Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
434d53b0 | 70 | #include <linux/bootmem.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
d7876a08 | 99 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 100 | */ |
d6322faf | 101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 102 | |
6aa645ea IM |
103 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
104 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
105 | ||
1da177e4 LT |
106 | /* |
107 | * These are the 'tuning knobs' of the scheduler: | |
108 | * | |
a4ec24b4 | 109 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
110 | * Timeslices get refilled after they expire. |
111 | */ | |
1da177e4 | 112 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 113 | |
d0b27fa7 PZ |
114 | /* |
115 | * single value that denotes runtime == period, ie unlimited time. | |
116 | */ | |
117 | #define RUNTIME_INF ((u64)~0ULL) | |
118 | ||
5517d86b ED |
119 | #ifdef CONFIG_SMP |
120 | /* | |
121 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
122 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
123 | */ | |
124 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
125 | { | |
126 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
127 | } | |
128 | ||
129 | /* | |
130 | * Each time a sched group cpu_power is changed, | |
131 | * we must compute its reciprocal value | |
132 | */ | |
133 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
134 | { | |
135 | sg->__cpu_power += val; | |
136 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
137 | } | |
138 | #endif | |
139 | ||
e05606d3 IM |
140 | static inline int rt_policy(int policy) |
141 | { | |
3f33a7ce | 142 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
143 | return 1; |
144 | return 0; | |
145 | } | |
146 | ||
147 | static inline int task_has_rt_policy(struct task_struct *p) | |
148 | { | |
149 | return rt_policy(p->policy); | |
150 | } | |
151 | ||
1da177e4 | 152 | /* |
6aa645ea | 153 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 154 | */ |
6aa645ea IM |
155 | struct rt_prio_array { |
156 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
157 | struct list_head queue[MAX_RT_PRIO]; | |
158 | }; | |
159 | ||
d0b27fa7 | 160 | struct rt_bandwidth { |
ea736ed5 IM |
161 | /* nests inside the rq lock: */ |
162 | spinlock_t rt_runtime_lock; | |
163 | ktime_t rt_period; | |
164 | u64 rt_runtime; | |
165 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
166 | }; |
167 | ||
168 | static struct rt_bandwidth def_rt_bandwidth; | |
169 | ||
170 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
171 | ||
172 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
173 | { | |
174 | struct rt_bandwidth *rt_b = | |
175 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
176 | ktime_t now; | |
177 | int overrun; | |
178 | int idle = 0; | |
179 | ||
180 | for (;;) { | |
181 | now = hrtimer_cb_get_time(timer); | |
182 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
183 | ||
184 | if (!overrun) | |
185 | break; | |
186 | ||
187 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
188 | } | |
189 | ||
190 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
191 | } | |
192 | ||
193 | static | |
194 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
195 | { | |
196 | rt_b->rt_period = ns_to_ktime(period); | |
197 | rt_b->rt_runtime = runtime; | |
198 | ||
ac086bc2 PZ |
199 | spin_lock_init(&rt_b->rt_runtime_lock); |
200 | ||
d0b27fa7 PZ |
201 | hrtimer_init(&rt_b->rt_period_timer, |
202 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
203 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
ccc7dadf | 204 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; |
d0b27fa7 PZ |
205 | } |
206 | ||
c8bfff6d KH |
207 | static inline int rt_bandwidth_enabled(void) |
208 | { | |
209 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
210 | } |
211 | ||
212 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
213 | { | |
214 | ktime_t now; | |
215 | ||
0b148fa0 | 216 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
217 | return; |
218 | ||
219 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
220 | return; | |
221 | ||
222 | spin_lock(&rt_b->rt_runtime_lock); | |
223 | for (;;) { | |
224 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
225 | break; | |
226 | ||
227 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
228 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
229 | hrtimer_start(&rt_b->rt_period_timer, | |
230 | rt_b->rt_period_timer.expires, | |
231 | HRTIMER_MODE_ABS); | |
232 | } | |
233 | spin_unlock(&rt_b->rt_runtime_lock); | |
234 | } | |
235 | ||
236 | #ifdef CONFIG_RT_GROUP_SCHED | |
237 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
238 | { | |
239 | hrtimer_cancel(&rt_b->rt_period_timer); | |
240 | } | |
241 | #endif | |
242 | ||
712555ee HC |
243 | /* |
244 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
245 | * detach_destroy_domains and partition_sched_domains. | |
246 | */ | |
247 | static DEFINE_MUTEX(sched_domains_mutex); | |
248 | ||
052f1dc7 | 249 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 250 | |
68318b8e SV |
251 | #include <linux/cgroup.h> |
252 | ||
29f59db3 SV |
253 | struct cfs_rq; |
254 | ||
6f505b16 PZ |
255 | static LIST_HEAD(task_groups); |
256 | ||
29f59db3 | 257 | /* task group related information */ |
4cf86d77 | 258 | struct task_group { |
052f1dc7 | 259 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
260 | struct cgroup_subsys_state css; |
261 | #endif | |
052f1dc7 PZ |
262 | |
263 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
264 | /* schedulable entities of this group on each cpu */ |
265 | struct sched_entity **se; | |
266 | /* runqueue "owned" by this group on each cpu */ | |
267 | struct cfs_rq **cfs_rq; | |
268 | unsigned long shares; | |
052f1dc7 PZ |
269 | #endif |
270 | ||
271 | #ifdef CONFIG_RT_GROUP_SCHED | |
272 | struct sched_rt_entity **rt_se; | |
273 | struct rt_rq **rt_rq; | |
274 | ||
d0b27fa7 | 275 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 276 | #endif |
6b2d7700 | 277 | |
ae8393e5 | 278 | struct rcu_head rcu; |
6f505b16 | 279 | struct list_head list; |
f473aa5e PZ |
280 | |
281 | struct task_group *parent; | |
282 | struct list_head siblings; | |
283 | struct list_head children; | |
29f59db3 SV |
284 | }; |
285 | ||
354d60c2 | 286 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
287 | |
288 | /* | |
289 | * Root task group. | |
290 | * Every UID task group (including init_task_group aka UID-0) will | |
291 | * be a child to this group. | |
292 | */ | |
293 | struct task_group root_task_group; | |
294 | ||
052f1dc7 | 295 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
296 | /* Default task group's sched entity on each cpu */ |
297 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
298 | /* Default task group's cfs_rq on each cpu */ | |
299 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 300 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
301 | |
302 | #ifdef CONFIG_RT_GROUP_SCHED | |
303 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
304 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 305 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 306 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 307 | #define root_task_group init_task_group |
9a7e0b18 | 308 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 309 | |
8ed36996 | 310 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
311 | * a task group's cpu shares. |
312 | */ | |
8ed36996 | 313 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 314 | |
052f1dc7 | 315 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
316 | #ifdef CONFIG_USER_SCHED |
317 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 318 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 319 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 320 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 321 | |
cb4ad1ff | 322 | /* |
2e084786 LJ |
323 | * A weight of 0 or 1 can cause arithmetics problems. |
324 | * A weight of a cfs_rq is the sum of weights of which entities | |
325 | * are queued on this cfs_rq, so a weight of a entity should not be | |
326 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
327 | * (The default weight is 1024 - so there's no practical |
328 | * limitation from this.) | |
329 | */ | |
18d95a28 | 330 | #define MIN_SHARES 2 |
2e084786 | 331 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 332 | |
052f1dc7 PZ |
333 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
334 | #endif | |
335 | ||
29f59db3 | 336 | /* Default task group. |
3a252015 | 337 | * Every task in system belong to this group at bootup. |
29f59db3 | 338 | */ |
434d53b0 | 339 | struct task_group init_task_group; |
29f59db3 SV |
340 | |
341 | /* return group to which a task belongs */ | |
4cf86d77 | 342 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 343 | { |
4cf86d77 | 344 | struct task_group *tg; |
9b5b7751 | 345 | |
052f1dc7 | 346 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 347 | tg = p->user->tg; |
052f1dc7 | 348 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
349 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
350 | struct task_group, css); | |
24e377a8 | 351 | #else |
41a2d6cf | 352 | tg = &init_task_group; |
24e377a8 | 353 | #endif |
9b5b7751 | 354 | return tg; |
29f59db3 SV |
355 | } |
356 | ||
357 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 358 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 359 | { |
052f1dc7 | 360 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
361 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
362 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 363 | #endif |
6f505b16 | 364 | |
052f1dc7 | 365 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
366 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
367 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 368 | #endif |
29f59db3 SV |
369 | } |
370 | ||
371 | #else | |
372 | ||
6f505b16 | 373 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
374 | static inline struct task_group *task_group(struct task_struct *p) |
375 | { | |
376 | return NULL; | |
377 | } | |
29f59db3 | 378 | |
052f1dc7 | 379 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 380 | |
6aa645ea IM |
381 | /* CFS-related fields in a runqueue */ |
382 | struct cfs_rq { | |
383 | struct load_weight load; | |
384 | unsigned long nr_running; | |
385 | ||
6aa645ea | 386 | u64 exec_clock; |
e9acbff6 | 387 | u64 min_vruntime; |
103638d9 | 388 | u64 pair_start; |
6aa645ea IM |
389 | |
390 | struct rb_root tasks_timeline; | |
391 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
392 | |
393 | struct list_head tasks; | |
394 | struct list_head *balance_iterator; | |
395 | ||
396 | /* | |
397 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
398 | * It is set to NULL otherwise (i.e when none are currently running). |
399 | */ | |
aa2ac252 | 400 | struct sched_entity *curr, *next; |
ddc97297 PZ |
401 | |
402 | unsigned long nr_spread_over; | |
403 | ||
62160e3f | 404 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
405 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
406 | ||
41a2d6cf IM |
407 | /* |
408 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
409 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
410 | * (like users, containers etc.) | |
411 | * | |
412 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
413 | * list is used during load balance. | |
414 | */ | |
41a2d6cf IM |
415 | struct list_head leaf_cfs_rq_list; |
416 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
417 | |
418 | #ifdef CONFIG_SMP | |
c09595f6 | 419 | /* |
c8cba857 | 420 | * the part of load.weight contributed by tasks |
c09595f6 | 421 | */ |
c8cba857 | 422 | unsigned long task_weight; |
c09595f6 | 423 | |
c8cba857 PZ |
424 | /* |
425 | * h_load = weight * f(tg) | |
426 | * | |
427 | * Where f(tg) is the recursive weight fraction assigned to | |
428 | * this group. | |
429 | */ | |
430 | unsigned long h_load; | |
c09595f6 | 431 | |
c8cba857 PZ |
432 | /* |
433 | * this cpu's part of tg->shares | |
434 | */ | |
435 | unsigned long shares; | |
f1d239f7 PZ |
436 | |
437 | /* | |
438 | * load.weight at the time we set shares | |
439 | */ | |
440 | unsigned long rq_weight; | |
c09595f6 | 441 | #endif |
6aa645ea IM |
442 | #endif |
443 | }; | |
1da177e4 | 444 | |
6aa645ea IM |
445 | /* Real-Time classes' related field in a runqueue: */ |
446 | struct rt_rq { | |
447 | struct rt_prio_array active; | |
63489e45 | 448 | unsigned long rt_nr_running; |
052f1dc7 | 449 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
450 | int highest_prio; /* highest queued rt task prio */ |
451 | #endif | |
fa85ae24 | 452 | #ifdef CONFIG_SMP |
73fe6aae | 453 | unsigned long rt_nr_migratory; |
a22d7fc1 | 454 | int overloaded; |
fa85ae24 | 455 | #endif |
6f505b16 | 456 | int rt_throttled; |
fa85ae24 | 457 | u64 rt_time; |
ac086bc2 | 458 | u64 rt_runtime; |
ea736ed5 | 459 | /* Nests inside the rq lock: */ |
ac086bc2 | 460 | spinlock_t rt_runtime_lock; |
6f505b16 | 461 | |
052f1dc7 | 462 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
463 | unsigned long rt_nr_boosted; |
464 | ||
6f505b16 PZ |
465 | struct rq *rq; |
466 | struct list_head leaf_rt_rq_list; | |
467 | struct task_group *tg; | |
468 | struct sched_rt_entity *rt_se; | |
469 | #endif | |
6aa645ea IM |
470 | }; |
471 | ||
57d885fe GH |
472 | #ifdef CONFIG_SMP |
473 | ||
474 | /* | |
475 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
476 | * variables. Each exclusive cpuset essentially defines an island domain by |
477 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
478 | * exclusive cpuset is created, we also create and attach a new root-domain |
479 | * object. | |
480 | * | |
57d885fe GH |
481 | */ |
482 | struct root_domain { | |
483 | atomic_t refcount; | |
484 | cpumask_t span; | |
485 | cpumask_t online; | |
637f5085 | 486 | |
0eab9146 | 487 | /* |
637f5085 GH |
488 | * The "RT overload" flag: it gets set if a CPU has more than |
489 | * one runnable RT task. | |
490 | */ | |
491 | cpumask_t rto_mask; | |
0eab9146 | 492 | atomic_t rto_count; |
6e0534f2 GH |
493 | #ifdef CONFIG_SMP |
494 | struct cpupri cpupri; | |
495 | #endif | |
57d885fe GH |
496 | }; |
497 | ||
dc938520 GH |
498 | /* |
499 | * By default the system creates a single root-domain with all cpus as | |
500 | * members (mimicking the global state we have today). | |
501 | */ | |
57d885fe GH |
502 | static struct root_domain def_root_domain; |
503 | ||
504 | #endif | |
505 | ||
1da177e4 LT |
506 | /* |
507 | * This is the main, per-CPU runqueue data structure. | |
508 | * | |
509 | * Locking rule: those places that want to lock multiple runqueues | |
510 | * (such as the load balancing or the thread migration code), lock | |
511 | * acquire operations must be ordered by ascending &runqueue. | |
512 | */ | |
70b97a7f | 513 | struct rq { |
d8016491 IM |
514 | /* runqueue lock: */ |
515 | spinlock_t lock; | |
1da177e4 LT |
516 | |
517 | /* | |
518 | * nr_running and cpu_load should be in the same cacheline because | |
519 | * remote CPUs use both these fields when doing load calculation. | |
520 | */ | |
521 | unsigned long nr_running; | |
6aa645ea IM |
522 | #define CPU_LOAD_IDX_MAX 5 |
523 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 524 | unsigned char idle_at_tick; |
46cb4b7c | 525 | #ifdef CONFIG_NO_HZ |
15934a37 | 526 | unsigned long last_tick_seen; |
46cb4b7c SS |
527 | unsigned char in_nohz_recently; |
528 | #endif | |
d8016491 IM |
529 | /* capture load from *all* tasks on this cpu: */ |
530 | struct load_weight load; | |
6aa645ea IM |
531 | unsigned long nr_load_updates; |
532 | u64 nr_switches; | |
533 | ||
534 | struct cfs_rq cfs; | |
6f505b16 | 535 | struct rt_rq rt; |
6f505b16 | 536 | |
6aa645ea | 537 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
538 | /* list of leaf cfs_rq on this cpu: */ |
539 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
540 | #endif |
541 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 542 | struct list_head leaf_rt_rq_list; |
1da177e4 | 543 | #endif |
1da177e4 LT |
544 | |
545 | /* | |
546 | * This is part of a global counter where only the total sum | |
547 | * over all CPUs matters. A task can increase this counter on | |
548 | * one CPU and if it got migrated afterwards it may decrease | |
549 | * it on another CPU. Always updated under the runqueue lock: | |
550 | */ | |
551 | unsigned long nr_uninterruptible; | |
552 | ||
36c8b586 | 553 | struct task_struct *curr, *idle; |
c9819f45 | 554 | unsigned long next_balance; |
1da177e4 | 555 | struct mm_struct *prev_mm; |
6aa645ea | 556 | |
3e51f33f | 557 | u64 clock; |
6aa645ea | 558 | |
1da177e4 LT |
559 | atomic_t nr_iowait; |
560 | ||
561 | #ifdef CONFIG_SMP | |
0eab9146 | 562 | struct root_domain *rd; |
1da177e4 LT |
563 | struct sched_domain *sd; |
564 | ||
565 | /* For active balancing */ | |
566 | int active_balance; | |
567 | int push_cpu; | |
d8016491 IM |
568 | /* cpu of this runqueue: */ |
569 | int cpu; | |
1f11eb6a | 570 | int online; |
1da177e4 | 571 | |
a8a51d5e | 572 | unsigned long avg_load_per_task; |
1da177e4 | 573 | |
36c8b586 | 574 | struct task_struct *migration_thread; |
1da177e4 LT |
575 | struct list_head migration_queue; |
576 | #endif | |
577 | ||
8f4d37ec | 578 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
579 | #ifdef CONFIG_SMP |
580 | int hrtick_csd_pending; | |
581 | struct call_single_data hrtick_csd; | |
582 | #endif | |
8f4d37ec PZ |
583 | struct hrtimer hrtick_timer; |
584 | #endif | |
585 | ||
1da177e4 LT |
586 | #ifdef CONFIG_SCHEDSTATS |
587 | /* latency stats */ | |
588 | struct sched_info rq_sched_info; | |
589 | ||
590 | /* sys_sched_yield() stats */ | |
480b9434 KC |
591 | unsigned int yld_exp_empty; |
592 | unsigned int yld_act_empty; | |
593 | unsigned int yld_both_empty; | |
594 | unsigned int yld_count; | |
1da177e4 LT |
595 | |
596 | /* schedule() stats */ | |
480b9434 KC |
597 | unsigned int sched_switch; |
598 | unsigned int sched_count; | |
599 | unsigned int sched_goidle; | |
1da177e4 LT |
600 | |
601 | /* try_to_wake_up() stats */ | |
480b9434 KC |
602 | unsigned int ttwu_count; |
603 | unsigned int ttwu_local; | |
b8efb561 IM |
604 | |
605 | /* BKL stats */ | |
480b9434 | 606 | unsigned int bkl_count; |
1da177e4 LT |
607 | #endif |
608 | }; | |
609 | ||
f34e3b61 | 610 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 611 | |
15afe09b | 612 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 613 | { |
15afe09b | 614 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
615 | } |
616 | ||
0a2966b4 CL |
617 | static inline int cpu_of(struct rq *rq) |
618 | { | |
619 | #ifdef CONFIG_SMP | |
620 | return rq->cpu; | |
621 | #else | |
622 | return 0; | |
623 | #endif | |
624 | } | |
625 | ||
674311d5 NP |
626 | /* |
627 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 628 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
629 | * |
630 | * The domain tree of any CPU may only be accessed from within | |
631 | * preempt-disabled sections. | |
632 | */ | |
48f24c4d IM |
633 | #define for_each_domain(cpu, __sd) \ |
634 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
635 | |
636 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
637 | #define this_rq() (&__get_cpu_var(runqueues)) | |
638 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
639 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
640 | ||
3e51f33f PZ |
641 | static inline void update_rq_clock(struct rq *rq) |
642 | { | |
643 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
644 | } | |
645 | ||
bf5c91ba IM |
646 | /* |
647 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
648 | */ | |
649 | #ifdef CONFIG_SCHED_DEBUG | |
650 | # define const_debug __read_mostly | |
651 | #else | |
652 | # define const_debug static const | |
653 | #endif | |
654 | ||
017730c1 IM |
655 | /** |
656 | * runqueue_is_locked | |
657 | * | |
658 | * Returns true if the current cpu runqueue is locked. | |
659 | * This interface allows printk to be called with the runqueue lock | |
660 | * held and know whether or not it is OK to wake up the klogd. | |
661 | */ | |
662 | int runqueue_is_locked(void) | |
663 | { | |
664 | int cpu = get_cpu(); | |
665 | struct rq *rq = cpu_rq(cpu); | |
666 | int ret; | |
667 | ||
668 | ret = spin_is_locked(&rq->lock); | |
669 | put_cpu(); | |
670 | return ret; | |
671 | } | |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Debugging: various feature bits | |
675 | */ | |
f00b45c1 PZ |
676 | |
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | __SCHED_FEAT_##name , | |
679 | ||
bf5c91ba | 680 | enum { |
f00b45c1 | 681 | #include "sched_features.h" |
bf5c91ba IM |
682 | }; |
683 | ||
f00b45c1 PZ |
684 | #undef SCHED_FEAT |
685 | ||
686 | #define SCHED_FEAT(name, enabled) \ | |
687 | (1UL << __SCHED_FEAT_##name) * enabled | | |
688 | ||
bf5c91ba | 689 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
690 | #include "sched_features.h" |
691 | 0; | |
692 | ||
693 | #undef SCHED_FEAT | |
694 | ||
695 | #ifdef CONFIG_SCHED_DEBUG | |
696 | #define SCHED_FEAT(name, enabled) \ | |
697 | #name , | |
698 | ||
983ed7a6 | 699 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
700 | #include "sched_features.h" |
701 | NULL | |
702 | }; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
983ed7a6 | 706 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
707 | { |
708 | filp->private_data = inode->i_private; | |
709 | return 0; | |
710 | } | |
711 | ||
712 | static ssize_t | |
713 | sched_feat_read(struct file *filp, char __user *ubuf, | |
714 | size_t cnt, loff_t *ppos) | |
715 | { | |
716 | char *buf; | |
717 | int r = 0; | |
718 | int len = 0; | |
719 | int i; | |
720 | ||
721 | for (i = 0; sched_feat_names[i]; i++) { | |
722 | len += strlen(sched_feat_names[i]); | |
723 | len += 4; | |
724 | } | |
725 | ||
726 | buf = kmalloc(len + 2, GFP_KERNEL); | |
727 | if (!buf) | |
728 | return -ENOMEM; | |
729 | ||
730 | for (i = 0; sched_feat_names[i]; i++) { | |
731 | if (sysctl_sched_features & (1UL << i)) | |
732 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
733 | else | |
c24b7c52 | 734 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
735 | } |
736 | ||
737 | r += sprintf(buf + r, "\n"); | |
738 | WARN_ON(r >= len + 2); | |
739 | ||
740 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
741 | ||
742 | kfree(buf); | |
743 | ||
744 | return r; | |
745 | } | |
746 | ||
747 | static ssize_t | |
748 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
749 | size_t cnt, loff_t *ppos) | |
750 | { | |
751 | char buf[64]; | |
752 | char *cmp = buf; | |
753 | int neg = 0; | |
754 | int i; | |
755 | ||
756 | if (cnt > 63) | |
757 | cnt = 63; | |
758 | ||
759 | if (copy_from_user(&buf, ubuf, cnt)) | |
760 | return -EFAULT; | |
761 | ||
762 | buf[cnt] = 0; | |
763 | ||
c24b7c52 | 764 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
765 | neg = 1; |
766 | cmp += 3; | |
767 | } | |
768 | ||
769 | for (i = 0; sched_feat_names[i]; i++) { | |
770 | int len = strlen(sched_feat_names[i]); | |
771 | ||
772 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
773 | if (neg) | |
774 | sysctl_sched_features &= ~(1UL << i); | |
775 | else | |
776 | sysctl_sched_features |= (1UL << i); | |
777 | break; | |
778 | } | |
779 | } | |
780 | ||
781 | if (!sched_feat_names[i]) | |
782 | return -EINVAL; | |
783 | ||
784 | filp->f_pos += cnt; | |
785 | ||
786 | return cnt; | |
787 | } | |
788 | ||
789 | static struct file_operations sched_feat_fops = { | |
790 | .open = sched_feat_open, | |
791 | .read = sched_feat_read, | |
792 | .write = sched_feat_write, | |
793 | }; | |
794 | ||
795 | static __init int sched_init_debug(void) | |
796 | { | |
f00b45c1 PZ |
797 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
798 | &sched_feat_fops); | |
799 | ||
800 | return 0; | |
801 | } | |
802 | late_initcall(sched_init_debug); | |
803 | ||
804 | #endif | |
805 | ||
806 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 807 | |
b82d9fdd PZ |
808 | /* |
809 | * Number of tasks to iterate in a single balance run. | |
810 | * Limited because this is done with IRQs disabled. | |
811 | */ | |
812 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
813 | ||
2398f2c6 PZ |
814 | /* |
815 | * ratelimit for updating the group shares. | |
55cd5340 | 816 | * default: 0.25ms |
2398f2c6 | 817 | */ |
55cd5340 | 818 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 819 | |
fa85ae24 | 820 | /* |
9f0c1e56 | 821 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
822 | * default: 1s |
823 | */ | |
9f0c1e56 | 824 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 825 | |
6892b75e IM |
826 | static __read_mostly int scheduler_running; |
827 | ||
9f0c1e56 PZ |
828 | /* |
829 | * part of the period that we allow rt tasks to run in us. | |
830 | * default: 0.95s | |
831 | */ | |
832 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 833 | |
d0b27fa7 PZ |
834 | static inline u64 global_rt_period(void) |
835 | { | |
836 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
837 | } | |
838 | ||
839 | static inline u64 global_rt_runtime(void) | |
840 | { | |
e26873bb | 841 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
842 | return RUNTIME_INF; |
843 | ||
844 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
845 | } | |
fa85ae24 | 846 | |
1da177e4 | 847 | #ifndef prepare_arch_switch |
4866cde0 NP |
848 | # define prepare_arch_switch(next) do { } while (0) |
849 | #endif | |
850 | #ifndef finish_arch_switch | |
851 | # define finish_arch_switch(prev) do { } while (0) | |
852 | #endif | |
853 | ||
051a1d1a DA |
854 | static inline int task_current(struct rq *rq, struct task_struct *p) |
855 | { | |
856 | return rq->curr == p; | |
857 | } | |
858 | ||
4866cde0 | 859 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 860 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 861 | { |
051a1d1a | 862 | return task_current(rq, p); |
4866cde0 NP |
863 | } |
864 | ||
70b97a7f | 865 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
866 | { |
867 | } | |
868 | ||
70b97a7f | 869 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 870 | { |
da04c035 IM |
871 | #ifdef CONFIG_DEBUG_SPINLOCK |
872 | /* this is a valid case when another task releases the spinlock */ | |
873 | rq->lock.owner = current; | |
874 | #endif | |
8a25d5de IM |
875 | /* |
876 | * If we are tracking spinlock dependencies then we have to | |
877 | * fix up the runqueue lock - which gets 'carried over' from | |
878 | * prev into current: | |
879 | */ | |
880 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
881 | ||
4866cde0 NP |
882 | spin_unlock_irq(&rq->lock); |
883 | } | |
884 | ||
885 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 886 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
887 | { |
888 | #ifdef CONFIG_SMP | |
889 | return p->oncpu; | |
890 | #else | |
051a1d1a | 891 | return task_current(rq, p); |
4866cde0 NP |
892 | #endif |
893 | } | |
894 | ||
70b97a7f | 895 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
896 | { |
897 | #ifdef CONFIG_SMP | |
898 | /* | |
899 | * We can optimise this out completely for !SMP, because the | |
900 | * SMP rebalancing from interrupt is the only thing that cares | |
901 | * here. | |
902 | */ | |
903 | next->oncpu = 1; | |
904 | #endif | |
905 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
906 | spin_unlock_irq(&rq->lock); | |
907 | #else | |
908 | spin_unlock(&rq->lock); | |
909 | #endif | |
910 | } | |
911 | ||
70b97a7f | 912 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
913 | { |
914 | #ifdef CONFIG_SMP | |
915 | /* | |
916 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
917 | * We must ensure this doesn't happen until the switch is completely | |
918 | * finished. | |
919 | */ | |
920 | smp_wmb(); | |
921 | prev->oncpu = 0; | |
922 | #endif | |
923 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
924 | local_irq_enable(); | |
1da177e4 | 925 | #endif |
4866cde0 NP |
926 | } |
927 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 928 | |
b29739f9 IM |
929 | /* |
930 | * __task_rq_lock - lock the runqueue a given task resides on. | |
931 | * Must be called interrupts disabled. | |
932 | */ | |
70b97a7f | 933 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
934 | __acquires(rq->lock) |
935 | { | |
3a5c359a AK |
936 | for (;;) { |
937 | struct rq *rq = task_rq(p); | |
938 | spin_lock(&rq->lock); | |
939 | if (likely(rq == task_rq(p))) | |
940 | return rq; | |
b29739f9 | 941 | spin_unlock(&rq->lock); |
b29739f9 | 942 | } |
b29739f9 IM |
943 | } |
944 | ||
1da177e4 LT |
945 | /* |
946 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 947 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
948 | * explicitly disabling preemption. |
949 | */ | |
70b97a7f | 950 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
951 | __acquires(rq->lock) |
952 | { | |
70b97a7f | 953 | struct rq *rq; |
1da177e4 | 954 | |
3a5c359a AK |
955 | for (;;) { |
956 | local_irq_save(*flags); | |
957 | rq = task_rq(p); | |
958 | spin_lock(&rq->lock); | |
959 | if (likely(rq == task_rq(p))) | |
960 | return rq; | |
1da177e4 | 961 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 962 | } |
1da177e4 LT |
963 | } |
964 | ||
a9957449 | 965 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
966 | __releases(rq->lock) |
967 | { | |
968 | spin_unlock(&rq->lock); | |
969 | } | |
970 | ||
70b97a7f | 971 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
972 | __releases(rq->lock) |
973 | { | |
974 | spin_unlock_irqrestore(&rq->lock, *flags); | |
975 | } | |
976 | ||
1da177e4 | 977 | /* |
cc2a73b5 | 978 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 979 | */ |
a9957449 | 980 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
981 | __acquires(rq->lock) |
982 | { | |
70b97a7f | 983 | struct rq *rq; |
1da177e4 LT |
984 | |
985 | local_irq_disable(); | |
986 | rq = this_rq(); | |
987 | spin_lock(&rq->lock); | |
988 | ||
989 | return rq; | |
990 | } | |
991 | ||
8f4d37ec PZ |
992 | #ifdef CONFIG_SCHED_HRTICK |
993 | /* | |
994 | * Use HR-timers to deliver accurate preemption points. | |
995 | * | |
996 | * Its all a bit involved since we cannot program an hrt while holding the | |
997 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
998 | * reschedule event. | |
999 | * | |
1000 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1001 | * rq->lock. | |
1002 | */ | |
8f4d37ec PZ |
1003 | |
1004 | /* | |
1005 | * Use hrtick when: | |
1006 | * - enabled by features | |
1007 | * - hrtimer is actually high res | |
1008 | */ | |
1009 | static inline int hrtick_enabled(struct rq *rq) | |
1010 | { | |
1011 | if (!sched_feat(HRTICK)) | |
1012 | return 0; | |
ba42059f | 1013 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1014 | return 0; |
8f4d37ec PZ |
1015 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1016 | } | |
1017 | ||
8f4d37ec PZ |
1018 | static void hrtick_clear(struct rq *rq) |
1019 | { | |
1020 | if (hrtimer_active(&rq->hrtick_timer)) | |
1021 | hrtimer_cancel(&rq->hrtick_timer); | |
1022 | } | |
1023 | ||
8f4d37ec PZ |
1024 | /* |
1025 | * High-resolution timer tick. | |
1026 | * Runs from hardirq context with interrupts disabled. | |
1027 | */ | |
1028 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1029 | { | |
1030 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1031 | ||
1032 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1033 | ||
1034 | spin_lock(&rq->lock); | |
3e51f33f | 1035 | update_rq_clock(rq); |
8f4d37ec PZ |
1036 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1037 | spin_unlock(&rq->lock); | |
1038 | ||
1039 | return HRTIMER_NORESTART; | |
1040 | } | |
1041 | ||
95e904c7 | 1042 | #ifdef CONFIG_SMP |
31656519 PZ |
1043 | /* |
1044 | * called from hardirq (IPI) context | |
1045 | */ | |
1046 | static void __hrtick_start(void *arg) | |
b328ca18 | 1047 | { |
31656519 | 1048 | struct rq *rq = arg; |
b328ca18 | 1049 | |
31656519 PZ |
1050 | spin_lock(&rq->lock); |
1051 | hrtimer_restart(&rq->hrtick_timer); | |
1052 | rq->hrtick_csd_pending = 0; | |
1053 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1054 | } |
1055 | ||
31656519 PZ |
1056 | /* |
1057 | * Called to set the hrtick timer state. | |
1058 | * | |
1059 | * called with rq->lock held and irqs disabled | |
1060 | */ | |
1061 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1062 | { |
31656519 PZ |
1063 | struct hrtimer *timer = &rq->hrtick_timer; |
1064 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1065 | |
31656519 PZ |
1066 | timer->expires = time; |
1067 | ||
1068 | if (rq == this_rq()) { | |
1069 | hrtimer_restart(timer); | |
1070 | } else if (!rq->hrtick_csd_pending) { | |
1071 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1072 | rq->hrtick_csd_pending = 1; | |
1073 | } | |
b328ca18 PZ |
1074 | } |
1075 | ||
1076 | static int | |
1077 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1078 | { | |
1079 | int cpu = (int)(long)hcpu; | |
1080 | ||
1081 | switch (action) { | |
1082 | case CPU_UP_CANCELED: | |
1083 | case CPU_UP_CANCELED_FROZEN: | |
1084 | case CPU_DOWN_PREPARE: | |
1085 | case CPU_DOWN_PREPARE_FROZEN: | |
1086 | case CPU_DEAD: | |
1087 | case CPU_DEAD_FROZEN: | |
31656519 | 1088 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1089 | return NOTIFY_OK; |
1090 | } | |
1091 | ||
1092 | return NOTIFY_DONE; | |
1093 | } | |
1094 | ||
fa748203 | 1095 | static __init void init_hrtick(void) |
b328ca18 PZ |
1096 | { |
1097 | hotcpu_notifier(hotplug_hrtick, 0); | |
1098 | } | |
31656519 PZ |
1099 | #else |
1100 | /* | |
1101 | * Called to set the hrtick timer state. | |
1102 | * | |
1103 | * called with rq->lock held and irqs disabled | |
1104 | */ | |
1105 | static void hrtick_start(struct rq *rq, u64 delay) | |
1106 | { | |
1107 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1108 | } | |
b328ca18 | 1109 | |
006c75f1 | 1110 | static inline void init_hrtick(void) |
8f4d37ec | 1111 | { |
8f4d37ec | 1112 | } |
31656519 | 1113 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1114 | |
31656519 | 1115 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1116 | { |
31656519 PZ |
1117 | #ifdef CONFIG_SMP |
1118 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1119 | |
31656519 PZ |
1120 | rq->hrtick_csd.flags = 0; |
1121 | rq->hrtick_csd.func = __hrtick_start; | |
1122 | rq->hrtick_csd.info = rq; | |
1123 | #endif | |
8f4d37ec | 1124 | |
31656519 PZ |
1125 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1126 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1127 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1128 | } |
006c75f1 | 1129 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1130 | static inline void hrtick_clear(struct rq *rq) |
1131 | { | |
1132 | } | |
1133 | ||
8f4d37ec PZ |
1134 | static inline void init_rq_hrtick(struct rq *rq) |
1135 | { | |
1136 | } | |
1137 | ||
b328ca18 PZ |
1138 | static inline void init_hrtick(void) |
1139 | { | |
1140 | } | |
006c75f1 | 1141 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1142 | |
c24d20db IM |
1143 | /* |
1144 | * resched_task - mark a task 'to be rescheduled now'. | |
1145 | * | |
1146 | * On UP this means the setting of the need_resched flag, on SMP it | |
1147 | * might also involve a cross-CPU call to trigger the scheduler on | |
1148 | * the target CPU. | |
1149 | */ | |
1150 | #ifdef CONFIG_SMP | |
1151 | ||
1152 | #ifndef tsk_is_polling | |
1153 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1154 | #endif | |
1155 | ||
31656519 | 1156 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1157 | { |
1158 | int cpu; | |
1159 | ||
1160 | assert_spin_locked(&task_rq(p)->lock); | |
1161 | ||
31656519 | 1162 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1163 | return; |
1164 | ||
31656519 | 1165 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1166 | |
1167 | cpu = task_cpu(p); | |
1168 | if (cpu == smp_processor_id()) | |
1169 | return; | |
1170 | ||
1171 | /* NEED_RESCHED must be visible before we test polling */ | |
1172 | smp_mb(); | |
1173 | if (!tsk_is_polling(p)) | |
1174 | smp_send_reschedule(cpu); | |
1175 | } | |
1176 | ||
1177 | static void resched_cpu(int cpu) | |
1178 | { | |
1179 | struct rq *rq = cpu_rq(cpu); | |
1180 | unsigned long flags; | |
1181 | ||
1182 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1183 | return; | |
1184 | resched_task(cpu_curr(cpu)); | |
1185 | spin_unlock_irqrestore(&rq->lock, flags); | |
1186 | } | |
06d8308c TG |
1187 | |
1188 | #ifdef CONFIG_NO_HZ | |
1189 | /* | |
1190 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1191 | * idle CPU then this timer might expire before the next timer event | |
1192 | * which is scheduled to wake up that CPU. In case of a completely | |
1193 | * idle system the next event might even be infinite time into the | |
1194 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1195 | * leaves the inner idle loop so the newly added timer is taken into | |
1196 | * account when the CPU goes back to idle and evaluates the timer | |
1197 | * wheel for the next timer event. | |
1198 | */ | |
1199 | void wake_up_idle_cpu(int cpu) | |
1200 | { | |
1201 | struct rq *rq = cpu_rq(cpu); | |
1202 | ||
1203 | if (cpu == smp_processor_id()) | |
1204 | return; | |
1205 | ||
1206 | /* | |
1207 | * This is safe, as this function is called with the timer | |
1208 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1209 | * to idle and has not yet set rq->curr to idle then it will | |
1210 | * be serialized on the timer wheel base lock and take the new | |
1211 | * timer into account automatically. | |
1212 | */ | |
1213 | if (rq->curr != rq->idle) | |
1214 | return; | |
1215 | ||
1216 | /* | |
1217 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1218 | * lockless. The worst case is that the other CPU runs the | |
1219 | * idle task through an additional NOOP schedule() | |
1220 | */ | |
1221 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1222 | ||
1223 | /* NEED_RESCHED must be visible before we test polling */ | |
1224 | smp_mb(); | |
1225 | if (!tsk_is_polling(rq->idle)) | |
1226 | smp_send_reschedule(cpu); | |
1227 | } | |
6d6bc0ad | 1228 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1229 | |
6d6bc0ad | 1230 | #else /* !CONFIG_SMP */ |
31656519 | 1231 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1232 | { |
1233 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1234 | set_tsk_need_resched(p); |
c24d20db | 1235 | } |
6d6bc0ad | 1236 | #endif /* CONFIG_SMP */ |
c24d20db | 1237 | |
45bf76df IM |
1238 | #if BITS_PER_LONG == 32 |
1239 | # define WMULT_CONST (~0UL) | |
1240 | #else | |
1241 | # define WMULT_CONST (1UL << 32) | |
1242 | #endif | |
1243 | ||
1244 | #define WMULT_SHIFT 32 | |
1245 | ||
194081eb IM |
1246 | /* |
1247 | * Shift right and round: | |
1248 | */ | |
cf2ab469 | 1249 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1250 | |
a7be37ac PZ |
1251 | /* |
1252 | * delta *= weight / lw | |
1253 | */ | |
cb1c4fc9 | 1254 | static unsigned long |
45bf76df IM |
1255 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1256 | struct load_weight *lw) | |
1257 | { | |
1258 | u64 tmp; | |
1259 | ||
7a232e03 LJ |
1260 | if (!lw->inv_weight) { |
1261 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1262 | lw->inv_weight = 1; | |
1263 | else | |
1264 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1265 | / (lw->weight+1); | |
1266 | } | |
45bf76df IM |
1267 | |
1268 | tmp = (u64)delta_exec * weight; | |
1269 | /* | |
1270 | * Check whether we'd overflow the 64-bit multiplication: | |
1271 | */ | |
194081eb | 1272 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1273 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1274 | WMULT_SHIFT/2); |
1275 | else | |
cf2ab469 | 1276 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1277 | |
ecf691da | 1278 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1279 | } |
1280 | ||
1091985b | 1281 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1282 | { |
1283 | lw->weight += inc; | |
e89996ae | 1284 | lw->inv_weight = 0; |
45bf76df IM |
1285 | } |
1286 | ||
1091985b | 1287 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1288 | { |
1289 | lw->weight -= dec; | |
e89996ae | 1290 | lw->inv_weight = 0; |
45bf76df IM |
1291 | } |
1292 | ||
2dd73a4f PW |
1293 | /* |
1294 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1295 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1296 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1297 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1298 | * scaled version of the new time slice allocation that they receive on time |
1299 | * slice expiry etc. | |
1300 | */ | |
1301 | ||
dd41f596 IM |
1302 | #define WEIGHT_IDLEPRIO 2 |
1303 | #define WMULT_IDLEPRIO (1 << 31) | |
1304 | ||
1305 | /* | |
1306 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1307 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1308 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1309 | * that remained on nice 0. | |
1310 | * | |
1311 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1312 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1313 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1314 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1315 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1316 | */ |
1317 | static const int prio_to_weight[40] = { | |
254753dc IM |
1318 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1319 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1320 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1321 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1322 | /* 0 */ 1024, 820, 655, 526, 423, | |
1323 | /* 5 */ 335, 272, 215, 172, 137, | |
1324 | /* 10 */ 110, 87, 70, 56, 45, | |
1325 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1326 | }; |
1327 | ||
5714d2de IM |
1328 | /* |
1329 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1330 | * | |
1331 | * In cases where the weight does not change often, we can use the | |
1332 | * precalculated inverse to speed up arithmetics by turning divisions | |
1333 | * into multiplications: | |
1334 | */ | |
dd41f596 | 1335 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1336 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1337 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1338 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1339 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1340 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1341 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1342 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1343 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1344 | }; |
2dd73a4f | 1345 | |
dd41f596 IM |
1346 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1347 | ||
1348 | /* | |
1349 | * runqueue iterator, to support SMP load-balancing between different | |
1350 | * scheduling classes, without having to expose their internal data | |
1351 | * structures to the load-balancing proper: | |
1352 | */ | |
1353 | struct rq_iterator { | |
1354 | void *arg; | |
1355 | struct task_struct *(*start)(void *); | |
1356 | struct task_struct *(*next)(void *); | |
1357 | }; | |
1358 | ||
e1d1484f PW |
1359 | #ifdef CONFIG_SMP |
1360 | static unsigned long | |
1361 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1362 | unsigned long max_load_move, struct sched_domain *sd, | |
1363 | enum cpu_idle_type idle, int *all_pinned, | |
1364 | int *this_best_prio, struct rq_iterator *iterator); | |
1365 | ||
1366 | static int | |
1367 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1368 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1369 | struct rq_iterator *iterator); | |
e1d1484f | 1370 | #endif |
dd41f596 | 1371 | |
d842de87 SV |
1372 | #ifdef CONFIG_CGROUP_CPUACCT |
1373 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1374 | #else | |
1375 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1376 | #endif | |
1377 | ||
18d95a28 PZ |
1378 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1379 | { | |
1380 | update_load_add(&rq->load, load); | |
1381 | } | |
1382 | ||
1383 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1384 | { | |
1385 | update_load_sub(&rq->load, load); | |
1386 | } | |
1387 | ||
7940ca36 | 1388 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1389 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1390 | |
1391 | /* | |
1392 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1393 | * leaving it for the final time. | |
1394 | */ | |
eb755805 | 1395 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1396 | { |
1397 | struct task_group *parent, *child; | |
eb755805 | 1398 | int ret; |
c09595f6 PZ |
1399 | |
1400 | rcu_read_lock(); | |
1401 | parent = &root_task_group; | |
1402 | down: | |
eb755805 PZ |
1403 | ret = (*down)(parent, data); |
1404 | if (ret) | |
1405 | goto out_unlock; | |
c09595f6 PZ |
1406 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1407 | parent = child; | |
1408 | goto down; | |
1409 | ||
1410 | up: | |
1411 | continue; | |
1412 | } | |
eb755805 PZ |
1413 | ret = (*up)(parent, data); |
1414 | if (ret) | |
1415 | goto out_unlock; | |
c09595f6 PZ |
1416 | |
1417 | child = parent; | |
1418 | parent = parent->parent; | |
1419 | if (parent) | |
1420 | goto up; | |
eb755805 | 1421 | out_unlock: |
c09595f6 | 1422 | rcu_read_unlock(); |
eb755805 PZ |
1423 | |
1424 | return ret; | |
c09595f6 PZ |
1425 | } |
1426 | ||
eb755805 PZ |
1427 | static int tg_nop(struct task_group *tg, void *data) |
1428 | { | |
1429 | return 0; | |
c09595f6 | 1430 | } |
eb755805 PZ |
1431 | #endif |
1432 | ||
1433 | #ifdef CONFIG_SMP | |
1434 | static unsigned long source_load(int cpu, int type); | |
1435 | static unsigned long target_load(int cpu, int type); | |
1436 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1437 | ||
1438 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1439 | { | |
1440 | struct rq *rq = cpu_rq(cpu); | |
1441 | ||
1442 | if (rq->nr_running) | |
1443 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1444 | ||
1445 | return rq->avg_load_per_task; | |
1446 | } | |
1447 | ||
1448 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1449 | |
c09595f6 PZ |
1450 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1451 | ||
1452 | /* | |
1453 | * Calculate and set the cpu's group shares. | |
1454 | */ | |
1455 | static void | |
b6a86c74 | 1456 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1457 | unsigned long sd_shares, unsigned long sd_rq_weight) |
18d95a28 | 1458 | { |
c09595f6 PZ |
1459 | int boost = 0; |
1460 | unsigned long shares; | |
1461 | unsigned long rq_weight; | |
1462 | ||
c8cba857 | 1463 | if (!tg->se[cpu]) |
c09595f6 PZ |
1464 | return; |
1465 | ||
c8cba857 | 1466 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1467 | |
1468 | /* | |
1469 | * If there are currently no tasks on the cpu pretend there is one of | |
1470 | * average load so that when a new task gets to run here it will not | |
1471 | * get delayed by group starvation. | |
1472 | */ | |
1473 | if (!rq_weight) { | |
1474 | boost = 1; | |
1475 | rq_weight = NICE_0_LOAD; | |
1476 | } | |
1477 | ||
c8cba857 PZ |
1478 | if (unlikely(rq_weight > sd_rq_weight)) |
1479 | rq_weight = sd_rq_weight; | |
1480 | ||
c09595f6 PZ |
1481 | /* |
1482 | * \Sum shares * rq_weight | |
1483 | * shares = ----------------------- | |
1484 | * \Sum rq_weight | |
1485 | * | |
1486 | */ | |
c8cba857 | 1487 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1488 | |
1489 | /* | |
1490 | * record the actual number of shares, not the boosted amount. | |
1491 | */ | |
c8cba857 | 1492 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
f1d239f7 | 1493 | tg->cfs_rq[cpu]->rq_weight = rq_weight; |
c09595f6 PZ |
1494 | |
1495 | if (shares < MIN_SHARES) | |
1496 | shares = MIN_SHARES; | |
1497 | else if (shares > MAX_SHARES) | |
1498 | shares = MAX_SHARES; | |
1499 | ||
c8cba857 | 1500 | __set_se_shares(tg->se[cpu], shares); |
18d95a28 | 1501 | } |
c09595f6 PZ |
1502 | |
1503 | /* | |
c8cba857 PZ |
1504 | * Re-compute the task group their per cpu shares over the given domain. |
1505 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1506 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1507 | */ |
eb755805 | 1508 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1509 | { |
c8cba857 PZ |
1510 | unsigned long rq_weight = 0; |
1511 | unsigned long shares = 0; | |
eb755805 | 1512 | struct sched_domain *sd = data; |
c8cba857 | 1513 | int i; |
c09595f6 | 1514 | |
c8cba857 PZ |
1515 | for_each_cpu_mask(i, sd->span) { |
1516 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1517 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1518 | } |
c09595f6 | 1519 | |
c8cba857 PZ |
1520 | if ((!shares && rq_weight) || shares > tg->shares) |
1521 | shares = tg->shares; | |
1522 | ||
1523 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1524 | shares = tg->shares; | |
c09595f6 | 1525 | |
cd80917e PZ |
1526 | if (!rq_weight) |
1527 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1528 | ||
c09595f6 PZ |
1529 | for_each_cpu_mask(i, sd->span) { |
1530 | struct rq *rq = cpu_rq(i); | |
1531 | unsigned long flags; | |
1532 | ||
1533 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1534 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1535 | spin_unlock_irqrestore(&rq->lock, flags); |
1536 | } | |
eb755805 PZ |
1537 | |
1538 | return 0; | |
c09595f6 PZ |
1539 | } |
1540 | ||
1541 | /* | |
c8cba857 PZ |
1542 | * Compute the cpu's hierarchical load factor for each task group. |
1543 | * This needs to be done in a top-down fashion because the load of a child | |
1544 | * group is a fraction of its parents load. | |
c09595f6 | 1545 | */ |
eb755805 | 1546 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1547 | { |
c8cba857 | 1548 | unsigned long load; |
eb755805 | 1549 | long cpu = (long)data; |
c09595f6 | 1550 | |
c8cba857 PZ |
1551 | if (!tg->parent) { |
1552 | load = cpu_rq(cpu)->load.weight; | |
1553 | } else { | |
1554 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1555 | load *= tg->cfs_rq[cpu]->shares; | |
1556 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1557 | } | |
c09595f6 | 1558 | |
c8cba857 | 1559 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1560 | |
eb755805 | 1561 | return 0; |
c09595f6 PZ |
1562 | } |
1563 | ||
c8cba857 | 1564 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1565 | { |
2398f2c6 PZ |
1566 | u64 now = cpu_clock(raw_smp_processor_id()); |
1567 | s64 elapsed = now - sd->last_update; | |
1568 | ||
1569 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1570 | sd->last_update = now; | |
eb755805 | 1571 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1572 | } |
4d8d595d PZ |
1573 | } |
1574 | ||
3e5459b4 PZ |
1575 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1576 | { | |
1577 | spin_unlock(&rq->lock); | |
1578 | update_shares(sd); | |
1579 | spin_lock(&rq->lock); | |
1580 | } | |
1581 | ||
eb755805 | 1582 | static void update_h_load(long cpu) |
c09595f6 | 1583 | { |
eb755805 | 1584 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1585 | } |
1586 | ||
c09595f6 PZ |
1587 | #else |
1588 | ||
c8cba857 | 1589 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1590 | { |
1591 | } | |
1592 | ||
3e5459b4 PZ |
1593 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1594 | { | |
1595 | } | |
1596 | ||
18d95a28 PZ |
1597 | #endif |
1598 | ||
18d95a28 PZ |
1599 | #endif |
1600 | ||
30432094 | 1601 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1602 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1603 | { | |
30432094 | 1604 | #ifdef CONFIG_SMP |
34e83e85 IM |
1605 | cfs_rq->shares = shares; |
1606 | #endif | |
1607 | } | |
30432094 | 1608 | #endif |
e7693a36 | 1609 | |
dd41f596 | 1610 | #include "sched_stats.h" |
dd41f596 | 1611 | #include "sched_idletask.c" |
5522d5d5 IM |
1612 | #include "sched_fair.c" |
1613 | #include "sched_rt.c" | |
dd41f596 IM |
1614 | #ifdef CONFIG_SCHED_DEBUG |
1615 | # include "sched_debug.c" | |
1616 | #endif | |
1617 | ||
1618 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1619 | #define for_each_class(class) \ |
1620 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1621 | |
c09595f6 | 1622 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1623 | { |
1624 | rq->nr_running++; | |
9c217245 IM |
1625 | } |
1626 | ||
c09595f6 | 1627 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1628 | { |
1629 | rq->nr_running--; | |
9c217245 IM |
1630 | } |
1631 | ||
45bf76df IM |
1632 | static void set_load_weight(struct task_struct *p) |
1633 | { | |
1634 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1635 | p->se.load.weight = prio_to_weight[0] * 2; |
1636 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1637 | return; | |
1638 | } | |
45bf76df | 1639 | |
dd41f596 IM |
1640 | /* |
1641 | * SCHED_IDLE tasks get minimal weight: | |
1642 | */ | |
1643 | if (p->policy == SCHED_IDLE) { | |
1644 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1645 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1646 | return; | |
1647 | } | |
71f8bd46 | 1648 | |
dd41f596 IM |
1649 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1650 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1651 | } |
1652 | ||
2087a1ad GH |
1653 | static void update_avg(u64 *avg, u64 sample) |
1654 | { | |
1655 | s64 diff = sample - *avg; | |
1656 | *avg += diff >> 3; | |
1657 | } | |
1658 | ||
8159f87e | 1659 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1660 | { |
dd41f596 | 1661 | sched_info_queued(p); |
fd390f6a | 1662 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1663 | p->se.on_rq = 1; |
71f8bd46 IM |
1664 | } |
1665 | ||
69be72c1 | 1666 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1667 | { |
2087a1ad GH |
1668 | if (sleep && p->se.last_wakeup) { |
1669 | update_avg(&p->se.avg_overlap, | |
1670 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1671 | p->se.last_wakeup = 0; | |
1672 | } | |
1673 | ||
46ac22ba | 1674 | sched_info_dequeued(p); |
f02231e5 | 1675 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1676 | p->se.on_rq = 0; |
71f8bd46 IM |
1677 | } |
1678 | ||
14531189 | 1679 | /* |
dd41f596 | 1680 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1681 | */ |
14531189 IM |
1682 | static inline int __normal_prio(struct task_struct *p) |
1683 | { | |
dd41f596 | 1684 | return p->static_prio; |
14531189 IM |
1685 | } |
1686 | ||
b29739f9 IM |
1687 | /* |
1688 | * Calculate the expected normal priority: i.e. priority | |
1689 | * without taking RT-inheritance into account. Might be | |
1690 | * boosted by interactivity modifiers. Changes upon fork, | |
1691 | * setprio syscalls, and whenever the interactivity | |
1692 | * estimator recalculates. | |
1693 | */ | |
36c8b586 | 1694 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1695 | { |
1696 | int prio; | |
1697 | ||
e05606d3 | 1698 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1699 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1700 | else | |
1701 | prio = __normal_prio(p); | |
1702 | return prio; | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * Calculate the current priority, i.e. the priority | |
1707 | * taken into account by the scheduler. This value might | |
1708 | * be boosted by RT tasks, or might be boosted by | |
1709 | * interactivity modifiers. Will be RT if the task got | |
1710 | * RT-boosted. If not then it returns p->normal_prio. | |
1711 | */ | |
36c8b586 | 1712 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1713 | { |
1714 | p->normal_prio = normal_prio(p); | |
1715 | /* | |
1716 | * If we are RT tasks or we were boosted to RT priority, | |
1717 | * keep the priority unchanged. Otherwise, update priority | |
1718 | * to the normal priority: | |
1719 | */ | |
1720 | if (!rt_prio(p->prio)) | |
1721 | return p->normal_prio; | |
1722 | return p->prio; | |
1723 | } | |
1724 | ||
1da177e4 | 1725 | /* |
dd41f596 | 1726 | * activate_task - move a task to the runqueue. |
1da177e4 | 1727 | */ |
dd41f596 | 1728 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1729 | { |
d9514f6c | 1730 | if (task_contributes_to_load(p)) |
dd41f596 | 1731 | rq->nr_uninterruptible--; |
1da177e4 | 1732 | |
8159f87e | 1733 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1734 | inc_nr_running(rq); |
1da177e4 LT |
1735 | } |
1736 | ||
1da177e4 LT |
1737 | /* |
1738 | * deactivate_task - remove a task from the runqueue. | |
1739 | */ | |
2e1cb74a | 1740 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1741 | { |
d9514f6c | 1742 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1743 | rq->nr_uninterruptible++; |
1744 | ||
69be72c1 | 1745 | dequeue_task(rq, p, sleep); |
c09595f6 | 1746 | dec_nr_running(rq); |
1da177e4 LT |
1747 | } |
1748 | ||
1da177e4 LT |
1749 | /** |
1750 | * task_curr - is this task currently executing on a CPU? | |
1751 | * @p: the task in question. | |
1752 | */ | |
36c8b586 | 1753 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1754 | { |
1755 | return cpu_curr(task_cpu(p)) == p; | |
1756 | } | |
1757 | ||
dd41f596 IM |
1758 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1759 | { | |
6f505b16 | 1760 | set_task_rq(p, cpu); |
dd41f596 | 1761 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1762 | /* |
1763 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1764 | * successfuly executed on another CPU. We must ensure that updates of | |
1765 | * per-task data have been completed by this moment. | |
1766 | */ | |
1767 | smp_wmb(); | |
dd41f596 | 1768 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1769 | #endif |
2dd73a4f PW |
1770 | } |
1771 | ||
cb469845 SR |
1772 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1773 | const struct sched_class *prev_class, | |
1774 | int oldprio, int running) | |
1775 | { | |
1776 | if (prev_class != p->sched_class) { | |
1777 | if (prev_class->switched_from) | |
1778 | prev_class->switched_from(rq, p, running); | |
1779 | p->sched_class->switched_to(rq, p, running); | |
1780 | } else | |
1781 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1782 | } | |
1783 | ||
1da177e4 | 1784 | #ifdef CONFIG_SMP |
c65cc870 | 1785 | |
e958b360 TG |
1786 | /* Used instead of source_load when we know the type == 0 */ |
1787 | static unsigned long weighted_cpuload(const int cpu) | |
1788 | { | |
1789 | return cpu_rq(cpu)->load.weight; | |
1790 | } | |
1791 | ||
cc367732 IM |
1792 | /* |
1793 | * Is this task likely cache-hot: | |
1794 | */ | |
e7693a36 | 1795 | static int |
cc367732 IM |
1796 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1797 | { | |
1798 | s64 delta; | |
1799 | ||
f540a608 IM |
1800 | /* |
1801 | * Buddy candidates are cache hot: | |
1802 | */ | |
d25ce4cd | 1803 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1804 | return 1; |
1805 | ||
cc367732 IM |
1806 | if (p->sched_class != &fair_sched_class) |
1807 | return 0; | |
1808 | ||
6bc1665b IM |
1809 | if (sysctl_sched_migration_cost == -1) |
1810 | return 1; | |
1811 | if (sysctl_sched_migration_cost == 0) | |
1812 | return 0; | |
1813 | ||
cc367732 IM |
1814 | delta = now - p->se.exec_start; |
1815 | ||
1816 | return delta < (s64)sysctl_sched_migration_cost; | |
1817 | } | |
1818 | ||
1819 | ||
dd41f596 | 1820 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1821 | { |
dd41f596 IM |
1822 | int old_cpu = task_cpu(p); |
1823 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1824 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1825 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1826 | u64 clock_offset; |
dd41f596 IM |
1827 | |
1828 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1829 | |
1830 | #ifdef CONFIG_SCHEDSTATS | |
1831 | if (p->se.wait_start) | |
1832 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1833 | if (p->se.sleep_start) |
1834 | p->se.sleep_start -= clock_offset; | |
1835 | if (p->se.block_start) | |
1836 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1837 | if (old_cpu != new_cpu) { |
1838 | schedstat_inc(p, se.nr_migrations); | |
1839 | if (task_hot(p, old_rq->clock, NULL)) | |
1840 | schedstat_inc(p, se.nr_forced2_migrations); | |
1841 | } | |
6cfb0d5d | 1842 | #endif |
2830cf8c SV |
1843 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1844 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1845 | |
1846 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1847 | } |
1848 | ||
70b97a7f | 1849 | struct migration_req { |
1da177e4 | 1850 | struct list_head list; |
1da177e4 | 1851 | |
36c8b586 | 1852 | struct task_struct *task; |
1da177e4 LT |
1853 | int dest_cpu; |
1854 | ||
1da177e4 | 1855 | struct completion done; |
70b97a7f | 1856 | }; |
1da177e4 LT |
1857 | |
1858 | /* | |
1859 | * The task's runqueue lock must be held. | |
1860 | * Returns true if you have to wait for migration thread. | |
1861 | */ | |
36c8b586 | 1862 | static int |
70b97a7f | 1863 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1864 | { |
70b97a7f | 1865 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1866 | |
1867 | /* | |
1868 | * If the task is not on a runqueue (and not running), then | |
1869 | * it is sufficient to simply update the task's cpu field. | |
1870 | */ | |
dd41f596 | 1871 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1872 | set_task_cpu(p, dest_cpu); |
1873 | return 0; | |
1874 | } | |
1875 | ||
1876 | init_completion(&req->done); | |
1da177e4 LT |
1877 | req->task = p; |
1878 | req->dest_cpu = dest_cpu; | |
1879 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1880 | |
1da177e4 LT |
1881 | return 1; |
1882 | } | |
1883 | ||
1884 | /* | |
1885 | * wait_task_inactive - wait for a thread to unschedule. | |
1886 | * | |
85ba2d86 RM |
1887 | * If @match_state is nonzero, it's the @p->state value just checked and |
1888 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1889 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1890 | * we return a positive number (its total switch count). If a second call | |
1891 | * a short while later returns the same number, the caller can be sure that | |
1892 | * @p has remained unscheduled the whole time. | |
1893 | * | |
1da177e4 LT |
1894 | * The caller must ensure that the task *will* unschedule sometime soon, |
1895 | * else this function might spin for a *long* time. This function can't | |
1896 | * be called with interrupts off, or it may introduce deadlock with | |
1897 | * smp_call_function() if an IPI is sent by the same process we are | |
1898 | * waiting to become inactive. | |
1899 | */ | |
85ba2d86 | 1900 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1901 | { |
1902 | unsigned long flags; | |
dd41f596 | 1903 | int running, on_rq; |
85ba2d86 | 1904 | unsigned long ncsw; |
70b97a7f | 1905 | struct rq *rq; |
1da177e4 | 1906 | |
3a5c359a AK |
1907 | for (;;) { |
1908 | /* | |
1909 | * We do the initial early heuristics without holding | |
1910 | * any task-queue locks at all. We'll only try to get | |
1911 | * the runqueue lock when things look like they will | |
1912 | * work out! | |
1913 | */ | |
1914 | rq = task_rq(p); | |
fa490cfd | 1915 | |
3a5c359a AK |
1916 | /* |
1917 | * If the task is actively running on another CPU | |
1918 | * still, just relax and busy-wait without holding | |
1919 | * any locks. | |
1920 | * | |
1921 | * NOTE! Since we don't hold any locks, it's not | |
1922 | * even sure that "rq" stays as the right runqueue! | |
1923 | * But we don't care, since "task_running()" will | |
1924 | * return false if the runqueue has changed and p | |
1925 | * is actually now running somewhere else! | |
1926 | */ | |
85ba2d86 RM |
1927 | while (task_running(rq, p)) { |
1928 | if (match_state && unlikely(p->state != match_state)) | |
1929 | return 0; | |
3a5c359a | 1930 | cpu_relax(); |
85ba2d86 | 1931 | } |
fa490cfd | 1932 | |
3a5c359a AK |
1933 | /* |
1934 | * Ok, time to look more closely! We need the rq | |
1935 | * lock now, to be *sure*. If we're wrong, we'll | |
1936 | * just go back and repeat. | |
1937 | */ | |
1938 | rq = task_rq_lock(p, &flags); | |
1939 | running = task_running(rq, p); | |
1940 | on_rq = p->se.on_rq; | |
85ba2d86 | 1941 | ncsw = 0; |
f31e11d8 | 1942 | if (!match_state || p->state == match_state) |
93dcf55f | 1943 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1944 | task_rq_unlock(rq, &flags); |
fa490cfd | 1945 | |
85ba2d86 RM |
1946 | /* |
1947 | * If it changed from the expected state, bail out now. | |
1948 | */ | |
1949 | if (unlikely(!ncsw)) | |
1950 | break; | |
1951 | ||
3a5c359a AK |
1952 | /* |
1953 | * Was it really running after all now that we | |
1954 | * checked with the proper locks actually held? | |
1955 | * | |
1956 | * Oops. Go back and try again.. | |
1957 | */ | |
1958 | if (unlikely(running)) { | |
1959 | cpu_relax(); | |
1960 | continue; | |
1961 | } | |
fa490cfd | 1962 | |
3a5c359a AK |
1963 | /* |
1964 | * It's not enough that it's not actively running, | |
1965 | * it must be off the runqueue _entirely_, and not | |
1966 | * preempted! | |
1967 | * | |
1968 | * So if it wa still runnable (but just not actively | |
1969 | * running right now), it's preempted, and we should | |
1970 | * yield - it could be a while. | |
1971 | */ | |
1972 | if (unlikely(on_rq)) { | |
1973 | schedule_timeout_uninterruptible(1); | |
1974 | continue; | |
1975 | } | |
fa490cfd | 1976 | |
3a5c359a AK |
1977 | /* |
1978 | * Ahh, all good. It wasn't running, and it wasn't | |
1979 | * runnable, which means that it will never become | |
1980 | * running in the future either. We're all done! | |
1981 | */ | |
1982 | break; | |
1983 | } | |
85ba2d86 RM |
1984 | |
1985 | return ncsw; | |
1da177e4 LT |
1986 | } |
1987 | ||
1988 | /*** | |
1989 | * kick_process - kick a running thread to enter/exit the kernel | |
1990 | * @p: the to-be-kicked thread | |
1991 | * | |
1992 | * Cause a process which is running on another CPU to enter | |
1993 | * kernel-mode, without any delay. (to get signals handled.) | |
1994 | * | |
1995 | * NOTE: this function doesnt have to take the runqueue lock, | |
1996 | * because all it wants to ensure is that the remote task enters | |
1997 | * the kernel. If the IPI races and the task has been migrated | |
1998 | * to another CPU then no harm is done and the purpose has been | |
1999 | * achieved as well. | |
2000 | */ | |
36c8b586 | 2001 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2002 | { |
2003 | int cpu; | |
2004 | ||
2005 | preempt_disable(); | |
2006 | cpu = task_cpu(p); | |
2007 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2008 | smp_send_reschedule(cpu); | |
2009 | preempt_enable(); | |
2010 | } | |
2011 | ||
2012 | /* | |
2dd73a4f PW |
2013 | * Return a low guess at the load of a migration-source cpu weighted |
2014 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2015 | * |
2016 | * We want to under-estimate the load of migration sources, to | |
2017 | * balance conservatively. | |
2018 | */ | |
a9957449 | 2019 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2020 | { |
70b97a7f | 2021 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2022 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2023 | |
93b75217 | 2024 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2025 | return total; |
b910472d | 2026 | |
dd41f596 | 2027 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2028 | } |
2029 | ||
2030 | /* | |
2dd73a4f PW |
2031 | * Return a high guess at the load of a migration-target cpu weighted |
2032 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2033 | */ |
a9957449 | 2034 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2035 | { |
70b97a7f | 2036 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2037 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2038 | |
93b75217 | 2039 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2040 | return total; |
3b0bd9bc | 2041 | |
dd41f596 | 2042 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2043 | } |
2044 | ||
147cbb4b NP |
2045 | /* |
2046 | * find_idlest_group finds and returns the least busy CPU group within the | |
2047 | * domain. | |
2048 | */ | |
2049 | static struct sched_group * | |
2050 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2051 | { | |
2052 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2053 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2054 | int load_idx = sd->forkexec_idx; | |
2055 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2056 | ||
2057 | do { | |
2058 | unsigned long load, avg_load; | |
2059 | int local_group; | |
2060 | int i; | |
2061 | ||
da5a5522 BD |
2062 | /* Skip over this group if it has no CPUs allowed */ |
2063 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2064 | continue; |
da5a5522 | 2065 | |
147cbb4b | 2066 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2067 | |
2068 | /* Tally up the load of all CPUs in the group */ | |
2069 | avg_load = 0; | |
2070 | ||
363ab6f1 | 2071 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2072 | /* Bias balancing toward cpus of our domain */ |
2073 | if (local_group) | |
2074 | load = source_load(i, load_idx); | |
2075 | else | |
2076 | load = target_load(i, load_idx); | |
2077 | ||
2078 | avg_load += load; | |
2079 | } | |
2080 | ||
2081 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2082 | avg_load = sg_div_cpu_power(group, |
2083 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2084 | |
2085 | if (local_group) { | |
2086 | this_load = avg_load; | |
2087 | this = group; | |
2088 | } else if (avg_load < min_load) { | |
2089 | min_load = avg_load; | |
2090 | idlest = group; | |
2091 | } | |
3a5c359a | 2092 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2093 | |
2094 | if (!idlest || 100*this_load < imbalance*min_load) | |
2095 | return NULL; | |
2096 | return idlest; | |
2097 | } | |
2098 | ||
2099 | /* | |
0feaece9 | 2100 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2101 | */ |
95cdf3b7 | 2102 | static int |
7c16ec58 MT |
2103 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2104 | cpumask_t *tmp) | |
147cbb4b NP |
2105 | { |
2106 | unsigned long load, min_load = ULONG_MAX; | |
2107 | int idlest = -1; | |
2108 | int i; | |
2109 | ||
da5a5522 | 2110 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2111 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2112 | |
363ab6f1 | 2113 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2114 | load = weighted_cpuload(i); |
147cbb4b NP |
2115 | |
2116 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2117 | min_load = load; | |
2118 | idlest = i; | |
2119 | } | |
2120 | } | |
2121 | ||
2122 | return idlest; | |
2123 | } | |
2124 | ||
476d139c NP |
2125 | /* |
2126 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2127 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2128 | * SD_BALANCE_EXEC. | |
2129 | * | |
2130 | * Balance, ie. select the least loaded group. | |
2131 | * | |
2132 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2133 | * | |
2134 | * preempt must be disabled. | |
2135 | */ | |
2136 | static int sched_balance_self(int cpu, int flag) | |
2137 | { | |
2138 | struct task_struct *t = current; | |
2139 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2140 | |
c96d145e | 2141 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2142 | /* |
2143 | * If power savings logic is enabled for a domain, stop there. | |
2144 | */ | |
5c45bf27 SS |
2145 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2146 | break; | |
476d139c NP |
2147 | if (tmp->flags & flag) |
2148 | sd = tmp; | |
c96d145e | 2149 | } |
476d139c | 2150 | |
039a1c41 PZ |
2151 | if (sd) |
2152 | update_shares(sd); | |
2153 | ||
476d139c | 2154 | while (sd) { |
7c16ec58 | 2155 | cpumask_t span, tmpmask; |
476d139c | 2156 | struct sched_group *group; |
1a848870 SS |
2157 | int new_cpu, weight; |
2158 | ||
2159 | if (!(sd->flags & flag)) { | |
2160 | sd = sd->child; | |
2161 | continue; | |
2162 | } | |
476d139c NP |
2163 | |
2164 | span = sd->span; | |
2165 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2166 | if (!group) { |
2167 | sd = sd->child; | |
2168 | continue; | |
2169 | } | |
476d139c | 2170 | |
7c16ec58 | 2171 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2172 | if (new_cpu == -1 || new_cpu == cpu) { |
2173 | /* Now try balancing at a lower domain level of cpu */ | |
2174 | sd = sd->child; | |
2175 | continue; | |
2176 | } | |
476d139c | 2177 | |
1a848870 | 2178 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2179 | cpu = new_cpu; |
476d139c NP |
2180 | sd = NULL; |
2181 | weight = cpus_weight(span); | |
2182 | for_each_domain(cpu, tmp) { | |
2183 | if (weight <= cpus_weight(tmp->span)) | |
2184 | break; | |
2185 | if (tmp->flags & flag) | |
2186 | sd = tmp; | |
2187 | } | |
2188 | /* while loop will break here if sd == NULL */ | |
2189 | } | |
2190 | ||
2191 | return cpu; | |
2192 | } | |
2193 | ||
2194 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2195 | |
1da177e4 LT |
2196 | /*** |
2197 | * try_to_wake_up - wake up a thread | |
2198 | * @p: the to-be-woken-up thread | |
2199 | * @state: the mask of task states that can be woken | |
2200 | * @sync: do a synchronous wakeup? | |
2201 | * | |
2202 | * Put it on the run-queue if it's not already there. The "current" | |
2203 | * thread is always on the run-queue (except when the actual | |
2204 | * re-schedule is in progress), and as such you're allowed to do | |
2205 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2206 | * runnable without the overhead of this. | |
2207 | * | |
2208 | * returns failure only if the task is already active. | |
2209 | */ | |
36c8b586 | 2210 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2211 | { |
cc367732 | 2212 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2213 | unsigned long flags; |
2214 | long old_state; | |
70b97a7f | 2215 | struct rq *rq; |
1da177e4 | 2216 | |
b85d0667 IM |
2217 | if (!sched_feat(SYNC_WAKEUPS)) |
2218 | sync = 0; | |
2219 | ||
2398f2c6 PZ |
2220 | #ifdef CONFIG_SMP |
2221 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2222 | struct sched_domain *sd; | |
2223 | ||
2224 | this_cpu = raw_smp_processor_id(); | |
2225 | cpu = task_cpu(p); | |
2226 | ||
2227 | for_each_domain(this_cpu, sd) { | |
2228 | if (cpu_isset(cpu, sd->span)) { | |
2229 | update_shares(sd); | |
2230 | break; | |
2231 | } | |
2232 | } | |
2233 | } | |
2234 | #endif | |
2235 | ||
04e2f174 | 2236 | smp_wmb(); |
1da177e4 LT |
2237 | rq = task_rq_lock(p, &flags); |
2238 | old_state = p->state; | |
2239 | if (!(old_state & state)) | |
2240 | goto out; | |
2241 | ||
dd41f596 | 2242 | if (p->se.on_rq) |
1da177e4 LT |
2243 | goto out_running; |
2244 | ||
2245 | cpu = task_cpu(p); | |
cc367732 | 2246 | orig_cpu = cpu; |
1da177e4 LT |
2247 | this_cpu = smp_processor_id(); |
2248 | ||
2249 | #ifdef CONFIG_SMP | |
2250 | if (unlikely(task_running(rq, p))) | |
2251 | goto out_activate; | |
2252 | ||
5d2f5a61 DA |
2253 | cpu = p->sched_class->select_task_rq(p, sync); |
2254 | if (cpu != orig_cpu) { | |
2255 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2256 | task_rq_unlock(rq, &flags); |
2257 | /* might preempt at this point */ | |
2258 | rq = task_rq_lock(p, &flags); | |
2259 | old_state = p->state; | |
2260 | if (!(old_state & state)) | |
2261 | goto out; | |
dd41f596 | 2262 | if (p->se.on_rq) |
1da177e4 LT |
2263 | goto out_running; |
2264 | ||
2265 | this_cpu = smp_processor_id(); | |
2266 | cpu = task_cpu(p); | |
2267 | } | |
2268 | ||
e7693a36 GH |
2269 | #ifdef CONFIG_SCHEDSTATS |
2270 | schedstat_inc(rq, ttwu_count); | |
2271 | if (cpu == this_cpu) | |
2272 | schedstat_inc(rq, ttwu_local); | |
2273 | else { | |
2274 | struct sched_domain *sd; | |
2275 | for_each_domain(this_cpu, sd) { | |
2276 | if (cpu_isset(cpu, sd->span)) { | |
2277 | schedstat_inc(sd, ttwu_wake_remote); | |
2278 | break; | |
2279 | } | |
2280 | } | |
2281 | } | |
6d6bc0ad | 2282 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2283 | |
1da177e4 LT |
2284 | out_activate: |
2285 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2286 | schedstat_inc(p, se.nr_wakeups); |
2287 | if (sync) | |
2288 | schedstat_inc(p, se.nr_wakeups_sync); | |
2289 | if (orig_cpu != cpu) | |
2290 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2291 | if (cpu == this_cpu) | |
2292 | schedstat_inc(p, se.nr_wakeups_local); | |
2293 | else | |
2294 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2295 | update_rq_clock(rq); |
dd41f596 | 2296 | activate_task(rq, p, 1); |
1da177e4 LT |
2297 | success = 1; |
2298 | ||
2299 | out_running: | |
5b82a1b0 MD |
2300 | trace_mark(kernel_sched_wakeup, |
2301 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2302 | p->pid, p->state, rq, p, rq->curr); | |
15afe09b | 2303 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2304 | |
1da177e4 | 2305 | p->state = TASK_RUNNING; |
9a897c5a SR |
2306 | #ifdef CONFIG_SMP |
2307 | if (p->sched_class->task_wake_up) | |
2308 | p->sched_class->task_wake_up(rq, p); | |
2309 | #endif | |
1da177e4 | 2310 | out: |
2087a1ad GH |
2311 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2312 | ||
1da177e4 LT |
2313 | task_rq_unlock(rq, &flags); |
2314 | ||
2315 | return success; | |
2316 | } | |
2317 | ||
7ad5b3a5 | 2318 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2319 | { |
d9514f6c | 2320 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2321 | } |
1da177e4 LT |
2322 | EXPORT_SYMBOL(wake_up_process); |
2323 | ||
7ad5b3a5 | 2324 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2325 | { |
2326 | return try_to_wake_up(p, state, 0); | |
2327 | } | |
2328 | ||
1da177e4 LT |
2329 | /* |
2330 | * Perform scheduler related setup for a newly forked process p. | |
2331 | * p is forked by current. | |
dd41f596 IM |
2332 | * |
2333 | * __sched_fork() is basic setup used by init_idle() too: | |
2334 | */ | |
2335 | static void __sched_fork(struct task_struct *p) | |
2336 | { | |
dd41f596 IM |
2337 | p->se.exec_start = 0; |
2338 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2339 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2340 | p->se.last_wakeup = 0; |
2341 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2342 | |
2343 | #ifdef CONFIG_SCHEDSTATS | |
2344 | p->se.wait_start = 0; | |
dd41f596 IM |
2345 | p->se.sum_sleep_runtime = 0; |
2346 | p->se.sleep_start = 0; | |
dd41f596 IM |
2347 | p->se.block_start = 0; |
2348 | p->se.sleep_max = 0; | |
2349 | p->se.block_max = 0; | |
2350 | p->se.exec_max = 0; | |
eba1ed4b | 2351 | p->se.slice_max = 0; |
dd41f596 | 2352 | p->se.wait_max = 0; |
6cfb0d5d | 2353 | #endif |
476d139c | 2354 | |
fa717060 | 2355 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2356 | p->se.on_rq = 0; |
4a55bd5e | 2357 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2358 | |
e107be36 AK |
2359 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2360 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2361 | #endif | |
2362 | ||
1da177e4 LT |
2363 | /* |
2364 | * We mark the process as running here, but have not actually | |
2365 | * inserted it onto the runqueue yet. This guarantees that | |
2366 | * nobody will actually run it, and a signal or other external | |
2367 | * event cannot wake it up and insert it on the runqueue either. | |
2368 | */ | |
2369 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2370 | } |
2371 | ||
2372 | /* | |
2373 | * fork()/clone()-time setup: | |
2374 | */ | |
2375 | void sched_fork(struct task_struct *p, int clone_flags) | |
2376 | { | |
2377 | int cpu = get_cpu(); | |
2378 | ||
2379 | __sched_fork(p); | |
2380 | ||
2381 | #ifdef CONFIG_SMP | |
2382 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2383 | #endif | |
02e4bac2 | 2384 | set_task_cpu(p, cpu); |
b29739f9 IM |
2385 | |
2386 | /* | |
2387 | * Make sure we do not leak PI boosting priority to the child: | |
2388 | */ | |
2389 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2390 | if (!rt_prio(p->prio)) |
2391 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2392 | |
52f17b6c | 2393 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2394 | if (likely(sched_info_on())) |
52f17b6c | 2395 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2396 | #endif |
d6077cb8 | 2397 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2398 | p->oncpu = 0; |
2399 | #endif | |
1da177e4 | 2400 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2401 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2402 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2403 | #endif |
476d139c | 2404 | put_cpu(); |
1da177e4 LT |
2405 | } |
2406 | ||
2407 | /* | |
2408 | * wake_up_new_task - wake up a newly created task for the first time. | |
2409 | * | |
2410 | * This function will do some initial scheduler statistics housekeeping | |
2411 | * that must be done for every newly created context, then puts the task | |
2412 | * on the runqueue and wakes it. | |
2413 | */ | |
7ad5b3a5 | 2414 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2415 | { |
2416 | unsigned long flags; | |
dd41f596 | 2417 | struct rq *rq; |
1da177e4 LT |
2418 | |
2419 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2420 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2421 | update_rq_clock(rq); |
1da177e4 LT |
2422 | |
2423 | p->prio = effective_prio(p); | |
2424 | ||
b9dca1e0 | 2425 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2426 | activate_task(rq, p, 0); |
1da177e4 | 2427 | } else { |
1da177e4 | 2428 | /* |
dd41f596 IM |
2429 | * Let the scheduling class do new task startup |
2430 | * management (if any): | |
1da177e4 | 2431 | */ |
ee0827d8 | 2432 | p->sched_class->task_new(rq, p); |
c09595f6 | 2433 | inc_nr_running(rq); |
1da177e4 | 2434 | } |
5b82a1b0 MD |
2435 | trace_mark(kernel_sched_wakeup_new, |
2436 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2437 | p->pid, p->state, rq, p, rq->curr); | |
15afe09b | 2438 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2439 | #ifdef CONFIG_SMP |
2440 | if (p->sched_class->task_wake_up) | |
2441 | p->sched_class->task_wake_up(rq, p); | |
2442 | #endif | |
dd41f596 | 2443 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2444 | } |
2445 | ||
e107be36 AK |
2446 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2447 | ||
2448 | /** | |
421cee29 RD |
2449 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2450 | * @notifier: notifier struct to register | |
e107be36 AK |
2451 | */ |
2452 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2453 | { | |
2454 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2455 | } | |
2456 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2457 | ||
2458 | /** | |
2459 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2460 | * @notifier: notifier struct to unregister |
e107be36 AK |
2461 | * |
2462 | * This is safe to call from within a preemption notifier. | |
2463 | */ | |
2464 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2465 | { | |
2466 | hlist_del(¬ifier->link); | |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2469 | ||
2470 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2471 | { | |
2472 | struct preempt_notifier *notifier; | |
2473 | struct hlist_node *node; | |
2474 | ||
2475 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2476 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2477 | } | |
2478 | ||
2479 | static void | |
2480 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2481 | struct task_struct *next) | |
2482 | { | |
2483 | struct preempt_notifier *notifier; | |
2484 | struct hlist_node *node; | |
2485 | ||
2486 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2487 | notifier->ops->sched_out(notifier, next); | |
2488 | } | |
2489 | ||
6d6bc0ad | 2490 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2491 | |
2492 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2493 | { | |
2494 | } | |
2495 | ||
2496 | static void | |
2497 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2498 | struct task_struct *next) | |
2499 | { | |
2500 | } | |
2501 | ||
6d6bc0ad | 2502 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2503 | |
4866cde0 NP |
2504 | /** |
2505 | * prepare_task_switch - prepare to switch tasks | |
2506 | * @rq: the runqueue preparing to switch | |
421cee29 | 2507 | * @prev: the current task that is being switched out |
4866cde0 NP |
2508 | * @next: the task we are going to switch to. |
2509 | * | |
2510 | * This is called with the rq lock held and interrupts off. It must | |
2511 | * be paired with a subsequent finish_task_switch after the context | |
2512 | * switch. | |
2513 | * | |
2514 | * prepare_task_switch sets up locking and calls architecture specific | |
2515 | * hooks. | |
2516 | */ | |
e107be36 AK |
2517 | static inline void |
2518 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2519 | struct task_struct *next) | |
4866cde0 | 2520 | { |
e107be36 | 2521 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2522 | prepare_lock_switch(rq, next); |
2523 | prepare_arch_switch(next); | |
2524 | } | |
2525 | ||
1da177e4 LT |
2526 | /** |
2527 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2528 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2529 | * @prev: the thread we just switched away from. |
2530 | * | |
4866cde0 NP |
2531 | * finish_task_switch must be called after the context switch, paired |
2532 | * with a prepare_task_switch call before the context switch. | |
2533 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2534 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2535 | * |
2536 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2537 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2538 | * with the lock held can cause deadlocks; see schedule() for |
2539 | * details.) | |
2540 | */ | |
a9957449 | 2541 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2542 | __releases(rq->lock) |
2543 | { | |
1da177e4 | 2544 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2545 | long prev_state; |
1da177e4 LT |
2546 | |
2547 | rq->prev_mm = NULL; | |
2548 | ||
2549 | /* | |
2550 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2551 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2552 | * schedule one last time. The schedule call will never return, and |
2553 | * the scheduled task must drop that reference. | |
c394cc9f | 2554 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2555 | * still held, otherwise prev could be scheduled on another cpu, die |
2556 | * there before we look at prev->state, and then the reference would | |
2557 | * be dropped twice. | |
2558 | * Manfred Spraul <manfred@colorfullife.com> | |
2559 | */ | |
55a101f8 | 2560 | prev_state = prev->state; |
4866cde0 NP |
2561 | finish_arch_switch(prev); |
2562 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2563 | #ifdef CONFIG_SMP |
2564 | if (current->sched_class->post_schedule) | |
2565 | current->sched_class->post_schedule(rq); | |
2566 | #endif | |
e8fa1362 | 2567 | |
e107be36 | 2568 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2569 | if (mm) |
2570 | mmdrop(mm); | |
c394cc9f | 2571 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2572 | /* |
2573 | * Remove function-return probe instances associated with this | |
2574 | * task and put them back on the free list. | |
9761eea8 | 2575 | */ |
c6fd91f0 | 2576 | kprobe_flush_task(prev); |
1da177e4 | 2577 | put_task_struct(prev); |
c6fd91f0 | 2578 | } |
1da177e4 LT |
2579 | } |
2580 | ||
2581 | /** | |
2582 | * schedule_tail - first thing a freshly forked thread must call. | |
2583 | * @prev: the thread we just switched away from. | |
2584 | */ | |
36c8b586 | 2585 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2586 | __releases(rq->lock) |
2587 | { | |
70b97a7f IM |
2588 | struct rq *rq = this_rq(); |
2589 | ||
4866cde0 NP |
2590 | finish_task_switch(rq, prev); |
2591 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2592 | /* In this case, finish_task_switch does not reenable preemption */ | |
2593 | preempt_enable(); | |
2594 | #endif | |
1da177e4 | 2595 | if (current->set_child_tid) |
b488893a | 2596 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2597 | } |
2598 | ||
2599 | /* | |
2600 | * context_switch - switch to the new MM and the new | |
2601 | * thread's register state. | |
2602 | */ | |
dd41f596 | 2603 | static inline void |
70b97a7f | 2604 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2605 | struct task_struct *next) |
1da177e4 | 2606 | { |
dd41f596 | 2607 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2608 | |
e107be36 | 2609 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2610 | trace_mark(kernel_sched_schedule, |
2611 | "prev_pid %d next_pid %d prev_state %ld " | |
2612 | "## rq %p prev %p next %p", | |
2613 | prev->pid, next->pid, prev->state, | |
2614 | rq, prev, next); | |
dd41f596 IM |
2615 | mm = next->mm; |
2616 | oldmm = prev->active_mm; | |
9226d125 ZA |
2617 | /* |
2618 | * For paravirt, this is coupled with an exit in switch_to to | |
2619 | * combine the page table reload and the switch backend into | |
2620 | * one hypercall. | |
2621 | */ | |
2622 | arch_enter_lazy_cpu_mode(); | |
2623 | ||
dd41f596 | 2624 | if (unlikely(!mm)) { |
1da177e4 LT |
2625 | next->active_mm = oldmm; |
2626 | atomic_inc(&oldmm->mm_count); | |
2627 | enter_lazy_tlb(oldmm, next); | |
2628 | } else | |
2629 | switch_mm(oldmm, mm, next); | |
2630 | ||
dd41f596 | 2631 | if (unlikely(!prev->mm)) { |
1da177e4 | 2632 | prev->active_mm = NULL; |
1da177e4 LT |
2633 | rq->prev_mm = oldmm; |
2634 | } | |
3a5f5e48 IM |
2635 | /* |
2636 | * Since the runqueue lock will be released by the next | |
2637 | * task (which is an invalid locking op but in the case | |
2638 | * of the scheduler it's an obvious special-case), so we | |
2639 | * do an early lockdep release here: | |
2640 | */ | |
2641 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2642 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2643 | #endif |
1da177e4 LT |
2644 | |
2645 | /* Here we just switch the register state and the stack. */ | |
2646 | switch_to(prev, next, prev); | |
2647 | ||
dd41f596 IM |
2648 | barrier(); |
2649 | /* | |
2650 | * this_rq must be evaluated again because prev may have moved | |
2651 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2652 | * frame will be invalid. | |
2653 | */ | |
2654 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2655 | } |
2656 | ||
2657 | /* | |
2658 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2659 | * | |
2660 | * externally visible scheduler statistics: current number of runnable | |
2661 | * threads, current number of uninterruptible-sleeping threads, total | |
2662 | * number of context switches performed since bootup. | |
2663 | */ | |
2664 | unsigned long nr_running(void) | |
2665 | { | |
2666 | unsigned long i, sum = 0; | |
2667 | ||
2668 | for_each_online_cpu(i) | |
2669 | sum += cpu_rq(i)->nr_running; | |
2670 | ||
2671 | return sum; | |
2672 | } | |
2673 | ||
2674 | unsigned long nr_uninterruptible(void) | |
2675 | { | |
2676 | unsigned long i, sum = 0; | |
2677 | ||
0a945022 | 2678 | for_each_possible_cpu(i) |
1da177e4 LT |
2679 | sum += cpu_rq(i)->nr_uninterruptible; |
2680 | ||
2681 | /* | |
2682 | * Since we read the counters lockless, it might be slightly | |
2683 | * inaccurate. Do not allow it to go below zero though: | |
2684 | */ | |
2685 | if (unlikely((long)sum < 0)) | |
2686 | sum = 0; | |
2687 | ||
2688 | return sum; | |
2689 | } | |
2690 | ||
2691 | unsigned long long nr_context_switches(void) | |
2692 | { | |
cc94abfc SR |
2693 | int i; |
2694 | unsigned long long sum = 0; | |
1da177e4 | 2695 | |
0a945022 | 2696 | for_each_possible_cpu(i) |
1da177e4 LT |
2697 | sum += cpu_rq(i)->nr_switches; |
2698 | ||
2699 | return sum; | |
2700 | } | |
2701 | ||
2702 | unsigned long nr_iowait(void) | |
2703 | { | |
2704 | unsigned long i, sum = 0; | |
2705 | ||
0a945022 | 2706 | for_each_possible_cpu(i) |
1da177e4 LT |
2707 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2708 | ||
2709 | return sum; | |
2710 | } | |
2711 | ||
db1b1fef JS |
2712 | unsigned long nr_active(void) |
2713 | { | |
2714 | unsigned long i, running = 0, uninterruptible = 0; | |
2715 | ||
2716 | for_each_online_cpu(i) { | |
2717 | running += cpu_rq(i)->nr_running; | |
2718 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2719 | } | |
2720 | ||
2721 | if (unlikely((long)uninterruptible < 0)) | |
2722 | uninterruptible = 0; | |
2723 | ||
2724 | return running + uninterruptible; | |
2725 | } | |
2726 | ||
48f24c4d | 2727 | /* |
dd41f596 IM |
2728 | * Update rq->cpu_load[] statistics. This function is usually called every |
2729 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2730 | */ |
dd41f596 | 2731 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2732 | { |
495eca49 | 2733 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2734 | int i, scale; |
2735 | ||
2736 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2737 | |
2738 | /* Update our load: */ | |
2739 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2740 | unsigned long old_load, new_load; | |
2741 | ||
2742 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2743 | ||
2744 | old_load = this_rq->cpu_load[i]; | |
2745 | new_load = this_load; | |
a25707f3 IM |
2746 | /* |
2747 | * Round up the averaging division if load is increasing. This | |
2748 | * prevents us from getting stuck on 9 if the load is 10, for | |
2749 | * example. | |
2750 | */ | |
2751 | if (new_load > old_load) | |
2752 | new_load += scale-1; | |
dd41f596 IM |
2753 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2754 | } | |
48f24c4d IM |
2755 | } |
2756 | ||
dd41f596 IM |
2757 | #ifdef CONFIG_SMP |
2758 | ||
1da177e4 LT |
2759 | /* |
2760 | * double_rq_lock - safely lock two runqueues | |
2761 | * | |
2762 | * Note this does not disable interrupts like task_rq_lock, | |
2763 | * you need to do so manually before calling. | |
2764 | */ | |
70b97a7f | 2765 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2766 | __acquires(rq1->lock) |
2767 | __acquires(rq2->lock) | |
2768 | { | |
054b9108 | 2769 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2770 | if (rq1 == rq2) { |
2771 | spin_lock(&rq1->lock); | |
2772 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2773 | } else { | |
c96d145e | 2774 | if (rq1 < rq2) { |
1da177e4 | 2775 | spin_lock(&rq1->lock); |
5e710e37 | 2776 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2777 | } else { |
2778 | spin_lock(&rq2->lock); | |
5e710e37 | 2779 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2780 | } |
2781 | } | |
6e82a3be IM |
2782 | update_rq_clock(rq1); |
2783 | update_rq_clock(rq2); | |
1da177e4 LT |
2784 | } |
2785 | ||
2786 | /* | |
2787 | * double_rq_unlock - safely unlock two runqueues | |
2788 | * | |
2789 | * Note this does not restore interrupts like task_rq_unlock, | |
2790 | * you need to do so manually after calling. | |
2791 | */ | |
70b97a7f | 2792 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2793 | __releases(rq1->lock) |
2794 | __releases(rq2->lock) | |
2795 | { | |
2796 | spin_unlock(&rq1->lock); | |
2797 | if (rq1 != rq2) | |
2798 | spin_unlock(&rq2->lock); | |
2799 | else | |
2800 | __release(rq2->lock); | |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2805 | */ | |
e8fa1362 | 2806 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2807 | __releases(this_rq->lock) |
2808 | __acquires(busiest->lock) | |
2809 | __acquires(this_rq->lock) | |
2810 | { | |
e8fa1362 SR |
2811 | int ret = 0; |
2812 | ||
054b9108 KK |
2813 | if (unlikely(!irqs_disabled())) { |
2814 | /* printk() doesn't work good under rq->lock */ | |
2815 | spin_unlock(&this_rq->lock); | |
2816 | BUG_ON(1); | |
2817 | } | |
1da177e4 | 2818 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2819 | if (busiest < this_rq) { |
1da177e4 LT |
2820 | spin_unlock(&this_rq->lock); |
2821 | spin_lock(&busiest->lock); | |
5e710e37 | 2822 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2823 | ret = 1; |
1da177e4 | 2824 | } else |
5e710e37 | 2825 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2826 | } |
e8fa1362 | 2827 | return ret; |
1da177e4 LT |
2828 | } |
2829 | ||
1b12bbc7 PZ |
2830 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2831 | __releases(busiest->lock) | |
2832 | { | |
2833 | spin_unlock(&busiest->lock); | |
2834 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2835 | } | |
2836 | ||
1da177e4 LT |
2837 | /* |
2838 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2839 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2840 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2841 | * the cpu_allowed mask is restored. |
2842 | */ | |
36c8b586 | 2843 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2844 | { |
70b97a7f | 2845 | struct migration_req req; |
1da177e4 | 2846 | unsigned long flags; |
70b97a7f | 2847 | struct rq *rq; |
1da177e4 LT |
2848 | |
2849 | rq = task_rq_lock(p, &flags); | |
2850 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2851 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2852 | goto out; |
2853 | ||
2854 | /* force the process onto the specified CPU */ | |
2855 | if (migrate_task(p, dest_cpu, &req)) { | |
2856 | /* Need to wait for migration thread (might exit: take ref). */ | |
2857 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2858 | |
1da177e4 LT |
2859 | get_task_struct(mt); |
2860 | task_rq_unlock(rq, &flags); | |
2861 | wake_up_process(mt); | |
2862 | put_task_struct(mt); | |
2863 | wait_for_completion(&req.done); | |
36c8b586 | 2864 | |
1da177e4 LT |
2865 | return; |
2866 | } | |
2867 | out: | |
2868 | task_rq_unlock(rq, &flags); | |
2869 | } | |
2870 | ||
2871 | /* | |
476d139c NP |
2872 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2873 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2874 | */ |
2875 | void sched_exec(void) | |
2876 | { | |
1da177e4 | 2877 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2878 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2879 | put_cpu(); |
476d139c NP |
2880 | if (new_cpu != this_cpu) |
2881 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2882 | } |
2883 | ||
2884 | /* | |
2885 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2886 | * Both runqueues must be locked. | |
2887 | */ | |
dd41f596 IM |
2888 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2889 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2890 | { |
2e1cb74a | 2891 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2892 | set_task_cpu(p, this_cpu); |
dd41f596 | 2893 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2894 | /* |
2895 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2896 | * to be always true for them. | |
2897 | */ | |
15afe09b | 2898 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2899 | } |
2900 | ||
2901 | /* | |
2902 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2903 | */ | |
858119e1 | 2904 | static |
70b97a7f | 2905 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2906 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2907 | int *all_pinned) |
1da177e4 LT |
2908 | { |
2909 | /* | |
2910 | * We do not migrate tasks that are: | |
2911 | * 1) running (obviously), or | |
2912 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2913 | * 3) are cache-hot on their current CPU. | |
2914 | */ | |
cc367732 IM |
2915 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2916 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2917 | return 0; |
cc367732 | 2918 | } |
81026794 NP |
2919 | *all_pinned = 0; |
2920 | ||
cc367732 IM |
2921 | if (task_running(rq, p)) { |
2922 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2923 | return 0; |
cc367732 | 2924 | } |
1da177e4 | 2925 | |
da84d961 IM |
2926 | /* |
2927 | * Aggressive migration if: | |
2928 | * 1) task is cache cold, or | |
2929 | * 2) too many balance attempts have failed. | |
2930 | */ | |
2931 | ||
6bc1665b IM |
2932 | if (!task_hot(p, rq->clock, sd) || |
2933 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2934 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2935 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2936 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2937 | schedstat_inc(p, se.nr_forced_migrations); |
2938 | } | |
da84d961 IM |
2939 | #endif |
2940 | return 1; | |
2941 | } | |
2942 | ||
cc367732 IM |
2943 | if (task_hot(p, rq->clock, sd)) { |
2944 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2945 | return 0; |
cc367732 | 2946 | } |
1da177e4 LT |
2947 | return 1; |
2948 | } | |
2949 | ||
e1d1484f PW |
2950 | static unsigned long |
2951 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2952 | unsigned long max_load_move, struct sched_domain *sd, | |
2953 | enum cpu_idle_type idle, int *all_pinned, | |
2954 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2955 | { |
051c6764 | 2956 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2957 | struct task_struct *p; |
2958 | long rem_load_move = max_load_move; | |
1da177e4 | 2959 | |
e1d1484f | 2960 | if (max_load_move == 0) |
1da177e4 LT |
2961 | goto out; |
2962 | ||
81026794 NP |
2963 | pinned = 1; |
2964 | ||
1da177e4 | 2965 | /* |
dd41f596 | 2966 | * Start the load-balancing iterator: |
1da177e4 | 2967 | */ |
dd41f596 IM |
2968 | p = iterator->start(iterator->arg); |
2969 | next: | |
b82d9fdd | 2970 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2971 | goto out; |
051c6764 PZ |
2972 | |
2973 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2974 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2975 | p = iterator->next(iterator->arg); |
2976 | goto next; | |
1da177e4 LT |
2977 | } |
2978 | ||
dd41f596 | 2979 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2980 | pulled++; |
dd41f596 | 2981 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2982 | |
2dd73a4f | 2983 | /* |
b82d9fdd | 2984 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2985 | */ |
e1d1484f | 2986 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2987 | if (p->prio < *this_best_prio) |
2988 | *this_best_prio = p->prio; | |
dd41f596 IM |
2989 | p = iterator->next(iterator->arg); |
2990 | goto next; | |
1da177e4 LT |
2991 | } |
2992 | out: | |
2993 | /* | |
e1d1484f | 2994 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2995 | * so we can safely collect pull_task() stats here rather than |
2996 | * inside pull_task(). | |
2997 | */ | |
2998 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2999 | |
3000 | if (all_pinned) | |
3001 | *all_pinned = pinned; | |
e1d1484f PW |
3002 | |
3003 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3004 | } |
3005 | ||
dd41f596 | 3006 | /* |
43010659 PW |
3007 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3008 | * this_rq, as part of a balancing operation within domain "sd". | |
3009 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3010 | * |
3011 | * Called with both runqueues locked. | |
3012 | */ | |
3013 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3014 | unsigned long max_load_move, |
dd41f596 IM |
3015 | struct sched_domain *sd, enum cpu_idle_type idle, |
3016 | int *all_pinned) | |
3017 | { | |
5522d5d5 | 3018 | const struct sched_class *class = sched_class_highest; |
43010659 | 3019 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3020 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3021 | |
3022 | do { | |
43010659 PW |
3023 | total_load_moved += |
3024 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3025 | max_load_move - total_load_moved, |
a4ac01c3 | 3026 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3027 | class = class->next; |
c4acb2c0 GH |
3028 | |
3029 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3030 | break; | |
3031 | ||
43010659 | 3032 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3033 | |
43010659 PW |
3034 | return total_load_moved > 0; |
3035 | } | |
3036 | ||
e1d1484f PW |
3037 | static int |
3038 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3039 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3040 | struct rq_iterator *iterator) | |
3041 | { | |
3042 | struct task_struct *p = iterator->start(iterator->arg); | |
3043 | int pinned = 0; | |
3044 | ||
3045 | while (p) { | |
3046 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3047 | pull_task(busiest, p, this_rq, this_cpu); | |
3048 | /* | |
3049 | * Right now, this is only the second place pull_task() | |
3050 | * is called, so we can safely collect pull_task() | |
3051 | * stats here rather than inside pull_task(). | |
3052 | */ | |
3053 | schedstat_inc(sd, lb_gained[idle]); | |
3054 | ||
3055 | return 1; | |
3056 | } | |
3057 | p = iterator->next(iterator->arg); | |
3058 | } | |
3059 | ||
3060 | return 0; | |
3061 | } | |
3062 | ||
43010659 PW |
3063 | /* |
3064 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3065 | * part of active balancing operations within "domain". | |
3066 | * Returns 1 if successful and 0 otherwise. | |
3067 | * | |
3068 | * Called with both runqueues locked. | |
3069 | */ | |
3070 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3071 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3072 | { | |
5522d5d5 | 3073 | const struct sched_class *class; |
43010659 PW |
3074 | |
3075 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3076 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3077 | return 1; |
3078 | ||
3079 | return 0; | |
dd41f596 IM |
3080 | } |
3081 | ||
1da177e4 LT |
3082 | /* |
3083 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3084 | * domain. It calculates and returns the amount of weighted load which |
3085 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3086 | */ |
3087 | static struct sched_group * | |
3088 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3089 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3090 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3091 | { |
3092 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3093 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3094 | unsigned long max_pull; |
2dd73a4f PW |
3095 | unsigned long busiest_load_per_task, busiest_nr_running; |
3096 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3097 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3098 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3099 | int power_savings_balance = 1; | |
3100 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3101 | unsigned long min_nr_running = ULONG_MAX; | |
3102 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3103 | #endif | |
1da177e4 LT |
3104 | |
3105 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3106 | busiest_load_per_task = busiest_nr_running = 0; |
3107 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3108 | |
d15bcfdb | 3109 | if (idle == CPU_NOT_IDLE) |
7897986b | 3110 | load_idx = sd->busy_idx; |
d15bcfdb | 3111 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3112 | load_idx = sd->newidle_idx; |
3113 | else | |
3114 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3115 | |
3116 | do { | |
908a7c1b | 3117 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3118 | int local_group; |
3119 | int i; | |
908a7c1b | 3120 | int __group_imb = 0; |
783609c6 | 3121 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3122 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3123 | unsigned long sum_avg_load_per_task; |
3124 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3125 | |
3126 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3127 | ||
783609c6 SS |
3128 | if (local_group) |
3129 | balance_cpu = first_cpu(group->cpumask); | |
3130 | ||
1da177e4 | 3131 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3132 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3133 | sum_avg_load_per_task = avg_load_per_task = 0; |
3134 | ||
908a7c1b KC |
3135 | max_cpu_load = 0; |
3136 | min_cpu_load = ~0UL; | |
1da177e4 | 3137 | |
363ab6f1 | 3138 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3139 | struct rq *rq; |
3140 | ||
3141 | if (!cpu_isset(i, *cpus)) | |
3142 | continue; | |
3143 | ||
3144 | rq = cpu_rq(i); | |
2dd73a4f | 3145 | |
9439aab8 | 3146 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3147 | *sd_idle = 0; |
3148 | ||
1da177e4 | 3149 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3150 | if (local_group) { |
3151 | if (idle_cpu(i) && !first_idle_cpu) { | |
3152 | first_idle_cpu = 1; | |
3153 | balance_cpu = i; | |
3154 | } | |
3155 | ||
a2000572 | 3156 | load = target_load(i, load_idx); |
908a7c1b | 3157 | } else { |
a2000572 | 3158 | load = source_load(i, load_idx); |
908a7c1b KC |
3159 | if (load > max_cpu_load) |
3160 | max_cpu_load = load; | |
3161 | if (min_cpu_load > load) | |
3162 | min_cpu_load = load; | |
3163 | } | |
1da177e4 LT |
3164 | |
3165 | avg_load += load; | |
2dd73a4f | 3166 | sum_nr_running += rq->nr_running; |
dd41f596 | 3167 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3168 | |
3169 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3170 | } |
3171 | ||
783609c6 SS |
3172 | /* |
3173 | * First idle cpu or the first cpu(busiest) in this sched group | |
3174 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3175 | * domains. In the newly idle case, we will allow all the cpu's |
3176 | * to do the newly idle load balance. | |
783609c6 | 3177 | */ |
9439aab8 SS |
3178 | if (idle != CPU_NEWLY_IDLE && local_group && |
3179 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3180 | *balance = 0; |
3181 | goto ret; | |
3182 | } | |
3183 | ||
1da177e4 | 3184 | total_load += avg_load; |
5517d86b | 3185 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3186 | |
3187 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3188 | avg_load = sg_div_cpu_power(group, |
3189 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3190 | |
408ed066 PZ |
3191 | |
3192 | /* | |
3193 | * Consider the group unbalanced when the imbalance is larger | |
3194 | * than the average weight of two tasks. | |
3195 | * | |
3196 | * APZ: with cgroup the avg task weight can vary wildly and | |
3197 | * might not be a suitable number - should we keep a | |
3198 | * normalized nr_running number somewhere that negates | |
3199 | * the hierarchy? | |
3200 | */ | |
3201 | avg_load_per_task = sg_div_cpu_power(group, | |
3202 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3203 | ||
3204 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3205 | __group_imb = 1; |
3206 | ||
5517d86b | 3207 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3208 | |
1da177e4 LT |
3209 | if (local_group) { |
3210 | this_load = avg_load; | |
3211 | this = group; | |
2dd73a4f PW |
3212 | this_nr_running = sum_nr_running; |
3213 | this_load_per_task = sum_weighted_load; | |
3214 | } else if (avg_load > max_load && | |
908a7c1b | 3215 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3216 | max_load = avg_load; |
3217 | busiest = group; | |
2dd73a4f PW |
3218 | busiest_nr_running = sum_nr_running; |
3219 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3220 | group_imb = __group_imb; |
1da177e4 | 3221 | } |
5c45bf27 SS |
3222 | |
3223 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3224 | /* | |
3225 | * Busy processors will not participate in power savings | |
3226 | * balance. | |
3227 | */ | |
dd41f596 IM |
3228 | if (idle == CPU_NOT_IDLE || |
3229 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3230 | goto group_next; | |
5c45bf27 SS |
3231 | |
3232 | /* | |
3233 | * If the local group is idle or completely loaded | |
3234 | * no need to do power savings balance at this domain | |
3235 | */ | |
3236 | if (local_group && (this_nr_running >= group_capacity || | |
3237 | !this_nr_running)) | |
3238 | power_savings_balance = 0; | |
3239 | ||
dd41f596 | 3240 | /* |
5c45bf27 SS |
3241 | * If a group is already running at full capacity or idle, |
3242 | * don't include that group in power savings calculations | |
dd41f596 IM |
3243 | */ |
3244 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3245 | || !sum_nr_running) |
dd41f596 | 3246 | goto group_next; |
5c45bf27 | 3247 | |
dd41f596 | 3248 | /* |
5c45bf27 | 3249 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3250 | * This is the group from where we need to pick up the load |
3251 | * for saving power | |
3252 | */ | |
3253 | if ((sum_nr_running < min_nr_running) || | |
3254 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3255 | first_cpu(group->cpumask) < |
3256 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3257 | group_min = group; |
3258 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3259 | min_load_per_task = sum_weighted_load / |
3260 | sum_nr_running; | |
dd41f596 | 3261 | } |
5c45bf27 | 3262 | |
dd41f596 | 3263 | /* |
5c45bf27 | 3264 | * Calculate the group which is almost near its |
dd41f596 IM |
3265 | * capacity but still has some space to pick up some load |
3266 | * from other group and save more power | |
3267 | */ | |
3268 | if (sum_nr_running <= group_capacity - 1) { | |
3269 | if (sum_nr_running > leader_nr_running || | |
3270 | (sum_nr_running == leader_nr_running && | |
3271 | first_cpu(group->cpumask) > | |
3272 | first_cpu(group_leader->cpumask))) { | |
3273 | group_leader = group; | |
3274 | leader_nr_running = sum_nr_running; | |
3275 | } | |
48f24c4d | 3276 | } |
5c45bf27 SS |
3277 | group_next: |
3278 | #endif | |
1da177e4 LT |
3279 | group = group->next; |
3280 | } while (group != sd->groups); | |
3281 | ||
2dd73a4f | 3282 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3283 | goto out_balanced; |
3284 | ||
3285 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3286 | ||
3287 | if (this_load >= avg_load || | |
3288 | 100*max_load <= sd->imbalance_pct*this_load) | |
3289 | goto out_balanced; | |
3290 | ||
2dd73a4f | 3291 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3292 | if (group_imb) |
3293 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3294 | ||
1da177e4 LT |
3295 | /* |
3296 | * We're trying to get all the cpus to the average_load, so we don't | |
3297 | * want to push ourselves above the average load, nor do we wish to | |
3298 | * reduce the max loaded cpu below the average load, as either of these | |
3299 | * actions would just result in more rebalancing later, and ping-pong | |
3300 | * tasks around. Thus we look for the minimum possible imbalance. | |
3301 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3302 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3303 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3304 | * appear as very large values with unsigned longs. |
3305 | */ | |
2dd73a4f PW |
3306 | if (max_load <= busiest_load_per_task) |
3307 | goto out_balanced; | |
3308 | ||
3309 | /* | |
3310 | * In the presence of smp nice balancing, certain scenarios can have | |
3311 | * max load less than avg load(as we skip the groups at or below | |
3312 | * its cpu_power, while calculating max_load..) | |
3313 | */ | |
3314 | if (max_load < avg_load) { | |
3315 | *imbalance = 0; | |
3316 | goto small_imbalance; | |
3317 | } | |
0c117f1b SS |
3318 | |
3319 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3320 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3321 | |
1da177e4 | 3322 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3323 | *imbalance = min(max_pull * busiest->__cpu_power, |
3324 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3325 | / SCHED_LOAD_SCALE; |
3326 | ||
2dd73a4f PW |
3327 | /* |
3328 | * if *imbalance is less than the average load per runnable task | |
3329 | * there is no gaurantee that any tasks will be moved so we'll have | |
3330 | * a think about bumping its value to force at least one task to be | |
3331 | * moved | |
3332 | */ | |
7fd0d2dd | 3333 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3334 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3335 | unsigned int imbn; |
3336 | ||
3337 | small_imbalance: | |
3338 | pwr_move = pwr_now = 0; | |
3339 | imbn = 2; | |
3340 | if (this_nr_running) { | |
3341 | this_load_per_task /= this_nr_running; | |
3342 | if (busiest_load_per_task > this_load_per_task) | |
3343 | imbn = 1; | |
3344 | } else | |
408ed066 | 3345 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3346 | |
408ed066 | 3347 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3348 | busiest_load_per_task * imbn) { |
2dd73a4f | 3349 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3350 | return busiest; |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * OK, we don't have enough imbalance to justify moving tasks, | |
3355 | * however we may be able to increase total CPU power used by | |
3356 | * moving them. | |
3357 | */ | |
3358 | ||
5517d86b ED |
3359 | pwr_now += busiest->__cpu_power * |
3360 | min(busiest_load_per_task, max_load); | |
3361 | pwr_now += this->__cpu_power * | |
3362 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3363 | pwr_now /= SCHED_LOAD_SCALE; |
3364 | ||
3365 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3366 | tmp = sg_div_cpu_power(busiest, |
3367 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3368 | if (max_load > tmp) |
5517d86b | 3369 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3370 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3371 | |
3372 | /* Amount of load we'd add */ | |
5517d86b | 3373 | if (max_load * busiest->__cpu_power < |
33859f7f | 3374 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3375 | tmp = sg_div_cpu_power(this, |
3376 | max_load * busiest->__cpu_power); | |
1da177e4 | 3377 | else |
5517d86b ED |
3378 | tmp = sg_div_cpu_power(this, |
3379 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3380 | pwr_move += this->__cpu_power * | |
3381 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3382 | pwr_move /= SCHED_LOAD_SCALE; |
3383 | ||
3384 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3385 | if (pwr_move > pwr_now) |
3386 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3387 | } |
3388 | ||
1da177e4 LT |
3389 | return busiest; |
3390 | ||
3391 | out_balanced: | |
5c45bf27 | 3392 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3393 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3394 | goto ret; |
1da177e4 | 3395 | |
5c45bf27 SS |
3396 | if (this == group_leader && group_leader != group_min) { |
3397 | *imbalance = min_load_per_task; | |
3398 | return group_min; | |
3399 | } | |
5c45bf27 | 3400 | #endif |
783609c6 | 3401 | ret: |
1da177e4 LT |
3402 | *imbalance = 0; |
3403 | return NULL; | |
3404 | } | |
3405 | ||
3406 | /* | |
3407 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3408 | */ | |
70b97a7f | 3409 | static struct rq * |
d15bcfdb | 3410 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3411 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3412 | { |
70b97a7f | 3413 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3414 | unsigned long max_load = 0; |
1da177e4 LT |
3415 | int i; |
3416 | ||
363ab6f1 | 3417 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3418 | unsigned long wl; |
0a2966b4 CL |
3419 | |
3420 | if (!cpu_isset(i, *cpus)) | |
3421 | continue; | |
3422 | ||
48f24c4d | 3423 | rq = cpu_rq(i); |
dd41f596 | 3424 | wl = weighted_cpuload(i); |
2dd73a4f | 3425 | |
dd41f596 | 3426 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3427 | continue; |
1da177e4 | 3428 | |
dd41f596 IM |
3429 | if (wl > max_load) { |
3430 | max_load = wl; | |
48f24c4d | 3431 | busiest = rq; |
1da177e4 LT |
3432 | } |
3433 | } | |
3434 | ||
3435 | return busiest; | |
3436 | } | |
3437 | ||
77391d71 NP |
3438 | /* |
3439 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3440 | * so long as it is large enough. | |
3441 | */ | |
3442 | #define MAX_PINNED_INTERVAL 512 | |
3443 | ||
1da177e4 LT |
3444 | /* |
3445 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3446 | * tasks if there is an imbalance. | |
1da177e4 | 3447 | */ |
70b97a7f | 3448 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3449 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3450 | int *balance, cpumask_t *cpus) |
1da177e4 | 3451 | { |
43010659 | 3452 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3453 | struct sched_group *group; |
1da177e4 | 3454 | unsigned long imbalance; |
70b97a7f | 3455 | struct rq *busiest; |
fe2eea3f | 3456 | unsigned long flags; |
5969fe06 | 3457 | |
7c16ec58 MT |
3458 | cpus_setall(*cpus); |
3459 | ||
89c4710e SS |
3460 | /* |
3461 | * When power savings policy is enabled for the parent domain, idle | |
3462 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3463 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3464 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3465 | */ |
d15bcfdb | 3466 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3467 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3468 | sd_idle = 1; |
1da177e4 | 3469 | |
2d72376b | 3470 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3471 | |
0a2966b4 | 3472 | redo: |
c8cba857 | 3473 | update_shares(sd); |
0a2966b4 | 3474 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3475 | cpus, balance); |
783609c6 | 3476 | |
06066714 | 3477 | if (*balance == 0) |
783609c6 | 3478 | goto out_balanced; |
783609c6 | 3479 | |
1da177e4 LT |
3480 | if (!group) { |
3481 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3482 | goto out_balanced; | |
3483 | } | |
3484 | ||
7c16ec58 | 3485 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3486 | if (!busiest) { |
3487 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3488 | goto out_balanced; | |
3489 | } | |
3490 | ||
db935dbd | 3491 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3492 | |
3493 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3494 | ||
43010659 | 3495 | ld_moved = 0; |
1da177e4 LT |
3496 | if (busiest->nr_running > 1) { |
3497 | /* | |
3498 | * Attempt to move tasks. If find_busiest_group has found | |
3499 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3500 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3501 | * correctly treated as an imbalance. |
3502 | */ | |
fe2eea3f | 3503 | local_irq_save(flags); |
e17224bf | 3504 | double_rq_lock(this_rq, busiest); |
43010659 | 3505 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3506 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3507 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3508 | local_irq_restore(flags); |
81026794 | 3509 | |
46cb4b7c SS |
3510 | /* |
3511 | * some other cpu did the load balance for us. | |
3512 | */ | |
43010659 | 3513 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3514 | resched_cpu(this_cpu); |
3515 | ||
81026794 | 3516 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3517 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3518 | cpu_clear(cpu_of(busiest), *cpus); |
3519 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3520 | goto redo; |
81026794 | 3521 | goto out_balanced; |
0a2966b4 | 3522 | } |
1da177e4 | 3523 | } |
81026794 | 3524 | |
43010659 | 3525 | if (!ld_moved) { |
1da177e4 LT |
3526 | schedstat_inc(sd, lb_failed[idle]); |
3527 | sd->nr_balance_failed++; | |
3528 | ||
3529 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3530 | |
fe2eea3f | 3531 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3532 | |
3533 | /* don't kick the migration_thread, if the curr | |
3534 | * task on busiest cpu can't be moved to this_cpu | |
3535 | */ | |
3536 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3537 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3538 | all_pinned = 1; |
3539 | goto out_one_pinned; | |
3540 | } | |
3541 | ||
1da177e4 LT |
3542 | if (!busiest->active_balance) { |
3543 | busiest->active_balance = 1; | |
3544 | busiest->push_cpu = this_cpu; | |
81026794 | 3545 | active_balance = 1; |
1da177e4 | 3546 | } |
fe2eea3f | 3547 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3548 | if (active_balance) |
1da177e4 LT |
3549 | wake_up_process(busiest->migration_thread); |
3550 | ||
3551 | /* | |
3552 | * We've kicked active balancing, reset the failure | |
3553 | * counter. | |
3554 | */ | |
39507451 | 3555 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3556 | } |
81026794 | 3557 | } else |
1da177e4 LT |
3558 | sd->nr_balance_failed = 0; |
3559 | ||
81026794 | 3560 | if (likely(!active_balance)) { |
1da177e4 LT |
3561 | /* We were unbalanced, so reset the balancing interval */ |
3562 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3563 | } else { |
3564 | /* | |
3565 | * If we've begun active balancing, start to back off. This | |
3566 | * case may not be covered by the all_pinned logic if there | |
3567 | * is only 1 task on the busy runqueue (because we don't call | |
3568 | * move_tasks). | |
3569 | */ | |
3570 | if (sd->balance_interval < sd->max_interval) | |
3571 | sd->balance_interval *= 2; | |
1da177e4 LT |
3572 | } |
3573 | ||
43010659 | 3574 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3575 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3576 | ld_moved = -1; |
3577 | ||
3578 | goto out; | |
1da177e4 LT |
3579 | |
3580 | out_balanced: | |
1da177e4 LT |
3581 | schedstat_inc(sd, lb_balanced[idle]); |
3582 | ||
16cfb1c0 | 3583 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3584 | |
3585 | out_one_pinned: | |
1da177e4 | 3586 | /* tune up the balancing interval */ |
77391d71 NP |
3587 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3588 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3589 | sd->balance_interval *= 2; |
3590 | ||
48f24c4d | 3591 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3592 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3593 | ld_moved = -1; |
3594 | else | |
3595 | ld_moved = 0; | |
3596 | out: | |
c8cba857 PZ |
3597 | if (ld_moved) |
3598 | update_shares(sd); | |
c09595f6 | 3599 | return ld_moved; |
1da177e4 LT |
3600 | } |
3601 | ||
3602 | /* | |
3603 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3604 | * tasks if there is an imbalance. | |
3605 | * | |
d15bcfdb | 3606 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3607 | * this_rq is locked. |
3608 | */ | |
48f24c4d | 3609 | static int |
7c16ec58 MT |
3610 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3611 | cpumask_t *cpus) | |
1da177e4 LT |
3612 | { |
3613 | struct sched_group *group; | |
70b97a7f | 3614 | struct rq *busiest = NULL; |
1da177e4 | 3615 | unsigned long imbalance; |
43010659 | 3616 | int ld_moved = 0; |
5969fe06 | 3617 | int sd_idle = 0; |
969bb4e4 | 3618 | int all_pinned = 0; |
7c16ec58 MT |
3619 | |
3620 | cpus_setall(*cpus); | |
5969fe06 | 3621 | |
89c4710e SS |
3622 | /* |
3623 | * When power savings policy is enabled for the parent domain, idle | |
3624 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3625 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3626 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3627 | */ |
3628 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3629 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3630 | sd_idle = 1; |
1da177e4 | 3631 | |
2d72376b | 3632 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3633 | redo: |
3e5459b4 | 3634 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3635 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3636 | &sd_idle, cpus, NULL); |
1da177e4 | 3637 | if (!group) { |
d15bcfdb | 3638 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3639 | goto out_balanced; |
1da177e4 LT |
3640 | } |
3641 | ||
7c16ec58 | 3642 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3643 | if (!busiest) { |
d15bcfdb | 3644 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3645 | goto out_balanced; |
1da177e4 LT |
3646 | } |
3647 | ||
db935dbd NP |
3648 | BUG_ON(busiest == this_rq); |
3649 | ||
d15bcfdb | 3650 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3651 | |
43010659 | 3652 | ld_moved = 0; |
d6d5cfaf NP |
3653 | if (busiest->nr_running > 1) { |
3654 | /* Attempt to move tasks */ | |
3655 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3656 | /* this_rq->clock is already updated */ |
3657 | update_rq_clock(busiest); | |
43010659 | 3658 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3659 | imbalance, sd, CPU_NEWLY_IDLE, |
3660 | &all_pinned); | |
1b12bbc7 | 3661 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3662 | |
969bb4e4 | 3663 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3664 | cpu_clear(cpu_of(busiest), *cpus); |
3665 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3666 | goto redo; |
3667 | } | |
d6d5cfaf NP |
3668 | } |
3669 | ||
43010659 | 3670 | if (!ld_moved) { |
d15bcfdb | 3671 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3672 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3673 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3674 | return -1; |
3675 | } else | |
16cfb1c0 | 3676 | sd->nr_balance_failed = 0; |
1da177e4 | 3677 | |
3e5459b4 | 3678 | update_shares_locked(this_rq, sd); |
43010659 | 3679 | return ld_moved; |
16cfb1c0 NP |
3680 | |
3681 | out_balanced: | |
d15bcfdb | 3682 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3683 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3684 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3685 | return -1; |
16cfb1c0 | 3686 | sd->nr_balance_failed = 0; |
48f24c4d | 3687 | |
16cfb1c0 | 3688 | return 0; |
1da177e4 LT |
3689 | } |
3690 | ||
3691 | /* | |
3692 | * idle_balance is called by schedule() if this_cpu is about to become | |
3693 | * idle. Attempts to pull tasks from other CPUs. | |
3694 | */ | |
70b97a7f | 3695 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3696 | { |
3697 | struct sched_domain *sd; | |
dd41f596 IM |
3698 | int pulled_task = -1; |
3699 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3700 | cpumask_t tmpmask; |
1da177e4 LT |
3701 | |
3702 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3703 | unsigned long interval; |
3704 | ||
3705 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3706 | continue; | |
3707 | ||
3708 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3709 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3710 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3711 | sd, &tmpmask); | |
92c4ca5c CL |
3712 | |
3713 | interval = msecs_to_jiffies(sd->balance_interval); | |
3714 | if (time_after(next_balance, sd->last_balance + interval)) | |
3715 | next_balance = sd->last_balance + interval; | |
3716 | if (pulled_task) | |
3717 | break; | |
1da177e4 | 3718 | } |
dd41f596 | 3719 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3720 | /* |
3721 | * We are going idle. next_balance may be set based on | |
3722 | * a busy processor. So reset next_balance. | |
3723 | */ | |
3724 | this_rq->next_balance = next_balance; | |
dd41f596 | 3725 | } |
1da177e4 LT |
3726 | } |
3727 | ||
3728 | /* | |
3729 | * active_load_balance is run by migration threads. It pushes running tasks | |
3730 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3731 | * running on each physical CPU where possible, and avoids physical / | |
3732 | * logical imbalances. | |
3733 | * | |
3734 | * Called with busiest_rq locked. | |
3735 | */ | |
70b97a7f | 3736 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3737 | { |
39507451 | 3738 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3739 | struct sched_domain *sd; |
3740 | struct rq *target_rq; | |
39507451 | 3741 | |
48f24c4d | 3742 | /* Is there any task to move? */ |
39507451 | 3743 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3744 | return; |
3745 | ||
3746 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3747 | |
3748 | /* | |
39507451 | 3749 | * This condition is "impossible", if it occurs |
41a2d6cf | 3750 | * we need to fix it. Originally reported by |
39507451 | 3751 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3752 | */ |
39507451 | 3753 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3754 | |
39507451 NP |
3755 | /* move a task from busiest_rq to target_rq */ |
3756 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3757 | update_rq_clock(busiest_rq); |
3758 | update_rq_clock(target_rq); | |
39507451 NP |
3759 | |
3760 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3761 | for_each_domain(target_cpu, sd) { |
39507451 | 3762 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3763 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3764 | break; |
c96d145e | 3765 | } |
39507451 | 3766 | |
48f24c4d | 3767 | if (likely(sd)) { |
2d72376b | 3768 | schedstat_inc(sd, alb_count); |
39507451 | 3769 | |
43010659 PW |
3770 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3771 | sd, CPU_IDLE)) | |
48f24c4d IM |
3772 | schedstat_inc(sd, alb_pushed); |
3773 | else | |
3774 | schedstat_inc(sd, alb_failed); | |
3775 | } | |
1b12bbc7 | 3776 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3777 | } |
3778 | ||
46cb4b7c SS |
3779 | #ifdef CONFIG_NO_HZ |
3780 | static struct { | |
3781 | atomic_t load_balancer; | |
41a2d6cf | 3782 | cpumask_t cpu_mask; |
46cb4b7c SS |
3783 | } nohz ____cacheline_aligned = { |
3784 | .load_balancer = ATOMIC_INIT(-1), | |
3785 | .cpu_mask = CPU_MASK_NONE, | |
3786 | }; | |
3787 | ||
7835b98b | 3788 | /* |
46cb4b7c SS |
3789 | * This routine will try to nominate the ilb (idle load balancing) |
3790 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3791 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3792 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3793 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3794 | * arrives... | |
3795 | * | |
3796 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3797 | * for idle load balancing. ilb owner will still be part of | |
3798 | * nohz.cpu_mask.. | |
7835b98b | 3799 | * |
46cb4b7c SS |
3800 | * While stopping the tick, this cpu will become the ilb owner if there |
3801 | * is no other owner. And will be the owner till that cpu becomes busy | |
3802 | * or if all cpus in the system stop their ticks at which point | |
3803 | * there is no need for ilb owner. | |
3804 | * | |
3805 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3806 | * next busy scheduler_tick() | |
3807 | */ | |
3808 | int select_nohz_load_balancer(int stop_tick) | |
3809 | { | |
3810 | int cpu = smp_processor_id(); | |
3811 | ||
3812 | if (stop_tick) { | |
3813 | cpu_set(cpu, nohz.cpu_mask); | |
3814 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3815 | ||
3816 | /* | |
3817 | * If we are going offline and still the leader, give up! | |
3818 | */ | |
e761b772 | 3819 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3820 | atomic_read(&nohz.load_balancer) == cpu) { |
3821 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3822 | BUG(); | |
3823 | return 0; | |
3824 | } | |
3825 | ||
3826 | /* time for ilb owner also to sleep */ | |
3827 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3828 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3829 | atomic_set(&nohz.load_balancer, -1); | |
3830 | return 0; | |
3831 | } | |
3832 | ||
3833 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3834 | /* make me the ilb owner */ | |
3835 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3836 | return 1; | |
3837 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3838 | return 1; | |
3839 | } else { | |
3840 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3841 | return 0; | |
3842 | ||
3843 | cpu_clear(cpu, nohz.cpu_mask); | |
3844 | ||
3845 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3846 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3847 | BUG(); | |
3848 | } | |
3849 | return 0; | |
3850 | } | |
3851 | #endif | |
3852 | ||
3853 | static DEFINE_SPINLOCK(balancing); | |
3854 | ||
3855 | /* | |
7835b98b CL |
3856 | * It checks each scheduling domain to see if it is due to be balanced, |
3857 | * and initiates a balancing operation if so. | |
3858 | * | |
3859 | * Balancing parameters are set up in arch_init_sched_domains. | |
3860 | */ | |
a9957449 | 3861 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3862 | { |
46cb4b7c SS |
3863 | int balance = 1; |
3864 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3865 | unsigned long interval; |
3866 | struct sched_domain *sd; | |
46cb4b7c | 3867 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3868 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3869 | int update_next_balance = 0; |
d07355f5 | 3870 | int need_serialize; |
7c16ec58 | 3871 | cpumask_t tmp; |
1da177e4 | 3872 | |
46cb4b7c | 3873 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3874 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3875 | continue; | |
3876 | ||
3877 | interval = sd->balance_interval; | |
d15bcfdb | 3878 | if (idle != CPU_IDLE) |
1da177e4 LT |
3879 | interval *= sd->busy_factor; |
3880 | ||
3881 | /* scale ms to jiffies */ | |
3882 | interval = msecs_to_jiffies(interval); | |
3883 | if (unlikely(!interval)) | |
3884 | interval = 1; | |
dd41f596 IM |
3885 | if (interval > HZ*NR_CPUS/10) |
3886 | interval = HZ*NR_CPUS/10; | |
3887 | ||
d07355f5 | 3888 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3889 | |
d07355f5 | 3890 | if (need_serialize) { |
08c183f3 CL |
3891 | if (!spin_trylock(&balancing)) |
3892 | goto out; | |
3893 | } | |
3894 | ||
c9819f45 | 3895 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3896 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3897 | /* |
3898 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3899 | * longer idle, or one of our SMT siblings is |
3900 | * not idle. | |
3901 | */ | |
d15bcfdb | 3902 | idle = CPU_NOT_IDLE; |
1da177e4 | 3903 | } |
1bd77f2d | 3904 | sd->last_balance = jiffies; |
1da177e4 | 3905 | } |
d07355f5 | 3906 | if (need_serialize) |
08c183f3 CL |
3907 | spin_unlock(&balancing); |
3908 | out: | |
f549da84 | 3909 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3910 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3911 | update_next_balance = 1; |
3912 | } | |
783609c6 SS |
3913 | |
3914 | /* | |
3915 | * Stop the load balance at this level. There is another | |
3916 | * CPU in our sched group which is doing load balancing more | |
3917 | * actively. | |
3918 | */ | |
3919 | if (!balance) | |
3920 | break; | |
1da177e4 | 3921 | } |
f549da84 SS |
3922 | |
3923 | /* | |
3924 | * next_balance will be updated only when there is a need. | |
3925 | * When the cpu is attached to null domain for ex, it will not be | |
3926 | * updated. | |
3927 | */ | |
3928 | if (likely(update_next_balance)) | |
3929 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3930 | } |
3931 | ||
3932 | /* | |
3933 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3934 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3935 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3936 | */ | |
3937 | static void run_rebalance_domains(struct softirq_action *h) | |
3938 | { | |
dd41f596 IM |
3939 | int this_cpu = smp_processor_id(); |
3940 | struct rq *this_rq = cpu_rq(this_cpu); | |
3941 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3942 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3943 | |
dd41f596 | 3944 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3945 | |
3946 | #ifdef CONFIG_NO_HZ | |
3947 | /* | |
3948 | * If this cpu is the owner for idle load balancing, then do the | |
3949 | * balancing on behalf of the other idle cpus whose ticks are | |
3950 | * stopped. | |
3951 | */ | |
dd41f596 IM |
3952 | if (this_rq->idle_at_tick && |
3953 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3954 | cpumask_t cpus = nohz.cpu_mask; |
3955 | struct rq *rq; | |
3956 | int balance_cpu; | |
3957 | ||
dd41f596 | 3958 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3959 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3960 | /* |
3961 | * If this cpu gets work to do, stop the load balancing | |
3962 | * work being done for other cpus. Next load | |
3963 | * balancing owner will pick it up. | |
3964 | */ | |
3965 | if (need_resched()) | |
3966 | break; | |
3967 | ||
de0cf899 | 3968 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3969 | |
3970 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3971 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3972 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3973 | } |
3974 | } | |
3975 | #endif | |
3976 | } | |
3977 | ||
3978 | /* | |
3979 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3980 | * | |
3981 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3982 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3983 | * if the whole system is idle. | |
3984 | */ | |
dd41f596 | 3985 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3986 | { |
46cb4b7c SS |
3987 | #ifdef CONFIG_NO_HZ |
3988 | /* | |
3989 | * If we were in the nohz mode recently and busy at the current | |
3990 | * scheduler tick, then check if we need to nominate new idle | |
3991 | * load balancer. | |
3992 | */ | |
3993 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3994 | rq->in_nohz_recently = 0; | |
3995 | ||
3996 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3997 | cpu_clear(cpu, nohz.cpu_mask); | |
3998 | atomic_set(&nohz.load_balancer, -1); | |
3999 | } | |
4000 | ||
4001 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4002 | /* | |
4003 | * simple selection for now: Nominate the | |
4004 | * first cpu in the nohz list to be the next | |
4005 | * ilb owner. | |
4006 | * | |
4007 | * TBD: Traverse the sched domains and nominate | |
4008 | * the nearest cpu in the nohz.cpu_mask. | |
4009 | */ | |
4010 | int ilb = first_cpu(nohz.cpu_mask); | |
4011 | ||
434d53b0 | 4012 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4013 | resched_cpu(ilb); |
4014 | } | |
4015 | } | |
4016 | ||
4017 | /* | |
4018 | * If this cpu is idle and doing idle load balancing for all the | |
4019 | * cpus with ticks stopped, is it time for that to stop? | |
4020 | */ | |
4021 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4022 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4023 | resched_cpu(cpu); | |
4024 | return; | |
4025 | } | |
4026 | ||
4027 | /* | |
4028 | * If this cpu is idle and the idle load balancing is done by | |
4029 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4030 | */ | |
4031 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4032 | cpu_isset(cpu, nohz.cpu_mask)) | |
4033 | return; | |
4034 | #endif | |
4035 | if (time_after_eq(jiffies, rq->next_balance)) | |
4036 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4037 | } |
dd41f596 IM |
4038 | |
4039 | #else /* CONFIG_SMP */ | |
4040 | ||
1da177e4 LT |
4041 | /* |
4042 | * on UP we do not need to balance between CPUs: | |
4043 | */ | |
70b97a7f | 4044 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4045 | { |
4046 | } | |
dd41f596 | 4047 | |
1da177e4 LT |
4048 | #endif |
4049 | ||
1da177e4 LT |
4050 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4051 | ||
4052 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4053 | ||
4054 | /* | |
f06febc9 FM |
4055 | * Return any ns on the sched_clock that have not yet been banked in |
4056 | * @p in case that task is currently running. | |
1da177e4 | 4057 | */ |
bb34d92f | 4058 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4059 | { |
1da177e4 | 4060 | unsigned long flags; |
41b86e9c | 4061 | struct rq *rq; |
bb34d92f | 4062 | u64 ns = 0; |
48f24c4d | 4063 | |
41b86e9c | 4064 | rq = task_rq_lock(p, &flags); |
1508487e | 4065 | |
051a1d1a | 4066 | if (task_current(rq, p)) { |
f06febc9 FM |
4067 | u64 delta_exec; |
4068 | ||
a8e504d2 IM |
4069 | update_rq_clock(rq); |
4070 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4071 | if ((s64)delta_exec > 0) |
bb34d92f | 4072 | ns = delta_exec; |
41b86e9c | 4073 | } |
48f24c4d | 4074 | |
41b86e9c | 4075 | task_rq_unlock(rq, &flags); |
48f24c4d | 4076 | |
1da177e4 LT |
4077 | return ns; |
4078 | } | |
4079 | ||
1da177e4 LT |
4080 | /* |
4081 | * Account user cpu time to a process. | |
4082 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4083 | * @cputime: the cpu time spent in user space since the last update |
4084 | */ | |
4085 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4086 | { | |
4087 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4088 | cputime64_t tmp; | |
4089 | ||
4090 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4091 | account_group_user_time(p, cputime); |
1da177e4 LT |
4092 | |
4093 | /* Add user time to cpustat. */ | |
4094 | tmp = cputime_to_cputime64(cputime); | |
4095 | if (TASK_NICE(p) > 0) | |
4096 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4097 | else | |
4098 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4099 | /* Account for user time used */ |
4100 | acct_update_integrals(p); | |
1da177e4 LT |
4101 | } |
4102 | ||
94886b84 LV |
4103 | /* |
4104 | * Account guest cpu time to a process. | |
4105 | * @p: the process that the cpu time gets accounted to | |
4106 | * @cputime: the cpu time spent in virtual machine since the last update | |
4107 | */ | |
f7402e03 | 4108 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4109 | { |
4110 | cputime64_t tmp; | |
4111 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4112 | ||
4113 | tmp = cputime_to_cputime64(cputime); | |
4114 | ||
4115 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4116 | account_group_user_time(p, cputime); |
94886b84 LV |
4117 | p->gtime = cputime_add(p->gtime, cputime); |
4118 | ||
4119 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4120 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4121 | } | |
4122 | ||
c66f08be MN |
4123 | /* |
4124 | * Account scaled user cpu time to a process. | |
4125 | * @p: the process that the cpu time gets accounted to | |
4126 | * @cputime: the cpu time spent in user space since the last update | |
4127 | */ | |
4128 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4129 | { | |
4130 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4131 | } | |
4132 | ||
1da177e4 LT |
4133 | /* |
4134 | * Account system cpu time to a process. | |
4135 | * @p: the process that the cpu time gets accounted to | |
4136 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4137 | * @cputime: the cpu time spent in kernel space since the last update | |
4138 | */ | |
4139 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4140 | cputime_t cputime) | |
4141 | { | |
4142 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4143 | struct rq *rq = this_rq(); |
1da177e4 LT |
4144 | cputime64_t tmp; |
4145 | ||
983ed7a6 HH |
4146 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4147 | account_guest_time(p, cputime); | |
4148 | return; | |
4149 | } | |
94886b84 | 4150 | |
1da177e4 | 4151 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4152 | account_group_system_time(p, cputime); |
1da177e4 LT |
4153 | |
4154 | /* Add system time to cpustat. */ | |
4155 | tmp = cputime_to_cputime64(cputime); | |
4156 | if (hardirq_count() - hardirq_offset) | |
4157 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4158 | else if (softirq_count()) | |
4159 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4160 | else if (p != rq->idle) |
1da177e4 | 4161 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4162 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4163 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4164 | else | |
4165 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4166 | /* Account for system time used */ | |
4167 | acct_update_integrals(p); | |
1da177e4 LT |
4168 | } |
4169 | ||
c66f08be MN |
4170 | /* |
4171 | * Account scaled system cpu time to a process. | |
4172 | * @p: the process that the cpu time gets accounted to | |
4173 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4174 | * @cputime: the cpu time spent in kernel space since the last update | |
4175 | */ | |
4176 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4177 | { | |
4178 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4179 | } | |
4180 | ||
1da177e4 LT |
4181 | /* |
4182 | * Account for involuntary wait time. | |
4183 | * @p: the process from which the cpu time has been stolen | |
4184 | * @steal: the cpu time spent in involuntary wait | |
4185 | */ | |
4186 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4187 | { | |
4188 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4189 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4190 | struct rq *rq = this_rq(); |
1da177e4 LT |
4191 | |
4192 | if (p == rq->idle) { | |
4193 | p->stime = cputime_add(p->stime, steal); | |
f06febc9 | 4194 | account_group_system_time(p, steal); |
1da177e4 LT |
4195 | if (atomic_read(&rq->nr_iowait) > 0) |
4196 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4197 | else | |
4198 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4199 | } else |
1da177e4 LT |
4200 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4201 | } | |
4202 | ||
49048622 BS |
4203 | /* |
4204 | * Use precise platform statistics if available: | |
4205 | */ | |
4206 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4207 | cputime_t task_utime(struct task_struct *p) | |
4208 | { | |
4209 | return p->utime; | |
4210 | } | |
4211 | ||
4212 | cputime_t task_stime(struct task_struct *p) | |
4213 | { | |
4214 | return p->stime; | |
4215 | } | |
4216 | #else | |
4217 | cputime_t task_utime(struct task_struct *p) | |
4218 | { | |
4219 | clock_t utime = cputime_to_clock_t(p->utime), | |
4220 | total = utime + cputime_to_clock_t(p->stime); | |
4221 | u64 temp; | |
4222 | ||
4223 | /* | |
4224 | * Use CFS's precise accounting: | |
4225 | */ | |
4226 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4227 | ||
4228 | if (total) { | |
4229 | temp *= utime; | |
4230 | do_div(temp, total); | |
4231 | } | |
4232 | utime = (clock_t)temp; | |
4233 | ||
4234 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4235 | return p->prev_utime; | |
4236 | } | |
4237 | ||
4238 | cputime_t task_stime(struct task_struct *p) | |
4239 | { | |
4240 | clock_t stime; | |
4241 | ||
4242 | /* | |
4243 | * Use CFS's precise accounting. (we subtract utime from | |
4244 | * the total, to make sure the total observed by userspace | |
4245 | * grows monotonically - apps rely on that): | |
4246 | */ | |
4247 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4248 | cputime_to_clock_t(task_utime(p)); | |
4249 | ||
4250 | if (stime >= 0) | |
4251 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4252 | ||
4253 | return p->prev_stime; | |
4254 | } | |
4255 | #endif | |
4256 | ||
4257 | inline cputime_t task_gtime(struct task_struct *p) | |
4258 | { | |
4259 | return p->gtime; | |
4260 | } | |
4261 | ||
7835b98b CL |
4262 | /* |
4263 | * This function gets called by the timer code, with HZ frequency. | |
4264 | * We call it with interrupts disabled. | |
4265 | * | |
4266 | * It also gets called by the fork code, when changing the parent's | |
4267 | * timeslices. | |
4268 | */ | |
4269 | void scheduler_tick(void) | |
4270 | { | |
7835b98b CL |
4271 | int cpu = smp_processor_id(); |
4272 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4273 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4274 | |
4275 | sched_clock_tick(); | |
dd41f596 IM |
4276 | |
4277 | spin_lock(&rq->lock); | |
3e51f33f | 4278 | update_rq_clock(rq); |
f1a438d8 | 4279 | update_cpu_load(rq); |
fa85ae24 | 4280 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4281 | spin_unlock(&rq->lock); |
7835b98b | 4282 | |
e418e1c2 | 4283 | #ifdef CONFIG_SMP |
dd41f596 IM |
4284 | rq->idle_at_tick = idle_cpu(cpu); |
4285 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4286 | #endif |
1da177e4 LT |
4287 | } |
4288 | ||
6cd8a4bb SR |
4289 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4290 | defined(CONFIG_PREEMPT_TRACER)) | |
4291 | ||
4292 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4293 | { | |
4294 | if (in_lock_functions(addr)) { | |
4295 | addr = CALLER_ADDR2; | |
4296 | if (in_lock_functions(addr)) | |
4297 | addr = CALLER_ADDR3; | |
4298 | } | |
4299 | return addr; | |
4300 | } | |
1da177e4 | 4301 | |
43627582 | 4302 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4303 | { |
6cd8a4bb | 4304 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4305 | /* |
4306 | * Underflow? | |
4307 | */ | |
9a11b49a IM |
4308 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4309 | return; | |
6cd8a4bb | 4310 | #endif |
1da177e4 | 4311 | preempt_count() += val; |
6cd8a4bb | 4312 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4313 | /* |
4314 | * Spinlock count overflowing soon? | |
4315 | */ | |
33859f7f MOS |
4316 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4317 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4318 | #endif |
4319 | if (preempt_count() == val) | |
4320 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4321 | } |
4322 | EXPORT_SYMBOL(add_preempt_count); | |
4323 | ||
43627582 | 4324 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4325 | { |
6cd8a4bb | 4326 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4327 | /* |
4328 | * Underflow? | |
4329 | */ | |
9a11b49a IM |
4330 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4331 | return; | |
1da177e4 LT |
4332 | /* |
4333 | * Is the spinlock portion underflowing? | |
4334 | */ | |
9a11b49a IM |
4335 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4336 | !(preempt_count() & PREEMPT_MASK))) | |
4337 | return; | |
6cd8a4bb | 4338 | #endif |
9a11b49a | 4339 | |
6cd8a4bb SR |
4340 | if (preempt_count() == val) |
4341 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4342 | preempt_count() -= val; |
4343 | } | |
4344 | EXPORT_SYMBOL(sub_preempt_count); | |
4345 | ||
4346 | #endif | |
4347 | ||
4348 | /* | |
dd41f596 | 4349 | * Print scheduling while atomic bug: |
1da177e4 | 4350 | */ |
dd41f596 | 4351 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4352 | { |
838225b4 SS |
4353 | struct pt_regs *regs = get_irq_regs(); |
4354 | ||
4355 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4356 | prev->comm, prev->pid, preempt_count()); | |
4357 | ||
dd41f596 | 4358 | debug_show_held_locks(prev); |
e21f5b15 | 4359 | print_modules(); |
dd41f596 IM |
4360 | if (irqs_disabled()) |
4361 | print_irqtrace_events(prev); | |
838225b4 SS |
4362 | |
4363 | if (regs) | |
4364 | show_regs(regs); | |
4365 | else | |
4366 | dump_stack(); | |
dd41f596 | 4367 | } |
1da177e4 | 4368 | |
dd41f596 IM |
4369 | /* |
4370 | * Various schedule()-time debugging checks and statistics: | |
4371 | */ | |
4372 | static inline void schedule_debug(struct task_struct *prev) | |
4373 | { | |
1da177e4 | 4374 | /* |
41a2d6cf | 4375 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4376 | * schedule() atomically, we ignore that path for now. |
4377 | * Otherwise, whine if we are scheduling when we should not be. | |
4378 | */ | |
3f33a7ce | 4379 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4380 | __schedule_bug(prev); |
4381 | ||
1da177e4 LT |
4382 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4383 | ||
2d72376b | 4384 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4385 | #ifdef CONFIG_SCHEDSTATS |
4386 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4387 | schedstat_inc(this_rq(), bkl_count); |
4388 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4389 | } |
4390 | #endif | |
dd41f596 IM |
4391 | } |
4392 | ||
4393 | /* | |
4394 | * Pick up the highest-prio task: | |
4395 | */ | |
4396 | static inline struct task_struct * | |
ff95f3df | 4397 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4398 | { |
5522d5d5 | 4399 | const struct sched_class *class; |
dd41f596 | 4400 | struct task_struct *p; |
1da177e4 LT |
4401 | |
4402 | /* | |
dd41f596 IM |
4403 | * Optimization: we know that if all tasks are in |
4404 | * the fair class we can call that function directly: | |
1da177e4 | 4405 | */ |
dd41f596 | 4406 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4407 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4408 | if (likely(p)) |
4409 | return p; | |
1da177e4 LT |
4410 | } |
4411 | ||
dd41f596 IM |
4412 | class = sched_class_highest; |
4413 | for ( ; ; ) { | |
fb8d4724 | 4414 | p = class->pick_next_task(rq); |
dd41f596 IM |
4415 | if (p) |
4416 | return p; | |
4417 | /* | |
4418 | * Will never be NULL as the idle class always | |
4419 | * returns a non-NULL p: | |
4420 | */ | |
4421 | class = class->next; | |
4422 | } | |
4423 | } | |
1da177e4 | 4424 | |
dd41f596 IM |
4425 | /* |
4426 | * schedule() is the main scheduler function. | |
4427 | */ | |
4428 | asmlinkage void __sched schedule(void) | |
4429 | { | |
4430 | struct task_struct *prev, *next; | |
67ca7bde | 4431 | unsigned long *switch_count; |
dd41f596 | 4432 | struct rq *rq; |
31656519 | 4433 | int cpu; |
dd41f596 IM |
4434 | |
4435 | need_resched: | |
4436 | preempt_disable(); | |
4437 | cpu = smp_processor_id(); | |
4438 | rq = cpu_rq(cpu); | |
4439 | rcu_qsctr_inc(cpu); | |
4440 | prev = rq->curr; | |
4441 | switch_count = &prev->nivcsw; | |
4442 | ||
4443 | release_kernel_lock(prev); | |
4444 | need_resched_nonpreemptible: | |
4445 | ||
4446 | schedule_debug(prev); | |
1da177e4 | 4447 | |
31656519 | 4448 | if (sched_feat(HRTICK)) |
f333fdc9 | 4449 | hrtick_clear(rq); |
8f4d37ec | 4450 | |
1e819950 IM |
4451 | /* |
4452 | * Do the rq-clock update outside the rq lock: | |
4453 | */ | |
4454 | local_irq_disable(); | |
3e51f33f | 4455 | update_rq_clock(rq); |
1e819950 IM |
4456 | spin_lock(&rq->lock); |
4457 | clear_tsk_need_resched(prev); | |
1da177e4 | 4458 | |
1da177e4 | 4459 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4460 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4461 | prev->state = TASK_RUNNING; |
16882c1e | 4462 | else |
2e1cb74a | 4463 | deactivate_task(rq, prev, 1); |
dd41f596 | 4464 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4465 | } |
4466 | ||
9a897c5a SR |
4467 | #ifdef CONFIG_SMP |
4468 | if (prev->sched_class->pre_schedule) | |
4469 | prev->sched_class->pre_schedule(rq, prev); | |
4470 | #endif | |
f65eda4f | 4471 | |
dd41f596 | 4472 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4473 | idle_balance(cpu, rq); |
1da177e4 | 4474 | |
31ee529c | 4475 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4476 | next = pick_next_task(rq, prev); |
1da177e4 | 4477 | |
1da177e4 | 4478 | if (likely(prev != next)) { |
673a90a1 DS |
4479 | sched_info_switch(prev, next); |
4480 | ||
1da177e4 LT |
4481 | rq->nr_switches++; |
4482 | rq->curr = next; | |
4483 | ++*switch_count; | |
4484 | ||
dd41f596 | 4485 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4486 | /* |
4487 | * the context switch might have flipped the stack from under | |
4488 | * us, hence refresh the local variables. | |
4489 | */ | |
4490 | cpu = smp_processor_id(); | |
4491 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4492 | } else |
4493 | spin_unlock_irq(&rq->lock); | |
4494 | ||
8f4d37ec | 4495 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4496 | goto need_resched_nonpreemptible; |
8f4d37ec | 4497 | |
1da177e4 LT |
4498 | preempt_enable_no_resched(); |
4499 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4500 | goto need_resched; | |
4501 | } | |
1da177e4 LT |
4502 | EXPORT_SYMBOL(schedule); |
4503 | ||
4504 | #ifdef CONFIG_PREEMPT | |
4505 | /* | |
2ed6e34f | 4506 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4507 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4508 | * occur there and call schedule directly. |
4509 | */ | |
4510 | asmlinkage void __sched preempt_schedule(void) | |
4511 | { | |
4512 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4513 | |
1da177e4 LT |
4514 | /* |
4515 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4516 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4517 | */ |
beed33a8 | 4518 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4519 | return; |
4520 | ||
3a5c359a AK |
4521 | do { |
4522 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4523 | schedule(); |
3a5c359a | 4524 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4525 | |
3a5c359a AK |
4526 | /* |
4527 | * Check again in case we missed a preemption opportunity | |
4528 | * between schedule and now. | |
4529 | */ | |
4530 | barrier(); | |
4531 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4532 | } |
1da177e4 LT |
4533 | EXPORT_SYMBOL(preempt_schedule); |
4534 | ||
4535 | /* | |
2ed6e34f | 4536 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4537 | * off of irq context. |
4538 | * Note, that this is called and return with irqs disabled. This will | |
4539 | * protect us against recursive calling from irq. | |
4540 | */ | |
4541 | asmlinkage void __sched preempt_schedule_irq(void) | |
4542 | { | |
4543 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4544 | |
2ed6e34f | 4545 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4546 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4547 | ||
3a5c359a AK |
4548 | do { |
4549 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4550 | local_irq_enable(); |
4551 | schedule(); | |
4552 | local_irq_disable(); | |
3a5c359a | 4553 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4554 | |
3a5c359a AK |
4555 | /* |
4556 | * Check again in case we missed a preemption opportunity | |
4557 | * between schedule and now. | |
4558 | */ | |
4559 | barrier(); | |
4560 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4561 | } |
4562 | ||
4563 | #endif /* CONFIG_PREEMPT */ | |
4564 | ||
95cdf3b7 IM |
4565 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4566 | void *key) | |
1da177e4 | 4567 | { |
48f24c4d | 4568 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4569 | } |
1da177e4 LT |
4570 | EXPORT_SYMBOL(default_wake_function); |
4571 | ||
4572 | /* | |
41a2d6cf IM |
4573 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4574 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4575 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4576 | * | |
4577 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4578 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4579 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4580 | */ | |
4581 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4582 | int nr_exclusive, int sync, void *key) | |
4583 | { | |
2e45874c | 4584 | wait_queue_t *curr, *next; |
1da177e4 | 4585 | |
2e45874c | 4586 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4587 | unsigned flags = curr->flags; |
4588 | ||
1da177e4 | 4589 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4590 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4591 | break; |
4592 | } | |
4593 | } | |
4594 | ||
4595 | /** | |
4596 | * __wake_up - wake up threads blocked on a waitqueue. | |
4597 | * @q: the waitqueue | |
4598 | * @mode: which threads | |
4599 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4600 | * @key: is directly passed to the wakeup function |
1da177e4 | 4601 | */ |
7ad5b3a5 | 4602 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4603 | int nr_exclusive, void *key) |
1da177e4 LT |
4604 | { |
4605 | unsigned long flags; | |
4606 | ||
4607 | spin_lock_irqsave(&q->lock, flags); | |
4608 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4609 | spin_unlock_irqrestore(&q->lock, flags); | |
4610 | } | |
1da177e4 LT |
4611 | EXPORT_SYMBOL(__wake_up); |
4612 | ||
4613 | /* | |
4614 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4615 | */ | |
7ad5b3a5 | 4616 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4617 | { |
4618 | __wake_up_common(q, mode, 1, 0, NULL); | |
4619 | } | |
4620 | ||
4621 | /** | |
67be2dd1 | 4622 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4623 | * @q: the waitqueue |
4624 | * @mode: which threads | |
4625 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4626 | * | |
4627 | * The sync wakeup differs that the waker knows that it will schedule | |
4628 | * away soon, so while the target thread will be woken up, it will not | |
4629 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4630 | * with each other. This can prevent needless bouncing between CPUs. | |
4631 | * | |
4632 | * On UP it can prevent extra preemption. | |
4633 | */ | |
7ad5b3a5 | 4634 | void |
95cdf3b7 | 4635 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4636 | { |
4637 | unsigned long flags; | |
4638 | int sync = 1; | |
4639 | ||
4640 | if (unlikely(!q)) | |
4641 | return; | |
4642 | ||
4643 | if (unlikely(!nr_exclusive)) | |
4644 | sync = 0; | |
4645 | ||
4646 | spin_lock_irqsave(&q->lock, flags); | |
4647 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4648 | spin_unlock_irqrestore(&q->lock, flags); | |
4649 | } | |
4650 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4651 | ||
65eb3dc6 KD |
4652 | /** |
4653 | * complete: - signals a single thread waiting on this completion | |
4654 | * @x: holds the state of this particular completion | |
4655 | * | |
4656 | * This will wake up a single thread waiting on this completion. Threads will be | |
4657 | * awakened in the same order in which they were queued. | |
4658 | * | |
4659 | * See also complete_all(), wait_for_completion() and related routines. | |
4660 | */ | |
b15136e9 | 4661 | void complete(struct completion *x) |
1da177e4 LT |
4662 | { |
4663 | unsigned long flags; | |
4664 | ||
4665 | spin_lock_irqsave(&x->wait.lock, flags); | |
4666 | x->done++; | |
d9514f6c | 4667 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4668 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4669 | } | |
4670 | EXPORT_SYMBOL(complete); | |
4671 | ||
65eb3dc6 KD |
4672 | /** |
4673 | * complete_all: - signals all threads waiting on this completion | |
4674 | * @x: holds the state of this particular completion | |
4675 | * | |
4676 | * This will wake up all threads waiting on this particular completion event. | |
4677 | */ | |
b15136e9 | 4678 | void complete_all(struct completion *x) |
1da177e4 LT |
4679 | { |
4680 | unsigned long flags; | |
4681 | ||
4682 | spin_lock_irqsave(&x->wait.lock, flags); | |
4683 | x->done += UINT_MAX/2; | |
d9514f6c | 4684 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4685 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4686 | } | |
4687 | EXPORT_SYMBOL(complete_all); | |
4688 | ||
8cbbe86d AK |
4689 | static inline long __sched |
4690 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4691 | { |
1da177e4 LT |
4692 | if (!x->done) { |
4693 | DECLARE_WAITQUEUE(wait, current); | |
4694 | ||
4695 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4696 | __add_wait_queue_tail(&x->wait, &wait); | |
4697 | do { | |
94d3d824 | 4698 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4699 | timeout = -ERESTARTSYS; |
4700 | break; | |
8cbbe86d AK |
4701 | } |
4702 | __set_current_state(state); | |
1da177e4 LT |
4703 | spin_unlock_irq(&x->wait.lock); |
4704 | timeout = schedule_timeout(timeout); | |
4705 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4706 | } while (!x->done && timeout); |
1da177e4 | 4707 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4708 | if (!x->done) |
4709 | return timeout; | |
1da177e4 LT |
4710 | } |
4711 | x->done--; | |
ea71a546 | 4712 | return timeout ?: 1; |
1da177e4 | 4713 | } |
1da177e4 | 4714 | |
8cbbe86d AK |
4715 | static long __sched |
4716 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4717 | { |
1da177e4 LT |
4718 | might_sleep(); |
4719 | ||
4720 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4721 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4722 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4723 | return timeout; |
4724 | } | |
1da177e4 | 4725 | |
65eb3dc6 KD |
4726 | /** |
4727 | * wait_for_completion: - waits for completion of a task | |
4728 | * @x: holds the state of this particular completion | |
4729 | * | |
4730 | * This waits to be signaled for completion of a specific task. It is NOT | |
4731 | * interruptible and there is no timeout. | |
4732 | * | |
4733 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4734 | * and interrupt capability. Also see complete(). | |
4735 | */ | |
b15136e9 | 4736 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4737 | { |
4738 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4739 | } |
8cbbe86d | 4740 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4741 | |
65eb3dc6 KD |
4742 | /** |
4743 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4744 | * @x: holds the state of this particular completion | |
4745 | * @timeout: timeout value in jiffies | |
4746 | * | |
4747 | * This waits for either a completion of a specific task to be signaled or for a | |
4748 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4749 | * interruptible. | |
4750 | */ | |
b15136e9 | 4751 | unsigned long __sched |
8cbbe86d | 4752 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4753 | { |
8cbbe86d | 4754 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4755 | } |
8cbbe86d | 4756 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4757 | |
65eb3dc6 KD |
4758 | /** |
4759 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4760 | * @x: holds the state of this particular completion | |
4761 | * | |
4762 | * This waits for completion of a specific task to be signaled. It is | |
4763 | * interruptible. | |
4764 | */ | |
8cbbe86d | 4765 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4766 | { |
51e97990 AK |
4767 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4768 | if (t == -ERESTARTSYS) | |
4769 | return t; | |
4770 | return 0; | |
0fec171c | 4771 | } |
8cbbe86d | 4772 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4773 | |
65eb3dc6 KD |
4774 | /** |
4775 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4776 | * @x: holds the state of this particular completion | |
4777 | * @timeout: timeout value in jiffies | |
4778 | * | |
4779 | * This waits for either a completion of a specific task to be signaled or for a | |
4780 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4781 | */ | |
b15136e9 | 4782 | unsigned long __sched |
8cbbe86d AK |
4783 | wait_for_completion_interruptible_timeout(struct completion *x, |
4784 | unsigned long timeout) | |
0fec171c | 4785 | { |
8cbbe86d | 4786 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4787 | } |
8cbbe86d | 4788 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4789 | |
65eb3dc6 KD |
4790 | /** |
4791 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4792 | * @x: holds the state of this particular completion | |
4793 | * | |
4794 | * This waits to be signaled for completion of a specific task. It can be | |
4795 | * interrupted by a kill signal. | |
4796 | */ | |
009e577e MW |
4797 | int __sched wait_for_completion_killable(struct completion *x) |
4798 | { | |
4799 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4800 | if (t == -ERESTARTSYS) | |
4801 | return t; | |
4802 | return 0; | |
4803 | } | |
4804 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4805 | ||
be4de352 DC |
4806 | /** |
4807 | * try_wait_for_completion - try to decrement a completion without blocking | |
4808 | * @x: completion structure | |
4809 | * | |
4810 | * Returns: 0 if a decrement cannot be done without blocking | |
4811 | * 1 if a decrement succeeded. | |
4812 | * | |
4813 | * If a completion is being used as a counting completion, | |
4814 | * attempt to decrement the counter without blocking. This | |
4815 | * enables us to avoid waiting if the resource the completion | |
4816 | * is protecting is not available. | |
4817 | */ | |
4818 | bool try_wait_for_completion(struct completion *x) | |
4819 | { | |
4820 | int ret = 1; | |
4821 | ||
4822 | spin_lock_irq(&x->wait.lock); | |
4823 | if (!x->done) | |
4824 | ret = 0; | |
4825 | else | |
4826 | x->done--; | |
4827 | spin_unlock_irq(&x->wait.lock); | |
4828 | return ret; | |
4829 | } | |
4830 | EXPORT_SYMBOL(try_wait_for_completion); | |
4831 | ||
4832 | /** | |
4833 | * completion_done - Test to see if a completion has any waiters | |
4834 | * @x: completion structure | |
4835 | * | |
4836 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4837 | * 1 if there are no waiters. | |
4838 | * | |
4839 | */ | |
4840 | bool completion_done(struct completion *x) | |
4841 | { | |
4842 | int ret = 1; | |
4843 | ||
4844 | spin_lock_irq(&x->wait.lock); | |
4845 | if (!x->done) | |
4846 | ret = 0; | |
4847 | spin_unlock_irq(&x->wait.lock); | |
4848 | return ret; | |
4849 | } | |
4850 | EXPORT_SYMBOL(completion_done); | |
4851 | ||
8cbbe86d AK |
4852 | static long __sched |
4853 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4854 | { |
0fec171c IM |
4855 | unsigned long flags; |
4856 | wait_queue_t wait; | |
4857 | ||
4858 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4859 | |
8cbbe86d | 4860 | __set_current_state(state); |
1da177e4 | 4861 | |
8cbbe86d AK |
4862 | spin_lock_irqsave(&q->lock, flags); |
4863 | __add_wait_queue(q, &wait); | |
4864 | spin_unlock(&q->lock); | |
4865 | timeout = schedule_timeout(timeout); | |
4866 | spin_lock_irq(&q->lock); | |
4867 | __remove_wait_queue(q, &wait); | |
4868 | spin_unlock_irqrestore(&q->lock, flags); | |
4869 | ||
4870 | return timeout; | |
4871 | } | |
4872 | ||
4873 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4874 | { | |
4875 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4876 | } |
1da177e4 LT |
4877 | EXPORT_SYMBOL(interruptible_sleep_on); |
4878 | ||
0fec171c | 4879 | long __sched |
95cdf3b7 | 4880 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4881 | { |
8cbbe86d | 4882 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4883 | } |
1da177e4 LT |
4884 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4885 | ||
0fec171c | 4886 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4887 | { |
8cbbe86d | 4888 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4889 | } |
1da177e4 LT |
4890 | EXPORT_SYMBOL(sleep_on); |
4891 | ||
0fec171c | 4892 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4893 | { |
8cbbe86d | 4894 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4895 | } |
1da177e4 LT |
4896 | EXPORT_SYMBOL(sleep_on_timeout); |
4897 | ||
b29739f9 IM |
4898 | #ifdef CONFIG_RT_MUTEXES |
4899 | ||
4900 | /* | |
4901 | * rt_mutex_setprio - set the current priority of a task | |
4902 | * @p: task | |
4903 | * @prio: prio value (kernel-internal form) | |
4904 | * | |
4905 | * This function changes the 'effective' priority of a task. It does | |
4906 | * not touch ->normal_prio like __setscheduler(). | |
4907 | * | |
4908 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4909 | */ | |
36c8b586 | 4910 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4911 | { |
4912 | unsigned long flags; | |
83b699ed | 4913 | int oldprio, on_rq, running; |
70b97a7f | 4914 | struct rq *rq; |
cb469845 | 4915 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4916 | |
4917 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4918 | ||
4919 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4920 | update_rq_clock(rq); |
b29739f9 | 4921 | |
d5f9f942 | 4922 | oldprio = p->prio; |
dd41f596 | 4923 | on_rq = p->se.on_rq; |
051a1d1a | 4924 | running = task_current(rq, p); |
0e1f3483 | 4925 | if (on_rq) |
69be72c1 | 4926 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4927 | if (running) |
4928 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4929 | |
4930 | if (rt_prio(prio)) | |
4931 | p->sched_class = &rt_sched_class; | |
4932 | else | |
4933 | p->sched_class = &fair_sched_class; | |
4934 | ||
b29739f9 IM |
4935 | p->prio = prio; |
4936 | ||
0e1f3483 HS |
4937 | if (running) |
4938 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4939 | if (on_rq) { |
8159f87e | 4940 | enqueue_task(rq, p, 0); |
cb469845 SR |
4941 | |
4942 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4943 | } |
4944 | task_rq_unlock(rq, &flags); | |
4945 | } | |
4946 | ||
4947 | #endif | |
4948 | ||
36c8b586 | 4949 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4950 | { |
dd41f596 | 4951 | int old_prio, delta, on_rq; |
1da177e4 | 4952 | unsigned long flags; |
70b97a7f | 4953 | struct rq *rq; |
1da177e4 LT |
4954 | |
4955 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4956 | return; | |
4957 | /* | |
4958 | * We have to be careful, if called from sys_setpriority(), | |
4959 | * the task might be in the middle of scheduling on another CPU. | |
4960 | */ | |
4961 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4962 | update_rq_clock(rq); |
1da177e4 LT |
4963 | /* |
4964 | * The RT priorities are set via sched_setscheduler(), but we still | |
4965 | * allow the 'normal' nice value to be set - but as expected | |
4966 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4967 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4968 | */ |
e05606d3 | 4969 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4970 | p->static_prio = NICE_TO_PRIO(nice); |
4971 | goto out_unlock; | |
4972 | } | |
dd41f596 | 4973 | on_rq = p->se.on_rq; |
c09595f6 | 4974 | if (on_rq) |
69be72c1 | 4975 | dequeue_task(rq, p, 0); |
1da177e4 | 4976 | |
1da177e4 | 4977 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4978 | set_load_weight(p); |
b29739f9 IM |
4979 | old_prio = p->prio; |
4980 | p->prio = effective_prio(p); | |
4981 | delta = p->prio - old_prio; | |
1da177e4 | 4982 | |
dd41f596 | 4983 | if (on_rq) { |
8159f87e | 4984 | enqueue_task(rq, p, 0); |
1da177e4 | 4985 | /* |
d5f9f942 AM |
4986 | * If the task increased its priority or is running and |
4987 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4988 | */ |
d5f9f942 | 4989 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4990 | resched_task(rq->curr); |
4991 | } | |
4992 | out_unlock: | |
4993 | task_rq_unlock(rq, &flags); | |
4994 | } | |
1da177e4 LT |
4995 | EXPORT_SYMBOL(set_user_nice); |
4996 | ||
e43379f1 MM |
4997 | /* |
4998 | * can_nice - check if a task can reduce its nice value | |
4999 | * @p: task | |
5000 | * @nice: nice value | |
5001 | */ | |
36c8b586 | 5002 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5003 | { |
024f4747 MM |
5004 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5005 | int nice_rlim = 20 - nice; | |
48f24c4d | 5006 | |
e43379f1 MM |
5007 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5008 | capable(CAP_SYS_NICE)); | |
5009 | } | |
5010 | ||
1da177e4 LT |
5011 | #ifdef __ARCH_WANT_SYS_NICE |
5012 | ||
5013 | /* | |
5014 | * sys_nice - change the priority of the current process. | |
5015 | * @increment: priority increment | |
5016 | * | |
5017 | * sys_setpriority is a more generic, but much slower function that | |
5018 | * does similar things. | |
5019 | */ | |
5020 | asmlinkage long sys_nice(int increment) | |
5021 | { | |
48f24c4d | 5022 | long nice, retval; |
1da177e4 LT |
5023 | |
5024 | /* | |
5025 | * Setpriority might change our priority at the same moment. | |
5026 | * We don't have to worry. Conceptually one call occurs first | |
5027 | * and we have a single winner. | |
5028 | */ | |
e43379f1 MM |
5029 | if (increment < -40) |
5030 | increment = -40; | |
1da177e4 LT |
5031 | if (increment > 40) |
5032 | increment = 40; | |
5033 | ||
5034 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5035 | if (nice < -20) | |
5036 | nice = -20; | |
5037 | if (nice > 19) | |
5038 | nice = 19; | |
5039 | ||
e43379f1 MM |
5040 | if (increment < 0 && !can_nice(current, nice)) |
5041 | return -EPERM; | |
5042 | ||
1da177e4 LT |
5043 | retval = security_task_setnice(current, nice); |
5044 | if (retval) | |
5045 | return retval; | |
5046 | ||
5047 | set_user_nice(current, nice); | |
5048 | return 0; | |
5049 | } | |
5050 | ||
5051 | #endif | |
5052 | ||
5053 | /** | |
5054 | * task_prio - return the priority value of a given task. | |
5055 | * @p: the task in question. | |
5056 | * | |
5057 | * This is the priority value as seen by users in /proc. | |
5058 | * RT tasks are offset by -200. Normal tasks are centered | |
5059 | * around 0, value goes from -16 to +15. | |
5060 | */ | |
36c8b586 | 5061 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5062 | { |
5063 | return p->prio - MAX_RT_PRIO; | |
5064 | } | |
5065 | ||
5066 | /** | |
5067 | * task_nice - return the nice value of a given task. | |
5068 | * @p: the task in question. | |
5069 | */ | |
36c8b586 | 5070 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5071 | { |
5072 | return TASK_NICE(p); | |
5073 | } | |
150d8bed | 5074 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5075 | |
5076 | /** | |
5077 | * idle_cpu - is a given cpu idle currently? | |
5078 | * @cpu: the processor in question. | |
5079 | */ | |
5080 | int idle_cpu(int cpu) | |
5081 | { | |
5082 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5083 | } | |
5084 | ||
1da177e4 LT |
5085 | /** |
5086 | * idle_task - return the idle task for a given cpu. | |
5087 | * @cpu: the processor in question. | |
5088 | */ | |
36c8b586 | 5089 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5090 | { |
5091 | return cpu_rq(cpu)->idle; | |
5092 | } | |
5093 | ||
5094 | /** | |
5095 | * find_process_by_pid - find a process with a matching PID value. | |
5096 | * @pid: the pid in question. | |
5097 | */ | |
a9957449 | 5098 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5099 | { |
228ebcbe | 5100 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5101 | } |
5102 | ||
5103 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5104 | static void |
5105 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5106 | { |
dd41f596 | 5107 | BUG_ON(p->se.on_rq); |
48f24c4d | 5108 | |
1da177e4 | 5109 | p->policy = policy; |
dd41f596 IM |
5110 | switch (p->policy) { |
5111 | case SCHED_NORMAL: | |
5112 | case SCHED_BATCH: | |
5113 | case SCHED_IDLE: | |
5114 | p->sched_class = &fair_sched_class; | |
5115 | break; | |
5116 | case SCHED_FIFO: | |
5117 | case SCHED_RR: | |
5118 | p->sched_class = &rt_sched_class; | |
5119 | break; | |
5120 | } | |
5121 | ||
1da177e4 | 5122 | p->rt_priority = prio; |
b29739f9 IM |
5123 | p->normal_prio = normal_prio(p); |
5124 | /* we are holding p->pi_lock already */ | |
5125 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5126 | set_load_weight(p); |
1da177e4 LT |
5127 | } |
5128 | ||
961ccddd RR |
5129 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5130 | struct sched_param *param, bool user) | |
1da177e4 | 5131 | { |
83b699ed | 5132 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5133 | unsigned long flags; |
cb469845 | 5134 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5135 | struct rq *rq; |
1da177e4 | 5136 | |
66e5393a SR |
5137 | /* may grab non-irq protected spin_locks */ |
5138 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5139 | recheck: |
5140 | /* double check policy once rq lock held */ | |
5141 | if (policy < 0) | |
5142 | policy = oldpolicy = p->policy; | |
5143 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5144 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5145 | policy != SCHED_IDLE) | |
b0a9499c | 5146 | return -EINVAL; |
1da177e4 LT |
5147 | /* |
5148 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5149 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5150 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5151 | */ |
5152 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5153 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5154 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5155 | return -EINVAL; |
e05606d3 | 5156 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5157 | return -EINVAL; |
5158 | ||
37e4ab3f OC |
5159 | /* |
5160 | * Allow unprivileged RT tasks to decrease priority: | |
5161 | */ | |
961ccddd | 5162 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5163 | if (rt_policy(policy)) { |
8dc3e909 | 5164 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5165 | |
5166 | if (!lock_task_sighand(p, &flags)) | |
5167 | return -ESRCH; | |
5168 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5169 | unlock_task_sighand(p, &flags); | |
5170 | ||
5171 | /* can't set/change the rt policy */ | |
5172 | if (policy != p->policy && !rlim_rtprio) | |
5173 | return -EPERM; | |
5174 | ||
5175 | /* can't increase priority */ | |
5176 | if (param->sched_priority > p->rt_priority && | |
5177 | param->sched_priority > rlim_rtprio) | |
5178 | return -EPERM; | |
5179 | } | |
dd41f596 IM |
5180 | /* |
5181 | * Like positive nice levels, dont allow tasks to | |
5182 | * move out of SCHED_IDLE either: | |
5183 | */ | |
5184 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5185 | return -EPERM; | |
5fe1d75f | 5186 | |
37e4ab3f OC |
5187 | /* can't change other user's priorities */ |
5188 | if ((current->euid != p->euid) && | |
5189 | (current->euid != p->uid)) | |
5190 | return -EPERM; | |
5191 | } | |
1da177e4 | 5192 | |
725aad24 | 5193 | if (user) { |
b68aa230 | 5194 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5195 | /* |
5196 | * Do not allow realtime tasks into groups that have no runtime | |
5197 | * assigned. | |
5198 | */ | |
9a7e0b18 PZ |
5199 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5200 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5201 | return -EPERM; |
b68aa230 PZ |
5202 | #endif |
5203 | ||
725aad24 JF |
5204 | retval = security_task_setscheduler(p, policy, param); |
5205 | if (retval) | |
5206 | return retval; | |
5207 | } | |
5208 | ||
b29739f9 IM |
5209 | /* |
5210 | * make sure no PI-waiters arrive (or leave) while we are | |
5211 | * changing the priority of the task: | |
5212 | */ | |
5213 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5214 | /* |
5215 | * To be able to change p->policy safely, the apropriate | |
5216 | * runqueue lock must be held. | |
5217 | */ | |
b29739f9 | 5218 | rq = __task_rq_lock(p); |
1da177e4 LT |
5219 | /* recheck policy now with rq lock held */ |
5220 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5221 | policy = oldpolicy = -1; | |
b29739f9 IM |
5222 | __task_rq_unlock(rq); |
5223 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5224 | goto recheck; |
5225 | } | |
2daa3577 | 5226 | update_rq_clock(rq); |
dd41f596 | 5227 | on_rq = p->se.on_rq; |
051a1d1a | 5228 | running = task_current(rq, p); |
0e1f3483 | 5229 | if (on_rq) |
2e1cb74a | 5230 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5231 | if (running) |
5232 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5233 | |
1da177e4 | 5234 | oldprio = p->prio; |
dd41f596 | 5235 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5236 | |
0e1f3483 HS |
5237 | if (running) |
5238 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5239 | if (on_rq) { |
5240 | activate_task(rq, p, 0); | |
cb469845 SR |
5241 | |
5242 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5243 | } |
b29739f9 IM |
5244 | __task_rq_unlock(rq); |
5245 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5246 | ||
95e02ca9 TG |
5247 | rt_mutex_adjust_pi(p); |
5248 | ||
1da177e4 LT |
5249 | return 0; |
5250 | } | |
961ccddd RR |
5251 | |
5252 | /** | |
5253 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5254 | * @p: the task in question. | |
5255 | * @policy: new policy. | |
5256 | * @param: structure containing the new RT priority. | |
5257 | * | |
5258 | * NOTE that the task may be already dead. | |
5259 | */ | |
5260 | int sched_setscheduler(struct task_struct *p, int policy, | |
5261 | struct sched_param *param) | |
5262 | { | |
5263 | return __sched_setscheduler(p, policy, param, true); | |
5264 | } | |
1da177e4 LT |
5265 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5266 | ||
961ccddd RR |
5267 | /** |
5268 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5269 | * @p: the task in question. | |
5270 | * @policy: new policy. | |
5271 | * @param: structure containing the new RT priority. | |
5272 | * | |
5273 | * Just like sched_setscheduler, only don't bother checking if the | |
5274 | * current context has permission. For example, this is needed in | |
5275 | * stop_machine(): we create temporary high priority worker threads, | |
5276 | * but our caller might not have that capability. | |
5277 | */ | |
5278 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5279 | struct sched_param *param) | |
5280 | { | |
5281 | return __sched_setscheduler(p, policy, param, false); | |
5282 | } | |
5283 | ||
95cdf3b7 IM |
5284 | static int |
5285 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5286 | { |
1da177e4 LT |
5287 | struct sched_param lparam; |
5288 | struct task_struct *p; | |
36c8b586 | 5289 | int retval; |
1da177e4 LT |
5290 | |
5291 | if (!param || pid < 0) | |
5292 | return -EINVAL; | |
5293 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5294 | return -EFAULT; | |
5fe1d75f ON |
5295 | |
5296 | rcu_read_lock(); | |
5297 | retval = -ESRCH; | |
1da177e4 | 5298 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5299 | if (p != NULL) |
5300 | retval = sched_setscheduler(p, policy, &lparam); | |
5301 | rcu_read_unlock(); | |
36c8b586 | 5302 | |
1da177e4 LT |
5303 | return retval; |
5304 | } | |
5305 | ||
5306 | /** | |
5307 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5308 | * @pid: the pid in question. | |
5309 | * @policy: new policy. | |
5310 | * @param: structure containing the new RT priority. | |
5311 | */ | |
41a2d6cf IM |
5312 | asmlinkage long |
5313 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5314 | { |
c21761f1 JB |
5315 | /* negative values for policy are not valid */ |
5316 | if (policy < 0) | |
5317 | return -EINVAL; | |
5318 | ||
1da177e4 LT |
5319 | return do_sched_setscheduler(pid, policy, param); |
5320 | } | |
5321 | ||
5322 | /** | |
5323 | * sys_sched_setparam - set/change the RT priority of a thread | |
5324 | * @pid: the pid in question. | |
5325 | * @param: structure containing the new RT priority. | |
5326 | */ | |
5327 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5328 | { | |
5329 | return do_sched_setscheduler(pid, -1, param); | |
5330 | } | |
5331 | ||
5332 | /** | |
5333 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5334 | * @pid: the pid in question. | |
5335 | */ | |
5336 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5337 | { | |
36c8b586 | 5338 | struct task_struct *p; |
3a5c359a | 5339 | int retval; |
1da177e4 LT |
5340 | |
5341 | if (pid < 0) | |
3a5c359a | 5342 | return -EINVAL; |
1da177e4 LT |
5343 | |
5344 | retval = -ESRCH; | |
5345 | read_lock(&tasklist_lock); | |
5346 | p = find_process_by_pid(pid); | |
5347 | if (p) { | |
5348 | retval = security_task_getscheduler(p); | |
5349 | if (!retval) | |
5350 | retval = p->policy; | |
5351 | } | |
5352 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5353 | return retval; |
5354 | } | |
5355 | ||
5356 | /** | |
5357 | * sys_sched_getscheduler - get the RT priority of a thread | |
5358 | * @pid: the pid in question. | |
5359 | * @param: structure containing the RT priority. | |
5360 | */ | |
5361 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5362 | { | |
5363 | struct sched_param lp; | |
36c8b586 | 5364 | struct task_struct *p; |
3a5c359a | 5365 | int retval; |
1da177e4 LT |
5366 | |
5367 | if (!param || pid < 0) | |
3a5c359a | 5368 | return -EINVAL; |
1da177e4 LT |
5369 | |
5370 | read_lock(&tasklist_lock); | |
5371 | p = find_process_by_pid(pid); | |
5372 | retval = -ESRCH; | |
5373 | if (!p) | |
5374 | goto out_unlock; | |
5375 | ||
5376 | retval = security_task_getscheduler(p); | |
5377 | if (retval) | |
5378 | goto out_unlock; | |
5379 | ||
5380 | lp.sched_priority = p->rt_priority; | |
5381 | read_unlock(&tasklist_lock); | |
5382 | ||
5383 | /* | |
5384 | * This one might sleep, we cannot do it with a spinlock held ... | |
5385 | */ | |
5386 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5387 | ||
1da177e4 LT |
5388 | return retval; |
5389 | ||
5390 | out_unlock: | |
5391 | read_unlock(&tasklist_lock); | |
5392 | return retval; | |
5393 | } | |
5394 | ||
b53e921b | 5395 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5396 | { |
1da177e4 | 5397 | cpumask_t cpus_allowed; |
b53e921b | 5398 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5399 | struct task_struct *p; |
5400 | int retval; | |
1da177e4 | 5401 | |
95402b38 | 5402 | get_online_cpus(); |
1da177e4 LT |
5403 | read_lock(&tasklist_lock); |
5404 | ||
5405 | p = find_process_by_pid(pid); | |
5406 | if (!p) { | |
5407 | read_unlock(&tasklist_lock); | |
95402b38 | 5408 | put_online_cpus(); |
1da177e4 LT |
5409 | return -ESRCH; |
5410 | } | |
5411 | ||
5412 | /* | |
5413 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5414 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5415 | * usage count and then drop tasklist_lock. |
5416 | */ | |
5417 | get_task_struct(p); | |
5418 | read_unlock(&tasklist_lock); | |
5419 | ||
5420 | retval = -EPERM; | |
5421 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5422 | !capable(CAP_SYS_NICE)) | |
5423 | goto out_unlock; | |
5424 | ||
e7834f8f DQ |
5425 | retval = security_task_setscheduler(p, 0, NULL); |
5426 | if (retval) | |
5427 | goto out_unlock; | |
5428 | ||
f9a86fcb | 5429 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5430 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5431 | again: |
7c16ec58 | 5432 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5433 | |
8707d8b8 | 5434 | if (!retval) { |
f9a86fcb | 5435 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5436 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5437 | /* | |
5438 | * We must have raced with a concurrent cpuset | |
5439 | * update. Just reset the cpus_allowed to the | |
5440 | * cpuset's cpus_allowed | |
5441 | */ | |
5442 | new_mask = cpus_allowed; | |
5443 | goto again; | |
5444 | } | |
5445 | } | |
1da177e4 LT |
5446 | out_unlock: |
5447 | put_task_struct(p); | |
95402b38 | 5448 | put_online_cpus(); |
1da177e4 LT |
5449 | return retval; |
5450 | } | |
5451 | ||
5452 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5453 | cpumask_t *new_mask) | |
5454 | { | |
5455 | if (len < sizeof(cpumask_t)) { | |
5456 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5457 | } else if (len > sizeof(cpumask_t)) { | |
5458 | len = sizeof(cpumask_t); | |
5459 | } | |
5460 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5461 | } | |
5462 | ||
5463 | /** | |
5464 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5465 | * @pid: pid of the process | |
5466 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5467 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5468 | */ | |
5469 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5470 | unsigned long __user *user_mask_ptr) | |
5471 | { | |
5472 | cpumask_t new_mask; | |
5473 | int retval; | |
5474 | ||
5475 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5476 | if (retval) | |
5477 | return retval; | |
5478 | ||
b53e921b | 5479 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5480 | } |
5481 | ||
1da177e4 LT |
5482 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5483 | { | |
36c8b586 | 5484 | struct task_struct *p; |
1da177e4 | 5485 | int retval; |
1da177e4 | 5486 | |
95402b38 | 5487 | get_online_cpus(); |
1da177e4 LT |
5488 | read_lock(&tasklist_lock); |
5489 | ||
5490 | retval = -ESRCH; | |
5491 | p = find_process_by_pid(pid); | |
5492 | if (!p) | |
5493 | goto out_unlock; | |
5494 | ||
e7834f8f DQ |
5495 | retval = security_task_getscheduler(p); |
5496 | if (retval) | |
5497 | goto out_unlock; | |
5498 | ||
2f7016d9 | 5499 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5500 | |
5501 | out_unlock: | |
5502 | read_unlock(&tasklist_lock); | |
95402b38 | 5503 | put_online_cpus(); |
1da177e4 | 5504 | |
9531b62f | 5505 | return retval; |
1da177e4 LT |
5506 | } |
5507 | ||
5508 | /** | |
5509 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5510 | * @pid: pid of the process | |
5511 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5512 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5513 | */ | |
5514 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5515 | unsigned long __user *user_mask_ptr) | |
5516 | { | |
5517 | int ret; | |
5518 | cpumask_t mask; | |
5519 | ||
5520 | if (len < sizeof(cpumask_t)) | |
5521 | return -EINVAL; | |
5522 | ||
5523 | ret = sched_getaffinity(pid, &mask); | |
5524 | if (ret < 0) | |
5525 | return ret; | |
5526 | ||
5527 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5528 | return -EFAULT; | |
5529 | ||
5530 | return sizeof(cpumask_t); | |
5531 | } | |
5532 | ||
5533 | /** | |
5534 | * sys_sched_yield - yield the current processor to other threads. | |
5535 | * | |
dd41f596 IM |
5536 | * This function yields the current CPU to other tasks. If there are no |
5537 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5538 | */ |
5539 | asmlinkage long sys_sched_yield(void) | |
5540 | { | |
70b97a7f | 5541 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5542 | |
2d72376b | 5543 | schedstat_inc(rq, yld_count); |
4530d7ab | 5544 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5545 | |
5546 | /* | |
5547 | * Since we are going to call schedule() anyway, there's | |
5548 | * no need to preempt or enable interrupts: | |
5549 | */ | |
5550 | __release(rq->lock); | |
8a25d5de | 5551 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5552 | _raw_spin_unlock(&rq->lock); |
5553 | preempt_enable_no_resched(); | |
5554 | ||
5555 | schedule(); | |
5556 | ||
5557 | return 0; | |
5558 | } | |
5559 | ||
e7b38404 | 5560 | static void __cond_resched(void) |
1da177e4 | 5561 | { |
8e0a43d8 IM |
5562 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5563 | __might_sleep(__FILE__, __LINE__); | |
5564 | #endif | |
5bbcfd90 IM |
5565 | /* |
5566 | * The BKS might be reacquired before we have dropped | |
5567 | * PREEMPT_ACTIVE, which could trigger a second | |
5568 | * cond_resched() call. | |
5569 | */ | |
1da177e4 LT |
5570 | do { |
5571 | add_preempt_count(PREEMPT_ACTIVE); | |
5572 | schedule(); | |
5573 | sub_preempt_count(PREEMPT_ACTIVE); | |
5574 | } while (need_resched()); | |
5575 | } | |
5576 | ||
02b67cc3 | 5577 | int __sched _cond_resched(void) |
1da177e4 | 5578 | { |
9414232f IM |
5579 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5580 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5581 | __cond_resched(); |
5582 | return 1; | |
5583 | } | |
5584 | return 0; | |
5585 | } | |
02b67cc3 | 5586 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5587 | |
5588 | /* | |
5589 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5590 | * call schedule, and on return reacquire the lock. | |
5591 | * | |
41a2d6cf | 5592 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5593 | * operations here to prevent schedule() from being called twice (once via |
5594 | * spin_unlock(), once by hand). | |
5595 | */ | |
95cdf3b7 | 5596 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5597 | { |
95c354fe | 5598 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5599 | int ret = 0; |
5600 | ||
95c354fe | 5601 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5602 | spin_unlock(lock); |
95c354fe NP |
5603 | if (resched && need_resched()) |
5604 | __cond_resched(); | |
5605 | else | |
5606 | cpu_relax(); | |
6df3cecb | 5607 | ret = 1; |
1da177e4 | 5608 | spin_lock(lock); |
1da177e4 | 5609 | } |
6df3cecb | 5610 | return ret; |
1da177e4 | 5611 | } |
1da177e4 LT |
5612 | EXPORT_SYMBOL(cond_resched_lock); |
5613 | ||
5614 | int __sched cond_resched_softirq(void) | |
5615 | { | |
5616 | BUG_ON(!in_softirq()); | |
5617 | ||
9414232f | 5618 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5619 | local_bh_enable(); |
1da177e4 LT |
5620 | __cond_resched(); |
5621 | local_bh_disable(); | |
5622 | return 1; | |
5623 | } | |
5624 | return 0; | |
5625 | } | |
1da177e4 LT |
5626 | EXPORT_SYMBOL(cond_resched_softirq); |
5627 | ||
1da177e4 LT |
5628 | /** |
5629 | * yield - yield the current processor to other threads. | |
5630 | * | |
72fd4a35 | 5631 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5632 | * thread runnable and calls sys_sched_yield(). |
5633 | */ | |
5634 | void __sched yield(void) | |
5635 | { | |
5636 | set_current_state(TASK_RUNNING); | |
5637 | sys_sched_yield(); | |
5638 | } | |
1da177e4 LT |
5639 | EXPORT_SYMBOL(yield); |
5640 | ||
5641 | /* | |
41a2d6cf | 5642 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5643 | * that process accounting knows that this is a task in IO wait state. |
5644 | * | |
5645 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5646 | * has set its backing_dev_info: the queue against which it should throttle) | |
5647 | */ | |
5648 | void __sched io_schedule(void) | |
5649 | { | |
70b97a7f | 5650 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5651 | |
0ff92245 | 5652 | delayacct_blkio_start(); |
1da177e4 LT |
5653 | atomic_inc(&rq->nr_iowait); |
5654 | schedule(); | |
5655 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5656 | delayacct_blkio_end(); |
1da177e4 | 5657 | } |
1da177e4 LT |
5658 | EXPORT_SYMBOL(io_schedule); |
5659 | ||
5660 | long __sched io_schedule_timeout(long timeout) | |
5661 | { | |
70b97a7f | 5662 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5663 | long ret; |
5664 | ||
0ff92245 | 5665 | delayacct_blkio_start(); |
1da177e4 LT |
5666 | atomic_inc(&rq->nr_iowait); |
5667 | ret = schedule_timeout(timeout); | |
5668 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5669 | delayacct_blkio_end(); |
1da177e4 LT |
5670 | return ret; |
5671 | } | |
5672 | ||
5673 | /** | |
5674 | * sys_sched_get_priority_max - return maximum RT priority. | |
5675 | * @policy: scheduling class. | |
5676 | * | |
5677 | * this syscall returns the maximum rt_priority that can be used | |
5678 | * by a given scheduling class. | |
5679 | */ | |
5680 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5681 | { | |
5682 | int ret = -EINVAL; | |
5683 | ||
5684 | switch (policy) { | |
5685 | case SCHED_FIFO: | |
5686 | case SCHED_RR: | |
5687 | ret = MAX_USER_RT_PRIO-1; | |
5688 | break; | |
5689 | case SCHED_NORMAL: | |
b0a9499c | 5690 | case SCHED_BATCH: |
dd41f596 | 5691 | case SCHED_IDLE: |
1da177e4 LT |
5692 | ret = 0; |
5693 | break; | |
5694 | } | |
5695 | return ret; | |
5696 | } | |
5697 | ||
5698 | /** | |
5699 | * sys_sched_get_priority_min - return minimum RT priority. | |
5700 | * @policy: scheduling class. | |
5701 | * | |
5702 | * this syscall returns the minimum rt_priority that can be used | |
5703 | * by a given scheduling class. | |
5704 | */ | |
5705 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5706 | { | |
5707 | int ret = -EINVAL; | |
5708 | ||
5709 | switch (policy) { | |
5710 | case SCHED_FIFO: | |
5711 | case SCHED_RR: | |
5712 | ret = 1; | |
5713 | break; | |
5714 | case SCHED_NORMAL: | |
b0a9499c | 5715 | case SCHED_BATCH: |
dd41f596 | 5716 | case SCHED_IDLE: |
1da177e4 LT |
5717 | ret = 0; |
5718 | } | |
5719 | return ret; | |
5720 | } | |
5721 | ||
5722 | /** | |
5723 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5724 | * @pid: pid of the process. | |
5725 | * @interval: userspace pointer to the timeslice value. | |
5726 | * | |
5727 | * this syscall writes the default timeslice value of a given process | |
5728 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5729 | */ | |
5730 | asmlinkage | |
5731 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5732 | { | |
36c8b586 | 5733 | struct task_struct *p; |
a4ec24b4 | 5734 | unsigned int time_slice; |
3a5c359a | 5735 | int retval; |
1da177e4 | 5736 | struct timespec t; |
1da177e4 LT |
5737 | |
5738 | if (pid < 0) | |
3a5c359a | 5739 | return -EINVAL; |
1da177e4 LT |
5740 | |
5741 | retval = -ESRCH; | |
5742 | read_lock(&tasklist_lock); | |
5743 | p = find_process_by_pid(pid); | |
5744 | if (!p) | |
5745 | goto out_unlock; | |
5746 | ||
5747 | retval = security_task_getscheduler(p); | |
5748 | if (retval) | |
5749 | goto out_unlock; | |
5750 | ||
77034937 IM |
5751 | /* |
5752 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5753 | * tasks that are on an otherwise idle runqueue: | |
5754 | */ | |
5755 | time_slice = 0; | |
5756 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5757 | time_slice = DEF_TIMESLICE; |
1868f958 | 5758 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5759 | struct sched_entity *se = &p->se; |
5760 | unsigned long flags; | |
5761 | struct rq *rq; | |
5762 | ||
5763 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5764 | if (rq->cfs.load.weight) |
5765 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5766 | task_rq_unlock(rq, &flags); |
5767 | } | |
1da177e4 | 5768 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5769 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5770 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5771 | return retval; |
3a5c359a | 5772 | |
1da177e4 LT |
5773 | out_unlock: |
5774 | read_unlock(&tasklist_lock); | |
5775 | return retval; | |
5776 | } | |
5777 | ||
7c731e0a | 5778 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5779 | |
82a1fcb9 | 5780 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5781 | { |
1da177e4 | 5782 | unsigned long free = 0; |
36c8b586 | 5783 | unsigned state; |
1da177e4 | 5784 | |
1da177e4 | 5785 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5786 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5787 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5788 | #if BITS_PER_LONG == 32 |
1da177e4 | 5789 | if (state == TASK_RUNNING) |
cc4ea795 | 5790 | printk(KERN_CONT " running "); |
1da177e4 | 5791 | else |
cc4ea795 | 5792 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5793 | #else |
5794 | if (state == TASK_RUNNING) | |
cc4ea795 | 5795 | printk(KERN_CONT " running task "); |
1da177e4 | 5796 | else |
cc4ea795 | 5797 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5798 | #endif |
5799 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5800 | { | |
10ebffde | 5801 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5802 | while (!*n) |
5803 | n++; | |
10ebffde | 5804 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5805 | } |
5806 | #endif | |
ba25f9dc | 5807 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5808 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5809 | |
5fb5e6de | 5810 | show_stack(p, NULL); |
1da177e4 LT |
5811 | } |
5812 | ||
e59e2ae2 | 5813 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5814 | { |
36c8b586 | 5815 | struct task_struct *g, *p; |
1da177e4 | 5816 | |
4bd77321 IM |
5817 | #if BITS_PER_LONG == 32 |
5818 | printk(KERN_INFO | |
5819 | " task PC stack pid father\n"); | |
1da177e4 | 5820 | #else |
4bd77321 IM |
5821 | printk(KERN_INFO |
5822 | " task PC stack pid father\n"); | |
1da177e4 LT |
5823 | #endif |
5824 | read_lock(&tasklist_lock); | |
5825 | do_each_thread(g, p) { | |
5826 | /* | |
5827 | * reset the NMI-timeout, listing all files on a slow | |
5828 | * console might take alot of time: | |
5829 | */ | |
5830 | touch_nmi_watchdog(); | |
39bc89fd | 5831 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5832 | sched_show_task(p); |
1da177e4 LT |
5833 | } while_each_thread(g, p); |
5834 | ||
04c9167f JF |
5835 | touch_all_softlockup_watchdogs(); |
5836 | ||
dd41f596 IM |
5837 | #ifdef CONFIG_SCHED_DEBUG |
5838 | sysrq_sched_debug_show(); | |
5839 | #endif | |
1da177e4 | 5840 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5841 | /* |
5842 | * Only show locks if all tasks are dumped: | |
5843 | */ | |
5844 | if (state_filter == -1) | |
5845 | debug_show_all_locks(); | |
1da177e4 LT |
5846 | } |
5847 | ||
1df21055 IM |
5848 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5849 | { | |
dd41f596 | 5850 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5851 | } |
5852 | ||
f340c0d1 IM |
5853 | /** |
5854 | * init_idle - set up an idle thread for a given CPU | |
5855 | * @idle: task in question | |
5856 | * @cpu: cpu the idle task belongs to | |
5857 | * | |
5858 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5859 | * flag, to make booting more robust. | |
5860 | */ | |
5c1e1767 | 5861 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5862 | { |
70b97a7f | 5863 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5864 | unsigned long flags; |
5865 | ||
dd41f596 IM |
5866 | __sched_fork(idle); |
5867 | idle->se.exec_start = sched_clock(); | |
5868 | ||
b29739f9 | 5869 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5870 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5871 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5872 | |
5873 | spin_lock_irqsave(&rq->lock, flags); | |
5874 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5875 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5876 | idle->oncpu = 1; | |
5877 | #endif | |
1da177e4 LT |
5878 | spin_unlock_irqrestore(&rq->lock, flags); |
5879 | ||
5880 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5881 | #if defined(CONFIG_PREEMPT) |
5882 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5883 | #else | |
a1261f54 | 5884 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5885 | #endif |
dd41f596 IM |
5886 | /* |
5887 | * The idle tasks have their own, simple scheduling class: | |
5888 | */ | |
5889 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5890 | } |
5891 | ||
5892 | /* | |
5893 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5894 | * indicates which cpus entered this state. This is used | |
5895 | * in the rcu update to wait only for active cpus. For system | |
5896 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5897 | * always be CPU_MASK_NONE. | |
5898 | */ | |
5899 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5900 | ||
19978ca6 IM |
5901 | /* |
5902 | * Increase the granularity value when there are more CPUs, | |
5903 | * because with more CPUs the 'effective latency' as visible | |
5904 | * to users decreases. But the relationship is not linear, | |
5905 | * so pick a second-best guess by going with the log2 of the | |
5906 | * number of CPUs. | |
5907 | * | |
5908 | * This idea comes from the SD scheduler of Con Kolivas: | |
5909 | */ | |
5910 | static inline void sched_init_granularity(void) | |
5911 | { | |
5912 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5913 | const unsigned long limit = 200000000; | |
5914 | ||
5915 | sysctl_sched_min_granularity *= factor; | |
5916 | if (sysctl_sched_min_granularity > limit) | |
5917 | sysctl_sched_min_granularity = limit; | |
5918 | ||
5919 | sysctl_sched_latency *= factor; | |
5920 | if (sysctl_sched_latency > limit) | |
5921 | sysctl_sched_latency = limit; | |
5922 | ||
5923 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5924 | |
5925 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5926 | } |
5927 | ||
1da177e4 LT |
5928 | #ifdef CONFIG_SMP |
5929 | /* | |
5930 | * This is how migration works: | |
5931 | * | |
70b97a7f | 5932 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5933 | * runqueue and wake up that CPU's migration thread. |
5934 | * 2) we down() the locked semaphore => thread blocks. | |
5935 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5936 | * thread off the CPU) | |
5937 | * 4) it gets the migration request and checks whether the migrated | |
5938 | * task is still in the wrong runqueue. | |
5939 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5940 | * it and puts it into the right queue. | |
5941 | * 6) migration thread up()s the semaphore. | |
5942 | * 7) we wake up and the migration is done. | |
5943 | */ | |
5944 | ||
5945 | /* | |
5946 | * Change a given task's CPU affinity. Migrate the thread to a | |
5947 | * proper CPU and schedule it away if the CPU it's executing on | |
5948 | * is removed from the allowed bitmask. | |
5949 | * | |
5950 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5951 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5952 | * call is not atomic; no spinlocks may be held. |
5953 | */ | |
cd8ba7cd | 5954 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5955 | { |
70b97a7f | 5956 | struct migration_req req; |
1da177e4 | 5957 | unsigned long flags; |
70b97a7f | 5958 | struct rq *rq; |
48f24c4d | 5959 | int ret = 0; |
1da177e4 LT |
5960 | |
5961 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5962 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5963 | ret = -EINVAL; |
5964 | goto out; | |
5965 | } | |
5966 | ||
9985b0ba DR |
5967 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5968 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5969 | ret = -EINVAL; | |
5970 | goto out; | |
5971 | } | |
5972 | ||
73fe6aae | 5973 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5974 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5975 | else { |
cd8ba7cd MT |
5976 | p->cpus_allowed = *new_mask; |
5977 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5978 | } |
5979 | ||
1da177e4 | 5980 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5981 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5982 | goto out; |
5983 | ||
cd8ba7cd | 5984 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5985 | /* Need help from migration thread: drop lock and wait. */ |
5986 | task_rq_unlock(rq, &flags); | |
5987 | wake_up_process(rq->migration_thread); | |
5988 | wait_for_completion(&req.done); | |
5989 | tlb_migrate_finish(p->mm); | |
5990 | return 0; | |
5991 | } | |
5992 | out: | |
5993 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5994 | |
1da177e4 LT |
5995 | return ret; |
5996 | } | |
cd8ba7cd | 5997 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5998 | |
5999 | /* | |
41a2d6cf | 6000 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6001 | * this because either it can't run here any more (set_cpus_allowed() |
6002 | * away from this CPU, or CPU going down), or because we're | |
6003 | * attempting to rebalance this task on exec (sched_exec). | |
6004 | * | |
6005 | * So we race with normal scheduler movements, but that's OK, as long | |
6006 | * as the task is no longer on this CPU. | |
efc30814 KK |
6007 | * |
6008 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6009 | */ |
efc30814 | 6010 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6011 | { |
70b97a7f | 6012 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6013 | int ret = 0, on_rq; |
1da177e4 | 6014 | |
e761b772 | 6015 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6016 | return ret; |
1da177e4 LT |
6017 | |
6018 | rq_src = cpu_rq(src_cpu); | |
6019 | rq_dest = cpu_rq(dest_cpu); | |
6020 | ||
6021 | double_rq_lock(rq_src, rq_dest); | |
6022 | /* Already moved. */ | |
6023 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6024 | goto done; |
1da177e4 LT |
6025 | /* Affinity changed (again). */ |
6026 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 6027 | goto fail; |
1da177e4 | 6028 | |
dd41f596 | 6029 | on_rq = p->se.on_rq; |
6e82a3be | 6030 | if (on_rq) |
2e1cb74a | 6031 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6032 | |
1da177e4 | 6033 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6034 | if (on_rq) { |
6035 | activate_task(rq_dest, p, 0); | |
15afe09b | 6036 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6037 | } |
b1e38734 | 6038 | done: |
efc30814 | 6039 | ret = 1; |
b1e38734 | 6040 | fail: |
1da177e4 | 6041 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6042 | return ret; |
1da177e4 LT |
6043 | } |
6044 | ||
6045 | /* | |
6046 | * migration_thread - this is a highprio system thread that performs | |
6047 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6048 | * another runqueue. | |
6049 | */ | |
95cdf3b7 | 6050 | static int migration_thread(void *data) |
1da177e4 | 6051 | { |
1da177e4 | 6052 | int cpu = (long)data; |
70b97a7f | 6053 | struct rq *rq; |
1da177e4 LT |
6054 | |
6055 | rq = cpu_rq(cpu); | |
6056 | BUG_ON(rq->migration_thread != current); | |
6057 | ||
6058 | set_current_state(TASK_INTERRUPTIBLE); | |
6059 | while (!kthread_should_stop()) { | |
70b97a7f | 6060 | struct migration_req *req; |
1da177e4 | 6061 | struct list_head *head; |
1da177e4 | 6062 | |
1da177e4 LT |
6063 | spin_lock_irq(&rq->lock); |
6064 | ||
6065 | if (cpu_is_offline(cpu)) { | |
6066 | spin_unlock_irq(&rq->lock); | |
6067 | goto wait_to_die; | |
6068 | } | |
6069 | ||
6070 | if (rq->active_balance) { | |
6071 | active_load_balance(rq, cpu); | |
6072 | rq->active_balance = 0; | |
6073 | } | |
6074 | ||
6075 | head = &rq->migration_queue; | |
6076 | ||
6077 | if (list_empty(head)) { | |
6078 | spin_unlock_irq(&rq->lock); | |
6079 | schedule(); | |
6080 | set_current_state(TASK_INTERRUPTIBLE); | |
6081 | continue; | |
6082 | } | |
70b97a7f | 6083 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6084 | list_del_init(head->next); |
6085 | ||
674311d5 NP |
6086 | spin_unlock(&rq->lock); |
6087 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6088 | local_irq_enable(); | |
1da177e4 LT |
6089 | |
6090 | complete(&req->done); | |
6091 | } | |
6092 | __set_current_state(TASK_RUNNING); | |
6093 | return 0; | |
6094 | ||
6095 | wait_to_die: | |
6096 | /* Wait for kthread_stop */ | |
6097 | set_current_state(TASK_INTERRUPTIBLE); | |
6098 | while (!kthread_should_stop()) { | |
6099 | schedule(); | |
6100 | set_current_state(TASK_INTERRUPTIBLE); | |
6101 | } | |
6102 | __set_current_state(TASK_RUNNING); | |
6103 | return 0; | |
6104 | } | |
6105 | ||
6106 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6107 | |
6108 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6109 | { | |
6110 | int ret; | |
6111 | ||
6112 | local_irq_disable(); | |
6113 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6114 | local_irq_enable(); | |
6115 | return ret; | |
6116 | } | |
6117 | ||
054b9108 | 6118 | /* |
3a4fa0a2 | 6119 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6120 | * NOTE: interrupts should be disabled by the caller |
6121 | */ | |
48f24c4d | 6122 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6123 | { |
efc30814 | 6124 | unsigned long flags; |
1da177e4 | 6125 | cpumask_t mask; |
70b97a7f IM |
6126 | struct rq *rq; |
6127 | int dest_cpu; | |
1da177e4 | 6128 | |
3a5c359a AK |
6129 | do { |
6130 | /* On same node? */ | |
6131 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6132 | cpus_and(mask, mask, p->cpus_allowed); | |
6133 | dest_cpu = any_online_cpu(mask); | |
6134 | ||
6135 | /* On any allowed CPU? */ | |
434d53b0 | 6136 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6137 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6138 | ||
6139 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6140 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6141 | cpumask_t cpus_allowed; |
6142 | ||
6143 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6144 | /* |
6145 | * Try to stay on the same cpuset, where the | |
6146 | * current cpuset may be a subset of all cpus. | |
6147 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6148 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6149 | * called within calls to cpuset_lock/cpuset_unlock. |
6150 | */ | |
3a5c359a | 6151 | rq = task_rq_lock(p, &flags); |
470fd646 | 6152 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6153 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6154 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6155 | |
3a5c359a AK |
6156 | /* |
6157 | * Don't tell them about moving exiting tasks or | |
6158 | * kernel threads (both mm NULL), since they never | |
6159 | * leave kernel. | |
6160 | */ | |
41a2d6cf | 6161 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6162 | printk(KERN_INFO "process %d (%s) no " |
6163 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6164 | task_pid_nr(p), p->comm, dead_cpu); |
6165 | } | |
3a5c359a | 6166 | } |
f7b4cddc | 6167 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6168 | } |
6169 | ||
6170 | /* | |
6171 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6172 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6173 | * for performance reasons the counter is not stricly tracking tasks to | |
6174 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6175 | * to keep the global sum constant after CPU-down: | |
6176 | */ | |
70b97a7f | 6177 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6178 | { |
7c16ec58 | 6179 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6180 | unsigned long flags; |
6181 | ||
6182 | local_irq_save(flags); | |
6183 | double_rq_lock(rq_src, rq_dest); | |
6184 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6185 | rq_src->nr_uninterruptible = 0; | |
6186 | double_rq_unlock(rq_src, rq_dest); | |
6187 | local_irq_restore(flags); | |
6188 | } | |
6189 | ||
6190 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6191 | static void migrate_live_tasks(int src_cpu) | |
6192 | { | |
48f24c4d | 6193 | struct task_struct *p, *t; |
1da177e4 | 6194 | |
f7b4cddc | 6195 | read_lock(&tasklist_lock); |
1da177e4 | 6196 | |
48f24c4d IM |
6197 | do_each_thread(t, p) { |
6198 | if (p == current) | |
1da177e4 LT |
6199 | continue; |
6200 | ||
48f24c4d IM |
6201 | if (task_cpu(p) == src_cpu) |
6202 | move_task_off_dead_cpu(src_cpu, p); | |
6203 | } while_each_thread(t, p); | |
1da177e4 | 6204 | |
f7b4cddc | 6205 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6206 | } |
6207 | ||
dd41f596 IM |
6208 | /* |
6209 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6210 | * It does so by boosting its priority to highest possible. |
6211 | * Used by CPU offline code. | |
1da177e4 LT |
6212 | */ |
6213 | void sched_idle_next(void) | |
6214 | { | |
48f24c4d | 6215 | int this_cpu = smp_processor_id(); |
70b97a7f | 6216 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6217 | struct task_struct *p = rq->idle; |
6218 | unsigned long flags; | |
6219 | ||
6220 | /* cpu has to be offline */ | |
48f24c4d | 6221 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6222 | |
48f24c4d IM |
6223 | /* |
6224 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6225 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6226 | */ |
6227 | spin_lock_irqsave(&rq->lock, flags); | |
6228 | ||
dd41f596 | 6229 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6230 | |
94bc9a7b DA |
6231 | update_rq_clock(rq); |
6232 | activate_task(rq, p, 0); | |
1da177e4 LT |
6233 | |
6234 | spin_unlock_irqrestore(&rq->lock, flags); | |
6235 | } | |
6236 | ||
48f24c4d IM |
6237 | /* |
6238 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6239 | * offline. |
6240 | */ | |
6241 | void idle_task_exit(void) | |
6242 | { | |
6243 | struct mm_struct *mm = current->active_mm; | |
6244 | ||
6245 | BUG_ON(cpu_online(smp_processor_id())); | |
6246 | ||
6247 | if (mm != &init_mm) | |
6248 | switch_mm(mm, &init_mm, current); | |
6249 | mmdrop(mm); | |
6250 | } | |
6251 | ||
054b9108 | 6252 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6253 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6254 | { |
70b97a7f | 6255 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6256 | |
6257 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6258 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6259 | |
6260 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6261 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6262 | |
48f24c4d | 6263 | get_task_struct(p); |
1da177e4 LT |
6264 | |
6265 | /* | |
6266 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6267 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6268 | * fine. |
6269 | */ | |
f7b4cddc | 6270 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6271 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6272 | spin_lock_irq(&rq->lock); |
1da177e4 | 6273 | |
48f24c4d | 6274 | put_task_struct(p); |
1da177e4 LT |
6275 | } |
6276 | ||
6277 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6278 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6279 | { | |
70b97a7f | 6280 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6281 | struct task_struct *next; |
48f24c4d | 6282 | |
dd41f596 IM |
6283 | for ( ; ; ) { |
6284 | if (!rq->nr_running) | |
6285 | break; | |
a8e504d2 | 6286 | update_rq_clock(rq); |
ff95f3df | 6287 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6288 | if (!next) |
6289 | break; | |
79c53799 | 6290 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6291 | migrate_dead(dead_cpu, next); |
e692ab53 | 6292 | |
1da177e4 LT |
6293 | } |
6294 | } | |
6295 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6296 | ||
e692ab53 NP |
6297 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6298 | ||
6299 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6300 | { |
6301 | .procname = "sched_domain", | |
c57baf1e | 6302 | .mode = 0555, |
e0361851 | 6303 | }, |
38605cae | 6304 | {0, }, |
e692ab53 NP |
6305 | }; |
6306 | ||
6307 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6308 | { |
c57baf1e | 6309 | .ctl_name = CTL_KERN, |
e0361851 | 6310 | .procname = "kernel", |
c57baf1e | 6311 | .mode = 0555, |
e0361851 AD |
6312 | .child = sd_ctl_dir, |
6313 | }, | |
38605cae | 6314 | {0, }, |
e692ab53 NP |
6315 | }; |
6316 | ||
6317 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6318 | { | |
6319 | struct ctl_table *entry = | |
5cf9f062 | 6320 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6321 | |
e692ab53 NP |
6322 | return entry; |
6323 | } | |
6324 | ||
6382bc90 MM |
6325 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6326 | { | |
cd790076 | 6327 | struct ctl_table *entry; |
6382bc90 | 6328 | |
cd790076 MM |
6329 | /* |
6330 | * In the intermediate directories, both the child directory and | |
6331 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6332 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6333 | * static strings and all have proc handlers. |
6334 | */ | |
6335 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6336 | if (entry->child) |
6337 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6338 | if (entry->proc_handler == NULL) |
6339 | kfree(entry->procname); | |
6340 | } | |
6382bc90 MM |
6341 | |
6342 | kfree(*tablep); | |
6343 | *tablep = NULL; | |
6344 | } | |
6345 | ||
e692ab53 | 6346 | static void |
e0361851 | 6347 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6348 | const char *procname, void *data, int maxlen, |
6349 | mode_t mode, proc_handler *proc_handler) | |
6350 | { | |
e692ab53 NP |
6351 | entry->procname = procname; |
6352 | entry->data = data; | |
6353 | entry->maxlen = maxlen; | |
6354 | entry->mode = mode; | |
6355 | entry->proc_handler = proc_handler; | |
6356 | } | |
6357 | ||
6358 | static struct ctl_table * | |
6359 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6360 | { | |
a5d8c348 | 6361 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6362 | |
ad1cdc1d MM |
6363 | if (table == NULL) |
6364 | return NULL; | |
6365 | ||
e0361851 | 6366 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6367 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6368 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6369 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6370 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6371 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6372 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6373 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6374 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6375 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6376 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6377 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6378 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6379 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6380 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6381 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6382 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6383 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6384 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6385 | &sd->cache_nice_tries, |
6386 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6387 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6388 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6389 | set_table_entry(&table[11], "name", sd->name, |
6390 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6391 | /* &table[12] is terminator */ | |
e692ab53 NP |
6392 | |
6393 | return table; | |
6394 | } | |
6395 | ||
9a4e7159 | 6396 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6397 | { |
6398 | struct ctl_table *entry, *table; | |
6399 | struct sched_domain *sd; | |
6400 | int domain_num = 0, i; | |
6401 | char buf[32]; | |
6402 | ||
6403 | for_each_domain(cpu, sd) | |
6404 | domain_num++; | |
6405 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6406 | if (table == NULL) |
6407 | return NULL; | |
e692ab53 NP |
6408 | |
6409 | i = 0; | |
6410 | for_each_domain(cpu, sd) { | |
6411 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6412 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6413 | entry->mode = 0555; |
e692ab53 NP |
6414 | entry->child = sd_alloc_ctl_domain_table(sd); |
6415 | entry++; | |
6416 | i++; | |
6417 | } | |
6418 | return table; | |
6419 | } | |
6420 | ||
6421 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6422 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6423 | { |
6424 | int i, cpu_num = num_online_cpus(); | |
6425 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6426 | char buf[32]; | |
6427 | ||
7378547f MM |
6428 | WARN_ON(sd_ctl_dir[0].child); |
6429 | sd_ctl_dir[0].child = entry; | |
6430 | ||
ad1cdc1d MM |
6431 | if (entry == NULL) |
6432 | return; | |
6433 | ||
97b6ea7b | 6434 | for_each_online_cpu(i) { |
e692ab53 | 6435 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6436 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6437 | entry->mode = 0555; |
e692ab53 | 6438 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6439 | entry++; |
e692ab53 | 6440 | } |
7378547f MM |
6441 | |
6442 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6443 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6444 | } | |
6382bc90 | 6445 | |
7378547f | 6446 | /* may be called multiple times per register */ |
6382bc90 MM |
6447 | static void unregister_sched_domain_sysctl(void) |
6448 | { | |
7378547f MM |
6449 | if (sd_sysctl_header) |
6450 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6451 | sd_sysctl_header = NULL; |
7378547f MM |
6452 | if (sd_ctl_dir[0].child) |
6453 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6454 | } |
e692ab53 | 6455 | #else |
6382bc90 MM |
6456 | static void register_sched_domain_sysctl(void) |
6457 | { | |
6458 | } | |
6459 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6460 | { |
6461 | } | |
6462 | #endif | |
6463 | ||
1f11eb6a GH |
6464 | static void set_rq_online(struct rq *rq) |
6465 | { | |
6466 | if (!rq->online) { | |
6467 | const struct sched_class *class; | |
6468 | ||
6469 | cpu_set(rq->cpu, rq->rd->online); | |
6470 | rq->online = 1; | |
6471 | ||
6472 | for_each_class(class) { | |
6473 | if (class->rq_online) | |
6474 | class->rq_online(rq); | |
6475 | } | |
6476 | } | |
6477 | } | |
6478 | ||
6479 | static void set_rq_offline(struct rq *rq) | |
6480 | { | |
6481 | if (rq->online) { | |
6482 | const struct sched_class *class; | |
6483 | ||
6484 | for_each_class(class) { | |
6485 | if (class->rq_offline) | |
6486 | class->rq_offline(rq); | |
6487 | } | |
6488 | ||
6489 | cpu_clear(rq->cpu, rq->rd->online); | |
6490 | rq->online = 0; | |
6491 | } | |
6492 | } | |
6493 | ||
1da177e4 LT |
6494 | /* |
6495 | * migration_call - callback that gets triggered when a CPU is added. | |
6496 | * Here we can start up the necessary migration thread for the new CPU. | |
6497 | */ | |
48f24c4d IM |
6498 | static int __cpuinit |
6499 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6500 | { |
1da177e4 | 6501 | struct task_struct *p; |
48f24c4d | 6502 | int cpu = (long)hcpu; |
1da177e4 | 6503 | unsigned long flags; |
70b97a7f | 6504 | struct rq *rq; |
1da177e4 LT |
6505 | |
6506 | switch (action) { | |
5be9361c | 6507 | |
1da177e4 | 6508 | case CPU_UP_PREPARE: |
8bb78442 | 6509 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6510 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6511 | if (IS_ERR(p)) |
6512 | return NOTIFY_BAD; | |
1da177e4 LT |
6513 | kthread_bind(p, cpu); |
6514 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6515 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6516 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6517 | task_rq_unlock(rq, &flags); |
6518 | cpu_rq(cpu)->migration_thread = p; | |
6519 | break; | |
48f24c4d | 6520 | |
1da177e4 | 6521 | case CPU_ONLINE: |
8bb78442 | 6522 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6523 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6524 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6525 | |
6526 | /* Update our root-domain */ | |
6527 | rq = cpu_rq(cpu); | |
6528 | spin_lock_irqsave(&rq->lock, flags); | |
6529 | if (rq->rd) { | |
6530 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6531 | |
6532 | set_rq_online(rq); | |
1f94ef59 GH |
6533 | } |
6534 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6535 | break; |
48f24c4d | 6536 | |
1da177e4 LT |
6537 | #ifdef CONFIG_HOTPLUG_CPU |
6538 | case CPU_UP_CANCELED: | |
8bb78442 | 6539 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6540 | if (!cpu_rq(cpu)->migration_thread) |
6541 | break; | |
41a2d6cf | 6542 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6543 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6544 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6545 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6546 | cpu_rq(cpu)->migration_thread = NULL; | |
6547 | break; | |
48f24c4d | 6548 | |
1da177e4 | 6549 | case CPU_DEAD: |
8bb78442 | 6550 | case CPU_DEAD_FROZEN: |
470fd646 | 6551 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6552 | migrate_live_tasks(cpu); |
6553 | rq = cpu_rq(cpu); | |
6554 | kthread_stop(rq->migration_thread); | |
6555 | rq->migration_thread = NULL; | |
6556 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6557 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6558 | update_rq_clock(rq); |
2e1cb74a | 6559 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6560 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6561 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6562 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6563 | migrate_dead_tasks(cpu); |
d2da272a | 6564 | spin_unlock_irq(&rq->lock); |
470fd646 | 6565 | cpuset_unlock(); |
1da177e4 LT |
6566 | migrate_nr_uninterruptible(rq); |
6567 | BUG_ON(rq->nr_running != 0); | |
6568 | ||
41a2d6cf IM |
6569 | /* |
6570 | * No need to migrate the tasks: it was best-effort if | |
6571 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6572 | * the requestors. | |
6573 | */ | |
1da177e4 LT |
6574 | spin_lock_irq(&rq->lock); |
6575 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6576 | struct migration_req *req; |
6577 | ||
1da177e4 | 6578 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6579 | struct migration_req, list); |
1da177e4 LT |
6580 | list_del_init(&req->list); |
6581 | complete(&req->done); | |
6582 | } | |
6583 | spin_unlock_irq(&rq->lock); | |
6584 | break; | |
57d885fe | 6585 | |
08f503b0 GH |
6586 | case CPU_DYING: |
6587 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6588 | /* Update our root-domain */ |
6589 | rq = cpu_rq(cpu); | |
6590 | spin_lock_irqsave(&rq->lock, flags); | |
6591 | if (rq->rd) { | |
6592 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6593 | set_rq_offline(rq); |
57d885fe GH |
6594 | } |
6595 | spin_unlock_irqrestore(&rq->lock, flags); | |
6596 | break; | |
1da177e4 LT |
6597 | #endif |
6598 | } | |
6599 | return NOTIFY_OK; | |
6600 | } | |
6601 | ||
6602 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6603 | * happens before everything else. | |
6604 | */ | |
26c2143b | 6605 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6606 | .notifier_call = migration_call, |
6607 | .priority = 10 | |
6608 | }; | |
6609 | ||
7babe8db | 6610 | static int __init migration_init(void) |
1da177e4 LT |
6611 | { |
6612 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6613 | int err; |
48f24c4d IM |
6614 | |
6615 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6616 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6617 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6618 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6619 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6620 | |
6621 | return err; | |
1da177e4 | 6622 | } |
7babe8db | 6623 | early_initcall(migration_init); |
1da177e4 LT |
6624 | #endif |
6625 | ||
6626 | #ifdef CONFIG_SMP | |
476f3534 | 6627 | |
3e9830dc | 6628 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6629 | |
099f98c8 GS |
6630 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6631 | { | |
6632 | switch (lvl) { | |
6633 | case SD_LV_NONE: | |
6634 | return "NONE"; | |
6635 | case SD_LV_SIBLING: | |
6636 | return "SIBLING"; | |
6637 | case SD_LV_MC: | |
6638 | return "MC"; | |
6639 | case SD_LV_CPU: | |
6640 | return "CPU"; | |
6641 | case SD_LV_NODE: | |
6642 | return "NODE"; | |
6643 | case SD_LV_ALLNODES: | |
6644 | return "ALLNODES"; | |
6645 | case SD_LV_MAX: | |
6646 | return "MAX"; | |
6647 | ||
6648 | } | |
6649 | return "MAX"; | |
6650 | } | |
6651 | ||
7c16ec58 MT |
6652 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6653 | cpumask_t *groupmask) | |
1da177e4 | 6654 | { |
4dcf6aff | 6655 | struct sched_group *group = sd->groups; |
434d53b0 | 6656 | char str[256]; |
1da177e4 | 6657 | |
434d53b0 | 6658 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6659 | cpus_clear(*groupmask); |
4dcf6aff IM |
6660 | |
6661 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6662 | ||
6663 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6664 | printk("does not load-balance\n"); | |
6665 | if (sd->parent) | |
6666 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6667 | " has parent"); | |
6668 | return -1; | |
41c7ce9a NP |
6669 | } |
6670 | ||
099f98c8 GS |
6671 | printk(KERN_CONT "span %s level %s\n", |
6672 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6673 | |
6674 | if (!cpu_isset(cpu, sd->span)) { | |
6675 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6676 | "CPU%d\n", cpu); | |
6677 | } | |
6678 | if (!cpu_isset(cpu, group->cpumask)) { | |
6679 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6680 | " CPU%d\n", cpu); | |
6681 | } | |
1da177e4 | 6682 | |
4dcf6aff | 6683 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6684 | do { |
4dcf6aff IM |
6685 | if (!group) { |
6686 | printk("\n"); | |
6687 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6688 | break; |
6689 | } | |
6690 | ||
4dcf6aff IM |
6691 | if (!group->__cpu_power) { |
6692 | printk(KERN_CONT "\n"); | |
6693 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6694 | "set\n"); | |
6695 | break; | |
6696 | } | |
1da177e4 | 6697 | |
4dcf6aff IM |
6698 | if (!cpus_weight(group->cpumask)) { |
6699 | printk(KERN_CONT "\n"); | |
6700 | printk(KERN_ERR "ERROR: empty group\n"); | |
6701 | break; | |
6702 | } | |
1da177e4 | 6703 | |
7c16ec58 | 6704 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6705 | printk(KERN_CONT "\n"); |
6706 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6707 | break; | |
6708 | } | |
1da177e4 | 6709 | |
7c16ec58 | 6710 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6711 | |
434d53b0 | 6712 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6713 | printk(KERN_CONT " %s", str); |
1da177e4 | 6714 | |
4dcf6aff IM |
6715 | group = group->next; |
6716 | } while (group != sd->groups); | |
6717 | printk(KERN_CONT "\n"); | |
1da177e4 | 6718 | |
7c16ec58 | 6719 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6720 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6721 | |
7c16ec58 | 6722 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6723 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6724 | "of domain->span\n"); | |
6725 | return 0; | |
6726 | } | |
1da177e4 | 6727 | |
4dcf6aff IM |
6728 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6729 | { | |
7c16ec58 | 6730 | cpumask_t *groupmask; |
4dcf6aff | 6731 | int level = 0; |
1da177e4 | 6732 | |
4dcf6aff IM |
6733 | if (!sd) { |
6734 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6735 | return; | |
6736 | } | |
1da177e4 | 6737 | |
4dcf6aff IM |
6738 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6739 | ||
7c16ec58 MT |
6740 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6741 | if (!groupmask) { | |
6742 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6743 | return; | |
6744 | } | |
6745 | ||
4dcf6aff | 6746 | for (;;) { |
7c16ec58 | 6747 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6748 | break; |
1da177e4 LT |
6749 | level++; |
6750 | sd = sd->parent; | |
33859f7f | 6751 | if (!sd) |
4dcf6aff IM |
6752 | break; |
6753 | } | |
7c16ec58 | 6754 | kfree(groupmask); |
1da177e4 | 6755 | } |
6d6bc0ad | 6756 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6757 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6758 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6759 | |
1a20ff27 | 6760 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6761 | { |
6762 | if (cpus_weight(sd->span) == 1) | |
6763 | return 1; | |
6764 | ||
6765 | /* Following flags need at least 2 groups */ | |
6766 | if (sd->flags & (SD_LOAD_BALANCE | | |
6767 | SD_BALANCE_NEWIDLE | | |
6768 | SD_BALANCE_FORK | | |
89c4710e SS |
6769 | SD_BALANCE_EXEC | |
6770 | SD_SHARE_CPUPOWER | | |
6771 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6772 | if (sd->groups != sd->groups->next) |
6773 | return 0; | |
6774 | } | |
6775 | ||
6776 | /* Following flags don't use groups */ | |
6777 | if (sd->flags & (SD_WAKE_IDLE | | |
6778 | SD_WAKE_AFFINE | | |
6779 | SD_WAKE_BALANCE)) | |
6780 | return 0; | |
6781 | ||
6782 | return 1; | |
6783 | } | |
6784 | ||
48f24c4d IM |
6785 | static int |
6786 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6787 | { |
6788 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6789 | ||
6790 | if (sd_degenerate(parent)) | |
6791 | return 1; | |
6792 | ||
6793 | if (!cpus_equal(sd->span, parent->span)) | |
6794 | return 0; | |
6795 | ||
6796 | /* Does parent contain flags not in child? */ | |
6797 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6798 | if (cflags & SD_WAKE_AFFINE) | |
6799 | pflags &= ~SD_WAKE_BALANCE; | |
6800 | /* Flags needing groups don't count if only 1 group in parent */ | |
6801 | if (parent->groups == parent->groups->next) { | |
6802 | pflags &= ~(SD_LOAD_BALANCE | | |
6803 | SD_BALANCE_NEWIDLE | | |
6804 | SD_BALANCE_FORK | | |
89c4710e SS |
6805 | SD_BALANCE_EXEC | |
6806 | SD_SHARE_CPUPOWER | | |
6807 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6808 | } |
6809 | if (~cflags & pflags) | |
6810 | return 0; | |
6811 | ||
6812 | return 1; | |
6813 | } | |
6814 | ||
57d885fe GH |
6815 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6816 | { | |
6817 | unsigned long flags; | |
57d885fe GH |
6818 | |
6819 | spin_lock_irqsave(&rq->lock, flags); | |
6820 | ||
6821 | if (rq->rd) { | |
6822 | struct root_domain *old_rd = rq->rd; | |
6823 | ||
1f11eb6a GH |
6824 | if (cpu_isset(rq->cpu, old_rd->online)) |
6825 | set_rq_offline(rq); | |
57d885fe | 6826 | |
dc938520 | 6827 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6828 | |
57d885fe GH |
6829 | if (atomic_dec_and_test(&old_rd->refcount)) |
6830 | kfree(old_rd); | |
6831 | } | |
6832 | ||
6833 | atomic_inc(&rd->refcount); | |
6834 | rq->rd = rd; | |
6835 | ||
dc938520 | 6836 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6837 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6838 | set_rq_online(rq); |
57d885fe GH |
6839 | |
6840 | spin_unlock_irqrestore(&rq->lock, flags); | |
6841 | } | |
6842 | ||
dc938520 | 6843 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6844 | { |
6845 | memset(rd, 0, sizeof(*rd)); | |
6846 | ||
dc938520 GH |
6847 | cpus_clear(rd->span); |
6848 | cpus_clear(rd->online); | |
6e0534f2 GH |
6849 | |
6850 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6851 | } |
6852 | ||
6853 | static void init_defrootdomain(void) | |
6854 | { | |
dc938520 | 6855 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6856 | atomic_set(&def_root_domain.refcount, 1); |
6857 | } | |
6858 | ||
dc938520 | 6859 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6860 | { |
6861 | struct root_domain *rd; | |
6862 | ||
6863 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6864 | if (!rd) | |
6865 | return NULL; | |
6866 | ||
dc938520 | 6867 | init_rootdomain(rd); |
57d885fe GH |
6868 | |
6869 | return rd; | |
6870 | } | |
6871 | ||
1da177e4 | 6872 | /* |
0eab9146 | 6873 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6874 | * hold the hotplug lock. |
6875 | */ | |
0eab9146 IM |
6876 | static void |
6877 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6878 | { |
70b97a7f | 6879 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6880 | struct sched_domain *tmp; |
6881 | ||
6882 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6883 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6884 | struct sched_domain *parent = tmp->parent; | |
6885 | if (!parent) | |
6886 | break; | |
1a848870 | 6887 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6888 | tmp->parent = parent->parent; |
1a848870 SS |
6889 | if (parent->parent) |
6890 | parent->parent->child = tmp; | |
6891 | } | |
245af2c7 SS |
6892 | } |
6893 | ||
1a848870 | 6894 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6895 | sd = sd->parent; |
1a848870 SS |
6896 | if (sd) |
6897 | sd->child = NULL; | |
6898 | } | |
1da177e4 LT |
6899 | |
6900 | sched_domain_debug(sd, cpu); | |
6901 | ||
57d885fe | 6902 | rq_attach_root(rq, rd); |
674311d5 | 6903 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6904 | } |
6905 | ||
6906 | /* cpus with isolated domains */ | |
67af63a6 | 6907 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6908 | |
6909 | /* Setup the mask of cpus configured for isolated domains */ | |
6910 | static int __init isolated_cpu_setup(char *str) | |
6911 | { | |
13b40c1e MT |
6912 | static int __initdata ints[NR_CPUS]; |
6913 | int i; | |
1da177e4 LT |
6914 | |
6915 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6916 | cpus_clear(cpu_isolated_map); | |
6917 | for (i = 1; i <= ints[0]; i++) | |
6918 | if (ints[i] < NR_CPUS) | |
6919 | cpu_set(ints[i], cpu_isolated_map); | |
6920 | return 1; | |
6921 | } | |
6922 | ||
8927f494 | 6923 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6924 | |
6925 | /* | |
6711cab4 SS |
6926 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6927 | * to a function which identifies what group(along with sched group) a CPU | |
6928 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6929 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6930 | * |
6931 | * init_sched_build_groups will build a circular linked list of the groups | |
6932 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6933 | * and ->cpu_power to 0. | |
6934 | */ | |
a616058b | 6935 | static void |
7c16ec58 | 6936 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6937 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6938 | struct sched_group **sg, |
6939 | cpumask_t *tmpmask), | |
6940 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6941 | { |
6942 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6943 | int i; |
6944 | ||
7c16ec58 MT |
6945 | cpus_clear(*covered); |
6946 | ||
363ab6f1 | 6947 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6948 | struct sched_group *sg; |
7c16ec58 | 6949 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6950 | int j; |
6951 | ||
7c16ec58 | 6952 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6953 | continue; |
6954 | ||
7c16ec58 | 6955 | cpus_clear(sg->cpumask); |
5517d86b | 6956 | sg->__cpu_power = 0; |
1da177e4 | 6957 | |
363ab6f1 | 6958 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6959 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6960 | continue; |
6961 | ||
7c16ec58 | 6962 | cpu_set(j, *covered); |
1da177e4 LT |
6963 | cpu_set(j, sg->cpumask); |
6964 | } | |
6965 | if (!first) | |
6966 | first = sg; | |
6967 | if (last) | |
6968 | last->next = sg; | |
6969 | last = sg; | |
6970 | } | |
6971 | last->next = first; | |
6972 | } | |
6973 | ||
9c1cfda2 | 6974 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6975 | |
9c1cfda2 | 6976 | #ifdef CONFIG_NUMA |
198e2f18 | 6977 | |
9c1cfda2 JH |
6978 | /** |
6979 | * find_next_best_node - find the next node to include in a sched_domain | |
6980 | * @node: node whose sched_domain we're building | |
6981 | * @used_nodes: nodes already in the sched_domain | |
6982 | * | |
41a2d6cf | 6983 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6984 | * finds the closest node not already in the @used_nodes map. |
6985 | * | |
6986 | * Should use nodemask_t. | |
6987 | */ | |
c5f59f08 | 6988 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6989 | { |
6990 | int i, n, val, min_val, best_node = 0; | |
6991 | ||
6992 | min_val = INT_MAX; | |
6993 | ||
076ac2af | 6994 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6995 | /* Start at @node */ |
076ac2af | 6996 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6997 | |
6998 | if (!nr_cpus_node(n)) | |
6999 | continue; | |
7000 | ||
7001 | /* Skip already used nodes */ | |
c5f59f08 | 7002 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7003 | continue; |
7004 | ||
7005 | /* Simple min distance search */ | |
7006 | val = node_distance(node, n); | |
7007 | ||
7008 | if (val < min_val) { | |
7009 | min_val = val; | |
7010 | best_node = n; | |
7011 | } | |
7012 | } | |
7013 | ||
c5f59f08 | 7014 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7015 | return best_node; |
7016 | } | |
7017 | ||
7018 | /** | |
7019 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7020 | * @node: node whose cpumask we're constructing | |
73486722 | 7021 | * @span: resulting cpumask |
9c1cfda2 | 7022 | * |
41a2d6cf | 7023 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7024 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7025 | * out optimally. | |
7026 | */ | |
4bdbaad3 | 7027 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 7028 | { |
c5f59f08 | 7029 | nodemask_t used_nodes; |
c5f59f08 | 7030 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7031 | int i; |
9c1cfda2 | 7032 | |
4bdbaad3 | 7033 | cpus_clear(*span); |
c5f59f08 | 7034 | nodes_clear(used_nodes); |
9c1cfda2 | 7035 | |
4bdbaad3 | 7036 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7037 | node_set(node, used_nodes); |
9c1cfda2 JH |
7038 | |
7039 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7040 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7041 | |
c5f59f08 | 7042 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7043 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7044 | } |
9c1cfda2 | 7045 | } |
6d6bc0ad | 7046 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7047 | |
5c45bf27 | 7048 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7049 | |
9c1cfda2 | 7050 | /* |
48f24c4d | 7051 | * SMT sched-domains: |
9c1cfda2 | 7052 | */ |
1da177e4 LT |
7053 | #ifdef CONFIG_SCHED_SMT |
7054 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7055 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7056 | |
41a2d6cf | 7057 | static int |
7c16ec58 MT |
7058 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7059 | cpumask_t *unused) | |
1da177e4 | 7060 | { |
6711cab4 SS |
7061 | if (sg) |
7062 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7063 | return cpu; |
7064 | } | |
6d6bc0ad | 7065 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7066 | |
48f24c4d IM |
7067 | /* |
7068 | * multi-core sched-domains: | |
7069 | */ | |
1e9f28fa SS |
7070 | #ifdef CONFIG_SCHED_MC |
7071 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7072 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7073 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7074 | |
7075 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7076 | static int |
7c16ec58 MT |
7077 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7078 | cpumask_t *mask) | |
1e9f28fa | 7079 | { |
6711cab4 | 7080 | int group; |
7c16ec58 MT |
7081 | |
7082 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7083 | cpus_and(*mask, *mask, *cpu_map); | |
7084 | group = first_cpu(*mask); | |
6711cab4 SS |
7085 | if (sg) |
7086 | *sg = &per_cpu(sched_group_core, group); | |
7087 | return group; | |
1e9f28fa SS |
7088 | } |
7089 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7090 | static int |
7c16ec58 MT |
7091 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7092 | cpumask_t *unused) | |
1e9f28fa | 7093 | { |
6711cab4 SS |
7094 | if (sg) |
7095 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7096 | return cpu; |
7097 | } | |
7098 | #endif | |
7099 | ||
1da177e4 | 7100 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7101 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7102 | |
41a2d6cf | 7103 | static int |
7c16ec58 MT |
7104 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7105 | cpumask_t *mask) | |
1da177e4 | 7106 | { |
6711cab4 | 7107 | int group; |
48f24c4d | 7108 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7109 | *mask = cpu_coregroup_map(cpu); |
7110 | cpus_and(*mask, *mask, *cpu_map); | |
7111 | group = first_cpu(*mask); | |
1e9f28fa | 7112 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7113 | *mask = per_cpu(cpu_sibling_map, cpu); |
7114 | cpus_and(*mask, *mask, *cpu_map); | |
7115 | group = first_cpu(*mask); | |
1da177e4 | 7116 | #else |
6711cab4 | 7117 | group = cpu; |
1da177e4 | 7118 | #endif |
6711cab4 SS |
7119 | if (sg) |
7120 | *sg = &per_cpu(sched_group_phys, group); | |
7121 | return group; | |
1da177e4 LT |
7122 | } |
7123 | ||
7124 | #ifdef CONFIG_NUMA | |
1da177e4 | 7125 | /* |
9c1cfda2 JH |
7126 | * The init_sched_build_groups can't handle what we want to do with node |
7127 | * groups, so roll our own. Now each node has its own list of groups which | |
7128 | * gets dynamically allocated. | |
1da177e4 | 7129 | */ |
9c1cfda2 | 7130 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7131 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7132 | |
9c1cfda2 | 7133 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7134 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7135 | |
6711cab4 | 7136 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7137 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7138 | { |
6711cab4 SS |
7139 | int group; |
7140 | ||
7c16ec58 MT |
7141 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7142 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7143 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7144 | |
7145 | if (sg) | |
7146 | *sg = &per_cpu(sched_group_allnodes, group); | |
7147 | return group; | |
1da177e4 | 7148 | } |
6711cab4 | 7149 | |
08069033 SS |
7150 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7151 | { | |
7152 | struct sched_group *sg = group_head; | |
7153 | int j; | |
7154 | ||
7155 | if (!sg) | |
7156 | return; | |
3a5c359a | 7157 | do { |
363ab6f1 | 7158 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7159 | struct sched_domain *sd; |
08069033 | 7160 | |
3a5c359a AK |
7161 | sd = &per_cpu(phys_domains, j); |
7162 | if (j != first_cpu(sd->groups->cpumask)) { | |
7163 | /* | |
7164 | * Only add "power" once for each | |
7165 | * physical package. | |
7166 | */ | |
7167 | continue; | |
7168 | } | |
08069033 | 7169 | |
3a5c359a AK |
7170 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7171 | } | |
7172 | sg = sg->next; | |
7173 | } while (sg != group_head); | |
08069033 | 7174 | } |
6d6bc0ad | 7175 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7176 | |
a616058b | 7177 | #ifdef CONFIG_NUMA |
51888ca2 | 7178 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7179 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7180 | { |
a616058b | 7181 | int cpu, i; |
51888ca2 | 7182 | |
363ab6f1 | 7183 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7184 | struct sched_group **sched_group_nodes |
7185 | = sched_group_nodes_bycpu[cpu]; | |
7186 | ||
51888ca2 SV |
7187 | if (!sched_group_nodes) |
7188 | continue; | |
7189 | ||
076ac2af | 7190 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7191 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7192 | ||
7c16ec58 MT |
7193 | *nodemask = node_to_cpumask(i); |
7194 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7195 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7196 | continue; |
7197 | ||
7198 | if (sg == NULL) | |
7199 | continue; | |
7200 | sg = sg->next; | |
7201 | next_sg: | |
7202 | oldsg = sg; | |
7203 | sg = sg->next; | |
7204 | kfree(oldsg); | |
7205 | if (oldsg != sched_group_nodes[i]) | |
7206 | goto next_sg; | |
7207 | } | |
7208 | kfree(sched_group_nodes); | |
7209 | sched_group_nodes_bycpu[cpu] = NULL; | |
7210 | } | |
51888ca2 | 7211 | } |
6d6bc0ad | 7212 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7213 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7214 | { |
7215 | } | |
6d6bc0ad | 7216 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7217 | |
89c4710e SS |
7218 | /* |
7219 | * Initialize sched groups cpu_power. | |
7220 | * | |
7221 | * cpu_power indicates the capacity of sched group, which is used while | |
7222 | * distributing the load between different sched groups in a sched domain. | |
7223 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7224 | * there are asymmetries in the topology. If there are asymmetries, group | |
7225 | * having more cpu_power will pickup more load compared to the group having | |
7226 | * less cpu_power. | |
7227 | * | |
7228 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7229 | * the maximum number of tasks a group can handle in the presence of other idle | |
7230 | * or lightly loaded groups in the same sched domain. | |
7231 | */ | |
7232 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7233 | { | |
7234 | struct sched_domain *child; | |
7235 | struct sched_group *group; | |
7236 | ||
7237 | WARN_ON(!sd || !sd->groups); | |
7238 | ||
7239 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7240 | return; | |
7241 | ||
7242 | child = sd->child; | |
7243 | ||
5517d86b ED |
7244 | sd->groups->__cpu_power = 0; |
7245 | ||
89c4710e SS |
7246 | /* |
7247 | * For perf policy, if the groups in child domain share resources | |
7248 | * (for example cores sharing some portions of the cache hierarchy | |
7249 | * or SMT), then set this domain groups cpu_power such that each group | |
7250 | * can handle only one task, when there are other idle groups in the | |
7251 | * same sched domain. | |
7252 | */ | |
7253 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7254 | (child->flags & | |
7255 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7256 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7257 | return; |
7258 | } | |
7259 | ||
89c4710e SS |
7260 | /* |
7261 | * add cpu_power of each child group to this groups cpu_power | |
7262 | */ | |
7263 | group = child->groups; | |
7264 | do { | |
5517d86b | 7265 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7266 | group = group->next; |
7267 | } while (group != child->groups); | |
7268 | } | |
7269 | ||
7c16ec58 MT |
7270 | /* |
7271 | * Initializers for schedule domains | |
7272 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7273 | */ | |
7274 | ||
a5d8c348 IM |
7275 | #ifdef CONFIG_SCHED_DEBUG |
7276 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7277 | #else | |
7278 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7279 | #endif | |
7280 | ||
7c16ec58 | 7281 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7282 | |
7c16ec58 MT |
7283 | #define SD_INIT_FUNC(type) \ |
7284 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7285 | { \ | |
7286 | memset(sd, 0, sizeof(*sd)); \ | |
7287 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7288 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7289 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7290 | } |
7291 | ||
7292 | SD_INIT_FUNC(CPU) | |
7293 | #ifdef CONFIG_NUMA | |
7294 | SD_INIT_FUNC(ALLNODES) | |
7295 | SD_INIT_FUNC(NODE) | |
7296 | #endif | |
7297 | #ifdef CONFIG_SCHED_SMT | |
7298 | SD_INIT_FUNC(SIBLING) | |
7299 | #endif | |
7300 | #ifdef CONFIG_SCHED_MC | |
7301 | SD_INIT_FUNC(MC) | |
7302 | #endif | |
7303 | ||
7304 | /* | |
7305 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7306 | * space as the usage in build_sched_domains() dictates. Used only | |
7307 | * if the amount of space is significant. | |
7308 | */ | |
7309 | struct allmasks { | |
7310 | cpumask_t tmpmask; /* make this one first */ | |
7311 | union { | |
7312 | cpumask_t nodemask; | |
7313 | cpumask_t this_sibling_map; | |
7314 | cpumask_t this_core_map; | |
7315 | }; | |
7316 | cpumask_t send_covered; | |
7317 | ||
7318 | #ifdef CONFIG_NUMA | |
7319 | cpumask_t domainspan; | |
7320 | cpumask_t covered; | |
7321 | cpumask_t notcovered; | |
7322 | #endif | |
7323 | }; | |
7324 | ||
7325 | #if NR_CPUS > 128 | |
7326 | #define SCHED_CPUMASK_ALLOC 1 | |
7327 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7328 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7329 | #else | |
7330 | #define SCHED_CPUMASK_ALLOC 0 | |
7331 | #define SCHED_CPUMASK_FREE(v) | |
7332 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7333 | #endif | |
7334 | ||
7335 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7336 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7337 | ||
1d3504fc HS |
7338 | static int default_relax_domain_level = -1; |
7339 | ||
7340 | static int __init setup_relax_domain_level(char *str) | |
7341 | { | |
30e0e178 LZ |
7342 | unsigned long val; |
7343 | ||
7344 | val = simple_strtoul(str, NULL, 0); | |
7345 | if (val < SD_LV_MAX) | |
7346 | default_relax_domain_level = val; | |
7347 | ||
1d3504fc HS |
7348 | return 1; |
7349 | } | |
7350 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7351 | ||
7352 | static void set_domain_attribute(struct sched_domain *sd, | |
7353 | struct sched_domain_attr *attr) | |
7354 | { | |
7355 | int request; | |
7356 | ||
7357 | if (!attr || attr->relax_domain_level < 0) { | |
7358 | if (default_relax_domain_level < 0) | |
7359 | return; | |
7360 | else | |
7361 | request = default_relax_domain_level; | |
7362 | } else | |
7363 | request = attr->relax_domain_level; | |
7364 | if (request < sd->level) { | |
7365 | /* turn off idle balance on this domain */ | |
7366 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7367 | } else { | |
7368 | /* turn on idle balance on this domain */ | |
7369 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7370 | } | |
7371 | } | |
7372 | ||
1da177e4 | 7373 | /* |
1a20ff27 DG |
7374 | * Build sched domains for a given set of cpus and attach the sched domains |
7375 | * to the individual cpus | |
1da177e4 | 7376 | */ |
1d3504fc HS |
7377 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7378 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7379 | { |
7380 | int i; | |
57d885fe | 7381 | struct root_domain *rd; |
7c16ec58 MT |
7382 | SCHED_CPUMASK_DECLARE(allmasks); |
7383 | cpumask_t *tmpmask; | |
d1b55138 JH |
7384 | #ifdef CONFIG_NUMA |
7385 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7386 | int sd_allnodes = 0; |
d1b55138 JH |
7387 | |
7388 | /* | |
7389 | * Allocate the per-node list of sched groups | |
7390 | */ | |
076ac2af | 7391 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7392 | GFP_KERNEL); |
d1b55138 JH |
7393 | if (!sched_group_nodes) { |
7394 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7395 | return -ENOMEM; |
d1b55138 | 7396 | } |
d1b55138 | 7397 | #endif |
1da177e4 | 7398 | |
dc938520 | 7399 | rd = alloc_rootdomain(); |
57d885fe GH |
7400 | if (!rd) { |
7401 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7402 | #ifdef CONFIG_NUMA |
7403 | kfree(sched_group_nodes); | |
7404 | #endif | |
57d885fe GH |
7405 | return -ENOMEM; |
7406 | } | |
7407 | ||
7c16ec58 MT |
7408 | #if SCHED_CPUMASK_ALLOC |
7409 | /* get space for all scratch cpumask variables */ | |
7410 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7411 | if (!allmasks) { | |
7412 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7413 | kfree(rd); | |
7414 | #ifdef CONFIG_NUMA | |
7415 | kfree(sched_group_nodes); | |
7416 | #endif | |
7417 | return -ENOMEM; | |
7418 | } | |
7419 | #endif | |
7420 | tmpmask = (cpumask_t *)allmasks; | |
7421 | ||
7422 | ||
7423 | #ifdef CONFIG_NUMA | |
7424 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7425 | #endif | |
7426 | ||
1da177e4 | 7427 | /* |
1a20ff27 | 7428 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7429 | */ |
363ab6f1 | 7430 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7431 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7432 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7433 | |
7c16ec58 MT |
7434 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7435 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7436 | |
7437 | #ifdef CONFIG_NUMA | |
dd41f596 | 7438 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7439 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7440 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7441 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7442 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7443 | sd->span = *cpu_map; |
7c16ec58 | 7444 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7445 | p = sd; |
6711cab4 | 7446 | sd_allnodes = 1; |
9c1cfda2 JH |
7447 | } else |
7448 | p = NULL; | |
7449 | ||
1da177e4 | 7450 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7451 | SD_INIT(sd, NODE); |
1d3504fc | 7452 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7453 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7454 | sd->parent = p; |
1a848870 SS |
7455 | if (p) |
7456 | p->child = sd; | |
9c1cfda2 | 7457 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7458 | #endif |
7459 | ||
7460 | p = sd; | |
7461 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7462 | SD_INIT(sd, CPU); |
1d3504fc | 7463 | set_domain_attribute(sd, attr); |
7c16ec58 | 7464 | sd->span = *nodemask; |
1da177e4 | 7465 | sd->parent = p; |
1a848870 SS |
7466 | if (p) |
7467 | p->child = sd; | |
7c16ec58 | 7468 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7469 | |
1e9f28fa SS |
7470 | #ifdef CONFIG_SCHED_MC |
7471 | p = sd; | |
7472 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7473 | SD_INIT(sd, MC); |
1d3504fc | 7474 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7475 | sd->span = cpu_coregroup_map(i); |
7476 | cpus_and(sd->span, sd->span, *cpu_map); | |
7477 | sd->parent = p; | |
1a848870 | 7478 | p->child = sd; |
7c16ec58 | 7479 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7480 | #endif |
7481 | ||
1da177e4 LT |
7482 | #ifdef CONFIG_SCHED_SMT |
7483 | p = sd; | |
7484 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7485 | SD_INIT(sd, SIBLING); |
1d3504fc | 7486 | set_domain_attribute(sd, attr); |
d5a7430d | 7487 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7488 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7489 | sd->parent = p; |
1a848870 | 7490 | p->child = sd; |
7c16ec58 | 7491 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7492 | #endif |
7493 | } | |
7494 | ||
7495 | #ifdef CONFIG_SCHED_SMT | |
7496 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7497 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7498 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7499 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7500 | ||
7501 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7502 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7503 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7504 | continue; |
7505 | ||
dd41f596 | 7506 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7507 | &cpu_to_cpu_group, |
7508 | send_covered, tmpmask); | |
1da177e4 LT |
7509 | } |
7510 | #endif | |
7511 | ||
1e9f28fa SS |
7512 | #ifdef CONFIG_SCHED_MC |
7513 | /* Set up multi-core groups */ | |
363ab6f1 | 7514 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7515 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7516 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7517 | ||
7518 | *this_core_map = cpu_coregroup_map(i); | |
7519 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7520 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7521 | continue; |
7c16ec58 | 7522 | |
dd41f596 | 7523 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7524 | &cpu_to_core_group, |
7525 | send_covered, tmpmask); | |
1e9f28fa SS |
7526 | } |
7527 | #endif | |
7528 | ||
1da177e4 | 7529 | /* Set up physical groups */ |
076ac2af | 7530 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7531 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7532 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7533 | |
7c16ec58 MT |
7534 | *nodemask = node_to_cpumask(i); |
7535 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7536 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7537 | continue; |
7538 | ||
7c16ec58 MT |
7539 | init_sched_build_groups(nodemask, cpu_map, |
7540 | &cpu_to_phys_group, | |
7541 | send_covered, tmpmask); | |
1da177e4 LT |
7542 | } |
7543 | ||
7544 | #ifdef CONFIG_NUMA | |
7545 | /* Set up node groups */ | |
7c16ec58 MT |
7546 | if (sd_allnodes) { |
7547 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7548 | ||
7549 | init_sched_build_groups(cpu_map, cpu_map, | |
7550 | &cpu_to_allnodes_group, | |
7551 | send_covered, tmpmask); | |
7552 | } | |
9c1cfda2 | 7553 | |
076ac2af | 7554 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7555 | /* Set up node groups */ |
7556 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7557 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7558 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7559 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7560 | int j; |
7561 | ||
7c16ec58 MT |
7562 | *nodemask = node_to_cpumask(i); |
7563 | cpus_clear(*covered); | |
7564 | ||
7565 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7566 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7567 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7568 | continue; |
d1b55138 | 7569 | } |
9c1cfda2 | 7570 | |
4bdbaad3 | 7571 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7572 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7573 | |
15f0b676 | 7574 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7575 | if (!sg) { |
7576 | printk(KERN_WARNING "Can not alloc domain group for " | |
7577 | "node %d\n", i); | |
7578 | goto error; | |
7579 | } | |
9c1cfda2 | 7580 | sched_group_nodes[i] = sg; |
363ab6f1 | 7581 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7582 | struct sched_domain *sd; |
9761eea8 | 7583 | |
9c1cfda2 JH |
7584 | sd = &per_cpu(node_domains, j); |
7585 | sd->groups = sg; | |
9c1cfda2 | 7586 | } |
5517d86b | 7587 | sg->__cpu_power = 0; |
7c16ec58 | 7588 | sg->cpumask = *nodemask; |
51888ca2 | 7589 | sg->next = sg; |
7c16ec58 | 7590 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7591 | prev = sg; |
7592 | ||
076ac2af | 7593 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7594 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7595 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7596 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7597 | |
7c16ec58 MT |
7598 | cpus_complement(*notcovered, *covered); |
7599 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7600 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7601 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7602 | break; |
7603 | ||
7c16ec58 MT |
7604 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7605 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7606 | continue; |
7607 | ||
15f0b676 SV |
7608 | sg = kmalloc_node(sizeof(struct sched_group), |
7609 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7610 | if (!sg) { |
7611 | printk(KERN_WARNING | |
7612 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7613 | goto error; |
9c1cfda2 | 7614 | } |
5517d86b | 7615 | sg->__cpu_power = 0; |
7c16ec58 | 7616 | sg->cpumask = *tmpmask; |
51888ca2 | 7617 | sg->next = prev->next; |
7c16ec58 | 7618 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7619 | prev->next = sg; |
7620 | prev = sg; | |
7621 | } | |
9c1cfda2 | 7622 | } |
1da177e4 LT |
7623 | #endif |
7624 | ||
7625 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7626 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7627 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7628 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7629 | ||
89c4710e | 7630 | init_sched_groups_power(i, sd); |
5c45bf27 | 7631 | } |
1da177e4 | 7632 | #endif |
1e9f28fa | 7633 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7634 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7635 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7636 | ||
89c4710e | 7637 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7638 | } |
7639 | #endif | |
1e9f28fa | 7640 | |
363ab6f1 | 7641 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7642 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7643 | ||
89c4710e | 7644 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7645 | } |
7646 | ||
9c1cfda2 | 7647 | #ifdef CONFIG_NUMA |
076ac2af | 7648 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7649 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7650 | |
6711cab4 SS |
7651 | if (sd_allnodes) { |
7652 | struct sched_group *sg; | |
f712c0c7 | 7653 | |
7c16ec58 MT |
7654 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7655 | tmpmask); | |
f712c0c7 SS |
7656 | init_numa_sched_groups_power(sg); |
7657 | } | |
9c1cfda2 JH |
7658 | #endif |
7659 | ||
1da177e4 | 7660 | /* Attach the domains */ |
363ab6f1 | 7661 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7662 | struct sched_domain *sd; |
7663 | #ifdef CONFIG_SCHED_SMT | |
7664 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7665 | #elif defined(CONFIG_SCHED_MC) |
7666 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7667 | #else |
7668 | sd = &per_cpu(phys_domains, i); | |
7669 | #endif | |
57d885fe | 7670 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7671 | } |
51888ca2 | 7672 | |
7c16ec58 | 7673 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7674 | return 0; |
7675 | ||
a616058b | 7676 | #ifdef CONFIG_NUMA |
51888ca2 | 7677 | error: |
7c16ec58 MT |
7678 | free_sched_groups(cpu_map, tmpmask); |
7679 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7680 | return -ENOMEM; |
a616058b | 7681 | #endif |
1da177e4 | 7682 | } |
029190c5 | 7683 | |
1d3504fc HS |
7684 | static int build_sched_domains(const cpumask_t *cpu_map) |
7685 | { | |
7686 | return __build_sched_domains(cpu_map, NULL); | |
7687 | } | |
7688 | ||
029190c5 PJ |
7689 | static cpumask_t *doms_cur; /* current sched domains */ |
7690 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7691 | static struct sched_domain_attr *dattr_cur; |
7692 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7693 | |
7694 | /* | |
7695 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7696 | * cpumask_t) fails, then fallback to a single sched domain, | |
7697 | * as determined by the single cpumask_t fallback_doms. | |
7698 | */ | |
7699 | static cpumask_t fallback_doms; | |
7700 | ||
22e52b07 HC |
7701 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7702 | { | |
7703 | } | |
7704 | ||
1a20ff27 | 7705 | /* |
41a2d6cf | 7706 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7707 | * For now this just excludes isolated cpus, but could be used to |
7708 | * exclude other special cases in the future. | |
1a20ff27 | 7709 | */ |
51888ca2 | 7710 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7711 | { |
7378547f MM |
7712 | int err; |
7713 | ||
22e52b07 | 7714 | arch_update_cpu_topology(); |
029190c5 PJ |
7715 | ndoms_cur = 1; |
7716 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7717 | if (!doms_cur) | |
7718 | doms_cur = &fallback_doms; | |
7719 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7720 | dattr_cur = NULL; |
7378547f | 7721 | err = build_sched_domains(doms_cur); |
6382bc90 | 7722 | register_sched_domain_sysctl(); |
7378547f MM |
7723 | |
7724 | return err; | |
1a20ff27 DG |
7725 | } |
7726 | ||
7c16ec58 MT |
7727 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7728 | cpumask_t *tmpmask) | |
1da177e4 | 7729 | { |
7c16ec58 | 7730 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7731 | } |
1da177e4 | 7732 | |
1a20ff27 DG |
7733 | /* |
7734 | * Detach sched domains from a group of cpus specified in cpu_map | |
7735 | * These cpus will now be attached to the NULL domain | |
7736 | */ | |
858119e1 | 7737 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7738 | { |
7c16ec58 | 7739 | cpumask_t tmpmask; |
1a20ff27 DG |
7740 | int i; |
7741 | ||
6382bc90 MM |
7742 | unregister_sched_domain_sysctl(); |
7743 | ||
363ab6f1 | 7744 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7745 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7746 | synchronize_sched(); |
7c16ec58 | 7747 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7748 | } |
7749 | ||
1d3504fc HS |
7750 | /* handle null as "default" */ |
7751 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7752 | struct sched_domain_attr *new, int idx_new) | |
7753 | { | |
7754 | struct sched_domain_attr tmp; | |
7755 | ||
7756 | /* fast path */ | |
7757 | if (!new && !cur) | |
7758 | return 1; | |
7759 | ||
7760 | tmp = SD_ATTR_INIT; | |
7761 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7762 | new ? (new + idx_new) : &tmp, | |
7763 | sizeof(struct sched_domain_attr)); | |
7764 | } | |
7765 | ||
029190c5 PJ |
7766 | /* |
7767 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7768 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7769 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7770 | * It destroys each deleted domain and builds each new domain. | |
7771 | * | |
7772 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7773 | * The masks don't intersect (don't overlap.) We should setup one |
7774 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7775 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7776 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7777 | * it as it is. | |
7778 | * | |
41a2d6cf IM |
7779 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7780 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7781 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7782 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7783 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 | 7784 | * |
dfb512ec MK |
7785 | * If doms_new==NULL it will be replaced with cpu_online_map. |
7786 | * ndoms_new==0 is a special case for destroying existing domains. | |
7787 | * It will not create the default domain. | |
7788 | * | |
029190c5 PJ |
7789 | * Call with hotplug lock held |
7790 | */ | |
1d3504fc HS |
7791 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7792 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7793 | { |
dfb512ec | 7794 | int i, j, n; |
029190c5 | 7795 | |
712555ee | 7796 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7797 | |
7378547f MM |
7798 | /* always unregister in case we don't destroy any domains */ |
7799 | unregister_sched_domain_sysctl(); | |
7800 | ||
dfb512ec | 7801 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7802 | |
7803 | /* Destroy deleted domains */ | |
7804 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7805 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7806 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7807 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7808 | goto match1; |
7809 | } | |
7810 | /* no match - a current sched domain not in new doms_new[] */ | |
7811 | detach_destroy_domains(doms_cur + i); | |
7812 | match1: | |
7813 | ; | |
7814 | } | |
7815 | ||
e761b772 MK |
7816 | if (doms_new == NULL) { |
7817 | ndoms_cur = 0; | |
e761b772 MK |
7818 | doms_new = &fallback_doms; |
7819 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7820 | dattr_new = NULL; | |
7821 | } | |
7822 | ||
029190c5 PJ |
7823 | /* Build new domains */ |
7824 | for (i = 0; i < ndoms_new; i++) { | |
7825 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7826 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7827 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7828 | goto match2; |
7829 | } | |
7830 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7831 | __build_sched_domains(doms_new + i, |
7832 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7833 | match2: |
7834 | ; | |
7835 | } | |
7836 | ||
7837 | /* Remember the new sched domains */ | |
7838 | if (doms_cur != &fallback_doms) | |
7839 | kfree(doms_cur); | |
1d3504fc | 7840 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7841 | doms_cur = doms_new; |
1d3504fc | 7842 | dattr_cur = dattr_new; |
029190c5 | 7843 | ndoms_cur = ndoms_new; |
7378547f MM |
7844 | |
7845 | register_sched_domain_sysctl(); | |
a1835615 | 7846 | |
712555ee | 7847 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7848 | } |
7849 | ||
5c45bf27 | 7850 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7851 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7852 | { |
95402b38 | 7853 | get_online_cpus(); |
dfb512ec MK |
7854 | |
7855 | /* Destroy domains first to force the rebuild */ | |
7856 | partition_sched_domains(0, NULL, NULL); | |
7857 | ||
e761b772 | 7858 | rebuild_sched_domains(); |
95402b38 | 7859 | put_online_cpus(); |
dfb512ec | 7860 | |
e761b772 | 7861 | return 0; |
5c45bf27 SS |
7862 | } |
7863 | ||
7864 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7865 | { | |
7866 | int ret; | |
7867 | ||
7868 | if (buf[0] != '0' && buf[0] != '1') | |
7869 | return -EINVAL; | |
7870 | ||
7871 | if (smt) | |
7872 | sched_smt_power_savings = (buf[0] == '1'); | |
7873 | else | |
7874 | sched_mc_power_savings = (buf[0] == '1'); | |
7875 | ||
7876 | ret = arch_reinit_sched_domains(); | |
7877 | ||
7878 | return ret ? ret : count; | |
7879 | } | |
7880 | ||
5c45bf27 | 7881 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7882 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7883 | char *page) | |
5c45bf27 SS |
7884 | { |
7885 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7886 | } | |
f718cd4a | 7887 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7888 | const char *buf, size_t count) |
5c45bf27 SS |
7889 | { |
7890 | return sched_power_savings_store(buf, count, 0); | |
7891 | } | |
f718cd4a AK |
7892 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7893 | sched_mc_power_savings_show, | |
7894 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7895 | #endif |
7896 | ||
7897 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7898 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7899 | char *page) | |
5c45bf27 SS |
7900 | { |
7901 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7902 | } | |
f718cd4a | 7903 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7904 | const char *buf, size_t count) |
5c45bf27 SS |
7905 | { |
7906 | return sched_power_savings_store(buf, count, 1); | |
7907 | } | |
f718cd4a AK |
7908 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7909 | sched_smt_power_savings_show, | |
6707de00 AB |
7910 | sched_smt_power_savings_store); |
7911 | #endif | |
7912 | ||
7913 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7914 | { | |
7915 | int err = 0; | |
7916 | ||
7917 | #ifdef CONFIG_SCHED_SMT | |
7918 | if (smt_capable()) | |
7919 | err = sysfs_create_file(&cls->kset.kobj, | |
7920 | &attr_sched_smt_power_savings.attr); | |
7921 | #endif | |
7922 | #ifdef CONFIG_SCHED_MC | |
7923 | if (!err && mc_capable()) | |
7924 | err = sysfs_create_file(&cls->kset.kobj, | |
7925 | &attr_sched_mc_power_savings.attr); | |
7926 | #endif | |
7927 | return err; | |
7928 | } | |
6d6bc0ad | 7929 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7930 | |
e761b772 | 7931 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7932 | /* |
e761b772 MK |
7933 | * Add online and remove offline CPUs from the scheduler domains. |
7934 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7935 | */ |
7936 | static int update_sched_domains(struct notifier_block *nfb, | |
7937 | unsigned long action, void *hcpu) | |
e761b772 MK |
7938 | { |
7939 | switch (action) { | |
7940 | case CPU_ONLINE: | |
7941 | case CPU_ONLINE_FROZEN: | |
7942 | case CPU_DEAD: | |
7943 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7944 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7945 | return NOTIFY_OK; |
7946 | ||
7947 | default: | |
7948 | return NOTIFY_DONE; | |
7949 | } | |
7950 | } | |
7951 | #endif | |
7952 | ||
7953 | static int update_runtime(struct notifier_block *nfb, | |
7954 | unsigned long action, void *hcpu) | |
1da177e4 | 7955 | { |
7def2be1 PZ |
7956 | int cpu = (int)(long)hcpu; |
7957 | ||
1da177e4 | 7958 | switch (action) { |
1da177e4 | 7959 | case CPU_DOWN_PREPARE: |
8bb78442 | 7960 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7961 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7962 | return NOTIFY_OK; |
7963 | ||
1da177e4 | 7964 | case CPU_DOWN_FAILED: |
8bb78442 | 7965 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7966 | case CPU_ONLINE: |
8bb78442 | 7967 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7968 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7969 | return NOTIFY_OK; |
7970 | ||
1da177e4 LT |
7971 | default: |
7972 | return NOTIFY_DONE; | |
7973 | } | |
1da177e4 | 7974 | } |
1da177e4 LT |
7975 | |
7976 | void __init sched_init_smp(void) | |
7977 | { | |
5c1e1767 NP |
7978 | cpumask_t non_isolated_cpus; |
7979 | ||
434d53b0 MT |
7980 | #if defined(CONFIG_NUMA) |
7981 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7982 | GFP_KERNEL); | |
7983 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7984 | #endif | |
95402b38 | 7985 | get_online_cpus(); |
712555ee | 7986 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7987 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7988 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7989 | if (cpus_empty(non_isolated_cpus)) |
7990 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7991 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7992 | put_online_cpus(); |
e761b772 MK |
7993 | |
7994 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7995 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7996 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7997 | #endif |
7998 | ||
7999 | /* RT runtime code needs to handle some hotplug events */ | |
8000 | hotcpu_notifier(update_runtime, 0); | |
8001 | ||
b328ca18 | 8002 | init_hrtick(); |
5c1e1767 NP |
8003 | |
8004 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 8005 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 8006 | BUG(); |
19978ca6 | 8007 | sched_init_granularity(); |
1da177e4 LT |
8008 | } |
8009 | #else | |
8010 | void __init sched_init_smp(void) | |
8011 | { | |
19978ca6 | 8012 | sched_init_granularity(); |
1da177e4 LT |
8013 | } |
8014 | #endif /* CONFIG_SMP */ | |
8015 | ||
8016 | int in_sched_functions(unsigned long addr) | |
8017 | { | |
1da177e4 LT |
8018 | return in_lock_functions(addr) || |
8019 | (addr >= (unsigned long)__sched_text_start | |
8020 | && addr < (unsigned long)__sched_text_end); | |
8021 | } | |
8022 | ||
a9957449 | 8023 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8024 | { |
8025 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8026 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8027 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8028 | cfs_rq->rq = rq; | |
8029 | #endif | |
67e9fb2a | 8030 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8031 | } |
8032 | ||
fa85ae24 PZ |
8033 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8034 | { | |
8035 | struct rt_prio_array *array; | |
8036 | int i; | |
8037 | ||
8038 | array = &rt_rq->active; | |
8039 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8040 | INIT_LIST_HEAD(array->queue + i); | |
8041 | __clear_bit(i, array->bitmap); | |
8042 | } | |
8043 | /* delimiter for bitsearch: */ | |
8044 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8045 | ||
052f1dc7 | 8046 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8047 | rt_rq->highest_prio = MAX_RT_PRIO; |
8048 | #endif | |
fa85ae24 PZ |
8049 | #ifdef CONFIG_SMP |
8050 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8051 | rt_rq->overloaded = 0; |
8052 | #endif | |
8053 | ||
8054 | rt_rq->rt_time = 0; | |
8055 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8056 | rt_rq->rt_runtime = 0; |
8057 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8058 | |
052f1dc7 | 8059 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8060 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8061 | rt_rq->rq = rq; |
8062 | #endif | |
fa85ae24 PZ |
8063 | } |
8064 | ||
6f505b16 | 8065 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8066 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8067 | struct sched_entity *se, int cpu, int add, | |
8068 | struct sched_entity *parent) | |
6f505b16 | 8069 | { |
ec7dc8ac | 8070 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8071 | tg->cfs_rq[cpu] = cfs_rq; |
8072 | init_cfs_rq(cfs_rq, rq); | |
8073 | cfs_rq->tg = tg; | |
8074 | if (add) | |
8075 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8076 | ||
8077 | tg->se[cpu] = se; | |
354d60c2 DG |
8078 | /* se could be NULL for init_task_group */ |
8079 | if (!se) | |
8080 | return; | |
8081 | ||
ec7dc8ac DG |
8082 | if (!parent) |
8083 | se->cfs_rq = &rq->cfs; | |
8084 | else | |
8085 | se->cfs_rq = parent->my_q; | |
8086 | ||
6f505b16 PZ |
8087 | se->my_q = cfs_rq; |
8088 | se->load.weight = tg->shares; | |
e05510d0 | 8089 | se->load.inv_weight = 0; |
ec7dc8ac | 8090 | se->parent = parent; |
6f505b16 | 8091 | } |
052f1dc7 | 8092 | #endif |
6f505b16 | 8093 | |
052f1dc7 | 8094 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8095 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8096 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8097 | struct sched_rt_entity *parent) | |
6f505b16 | 8098 | { |
ec7dc8ac DG |
8099 | struct rq *rq = cpu_rq(cpu); |
8100 | ||
6f505b16 PZ |
8101 | tg->rt_rq[cpu] = rt_rq; |
8102 | init_rt_rq(rt_rq, rq); | |
8103 | rt_rq->tg = tg; | |
8104 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8105 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8106 | if (add) |
8107 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8108 | ||
8109 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8110 | if (!rt_se) |
8111 | return; | |
8112 | ||
ec7dc8ac DG |
8113 | if (!parent) |
8114 | rt_se->rt_rq = &rq->rt; | |
8115 | else | |
8116 | rt_se->rt_rq = parent->my_q; | |
8117 | ||
6f505b16 | 8118 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8119 | rt_se->parent = parent; |
6f505b16 PZ |
8120 | INIT_LIST_HEAD(&rt_se->run_list); |
8121 | } | |
8122 | #endif | |
8123 | ||
1da177e4 LT |
8124 | void __init sched_init(void) |
8125 | { | |
dd41f596 | 8126 | int i, j; |
434d53b0 MT |
8127 | unsigned long alloc_size = 0, ptr; |
8128 | ||
8129 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8130 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8131 | #endif | |
8132 | #ifdef CONFIG_RT_GROUP_SCHED | |
8133 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8134 | #endif |
8135 | #ifdef CONFIG_USER_SCHED | |
8136 | alloc_size *= 2; | |
434d53b0 MT |
8137 | #endif |
8138 | /* | |
8139 | * As sched_init() is called before page_alloc is setup, | |
8140 | * we use alloc_bootmem(). | |
8141 | */ | |
8142 | if (alloc_size) { | |
5a9d3225 | 8143 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8144 | |
8145 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8146 | init_task_group.se = (struct sched_entity **)ptr; | |
8147 | ptr += nr_cpu_ids * sizeof(void **); | |
8148 | ||
8149 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8150 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8151 | |
8152 | #ifdef CONFIG_USER_SCHED | |
8153 | root_task_group.se = (struct sched_entity **)ptr; | |
8154 | ptr += nr_cpu_ids * sizeof(void **); | |
8155 | ||
8156 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8157 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8158 | #endif /* CONFIG_USER_SCHED */ |
8159 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8160 | #ifdef CONFIG_RT_GROUP_SCHED |
8161 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8162 | ptr += nr_cpu_ids * sizeof(void **); | |
8163 | ||
8164 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8165 | ptr += nr_cpu_ids * sizeof(void **); |
8166 | ||
8167 | #ifdef CONFIG_USER_SCHED | |
8168 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8169 | ptr += nr_cpu_ids * sizeof(void **); | |
8170 | ||
8171 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8172 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8173 | #endif /* CONFIG_USER_SCHED */ |
8174 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8175 | } |
dd41f596 | 8176 | |
57d885fe GH |
8177 | #ifdef CONFIG_SMP |
8178 | init_defrootdomain(); | |
8179 | #endif | |
8180 | ||
d0b27fa7 PZ |
8181 | init_rt_bandwidth(&def_rt_bandwidth, |
8182 | global_rt_period(), global_rt_runtime()); | |
8183 | ||
8184 | #ifdef CONFIG_RT_GROUP_SCHED | |
8185 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8186 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8187 | #ifdef CONFIG_USER_SCHED |
8188 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8189 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8190 | #endif /* CONFIG_USER_SCHED */ |
8191 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8192 | |
052f1dc7 | 8193 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8194 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8195 | INIT_LIST_HEAD(&init_task_group.children); |
8196 | ||
8197 | #ifdef CONFIG_USER_SCHED | |
8198 | INIT_LIST_HEAD(&root_task_group.children); | |
8199 | init_task_group.parent = &root_task_group; | |
8200 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8201 | #endif /* CONFIG_USER_SCHED */ |
8202 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8203 | |
0a945022 | 8204 | for_each_possible_cpu(i) { |
70b97a7f | 8205 | struct rq *rq; |
1da177e4 LT |
8206 | |
8207 | rq = cpu_rq(i); | |
8208 | spin_lock_init(&rq->lock); | |
7897986b | 8209 | rq->nr_running = 0; |
dd41f596 | 8210 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8211 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8212 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8213 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8214 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8215 | #ifdef CONFIG_CGROUP_SCHED |
8216 | /* | |
8217 | * How much cpu bandwidth does init_task_group get? | |
8218 | * | |
8219 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8220 | * gets 100% of the cpu resources in the system. This overall | |
8221 | * system cpu resource is divided among the tasks of | |
8222 | * init_task_group and its child task-groups in a fair manner, | |
8223 | * based on each entity's (task or task-group's) weight | |
8224 | * (se->load.weight). | |
8225 | * | |
8226 | * In other words, if init_task_group has 10 tasks of weight | |
8227 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8228 | * then A0's share of the cpu resource is: | |
8229 | * | |
8230 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8231 | * | |
8232 | * We achieve this by letting init_task_group's tasks sit | |
8233 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8234 | */ | |
ec7dc8ac | 8235 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8236 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8237 | root_task_group.shares = NICE_0_LOAD; |
8238 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8239 | /* |
8240 | * In case of task-groups formed thr' the user id of tasks, | |
8241 | * init_task_group represents tasks belonging to root user. | |
8242 | * Hence it forms a sibling of all subsequent groups formed. | |
8243 | * In this case, init_task_group gets only a fraction of overall | |
8244 | * system cpu resource, based on the weight assigned to root | |
8245 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8246 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8247 | * (init_cfs_rq) and having one entity represent this group of | |
8248 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8249 | */ | |
ec7dc8ac | 8250 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8251 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8252 | &per_cpu(init_sched_entity, i), i, 1, |
8253 | root_task_group.se[i]); | |
6f505b16 | 8254 | |
052f1dc7 | 8255 | #endif |
354d60c2 DG |
8256 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8257 | ||
8258 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8259 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8260 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8261 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8262 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8263 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8264 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8265 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8266 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8267 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8268 | root_task_group.rt_se[i]); | |
354d60c2 | 8269 | #endif |
dd41f596 | 8270 | #endif |
1da177e4 | 8271 | |
dd41f596 IM |
8272 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8273 | rq->cpu_load[j] = 0; | |
1da177e4 | 8274 | #ifdef CONFIG_SMP |
41c7ce9a | 8275 | rq->sd = NULL; |
57d885fe | 8276 | rq->rd = NULL; |
1da177e4 | 8277 | rq->active_balance = 0; |
dd41f596 | 8278 | rq->next_balance = jiffies; |
1da177e4 | 8279 | rq->push_cpu = 0; |
0a2966b4 | 8280 | rq->cpu = i; |
1f11eb6a | 8281 | rq->online = 0; |
1da177e4 LT |
8282 | rq->migration_thread = NULL; |
8283 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8284 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8285 | #endif |
8f4d37ec | 8286 | init_rq_hrtick(rq); |
1da177e4 | 8287 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8288 | } |
8289 | ||
2dd73a4f | 8290 | set_load_weight(&init_task); |
b50f60ce | 8291 | |
e107be36 AK |
8292 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8293 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8294 | #endif | |
8295 | ||
c9819f45 | 8296 | #ifdef CONFIG_SMP |
962cf36c | 8297 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8298 | #endif |
8299 | ||
b50f60ce HC |
8300 | #ifdef CONFIG_RT_MUTEXES |
8301 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8302 | #endif | |
8303 | ||
1da177e4 LT |
8304 | /* |
8305 | * The boot idle thread does lazy MMU switching as well: | |
8306 | */ | |
8307 | atomic_inc(&init_mm.mm_count); | |
8308 | enter_lazy_tlb(&init_mm, current); | |
8309 | ||
8310 | /* | |
8311 | * Make us the idle thread. Technically, schedule() should not be | |
8312 | * called from this thread, however somewhere below it might be, | |
8313 | * but because we are the idle thread, we just pick up running again | |
8314 | * when this runqueue becomes "idle". | |
8315 | */ | |
8316 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8317 | /* |
8318 | * During early bootup we pretend to be a normal task: | |
8319 | */ | |
8320 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8321 | |
8322 | scheduler_running = 1; | |
1da177e4 LT |
8323 | } |
8324 | ||
8325 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8326 | void __might_sleep(char *file, int line) | |
8327 | { | |
48f24c4d | 8328 | #ifdef in_atomic |
1da177e4 LT |
8329 | static unsigned long prev_jiffy; /* ratelimiting */ |
8330 | ||
aef745fc IM |
8331 | if ((!in_atomic() && !irqs_disabled()) || |
8332 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8333 | return; | |
8334 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8335 | return; | |
8336 | prev_jiffy = jiffies; | |
8337 | ||
8338 | printk(KERN_ERR | |
8339 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8340 | file, line); | |
8341 | printk(KERN_ERR | |
8342 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8343 | in_atomic(), irqs_disabled(), | |
8344 | current->pid, current->comm); | |
8345 | ||
8346 | debug_show_held_locks(current); | |
8347 | if (irqs_disabled()) | |
8348 | print_irqtrace_events(current); | |
8349 | dump_stack(); | |
1da177e4 LT |
8350 | #endif |
8351 | } | |
8352 | EXPORT_SYMBOL(__might_sleep); | |
8353 | #endif | |
8354 | ||
8355 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8356 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8357 | { | |
8358 | int on_rq; | |
3e51f33f | 8359 | |
3a5e4dc1 AK |
8360 | update_rq_clock(rq); |
8361 | on_rq = p->se.on_rq; | |
8362 | if (on_rq) | |
8363 | deactivate_task(rq, p, 0); | |
8364 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8365 | if (on_rq) { | |
8366 | activate_task(rq, p, 0); | |
8367 | resched_task(rq->curr); | |
8368 | } | |
8369 | } | |
8370 | ||
1da177e4 LT |
8371 | void normalize_rt_tasks(void) |
8372 | { | |
a0f98a1c | 8373 | struct task_struct *g, *p; |
1da177e4 | 8374 | unsigned long flags; |
70b97a7f | 8375 | struct rq *rq; |
1da177e4 | 8376 | |
4cf5d77a | 8377 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8378 | do_each_thread(g, p) { |
178be793 IM |
8379 | /* |
8380 | * Only normalize user tasks: | |
8381 | */ | |
8382 | if (!p->mm) | |
8383 | continue; | |
8384 | ||
6cfb0d5d | 8385 | p->se.exec_start = 0; |
6cfb0d5d | 8386 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8387 | p->se.wait_start = 0; |
dd41f596 | 8388 | p->se.sleep_start = 0; |
dd41f596 | 8389 | p->se.block_start = 0; |
6cfb0d5d | 8390 | #endif |
dd41f596 IM |
8391 | |
8392 | if (!rt_task(p)) { | |
8393 | /* | |
8394 | * Renice negative nice level userspace | |
8395 | * tasks back to 0: | |
8396 | */ | |
8397 | if (TASK_NICE(p) < 0 && p->mm) | |
8398 | set_user_nice(p, 0); | |
1da177e4 | 8399 | continue; |
dd41f596 | 8400 | } |
1da177e4 | 8401 | |
4cf5d77a | 8402 | spin_lock(&p->pi_lock); |
b29739f9 | 8403 | rq = __task_rq_lock(p); |
1da177e4 | 8404 | |
178be793 | 8405 | normalize_task(rq, p); |
3a5e4dc1 | 8406 | |
b29739f9 | 8407 | __task_rq_unlock(rq); |
4cf5d77a | 8408 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8409 | } while_each_thread(g, p); |
8410 | ||
4cf5d77a | 8411 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8412 | } |
8413 | ||
8414 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8415 | |
8416 | #ifdef CONFIG_IA64 | |
8417 | /* | |
8418 | * These functions are only useful for the IA64 MCA handling. | |
8419 | * | |
8420 | * They can only be called when the whole system has been | |
8421 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8422 | * activity can take place. Using them for anything else would | |
8423 | * be a serious bug, and as a result, they aren't even visible | |
8424 | * under any other configuration. | |
8425 | */ | |
8426 | ||
8427 | /** | |
8428 | * curr_task - return the current task for a given cpu. | |
8429 | * @cpu: the processor in question. | |
8430 | * | |
8431 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8432 | */ | |
36c8b586 | 8433 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8434 | { |
8435 | return cpu_curr(cpu); | |
8436 | } | |
8437 | ||
8438 | /** | |
8439 | * set_curr_task - set the current task for a given cpu. | |
8440 | * @cpu: the processor in question. | |
8441 | * @p: the task pointer to set. | |
8442 | * | |
8443 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8444 | * are serviced on a separate stack. It allows the architecture to switch the |
8445 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8446 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8447 | * and caller must save the original value of the current task (see | |
8448 | * curr_task() above) and restore that value before reenabling interrupts and | |
8449 | * re-starting the system. | |
8450 | * | |
8451 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8452 | */ | |
36c8b586 | 8453 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8454 | { |
8455 | cpu_curr(cpu) = p; | |
8456 | } | |
8457 | ||
8458 | #endif | |
29f59db3 | 8459 | |
bccbe08a PZ |
8460 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8461 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8462 | { |
8463 | int i; | |
8464 | ||
8465 | for_each_possible_cpu(i) { | |
8466 | if (tg->cfs_rq) | |
8467 | kfree(tg->cfs_rq[i]); | |
8468 | if (tg->se) | |
8469 | kfree(tg->se[i]); | |
6f505b16 PZ |
8470 | } |
8471 | ||
8472 | kfree(tg->cfs_rq); | |
8473 | kfree(tg->se); | |
6f505b16 PZ |
8474 | } |
8475 | ||
ec7dc8ac DG |
8476 | static |
8477 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8478 | { |
29f59db3 | 8479 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8480 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8481 | struct rq *rq; |
29f59db3 SV |
8482 | int i; |
8483 | ||
434d53b0 | 8484 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8485 | if (!tg->cfs_rq) |
8486 | goto err; | |
434d53b0 | 8487 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8488 | if (!tg->se) |
8489 | goto err; | |
052f1dc7 PZ |
8490 | |
8491 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8492 | |
8493 | for_each_possible_cpu(i) { | |
9b5b7751 | 8494 | rq = cpu_rq(i); |
29f59db3 | 8495 | |
6f505b16 PZ |
8496 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8497 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8498 | if (!cfs_rq) |
8499 | goto err; | |
8500 | ||
6f505b16 PZ |
8501 | se = kmalloc_node(sizeof(struct sched_entity), |
8502 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8503 | if (!se) |
8504 | goto err; | |
8505 | ||
ec7dc8ac DG |
8506 | parent_se = parent ? parent->se[i] : NULL; |
8507 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8508 | } |
8509 | ||
8510 | return 1; | |
8511 | ||
8512 | err: | |
8513 | return 0; | |
8514 | } | |
8515 | ||
8516 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8517 | { | |
8518 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8519 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8520 | } | |
8521 | ||
8522 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8523 | { | |
8524 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8525 | } | |
6d6bc0ad | 8526 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8527 | static inline void free_fair_sched_group(struct task_group *tg) |
8528 | { | |
8529 | } | |
8530 | ||
ec7dc8ac DG |
8531 | static inline |
8532 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8533 | { |
8534 | return 1; | |
8535 | } | |
8536 | ||
8537 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8538 | { | |
8539 | } | |
8540 | ||
8541 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8542 | { | |
8543 | } | |
6d6bc0ad | 8544 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8545 | |
8546 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8547 | static void free_rt_sched_group(struct task_group *tg) |
8548 | { | |
8549 | int i; | |
8550 | ||
d0b27fa7 PZ |
8551 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8552 | ||
bccbe08a PZ |
8553 | for_each_possible_cpu(i) { |
8554 | if (tg->rt_rq) | |
8555 | kfree(tg->rt_rq[i]); | |
8556 | if (tg->rt_se) | |
8557 | kfree(tg->rt_se[i]); | |
8558 | } | |
8559 | ||
8560 | kfree(tg->rt_rq); | |
8561 | kfree(tg->rt_se); | |
8562 | } | |
8563 | ||
ec7dc8ac DG |
8564 | static |
8565 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8566 | { |
8567 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8568 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8569 | struct rq *rq; |
8570 | int i; | |
8571 | ||
434d53b0 | 8572 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8573 | if (!tg->rt_rq) |
8574 | goto err; | |
434d53b0 | 8575 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8576 | if (!tg->rt_se) |
8577 | goto err; | |
8578 | ||
d0b27fa7 PZ |
8579 | init_rt_bandwidth(&tg->rt_bandwidth, |
8580 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8581 | |
8582 | for_each_possible_cpu(i) { | |
8583 | rq = cpu_rq(i); | |
8584 | ||
6f505b16 PZ |
8585 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8586 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8587 | if (!rt_rq) | |
8588 | goto err; | |
29f59db3 | 8589 | |
6f505b16 PZ |
8590 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8591 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8592 | if (!rt_se) | |
8593 | goto err; | |
29f59db3 | 8594 | |
ec7dc8ac DG |
8595 | parent_se = parent ? parent->rt_se[i] : NULL; |
8596 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8597 | } |
8598 | ||
bccbe08a PZ |
8599 | return 1; |
8600 | ||
8601 | err: | |
8602 | return 0; | |
8603 | } | |
8604 | ||
8605 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8606 | { | |
8607 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8608 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8609 | } | |
8610 | ||
8611 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8612 | { | |
8613 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8614 | } | |
6d6bc0ad | 8615 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8616 | static inline void free_rt_sched_group(struct task_group *tg) |
8617 | { | |
8618 | } | |
8619 | ||
ec7dc8ac DG |
8620 | static inline |
8621 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8622 | { |
8623 | return 1; | |
8624 | } | |
8625 | ||
8626 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8627 | { | |
8628 | } | |
8629 | ||
8630 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8631 | { | |
8632 | } | |
6d6bc0ad | 8633 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8634 | |
d0b27fa7 | 8635 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8636 | static void free_sched_group(struct task_group *tg) |
8637 | { | |
8638 | free_fair_sched_group(tg); | |
8639 | free_rt_sched_group(tg); | |
8640 | kfree(tg); | |
8641 | } | |
8642 | ||
8643 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8644 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8645 | { |
8646 | struct task_group *tg; | |
8647 | unsigned long flags; | |
8648 | int i; | |
8649 | ||
8650 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8651 | if (!tg) | |
8652 | return ERR_PTR(-ENOMEM); | |
8653 | ||
ec7dc8ac | 8654 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8655 | goto err; |
8656 | ||
ec7dc8ac | 8657 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8658 | goto err; |
8659 | ||
8ed36996 | 8660 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8661 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8662 | register_fair_sched_group(tg, i); |
8663 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8664 | } |
6f505b16 | 8665 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8666 | |
8667 | WARN_ON(!parent); /* root should already exist */ | |
8668 | ||
8669 | tg->parent = parent; | |
f473aa5e | 8670 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8671 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8672 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8673 | |
9b5b7751 | 8674 | return tg; |
29f59db3 SV |
8675 | |
8676 | err: | |
6f505b16 | 8677 | free_sched_group(tg); |
29f59db3 SV |
8678 | return ERR_PTR(-ENOMEM); |
8679 | } | |
8680 | ||
9b5b7751 | 8681 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8682 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8683 | { |
29f59db3 | 8684 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8685 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8686 | } |
8687 | ||
9b5b7751 | 8688 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8689 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8690 | { |
8ed36996 | 8691 | unsigned long flags; |
9b5b7751 | 8692 | int i; |
29f59db3 | 8693 | |
8ed36996 | 8694 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8695 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8696 | unregister_fair_sched_group(tg, i); |
8697 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8698 | } |
6f505b16 | 8699 | list_del_rcu(&tg->list); |
f473aa5e | 8700 | list_del_rcu(&tg->siblings); |
8ed36996 | 8701 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8702 | |
9b5b7751 | 8703 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8704 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8705 | } |
8706 | ||
9b5b7751 | 8707 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8708 | * The caller of this function should have put the task in its new group |
8709 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8710 | * reflect its new group. | |
9b5b7751 SV |
8711 | */ |
8712 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8713 | { |
8714 | int on_rq, running; | |
8715 | unsigned long flags; | |
8716 | struct rq *rq; | |
8717 | ||
8718 | rq = task_rq_lock(tsk, &flags); | |
8719 | ||
29f59db3 SV |
8720 | update_rq_clock(rq); |
8721 | ||
051a1d1a | 8722 | running = task_current(rq, tsk); |
29f59db3 SV |
8723 | on_rq = tsk->se.on_rq; |
8724 | ||
0e1f3483 | 8725 | if (on_rq) |
29f59db3 | 8726 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8727 | if (unlikely(running)) |
8728 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8729 | |
6f505b16 | 8730 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8731 | |
810b3817 PZ |
8732 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8733 | if (tsk->sched_class->moved_group) | |
8734 | tsk->sched_class->moved_group(tsk); | |
8735 | #endif | |
8736 | ||
0e1f3483 HS |
8737 | if (unlikely(running)) |
8738 | tsk->sched_class->set_curr_task(rq); | |
8739 | if (on_rq) | |
7074badb | 8740 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8741 | |
29f59db3 SV |
8742 | task_rq_unlock(rq, &flags); |
8743 | } | |
6d6bc0ad | 8744 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8745 | |
052f1dc7 | 8746 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8747 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8748 | { |
8749 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8750 | int on_rq; |
8751 | ||
29f59db3 | 8752 | on_rq = se->on_rq; |
62fb1851 | 8753 | if (on_rq) |
29f59db3 SV |
8754 | dequeue_entity(cfs_rq, se, 0); |
8755 | ||
8756 | se->load.weight = shares; | |
e05510d0 | 8757 | se->load.inv_weight = 0; |
29f59db3 | 8758 | |
62fb1851 | 8759 | if (on_rq) |
29f59db3 | 8760 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8761 | } |
62fb1851 | 8762 | |
c09595f6 PZ |
8763 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8764 | { | |
8765 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8766 | struct rq *rq = cfs_rq->rq; | |
8767 | unsigned long flags; | |
8768 | ||
8769 | spin_lock_irqsave(&rq->lock, flags); | |
8770 | __set_se_shares(se, shares); | |
8771 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8772 | } |
8773 | ||
8ed36996 PZ |
8774 | static DEFINE_MUTEX(shares_mutex); |
8775 | ||
4cf86d77 | 8776 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8777 | { |
8778 | int i; | |
8ed36996 | 8779 | unsigned long flags; |
c61935fd | 8780 | |
ec7dc8ac DG |
8781 | /* |
8782 | * We can't change the weight of the root cgroup. | |
8783 | */ | |
8784 | if (!tg->se[0]) | |
8785 | return -EINVAL; | |
8786 | ||
18d95a28 PZ |
8787 | if (shares < MIN_SHARES) |
8788 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8789 | else if (shares > MAX_SHARES) |
8790 | shares = MAX_SHARES; | |
62fb1851 | 8791 | |
8ed36996 | 8792 | mutex_lock(&shares_mutex); |
9b5b7751 | 8793 | if (tg->shares == shares) |
5cb350ba | 8794 | goto done; |
29f59db3 | 8795 | |
8ed36996 | 8796 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8797 | for_each_possible_cpu(i) |
8798 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8799 | list_del_rcu(&tg->siblings); |
8ed36996 | 8800 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8801 | |
8802 | /* wait for any ongoing reference to this group to finish */ | |
8803 | synchronize_sched(); | |
8804 | ||
8805 | /* | |
8806 | * Now we are free to modify the group's share on each cpu | |
8807 | * w/o tripping rebalance_share or load_balance_fair. | |
8808 | */ | |
9b5b7751 | 8809 | tg->shares = shares; |
c09595f6 PZ |
8810 | for_each_possible_cpu(i) { |
8811 | /* | |
8812 | * force a rebalance | |
8813 | */ | |
8814 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8815 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8816 | } |
29f59db3 | 8817 | |
6b2d7700 SV |
8818 | /* |
8819 | * Enable load balance activity on this group, by inserting it back on | |
8820 | * each cpu's rq->leaf_cfs_rq_list. | |
8821 | */ | |
8ed36996 | 8822 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8823 | for_each_possible_cpu(i) |
8824 | register_fair_sched_group(tg, i); | |
f473aa5e | 8825 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8826 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8827 | done: |
8ed36996 | 8828 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8829 | return 0; |
29f59db3 SV |
8830 | } |
8831 | ||
5cb350ba DG |
8832 | unsigned long sched_group_shares(struct task_group *tg) |
8833 | { | |
8834 | return tg->shares; | |
8835 | } | |
052f1dc7 | 8836 | #endif |
5cb350ba | 8837 | |
052f1dc7 | 8838 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8839 | /* |
9f0c1e56 | 8840 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8841 | */ |
9f0c1e56 PZ |
8842 | static DEFINE_MUTEX(rt_constraints_mutex); |
8843 | ||
8844 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8845 | { | |
8846 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8847 | return 1ULL << 20; |
9f0c1e56 | 8848 | |
9a7e0b18 | 8849 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8850 | } |
8851 | ||
9a7e0b18 PZ |
8852 | /* Must be called with tasklist_lock held */ |
8853 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8854 | { |
9a7e0b18 | 8855 | struct task_struct *g, *p; |
b40b2e8e | 8856 | |
9a7e0b18 PZ |
8857 | do_each_thread(g, p) { |
8858 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8859 | return 1; | |
8860 | } while_each_thread(g, p); | |
b40b2e8e | 8861 | |
9a7e0b18 PZ |
8862 | return 0; |
8863 | } | |
b40b2e8e | 8864 | |
9a7e0b18 PZ |
8865 | struct rt_schedulable_data { |
8866 | struct task_group *tg; | |
8867 | u64 rt_period; | |
8868 | u64 rt_runtime; | |
8869 | }; | |
b40b2e8e | 8870 | |
9a7e0b18 PZ |
8871 | static int tg_schedulable(struct task_group *tg, void *data) |
8872 | { | |
8873 | struct rt_schedulable_data *d = data; | |
8874 | struct task_group *child; | |
8875 | unsigned long total, sum = 0; | |
8876 | u64 period, runtime; | |
b40b2e8e | 8877 | |
9a7e0b18 PZ |
8878 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8879 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8880 | |
9a7e0b18 PZ |
8881 | if (tg == d->tg) { |
8882 | period = d->rt_period; | |
8883 | runtime = d->rt_runtime; | |
b40b2e8e | 8884 | } |
b40b2e8e | 8885 | |
4653f803 PZ |
8886 | /* |
8887 | * Cannot have more runtime than the period. | |
8888 | */ | |
8889 | if (runtime > period && runtime != RUNTIME_INF) | |
8890 | return -EINVAL; | |
6f505b16 | 8891 | |
4653f803 PZ |
8892 | /* |
8893 | * Ensure we don't starve existing RT tasks. | |
8894 | */ | |
9a7e0b18 PZ |
8895 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8896 | return -EBUSY; | |
6f505b16 | 8897 | |
9a7e0b18 | 8898 | total = to_ratio(period, runtime); |
6f505b16 | 8899 | |
4653f803 PZ |
8900 | /* |
8901 | * Nobody can have more than the global setting allows. | |
8902 | */ | |
8903 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8904 | return -EINVAL; | |
6f505b16 | 8905 | |
4653f803 PZ |
8906 | /* |
8907 | * The sum of our children's runtime should not exceed our own. | |
8908 | */ | |
9a7e0b18 PZ |
8909 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8910 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8911 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8912 | |
9a7e0b18 PZ |
8913 | if (child == d->tg) { |
8914 | period = d->rt_period; | |
8915 | runtime = d->rt_runtime; | |
8916 | } | |
6f505b16 | 8917 | |
9a7e0b18 | 8918 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8919 | } |
6f505b16 | 8920 | |
9a7e0b18 PZ |
8921 | if (sum > total) |
8922 | return -EINVAL; | |
8923 | ||
8924 | return 0; | |
6f505b16 PZ |
8925 | } |
8926 | ||
9a7e0b18 | 8927 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8928 | { |
9a7e0b18 PZ |
8929 | struct rt_schedulable_data data = { |
8930 | .tg = tg, | |
8931 | .rt_period = period, | |
8932 | .rt_runtime = runtime, | |
8933 | }; | |
8934 | ||
8935 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8936 | } |
8937 | ||
d0b27fa7 PZ |
8938 | static int tg_set_bandwidth(struct task_group *tg, |
8939 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8940 | { |
ac086bc2 | 8941 | int i, err = 0; |
9f0c1e56 | 8942 | |
9f0c1e56 | 8943 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8944 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8945 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8946 | if (err) | |
9f0c1e56 | 8947 | goto unlock; |
ac086bc2 PZ |
8948 | |
8949 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8950 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8951 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8952 | |
8953 | for_each_possible_cpu(i) { | |
8954 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8955 | ||
8956 | spin_lock(&rt_rq->rt_runtime_lock); | |
8957 | rt_rq->rt_runtime = rt_runtime; | |
8958 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8959 | } | |
8960 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8961 | unlock: |
521f1a24 | 8962 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8963 | mutex_unlock(&rt_constraints_mutex); |
8964 | ||
8965 | return err; | |
6f505b16 PZ |
8966 | } |
8967 | ||
d0b27fa7 PZ |
8968 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8969 | { | |
8970 | u64 rt_runtime, rt_period; | |
8971 | ||
8972 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8973 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8974 | if (rt_runtime_us < 0) | |
8975 | rt_runtime = RUNTIME_INF; | |
8976 | ||
8977 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8978 | } | |
8979 | ||
9f0c1e56 PZ |
8980 | long sched_group_rt_runtime(struct task_group *tg) |
8981 | { | |
8982 | u64 rt_runtime_us; | |
8983 | ||
d0b27fa7 | 8984 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8985 | return -1; |
8986 | ||
d0b27fa7 | 8987 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8988 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8989 | return rt_runtime_us; | |
8990 | } | |
d0b27fa7 PZ |
8991 | |
8992 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8993 | { | |
8994 | u64 rt_runtime, rt_period; | |
8995 | ||
8996 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8997 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8998 | ||
619b0488 R |
8999 | if (rt_period == 0) |
9000 | return -EINVAL; | |
9001 | ||
d0b27fa7 PZ |
9002 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9003 | } | |
9004 | ||
9005 | long sched_group_rt_period(struct task_group *tg) | |
9006 | { | |
9007 | u64 rt_period_us; | |
9008 | ||
9009 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9010 | do_div(rt_period_us, NSEC_PER_USEC); | |
9011 | return rt_period_us; | |
9012 | } | |
9013 | ||
9014 | static int sched_rt_global_constraints(void) | |
9015 | { | |
4653f803 | 9016 | u64 runtime, period; |
d0b27fa7 PZ |
9017 | int ret = 0; |
9018 | ||
ec5d4989 HS |
9019 | if (sysctl_sched_rt_period <= 0) |
9020 | return -EINVAL; | |
9021 | ||
4653f803 PZ |
9022 | runtime = global_rt_runtime(); |
9023 | period = global_rt_period(); | |
9024 | ||
9025 | /* | |
9026 | * Sanity check on the sysctl variables. | |
9027 | */ | |
9028 | if (runtime > period && runtime != RUNTIME_INF) | |
9029 | return -EINVAL; | |
10b612f4 | 9030 | |
d0b27fa7 | 9031 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9032 | read_lock(&tasklist_lock); |
4653f803 | 9033 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9034 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9035 | mutex_unlock(&rt_constraints_mutex); |
9036 | ||
9037 | return ret; | |
9038 | } | |
6d6bc0ad | 9039 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9040 | static int sched_rt_global_constraints(void) |
9041 | { | |
ac086bc2 PZ |
9042 | unsigned long flags; |
9043 | int i; | |
9044 | ||
ec5d4989 HS |
9045 | if (sysctl_sched_rt_period <= 0) |
9046 | return -EINVAL; | |
9047 | ||
ac086bc2 PZ |
9048 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9049 | for_each_possible_cpu(i) { | |
9050 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9051 | ||
9052 | spin_lock(&rt_rq->rt_runtime_lock); | |
9053 | rt_rq->rt_runtime = global_rt_runtime(); | |
9054 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9055 | } | |
9056 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9057 | ||
d0b27fa7 PZ |
9058 | return 0; |
9059 | } | |
6d6bc0ad | 9060 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9061 | |
9062 | int sched_rt_handler(struct ctl_table *table, int write, | |
9063 | struct file *filp, void __user *buffer, size_t *lenp, | |
9064 | loff_t *ppos) | |
9065 | { | |
9066 | int ret; | |
9067 | int old_period, old_runtime; | |
9068 | static DEFINE_MUTEX(mutex); | |
9069 | ||
9070 | mutex_lock(&mutex); | |
9071 | old_period = sysctl_sched_rt_period; | |
9072 | old_runtime = sysctl_sched_rt_runtime; | |
9073 | ||
9074 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9075 | ||
9076 | if (!ret && write) { | |
9077 | ret = sched_rt_global_constraints(); | |
9078 | if (ret) { | |
9079 | sysctl_sched_rt_period = old_period; | |
9080 | sysctl_sched_rt_runtime = old_runtime; | |
9081 | } else { | |
9082 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9083 | def_rt_bandwidth.rt_period = | |
9084 | ns_to_ktime(global_rt_period()); | |
9085 | } | |
9086 | } | |
9087 | mutex_unlock(&mutex); | |
9088 | ||
9089 | return ret; | |
9090 | } | |
68318b8e | 9091 | |
052f1dc7 | 9092 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9093 | |
9094 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9095 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9096 | { |
2b01dfe3 PM |
9097 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9098 | struct task_group, css); | |
68318b8e SV |
9099 | } |
9100 | ||
9101 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9102 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9103 | { |
ec7dc8ac | 9104 | struct task_group *tg, *parent; |
68318b8e | 9105 | |
2b01dfe3 | 9106 | if (!cgrp->parent) { |
68318b8e | 9107 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9108 | return &init_task_group.css; |
9109 | } | |
9110 | ||
ec7dc8ac DG |
9111 | parent = cgroup_tg(cgrp->parent); |
9112 | tg = sched_create_group(parent); | |
68318b8e SV |
9113 | if (IS_ERR(tg)) |
9114 | return ERR_PTR(-ENOMEM); | |
9115 | ||
68318b8e SV |
9116 | return &tg->css; |
9117 | } | |
9118 | ||
41a2d6cf IM |
9119 | static void |
9120 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9121 | { |
2b01dfe3 | 9122 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9123 | |
9124 | sched_destroy_group(tg); | |
9125 | } | |
9126 | ||
41a2d6cf IM |
9127 | static int |
9128 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9129 | struct task_struct *tsk) | |
68318b8e | 9130 | { |
b68aa230 PZ |
9131 | #ifdef CONFIG_RT_GROUP_SCHED |
9132 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9133 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9134 | return -EINVAL; |
9135 | #else | |
68318b8e SV |
9136 | /* We don't support RT-tasks being in separate groups */ |
9137 | if (tsk->sched_class != &fair_sched_class) | |
9138 | return -EINVAL; | |
b68aa230 | 9139 | #endif |
68318b8e SV |
9140 | |
9141 | return 0; | |
9142 | } | |
9143 | ||
9144 | static void | |
2b01dfe3 | 9145 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9146 | struct cgroup *old_cont, struct task_struct *tsk) |
9147 | { | |
9148 | sched_move_task(tsk); | |
9149 | } | |
9150 | ||
052f1dc7 | 9151 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9152 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9153 | u64 shareval) |
68318b8e | 9154 | { |
2b01dfe3 | 9155 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9156 | } |
9157 | ||
f4c753b7 | 9158 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9159 | { |
2b01dfe3 | 9160 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9161 | |
9162 | return (u64) tg->shares; | |
9163 | } | |
6d6bc0ad | 9164 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9165 | |
052f1dc7 | 9166 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9167 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9168 | s64 val) |
6f505b16 | 9169 | { |
06ecb27c | 9170 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9171 | } |
9172 | ||
06ecb27c | 9173 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9174 | { |
06ecb27c | 9175 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9176 | } |
d0b27fa7 PZ |
9177 | |
9178 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9179 | u64 rt_period_us) | |
9180 | { | |
9181 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9182 | } | |
9183 | ||
9184 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9185 | { | |
9186 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9187 | } | |
6d6bc0ad | 9188 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9189 | |
fe5c7cc2 | 9190 | static struct cftype cpu_files[] = { |
052f1dc7 | 9191 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9192 | { |
9193 | .name = "shares", | |
f4c753b7 PM |
9194 | .read_u64 = cpu_shares_read_u64, |
9195 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9196 | }, |
052f1dc7 PZ |
9197 | #endif |
9198 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9199 | { |
9f0c1e56 | 9200 | .name = "rt_runtime_us", |
06ecb27c PM |
9201 | .read_s64 = cpu_rt_runtime_read, |
9202 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9203 | }, |
d0b27fa7 PZ |
9204 | { |
9205 | .name = "rt_period_us", | |
f4c753b7 PM |
9206 | .read_u64 = cpu_rt_period_read_uint, |
9207 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9208 | }, |
052f1dc7 | 9209 | #endif |
68318b8e SV |
9210 | }; |
9211 | ||
9212 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9213 | { | |
fe5c7cc2 | 9214 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9215 | } |
9216 | ||
9217 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9218 | .name = "cpu", |
9219 | .create = cpu_cgroup_create, | |
9220 | .destroy = cpu_cgroup_destroy, | |
9221 | .can_attach = cpu_cgroup_can_attach, | |
9222 | .attach = cpu_cgroup_attach, | |
9223 | .populate = cpu_cgroup_populate, | |
9224 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9225 | .early_init = 1, |
9226 | }; | |
9227 | ||
052f1dc7 | 9228 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9229 | |
9230 | #ifdef CONFIG_CGROUP_CPUACCT | |
9231 | ||
9232 | /* | |
9233 | * CPU accounting code for task groups. | |
9234 | * | |
9235 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9236 | * (balbir@in.ibm.com). | |
9237 | */ | |
9238 | ||
9239 | /* track cpu usage of a group of tasks */ | |
9240 | struct cpuacct { | |
9241 | struct cgroup_subsys_state css; | |
9242 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9243 | u64 *cpuusage; | |
9244 | }; | |
9245 | ||
9246 | struct cgroup_subsys cpuacct_subsys; | |
9247 | ||
9248 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9249 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9250 | { |
32cd756a | 9251 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9252 | struct cpuacct, css); |
9253 | } | |
9254 | ||
9255 | /* return cpu accounting group to which this task belongs */ | |
9256 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9257 | { | |
9258 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9259 | struct cpuacct, css); | |
9260 | } | |
9261 | ||
9262 | /* create a new cpu accounting group */ | |
9263 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9264 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9265 | { |
9266 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9267 | ||
9268 | if (!ca) | |
9269 | return ERR_PTR(-ENOMEM); | |
9270 | ||
9271 | ca->cpuusage = alloc_percpu(u64); | |
9272 | if (!ca->cpuusage) { | |
9273 | kfree(ca); | |
9274 | return ERR_PTR(-ENOMEM); | |
9275 | } | |
9276 | ||
9277 | return &ca->css; | |
9278 | } | |
9279 | ||
9280 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9281 | static void |
32cd756a | 9282 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9283 | { |
32cd756a | 9284 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9285 | |
9286 | free_percpu(ca->cpuusage); | |
9287 | kfree(ca); | |
9288 | } | |
9289 | ||
9290 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9291 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9292 | { |
32cd756a | 9293 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9294 | u64 totalcpuusage = 0; |
9295 | int i; | |
9296 | ||
9297 | for_each_possible_cpu(i) { | |
9298 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9299 | ||
9300 | /* | |
9301 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9302 | * platforms. | |
9303 | */ | |
9304 | spin_lock_irq(&cpu_rq(i)->lock); | |
9305 | totalcpuusage += *cpuusage; | |
9306 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9307 | } | |
9308 | ||
9309 | return totalcpuusage; | |
9310 | } | |
9311 | ||
0297b803 DG |
9312 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9313 | u64 reset) | |
9314 | { | |
9315 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9316 | int err = 0; | |
9317 | int i; | |
9318 | ||
9319 | if (reset) { | |
9320 | err = -EINVAL; | |
9321 | goto out; | |
9322 | } | |
9323 | ||
9324 | for_each_possible_cpu(i) { | |
9325 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9326 | ||
9327 | spin_lock_irq(&cpu_rq(i)->lock); | |
9328 | *cpuusage = 0; | |
9329 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9330 | } | |
9331 | out: | |
9332 | return err; | |
9333 | } | |
9334 | ||
d842de87 SV |
9335 | static struct cftype files[] = { |
9336 | { | |
9337 | .name = "usage", | |
f4c753b7 PM |
9338 | .read_u64 = cpuusage_read, |
9339 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9340 | }, |
9341 | }; | |
9342 | ||
32cd756a | 9343 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9344 | { |
32cd756a | 9345 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9346 | } |
9347 | ||
9348 | /* | |
9349 | * charge this task's execution time to its accounting group. | |
9350 | * | |
9351 | * called with rq->lock held. | |
9352 | */ | |
9353 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9354 | { | |
9355 | struct cpuacct *ca; | |
9356 | ||
9357 | if (!cpuacct_subsys.active) | |
9358 | return; | |
9359 | ||
9360 | ca = task_ca(tsk); | |
9361 | if (ca) { | |
9362 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9363 | ||
9364 | *cpuusage += cputime; | |
9365 | } | |
9366 | } | |
9367 | ||
9368 | struct cgroup_subsys cpuacct_subsys = { | |
9369 | .name = "cpuacct", | |
9370 | .create = cpuacct_create, | |
9371 | .destroy = cpuacct_destroy, | |
9372 | .populate = cpuacct_populate, | |
9373 | .subsys_id = cpuacct_subsys_id, | |
9374 | }; | |
9375 | #endif /* CONFIG_CGROUP_CPUACCT */ |