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); | |
cc584b21 AV |
229 | hrtimer_start_expires(&rt_b->rt_period_timer, |
230 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
231 | } |
232 | spin_unlock(&rt_b->rt_runtime_lock); | |
233 | } | |
234 | ||
235 | #ifdef CONFIG_RT_GROUP_SCHED | |
236 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
237 | { | |
238 | hrtimer_cancel(&rt_b->rt_period_timer); | |
239 | } | |
240 | #endif | |
241 | ||
712555ee HC |
242 | /* |
243 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
244 | * detach_destroy_domains and partition_sched_domains. | |
245 | */ | |
246 | static DEFINE_MUTEX(sched_domains_mutex); | |
247 | ||
052f1dc7 | 248 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 249 | |
68318b8e SV |
250 | #include <linux/cgroup.h> |
251 | ||
29f59db3 SV |
252 | struct cfs_rq; |
253 | ||
6f505b16 PZ |
254 | static LIST_HEAD(task_groups); |
255 | ||
29f59db3 | 256 | /* task group related information */ |
4cf86d77 | 257 | struct task_group { |
052f1dc7 | 258 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
259 | struct cgroup_subsys_state css; |
260 | #endif | |
052f1dc7 PZ |
261 | |
262 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
263 | /* schedulable entities of this group on each cpu */ |
264 | struct sched_entity **se; | |
265 | /* runqueue "owned" by this group on each cpu */ | |
266 | struct cfs_rq **cfs_rq; | |
267 | unsigned long shares; | |
052f1dc7 PZ |
268 | #endif |
269 | ||
270 | #ifdef CONFIG_RT_GROUP_SCHED | |
271 | struct sched_rt_entity **rt_se; | |
272 | struct rt_rq **rt_rq; | |
273 | ||
d0b27fa7 | 274 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 275 | #endif |
6b2d7700 | 276 | |
ae8393e5 | 277 | struct rcu_head rcu; |
6f505b16 | 278 | struct list_head list; |
f473aa5e PZ |
279 | |
280 | struct task_group *parent; | |
281 | struct list_head siblings; | |
282 | struct list_head children; | |
29f59db3 SV |
283 | }; |
284 | ||
354d60c2 | 285 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
286 | |
287 | /* | |
288 | * Root task group. | |
289 | * Every UID task group (including init_task_group aka UID-0) will | |
290 | * be a child to this group. | |
291 | */ | |
292 | struct task_group root_task_group; | |
293 | ||
052f1dc7 | 294 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
295 | /* Default task group's sched entity on each cpu */ |
296 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
297 | /* Default task group's cfs_rq on each cpu */ | |
298 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 299 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
300 | |
301 | #ifdef CONFIG_RT_GROUP_SCHED | |
302 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
303 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 304 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 305 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 306 | #define root_task_group init_task_group |
9a7e0b18 | 307 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 308 | |
8ed36996 | 309 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
310 | * a task group's cpu shares. |
311 | */ | |
8ed36996 | 312 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 313 | |
052f1dc7 | 314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
315 | #ifdef CONFIG_USER_SCHED |
316 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 317 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 318 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 319 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 320 | |
cb4ad1ff | 321 | /* |
2e084786 LJ |
322 | * A weight of 0 or 1 can cause arithmetics problems. |
323 | * A weight of a cfs_rq is the sum of weights of which entities | |
324 | * are queued on this cfs_rq, so a weight of a entity should not be | |
325 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
326 | * (The default weight is 1024 - so there's no practical |
327 | * limitation from this.) | |
328 | */ | |
18d95a28 | 329 | #define MIN_SHARES 2 |
2e084786 | 330 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 331 | |
052f1dc7 PZ |
332 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
333 | #endif | |
334 | ||
29f59db3 | 335 | /* Default task group. |
3a252015 | 336 | * Every task in system belong to this group at bootup. |
29f59db3 | 337 | */ |
434d53b0 | 338 | struct task_group init_task_group; |
29f59db3 SV |
339 | |
340 | /* return group to which a task belongs */ | |
4cf86d77 | 341 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 342 | { |
4cf86d77 | 343 | struct task_group *tg; |
9b5b7751 | 344 | |
052f1dc7 | 345 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 346 | tg = p->user->tg; |
052f1dc7 | 347 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
348 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
349 | struct task_group, css); | |
24e377a8 | 350 | #else |
41a2d6cf | 351 | tg = &init_task_group; |
24e377a8 | 352 | #endif |
9b5b7751 | 353 | return tg; |
29f59db3 SV |
354 | } |
355 | ||
356 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 357 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 358 | { |
052f1dc7 | 359 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
360 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
361 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 362 | #endif |
6f505b16 | 363 | |
052f1dc7 | 364 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
365 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
366 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 367 | #endif |
29f59db3 SV |
368 | } |
369 | ||
370 | #else | |
371 | ||
6f505b16 | 372 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
373 | static inline struct task_group *task_group(struct task_struct *p) |
374 | { | |
375 | return NULL; | |
376 | } | |
29f59db3 | 377 | |
052f1dc7 | 378 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 379 | |
6aa645ea IM |
380 | /* CFS-related fields in a runqueue */ |
381 | struct cfs_rq { | |
382 | struct load_weight load; | |
383 | unsigned long nr_running; | |
384 | ||
6aa645ea | 385 | u64 exec_clock; |
e9acbff6 | 386 | u64 min_vruntime; |
103638d9 | 387 | u64 pair_start; |
6aa645ea IM |
388 | |
389 | struct rb_root tasks_timeline; | |
390 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
391 | |
392 | struct list_head tasks; | |
393 | struct list_head *balance_iterator; | |
394 | ||
395 | /* | |
396 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
397 | * It is set to NULL otherwise (i.e when none are currently running). |
398 | */ | |
aa2ac252 | 399 | struct sched_entity *curr, *next; |
ddc97297 PZ |
400 | |
401 | unsigned long nr_spread_over; | |
402 | ||
62160e3f | 403 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
404 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
405 | ||
41a2d6cf IM |
406 | /* |
407 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
408 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
409 | * (like users, containers etc.) | |
410 | * | |
411 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
412 | * list is used during load balance. | |
413 | */ | |
41a2d6cf IM |
414 | struct list_head leaf_cfs_rq_list; |
415 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
416 | |
417 | #ifdef CONFIG_SMP | |
c09595f6 | 418 | /* |
c8cba857 | 419 | * the part of load.weight contributed by tasks |
c09595f6 | 420 | */ |
c8cba857 | 421 | unsigned long task_weight; |
c09595f6 | 422 | |
c8cba857 PZ |
423 | /* |
424 | * h_load = weight * f(tg) | |
425 | * | |
426 | * Where f(tg) is the recursive weight fraction assigned to | |
427 | * this group. | |
428 | */ | |
429 | unsigned long h_load; | |
c09595f6 | 430 | |
c8cba857 PZ |
431 | /* |
432 | * this cpu's part of tg->shares | |
433 | */ | |
434 | unsigned long shares; | |
f1d239f7 PZ |
435 | |
436 | /* | |
437 | * load.weight at the time we set shares | |
438 | */ | |
439 | unsigned long rq_weight; | |
c09595f6 | 440 | #endif |
6aa645ea IM |
441 | #endif |
442 | }; | |
1da177e4 | 443 | |
6aa645ea IM |
444 | /* Real-Time classes' related field in a runqueue: */ |
445 | struct rt_rq { | |
446 | struct rt_prio_array active; | |
63489e45 | 447 | unsigned long rt_nr_running; |
052f1dc7 | 448 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
449 | int highest_prio; /* highest queued rt task prio */ |
450 | #endif | |
fa85ae24 | 451 | #ifdef CONFIG_SMP |
73fe6aae | 452 | unsigned long rt_nr_migratory; |
a22d7fc1 | 453 | int overloaded; |
fa85ae24 | 454 | #endif |
6f505b16 | 455 | int rt_throttled; |
fa85ae24 | 456 | u64 rt_time; |
ac086bc2 | 457 | u64 rt_runtime; |
ea736ed5 | 458 | /* Nests inside the rq lock: */ |
ac086bc2 | 459 | spinlock_t rt_runtime_lock; |
6f505b16 | 460 | |
052f1dc7 | 461 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
462 | unsigned long rt_nr_boosted; |
463 | ||
6f505b16 PZ |
464 | struct rq *rq; |
465 | struct list_head leaf_rt_rq_list; | |
466 | struct task_group *tg; | |
467 | struct sched_rt_entity *rt_se; | |
468 | #endif | |
6aa645ea IM |
469 | }; |
470 | ||
57d885fe GH |
471 | #ifdef CONFIG_SMP |
472 | ||
473 | /* | |
474 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
475 | * variables. Each exclusive cpuset essentially defines an island domain by |
476 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
477 | * exclusive cpuset is created, we also create and attach a new root-domain |
478 | * object. | |
479 | * | |
57d885fe GH |
480 | */ |
481 | struct root_domain { | |
482 | atomic_t refcount; | |
483 | cpumask_t span; | |
484 | cpumask_t online; | |
637f5085 | 485 | |
0eab9146 | 486 | /* |
637f5085 GH |
487 | * The "RT overload" flag: it gets set if a CPU has more than |
488 | * one runnable RT task. | |
489 | */ | |
490 | cpumask_t rto_mask; | |
0eab9146 | 491 | atomic_t rto_count; |
6e0534f2 GH |
492 | #ifdef CONFIG_SMP |
493 | struct cpupri cpupri; | |
494 | #endif | |
57d885fe GH |
495 | }; |
496 | ||
dc938520 GH |
497 | /* |
498 | * By default the system creates a single root-domain with all cpus as | |
499 | * members (mimicking the global state we have today). | |
500 | */ | |
57d885fe GH |
501 | static struct root_domain def_root_domain; |
502 | ||
503 | #endif | |
504 | ||
1da177e4 LT |
505 | /* |
506 | * This is the main, per-CPU runqueue data structure. | |
507 | * | |
508 | * Locking rule: those places that want to lock multiple runqueues | |
509 | * (such as the load balancing or the thread migration code), lock | |
510 | * acquire operations must be ordered by ascending &runqueue. | |
511 | */ | |
70b97a7f | 512 | struct rq { |
d8016491 IM |
513 | /* runqueue lock: */ |
514 | spinlock_t lock; | |
1da177e4 LT |
515 | |
516 | /* | |
517 | * nr_running and cpu_load should be in the same cacheline because | |
518 | * remote CPUs use both these fields when doing load calculation. | |
519 | */ | |
520 | unsigned long nr_running; | |
6aa645ea IM |
521 | #define CPU_LOAD_IDX_MAX 5 |
522 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 523 | unsigned char idle_at_tick; |
46cb4b7c | 524 | #ifdef CONFIG_NO_HZ |
15934a37 | 525 | unsigned long last_tick_seen; |
46cb4b7c SS |
526 | unsigned char in_nohz_recently; |
527 | #endif | |
d8016491 IM |
528 | /* capture load from *all* tasks on this cpu: */ |
529 | struct load_weight load; | |
6aa645ea IM |
530 | unsigned long nr_load_updates; |
531 | u64 nr_switches; | |
532 | ||
533 | struct cfs_rq cfs; | |
6f505b16 | 534 | struct rt_rq rt; |
6f505b16 | 535 | |
6aa645ea | 536 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
537 | /* list of leaf cfs_rq on this cpu: */ |
538 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
539 | #endif |
540 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 541 | struct list_head leaf_rt_rq_list; |
1da177e4 | 542 | #endif |
1da177e4 LT |
543 | |
544 | /* | |
545 | * This is part of a global counter where only the total sum | |
546 | * over all CPUs matters. A task can increase this counter on | |
547 | * one CPU and if it got migrated afterwards it may decrease | |
548 | * it on another CPU. Always updated under the runqueue lock: | |
549 | */ | |
550 | unsigned long nr_uninterruptible; | |
551 | ||
36c8b586 | 552 | struct task_struct *curr, *idle; |
c9819f45 | 553 | unsigned long next_balance; |
1da177e4 | 554 | struct mm_struct *prev_mm; |
6aa645ea | 555 | |
3e51f33f | 556 | u64 clock; |
6aa645ea | 557 | |
1da177e4 LT |
558 | atomic_t nr_iowait; |
559 | ||
560 | #ifdef CONFIG_SMP | |
0eab9146 | 561 | struct root_domain *rd; |
1da177e4 LT |
562 | struct sched_domain *sd; |
563 | ||
564 | /* For active balancing */ | |
565 | int active_balance; | |
566 | int push_cpu; | |
d8016491 IM |
567 | /* cpu of this runqueue: */ |
568 | int cpu; | |
1f11eb6a | 569 | int online; |
1da177e4 | 570 | |
a8a51d5e | 571 | unsigned long avg_load_per_task; |
1da177e4 | 572 | |
36c8b586 | 573 | struct task_struct *migration_thread; |
1da177e4 LT |
574 | struct list_head migration_queue; |
575 | #endif | |
576 | ||
8f4d37ec | 577 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
578 | #ifdef CONFIG_SMP |
579 | int hrtick_csd_pending; | |
580 | struct call_single_data hrtick_csd; | |
581 | #endif | |
8f4d37ec PZ |
582 | struct hrtimer hrtick_timer; |
583 | #endif | |
584 | ||
1da177e4 LT |
585 | #ifdef CONFIG_SCHEDSTATS |
586 | /* latency stats */ | |
587 | struct sched_info rq_sched_info; | |
588 | ||
589 | /* sys_sched_yield() stats */ | |
480b9434 KC |
590 | unsigned int yld_exp_empty; |
591 | unsigned int yld_act_empty; | |
592 | unsigned int yld_both_empty; | |
593 | unsigned int yld_count; | |
1da177e4 LT |
594 | |
595 | /* schedule() stats */ | |
480b9434 KC |
596 | unsigned int sched_switch; |
597 | unsigned int sched_count; | |
598 | unsigned int sched_goidle; | |
1da177e4 LT |
599 | |
600 | /* try_to_wake_up() stats */ | |
480b9434 KC |
601 | unsigned int ttwu_count; |
602 | unsigned int ttwu_local; | |
b8efb561 IM |
603 | |
604 | /* BKL stats */ | |
480b9434 | 605 | unsigned int bkl_count; |
1da177e4 LT |
606 | #endif |
607 | }; | |
608 | ||
f34e3b61 | 609 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 610 | |
15afe09b | 611 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 612 | { |
15afe09b | 613 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
614 | } |
615 | ||
0a2966b4 CL |
616 | static inline int cpu_of(struct rq *rq) |
617 | { | |
618 | #ifdef CONFIG_SMP | |
619 | return rq->cpu; | |
620 | #else | |
621 | return 0; | |
622 | #endif | |
623 | } | |
624 | ||
674311d5 NP |
625 | /* |
626 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 627 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
628 | * |
629 | * The domain tree of any CPU may only be accessed from within | |
630 | * preempt-disabled sections. | |
631 | */ | |
48f24c4d IM |
632 | #define for_each_domain(cpu, __sd) \ |
633 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
634 | |
635 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
636 | #define this_rq() (&__get_cpu_var(runqueues)) | |
637 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
638 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
639 | ||
3e51f33f PZ |
640 | static inline void update_rq_clock(struct rq *rq) |
641 | { | |
642 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
643 | } | |
644 | ||
bf5c91ba IM |
645 | /* |
646 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
647 | */ | |
648 | #ifdef CONFIG_SCHED_DEBUG | |
649 | # define const_debug __read_mostly | |
650 | #else | |
651 | # define const_debug static const | |
652 | #endif | |
653 | ||
017730c1 IM |
654 | /** |
655 | * runqueue_is_locked | |
656 | * | |
657 | * Returns true if the current cpu runqueue is locked. | |
658 | * This interface allows printk to be called with the runqueue lock | |
659 | * held and know whether or not it is OK to wake up the klogd. | |
660 | */ | |
661 | int runqueue_is_locked(void) | |
662 | { | |
663 | int cpu = get_cpu(); | |
664 | struct rq *rq = cpu_rq(cpu); | |
665 | int ret; | |
666 | ||
667 | ret = spin_is_locked(&rq->lock); | |
668 | put_cpu(); | |
669 | return ret; | |
670 | } | |
671 | ||
bf5c91ba IM |
672 | /* |
673 | * Debugging: various feature bits | |
674 | */ | |
f00b45c1 PZ |
675 | |
676 | #define SCHED_FEAT(name, enabled) \ | |
677 | __SCHED_FEAT_##name , | |
678 | ||
bf5c91ba | 679 | enum { |
f00b45c1 | 680 | #include "sched_features.h" |
bf5c91ba IM |
681 | }; |
682 | ||
f00b45c1 PZ |
683 | #undef SCHED_FEAT |
684 | ||
685 | #define SCHED_FEAT(name, enabled) \ | |
686 | (1UL << __SCHED_FEAT_##name) * enabled | | |
687 | ||
bf5c91ba | 688 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
689 | #include "sched_features.h" |
690 | 0; | |
691 | ||
692 | #undef SCHED_FEAT | |
693 | ||
694 | #ifdef CONFIG_SCHED_DEBUG | |
695 | #define SCHED_FEAT(name, enabled) \ | |
696 | #name , | |
697 | ||
983ed7a6 | 698 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
699 | #include "sched_features.h" |
700 | NULL | |
701 | }; | |
702 | ||
703 | #undef SCHED_FEAT | |
704 | ||
983ed7a6 | 705 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
706 | { |
707 | filp->private_data = inode->i_private; | |
708 | return 0; | |
709 | } | |
710 | ||
711 | static ssize_t | |
712 | sched_feat_read(struct file *filp, char __user *ubuf, | |
713 | size_t cnt, loff_t *ppos) | |
714 | { | |
715 | char *buf; | |
716 | int r = 0; | |
717 | int len = 0; | |
718 | int i; | |
719 | ||
720 | for (i = 0; sched_feat_names[i]; i++) { | |
721 | len += strlen(sched_feat_names[i]); | |
722 | len += 4; | |
723 | } | |
724 | ||
725 | buf = kmalloc(len + 2, GFP_KERNEL); | |
726 | if (!buf) | |
727 | return -ENOMEM; | |
728 | ||
729 | for (i = 0; sched_feat_names[i]; i++) { | |
730 | if (sysctl_sched_features & (1UL << i)) | |
731 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
732 | else | |
c24b7c52 | 733 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
734 | } |
735 | ||
736 | r += sprintf(buf + r, "\n"); | |
737 | WARN_ON(r >= len + 2); | |
738 | ||
739 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
740 | ||
741 | kfree(buf); | |
742 | ||
743 | return r; | |
744 | } | |
745 | ||
746 | static ssize_t | |
747 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
748 | size_t cnt, loff_t *ppos) | |
749 | { | |
750 | char buf[64]; | |
751 | char *cmp = buf; | |
752 | int neg = 0; | |
753 | int i; | |
754 | ||
755 | if (cnt > 63) | |
756 | cnt = 63; | |
757 | ||
758 | if (copy_from_user(&buf, ubuf, cnt)) | |
759 | return -EFAULT; | |
760 | ||
761 | buf[cnt] = 0; | |
762 | ||
c24b7c52 | 763 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
764 | neg = 1; |
765 | cmp += 3; | |
766 | } | |
767 | ||
768 | for (i = 0; sched_feat_names[i]; i++) { | |
769 | int len = strlen(sched_feat_names[i]); | |
770 | ||
771 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
772 | if (neg) | |
773 | sysctl_sched_features &= ~(1UL << i); | |
774 | else | |
775 | sysctl_sched_features |= (1UL << i); | |
776 | break; | |
777 | } | |
778 | } | |
779 | ||
780 | if (!sched_feat_names[i]) | |
781 | return -EINVAL; | |
782 | ||
783 | filp->f_pos += cnt; | |
784 | ||
785 | return cnt; | |
786 | } | |
787 | ||
788 | static struct file_operations sched_feat_fops = { | |
789 | .open = sched_feat_open, | |
790 | .read = sched_feat_read, | |
791 | .write = sched_feat_write, | |
792 | }; | |
793 | ||
794 | static __init int sched_init_debug(void) | |
795 | { | |
f00b45c1 PZ |
796 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
797 | &sched_feat_fops); | |
798 | ||
799 | return 0; | |
800 | } | |
801 | late_initcall(sched_init_debug); | |
802 | ||
803 | #endif | |
804 | ||
805 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 806 | |
b82d9fdd PZ |
807 | /* |
808 | * Number of tasks to iterate in a single balance run. | |
809 | * Limited because this is done with IRQs disabled. | |
810 | */ | |
811 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
812 | ||
2398f2c6 PZ |
813 | /* |
814 | * ratelimit for updating the group shares. | |
55cd5340 | 815 | * default: 0.25ms |
2398f2c6 | 816 | */ |
55cd5340 | 817 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 818 | |
fa85ae24 | 819 | /* |
9f0c1e56 | 820 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
821 | * default: 1s |
822 | */ | |
9f0c1e56 | 823 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 824 | |
6892b75e IM |
825 | static __read_mostly int scheduler_running; |
826 | ||
9f0c1e56 PZ |
827 | /* |
828 | * part of the period that we allow rt tasks to run in us. | |
829 | * default: 0.95s | |
830 | */ | |
831 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 832 | |
d0b27fa7 PZ |
833 | static inline u64 global_rt_period(void) |
834 | { | |
835 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
836 | } | |
837 | ||
838 | static inline u64 global_rt_runtime(void) | |
839 | { | |
e26873bb | 840 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
841 | return RUNTIME_INF; |
842 | ||
843 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
844 | } | |
fa85ae24 | 845 | |
1da177e4 | 846 | #ifndef prepare_arch_switch |
4866cde0 NP |
847 | # define prepare_arch_switch(next) do { } while (0) |
848 | #endif | |
849 | #ifndef finish_arch_switch | |
850 | # define finish_arch_switch(prev) do { } while (0) | |
851 | #endif | |
852 | ||
051a1d1a DA |
853 | static inline int task_current(struct rq *rq, struct task_struct *p) |
854 | { | |
855 | return rq->curr == p; | |
856 | } | |
857 | ||
4866cde0 | 858 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 859 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 860 | { |
051a1d1a | 861 | return task_current(rq, p); |
4866cde0 NP |
862 | } |
863 | ||
70b97a7f | 864 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
865 | { |
866 | } | |
867 | ||
70b97a7f | 868 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 869 | { |
da04c035 IM |
870 | #ifdef CONFIG_DEBUG_SPINLOCK |
871 | /* this is a valid case when another task releases the spinlock */ | |
872 | rq->lock.owner = current; | |
873 | #endif | |
8a25d5de IM |
874 | /* |
875 | * If we are tracking spinlock dependencies then we have to | |
876 | * fix up the runqueue lock - which gets 'carried over' from | |
877 | * prev into current: | |
878 | */ | |
879 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
880 | ||
4866cde0 NP |
881 | spin_unlock_irq(&rq->lock); |
882 | } | |
883 | ||
884 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 885 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
886 | { |
887 | #ifdef CONFIG_SMP | |
888 | return p->oncpu; | |
889 | #else | |
051a1d1a | 890 | return task_current(rq, p); |
4866cde0 NP |
891 | #endif |
892 | } | |
893 | ||
70b97a7f | 894 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
895 | { |
896 | #ifdef CONFIG_SMP | |
897 | /* | |
898 | * We can optimise this out completely for !SMP, because the | |
899 | * SMP rebalancing from interrupt is the only thing that cares | |
900 | * here. | |
901 | */ | |
902 | next->oncpu = 1; | |
903 | #endif | |
904 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
905 | spin_unlock_irq(&rq->lock); | |
906 | #else | |
907 | spin_unlock(&rq->lock); | |
908 | #endif | |
909 | } | |
910 | ||
70b97a7f | 911 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
912 | { |
913 | #ifdef CONFIG_SMP | |
914 | /* | |
915 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
916 | * We must ensure this doesn't happen until the switch is completely | |
917 | * finished. | |
918 | */ | |
919 | smp_wmb(); | |
920 | prev->oncpu = 0; | |
921 | #endif | |
922 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
923 | local_irq_enable(); | |
1da177e4 | 924 | #endif |
4866cde0 NP |
925 | } |
926 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 927 | |
b29739f9 IM |
928 | /* |
929 | * __task_rq_lock - lock the runqueue a given task resides on. | |
930 | * Must be called interrupts disabled. | |
931 | */ | |
70b97a7f | 932 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
933 | __acquires(rq->lock) |
934 | { | |
3a5c359a AK |
935 | for (;;) { |
936 | struct rq *rq = task_rq(p); | |
937 | spin_lock(&rq->lock); | |
938 | if (likely(rq == task_rq(p))) | |
939 | return rq; | |
b29739f9 | 940 | spin_unlock(&rq->lock); |
b29739f9 | 941 | } |
b29739f9 IM |
942 | } |
943 | ||
1da177e4 LT |
944 | /* |
945 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 946 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
947 | * explicitly disabling preemption. |
948 | */ | |
70b97a7f | 949 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
950 | __acquires(rq->lock) |
951 | { | |
70b97a7f | 952 | struct rq *rq; |
1da177e4 | 953 | |
3a5c359a AK |
954 | for (;;) { |
955 | local_irq_save(*flags); | |
956 | rq = task_rq(p); | |
957 | spin_lock(&rq->lock); | |
958 | if (likely(rq == task_rq(p))) | |
959 | return rq; | |
1da177e4 | 960 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 961 | } |
1da177e4 LT |
962 | } |
963 | ||
a9957449 | 964 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
965 | __releases(rq->lock) |
966 | { | |
967 | spin_unlock(&rq->lock); | |
968 | } | |
969 | ||
70b97a7f | 970 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
971 | __releases(rq->lock) |
972 | { | |
973 | spin_unlock_irqrestore(&rq->lock, *flags); | |
974 | } | |
975 | ||
1da177e4 | 976 | /* |
cc2a73b5 | 977 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 978 | */ |
a9957449 | 979 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
980 | __acquires(rq->lock) |
981 | { | |
70b97a7f | 982 | struct rq *rq; |
1da177e4 LT |
983 | |
984 | local_irq_disable(); | |
985 | rq = this_rq(); | |
986 | spin_lock(&rq->lock); | |
987 | ||
988 | return rq; | |
989 | } | |
990 | ||
8f4d37ec PZ |
991 | #ifdef CONFIG_SCHED_HRTICK |
992 | /* | |
993 | * Use HR-timers to deliver accurate preemption points. | |
994 | * | |
995 | * Its all a bit involved since we cannot program an hrt while holding the | |
996 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
997 | * reschedule event. | |
998 | * | |
999 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1000 | * rq->lock. | |
1001 | */ | |
8f4d37ec PZ |
1002 | |
1003 | /* | |
1004 | * Use hrtick when: | |
1005 | * - enabled by features | |
1006 | * - hrtimer is actually high res | |
1007 | */ | |
1008 | static inline int hrtick_enabled(struct rq *rq) | |
1009 | { | |
1010 | if (!sched_feat(HRTICK)) | |
1011 | return 0; | |
ba42059f | 1012 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1013 | return 0; |
8f4d37ec PZ |
1014 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1015 | } | |
1016 | ||
8f4d37ec PZ |
1017 | static void hrtick_clear(struct rq *rq) |
1018 | { | |
1019 | if (hrtimer_active(&rq->hrtick_timer)) | |
1020 | hrtimer_cancel(&rq->hrtick_timer); | |
1021 | } | |
1022 | ||
8f4d37ec PZ |
1023 | /* |
1024 | * High-resolution timer tick. | |
1025 | * Runs from hardirq context with interrupts disabled. | |
1026 | */ | |
1027 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1028 | { | |
1029 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1030 | ||
1031 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1032 | ||
1033 | spin_lock(&rq->lock); | |
3e51f33f | 1034 | update_rq_clock(rq); |
8f4d37ec PZ |
1035 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1036 | spin_unlock(&rq->lock); | |
1037 | ||
1038 | return HRTIMER_NORESTART; | |
1039 | } | |
1040 | ||
95e904c7 | 1041 | #ifdef CONFIG_SMP |
31656519 PZ |
1042 | /* |
1043 | * called from hardirq (IPI) context | |
1044 | */ | |
1045 | static void __hrtick_start(void *arg) | |
b328ca18 | 1046 | { |
31656519 | 1047 | struct rq *rq = arg; |
b328ca18 | 1048 | |
31656519 PZ |
1049 | spin_lock(&rq->lock); |
1050 | hrtimer_restart(&rq->hrtick_timer); | |
1051 | rq->hrtick_csd_pending = 0; | |
1052 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1053 | } |
1054 | ||
31656519 PZ |
1055 | /* |
1056 | * Called to set the hrtick timer state. | |
1057 | * | |
1058 | * called with rq->lock held and irqs disabled | |
1059 | */ | |
1060 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1061 | { |
31656519 PZ |
1062 | struct hrtimer *timer = &rq->hrtick_timer; |
1063 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1064 | |
cc584b21 | 1065 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1066 | |
1067 | if (rq == this_rq()) { | |
1068 | hrtimer_restart(timer); | |
1069 | } else if (!rq->hrtick_csd_pending) { | |
1070 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1071 | rq->hrtick_csd_pending = 1; | |
1072 | } | |
b328ca18 PZ |
1073 | } |
1074 | ||
1075 | static int | |
1076 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1077 | { | |
1078 | int cpu = (int)(long)hcpu; | |
1079 | ||
1080 | switch (action) { | |
1081 | case CPU_UP_CANCELED: | |
1082 | case CPU_UP_CANCELED_FROZEN: | |
1083 | case CPU_DOWN_PREPARE: | |
1084 | case CPU_DOWN_PREPARE_FROZEN: | |
1085 | case CPU_DEAD: | |
1086 | case CPU_DEAD_FROZEN: | |
31656519 | 1087 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1088 | return NOTIFY_OK; |
1089 | } | |
1090 | ||
1091 | return NOTIFY_DONE; | |
1092 | } | |
1093 | ||
fa748203 | 1094 | static __init void init_hrtick(void) |
b328ca18 PZ |
1095 | { |
1096 | hotcpu_notifier(hotplug_hrtick, 0); | |
1097 | } | |
31656519 PZ |
1098 | #else |
1099 | /* | |
1100 | * Called to set the hrtick timer state. | |
1101 | * | |
1102 | * called with rq->lock held and irqs disabled | |
1103 | */ | |
1104 | static void hrtick_start(struct rq *rq, u64 delay) | |
1105 | { | |
1106 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1107 | } | |
b328ca18 | 1108 | |
006c75f1 | 1109 | static inline void init_hrtick(void) |
8f4d37ec | 1110 | { |
8f4d37ec | 1111 | } |
31656519 | 1112 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1113 | |
31656519 | 1114 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1115 | { |
31656519 PZ |
1116 | #ifdef CONFIG_SMP |
1117 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1118 | |
31656519 PZ |
1119 | rq->hrtick_csd.flags = 0; |
1120 | rq->hrtick_csd.func = __hrtick_start; | |
1121 | rq->hrtick_csd.info = rq; | |
1122 | #endif | |
8f4d37ec | 1123 | |
31656519 PZ |
1124 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1125 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1126 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1127 | } |
006c75f1 | 1128 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1129 | static inline void hrtick_clear(struct rq *rq) |
1130 | { | |
1131 | } | |
1132 | ||
8f4d37ec PZ |
1133 | static inline void init_rq_hrtick(struct rq *rq) |
1134 | { | |
1135 | } | |
1136 | ||
b328ca18 PZ |
1137 | static inline void init_hrtick(void) |
1138 | { | |
1139 | } | |
006c75f1 | 1140 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1141 | |
c24d20db IM |
1142 | /* |
1143 | * resched_task - mark a task 'to be rescheduled now'. | |
1144 | * | |
1145 | * On UP this means the setting of the need_resched flag, on SMP it | |
1146 | * might also involve a cross-CPU call to trigger the scheduler on | |
1147 | * the target CPU. | |
1148 | */ | |
1149 | #ifdef CONFIG_SMP | |
1150 | ||
1151 | #ifndef tsk_is_polling | |
1152 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1153 | #endif | |
1154 | ||
31656519 | 1155 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1156 | { |
1157 | int cpu; | |
1158 | ||
1159 | assert_spin_locked(&task_rq(p)->lock); | |
1160 | ||
31656519 | 1161 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1162 | return; |
1163 | ||
31656519 | 1164 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1165 | |
1166 | cpu = task_cpu(p); | |
1167 | if (cpu == smp_processor_id()) | |
1168 | return; | |
1169 | ||
1170 | /* NEED_RESCHED must be visible before we test polling */ | |
1171 | smp_mb(); | |
1172 | if (!tsk_is_polling(p)) | |
1173 | smp_send_reschedule(cpu); | |
1174 | } | |
1175 | ||
1176 | static void resched_cpu(int cpu) | |
1177 | { | |
1178 | struct rq *rq = cpu_rq(cpu); | |
1179 | unsigned long flags; | |
1180 | ||
1181 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1182 | return; | |
1183 | resched_task(cpu_curr(cpu)); | |
1184 | spin_unlock_irqrestore(&rq->lock, flags); | |
1185 | } | |
06d8308c TG |
1186 | |
1187 | #ifdef CONFIG_NO_HZ | |
1188 | /* | |
1189 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1190 | * idle CPU then this timer might expire before the next timer event | |
1191 | * which is scheduled to wake up that CPU. In case of a completely | |
1192 | * idle system the next event might even be infinite time into the | |
1193 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1194 | * leaves the inner idle loop so the newly added timer is taken into | |
1195 | * account when the CPU goes back to idle and evaluates the timer | |
1196 | * wheel for the next timer event. | |
1197 | */ | |
1198 | void wake_up_idle_cpu(int cpu) | |
1199 | { | |
1200 | struct rq *rq = cpu_rq(cpu); | |
1201 | ||
1202 | if (cpu == smp_processor_id()) | |
1203 | return; | |
1204 | ||
1205 | /* | |
1206 | * This is safe, as this function is called with the timer | |
1207 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1208 | * to idle and has not yet set rq->curr to idle then it will | |
1209 | * be serialized on the timer wheel base lock and take the new | |
1210 | * timer into account automatically. | |
1211 | */ | |
1212 | if (rq->curr != rq->idle) | |
1213 | return; | |
1214 | ||
1215 | /* | |
1216 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1217 | * lockless. The worst case is that the other CPU runs the | |
1218 | * idle task through an additional NOOP schedule() | |
1219 | */ | |
1220 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1221 | ||
1222 | /* NEED_RESCHED must be visible before we test polling */ | |
1223 | smp_mb(); | |
1224 | if (!tsk_is_polling(rq->idle)) | |
1225 | smp_send_reschedule(cpu); | |
1226 | } | |
6d6bc0ad | 1227 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1228 | |
6d6bc0ad | 1229 | #else /* !CONFIG_SMP */ |
31656519 | 1230 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1231 | { |
1232 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1233 | set_tsk_need_resched(p); |
c24d20db | 1234 | } |
6d6bc0ad | 1235 | #endif /* CONFIG_SMP */ |
c24d20db | 1236 | |
45bf76df IM |
1237 | #if BITS_PER_LONG == 32 |
1238 | # define WMULT_CONST (~0UL) | |
1239 | #else | |
1240 | # define WMULT_CONST (1UL << 32) | |
1241 | #endif | |
1242 | ||
1243 | #define WMULT_SHIFT 32 | |
1244 | ||
194081eb IM |
1245 | /* |
1246 | * Shift right and round: | |
1247 | */ | |
cf2ab469 | 1248 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1249 | |
a7be37ac PZ |
1250 | /* |
1251 | * delta *= weight / lw | |
1252 | */ | |
cb1c4fc9 | 1253 | static unsigned long |
45bf76df IM |
1254 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1255 | struct load_weight *lw) | |
1256 | { | |
1257 | u64 tmp; | |
1258 | ||
7a232e03 LJ |
1259 | if (!lw->inv_weight) { |
1260 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1261 | lw->inv_weight = 1; | |
1262 | else | |
1263 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1264 | / (lw->weight+1); | |
1265 | } | |
45bf76df IM |
1266 | |
1267 | tmp = (u64)delta_exec * weight; | |
1268 | /* | |
1269 | * Check whether we'd overflow the 64-bit multiplication: | |
1270 | */ | |
194081eb | 1271 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1272 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1273 | WMULT_SHIFT/2); |
1274 | else | |
cf2ab469 | 1275 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1276 | |
ecf691da | 1277 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1278 | } |
1279 | ||
1091985b | 1280 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1281 | { |
1282 | lw->weight += inc; | |
e89996ae | 1283 | lw->inv_weight = 0; |
45bf76df IM |
1284 | } |
1285 | ||
1091985b | 1286 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1287 | { |
1288 | lw->weight -= dec; | |
e89996ae | 1289 | lw->inv_weight = 0; |
45bf76df IM |
1290 | } |
1291 | ||
2dd73a4f PW |
1292 | /* |
1293 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1294 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1295 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1296 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1297 | * scaled version of the new time slice allocation that they receive on time |
1298 | * slice expiry etc. | |
1299 | */ | |
1300 | ||
dd41f596 IM |
1301 | #define WEIGHT_IDLEPRIO 2 |
1302 | #define WMULT_IDLEPRIO (1 << 31) | |
1303 | ||
1304 | /* | |
1305 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1306 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1307 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1308 | * that remained on nice 0. | |
1309 | * | |
1310 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1311 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1312 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1313 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1314 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1315 | */ |
1316 | static const int prio_to_weight[40] = { | |
254753dc IM |
1317 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1318 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1319 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1320 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1321 | /* 0 */ 1024, 820, 655, 526, 423, | |
1322 | /* 5 */ 335, 272, 215, 172, 137, | |
1323 | /* 10 */ 110, 87, 70, 56, 45, | |
1324 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1325 | }; |
1326 | ||
5714d2de IM |
1327 | /* |
1328 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1329 | * | |
1330 | * In cases where the weight does not change often, we can use the | |
1331 | * precalculated inverse to speed up arithmetics by turning divisions | |
1332 | * into multiplications: | |
1333 | */ | |
dd41f596 | 1334 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1335 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1336 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1337 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1338 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1339 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1340 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1341 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1342 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1343 | }; |
2dd73a4f | 1344 | |
dd41f596 IM |
1345 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1346 | ||
1347 | /* | |
1348 | * runqueue iterator, to support SMP load-balancing between different | |
1349 | * scheduling classes, without having to expose their internal data | |
1350 | * structures to the load-balancing proper: | |
1351 | */ | |
1352 | struct rq_iterator { | |
1353 | void *arg; | |
1354 | struct task_struct *(*start)(void *); | |
1355 | struct task_struct *(*next)(void *); | |
1356 | }; | |
1357 | ||
e1d1484f PW |
1358 | #ifdef CONFIG_SMP |
1359 | static unsigned long | |
1360 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1361 | unsigned long max_load_move, struct sched_domain *sd, | |
1362 | enum cpu_idle_type idle, int *all_pinned, | |
1363 | int *this_best_prio, struct rq_iterator *iterator); | |
1364 | ||
1365 | static int | |
1366 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1367 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1368 | struct rq_iterator *iterator); | |
e1d1484f | 1369 | #endif |
dd41f596 | 1370 | |
d842de87 SV |
1371 | #ifdef CONFIG_CGROUP_CPUACCT |
1372 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1373 | #else | |
1374 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1375 | #endif | |
1376 | ||
18d95a28 PZ |
1377 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1378 | { | |
1379 | update_load_add(&rq->load, load); | |
1380 | } | |
1381 | ||
1382 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1383 | { | |
1384 | update_load_sub(&rq->load, load); | |
1385 | } | |
1386 | ||
7940ca36 | 1387 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1388 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1389 | |
1390 | /* | |
1391 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1392 | * leaving it for the final time. | |
1393 | */ | |
eb755805 | 1394 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1395 | { |
1396 | struct task_group *parent, *child; | |
eb755805 | 1397 | int ret; |
c09595f6 PZ |
1398 | |
1399 | rcu_read_lock(); | |
1400 | parent = &root_task_group; | |
1401 | down: | |
eb755805 PZ |
1402 | ret = (*down)(parent, data); |
1403 | if (ret) | |
1404 | goto out_unlock; | |
c09595f6 PZ |
1405 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1406 | parent = child; | |
1407 | goto down; | |
1408 | ||
1409 | up: | |
1410 | continue; | |
1411 | } | |
eb755805 PZ |
1412 | ret = (*up)(parent, data); |
1413 | if (ret) | |
1414 | goto out_unlock; | |
c09595f6 PZ |
1415 | |
1416 | child = parent; | |
1417 | parent = parent->parent; | |
1418 | if (parent) | |
1419 | goto up; | |
eb755805 | 1420 | out_unlock: |
c09595f6 | 1421 | rcu_read_unlock(); |
eb755805 PZ |
1422 | |
1423 | return ret; | |
c09595f6 PZ |
1424 | } |
1425 | ||
eb755805 PZ |
1426 | static int tg_nop(struct task_group *tg, void *data) |
1427 | { | |
1428 | return 0; | |
c09595f6 | 1429 | } |
eb755805 PZ |
1430 | #endif |
1431 | ||
1432 | #ifdef CONFIG_SMP | |
1433 | static unsigned long source_load(int cpu, int type); | |
1434 | static unsigned long target_load(int cpu, int type); | |
1435 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1436 | ||
1437 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1438 | { | |
1439 | struct rq *rq = cpu_rq(cpu); | |
1440 | ||
1441 | if (rq->nr_running) | |
1442 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1443 | ||
1444 | return rq->avg_load_per_task; | |
1445 | } | |
1446 | ||
1447 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1448 | |
c09595f6 PZ |
1449 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1450 | ||
1451 | /* | |
1452 | * Calculate and set the cpu's group shares. | |
1453 | */ | |
1454 | static void | |
b6a86c74 | 1455 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1456 | unsigned long sd_shares, unsigned long sd_rq_weight) |
18d95a28 | 1457 | { |
c09595f6 PZ |
1458 | int boost = 0; |
1459 | unsigned long shares; | |
1460 | unsigned long rq_weight; | |
1461 | ||
c8cba857 | 1462 | if (!tg->se[cpu]) |
c09595f6 PZ |
1463 | return; |
1464 | ||
c8cba857 | 1465 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1466 | |
1467 | /* | |
1468 | * If there are currently no tasks on the cpu pretend there is one of | |
1469 | * average load so that when a new task gets to run here it will not | |
1470 | * get delayed by group starvation. | |
1471 | */ | |
1472 | if (!rq_weight) { | |
1473 | boost = 1; | |
1474 | rq_weight = NICE_0_LOAD; | |
1475 | } | |
1476 | ||
c8cba857 PZ |
1477 | if (unlikely(rq_weight > sd_rq_weight)) |
1478 | rq_weight = sd_rq_weight; | |
1479 | ||
c09595f6 PZ |
1480 | /* |
1481 | * \Sum shares * rq_weight | |
1482 | * shares = ----------------------- | |
1483 | * \Sum rq_weight | |
1484 | * | |
1485 | */ | |
c8cba857 | 1486 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1487 | |
1488 | /* | |
1489 | * record the actual number of shares, not the boosted amount. | |
1490 | */ | |
c8cba857 | 1491 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
f1d239f7 | 1492 | tg->cfs_rq[cpu]->rq_weight = rq_weight; |
c09595f6 PZ |
1493 | |
1494 | if (shares < MIN_SHARES) | |
1495 | shares = MIN_SHARES; | |
1496 | else if (shares > MAX_SHARES) | |
1497 | shares = MAX_SHARES; | |
1498 | ||
c8cba857 | 1499 | __set_se_shares(tg->se[cpu], shares); |
18d95a28 | 1500 | } |
c09595f6 PZ |
1501 | |
1502 | /* | |
c8cba857 PZ |
1503 | * Re-compute the task group their per cpu shares over the given domain. |
1504 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1505 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1506 | */ |
eb755805 | 1507 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1508 | { |
c8cba857 PZ |
1509 | unsigned long rq_weight = 0; |
1510 | unsigned long shares = 0; | |
eb755805 | 1511 | struct sched_domain *sd = data; |
c8cba857 | 1512 | int i; |
c09595f6 | 1513 | |
c8cba857 PZ |
1514 | for_each_cpu_mask(i, sd->span) { |
1515 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1516 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1517 | } |
c09595f6 | 1518 | |
c8cba857 PZ |
1519 | if ((!shares && rq_weight) || shares > tg->shares) |
1520 | shares = tg->shares; | |
1521 | ||
1522 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1523 | shares = tg->shares; | |
c09595f6 | 1524 | |
cd80917e PZ |
1525 | if (!rq_weight) |
1526 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1527 | ||
c09595f6 PZ |
1528 | for_each_cpu_mask(i, sd->span) { |
1529 | struct rq *rq = cpu_rq(i); | |
1530 | unsigned long flags; | |
1531 | ||
1532 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1533 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1534 | spin_unlock_irqrestore(&rq->lock, flags); |
1535 | } | |
eb755805 PZ |
1536 | |
1537 | return 0; | |
c09595f6 PZ |
1538 | } |
1539 | ||
1540 | /* | |
c8cba857 PZ |
1541 | * Compute the cpu's hierarchical load factor for each task group. |
1542 | * This needs to be done in a top-down fashion because the load of a child | |
1543 | * group is a fraction of its parents load. | |
c09595f6 | 1544 | */ |
eb755805 | 1545 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1546 | { |
c8cba857 | 1547 | unsigned long load; |
eb755805 | 1548 | long cpu = (long)data; |
c09595f6 | 1549 | |
c8cba857 PZ |
1550 | if (!tg->parent) { |
1551 | load = cpu_rq(cpu)->load.weight; | |
1552 | } else { | |
1553 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1554 | load *= tg->cfs_rq[cpu]->shares; | |
1555 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1556 | } | |
c09595f6 | 1557 | |
c8cba857 | 1558 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1559 | |
eb755805 | 1560 | return 0; |
c09595f6 PZ |
1561 | } |
1562 | ||
c8cba857 | 1563 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1564 | { |
2398f2c6 PZ |
1565 | u64 now = cpu_clock(raw_smp_processor_id()); |
1566 | s64 elapsed = now - sd->last_update; | |
1567 | ||
1568 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1569 | sd->last_update = now; | |
eb755805 | 1570 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1571 | } |
4d8d595d PZ |
1572 | } |
1573 | ||
3e5459b4 PZ |
1574 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1575 | { | |
1576 | spin_unlock(&rq->lock); | |
1577 | update_shares(sd); | |
1578 | spin_lock(&rq->lock); | |
1579 | } | |
1580 | ||
eb755805 | 1581 | static void update_h_load(long cpu) |
c09595f6 | 1582 | { |
eb755805 | 1583 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1584 | } |
1585 | ||
c09595f6 PZ |
1586 | #else |
1587 | ||
c8cba857 | 1588 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1589 | { |
1590 | } | |
1591 | ||
3e5459b4 PZ |
1592 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1593 | { | |
1594 | } | |
1595 | ||
18d95a28 PZ |
1596 | #endif |
1597 | ||
18d95a28 PZ |
1598 | #endif |
1599 | ||
30432094 | 1600 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1601 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1602 | { | |
30432094 | 1603 | #ifdef CONFIG_SMP |
34e83e85 IM |
1604 | cfs_rq->shares = shares; |
1605 | #endif | |
1606 | } | |
30432094 | 1607 | #endif |
e7693a36 | 1608 | |
dd41f596 | 1609 | #include "sched_stats.h" |
dd41f596 | 1610 | #include "sched_idletask.c" |
5522d5d5 IM |
1611 | #include "sched_fair.c" |
1612 | #include "sched_rt.c" | |
dd41f596 IM |
1613 | #ifdef CONFIG_SCHED_DEBUG |
1614 | # include "sched_debug.c" | |
1615 | #endif | |
1616 | ||
1617 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1618 | #define for_each_class(class) \ |
1619 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1620 | |
c09595f6 | 1621 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1622 | { |
1623 | rq->nr_running++; | |
9c217245 IM |
1624 | } |
1625 | ||
c09595f6 | 1626 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1627 | { |
1628 | rq->nr_running--; | |
9c217245 IM |
1629 | } |
1630 | ||
45bf76df IM |
1631 | static void set_load_weight(struct task_struct *p) |
1632 | { | |
1633 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1634 | p->se.load.weight = prio_to_weight[0] * 2; |
1635 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1636 | return; | |
1637 | } | |
45bf76df | 1638 | |
dd41f596 IM |
1639 | /* |
1640 | * SCHED_IDLE tasks get minimal weight: | |
1641 | */ | |
1642 | if (p->policy == SCHED_IDLE) { | |
1643 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1644 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1645 | return; | |
1646 | } | |
71f8bd46 | 1647 | |
dd41f596 IM |
1648 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1649 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1650 | } |
1651 | ||
2087a1ad GH |
1652 | static void update_avg(u64 *avg, u64 sample) |
1653 | { | |
1654 | s64 diff = sample - *avg; | |
1655 | *avg += diff >> 3; | |
1656 | } | |
1657 | ||
8159f87e | 1658 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1659 | { |
dd41f596 | 1660 | sched_info_queued(p); |
fd390f6a | 1661 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1662 | p->se.on_rq = 1; |
71f8bd46 IM |
1663 | } |
1664 | ||
69be72c1 | 1665 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1666 | { |
2087a1ad GH |
1667 | if (sleep && p->se.last_wakeup) { |
1668 | update_avg(&p->se.avg_overlap, | |
1669 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1670 | p->se.last_wakeup = 0; | |
1671 | } | |
1672 | ||
46ac22ba | 1673 | sched_info_dequeued(p); |
f02231e5 | 1674 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1675 | p->se.on_rq = 0; |
71f8bd46 IM |
1676 | } |
1677 | ||
14531189 | 1678 | /* |
dd41f596 | 1679 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1680 | */ |
14531189 IM |
1681 | static inline int __normal_prio(struct task_struct *p) |
1682 | { | |
dd41f596 | 1683 | return p->static_prio; |
14531189 IM |
1684 | } |
1685 | ||
b29739f9 IM |
1686 | /* |
1687 | * Calculate the expected normal priority: i.e. priority | |
1688 | * without taking RT-inheritance into account. Might be | |
1689 | * boosted by interactivity modifiers. Changes upon fork, | |
1690 | * setprio syscalls, and whenever the interactivity | |
1691 | * estimator recalculates. | |
1692 | */ | |
36c8b586 | 1693 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1694 | { |
1695 | int prio; | |
1696 | ||
e05606d3 | 1697 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1698 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1699 | else | |
1700 | prio = __normal_prio(p); | |
1701 | return prio; | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * Calculate the current priority, i.e. the priority | |
1706 | * taken into account by the scheduler. This value might | |
1707 | * be boosted by RT tasks, or might be boosted by | |
1708 | * interactivity modifiers. Will be RT if the task got | |
1709 | * RT-boosted. If not then it returns p->normal_prio. | |
1710 | */ | |
36c8b586 | 1711 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1712 | { |
1713 | p->normal_prio = normal_prio(p); | |
1714 | /* | |
1715 | * If we are RT tasks or we were boosted to RT priority, | |
1716 | * keep the priority unchanged. Otherwise, update priority | |
1717 | * to the normal priority: | |
1718 | */ | |
1719 | if (!rt_prio(p->prio)) | |
1720 | return p->normal_prio; | |
1721 | return p->prio; | |
1722 | } | |
1723 | ||
1da177e4 | 1724 | /* |
dd41f596 | 1725 | * activate_task - move a task to the runqueue. |
1da177e4 | 1726 | */ |
dd41f596 | 1727 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1728 | { |
d9514f6c | 1729 | if (task_contributes_to_load(p)) |
dd41f596 | 1730 | rq->nr_uninterruptible--; |
1da177e4 | 1731 | |
8159f87e | 1732 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1733 | inc_nr_running(rq); |
1da177e4 LT |
1734 | } |
1735 | ||
1da177e4 LT |
1736 | /* |
1737 | * deactivate_task - remove a task from the runqueue. | |
1738 | */ | |
2e1cb74a | 1739 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1740 | { |
d9514f6c | 1741 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1742 | rq->nr_uninterruptible++; |
1743 | ||
69be72c1 | 1744 | dequeue_task(rq, p, sleep); |
c09595f6 | 1745 | dec_nr_running(rq); |
1da177e4 LT |
1746 | } |
1747 | ||
1da177e4 LT |
1748 | /** |
1749 | * task_curr - is this task currently executing on a CPU? | |
1750 | * @p: the task in question. | |
1751 | */ | |
36c8b586 | 1752 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1753 | { |
1754 | return cpu_curr(task_cpu(p)) == p; | |
1755 | } | |
1756 | ||
dd41f596 IM |
1757 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1758 | { | |
6f505b16 | 1759 | set_task_rq(p, cpu); |
dd41f596 | 1760 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1761 | /* |
1762 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1763 | * successfuly executed on another CPU. We must ensure that updates of | |
1764 | * per-task data have been completed by this moment. | |
1765 | */ | |
1766 | smp_wmb(); | |
dd41f596 | 1767 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1768 | #endif |
2dd73a4f PW |
1769 | } |
1770 | ||
cb469845 SR |
1771 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1772 | const struct sched_class *prev_class, | |
1773 | int oldprio, int running) | |
1774 | { | |
1775 | if (prev_class != p->sched_class) { | |
1776 | if (prev_class->switched_from) | |
1777 | prev_class->switched_from(rq, p, running); | |
1778 | p->sched_class->switched_to(rq, p, running); | |
1779 | } else | |
1780 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1781 | } | |
1782 | ||
1da177e4 | 1783 | #ifdef CONFIG_SMP |
c65cc870 | 1784 | |
e958b360 TG |
1785 | /* Used instead of source_load when we know the type == 0 */ |
1786 | static unsigned long weighted_cpuload(const int cpu) | |
1787 | { | |
1788 | return cpu_rq(cpu)->load.weight; | |
1789 | } | |
1790 | ||
cc367732 IM |
1791 | /* |
1792 | * Is this task likely cache-hot: | |
1793 | */ | |
e7693a36 | 1794 | static int |
cc367732 IM |
1795 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1796 | { | |
1797 | s64 delta; | |
1798 | ||
f540a608 IM |
1799 | /* |
1800 | * Buddy candidates are cache hot: | |
1801 | */ | |
d25ce4cd | 1802 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1803 | return 1; |
1804 | ||
cc367732 IM |
1805 | if (p->sched_class != &fair_sched_class) |
1806 | return 0; | |
1807 | ||
6bc1665b IM |
1808 | if (sysctl_sched_migration_cost == -1) |
1809 | return 1; | |
1810 | if (sysctl_sched_migration_cost == 0) | |
1811 | return 0; | |
1812 | ||
cc367732 IM |
1813 | delta = now - p->se.exec_start; |
1814 | ||
1815 | return delta < (s64)sysctl_sched_migration_cost; | |
1816 | } | |
1817 | ||
1818 | ||
dd41f596 | 1819 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1820 | { |
dd41f596 IM |
1821 | int old_cpu = task_cpu(p); |
1822 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1823 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1824 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1825 | u64 clock_offset; |
dd41f596 IM |
1826 | |
1827 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1828 | |
1829 | #ifdef CONFIG_SCHEDSTATS | |
1830 | if (p->se.wait_start) | |
1831 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1832 | if (p->se.sleep_start) |
1833 | p->se.sleep_start -= clock_offset; | |
1834 | if (p->se.block_start) | |
1835 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1836 | if (old_cpu != new_cpu) { |
1837 | schedstat_inc(p, se.nr_migrations); | |
1838 | if (task_hot(p, old_rq->clock, NULL)) | |
1839 | schedstat_inc(p, se.nr_forced2_migrations); | |
1840 | } | |
6cfb0d5d | 1841 | #endif |
2830cf8c SV |
1842 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1843 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1844 | |
1845 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1846 | } |
1847 | ||
70b97a7f | 1848 | struct migration_req { |
1da177e4 | 1849 | struct list_head list; |
1da177e4 | 1850 | |
36c8b586 | 1851 | struct task_struct *task; |
1da177e4 LT |
1852 | int dest_cpu; |
1853 | ||
1da177e4 | 1854 | struct completion done; |
70b97a7f | 1855 | }; |
1da177e4 LT |
1856 | |
1857 | /* | |
1858 | * The task's runqueue lock must be held. | |
1859 | * Returns true if you have to wait for migration thread. | |
1860 | */ | |
36c8b586 | 1861 | static int |
70b97a7f | 1862 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1863 | { |
70b97a7f | 1864 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1865 | |
1866 | /* | |
1867 | * If the task is not on a runqueue (and not running), then | |
1868 | * it is sufficient to simply update the task's cpu field. | |
1869 | */ | |
dd41f596 | 1870 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1871 | set_task_cpu(p, dest_cpu); |
1872 | return 0; | |
1873 | } | |
1874 | ||
1875 | init_completion(&req->done); | |
1da177e4 LT |
1876 | req->task = p; |
1877 | req->dest_cpu = dest_cpu; | |
1878 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1879 | |
1da177e4 LT |
1880 | return 1; |
1881 | } | |
1882 | ||
1883 | /* | |
1884 | * wait_task_inactive - wait for a thread to unschedule. | |
1885 | * | |
85ba2d86 RM |
1886 | * If @match_state is nonzero, it's the @p->state value just checked and |
1887 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1888 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1889 | * we return a positive number (its total switch count). If a second call | |
1890 | * a short while later returns the same number, the caller can be sure that | |
1891 | * @p has remained unscheduled the whole time. | |
1892 | * | |
1da177e4 LT |
1893 | * The caller must ensure that the task *will* unschedule sometime soon, |
1894 | * else this function might spin for a *long* time. This function can't | |
1895 | * be called with interrupts off, or it may introduce deadlock with | |
1896 | * smp_call_function() if an IPI is sent by the same process we are | |
1897 | * waiting to become inactive. | |
1898 | */ | |
85ba2d86 | 1899 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1900 | { |
1901 | unsigned long flags; | |
dd41f596 | 1902 | int running, on_rq; |
85ba2d86 | 1903 | unsigned long ncsw; |
70b97a7f | 1904 | struct rq *rq; |
1da177e4 | 1905 | |
3a5c359a AK |
1906 | for (;;) { |
1907 | /* | |
1908 | * We do the initial early heuristics without holding | |
1909 | * any task-queue locks at all. We'll only try to get | |
1910 | * the runqueue lock when things look like they will | |
1911 | * work out! | |
1912 | */ | |
1913 | rq = task_rq(p); | |
fa490cfd | 1914 | |
3a5c359a AK |
1915 | /* |
1916 | * If the task is actively running on another CPU | |
1917 | * still, just relax and busy-wait without holding | |
1918 | * any locks. | |
1919 | * | |
1920 | * NOTE! Since we don't hold any locks, it's not | |
1921 | * even sure that "rq" stays as the right runqueue! | |
1922 | * But we don't care, since "task_running()" will | |
1923 | * return false if the runqueue has changed and p | |
1924 | * is actually now running somewhere else! | |
1925 | */ | |
85ba2d86 RM |
1926 | while (task_running(rq, p)) { |
1927 | if (match_state && unlikely(p->state != match_state)) | |
1928 | return 0; | |
3a5c359a | 1929 | cpu_relax(); |
85ba2d86 | 1930 | } |
fa490cfd | 1931 | |
3a5c359a AK |
1932 | /* |
1933 | * Ok, time to look more closely! We need the rq | |
1934 | * lock now, to be *sure*. If we're wrong, we'll | |
1935 | * just go back and repeat. | |
1936 | */ | |
1937 | rq = task_rq_lock(p, &flags); | |
1938 | running = task_running(rq, p); | |
1939 | on_rq = p->se.on_rq; | |
85ba2d86 | 1940 | ncsw = 0; |
f31e11d8 | 1941 | if (!match_state || p->state == match_state) |
93dcf55f | 1942 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1943 | task_rq_unlock(rq, &flags); |
fa490cfd | 1944 | |
85ba2d86 RM |
1945 | /* |
1946 | * If it changed from the expected state, bail out now. | |
1947 | */ | |
1948 | if (unlikely(!ncsw)) | |
1949 | break; | |
1950 | ||
3a5c359a AK |
1951 | /* |
1952 | * Was it really running after all now that we | |
1953 | * checked with the proper locks actually held? | |
1954 | * | |
1955 | * Oops. Go back and try again.. | |
1956 | */ | |
1957 | if (unlikely(running)) { | |
1958 | cpu_relax(); | |
1959 | continue; | |
1960 | } | |
fa490cfd | 1961 | |
3a5c359a AK |
1962 | /* |
1963 | * It's not enough that it's not actively running, | |
1964 | * it must be off the runqueue _entirely_, and not | |
1965 | * preempted! | |
1966 | * | |
1967 | * So if it wa still runnable (but just not actively | |
1968 | * running right now), it's preempted, and we should | |
1969 | * yield - it could be a while. | |
1970 | */ | |
1971 | if (unlikely(on_rq)) { | |
1972 | schedule_timeout_uninterruptible(1); | |
1973 | continue; | |
1974 | } | |
fa490cfd | 1975 | |
3a5c359a AK |
1976 | /* |
1977 | * Ahh, all good. It wasn't running, and it wasn't | |
1978 | * runnable, which means that it will never become | |
1979 | * running in the future either. We're all done! | |
1980 | */ | |
1981 | break; | |
1982 | } | |
85ba2d86 RM |
1983 | |
1984 | return ncsw; | |
1da177e4 LT |
1985 | } |
1986 | ||
1987 | /*** | |
1988 | * kick_process - kick a running thread to enter/exit the kernel | |
1989 | * @p: the to-be-kicked thread | |
1990 | * | |
1991 | * Cause a process which is running on another CPU to enter | |
1992 | * kernel-mode, without any delay. (to get signals handled.) | |
1993 | * | |
1994 | * NOTE: this function doesnt have to take the runqueue lock, | |
1995 | * because all it wants to ensure is that the remote task enters | |
1996 | * the kernel. If the IPI races and the task has been migrated | |
1997 | * to another CPU then no harm is done and the purpose has been | |
1998 | * achieved as well. | |
1999 | */ | |
36c8b586 | 2000 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2001 | { |
2002 | int cpu; | |
2003 | ||
2004 | preempt_disable(); | |
2005 | cpu = task_cpu(p); | |
2006 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2007 | smp_send_reschedule(cpu); | |
2008 | preempt_enable(); | |
2009 | } | |
2010 | ||
2011 | /* | |
2dd73a4f PW |
2012 | * Return a low guess at the load of a migration-source cpu weighted |
2013 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2014 | * |
2015 | * We want to under-estimate the load of migration sources, to | |
2016 | * balance conservatively. | |
2017 | */ | |
a9957449 | 2018 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2019 | { |
70b97a7f | 2020 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2021 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2022 | |
93b75217 | 2023 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2024 | return total; |
b910472d | 2025 | |
dd41f596 | 2026 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2027 | } |
2028 | ||
2029 | /* | |
2dd73a4f PW |
2030 | * Return a high guess at the load of a migration-target cpu weighted |
2031 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2032 | */ |
a9957449 | 2033 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2034 | { |
70b97a7f | 2035 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2036 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2037 | |
93b75217 | 2038 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2039 | return total; |
3b0bd9bc | 2040 | |
dd41f596 | 2041 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2042 | } |
2043 | ||
147cbb4b NP |
2044 | /* |
2045 | * find_idlest_group finds and returns the least busy CPU group within the | |
2046 | * domain. | |
2047 | */ | |
2048 | static struct sched_group * | |
2049 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2050 | { | |
2051 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2052 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2053 | int load_idx = sd->forkexec_idx; | |
2054 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2055 | ||
2056 | do { | |
2057 | unsigned long load, avg_load; | |
2058 | int local_group; | |
2059 | int i; | |
2060 | ||
da5a5522 BD |
2061 | /* Skip over this group if it has no CPUs allowed */ |
2062 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2063 | continue; |
da5a5522 | 2064 | |
147cbb4b | 2065 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2066 | |
2067 | /* Tally up the load of all CPUs in the group */ | |
2068 | avg_load = 0; | |
2069 | ||
363ab6f1 | 2070 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2071 | /* Bias balancing toward cpus of our domain */ |
2072 | if (local_group) | |
2073 | load = source_load(i, load_idx); | |
2074 | else | |
2075 | load = target_load(i, load_idx); | |
2076 | ||
2077 | avg_load += load; | |
2078 | } | |
2079 | ||
2080 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2081 | avg_load = sg_div_cpu_power(group, |
2082 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2083 | |
2084 | if (local_group) { | |
2085 | this_load = avg_load; | |
2086 | this = group; | |
2087 | } else if (avg_load < min_load) { | |
2088 | min_load = avg_load; | |
2089 | idlest = group; | |
2090 | } | |
3a5c359a | 2091 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2092 | |
2093 | if (!idlest || 100*this_load < imbalance*min_load) | |
2094 | return NULL; | |
2095 | return idlest; | |
2096 | } | |
2097 | ||
2098 | /* | |
0feaece9 | 2099 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2100 | */ |
95cdf3b7 | 2101 | static int |
7c16ec58 MT |
2102 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2103 | cpumask_t *tmp) | |
147cbb4b NP |
2104 | { |
2105 | unsigned long load, min_load = ULONG_MAX; | |
2106 | int idlest = -1; | |
2107 | int i; | |
2108 | ||
da5a5522 | 2109 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2110 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2111 | |
363ab6f1 | 2112 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2113 | load = weighted_cpuload(i); |
147cbb4b NP |
2114 | |
2115 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2116 | min_load = load; | |
2117 | idlest = i; | |
2118 | } | |
2119 | } | |
2120 | ||
2121 | return idlest; | |
2122 | } | |
2123 | ||
476d139c NP |
2124 | /* |
2125 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2126 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2127 | * SD_BALANCE_EXEC. | |
2128 | * | |
2129 | * Balance, ie. select the least loaded group. | |
2130 | * | |
2131 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2132 | * | |
2133 | * preempt must be disabled. | |
2134 | */ | |
2135 | static int sched_balance_self(int cpu, int flag) | |
2136 | { | |
2137 | struct task_struct *t = current; | |
2138 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2139 | |
c96d145e | 2140 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2141 | /* |
2142 | * If power savings logic is enabled for a domain, stop there. | |
2143 | */ | |
5c45bf27 SS |
2144 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2145 | break; | |
476d139c NP |
2146 | if (tmp->flags & flag) |
2147 | sd = tmp; | |
c96d145e | 2148 | } |
476d139c | 2149 | |
039a1c41 PZ |
2150 | if (sd) |
2151 | update_shares(sd); | |
2152 | ||
476d139c | 2153 | while (sd) { |
7c16ec58 | 2154 | cpumask_t span, tmpmask; |
476d139c | 2155 | struct sched_group *group; |
1a848870 SS |
2156 | int new_cpu, weight; |
2157 | ||
2158 | if (!(sd->flags & flag)) { | |
2159 | sd = sd->child; | |
2160 | continue; | |
2161 | } | |
476d139c NP |
2162 | |
2163 | span = sd->span; | |
2164 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2165 | if (!group) { |
2166 | sd = sd->child; | |
2167 | continue; | |
2168 | } | |
476d139c | 2169 | |
7c16ec58 | 2170 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2171 | if (new_cpu == -1 || new_cpu == cpu) { |
2172 | /* Now try balancing at a lower domain level of cpu */ | |
2173 | sd = sd->child; | |
2174 | continue; | |
2175 | } | |
476d139c | 2176 | |
1a848870 | 2177 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2178 | cpu = new_cpu; |
476d139c NP |
2179 | sd = NULL; |
2180 | weight = cpus_weight(span); | |
2181 | for_each_domain(cpu, tmp) { | |
2182 | if (weight <= cpus_weight(tmp->span)) | |
2183 | break; | |
2184 | if (tmp->flags & flag) | |
2185 | sd = tmp; | |
2186 | } | |
2187 | /* while loop will break here if sd == NULL */ | |
2188 | } | |
2189 | ||
2190 | return cpu; | |
2191 | } | |
2192 | ||
2193 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2194 | |
1da177e4 LT |
2195 | /*** |
2196 | * try_to_wake_up - wake up a thread | |
2197 | * @p: the to-be-woken-up thread | |
2198 | * @state: the mask of task states that can be woken | |
2199 | * @sync: do a synchronous wakeup? | |
2200 | * | |
2201 | * Put it on the run-queue if it's not already there. The "current" | |
2202 | * thread is always on the run-queue (except when the actual | |
2203 | * re-schedule is in progress), and as such you're allowed to do | |
2204 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2205 | * runnable without the overhead of this. | |
2206 | * | |
2207 | * returns failure only if the task is already active. | |
2208 | */ | |
36c8b586 | 2209 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2210 | { |
cc367732 | 2211 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2212 | unsigned long flags; |
2213 | long old_state; | |
70b97a7f | 2214 | struct rq *rq; |
1da177e4 | 2215 | |
b85d0667 IM |
2216 | if (!sched_feat(SYNC_WAKEUPS)) |
2217 | sync = 0; | |
2218 | ||
2398f2c6 PZ |
2219 | #ifdef CONFIG_SMP |
2220 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2221 | struct sched_domain *sd; | |
2222 | ||
2223 | this_cpu = raw_smp_processor_id(); | |
2224 | cpu = task_cpu(p); | |
2225 | ||
2226 | for_each_domain(this_cpu, sd) { | |
2227 | if (cpu_isset(cpu, sd->span)) { | |
2228 | update_shares(sd); | |
2229 | break; | |
2230 | } | |
2231 | } | |
2232 | } | |
2233 | #endif | |
2234 | ||
04e2f174 | 2235 | smp_wmb(); |
1da177e4 LT |
2236 | rq = task_rq_lock(p, &flags); |
2237 | old_state = p->state; | |
2238 | if (!(old_state & state)) | |
2239 | goto out; | |
2240 | ||
dd41f596 | 2241 | if (p->se.on_rq) |
1da177e4 LT |
2242 | goto out_running; |
2243 | ||
2244 | cpu = task_cpu(p); | |
cc367732 | 2245 | orig_cpu = cpu; |
1da177e4 LT |
2246 | this_cpu = smp_processor_id(); |
2247 | ||
2248 | #ifdef CONFIG_SMP | |
2249 | if (unlikely(task_running(rq, p))) | |
2250 | goto out_activate; | |
2251 | ||
5d2f5a61 DA |
2252 | cpu = p->sched_class->select_task_rq(p, sync); |
2253 | if (cpu != orig_cpu) { | |
2254 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2255 | task_rq_unlock(rq, &flags); |
2256 | /* might preempt at this point */ | |
2257 | rq = task_rq_lock(p, &flags); | |
2258 | old_state = p->state; | |
2259 | if (!(old_state & state)) | |
2260 | goto out; | |
dd41f596 | 2261 | if (p->se.on_rq) |
1da177e4 LT |
2262 | goto out_running; |
2263 | ||
2264 | this_cpu = smp_processor_id(); | |
2265 | cpu = task_cpu(p); | |
2266 | } | |
2267 | ||
e7693a36 GH |
2268 | #ifdef CONFIG_SCHEDSTATS |
2269 | schedstat_inc(rq, ttwu_count); | |
2270 | if (cpu == this_cpu) | |
2271 | schedstat_inc(rq, ttwu_local); | |
2272 | else { | |
2273 | struct sched_domain *sd; | |
2274 | for_each_domain(this_cpu, sd) { | |
2275 | if (cpu_isset(cpu, sd->span)) { | |
2276 | schedstat_inc(sd, ttwu_wake_remote); | |
2277 | break; | |
2278 | } | |
2279 | } | |
2280 | } | |
6d6bc0ad | 2281 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2282 | |
1da177e4 LT |
2283 | out_activate: |
2284 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2285 | schedstat_inc(p, se.nr_wakeups); |
2286 | if (sync) | |
2287 | schedstat_inc(p, se.nr_wakeups_sync); | |
2288 | if (orig_cpu != cpu) | |
2289 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2290 | if (cpu == this_cpu) | |
2291 | schedstat_inc(p, se.nr_wakeups_local); | |
2292 | else | |
2293 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2294 | update_rq_clock(rq); |
dd41f596 | 2295 | activate_task(rq, p, 1); |
1da177e4 LT |
2296 | success = 1; |
2297 | ||
2298 | out_running: | |
5b82a1b0 MD |
2299 | trace_mark(kernel_sched_wakeup, |
2300 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2301 | p->pid, p->state, rq, p, rq->curr); | |
15afe09b | 2302 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2303 | |
1da177e4 | 2304 | p->state = TASK_RUNNING; |
9a897c5a SR |
2305 | #ifdef CONFIG_SMP |
2306 | if (p->sched_class->task_wake_up) | |
2307 | p->sched_class->task_wake_up(rq, p); | |
2308 | #endif | |
1da177e4 | 2309 | out: |
2087a1ad GH |
2310 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2311 | ||
1da177e4 LT |
2312 | task_rq_unlock(rq, &flags); |
2313 | ||
2314 | return success; | |
2315 | } | |
2316 | ||
7ad5b3a5 | 2317 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2318 | { |
d9514f6c | 2319 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2320 | } |
1da177e4 LT |
2321 | EXPORT_SYMBOL(wake_up_process); |
2322 | ||
7ad5b3a5 | 2323 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2324 | { |
2325 | return try_to_wake_up(p, state, 0); | |
2326 | } | |
2327 | ||
1da177e4 LT |
2328 | /* |
2329 | * Perform scheduler related setup for a newly forked process p. | |
2330 | * p is forked by current. | |
dd41f596 IM |
2331 | * |
2332 | * __sched_fork() is basic setup used by init_idle() too: | |
2333 | */ | |
2334 | static void __sched_fork(struct task_struct *p) | |
2335 | { | |
dd41f596 IM |
2336 | p->se.exec_start = 0; |
2337 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2338 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2339 | p->se.last_wakeup = 0; |
2340 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2341 | |
2342 | #ifdef CONFIG_SCHEDSTATS | |
2343 | p->se.wait_start = 0; | |
dd41f596 IM |
2344 | p->se.sum_sleep_runtime = 0; |
2345 | p->se.sleep_start = 0; | |
dd41f596 IM |
2346 | p->se.block_start = 0; |
2347 | p->se.sleep_max = 0; | |
2348 | p->se.block_max = 0; | |
2349 | p->se.exec_max = 0; | |
eba1ed4b | 2350 | p->se.slice_max = 0; |
dd41f596 | 2351 | p->se.wait_max = 0; |
6cfb0d5d | 2352 | #endif |
476d139c | 2353 | |
fa717060 | 2354 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2355 | p->se.on_rq = 0; |
4a55bd5e | 2356 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2357 | |
e107be36 AK |
2358 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2359 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2360 | #endif | |
2361 | ||
1da177e4 LT |
2362 | /* |
2363 | * We mark the process as running here, but have not actually | |
2364 | * inserted it onto the runqueue yet. This guarantees that | |
2365 | * nobody will actually run it, and a signal or other external | |
2366 | * event cannot wake it up and insert it on the runqueue either. | |
2367 | */ | |
2368 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2369 | } |
2370 | ||
2371 | /* | |
2372 | * fork()/clone()-time setup: | |
2373 | */ | |
2374 | void sched_fork(struct task_struct *p, int clone_flags) | |
2375 | { | |
2376 | int cpu = get_cpu(); | |
2377 | ||
2378 | __sched_fork(p); | |
2379 | ||
2380 | #ifdef CONFIG_SMP | |
2381 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2382 | #endif | |
02e4bac2 | 2383 | set_task_cpu(p, cpu); |
b29739f9 IM |
2384 | |
2385 | /* | |
2386 | * Make sure we do not leak PI boosting priority to the child: | |
2387 | */ | |
2388 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2389 | if (!rt_prio(p->prio)) |
2390 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2391 | |
52f17b6c | 2392 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2393 | if (likely(sched_info_on())) |
52f17b6c | 2394 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2395 | #endif |
d6077cb8 | 2396 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2397 | p->oncpu = 0; |
2398 | #endif | |
1da177e4 | 2399 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2400 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2401 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2402 | #endif |
476d139c | 2403 | put_cpu(); |
1da177e4 LT |
2404 | } |
2405 | ||
2406 | /* | |
2407 | * wake_up_new_task - wake up a newly created task for the first time. | |
2408 | * | |
2409 | * This function will do some initial scheduler statistics housekeeping | |
2410 | * that must be done for every newly created context, then puts the task | |
2411 | * on the runqueue and wakes it. | |
2412 | */ | |
7ad5b3a5 | 2413 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2414 | { |
2415 | unsigned long flags; | |
dd41f596 | 2416 | struct rq *rq; |
1da177e4 LT |
2417 | |
2418 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2419 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2420 | update_rq_clock(rq); |
1da177e4 LT |
2421 | |
2422 | p->prio = effective_prio(p); | |
2423 | ||
b9dca1e0 | 2424 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2425 | activate_task(rq, p, 0); |
1da177e4 | 2426 | } else { |
1da177e4 | 2427 | /* |
dd41f596 IM |
2428 | * Let the scheduling class do new task startup |
2429 | * management (if any): | |
1da177e4 | 2430 | */ |
ee0827d8 | 2431 | p->sched_class->task_new(rq, p); |
c09595f6 | 2432 | inc_nr_running(rq); |
1da177e4 | 2433 | } |
5b82a1b0 MD |
2434 | trace_mark(kernel_sched_wakeup_new, |
2435 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2436 | p->pid, p->state, rq, p, rq->curr); | |
15afe09b | 2437 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2438 | #ifdef CONFIG_SMP |
2439 | if (p->sched_class->task_wake_up) | |
2440 | p->sched_class->task_wake_up(rq, p); | |
2441 | #endif | |
dd41f596 | 2442 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2443 | } |
2444 | ||
e107be36 AK |
2445 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2446 | ||
2447 | /** | |
421cee29 RD |
2448 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2449 | * @notifier: notifier struct to register | |
e107be36 AK |
2450 | */ |
2451 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2452 | { | |
2453 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2454 | } | |
2455 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2456 | ||
2457 | /** | |
2458 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2459 | * @notifier: notifier struct to unregister |
e107be36 AK |
2460 | * |
2461 | * This is safe to call from within a preemption notifier. | |
2462 | */ | |
2463 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2464 | { | |
2465 | hlist_del(¬ifier->link); | |
2466 | } | |
2467 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2468 | ||
2469 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2470 | { | |
2471 | struct preempt_notifier *notifier; | |
2472 | struct hlist_node *node; | |
2473 | ||
2474 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2475 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2476 | } | |
2477 | ||
2478 | static void | |
2479 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2480 | struct task_struct *next) | |
2481 | { | |
2482 | struct preempt_notifier *notifier; | |
2483 | struct hlist_node *node; | |
2484 | ||
2485 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2486 | notifier->ops->sched_out(notifier, next); | |
2487 | } | |
2488 | ||
6d6bc0ad | 2489 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2490 | |
2491 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2492 | { | |
2493 | } | |
2494 | ||
2495 | static void | |
2496 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2497 | struct task_struct *next) | |
2498 | { | |
2499 | } | |
2500 | ||
6d6bc0ad | 2501 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2502 | |
4866cde0 NP |
2503 | /** |
2504 | * prepare_task_switch - prepare to switch tasks | |
2505 | * @rq: the runqueue preparing to switch | |
421cee29 | 2506 | * @prev: the current task that is being switched out |
4866cde0 NP |
2507 | * @next: the task we are going to switch to. |
2508 | * | |
2509 | * This is called with the rq lock held and interrupts off. It must | |
2510 | * be paired with a subsequent finish_task_switch after the context | |
2511 | * switch. | |
2512 | * | |
2513 | * prepare_task_switch sets up locking and calls architecture specific | |
2514 | * hooks. | |
2515 | */ | |
e107be36 AK |
2516 | static inline void |
2517 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2518 | struct task_struct *next) | |
4866cde0 | 2519 | { |
e107be36 | 2520 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2521 | prepare_lock_switch(rq, next); |
2522 | prepare_arch_switch(next); | |
2523 | } | |
2524 | ||
1da177e4 LT |
2525 | /** |
2526 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2527 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2528 | * @prev: the thread we just switched away from. |
2529 | * | |
4866cde0 NP |
2530 | * finish_task_switch must be called after the context switch, paired |
2531 | * with a prepare_task_switch call before the context switch. | |
2532 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2533 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2534 | * |
2535 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2536 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2537 | * with the lock held can cause deadlocks; see schedule() for |
2538 | * details.) | |
2539 | */ | |
a9957449 | 2540 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2541 | __releases(rq->lock) |
2542 | { | |
1da177e4 | 2543 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2544 | long prev_state; |
1da177e4 LT |
2545 | |
2546 | rq->prev_mm = NULL; | |
2547 | ||
2548 | /* | |
2549 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2550 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2551 | * schedule one last time. The schedule call will never return, and |
2552 | * the scheduled task must drop that reference. | |
c394cc9f | 2553 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2554 | * still held, otherwise prev could be scheduled on another cpu, die |
2555 | * there before we look at prev->state, and then the reference would | |
2556 | * be dropped twice. | |
2557 | * Manfred Spraul <manfred@colorfullife.com> | |
2558 | */ | |
55a101f8 | 2559 | prev_state = prev->state; |
4866cde0 NP |
2560 | finish_arch_switch(prev); |
2561 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2562 | #ifdef CONFIG_SMP |
2563 | if (current->sched_class->post_schedule) | |
2564 | current->sched_class->post_schedule(rq); | |
2565 | #endif | |
e8fa1362 | 2566 | |
e107be36 | 2567 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2568 | if (mm) |
2569 | mmdrop(mm); | |
c394cc9f | 2570 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2571 | /* |
2572 | * Remove function-return probe instances associated with this | |
2573 | * task and put them back on the free list. | |
9761eea8 | 2574 | */ |
c6fd91f0 | 2575 | kprobe_flush_task(prev); |
1da177e4 | 2576 | put_task_struct(prev); |
c6fd91f0 | 2577 | } |
1da177e4 LT |
2578 | } |
2579 | ||
2580 | /** | |
2581 | * schedule_tail - first thing a freshly forked thread must call. | |
2582 | * @prev: the thread we just switched away from. | |
2583 | */ | |
36c8b586 | 2584 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2585 | __releases(rq->lock) |
2586 | { | |
70b97a7f IM |
2587 | struct rq *rq = this_rq(); |
2588 | ||
4866cde0 NP |
2589 | finish_task_switch(rq, prev); |
2590 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2591 | /* In this case, finish_task_switch does not reenable preemption */ | |
2592 | preempt_enable(); | |
2593 | #endif | |
1da177e4 | 2594 | if (current->set_child_tid) |
b488893a | 2595 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2596 | } |
2597 | ||
2598 | /* | |
2599 | * context_switch - switch to the new MM and the new | |
2600 | * thread's register state. | |
2601 | */ | |
dd41f596 | 2602 | static inline void |
70b97a7f | 2603 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2604 | struct task_struct *next) |
1da177e4 | 2605 | { |
dd41f596 | 2606 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2607 | |
e107be36 | 2608 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2609 | trace_mark(kernel_sched_schedule, |
2610 | "prev_pid %d next_pid %d prev_state %ld " | |
2611 | "## rq %p prev %p next %p", | |
2612 | prev->pid, next->pid, prev->state, | |
2613 | rq, prev, next); | |
dd41f596 IM |
2614 | mm = next->mm; |
2615 | oldmm = prev->active_mm; | |
9226d125 ZA |
2616 | /* |
2617 | * For paravirt, this is coupled with an exit in switch_to to | |
2618 | * combine the page table reload and the switch backend into | |
2619 | * one hypercall. | |
2620 | */ | |
2621 | arch_enter_lazy_cpu_mode(); | |
2622 | ||
dd41f596 | 2623 | if (unlikely(!mm)) { |
1da177e4 LT |
2624 | next->active_mm = oldmm; |
2625 | atomic_inc(&oldmm->mm_count); | |
2626 | enter_lazy_tlb(oldmm, next); | |
2627 | } else | |
2628 | switch_mm(oldmm, mm, next); | |
2629 | ||
dd41f596 | 2630 | if (unlikely(!prev->mm)) { |
1da177e4 | 2631 | prev->active_mm = NULL; |
1da177e4 LT |
2632 | rq->prev_mm = oldmm; |
2633 | } | |
3a5f5e48 IM |
2634 | /* |
2635 | * Since the runqueue lock will be released by the next | |
2636 | * task (which is an invalid locking op but in the case | |
2637 | * of the scheduler it's an obvious special-case), so we | |
2638 | * do an early lockdep release here: | |
2639 | */ | |
2640 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2641 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2642 | #endif |
1da177e4 LT |
2643 | |
2644 | /* Here we just switch the register state and the stack. */ | |
2645 | switch_to(prev, next, prev); | |
2646 | ||
dd41f596 IM |
2647 | barrier(); |
2648 | /* | |
2649 | * this_rq must be evaluated again because prev may have moved | |
2650 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2651 | * frame will be invalid. | |
2652 | */ | |
2653 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2654 | } |
2655 | ||
2656 | /* | |
2657 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2658 | * | |
2659 | * externally visible scheduler statistics: current number of runnable | |
2660 | * threads, current number of uninterruptible-sleeping threads, total | |
2661 | * number of context switches performed since bootup. | |
2662 | */ | |
2663 | unsigned long nr_running(void) | |
2664 | { | |
2665 | unsigned long i, sum = 0; | |
2666 | ||
2667 | for_each_online_cpu(i) | |
2668 | sum += cpu_rq(i)->nr_running; | |
2669 | ||
2670 | return sum; | |
2671 | } | |
2672 | ||
2673 | unsigned long nr_uninterruptible(void) | |
2674 | { | |
2675 | unsigned long i, sum = 0; | |
2676 | ||
0a945022 | 2677 | for_each_possible_cpu(i) |
1da177e4 LT |
2678 | sum += cpu_rq(i)->nr_uninterruptible; |
2679 | ||
2680 | /* | |
2681 | * Since we read the counters lockless, it might be slightly | |
2682 | * inaccurate. Do not allow it to go below zero though: | |
2683 | */ | |
2684 | if (unlikely((long)sum < 0)) | |
2685 | sum = 0; | |
2686 | ||
2687 | return sum; | |
2688 | } | |
2689 | ||
2690 | unsigned long long nr_context_switches(void) | |
2691 | { | |
cc94abfc SR |
2692 | int i; |
2693 | unsigned long long sum = 0; | |
1da177e4 | 2694 | |
0a945022 | 2695 | for_each_possible_cpu(i) |
1da177e4 LT |
2696 | sum += cpu_rq(i)->nr_switches; |
2697 | ||
2698 | return sum; | |
2699 | } | |
2700 | ||
2701 | unsigned long nr_iowait(void) | |
2702 | { | |
2703 | unsigned long i, sum = 0; | |
2704 | ||
0a945022 | 2705 | for_each_possible_cpu(i) |
1da177e4 LT |
2706 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2707 | ||
2708 | return sum; | |
2709 | } | |
2710 | ||
db1b1fef JS |
2711 | unsigned long nr_active(void) |
2712 | { | |
2713 | unsigned long i, running = 0, uninterruptible = 0; | |
2714 | ||
2715 | for_each_online_cpu(i) { | |
2716 | running += cpu_rq(i)->nr_running; | |
2717 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2718 | } | |
2719 | ||
2720 | if (unlikely((long)uninterruptible < 0)) | |
2721 | uninterruptible = 0; | |
2722 | ||
2723 | return running + uninterruptible; | |
2724 | } | |
2725 | ||
48f24c4d | 2726 | /* |
dd41f596 IM |
2727 | * Update rq->cpu_load[] statistics. This function is usually called every |
2728 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2729 | */ |
dd41f596 | 2730 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2731 | { |
495eca49 | 2732 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2733 | int i, scale; |
2734 | ||
2735 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2736 | |
2737 | /* Update our load: */ | |
2738 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2739 | unsigned long old_load, new_load; | |
2740 | ||
2741 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2742 | ||
2743 | old_load = this_rq->cpu_load[i]; | |
2744 | new_load = this_load; | |
a25707f3 IM |
2745 | /* |
2746 | * Round up the averaging division if load is increasing. This | |
2747 | * prevents us from getting stuck on 9 if the load is 10, for | |
2748 | * example. | |
2749 | */ | |
2750 | if (new_load > old_load) | |
2751 | new_load += scale-1; | |
dd41f596 IM |
2752 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2753 | } | |
48f24c4d IM |
2754 | } |
2755 | ||
dd41f596 IM |
2756 | #ifdef CONFIG_SMP |
2757 | ||
1da177e4 LT |
2758 | /* |
2759 | * double_rq_lock - safely lock two runqueues | |
2760 | * | |
2761 | * Note this does not disable interrupts like task_rq_lock, | |
2762 | * you need to do so manually before calling. | |
2763 | */ | |
70b97a7f | 2764 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2765 | __acquires(rq1->lock) |
2766 | __acquires(rq2->lock) | |
2767 | { | |
054b9108 | 2768 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2769 | if (rq1 == rq2) { |
2770 | spin_lock(&rq1->lock); | |
2771 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2772 | } else { | |
c96d145e | 2773 | if (rq1 < rq2) { |
1da177e4 | 2774 | spin_lock(&rq1->lock); |
5e710e37 | 2775 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2776 | } else { |
2777 | spin_lock(&rq2->lock); | |
5e710e37 | 2778 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2779 | } |
2780 | } | |
6e82a3be IM |
2781 | update_rq_clock(rq1); |
2782 | update_rq_clock(rq2); | |
1da177e4 LT |
2783 | } |
2784 | ||
2785 | /* | |
2786 | * double_rq_unlock - safely unlock two runqueues | |
2787 | * | |
2788 | * Note this does not restore interrupts like task_rq_unlock, | |
2789 | * you need to do so manually after calling. | |
2790 | */ | |
70b97a7f | 2791 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2792 | __releases(rq1->lock) |
2793 | __releases(rq2->lock) | |
2794 | { | |
2795 | spin_unlock(&rq1->lock); | |
2796 | if (rq1 != rq2) | |
2797 | spin_unlock(&rq2->lock); | |
2798 | else | |
2799 | __release(rq2->lock); | |
2800 | } | |
2801 | ||
2802 | /* | |
2803 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2804 | */ | |
e8fa1362 | 2805 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2806 | __releases(this_rq->lock) |
2807 | __acquires(busiest->lock) | |
2808 | __acquires(this_rq->lock) | |
2809 | { | |
e8fa1362 SR |
2810 | int ret = 0; |
2811 | ||
054b9108 KK |
2812 | if (unlikely(!irqs_disabled())) { |
2813 | /* printk() doesn't work good under rq->lock */ | |
2814 | spin_unlock(&this_rq->lock); | |
2815 | BUG_ON(1); | |
2816 | } | |
1da177e4 | 2817 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2818 | if (busiest < this_rq) { |
1da177e4 LT |
2819 | spin_unlock(&this_rq->lock); |
2820 | spin_lock(&busiest->lock); | |
5e710e37 | 2821 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2822 | ret = 1; |
1da177e4 | 2823 | } else |
5e710e37 | 2824 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2825 | } |
e8fa1362 | 2826 | return ret; |
1da177e4 LT |
2827 | } |
2828 | ||
1b12bbc7 PZ |
2829 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2830 | __releases(busiest->lock) | |
2831 | { | |
2832 | spin_unlock(&busiest->lock); | |
2833 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2834 | } | |
2835 | ||
1da177e4 LT |
2836 | /* |
2837 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2838 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2839 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2840 | * the cpu_allowed mask is restored. |
2841 | */ | |
36c8b586 | 2842 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2843 | { |
70b97a7f | 2844 | struct migration_req req; |
1da177e4 | 2845 | unsigned long flags; |
70b97a7f | 2846 | struct rq *rq; |
1da177e4 LT |
2847 | |
2848 | rq = task_rq_lock(p, &flags); | |
2849 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2850 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2851 | goto out; |
2852 | ||
2853 | /* force the process onto the specified CPU */ | |
2854 | if (migrate_task(p, dest_cpu, &req)) { | |
2855 | /* Need to wait for migration thread (might exit: take ref). */ | |
2856 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2857 | |
1da177e4 LT |
2858 | get_task_struct(mt); |
2859 | task_rq_unlock(rq, &flags); | |
2860 | wake_up_process(mt); | |
2861 | put_task_struct(mt); | |
2862 | wait_for_completion(&req.done); | |
36c8b586 | 2863 | |
1da177e4 LT |
2864 | return; |
2865 | } | |
2866 | out: | |
2867 | task_rq_unlock(rq, &flags); | |
2868 | } | |
2869 | ||
2870 | /* | |
476d139c NP |
2871 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2872 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2873 | */ |
2874 | void sched_exec(void) | |
2875 | { | |
1da177e4 | 2876 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2877 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2878 | put_cpu(); |
476d139c NP |
2879 | if (new_cpu != this_cpu) |
2880 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2881 | } |
2882 | ||
2883 | /* | |
2884 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2885 | * Both runqueues must be locked. | |
2886 | */ | |
dd41f596 IM |
2887 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2888 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2889 | { |
2e1cb74a | 2890 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2891 | set_task_cpu(p, this_cpu); |
dd41f596 | 2892 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2893 | /* |
2894 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2895 | * to be always true for them. | |
2896 | */ | |
15afe09b | 2897 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2898 | } |
2899 | ||
2900 | /* | |
2901 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2902 | */ | |
858119e1 | 2903 | static |
70b97a7f | 2904 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2905 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2906 | int *all_pinned) |
1da177e4 LT |
2907 | { |
2908 | /* | |
2909 | * We do not migrate tasks that are: | |
2910 | * 1) running (obviously), or | |
2911 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2912 | * 3) are cache-hot on their current CPU. | |
2913 | */ | |
cc367732 IM |
2914 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2915 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2916 | return 0; |
cc367732 | 2917 | } |
81026794 NP |
2918 | *all_pinned = 0; |
2919 | ||
cc367732 IM |
2920 | if (task_running(rq, p)) { |
2921 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2922 | return 0; |
cc367732 | 2923 | } |
1da177e4 | 2924 | |
da84d961 IM |
2925 | /* |
2926 | * Aggressive migration if: | |
2927 | * 1) task is cache cold, or | |
2928 | * 2) too many balance attempts have failed. | |
2929 | */ | |
2930 | ||
6bc1665b IM |
2931 | if (!task_hot(p, rq->clock, sd) || |
2932 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2933 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2934 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2935 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2936 | schedstat_inc(p, se.nr_forced_migrations); |
2937 | } | |
da84d961 IM |
2938 | #endif |
2939 | return 1; | |
2940 | } | |
2941 | ||
cc367732 IM |
2942 | if (task_hot(p, rq->clock, sd)) { |
2943 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2944 | return 0; |
cc367732 | 2945 | } |
1da177e4 LT |
2946 | return 1; |
2947 | } | |
2948 | ||
e1d1484f PW |
2949 | static unsigned long |
2950 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2951 | unsigned long max_load_move, struct sched_domain *sd, | |
2952 | enum cpu_idle_type idle, int *all_pinned, | |
2953 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2954 | { |
051c6764 | 2955 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2956 | struct task_struct *p; |
2957 | long rem_load_move = max_load_move; | |
1da177e4 | 2958 | |
e1d1484f | 2959 | if (max_load_move == 0) |
1da177e4 LT |
2960 | goto out; |
2961 | ||
81026794 NP |
2962 | pinned = 1; |
2963 | ||
1da177e4 | 2964 | /* |
dd41f596 | 2965 | * Start the load-balancing iterator: |
1da177e4 | 2966 | */ |
dd41f596 IM |
2967 | p = iterator->start(iterator->arg); |
2968 | next: | |
b82d9fdd | 2969 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2970 | goto out; |
051c6764 PZ |
2971 | |
2972 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2973 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2974 | p = iterator->next(iterator->arg); |
2975 | goto next; | |
1da177e4 LT |
2976 | } |
2977 | ||
dd41f596 | 2978 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2979 | pulled++; |
dd41f596 | 2980 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2981 | |
2dd73a4f | 2982 | /* |
b82d9fdd | 2983 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2984 | */ |
e1d1484f | 2985 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2986 | if (p->prio < *this_best_prio) |
2987 | *this_best_prio = p->prio; | |
dd41f596 IM |
2988 | p = iterator->next(iterator->arg); |
2989 | goto next; | |
1da177e4 LT |
2990 | } |
2991 | out: | |
2992 | /* | |
e1d1484f | 2993 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2994 | * so we can safely collect pull_task() stats here rather than |
2995 | * inside pull_task(). | |
2996 | */ | |
2997 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2998 | |
2999 | if (all_pinned) | |
3000 | *all_pinned = pinned; | |
e1d1484f PW |
3001 | |
3002 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3003 | } |
3004 | ||
dd41f596 | 3005 | /* |
43010659 PW |
3006 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3007 | * this_rq, as part of a balancing operation within domain "sd". | |
3008 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3009 | * |
3010 | * Called with both runqueues locked. | |
3011 | */ | |
3012 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3013 | unsigned long max_load_move, |
dd41f596 IM |
3014 | struct sched_domain *sd, enum cpu_idle_type idle, |
3015 | int *all_pinned) | |
3016 | { | |
5522d5d5 | 3017 | const struct sched_class *class = sched_class_highest; |
43010659 | 3018 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3019 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3020 | |
3021 | do { | |
43010659 PW |
3022 | total_load_moved += |
3023 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3024 | max_load_move - total_load_moved, |
a4ac01c3 | 3025 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3026 | class = class->next; |
c4acb2c0 GH |
3027 | |
3028 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3029 | break; | |
3030 | ||
43010659 | 3031 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3032 | |
43010659 PW |
3033 | return total_load_moved > 0; |
3034 | } | |
3035 | ||
e1d1484f PW |
3036 | static int |
3037 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3038 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3039 | struct rq_iterator *iterator) | |
3040 | { | |
3041 | struct task_struct *p = iterator->start(iterator->arg); | |
3042 | int pinned = 0; | |
3043 | ||
3044 | while (p) { | |
3045 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3046 | pull_task(busiest, p, this_rq, this_cpu); | |
3047 | /* | |
3048 | * Right now, this is only the second place pull_task() | |
3049 | * is called, so we can safely collect pull_task() | |
3050 | * stats here rather than inside pull_task(). | |
3051 | */ | |
3052 | schedstat_inc(sd, lb_gained[idle]); | |
3053 | ||
3054 | return 1; | |
3055 | } | |
3056 | p = iterator->next(iterator->arg); | |
3057 | } | |
3058 | ||
3059 | return 0; | |
3060 | } | |
3061 | ||
43010659 PW |
3062 | /* |
3063 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3064 | * part of active balancing operations within "domain". | |
3065 | * Returns 1 if successful and 0 otherwise. | |
3066 | * | |
3067 | * Called with both runqueues locked. | |
3068 | */ | |
3069 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3070 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3071 | { | |
5522d5d5 | 3072 | const struct sched_class *class; |
43010659 PW |
3073 | |
3074 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3075 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3076 | return 1; |
3077 | ||
3078 | return 0; | |
dd41f596 IM |
3079 | } |
3080 | ||
1da177e4 LT |
3081 | /* |
3082 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3083 | * domain. It calculates and returns the amount of weighted load which |
3084 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3085 | */ |
3086 | static struct sched_group * | |
3087 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3088 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3089 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3090 | { |
3091 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3092 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3093 | unsigned long max_pull; |
2dd73a4f PW |
3094 | unsigned long busiest_load_per_task, busiest_nr_running; |
3095 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3096 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3097 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3098 | int power_savings_balance = 1; | |
3099 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3100 | unsigned long min_nr_running = ULONG_MAX; | |
3101 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3102 | #endif | |
1da177e4 LT |
3103 | |
3104 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3105 | busiest_load_per_task = busiest_nr_running = 0; |
3106 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3107 | |
d15bcfdb | 3108 | if (idle == CPU_NOT_IDLE) |
7897986b | 3109 | load_idx = sd->busy_idx; |
d15bcfdb | 3110 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3111 | load_idx = sd->newidle_idx; |
3112 | else | |
3113 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3114 | |
3115 | do { | |
908a7c1b | 3116 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3117 | int local_group; |
3118 | int i; | |
908a7c1b | 3119 | int __group_imb = 0; |
783609c6 | 3120 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3121 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3122 | unsigned long sum_avg_load_per_task; |
3123 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3124 | |
3125 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3126 | ||
783609c6 SS |
3127 | if (local_group) |
3128 | balance_cpu = first_cpu(group->cpumask); | |
3129 | ||
1da177e4 | 3130 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3131 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3132 | sum_avg_load_per_task = avg_load_per_task = 0; |
3133 | ||
908a7c1b KC |
3134 | max_cpu_load = 0; |
3135 | min_cpu_load = ~0UL; | |
1da177e4 | 3136 | |
363ab6f1 | 3137 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3138 | struct rq *rq; |
3139 | ||
3140 | if (!cpu_isset(i, *cpus)) | |
3141 | continue; | |
3142 | ||
3143 | rq = cpu_rq(i); | |
2dd73a4f | 3144 | |
9439aab8 | 3145 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3146 | *sd_idle = 0; |
3147 | ||
1da177e4 | 3148 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3149 | if (local_group) { |
3150 | if (idle_cpu(i) && !first_idle_cpu) { | |
3151 | first_idle_cpu = 1; | |
3152 | balance_cpu = i; | |
3153 | } | |
3154 | ||
a2000572 | 3155 | load = target_load(i, load_idx); |
908a7c1b | 3156 | } else { |
a2000572 | 3157 | load = source_load(i, load_idx); |
908a7c1b KC |
3158 | if (load > max_cpu_load) |
3159 | max_cpu_load = load; | |
3160 | if (min_cpu_load > load) | |
3161 | min_cpu_load = load; | |
3162 | } | |
1da177e4 LT |
3163 | |
3164 | avg_load += load; | |
2dd73a4f | 3165 | sum_nr_running += rq->nr_running; |
dd41f596 | 3166 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3167 | |
3168 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3169 | } |
3170 | ||
783609c6 SS |
3171 | /* |
3172 | * First idle cpu or the first cpu(busiest) in this sched group | |
3173 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3174 | * domains. In the newly idle case, we will allow all the cpu's |
3175 | * to do the newly idle load balance. | |
783609c6 | 3176 | */ |
9439aab8 SS |
3177 | if (idle != CPU_NEWLY_IDLE && local_group && |
3178 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3179 | *balance = 0; |
3180 | goto ret; | |
3181 | } | |
3182 | ||
1da177e4 | 3183 | total_load += avg_load; |
5517d86b | 3184 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3185 | |
3186 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3187 | avg_load = sg_div_cpu_power(group, |
3188 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3189 | |
408ed066 PZ |
3190 | |
3191 | /* | |
3192 | * Consider the group unbalanced when the imbalance is larger | |
3193 | * than the average weight of two tasks. | |
3194 | * | |
3195 | * APZ: with cgroup the avg task weight can vary wildly and | |
3196 | * might not be a suitable number - should we keep a | |
3197 | * normalized nr_running number somewhere that negates | |
3198 | * the hierarchy? | |
3199 | */ | |
3200 | avg_load_per_task = sg_div_cpu_power(group, | |
3201 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3202 | ||
3203 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3204 | __group_imb = 1; |
3205 | ||
5517d86b | 3206 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3207 | |
1da177e4 LT |
3208 | if (local_group) { |
3209 | this_load = avg_load; | |
3210 | this = group; | |
2dd73a4f PW |
3211 | this_nr_running = sum_nr_running; |
3212 | this_load_per_task = sum_weighted_load; | |
3213 | } else if (avg_load > max_load && | |
908a7c1b | 3214 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3215 | max_load = avg_load; |
3216 | busiest = group; | |
2dd73a4f PW |
3217 | busiest_nr_running = sum_nr_running; |
3218 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3219 | group_imb = __group_imb; |
1da177e4 | 3220 | } |
5c45bf27 SS |
3221 | |
3222 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3223 | /* | |
3224 | * Busy processors will not participate in power savings | |
3225 | * balance. | |
3226 | */ | |
dd41f596 IM |
3227 | if (idle == CPU_NOT_IDLE || |
3228 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3229 | goto group_next; | |
5c45bf27 SS |
3230 | |
3231 | /* | |
3232 | * If the local group is idle or completely loaded | |
3233 | * no need to do power savings balance at this domain | |
3234 | */ | |
3235 | if (local_group && (this_nr_running >= group_capacity || | |
3236 | !this_nr_running)) | |
3237 | power_savings_balance = 0; | |
3238 | ||
dd41f596 | 3239 | /* |
5c45bf27 SS |
3240 | * If a group is already running at full capacity or idle, |
3241 | * don't include that group in power savings calculations | |
dd41f596 IM |
3242 | */ |
3243 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3244 | || !sum_nr_running) |
dd41f596 | 3245 | goto group_next; |
5c45bf27 | 3246 | |
dd41f596 | 3247 | /* |
5c45bf27 | 3248 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3249 | * This is the group from where we need to pick up the load |
3250 | * for saving power | |
3251 | */ | |
3252 | if ((sum_nr_running < min_nr_running) || | |
3253 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3254 | first_cpu(group->cpumask) < |
3255 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3256 | group_min = group; |
3257 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3258 | min_load_per_task = sum_weighted_load / |
3259 | sum_nr_running; | |
dd41f596 | 3260 | } |
5c45bf27 | 3261 | |
dd41f596 | 3262 | /* |
5c45bf27 | 3263 | * Calculate the group which is almost near its |
dd41f596 IM |
3264 | * capacity but still has some space to pick up some load |
3265 | * from other group and save more power | |
3266 | */ | |
3267 | if (sum_nr_running <= group_capacity - 1) { | |
3268 | if (sum_nr_running > leader_nr_running || | |
3269 | (sum_nr_running == leader_nr_running && | |
3270 | first_cpu(group->cpumask) > | |
3271 | first_cpu(group_leader->cpumask))) { | |
3272 | group_leader = group; | |
3273 | leader_nr_running = sum_nr_running; | |
3274 | } | |
48f24c4d | 3275 | } |
5c45bf27 SS |
3276 | group_next: |
3277 | #endif | |
1da177e4 LT |
3278 | group = group->next; |
3279 | } while (group != sd->groups); | |
3280 | ||
2dd73a4f | 3281 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3282 | goto out_balanced; |
3283 | ||
3284 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3285 | ||
3286 | if (this_load >= avg_load || | |
3287 | 100*max_load <= sd->imbalance_pct*this_load) | |
3288 | goto out_balanced; | |
3289 | ||
2dd73a4f | 3290 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3291 | if (group_imb) |
3292 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3293 | ||
1da177e4 LT |
3294 | /* |
3295 | * We're trying to get all the cpus to the average_load, so we don't | |
3296 | * want to push ourselves above the average load, nor do we wish to | |
3297 | * reduce the max loaded cpu below the average load, as either of these | |
3298 | * actions would just result in more rebalancing later, and ping-pong | |
3299 | * tasks around. Thus we look for the minimum possible imbalance. | |
3300 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3301 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3302 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3303 | * appear as very large values with unsigned longs. |
3304 | */ | |
2dd73a4f PW |
3305 | if (max_load <= busiest_load_per_task) |
3306 | goto out_balanced; | |
3307 | ||
3308 | /* | |
3309 | * In the presence of smp nice balancing, certain scenarios can have | |
3310 | * max load less than avg load(as we skip the groups at or below | |
3311 | * its cpu_power, while calculating max_load..) | |
3312 | */ | |
3313 | if (max_load < avg_load) { | |
3314 | *imbalance = 0; | |
3315 | goto small_imbalance; | |
3316 | } | |
0c117f1b SS |
3317 | |
3318 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3319 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3320 | |
1da177e4 | 3321 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3322 | *imbalance = min(max_pull * busiest->__cpu_power, |
3323 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3324 | / SCHED_LOAD_SCALE; |
3325 | ||
2dd73a4f PW |
3326 | /* |
3327 | * if *imbalance is less than the average load per runnable task | |
3328 | * there is no gaurantee that any tasks will be moved so we'll have | |
3329 | * a think about bumping its value to force at least one task to be | |
3330 | * moved | |
3331 | */ | |
7fd0d2dd | 3332 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3333 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3334 | unsigned int imbn; |
3335 | ||
3336 | small_imbalance: | |
3337 | pwr_move = pwr_now = 0; | |
3338 | imbn = 2; | |
3339 | if (this_nr_running) { | |
3340 | this_load_per_task /= this_nr_running; | |
3341 | if (busiest_load_per_task > this_load_per_task) | |
3342 | imbn = 1; | |
3343 | } else | |
408ed066 | 3344 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3345 | |
408ed066 | 3346 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3347 | busiest_load_per_task * imbn) { |
2dd73a4f | 3348 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3349 | return busiest; |
3350 | } | |
3351 | ||
3352 | /* | |
3353 | * OK, we don't have enough imbalance to justify moving tasks, | |
3354 | * however we may be able to increase total CPU power used by | |
3355 | * moving them. | |
3356 | */ | |
3357 | ||
5517d86b ED |
3358 | pwr_now += busiest->__cpu_power * |
3359 | min(busiest_load_per_task, max_load); | |
3360 | pwr_now += this->__cpu_power * | |
3361 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3362 | pwr_now /= SCHED_LOAD_SCALE; |
3363 | ||
3364 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3365 | tmp = sg_div_cpu_power(busiest, |
3366 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3367 | if (max_load > tmp) |
5517d86b | 3368 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3369 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3370 | |
3371 | /* Amount of load we'd add */ | |
5517d86b | 3372 | if (max_load * busiest->__cpu_power < |
33859f7f | 3373 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3374 | tmp = sg_div_cpu_power(this, |
3375 | max_load * busiest->__cpu_power); | |
1da177e4 | 3376 | else |
5517d86b ED |
3377 | tmp = sg_div_cpu_power(this, |
3378 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3379 | pwr_move += this->__cpu_power * | |
3380 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3381 | pwr_move /= SCHED_LOAD_SCALE; |
3382 | ||
3383 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3384 | if (pwr_move > pwr_now) |
3385 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3386 | } |
3387 | ||
1da177e4 LT |
3388 | return busiest; |
3389 | ||
3390 | out_balanced: | |
5c45bf27 | 3391 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3392 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3393 | goto ret; |
1da177e4 | 3394 | |
5c45bf27 SS |
3395 | if (this == group_leader && group_leader != group_min) { |
3396 | *imbalance = min_load_per_task; | |
3397 | return group_min; | |
3398 | } | |
5c45bf27 | 3399 | #endif |
783609c6 | 3400 | ret: |
1da177e4 LT |
3401 | *imbalance = 0; |
3402 | return NULL; | |
3403 | } | |
3404 | ||
3405 | /* | |
3406 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3407 | */ | |
70b97a7f | 3408 | static struct rq * |
d15bcfdb | 3409 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3410 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3411 | { |
70b97a7f | 3412 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3413 | unsigned long max_load = 0; |
1da177e4 LT |
3414 | int i; |
3415 | ||
363ab6f1 | 3416 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3417 | unsigned long wl; |
0a2966b4 CL |
3418 | |
3419 | if (!cpu_isset(i, *cpus)) | |
3420 | continue; | |
3421 | ||
48f24c4d | 3422 | rq = cpu_rq(i); |
dd41f596 | 3423 | wl = weighted_cpuload(i); |
2dd73a4f | 3424 | |
dd41f596 | 3425 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3426 | continue; |
1da177e4 | 3427 | |
dd41f596 IM |
3428 | if (wl > max_load) { |
3429 | max_load = wl; | |
48f24c4d | 3430 | busiest = rq; |
1da177e4 LT |
3431 | } |
3432 | } | |
3433 | ||
3434 | return busiest; | |
3435 | } | |
3436 | ||
77391d71 NP |
3437 | /* |
3438 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3439 | * so long as it is large enough. | |
3440 | */ | |
3441 | #define MAX_PINNED_INTERVAL 512 | |
3442 | ||
1da177e4 LT |
3443 | /* |
3444 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3445 | * tasks if there is an imbalance. | |
1da177e4 | 3446 | */ |
70b97a7f | 3447 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3448 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3449 | int *balance, cpumask_t *cpus) |
1da177e4 | 3450 | { |
43010659 | 3451 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3452 | struct sched_group *group; |
1da177e4 | 3453 | unsigned long imbalance; |
70b97a7f | 3454 | struct rq *busiest; |
fe2eea3f | 3455 | unsigned long flags; |
5969fe06 | 3456 | |
7c16ec58 MT |
3457 | cpus_setall(*cpus); |
3458 | ||
89c4710e SS |
3459 | /* |
3460 | * When power savings policy is enabled for the parent domain, idle | |
3461 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3462 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3463 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3464 | */ |
d15bcfdb | 3465 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3466 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3467 | sd_idle = 1; |
1da177e4 | 3468 | |
2d72376b | 3469 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3470 | |
0a2966b4 | 3471 | redo: |
c8cba857 | 3472 | update_shares(sd); |
0a2966b4 | 3473 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3474 | cpus, balance); |
783609c6 | 3475 | |
06066714 | 3476 | if (*balance == 0) |
783609c6 | 3477 | goto out_balanced; |
783609c6 | 3478 | |
1da177e4 LT |
3479 | if (!group) { |
3480 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3481 | goto out_balanced; | |
3482 | } | |
3483 | ||
7c16ec58 | 3484 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3485 | if (!busiest) { |
3486 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3487 | goto out_balanced; | |
3488 | } | |
3489 | ||
db935dbd | 3490 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3491 | |
3492 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3493 | ||
43010659 | 3494 | ld_moved = 0; |
1da177e4 LT |
3495 | if (busiest->nr_running > 1) { |
3496 | /* | |
3497 | * Attempt to move tasks. If find_busiest_group has found | |
3498 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3499 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3500 | * correctly treated as an imbalance. |
3501 | */ | |
fe2eea3f | 3502 | local_irq_save(flags); |
e17224bf | 3503 | double_rq_lock(this_rq, busiest); |
43010659 | 3504 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3505 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3506 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3507 | local_irq_restore(flags); |
81026794 | 3508 | |
46cb4b7c SS |
3509 | /* |
3510 | * some other cpu did the load balance for us. | |
3511 | */ | |
43010659 | 3512 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3513 | resched_cpu(this_cpu); |
3514 | ||
81026794 | 3515 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3516 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3517 | cpu_clear(cpu_of(busiest), *cpus); |
3518 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3519 | goto redo; |
81026794 | 3520 | goto out_balanced; |
0a2966b4 | 3521 | } |
1da177e4 | 3522 | } |
81026794 | 3523 | |
43010659 | 3524 | if (!ld_moved) { |
1da177e4 LT |
3525 | schedstat_inc(sd, lb_failed[idle]); |
3526 | sd->nr_balance_failed++; | |
3527 | ||
3528 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3529 | |
fe2eea3f | 3530 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3531 | |
3532 | /* don't kick the migration_thread, if the curr | |
3533 | * task on busiest cpu can't be moved to this_cpu | |
3534 | */ | |
3535 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3536 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3537 | all_pinned = 1; |
3538 | goto out_one_pinned; | |
3539 | } | |
3540 | ||
1da177e4 LT |
3541 | if (!busiest->active_balance) { |
3542 | busiest->active_balance = 1; | |
3543 | busiest->push_cpu = this_cpu; | |
81026794 | 3544 | active_balance = 1; |
1da177e4 | 3545 | } |
fe2eea3f | 3546 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3547 | if (active_balance) |
1da177e4 LT |
3548 | wake_up_process(busiest->migration_thread); |
3549 | ||
3550 | /* | |
3551 | * We've kicked active balancing, reset the failure | |
3552 | * counter. | |
3553 | */ | |
39507451 | 3554 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3555 | } |
81026794 | 3556 | } else |
1da177e4 LT |
3557 | sd->nr_balance_failed = 0; |
3558 | ||
81026794 | 3559 | if (likely(!active_balance)) { |
1da177e4 LT |
3560 | /* We were unbalanced, so reset the balancing interval */ |
3561 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3562 | } else { |
3563 | /* | |
3564 | * If we've begun active balancing, start to back off. This | |
3565 | * case may not be covered by the all_pinned logic if there | |
3566 | * is only 1 task on the busy runqueue (because we don't call | |
3567 | * move_tasks). | |
3568 | */ | |
3569 | if (sd->balance_interval < sd->max_interval) | |
3570 | sd->balance_interval *= 2; | |
1da177e4 LT |
3571 | } |
3572 | ||
43010659 | 3573 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3574 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3575 | ld_moved = -1; |
3576 | ||
3577 | goto out; | |
1da177e4 LT |
3578 | |
3579 | out_balanced: | |
1da177e4 LT |
3580 | schedstat_inc(sd, lb_balanced[idle]); |
3581 | ||
16cfb1c0 | 3582 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3583 | |
3584 | out_one_pinned: | |
1da177e4 | 3585 | /* tune up the balancing interval */ |
77391d71 NP |
3586 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3587 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3588 | sd->balance_interval *= 2; |
3589 | ||
48f24c4d | 3590 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3591 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3592 | ld_moved = -1; |
3593 | else | |
3594 | ld_moved = 0; | |
3595 | out: | |
c8cba857 PZ |
3596 | if (ld_moved) |
3597 | update_shares(sd); | |
c09595f6 | 3598 | return ld_moved; |
1da177e4 LT |
3599 | } |
3600 | ||
3601 | /* | |
3602 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3603 | * tasks if there is an imbalance. | |
3604 | * | |
d15bcfdb | 3605 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3606 | * this_rq is locked. |
3607 | */ | |
48f24c4d | 3608 | static int |
7c16ec58 MT |
3609 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3610 | cpumask_t *cpus) | |
1da177e4 LT |
3611 | { |
3612 | struct sched_group *group; | |
70b97a7f | 3613 | struct rq *busiest = NULL; |
1da177e4 | 3614 | unsigned long imbalance; |
43010659 | 3615 | int ld_moved = 0; |
5969fe06 | 3616 | int sd_idle = 0; |
969bb4e4 | 3617 | int all_pinned = 0; |
7c16ec58 MT |
3618 | |
3619 | cpus_setall(*cpus); | |
5969fe06 | 3620 | |
89c4710e SS |
3621 | /* |
3622 | * When power savings policy is enabled for the parent domain, idle | |
3623 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3624 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3625 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3626 | */ |
3627 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3628 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3629 | sd_idle = 1; |
1da177e4 | 3630 | |
2d72376b | 3631 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3632 | redo: |
3e5459b4 | 3633 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3634 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3635 | &sd_idle, cpus, NULL); |
1da177e4 | 3636 | if (!group) { |
d15bcfdb | 3637 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3638 | goto out_balanced; |
1da177e4 LT |
3639 | } |
3640 | ||
7c16ec58 | 3641 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3642 | if (!busiest) { |
d15bcfdb | 3643 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3644 | goto out_balanced; |
1da177e4 LT |
3645 | } |
3646 | ||
db935dbd NP |
3647 | BUG_ON(busiest == this_rq); |
3648 | ||
d15bcfdb | 3649 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3650 | |
43010659 | 3651 | ld_moved = 0; |
d6d5cfaf NP |
3652 | if (busiest->nr_running > 1) { |
3653 | /* Attempt to move tasks */ | |
3654 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3655 | /* this_rq->clock is already updated */ |
3656 | update_rq_clock(busiest); | |
43010659 | 3657 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3658 | imbalance, sd, CPU_NEWLY_IDLE, |
3659 | &all_pinned); | |
1b12bbc7 | 3660 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3661 | |
969bb4e4 | 3662 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3663 | cpu_clear(cpu_of(busiest), *cpus); |
3664 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3665 | goto redo; |
3666 | } | |
d6d5cfaf NP |
3667 | } |
3668 | ||
43010659 | 3669 | if (!ld_moved) { |
d15bcfdb | 3670 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3671 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3672 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3673 | return -1; |
3674 | } else | |
16cfb1c0 | 3675 | sd->nr_balance_failed = 0; |
1da177e4 | 3676 | |
3e5459b4 | 3677 | update_shares_locked(this_rq, sd); |
43010659 | 3678 | return ld_moved; |
16cfb1c0 NP |
3679 | |
3680 | out_balanced: | |
d15bcfdb | 3681 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3682 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3683 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3684 | return -1; |
16cfb1c0 | 3685 | sd->nr_balance_failed = 0; |
48f24c4d | 3686 | |
16cfb1c0 | 3687 | return 0; |
1da177e4 LT |
3688 | } |
3689 | ||
3690 | /* | |
3691 | * idle_balance is called by schedule() if this_cpu is about to become | |
3692 | * idle. Attempts to pull tasks from other CPUs. | |
3693 | */ | |
70b97a7f | 3694 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3695 | { |
3696 | struct sched_domain *sd; | |
dd41f596 IM |
3697 | int pulled_task = -1; |
3698 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3699 | cpumask_t tmpmask; |
1da177e4 LT |
3700 | |
3701 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3702 | unsigned long interval; |
3703 | ||
3704 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3705 | continue; | |
3706 | ||
3707 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3708 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3709 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3710 | sd, &tmpmask); | |
92c4ca5c CL |
3711 | |
3712 | interval = msecs_to_jiffies(sd->balance_interval); | |
3713 | if (time_after(next_balance, sd->last_balance + interval)) | |
3714 | next_balance = sd->last_balance + interval; | |
3715 | if (pulled_task) | |
3716 | break; | |
1da177e4 | 3717 | } |
dd41f596 | 3718 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3719 | /* |
3720 | * We are going idle. next_balance may be set based on | |
3721 | * a busy processor. So reset next_balance. | |
3722 | */ | |
3723 | this_rq->next_balance = next_balance; | |
dd41f596 | 3724 | } |
1da177e4 LT |
3725 | } |
3726 | ||
3727 | /* | |
3728 | * active_load_balance is run by migration threads. It pushes running tasks | |
3729 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3730 | * running on each physical CPU where possible, and avoids physical / | |
3731 | * logical imbalances. | |
3732 | * | |
3733 | * Called with busiest_rq locked. | |
3734 | */ | |
70b97a7f | 3735 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3736 | { |
39507451 | 3737 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3738 | struct sched_domain *sd; |
3739 | struct rq *target_rq; | |
39507451 | 3740 | |
48f24c4d | 3741 | /* Is there any task to move? */ |
39507451 | 3742 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3743 | return; |
3744 | ||
3745 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3746 | |
3747 | /* | |
39507451 | 3748 | * This condition is "impossible", if it occurs |
41a2d6cf | 3749 | * we need to fix it. Originally reported by |
39507451 | 3750 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3751 | */ |
39507451 | 3752 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3753 | |
39507451 NP |
3754 | /* move a task from busiest_rq to target_rq */ |
3755 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3756 | update_rq_clock(busiest_rq); |
3757 | update_rq_clock(target_rq); | |
39507451 NP |
3758 | |
3759 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3760 | for_each_domain(target_cpu, sd) { |
39507451 | 3761 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3762 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3763 | break; |
c96d145e | 3764 | } |
39507451 | 3765 | |
48f24c4d | 3766 | if (likely(sd)) { |
2d72376b | 3767 | schedstat_inc(sd, alb_count); |
39507451 | 3768 | |
43010659 PW |
3769 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3770 | sd, CPU_IDLE)) | |
48f24c4d IM |
3771 | schedstat_inc(sd, alb_pushed); |
3772 | else | |
3773 | schedstat_inc(sd, alb_failed); | |
3774 | } | |
1b12bbc7 | 3775 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3776 | } |
3777 | ||
46cb4b7c SS |
3778 | #ifdef CONFIG_NO_HZ |
3779 | static struct { | |
3780 | atomic_t load_balancer; | |
41a2d6cf | 3781 | cpumask_t cpu_mask; |
46cb4b7c SS |
3782 | } nohz ____cacheline_aligned = { |
3783 | .load_balancer = ATOMIC_INIT(-1), | |
3784 | .cpu_mask = CPU_MASK_NONE, | |
3785 | }; | |
3786 | ||
7835b98b | 3787 | /* |
46cb4b7c SS |
3788 | * This routine will try to nominate the ilb (idle load balancing) |
3789 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3790 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3791 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3792 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3793 | * arrives... | |
3794 | * | |
3795 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3796 | * for idle load balancing. ilb owner will still be part of | |
3797 | * nohz.cpu_mask.. | |
7835b98b | 3798 | * |
46cb4b7c SS |
3799 | * While stopping the tick, this cpu will become the ilb owner if there |
3800 | * is no other owner. And will be the owner till that cpu becomes busy | |
3801 | * or if all cpus in the system stop their ticks at which point | |
3802 | * there is no need for ilb owner. | |
3803 | * | |
3804 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3805 | * next busy scheduler_tick() | |
3806 | */ | |
3807 | int select_nohz_load_balancer(int stop_tick) | |
3808 | { | |
3809 | int cpu = smp_processor_id(); | |
3810 | ||
3811 | if (stop_tick) { | |
3812 | cpu_set(cpu, nohz.cpu_mask); | |
3813 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3814 | ||
3815 | /* | |
3816 | * If we are going offline and still the leader, give up! | |
3817 | */ | |
e761b772 | 3818 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3819 | atomic_read(&nohz.load_balancer) == cpu) { |
3820 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3821 | BUG(); | |
3822 | return 0; | |
3823 | } | |
3824 | ||
3825 | /* time for ilb owner also to sleep */ | |
3826 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3827 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3828 | atomic_set(&nohz.load_balancer, -1); | |
3829 | return 0; | |
3830 | } | |
3831 | ||
3832 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3833 | /* make me the ilb owner */ | |
3834 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3835 | return 1; | |
3836 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3837 | return 1; | |
3838 | } else { | |
3839 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3840 | return 0; | |
3841 | ||
3842 | cpu_clear(cpu, nohz.cpu_mask); | |
3843 | ||
3844 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3845 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3846 | BUG(); | |
3847 | } | |
3848 | return 0; | |
3849 | } | |
3850 | #endif | |
3851 | ||
3852 | static DEFINE_SPINLOCK(balancing); | |
3853 | ||
3854 | /* | |
7835b98b CL |
3855 | * It checks each scheduling domain to see if it is due to be balanced, |
3856 | * and initiates a balancing operation if so. | |
3857 | * | |
3858 | * Balancing parameters are set up in arch_init_sched_domains. | |
3859 | */ | |
a9957449 | 3860 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3861 | { |
46cb4b7c SS |
3862 | int balance = 1; |
3863 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3864 | unsigned long interval; |
3865 | struct sched_domain *sd; | |
46cb4b7c | 3866 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3867 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3868 | int update_next_balance = 0; |
d07355f5 | 3869 | int need_serialize; |
7c16ec58 | 3870 | cpumask_t tmp; |
1da177e4 | 3871 | |
46cb4b7c | 3872 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3873 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3874 | continue; | |
3875 | ||
3876 | interval = sd->balance_interval; | |
d15bcfdb | 3877 | if (idle != CPU_IDLE) |
1da177e4 LT |
3878 | interval *= sd->busy_factor; |
3879 | ||
3880 | /* scale ms to jiffies */ | |
3881 | interval = msecs_to_jiffies(interval); | |
3882 | if (unlikely(!interval)) | |
3883 | interval = 1; | |
dd41f596 IM |
3884 | if (interval > HZ*NR_CPUS/10) |
3885 | interval = HZ*NR_CPUS/10; | |
3886 | ||
d07355f5 | 3887 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3888 | |
d07355f5 | 3889 | if (need_serialize) { |
08c183f3 CL |
3890 | if (!spin_trylock(&balancing)) |
3891 | goto out; | |
3892 | } | |
3893 | ||
c9819f45 | 3894 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3895 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3896 | /* |
3897 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3898 | * longer idle, or one of our SMT siblings is |
3899 | * not idle. | |
3900 | */ | |
d15bcfdb | 3901 | idle = CPU_NOT_IDLE; |
1da177e4 | 3902 | } |
1bd77f2d | 3903 | sd->last_balance = jiffies; |
1da177e4 | 3904 | } |
d07355f5 | 3905 | if (need_serialize) |
08c183f3 CL |
3906 | spin_unlock(&balancing); |
3907 | out: | |
f549da84 | 3908 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3909 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3910 | update_next_balance = 1; |
3911 | } | |
783609c6 SS |
3912 | |
3913 | /* | |
3914 | * Stop the load balance at this level. There is another | |
3915 | * CPU in our sched group which is doing load balancing more | |
3916 | * actively. | |
3917 | */ | |
3918 | if (!balance) | |
3919 | break; | |
1da177e4 | 3920 | } |
f549da84 SS |
3921 | |
3922 | /* | |
3923 | * next_balance will be updated only when there is a need. | |
3924 | * When the cpu is attached to null domain for ex, it will not be | |
3925 | * updated. | |
3926 | */ | |
3927 | if (likely(update_next_balance)) | |
3928 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3929 | } |
3930 | ||
3931 | /* | |
3932 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3933 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3934 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3935 | */ | |
3936 | static void run_rebalance_domains(struct softirq_action *h) | |
3937 | { | |
dd41f596 IM |
3938 | int this_cpu = smp_processor_id(); |
3939 | struct rq *this_rq = cpu_rq(this_cpu); | |
3940 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3941 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3942 | |
dd41f596 | 3943 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3944 | |
3945 | #ifdef CONFIG_NO_HZ | |
3946 | /* | |
3947 | * If this cpu is the owner for idle load balancing, then do the | |
3948 | * balancing on behalf of the other idle cpus whose ticks are | |
3949 | * stopped. | |
3950 | */ | |
dd41f596 IM |
3951 | if (this_rq->idle_at_tick && |
3952 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3953 | cpumask_t cpus = nohz.cpu_mask; |
3954 | struct rq *rq; | |
3955 | int balance_cpu; | |
3956 | ||
dd41f596 | 3957 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3958 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3959 | /* |
3960 | * If this cpu gets work to do, stop the load balancing | |
3961 | * work being done for other cpus. Next load | |
3962 | * balancing owner will pick it up. | |
3963 | */ | |
3964 | if (need_resched()) | |
3965 | break; | |
3966 | ||
de0cf899 | 3967 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3968 | |
3969 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3970 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3971 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3972 | } |
3973 | } | |
3974 | #endif | |
3975 | } | |
3976 | ||
3977 | /* | |
3978 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3979 | * | |
3980 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3981 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3982 | * if the whole system is idle. | |
3983 | */ | |
dd41f596 | 3984 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3985 | { |
46cb4b7c SS |
3986 | #ifdef CONFIG_NO_HZ |
3987 | /* | |
3988 | * If we were in the nohz mode recently and busy at the current | |
3989 | * scheduler tick, then check if we need to nominate new idle | |
3990 | * load balancer. | |
3991 | */ | |
3992 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3993 | rq->in_nohz_recently = 0; | |
3994 | ||
3995 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3996 | cpu_clear(cpu, nohz.cpu_mask); | |
3997 | atomic_set(&nohz.load_balancer, -1); | |
3998 | } | |
3999 | ||
4000 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4001 | /* | |
4002 | * simple selection for now: Nominate the | |
4003 | * first cpu in the nohz list to be the next | |
4004 | * ilb owner. | |
4005 | * | |
4006 | * TBD: Traverse the sched domains and nominate | |
4007 | * the nearest cpu in the nohz.cpu_mask. | |
4008 | */ | |
4009 | int ilb = first_cpu(nohz.cpu_mask); | |
4010 | ||
434d53b0 | 4011 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4012 | resched_cpu(ilb); |
4013 | } | |
4014 | } | |
4015 | ||
4016 | /* | |
4017 | * If this cpu is idle and doing idle load balancing for all the | |
4018 | * cpus with ticks stopped, is it time for that to stop? | |
4019 | */ | |
4020 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4021 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4022 | resched_cpu(cpu); | |
4023 | return; | |
4024 | } | |
4025 | ||
4026 | /* | |
4027 | * If this cpu is idle and the idle load balancing is done by | |
4028 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4029 | */ | |
4030 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4031 | cpu_isset(cpu, nohz.cpu_mask)) | |
4032 | return; | |
4033 | #endif | |
4034 | if (time_after_eq(jiffies, rq->next_balance)) | |
4035 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4036 | } |
dd41f596 IM |
4037 | |
4038 | #else /* CONFIG_SMP */ | |
4039 | ||
1da177e4 LT |
4040 | /* |
4041 | * on UP we do not need to balance between CPUs: | |
4042 | */ | |
70b97a7f | 4043 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4044 | { |
4045 | } | |
dd41f596 | 4046 | |
1da177e4 LT |
4047 | #endif |
4048 | ||
1da177e4 LT |
4049 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4050 | ||
4051 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4052 | ||
4053 | /* | |
41b86e9c IM |
4054 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4055 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4056 | */ |
41b86e9c | 4057 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4058 | { |
1da177e4 | 4059 | unsigned long flags; |
41b86e9c IM |
4060 | u64 ns, delta_exec; |
4061 | struct rq *rq; | |
48f24c4d | 4062 | |
41b86e9c IM |
4063 | rq = task_rq_lock(p, &flags); |
4064 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4065 | if (task_current(rq, p)) { |
a8e504d2 IM |
4066 | update_rq_clock(rq); |
4067 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4068 | if ((s64)delta_exec > 0) |
4069 | ns += delta_exec; | |
4070 | } | |
4071 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4072 | |
1da177e4 LT |
4073 | return ns; |
4074 | } | |
4075 | ||
1da177e4 LT |
4076 | /* |
4077 | * Account user cpu time to a process. | |
4078 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4079 | * @cputime: the cpu time spent in user space since the last update |
4080 | */ | |
4081 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4082 | { | |
4083 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4084 | cputime64_t tmp; | |
4085 | ||
4086 | p->utime = cputime_add(p->utime, cputime); | |
4087 | ||
4088 | /* Add user time to cpustat. */ | |
4089 | tmp = cputime_to_cputime64(cputime); | |
4090 | if (TASK_NICE(p) > 0) | |
4091 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4092 | else | |
4093 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4094 | /* Account for user time used */ |
4095 | acct_update_integrals(p); | |
1da177e4 LT |
4096 | } |
4097 | ||
94886b84 LV |
4098 | /* |
4099 | * Account guest cpu time to a process. | |
4100 | * @p: the process that the cpu time gets accounted to | |
4101 | * @cputime: the cpu time spent in virtual machine since the last update | |
4102 | */ | |
f7402e03 | 4103 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4104 | { |
4105 | cputime64_t tmp; | |
4106 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4107 | ||
4108 | tmp = cputime_to_cputime64(cputime); | |
4109 | ||
4110 | p->utime = cputime_add(p->utime, cputime); | |
4111 | p->gtime = cputime_add(p->gtime, cputime); | |
4112 | ||
4113 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4114 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4115 | } | |
4116 | ||
c66f08be MN |
4117 | /* |
4118 | * Account scaled user cpu time to a process. | |
4119 | * @p: the process that the cpu time gets accounted to | |
4120 | * @cputime: the cpu time spent in user space since the last update | |
4121 | */ | |
4122 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4123 | { | |
4124 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4125 | } | |
4126 | ||
1da177e4 LT |
4127 | /* |
4128 | * Account system cpu time to a process. | |
4129 | * @p: the process that the cpu time gets accounted to | |
4130 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4131 | * @cputime: the cpu time spent in kernel space since the last update | |
4132 | */ | |
4133 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4134 | cputime_t cputime) | |
4135 | { | |
4136 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4137 | struct rq *rq = this_rq(); |
1da177e4 LT |
4138 | cputime64_t tmp; |
4139 | ||
983ed7a6 HH |
4140 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4141 | account_guest_time(p, cputime); | |
4142 | return; | |
4143 | } | |
94886b84 | 4144 | |
1da177e4 LT |
4145 | p->stime = cputime_add(p->stime, cputime); |
4146 | ||
4147 | /* Add system time to cpustat. */ | |
4148 | tmp = cputime_to_cputime64(cputime); | |
4149 | if (hardirq_count() - hardirq_offset) | |
4150 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4151 | else if (softirq_count()) | |
4152 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4153 | else if (p != rq->idle) |
1da177e4 | 4154 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4155 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4156 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4157 | else | |
4158 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4159 | /* Account for system time used */ | |
4160 | acct_update_integrals(p); | |
1da177e4 LT |
4161 | } |
4162 | ||
c66f08be MN |
4163 | /* |
4164 | * Account scaled system cpu time to a process. | |
4165 | * @p: the process that the cpu time gets accounted to | |
4166 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4167 | * @cputime: the cpu time spent in kernel space since the last update | |
4168 | */ | |
4169 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4170 | { | |
4171 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4172 | } | |
4173 | ||
1da177e4 LT |
4174 | /* |
4175 | * Account for involuntary wait time. | |
4176 | * @p: the process from which the cpu time has been stolen | |
4177 | * @steal: the cpu time spent in involuntary wait | |
4178 | */ | |
4179 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4180 | { | |
4181 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4182 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4183 | struct rq *rq = this_rq(); |
1da177e4 LT |
4184 | |
4185 | if (p == rq->idle) { | |
4186 | p->stime = cputime_add(p->stime, steal); | |
4187 | if (atomic_read(&rq->nr_iowait) > 0) | |
4188 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4189 | else | |
4190 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4191 | } else |
1da177e4 LT |
4192 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4193 | } | |
4194 | ||
49048622 BS |
4195 | /* |
4196 | * Use precise platform statistics if available: | |
4197 | */ | |
4198 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4199 | cputime_t task_utime(struct task_struct *p) | |
4200 | { | |
4201 | return p->utime; | |
4202 | } | |
4203 | ||
4204 | cputime_t task_stime(struct task_struct *p) | |
4205 | { | |
4206 | return p->stime; | |
4207 | } | |
4208 | #else | |
4209 | cputime_t task_utime(struct task_struct *p) | |
4210 | { | |
4211 | clock_t utime = cputime_to_clock_t(p->utime), | |
4212 | total = utime + cputime_to_clock_t(p->stime); | |
4213 | u64 temp; | |
4214 | ||
4215 | /* | |
4216 | * Use CFS's precise accounting: | |
4217 | */ | |
4218 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4219 | ||
4220 | if (total) { | |
4221 | temp *= utime; | |
4222 | do_div(temp, total); | |
4223 | } | |
4224 | utime = (clock_t)temp; | |
4225 | ||
4226 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4227 | return p->prev_utime; | |
4228 | } | |
4229 | ||
4230 | cputime_t task_stime(struct task_struct *p) | |
4231 | { | |
4232 | clock_t stime; | |
4233 | ||
4234 | /* | |
4235 | * Use CFS's precise accounting. (we subtract utime from | |
4236 | * the total, to make sure the total observed by userspace | |
4237 | * grows monotonically - apps rely on that): | |
4238 | */ | |
4239 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4240 | cputime_to_clock_t(task_utime(p)); | |
4241 | ||
4242 | if (stime >= 0) | |
4243 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4244 | ||
4245 | return p->prev_stime; | |
4246 | } | |
4247 | #endif | |
4248 | ||
4249 | inline cputime_t task_gtime(struct task_struct *p) | |
4250 | { | |
4251 | return p->gtime; | |
4252 | } | |
4253 | ||
7835b98b CL |
4254 | /* |
4255 | * This function gets called by the timer code, with HZ frequency. | |
4256 | * We call it with interrupts disabled. | |
4257 | * | |
4258 | * It also gets called by the fork code, when changing the parent's | |
4259 | * timeslices. | |
4260 | */ | |
4261 | void scheduler_tick(void) | |
4262 | { | |
7835b98b CL |
4263 | int cpu = smp_processor_id(); |
4264 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4265 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4266 | |
4267 | sched_clock_tick(); | |
dd41f596 IM |
4268 | |
4269 | spin_lock(&rq->lock); | |
3e51f33f | 4270 | update_rq_clock(rq); |
f1a438d8 | 4271 | update_cpu_load(rq); |
fa85ae24 | 4272 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4273 | spin_unlock(&rq->lock); |
7835b98b | 4274 | |
e418e1c2 | 4275 | #ifdef CONFIG_SMP |
dd41f596 IM |
4276 | rq->idle_at_tick = idle_cpu(cpu); |
4277 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4278 | #endif |
1da177e4 LT |
4279 | } |
4280 | ||
6cd8a4bb SR |
4281 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4282 | defined(CONFIG_PREEMPT_TRACER)) | |
4283 | ||
4284 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4285 | { | |
4286 | if (in_lock_functions(addr)) { | |
4287 | addr = CALLER_ADDR2; | |
4288 | if (in_lock_functions(addr)) | |
4289 | addr = CALLER_ADDR3; | |
4290 | } | |
4291 | return addr; | |
4292 | } | |
1da177e4 | 4293 | |
43627582 | 4294 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4295 | { |
6cd8a4bb | 4296 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4297 | /* |
4298 | * Underflow? | |
4299 | */ | |
9a11b49a IM |
4300 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4301 | return; | |
6cd8a4bb | 4302 | #endif |
1da177e4 | 4303 | preempt_count() += val; |
6cd8a4bb | 4304 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4305 | /* |
4306 | * Spinlock count overflowing soon? | |
4307 | */ | |
33859f7f MOS |
4308 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4309 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4310 | #endif |
4311 | if (preempt_count() == val) | |
4312 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4313 | } |
4314 | EXPORT_SYMBOL(add_preempt_count); | |
4315 | ||
43627582 | 4316 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4317 | { |
6cd8a4bb | 4318 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4319 | /* |
4320 | * Underflow? | |
4321 | */ | |
9a11b49a IM |
4322 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4323 | return; | |
1da177e4 LT |
4324 | /* |
4325 | * Is the spinlock portion underflowing? | |
4326 | */ | |
9a11b49a IM |
4327 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4328 | !(preempt_count() & PREEMPT_MASK))) | |
4329 | return; | |
6cd8a4bb | 4330 | #endif |
9a11b49a | 4331 | |
6cd8a4bb SR |
4332 | if (preempt_count() == val) |
4333 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4334 | preempt_count() -= val; |
4335 | } | |
4336 | EXPORT_SYMBOL(sub_preempt_count); | |
4337 | ||
4338 | #endif | |
4339 | ||
4340 | /* | |
dd41f596 | 4341 | * Print scheduling while atomic bug: |
1da177e4 | 4342 | */ |
dd41f596 | 4343 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4344 | { |
838225b4 SS |
4345 | struct pt_regs *regs = get_irq_regs(); |
4346 | ||
4347 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4348 | prev->comm, prev->pid, preempt_count()); | |
4349 | ||
dd41f596 | 4350 | debug_show_held_locks(prev); |
e21f5b15 | 4351 | print_modules(); |
dd41f596 IM |
4352 | if (irqs_disabled()) |
4353 | print_irqtrace_events(prev); | |
838225b4 SS |
4354 | |
4355 | if (regs) | |
4356 | show_regs(regs); | |
4357 | else | |
4358 | dump_stack(); | |
dd41f596 | 4359 | } |
1da177e4 | 4360 | |
dd41f596 IM |
4361 | /* |
4362 | * Various schedule()-time debugging checks and statistics: | |
4363 | */ | |
4364 | static inline void schedule_debug(struct task_struct *prev) | |
4365 | { | |
1da177e4 | 4366 | /* |
41a2d6cf | 4367 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4368 | * schedule() atomically, we ignore that path for now. |
4369 | * Otherwise, whine if we are scheduling when we should not be. | |
4370 | */ | |
3f33a7ce | 4371 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4372 | __schedule_bug(prev); |
4373 | ||
1da177e4 LT |
4374 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4375 | ||
2d72376b | 4376 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4377 | #ifdef CONFIG_SCHEDSTATS |
4378 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4379 | schedstat_inc(this_rq(), bkl_count); |
4380 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4381 | } |
4382 | #endif | |
dd41f596 IM |
4383 | } |
4384 | ||
4385 | /* | |
4386 | * Pick up the highest-prio task: | |
4387 | */ | |
4388 | static inline struct task_struct * | |
ff95f3df | 4389 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4390 | { |
5522d5d5 | 4391 | const struct sched_class *class; |
dd41f596 | 4392 | struct task_struct *p; |
1da177e4 LT |
4393 | |
4394 | /* | |
dd41f596 IM |
4395 | * Optimization: we know that if all tasks are in |
4396 | * the fair class we can call that function directly: | |
1da177e4 | 4397 | */ |
dd41f596 | 4398 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4399 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4400 | if (likely(p)) |
4401 | return p; | |
1da177e4 LT |
4402 | } |
4403 | ||
dd41f596 IM |
4404 | class = sched_class_highest; |
4405 | for ( ; ; ) { | |
fb8d4724 | 4406 | p = class->pick_next_task(rq); |
dd41f596 IM |
4407 | if (p) |
4408 | return p; | |
4409 | /* | |
4410 | * Will never be NULL as the idle class always | |
4411 | * returns a non-NULL p: | |
4412 | */ | |
4413 | class = class->next; | |
4414 | } | |
4415 | } | |
1da177e4 | 4416 | |
dd41f596 IM |
4417 | /* |
4418 | * schedule() is the main scheduler function. | |
4419 | */ | |
4420 | asmlinkage void __sched schedule(void) | |
4421 | { | |
4422 | struct task_struct *prev, *next; | |
67ca7bde | 4423 | unsigned long *switch_count; |
dd41f596 | 4424 | struct rq *rq; |
31656519 | 4425 | int cpu; |
dd41f596 IM |
4426 | |
4427 | need_resched: | |
4428 | preempt_disable(); | |
4429 | cpu = smp_processor_id(); | |
4430 | rq = cpu_rq(cpu); | |
4431 | rcu_qsctr_inc(cpu); | |
4432 | prev = rq->curr; | |
4433 | switch_count = &prev->nivcsw; | |
4434 | ||
4435 | release_kernel_lock(prev); | |
4436 | need_resched_nonpreemptible: | |
4437 | ||
4438 | schedule_debug(prev); | |
1da177e4 | 4439 | |
31656519 | 4440 | if (sched_feat(HRTICK)) |
f333fdc9 | 4441 | hrtick_clear(rq); |
8f4d37ec | 4442 | |
1e819950 IM |
4443 | /* |
4444 | * Do the rq-clock update outside the rq lock: | |
4445 | */ | |
4446 | local_irq_disable(); | |
3e51f33f | 4447 | update_rq_clock(rq); |
1e819950 IM |
4448 | spin_lock(&rq->lock); |
4449 | clear_tsk_need_resched(prev); | |
1da177e4 | 4450 | |
1da177e4 | 4451 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4452 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4453 | prev->state = TASK_RUNNING; |
16882c1e | 4454 | else |
2e1cb74a | 4455 | deactivate_task(rq, prev, 1); |
dd41f596 | 4456 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4457 | } |
4458 | ||
9a897c5a SR |
4459 | #ifdef CONFIG_SMP |
4460 | if (prev->sched_class->pre_schedule) | |
4461 | prev->sched_class->pre_schedule(rq, prev); | |
4462 | #endif | |
f65eda4f | 4463 | |
dd41f596 | 4464 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4465 | idle_balance(cpu, rq); |
1da177e4 | 4466 | |
31ee529c | 4467 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4468 | next = pick_next_task(rq, prev); |
1da177e4 | 4469 | |
1da177e4 | 4470 | if (likely(prev != next)) { |
673a90a1 DS |
4471 | sched_info_switch(prev, next); |
4472 | ||
1da177e4 LT |
4473 | rq->nr_switches++; |
4474 | rq->curr = next; | |
4475 | ++*switch_count; | |
4476 | ||
dd41f596 | 4477 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4478 | /* |
4479 | * the context switch might have flipped the stack from under | |
4480 | * us, hence refresh the local variables. | |
4481 | */ | |
4482 | cpu = smp_processor_id(); | |
4483 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4484 | } else |
4485 | spin_unlock_irq(&rq->lock); | |
4486 | ||
8f4d37ec | 4487 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4488 | goto need_resched_nonpreemptible; |
8f4d37ec | 4489 | |
1da177e4 LT |
4490 | preempt_enable_no_resched(); |
4491 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4492 | goto need_resched; | |
4493 | } | |
1da177e4 LT |
4494 | EXPORT_SYMBOL(schedule); |
4495 | ||
4496 | #ifdef CONFIG_PREEMPT | |
4497 | /* | |
2ed6e34f | 4498 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4499 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4500 | * occur there and call schedule directly. |
4501 | */ | |
4502 | asmlinkage void __sched preempt_schedule(void) | |
4503 | { | |
4504 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4505 | |
1da177e4 LT |
4506 | /* |
4507 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4508 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4509 | */ |
beed33a8 | 4510 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4511 | return; |
4512 | ||
3a5c359a AK |
4513 | do { |
4514 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4515 | schedule(); |
3a5c359a | 4516 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4517 | |
3a5c359a AK |
4518 | /* |
4519 | * Check again in case we missed a preemption opportunity | |
4520 | * between schedule and now. | |
4521 | */ | |
4522 | barrier(); | |
4523 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4524 | } |
1da177e4 LT |
4525 | EXPORT_SYMBOL(preempt_schedule); |
4526 | ||
4527 | /* | |
2ed6e34f | 4528 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4529 | * off of irq context. |
4530 | * Note, that this is called and return with irqs disabled. This will | |
4531 | * protect us against recursive calling from irq. | |
4532 | */ | |
4533 | asmlinkage void __sched preempt_schedule_irq(void) | |
4534 | { | |
4535 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4536 | |
2ed6e34f | 4537 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4538 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4539 | ||
3a5c359a AK |
4540 | do { |
4541 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4542 | local_irq_enable(); |
4543 | schedule(); | |
4544 | local_irq_disable(); | |
3a5c359a | 4545 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4546 | |
3a5c359a AK |
4547 | /* |
4548 | * Check again in case we missed a preemption opportunity | |
4549 | * between schedule and now. | |
4550 | */ | |
4551 | barrier(); | |
4552 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4553 | } |
4554 | ||
4555 | #endif /* CONFIG_PREEMPT */ | |
4556 | ||
95cdf3b7 IM |
4557 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4558 | void *key) | |
1da177e4 | 4559 | { |
48f24c4d | 4560 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4561 | } |
1da177e4 LT |
4562 | EXPORT_SYMBOL(default_wake_function); |
4563 | ||
4564 | /* | |
41a2d6cf IM |
4565 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4566 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4567 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4568 | * | |
4569 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4570 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4571 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4572 | */ | |
4573 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4574 | int nr_exclusive, int sync, void *key) | |
4575 | { | |
2e45874c | 4576 | wait_queue_t *curr, *next; |
1da177e4 | 4577 | |
2e45874c | 4578 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4579 | unsigned flags = curr->flags; |
4580 | ||
1da177e4 | 4581 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4582 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4583 | break; |
4584 | } | |
4585 | } | |
4586 | ||
4587 | /** | |
4588 | * __wake_up - wake up threads blocked on a waitqueue. | |
4589 | * @q: the waitqueue | |
4590 | * @mode: which threads | |
4591 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4592 | * @key: is directly passed to the wakeup function |
1da177e4 | 4593 | */ |
7ad5b3a5 | 4594 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4595 | int nr_exclusive, void *key) |
1da177e4 LT |
4596 | { |
4597 | unsigned long flags; | |
4598 | ||
4599 | spin_lock_irqsave(&q->lock, flags); | |
4600 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4601 | spin_unlock_irqrestore(&q->lock, flags); | |
4602 | } | |
1da177e4 LT |
4603 | EXPORT_SYMBOL(__wake_up); |
4604 | ||
4605 | /* | |
4606 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4607 | */ | |
7ad5b3a5 | 4608 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4609 | { |
4610 | __wake_up_common(q, mode, 1, 0, NULL); | |
4611 | } | |
4612 | ||
4613 | /** | |
67be2dd1 | 4614 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4615 | * @q: the waitqueue |
4616 | * @mode: which threads | |
4617 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4618 | * | |
4619 | * The sync wakeup differs that the waker knows that it will schedule | |
4620 | * away soon, so while the target thread will be woken up, it will not | |
4621 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4622 | * with each other. This can prevent needless bouncing between CPUs. | |
4623 | * | |
4624 | * On UP it can prevent extra preemption. | |
4625 | */ | |
7ad5b3a5 | 4626 | void |
95cdf3b7 | 4627 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4628 | { |
4629 | unsigned long flags; | |
4630 | int sync = 1; | |
4631 | ||
4632 | if (unlikely(!q)) | |
4633 | return; | |
4634 | ||
4635 | if (unlikely(!nr_exclusive)) | |
4636 | sync = 0; | |
4637 | ||
4638 | spin_lock_irqsave(&q->lock, flags); | |
4639 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4640 | spin_unlock_irqrestore(&q->lock, flags); | |
4641 | } | |
4642 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4643 | ||
65eb3dc6 KD |
4644 | /** |
4645 | * complete: - signals a single thread waiting on this completion | |
4646 | * @x: holds the state of this particular completion | |
4647 | * | |
4648 | * This will wake up a single thread waiting on this completion. Threads will be | |
4649 | * awakened in the same order in which they were queued. | |
4650 | * | |
4651 | * See also complete_all(), wait_for_completion() and related routines. | |
4652 | */ | |
b15136e9 | 4653 | void complete(struct completion *x) |
1da177e4 LT |
4654 | { |
4655 | unsigned long flags; | |
4656 | ||
4657 | spin_lock_irqsave(&x->wait.lock, flags); | |
4658 | x->done++; | |
d9514f6c | 4659 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4660 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4661 | } | |
4662 | EXPORT_SYMBOL(complete); | |
4663 | ||
65eb3dc6 KD |
4664 | /** |
4665 | * complete_all: - signals all threads waiting on this completion | |
4666 | * @x: holds the state of this particular completion | |
4667 | * | |
4668 | * This will wake up all threads waiting on this particular completion event. | |
4669 | */ | |
b15136e9 | 4670 | void complete_all(struct completion *x) |
1da177e4 LT |
4671 | { |
4672 | unsigned long flags; | |
4673 | ||
4674 | spin_lock_irqsave(&x->wait.lock, flags); | |
4675 | x->done += UINT_MAX/2; | |
d9514f6c | 4676 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4677 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4678 | } | |
4679 | EXPORT_SYMBOL(complete_all); | |
4680 | ||
8cbbe86d AK |
4681 | static inline long __sched |
4682 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4683 | { |
1da177e4 LT |
4684 | if (!x->done) { |
4685 | DECLARE_WAITQUEUE(wait, current); | |
4686 | ||
4687 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4688 | __add_wait_queue_tail(&x->wait, &wait); | |
4689 | do { | |
94d3d824 | 4690 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4691 | timeout = -ERESTARTSYS; |
4692 | break; | |
8cbbe86d AK |
4693 | } |
4694 | __set_current_state(state); | |
1da177e4 LT |
4695 | spin_unlock_irq(&x->wait.lock); |
4696 | timeout = schedule_timeout(timeout); | |
4697 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4698 | } while (!x->done && timeout); |
1da177e4 | 4699 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4700 | if (!x->done) |
4701 | return timeout; | |
1da177e4 LT |
4702 | } |
4703 | x->done--; | |
ea71a546 | 4704 | return timeout ?: 1; |
1da177e4 | 4705 | } |
1da177e4 | 4706 | |
8cbbe86d AK |
4707 | static long __sched |
4708 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4709 | { |
1da177e4 LT |
4710 | might_sleep(); |
4711 | ||
4712 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4713 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4714 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4715 | return timeout; |
4716 | } | |
1da177e4 | 4717 | |
65eb3dc6 KD |
4718 | /** |
4719 | * wait_for_completion: - waits for completion of a task | |
4720 | * @x: holds the state of this particular completion | |
4721 | * | |
4722 | * This waits to be signaled for completion of a specific task. It is NOT | |
4723 | * interruptible and there is no timeout. | |
4724 | * | |
4725 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4726 | * and interrupt capability. Also see complete(). | |
4727 | */ | |
b15136e9 | 4728 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4729 | { |
4730 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4731 | } |
8cbbe86d | 4732 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4733 | |
65eb3dc6 KD |
4734 | /** |
4735 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4736 | * @x: holds the state of this particular completion | |
4737 | * @timeout: timeout value in jiffies | |
4738 | * | |
4739 | * This waits for either a completion of a specific task to be signaled or for a | |
4740 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4741 | * interruptible. | |
4742 | */ | |
b15136e9 | 4743 | unsigned long __sched |
8cbbe86d | 4744 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4745 | { |
8cbbe86d | 4746 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4747 | } |
8cbbe86d | 4748 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4749 | |
65eb3dc6 KD |
4750 | /** |
4751 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4752 | * @x: holds the state of this particular completion | |
4753 | * | |
4754 | * This waits for completion of a specific task to be signaled. It is | |
4755 | * interruptible. | |
4756 | */ | |
8cbbe86d | 4757 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4758 | { |
51e97990 AK |
4759 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4760 | if (t == -ERESTARTSYS) | |
4761 | return t; | |
4762 | return 0; | |
0fec171c | 4763 | } |
8cbbe86d | 4764 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4765 | |
65eb3dc6 KD |
4766 | /** |
4767 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4768 | * @x: holds the state of this particular completion | |
4769 | * @timeout: timeout value in jiffies | |
4770 | * | |
4771 | * This waits for either a completion of a specific task to be signaled or for a | |
4772 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4773 | */ | |
b15136e9 | 4774 | unsigned long __sched |
8cbbe86d AK |
4775 | wait_for_completion_interruptible_timeout(struct completion *x, |
4776 | unsigned long timeout) | |
0fec171c | 4777 | { |
8cbbe86d | 4778 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4779 | } |
8cbbe86d | 4780 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4781 | |
65eb3dc6 KD |
4782 | /** |
4783 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4784 | * @x: holds the state of this particular completion | |
4785 | * | |
4786 | * This waits to be signaled for completion of a specific task. It can be | |
4787 | * interrupted by a kill signal. | |
4788 | */ | |
009e577e MW |
4789 | int __sched wait_for_completion_killable(struct completion *x) |
4790 | { | |
4791 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4792 | if (t == -ERESTARTSYS) | |
4793 | return t; | |
4794 | return 0; | |
4795 | } | |
4796 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4797 | ||
be4de352 DC |
4798 | /** |
4799 | * try_wait_for_completion - try to decrement a completion without blocking | |
4800 | * @x: completion structure | |
4801 | * | |
4802 | * Returns: 0 if a decrement cannot be done without blocking | |
4803 | * 1 if a decrement succeeded. | |
4804 | * | |
4805 | * If a completion is being used as a counting completion, | |
4806 | * attempt to decrement the counter without blocking. This | |
4807 | * enables us to avoid waiting if the resource the completion | |
4808 | * is protecting is not available. | |
4809 | */ | |
4810 | bool try_wait_for_completion(struct completion *x) | |
4811 | { | |
4812 | int ret = 1; | |
4813 | ||
4814 | spin_lock_irq(&x->wait.lock); | |
4815 | if (!x->done) | |
4816 | ret = 0; | |
4817 | else | |
4818 | x->done--; | |
4819 | spin_unlock_irq(&x->wait.lock); | |
4820 | return ret; | |
4821 | } | |
4822 | EXPORT_SYMBOL(try_wait_for_completion); | |
4823 | ||
4824 | /** | |
4825 | * completion_done - Test to see if a completion has any waiters | |
4826 | * @x: completion structure | |
4827 | * | |
4828 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4829 | * 1 if there are no waiters. | |
4830 | * | |
4831 | */ | |
4832 | bool completion_done(struct completion *x) | |
4833 | { | |
4834 | int ret = 1; | |
4835 | ||
4836 | spin_lock_irq(&x->wait.lock); | |
4837 | if (!x->done) | |
4838 | ret = 0; | |
4839 | spin_unlock_irq(&x->wait.lock); | |
4840 | return ret; | |
4841 | } | |
4842 | EXPORT_SYMBOL(completion_done); | |
4843 | ||
8cbbe86d AK |
4844 | static long __sched |
4845 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4846 | { |
0fec171c IM |
4847 | unsigned long flags; |
4848 | wait_queue_t wait; | |
4849 | ||
4850 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4851 | |
8cbbe86d | 4852 | __set_current_state(state); |
1da177e4 | 4853 | |
8cbbe86d AK |
4854 | spin_lock_irqsave(&q->lock, flags); |
4855 | __add_wait_queue(q, &wait); | |
4856 | spin_unlock(&q->lock); | |
4857 | timeout = schedule_timeout(timeout); | |
4858 | spin_lock_irq(&q->lock); | |
4859 | __remove_wait_queue(q, &wait); | |
4860 | spin_unlock_irqrestore(&q->lock, flags); | |
4861 | ||
4862 | return timeout; | |
4863 | } | |
4864 | ||
4865 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4866 | { | |
4867 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4868 | } |
1da177e4 LT |
4869 | EXPORT_SYMBOL(interruptible_sleep_on); |
4870 | ||
0fec171c | 4871 | long __sched |
95cdf3b7 | 4872 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4873 | { |
8cbbe86d | 4874 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4875 | } |
1da177e4 LT |
4876 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4877 | ||
0fec171c | 4878 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4879 | { |
8cbbe86d | 4880 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4881 | } |
1da177e4 LT |
4882 | EXPORT_SYMBOL(sleep_on); |
4883 | ||
0fec171c | 4884 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4885 | { |
8cbbe86d | 4886 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4887 | } |
1da177e4 LT |
4888 | EXPORT_SYMBOL(sleep_on_timeout); |
4889 | ||
b29739f9 IM |
4890 | #ifdef CONFIG_RT_MUTEXES |
4891 | ||
4892 | /* | |
4893 | * rt_mutex_setprio - set the current priority of a task | |
4894 | * @p: task | |
4895 | * @prio: prio value (kernel-internal form) | |
4896 | * | |
4897 | * This function changes the 'effective' priority of a task. It does | |
4898 | * not touch ->normal_prio like __setscheduler(). | |
4899 | * | |
4900 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4901 | */ | |
36c8b586 | 4902 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4903 | { |
4904 | unsigned long flags; | |
83b699ed | 4905 | int oldprio, on_rq, running; |
70b97a7f | 4906 | struct rq *rq; |
cb469845 | 4907 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4908 | |
4909 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4910 | ||
4911 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4912 | update_rq_clock(rq); |
b29739f9 | 4913 | |
d5f9f942 | 4914 | oldprio = p->prio; |
dd41f596 | 4915 | on_rq = p->se.on_rq; |
051a1d1a | 4916 | running = task_current(rq, p); |
0e1f3483 | 4917 | if (on_rq) |
69be72c1 | 4918 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4919 | if (running) |
4920 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4921 | |
4922 | if (rt_prio(prio)) | |
4923 | p->sched_class = &rt_sched_class; | |
4924 | else | |
4925 | p->sched_class = &fair_sched_class; | |
4926 | ||
b29739f9 IM |
4927 | p->prio = prio; |
4928 | ||
0e1f3483 HS |
4929 | if (running) |
4930 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4931 | if (on_rq) { |
8159f87e | 4932 | enqueue_task(rq, p, 0); |
cb469845 SR |
4933 | |
4934 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4935 | } |
4936 | task_rq_unlock(rq, &flags); | |
4937 | } | |
4938 | ||
4939 | #endif | |
4940 | ||
36c8b586 | 4941 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4942 | { |
dd41f596 | 4943 | int old_prio, delta, on_rq; |
1da177e4 | 4944 | unsigned long flags; |
70b97a7f | 4945 | struct rq *rq; |
1da177e4 LT |
4946 | |
4947 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4948 | return; | |
4949 | /* | |
4950 | * We have to be careful, if called from sys_setpriority(), | |
4951 | * the task might be in the middle of scheduling on another CPU. | |
4952 | */ | |
4953 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4954 | update_rq_clock(rq); |
1da177e4 LT |
4955 | /* |
4956 | * The RT priorities are set via sched_setscheduler(), but we still | |
4957 | * allow the 'normal' nice value to be set - but as expected | |
4958 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4959 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4960 | */ |
e05606d3 | 4961 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4962 | p->static_prio = NICE_TO_PRIO(nice); |
4963 | goto out_unlock; | |
4964 | } | |
dd41f596 | 4965 | on_rq = p->se.on_rq; |
c09595f6 | 4966 | if (on_rq) |
69be72c1 | 4967 | dequeue_task(rq, p, 0); |
1da177e4 | 4968 | |
1da177e4 | 4969 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4970 | set_load_weight(p); |
b29739f9 IM |
4971 | old_prio = p->prio; |
4972 | p->prio = effective_prio(p); | |
4973 | delta = p->prio - old_prio; | |
1da177e4 | 4974 | |
dd41f596 | 4975 | if (on_rq) { |
8159f87e | 4976 | enqueue_task(rq, p, 0); |
1da177e4 | 4977 | /* |
d5f9f942 AM |
4978 | * If the task increased its priority or is running and |
4979 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4980 | */ |
d5f9f942 | 4981 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4982 | resched_task(rq->curr); |
4983 | } | |
4984 | out_unlock: | |
4985 | task_rq_unlock(rq, &flags); | |
4986 | } | |
1da177e4 LT |
4987 | EXPORT_SYMBOL(set_user_nice); |
4988 | ||
e43379f1 MM |
4989 | /* |
4990 | * can_nice - check if a task can reduce its nice value | |
4991 | * @p: task | |
4992 | * @nice: nice value | |
4993 | */ | |
36c8b586 | 4994 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4995 | { |
024f4747 MM |
4996 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4997 | int nice_rlim = 20 - nice; | |
48f24c4d | 4998 | |
e43379f1 MM |
4999 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5000 | capable(CAP_SYS_NICE)); | |
5001 | } | |
5002 | ||
1da177e4 LT |
5003 | #ifdef __ARCH_WANT_SYS_NICE |
5004 | ||
5005 | /* | |
5006 | * sys_nice - change the priority of the current process. | |
5007 | * @increment: priority increment | |
5008 | * | |
5009 | * sys_setpriority is a more generic, but much slower function that | |
5010 | * does similar things. | |
5011 | */ | |
5012 | asmlinkage long sys_nice(int increment) | |
5013 | { | |
48f24c4d | 5014 | long nice, retval; |
1da177e4 LT |
5015 | |
5016 | /* | |
5017 | * Setpriority might change our priority at the same moment. | |
5018 | * We don't have to worry. Conceptually one call occurs first | |
5019 | * and we have a single winner. | |
5020 | */ | |
e43379f1 MM |
5021 | if (increment < -40) |
5022 | increment = -40; | |
1da177e4 LT |
5023 | if (increment > 40) |
5024 | increment = 40; | |
5025 | ||
5026 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5027 | if (nice < -20) | |
5028 | nice = -20; | |
5029 | if (nice > 19) | |
5030 | nice = 19; | |
5031 | ||
e43379f1 MM |
5032 | if (increment < 0 && !can_nice(current, nice)) |
5033 | return -EPERM; | |
5034 | ||
1da177e4 LT |
5035 | retval = security_task_setnice(current, nice); |
5036 | if (retval) | |
5037 | return retval; | |
5038 | ||
5039 | set_user_nice(current, nice); | |
5040 | return 0; | |
5041 | } | |
5042 | ||
5043 | #endif | |
5044 | ||
5045 | /** | |
5046 | * task_prio - return the priority value of a given task. | |
5047 | * @p: the task in question. | |
5048 | * | |
5049 | * This is the priority value as seen by users in /proc. | |
5050 | * RT tasks are offset by -200. Normal tasks are centered | |
5051 | * around 0, value goes from -16 to +15. | |
5052 | */ | |
36c8b586 | 5053 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5054 | { |
5055 | return p->prio - MAX_RT_PRIO; | |
5056 | } | |
5057 | ||
5058 | /** | |
5059 | * task_nice - return the nice value of a given task. | |
5060 | * @p: the task in question. | |
5061 | */ | |
36c8b586 | 5062 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5063 | { |
5064 | return TASK_NICE(p); | |
5065 | } | |
150d8bed | 5066 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5067 | |
5068 | /** | |
5069 | * idle_cpu - is a given cpu idle currently? | |
5070 | * @cpu: the processor in question. | |
5071 | */ | |
5072 | int idle_cpu(int cpu) | |
5073 | { | |
5074 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5075 | } | |
5076 | ||
1da177e4 LT |
5077 | /** |
5078 | * idle_task - return the idle task for a given cpu. | |
5079 | * @cpu: the processor in question. | |
5080 | */ | |
36c8b586 | 5081 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5082 | { |
5083 | return cpu_rq(cpu)->idle; | |
5084 | } | |
5085 | ||
5086 | /** | |
5087 | * find_process_by_pid - find a process with a matching PID value. | |
5088 | * @pid: the pid in question. | |
5089 | */ | |
a9957449 | 5090 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5091 | { |
228ebcbe | 5092 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5093 | } |
5094 | ||
5095 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5096 | static void |
5097 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5098 | { |
dd41f596 | 5099 | BUG_ON(p->se.on_rq); |
48f24c4d | 5100 | |
1da177e4 | 5101 | p->policy = policy; |
dd41f596 IM |
5102 | switch (p->policy) { |
5103 | case SCHED_NORMAL: | |
5104 | case SCHED_BATCH: | |
5105 | case SCHED_IDLE: | |
5106 | p->sched_class = &fair_sched_class; | |
5107 | break; | |
5108 | case SCHED_FIFO: | |
5109 | case SCHED_RR: | |
5110 | p->sched_class = &rt_sched_class; | |
5111 | break; | |
5112 | } | |
5113 | ||
1da177e4 | 5114 | p->rt_priority = prio; |
b29739f9 IM |
5115 | p->normal_prio = normal_prio(p); |
5116 | /* we are holding p->pi_lock already */ | |
5117 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5118 | set_load_weight(p); |
1da177e4 LT |
5119 | } |
5120 | ||
961ccddd RR |
5121 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5122 | struct sched_param *param, bool user) | |
1da177e4 | 5123 | { |
83b699ed | 5124 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5125 | unsigned long flags; |
cb469845 | 5126 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5127 | struct rq *rq; |
1da177e4 | 5128 | |
66e5393a SR |
5129 | /* may grab non-irq protected spin_locks */ |
5130 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5131 | recheck: |
5132 | /* double check policy once rq lock held */ | |
5133 | if (policy < 0) | |
5134 | policy = oldpolicy = p->policy; | |
5135 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5136 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5137 | policy != SCHED_IDLE) | |
b0a9499c | 5138 | return -EINVAL; |
1da177e4 LT |
5139 | /* |
5140 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5141 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5142 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5143 | */ |
5144 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5145 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5146 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5147 | return -EINVAL; |
e05606d3 | 5148 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5149 | return -EINVAL; |
5150 | ||
37e4ab3f OC |
5151 | /* |
5152 | * Allow unprivileged RT tasks to decrease priority: | |
5153 | */ | |
961ccddd | 5154 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5155 | if (rt_policy(policy)) { |
8dc3e909 | 5156 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5157 | |
5158 | if (!lock_task_sighand(p, &flags)) | |
5159 | return -ESRCH; | |
5160 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5161 | unlock_task_sighand(p, &flags); | |
5162 | ||
5163 | /* can't set/change the rt policy */ | |
5164 | if (policy != p->policy && !rlim_rtprio) | |
5165 | return -EPERM; | |
5166 | ||
5167 | /* can't increase priority */ | |
5168 | if (param->sched_priority > p->rt_priority && | |
5169 | param->sched_priority > rlim_rtprio) | |
5170 | return -EPERM; | |
5171 | } | |
dd41f596 IM |
5172 | /* |
5173 | * Like positive nice levels, dont allow tasks to | |
5174 | * move out of SCHED_IDLE either: | |
5175 | */ | |
5176 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5177 | return -EPERM; | |
5fe1d75f | 5178 | |
37e4ab3f OC |
5179 | /* can't change other user's priorities */ |
5180 | if ((current->euid != p->euid) && | |
5181 | (current->euid != p->uid)) | |
5182 | return -EPERM; | |
5183 | } | |
1da177e4 | 5184 | |
725aad24 | 5185 | if (user) { |
b68aa230 | 5186 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5187 | /* |
5188 | * Do not allow realtime tasks into groups that have no runtime | |
5189 | * assigned. | |
5190 | */ | |
9a7e0b18 PZ |
5191 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5192 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5193 | return -EPERM; |
b68aa230 PZ |
5194 | #endif |
5195 | ||
725aad24 JF |
5196 | retval = security_task_setscheduler(p, policy, param); |
5197 | if (retval) | |
5198 | return retval; | |
5199 | } | |
5200 | ||
b29739f9 IM |
5201 | /* |
5202 | * make sure no PI-waiters arrive (or leave) while we are | |
5203 | * changing the priority of the task: | |
5204 | */ | |
5205 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5206 | /* |
5207 | * To be able to change p->policy safely, the apropriate | |
5208 | * runqueue lock must be held. | |
5209 | */ | |
b29739f9 | 5210 | rq = __task_rq_lock(p); |
1da177e4 LT |
5211 | /* recheck policy now with rq lock held */ |
5212 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5213 | policy = oldpolicy = -1; | |
b29739f9 IM |
5214 | __task_rq_unlock(rq); |
5215 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5216 | goto recheck; |
5217 | } | |
2daa3577 | 5218 | update_rq_clock(rq); |
dd41f596 | 5219 | on_rq = p->se.on_rq; |
051a1d1a | 5220 | running = task_current(rq, p); |
0e1f3483 | 5221 | if (on_rq) |
2e1cb74a | 5222 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5223 | if (running) |
5224 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5225 | |
1da177e4 | 5226 | oldprio = p->prio; |
dd41f596 | 5227 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5228 | |
0e1f3483 HS |
5229 | if (running) |
5230 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5231 | if (on_rq) { |
5232 | activate_task(rq, p, 0); | |
cb469845 SR |
5233 | |
5234 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5235 | } |
b29739f9 IM |
5236 | __task_rq_unlock(rq); |
5237 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5238 | ||
95e02ca9 TG |
5239 | rt_mutex_adjust_pi(p); |
5240 | ||
1da177e4 LT |
5241 | return 0; |
5242 | } | |
961ccddd RR |
5243 | |
5244 | /** | |
5245 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5246 | * @p: the task in question. | |
5247 | * @policy: new policy. | |
5248 | * @param: structure containing the new RT priority. | |
5249 | * | |
5250 | * NOTE that the task may be already dead. | |
5251 | */ | |
5252 | int sched_setscheduler(struct task_struct *p, int policy, | |
5253 | struct sched_param *param) | |
5254 | { | |
5255 | return __sched_setscheduler(p, policy, param, true); | |
5256 | } | |
1da177e4 LT |
5257 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5258 | ||
961ccddd RR |
5259 | /** |
5260 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5261 | * @p: the task in question. | |
5262 | * @policy: new policy. | |
5263 | * @param: structure containing the new RT priority. | |
5264 | * | |
5265 | * Just like sched_setscheduler, only don't bother checking if the | |
5266 | * current context has permission. For example, this is needed in | |
5267 | * stop_machine(): we create temporary high priority worker threads, | |
5268 | * but our caller might not have that capability. | |
5269 | */ | |
5270 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5271 | struct sched_param *param) | |
5272 | { | |
5273 | return __sched_setscheduler(p, policy, param, false); | |
5274 | } | |
5275 | ||
95cdf3b7 IM |
5276 | static int |
5277 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5278 | { |
1da177e4 LT |
5279 | struct sched_param lparam; |
5280 | struct task_struct *p; | |
36c8b586 | 5281 | int retval; |
1da177e4 LT |
5282 | |
5283 | if (!param || pid < 0) | |
5284 | return -EINVAL; | |
5285 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5286 | return -EFAULT; | |
5fe1d75f ON |
5287 | |
5288 | rcu_read_lock(); | |
5289 | retval = -ESRCH; | |
1da177e4 | 5290 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5291 | if (p != NULL) |
5292 | retval = sched_setscheduler(p, policy, &lparam); | |
5293 | rcu_read_unlock(); | |
36c8b586 | 5294 | |
1da177e4 LT |
5295 | return retval; |
5296 | } | |
5297 | ||
5298 | /** | |
5299 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5300 | * @pid: the pid in question. | |
5301 | * @policy: new policy. | |
5302 | * @param: structure containing the new RT priority. | |
5303 | */ | |
41a2d6cf IM |
5304 | asmlinkage long |
5305 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5306 | { |
c21761f1 JB |
5307 | /* negative values for policy are not valid */ |
5308 | if (policy < 0) | |
5309 | return -EINVAL; | |
5310 | ||
1da177e4 LT |
5311 | return do_sched_setscheduler(pid, policy, param); |
5312 | } | |
5313 | ||
5314 | /** | |
5315 | * sys_sched_setparam - set/change the RT priority of a thread | |
5316 | * @pid: the pid in question. | |
5317 | * @param: structure containing the new RT priority. | |
5318 | */ | |
5319 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5320 | { | |
5321 | return do_sched_setscheduler(pid, -1, param); | |
5322 | } | |
5323 | ||
5324 | /** | |
5325 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5326 | * @pid: the pid in question. | |
5327 | */ | |
5328 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5329 | { | |
36c8b586 | 5330 | struct task_struct *p; |
3a5c359a | 5331 | int retval; |
1da177e4 LT |
5332 | |
5333 | if (pid < 0) | |
3a5c359a | 5334 | return -EINVAL; |
1da177e4 LT |
5335 | |
5336 | retval = -ESRCH; | |
5337 | read_lock(&tasklist_lock); | |
5338 | p = find_process_by_pid(pid); | |
5339 | if (p) { | |
5340 | retval = security_task_getscheduler(p); | |
5341 | if (!retval) | |
5342 | retval = p->policy; | |
5343 | } | |
5344 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5345 | return retval; |
5346 | } | |
5347 | ||
5348 | /** | |
5349 | * sys_sched_getscheduler - get the RT priority of a thread | |
5350 | * @pid: the pid in question. | |
5351 | * @param: structure containing the RT priority. | |
5352 | */ | |
5353 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5354 | { | |
5355 | struct sched_param lp; | |
36c8b586 | 5356 | struct task_struct *p; |
3a5c359a | 5357 | int retval; |
1da177e4 LT |
5358 | |
5359 | if (!param || pid < 0) | |
3a5c359a | 5360 | return -EINVAL; |
1da177e4 LT |
5361 | |
5362 | read_lock(&tasklist_lock); | |
5363 | p = find_process_by_pid(pid); | |
5364 | retval = -ESRCH; | |
5365 | if (!p) | |
5366 | goto out_unlock; | |
5367 | ||
5368 | retval = security_task_getscheduler(p); | |
5369 | if (retval) | |
5370 | goto out_unlock; | |
5371 | ||
5372 | lp.sched_priority = p->rt_priority; | |
5373 | read_unlock(&tasklist_lock); | |
5374 | ||
5375 | /* | |
5376 | * This one might sleep, we cannot do it with a spinlock held ... | |
5377 | */ | |
5378 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5379 | ||
1da177e4 LT |
5380 | return retval; |
5381 | ||
5382 | out_unlock: | |
5383 | read_unlock(&tasklist_lock); | |
5384 | return retval; | |
5385 | } | |
5386 | ||
b53e921b | 5387 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5388 | { |
1da177e4 | 5389 | cpumask_t cpus_allowed; |
b53e921b | 5390 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5391 | struct task_struct *p; |
5392 | int retval; | |
1da177e4 | 5393 | |
95402b38 | 5394 | get_online_cpus(); |
1da177e4 LT |
5395 | read_lock(&tasklist_lock); |
5396 | ||
5397 | p = find_process_by_pid(pid); | |
5398 | if (!p) { | |
5399 | read_unlock(&tasklist_lock); | |
95402b38 | 5400 | put_online_cpus(); |
1da177e4 LT |
5401 | return -ESRCH; |
5402 | } | |
5403 | ||
5404 | /* | |
5405 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5406 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5407 | * usage count and then drop tasklist_lock. |
5408 | */ | |
5409 | get_task_struct(p); | |
5410 | read_unlock(&tasklist_lock); | |
5411 | ||
5412 | retval = -EPERM; | |
5413 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5414 | !capable(CAP_SYS_NICE)) | |
5415 | goto out_unlock; | |
5416 | ||
e7834f8f DQ |
5417 | retval = security_task_setscheduler(p, 0, NULL); |
5418 | if (retval) | |
5419 | goto out_unlock; | |
5420 | ||
f9a86fcb | 5421 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5422 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5423 | again: |
7c16ec58 | 5424 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5425 | |
8707d8b8 | 5426 | if (!retval) { |
f9a86fcb | 5427 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5428 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5429 | /* | |
5430 | * We must have raced with a concurrent cpuset | |
5431 | * update. Just reset the cpus_allowed to the | |
5432 | * cpuset's cpus_allowed | |
5433 | */ | |
5434 | new_mask = cpus_allowed; | |
5435 | goto again; | |
5436 | } | |
5437 | } | |
1da177e4 LT |
5438 | out_unlock: |
5439 | put_task_struct(p); | |
95402b38 | 5440 | put_online_cpus(); |
1da177e4 LT |
5441 | return retval; |
5442 | } | |
5443 | ||
5444 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5445 | cpumask_t *new_mask) | |
5446 | { | |
5447 | if (len < sizeof(cpumask_t)) { | |
5448 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5449 | } else if (len > sizeof(cpumask_t)) { | |
5450 | len = sizeof(cpumask_t); | |
5451 | } | |
5452 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5453 | } | |
5454 | ||
5455 | /** | |
5456 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5457 | * @pid: pid of the process | |
5458 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5459 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5460 | */ | |
5461 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5462 | unsigned long __user *user_mask_ptr) | |
5463 | { | |
5464 | cpumask_t new_mask; | |
5465 | int retval; | |
5466 | ||
5467 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5468 | if (retval) | |
5469 | return retval; | |
5470 | ||
b53e921b | 5471 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5472 | } |
5473 | ||
1da177e4 LT |
5474 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5475 | { | |
36c8b586 | 5476 | struct task_struct *p; |
1da177e4 | 5477 | int retval; |
1da177e4 | 5478 | |
95402b38 | 5479 | get_online_cpus(); |
1da177e4 LT |
5480 | read_lock(&tasklist_lock); |
5481 | ||
5482 | retval = -ESRCH; | |
5483 | p = find_process_by_pid(pid); | |
5484 | if (!p) | |
5485 | goto out_unlock; | |
5486 | ||
e7834f8f DQ |
5487 | retval = security_task_getscheduler(p); |
5488 | if (retval) | |
5489 | goto out_unlock; | |
5490 | ||
2f7016d9 | 5491 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5492 | |
5493 | out_unlock: | |
5494 | read_unlock(&tasklist_lock); | |
95402b38 | 5495 | put_online_cpus(); |
1da177e4 | 5496 | |
9531b62f | 5497 | return retval; |
1da177e4 LT |
5498 | } |
5499 | ||
5500 | /** | |
5501 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5502 | * @pid: pid of the process | |
5503 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5504 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5505 | */ | |
5506 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5507 | unsigned long __user *user_mask_ptr) | |
5508 | { | |
5509 | int ret; | |
5510 | cpumask_t mask; | |
5511 | ||
5512 | if (len < sizeof(cpumask_t)) | |
5513 | return -EINVAL; | |
5514 | ||
5515 | ret = sched_getaffinity(pid, &mask); | |
5516 | if (ret < 0) | |
5517 | return ret; | |
5518 | ||
5519 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5520 | return -EFAULT; | |
5521 | ||
5522 | return sizeof(cpumask_t); | |
5523 | } | |
5524 | ||
5525 | /** | |
5526 | * sys_sched_yield - yield the current processor to other threads. | |
5527 | * | |
dd41f596 IM |
5528 | * This function yields the current CPU to other tasks. If there are no |
5529 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5530 | */ |
5531 | asmlinkage long sys_sched_yield(void) | |
5532 | { | |
70b97a7f | 5533 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5534 | |
2d72376b | 5535 | schedstat_inc(rq, yld_count); |
4530d7ab | 5536 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5537 | |
5538 | /* | |
5539 | * Since we are going to call schedule() anyway, there's | |
5540 | * no need to preempt or enable interrupts: | |
5541 | */ | |
5542 | __release(rq->lock); | |
8a25d5de | 5543 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5544 | _raw_spin_unlock(&rq->lock); |
5545 | preempt_enable_no_resched(); | |
5546 | ||
5547 | schedule(); | |
5548 | ||
5549 | return 0; | |
5550 | } | |
5551 | ||
e7b38404 | 5552 | static void __cond_resched(void) |
1da177e4 | 5553 | { |
8e0a43d8 IM |
5554 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5555 | __might_sleep(__FILE__, __LINE__); | |
5556 | #endif | |
5bbcfd90 IM |
5557 | /* |
5558 | * The BKS might be reacquired before we have dropped | |
5559 | * PREEMPT_ACTIVE, which could trigger a second | |
5560 | * cond_resched() call. | |
5561 | */ | |
1da177e4 LT |
5562 | do { |
5563 | add_preempt_count(PREEMPT_ACTIVE); | |
5564 | schedule(); | |
5565 | sub_preempt_count(PREEMPT_ACTIVE); | |
5566 | } while (need_resched()); | |
5567 | } | |
5568 | ||
02b67cc3 | 5569 | int __sched _cond_resched(void) |
1da177e4 | 5570 | { |
9414232f IM |
5571 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5572 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5573 | __cond_resched(); |
5574 | return 1; | |
5575 | } | |
5576 | return 0; | |
5577 | } | |
02b67cc3 | 5578 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5579 | |
5580 | /* | |
5581 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5582 | * call schedule, and on return reacquire the lock. | |
5583 | * | |
41a2d6cf | 5584 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5585 | * operations here to prevent schedule() from being called twice (once via |
5586 | * spin_unlock(), once by hand). | |
5587 | */ | |
95cdf3b7 | 5588 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5589 | { |
95c354fe | 5590 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5591 | int ret = 0; |
5592 | ||
95c354fe | 5593 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5594 | spin_unlock(lock); |
95c354fe NP |
5595 | if (resched && need_resched()) |
5596 | __cond_resched(); | |
5597 | else | |
5598 | cpu_relax(); | |
6df3cecb | 5599 | ret = 1; |
1da177e4 | 5600 | spin_lock(lock); |
1da177e4 | 5601 | } |
6df3cecb | 5602 | return ret; |
1da177e4 | 5603 | } |
1da177e4 LT |
5604 | EXPORT_SYMBOL(cond_resched_lock); |
5605 | ||
5606 | int __sched cond_resched_softirq(void) | |
5607 | { | |
5608 | BUG_ON(!in_softirq()); | |
5609 | ||
9414232f | 5610 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5611 | local_bh_enable(); |
1da177e4 LT |
5612 | __cond_resched(); |
5613 | local_bh_disable(); | |
5614 | return 1; | |
5615 | } | |
5616 | return 0; | |
5617 | } | |
1da177e4 LT |
5618 | EXPORT_SYMBOL(cond_resched_softirq); |
5619 | ||
1da177e4 LT |
5620 | /** |
5621 | * yield - yield the current processor to other threads. | |
5622 | * | |
72fd4a35 | 5623 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5624 | * thread runnable and calls sys_sched_yield(). |
5625 | */ | |
5626 | void __sched yield(void) | |
5627 | { | |
5628 | set_current_state(TASK_RUNNING); | |
5629 | sys_sched_yield(); | |
5630 | } | |
1da177e4 LT |
5631 | EXPORT_SYMBOL(yield); |
5632 | ||
5633 | /* | |
41a2d6cf | 5634 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5635 | * that process accounting knows that this is a task in IO wait state. |
5636 | * | |
5637 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5638 | * has set its backing_dev_info: the queue against which it should throttle) | |
5639 | */ | |
5640 | void __sched io_schedule(void) | |
5641 | { | |
70b97a7f | 5642 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5643 | |
0ff92245 | 5644 | delayacct_blkio_start(); |
1da177e4 LT |
5645 | atomic_inc(&rq->nr_iowait); |
5646 | schedule(); | |
5647 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5648 | delayacct_blkio_end(); |
1da177e4 | 5649 | } |
1da177e4 LT |
5650 | EXPORT_SYMBOL(io_schedule); |
5651 | ||
5652 | long __sched io_schedule_timeout(long timeout) | |
5653 | { | |
70b97a7f | 5654 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5655 | long ret; |
5656 | ||
0ff92245 | 5657 | delayacct_blkio_start(); |
1da177e4 LT |
5658 | atomic_inc(&rq->nr_iowait); |
5659 | ret = schedule_timeout(timeout); | |
5660 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5661 | delayacct_blkio_end(); |
1da177e4 LT |
5662 | return ret; |
5663 | } | |
5664 | ||
5665 | /** | |
5666 | * sys_sched_get_priority_max - return maximum RT priority. | |
5667 | * @policy: scheduling class. | |
5668 | * | |
5669 | * this syscall returns the maximum rt_priority that can be used | |
5670 | * by a given scheduling class. | |
5671 | */ | |
5672 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5673 | { | |
5674 | int ret = -EINVAL; | |
5675 | ||
5676 | switch (policy) { | |
5677 | case SCHED_FIFO: | |
5678 | case SCHED_RR: | |
5679 | ret = MAX_USER_RT_PRIO-1; | |
5680 | break; | |
5681 | case SCHED_NORMAL: | |
b0a9499c | 5682 | case SCHED_BATCH: |
dd41f596 | 5683 | case SCHED_IDLE: |
1da177e4 LT |
5684 | ret = 0; |
5685 | break; | |
5686 | } | |
5687 | return ret; | |
5688 | } | |
5689 | ||
5690 | /** | |
5691 | * sys_sched_get_priority_min - return minimum RT priority. | |
5692 | * @policy: scheduling class. | |
5693 | * | |
5694 | * this syscall returns the minimum rt_priority that can be used | |
5695 | * by a given scheduling class. | |
5696 | */ | |
5697 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5698 | { | |
5699 | int ret = -EINVAL; | |
5700 | ||
5701 | switch (policy) { | |
5702 | case SCHED_FIFO: | |
5703 | case SCHED_RR: | |
5704 | ret = 1; | |
5705 | break; | |
5706 | case SCHED_NORMAL: | |
b0a9499c | 5707 | case SCHED_BATCH: |
dd41f596 | 5708 | case SCHED_IDLE: |
1da177e4 LT |
5709 | ret = 0; |
5710 | } | |
5711 | return ret; | |
5712 | } | |
5713 | ||
5714 | /** | |
5715 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5716 | * @pid: pid of the process. | |
5717 | * @interval: userspace pointer to the timeslice value. | |
5718 | * | |
5719 | * this syscall writes the default timeslice value of a given process | |
5720 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5721 | */ | |
5722 | asmlinkage | |
5723 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5724 | { | |
36c8b586 | 5725 | struct task_struct *p; |
a4ec24b4 | 5726 | unsigned int time_slice; |
3a5c359a | 5727 | int retval; |
1da177e4 | 5728 | struct timespec t; |
1da177e4 LT |
5729 | |
5730 | if (pid < 0) | |
3a5c359a | 5731 | return -EINVAL; |
1da177e4 LT |
5732 | |
5733 | retval = -ESRCH; | |
5734 | read_lock(&tasklist_lock); | |
5735 | p = find_process_by_pid(pid); | |
5736 | if (!p) | |
5737 | goto out_unlock; | |
5738 | ||
5739 | retval = security_task_getscheduler(p); | |
5740 | if (retval) | |
5741 | goto out_unlock; | |
5742 | ||
77034937 IM |
5743 | /* |
5744 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5745 | * tasks that are on an otherwise idle runqueue: | |
5746 | */ | |
5747 | time_slice = 0; | |
5748 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5749 | time_slice = DEF_TIMESLICE; |
1868f958 | 5750 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5751 | struct sched_entity *se = &p->se; |
5752 | unsigned long flags; | |
5753 | struct rq *rq; | |
5754 | ||
5755 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5756 | if (rq->cfs.load.weight) |
5757 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5758 | task_rq_unlock(rq, &flags); |
5759 | } | |
1da177e4 | 5760 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5761 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5762 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5763 | return retval; |
3a5c359a | 5764 | |
1da177e4 LT |
5765 | out_unlock: |
5766 | read_unlock(&tasklist_lock); | |
5767 | return retval; | |
5768 | } | |
5769 | ||
7c731e0a | 5770 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5771 | |
82a1fcb9 | 5772 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5773 | { |
1da177e4 | 5774 | unsigned long free = 0; |
36c8b586 | 5775 | unsigned state; |
1da177e4 | 5776 | |
1da177e4 | 5777 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5778 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5779 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5780 | #if BITS_PER_LONG == 32 |
1da177e4 | 5781 | if (state == TASK_RUNNING) |
cc4ea795 | 5782 | printk(KERN_CONT " running "); |
1da177e4 | 5783 | else |
cc4ea795 | 5784 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5785 | #else |
5786 | if (state == TASK_RUNNING) | |
cc4ea795 | 5787 | printk(KERN_CONT " running task "); |
1da177e4 | 5788 | else |
cc4ea795 | 5789 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5790 | #endif |
5791 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5792 | { | |
10ebffde | 5793 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5794 | while (!*n) |
5795 | n++; | |
10ebffde | 5796 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5797 | } |
5798 | #endif | |
ba25f9dc | 5799 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5800 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5801 | |
5fb5e6de | 5802 | show_stack(p, NULL); |
1da177e4 LT |
5803 | } |
5804 | ||
e59e2ae2 | 5805 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5806 | { |
36c8b586 | 5807 | struct task_struct *g, *p; |
1da177e4 | 5808 | |
4bd77321 IM |
5809 | #if BITS_PER_LONG == 32 |
5810 | printk(KERN_INFO | |
5811 | " task PC stack pid father\n"); | |
1da177e4 | 5812 | #else |
4bd77321 IM |
5813 | printk(KERN_INFO |
5814 | " task PC stack pid father\n"); | |
1da177e4 LT |
5815 | #endif |
5816 | read_lock(&tasklist_lock); | |
5817 | do_each_thread(g, p) { | |
5818 | /* | |
5819 | * reset the NMI-timeout, listing all files on a slow | |
5820 | * console might take alot of time: | |
5821 | */ | |
5822 | touch_nmi_watchdog(); | |
39bc89fd | 5823 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5824 | sched_show_task(p); |
1da177e4 LT |
5825 | } while_each_thread(g, p); |
5826 | ||
04c9167f JF |
5827 | touch_all_softlockup_watchdogs(); |
5828 | ||
dd41f596 IM |
5829 | #ifdef CONFIG_SCHED_DEBUG |
5830 | sysrq_sched_debug_show(); | |
5831 | #endif | |
1da177e4 | 5832 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5833 | /* |
5834 | * Only show locks if all tasks are dumped: | |
5835 | */ | |
5836 | if (state_filter == -1) | |
5837 | debug_show_all_locks(); | |
1da177e4 LT |
5838 | } |
5839 | ||
1df21055 IM |
5840 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5841 | { | |
dd41f596 | 5842 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5843 | } |
5844 | ||
f340c0d1 IM |
5845 | /** |
5846 | * init_idle - set up an idle thread for a given CPU | |
5847 | * @idle: task in question | |
5848 | * @cpu: cpu the idle task belongs to | |
5849 | * | |
5850 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5851 | * flag, to make booting more robust. | |
5852 | */ | |
5c1e1767 | 5853 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5854 | { |
70b97a7f | 5855 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5856 | unsigned long flags; |
5857 | ||
dd41f596 IM |
5858 | __sched_fork(idle); |
5859 | idle->se.exec_start = sched_clock(); | |
5860 | ||
b29739f9 | 5861 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5862 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5863 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5864 | |
5865 | spin_lock_irqsave(&rq->lock, flags); | |
5866 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5867 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5868 | idle->oncpu = 1; | |
5869 | #endif | |
1da177e4 LT |
5870 | spin_unlock_irqrestore(&rq->lock, flags); |
5871 | ||
5872 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5873 | #if defined(CONFIG_PREEMPT) |
5874 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5875 | #else | |
a1261f54 | 5876 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5877 | #endif |
dd41f596 IM |
5878 | /* |
5879 | * The idle tasks have their own, simple scheduling class: | |
5880 | */ | |
5881 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5882 | } |
5883 | ||
5884 | /* | |
5885 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5886 | * indicates which cpus entered this state. This is used | |
5887 | * in the rcu update to wait only for active cpus. For system | |
5888 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5889 | * always be CPU_MASK_NONE. | |
5890 | */ | |
5891 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5892 | ||
19978ca6 IM |
5893 | /* |
5894 | * Increase the granularity value when there are more CPUs, | |
5895 | * because with more CPUs the 'effective latency' as visible | |
5896 | * to users decreases. But the relationship is not linear, | |
5897 | * so pick a second-best guess by going with the log2 of the | |
5898 | * number of CPUs. | |
5899 | * | |
5900 | * This idea comes from the SD scheduler of Con Kolivas: | |
5901 | */ | |
5902 | static inline void sched_init_granularity(void) | |
5903 | { | |
5904 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5905 | const unsigned long limit = 200000000; | |
5906 | ||
5907 | sysctl_sched_min_granularity *= factor; | |
5908 | if (sysctl_sched_min_granularity > limit) | |
5909 | sysctl_sched_min_granularity = limit; | |
5910 | ||
5911 | sysctl_sched_latency *= factor; | |
5912 | if (sysctl_sched_latency > limit) | |
5913 | sysctl_sched_latency = limit; | |
5914 | ||
5915 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5916 | |
5917 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5918 | } |
5919 | ||
1da177e4 LT |
5920 | #ifdef CONFIG_SMP |
5921 | /* | |
5922 | * This is how migration works: | |
5923 | * | |
70b97a7f | 5924 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5925 | * runqueue and wake up that CPU's migration thread. |
5926 | * 2) we down() the locked semaphore => thread blocks. | |
5927 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5928 | * thread off the CPU) | |
5929 | * 4) it gets the migration request and checks whether the migrated | |
5930 | * task is still in the wrong runqueue. | |
5931 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5932 | * it and puts it into the right queue. | |
5933 | * 6) migration thread up()s the semaphore. | |
5934 | * 7) we wake up and the migration is done. | |
5935 | */ | |
5936 | ||
5937 | /* | |
5938 | * Change a given task's CPU affinity. Migrate the thread to a | |
5939 | * proper CPU and schedule it away if the CPU it's executing on | |
5940 | * is removed from the allowed bitmask. | |
5941 | * | |
5942 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5943 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5944 | * call is not atomic; no spinlocks may be held. |
5945 | */ | |
cd8ba7cd | 5946 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5947 | { |
70b97a7f | 5948 | struct migration_req req; |
1da177e4 | 5949 | unsigned long flags; |
70b97a7f | 5950 | struct rq *rq; |
48f24c4d | 5951 | int ret = 0; |
1da177e4 LT |
5952 | |
5953 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5954 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5955 | ret = -EINVAL; |
5956 | goto out; | |
5957 | } | |
5958 | ||
9985b0ba DR |
5959 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5960 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5961 | ret = -EINVAL; | |
5962 | goto out; | |
5963 | } | |
5964 | ||
73fe6aae | 5965 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5966 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5967 | else { |
cd8ba7cd MT |
5968 | p->cpus_allowed = *new_mask; |
5969 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5970 | } |
5971 | ||
1da177e4 | 5972 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5973 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5974 | goto out; |
5975 | ||
cd8ba7cd | 5976 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5977 | /* Need help from migration thread: drop lock and wait. */ |
5978 | task_rq_unlock(rq, &flags); | |
5979 | wake_up_process(rq->migration_thread); | |
5980 | wait_for_completion(&req.done); | |
5981 | tlb_migrate_finish(p->mm); | |
5982 | return 0; | |
5983 | } | |
5984 | out: | |
5985 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5986 | |
1da177e4 LT |
5987 | return ret; |
5988 | } | |
cd8ba7cd | 5989 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5990 | |
5991 | /* | |
41a2d6cf | 5992 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5993 | * this because either it can't run here any more (set_cpus_allowed() |
5994 | * away from this CPU, or CPU going down), or because we're | |
5995 | * attempting to rebalance this task on exec (sched_exec). | |
5996 | * | |
5997 | * So we race with normal scheduler movements, but that's OK, as long | |
5998 | * as the task is no longer on this CPU. | |
efc30814 KK |
5999 | * |
6000 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6001 | */ |
efc30814 | 6002 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6003 | { |
70b97a7f | 6004 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6005 | int ret = 0, on_rq; |
1da177e4 | 6006 | |
e761b772 | 6007 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6008 | return ret; |
1da177e4 LT |
6009 | |
6010 | rq_src = cpu_rq(src_cpu); | |
6011 | rq_dest = cpu_rq(dest_cpu); | |
6012 | ||
6013 | double_rq_lock(rq_src, rq_dest); | |
6014 | /* Already moved. */ | |
6015 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6016 | goto done; |
1da177e4 LT |
6017 | /* Affinity changed (again). */ |
6018 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 6019 | goto fail; |
1da177e4 | 6020 | |
dd41f596 | 6021 | on_rq = p->se.on_rq; |
6e82a3be | 6022 | if (on_rq) |
2e1cb74a | 6023 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6024 | |
1da177e4 | 6025 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6026 | if (on_rq) { |
6027 | activate_task(rq_dest, p, 0); | |
15afe09b | 6028 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6029 | } |
b1e38734 | 6030 | done: |
efc30814 | 6031 | ret = 1; |
b1e38734 | 6032 | fail: |
1da177e4 | 6033 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6034 | return ret; |
1da177e4 LT |
6035 | } |
6036 | ||
6037 | /* | |
6038 | * migration_thread - this is a highprio system thread that performs | |
6039 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6040 | * another runqueue. | |
6041 | */ | |
95cdf3b7 | 6042 | static int migration_thread(void *data) |
1da177e4 | 6043 | { |
1da177e4 | 6044 | int cpu = (long)data; |
70b97a7f | 6045 | struct rq *rq; |
1da177e4 LT |
6046 | |
6047 | rq = cpu_rq(cpu); | |
6048 | BUG_ON(rq->migration_thread != current); | |
6049 | ||
6050 | set_current_state(TASK_INTERRUPTIBLE); | |
6051 | while (!kthread_should_stop()) { | |
70b97a7f | 6052 | struct migration_req *req; |
1da177e4 | 6053 | struct list_head *head; |
1da177e4 | 6054 | |
1da177e4 LT |
6055 | spin_lock_irq(&rq->lock); |
6056 | ||
6057 | if (cpu_is_offline(cpu)) { | |
6058 | spin_unlock_irq(&rq->lock); | |
6059 | goto wait_to_die; | |
6060 | } | |
6061 | ||
6062 | if (rq->active_balance) { | |
6063 | active_load_balance(rq, cpu); | |
6064 | rq->active_balance = 0; | |
6065 | } | |
6066 | ||
6067 | head = &rq->migration_queue; | |
6068 | ||
6069 | if (list_empty(head)) { | |
6070 | spin_unlock_irq(&rq->lock); | |
6071 | schedule(); | |
6072 | set_current_state(TASK_INTERRUPTIBLE); | |
6073 | continue; | |
6074 | } | |
70b97a7f | 6075 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6076 | list_del_init(head->next); |
6077 | ||
674311d5 NP |
6078 | spin_unlock(&rq->lock); |
6079 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6080 | local_irq_enable(); | |
1da177e4 LT |
6081 | |
6082 | complete(&req->done); | |
6083 | } | |
6084 | __set_current_state(TASK_RUNNING); | |
6085 | return 0; | |
6086 | ||
6087 | wait_to_die: | |
6088 | /* Wait for kthread_stop */ | |
6089 | set_current_state(TASK_INTERRUPTIBLE); | |
6090 | while (!kthread_should_stop()) { | |
6091 | schedule(); | |
6092 | set_current_state(TASK_INTERRUPTIBLE); | |
6093 | } | |
6094 | __set_current_state(TASK_RUNNING); | |
6095 | return 0; | |
6096 | } | |
6097 | ||
6098 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6099 | |
6100 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6101 | { | |
6102 | int ret; | |
6103 | ||
6104 | local_irq_disable(); | |
6105 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6106 | local_irq_enable(); | |
6107 | return ret; | |
6108 | } | |
6109 | ||
054b9108 | 6110 | /* |
3a4fa0a2 | 6111 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6112 | * NOTE: interrupts should be disabled by the caller |
6113 | */ | |
48f24c4d | 6114 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6115 | { |
efc30814 | 6116 | unsigned long flags; |
1da177e4 | 6117 | cpumask_t mask; |
70b97a7f IM |
6118 | struct rq *rq; |
6119 | int dest_cpu; | |
1da177e4 | 6120 | |
3a5c359a AK |
6121 | do { |
6122 | /* On same node? */ | |
6123 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6124 | cpus_and(mask, mask, p->cpus_allowed); | |
6125 | dest_cpu = any_online_cpu(mask); | |
6126 | ||
6127 | /* On any allowed CPU? */ | |
434d53b0 | 6128 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6129 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6130 | ||
6131 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6132 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6133 | cpumask_t cpus_allowed; |
6134 | ||
6135 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6136 | /* |
6137 | * Try to stay on the same cpuset, where the | |
6138 | * current cpuset may be a subset of all cpus. | |
6139 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6140 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6141 | * called within calls to cpuset_lock/cpuset_unlock. |
6142 | */ | |
3a5c359a | 6143 | rq = task_rq_lock(p, &flags); |
470fd646 | 6144 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6145 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6146 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6147 | |
3a5c359a AK |
6148 | /* |
6149 | * Don't tell them about moving exiting tasks or | |
6150 | * kernel threads (both mm NULL), since they never | |
6151 | * leave kernel. | |
6152 | */ | |
41a2d6cf | 6153 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6154 | printk(KERN_INFO "process %d (%s) no " |
6155 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6156 | task_pid_nr(p), p->comm, dead_cpu); |
6157 | } | |
3a5c359a | 6158 | } |
f7b4cddc | 6159 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6160 | } |
6161 | ||
6162 | /* | |
6163 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6164 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6165 | * for performance reasons the counter is not stricly tracking tasks to | |
6166 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6167 | * to keep the global sum constant after CPU-down: | |
6168 | */ | |
70b97a7f | 6169 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6170 | { |
7c16ec58 | 6171 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6172 | unsigned long flags; |
6173 | ||
6174 | local_irq_save(flags); | |
6175 | double_rq_lock(rq_src, rq_dest); | |
6176 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6177 | rq_src->nr_uninterruptible = 0; | |
6178 | double_rq_unlock(rq_src, rq_dest); | |
6179 | local_irq_restore(flags); | |
6180 | } | |
6181 | ||
6182 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6183 | static void migrate_live_tasks(int src_cpu) | |
6184 | { | |
48f24c4d | 6185 | struct task_struct *p, *t; |
1da177e4 | 6186 | |
f7b4cddc | 6187 | read_lock(&tasklist_lock); |
1da177e4 | 6188 | |
48f24c4d IM |
6189 | do_each_thread(t, p) { |
6190 | if (p == current) | |
1da177e4 LT |
6191 | continue; |
6192 | ||
48f24c4d IM |
6193 | if (task_cpu(p) == src_cpu) |
6194 | move_task_off_dead_cpu(src_cpu, p); | |
6195 | } while_each_thread(t, p); | |
1da177e4 | 6196 | |
f7b4cddc | 6197 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6198 | } |
6199 | ||
dd41f596 IM |
6200 | /* |
6201 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6202 | * It does so by boosting its priority to highest possible. |
6203 | * Used by CPU offline code. | |
1da177e4 LT |
6204 | */ |
6205 | void sched_idle_next(void) | |
6206 | { | |
48f24c4d | 6207 | int this_cpu = smp_processor_id(); |
70b97a7f | 6208 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6209 | struct task_struct *p = rq->idle; |
6210 | unsigned long flags; | |
6211 | ||
6212 | /* cpu has to be offline */ | |
48f24c4d | 6213 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6214 | |
48f24c4d IM |
6215 | /* |
6216 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6217 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6218 | */ |
6219 | spin_lock_irqsave(&rq->lock, flags); | |
6220 | ||
dd41f596 | 6221 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6222 | |
94bc9a7b DA |
6223 | update_rq_clock(rq); |
6224 | activate_task(rq, p, 0); | |
1da177e4 LT |
6225 | |
6226 | spin_unlock_irqrestore(&rq->lock, flags); | |
6227 | } | |
6228 | ||
48f24c4d IM |
6229 | /* |
6230 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6231 | * offline. |
6232 | */ | |
6233 | void idle_task_exit(void) | |
6234 | { | |
6235 | struct mm_struct *mm = current->active_mm; | |
6236 | ||
6237 | BUG_ON(cpu_online(smp_processor_id())); | |
6238 | ||
6239 | if (mm != &init_mm) | |
6240 | switch_mm(mm, &init_mm, current); | |
6241 | mmdrop(mm); | |
6242 | } | |
6243 | ||
054b9108 | 6244 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6245 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6246 | { |
70b97a7f | 6247 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6248 | |
6249 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6250 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6251 | |
6252 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6253 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6254 | |
48f24c4d | 6255 | get_task_struct(p); |
1da177e4 LT |
6256 | |
6257 | /* | |
6258 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6259 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6260 | * fine. |
6261 | */ | |
f7b4cddc | 6262 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6263 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6264 | spin_lock_irq(&rq->lock); |
1da177e4 | 6265 | |
48f24c4d | 6266 | put_task_struct(p); |
1da177e4 LT |
6267 | } |
6268 | ||
6269 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6270 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6271 | { | |
70b97a7f | 6272 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6273 | struct task_struct *next; |
48f24c4d | 6274 | |
dd41f596 IM |
6275 | for ( ; ; ) { |
6276 | if (!rq->nr_running) | |
6277 | break; | |
a8e504d2 | 6278 | update_rq_clock(rq); |
ff95f3df | 6279 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6280 | if (!next) |
6281 | break; | |
79c53799 | 6282 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6283 | migrate_dead(dead_cpu, next); |
e692ab53 | 6284 | |
1da177e4 LT |
6285 | } |
6286 | } | |
6287 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6288 | ||
e692ab53 NP |
6289 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6290 | ||
6291 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6292 | { |
6293 | .procname = "sched_domain", | |
c57baf1e | 6294 | .mode = 0555, |
e0361851 | 6295 | }, |
38605cae | 6296 | {0, }, |
e692ab53 NP |
6297 | }; |
6298 | ||
6299 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6300 | { |
c57baf1e | 6301 | .ctl_name = CTL_KERN, |
e0361851 | 6302 | .procname = "kernel", |
c57baf1e | 6303 | .mode = 0555, |
e0361851 AD |
6304 | .child = sd_ctl_dir, |
6305 | }, | |
38605cae | 6306 | {0, }, |
e692ab53 NP |
6307 | }; |
6308 | ||
6309 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6310 | { | |
6311 | struct ctl_table *entry = | |
5cf9f062 | 6312 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6313 | |
e692ab53 NP |
6314 | return entry; |
6315 | } | |
6316 | ||
6382bc90 MM |
6317 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6318 | { | |
cd790076 | 6319 | struct ctl_table *entry; |
6382bc90 | 6320 | |
cd790076 MM |
6321 | /* |
6322 | * In the intermediate directories, both the child directory and | |
6323 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6324 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6325 | * static strings and all have proc handlers. |
6326 | */ | |
6327 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6328 | if (entry->child) |
6329 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6330 | if (entry->proc_handler == NULL) |
6331 | kfree(entry->procname); | |
6332 | } | |
6382bc90 MM |
6333 | |
6334 | kfree(*tablep); | |
6335 | *tablep = NULL; | |
6336 | } | |
6337 | ||
e692ab53 | 6338 | static void |
e0361851 | 6339 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6340 | const char *procname, void *data, int maxlen, |
6341 | mode_t mode, proc_handler *proc_handler) | |
6342 | { | |
e692ab53 NP |
6343 | entry->procname = procname; |
6344 | entry->data = data; | |
6345 | entry->maxlen = maxlen; | |
6346 | entry->mode = mode; | |
6347 | entry->proc_handler = proc_handler; | |
6348 | } | |
6349 | ||
6350 | static struct ctl_table * | |
6351 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6352 | { | |
a5d8c348 | 6353 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6354 | |
ad1cdc1d MM |
6355 | if (table == NULL) |
6356 | return NULL; | |
6357 | ||
e0361851 | 6358 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6359 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6360 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6361 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6362 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6363 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6364 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6365 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6366 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6367 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6368 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6369 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6370 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6371 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6372 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6373 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6374 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6375 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6376 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6377 | &sd->cache_nice_tries, |
6378 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6379 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6380 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6381 | set_table_entry(&table[11], "name", sd->name, |
6382 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6383 | /* &table[12] is terminator */ | |
e692ab53 NP |
6384 | |
6385 | return table; | |
6386 | } | |
6387 | ||
9a4e7159 | 6388 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6389 | { |
6390 | struct ctl_table *entry, *table; | |
6391 | struct sched_domain *sd; | |
6392 | int domain_num = 0, i; | |
6393 | char buf[32]; | |
6394 | ||
6395 | for_each_domain(cpu, sd) | |
6396 | domain_num++; | |
6397 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6398 | if (table == NULL) |
6399 | return NULL; | |
e692ab53 NP |
6400 | |
6401 | i = 0; | |
6402 | for_each_domain(cpu, sd) { | |
6403 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6404 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6405 | entry->mode = 0555; |
e692ab53 NP |
6406 | entry->child = sd_alloc_ctl_domain_table(sd); |
6407 | entry++; | |
6408 | i++; | |
6409 | } | |
6410 | return table; | |
6411 | } | |
6412 | ||
6413 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6414 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6415 | { |
6416 | int i, cpu_num = num_online_cpus(); | |
6417 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6418 | char buf[32]; | |
6419 | ||
7378547f MM |
6420 | WARN_ON(sd_ctl_dir[0].child); |
6421 | sd_ctl_dir[0].child = entry; | |
6422 | ||
ad1cdc1d MM |
6423 | if (entry == NULL) |
6424 | return; | |
6425 | ||
97b6ea7b | 6426 | for_each_online_cpu(i) { |
e692ab53 | 6427 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6428 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6429 | entry->mode = 0555; |
e692ab53 | 6430 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6431 | entry++; |
e692ab53 | 6432 | } |
7378547f MM |
6433 | |
6434 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6435 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6436 | } | |
6382bc90 | 6437 | |
7378547f | 6438 | /* may be called multiple times per register */ |
6382bc90 MM |
6439 | static void unregister_sched_domain_sysctl(void) |
6440 | { | |
7378547f MM |
6441 | if (sd_sysctl_header) |
6442 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6443 | sd_sysctl_header = NULL; |
7378547f MM |
6444 | if (sd_ctl_dir[0].child) |
6445 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6446 | } |
e692ab53 | 6447 | #else |
6382bc90 MM |
6448 | static void register_sched_domain_sysctl(void) |
6449 | { | |
6450 | } | |
6451 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6452 | { |
6453 | } | |
6454 | #endif | |
6455 | ||
1f11eb6a GH |
6456 | static void set_rq_online(struct rq *rq) |
6457 | { | |
6458 | if (!rq->online) { | |
6459 | const struct sched_class *class; | |
6460 | ||
6461 | cpu_set(rq->cpu, rq->rd->online); | |
6462 | rq->online = 1; | |
6463 | ||
6464 | for_each_class(class) { | |
6465 | if (class->rq_online) | |
6466 | class->rq_online(rq); | |
6467 | } | |
6468 | } | |
6469 | } | |
6470 | ||
6471 | static void set_rq_offline(struct rq *rq) | |
6472 | { | |
6473 | if (rq->online) { | |
6474 | const struct sched_class *class; | |
6475 | ||
6476 | for_each_class(class) { | |
6477 | if (class->rq_offline) | |
6478 | class->rq_offline(rq); | |
6479 | } | |
6480 | ||
6481 | cpu_clear(rq->cpu, rq->rd->online); | |
6482 | rq->online = 0; | |
6483 | } | |
6484 | } | |
6485 | ||
1da177e4 LT |
6486 | /* |
6487 | * migration_call - callback that gets triggered when a CPU is added. | |
6488 | * Here we can start up the necessary migration thread for the new CPU. | |
6489 | */ | |
48f24c4d IM |
6490 | static int __cpuinit |
6491 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6492 | { |
1da177e4 | 6493 | struct task_struct *p; |
48f24c4d | 6494 | int cpu = (long)hcpu; |
1da177e4 | 6495 | unsigned long flags; |
70b97a7f | 6496 | struct rq *rq; |
1da177e4 LT |
6497 | |
6498 | switch (action) { | |
5be9361c | 6499 | |
1da177e4 | 6500 | case CPU_UP_PREPARE: |
8bb78442 | 6501 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6502 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6503 | if (IS_ERR(p)) |
6504 | return NOTIFY_BAD; | |
1da177e4 LT |
6505 | kthread_bind(p, cpu); |
6506 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6507 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6508 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6509 | task_rq_unlock(rq, &flags); |
6510 | cpu_rq(cpu)->migration_thread = p; | |
6511 | break; | |
48f24c4d | 6512 | |
1da177e4 | 6513 | case CPU_ONLINE: |
8bb78442 | 6514 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6515 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6516 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6517 | |
6518 | /* Update our root-domain */ | |
6519 | rq = cpu_rq(cpu); | |
6520 | spin_lock_irqsave(&rq->lock, flags); | |
6521 | if (rq->rd) { | |
6522 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6523 | |
6524 | set_rq_online(rq); | |
1f94ef59 GH |
6525 | } |
6526 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6527 | break; |
48f24c4d | 6528 | |
1da177e4 LT |
6529 | #ifdef CONFIG_HOTPLUG_CPU |
6530 | case CPU_UP_CANCELED: | |
8bb78442 | 6531 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6532 | if (!cpu_rq(cpu)->migration_thread) |
6533 | break; | |
41a2d6cf | 6534 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6535 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6536 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6537 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6538 | cpu_rq(cpu)->migration_thread = NULL; | |
6539 | break; | |
48f24c4d | 6540 | |
1da177e4 | 6541 | case CPU_DEAD: |
8bb78442 | 6542 | case CPU_DEAD_FROZEN: |
470fd646 | 6543 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6544 | migrate_live_tasks(cpu); |
6545 | rq = cpu_rq(cpu); | |
6546 | kthread_stop(rq->migration_thread); | |
6547 | rq->migration_thread = NULL; | |
6548 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6549 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6550 | update_rq_clock(rq); |
2e1cb74a | 6551 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6552 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6553 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6554 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6555 | migrate_dead_tasks(cpu); |
d2da272a | 6556 | spin_unlock_irq(&rq->lock); |
470fd646 | 6557 | cpuset_unlock(); |
1da177e4 LT |
6558 | migrate_nr_uninterruptible(rq); |
6559 | BUG_ON(rq->nr_running != 0); | |
6560 | ||
41a2d6cf IM |
6561 | /* |
6562 | * No need to migrate the tasks: it was best-effort if | |
6563 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6564 | * the requestors. | |
6565 | */ | |
1da177e4 LT |
6566 | spin_lock_irq(&rq->lock); |
6567 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6568 | struct migration_req *req; |
6569 | ||
1da177e4 | 6570 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6571 | struct migration_req, list); |
1da177e4 LT |
6572 | list_del_init(&req->list); |
6573 | complete(&req->done); | |
6574 | } | |
6575 | spin_unlock_irq(&rq->lock); | |
6576 | break; | |
57d885fe | 6577 | |
08f503b0 GH |
6578 | case CPU_DYING: |
6579 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6580 | /* Update our root-domain */ |
6581 | rq = cpu_rq(cpu); | |
6582 | spin_lock_irqsave(&rq->lock, flags); | |
6583 | if (rq->rd) { | |
6584 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6585 | set_rq_offline(rq); |
57d885fe GH |
6586 | } |
6587 | spin_unlock_irqrestore(&rq->lock, flags); | |
6588 | break; | |
1da177e4 LT |
6589 | #endif |
6590 | } | |
6591 | return NOTIFY_OK; | |
6592 | } | |
6593 | ||
6594 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6595 | * happens before everything else. | |
6596 | */ | |
26c2143b | 6597 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6598 | .notifier_call = migration_call, |
6599 | .priority = 10 | |
6600 | }; | |
6601 | ||
7babe8db | 6602 | static int __init migration_init(void) |
1da177e4 LT |
6603 | { |
6604 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6605 | int err; |
48f24c4d IM |
6606 | |
6607 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6608 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6609 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6610 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6611 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6612 | |
6613 | return err; | |
1da177e4 | 6614 | } |
7babe8db | 6615 | early_initcall(migration_init); |
1da177e4 LT |
6616 | #endif |
6617 | ||
6618 | #ifdef CONFIG_SMP | |
476f3534 | 6619 | |
3e9830dc | 6620 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6621 | |
099f98c8 GS |
6622 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6623 | { | |
6624 | switch (lvl) { | |
6625 | case SD_LV_NONE: | |
6626 | return "NONE"; | |
6627 | case SD_LV_SIBLING: | |
6628 | return "SIBLING"; | |
6629 | case SD_LV_MC: | |
6630 | return "MC"; | |
6631 | case SD_LV_CPU: | |
6632 | return "CPU"; | |
6633 | case SD_LV_NODE: | |
6634 | return "NODE"; | |
6635 | case SD_LV_ALLNODES: | |
6636 | return "ALLNODES"; | |
6637 | case SD_LV_MAX: | |
6638 | return "MAX"; | |
6639 | ||
6640 | } | |
6641 | return "MAX"; | |
6642 | } | |
6643 | ||
7c16ec58 MT |
6644 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6645 | cpumask_t *groupmask) | |
1da177e4 | 6646 | { |
4dcf6aff | 6647 | struct sched_group *group = sd->groups; |
434d53b0 | 6648 | char str[256]; |
1da177e4 | 6649 | |
434d53b0 | 6650 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6651 | cpus_clear(*groupmask); |
4dcf6aff IM |
6652 | |
6653 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6654 | ||
6655 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6656 | printk("does not load-balance\n"); | |
6657 | if (sd->parent) | |
6658 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6659 | " has parent"); | |
6660 | return -1; | |
41c7ce9a NP |
6661 | } |
6662 | ||
099f98c8 GS |
6663 | printk(KERN_CONT "span %s level %s\n", |
6664 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6665 | |
6666 | if (!cpu_isset(cpu, sd->span)) { | |
6667 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6668 | "CPU%d\n", cpu); | |
6669 | } | |
6670 | if (!cpu_isset(cpu, group->cpumask)) { | |
6671 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6672 | " CPU%d\n", cpu); | |
6673 | } | |
1da177e4 | 6674 | |
4dcf6aff | 6675 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6676 | do { |
4dcf6aff IM |
6677 | if (!group) { |
6678 | printk("\n"); | |
6679 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6680 | break; |
6681 | } | |
6682 | ||
4dcf6aff IM |
6683 | if (!group->__cpu_power) { |
6684 | printk(KERN_CONT "\n"); | |
6685 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6686 | "set\n"); | |
6687 | break; | |
6688 | } | |
1da177e4 | 6689 | |
4dcf6aff IM |
6690 | if (!cpus_weight(group->cpumask)) { |
6691 | printk(KERN_CONT "\n"); | |
6692 | printk(KERN_ERR "ERROR: empty group\n"); | |
6693 | break; | |
6694 | } | |
1da177e4 | 6695 | |
7c16ec58 | 6696 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6697 | printk(KERN_CONT "\n"); |
6698 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6699 | break; | |
6700 | } | |
1da177e4 | 6701 | |
7c16ec58 | 6702 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6703 | |
434d53b0 | 6704 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6705 | printk(KERN_CONT " %s", str); |
1da177e4 | 6706 | |
4dcf6aff IM |
6707 | group = group->next; |
6708 | } while (group != sd->groups); | |
6709 | printk(KERN_CONT "\n"); | |
1da177e4 | 6710 | |
7c16ec58 | 6711 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6712 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6713 | |
7c16ec58 | 6714 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6715 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6716 | "of domain->span\n"); | |
6717 | return 0; | |
6718 | } | |
1da177e4 | 6719 | |
4dcf6aff IM |
6720 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6721 | { | |
7c16ec58 | 6722 | cpumask_t *groupmask; |
4dcf6aff | 6723 | int level = 0; |
1da177e4 | 6724 | |
4dcf6aff IM |
6725 | if (!sd) { |
6726 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6727 | return; | |
6728 | } | |
1da177e4 | 6729 | |
4dcf6aff IM |
6730 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6731 | ||
7c16ec58 MT |
6732 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6733 | if (!groupmask) { | |
6734 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6735 | return; | |
6736 | } | |
6737 | ||
4dcf6aff | 6738 | for (;;) { |
7c16ec58 | 6739 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6740 | break; |
1da177e4 LT |
6741 | level++; |
6742 | sd = sd->parent; | |
33859f7f | 6743 | if (!sd) |
4dcf6aff IM |
6744 | break; |
6745 | } | |
7c16ec58 | 6746 | kfree(groupmask); |
1da177e4 | 6747 | } |
6d6bc0ad | 6748 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6749 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6750 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6751 | |
1a20ff27 | 6752 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6753 | { |
6754 | if (cpus_weight(sd->span) == 1) | |
6755 | return 1; | |
6756 | ||
6757 | /* Following flags need at least 2 groups */ | |
6758 | if (sd->flags & (SD_LOAD_BALANCE | | |
6759 | SD_BALANCE_NEWIDLE | | |
6760 | SD_BALANCE_FORK | | |
89c4710e SS |
6761 | SD_BALANCE_EXEC | |
6762 | SD_SHARE_CPUPOWER | | |
6763 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6764 | if (sd->groups != sd->groups->next) |
6765 | return 0; | |
6766 | } | |
6767 | ||
6768 | /* Following flags don't use groups */ | |
6769 | if (sd->flags & (SD_WAKE_IDLE | | |
6770 | SD_WAKE_AFFINE | | |
6771 | SD_WAKE_BALANCE)) | |
6772 | return 0; | |
6773 | ||
6774 | return 1; | |
6775 | } | |
6776 | ||
48f24c4d IM |
6777 | static int |
6778 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6779 | { |
6780 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6781 | ||
6782 | if (sd_degenerate(parent)) | |
6783 | return 1; | |
6784 | ||
6785 | if (!cpus_equal(sd->span, parent->span)) | |
6786 | return 0; | |
6787 | ||
6788 | /* Does parent contain flags not in child? */ | |
6789 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6790 | if (cflags & SD_WAKE_AFFINE) | |
6791 | pflags &= ~SD_WAKE_BALANCE; | |
6792 | /* Flags needing groups don't count if only 1 group in parent */ | |
6793 | if (parent->groups == parent->groups->next) { | |
6794 | pflags &= ~(SD_LOAD_BALANCE | | |
6795 | SD_BALANCE_NEWIDLE | | |
6796 | SD_BALANCE_FORK | | |
89c4710e SS |
6797 | SD_BALANCE_EXEC | |
6798 | SD_SHARE_CPUPOWER | | |
6799 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6800 | } |
6801 | if (~cflags & pflags) | |
6802 | return 0; | |
6803 | ||
6804 | return 1; | |
6805 | } | |
6806 | ||
57d885fe GH |
6807 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6808 | { | |
6809 | unsigned long flags; | |
57d885fe GH |
6810 | |
6811 | spin_lock_irqsave(&rq->lock, flags); | |
6812 | ||
6813 | if (rq->rd) { | |
6814 | struct root_domain *old_rd = rq->rd; | |
6815 | ||
1f11eb6a GH |
6816 | if (cpu_isset(rq->cpu, old_rd->online)) |
6817 | set_rq_offline(rq); | |
57d885fe | 6818 | |
dc938520 | 6819 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6820 | |
57d885fe GH |
6821 | if (atomic_dec_and_test(&old_rd->refcount)) |
6822 | kfree(old_rd); | |
6823 | } | |
6824 | ||
6825 | atomic_inc(&rd->refcount); | |
6826 | rq->rd = rd; | |
6827 | ||
dc938520 | 6828 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6829 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6830 | set_rq_online(rq); |
57d885fe GH |
6831 | |
6832 | spin_unlock_irqrestore(&rq->lock, flags); | |
6833 | } | |
6834 | ||
dc938520 | 6835 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6836 | { |
6837 | memset(rd, 0, sizeof(*rd)); | |
6838 | ||
dc938520 GH |
6839 | cpus_clear(rd->span); |
6840 | cpus_clear(rd->online); | |
6e0534f2 GH |
6841 | |
6842 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6843 | } |
6844 | ||
6845 | static void init_defrootdomain(void) | |
6846 | { | |
dc938520 | 6847 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6848 | atomic_set(&def_root_domain.refcount, 1); |
6849 | } | |
6850 | ||
dc938520 | 6851 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6852 | { |
6853 | struct root_domain *rd; | |
6854 | ||
6855 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6856 | if (!rd) | |
6857 | return NULL; | |
6858 | ||
dc938520 | 6859 | init_rootdomain(rd); |
57d885fe GH |
6860 | |
6861 | return rd; | |
6862 | } | |
6863 | ||
1da177e4 | 6864 | /* |
0eab9146 | 6865 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6866 | * hold the hotplug lock. |
6867 | */ | |
0eab9146 IM |
6868 | static void |
6869 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6870 | { |
70b97a7f | 6871 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6872 | struct sched_domain *tmp; |
6873 | ||
6874 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6875 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6876 | struct sched_domain *parent = tmp->parent; | |
6877 | if (!parent) | |
6878 | break; | |
1a848870 | 6879 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6880 | tmp->parent = parent->parent; |
1a848870 SS |
6881 | if (parent->parent) |
6882 | parent->parent->child = tmp; | |
6883 | } | |
245af2c7 SS |
6884 | } |
6885 | ||
1a848870 | 6886 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6887 | sd = sd->parent; |
1a848870 SS |
6888 | if (sd) |
6889 | sd->child = NULL; | |
6890 | } | |
1da177e4 LT |
6891 | |
6892 | sched_domain_debug(sd, cpu); | |
6893 | ||
57d885fe | 6894 | rq_attach_root(rq, rd); |
674311d5 | 6895 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6896 | } |
6897 | ||
6898 | /* cpus with isolated domains */ | |
67af63a6 | 6899 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6900 | |
6901 | /* Setup the mask of cpus configured for isolated domains */ | |
6902 | static int __init isolated_cpu_setup(char *str) | |
6903 | { | |
13b40c1e MT |
6904 | static int __initdata ints[NR_CPUS]; |
6905 | int i; | |
1da177e4 LT |
6906 | |
6907 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6908 | cpus_clear(cpu_isolated_map); | |
6909 | for (i = 1; i <= ints[0]; i++) | |
6910 | if (ints[i] < NR_CPUS) | |
6911 | cpu_set(ints[i], cpu_isolated_map); | |
6912 | return 1; | |
6913 | } | |
6914 | ||
8927f494 | 6915 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6916 | |
6917 | /* | |
6711cab4 SS |
6918 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6919 | * to a function which identifies what group(along with sched group) a CPU | |
6920 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6921 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6922 | * |
6923 | * init_sched_build_groups will build a circular linked list of the groups | |
6924 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6925 | * and ->cpu_power to 0. | |
6926 | */ | |
a616058b | 6927 | static void |
7c16ec58 | 6928 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6929 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6930 | struct sched_group **sg, |
6931 | cpumask_t *tmpmask), | |
6932 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6933 | { |
6934 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6935 | int i; |
6936 | ||
7c16ec58 MT |
6937 | cpus_clear(*covered); |
6938 | ||
363ab6f1 | 6939 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6940 | struct sched_group *sg; |
7c16ec58 | 6941 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6942 | int j; |
6943 | ||
7c16ec58 | 6944 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6945 | continue; |
6946 | ||
7c16ec58 | 6947 | cpus_clear(sg->cpumask); |
5517d86b | 6948 | sg->__cpu_power = 0; |
1da177e4 | 6949 | |
363ab6f1 | 6950 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6951 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6952 | continue; |
6953 | ||
7c16ec58 | 6954 | cpu_set(j, *covered); |
1da177e4 LT |
6955 | cpu_set(j, sg->cpumask); |
6956 | } | |
6957 | if (!first) | |
6958 | first = sg; | |
6959 | if (last) | |
6960 | last->next = sg; | |
6961 | last = sg; | |
6962 | } | |
6963 | last->next = first; | |
6964 | } | |
6965 | ||
9c1cfda2 | 6966 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6967 | |
9c1cfda2 | 6968 | #ifdef CONFIG_NUMA |
198e2f18 | 6969 | |
9c1cfda2 JH |
6970 | /** |
6971 | * find_next_best_node - find the next node to include in a sched_domain | |
6972 | * @node: node whose sched_domain we're building | |
6973 | * @used_nodes: nodes already in the sched_domain | |
6974 | * | |
41a2d6cf | 6975 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6976 | * finds the closest node not already in the @used_nodes map. |
6977 | * | |
6978 | * Should use nodemask_t. | |
6979 | */ | |
c5f59f08 | 6980 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6981 | { |
6982 | int i, n, val, min_val, best_node = 0; | |
6983 | ||
6984 | min_val = INT_MAX; | |
6985 | ||
076ac2af | 6986 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6987 | /* Start at @node */ |
076ac2af | 6988 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6989 | |
6990 | if (!nr_cpus_node(n)) | |
6991 | continue; | |
6992 | ||
6993 | /* Skip already used nodes */ | |
c5f59f08 | 6994 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6995 | continue; |
6996 | ||
6997 | /* Simple min distance search */ | |
6998 | val = node_distance(node, n); | |
6999 | ||
7000 | if (val < min_val) { | |
7001 | min_val = val; | |
7002 | best_node = n; | |
7003 | } | |
7004 | } | |
7005 | ||
c5f59f08 | 7006 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7007 | return best_node; |
7008 | } | |
7009 | ||
7010 | /** | |
7011 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7012 | * @node: node whose cpumask we're constructing | |
73486722 | 7013 | * @span: resulting cpumask |
9c1cfda2 | 7014 | * |
41a2d6cf | 7015 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7016 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7017 | * out optimally. | |
7018 | */ | |
4bdbaad3 | 7019 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 7020 | { |
c5f59f08 | 7021 | nodemask_t used_nodes; |
c5f59f08 | 7022 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7023 | int i; |
9c1cfda2 | 7024 | |
4bdbaad3 | 7025 | cpus_clear(*span); |
c5f59f08 | 7026 | nodes_clear(used_nodes); |
9c1cfda2 | 7027 | |
4bdbaad3 | 7028 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7029 | node_set(node, used_nodes); |
9c1cfda2 JH |
7030 | |
7031 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7032 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7033 | |
c5f59f08 | 7034 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7035 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7036 | } |
9c1cfda2 | 7037 | } |
6d6bc0ad | 7038 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7039 | |
5c45bf27 | 7040 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7041 | |
9c1cfda2 | 7042 | /* |
48f24c4d | 7043 | * SMT sched-domains: |
9c1cfda2 | 7044 | */ |
1da177e4 LT |
7045 | #ifdef CONFIG_SCHED_SMT |
7046 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7047 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7048 | |
41a2d6cf | 7049 | static int |
7c16ec58 MT |
7050 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7051 | cpumask_t *unused) | |
1da177e4 | 7052 | { |
6711cab4 SS |
7053 | if (sg) |
7054 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7055 | return cpu; |
7056 | } | |
6d6bc0ad | 7057 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7058 | |
48f24c4d IM |
7059 | /* |
7060 | * multi-core sched-domains: | |
7061 | */ | |
1e9f28fa SS |
7062 | #ifdef CONFIG_SCHED_MC |
7063 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7064 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7065 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7066 | |
7067 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7068 | static int |
7c16ec58 MT |
7069 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7070 | cpumask_t *mask) | |
1e9f28fa | 7071 | { |
6711cab4 | 7072 | int group; |
7c16ec58 MT |
7073 | |
7074 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7075 | cpus_and(*mask, *mask, *cpu_map); | |
7076 | group = first_cpu(*mask); | |
6711cab4 SS |
7077 | if (sg) |
7078 | *sg = &per_cpu(sched_group_core, group); | |
7079 | return group; | |
1e9f28fa SS |
7080 | } |
7081 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7082 | static int |
7c16ec58 MT |
7083 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7084 | cpumask_t *unused) | |
1e9f28fa | 7085 | { |
6711cab4 SS |
7086 | if (sg) |
7087 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7088 | return cpu; |
7089 | } | |
7090 | #endif | |
7091 | ||
1da177e4 | 7092 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7093 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7094 | |
41a2d6cf | 7095 | static int |
7c16ec58 MT |
7096 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7097 | cpumask_t *mask) | |
1da177e4 | 7098 | { |
6711cab4 | 7099 | int group; |
48f24c4d | 7100 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7101 | *mask = cpu_coregroup_map(cpu); |
7102 | cpus_and(*mask, *mask, *cpu_map); | |
7103 | group = first_cpu(*mask); | |
1e9f28fa | 7104 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7105 | *mask = per_cpu(cpu_sibling_map, cpu); |
7106 | cpus_and(*mask, *mask, *cpu_map); | |
7107 | group = first_cpu(*mask); | |
1da177e4 | 7108 | #else |
6711cab4 | 7109 | group = cpu; |
1da177e4 | 7110 | #endif |
6711cab4 SS |
7111 | if (sg) |
7112 | *sg = &per_cpu(sched_group_phys, group); | |
7113 | return group; | |
1da177e4 LT |
7114 | } |
7115 | ||
7116 | #ifdef CONFIG_NUMA | |
1da177e4 | 7117 | /* |
9c1cfda2 JH |
7118 | * The init_sched_build_groups can't handle what we want to do with node |
7119 | * groups, so roll our own. Now each node has its own list of groups which | |
7120 | * gets dynamically allocated. | |
1da177e4 | 7121 | */ |
9c1cfda2 | 7122 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7123 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7124 | |
9c1cfda2 | 7125 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7126 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7127 | |
6711cab4 | 7128 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7129 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7130 | { |
6711cab4 SS |
7131 | int group; |
7132 | ||
7c16ec58 MT |
7133 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7134 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7135 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7136 | |
7137 | if (sg) | |
7138 | *sg = &per_cpu(sched_group_allnodes, group); | |
7139 | return group; | |
1da177e4 | 7140 | } |
6711cab4 | 7141 | |
08069033 SS |
7142 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7143 | { | |
7144 | struct sched_group *sg = group_head; | |
7145 | int j; | |
7146 | ||
7147 | if (!sg) | |
7148 | return; | |
3a5c359a | 7149 | do { |
363ab6f1 | 7150 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7151 | struct sched_domain *sd; |
08069033 | 7152 | |
3a5c359a AK |
7153 | sd = &per_cpu(phys_domains, j); |
7154 | if (j != first_cpu(sd->groups->cpumask)) { | |
7155 | /* | |
7156 | * Only add "power" once for each | |
7157 | * physical package. | |
7158 | */ | |
7159 | continue; | |
7160 | } | |
08069033 | 7161 | |
3a5c359a AK |
7162 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7163 | } | |
7164 | sg = sg->next; | |
7165 | } while (sg != group_head); | |
08069033 | 7166 | } |
6d6bc0ad | 7167 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7168 | |
a616058b | 7169 | #ifdef CONFIG_NUMA |
51888ca2 | 7170 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7171 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7172 | { |
a616058b | 7173 | int cpu, i; |
51888ca2 | 7174 | |
363ab6f1 | 7175 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7176 | struct sched_group **sched_group_nodes |
7177 | = sched_group_nodes_bycpu[cpu]; | |
7178 | ||
51888ca2 SV |
7179 | if (!sched_group_nodes) |
7180 | continue; | |
7181 | ||
076ac2af | 7182 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7183 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7184 | ||
7c16ec58 MT |
7185 | *nodemask = node_to_cpumask(i); |
7186 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7187 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7188 | continue; |
7189 | ||
7190 | if (sg == NULL) | |
7191 | continue; | |
7192 | sg = sg->next; | |
7193 | next_sg: | |
7194 | oldsg = sg; | |
7195 | sg = sg->next; | |
7196 | kfree(oldsg); | |
7197 | if (oldsg != sched_group_nodes[i]) | |
7198 | goto next_sg; | |
7199 | } | |
7200 | kfree(sched_group_nodes); | |
7201 | sched_group_nodes_bycpu[cpu] = NULL; | |
7202 | } | |
51888ca2 | 7203 | } |
6d6bc0ad | 7204 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7205 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7206 | { |
7207 | } | |
6d6bc0ad | 7208 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7209 | |
89c4710e SS |
7210 | /* |
7211 | * Initialize sched groups cpu_power. | |
7212 | * | |
7213 | * cpu_power indicates the capacity of sched group, which is used while | |
7214 | * distributing the load between different sched groups in a sched domain. | |
7215 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7216 | * there are asymmetries in the topology. If there are asymmetries, group | |
7217 | * having more cpu_power will pickup more load compared to the group having | |
7218 | * less cpu_power. | |
7219 | * | |
7220 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7221 | * the maximum number of tasks a group can handle in the presence of other idle | |
7222 | * or lightly loaded groups in the same sched domain. | |
7223 | */ | |
7224 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7225 | { | |
7226 | struct sched_domain *child; | |
7227 | struct sched_group *group; | |
7228 | ||
7229 | WARN_ON(!sd || !sd->groups); | |
7230 | ||
7231 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7232 | return; | |
7233 | ||
7234 | child = sd->child; | |
7235 | ||
5517d86b ED |
7236 | sd->groups->__cpu_power = 0; |
7237 | ||
89c4710e SS |
7238 | /* |
7239 | * For perf policy, if the groups in child domain share resources | |
7240 | * (for example cores sharing some portions of the cache hierarchy | |
7241 | * or SMT), then set this domain groups cpu_power such that each group | |
7242 | * can handle only one task, when there are other idle groups in the | |
7243 | * same sched domain. | |
7244 | */ | |
7245 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7246 | (child->flags & | |
7247 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7248 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7249 | return; |
7250 | } | |
7251 | ||
89c4710e SS |
7252 | /* |
7253 | * add cpu_power of each child group to this groups cpu_power | |
7254 | */ | |
7255 | group = child->groups; | |
7256 | do { | |
5517d86b | 7257 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7258 | group = group->next; |
7259 | } while (group != child->groups); | |
7260 | } | |
7261 | ||
7c16ec58 MT |
7262 | /* |
7263 | * Initializers for schedule domains | |
7264 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7265 | */ | |
7266 | ||
a5d8c348 IM |
7267 | #ifdef CONFIG_SCHED_DEBUG |
7268 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7269 | #else | |
7270 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7271 | #endif | |
7272 | ||
7c16ec58 | 7273 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7274 | |
7c16ec58 MT |
7275 | #define SD_INIT_FUNC(type) \ |
7276 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7277 | { \ | |
7278 | memset(sd, 0, sizeof(*sd)); \ | |
7279 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7280 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7281 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7282 | } |
7283 | ||
7284 | SD_INIT_FUNC(CPU) | |
7285 | #ifdef CONFIG_NUMA | |
7286 | SD_INIT_FUNC(ALLNODES) | |
7287 | SD_INIT_FUNC(NODE) | |
7288 | #endif | |
7289 | #ifdef CONFIG_SCHED_SMT | |
7290 | SD_INIT_FUNC(SIBLING) | |
7291 | #endif | |
7292 | #ifdef CONFIG_SCHED_MC | |
7293 | SD_INIT_FUNC(MC) | |
7294 | #endif | |
7295 | ||
7296 | /* | |
7297 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7298 | * space as the usage in build_sched_domains() dictates. Used only | |
7299 | * if the amount of space is significant. | |
7300 | */ | |
7301 | struct allmasks { | |
7302 | cpumask_t tmpmask; /* make this one first */ | |
7303 | union { | |
7304 | cpumask_t nodemask; | |
7305 | cpumask_t this_sibling_map; | |
7306 | cpumask_t this_core_map; | |
7307 | }; | |
7308 | cpumask_t send_covered; | |
7309 | ||
7310 | #ifdef CONFIG_NUMA | |
7311 | cpumask_t domainspan; | |
7312 | cpumask_t covered; | |
7313 | cpumask_t notcovered; | |
7314 | #endif | |
7315 | }; | |
7316 | ||
7317 | #if NR_CPUS > 128 | |
7318 | #define SCHED_CPUMASK_ALLOC 1 | |
7319 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7320 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7321 | #else | |
7322 | #define SCHED_CPUMASK_ALLOC 0 | |
7323 | #define SCHED_CPUMASK_FREE(v) | |
7324 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7325 | #endif | |
7326 | ||
7327 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7328 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7329 | ||
1d3504fc HS |
7330 | static int default_relax_domain_level = -1; |
7331 | ||
7332 | static int __init setup_relax_domain_level(char *str) | |
7333 | { | |
30e0e178 LZ |
7334 | unsigned long val; |
7335 | ||
7336 | val = simple_strtoul(str, NULL, 0); | |
7337 | if (val < SD_LV_MAX) | |
7338 | default_relax_domain_level = val; | |
7339 | ||
1d3504fc HS |
7340 | return 1; |
7341 | } | |
7342 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7343 | ||
7344 | static void set_domain_attribute(struct sched_domain *sd, | |
7345 | struct sched_domain_attr *attr) | |
7346 | { | |
7347 | int request; | |
7348 | ||
7349 | if (!attr || attr->relax_domain_level < 0) { | |
7350 | if (default_relax_domain_level < 0) | |
7351 | return; | |
7352 | else | |
7353 | request = default_relax_domain_level; | |
7354 | } else | |
7355 | request = attr->relax_domain_level; | |
7356 | if (request < sd->level) { | |
7357 | /* turn off idle balance on this domain */ | |
7358 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7359 | } else { | |
7360 | /* turn on idle balance on this domain */ | |
7361 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7362 | } | |
7363 | } | |
7364 | ||
1da177e4 | 7365 | /* |
1a20ff27 DG |
7366 | * Build sched domains for a given set of cpus and attach the sched domains |
7367 | * to the individual cpus | |
1da177e4 | 7368 | */ |
1d3504fc HS |
7369 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7370 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7371 | { |
7372 | int i; | |
57d885fe | 7373 | struct root_domain *rd; |
7c16ec58 MT |
7374 | SCHED_CPUMASK_DECLARE(allmasks); |
7375 | cpumask_t *tmpmask; | |
d1b55138 JH |
7376 | #ifdef CONFIG_NUMA |
7377 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7378 | int sd_allnodes = 0; |
d1b55138 JH |
7379 | |
7380 | /* | |
7381 | * Allocate the per-node list of sched groups | |
7382 | */ | |
076ac2af | 7383 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7384 | GFP_KERNEL); |
d1b55138 JH |
7385 | if (!sched_group_nodes) { |
7386 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7387 | return -ENOMEM; |
d1b55138 | 7388 | } |
d1b55138 | 7389 | #endif |
1da177e4 | 7390 | |
dc938520 | 7391 | rd = alloc_rootdomain(); |
57d885fe GH |
7392 | if (!rd) { |
7393 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7394 | #ifdef CONFIG_NUMA |
7395 | kfree(sched_group_nodes); | |
7396 | #endif | |
57d885fe GH |
7397 | return -ENOMEM; |
7398 | } | |
7399 | ||
7c16ec58 MT |
7400 | #if SCHED_CPUMASK_ALLOC |
7401 | /* get space for all scratch cpumask variables */ | |
7402 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7403 | if (!allmasks) { | |
7404 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7405 | kfree(rd); | |
7406 | #ifdef CONFIG_NUMA | |
7407 | kfree(sched_group_nodes); | |
7408 | #endif | |
7409 | return -ENOMEM; | |
7410 | } | |
7411 | #endif | |
7412 | tmpmask = (cpumask_t *)allmasks; | |
7413 | ||
7414 | ||
7415 | #ifdef CONFIG_NUMA | |
7416 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7417 | #endif | |
7418 | ||
1da177e4 | 7419 | /* |
1a20ff27 | 7420 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7421 | */ |
363ab6f1 | 7422 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7423 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7424 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7425 | |
7c16ec58 MT |
7426 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7427 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7428 | |
7429 | #ifdef CONFIG_NUMA | |
dd41f596 | 7430 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7431 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7432 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7433 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7434 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7435 | sd->span = *cpu_map; |
7c16ec58 | 7436 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7437 | p = sd; |
6711cab4 | 7438 | sd_allnodes = 1; |
9c1cfda2 JH |
7439 | } else |
7440 | p = NULL; | |
7441 | ||
1da177e4 | 7442 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7443 | SD_INIT(sd, NODE); |
1d3504fc | 7444 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7445 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7446 | sd->parent = p; |
1a848870 SS |
7447 | if (p) |
7448 | p->child = sd; | |
9c1cfda2 | 7449 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7450 | #endif |
7451 | ||
7452 | p = sd; | |
7453 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7454 | SD_INIT(sd, CPU); |
1d3504fc | 7455 | set_domain_attribute(sd, attr); |
7c16ec58 | 7456 | sd->span = *nodemask; |
1da177e4 | 7457 | sd->parent = p; |
1a848870 SS |
7458 | if (p) |
7459 | p->child = sd; | |
7c16ec58 | 7460 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7461 | |
1e9f28fa SS |
7462 | #ifdef CONFIG_SCHED_MC |
7463 | p = sd; | |
7464 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7465 | SD_INIT(sd, MC); |
1d3504fc | 7466 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7467 | sd->span = cpu_coregroup_map(i); |
7468 | cpus_and(sd->span, sd->span, *cpu_map); | |
7469 | sd->parent = p; | |
1a848870 | 7470 | p->child = sd; |
7c16ec58 | 7471 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7472 | #endif |
7473 | ||
1da177e4 LT |
7474 | #ifdef CONFIG_SCHED_SMT |
7475 | p = sd; | |
7476 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7477 | SD_INIT(sd, SIBLING); |
1d3504fc | 7478 | set_domain_attribute(sd, attr); |
d5a7430d | 7479 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7480 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7481 | sd->parent = p; |
1a848870 | 7482 | p->child = sd; |
7c16ec58 | 7483 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7484 | #endif |
7485 | } | |
7486 | ||
7487 | #ifdef CONFIG_SCHED_SMT | |
7488 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7489 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7490 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7491 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7492 | ||
7493 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7494 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7495 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7496 | continue; |
7497 | ||
dd41f596 | 7498 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7499 | &cpu_to_cpu_group, |
7500 | send_covered, tmpmask); | |
1da177e4 LT |
7501 | } |
7502 | #endif | |
7503 | ||
1e9f28fa SS |
7504 | #ifdef CONFIG_SCHED_MC |
7505 | /* Set up multi-core groups */ | |
363ab6f1 | 7506 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7507 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7508 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7509 | ||
7510 | *this_core_map = cpu_coregroup_map(i); | |
7511 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7512 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7513 | continue; |
7c16ec58 | 7514 | |
dd41f596 | 7515 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7516 | &cpu_to_core_group, |
7517 | send_covered, tmpmask); | |
1e9f28fa SS |
7518 | } |
7519 | #endif | |
7520 | ||
1da177e4 | 7521 | /* Set up physical groups */ |
076ac2af | 7522 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7523 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7524 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7525 | |
7c16ec58 MT |
7526 | *nodemask = node_to_cpumask(i); |
7527 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7528 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7529 | continue; |
7530 | ||
7c16ec58 MT |
7531 | init_sched_build_groups(nodemask, cpu_map, |
7532 | &cpu_to_phys_group, | |
7533 | send_covered, tmpmask); | |
1da177e4 LT |
7534 | } |
7535 | ||
7536 | #ifdef CONFIG_NUMA | |
7537 | /* Set up node groups */ | |
7c16ec58 MT |
7538 | if (sd_allnodes) { |
7539 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7540 | ||
7541 | init_sched_build_groups(cpu_map, cpu_map, | |
7542 | &cpu_to_allnodes_group, | |
7543 | send_covered, tmpmask); | |
7544 | } | |
9c1cfda2 | 7545 | |
076ac2af | 7546 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7547 | /* Set up node groups */ |
7548 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7549 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7550 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7551 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7552 | int j; |
7553 | ||
7c16ec58 MT |
7554 | *nodemask = node_to_cpumask(i); |
7555 | cpus_clear(*covered); | |
7556 | ||
7557 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7558 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7559 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7560 | continue; |
d1b55138 | 7561 | } |
9c1cfda2 | 7562 | |
4bdbaad3 | 7563 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7564 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7565 | |
15f0b676 | 7566 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7567 | if (!sg) { |
7568 | printk(KERN_WARNING "Can not alloc domain group for " | |
7569 | "node %d\n", i); | |
7570 | goto error; | |
7571 | } | |
9c1cfda2 | 7572 | sched_group_nodes[i] = sg; |
363ab6f1 | 7573 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7574 | struct sched_domain *sd; |
9761eea8 | 7575 | |
9c1cfda2 JH |
7576 | sd = &per_cpu(node_domains, j); |
7577 | sd->groups = sg; | |
9c1cfda2 | 7578 | } |
5517d86b | 7579 | sg->__cpu_power = 0; |
7c16ec58 | 7580 | sg->cpumask = *nodemask; |
51888ca2 | 7581 | sg->next = sg; |
7c16ec58 | 7582 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7583 | prev = sg; |
7584 | ||
076ac2af | 7585 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7586 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7587 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7588 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7589 | |
7c16ec58 MT |
7590 | cpus_complement(*notcovered, *covered); |
7591 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7592 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7593 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7594 | break; |
7595 | ||
7c16ec58 MT |
7596 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7597 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7598 | continue; |
7599 | ||
15f0b676 SV |
7600 | sg = kmalloc_node(sizeof(struct sched_group), |
7601 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7602 | if (!sg) { |
7603 | printk(KERN_WARNING | |
7604 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7605 | goto error; |
9c1cfda2 | 7606 | } |
5517d86b | 7607 | sg->__cpu_power = 0; |
7c16ec58 | 7608 | sg->cpumask = *tmpmask; |
51888ca2 | 7609 | sg->next = prev->next; |
7c16ec58 | 7610 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7611 | prev->next = sg; |
7612 | prev = sg; | |
7613 | } | |
9c1cfda2 | 7614 | } |
1da177e4 LT |
7615 | #endif |
7616 | ||
7617 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7618 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7619 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7620 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7621 | ||
89c4710e | 7622 | init_sched_groups_power(i, sd); |
5c45bf27 | 7623 | } |
1da177e4 | 7624 | #endif |
1e9f28fa | 7625 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7626 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7627 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7628 | ||
89c4710e | 7629 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7630 | } |
7631 | #endif | |
1e9f28fa | 7632 | |
363ab6f1 | 7633 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7634 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7635 | ||
89c4710e | 7636 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7637 | } |
7638 | ||
9c1cfda2 | 7639 | #ifdef CONFIG_NUMA |
076ac2af | 7640 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7641 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7642 | |
6711cab4 SS |
7643 | if (sd_allnodes) { |
7644 | struct sched_group *sg; | |
f712c0c7 | 7645 | |
7c16ec58 MT |
7646 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7647 | tmpmask); | |
f712c0c7 SS |
7648 | init_numa_sched_groups_power(sg); |
7649 | } | |
9c1cfda2 JH |
7650 | #endif |
7651 | ||
1da177e4 | 7652 | /* Attach the domains */ |
363ab6f1 | 7653 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7654 | struct sched_domain *sd; |
7655 | #ifdef CONFIG_SCHED_SMT | |
7656 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7657 | #elif defined(CONFIG_SCHED_MC) |
7658 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7659 | #else |
7660 | sd = &per_cpu(phys_domains, i); | |
7661 | #endif | |
57d885fe | 7662 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7663 | } |
51888ca2 | 7664 | |
7c16ec58 | 7665 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7666 | return 0; |
7667 | ||
a616058b | 7668 | #ifdef CONFIG_NUMA |
51888ca2 | 7669 | error: |
7c16ec58 MT |
7670 | free_sched_groups(cpu_map, tmpmask); |
7671 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7672 | return -ENOMEM; |
a616058b | 7673 | #endif |
1da177e4 | 7674 | } |
029190c5 | 7675 | |
1d3504fc HS |
7676 | static int build_sched_domains(const cpumask_t *cpu_map) |
7677 | { | |
7678 | return __build_sched_domains(cpu_map, NULL); | |
7679 | } | |
7680 | ||
029190c5 PJ |
7681 | static cpumask_t *doms_cur; /* current sched domains */ |
7682 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7683 | static struct sched_domain_attr *dattr_cur; |
7684 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7685 | |
7686 | /* | |
7687 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7688 | * cpumask_t) fails, then fallback to a single sched domain, | |
7689 | * as determined by the single cpumask_t fallback_doms. | |
7690 | */ | |
7691 | static cpumask_t fallback_doms; | |
7692 | ||
22e52b07 HC |
7693 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7694 | { | |
7695 | } | |
7696 | ||
1a20ff27 | 7697 | /* |
41a2d6cf | 7698 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7699 | * For now this just excludes isolated cpus, but could be used to |
7700 | * exclude other special cases in the future. | |
1a20ff27 | 7701 | */ |
51888ca2 | 7702 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7703 | { |
7378547f MM |
7704 | int err; |
7705 | ||
22e52b07 | 7706 | arch_update_cpu_topology(); |
029190c5 PJ |
7707 | ndoms_cur = 1; |
7708 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7709 | if (!doms_cur) | |
7710 | doms_cur = &fallback_doms; | |
7711 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7712 | dattr_cur = NULL; |
7378547f | 7713 | err = build_sched_domains(doms_cur); |
6382bc90 | 7714 | register_sched_domain_sysctl(); |
7378547f MM |
7715 | |
7716 | return err; | |
1a20ff27 DG |
7717 | } |
7718 | ||
7c16ec58 MT |
7719 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7720 | cpumask_t *tmpmask) | |
1da177e4 | 7721 | { |
7c16ec58 | 7722 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7723 | } |
1da177e4 | 7724 | |
1a20ff27 DG |
7725 | /* |
7726 | * Detach sched domains from a group of cpus specified in cpu_map | |
7727 | * These cpus will now be attached to the NULL domain | |
7728 | */ | |
858119e1 | 7729 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7730 | { |
7c16ec58 | 7731 | cpumask_t tmpmask; |
1a20ff27 DG |
7732 | int i; |
7733 | ||
6382bc90 MM |
7734 | unregister_sched_domain_sysctl(); |
7735 | ||
363ab6f1 | 7736 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7737 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7738 | synchronize_sched(); |
7c16ec58 | 7739 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7740 | } |
7741 | ||
1d3504fc HS |
7742 | /* handle null as "default" */ |
7743 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7744 | struct sched_domain_attr *new, int idx_new) | |
7745 | { | |
7746 | struct sched_domain_attr tmp; | |
7747 | ||
7748 | /* fast path */ | |
7749 | if (!new && !cur) | |
7750 | return 1; | |
7751 | ||
7752 | tmp = SD_ATTR_INIT; | |
7753 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7754 | new ? (new + idx_new) : &tmp, | |
7755 | sizeof(struct sched_domain_attr)); | |
7756 | } | |
7757 | ||
029190c5 PJ |
7758 | /* |
7759 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7760 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7761 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7762 | * It destroys each deleted domain and builds each new domain. | |
7763 | * | |
7764 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7765 | * The masks don't intersect (don't overlap.) We should setup one |
7766 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7767 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7768 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7769 | * it as it is. | |
7770 | * | |
41a2d6cf IM |
7771 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7772 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7773 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7774 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7775 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 | 7776 | * |
dfb512ec MK |
7777 | * If doms_new==NULL it will be replaced with cpu_online_map. |
7778 | * ndoms_new==0 is a special case for destroying existing domains. | |
7779 | * It will not create the default domain. | |
7780 | * | |
029190c5 PJ |
7781 | * Call with hotplug lock held |
7782 | */ | |
1d3504fc HS |
7783 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7784 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7785 | { |
dfb512ec | 7786 | int i, j, n; |
029190c5 | 7787 | |
712555ee | 7788 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7789 | |
7378547f MM |
7790 | /* always unregister in case we don't destroy any domains */ |
7791 | unregister_sched_domain_sysctl(); | |
7792 | ||
dfb512ec | 7793 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7794 | |
7795 | /* Destroy deleted domains */ | |
7796 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7797 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7798 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7799 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7800 | goto match1; |
7801 | } | |
7802 | /* no match - a current sched domain not in new doms_new[] */ | |
7803 | detach_destroy_domains(doms_cur + i); | |
7804 | match1: | |
7805 | ; | |
7806 | } | |
7807 | ||
e761b772 MK |
7808 | if (doms_new == NULL) { |
7809 | ndoms_cur = 0; | |
e761b772 MK |
7810 | doms_new = &fallback_doms; |
7811 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7812 | dattr_new = NULL; | |
7813 | } | |
7814 | ||
029190c5 PJ |
7815 | /* Build new domains */ |
7816 | for (i = 0; i < ndoms_new; i++) { | |
7817 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7818 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7819 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7820 | goto match2; |
7821 | } | |
7822 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7823 | __build_sched_domains(doms_new + i, |
7824 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7825 | match2: |
7826 | ; | |
7827 | } | |
7828 | ||
7829 | /* Remember the new sched domains */ | |
7830 | if (doms_cur != &fallback_doms) | |
7831 | kfree(doms_cur); | |
1d3504fc | 7832 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7833 | doms_cur = doms_new; |
1d3504fc | 7834 | dattr_cur = dattr_new; |
029190c5 | 7835 | ndoms_cur = ndoms_new; |
7378547f MM |
7836 | |
7837 | register_sched_domain_sysctl(); | |
a1835615 | 7838 | |
712555ee | 7839 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7840 | } |
7841 | ||
5c45bf27 | 7842 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7843 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7844 | { |
95402b38 | 7845 | get_online_cpus(); |
dfb512ec MK |
7846 | |
7847 | /* Destroy domains first to force the rebuild */ | |
7848 | partition_sched_domains(0, NULL, NULL); | |
7849 | ||
e761b772 | 7850 | rebuild_sched_domains(); |
95402b38 | 7851 | put_online_cpus(); |
dfb512ec | 7852 | |
e761b772 | 7853 | return 0; |
5c45bf27 SS |
7854 | } |
7855 | ||
7856 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7857 | { | |
7858 | int ret; | |
7859 | ||
7860 | if (buf[0] != '0' && buf[0] != '1') | |
7861 | return -EINVAL; | |
7862 | ||
7863 | if (smt) | |
7864 | sched_smt_power_savings = (buf[0] == '1'); | |
7865 | else | |
7866 | sched_mc_power_savings = (buf[0] == '1'); | |
7867 | ||
7868 | ret = arch_reinit_sched_domains(); | |
7869 | ||
7870 | return ret ? ret : count; | |
7871 | } | |
7872 | ||
5c45bf27 | 7873 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7874 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7875 | char *page) | |
5c45bf27 SS |
7876 | { |
7877 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7878 | } | |
f718cd4a | 7879 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7880 | const char *buf, size_t count) |
5c45bf27 SS |
7881 | { |
7882 | return sched_power_savings_store(buf, count, 0); | |
7883 | } | |
f718cd4a AK |
7884 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7885 | sched_mc_power_savings_show, | |
7886 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7887 | #endif |
7888 | ||
7889 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7890 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7891 | char *page) | |
5c45bf27 SS |
7892 | { |
7893 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7894 | } | |
f718cd4a | 7895 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7896 | const char *buf, size_t count) |
5c45bf27 SS |
7897 | { |
7898 | return sched_power_savings_store(buf, count, 1); | |
7899 | } | |
f718cd4a AK |
7900 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7901 | sched_smt_power_savings_show, | |
6707de00 AB |
7902 | sched_smt_power_savings_store); |
7903 | #endif | |
7904 | ||
7905 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7906 | { | |
7907 | int err = 0; | |
7908 | ||
7909 | #ifdef CONFIG_SCHED_SMT | |
7910 | if (smt_capable()) | |
7911 | err = sysfs_create_file(&cls->kset.kobj, | |
7912 | &attr_sched_smt_power_savings.attr); | |
7913 | #endif | |
7914 | #ifdef CONFIG_SCHED_MC | |
7915 | if (!err && mc_capable()) | |
7916 | err = sysfs_create_file(&cls->kset.kobj, | |
7917 | &attr_sched_mc_power_savings.attr); | |
7918 | #endif | |
7919 | return err; | |
7920 | } | |
6d6bc0ad | 7921 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7922 | |
e761b772 | 7923 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7924 | /* |
e761b772 MK |
7925 | * Add online and remove offline CPUs from the scheduler domains. |
7926 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7927 | */ |
7928 | static int update_sched_domains(struct notifier_block *nfb, | |
7929 | unsigned long action, void *hcpu) | |
e761b772 MK |
7930 | { |
7931 | switch (action) { | |
7932 | case CPU_ONLINE: | |
7933 | case CPU_ONLINE_FROZEN: | |
7934 | case CPU_DEAD: | |
7935 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7936 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7937 | return NOTIFY_OK; |
7938 | ||
7939 | default: | |
7940 | return NOTIFY_DONE; | |
7941 | } | |
7942 | } | |
7943 | #endif | |
7944 | ||
7945 | static int update_runtime(struct notifier_block *nfb, | |
7946 | unsigned long action, void *hcpu) | |
1da177e4 | 7947 | { |
7def2be1 PZ |
7948 | int cpu = (int)(long)hcpu; |
7949 | ||
1da177e4 | 7950 | switch (action) { |
1da177e4 | 7951 | case CPU_DOWN_PREPARE: |
8bb78442 | 7952 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7953 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7954 | return NOTIFY_OK; |
7955 | ||
1da177e4 | 7956 | case CPU_DOWN_FAILED: |
8bb78442 | 7957 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7958 | case CPU_ONLINE: |
8bb78442 | 7959 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7960 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7961 | return NOTIFY_OK; |
7962 | ||
1da177e4 LT |
7963 | default: |
7964 | return NOTIFY_DONE; | |
7965 | } | |
1da177e4 | 7966 | } |
1da177e4 LT |
7967 | |
7968 | void __init sched_init_smp(void) | |
7969 | { | |
5c1e1767 NP |
7970 | cpumask_t non_isolated_cpus; |
7971 | ||
434d53b0 MT |
7972 | #if defined(CONFIG_NUMA) |
7973 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7974 | GFP_KERNEL); | |
7975 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7976 | #endif | |
95402b38 | 7977 | get_online_cpus(); |
712555ee | 7978 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7979 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7980 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7981 | if (cpus_empty(non_isolated_cpus)) |
7982 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7983 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7984 | put_online_cpus(); |
e761b772 MK |
7985 | |
7986 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7987 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7988 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7989 | #endif |
7990 | ||
7991 | /* RT runtime code needs to handle some hotplug events */ | |
7992 | hotcpu_notifier(update_runtime, 0); | |
7993 | ||
b328ca18 | 7994 | init_hrtick(); |
5c1e1767 NP |
7995 | |
7996 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7997 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7998 | BUG(); |
19978ca6 | 7999 | sched_init_granularity(); |
1da177e4 LT |
8000 | } |
8001 | #else | |
8002 | void __init sched_init_smp(void) | |
8003 | { | |
19978ca6 | 8004 | sched_init_granularity(); |
1da177e4 LT |
8005 | } |
8006 | #endif /* CONFIG_SMP */ | |
8007 | ||
8008 | int in_sched_functions(unsigned long addr) | |
8009 | { | |
1da177e4 LT |
8010 | return in_lock_functions(addr) || |
8011 | (addr >= (unsigned long)__sched_text_start | |
8012 | && addr < (unsigned long)__sched_text_end); | |
8013 | } | |
8014 | ||
a9957449 | 8015 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8016 | { |
8017 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8018 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8019 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8020 | cfs_rq->rq = rq; | |
8021 | #endif | |
67e9fb2a | 8022 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8023 | } |
8024 | ||
fa85ae24 PZ |
8025 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8026 | { | |
8027 | struct rt_prio_array *array; | |
8028 | int i; | |
8029 | ||
8030 | array = &rt_rq->active; | |
8031 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8032 | INIT_LIST_HEAD(array->queue + i); | |
8033 | __clear_bit(i, array->bitmap); | |
8034 | } | |
8035 | /* delimiter for bitsearch: */ | |
8036 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8037 | ||
052f1dc7 | 8038 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8039 | rt_rq->highest_prio = MAX_RT_PRIO; |
8040 | #endif | |
fa85ae24 PZ |
8041 | #ifdef CONFIG_SMP |
8042 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8043 | rt_rq->overloaded = 0; |
8044 | #endif | |
8045 | ||
8046 | rt_rq->rt_time = 0; | |
8047 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8048 | rt_rq->rt_runtime = 0; |
8049 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8050 | |
052f1dc7 | 8051 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8052 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8053 | rt_rq->rq = rq; |
8054 | #endif | |
fa85ae24 PZ |
8055 | } |
8056 | ||
6f505b16 | 8057 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8058 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8059 | struct sched_entity *se, int cpu, int add, | |
8060 | struct sched_entity *parent) | |
6f505b16 | 8061 | { |
ec7dc8ac | 8062 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8063 | tg->cfs_rq[cpu] = cfs_rq; |
8064 | init_cfs_rq(cfs_rq, rq); | |
8065 | cfs_rq->tg = tg; | |
8066 | if (add) | |
8067 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8068 | ||
8069 | tg->se[cpu] = se; | |
354d60c2 DG |
8070 | /* se could be NULL for init_task_group */ |
8071 | if (!se) | |
8072 | return; | |
8073 | ||
ec7dc8ac DG |
8074 | if (!parent) |
8075 | se->cfs_rq = &rq->cfs; | |
8076 | else | |
8077 | se->cfs_rq = parent->my_q; | |
8078 | ||
6f505b16 PZ |
8079 | se->my_q = cfs_rq; |
8080 | se->load.weight = tg->shares; | |
e05510d0 | 8081 | se->load.inv_weight = 0; |
ec7dc8ac | 8082 | se->parent = parent; |
6f505b16 | 8083 | } |
052f1dc7 | 8084 | #endif |
6f505b16 | 8085 | |
052f1dc7 | 8086 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8087 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8088 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8089 | struct sched_rt_entity *parent) | |
6f505b16 | 8090 | { |
ec7dc8ac DG |
8091 | struct rq *rq = cpu_rq(cpu); |
8092 | ||
6f505b16 PZ |
8093 | tg->rt_rq[cpu] = rt_rq; |
8094 | init_rt_rq(rt_rq, rq); | |
8095 | rt_rq->tg = tg; | |
8096 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8097 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8098 | if (add) |
8099 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8100 | ||
8101 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8102 | if (!rt_se) |
8103 | return; | |
8104 | ||
ec7dc8ac DG |
8105 | if (!parent) |
8106 | rt_se->rt_rq = &rq->rt; | |
8107 | else | |
8108 | rt_se->rt_rq = parent->my_q; | |
8109 | ||
6f505b16 | 8110 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8111 | rt_se->parent = parent; |
6f505b16 PZ |
8112 | INIT_LIST_HEAD(&rt_se->run_list); |
8113 | } | |
8114 | #endif | |
8115 | ||
1da177e4 LT |
8116 | void __init sched_init(void) |
8117 | { | |
dd41f596 | 8118 | int i, j; |
434d53b0 MT |
8119 | unsigned long alloc_size = 0, ptr; |
8120 | ||
8121 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8122 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8123 | #endif | |
8124 | #ifdef CONFIG_RT_GROUP_SCHED | |
8125 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8126 | #endif |
8127 | #ifdef CONFIG_USER_SCHED | |
8128 | alloc_size *= 2; | |
434d53b0 MT |
8129 | #endif |
8130 | /* | |
8131 | * As sched_init() is called before page_alloc is setup, | |
8132 | * we use alloc_bootmem(). | |
8133 | */ | |
8134 | if (alloc_size) { | |
5a9d3225 | 8135 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8136 | |
8137 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8138 | init_task_group.se = (struct sched_entity **)ptr; | |
8139 | ptr += nr_cpu_ids * sizeof(void **); | |
8140 | ||
8141 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8142 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8143 | |
8144 | #ifdef CONFIG_USER_SCHED | |
8145 | root_task_group.se = (struct sched_entity **)ptr; | |
8146 | ptr += nr_cpu_ids * sizeof(void **); | |
8147 | ||
8148 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8149 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8150 | #endif /* CONFIG_USER_SCHED */ |
8151 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8152 | #ifdef CONFIG_RT_GROUP_SCHED |
8153 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8154 | ptr += nr_cpu_ids * sizeof(void **); | |
8155 | ||
8156 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8157 | ptr += nr_cpu_ids * sizeof(void **); |
8158 | ||
8159 | #ifdef CONFIG_USER_SCHED | |
8160 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8161 | ptr += nr_cpu_ids * sizeof(void **); | |
8162 | ||
8163 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8164 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8165 | #endif /* CONFIG_USER_SCHED */ |
8166 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8167 | } |
dd41f596 | 8168 | |
57d885fe GH |
8169 | #ifdef CONFIG_SMP |
8170 | init_defrootdomain(); | |
8171 | #endif | |
8172 | ||
d0b27fa7 PZ |
8173 | init_rt_bandwidth(&def_rt_bandwidth, |
8174 | global_rt_period(), global_rt_runtime()); | |
8175 | ||
8176 | #ifdef CONFIG_RT_GROUP_SCHED | |
8177 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8178 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8179 | #ifdef CONFIG_USER_SCHED |
8180 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8181 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8182 | #endif /* CONFIG_USER_SCHED */ |
8183 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8184 | |
052f1dc7 | 8185 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8186 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8187 | INIT_LIST_HEAD(&init_task_group.children); |
8188 | ||
8189 | #ifdef CONFIG_USER_SCHED | |
8190 | INIT_LIST_HEAD(&root_task_group.children); | |
8191 | init_task_group.parent = &root_task_group; | |
8192 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8193 | #endif /* CONFIG_USER_SCHED */ |
8194 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8195 | |
0a945022 | 8196 | for_each_possible_cpu(i) { |
70b97a7f | 8197 | struct rq *rq; |
1da177e4 LT |
8198 | |
8199 | rq = cpu_rq(i); | |
8200 | spin_lock_init(&rq->lock); | |
7897986b | 8201 | rq->nr_running = 0; |
dd41f596 | 8202 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8203 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8204 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8205 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8206 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8207 | #ifdef CONFIG_CGROUP_SCHED |
8208 | /* | |
8209 | * How much cpu bandwidth does init_task_group get? | |
8210 | * | |
8211 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8212 | * gets 100% of the cpu resources in the system. This overall | |
8213 | * system cpu resource is divided among the tasks of | |
8214 | * init_task_group and its child task-groups in a fair manner, | |
8215 | * based on each entity's (task or task-group's) weight | |
8216 | * (se->load.weight). | |
8217 | * | |
8218 | * In other words, if init_task_group has 10 tasks of weight | |
8219 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8220 | * then A0's share of the cpu resource is: | |
8221 | * | |
8222 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8223 | * | |
8224 | * We achieve this by letting init_task_group's tasks sit | |
8225 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8226 | */ | |
ec7dc8ac | 8227 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8228 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8229 | root_task_group.shares = NICE_0_LOAD; |
8230 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8231 | /* |
8232 | * In case of task-groups formed thr' the user id of tasks, | |
8233 | * init_task_group represents tasks belonging to root user. | |
8234 | * Hence it forms a sibling of all subsequent groups formed. | |
8235 | * In this case, init_task_group gets only a fraction of overall | |
8236 | * system cpu resource, based on the weight assigned to root | |
8237 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8238 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8239 | * (init_cfs_rq) and having one entity represent this group of | |
8240 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8241 | */ | |
ec7dc8ac | 8242 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8243 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8244 | &per_cpu(init_sched_entity, i), i, 1, |
8245 | root_task_group.se[i]); | |
6f505b16 | 8246 | |
052f1dc7 | 8247 | #endif |
354d60c2 DG |
8248 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8249 | ||
8250 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8251 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8252 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8253 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8254 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8255 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8256 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8257 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8258 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8259 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8260 | root_task_group.rt_se[i]); | |
354d60c2 | 8261 | #endif |
dd41f596 | 8262 | #endif |
1da177e4 | 8263 | |
dd41f596 IM |
8264 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8265 | rq->cpu_load[j] = 0; | |
1da177e4 | 8266 | #ifdef CONFIG_SMP |
41c7ce9a | 8267 | rq->sd = NULL; |
57d885fe | 8268 | rq->rd = NULL; |
1da177e4 | 8269 | rq->active_balance = 0; |
dd41f596 | 8270 | rq->next_balance = jiffies; |
1da177e4 | 8271 | rq->push_cpu = 0; |
0a2966b4 | 8272 | rq->cpu = i; |
1f11eb6a | 8273 | rq->online = 0; |
1da177e4 LT |
8274 | rq->migration_thread = NULL; |
8275 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8276 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8277 | #endif |
8f4d37ec | 8278 | init_rq_hrtick(rq); |
1da177e4 | 8279 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8280 | } |
8281 | ||
2dd73a4f | 8282 | set_load_weight(&init_task); |
b50f60ce | 8283 | |
e107be36 AK |
8284 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8285 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8286 | #endif | |
8287 | ||
c9819f45 | 8288 | #ifdef CONFIG_SMP |
962cf36c | 8289 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8290 | #endif |
8291 | ||
b50f60ce HC |
8292 | #ifdef CONFIG_RT_MUTEXES |
8293 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8294 | #endif | |
8295 | ||
1da177e4 LT |
8296 | /* |
8297 | * The boot idle thread does lazy MMU switching as well: | |
8298 | */ | |
8299 | atomic_inc(&init_mm.mm_count); | |
8300 | enter_lazy_tlb(&init_mm, current); | |
8301 | ||
8302 | /* | |
8303 | * Make us the idle thread. Technically, schedule() should not be | |
8304 | * called from this thread, however somewhere below it might be, | |
8305 | * but because we are the idle thread, we just pick up running again | |
8306 | * when this runqueue becomes "idle". | |
8307 | */ | |
8308 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8309 | /* |
8310 | * During early bootup we pretend to be a normal task: | |
8311 | */ | |
8312 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8313 | |
8314 | scheduler_running = 1; | |
1da177e4 LT |
8315 | } |
8316 | ||
8317 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8318 | void __might_sleep(char *file, int line) | |
8319 | { | |
48f24c4d | 8320 | #ifdef in_atomic |
1da177e4 LT |
8321 | static unsigned long prev_jiffy; /* ratelimiting */ |
8322 | ||
aef745fc IM |
8323 | if ((!in_atomic() && !irqs_disabled()) || |
8324 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8325 | return; | |
8326 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8327 | return; | |
8328 | prev_jiffy = jiffies; | |
8329 | ||
8330 | printk(KERN_ERR | |
8331 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8332 | file, line); | |
8333 | printk(KERN_ERR | |
8334 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8335 | in_atomic(), irqs_disabled(), | |
8336 | current->pid, current->comm); | |
8337 | ||
8338 | debug_show_held_locks(current); | |
8339 | if (irqs_disabled()) | |
8340 | print_irqtrace_events(current); | |
8341 | dump_stack(); | |
1da177e4 LT |
8342 | #endif |
8343 | } | |
8344 | EXPORT_SYMBOL(__might_sleep); | |
8345 | #endif | |
8346 | ||
8347 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8348 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8349 | { | |
8350 | int on_rq; | |
3e51f33f | 8351 | |
3a5e4dc1 AK |
8352 | update_rq_clock(rq); |
8353 | on_rq = p->se.on_rq; | |
8354 | if (on_rq) | |
8355 | deactivate_task(rq, p, 0); | |
8356 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8357 | if (on_rq) { | |
8358 | activate_task(rq, p, 0); | |
8359 | resched_task(rq->curr); | |
8360 | } | |
8361 | } | |
8362 | ||
1da177e4 LT |
8363 | void normalize_rt_tasks(void) |
8364 | { | |
a0f98a1c | 8365 | struct task_struct *g, *p; |
1da177e4 | 8366 | unsigned long flags; |
70b97a7f | 8367 | struct rq *rq; |
1da177e4 | 8368 | |
4cf5d77a | 8369 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8370 | do_each_thread(g, p) { |
178be793 IM |
8371 | /* |
8372 | * Only normalize user tasks: | |
8373 | */ | |
8374 | if (!p->mm) | |
8375 | continue; | |
8376 | ||
6cfb0d5d | 8377 | p->se.exec_start = 0; |
6cfb0d5d | 8378 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8379 | p->se.wait_start = 0; |
dd41f596 | 8380 | p->se.sleep_start = 0; |
dd41f596 | 8381 | p->se.block_start = 0; |
6cfb0d5d | 8382 | #endif |
dd41f596 IM |
8383 | |
8384 | if (!rt_task(p)) { | |
8385 | /* | |
8386 | * Renice negative nice level userspace | |
8387 | * tasks back to 0: | |
8388 | */ | |
8389 | if (TASK_NICE(p) < 0 && p->mm) | |
8390 | set_user_nice(p, 0); | |
1da177e4 | 8391 | continue; |
dd41f596 | 8392 | } |
1da177e4 | 8393 | |
4cf5d77a | 8394 | spin_lock(&p->pi_lock); |
b29739f9 | 8395 | rq = __task_rq_lock(p); |
1da177e4 | 8396 | |
178be793 | 8397 | normalize_task(rq, p); |
3a5e4dc1 | 8398 | |
b29739f9 | 8399 | __task_rq_unlock(rq); |
4cf5d77a | 8400 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8401 | } while_each_thread(g, p); |
8402 | ||
4cf5d77a | 8403 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8404 | } |
8405 | ||
8406 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8407 | |
8408 | #ifdef CONFIG_IA64 | |
8409 | /* | |
8410 | * These functions are only useful for the IA64 MCA handling. | |
8411 | * | |
8412 | * They can only be called when the whole system has been | |
8413 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8414 | * activity can take place. Using them for anything else would | |
8415 | * be a serious bug, and as a result, they aren't even visible | |
8416 | * under any other configuration. | |
8417 | */ | |
8418 | ||
8419 | /** | |
8420 | * curr_task - return the current task for a given cpu. | |
8421 | * @cpu: the processor in question. | |
8422 | * | |
8423 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8424 | */ | |
36c8b586 | 8425 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8426 | { |
8427 | return cpu_curr(cpu); | |
8428 | } | |
8429 | ||
8430 | /** | |
8431 | * set_curr_task - set the current task for a given cpu. | |
8432 | * @cpu: the processor in question. | |
8433 | * @p: the task pointer to set. | |
8434 | * | |
8435 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8436 | * are serviced on a separate stack. It allows the architecture to switch the |
8437 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8438 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8439 | * and caller must save the original value of the current task (see | |
8440 | * curr_task() above) and restore that value before reenabling interrupts and | |
8441 | * re-starting the system. | |
8442 | * | |
8443 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8444 | */ | |
36c8b586 | 8445 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8446 | { |
8447 | cpu_curr(cpu) = p; | |
8448 | } | |
8449 | ||
8450 | #endif | |
29f59db3 | 8451 | |
bccbe08a PZ |
8452 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8453 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8454 | { |
8455 | int i; | |
8456 | ||
8457 | for_each_possible_cpu(i) { | |
8458 | if (tg->cfs_rq) | |
8459 | kfree(tg->cfs_rq[i]); | |
8460 | if (tg->se) | |
8461 | kfree(tg->se[i]); | |
6f505b16 PZ |
8462 | } |
8463 | ||
8464 | kfree(tg->cfs_rq); | |
8465 | kfree(tg->se); | |
6f505b16 PZ |
8466 | } |
8467 | ||
ec7dc8ac DG |
8468 | static |
8469 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8470 | { |
29f59db3 | 8471 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8472 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8473 | struct rq *rq; |
29f59db3 SV |
8474 | int i; |
8475 | ||
434d53b0 | 8476 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8477 | if (!tg->cfs_rq) |
8478 | goto err; | |
434d53b0 | 8479 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8480 | if (!tg->se) |
8481 | goto err; | |
052f1dc7 PZ |
8482 | |
8483 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8484 | |
8485 | for_each_possible_cpu(i) { | |
9b5b7751 | 8486 | rq = cpu_rq(i); |
29f59db3 | 8487 | |
6f505b16 PZ |
8488 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8489 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8490 | if (!cfs_rq) |
8491 | goto err; | |
8492 | ||
6f505b16 PZ |
8493 | se = kmalloc_node(sizeof(struct sched_entity), |
8494 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8495 | if (!se) |
8496 | goto err; | |
8497 | ||
ec7dc8ac DG |
8498 | parent_se = parent ? parent->se[i] : NULL; |
8499 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8500 | } |
8501 | ||
8502 | return 1; | |
8503 | ||
8504 | err: | |
8505 | return 0; | |
8506 | } | |
8507 | ||
8508 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8509 | { | |
8510 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8511 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8512 | } | |
8513 | ||
8514 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8515 | { | |
8516 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8517 | } | |
6d6bc0ad | 8518 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8519 | static inline void free_fair_sched_group(struct task_group *tg) |
8520 | { | |
8521 | } | |
8522 | ||
ec7dc8ac DG |
8523 | static inline |
8524 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8525 | { |
8526 | return 1; | |
8527 | } | |
8528 | ||
8529 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8530 | { | |
8531 | } | |
8532 | ||
8533 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8534 | { | |
8535 | } | |
6d6bc0ad | 8536 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8537 | |
8538 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8539 | static void free_rt_sched_group(struct task_group *tg) |
8540 | { | |
8541 | int i; | |
8542 | ||
d0b27fa7 PZ |
8543 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8544 | ||
bccbe08a PZ |
8545 | for_each_possible_cpu(i) { |
8546 | if (tg->rt_rq) | |
8547 | kfree(tg->rt_rq[i]); | |
8548 | if (tg->rt_se) | |
8549 | kfree(tg->rt_se[i]); | |
8550 | } | |
8551 | ||
8552 | kfree(tg->rt_rq); | |
8553 | kfree(tg->rt_se); | |
8554 | } | |
8555 | ||
ec7dc8ac DG |
8556 | static |
8557 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8558 | { |
8559 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8560 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8561 | struct rq *rq; |
8562 | int i; | |
8563 | ||
434d53b0 | 8564 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8565 | if (!tg->rt_rq) |
8566 | goto err; | |
434d53b0 | 8567 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8568 | if (!tg->rt_se) |
8569 | goto err; | |
8570 | ||
d0b27fa7 PZ |
8571 | init_rt_bandwidth(&tg->rt_bandwidth, |
8572 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8573 | |
8574 | for_each_possible_cpu(i) { | |
8575 | rq = cpu_rq(i); | |
8576 | ||
6f505b16 PZ |
8577 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8578 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8579 | if (!rt_rq) | |
8580 | goto err; | |
29f59db3 | 8581 | |
6f505b16 PZ |
8582 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8583 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8584 | if (!rt_se) | |
8585 | goto err; | |
29f59db3 | 8586 | |
ec7dc8ac DG |
8587 | parent_se = parent ? parent->rt_se[i] : NULL; |
8588 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8589 | } |
8590 | ||
bccbe08a PZ |
8591 | return 1; |
8592 | ||
8593 | err: | |
8594 | return 0; | |
8595 | } | |
8596 | ||
8597 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8598 | { | |
8599 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8600 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8601 | } | |
8602 | ||
8603 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8604 | { | |
8605 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8606 | } | |
6d6bc0ad | 8607 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8608 | static inline void free_rt_sched_group(struct task_group *tg) |
8609 | { | |
8610 | } | |
8611 | ||
ec7dc8ac DG |
8612 | static inline |
8613 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8614 | { |
8615 | return 1; | |
8616 | } | |
8617 | ||
8618 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8619 | { | |
8620 | } | |
8621 | ||
8622 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8623 | { | |
8624 | } | |
6d6bc0ad | 8625 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8626 | |
d0b27fa7 | 8627 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8628 | static void free_sched_group(struct task_group *tg) |
8629 | { | |
8630 | free_fair_sched_group(tg); | |
8631 | free_rt_sched_group(tg); | |
8632 | kfree(tg); | |
8633 | } | |
8634 | ||
8635 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8636 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8637 | { |
8638 | struct task_group *tg; | |
8639 | unsigned long flags; | |
8640 | int i; | |
8641 | ||
8642 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8643 | if (!tg) | |
8644 | return ERR_PTR(-ENOMEM); | |
8645 | ||
ec7dc8ac | 8646 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8647 | goto err; |
8648 | ||
ec7dc8ac | 8649 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8650 | goto err; |
8651 | ||
8ed36996 | 8652 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8653 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8654 | register_fair_sched_group(tg, i); |
8655 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8656 | } |
6f505b16 | 8657 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8658 | |
8659 | WARN_ON(!parent); /* root should already exist */ | |
8660 | ||
8661 | tg->parent = parent; | |
f473aa5e | 8662 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8663 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8664 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8665 | |
9b5b7751 | 8666 | return tg; |
29f59db3 SV |
8667 | |
8668 | err: | |
6f505b16 | 8669 | free_sched_group(tg); |
29f59db3 SV |
8670 | return ERR_PTR(-ENOMEM); |
8671 | } | |
8672 | ||
9b5b7751 | 8673 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8674 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8675 | { |
29f59db3 | 8676 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8677 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8678 | } |
8679 | ||
9b5b7751 | 8680 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8681 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8682 | { |
8ed36996 | 8683 | unsigned long flags; |
9b5b7751 | 8684 | int i; |
29f59db3 | 8685 | |
8ed36996 | 8686 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8687 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8688 | unregister_fair_sched_group(tg, i); |
8689 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8690 | } |
6f505b16 | 8691 | list_del_rcu(&tg->list); |
f473aa5e | 8692 | list_del_rcu(&tg->siblings); |
8ed36996 | 8693 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8694 | |
9b5b7751 | 8695 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8696 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8697 | } |
8698 | ||
9b5b7751 | 8699 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8700 | * The caller of this function should have put the task in its new group |
8701 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8702 | * reflect its new group. | |
9b5b7751 SV |
8703 | */ |
8704 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8705 | { |
8706 | int on_rq, running; | |
8707 | unsigned long flags; | |
8708 | struct rq *rq; | |
8709 | ||
8710 | rq = task_rq_lock(tsk, &flags); | |
8711 | ||
29f59db3 SV |
8712 | update_rq_clock(rq); |
8713 | ||
051a1d1a | 8714 | running = task_current(rq, tsk); |
29f59db3 SV |
8715 | on_rq = tsk->se.on_rq; |
8716 | ||
0e1f3483 | 8717 | if (on_rq) |
29f59db3 | 8718 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8719 | if (unlikely(running)) |
8720 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8721 | |
6f505b16 | 8722 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8723 | |
810b3817 PZ |
8724 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8725 | if (tsk->sched_class->moved_group) | |
8726 | tsk->sched_class->moved_group(tsk); | |
8727 | #endif | |
8728 | ||
0e1f3483 HS |
8729 | if (unlikely(running)) |
8730 | tsk->sched_class->set_curr_task(rq); | |
8731 | if (on_rq) | |
7074badb | 8732 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8733 | |
29f59db3 SV |
8734 | task_rq_unlock(rq, &flags); |
8735 | } | |
6d6bc0ad | 8736 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8737 | |
052f1dc7 | 8738 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8739 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8740 | { |
8741 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8742 | int on_rq; |
8743 | ||
29f59db3 | 8744 | on_rq = se->on_rq; |
62fb1851 | 8745 | if (on_rq) |
29f59db3 SV |
8746 | dequeue_entity(cfs_rq, se, 0); |
8747 | ||
8748 | se->load.weight = shares; | |
e05510d0 | 8749 | se->load.inv_weight = 0; |
29f59db3 | 8750 | |
62fb1851 | 8751 | if (on_rq) |
29f59db3 | 8752 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8753 | } |
62fb1851 | 8754 | |
c09595f6 PZ |
8755 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8756 | { | |
8757 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8758 | struct rq *rq = cfs_rq->rq; | |
8759 | unsigned long flags; | |
8760 | ||
8761 | spin_lock_irqsave(&rq->lock, flags); | |
8762 | __set_se_shares(se, shares); | |
8763 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8764 | } |
8765 | ||
8ed36996 PZ |
8766 | static DEFINE_MUTEX(shares_mutex); |
8767 | ||
4cf86d77 | 8768 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8769 | { |
8770 | int i; | |
8ed36996 | 8771 | unsigned long flags; |
c61935fd | 8772 | |
ec7dc8ac DG |
8773 | /* |
8774 | * We can't change the weight of the root cgroup. | |
8775 | */ | |
8776 | if (!tg->se[0]) | |
8777 | return -EINVAL; | |
8778 | ||
18d95a28 PZ |
8779 | if (shares < MIN_SHARES) |
8780 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8781 | else if (shares > MAX_SHARES) |
8782 | shares = MAX_SHARES; | |
62fb1851 | 8783 | |
8ed36996 | 8784 | mutex_lock(&shares_mutex); |
9b5b7751 | 8785 | if (tg->shares == shares) |
5cb350ba | 8786 | goto done; |
29f59db3 | 8787 | |
8ed36996 | 8788 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8789 | for_each_possible_cpu(i) |
8790 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8791 | list_del_rcu(&tg->siblings); |
8ed36996 | 8792 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8793 | |
8794 | /* wait for any ongoing reference to this group to finish */ | |
8795 | synchronize_sched(); | |
8796 | ||
8797 | /* | |
8798 | * Now we are free to modify the group's share on each cpu | |
8799 | * w/o tripping rebalance_share or load_balance_fair. | |
8800 | */ | |
9b5b7751 | 8801 | tg->shares = shares; |
c09595f6 PZ |
8802 | for_each_possible_cpu(i) { |
8803 | /* | |
8804 | * force a rebalance | |
8805 | */ | |
8806 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8807 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8808 | } |
29f59db3 | 8809 | |
6b2d7700 SV |
8810 | /* |
8811 | * Enable load balance activity on this group, by inserting it back on | |
8812 | * each cpu's rq->leaf_cfs_rq_list. | |
8813 | */ | |
8ed36996 | 8814 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8815 | for_each_possible_cpu(i) |
8816 | register_fair_sched_group(tg, i); | |
f473aa5e | 8817 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8818 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8819 | done: |
8ed36996 | 8820 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8821 | return 0; |
29f59db3 SV |
8822 | } |
8823 | ||
5cb350ba DG |
8824 | unsigned long sched_group_shares(struct task_group *tg) |
8825 | { | |
8826 | return tg->shares; | |
8827 | } | |
052f1dc7 | 8828 | #endif |
5cb350ba | 8829 | |
052f1dc7 | 8830 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8831 | /* |
9f0c1e56 | 8832 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8833 | */ |
9f0c1e56 PZ |
8834 | static DEFINE_MUTEX(rt_constraints_mutex); |
8835 | ||
8836 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8837 | { | |
8838 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8839 | return 1ULL << 20; |
9f0c1e56 | 8840 | |
9a7e0b18 | 8841 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8842 | } |
8843 | ||
9a7e0b18 PZ |
8844 | /* Must be called with tasklist_lock held */ |
8845 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8846 | { |
9a7e0b18 | 8847 | struct task_struct *g, *p; |
b40b2e8e | 8848 | |
9a7e0b18 PZ |
8849 | do_each_thread(g, p) { |
8850 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8851 | return 1; | |
8852 | } while_each_thread(g, p); | |
b40b2e8e | 8853 | |
9a7e0b18 PZ |
8854 | return 0; |
8855 | } | |
b40b2e8e | 8856 | |
9a7e0b18 PZ |
8857 | struct rt_schedulable_data { |
8858 | struct task_group *tg; | |
8859 | u64 rt_period; | |
8860 | u64 rt_runtime; | |
8861 | }; | |
b40b2e8e | 8862 | |
9a7e0b18 PZ |
8863 | static int tg_schedulable(struct task_group *tg, void *data) |
8864 | { | |
8865 | struct rt_schedulable_data *d = data; | |
8866 | struct task_group *child; | |
8867 | unsigned long total, sum = 0; | |
8868 | u64 period, runtime; | |
b40b2e8e | 8869 | |
9a7e0b18 PZ |
8870 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8871 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8872 | |
9a7e0b18 PZ |
8873 | if (tg == d->tg) { |
8874 | period = d->rt_period; | |
8875 | runtime = d->rt_runtime; | |
b40b2e8e | 8876 | } |
b40b2e8e | 8877 | |
4653f803 PZ |
8878 | /* |
8879 | * Cannot have more runtime than the period. | |
8880 | */ | |
8881 | if (runtime > period && runtime != RUNTIME_INF) | |
8882 | return -EINVAL; | |
6f505b16 | 8883 | |
4653f803 PZ |
8884 | /* |
8885 | * Ensure we don't starve existing RT tasks. | |
8886 | */ | |
9a7e0b18 PZ |
8887 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8888 | return -EBUSY; | |
6f505b16 | 8889 | |
9a7e0b18 | 8890 | total = to_ratio(period, runtime); |
6f505b16 | 8891 | |
4653f803 PZ |
8892 | /* |
8893 | * Nobody can have more than the global setting allows. | |
8894 | */ | |
8895 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8896 | return -EINVAL; | |
6f505b16 | 8897 | |
4653f803 PZ |
8898 | /* |
8899 | * The sum of our children's runtime should not exceed our own. | |
8900 | */ | |
9a7e0b18 PZ |
8901 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8902 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8903 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8904 | |
9a7e0b18 PZ |
8905 | if (child == d->tg) { |
8906 | period = d->rt_period; | |
8907 | runtime = d->rt_runtime; | |
8908 | } | |
6f505b16 | 8909 | |
9a7e0b18 | 8910 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8911 | } |
6f505b16 | 8912 | |
9a7e0b18 PZ |
8913 | if (sum > total) |
8914 | return -EINVAL; | |
8915 | ||
8916 | return 0; | |
6f505b16 PZ |
8917 | } |
8918 | ||
9a7e0b18 | 8919 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8920 | { |
9a7e0b18 PZ |
8921 | struct rt_schedulable_data data = { |
8922 | .tg = tg, | |
8923 | .rt_period = period, | |
8924 | .rt_runtime = runtime, | |
8925 | }; | |
8926 | ||
8927 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8928 | } |
8929 | ||
d0b27fa7 PZ |
8930 | static int tg_set_bandwidth(struct task_group *tg, |
8931 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8932 | { |
ac086bc2 | 8933 | int i, err = 0; |
9f0c1e56 | 8934 | |
9f0c1e56 | 8935 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8936 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8937 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8938 | if (err) | |
9f0c1e56 | 8939 | goto unlock; |
ac086bc2 PZ |
8940 | |
8941 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8942 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8943 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8944 | |
8945 | for_each_possible_cpu(i) { | |
8946 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8947 | ||
8948 | spin_lock(&rt_rq->rt_runtime_lock); | |
8949 | rt_rq->rt_runtime = rt_runtime; | |
8950 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8951 | } | |
8952 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8953 | unlock: |
521f1a24 | 8954 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8955 | mutex_unlock(&rt_constraints_mutex); |
8956 | ||
8957 | return err; | |
6f505b16 PZ |
8958 | } |
8959 | ||
d0b27fa7 PZ |
8960 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8961 | { | |
8962 | u64 rt_runtime, rt_period; | |
8963 | ||
8964 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8965 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8966 | if (rt_runtime_us < 0) | |
8967 | rt_runtime = RUNTIME_INF; | |
8968 | ||
8969 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8970 | } | |
8971 | ||
9f0c1e56 PZ |
8972 | long sched_group_rt_runtime(struct task_group *tg) |
8973 | { | |
8974 | u64 rt_runtime_us; | |
8975 | ||
d0b27fa7 | 8976 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8977 | return -1; |
8978 | ||
d0b27fa7 | 8979 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8980 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8981 | return rt_runtime_us; | |
8982 | } | |
d0b27fa7 PZ |
8983 | |
8984 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8985 | { | |
8986 | u64 rt_runtime, rt_period; | |
8987 | ||
8988 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8989 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8990 | ||
619b0488 R |
8991 | if (rt_period == 0) |
8992 | return -EINVAL; | |
8993 | ||
d0b27fa7 PZ |
8994 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8995 | } | |
8996 | ||
8997 | long sched_group_rt_period(struct task_group *tg) | |
8998 | { | |
8999 | u64 rt_period_us; | |
9000 | ||
9001 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9002 | do_div(rt_period_us, NSEC_PER_USEC); | |
9003 | return rt_period_us; | |
9004 | } | |
9005 | ||
9006 | static int sched_rt_global_constraints(void) | |
9007 | { | |
4653f803 | 9008 | u64 runtime, period; |
d0b27fa7 PZ |
9009 | int ret = 0; |
9010 | ||
ec5d4989 HS |
9011 | if (sysctl_sched_rt_period <= 0) |
9012 | return -EINVAL; | |
9013 | ||
4653f803 PZ |
9014 | runtime = global_rt_runtime(); |
9015 | period = global_rt_period(); | |
9016 | ||
9017 | /* | |
9018 | * Sanity check on the sysctl variables. | |
9019 | */ | |
9020 | if (runtime > period && runtime != RUNTIME_INF) | |
9021 | return -EINVAL; | |
10b612f4 | 9022 | |
d0b27fa7 | 9023 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9024 | read_lock(&tasklist_lock); |
4653f803 | 9025 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9026 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9027 | mutex_unlock(&rt_constraints_mutex); |
9028 | ||
9029 | return ret; | |
9030 | } | |
6d6bc0ad | 9031 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9032 | static int sched_rt_global_constraints(void) |
9033 | { | |
ac086bc2 PZ |
9034 | unsigned long flags; |
9035 | int i; | |
9036 | ||
ec5d4989 HS |
9037 | if (sysctl_sched_rt_period <= 0) |
9038 | return -EINVAL; | |
9039 | ||
ac086bc2 PZ |
9040 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9041 | for_each_possible_cpu(i) { | |
9042 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9043 | ||
9044 | spin_lock(&rt_rq->rt_runtime_lock); | |
9045 | rt_rq->rt_runtime = global_rt_runtime(); | |
9046 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9047 | } | |
9048 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9049 | ||
d0b27fa7 PZ |
9050 | return 0; |
9051 | } | |
6d6bc0ad | 9052 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9053 | |
9054 | int sched_rt_handler(struct ctl_table *table, int write, | |
9055 | struct file *filp, void __user *buffer, size_t *lenp, | |
9056 | loff_t *ppos) | |
9057 | { | |
9058 | int ret; | |
9059 | int old_period, old_runtime; | |
9060 | static DEFINE_MUTEX(mutex); | |
9061 | ||
9062 | mutex_lock(&mutex); | |
9063 | old_period = sysctl_sched_rt_period; | |
9064 | old_runtime = sysctl_sched_rt_runtime; | |
9065 | ||
9066 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9067 | ||
9068 | if (!ret && write) { | |
9069 | ret = sched_rt_global_constraints(); | |
9070 | if (ret) { | |
9071 | sysctl_sched_rt_period = old_period; | |
9072 | sysctl_sched_rt_runtime = old_runtime; | |
9073 | } else { | |
9074 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9075 | def_rt_bandwidth.rt_period = | |
9076 | ns_to_ktime(global_rt_period()); | |
9077 | } | |
9078 | } | |
9079 | mutex_unlock(&mutex); | |
9080 | ||
9081 | return ret; | |
9082 | } | |
68318b8e | 9083 | |
052f1dc7 | 9084 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9085 | |
9086 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9087 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9088 | { |
2b01dfe3 PM |
9089 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9090 | struct task_group, css); | |
68318b8e SV |
9091 | } |
9092 | ||
9093 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9094 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9095 | { |
ec7dc8ac | 9096 | struct task_group *tg, *parent; |
68318b8e | 9097 | |
2b01dfe3 | 9098 | if (!cgrp->parent) { |
68318b8e | 9099 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9100 | return &init_task_group.css; |
9101 | } | |
9102 | ||
ec7dc8ac DG |
9103 | parent = cgroup_tg(cgrp->parent); |
9104 | tg = sched_create_group(parent); | |
68318b8e SV |
9105 | if (IS_ERR(tg)) |
9106 | return ERR_PTR(-ENOMEM); | |
9107 | ||
68318b8e SV |
9108 | return &tg->css; |
9109 | } | |
9110 | ||
41a2d6cf IM |
9111 | static void |
9112 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9113 | { |
2b01dfe3 | 9114 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9115 | |
9116 | sched_destroy_group(tg); | |
9117 | } | |
9118 | ||
41a2d6cf IM |
9119 | static int |
9120 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9121 | struct task_struct *tsk) | |
68318b8e | 9122 | { |
b68aa230 PZ |
9123 | #ifdef CONFIG_RT_GROUP_SCHED |
9124 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9125 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9126 | return -EINVAL; |
9127 | #else | |
68318b8e SV |
9128 | /* We don't support RT-tasks being in separate groups */ |
9129 | if (tsk->sched_class != &fair_sched_class) | |
9130 | return -EINVAL; | |
b68aa230 | 9131 | #endif |
68318b8e SV |
9132 | |
9133 | return 0; | |
9134 | } | |
9135 | ||
9136 | static void | |
2b01dfe3 | 9137 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9138 | struct cgroup *old_cont, struct task_struct *tsk) |
9139 | { | |
9140 | sched_move_task(tsk); | |
9141 | } | |
9142 | ||
052f1dc7 | 9143 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9144 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9145 | u64 shareval) |
68318b8e | 9146 | { |
2b01dfe3 | 9147 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9148 | } |
9149 | ||
f4c753b7 | 9150 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9151 | { |
2b01dfe3 | 9152 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9153 | |
9154 | return (u64) tg->shares; | |
9155 | } | |
6d6bc0ad | 9156 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9157 | |
052f1dc7 | 9158 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9159 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9160 | s64 val) |
6f505b16 | 9161 | { |
06ecb27c | 9162 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9163 | } |
9164 | ||
06ecb27c | 9165 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9166 | { |
06ecb27c | 9167 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9168 | } |
d0b27fa7 PZ |
9169 | |
9170 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9171 | u64 rt_period_us) | |
9172 | { | |
9173 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9174 | } | |
9175 | ||
9176 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9177 | { | |
9178 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9179 | } | |
6d6bc0ad | 9180 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9181 | |
fe5c7cc2 | 9182 | static struct cftype cpu_files[] = { |
052f1dc7 | 9183 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9184 | { |
9185 | .name = "shares", | |
f4c753b7 PM |
9186 | .read_u64 = cpu_shares_read_u64, |
9187 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9188 | }, |
052f1dc7 PZ |
9189 | #endif |
9190 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9191 | { |
9f0c1e56 | 9192 | .name = "rt_runtime_us", |
06ecb27c PM |
9193 | .read_s64 = cpu_rt_runtime_read, |
9194 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9195 | }, |
d0b27fa7 PZ |
9196 | { |
9197 | .name = "rt_period_us", | |
f4c753b7 PM |
9198 | .read_u64 = cpu_rt_period_read_uint, |
9199 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9200 | }, |
052f1dc7 | 9201 | #endif |
68318b8e SV |
9202 | }; |
9203 | ||
9204 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9205 | { | |
fe5c7cc2 | 9206 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9207 | } |
9208 | ||
9209 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9210 | .name = "cpu", |
9211 | .create = cpu_cgroup_create, | |
9212 | .destroy = cpu_cgroup_destroy, | |
9213 | .can_attach = cpu_cgroup_can_attach, | |
9214 | .attach = cpu_cgroup_attach, | |
9215 | .populate = cpu_cgroup_populate, | |
9216 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9217 | .early_init = 1, |
9218 | }; | |
9219 | ||
052f1dc7 | 9220 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9221 | |
9222 | #ifdef CONFIG_CGROUP_CPUACCT | |
9223 | ||
9224 | /* | |
9225 | * CPU accounting code for task groups. | |
9226 | * | |
9227 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9228 | * (balbir@in.ibm.com). | |
9229 | */ | |
9230 | ||
9231 | /* track cpu usage of a group of tasks */ | |
9232 | struct cpuacct { | |
9233 | struct cgroup_subsys_state css; | |
9234 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9235 | u64 *cpuusage; | |
9236 | }; | |
9237 | ||
9238 | struct cgroup_subsys cpuacct_subsys; | |
9239 | ||
9240 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9241 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9242 | { |
32cd756a | 9243 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9244 | struct cpuacct, css); |
9245 | } | |
9246 | ||
9247 | /* return cpu accounting group to which this task belongs */ | |
9248 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9249 | { | |
9250 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9251 | struct cpuacct, css); | |
9252 | } | |
9253 | ||
9254 | /* create a new cpu accounting group */ | |
9255 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9256 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9257 | { |
9258 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9259 | ||
9260 | if (!ca) | |
9261 | return ERR_PTR(-ENOMEM); | |
9262 | ||
9263 | ca->cpuusage = alloc_percpu(u64); | |
9264 | if (!ca->cpuusage) { | |
9265 | kfree(ca); | |
9266 | return ERR_PTR(-ENOMEM); | |
9267 | } | |
9268 | ||
9269 | return &ca->css; | |
9270 | } | |
9271 | ||
9272 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9273 | static void |
32cd756a | 9274 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9275 | { |
32cd756a | 9276 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9277 | |
9278 | free_percpu(ca->cpuusage); | |
9279 | kfree(ca); | |
9280 | } | |
9281 | ||
9282 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9283 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9284 | { |
32cd756a | 9285 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9286 | u64 totalcpuusage = 0; |
9287 | int i; | |
9288 | ||
9289 | for_each_possible_cpu(i) { | |
9290 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9291 | ||
9292 | /* | |
9293 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9294 | * platforms. | |
9295 | */ | |
9296 | spin_lock_irq(&cpu_rq(i)->lock); | |
9297 | totalcpuusage += *cpuusage; | |
9298 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9299 | } | |
9300 | ||
9301 | return totalcpuusage; | |
9302 | } | |
9303 | ||
0297b803 DG |
9304 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9305 | u64 reset) | |
9306 | { | |
9307 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9308 | int err = 0; | |
9309 | int i; | |
9310 | ||
9311 | if (reset) { | |
9312 | err = -EINVAL; | |
9313 | goto out; | |
9314 | } | |
9315 | ||
9316 | for_each_possible_cpu(i) { | |
9317 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9318 | ||
9319 | spin_lock_irq(&cpu_rq(i)->lock); | |
9320 | *cpuusage = 0; | |
9321 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9322 | } | |
9323 | out: | |
9324 | return err; | |
9325 | } | |
9326 | ||
d842de87 SV |
9327 | static struct cftype files[] = { |
9328 | { | |
9329 | .name = "usage", | |
f4c753b7 PM |
9330 | .read_u64 = cpuusage_read, |
9331 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9332 | }, |
9333 | }; | |
9334 | ||
32cd756a | 9335 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9336 | { |
32cd756a | 9337 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9338 | } |
9339 | ||
9340 | /* | |
9341 | * charge this task's execution time to its accounting group. | |
9342 | * | |
9343 | * called with rq->lock held. | |
9344 | */ | |
9345 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9346 | { | |
9347 | struct cpuacct *ca; | |
9348 | ||
9349 | if (!cpuacct_subsys.active) | |
9350 | return; | |
9351 | ||
9352 | ca = task_ca(tsk); | |
9353 | if (ca) { | |
9354 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9355 | ||
9356 | *cpuusage += cputime; | |
9357 | } | |
9358 | } | |
9359 | ||
9360 | struct cgroup_subsys cpuacct_subsys = { | |
9361 | .name = "cpuacct", | |
9362 | .create = cpuacct_create, | |
9363 | .destroy = cpuacct_destroy, | |
9364 | .populate = cpuacct_populate, | |
9365 | .subsys_id = cpuacct_subsys_id, | |
9366 | }; | |
9367 | #endif /* CONFIG_CGROUP_CPUACCT */ |