | 3713 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3714 | return; |
3715 | ||
3716 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3717 | |
3718 | /* | |
39507451 | 3719 | * This condition is "impossible", if it occurs |
41a2d6cf | 3720 | * we need to fix it. Originally reported by |
39507451 | 3721 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3722 | */ |
39507451 | 3723 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3724 | |
39507451 NP |
3725 | /* move a task from busiest_rq to target_rq */ |
3726 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3727 | update_rq_clock(busiest_rq); |
3728 | update_rq_clock(target_rq); | |
39507451 NP |
3729 | |
3730 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3731 | for_each_domain(target_cpu, sd) { |
39507451 | 3732 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3733 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3734 | break; |
c96d145e | 3735 | } |
39507451 | 3736 | |
48f24c4d | 3737 | if (likely(sd)) { |
2d72376b | 3738 | schedstat_inc(sd, alb_count); |
39507451 | 3739 | |
43010659 PW |
3740 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3741 | sd, CPU_IDLE)) | |
48f24c4d IM |
3742 | schedstat_inc(sd, alb_pushed); |
3743 | else | |
3744 | schedstat_inc(sd, alb_failed); | |
3745 | } | |
39507451 | 3746 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3747 | } |
3748 | ||
46cb4b7c SS |
3749 | #ifdef CONFIG_NO_HZ |
3750 | static struct { | |
3751 | atomic_t load_balancer; | |
41a2d6cf | 3752 | cpumask_t cpu_mask; |
46cb4b7c SS |
3753 | } nohz ____cacheline_aligned = { |
3754 | .load_balancer = ATOMIC_INIT(-1), | |
3755 | .cpu_mask = CPU_MASK_NONE, | |
3756 | }; | |
3757 | ||
7835b98b | 3758 | /* |
46cb4b7c SS |
3759 | * This routine will try to nominate the ilb (idle load balancing) |
3760 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3761 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3762 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3763 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3764 | * arrives... | |
3765 | * | |
3766 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3767 | * for idle load balancing. ilb owner will still be part of | |
3768 | * nohz.cpu_mask.. | |
7835b98b | 3769 | * |
46cb4b7c SS |
3770 | * While stopping the tick, this cpu will become the ilb owner if there |
3771 | * is no other owner. And will be the owner till that cpu becomes busy | |
3772 | * or if all cpus in the system stop their ticks at which point | |
3773 | * there is no need for ilb owner. | |
3774 | * | |
3775 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3776 | * next busy scheduler_tick() | |
3777 | */ | |
3778 | int select_nohz_load_balancer(int stop_tick) | |
3779 | { | |
3780 | int cpu = smp_processor_id(); | |
3781 | ||
3782 | if (stop_tick) { | |
3783 | cpu_set(cpu, nohz.cpu_mask); | |
3784 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3785 | ||
3786 | /* | |
3787 | * If we are going offline and still the leader, give up! | |
3788 | */ | |
3789 | if (cpu_is_offline(cpu) && | |
3790 | atomic_read(&nohz.load_balancer) == cpu) { | |
3791 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3792 | BUG(); | |
3793 | return 0; | |
3794 | } | |
3795 | ||
3796 | /* time for ilb owner also to sleep */ | |
3797 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3798 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3799 | atomic_set(&nohz.load_balancer, -1); | |
3800 | return 0; | |
3801 | } | |
3802 | ||
3803 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3804 | /* make me the ilb owner */ | |
3805 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3806 | return 1; | |
3807 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3808 | return 1; | |
3809 | } else { | |
3810 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3811 | return 0; | |
3812 | ||
3813 | cpu_clear(cpu, nohz.cpu_mask); | |
3814 | ||
3815 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3816 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3817 | BUG(); | |
3818 | } | |
3819 | return 0; | |
3820 | } | |
3821 | #endif | |
3822 | ||
3823 | static DEFINE_SPINLOCK(balancing); | |
3824 | ||
3825 | /* | |
7835b98b CL |
3826 | * It checks each scheduling domain to see if it is due to be balanced, |
3827 | * and initiates a balancing operation if so. | |
3828 | * | |
3829 | * Balancing parameters are set up in arch_init_sched_domains. | |
3830 | */ | |
a9957449 | 3831 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3832 | { |
46cb4b7c SS |
3833 | int balance = 1; |
3834 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3835 | unsigned long interval; |
3836 | struct sched_domain *sd; | |
46cb4b7c | 3837 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3838 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3839 | int update_next_balance = 0; |
d07355f5 | 3840 | int need_serialize; |
7c16ec58 | 3841 | cpumask_t tmp; |
1da177e4 | 3842 | |
46cb4b7c | 3843 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3844 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3845 | continue; | |
3846 | ||
3847 | interval = sd->balance_interval; | |
d15bcfdb | 3848 | if (idle != CPU_IDLE) |
1da177e4 LT |
3849 | interval *= sd->busy_factor; |
3850 | ||
3851 | /* scale ms to jiffies */ | |
3852 | interval = msecs_to_jiffies(interval); | |
3853 | if (unlikely(!interval)) | |
3854 | interval = 1; | |
dd41f596 IM |
3855 | if (interval > HZ*NR_CPUS/10) |
3856 | interval = HZ*NR_CPUS/10; | |
3857 | ||
d07355f5 | 3858 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3859 | |
d07355f5 | 3860 | if (need_serialize) { |
08c183f3 CL |
3861 | if (!spin_trylock(&balancing)) |
3862 | goto out; | |
3863 | } | |
3864 | ||
c9819f45 | 3865 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3866 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3867 | /* |
3868 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3869 | * longer idle, or one of our SMT siblings is |
3870 | * not idle. | |
3871 | */ | |
d15bcfdb | 3872 | idle = CPU_NOT_IDLE; |
1da177e4 | 3873 | } |
1bd77f2d | 3874 | sd->last_balance = jiffies; |
1da177e4 | 3875 | } |
d07355f5 | 3876 | if (need_serialize) |
08c183f3 CL |
3877 | spin_unlock(&balancing); |
3878 | out: | |
f549da84 | 3879 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3880 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3881 | update_next_balance = 1; |
3882 | } | |
783609c6 SS |
3883 | |
3884 | /* | |
3885 | * Stop the load balance at this level. There is another | |
3886 | * CPU in our sched group which is doing load balancing more | |
3887 | * actively. | |
3888 | */ | |
3889 | if (!balance) | |
3890 | break; | |
1da177e4 | 3891 | } |
f549da84 SS |
3892 | |
3893 | /* | |
3894 | * next_balance will be updated only when there is a need. | |
3895 | * When the cpu is attached to null domain for ex, it will not be | |
3896 | * updated. | |
3897 | */ | |
3898 | if (likely(update_next_balance)) | |
3899 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3900 | } |
3901 | ||
3902 | /* | |
3903 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3904 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3905 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3906 | */ | |
3907 | static void run_rebalance_domains(struct softirq_action *h) | |
3908 | { | |
dd41f596 IM |
3909 | int this_cpu = smp_processor_id(); |
3910 | struct rq *this_rq = cpu_rq(this_cpu); | |
3911 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3912 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3913 | |
dd41f596 | 3914 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3915 | |
3916 | #ifdef CONFIG_NO_HZ | |
3917 | /* | |
3918 | * If this cpu is the owner for idle load balancing, then do the | |
3919 | * balancing on behalf of the other idle cpus whose ticks are | |
3920 | * stopped. | |
3921 | */ | |
dd41f596 IM |
3922 | if (this_rq->idle_at_tick && |
3923 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3924 | cpumask_t cpus = nohz.cpu_mask; |
3925 | struct rq *rq; | |
3926 | int balance_cpu; | |
3927 | ||
dd41f596 | 3928 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3929 | for_each_cpu_mask(balance_cpu, cpus) { |
3930 | /* | |
3931 | * If this cpu gets work to do, stop the load balancing | |
3932 | * work being done for other cpus. Next load | |
3933 | * balancing owner will pick it up. | |
3934 | */ | |
3935 | if (need_resched()) | |
3936 | break; | |
3937 | ||
de0cf899 | 3938 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3939 | |
3940 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3941 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3942 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3943 | } |
3944 | } | |
3945 | #endif | |
3946 | } | |
3947 | ||
3948 | /* | |
3949 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3950 | * | |
3951 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3952 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3953 | * if the whole system is idle. | |
3954 | */ | |
dd41f596 | 3955 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3956 | { |
46cb4b7c SS |
3957 | #ifdef CONFIG_NO_HZ |
3958 | /* | |
3959 | * If we were in the nohz mode recently and busy at the current | |
3960 | * scheduler tick, then check if we need to nominate new idle | |
3961 | * load balancer. | |
3962 | */ | |
3963 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3964 | rq->in_nohz_recently = 0; | |
3965 | ||
3966 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3967 | cpu_clear(cpu, nohz.cpu_mask); | |
3968 | atomic_set(&nohz.load_balancer, -1); | |
3969 | } | |
3970 | ||
3971 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3972 | /* | |
3973 | * simple selection for now: Nominate the | |
3974 | * first cpu in the nohz list to be the next | |
3975 | * ilb owner. | |
3976 | * | |
3977 | * TBD: Traverse the sched domains and nominate | |
3978 | * the nearest cpu in the nohz.cpu_mask. | |
3979 | */ | |
3980 | int ilb = first_cpu(nohz.cpu_mask); | |
3981 | ||
434d53b0 | 3982 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3983 | resched_cpu(ilb); |
3984 | } | |
3985 | } | |
3986 | ||
3987 | /* | |
3988 | * If this cpu is idle and doing idle load balancing for all the | |
3989 | * cpus with ticks stopped, is it time for that to stop? | |
3990 | */ | |
3991 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3992 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3993 | resched_cpu(cpu); | |
3994 | return; | |
3995 | } | |
3996 | ||
3997 | /* | |
3998 | * If this cpu is idle and the idle load balancing is done by | |
3999 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4000 | */ | |
4001 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4002 | cpu_isset(cpu, nohz.cpu_mask)) | |
4003 | return; | |
4004 | #endif | |
4005 | if (time_after_eq(jiffies, rq->next_balance)) | |
4006 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4007 | } |
dd41f596 IM |
4008 | |
4009 | #else /* CONFIG_SMP */ | |
4010 | ||
1da177e4 LT |
4011 | /* |
4012 | * on UP we do not need to balance between CPUs: | |
4013 | */ | |
70b97a7f | 4014 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4015 | { |
4016 | } | |
dd41f596 | 4017 | |
1da177e4 LT |
4018 | #endif |
4019 | ||
1da177e4 LT |
4020 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4021 | ||
4022 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4023 | ||
4024 | /* | |
41b86e9c IM |
4025 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4026 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4027 | */ |
41b86e9c | 4028 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4029 | { |
1da177e4 | 4030 | unsigned long flags; |
41b86e9c IM |
4031 | u64 ns, delta_exec; |
4032 | struct rq *rq; | |
48f24c4d | 4033 | |
41b86e9c IM |
4034 | rq = task_rq_lock(p, &flags); |
4035 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4036 | if (task_current(rq, p)) { |
a8e504d2 IM |
4037 | update_rq_clock(rq); |
4038 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4039 | if ((s64)delta_exec > 0) |
4040 | ns += delta_exec; | |
4041 | } | |
4042 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4043 | |
1da177e4 LT |
4044 | return ns; |
4045 | } | |
4046 | ||
1da177e4 LT |
4047 | /* |
4048 | * Account user cpu time to a process. | |
4049 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4050 | * @cputime: the cpu time spent in user space since the last update |
4051 | */ | |
4052 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4053 | { | |
4054 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4055 | cputime64_t tmp; | |
4056 | ||
4057 | p->utime = cputime_add(p->utime, cputime); | |
4058 | ||
4059 | /* Add user time to cpustat. */ | |
4060 | tmp = cputime_to_cputime64(cputime); | |
4061 | if (TASK_NICE(p) > 0) | |
4062 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4063 | else | |
4064 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4065 | } | |
4066 | ||
94886b84 LV |
4067 | /* |
4068 | * Account guest cpu time to a process. | |
4069 | * @p: the process that the cpu time gets accounted to | |
4070 | * @cputime: the cpu time spent in virtual machine since the last update | |
4071 | */ | |
f7402e03 | 4072 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4073 | { |
4074 | cputime64_t tmp; | |
4075 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4076 | ||
4077 | tmp = cputime_to_cputime64(cputime); | |
4078 | ||
4079 | p->utime = cputime_add(p->utime, cputime); | |
4080 | p->gtime = cputime_add(p->gtime, cputime); | |
4081 | ||
4082 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4083 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4084 | } | |
4085 | ||
c66f08be MN |
4086 | /* |
4087 | * Account scaled user cpu time to a process. | |
4088 | * @p: the process that the cpu time gets accounted to | |
4089 | * @cputime: the cpu time spent in user space since the last update | |
4090 | */ | |
4091 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4092 | { | |
4093 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4094 | } | |
4095 | ||
1da177e4 LT |
4096 | /* |
4097 | * Account system cpu time to a process. | |
4098 | * @p: the process that the cpu time gets accounted to | |
4099 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4100 | * @cputime: the cpu time spent in kernel space since the last update | |
4101 | */ | |
4102 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4103 | cputime_t cputime) | |
4104 | { | |
4105 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4106 | struct rq *rq = this_rq(); |
1da177e4 LT |
4107 | cputime64_t tmp; |
4108 | ||
983ed7a6 HH |
4109 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4110 | account_guest_time(p, cputime); | |
4111 | return; | |
4112 | } | |
94886b84 | 4113 | |
1da177e4 LT |
4114 | p->stime = cputime_add(p->stime, cputime); |
4115 | ||
4116 | /* Add system time to cpustat. */ | |
4117 | tmp = cputime_to_cputime64(cputime); | |
4118 | if (hardirq_count() - hardirq_offset) | |
4119 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4120 | else if (softirq_count()) | |
4121 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4122 | else if (p != rq->idle) |
1da177e4 | 4123 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4124 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4125 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4126 | else | |
4127 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4128 | /* Account for system time used */ | |
4129 | acct_update_integrals(p); | |
1da177e4 LT |
4130 | } |
4131 | ||
c66f08be MN |
4132 | /* |
4133 | * Account scaled system cpu time to a process. | |
4134 | * @p: the process that the cpu time gets accounted to | |
4135 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4136 | * @cputime: the cpu time spent in kernel space since the last update | |
4137 | */ | |
4138 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4139 | { | |
4140 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4141 | } | |
4142 | ||
1da177e4 LT |
4143 | /* |
4144 | * Account for involuntary wait time. | |
4145 | * @p: the process from which the cpu time has been stolen | |
4146 | * @steal: the cpu time spent in involuntary wait | |
4147 | */ | |
4148 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4149 | { | |
4150 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4151 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4152 | struct rq *rq = this_rq(); |
1da177e4 LT |
4153 | |
4154 | if (p == rq->idle) { | |
4155 | p->stime = cputime_add(p->stime, steal); | |
4156 | if (atomic_read(&rq->nr_iowait) > 0) | |
4157 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4158 | else | |
4159 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4160 | } else |
1da177e4 LT |
4161 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4162 | } | |
4163 | ||
7835b98b CL |
4164 | /* |
4165 | * This function gets called by the timer code, with HZ frequency. | |
4166 | * We call it with interrupts disabled. | |
4167 | * | |
4168 | * It also gets called by the fork code, when changing the parent's | |
4169 | * timeslices. | |
4170 | */ | |
4171 | void scheduler_tick(void) | |
4172 | { | |
7835b98b CL |
4173 | int cpu = smp_processor_id(); |
4174 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4175 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4176 | |
4177 | sched_clock_tick(); | |
dd41f596 IM |
4178 | |
4179 | spin_lock(&rq->lock); | |
3e51f33f | 4180 | update_rq_clock(rq); |
f1a438d8 | 4181 | update_cpu_load(rq); |
fa85ae24 | 4182 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4183 | spin_unlock(&rq->lock); |
7835b98b | 4184 | |
e418e1c2 | 4185 | #ifdef CONFIG_SMP |
dd41f596 IM |
4186 | rq->idle_at_tick = idle_cpu(cpu); |
4187 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4188 | #endif |
1da177e4 LT |
4189 | } |
4190 | ||
1da177e4 LT |
4191 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
4192 | ||
43627582 | 4193 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
4194 | { |
4195 | /* | |
4196 | * Underflow? | |
4197 | */ | |
9a11b49a IM |
4198 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4199 | return; | |
1da177e4 LT |
4200 | preempt_count() += val; |
4201 | /* | |
4202 | * Spinlock count overflowing soon? | |
4203 | */ | |
33859f7f MOS |
4204 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4205 | PREEMPT_MASK - 10); | |
1da177e4 LT |
4206 | } |
4207 | EXPORT_SYMBOL(add_preempt_count); | |
4208 | ||
43627582 | 4209 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
4210 | { |
4211 | /* | |
4212 | * Underflow? | |
4213 | */ | |
9a11b49a IM |
4214 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4215 | return; | |
1da177e4 LT |
4216 | /* |
4217 | * Is the spinlock portion underflowing? | |
4218 | */ | |
9a11b49a IM |
4219 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4220 | !(preempt_count() & PREEMPT_MASK))) | |
4221 | return; | |
4222 | ||
1da177e4 LT |
4223 | preempt_count() -= val; |
4224 | } | |
4225 | EXPORT_SYMBOL(sub_preempt_count); | |
4226 | ||
4227 | #endif | |
4228 | ||
4229 | /* | |
dd41f596 | 4230 | * Print scheduling while atomic bug: |
1da177e4 | 4231 | */ |
dd41f596 | 4232 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4233 | { |
838225b4 SS |
4234 | struct pt_regs *regs = get_irq_regs(); |
4235 | ||
4236 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4237 | prev->comm, prev->pid, preempt_count()); | |
4238 | ||
dd41f596 | 4239 | debug_show_held_locks(prev); |
e21f5b15 | 4240 | print_modules(); |
dd41f596 IM |
4241 | if (irqs_disabled()) |
4242 | print_irqtrace_events(prev); | |
838225b4 SS |
4243 | |
4244 | if (regs) | |
4245 | show_regs(regs); | |
4246 | else | |
4247 | dump_stack(); | |
dd41f596 | 4248 | } |
1da177e4 | 4249 | |
dd41f596 IM |
4250 | /* |
4251 | * Various schedule()-time debugging checks and statistics: | |
4252 | */ | |
4253 | static inline void schedule_debug(struct task_struct *prev) | |
4254 | { | |
1da177e4 | 4255 | /* |
41a2d6cf | 4256 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4257 | * schedule() atomically, we ignore that path for now. |
4258 | * Otherwise, whine if we are scheduling when we should not be. | |
4259 | */ | |
3f33a7ce | 4260 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4261 | __schedule_bug(prev); |
4262 | ||
1da177e4 LT |
4263 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4264 | ||
2d72376b | 4265 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4266 | #ifdef CONFIG_SCHEDSTATS |
4267 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4268 | schedstat_inc(this_rq(), bkl_count); |
4269 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4270 | } |
4271 | #endif | |
dd41f596 IM |
4272 | } |
4273 | ||
4274 | /* | |
4275 | * Pick up the highest-prio task: | |
4276 | */ | |
4277 | static inline struct task_struct * | |
ff95f3df | 4278 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4279 | { |
5522d5d5 | 4280 | const struct sched_class *class; |
dd41f596 | 4281 | struct task_struct *p; |
1da177e4 LT |
4282 | |
4283 | /* | |
dd41f596 IM |
4284 | * Optimization: we know that if all tasks are in |
4285 | * the fair class we can call that function directly: | |
1da177e4 | 4286 | */ |
dd41f596 | 4287 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4288 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4289 | if (likely(p)) |
4290 | return p; | |
1da177e4 LT |
4291 | } |
4292 | ||
dd41f596 IM |
4293 | class = sched_class_highest; |
4294 | for ( ; ; ) { | |
fb8d4724 | 4295 | p = class->pick_next_task(rq); |
dd41f596 IM |
4296 | if (p) |
4297 | return p; | |
4298 | /* | |
4299 | * Will never be NULL as the idle class always | |
4300 | * returns a non-NULL p: | |
4301 | */ | |
4302 | class = class->next; | |
4303 | } | |
4304 | } | |
1da177e4 | 4305 | |
dd41f596 IM |
4306 | /* |
4307 | * schedule() is the main scheduler function. | |
4308 | */ | |
4309 | asmlinkage void __sched schedule(void) | |
4310 | { | |
4311 | struct task_struct *prev, *next; | |
67ca7bde | 4312 | unsigned long *switch_count; |
dd41f596 | 4313 | struct rq *rq; |
f333fdc9 | 4314 | int cpu, hrtick = sched_feat(HRTICK); |
dd41f596 IM |
4315 | |
4316 | need_resched: | |
4317 | preempt_disable(); | |
4318 | cpu = smp_processor_id(); | |
4319 | rq = cpu_rq(cpu); | |
4320 | rcu_qsctr_inc(cpu); | |
4321 | prev = rq->curr; | |
4322 | switch_count = &prev->nivcsw; | |
4323 | ||
4324 | release_kernel_lock(prev); | |
4325 | need_resched_nonpreemptible: | |
4326 | ||
4327 | schedule_debug(prev); | |
1da177e4 | 4328 | |
f333fdc9 MG |
4329 | if (hrtick) |
4330 | hrtick_clear(rq); | |
8f4d37ec | 4331 | |
1e819950 IM |
4332 | /* |
4333 | * Do the rq-clock update outside the rq lock: | |
4334 | */ | |
4335 | local_irq_disable(); | |
3e51f33f | 4336 | update_rq_clock(rq); |
1e819950 IM |
4337 | spin_lock(&rq->lock); |
4338 | clear_tsk_need_resched(prev); | |
1da177e4 | 4339 | |
1da177e4 | 4340 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4341 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4342 | prev->state = TASK_RUNNING; |
16882c1e | 4343 | else |
2e1cb74a | 4344 | deactivate_task(rq, prev, 1); |
dd41f596 | 4345 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4346 | } |
4347 | ||
9a897c5a SR |
4348 | #ifdef CONFIG_SMP |
4349 | if (prev->sched_class->pre_schedule) | |
4350 | prev->sched_class->pre_schedule(rq, prev); | |
4351 | #endif | |
f65eda4f | 4352 | |
dd41f596 | 4353 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4354 | idle_balance(cpu, rq); |
1da177e4 | 4355 | |
31ee529c | 4356 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4357 | next = pick_next_task(rq, prev); |
1da177e4 | 4358 | |
1da177e4 | 4359 | if (likely(prev != next)) { |
673a90a1 DS |
4360 | sched_info_switch(prev, next); |
4361 | ||
1da177e4 LT |
4362 | rq->nr_switches++; |
4363 | rq->curr = next; | |
4364 | ++*switch_count; | |
4365 | ||
dd41f596 | 4366 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4367 | /* |
4368 | * the context switch might have flipped the stack from under | |
4369 | * us, hence refresh the local variables. | |
4370 | */ | |
4371 | cpu = smp_processor_id(); | |
4372 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4373 | } else |
4374 | spin_unlock_irq(&rq->lock); | |
4375 | ||
f333fdc9 MG |
4376 | if (hrtick) |
4377 | hrtick_set(rq); | |
8f4d37ec PZ |
4378 | |
4379 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4380 | goto need_resched_nonpreemptible; |
8f4d37ec | 4381 | |
1da177e4 LT |
4382 | preempt_enable_no_resched(); |
4383 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4384 | goto need_resched; | |
4385 | } | |
1da177e4 LT |
4386 | EXPORT_SYMBOL(schedule); |
4387 | ||
4388 | #ifdef CONFIG_PREEMPT | |
4389 | /* | |
2ed6e34f | 4390 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4391 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4392 | * occur there and call schedule directly. |
4393 | */ | |
4394 | asmlinkage void __sched preempt_schedule(void) | |
4395 | { | |
4396 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4397 | |
1da177e4 LT |
4398 | /* |
4399 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4400 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4401 | */ |
beed33a8 | 4402 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4403 | return; |
4404 | ||
3a5c359a AK |
4405 | do { |
4406 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4407 | schedule(); |
3a5c359a | 4408 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4409 | |
3a5c359a AK |
4410 | /* |
4411 | * Check again in case we missed a preemption opportunity | |
4412 | * between schedule and now. | |
4413 | */ | |
4414 | barrier(); | |
4415 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4416 | } |
1da177e4 LT |
4417 | EXPORT_SYMBOL(preempt_schedule); |
4418 | ||
4419 | /* | |
2ed6e34f | 4420 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4421 | * off of irq context. |
4422 | * Note, that this is called and return with irqs disabled. This will | |
4423 | * protect us against recursive calling from irq. | |
4424 | */ | |
4425 | asmlinkage void __sched preempt_schedule_irq(void) | |
4426 | { | |
4427 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4428 | |
2ed6e34f | 4429 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4430 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4431 | ||
3a5c359a AK |
4432 | do { |
4433 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4434 | local_irq_enable(); |
4435 | schedule(); | |
4436 | local_irq_disable(); | |
3a5c359a | 4437 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4438 | |
3a5c359a AK |
4439 | /* |
4440 | * Check again in case we missed a preemption opportunity | |
4441 | * between schedule and now. | |
4442 | */ | |
4443 | barrier(); | |
4444 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4445 | } |
4446 | ||
4447 | #endif /* CONFIG_PREEMPT */ | |
4448 | ||
95cdf3b7 IM |
4449 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4450 | void *key) | |
1da177e4 | 4451 | { |
48f24c4d | 4452 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4453 | } |
1da177e4 LT |
4454 | EXPORT_SYMBOL(default_wake_function); |
4455 | ||
4456 | /* | |
41a2d6cf IM |
4457 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4458 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4459 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4460 | * | |
4461 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4462 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4463 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4464 | */ | |
4465 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4466 | int nr_exclusive, int sync, void *key) | |
4467 | { | |
2e45874c | 4468 | wait_queue_t *curr, *next; |
1da177e4 | 4469 | |
2e45874c | 4470 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4471 | unsigned flags = curr->flags; |
4472 | ||
1da177e4 | 4473 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4474 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4475 | break; |
4476 | } | |
4477 | } | |
4478 | ||
4479 | /** | |
4480 | * __wake_up - wake up threads blocked on a waitqueue. | |
4481 | * @q: the waitqueue | |
4482 | * @mode: which threads | |
4483 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4484 | * @key: is directly passed to the wakeup function |
1da177e4 | 4485 | */ |
7ad5b3a5 | 4486 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4487 | int nr_exclusive, void *key) |
1da177e4 LT |
4488 | { |
4489 | unsigned long flags; | |
4490 | ||
4491 | spin_lock_irqsave(&q->lock, flags); | |
4492 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4493 | spin_unlock_irqrestore(&q->lock, flags); | |
4494 | } | |
1da177e4 LT |
4495 | EXPORT_SYMBOL(__wake_up); |
4496 | ||
4497 | /* | |
4498 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4499 | */ | |
7ad5b3a5 | 4500 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4501 | { |
4502 | __wake_up_common(q, mode, 1, 0, NULL); | |
4503 | } | |
4504 | ||
4505 | /** | |
67be2dd1 | 4506 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4507 | * @q: the waitqueue |
4508 | * @mode: which threads | |
4509 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4510 | * | |
4511 | * The sync wakeup differs that the waker knows that it will schedule | |
4512 | * away soon, so while the target thread will be woken up, it will not | |
4513 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4514 | * with each other. This can prevent needless bouncing between CPUs. | |
4515 | * | |
4516 | * On UP it can prevent extra preemption. | |
4517 | */ | |
7ad5b3a5 | 4518 | void |
95cdf3b7 | 4519 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4520 | { |
4521 | unsigned long flags; | |
4522 | int sync = 1; | |
4523 | ||
4524 | if (unlikely(!q)) | |
4525 | return; | |
4526 | ||
4527 | if (unlikely(!nr_exclusive)) | |
4528 | sync = 0; | |
4529 | ||
4530 | spin_lock_irqsave(&q->lock, flags); | |
4531 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4532 | spin_unlock_irqrestore(&q->lock, flags); | |
4533 | } | |
4534 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4535 | ||
b15136e9 | 4536 | void complete(struct completion *x) |
1da177e4 LT |
4537 | { |
4538 | unsigned long flags; | |
4539 | ||
4540 | spin_lock_irqsave(&x->wait.lock, flags); | |
4541 | x->done++; | |
d9514f6c | 4542 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4543 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4544 | } | |
4545 | EXPORT_SYMBOL(complete); | |
4546 | ||
b15136e9 | 4547 | void complete_all(struct completion *x) |
1da177e4 LT |
4548 | { |
4549 | unsigned long flags; | |
4550 | ||
4551 | spin_lock_irqsave(&x->wait.lock, flags); | |
4552 | x->done += UINT_MAX/2; | |
d9514f6c | 4553 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4554 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4555 | } | |
4556 | EXPORT_SYMBOL(complete_all); | |
4557 | ||
8cbbe86d AK |
4558 | static inline long __sched |
4559 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4560 | { |
1da177e4 LT |
4561 | if (!x->done) { |
4562 | DECLARE_WAITQUEUE(wait, current); | |
4563 | ||
4564 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4565 | __add_wait_queue_tail(&x->wait, &wait); | |
4566 | do { | |
009e577e MW |
4567 | if ((state == TASK_INTERRUPTIBLE && |
4568 | signal_pending(current)) || | |
4569 | (state == TASK_KILLABLE && | |
4570 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4571 | timeout = -ERESTARTSYS; |
4572 | break; | |
8cbbe86d AK |
4573 | } |
4574 | __set_current_state(state); | |
1da177e4 LT |
4575 | spin_unlock_irq(&x->wait.lock); |
4576 | timeout = schedule_timeout(timeout); | |
4577 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4578 | } while (!x->done && timeout); |
1da177e4 | 4579 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4580 | if (!x->done) |
4581 | return timeout; | |
1da177e4 LT |
4582 | } |
4583 | x->done--; | |
ea71a546 | 4584 | return timeout ?: 1; |
1da177e4 | 4585 | } |
1da177e4 | 4586 | |
8cbbe86d AK |
4587 | static long __sched |
4588 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4589 | { |
1da177e4 LT |
4590 | might_sleep(); |
4591 | ||
4592 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4593 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4594 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4595 | return timeout; |
4596 | } | |
1da177e4 | 4597 | |
b15136e9 | 4598 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4599 | { |
4600 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4601 | } |
8cbbe86d | 4602 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4603 | |
b15136e9 | 4604 | unsigned long __sched |
8cbbe86d | 4605 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4606 | { |
8cbbe86d | 4607 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4608 | } |
8cbbe86d | 4609 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4610 | |
8cbbe86d | 4611 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4612 | { |
51e97990 AK |
4613 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4614 | if (t == -ERESTARTSYS) | |
4615 | return t; | |
4616 | return 0; | |
0fec171c | 4617 | } |
8cbbe86d | 4618 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4619 | |
b15136e9 | 4620 | unsigned long __sched |
8cbbe86d AK |
4621 | wait_for_completion_interruptible_timeout(struct completion *x, |
4622 | unsigned long timeout) | |
0fec171c | 4623 | { |
8cbbe86d | 4624 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4625 | } |
8cbbe86d | 4626 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4627 | |
009e577e MW |
4628 | int __sched wait_for_completion_killable(struct completion *x) |
4629 | { | |
4630 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4631 | if (t == -ERESTARTSYS) | |
4632 | return t; | |
4633 | return 0; | |
4634 | } | |
4635 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4636 | ||
8cbbe86d AK |
4637 | static long __sched |
4638 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4639 | { |
0fec171c IM |
4640 | unsigned long flags; |
4641 | wait_queue_t wait; | |
4642 | ||
4643 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4644 | |
8cbbe86d | 4645 | __set_current_state(state); |
1da177e4 | 4646 | |
8cbbe86d AK |
4647 | spin_lock_irqsave(&q->lock, flags); |
4648 | __add_wait_queue(q, &wait); | |
4649 | spin_unlock(&q->lock); | |
4650 | timeout = schedule_timeout(timeout); | |
4651 | spin_lock_irq(&q->lock); | |
4652 | __remove_wait_queue(q, &wait); | |
4653 | spin_unlock_irqrestore(&q->lock, flags); | |
4654 | ||
4655 | return timeout; | |
4656 | } | |
4657 | ||
4658 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4659 | { | |
4660 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4661 | } |
1da177e4 LT |
4662 | EXPORT_SYMBOL(interruptible_sleep_on); |
4663 | ||
0fec171c | 4664 | long __sched |
95cdf3b7 | 4665 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4666 | { |
8cbbe86d | 4667 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4668 | } |
1da177e4 LT |
4669 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4670 | ||
0fec171c | 4671 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4672 | { |
8cbbe86d | 4673 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4674 | } |
1da177e4 LT |
4675 | EXPORT_SYMBOL(sleep_on); |
4676 | ||
0fec171c | 4677 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4678 | { |
8cbbe86d | 4679 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4680 | } |
1da177e4 LT |
4681 | EXPORT_SYMBOL(sleep_on_timeout); |
4682 | ||
b29739f9 IM |
4683 | #ifdef CONFIG_RT_MUTEXES |
4684 | ||
4685 | /* | |
4686 | * rt_mutex_setprio - set the current priority of a task | |
4687 | * @p: task | |
4688 | * @prio: prio value (kernel-internal form) | |
4689 | * | |
4690 | * This function changes the 'effective' priority of a task. It does | |
4691 | * not touch ->normal_prio like __setscheduler(). | |
4692 | * | |
4693 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4694 | */ | |
36c8b586 | 4695 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4696 | { |
4697 | unsigned long flags; | |
83b699ed | 4698 | int oldprio, on_rq, running; |
70b97a7f | 4699 | struct rq *rq; |
cb469845 | 4700 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4701 | |
4702 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4703 | ||
4704 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4705 | update_rq_clock(rq); |
b29739f9 | 4706 | |
d5f9f942 | 4707 | oldprio = p->prio; |
dd41f596 | 4708 | on_rq = p->se.on_rq; |
051a1d1a | 4709 | running = task_current(rq, p); |
0e1f3483 | 4710 | if (on_rq) |
69be72c1 | 4711 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4712 | if (running) |
4713 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4714 | |
4715 | if (rt_prio(prio)) | |
4716 | p->sched_class = &rt_sched_class; | |
4717 | else | |
4718 | p->sched_class = &fair_sched_class; | |
4719 | ||
b29739f9 IM |
4720 | p->prio = prio; |
4721 | ||
0e1f3483 HS |
4722 | if (running) |
4723 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4724 | if (on_rq) { |
8159f87e | 4725 | enqueue_task(rq, p, 0); |
cb469845 SR |
4726 | |
4727 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4728 | } |
4729 | task_rq_unlock(rq, &flags); | |
4730 | } | |
4731 | ||
4732 | #endif | |
4733 | ||
36c8b586 | 4734 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4735 | { |
dd41f596 | 4736 | int old_prio, delta, on_rq; |
1da177e4 | 4737 | unsigned long flags; |
70b97a7f | 4738 | struct rq *rq; |
1da177e4 LT |
4739 | |
4740 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4741 | return; | |
4742 | /* | |
4743 | * We have to be careful, if called from sys_setpriority(), | |
4744 | * the task might be in the middle of scheduling on another CPU. | |
4745 | */ | |
4746 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4747 | update_rq_clock(rq); |
1da177e4 LT |
4748 | /* |
4749 | * The RT priorities are set via sched_setscheduler(), but we still | |
4750 | * allow the 'normal' nice value to be set - but as expected | |
4751 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4752 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4753 | */ |
e05606d3 | 4754 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4755 | p->static_prio = NICE_TO_PRIO(nice); |
4756 | goto out_unlock; | |
4757 | } | |
dd41f596 | 4758 | on_rq = p->se.on_rq; |
c09595f6 | 4759 | if (on_rq) |
69be72c1 | 4760 | dequeue_task(rq, p, 0); |
1da177e4 | 4761 | |
1da177e4 | 4762 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4763 | set_load_weight(p); |
b29739f9 IM |
4764 | old_prio = p->prio; |
4765 | p->prio = effective_prio(p); | |
4766 | delta = p->prio - old_prio; | |
1da177e4 | 4767 | |
dd41f596 | 4768 | if (on_rq) { |
8159f87e | 4769 | enqueue_task(rq, p, 0); |
1da177e4 | 4770 | /* |
d5f9f942 AM |
4771 | * If the task increased its priority or is running and |
4772 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4773 | */ |
d5f9f942 | 4774 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4775 | resched_task(rq->curr); |
4776 | } | |
4777 | out_unlock: | |
4778 | task_rq_unlock(rq, &flags); | |
4779 | } | |
1da177e4 LT |
4780 | EXPORT_SYMBOL(set_user_nice); |
4781 | ||
e43379f1 MM |
4782 | /* |
4783 | * can_nice - check if a task can reduce its nice value | |
4784 | * @p: task | |
4785 | * @nice: nice value | |
4786 | */ | |
36c8b586 | 4787 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4788 | { |
024f4747 MM |
4789 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4790 | int nice_rlim = 20 - nice; | |
48f24c4d | 4791 | |
e43379f1 MM |
4792 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4793 | capable(CAP_SYS_NICE)); | |
4794 | } | |
4795 | ||
1da177e4 LT |
4796 | #ifdef __ARCH_WANT_SYS_NICE |
4797 | ||
4798 | /* | |
4799 | * sys_nice - change the priority of the current process. | |
4800 | * @increment: priority increment | |
4801 | * | |
4802 | * sys_setpriority is a more generic, but much slower function that | |
4803 | * does similar things. | |
4804 | */ | |
4805 | asmlinkage long sys_nice(int increment) | |
4806 | { | |
48f24c4d | 4807 | long nice, retval; |
1da177e4 LT |
4808 | |
4809 | /* | |
4810 | * Setpriority might change our priority at the same moment. | |
4811 | * We don't have to worry. Conceptually one call occurs first | |
4812 | * and we have a single winner. | |
4813 | */ | |
e43379f1 MM |
4814 | if (increment < -40) |
4815 | increment = -40; | |
1da177e4 LT |
4816 | if (increment > 40) |
4817 | increment = 40; | |
4818 | ||
4819 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4820 | if (nice < -20) | |
4821 | nice = -20; | |
4822 | if (nice > 19) | |
4823 | nice = 19; | |
4824 | ||
e43379f1 MM |
4825 | if (increment < 0 && !can_nice(current, nice)) |
4826 | return -EPERM; | |
4827 | ||
1da177e4 LT |
4828 | retval = security_task_setnice(current, nice); |
4829 | if (retval) | |
4830 | return retval; | |
4831 | ||
4832 | set_user_nice(current, nice); | |
4833 | return 0; | |
4834 | } | |
4835 | ||
4836 | #endif | |
4837 | ||
4838 | /** | |
4839 | * task_prio - return the priority value of a given task. | |
4840 | * @p: the task in question. | |
4841 | * | |
4842 | * This is the priority value as seen by users in /proc. | |
4843 | * RT tasks are offset by -200. Normal tasks are centered | |
4844 | * around 0, value goes from -16 to +15. | |
4845 | */ | |
36c8b586 | 4846 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4847 | { |
4848 | return p->prio - MAX_RT_PRIO; | |
4849 | } | |
4850 | ||
4851 | /** | |
4852 | * task_nice - return the nice value of a given task. | |
4853 | * @p: the task in question. | |
4854 | */ | |
36c8b586 | 4855 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4856 | { |
4857 | return TASK_NICE(p); | |
4858 | } | |
150d8bed | 4859 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4860 | |
4861 | /** | |
4862 | * idle_cpu - is a given cpu idle currently? | |
4863 | * @cpu: the processor in question. | |
4864 | */ | |
4865 | int idle_cpu(int cpu) | |
4866 | { | |
4867 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4868 | } | |
4869 | ||
1da177e4 LT |
4870 | /** |
4871 | * idle_task - return the idle task for a given cpu. | |
4872 | * @cpu: the processor in question. | |
4873 | */ | |
36c8b586 | 4874 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4875 | { |
4876 | return cpu_rq(cpu)->idle; | |
4877 | } | |
4878 | ||
4879 | /** | |
4880 | * find_process_by_pid - find a process with a matching PID value. | |
4881 | * @pid: the pid in question. | |
4882 | */ | |
a9957449 | 4883 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4884 | { |
228ebcbe | 4885 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4886 | } |
4887 | ||
4888 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4889 | static void |
4890 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4891 | { |
dd41f596 | 4892 | BUG_ON(p->se.on_rq); |
48f24c4d | 4893 | |
1da177e4 | 4894 | p->policy = policy; |
dd41f596 IM |
4895 | switch (p->policy) { |
4896 | case SCHED_NORMAL: | |
4897 | case SCHED_BATCH: | |
4898 | case SCHED_IDLE: | |
4899 | p->sched_class = &fair_sched_class; | |
4900 | break; | |
4901 | case SCHED_FIFO: | |
4902 | case SCHED_RR: | |
4903 | p->sched_class = &rt_sched_class; | |
4904 | break; | |
4905 | } | |
4906 | ||
1da177e4 | 4907 | p->rt_priority = prio; |
b29739f9 IM |
4908 | p->normal_prio = normal_prio(p); |
4909 | /* we are holding p->pi_lock already */ | |
4910 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4911 | set_load_weight(p); |
1da177e4 LT |
4912 | } |
4913 | ||
4914 | /** | |
72fd4a35 | 4915 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4916 | * @p: the task in question. |
4917 | * @policy: new policy. | |
4918 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4919 | * |
72fd4a35 | 4920 | * NOTE that the task may be already dead. |
1da177e4 | 4921 | */ |
95cdf3b7 IM |
4922 | int sched_setscheduler(struct task_struct *p, int policy, |
4923 | struct sched_param *param) | |
1da177e4 | 4924 | { |
83b699ed | 4925 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4926 | unsigned long flags; |
cb469845 | 4927 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4928 | struct rq *rq; |
1da177e4 | 4929 | |
66e5393a SR |
4930 | /* may grab non-irq protected spin_locks */ |
4931 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4932 | recheck: |
4933 | /* double check policy once rq lock held */ | |
4934 | if (policy < 0) | |
4935 | policy = oldpolicy = p->policy; | |
4936 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4937 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4938 | policy != SCHED_IDLE) | |
b0a9499c | 4939 | return -EINVAL; |
1da177e4 LT |
4940 | /* |
4941 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4942 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4943 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4944 | */ |
4945 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4946 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4947 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4948 | return -EINVAL; |
e05606d3 | 4949 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4950 | return -EINVAL; |
4951 | ||
37e4ab3f OC |
4952 | /* |
4953 | * Allow unprivileged RT tasks to decrease priority: | |
4954 | */ | |
4955 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4956 | if (rt_policy(policy)) { |
8dc3e909 | 4957 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4958 | |
4959 | if (!lock_task_sighand(p, &flags)) | |
4960 | return -ESRCH; | |
4961 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4962 | unlock_task_sighand(p, &flags); | |
4963 | ||
4964 | /* can't set/change the rt policy */ | |
4965 | if (policy != p->policy && !rlim_rtprio) | |
4966 | return -EPERM; | |
4967 | ||
4968 | /* can't increase priority */ | |
4969 | if (param->sched_priority > p->rt_priority && | |
4970 | param->sched_priority > rlim_rtprio) | |
4971 | return -EPERM; | |
4972 | } | |
dd41f596 IM |
4973 | /* |
4974 | * Like positive nice levels, dont allow tasks to | |
4975 | * move out of SCHED_IDLE either: | |
4976 | */ | |
4977 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4978 | return -EPERM; | |
5fe1d75f | 4979 | |
37e4ab3f OC |
4980 | /* can't change other user's priorities */ |
4981 | if ((current->euid != p->euid) && | |
4982 | (current->euid != p->uid)) | |
4983 | return -EPERM; | |
4984 | } | |
1da177e4 | 4985 | |
b68aa230 PZ |
4986 | #ifdef CONFIG_RT_GROUP_SCHED |
4987 | /* | |
4988 | * Do not allow realtime tasks into groups that have no runtime | |
4989 | * assigned. | |
4990 | */ | |
d0b27fa7 | 4991 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4992 | return -EPERM; |
4993 | #endif | |
4994 | ||
1da177e4 LT |
4995 | retval = security_task_setscheduler(p, policy, param); |
4996 | if (retval) | |
4997 | return retval; | |
b29739f9 IM |
4998 | /* |
4999 | * make sure no PI-waiters arrive (or leave) while we are | |
5000 | * changing the priority of the task: | |
5001 | */ | |
5002 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5003 | /* |
5004 | * To be able to change p->policy safely, the apropriate | |
5005 | * runqueue lock must be held. | |
5006 | */ | |
b29739f9 | 5007 | rq = __task_rq_lock(p); |
1da177e4 LT |
5008 | /* recheck policy now with rq lock held */ |
5009 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5010 | policy = oldpolicy = -1; | |
b29739f9 IM |
5011 | __task_rq_unlock(rq); |
5012 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5013 | goto recheck; |
5014 | } | |
2daa3577 | 5015 | update_rq_clock(rq); |
dd41f596 | 5016 | on_rq = p->se.on_rq; |
051a1d1a | 5017 | running = task_current(rq, p); |
0e1f3483 | 5018 | if (on_rq) |
2e1cb74a | 5019 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5020 | if (running) |
5021 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5022 | |
1da177e4 | 5023 | oldprio = p->prio; |
dd41f596 | 5024 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5025 | |
0e1f3483 HS |
5026 | if (running) |
5027 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5028 | if (on_rq) { |
5029 | activate_task(rq, p, 0); | |
cb469845 SR |
5030 | |
5031 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5032 | } |
b29739f9 IM |
5033 | __task_rq_unlock(rq); |
5034 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5035 | ||
95e02ca9 TG |
5036 | rt_mutex_adjust_pi(p); |
5037 | ||
1da177e4 LT |
5038 | return 0; |
5039 | } | |
5040 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
5041 | ||
95cdf3b7 IM |
5042 | static int |
5043 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5044 | { |
1da177e4 LT |
5045 | struct sched_param lparam; |
5046 | struct task_struct *p; | |
36c8b586 | 5047 | int retval; |
1da177e4 LT |
5048 | |
5049 | if (!param || pid < 0) | |
5050 | return -EINVAL; | |
5051 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5052 | return -EFAULT; | |
5fe1d75f ON |
5053 | |
5054 | rcu_read_lock(); | |
5055 | retval = -ESRCH; | |
1da177e4 | 5056 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5057 | if (p != NULL) |
5058 | retval = sched_setscheduler(p, policy, &lparam); | |
5059 | rcu_read_unlock(); | |
36c8b586 | 5060 | |
1da177e4 LT |
5061 | return retval; |
5062 | } | |
5063 | ||
5064 | /** | |
5065 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5066 | * @pid: the pid in question. | |
5067 | * @policy: new policy. | |
5068 | * @param: structure containing the new RT priority. | |
5069 | */ | |
41a2d6cf IM |
5070 | asmlinkage long |
5071 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5072 | { |
c21761f1 JB |
5073 | /* negative values for policy are not valid */ |
5074 | if (policy < 0) | |
5075 | return -EINVAL; | |
5076 | ||
1da177e4 LT |
5077 | return do_sched_setscheduler(pid, policy, param); |
5078 | } | |
5079 | ||
5080 | /** | |
5081 | * sys_sched_setparam - set/change the RT priority of a thread | |
5082 | * @pid: the pid in question. | |
5083 | * @param: structure containing the new RT priority. | |
5084 | */ | |
5085 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5086 | { | |
5087 | return do_sched_setscheduler(pid, -1, param); | |
5088 | } | |
5089 | ||
5090 | /** | |
5091 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5092 | * @pid: the pid in question. | |
5093 | */ | |
5094 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5095 | { | |
36c8b586 | 5096 | struct task_struct *p; |
3a5c359a | 5097 | int retval; |
1da177e4 LT |
5098 | |
5099 | if (pid < 0) | |
3a5c359a | 5100 | return -EINVAL; |
1da177e4 LT |
5101 | |
5102 | retval = -ESRCH; | |
5103 | read_lock(&tasklist_lock); | |
5104 | p = find_process_by_pid(pid); | |
5105 | if (p) { | |
5106 | retval = security_task_getscheduler(p); | |
5107 | if (!retval) | |
5108 | retval = p->policy; | |
5109 | } | |
5110 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5111 | return retval; |
5112 | } | |
5113 | ||
5114 | /** | |
5115 | * sys_sched_getscheduler - get the RT priority of a thread | |
5116 | * @pid: the pid in question. | |
5117 | * @param: structure containing the RT priority. | |
5118 | */ | |
5119 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5120 | { | |
5121 | struct sched_param lp; | |
36c8b586 | 5122 | struct task_struct *p; |
3a5c359a | 5123 | int retval; |
1da177e4 LT |
5124 | |
5125 | if (!param || pid < 0) | |
3a5c359a | 5126 | return -EINVAL; |
1da177e4 LT |
5127 | |
5128 | read_lock(&tasklist_lock); | |
5129 | p = find_process_by_pid(pid); | |
5130 | retval = -ESRCH; | |
5131 | if (!p) | |
5132 | goto out_unlock; | |
5133 | ||
5134 | retval = security_task_getscheduler(p); | |
5135 | if (retval) | |
5136 | goto out_unlock; | |
5137 | ||
5138 | lp.sched_priority = p->rt_priority; | |
5139 | read_unlock(&tasklist_lock); | |
5140 | ||
5141 | /* | |
5142 | * This one might sleep, we cannot do it with a spinlock held ... | |
5143 | */ | |
5144 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5145 | ||
1da177e4 LT |
5146 | return retval; |
5147 | ||
5148 | out_unlock: | |
5149 | read_unlock(&tasklist_lock); | |
5150 | return retval; | |
5151 | } | |
5152 | ||
b53e921b | 5153 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5154 | { |
1da177e4 | 5155 | cpumask_t cpus_allowed; |
b53e921b | 5156 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5157 | struct task_struct *p; |
5158 | int retval; | |
1da177e4 | 5159 | |
95402b38 | 5160 | get_online_cpus(); |
1da177e4 LT |
5161 | read_lock(&tasklist_lock); |
5162 | ||
5163 | p = find_process_by_pid(pid); | |
5164 | if (!p) { | |
5165 | read_unlock(&tasklist_lock); | |
95402b38 | 5166 | put_online_cpus(); |
1da177e4 LT |
5167 | return -ESRCH; |
5168 | } | |
5169 | ||
5170 | /* | |
5171 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5172 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5173 | * usage count and then drop tasklist_lock. |
5174 | */ | |
5175 | get_task_struct(p); | |
5176 | read_unlock(&tasklist_lock); | |
5177 | ||
5178 | retval = -EPERM; | |
5179 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5180 | !capable(CAP_SYS_NICE)) | |
5181 | goto out_unlock; | |
5182 | ||
e7834f8f DQ |
5183 | retval = security_task_setscheduler(p, 0, NULL); |
5184 | if (retval) | |
5185 | goto out_unlock; | |
5186 | ||
f9a86fcb | 5187 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5188 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5189 | again: |
7c16ec58 | 5190 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5191 | |
8707d8b8 | 5192 | if (!retval) { |
f9a86fcb | 5193 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5194 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5195 | /* | |
5196 | * We must have raced with a concurrent cpuset | |
5197 | * update. Just reset the cpus_allowed to the | |
5198 | * cpuset's cpus_allowed | |
5199 | */ | |
5200 | new_mask = cpus_allowed; | |
5201 | goto again; | |
5202 | } | |
5203 | } | |
1da177e4 LT |
5204 | out_unlock: |
5205 | put_task_struct(p); | |
95402b38 | 5206 | put_online_cpus(); |
1da177e4 LT |
5207 | return retval; |
5208 | } | |
5209 | ||
5210 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5211 | cpumask_t *new_mask) | |
5212 | { | |
5213 | if (len < sizeof(cpumask_t)) { | |
5214 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5215 | } else if (len > sizeof(cpumask_t)) { | |
5216 | len = sizeof(cpumask_t); | |
5217 | } | |
5218 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5219 | } | |
5220 | ||
5221 | /** | |
5222 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5223 | * @pid: pid of the process | |
5224 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5225 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5226 | */ | |
5227 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5228 | unsigned long __user *user_mask_ptr) | |
5229 | { | |
5230 | cpumask_t new_mask; | |
5231 | int retval; | |
5232 | ||
5233 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5234 | if (retval) | |
5235 | return retval; | |
5236 | ||
b53e921b | 5237 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5238 | } |
5239 | ||
1da177e4 LT |
5240 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5241 | { | |
36c8b586 | 5242 | struct task_struct *p; |
1da177e4 | 5243 | int retval; |
1da177e4 | 5244 | |
95402b38 | 5245 | get_online_cpus(); |
1da177e4 LT |
5246 | read_lock(&tasklist_lock); |
5247 | ||
5248 | retval = -ESRCH; | |
5249 | p = find_process_by_pid(pid); | |
5250 | if (!p) | |
5251 | goto out_unlock; | |
5252 | ||
e7834f8f DQ |
5253 | retval = security_task_getscheduler(p); |
5254 | if (retval) | |
5255 | goto out_unlock; | |
5256 | ||
2f7016d9 | 5257 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5258 | |
5259 | out_unlock: | |
5260 | read_unlock(&tasklist_lock); | |
95402b38 | 5261 | put_online_cpus(); |
1da177e4 | 5262 | |
9531b62f | 5263 | return retval; |
1da177e4 LT |
5264 | } |
5265 | ||
5266 | /** | |
5267 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5268 | * @pid: pid of the process | |
5269 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5270 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5271 | */ | |
5272 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5273 | unsigned long __user *user_mask_ptr) | |
5274 | { | |
5275 | int ret; | |
5276 | cpumask_t mask; | |
5277 | ||
5278 | if (len < sizeof(cpumask_t)) | |
5279 | return -EINVAL; | |
5280 | ||
5281 | ret = sched_getaffinity(pid, &mask); | |
5282 | if (ret < 0) | |
5283 | return ret; | |
5284 | ||
5285 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5286 | return -EFAULT; | |
5287 | ||
5288 | return sizeof(cpumask_t); | |
5289 | } | |
5290 | ||
5291 | /** | |
5292 | * sys_sched_yield - yield the current processor to other threads. | |
5293 | * | |
dd41f596 IM |
5294 | * This function yields the current CPU to other tasks. If there are no |
5295 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5296 | */ |
5297 | asmlinkage long sys_sched_yield(void) | |
5298 | { | |
70b97a7f | 5299 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5300 | |
2d72376b | 5301 | schedstat_inc(rq, yld_count); |
4530d7ab | 5302 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5303 | |
5304 | /* | |
5305 | * Since we are going to call schedule() anyway, there's | |
5306 | * no need to preempt or enable interrupts: | |
5307 | */ | |
5308 | __release(rq->lock); | |
8a25d5de | 5309 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5310 | _raw_spin_unlock(&rq->lock); |
5311 | preempt_enable_no_resched(); | |
5312 | ||
5313 | schedule(); | |
5314 | ||
5315 | return 0; | |
5316 | } | |
5317 | ||
e7b38404 | 5318 | static void __cond_resched(void) |
1da177e4 | 5319 | { |
8e0a43d8 IM |
5320 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5321 | __might_sleep(__FILE__, __LINE__); | |
5322 | #endif | |
5bbcfd90 IM |
5323 | /* |
5324 | * The BKS might be reacquired before we have dropped | |
5325 | * PREEMPT_ACTIVE, which could trigger a second | |
5326 | * cond_resched() call. | |
5327 | */ | |
1da177e4 LT |
5328 | do { |
5329 | add_preempt_count(PREEMPT_ACTIVE); | |
5330 | schedule(); | |
5331 | sub_preempt_count(PREEMPT_ACTIVE); | |
5332 | } while (need_resched()); | |
5333 | } | |
5334 | ||
02b67cc3 | 5335 | int __sched _cond_resched(void) |
1da177e4 | 5336 | { |
9414232f IM |
5337 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5338 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5339 | __cond_resched(); |
5340 | return 1; | |
5341 | } | |
5342 | return 0; | |
5343 | } | |
02b67cc3 | 5344 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5345 | |
5346 | /* | |
5347 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5348 | * call schedule, and on return reacquire the lock. | |
5349 | * | |
41a2d6cf | 5350 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5351 | * operations here to prevent schedule() from being called twice (once via |
5352 | * spin_unlock(), once by hand). | |
5353 | */ | |
95cdf3b7 | 5354 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5355 | { |
95c354fe | 5356 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5357 | int ret = 0; |
5358 | ||
95c354fe | 5359 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5360 | spin_unlock(lock); |
95c354fe NP |
5361 | if (resched && need_resched()) |
5362 | __cond_resched(); | |
5363 | else | |
5364 | cpu_relax(); | |
6df3cecb | 5365 | ret = 1; |
1da177e4 | 5366 | spin_lock(lock); |
1da177e4 | 5367 | } |
6df3cecb | 5368 | return ret; |
1da177e4 | 5369 | } |
1da177e4 LT |
5370 | EXPORT_SYMBOL(cond_resched_lock); |
5371 | ||
5372 | int __sched cond_resched_softirq(void) | |
5373 | { | |
5374 | BUG_ON(!in_softirq()); | |
5375 | ||
9414232f | 5376 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5377 | local_bh_enable(); |
1da177e4 LT |
5378 | __cond_resched(); |
5379 | local_bh_disable(); | |
5380 | return 1; | |
5381 | } | |
5382 | return 0; | |
5383 | } | |
1da177e4 LT |
5384 | EXPORT_SYMBOL(cond_resched_softirq); |
5385 | ||
1da177e4 LT |
5386 | /** |
5387 | * yield - yield the current processor to other threads. | |
5388 | * | |
72fd4a35 | 5389 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5390 | * thread runnable and calls sys_sched_yield(). |
5391 | */ | |
5392 | void __sched yield(void) | |
5393 | { | |
5394 | set_current_state(TASK_RUNNING); | |
5395 | sys_sched_yield(); | |
5396 | } | |
1da177e4 LT |
5397 | EXPORT_SYMBOL(yield); |
5398 | ||
5399 | /* | |
41a2d6cf | 5400 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5401 | * that process accounting knows that this is a task in IO wait state. |
5402 | * | |
5403 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5404 | * has set its backing_dev_info: the queue against which it should throttle) | |
5405 | */ | |
5406 | void __sched io_schedule(void) | |
5407 | { | |
70b97a7f | 5408 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5409 | |
0ff92245 | 5410 | delayacct_blkio_start(); |
1da177e4 LT |
5411 | atomic_inc(&rq->nr_iowait); |
5412 | schedule(); | |
5413 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5414 | delayacct_blkio_end(); |
1da177e4 | 5415 | } |
1da177e4 LT |
5416 | EXPORT_SYMBOL(io_schedule); |
5417 | ||
5418 | long __sched io_schedule_timeout(long timeout) | |
5419 | { | |
70b97a7f | 5420 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5421 | long ret; |
5422 | ||
0ff92245 | 5423 | delayacct_blkio_start(); |
1da177e4 LT |
5424 | atomic_inc(&rq->nr_iowait); |
5425 | ret = schedule_timeout(timeout); | |
5426 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5427 | delayacct_blkio_end(); |
1da177e4 LT |
5428 | return ret; |
5429 | } | |
5430 | ||
5431 | /** | |
5432 | * sys_sched_get_priority_max - return maximum RT priority. | |
5433 | * @policy: scheduling class. | |
5434 | * | |
5435 | * this syscall returns the maximum rt_priority that can be used | |
5436 | * by a given scheduling class. | |
5437 | */ | |
5438 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5439 | { | |
5440 | int ret = -EINVAL; | |
5441 | ||
5442 | switch (policy) { | |
5443 | case SCHED_FIFO: | |
5444 | case SCHED_RR: | |
5445 | ret = MAX_USER_RT_PRIO-1; | |
5446 | break; | |
5447 | case SCHED_NORMAL: | |
b0a9499c | 5448 | case SCHED_BATCH: |
dd41f596 | 5449 | case SCHED_IDLE: |
1da177e4 LT |
5450 | ret = 0; |
5451 | break; | |
5452 | } | |
5453 | return ret; | |
5454 | } | |
5455 | ||
5456 | /** | |
5457 | * sys_sched_get_priority_min - return minimum RT priority. | |
5458 | * @policy: scheduling class. | |
5459 | * | |
5460 | * this syscall returns the minimum rt_priority that can be used | |
5461 | * by a given scheduling class. | |
5462 | */ | |
5463 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5464 | { | |
5465 | int ret = -EINVAL; | |
5466 | ||
5467 | switch (policy) { | |
5468 | case SCHED_FIFO: | |
5469 | case SCHED_RR: | |
5470 | ret = 1; | |
5471 | break; | |
5472 | case SCHED_NORMAL: | |
b0a9499c | 5473 | case SCHED_BATCH: |
dd41f596 | 5474 | case SCHED_IDLE: |
1da177e4 LT |
5475 | ret = 0; |
5476 | } | |
5477 | return ret; | |
5478 | } | |
5479 | ||
5480 | /** | |
5481 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5482 | * @pid: pid of the process. | |
5483 | * @interval: userspace pointer to the timeslice value. | |
5484 | * | |
5485 | * this syscall writes the default timeslice value of a given process | |
5486 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5487 | */ | |
5488 | asmlinkage | |
5489 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5490 | { | |
36c8b586 | 5491 | struct task_struct *p; |
a4ec24b4 | 5492 | unsigned int time_slice; |
3a5c359a | 5493 | int retval; |
1da177e4 | 5494 | struct timespec t; |
1da177e4 LT |
5495 | |
5496 | if (pid < 0) | |
3a5c359a | 5497 | return -EINVAL; |
1da177e4 LT |
5498 | |
5499 | retval = -ESRCH; | |
5500 | read_lock(&tasklist_lock); | |
5501 | p = find_process_by_pid(pid); | |
5502 | if (!p) | |
5503 | goto out_unlock; | |
5504 | ||
5505 | retval = security_task_getscheduler(p); | |
5506 | if (retval) | |
5507 | goto out_unlock; | |
5508 | ||
77034937 IM |
5509 | /* |
5510 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5511 | * tasks that are on an otherwise idle runqueue: | |
5512 | */ | |
5513 | time_slice = 0; | |
5514 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5515 | time_slice = DEF_TIMESLICE; |
1868f958 | 5516 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5517 | struct sched_entity *se = &p->se; |
5518 | unsigned long flags; | |
5519 | struct rq *rq; | |
5520 | ||
5521 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5522 | if (rq->cfs.load.weight) |
5523 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5524 | task_rq_unlock(rq, &flags); |
5525 | } | |
1da177e4 | 5526 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5527 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5528 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5529 | return retval; |
3a5c359a | 5530 | |
1da177e4 LT |
5531 | out_unlock: |
5532 | read_unlock(&tasklist_lock); | |
5533 | return retval; | |
5534 | } | |
5535 | ||
2ed6e34f | 5536 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5537 | |
82a1fcb9 | 5538 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5539 | { |
1da177e4 | 5540 | unsigned long free = 0; |
36c8b586 | 5541 | unsigned state; |
1da177e4 | 5542 | |
1da177e4 | 5543 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5544 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5545 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5546 | #if BITS_PER_LONG == 32 |
1da177e4 | 5547 | if (state == TASK_RUNNING) |
cc4ea795 | 5548 | printk(KERN_CONT " running "); |
1da177e4 | 5549 | else |
cc4ea795 | 5550 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5551 | #else |
5552 | if (state == TASK_RUNNING) | |
cc4ea795 | 5553 | printk(KERN_CONT " running task "); |
1da177e4 | 5554 | else |
cc4ea795 | 5555 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5556 | #endif |
5557 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5558 | { | |
10ebffde | 5559 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5560 | while (!*n) |
5561 | n++; | |
10ebffde | 5562 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5563 | } |
5564 | #endif | |
ba25f9dc | 5565 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5566 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5567 | |
5fb5e6de | 5568 | show_stack(p, NULL); |
1da177e4 LT |
5569 | } |
5570 | ||
e59e2ae2 | 5571 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5572 | { |
36c8b586 | 5573 | struct task_struct *g, *p; |
1da177e4 | 5574 | |
4bd77321 IM |
5575 | #if BITS_PER_LONG == 32 |
5576 | printk(KERN_INFO | |
5577 | " task PC stack pid father\n"); | |
1da177e4 | 5578 | #else |
4bd77321 IM |
5579 | printk(KERN_INFO |
5580 | " task PC stack pid father\n"); | |
1da177e4 LT |
5581 | #endif |
5582 | read_lock(&tasklist_lock); | |
5583 | do_each_thread(g, p) { | |
5584 | /* | |
5585 | * reset the NMI-timeout, listing all files on a slow | |
5586 | * console might take alot of time: | |
5587 | */ | |
5588 | touch_nmi_watchdog(); | |
39bc89fd | 5589 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5590 | sched_show_task(p); |
1da177e4 LT |
5591 | } while_each_thread(g, p); |
5592 | ||
04c9167f JF |
5593 | touch_all_softlockup_watchdogs(); |
5594 | ||
dd41f596 IM |
5595 | #ifdef CONFIG_SCHED_DEBUG |
5596 | sysrq_sched_debug_show(); | |
5597 | #endif | |
1da177e4 | 5598 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5599 | /* |
5600 | * Only show locks if all tasks are dumped: | |
5601 | */ | |
5602 | if (state_filter == -1) | |
5603 | debug_show_all_locks(); | |
1da177e4 LT |
5604 | } |
5605 | ||
1df21055 IM |
5606 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5607 | { | |
dd41f596 | 5608 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5609 | } |
5610 | ||
f340c0d1 IM |
5611 | /** |
5612 | * init_idle - set up an idle thread for a given CPU | |
5613 | * @idle: task in question | |
5614 | * @cpu: cpu the idle task belongs to | |
5615 | * | |
5616 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5617 | * flag, to make booting more robust. | |
5618 | */ | |
5c1e1767 | 5619 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5620 | { |
70b97a7f | 5621 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5622 | unsigned long flags; |
5623 | ||
dd41f596 IM |
5624 | __sched_fork(idle); |
5625 | idle->se.exec_start = sched_clock(); | |
5626 | ||
b29739f9 | 5627 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5628 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5629 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5630 | |
5631 | spin_lock_irqsave(&rq->lock, flags); | |
5632 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5633 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5634 | idle->oncpu = 1; | |
5635 | #endif | |
1da177e4 LT |
5636 | spin_unlock_irqrestore(&rq->lock, flags); |
5637 | ||
5638 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5639 | #if defined(CONFIG_PREEMPT) |
5640 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5641 | #else | |
a1261f54 | 5642 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5643 | #endif |
dd41f596 IM |
5644 | /* |
5645 | * The idle tasks have their own, simple scheduling class: | |
5646 | */ | |
5647 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5648 | } |
5649 | ||
5650 | /* | |
5651 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5652 | * indicates which cpus entered this state. This is used | |
5653 | * in the rcu update to wait only for active cpus. For system | |
5654 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5655 | * always be CPU_MASK_NONE. | |
5656 | */ | |
5657 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5658 | ||
19978ca6 IM |
5659 | /* |
5660 | * Increase the granularity value when there are more CPUs, | |
5661 | * because with more CPUs the 'effective latency' as visible | |
5662 | * to users decreases. But the relationship is not linear, | |
5663 | * so pick a second-best guess by going with the log2 of the | |
5664 | * number of CPUs. | |
5665 | * | |
5666 | * This idea comes from the SD scheduler of Con Kolivas: | |
5667 | */ | |
5668 | static inline void sched_init_granularity(void) | |
5669 | { | |
5670 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5671 | const unsigned long limit = 200000000; | |
5672 | ||
5673 | sysctl_sched_min_granularity *= factor; | |
5674 | if (sysctl_sched_min_granularity > limit) | |
5675 | sysctl_sched_min_granularity = limit; | |
5676 | ||
5677 | sysctl_sched_latency *= factor; | |
5678 | if (sysctl_sched_latency > limit) | |
5679 | sysctl_sched_latency = limit; | |
5680 | ||
5681 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5682 | } |
5683 | ||
1da177e4 LT |
5684 | #ifdef CONFIG_SMP |
5685 | /* | |
5686 | * This is how migration works: | |
5687 | * | |
70b97a7f | 5688 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5689 | * runqueue and wake up that CPU's migration thread. |
5690 | * 2) we down() the locked semaphore => thread blocks. | |
5691 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5692 | * thread off the CPU) | |
5693 | * 4) it gets the migration request and checks whether the migrated | |
5694 | * task is still in the wrong runqueue. | |
5695 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5696 | * it and puts it into the right queue. | |
5697 | * 6) migration thread up()s the semaphore. | |
5698 | * 7) we wake up and the migration is done. | |
5699 | */ | |
5700 | ||
5701 | /* | |
5702 | * Change a given task's CPU affinity. Migrate the thread to a | |
5703 | * proper CPU and schedule it away if the CPU it's executing on | |
5704 | * is removed from the allowed bitmask. | |
5705 | * | |
5706 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5707 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5708 | * call is not atomic; no spinlocks may be held. |
5709 | */ | |
cd8ba7cd | 5710 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5711 | { |
70b97a7f | 5712 | struct migration_req req; |
1da177e4 | 5713 | unsigned long flags; |
70b97a7f | 5714 | struct rq *rq; |
48f24c4d | 5715 | int ret = 0; |
1da177e4 LT |
5716 | |
5717 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5718 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5719 | ret = -EINVAL; |
5720 | goto out; | |
5721 | } | |
5722 | ||
9985b0ba DR |
5723 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5724 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5725 | ret = -EINVAL; | |
5726 | goto out; | |
5727 | } | |
5728 | ||
73fe6aae | 5729 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5730 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5731 | else { |
cd8ba7cd MT |
5732 | p->cpus_allowed = *new_mask; |
5733 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5734 | } |
5735 | ||
1da177e4 | 5736 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5737 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5738 | goto out; |
5739 | ||
cd8ba7cd | 5740 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5741 | /* Need help from migration thread: drop lock and wait. */ |
5742 | task_rq_unlock(rq, &flags); | |
5743 | wake_up_process(rq->migration_thread); | |
5744 | wait_for_completion(&req.done); | |
5745 | tlb_migrate_finish(p->mm); | |
5746 | return 0; | |
5747 | } | |
5748 | out: | |
5749 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5750 | |
1da177e4 LT |
5751 | return ret; |
5752 | } | |
cd8ba7cd | 5753 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5754 | |
5755 | /* | |
41a2d6cf | 5756 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5757 | * this because either it can't run here any more (set_cpus_allowed() |
5758 | * away from this CPU, or CPU going down), or because we're | |
5759 | * attempting to rebalance this task on exec (sched_exec). | |
5760 | * | |
5761 | * So we race with normal scheduler movements, but that's OK, as long | |
5762 | * as the task is no longer on this CPU. | |
efc30814 KK |
5763 | * |
5764 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5765 | */ |
efc30814 | 5766 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5767 | { |
70b97a7f | 5768 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5769 | int ret = 0, on_rq; |
1da177e4 LT |
5770 | |
5771 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5772 | return ret; |
1da177e4 LT |
5773 | |
5774 | rq_src = cpu_rq(src_cpu); | |
5775 | rq_dest = cpu_rq(dest_cpu); | |
5776 | ||
5777 | double_rq_lock(rq_src, rq_dest); | |
5778 | /* Already moved. */ | |
5779 | if (task_cpu(p) != src_cpu) | |
5780 | goto out; | |
5781 | /* Affinity changed (again). */ | |
5782 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5783 | goto out; | |
5784 | ||
dd41f596 | 5785 | on_rq = p->se.on_rq; |
6e82a3be | 5786 | if (on_rq) |
2e1cb74a | 5787 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5788 | |
1da177e4 | 5789 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5790 | if (on_rq) { |
5791 | activate_task(rq_dest, p, 0); | |
5792 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5793 | } |
efc30814 | 5794 | ret = 1; |
1da177e4 LT |
5795 | out: |
5796 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5797 | return ret; |
1da177e4 LT |
5798 | } |
5799 | ||
5800 | /* | |
5801 | * migration_thread - this is a highprio system thread that performs | |
5802 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5803 | * another runqueue. | |
5804 | */ | |
95cdf3b7 | 5805 | static int migration_thread(void *data) |
1da177e4 | 5806 | { |
1da177e4 | 5807 | int cpu = (long)data; |
70b97a7f | 5808 | struct rq *rq; |
1da177e4 LT |
5809 | |
5810 | rq = cpu_rq(cpu); | |
5811 | BUG_ON(rq->migration_thread != current); | |
5812 | ||
5813 | set_current_state(TASK_INTERRUPTIBLE); | |
5814 | while (!kthread_should_stop()) { | |
70b97a7f | 5815 | struct migration_req *req; |
1da177e4 | 5816 | struct list_head *head; |
1da177e4 | 5817 | |
1da177e4 LT |
5818 | spin_lock_irq(&rq->lock); |
5819 | ||
5820 | if (cpu_is_offline(cpu)) { | |
5821 | spin_unlock_irq(&rq->lock); | |
5822 | goto wait_to_die; | |
5823 | } | |
5824 | ||
5825 | if (rq->active_balance) { | |
5826 | active_load_balance(rq, cpu); | |
5827 | rq->active_balance = 0; | |
5828 | } | |
5829 | ||
5830 | head = &rq->migration_queue; | |
5831 | ||
5832 | if (list_empty(head)) { | |
5833 | spin_unlock_irq(&rq->lock); | |
5834 | schedule(); | |
5835 | set_current_state(TASK_INTERRUPTIBLE); | |
5836 | continue; | |
5837 | } | |
70b97a7f | 5838 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5839 | list_del_init(head->next); |
5840 | ||
674311d5 NP |
5841 | spin_unlock(&rq->lock); |
5842 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5843 | local_irq_enable(); | |
1da177e4 LT |
5844 | |
5845 | complete(&req->done); | |
5846 | } | |
5847 | __set_current_state(TASK_RUNNING); | |
5848 | return 0; | |
5849 | ||
5850 | wait_to_die: | |
5851 | /* Wait for kthread_stop */ | |
5852 | set_current_state(TASK_INTERRUPTIBLE); | |
5853 | while (!kthread_should_stop()) { | |
5854 | schedule(); | |
5855 | set_current_state(TASK_INTERRUPTIBLE); | |
5856 | } | |
5857 | __set_current_state(TASK_RUNNING); | |
5858 | return 0; | |
5859 | } | |
5860 | ||
5861 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5862 | |
5863 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5864 | { | |
5865 | int ret; | |
5866 | ||
5867 | local_irq_disable(); | |
5868 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5869 | local_irq_enable(); | |
5870 | return ret; | |
5871 | } | |
5872 | ||
054b9108 | 5873 | /* |
3a4fa0a2 | 5874 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5875 | * NOTE: interrupts should be disabled by the caller |
5876 | */ | |
48f24c4d | 5877 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5878 | { |
efc30814 | 5879 | unsigned long flags; |
1da177e4 | 5880 | cpumask_t mask; |
70b97a7f IM |
5881 | struct rq *rq; |
5882 | int dest_cpu; | |
1da177e4 | 5883 | |
3a5c359a AK |
5884 | do { |
5885 | /* On same node? */ | |
5886 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5887 | cpus_and(mask, mask, p->cpus_allowed); | |
5888 | dest_cpu = any_online_cpu(mask); | |
5889 | ||
5890 | /* On any allowed CPU? */ | |
434d53b0 | 5891 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5892 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5893 | ||
5894 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5895 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5896 | cpumask_t cpus_allowed; |
5897 | ||
5898 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5899 | /* |
5900 | * Try to stay on the same cpuset, where the | |
5901 | * current cpuset may be a subset of all cpus. | |
5902 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5903 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5904 | * called within calls to cpuset_lock/cpuset_unlock. |
5905 | */ | |
3a5c359a | 5906 | rq = task_rq_lock(p, &flags); |
470fd646 | 5907 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5908 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5909 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5910 | |
3a5c359a AK |
5911 | /* |
5912 | * Don't tell them about moving exiting tasks or | |
5913 | * kernel threads (both mm NULL), since they never | |
5914 | * leave kernel. | |
5915 | */ | |
41a2d6cf | 5916 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5917 | printk(KERN_INFO "process %d (%s) no " |
5918 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5919 | task_pid_nr(p), p->comm, dead_cpu); |
5920 | } | |
3a5c359a | 5921 | } |
f7b4cddc | 5922 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5923 | } |
5924 | ||
5925 | /* | |
5926 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5927 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5928 | * for performance reasons the counter is not stricly tracking tasks to | |
5929 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5930 | * to keep the global sum constant after CPU-down: | |
5931 | */ | |
70b97a7f | 5932 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5933 | { |
7c16ec58 | 5934 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5935 | unsigned long flags; |
5936 | ||
5937 | local_irq_save(flags); | |
5938 | double_rq_lock(rq_src, rq_dest); | |
5939 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5940 | rq_src->nr_uninterruptible = 0; | |
5941 | double_rq_unlock(rq_src, rq_dest); | |
5942 | local_irq_restore(flags); | |
5943 | } | |
5944 | ||
5945 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5946 | static void migrate_live_tasks(int src_cpu) | |
5947 | { | |
48f24c4d | 5948 | struct task_struct *p, *t; |
1da177e4 | 5949 | |
f7b4cddc | 5950 | read_lock(&tasklist_lock); |
1da177e4 | 5951 | |
48f24c4d IM |
5952 | do_each_thread(t, p) { |
5953 | if (p == current) | |
1da177e4 LT |
5954 | continue; |
5955 | ||
48f24c4d IM |
5956 | if (task_cpu(p) == src_cpu) |
5957 | move_task_off_dead_cpu(src_cpu, p); | |
5958 | } while_each_thread(t, p); | |
1da177e4 | 5959 | |
f7b4cddc | 5960 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5961 | } |
5962 | ||
dd41f596 IM |
5963 | /* |
5964 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5965 | * It does so by boosting its priority to highest possible. |
5966 | * Used by CPU offline code. | |
1da177e4 LT |
5967 | */ |
5968 | void sched_idle_next(void) | |
5969 | { | |
48f24c4d | 5970 | int this_cpu = smp_processor_id(); |
70b97a7f | 5971 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5972 | struct task_struct *p = rq->idle; |
5973 | unsigned long flags; | |
5974 | ||
5975 | /* cpu has to be offline */ | |
48f24c4d | 5976 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5977 | |
48f24c4d IM |
5978 | /* |
5979 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5980 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5981 | */ |
5982 | spin_lock_irqsave(&rq->lock, flags); | |
5983 | ||
dd41f596 | 5984 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5985 | |
94bc9a7b DA |
5986 | update_rq_clock(rq); |
5987 | activate_task(rq, p, 0); | |
1da177e4 LT |
5988 | |
5989 | spin_unlock_irqrestore(&rq->lock, flags); | |
5990 | } | |
5991 | ||
48f24c4d IM |
5992 | /* |
5993 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5994 | * offline. |
5995 | */ | |
5996 | void idle_task_exit(void) | |
5997 | { | |
5998 | struct mm_struct *mm = current->active_mm; | |
5999 | ||
6000 | BUG_ON(cpu_online(smp_processor_id())); | |
6001 | ||
6002 | if (mm != &init_mm) | |
6003 | switch_mm(mm, &init_mm, current); | |
6004 | mmdrop(mm); | |
6005 | } | |
6006 | ||
054b9108 | 6007 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6008 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6009 | { |
70b97a7f | 6010 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6011 | |
6012 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6013 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6014 | |
6015 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6016 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6017 | |
48f24c4d | 6018 | get_task_struct(p); |
1da177e4 LT |
6019 | |
6020 | /* | |
6021 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6022 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6023 | * fine. |
6024 | */ | |
f7b4cddc | 6025 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6026 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6027 | spin_lock_irq(&rq->lock); |
1da177e4 | 6028 | |
48f24c4d | 6029 | put_task_struct(p); |
1da177e4 LT |
6030 | } |
6031 | ||
6032 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6033 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6034 | { | |
70b97a7f | 6035 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6036 | struct task_struct *next; |
48f24c4d | 6037 | |
dd41f596 IM |
6038 | for ( ; ; ) { |
6039 | if (!rq->nr_running) | |
6040 | break; | |
a8e504d2 | 6041 | update_rq_clock(rq); |
ff95f3df | 6042 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6043 | if (!next) |
6044 | break; | |
6045 | migrate_dead(dead_cpu, next); | |
e692ab53 | 6046 | |
1da177e4 LT |
6047 | } |
6048 | } | |
6049 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6050 | ||
e692ab53 NP |
6051 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6052 | ||
6053 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6054 | { |
6055 | .procname = "sched_domain", | |
c57baf1e | 6056 | .mode = 0555, |
e0361851 | 6057 | }, |
38605cae | 6058 | {0, }, |
e692ab53 NP |
6059 | }; |
6060 | ||
6061 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6062 | { |
c57baf1e | 6063 | .ctl_name = CTL_KERN, |
e0361851 | 6064 | .procname = "kernel", |
c57baf1e | 6065 | .mode = 0555, |
e0361851 AD |
6066 | .child = sd_ctl_dir, |
6067 | }, | |
38605cae | 6068 | {0, }, |
e692ab53 NP |
6069 | }; |
6070 | ||
6071 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6072 | { | |
6073 | struct ctl_table *entry = | |
5cf9f062 | 6074 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6075 | |
e692ab53 NP |
6076 | return entry; |
6077 | } | |
6078 | ||
6382bc90 MM |
6079 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6080 | { | |
cd790076 | 6081 | struct ctl_table *entry; |
6382bc90 | 6082 | |
cd790076 MM |
6083 | /* |
6084 | * In the intermediate directories, both the child directory and | |
6085 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6086 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6087 | * static strings and all have proc handlers. |
6088 | */ | |
6089 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6090 | if (entry->child) |
6091 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6092 | if (entry->proc_handler == NULL) |
6093 | kfree(entry->procname); | |
6094 | } | |
6382bc90 MM |
6095 | |
6096 | kfree(*tablep); | |
6097 | *tablep = NULL; | |
6098 | } | |
6099 | ||
e692ab53 | 6100 | static void |
e0361851 | 6101 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6102 | const char *procname, void *data, int maxlen, |
6103 | mode_t mode, proc_handler *proc_handler) | |
6104 | { | |
e692ab53 NP |
6105 | entry->procname = procname; |
6106 | entry->data = data; | |
6107 | entry->maxlen = maxlen; | |
6108 | entry->mode = mode; | |
6109 | entry->proc_handler = proc_handler; | |
6110 | } | |
6111 | ||
6112 | static struct ctl_table * | |
6113 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6114 | { | |
ace8b3d6 | 6115 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6116 | |
ad1cdc1d MM |
6117 | if (table == NULL) |
6118 | return NULL; | |
6119 | ||
e0361851 | 6120 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6121 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6122 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6123 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6124 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6125 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6126 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6127 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6128 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6129 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6130 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6131 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6132 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6133 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6134 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6135 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6136 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6137 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6138 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6139 | &sd->cache_nice_tries, |
6140 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6141 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6142 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6143 | /* &table[11] is terminator */ |
e692ab53 NP |
6144 | |
6145 | return table; | |
6146 | } | |
6147 | ||
9a4e7159 | 6148 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6149 | { |
6150 | struct ctl_table *entry, *table; | |
6151 | struct sched_domain *sd; | |
6152 | int domain_num = 0, i; | |
6153 | char buf[32]; | |
6154 | ||
6155 | for_each_domain(cpu, sd) | |
6156 | domain_num++; | |
6157 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6158 | if (table == NULL) |
6159 | return NULL; | |
e692ab53 NP |
6160 | |
6161 | i = 0; | |
6162 | for_each_domain(cpu, sd) { | |
6163 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6164 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6165 | entry->mode = 0555; |
e692ab53 NP |
6166 | entry->child = sd_alloc_ctl_domain_table(sd); |
6167 | entry++; | |
6168 | i++; | |
6169 | } | |
6170 | return table; | |
6171 | } | |
6172 | ||
6173 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6174 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6175 | { |
6176 | int i, cpu_num = num_online_cpus(); | |
6177 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6178 | char buf[32]; | |
6179 | ||
7378547f MM |
6180 | WARN_ON(sd_ctl_dir[0].child); |
6181 | sd_ctl_dir[0].child = entry; | |
6182 | ||
ad1cdc1d MM |
6183 | if (entry == NULL) |
6184 | return; | |
6185 | ||
97b6ea7b | 6186 | for_each_online_cpu(i) { |
e692ab53 | 6187 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6188 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6189 | entry->mode = 0555; |
e692ab53 | 6190 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6191 | entry++; |
e692ab53 | 6192 | } |
7378547f MM |
6193 | |
6194 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6195 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6196 | } | |
6382bc90 | 6197 | |
7378547f | 6198 | /* may be called multiple times per register */ |
6382bc90 MM |
6199 | static void unregister_sched_domain_sysctl(void) |
6200 | { | |
7378547f MM |
6201 | if (sd_sysctl_header) |
6202 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6203 | sd_sysctl_header = NULL; |
7378547f MM |
6204 | if (sd_ctl_dir[0].child) |
6205 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6206 | } |
e692ab53 | 6207 | #else |
6382bc90 MM |
6208 | static void register_sched_domain_sysctl(void) |
6209 | { | |
6210 | } | |
6211 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6212 | { |
6213 | } | |
6214 | #endif | |
6215 | ||
1f11eb6a GH |
6216 | static void set_rq_online(struct rq *rq) |
6217 | { | |
6218 | if (!rq->online) { | |
6219 | const struct sched_class *class; | |
6220 | ||
6221 | cpu_set(rq->cpu, rq->rd->online); | |
6222 | rq->online = 1; | |
6223 | ||
6224 | for_each_class(class) { | |
6225 | if (class->rq_online) | |
6226 | class->rq_online(rq); | |
6227 | } | |
6228 | } | |
6229 | } | |
6230 | ||
6231 | static void set_rq_offline(struct rq *rq) | |
6232 | { | |
6233 | if (rq->online) { | |
6234 | const struct sched_class *class; | |
6235 | ||
6236 | for_each_class(class) { | |
6237 | if (class->rq_offline) | |
6238 | class->rq_offline(rq); | |
6239 | } | |
6240 | ||
6241 | cpu_clear(rq->cpu, rq->rd->online); | |
6242 | rq->online = 0; | |
6243 | } | |
6244 | } | |
6245 | ||
1da177e4 LT |
6246 | /* |
6247 | * migration_call - callback that gets triggered when a CPU is added. | |
6248 | * Here we can start up the necessary migration thread for the new CPU. | |
6249 | */ | |
48f24c4d IM |
6250 | static int __cpuinit |
6251 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6252 | { |
1da177e4 | 6253 | struct task_struct *p; |
48f24c4d | 6254 | int cpu = (long)hcpu; |
1da177e4 | 6255 | unsigned long flags; |
70b97a7f | 6256 | struct rq *rq; |
1da177e4 LT |
6257 | |
6258 | switch (action) { | |
5be9361c | 6259 | |
1da177e4 | 6260 | case CPU_UP_PREPARE: |
8bb78442 | 6261 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6262 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6263 | if (IS_ERR(p)) |
6264 | return NOTIFY_BAD; | |
1da177e4 LT |
6265 | kthread_bind(p, cpu); |
6266 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6267 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6268 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6269 | task_rq_unlock(rq, &flags); |
6270 | cpu_rq(cpu)->migration_thread = p; | |
6271 | break; | |
48f24c4d | 6272 | |
1da177e4 | 6273 | case CPU_ONLINE: |
8bb78442 | 6274 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6275 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6276 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6277 | |
6278 | /* Update our root-domain */ | |
6279 | rq = cpu_rq(cpu); | |
6280 | spin_lock_irqsave(&rq->lock, flags); | |
6281 | if (rq->rd) { | |
6282 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6283 | |
6284 | set_rq_online(rq); | |
1f94ef59 GH |
6285 | } |
6286 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6287 | break; |
48f24c4d | 6288 | |
1da177e4 LT |
6289 | #ifdef CONFIG_HOTPLUG_CPU |
6290 | case CPU_UP_CANCELED: | |
8bb78442 | 6291 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6292 | if (!cpu_rq(cpu)->migration_thread) |
6293 | break; | |
41a2d6cf | 6294 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6295 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6296 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6297 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6298 | cpu_rq(cpu)->migration_thread = NULL; | |
6299 | break; | |
48f24c4d | 6300 | |
1da177e4 | 6301 | case CPU_DEAD: |
8bb78442 | 6302 | case CPU_DEAD_FROZEN: |
470fd646 | 6303 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6304 | migrate_live_tasks(cpu); |
6305 | rq = cpu_rq(cpu); | |
6306 | kthread_stop(rq->migration_thread); | |
6307 | rq->migration_thread = NULL; | |
6308 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6309 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6310 | update_rq_clock(rq); |
2e1cb74a | 6311 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6312 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6313 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6314 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6315 | migrate_dead_tasks(cpu); |
d2da272a | 6316 | spin_unlock_irq(&rq->lock); |
470fd646 | 6317 | cpuset_unlock(); |
1da177e4 LT |
6318 | migrate_nr_uninterruptible(rq); |
6319 | BUG_ON(rq->nr_running != 0); | |
6320 | ||
41a2d6cf IM |
6321 | /* |
6322 | * No need to migrate the tasks: it was best-effort if | |
6323 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6324 | * the requestors. | |
6325 | */ | |
1da177e4 LT |
6326 | spin_lock_irq(&rq->lock); |
6327 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6328 | struct migration_req *req; |
6329 | ||
1da177e4 | 6330 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6331 | struct migration_req, list); |
1da177e4 LT |
6332 | list_del_init(&req->list); |
6333 | complete(&req->done); | |
6334 | } | |
6335 | spin_unlock_irq(&rq->lock); | |
6336 | break; | |
57d885fe | 6337 | |
08f503b0 GH |
6338 | case CPU_DYING: |
6339 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6340 | /* Update our root-domain */ |
6341 | rq = cpu_rq(cpu); | |
6342 | spin_lock_irqsave(&rq->lock, flags); | |
6343 | if (rq->rd) { | |
6344 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6345 | set_rq_offline(rq); |
57d885fe GH |
6346 | } |
6347 | spin_unlock_irqrestore(&rq->lock, flags); | |
6348 | break; | |
1da177e4 LT |
6349 | #endif |
6350 | } | |
6351 | return NOTIFY_OK; | |
6352 | } | |
6353 | ||
6354 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6355 | * happens before everything else. | |
6356 | */ | |
26c2143b | 6357 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6358 | .notifier_call = migration_call, |
6359 | .priority = 10 | |
6360 | }; | |
6361 | ||
e6fe6649 | 6362 | void __init migration_init(void) |
1da177e4 LT |
6363 | { |
6364 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6365 | int err; |
48f24c4d IM |
6366 | |
6367 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6368 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6369 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6370 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6371 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6372 | } |
6373 | #endif | |
6374 | ||
6375 | #ifdef CONFIG_SMP | |
476f3534 | 6376 | |
3e9830dc | 6377 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6378 | |
099f98c8 GS |
6379 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6380 | { | |
6381 | switch (lvl) { | |
6382 | case SD_LV_NONE: | |
6383 | return "NONE"; | |
6384 | case SD_LV_SIBLING: | |
6385 | return "SIBLING"; | |
6386 | case SD_LV_MC: | |
6387 | return "MC"; | |
6388 | case SD_LV_CPU: | |
6389 | return "CPU"; | |
6390 | case SD_LV_NODE: | |
6391 | return "NODE"; | |
6392 | case SD_LV_ALLNODES: | |
6393 | return "ALLNODES"; | |
6394 | case SD_LV_MAX: | |
6395 | return "MAX"; | |
6396 | ||
6397 | } | |
6398 | return "MAX"; | |
6399 | } | |
6400 | ||
7c16ec58 MT |
6401 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6402 | cpumask_t *groupmask) | |
1da177e4 | 6403 | { |
4dcf6aff | 6404 | struct sched_group *group = sd->groups; |
434d53b0 | 6405 | char str[256]; |
1da177e4 | 6406 | |
434d53b0 | 6407 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6408 | cpus_clear(*groupmask); |
4dcf6aff IM |
6409 | |
6410 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6411 | ||
6412 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6413 | printk("does not load-balance\n"); | |
6414 | if (sd->parent) | |
6415 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6416 | " has parent"); | |
6417 | return -1; | |
41c7ce9a NP |
6418 | } |
6419 | ||
099f98c8 GS |
6420 | printk(KERN_CONT "span %s level %s\n", |
6421 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6422 | |
6423 | if (!cpu_isset(cpu, sd->span)) { | |
6424 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6425 | "CPU%d\n", cpu); | |
6426 | } | |
6427 | if (!cpu_isset(cpu, group->cpumask)) { | |
6428 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6429 | " CPU%d\n", cpu); | |
6430 | } | |
1da177e4 | 6431 | |
4dcf6aff | 6432 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6433 | do { |
4dcf6aff IM |
6434 | if (!group) { |
6435 | printk("\n"); | |
6436 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6437 | break; |
6438 | } | |
6439 | ||
4dcf6aff IM |
6440 | if (!group->__cpu_power) { |
6441 | printk(KERN_CONT "\n"); | |
6442 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6443 | "set\n"); | |
6444 | break; | |
6445 | } | |
1da177e4 | 6446 | |
4dcf6aff IM |
6447 | if (!cpus_weight(group->cpumask)) { |
6448 | printk(KERN_CONT "\n"); | |
6449 | printk(KERN_ERR "ERROR: empty group\n"); | |
6450 | break; | |
6451 | } | |
1da177e4 | 6452 | |
7c16ec58 | 6453 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6454 | printk(KERN_CONT "\n"); |
6455 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6456 | break; | |
6457 | } | |
1da177e4 | 6458 | |
7c16ec58 | 6459 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6460 | |
434d53b0 | 6461 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6462 | printk(KERN_CONT " %s", str); |
1da177e4 | 6463 | |
4dcf6aff IM |
6464 | group = group->next; |
6465 | } while (group != sd->groups); | |
6466 | printk(KERN_CONT "\n"); | |
1da177e4 | 6467 | |
7c16ec58 | 6468 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6469 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6470 | |
7c16ec58 | 6471 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6472 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6473 | "of domain->span\n"); | |
6474 | return 0; | |
6475 | } | |
1da177e4 | 6476 | |
4dcf6aff IM |
6477 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6478 | { | |
7c16ec58 | 6479 | cpumask_t *groupmask; |
4dcf6aff | 6480 | int level = 0; |
1da177e4 | 6481 | |
4dcf6aff IM |
6482 | if (!sd) { |
6483 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6484 | return; | |
6485 | } | |
1da177e4 | 6486 | |
4dcf6aff IM |
6487 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6488 | ||
7c16ec58 MT |
6489 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6490 | if (!groupmask) { | |
6491 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6492 | return; | |
6493 | } | |
6494 | ||
4dcf6aff | 6495 | for (;;) { |
7c16ec58 | 6496 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6497 | break; |
1da177e4 LT |
6498 | level++; |
6499 | sd = sd->parent; | |
33859f7f | 6500 | if (!sd) |
4dcf6aff IM |
6501 | break; |
6502 | } | |
7c16ec58 | 6503 | kfree(groupmask); |
1da177e4 | 6504 | } |
6d6bc0ad | 6505 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6506 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6507 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6508 | |
1a20ff27 | 6509 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6510 | { |
6511 | if (cpus_weight(sd->span) == 1) | |
6512 | return 1; | |
6513 | ||
6514 | /* Following flags need at least 2 groups */ | |
6515 | if (sd->flags & (SD_LOAD_BALANCE | | |
6516 | SD_BALANCE_NEWIDLE | | |
6517 | SD_BALANCE_FORK | | |
89c4710e SS |
6518 | SD_BALANCE_EXEC | |
6519 | SD_SHARE_CPUPOWER | | |
6520 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6521 | if (sd->groups != sd->groups->next) |
6522 | return 0; | |
6523 | } | |
6524 | ||
6525 | /* Following flags don't use groups */ | |
6526 | if (sd->flags & (SD_WAKE_IDLE | | |
6527 | SD_WAKE_AFFINE | | |
6528 | SD_WAKE_BALANCE)) | |
6529 | return 0; | |
6530 | ||
6531 | return 1; | |
6532 | } | |
6533 | ||
48f24c4d IM |
6534 | static int |
6535 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6536 | { |
6537 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6538 | ||
6539 | if (sd_degenerate(parent)) | |
6540 | return 1; | |
6541 | ||
6542 | if (!cpus_equal(sd->span, parent->span)) | |
6543 | return 0; | |
6544 | ||
6545 | /* Does parent contain flags not in child? */ | |
6546 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6547 | if (cflags & SD_WAKE_AFFINE) | |
6548 | pflags &= ~SD_WAKE_BALANCE; | |
6549 | /* Flags needing groups don't count if only 1 group in parent */ | |
6550 | if (parent->groups == parent->groups->next) { | |
6551 | pflags &= ~(SD_LOAD_BALANCE | | |
6552 | SD_BALANCE_NEWIDLE | | |
6553 | SD_BALANCE_FORK | | |
89c4710e SS |
6554 | SD_BALANCE_EXEC | |
6555 | SD_SHARE_CPUPOWER | | |
6556 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6557 | } |
6558 | if (~cflags & pflags) | |
6559 | return 0; | |
6560 | ||
6561 | return 1; | |
6562 | } | |
6563 | ||
57d885fe GH |
6564 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6565 | { | |
6566 | unsigned long flags; | |
57d885fe GH |
6567 | |
6568 | spin_lock_irqsave(&rq->lock, flags); | |
6569 | ||
6570 | if (rq->rd) { | |
6571 | struct root_domain *old_rd = rq->rd; | |
6572 | ||
1f11eb6a GH |
6573 | if (cpu_isset(rq->cpu, old_rd->online)) |
6574 | set_rq_offline(rq); | |
57d885fe | 6575 | |
dc938520 | 6576 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6577 | |
57d885fe GH |
6578 | if (atomic_dec_and_test(&old_rd->refcount)) |
6579 | kfree(old_rd); | |
6580 | } | |
6581 | ||
6582 | atomic_inc(&rd->refcount); | |
6583 | rq->rd = rd; | |
6584 | ||
dc938520 | 6585 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6586 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6587 | set_rq_online(rq); |
57d885fe GH |
6588 | |
6589 | spin_unlock_irqrestore(&rq->lock, flags); | |
6590 | } | |
6591 | ||
dc938520 | 6592 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6593 | { |
6594 | memset(rd, 0, sizeof(*rd)); | |
6595 | ||
dc938520 GH |
6596 | cpus_clear(rd->span); |
6597 | cpus_clear(rd->online); | |
6e0534f2 GH |
6598 | |
6599 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6600 | } |
6601 | ||
6602 | static void init_defrootdomain(void) | |
6603 | { | |
dc938520 | 6604 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6605 | atomic_set(&def_root_domain.refcount, 1); |
6606 | } | |
6607 | ||
dc938520 | 6608 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6609 | { |
6610 | struct root_domain *rd; | |
6611 | ||
6612 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6613 | if (!rd) | |
6614 | return NULL; | |
6615 | ||
dc938520 | 6616 | init_rootdomain(rd); |
57d885fe GH |
6617 | |
6618 | return rd; | |
6619 | } | |
6620 | ||
1da177e4 | 6621 | /* |
0eab9146 | 6622 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6623 | * hold the hotplug lock. |
6624 | */ | |
0eab9146 IM |
6625 | static void |
6626 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6627 | { |
70b97a7f | 6628 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6629 | struct sched_domain *tmp; |
6630 | ||
6631 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6632 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6633 | struct sched_domain *parent = tmp->parent; | |
6634 | if (!parent) | |
6635 | break; | |
1a848870 | 6636 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6637 | tmp->parent = parent->parent; |
1a848870 SS |
6638 | if (parent->parent) |
6639 | parent->parent->child = tmp; | |
6640 | } | |
245af2c7 SS |
6641 | } |
6642 | ||
1a848870 | 6643 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6644 | sd = sd->parent; |
1a848870 SS |
6645 | if (sd) |
6646 | sd->child = NULL; | |
6647 | } | |
1da177e4 LT |
6648 | |
6649 | sched_domain_debug(sd, cpu); | |
6650 | ||
57d885fe | 6651 | rq_attach_root(rq, rd); |
674311d5 | 6652 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6653 | } |
6654 | ||
6655 | /* cpus with isolated domains */ | |
67af63a6 | 6656 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6657 | |
6658 | /* Setup the mask of cpus configured for isolated domains */ | |
6659 | static int __init isolated_cpu_setup(char *str) | |
6660 | { | |
6661 | int ints[NR_CPUS], i; | |
6662 | ||
6663 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6664 | cpus_clear(cpu_isolated_map); | |
6665 | for (i = 1; i <= ints[0]; i++) | |
6666 | if (ints[i] < NR_CPUS) | |
6667 | cpu_set(ints[i], cpu_isolated_map); | |
6668 | return 1; | |
6669 | } | |
6670 | ||
8927f494 | 6671 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6672 | |
6673 | /* | |
6711cab4 SS |
6674 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6675 | * to a function which identifies what group(along with sched group) a CPU | |
6676 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6677 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6678 | * |
6679 | * init_sched_build_groups will build a circular linked list of the groups | |
6680 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6681 | * and ->cpu_power to 0. | |
6682 | */ | |
a616058b | 6683 | static void |
7c16ec58 | 6684 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6685 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6686 | struct sched_group **sg, |
6687 | cpumask_t *tmpmask), | |
6688 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6689 | { |
6690 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6691 | int i; |
6692 | ||
7c16ec58 MT |
6693 | cpus_clear(*covered); |
6694 | ||
6695 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6696 | struct sched_group *sg; |
7c16ec58 | 6697 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6698 | int j; |
6699 | ||
7c16ec58 | 6700 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6701 | continue; |
6702 | ||
7c16ec58 | 6703 | cpus_clear(sg->cpumask); |
5517d86b | 6704 | sg->__cpu_power = 0; |
1da177e4 | 6705 | |
7c16ec58 MT |
6706 | for_each_cpu_mask(j, *span) { |
6707 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6708 | continue; |
6709 | ||
7c16ec58 | 6710 | cpu_set(j, *covered); |
1da177e4 LT |
6711 | cpu_set(j, sg->cpumask); |
6712 | } | |
6713 | if (!first) | |
6714 | first = sg; | |
6715 | if (last) | |
6716 | last->next = sg; | |
6717 | last = sg; | |
6718 | } | |
6719 | last->next = first; | |
6720 | } | |
6721 | ||
9c1cfda2 | 6722 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6723 | |
9c1cfda2 | 6724 | #ifdef CONFIG_NUMA |
198e2f18 | 6725 | |
9c1cfda2 JH |
6726 | /** |
6727 | * find_next_best_node - find the next node to include in a sched_domain | |
6728 | * @node: node whose sched_domain we're building | |
6729 | * @used_nodes: nodes already in the sched_domain | |
6730 | * | |
41a2d6cf | 6731 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6732 | * finds the closest node not already in the @used_nodes map. |
6733 | * | |
6734 | * Should use nodemask_t. | |
6735 | */ | |
c5f59f08 | 6736 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6737 | { |
6738 | int i, n, val, min_val, best_node = 0; | |
6739 | ||
6740 | min_val = INT_MAX; | |
6741 | ||
6742 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6743 | /* Start at @node */ | |
6744 | n = (node + i) % MAX_NUMNODES; | |
6745 | ||
6746 | if (!nr_cpus_node(n)) | |
6747 | continue; | |
6748 | ||
6749 | /* Skip already used nodes */ | |
c5f59f08 | 6750 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6751 | continue; |
6752 | ||
6753 | /* Simple min distance search */ | |
6754 | val = node_distance(node, n); | |
6755 | ||
6756 | if (val < min_val) { | |
6757 | min_val = val; | |
6758 | best_node = n; | |
6759 | } | |
6760 | } | |
6761 | ||
c5f59f08 | 6762 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6763 | return best_node; |
6764 | } | |
6765 | ||
6766 | /** | |
6767 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6768 | * @node: node whose cpumask we're constructing | |
73486722 | 6769 | * @span: resulting cpumask |
9c1cfda2 | 6770 | * |
41a2d6cf | 6771 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6772 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6773 | * out optimally. | |
6774 | */ | |
4bdbaad3 | 6775 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6776 | { |
c5f59f08 | 6777 | nodemask_t used_nodes; |
c5f59f08 | 6778 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6779 | int i; |
9c1cfda2 | 6780 | |
4bdbaad3 | 6781 | cpus_clear(*span); |
c5f59f08 | 6782 | nodes_clear(used_nodes); |
9c1cfda2 | 6783 | |
4bdbaad3 | 6784 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6785 | node_set(node, used_nodes); |
9c1cfda2 JH |
6786 | |
6787 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6788 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6789 | |
c5f59f08 | 6790 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6791 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6792 | } |
9c1cfda2 | 6793 | } |
6d6bc0ad | 6794 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6795 | |
5c45bf27 | 6796 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6797 | |
9c1cfda2 | 6798 | /* |
48f24c4d | 6799 | * SMT sched-domains: |
9c1cfda2 | 6800 | */ |
1da177e4 LT |
6801 | #ifdef CONFIG_SCHED_SMT |
6802 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6803 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6804 | |
41a2d6cf | 6805 | static int |
7c16ec58 MT |
6806 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6807 | cpumask_t *unused) | |
1da177e4 | 6808 | { |
6711cab4 SS |
6809 | if (sg) |
6810 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6811 | return cpu; |
6812 | } | |
6d6bc0ad | 6813 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6814 | |
48f24c4d IM |
6815 | /* |
6816 | * multi-core sched-domains: | |
6817 | */ | |
1e9f28fa SS |
6818 | #ifdef CONFIG_SCHED_MC |
6819 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6820 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6821 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6822 | |
6823 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6824 | static int |
7c16ec58 MT |
6825 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6826 | cpumask_t *mask) | |
1e9f28fa | 6827 | { |
6711cab4 | 6828 | int group; |
7c16ec58 MT |
6829 | |
6830 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6831 | cpus_and(*mask, *mask, *cpu_map); | |
6832 | group = first_cpu(*mask); | |
6711cab4 SS |
6833 | if (sg) |
6834 | *sg = &per_cpu(sched_group_core, group); | |
6835 | return group; | |
1e9f28fa SS |
6836 | } |
6837 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6838 | static int |
7c16ec58 MT |
6839 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6840 | cpumask_t *unused) | |
1e9f28fa | 6841 | { |
6711cab4 SS |
6842 | if (sg) |
6843 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6844 | return cpu; |
6845 | } | |
6846 | #endif | |
6847 | ||
1da177e4 | 6848 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6849 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6850 | |
41a2d6cf | 6851 | static int |
7c16ec58 MT |
6852 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6853 | cpumask_t *mask) | |
1da177e4 | 6854 | { |
6711cab4 | 6855 | int group; |
48f24c4d | 6856 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6857 | *mask = cpu_coregroup_map(cpu); |
6858 | cpus_and(*mask, *mask, *cpu_map); | |
6859 | group = first_cpu(*mask); | |
1e9f28fa | 6860 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6861 | *mask = per_cpu(cpu_sibling_map, cpu); |
6862 | cpus_and(*mask, *mask, *cpu_map); | |
6863 | group = first_cpu(*mask); | |
1da177e4 | 6864 | #else |
6711cab4 | 6865 | group = cpu; |
1da177e4 | 6866 | #endif |
6711cab4 SS |
6867 | if (sg) |
6868 | *sg = &per_cpu(sched_group_phys, group); | |
6869 | return group; | |
1da177e4 LT |
6870 | } |
6871 | ||
6872 | #ifdef CONFIG_NUMA | |
1da177e4 | 6873 | /* |
9c1cfda2 JH |
6874 | * The init_sched_build_groups can't handle what we want to do with node |
6875 | * groups, so roll our own. Now each node has its own list of groups which | |
6876 | * gets dynamically allocated. | |
1da177e4 | 6877 | */ |
9c1cfda2 | 6878 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6879 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6880 | |
9c1cfda2 | 6881 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6882 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6883 | |
6711cab4 | 6884 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6885 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6886 | { |
6711cab4 SS |
6887 | int group; |
6888 | ||
7c16ec58 MT |
6889 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6890 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6891 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6892 | |
6893 | if (sg) | |
6894 | *sg = &per_cpu(sched_group_allnodes, group); | |
6895 | return group; | |
1da177e4 | 6896 | } |
6711cab4 | 6897 | |
08069033 SS |
6898 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6899 | { | |
6900 | struct sched_group *sg = group_head; | |
6901 | int j; | |
6902 | ||
6903 | if (!sg) | |
6904 | return; | |
3a5c359a AK |
6905 | do { |
6906 | for_each_cpu_mask(j, sg->cpumask) { | |
6907 | struct sched_domain *sd; | |
08069033 | 6908 | |
3a5c359a AK |
6909 | sd = &per_cpu(phys_domains, j); |
6910 | if (j != first_cpu(sd->groups->cpumask)) { | |
6911 | /* | |
6912 | * Only add "power" once for each | |
6913 | * physical package. | |
6914 | */ | |
6915 | continue; | |
6916 | } | |
08069033 | 6917 | |
3a5c359a AK |
6918 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6919 | } | |
6920 | sg = sg->next; | |
6921 | } while (sg != group_head); | |
08069033 | 6922 | } |
6d6bc0ad | 6923 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6924 | |
a616058b | 6925 | #ifdef CONFIG_NUMA |
51888ca2 | 6926 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6927 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6928 | { |
a616058b | 6929 | int cpu, i; |
51888ca2 SV |
6930 | |
6931 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6932 | struct sched_group **sched_group_nodes |
6933 | = sched_group_nodes_bycpu[cpu]; | |
6934 | ||
51888ca2 SV |
6935 | if (!sched_group_nodes) |
6936 | continue; | |
6937 | ||
6938 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6939 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6940 | ||
7c16ec58 MT |
6941 | *nodemask = node_to_cpumask(i); |
6942 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6943 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6944 | continue; |
6945 | ||
6946 | if (sg == NULL) | |
6947 | continue; | |
6948 | sg = sg->next; | |
6949 | next_sg: | |
6950 | oldsg = sg; | |
6951 | sg = sg->next; | |
6952 | kfree(oldsg); | |
6953 | if (oldsg != sched_group_nodes[i]) | |
6954 | goto next_sg; | |
6955 | } | |
6956 | kfree(sched_group_nodes); | |
6957 | sched_group_nodes_bycpu[cpu] = NULL; | |
6958 | } | |
51888ca2 | 6959 | } |
6d6bc0ad | 6960 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 6961 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6962 | { |
6963 | } | |
6d6bc0ad | 6964 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6965 | |
89c4710e SS |
6966 | /* |
6967 | * Initialize sched groups cpu_power. | |
6968 | * | |
6969 | * cpu_power indicates the capacity of sched group, which is used while | |
6970 | * distributing the load between different sched groups in a sched domain. | |
6971 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6972 | * there are asymmetries in the topology. If there are asymmetries, group | |
6973 | * having more cpu_power will pickup more load compared to the group having | |
6974 | * less cpu_power. | |
6975 | * | |
6976 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6977 | * the maximum number of tasks a group can handle in the presence of other idle | |
6978 | * or lightly loaded groups in the same sched domain. | |
6979 | */ | |
6980 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6981 | { | |
6982 | struct sched_domain *child; | |
6983 | struct sched_group *group; | |
6984 | ||
6985 | WARN_ON(!sd || !sd->groups); | |
6986 | ||
6987 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6988 | return; | |
6989 | ||
6990 | child = sd->child; | |
6991 | ||
5517d86b ED |
6992 | sd->groups->__cpu_power = 0; |
6993 | ||
89c4710e SS |
6994 | /* |
6995 | * For perf policy, if the groups in child domain share resources | |
6996 | * (for example cores sharing some portions of the cache hierarchy | |
6997 | * or SMT), then set this domain groups cpu_power such that each group | |
6998 | * can handle only one task, when there are other idle groups in the | |
6999 | * same sched domain. | |
7000 | */ | |
7001 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7002 | (child->flags & | |
7003 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7004 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7005 | return; |
7006 | } | |
7007 | ||
89c4710e SS |
7008 | /* |
7009 | * add cpu_power of each child group to this groups cpu_power | |
7010 | */ | |
7011 | group = child->groups; | |
7012 | do { | |
5517d86b | 7013 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7014 | group = group->next; |
7015 | } while (group != child->groups); | |
7016 | } | |
7017 | ||
7c16ec58 MT |
7018 | /* |
7019 | * Initializers for schedule domains | |
7020 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7021 | */ | |
7022 | ||
7023 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7024 | #define SD_INIT_FUNC(type) \ | |
7025 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7026 | { \ | |
7027 | memset(sd, 0, sizeof(*sd)); \ | |
7028 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7029 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7030 | } |
7031 | ||
7032 | SD_INIT_FUNC(CPU) | |
7033 | #ifdef CONFIG_NUMA | |
7034 | SD_INIT_FUNC(ALLNODES) | |
7035 | SD_INIT_FUNC(NODE) | |
7036 | #endif | |
7037 | #ifdef CONFIG_SCHED_SMT | |
7038 | SD_INIT_FUNC(SIBLING) | |
7039 | #endif | |
7040 | #ifdef CONFIG_SCHED_MC | |
7041 | SD_INIT_FUNC(MC) | |
7042 | #endif | |
7043 | ||
7044 | /* | |
7045 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7046 | * space as the usage in build_sched_domains() dictates. Used only | |
7047 | * if the amount of space is significant. | |
7048 | */ | |
7049 | struct allmasks { | |
7050 | cpumask_t tmpmask; /* make this one first */ | |
7051 | union { | |
7052 | cpumask_t nodemask; | |
7053 | cpumask_t this_sibling_map; | |
7054 | cpumask_t this_core_map; | |
7055 | }; | |
7056 | cpumask_t send_covered; | |
7057 | ||
7058 | #ifdef CONFIG_NUMA | |
7059 | cpumask_t domainspan; | |
7060 | cpumask_t covered; | |
7061 | cpumask_t notcovered; | |
7062 | #endif | |
7063 | }; | |
7064 | ||
7065 | #if NR_CPUS > 128 | |
7066 | #define SCHED_CPUMASK_ALLOC 1 | |
7067 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7068 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7069 | #else | |
7070 | #define SCHED_CPUMASK_ALLOC 0 | |
7071 | #define SCHED_CPUMASK_FREE(v) | |
7072 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7073 | #endif | |
7074 | ||
7075 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7076 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7077 | ||
1d3504fc HS |
7078 | static int default_relax_domain_level = -1; |
7079 | ||
7080 | static int __init setup_relax_domain_level(char *str) | |
7081 | { | |
30e0e178 LZ |
7082 | unsigned long val; |
7083 | ||
7084 | val = simple_strtoul(str, NULL, 0); | |
7085 | if (val < SD_LV_MAX) | |
7086 | default_relax_domain_level = val; | |
7087 | ||
1d3504fc HS |
7088 | return 1; |
7089 | } | |
7090 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7091 | ||
7092 | static void set_domain_attribute(struct sched_domain *sd, | |
7093 | struct sched_domain_attr *attr) | |
7094 | { | |
7095 | int request; | |
7096 | ||
7097 | if (!attr || attr->relax_domain_level < 0) { | |
7098 | if (default_relax_domain_level < 0) | |
7099 | return; | |
7100 | else | |
7101 | request = default_relax_domain_level; | |
7102 | } else | |
7103 | request = attr->relax_domain_level; | |
7104 | if (request < sd->level) { | |
7105 | /* turn off idle balance on this domain */ | |
7106 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7107 | } else { | |
7108 | /* turn on idle balance on this domain */ | |
7109 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7110 | } | |
7111 | } | |
7112 | ||
1da177e4 | 7113 | /* |
1a20ff27 DG |
7114 | * Build sched domains for a given set of cpus and attach the sched domains |
7115 | * to the individual cpus | |
1da177e4 | 7116 | */ |
1d3504fc HS |
7117 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7118 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7119 | { |
7120 | int i; | |
57d885fe | 7121 | struct root_domain *rd; |
7c16ec58 MT |
7122 | SCHED_CPUMASK_DECLARE(allmasks); |
7123 | cpumask_t *tmpmask; | |
d1b55138 JH |
7124 | #ifdef CONFIG_NUMA |
7125 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7126 | int sd_allnodes = 0; |
d1b55138 JH |
7127 | |
7128 | /* | |
7129 | * Allocate the per-node list of sched groups | |
7130 | */ | |
5cf9f062 | 7131 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 7132 | GFP_KERNEL); |
d1b55138 JH |
7133 | if (!sched_group_nodes) { |
7134 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7135 | return -ENOMEM; |
d1b55138 | 7136 | } |
d1b55138 | 7137 | #endif |
1da177e4 | 7138 | |
dc938520 | 7139 | rd = alloc_rootdomain(); |
57d885fe GH |
7140 | if (!rd) { |
7141 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7142 | #ifdef CONFIG_NUMA |
7143 | kfree(sched_group_nodes); | |
7144 | #endif | |
57d885fe GH |
7145 | return -ENOMEM; |
7146 | } | |
7147 | ||
7c16ec58 MT |
7148 | #if SCHED_CPUMASK_ALLOC |
7149 | /* get space for all scratch cpumask variables */ | |
7150 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7151 | if (!allmasks) { | |
7152 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7153 | kfree(rd); | |
7154 | #ifdef CONFIG_NUMA | |
7155 | kfree(sched_group_nodes); | |
7156 | #endif | |
7157 | return -ENOMEM; | |
7158 | } | |
7159 | #endif | |
7160 | tmpmask = (cpumask_t *)allmasks; | |
7161 | ||
7162 | ||
7163 | #ifdef CONFIG_NUMA | |
7164 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7165 | #endif | |
7166 | ||
1da177e4 | 7167 | /* |
1a20ff27 | 7168 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7169 | */ |
1a20ff27 | 7170 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 7171 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7172 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7173 | |
7c16ec58 MT |
7174 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7175 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7176 | |
7177 | #ifdef CONFIG_NUMA | |
dd41f596 | 7178 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7179 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7180 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7181 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7182 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7183 | sd->span = *cpu_map; |
7c16ec58 | 7184 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7185 | p = sd; |
6711cab4 | 7186 | sd_allnodes = 1; |
9c1cfda2 JH |
7187 | } else |
7188 | p = NULL; | |
7189 | ||
1da177e4 | 7190 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7191 | SD_INIT(sd, NODE); |
1d3504fc | 7192 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7193 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7194 | sd->parent = p; |
1a848870 SS |
7195 | if (p) |
7196 | p->child = sd; | |
9c1cfda2 | 7197 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7198 | #endif |
7199 | ||
7200 | p = sd; | |
7201 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7202 | SD_INIT(sd, CPU); |
1d3504fc | 7203 | set_domain_attribute(sd, attr); |
7c16ec58 | 7204 | sd->span = *nodemask; |
1da177e4 | 7205 | sd->parent = p; |
1a848870 SS |
7206 | if (p) |
7207 | p->child = sd; | |
7c16ec58 | 7208 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7209 | |
1e9f28fa SS |
7210 | #ifdef CONFIG_SCHED_MC |
7211 | p = sd; | |
7212 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7213 | SD_INIT(sd, MC); |
1d3504fc | 7214 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7215 | sd->span = cpu_coregroup_map(i); |
7216 | cpus_and(sd->span, sd->span, *cpu_map); | |
7217 | sd->parent = p; | |
1a848870 | 7218 | p->child = sd; |
7c16ec58 | 7219 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7220 | #endif |
7221 | ||
1da177e4 LT |
7222 | #ifdef CONFIG_SCHED_SMT |
7223 | p = sd; | |
7224 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7225 | SD_INIT(sd, SIBLING); |
1d3504fc | 7226 | set_domain_attribute(sd, attr); |
d5a7430d | 7227 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7228 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7229 | sd->parent = p; |
1a848870 | 7230 | p->child = sd; |
7c16ec58 | 7231 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7232 | #endif |
7233 | } | |
7234 | ||
7235 | #ifdef CONFIG_SCHED_SMT | |
7236 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 7237 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
7238 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7239 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7240 | ||
7241 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7242 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7243 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7244 | continue; |
7245 | ||
dd41f596 | 7246 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7247 | &cpu_to_cpu_group, |
7248 | send_covered, tmpmask); | |
1da177e4 LT |
7249 | } |
7250 | #endif | |
7251 | ||
1e9f28fa SS |
7252 | #ifdef CONFIG_SCHED_MC |
7253 | /* Set up multi-core groups */ | |
7254 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
7255 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7256 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7257 | ||
7258 | *this_core_map = cpu_coregroup_map(i); | |
7259 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7260 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7261 | continue; |
7c16ec58 | 7262 | |
dd41f596 | 7263 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7264 | &cpu_to_core_group, |
7265 | send_covered, tmpmask); | |
1e9f28fa SS |
7266 | } |
7267 | #endif | |
7268 | ||
1da177e4 LT |
7269 | /* Set up physical groups */ |
7270 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
7271 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7272 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7273 | |
7c16ec58 MT |
7274 | *nodemask = node_to_cpumask(i); |
7275 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7276 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7277 | continue; |
7278 | ||
7c16ec58 MT |
7279 | init_sched_build_groups(nodemask, cpu_map, |
7280 | &cpu_to_phys_group, | |
7281 | send_covered, tmpmask); | |
1da177e4 LT |
7282 | } |
7283 | ||
7284 | #ifdef CONFIG_NUMA | |
7285 | /* Set up node groups */ | |
7c16ec58 MT |
7286 | if (sd_allnodes) { |
7287 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7288 | ||
7289 | init_sched_build_groups(cpu_map, cpu_map, | |
7290 | &cpu_to_allnodes_group, | |
7291 | send_covered, tmpmask); | |
7292 | } | |
9c1cfda2 JH |
7293 | |
7294 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7295 | /* Set up node groups */ | |
7296 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7297 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7298 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7299 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7300 | int j; |
7301 | ||
7c16ec58 MT |
7302 | *nodemask = node_to_cpumask(i); |
7303 | cpus_clear(*covered); | |
7304 | ||
7305 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7306 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7307 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7308 | continue; |
d1b55138 | 7309 | } |
9c1cfda2 | 7310 | |
4bdbaad3 | 7311 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7312 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7313 | |
15f0b676 | 7314 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7315 | if (!sg) { |
7316 | printk(KERN_WARNING "Can not alloc domain group for " | |
7317 | "node %d\n", i); | |
7318 | goto error; | |
7319 | } | |
9c1cfda2 | 7320 | sched_group_nodes[i] = sg; |
7c16ec58 | 7321 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7322 | struct sched_domain *sd; |
9761eea8 | 7323 | |
9c1cfda2 JH |
7324 | sd = &per_cpu(node_domains, j); |
7325 | sd->groups = sg; | |
9c1cfda2 | 7326 | } |
5517d86b | 7327 | sg->__cpu_power = 0; |
7c16ec58 | 7328 | sg->cpumask = *nodemask; |
51888ca2 | 7329 | sg->next = sg; |
7c16ec58 | 7330 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7331 | prev = sg; |
7332 | ||
7333 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7334 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7335 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7336 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7337 | |
7c16ec58 MT |
7338 | cpus_complement(*notcovered, *covered); |
7339 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7340 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7341 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7342 | break; |
7343 | ||
7c16ec58 MT |
7344 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7345 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7346 | continue; |
7347 | ||
15f0b676 SV |
7348 | sg = kmalloc_node(sizeof(struct sched_group), |
7349 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7350 | if (!sg) { |
7351 | printk(KERN_WARNING | |
7352 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7353 | goto error; |
9c1cfda2 | 7354 | } |
5517d86b | 7355 | sg->__cpu_power = 0; |
7c16ec58 | 7356 | sg->cpumask = *tmpmask; |
51888ca2 | 7357 | sg->next = prev->next; |
7c16ec58 | 7358 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7359 | prev->next = sg; |
7360 | prev = sg; | |
7361 | } | |
9c1cfda2 | 7362 | } |
1da177e4 LT |
7363 | #endif |
7364 | ||
7365 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7366 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7367 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7368 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7369 | ||
89c4710e | 7370 | init_sched_groups_power(i, sd); |
5c45bf27 | 7371 | } |
1da177e4 | 7372 | #endif |
1e9f28fa | 7373 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7374 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7375 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7376 | ||
89c4710e | 7377 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7378 | } |
7379 | #endif | |
1e9f28fa | 7380 | |
5c45bf27 | 7381 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7382 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7383 | ||
89c4710e | 7384 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7385 | } |
7386 | ||
9c1cfda2 | 7387 | #ifdef CONFIG_NUMA |
08069033 SS |
7388 | for (i = 0; i < MAX_NUMNODES; i++) |
7389 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7390 | |
6711cab4 SS |
7391 | if (sd_allnodes) { |
7392 | struct sched_group *sg; | |
f712c0c7 | 7393 | |
7c16ec58 MT |
7394 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7395 | tmpmask); | |
f712c0c7 SS |
7396 | init_numa_sched_groups_power(sg); |
7397 | } | |
9c1cfda2 JH |
7398 | #endif |
7399 | ||
1da177e4 | 7400 | /* Attach the domains */ |
1a20ff27 | 7401 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7402 | struct sched_domain *sd; |
7403 | #ifdef CONFIG_SCHED_SMT | |
7404 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7405 | #elif defined(CONFIG_SCHED_MC) |
7406 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7407 | #else |
7408 | sd = &per_cpu(phys_domains, i); | |
7409 | #endif | |
57d885fe | 7410 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7411 | } |
51888ca2 | 7412 | |
7c16ec58 | 7413 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7414 | return 0; |
7415 | ||
a616058b | 7416 | #ifdef CONFIG_NUMA |
51888ca2 | 7417 | error: |
7c16ec58 MT |
7418 | free_sched_groups(cpu_map, tmpmask); |
7419 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7420 | return -ENOMEM; |
a616058b | 7421 | #endif |
1da177e4 | 7422 | } |
029190c5 | 7423 | |
1d3504fc HS |
7424 | static int build_sched_domains(const cpumask_t *cpu_map) |
7425 | { | |
7426 | return __build_sched_domains(cpu_map, NULL); | |
7427 | } | |
7428 | ||
029190c5 PJ |
7429 | static cpumask_t *doms_cur; /* current sched domains */ |
7430 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7431 | static struct sched_domain_attr *dattr_cur; |
7432 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7433 | |
7434 | /* | |
7435 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7436 | * cpumask_t) fails, then fallback to a single sched domain, | |
7437 | * as determined by the single cpumask_t fallback_doms. | |
7438 | */ | |
7439 | static cpumask_t fallback_doms; | |
7440 | ||
22e52b07 HC |
7441 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7442 | { | |
7443 | } | |
7444 | ||
5c8e1ed1 MK |
7445 | /* |
7446 | * Free current domain masks. | |
7447 | * Called after all cpus are attached to NULL domain. | |
7448 | */ | |
7449 | static void free_sched_domains(void) | |
7450 | { | |
7451 | ndoms_cur = 0; | |
7452 | if (doms_cur != &fallback_doms) | |
7453 | kfree(doms_cur); | |
7454 | doms_cur = &fallback_doms; | |
7455 | } | |
7456 | ||
1a20ff27 | 7457 | /* |
41a2d6cf | 7458 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7459 | * For now this just excludes isolated cpus, but could be used to |
7460 | * exclude other special cases in the future. | |
1a20ff27 | 7461 | */ |
51888ca2 | 7462 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7463 | { |
7378547f MM |
7464 | int err; |
7465 | ||
22e52b07 | 7466 | arch_update_cpu_topology(); |
029190c5 PJ |
7467 | ndoms_cur = 1; |
7468 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7469 | if (!doms_cur) | |
7470 | doms_cur = &fallback_doms; | |
7471 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7472 | dattr_cur = NULL; |
7378547f | 7473 | err = build_sched_domains(doms_cur); |
6382bc90 | 7474 | register_sched_domain_sysctl(); |
7378547f MM |
7475 | |
7476 | return err; | |
1a20ff27 DG |
7477 | } |
7478 | ||
7c16ec58 MT |
7479 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7480 | cpumask_t *tmpmask) | |
1da177e4 | 7481 | { |
7c16ec58 | 7482 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7483 | } |
1da177e4 | 7484 | |
1a20ff27 DG |
7485 | /* |
7486 | * Detach sched domains from a group of cpus specified in cpu_map | |
7487 | * These cpus will now be attached to the NULL domain | |
7488 | */ | |
858119e1 | 7489 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7490 | { |
7c16ec58 | 7491 | cpumask_t tmpmask; |
1a20ff27 DG |
7492 | int i; |
7493 | ||
6382bc90 MM |
7494 | unregister_sched_domain_sysctl(); |
7495 | ||
1a20ff27 | 7496 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7497 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7498 | synchronize_sched(); |
7c16ec58 | 7499 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7500 | } |
7501 | ||
1d3504fc HS |
7502 | /* handle null as "default" */ |
7503 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7504 | struct sched_domain_attr *new, int idx_new) | |
7505 | { | |
7506 | struct sched_domain_attr tmp; | |
7507 | ||
7508 | /* fast path */ | |
7509 | if (!new && !cur) | |
7510 | return 1; | |
7511 | ||
7512 | tmp = SD_ATTR_INIT; | |
7513 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7514 | new ? (new + idx_new) : &tmp, | |
7515 | sizeof(struct sched_domain_attr)); | |
7516 | } | |
7517 | ||
029190c5 PJ |
7518 | /* |
7519 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7520 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7521 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7522 | * It destroys each deleted domain and builds each new domain. | |
7523 | * | |
7524 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7525 | * The masks don't intersect (don't overlap.) We should setup one |
7526 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7527 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7528 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7529 | * it as it is. | |
7530 | * | |
41a2d6cf IM |
7531 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7532 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7533 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7534 | * and partition_sched_domains() will fallback to the single partition | |
7535 | * 'fallback_doms'. | |
7536 | * | |
7537 | * Call with hotplug lock held | |
7538 | */ | |
1d3504fc HS |
7539 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7540 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7541 | { |
7542 | int i, j; | |
7543 | ||
712555ee | 7544 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7545 | |
7378547f MM |
7546 | /* always unregister in case we don't destroy any domains */ |
7547 | unregister_sched_domain_sysctl(); | |
7548 | ||
029190c5 PJ |
7549 | if (doms_new == NULL) { |
7550 | ndoms_new = 1; | |
7551 | doms_new = &fallback_doms; | |
7552 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
1d3504fc | 7553 | dattr_new = NULL; |
029190c5 PJ |
7554 | } |
7555 | ||
7556 | /* Destroy deleted domains */ | |
7557 | for (i = 0; i < ndoms_cur; i++) { | |
7558 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7559 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7560 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7561 | goto match1; |
7562 | } | |
7563 | /* no match - a current sched domain not in new doms_new[] */ | |
7564 | detach_destroy_domains(doms_cur + i); | |
7565 | match1: | |
7566 | ; | |
7567 | } | |
7568 | ||
7569 | /* Build new domains */ | |
7570 | for (i = 0; i < ndoms_new; i++) { | |
7571 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7572 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7573 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7574 | goto match2; |
7575 | } | |
7576 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7577 | __build_sched_domains(doms_new + i, |
7578 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7579 | match2: |
7580 | ; | |
7581 | } | |
7582 | ||
7583 | /* Remember the new sched domains */ | |
7584 | if (doms_cur != &fallback_doms) | |
7585 | kfree(doms_cur); | |
1d3504fc | 7586 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7587 | doms_cur = doms_new; |
1d3504fc | 7588 | dattr_cur = dattr_new; |
029190c5 | 7589 | ndoms_cur = ndoms_new; |
7378547f MM |
7590 | |
7591 | register_sched_domain_sysctl(); | |
a1835615 | 7592 | |
712555ee | 7593 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7594 | } |
7595 | ||
5c45bf27 | 7596 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7597 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7598 | { |
7599 | int err; | |
7600 | ||
95402b38 | 7601 | get_online_cpus(); |
712555ee | 7602 | mutex_lock(&sched_domains_mutex); |
5c45bf27 | 7603 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7604 | free_sched_domains(); |
5c45bf27 | 7605 | err = arch_init_sched_domains(&cpu_online_map); |
712555ee | 7606 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7607 | put_online_cpus(); |
5c45bf27 SS |
7608 | |
7609 | return err; | |
7610 | } | |
7611 | ||
7612 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7613 | { | |
7614 | int ret; | |
7615 | ||
7616 | if (buf[0] != '0' && buf[0] != '1') | |
7617 | return -EINVAL; | |
7618 | ||
7619 | if (smt) | |
7620 | sched_smt_power_savings = (buf[0] == '1'); | |
7621 | else | |
7622 | sched_mc_power_savings = (buf[0] == '1'); | |
7623 | ||
7624 | ret = arch_reinit_sched_domains(); | |
7625 | ||
7626 | return ret ? ret : count; | |
7627 | } | |
7628 | ||
5c45bf27 SS |
7629 | #ifdef CONFIG_SCHED_MC |
7630 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7631 | { | |
7632 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7633 | } | |
48f24c4d IM |
7634 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7635 | const char *buf, size_t count) | |
5c45bf27 SS |
7636 | { |
7637 | return sched_power_savings_store(buf, count, 0); | |
7638 | } | |
6707de00 AB |
7639 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7640 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7641 | #endif |
7642 | ||
7643 | #ifdef CONFIG_SCHED_SMT | |
7644 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7645 | { | |
7646 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7647 | } | |
48f24c4d IM |
7648 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7649 | const char *buf, size_t count) | |
5c45bf27 SS |
7650 | { |
7651 | return sched_power_savings_store(buf, count, 1); | |
7652 | } | |
6707de00 AB |
7653 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7654 | sched_smt_power_savings_store); | |
7655 | #endif | |
7656 | ||
7657 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7658 | { | |
7659 | int err = 0; | |
7660 | ||
7661 | #ifdef CONFIG_SCHED_SMT | |
7662 | if (smt_capable()) | |
7663 | err = sysfs_create_file(&cls->kset.kobj, | |
7664 | &attr_sched_smt_power_savings.attr); | |
7665 | #endif | |
7666 | #ifdef CONFIG_SCHED_MC | |
7667 | if (!err && mc_capable()) | |
7668 | err = sysfs_create_file(&cls->kset.kobj, | |
7669 | &attr_sched_mc_power_savings.attr); | |
7670 | #endif | |
7671 | return err; | |
7672 | } | |
6d6bc0ad | 7673 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7674 | |
1da177e4 | 7675 | /* |
41a2d6cf | 7676 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7677 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7678 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7679 | * which will prevent rebalancing while the sched domains are recalculated. |
7680 | */ | |
7681 | static int update_sched_domains(struct notifier_block *nfb, | |
7682 | unsigned long action, void *hcpu) | |
7683 | { | |
7def2be1 PZ |
7684 | int cpu = (int)(long)hcpu; |
7685 | ||
1da177e4 | 7686 | switch (action) { |
1da177e4 | 7687 | case CPU_DOWN_PREPARE: |
8bb78442 | 7688 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 PZ |
7689 | disable_runtime(cpu_rq(cpu)); |
7690 | /* fall-through */ | |
7691 | case CPU_UP_PREPARE: | |
7692 | case CPU_UP_PREPARE_FROZEN: | |
1a20ff27 | 7693 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7694 | free_sched_domains(); |
1da177e4 LT |
7695 | return NOTIFY_OK; |
7696 | ||
7def2be1 | 7697 | |
1da177e4 | 7698 | case CPU_DOWN_FAILED: |
8bb78442 | 7699 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7700 | case CPU_ONLINE: |
8bb78442 | 7701 | case CPU_ONLINE_FROZEN: |
7def2be1 PZ |
7702 | enable_runtime(cpu_rq(cpu)); |
7703 | /* fall-through */ | |
7704 | case CPU_UP_CANCELED: | |
7705 | case CPU_UP_CANCELED_FROZEN: | |
1da177e4 | 7706 | case CPU_DEAD: |
8bb78442 | 7707 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7708 | /* |
7709 | * Fall through and re-initialise the domains. | |
7710 | */ | |
7711 | break; | |
7712 | default: | |
7713 | return NOTIFY_DONE; | |
7714 | } | |
7715 | ||
5c8e1ed1 MK |
7716 | #ifndef CONFIG_CPUSETS |
7717 | /* | |
7718 | * Create default domain partitioning if cpusets are disabled. | |
7719 | * Otherwise we let cpusets rebuild the domains based on the | |
7720 | * current setup. | |
7721 | */ | |
7722 | ||
1da177e4 | 7723 | /* The hotplug lock is already held by cpu_up/cpu_down */ |
1a20ff27 | 7724 | arch_init_sched_domains(&cpu_online_map); |
5c8e1ed1 | 7725 | #endif |
1da177e4 LT |
7726 | |
7727 | return NOTIFY_OK; | |
7728 | } | |
1da177e4 LT |
7729 | |
7730 | void __init sched_init_smp(void) | |
7731 | { | |
5c1e1767 NP |
7732 | cpumask_t non_isolated_cpus; |
7733 | ||
434d53b0 MT |
7734 | #if defined(CONFIG_NUMA) |
7735 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7736 | GFP_KERNEL); | |
7737 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7738 | #endif | |
95402b38 | 7739 | get_online_cpus(); |
712555ee | 7740 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7741 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7742 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7743 | if (cpus_empty(non_isolated_cpus)) |
7744 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7745 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7746 | put_online_cpus(); |
1da177e4 LT |
7747 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7748 | hotcpu_notifier(update_sched_domains, 0); | |
b328ca18 | 7749 | init_hrtick(); |
5c1e1767 NP |
7750 | |
7751 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7752 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7753 | BUG(); |
19978ca6 | 7754 | sched_init_granularity(); |
1da177e4 LT |
7755 | } |
7756 | #else | |
7757 | void __init sched_init_smp(void) | |
7758 | { | |
19978ca6 | 7759 | sched_init_granularity(); |
1da177e4 LT |
7760 | } |
7761 | #endif /* CONFIG_SMP */ | |
7762 | ||
7763 | int in_sched_functions(unsigned long addr) | |
7764 | { | |
1da177e4 LT |
7765 | return in_lock_functions(addr) || |
7766 | (addr >= (unsigned long)__sched_text_start | |
7767 | && addr < (unsigned long)__sched_text_end); | |
7768 | } | |
7769 | ||
a9957449 | 7770 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7771 | { |
7772 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7773 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7774 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7775 | cfs_rq->rq = rq; | |
7776 | #endif | |
67e9fb2a | 7777 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7778 | } |
7779 | ||
fa85ae24 PZ |
7780 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7781 | { | |
7782 | struct rt_prio_array *array; | |
7783 | int i; | |
7784 | ||
7785 | array = &rt_rq->active; | |
7786 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
20b6331b | 7787 | INIT_LIST_HEAD(array->queue + i); |
fa85ae24 PZ |
7788 | __clear_bit(i, array->bitmap); |
7789 | } | |
7790 | /* delimiter for bitsearch: */ | |
7791 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7792 | ||
052f1dc7 | 7793 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7794 | rt_rq->highest_prio = MAX_RT_PRIO; |
7795 | #endif | |
fa85ae24 PZ |
7796 | #ifdef CONFIG_SMP |
7797 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7798 | rt_rq->overloaded = 0; |
7799 | #endif | |
7800 | ||
7801 | rt_rq->rt_time = 0; | |
7802 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7803 | rt_rq->rt_runtime = 0; |
7804 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7805 | |
052f1dc7 | 7806 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7807 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7808 | rt_rq->rq = rq; |
7809 | #endif | |
fa85ae24 PZ |
7810 | } |
7811 | ||
6f505b16 | 7812 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7813 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7814 | struct sched_entity *se, int cpu, int add, | |
7815 | struct sched_entity *parent) | |
6f505b16 | 7816 | { |
ec7dc8ac | 7817 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7818 | tg->cfs_rq[cpu] = cfs_rq; |
7819 | init_cfs_rq(cfs_rq, rq); | |
7820 | cfs_rq->tg = tg; | |
7821 | if (add) | |
7822 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7823 | ||
7824 | tg->se[cpu] = se; | |
354d60c2 DG |
7825 | /* se could be NULL for init_task_group */ |
7826 | if (!se) | |
7827 | return; | |
7828 | ||
ec7dc8ac DG |
7829 | if (!parent) |
7830 | se->cfs_rq = &rq->cfs; | |
7831 | else | |
7832 | se->cfs_rq = parent->my_q; | |
7833 | ||
6f505b16 PZ |
7834 | se->my_q = cfs_rq; |
7835 | se->load.weight = tg->shares; | |
e05510d0 | 7836 | se->load.inv_weight = 0; |
ec7dc8ac | 7837 | se->parent = parent; |
6f505b16 | 7838 | } |
052f1dc7 | 7839 | #endif |
6f505b16 | 7840 | |
052f1dc7 | 7841 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7842 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7843 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7844 | struct sched_rt_entity *parent) | |
6f505b16 | 7845 | { |
ec7dc8ac DG |
7846 | struct rq *rq = cpu_rq(cpu); |
7847 | ||
6f505b16 PZ |
7848 | tg->rt_rq[cpu] = rt_rq; |
7849 | init_rt_rq(rt_rq, rq); | |
7850 | rt_rq->tg = tg; | |
7851 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7852 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7853 | if (add) |
7854 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7855 | ||
7856 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7857 | if (!rt_se) |
7858 | return; | |
7859 | ||
ec7dc8ac DG |
7860 | if (!parent) |
7861 | rt_se->rt_rq = &rq->rt; | |
7862 | else | |
7863 | rt_se->rt_rq = parent->my_q; | |
7864 | ||
6f505b16 | 7865 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7866 | rt_se->parent = parent; |
6f505b16 PZ |
7867 | INIT_LIST_HEAD(&rt_se->run_list); |
7868 | } | |
7869 | #endif | |
7870 | ||
1da177e4 LT |
7871 | void __init sched_init(void) |
7872 | { | |
dd41f596 | 7873 | int i, j; |
434d53b0 MT |
7874 | unsigned long alloc_size = 0, ptr; |
7875 | ||
7876 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7877 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7878 | #endif | |
7879 | #ifdef CONFIG_RT_GROUP_SCHED | |
7880 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7881 | #endif |
7882 | #ifdef CONFIG_USER_SCHED | |
7883 | alloc_size *= 2; | |
434d53b0 MT |
7884 | #endif |
7885 | /* | |
7886 | * As sched_init() is called before page_alloc is setup, | |
7887 | * we use alloc_bootmem(). | |
7888 | */ | |
7889 | if (alloc_size) { | |
5a9d3225 | 7890 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7891 | |
7892 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7893 | init_task_group.se = (struct sched_entity **)ptr; | |
7894 | ptr += nr_cpu_ids * sizeof(void **); | |
7895 | ||
7896 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7897 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7898 | |
7899 | #ifdef CONFIG_USER_SCHED | |
7900 | root_task_group.se = (struct sched_entity **)ptr; | |
7901 | ptr += nr_cpu_ids * sizeof(void **); | |
7902 | ||
7903 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7904 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7905 | #endif /* CONFIG_USER_SCHED */ |
7906 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7907 | #ifdef CONFIG_RT_GROUP_SCHED |
7908 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7909 | ptr += nr_cpu_ids * sizeof(void **); | |
7910 | ||
7911 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7912 | ptr += nr_cpu_ids * sizeof(void **); |
7913 | ||
7914 | #ifdef CONFIG_USER_SCHED | |
7915 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7916 | ptr += nr_cpu_ids * sizeof(void **); | |
7917 | ||
7918 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7919 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7920 | #endif /* CONFIG_USER_SCHED */ |
7921 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7922 | } |
dd41f596 | 7923 | |
57d885fe GH |
7924 | #ifdef CONFIG_SMP |
7925 | init_defrootdomain(); | |
7926 | #endif | |
7927 | ||
d0b27fa7 PZ |
7928 | init_rt_bandwidth(&def_rt_bandwidth, |
7929 | global_rt_period(), global_rt_runtime()); | |
7930 | ||
7931 | #ifdef CONFIG_RT_GROUP_SCHED | |
7932 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7933 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7934 | #ifdef CONFIG_USER_SCHED |
7935 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7936 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7937 | #endif /* CONFIG_USER_SCHED */ |
7938 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7939 | |
052f1dc7 | 7940 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7941 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7942 | INIT_LIST_HEAD(&init_task_group.children); |
7943 | ||
7944 | #ifdef CONFIG_USER_SCHED | |
7945 | INIT_LIST_HEAD(&root_task_group.children); | |
7946 | init_task_group.parent = &root_task_group; | |
7947 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7948 | #endif /* CONFIG_USER_SCHED */ |
7949 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7950 | |
0a945022 | 7951 | for_each_possible_cpu(i) { |
70b97a7f | 7952 | struct rq *rq; |
1da177e4 LT |
7953 | |
7954 | rq = cpu_rq(i); | |
7955 | spin_lock_init(&rq->lock); | |
fcb99371 | 7956 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7957 | rq->nr_running = 0; |
dd41f596 | 7958 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7959 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7960 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7961 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7962 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7963 | #ifdef CONFIG_CGROUP_SCHED |
7964 | /* | |
7965 | * How much cpu bandwidth does init_task_group get? | |
7966 | * | |
7967 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7968 | * gets 100% of the cpu resources in the system. This overall | |
7969 | * system cpu resource is divided among the tasks of | |
7970 | * init_task_group and its child task-groups in a fair manner, | |
7971 | * based on each entity's (task or task-group's) weight | |
7972 | * (se->load.weight). | |
7973 | * | |
7974 | * In other words, if init_task_group has 10 tasks of weight | |
7975 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7976 | * then A0's share of the cpu resource is: | |
7977 | * | |
7978 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7979 | * | |
7980 | * We achieve this by letting init_task_group's tasks sit | |
7981 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7982 | */ | |
ec7dc8ac | 7983 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7984 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
7985 | root_task_group.shares = NICE_0_LOAD; |
7986 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
7987 | /* |
7988 | * In case of task-groups formed thr' the user id of tasks, | |
7989 | * init_task_group represents tasks belonging to root user. | |
7990 | * Hence it forms a sibling of all subsequent groups formed. | |
7991 | * In this case, init_task_group gets only a fraction of overall | |
7992 | * system cpu resource, based on the weight assigned to root | |
7993 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
7994 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
7995 | * (init_cfs_rq) and having one entity represent this group of | |
7996 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
7997 | */ | |
ec7dc8ac | 7998 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 7999 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8000 | &per_cpu(init_sched_entity, i), i, 1, |
8001 | root_task_group.se[i]); | |
6f505b16 | 8002 | |
052f1dc7 | 8003 | #endif |
354d60c2 DG |
8004 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8005 | ||
8006 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8007 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8008 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8009 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8010 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8011 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8012 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8013 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8014 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8015 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8016 | root_task_group.rt_se[i]); | |
354d60c2 | 8017 | #endif |
dd41f596 | 8018 | #endif |
1da177e4 | 8019 | |
dd41f596 IM |
8020 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8021 | rq->cpu_load[j] = 0; | |
1da177e4 | 8022 | #ifdef CONFIG_SMP |
41c7ce9a | 8023 | rq->sd = NULL; |
57d885fe | 8024 | rq->rd = NULL; |
1da177e4 | 8025 | rq->active_balance = 0; |
dd41f596 | 8026 | rq->next_balance = jiffies; |
1da177e4 | 8027 | rq->push_cpu = 0; |
0a2966b4 | 8028 | rq->cpu = i; |
1f11eb6a | 8029 | rq->online = 0; |
1da177e4 LT |
8030 | rq->migration_thread = NULL; |
8031 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8032 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8033 | #endif |
8f4d37ec | 8034 | init_rq_hrtick(rq); |
1da177e4 | 8035 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8036 | } |
8037 | ||
2dd73a4f | 8038 | set_load_weight(&init_task); |
b50f60ce | 8039 | |
e107be36 AK |
8040 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8041 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8042 | #endif | |
8043 | ||
c9819f45 CL |
8044 | #ifdef CONFIG_SMP |
8045 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
8046 | #endif | |
8047 | ||
b50f60ce HC |
8048 | #ifdef CONFIG_RT_MUTEXES |
8049 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8050 | #endif | |
8051 | ||
1da177e4 LT |
8052 | /* |
8053 | * The boot idle thread does lazy MMU switching as well: | |
8054 | */ | |
8055 | atomic_inc(&init_mm.mm_count); | |
8056 | enter_lazy_tlb(&init_mm, current); | |
8057 | ||
8058 | /* | |
8059 | * Make us the idle thread. Technically, schedule() should not be | |
8060 | * called from this thread, however somewhere below it might be, | |
8061 | * but because we are the idle thread, we just pick up running again | |
8062 | * when this runqueue becomes "idle". | |
8063 | */ | |
8064 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8065 | /* |
8066 | * During early bootup we pretend to be a normal task: | |
8067 | */ | |
8068 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8069 | |
8070 | scheduler_running = 1; | |
1da177e4 LT |
8071 | } |
8072 | ||
8073 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8074 | void __might_sleep(char *file, int line) | |
8075 | { | |
48f24c4d | 8076 | #ifdef in_atomic |
1da177e4 LT |
8077 | static unsigned long prev_jiffy; /* ratelimiting */ |
8078 | ||
8079 | if ((in_atomic() || irqs_disabled()) && | |
8080 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8081 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8082 | return; | |
8083 | prev_jiffy = jiffies; | |
91368d73 | 8084 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8085 | " context at %s:%d\n", file, line); |
8086 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8087 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8088 | debug_show_held_locks(current); |
3117df04 IM |
8089 | if (irqs_disabled()) |
8090 | print_irqtrace_events(current); | |
1da177e4 LT |
8091 | dump_stack(); |
8092 | } | |
8093 | #endif | |
8094 | } | |
8095 | EXPORT_SYMBOL(__might_sleep); | |
8096 | #endif | |
8097 | ||
8098 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8099 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8100 | { | |
8101 | int on_rq; | |
3e51f33f | 8102 | |
3a5e4dc1 AK |
8103 | update_rq_clock(rq); |
8104 | on_rq = p->se.on_rq; | |
8105 | if (on_rq) | |
8106 | deactivate_task(rq, p, 0); | |
8107 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8108 | if (on_rq) { | |
8109 | activate_task(rq, p, 0); | |
8110 | resched_task(rq->curr); | |
8111 | } | |
8112 | } | |
8113 | ||
1da177e4 LT |
8114 | void normalize_rt_tasks(void) |
8115 | { | |
a0f98a1c | 8116 | struct task_struct *g, *p; |
1da177e4 | 8117 | unsigned long flags; |
70b97a7f | 8118 | struct rq *rq; |
1da177e4 | 8119 | |
4cf5d77a | 8120 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8121 | do_each_thread(g, p) { |
178be793 IM |
8122 | /* |
8123 | * Only normalize user tasks: | |
8124 | */ | |
8125 | if (!p->mm) | |
8126 | continue; | |
8127 | ||
6cfb0d5d | 8128 | p->se.exec_start = 0; |
6cfb0d5d | 8129 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8130 | p->se.wait_start = 0; |
dd41f596 | 8131 | p->se.sleep_start = 0; |
dd41f596 | 8132 | p->se.block_start = 0; |
6cfb0d5d | 8133 | #endif |
dd41f596 IM |
8134 | |
8135 | if (!rt_task(p)) { | |
8136 | /* | |
8137 | * Renice negative nice level userspace | |
8138 | * tasks back to 0: | |
8139 | */ | |
8140 | if (TASK_NICE(p) < 0 && p->mm) | |
8141 | set_user_nice(p, 0); | |
1da177e4 | 8142 | continue; |
dd41f596 | 8143 | } |
1da177e4 | 8144 | |
4cf5d77a | 8145 | spin_lock(&p->pi_lock); |
b29739f9 | 8146 | rq = __task_rq_lock(p); |
1da177e4 | 8147 | |
178be793 | 8148 | normalize_task(rq, p); |
3a5e4dc1 | 8149 | |
b29739f9 | 8150 | __task_rq_unlock(rq); |
4cf5d77a | 8151 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8152 | } while_each_thread(g, p); |
8153 | ||
4cf5d77a | 8154 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8155 | } |
8156 | ||
8157 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8158 | |
8159 | #ifdef CONFIG_IA64 | |
8160 | /* | |
8161 | * These functions are only useful for the IA64 MCA handling. | |
8162 | * | |
8163 | * They can only be called when the whole system has been | |
8164 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8165 | * activity can take place. Using them for anything else would | |
8166 | * be a serious bug, and as a result, they aren't even visible | |
8167 | * under any other configuration. | |
8168 | */ | |
8169 | ||
8170 | /** | |
8171 | * curr_task - return the current task for a given cpu. | |
8172 | * @cpu: the processor in question. | |
8173 | * | |
8174 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8175 | */ | |
36c8b586 | 8176 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8177 | { |
8178 | return cpu_curr(cpu); | |
8179 | } | |
8180 | ||
8181 | /** | |
8182 | * set_curr_task - set the current task for a given cpu. | |
8183 | * @cpu: the processor in question. | |
8184 | * @p: the task pointer to set. | |
8185 | * | |
8186 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8187 | * are serviced on a separate stack. It allows the architecture to switch the |
8188 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8189 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8190 | * and caller must save the original value of the current task (see | |
8191 | * curr_task() above) and restore that value before reenabling interrupts and | |
8192 | * re-starting the system. | |
8193 | * | |
8194 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8195 | */ | |
36c8b586 | 8196 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8197 | { |
8198 | cpu_curr(cpu) = p; | |
8199 | } | |
8200 | ||
8201 | #endif | |
29f59db3 | 8202 | |
bccbe08a PZ |
8203 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8204 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8205 | { |
8206 | int i; | |
8207 | ||
8208 | for_each_possible_cpu(i) { | |
8209 | if (tg->cfs_rq) | |
8210 | kfree(tg->cfs_rq[i]); | |
8211 | if (tg->se) | |
8212 | kfree(tg->se[i]); | |
6f505b16 PZ |
8213 | } |
8214 | ||
8215 | kfree(tg->cfs_rq); | |
8216 | kfree(tg->se); | |
6f505b16 PZ |
8217 | } |
8218 | ||
ec7dc8ac DG |
8219 | static |
8220 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8221 | { |
29f59db3 | 8222 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8223 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8224 | struct rq *rq; |
29f59db3 SV |
8225 | int i; |
8226 | ||
434d53b0 | 8227 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8228 | if (!tg->cfs_rq) |
8229 | goto err; | |
434d53b0 | 8230 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8231 | if (!tg->se) |
8232 | goto err; | |
052f1dc7 PZ |
8233 | |
8234 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8235 | |
8236 | for_each_possible_cpu(i) { | |
9b5b7751 | 8237 | rq = cpu_rq(i); |
29f59db3 | 8238 | |
6f505b16 PZ |
8239 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8240 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8241 | if (!cfs_rq) |
8242 | goto err; | |
8243 | ||
6f505b16 PZ |
8244 | se = kmalloc_node(sizeof(struct sched_entity), |
8245 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8246 | if (!se) |
8247 | goto err; | |
8248 | ||
ec7dc8ac DG |
8249 | parent_se = parent ? parent->se[i] : NULL; |
8250 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8251 | } |
8252 | ||
8253 | return 1; | |
8254 | ||
8255 | err: | |
8256 | return 0; | |
8257 | } | |
8258 | ||
8259 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8260 | { | |
8261 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8262 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8263 | } | |
8264 | ||
8265 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8266 | { | |
8267 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8268 | } | |
6d6bc0ad | 8269 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8270 | static inline void free_fair_sched_group(struct task_group *tg) |
8271 | { | |
8272 | } | |
8273 | ||
ec7dc8ac DG |
8274 | static inline |
8275 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8276 | { |
8277 | return 1; | |
8278 | } | |
8279 | ||
8280 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8281 | { | |
8282 | } | |
8283 | ||
8284 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8285 | { | |
8286 | } | |
6d6bc0ad | 8287 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8288 | |
8289 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8290 | static void free_rt_sched_group(struct task_group *tg) |
8291 | { | |
8292 | int i; | |
8293 | ||
d0b27fa7 PZ |
8294 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8295 | ||
bccbe08a PZ |
8296 | for_each_possible_cpu(i) { |
8297 | if (tg->rt_rq) | |
8298 | kfree(tg->rt_rq[i]); | |
8299 | if (tg->rt_se) | |
8300 | kfree(tg->rt_se[i]); | |
8301 | } | |
8302 | ||
8303 | kfree(tg->rt_rq); | |
8304 | kfree(tg->rt_se); | |
8305 | } | |
8306 | ||
ec7dc8ac DG |
8307 | static |
8308 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8309 | { |
8310 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8311 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8312 | struct rq *rq; |
8313 | int i; | |
8314 | ||
434d53b0 | 8315 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8316 | if (!tg->rt_rq) |
8317 | goto err; | |
434d53b0 | 8318 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8319 | if (!tg->rt_se) |
8320 | goto err; | |
8321 | ||
d0b27fa7 PZ |
8322 | init_rt_bandwidth(&tg->rt_bandwidth, |
8323 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8324 | |
8325 | for_each_possible_cpu(i) { | |
8326 | rq = cpu_rq(i); | |
8327 | ||
6f505b16 PZ |
8328 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8329 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8330 | if (!rt_rq) | |
8331 | goto err; | |
29f59db3 | 8332 | |
6f505b16 PZ |
8333 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8334 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8335 | if (!rt_se) | |
8336 | goto err; | |
29f59db3 | 8337 | |
ec7dc8ac DG |
8338 | parent_se = parent ? parent->rt_se[i] : NULL; |
8339 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8340 | } |
8341 | ||
bccbe08a PZ |
8342 | return 1; |
8343 | ||
8344 | err: | |
8345 | return 0; | |
8346 | } | |
8347 | ||
8348 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8349 | { | |
8350 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8351 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8352 | } | |
8353 | ||
8354 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8355 | { | |
8356 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8357 | } | |
6d6bc0ad | 8358 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8359 | static inline void free_rt_sched_group(struct task_group *tg) |
8360 | { | |
8361 | } | |
8362 | ||
ec7dc8ac DG |
8363 | static inline |
8364 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8365 | { |
8366 | return 1; | |
8367 | } | |
8368 | ||
8369 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8370 | { | |
8371 | } | |
8372 | ||
8373 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8374 | { | |
8375 | } | |
6d6bc0ad | 8376 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8377 | |
d0b27fa7 | 8378 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8379 | static void free_sched_group(struct task_group *tg) |
8380 | { | |
8381 | free_fair_sched_group(tg); | |
8382 | free_rt_sched_group(tg); | |
8383 | kfree(tg); | |
8384 | } | |
8385 | ||
8386 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8387 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8388 | { |
8389 | struct task_group *tg; | |
8390 | unsigned long flags; | |
8391 | int i; | |
8392 | ||
8393 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8394 | if (!tg) | |
8395 | return ERR_PTR(-ENOMEM); | |
8396 | ||
ec7dc8ac | 8397 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8398 | goto err; |
8399 | ||
ec7dc8ac | 8400 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8401 | goto err; |
8402 | ||
8ed36996 | 8403 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8404 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8405 | register_fair_sched_group(tg, i); |
8406 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8407 | } |
6f505b16 | 8408 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8409 | |
8410 | WARN_ON(!parent); /* root should already exist */ | |
8411 | ||
8412 | tg->parent = parent; | |
8413 | list_add_rcu(&tg->siblings, &parent->children); | |
8414 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8415 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8416 | |
9b5b7751 | 8417 | return tg; |
29f59db3 SV |
8418 | |
8419 | err: | |
6f505b16 | 8420 | free_sched_group(tg); |
29f59db3 SV |
8421 | return ERR_PTR(-ENOMEM); |
8422 | } | |
8423 | ||
9b5b7751 | 8424 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8425 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8426 | { |
29f59db3 | 8427 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8428 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8429 | } |
8430 | ||
9b5b7751 | 8431 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8432 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8433 | { |
8ed36996 | 8434 | unsigned long flags; |
9b5b7751 | 8435 | int i; |
29f59db3 | 8436 | |
8ed36996 | 8437 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8438 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8439 | unregister_fair_sched_group(tg, i); |
8440 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8441 | } |
6f505b16 | 8442 | list_del_rcu(&tg->list); |
f473aa5e | 8443 | list_del_rcu(&tg->siblings); |
8ed36996 | 8444 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8445 | |
9b5b7751 | 8446 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8447 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8448 | } |
8449 | ||
9b5b7751 | 8450 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8451 | * The caller of this function should have put the task in its new group |
8452 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8453 | * reflect its new group. | |
9b5b7751 SV |
8454 | */ |
8455 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8456 | { |
8457 | int on_rq, running; | |
8458 | unsigned long flags; | |
8459 | struct rq *rq; | |
8460 | ||
8461 | rq = task_rq_lock(tsk, &flags); | |
8462 | ||
29f59db3 SV |
8463 | update_rq_clock(rq); |
8464 | ||
051a1d1a | 8465 | running = task_current(rq, tsk); |
29f59db3 SV |
8466 | on_rq = tsk->se.on_rq; |
8467 | ||
0e1f3483 | 8468 | if (on_rq) |
29f59db3 | 8469 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8470 | if (unlikely(running)) |
8471 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8472 | |
6f505b16 | 8473 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8474 | |
810b3817 PZ |
8475 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8476 | if (tsk->sched_class->moved_group) | |
8477 | tsk->sched_class->moved_group(tsk); | |
8478 | #endif | |
8479 | ||
0e1f3483 HS |
8480 | if (unlikely(running)) |
8481 | tsk->sched_class->set_curr_task(rq); | |
8482 | if (on_rq) | |
7074badb | 8483 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8484 | |
29f59db3 SV |
8485 | task_rq_unlock(rq, &flags); |
8486 | } | |
6d6bc0ad | 8487 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8488 | |
052f1dc7 | 8489 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8490 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8491 | { |
8492 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8493 | int on_rq; |
8494 | ||
29f59db3 | 8495 | on_rq = se->on_rq; |
62fb1851 | 8496 | if (on_rq) |
29f59db3 SV |
8497 | dequeue_entity(cfs_rq, se, 0); |
8498 | ||
8499 | se->load.weight = shares; | |
e05510d0 | 8500 | se->load.inv_weight = 0; |
29f59db3 | 8501 | |
62fb1851 | 8502 | if (on_rq) |
29f59db3 | 8503 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8504 | } |
62fb1851 | 8505 | |
c09595f6 PZ |
8506 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8507 | { | |
8508 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8509 | struct rq *rq = cfs_rq->rq; | |
8510 | unsigned long flags; | |
8511 | ||
8512 | spin_lock_irqsave(&rq->lock, flags); | |
8513 | __set_se_shares(se, shares); | |
8514 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8515 | } |
8516 | ||
8ed36996 PZ |
8517 | static DEFINE_MUTEX(shares_mutex); |
8518 | ||
4cf86d77 | 8519 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8520 | { |
8521 | int i; | |
8ed36996 | 8522 | unsigned long flags; |
c61935fd | 8523 | |
ec7dc8ac DG |
8524 | /* |
8525 | * We can't change the weight of the root cgroup. | |
8526 | */ | |
8527 | if (!tg->se[0]) | |
8528 | return -EINVAL; | |
8529 | ||
18d95a28 PZ |
8530 | if (shares < MIN_SHARES) |
8531 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8532 | else if (shares > MAX_SHARES) |
8533 | shares = MAX_SHARES; | |
62fb1851 | 8534 | |
8ed36996 | 8535 | mutex_lock(&shares_mutex); |
9b5b7751 | 8536 | if (tg->shares == shares) |
5cb350ba | 8537 | goto done; |
29f59db3 | 8538 | |
8ed36996 | 8539 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8540 | for_each_possible_cpu(i) |
8541 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8542 | list_del_rcu(&tg->siblings); |
8ed36996 | 8543 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8544 | |
8545 | /* wait for any ongoing reference to this group to finish */ | |
8546 | synchronize_sched(); | |
8547 | ||
8548 | /* | |
8549 | * Now we are free to modify the group's share on each cpu | |
8550 | * w/o tripping rebalance_share or load_balance_fair. | |
8551 | */ | |
9b5b7751 | 8552 | tg->shares = shares; |
c09595f6 PZ |
8553 | for_each_possible_cpu(i) { |
8554 | /* | |
8555 | * force a rebalance | |
8556 | */ | |
8557 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8558 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8559 | } |
29f59db3 | 8560 | |
6b2d7700 SV |
8561 | /* |
8562 | * Enable load balance activity on this group, by inserting it back on | |
8563 | * each cpu's rq->leaf_cfs_rq_list. | |
8564 | */ | |
8ed36996 | 8565 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8566 | for_each_possible_cpu(i) |
8567 | register_fair_sched_group(tg, i); | |
f473aa5e | 8568 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8569 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8570 | done: |
8ed36996 | 8571 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8572 | return 0; |
29f59db3 SV |
8573 | } |
8574 | ||
5cb350ba DG |
8575 | unsigned long sched_group_shares(struct task_group *tg) |
8576 | { | |
8577 | return tg->shares; | |
8578 | } | |
052f1dc7 | 8579 | #endif |
5cb350ba | 8580 | |
052f1dc7 | 8581 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8582 | /* |
9f0c1e56 | 8583 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8584 | */ |
9f0c1e56 PZ |
8585 | static DEFINE_MUTEX(rt_constraints_mutex); |
8586 | ||
8587 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8588 | { | |
8589 | if (runtime == RUNTIME_INF) | |
8590 | return 1ULL << 16; | |
8591 | ||
6f6d6a1a | 8592 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8593 | } |
8594 | ||
b40b2e8e PZ |
8595 | #ifdef CONFIG_CGROUP_SCHED |
8596 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8597 | { | |
10b612f4 | 8598 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8599 | unsigned long total = 0; |
8600 | ||
8601 | if (!parent) { | |
8602 | if (global_rt_period() < period) | |
8603 | return 0; | |
8604 | ||
8605 | return to_ratio(period, runtime) < | |
8606 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8607 | } | |
8608 | ||
8609 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8610 | return 0; | |
8611 | ||
8612 | rcu_read_lock(); | |
8613 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8614 | if (tgi == tg) | |
8615 | continue; | |
8616 | ||
8617 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8618 | tgi->rt_bandwidth.rt_runtime); | |
8619 | } | |
8620 | rcu_read_unlock(); | |
8621 | ||
10b612f4 | 8622 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8623 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8624 | parent->rt_bandwidth.rt_runtime); | |
8625 | } | |
8626 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8627 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8628 | { |
8629 | struct task_group *tgi; | |
8630 | unsigned long total = 0; | |
9f0c1e56 | 8631 | unsigned long global_ratio = |
d0b27fa7 | 8632 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8633 | |
8634 | rcu_read_lock(); | |
9f0c1e56 PZ |
8635 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8636 | if (tgi == tg) | |
8637 | continue; | |
6f505b16 | 8638 | |
d0b27fa7 PZ |
8639 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8640 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8641 | } |
8642 | rcu_read_unlock(); | |
6f505b16 | 8643 | |
9f0c1e56 | 8644 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8645 | } |
b40b2e8e | 8646 | #endif |
6f505b16 | 8647 | |
521f1a24 DG |
8648 | /* Must be called with tasklist_lock held */ |
8649 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8650 | { | |
8651 | struct task_struct *g, *p; | |
8652 | do_each_thread(g, p) { | |
8653 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8654 | return 1; | |
8655 | } while_each_thread(g, p); | |
8656 | return 0; | |
8657 | } | |
8658 | ||
d0b27fa7 PZ |
8659 | static int tg_set_bandwidth(struct task_group *tg, |
8660 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8661 | { |
ac086bc2 | 8662 | int i, err = 0; |
9f0c1e56 | 8663 | |
9f0c1e56 | 8664 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8665 | read_lock(&tasklist_lock); |
ac086bc2 | 8666 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8667 | err = -EBUSY; |
8668 | goto unlock; | |
8669 | } | |
9f0c1e56 PZ |
8670 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8671 | err = -EINVAL; | |
8672 | goto unlock; | |
8673 | } | |
ac086bc2 PZ |
8674 | |
8675 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8676 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8677 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8678 | |
8679 | for_each_possible_cpu(i) { | |
8680 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8681 | ||
8682 | spin_lock(&rt_rq->rt_runtime_lock); | |
8683 | rt_rq->rt_runtime = rt_runtime; | |
8684 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8685 | } | |
8686 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8687 | unlock: |
521f1a24 | 8688 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8689 | mutex_unlock(&rt_constraints_mutex); |
8690 | ||
8691 | return err; | |
6f505b16 PZ |
8692 | } |
8693 | ||
d0b27fa7 PZ |
8694 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8695 | { | |
8696 | u64 rt_runtime, rt_period; | |
8697 | ||
8698 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8699 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8700 | if (rt_runtime_us < 0) | |
8701 | rt_runtime = RUNTIME_INF; | |
8702 | ||
8703 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8704 | } | |
8705 | ||
9f0c1e56 PZ |
8706 | long sched_group_rt_runtime(struct task_group *tg) |
8707 | { | |
8708 | u64 rt_runtime_us; | |
8709 | ||
d0b27fa7 | 8710 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8711 | return -1; |
8712 | ||
d0b27fa7 | 8713 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8714 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8715 | return rt_runtime_us; | |
8716 | } | |
d0b27fa7 PZ |
8717 | |
8718 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8719 | { | |
8720 | u64 rt_runtime, rt_period; | |
8721 | ||
8722 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8723 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8724 | ||
8725 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8726 | } | |
8727 | ||
8728 | long sched_group_rt_period(struct task_group *tg) | |
8729 | { | |
8730 | u64 rt_period_us; | |
8731 | ||
8732 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8733 | do_div(rt_period_us, NSEC_PER_USEC); | |
8734 | return rt_period_us; | |
8735 | } | |
8736 | ||
8737 | static int sched_rt_global_constraints(void) | |
8738 | { | |
10b612f4 PZ |
8739 | struct task_group *tg = &root_task_group; |
8740 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8741 | int ret = 0; |
8742 | ||
10b612f4 PZ |
8743 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8744 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8745 | ||
d0b27fa7 | 8746 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8747 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8748 | ret = -EINVAL; |
8749 | mutex_unlock(&rt_constraints_mutex); | |
8750 | ||
8751 | return ret; | |
8752 | } | |
6d6bc0ad | 8753 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8754 | static int sched_rt_global_constraints(void) |
8755 | { | |
ac086bc2 PZ |
8756 | unsigned long flags; |
8757 | int i; | |
8758 | ||
8759 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8760 | for_each_possible_cpu(i) { | |
8761 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8762 | ||
8763 | spin_lock(&rt_rq->rt_runtime_lock); | |
8764 | rt_rq->rt_runtime = global_rt_runtime(); | |
8765 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8766 | } | |
8767 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8768 | ||
d0b27fa7 PZ |
8769 | return 0; |
8770 | } | |
6d6bc0ad | 8771 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8772 | |
8773 | int sched_rt_handler(struct ctl_table *table, int write, | |
8774 | struct file *filp, void __user *buffer, size_t *lenp, | |
8775 | loff_t *ppos) | |
8776 | { | |
8777 | int ret; | |
8778 | int old_period, old_runtime; | |
8779 | static DEFINE_MUTEX(mutex); | |
8780 | ||
8781 | mutex_lock(&mutex); | |
8782 | old_period = sysctl_sched_rt_period; | |
8783 | old_runtime = sysctl_sched_rt_runtime; | |
8784 | ||
8785 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8786 | ||
8787 | if (!ret && write) { | |
8788 | ret = sched_rt_global_constraints(); | |
8789 | if (ret) { | |
8790 | sysctl_sched_rt_period = old_period; | |
8791 | sysctl_sched_rt_runtime = old_runtime; | |
8792 | } else { | |
8793 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8794 | def_rt_bandwidth.rt_period = | |
8795 | ns_to_ktime(global_rt_period()); | |
8796 | } | |
8797 | } | |
8798 | mutex_unlock(&mutex); | |
8799 | ||
8800 | return ret; | |
8801 | } | |
68318b8e | 8802 | |
052f1dc7 | 8803 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8804 | |
8805 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8806 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8807 | { |
2b01dfe3 PM |
8808 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8809 | struct task_group, css); | |
68318b8e SV |
8810 | } |
8811 | ||
8812 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8813 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8814 | { |
ec7dc8ac | 8815 | struct task_group *tg, *parent; |
68318b8e | 8816 | |
2b01dfe3 | 8817 | if (!cgrp->parent) { |
68318b8e | 8818 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8819 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8820 | return &init_task_group.css; |
8821 | } | |
8822 | ||
ec7dc8ac DG |
8823 | parent = cgroup_tg(cgrp->parent); |
8824 | tg = sched_create_group(parent); | |
68318b8e SV |
8825 | if (IS_ERR(tg)) |
8826 | return ERR_PTR(-ENOMEM); | |
8827 | ||
8828 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8829 | tg->css.cgroup = cgrp; |
68318b8e SV |
8830 | |
8831 | return &tg->css; | |
8832 | } | |
8833 | ||
41a2d6cf IM |
8834 | static void |
8835 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8836 | { |
2b01dfe3 | 8837 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8838 | |
8839 | sched_destroy_group(tg); | |
8840 | } | |
8841 | ||
41a2d6cf IM |
8842 | static int |
8843 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8844 | struct task_struct *tsk) | |
68318b8e | 8845 | { |
b68aa230 PZ |
8846 | #ifdef CONFIG_RT_GROUP_SCHED |
8847 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8848 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8849 | return -EINVAL; |
8850 | #else | |
68318b8e SV |
8851 | /* We don't support RT-tasks being in separate groups */ |
8852 | if (tsk->sched_class != &fair_sched_class) | |
8853 | return -EINVAL; | |
b68aa230 | 8854 | #endif |
68318b8e SV |
8855 | |
8856 | return 0; | |
8857 | } | |
8858 | ||
8859 | static void | |
2b01dfe3 | 8860 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8861 | struct cgroup *old_cont, struct task_struct *tsk) |
8862 | { | |
8863 | sched_move_task(tsk); | |
8864 | } | |
8865 | ||
052f1dc7 | 8866 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8867 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8868 | u64 shareval) |
68318b8e | 8869 | { |
2b01dfe3 | 8870 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8871 | } |
8872 | ||
f4c753b7 | 8873 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8874 | { |
2b01dfe3 | 8875 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8876 | |
8877 | return (u64) tg->shares; | |
8878 | } | |
6d6bc0ad | 8879 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8880 | |
052f1dc7 | 8881 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8882 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8883 | s64 val) |
6f505b16 | 8884 | { |
06ecb27c | 8885 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8886 | } |
8887 | ||
06ecb27c | 8888 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8889 | { |
06ecb27c | 8890 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8891 | } |
d0b27fa7 PZ |
8892 | |
8893 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8894 | u64 rt_period_us) | |
8895 | { | |
8896 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8897 | } | |
8898 | ||
8899 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8900 | { | |
8901 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8902 | } | |
6d6bc0ad | 8903 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8904 | |
fe5c7cc2 | 8905 | static struct cftype cpu_files[] = { |
052f1dc7 | 8906 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8907 | { |
8908 | .name = "shares", | |
f4c753b7 PM |
8909 | .read_u64 = cpu_shares_read_u64, |
8910 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8911 | }, |
052f1dc7 PZ |
8912 | #endif |
8913 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8914 | { |
9f0c1e56 | 8915 | .name = "rt_runtime_us", |
06ecb27c PM |
8916 | .read_s64 = cpu_rt_runtime_read, |
8917 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8918 | }, |
d0b27fa7 PZ |
8919 | { |
8920 | .name = "rt_period_us", | |
f4c753b7 PM |
8921 | .read_u64 = cpu_rt_period_read_uint, |
8922 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8923 | }, |
052f1dc7 | 8924 | #endif |
68318b8e SV |
8925 | }; |
8926 | ||
8927 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8928 | { | |
fe5c7cc2 | 8929 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8930 | } |
8931 | ||
8932 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8933 | .name = "cpu", |
8934 | .create = cpu_cgroup_create, | |
8935 | .destroy = cpu_cgroup_destroy, | |
8936 | .can_attach = cpu_cgroup_can_attach, | |
8937 | .attach = cpu_cgroup_attach, | |
8938 | .populate = cpu_cgroup_populate, | |
8939 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8940 | .early_init = 1, |
8941 | }; | |
8942 | ||
052f1dc7 | 8943 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8944 | |
8945 | #ifdef CONFIG_CGROUP_CPUACCT | |
8946 | ||
8947 | /* | |
8948 | * CPU accounting code for task groups. | |
8949 | * | |
8950 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8951 | * (balbir@in.ibm.com). | |
8952 | */ | |
8953 | ||
8954 | /* track cpu usage of a group of tasks */ | |
8955 | struct cpuacct { | |
8956 | struct cgroup_subsys_state css; | |
8957 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8958 | u64 *cpuusage; | |
8959 | }; | |
8960 | ||
8961 | struct cgroup_subsys cpuacct_subsys; | |
8962 | ||
8963 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8964 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8965 | { |
32cd756a | 8966 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8967 | struct cpuacct, css); |
8968 | } | |
8969 | ||
8970 | /* return cpu accounting group to which this task belongs */ | |
8971 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8972 | { | |
8973 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8974 | struct cpuacct, css); | |
8975 | } | |
8976 | ||
8977 | /* create a new cpu accounting group */ | |
8978 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8979 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8980 | { |
8981 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8982 | ||
8983 | if (!ca) | |
8984 | return ERR_PTR(-ENOMEM); | |
8985 | ||
8986 | ca->cpuusage = alloc_percpu(u64); | |
8987 | if (!ca->cpuusage) { | |
8988 | kfree(ca); | |
8989 | return ERR_PTR(-ENOMEM); | |
8990 | } | |
8991 | ||
8992 | return &ca->css; | |
8993 | } | |
8994 | ||
8995 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8996 | static void |
32cd756a | 8997 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8998 | { |
32cd756a | 8999 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9000 | |
9001 | free_percpu(ca->cpuusage); | |
9002 | kfree(ca); | |
9003 | } | |
9004 | ||
9005 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9006 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9007 | { |
32cd756a | 9008 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9009 | u64 totalcpuusage = 0; |
9010 | int i; | |
9011 | ||
9012 | for_each_possible_cpu(i) { | |
9013 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9014 | ||
9015 | /* | |
9016 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9017 | * platforms. | |
9018 | */ | |
9019 | spin_lock_irq(&cpu_rq(i)->lock); | |
9020 | totalcpuusage += *cpuusage; | |
9021 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9022 | } | |
9023 | ||
9024 | return totalcpuusage; | |
9025 | } | |
9026 | ||
0297b803 DG |
9027 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9028 | u64 reset) | |
9029 | { | |
9030 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9031 | int err = 0; | |
9032 | int i; | |
9033 | ||
9034 | if (reset) { | |
9035 | err = -EINVAL; | |
9036 | goto out; | |
9037 | } | |
9038 | ||
9039 | for_each_possible_cpu(i) { | |
9040 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9041 | ||
9042 | spin_lock_irq(&cpu_rq(i)->lock); | |
9043 | *cpuusage = 0; | |
9044 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9045 | } | |
9046 | out: | |
9047 | return err; | |
9048 | } | |
9049 | ||
d842de87 SV |
9050 | static struct cftype files[] = { |
9051 | { | |
9052 | .name = "usage", | |
f4c753b7 PM |
9053 | .read_u64 = cpuusage_read, |
9054 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9055 | }, |
9056 | }; | |
9057 | ||
32cd756a | 9058 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9059 | { |
32cd756a | 9060 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9061 | } |
9062 | ||
9063 | /* | |
9064 | * charge this task's execution time to its accounting group. | |
9065 | * | |
9066 | * called with rq->lock held. | |
9067 | */ | |
9068 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9069 | { | |
9070 | struct cpuacct *ca; | |
9071 | ||
9072 | if (!cpuacct_subsys.active) | |
9073 | return; | |
9074 | ||
9075 | ca = task_ca(tsk); | |
9076 | if (ca) { | |
9077 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9078 | ||
9079 | *cpuusage += cputime; | |
9080 | } | |
9081 | } | |
9082 | ||
9083 | struct cgroup_subsys cpuacct_subsys = { | |
9084 | .name = "cpuacct", | |
9085 | .create = cpuacct_create, | |
9086 | .destroy = cpuacct_destroy, | |
9087 | .populate = cpuacct_populate, | |
9088 | .subsys_id = cpuacct_subsys_id, | |
9089 | }; | |
9090 | #endif /* CONFIG_CGROUP_CPUACCT */ |