]>
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> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
5517d86b | 66 | #include <linux/reciprocal_div.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
434d53b0 | 71 | #include <linux/bootmem.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
0a16b607 | 75 | #include <trace/sched.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
d0b27fa7 PZ |
116 | /* |
117 | * single value that denotes runtime == period, ie unlimited time. | |
118 | */ | |
119 | #define RUNTIME_INF ((u64)~0ULL) | |
120 | ||
5517d86b ED |
121 | #ifdef CONFIG_SMP |
122 | /* | |
123 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
124 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
125 | */ | |
126 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
127 | { | |
128 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
129 | } | |
130 | ||
131 | /* | |
132 | * Each time a sched group cpu_power is changed, | |
133 | * we must compute its reciprocal value | |
134 | */ | |
135 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
136 | { | |
137 | sg->__cpu_power += val; | |
138 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
139 | } | |
140 | #endif | |
141 | ||
e05606d3 IM |
142 | static inline int rt_policy(int policy) |
143 | { | |
3f33a7ce | 144 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
145 | return 1; |
146 | return 0; | |
147 | } | |
148 | ||
149 | static inline int task_has_rt_policy(struct task_struct *p) | |
150 | { | |
151 | return rt_policy(p->policy); | |
152 | } | |
153 | ||
1da177e4 | 154 | /* |
6aa645ea | 155 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 156 | */ |
6aa645ea IM |
157 | struct rt_prio_array { |
158 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
159 | struct list_head queue[MAX_RT_PRIO]; | |
160 | }; | |
161 | ||
d0b27fa7 | 162 | struct rt_bandwidth { |
ea736ed5 IM |
163 | /* nests inside the rq lock: */ |
164 | spinlock_t rt_runtime_lock; | |
165 | ktime_t rt_period; | |
166 | u64 rt_runtime; | |
167 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
168 | }; |
169 | ||
170 | static struct rt_bandwidth def_rt_bandwidth; | |
171 | ||
172 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
173 | ||
174 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
175 | { | |
176 | struct rt_bandwidth *rt_b = | |
177 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
178 | ktime_t now; | |
179 | int overrun; | |
180 | int idle = 0; | |
181 | ||
182 | for (;;) { | |
183 | now = hrtimer_cb_get_time(timer); | |
184 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
185 | ||
186 | if (!overrun) | |
187 | break; | |
188 | ||
189 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
190 | } | |
191 | ||
192 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
193 | } | |
194 | ||
195 | static | |
196 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
197 | { | |
198 | rt_b->rt_period = ns_to_ktime(period); | |
199 | rt_b->rt_runtime = runtime; | |
200 | ||
ac086bc2 PZ |
201 | spin_lock_init(&rt_b->rt_runtime_lock); |
202 | ||
d0b27fa7 PZ |
203 | hrtimer_init(&rt_b->rt_period_timer, |
204 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
205 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
ccc7dadf | 206 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; |
d0b27fa7 PZ |
207 | } |
208 | ||
c8bfff6d KH |
209 | static inline int rt_bandwidth_enabled(void) |
210 | { | |
211 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
212 | } |
213 | ||
214 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
215 | { | |
216 | ktime_t now; | |
217 | ||
0b148fa0 | 218 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
219 | return; |
220 | ||
221 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
222 | return; | |
223 | ||
224 | spin_lock(&rt_b->rt_runtime_lock); | |
225 | for (;;) { | |
226 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
227 | break; | |
228 | ||
229 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
230 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
231 | hrtimer_start_expires(&rt_b->rt_period_timer, |
232 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
233 | } |
234 | spin_unlock(&rt_b->rt_runtime_lock); | |
235 | } | |
236 | ||
237 | #ifdef CONFIG_RT_GROUP_SCHED | |
238 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
239 | { | |
240 | hrtimer_cancel(&rt_b->rt_period_timer); | |
241 | } | |
242 | #endif | |
243 | ||
712555ee HC |
244 | /* |
245 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
246 | * detach_destroy_domains and partition_sched_domains. | |
247 | */ | |
248 | static DEFINE_MUTEX(sched_domains_mutex); | |
249 | ||
052f1dc7 | 250 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 251 | |
68318b8e SV |
252 | #include <linux/cgroup.h> |
253 | ||
29f59db3 SV |
254 | struct cfs_rq; |
255 | ||
6f505b16 PZ |
256 | static LIST_HEAD(task_groups); |
257 | ||
29f59db3 | 258 | /* task group related information */ |
4cf86d77 | 259 | struct task_group { |
052f1dc7 | 260 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
261 | struct cgroup_subsys_state css; |
262 | #endif | |
052f1dc7 PZ |
263 | |
264 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
265 | /* schedulable entities of this group on each cpu */ |
266 | struct sched_entity **se; | |
267 | /* runqueue "owned" by this group on each cpu */ | |
268 | struct cfs_rq **cfs_rq; | |
269 | unsigned long shares; | |
052f1dc7 PZ |
270 | #endif |
271 | ||
272 | #ifdef CONFIG_RT_GROUP_SCHED | |
273 | struct sched_rt_entity **rt_se; | |
274 | struct rt_rq **rt_rq; | |
275 | ||
d0b27fa7 | 276 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 277 | #endif |
6b2d7700 | 278 | |
ae8393e5 | 279 | struct rcu_head rcu; |
6f505b16 | 280 | struct list_head list; |
f473aa5e PZ |
281 | |
282 | struct task_group *parent; | |
283 | struct list_head siblings; | |
284 | struct list_head children; | |
29f59db3 SV |
285 | }; |
286 | ||
354d60c2 | 287 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
288 | |
289 | /* | |
290 | * Root task group. | |
291 | * Every UID task group (including init_task_group aka UID-0) will | |
292 | * be a child to this group. | |
293 | */ | |
294 | struct task_group root_task_group; | |
295 | ||
052f1dc7 | 296 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
297 | /* Default task group's sched entity on each cpu */ |
298 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
299 | /* Default task group's cfs_rq on each cpu */ | |
300 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 301 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
302 | |
303 | #ifdef CONFIG_RT_GROUP_SCHED | |
304 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
305 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 306 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 307 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 308 | #define root_task_group init_task_group |
9a7e0b18 | 309 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 310 | |
8ed36996 | 311 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
312 | * a task group's cpu shares. |
313 | */ | |
8ed36996 | 314 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 315 | |
052f1dc7 | 316 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
317 | #ifdef CONFIG_USER_SCHED |
318 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 319 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 320 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 321 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 322 | |
cb4ad1ff | 323 | /* |
2e084786 LJ |
324 | * A weight of 0 or 1 can cause arithmetics problems. |
325 | * A weight of a cfs_rq is the sum of weights of which entities | |
326 | * are queued on this cfs_rq, so a weight of a entity should not be | |
327 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
328 | * (The default weight is 1024 - so there's no practical |
329 | * limitation from this.) | |
330 | */ | |
18d95a28 | 331 | #define MIN_SHARES 2 |
2e084786 | 332 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 333 | |
052f1dc7 PZ |
334 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
335 | #endif | |
336 | ||
29f59db3 | 337 | /* Default task group. |
3a252015 | 338 | * Every task in system belong to this group at bootup. |
29f59db3 | 339 | */ |
434d53b0 | 340 | struct task_group init_task_group; |
29f59db3 SV |
341 | |
342 | /* return group to which a task belongs */ | |
4cf86d77 | 343 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 344 | { |
4cf86d77 | 345 | struct task_group *tg; |
9b5b7751 | 346 | |
052f1dc7 | 347 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 348 | tg = p->user->tg; |
052f1dc7 | 349 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
350 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
351 | struct task_group, css); | |
24e377a8 | 352 | #else |
41a2d6cf | 353 | tg = &init_task_group; |
24e377a8 | 354 | #endif |
9b5b7751 | 355 | return tg; |
29f59db3 SV |
356 | } |
357 | ||
358 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 359 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 360 | { |
052f1dc7 | 361 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
362 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
363 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 364 | #endif |
6f505b16 | 365 | |
052f1dc7 | 366 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
367 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
368 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 369 | #endif |
29f59db3 SV |
370 | } |
371 | ||
372 | #else | |
373 | ||
6f505b16 | 374 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
375 | static inline struct task_group *task_group(struct task_struct *p) |
376 | { | |
377 | return NULL; | |
378 | } | |
29f59db3 | 379 | |
052f1dc7 | 380 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 381 | |
6aa645ea IM |
382 | /* CFS-related fields in a runqueue */ |
383 | struct cfs_rq { | |
384 | struct load_weight load; | |
385 | unsigned long nr_running; | |
386 | ||
6aa645ea | 387 | u64 exec_clock; |
e9acbff6 | 388 | u64 min_vruntime; |
6aa645ea IM |
389 | |
390 | struct rb_root tasks_timeline; | |
391 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
392 | |
393 | struct list_head tasks; | |
394 | struct list_head *balance_iterator; | |
395 | ||
396 | /* | |
397 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
398 | * It is set to NULL otherwise (i.e when none are currently running). |
399 | */ | |
4793241b | 400 | struct sched_entity *curr, *next, *last; |
ddc97297 PZ |
401 | |
402 | unsigned long nr_spread_over; | |
403 | ||
62160e3f | 404 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
405 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
406 | ||
41a2d6cf IM |
407 | /* |
408 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
409 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
410 | * (like users, containers etc.) | |
411 | * | |
412 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
413 | * list is used during load balance. | |
414 | */ | |
41a2d6cf IM |
415 | struct list_head leaf_cfs_rq_list; |
416 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
417 | |
418 | #ifdef CONFIG_SMP | |
c09595f6 | 419 | /* |
c8cba857 | 420 | * the part of load.weight contributed by tasks |
c09595f6 | 421 | */ |
c8cba857 | 422 | unsigned long task_weight; |
c09595f6 | 423 | |
c8cba857 PZ |
424 | /* |
425 | * h_load = weight * f(tg) | |
426 | * | |
427 | * Where f(tg) is the recursive weight fraction assigned to | |
428 | * this group. | |
429 | */ | |
430 | unsigned long h_load; | |
c09595f6 | 431 | |
c8cba857 PZ |
432 | /* |
433 | * this cpu's part of tg->shares | |
434 | */ | |
435 | unsigned long shares; | |
f1d239f7 PZ |
436 | |
437 | /* | |
438 | * load.weight at the time we set shares | |
439 | */ | |
440 | unsigned long rq_weight; | |
c09595f6 | 441 | #endif |
6aa645ea IM |
442 | #endif |
443 | }; | |
1da177e4 | 444 | |
6aa645ea IM |
445 | /* Real-Time classes' related field in a runqueue: */ |
446 | struct rt_rq { | |
447 | struct rt_prio_array active; | |
63489e45 | 448 | unsigned long rt_nr_running; |
052f1dc7 | 449 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
450 | int highest_prio; /* highest queued rt task prio */ |
451 | #endif | |
fa85ae24 | 452 | #ifdef CONFIG_SMP |
73fe6aae | 453 | unsigned long rt_nr_migratory; |
a22d7fc1 | 454 | int overloaded; |
fa85ae24 | 455 | #endif |
6f505b16 | 456 | int rt_throttled; |
fa85ae24 | 457 | u64 rt_time; |
ac086bc2 | 458 | u64 rt_runtime; |
ea736ed5 | 459 | /* Nests inside the rq lock: */ |
ac086bc2 | 460 | spinlock_t rt_runtime_lock; |
6f505b16 | 461 | |
052f1dc7 | 462 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
463 | unsigned long rt_nr_boosted; |
464 | ||
6f505b16 PZ |
465 | struct rq *rq; |
466 | struct list_head leaf_rt_rq_list; | |
467 | struct task_group *tg; | |
468 | struct sched_rt_entity *rt_se; | |
469 | #endif | |
6aa645ea IM |
470 | }; |
471 | ||
57d885fe GH |
472 | #ifdef CONFIG_SMP |
473 | ||
474 | /* | |
475 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
476 | * variables. Each exclusive cpuset essentially defines an island domain by |
477 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
478 | * exclusive cpuset is created, we also create and attach a new root-domain |
479 | * object. | |
480 | * | |
57d885fe GH |
481 | */ |
482 | struct root_domain { | |
483 | atomic_t refcount; | |
484 | cpumask_t span; | |
485 | cpumask_t online; | |
637f5085 | 486 | |
0eab9146 | 487 | /* |
637f5085 GH |
488 | * The "RT overload" flag: it gets set if a CPU has more than |
489 | * one runnable RT task. | |
490 | */ | |
491 | cpumask_t rto_mask; | |
0eab9146 | 492 | atomic_t rto_count; |
6e0534f2 GH |
493 | #ifdef CONFIG_SMP |
494 | struct cpupri cpupri; | |
495 | #endif | |
57d885fe GH |
496 | }; |
497 | ||
dc938520 GH |
498 | /* |
499 | * By default the system creates a single root-domain with all cpus as | |
500 | * members (mimicking the global state we have today). | |
501 | */ | |
57d885fe GH |
502 | static struct root_domain def_root_domain; |
503 | ||
504 | #endif | |
505 | ||
1da177e4 LT |
506 | /* |
507 | * This is the main, per-CPU runqueue data structure. | |
508 | * | |
509 | * Locking rule: those places that want to lock multiple runqueues | |
510 | * (such as the load balancing or the thread migration code), lock | |
511 | * acquire operations must be ordered by ascending &runqueue. | |
512 | */ | |
70b97a7f | 513 | struct rq { |
d8016491 IM |
514 | /* runqueue lock: */ |
515 | spinlock_t lock; | |
1da177e4 LT |
516 | |
517 | /* | |
518 | * nr_running and cpu_load should be in the same cacheline because | |
519 | * remote CPUs use both these fields when doing load calculation. | |
520 | */ | |
521 | unsigned long nr_running; | |
6aa645ea IM |
522 | #define CPU_LOAD_IDX_MAX 5 |
523 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 524 | unsigned char idle_at_tick; |
46cb4b7c | 525 | #ifdef CONFIG_NO_HZ |
15934a37 | 526 | unsigned long last_tick_seen; |
46cb4b7c SS |
527 | unsigned char in_nohz_recently; |
528 | #endif | |
d8016491 IM |
529 | /* capture load from *all* tasks on this cpu: */ |
530 | struct load_weight load; | |
6aa645ea IM |
531 | unsigned long nr_load_updates; |
532 | u64 nr_switches; | |
533 | ||
534 | struct cfs_rq cfs; | |
6f505b16 | 535 | struct rt_rq rt; |
6f505b16 | 536 | |
6aa645ea | 537 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
538 | /* list of leaf cfs_rq on this cpu: */ |
539 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
540 | #endif |
541 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 542 | struct list_head leaf_rt_rq_list; |
1da177e4 | 543 | #endif |
1da177e4 LT |
544 | |
545 | /* | |
546 | * This is part of a global counter where only the total sum | |
547 | * over all CPUs matters. A task can increase this counter on | |
548 | * one CPU and if it got migrated afterwards it may decrease | |
549 | * it on another CPU. Always updated under the runqueue lock: | |
550 | */ | |
551 | unsigned long nr_uninterruptible; | |
552 | ||
36c8b586 | 553 | struct task_struct *curr, *idle; |
c9819f45 | 554 | unsigned long next_balance; |
1da177e4 | 555 | struct mm_struct *prev_mm; |
6aa645ea | 556 | |
3e51f33f | 557 | u64 clock; |
6aa645ea | 558 | |
1da177e4 LT |
559 | atomic_t nr_iowait; |
560 | ||
561 | #ifdef CONFIG_SMP | |
0eab9146 | 562 | struct root_domain *rd; |
1da177e4 LT |
563 | struct sched_domain *sd; |
564 | ||
565 | /* For active balancing */ | |
566 | int active_balance; | |
567 | int push_cpu; | |
d8016491 IM |
568 | /* cpu of this runqueue: */ |
569 | int cpu; | |
1f11eb6a | 570 | int online; |
1da177e4 | 571 | |
a8a51d5e | 572 | unsigned long avg_load_per_task; |
1da177e4 | 573 | |
36c8b586 | 574 | struct task_struct *migration_thread; |
1da177e4 LT |
575 | struct list_head migration_queue; |
576 | #endif | |
577 | ||
8f4d37ec | 578 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
579 | #ifdef CONFIG_SMP |
580 | int hrtick_csd_pending; | |
581 | struct call_single_data hrtick_csd; | |
582 | #endif | |
8f4d37ec PZ |
583 | struct hrtimer hrtick_timer; |
584 | #endif | |
585 | ||
1da177e4 LT |
586 | #ifdef CONFIG_SCHEDSTATS |
587 | /* latency stats */ | |
588 | struct sched_info rq_sched_info; | |
589 | ||
590 | /* sys_sched_yield() stats */ | |
480b9434 KC |
591 | unsigned int yld_exp_empty; |
592 | unsigned int yld_act_empty; | |
593 | unsigned int yld_both_empty; | |
594 | unsigned int yld_count; | |
1da177e4 LT |
595 | |
596 | /* schedule() stats */ | |
480b9434 KC |
597 | unsigned int sched_switch; |
598 | unsigned int sched_count; | |
599 | unsigned int sched_goidle; | |
1da177e4 LT |
600 | |
601 | /* try_to_wake_up() stats */ | |
480b9434 KC |
602 | unsigned int ttwu_count; |
603 | unsigned int ttwu_local; | |
b8efb561 IM |
604 | |
605 | /* BKL stats */ | |
480b9434 | 606 | unsigned int bkl_count; |
1da177e4 LT |
607 | #endif |
608 | }; | |
609 | ||
f34e3b61 | 610 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 611 | |
15afe09b | 612 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 613 | { |
15afe09b | 614 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
615 | } |
616 | ||
0a2966b4 CL |
617 | static inline int cpu_of(struct rq *rq) |
618 | { | |
619 | #ifdef CONFIG_SMP | |
620 | return rq->cpu; | |
621 | #else | |
622 | return 0; | |
623 | #endif | |
624 | } | |
625 | ||
674311d5 NP |
626 | /* |
627 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 628 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
629 | * |
630 | * The domain tree of any CPU may only be accessed from within | |
631 | * preempt-disabled sections. | |
632 | */ | |
48f24c4d IM |
633 | #define for_each_domain(cpu, __sd) \ |
634 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
635 | |
636 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
637 | #define this_rq() (&__get_cpu_var(runqueues)) | |
638 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
639 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
640 | ||
3e51f33f PZ |
641 | static inline void update_rq_clock(struct rq *rq) |
642 | { | |
643 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
644 | } | |
645 | ||
bf5c91ba IM |
646 | /* |
647 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
648 | */ | |
649 | #ifdef CONFIG_SCHED_DEBUG | |
650 | # define const_debug __read_mostly | |
651 | #else | |
652 | # define const_debug static const | |
653 | #endif | |
654 | ||
017730c1 IM |
655 | /** |
656 | * runqueue_is_locked | |
657 | * | |
658 | * Returns true if the current cpu runqueue is locked. | |
659 | * This interface allows printk to be called with the runqueue lock | |
660 | * held and know whether or not it is OK to wake up the klogd. | |
661 | */ | |
662 | int runqueue_is_locked(void) | |
663 | { | |
664 | int cpu = get_cpu(); | |
665 | struct rq *rq = cpu_rq(cpu); | |
666 | int ret; | |
667 | ||
668 | ret = spin_is_locked(&rq->lock); | |
669 | put_cpu(); | |
670 | return ret; | |
671 | } | |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Debugging: various feature bits | |
675 | */ | |
f00b45c1 PZ |
676 | |
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | __SCHED_FEAT_##name , | |
679 | ||
bf5c91ba | 680 | enum { |
f00b45c1 | 681 | #include "sched_features.h" |
bf5c91ba IM |
682 | }; |
683 | ||
f00b45c1 PZ |
684 | #undef SCHED_FEAT |
685 | ||
686 | #define SCHED_FEAT(name, enabled) \ | |
687 | (1UL << __SCHED_FEAT_##name) * enabled | | |
688 | ||
bf5c91ba | 689 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
690 | #include "sched_features.h" |
691 | 0; | |
692 | ||
693 | #undef SCHED_FEAT | |
694 | ||
695 | #ifdef CONFIG_SCHED_DEBUG | |
696 | #define SCHED_FEAT(name, enabled) \ | |
697 | #name , | |
698 | ||
983ed7a6 | 699 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
700 | #include "sched_features.h" |
701 | NULL | |
702 | }; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
983ed7a6 | 706 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
707 | { |
708 | filp->private_data = inode->i_private; | |
709 | return 0; | |
710 | } | |
711 | ||
712 | static ssize_t | |
713 | sched_feat_read(struct file *filp, char __user *ubuf, | |
714 | size_t cnt, loff_t *ppos) | |
715 | { | |
716 | char *buf; | |
717 | int r = 0; | |
718 | int len = 0; | |
719 | int i; | |
720 | ||
721 | for (i = 0; sched_feat_names[i]; i++) { | |
722 | len += strlen(sched_feat_names[i]); | |
723 | len += 4; | |
724 | } | |
725 | ||
726 | buf = kmalloc(len + 2, GFP_KERNEL); | |
727 | if (!buf) | |
728 | return -ENOMEM; | |
729 | ||
730 | for (i = 0; sched_feat_names[i]; i++) { | |
731 | if (sysctl_sched_features & (1UL << i)) | |
732 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
733 | else | |
c24b7c52 | 734 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
735 | } |
736 | ||
737 | r += sprintf(buf + r, "\n"); | |
738 | WARN_ON(r >= len + 2); | |
739 | ||
740 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
741 | ||
742 | kfree(buf); | |
743 | ||
744 | return r; | |
745 | } | |
746 | ||
747 | static ssize_t | |
748 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
749 | size_t cnt, loff_t *ppos) | |
750 | { | |
751 | char buf[64]; | |
752 | char *cmp = buf; | |
753 | int neg = 0; | |
754 | int i; | |
755 | ||
756 | if (cnt > 63) | |
757 | cnt = 63; | |
758 | ||
759 | if (copy_from_user(&buf, ubuf, cnt)) | |
760 | return -EFAULT; | |
761 | ||
762 | buf[cnt] = 0; | |
763 | ||
c24b7c52 | 764 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
765 | neg = 1; |
766 | cmp += 3; | |
767 | } | |
768 | ||
769 | for (i = 0; sched_feat_names[i]; i++) { | |
770 | int len = strlen(sched_feat_names[i]); | |
771 | ||
772 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
773 | if (neg) | |
774 | sysctl_sched_features &= ~(1UL << i); | |
775 | else | |
776 | sysctl_sched_features |= (1UL << i); | |
777 | break; | |
778 | } | |
779 | } | |
780 | ||
781 | if (!sched_feat_names[i]) | |
782 | return -EINVAL; | |
783 | ||
784 | filp->f_pos += cnt; | |
785 | ||
786 | return cnt; | |
787 | } | |
788 | ||
789 | static struct file_operations sched_feat_fops = { | |
790 | .open = sched_feat_open, | |
791 | .read = sched_feat_read, | |
792 | .write = sched_feat_write, | |
793 | }; | |
794 | ||
795 | static __init int sched_init_debug(void) | |
796 | { | |
f00b45c1 PZ |
797 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
798 | &sched_feat_fops); | |
799 | ||
800 | return 0; | |
801 | } | |
802 | late_initcall(sched_init_debug); | |
803 | ||
804 | #endif | |
805 | ||
806 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 807 | |
b82d9fdd PZ |
808 | /* |
809 | * Number of tasks to iterate in a single balance run. | |
810 | * Limited because this is done with IRQs disabled. | |
811 | */ | |
812 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
813 | ||
2398f2c6 PZ |
814 | /* |
815 | * ratelimit for updating the group shares. | |
55cd5340 | 816 | * default: 0.25ms |
2398f2c6 | 817 | */ |
55cd5340 | 818 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 819 | |
ffda12a1 PZ |
820 | /* |
821 | * Inject some fuzzyness into changing the per-cpu group shares | |
822 | * this avoids remote rq-locks at the expense of fairness. | |
823 | * default: 4 | |
824 | */ | |
825 | unsigned int sysctl_sched_shares_thresh = 4; | |
826 | ||
fa85ae24 | 827 | /* |
9f0c1e56 | 828 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
829 | * default: 1s |
830 | */ | |
9f0c1e56 | 831 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 832 | |
6892b75e IM |
833 | static __read_mostly int scheduler_running; |
834 | ||
9f0c1e56 PZ |
835 | /* |
836 | * part of the period that we allow rt tasks to run in us. | |
837 | * default: 0.95s | |
838 | */ | |
839 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 840 | |
d0b27fa7 PZ |
841 | static inline u64 global_rt_period(void) |
842 | { | |
843 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
844 | } | |
845 | ||
846 | static inline u64 global_rt_runtime(void) | |
847 | { | |
e26873bb | 848 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
849 | return RUNTIME_INF; |
850 | ||
851 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
852 | } | |
fa85ae24 | 853 | |
1da177e4 | 854 | #ifndef prepare_arch_switch |
4866cde0 NP |
855 | # define prepare_arch_switch(next) do { } while (0) |
856 | #endif | |
857 | #ifndef finish_arch_switch | |
858 | # define finish_arch_switch(prev) do { } while (0) | |
859 | #endif | |
860 | ||
051a1d1a DA |
861 | static inline int task_current(struct rq *rq, struct task_struct *p) |
862 | { | |
863 | return rq->curr == p; | |
864 | } | |
865 | ||
4866cde0 | 866 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 867 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 868 | { |
051a1d1a | 869 | return task_current(rq, p); |
4866cde0 NP |
870 | } |
871 | ||
70b97a7f | 872 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
873 | { |
874 | } | |
875 | ||
70b97a7f | 876 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 877 | { |
da04c035 IM |
878 | #ifdef CONFIG_DEBUG_SPINLOCK |
879 | /* this is a valid case when another task releases the spinlock */ | |
880 | rq->lock.owner = current; | |
881 | #endif | |
8a25d5de IM |
882 | /* |
883 | * If we are tracking spinlock dependencies then we have to | |
884 | * fix up the runqueue lock - which gets 'carried over' from | |
885 | * prev into current: | |
886 | */ | |
887 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
888 | ||
4866cde0 NP |
889 | spin_unlock_irq(&rq->lock); |
890 | } | |
891 | ||
892 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 893 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
894 | { |
895 | #ifdef CONFIG_SMP | |
896 | return p->oncpu; | |
897 | #else | |
051a1d1a | 898 | return task_current(rq, p); |
4866cde0 NP |
899 | #endif |
900 | } | |
901 | ||
70b97a7f | 902 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
903 | { |
904 | #ifdef CONFIG_SMP | |
905 | /* | |
906 | * We can optimise this out completely for !SMP, because the | |
907 | * SMP rebalancing from interrupt is the only thing that cares | |
908 | * here. | |
909 | */ | |
910 | next->oncpu = 1; | |
911 | #endif | |
912 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
913 | spin_unlock_irq(&rq->lock); | |
914 | #else | |
915 | spin_unlock(&rq->lock); | |
916 | #endif | |
917 | } | |
918 | ||
70b97a7f | 919 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
920 | { |
921 | #ifdef CONFIG_SMP | |
922 | /* | |
923 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
924 | * We must ensure this doesn't happen until the switch is completely | |
925 | * finished. | |
926 | */ | |
927 | smp_wmb(); | |
928 | prev->oncpu = 0; | |
929 | #endif | |
930 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
931 | local_irq_enable(); | |
1da177e4 | 932 | #endif |
4866cde0 NP |
933 | } |
934 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 935 | |
b29739f9 IM |
936 | /* |
937 | * __task_rq_lock - lock the runqueue a given task resides on. | |
938 | * Must be called interrupts disabled. | |
939 | */ | |
70b97a7f | 940 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
941 | __acquires(rq->lock) |
942 | { | |
3a5c359a AK |
943 | for (;;) { |
944 | struct rq *rq = task_rq(p); | |
945 | spin_lock(&rq->lock); | |
946 | if (likely(rq == task_rq(p))) | |
947 | return rq; | |
b29739f9 | 948 | spin_unlock(&rq->lock); |
b29739f9 | 949 | } |
b29739f9 IM |
950 | } |
951 | ||
1da177e4 LT |
952 | /* |
953 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 954 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
955 | * explicitly disabling preemption. |
956 | */ | |
70b97a7f | 957 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
958 | __acquires(rq->lock) |
959 | { | |
70b97a7f | 960 | struct rq *rq; |
1da177e4 | 961 | |
3a5c359a AK |
962 | for (;;) { |
963 | local_irq_save(*flags); | |
964 | rq = task_rq(p); | |
965 | spin_lock(&rq->lock); | |
966 | if (likely(rq == task_rq(p))) | |
967 | return rq; | |
1da177e4 | 968 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 969 | } |
1da177e4 LT |
970 | } |
971 | ||
a9957449 | 972 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
973 | __releases(rq->lock) |
974 | { | |
975 | spin_unlock(&rq->lock); | |
976 | } | |
977 | ||
70b97a7f | 978 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
979 | __releases(rq->lock) |
980 | { | |
981 | spin_unlock_irqrestore(&rq->lock, *flags); | |
982 | } | |
983 | ||
1da177e4 | 984 | /* |
cc2a73b5 | 985 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 986 | */ |
a9957449 | 987 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
988 | __acquires(rq->lock) |
989 | { | |
70b97a7f | 990 | struct rq *rq; |
1da177e4 LT |
991 | |
992 | local_irq_disable(); | |
993 | rq = this_rq(); | |
994 | spin_lock(&rq->lock); | |
995 | ||
996 | return rq; | |
997 | } | |
998 | ||
8f4d37ec PZ |
999 | #ifdef CONFIG_SCHED_HRTICK |
1000 | /* | |
1001 | * Use HR-timers to deliver accurate preemption points. | |
1002 | * | |
1003 | * Its all a bit involved since we cannot program an hrt while holding the | |
1004 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1005 | * reschedule event. | |
1006 | * | |
1007 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1008 | * rq->lock. | |
1009 | */ | |
8f4d37ec PZ |
1010 | |
1011 | /* | |
1012 | * Use hrtick when: | |
1013 | * - enabled by features | |
1014 | * - hrtimer is actually high res | |
1015 | */ | |
1016 | static inline int hrtick_enabled(struct rq *rq) | |
1017 | { | |
1018 | if (!sched_feat(HRTICK)) | |
1019 | return 0; | |
ba42059f | 1020 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1021 | return 0; |
8f4d37ec PZ |
1022 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1023 | } | |
1024 | ||
8f4d37ec PZ |
1025 | static void hrtick_clear(struct rq *rq) |
1026 | { | |
1027 | if (hrtimer_active(&rq->hrtick_timer)) | |
1028 | hrtimer_cancel(&rq->hrtick_timer); | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | /* |
1032 | * High-resolution timer tick. | |
1033 | * Runs from hardirq context with interrupts disabled. | |
1034 | */ | |
1035 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1036 | { | |
1037 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1038 | ||
1039 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1040 | ||
1041 | spin_lock(&rq->lock); | |
3e51f33f | 1042 | update_rq_clock(rq); |
8f4d37ec PZ |
1043 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1044 | spin_unlock(&rq->lock); | |
1045 | ||
1046 | return HRTIMER_NORESTART; | |
1047 | } | |
1048 | ||
95e904c7 | 1049 | #ifdef CONFIG_SMP |
31656519 PZ |
1050 | /* |
1051 | * called from hardirq (IPI) context | |
1052 | */ | |
1053 | static void __hrtick_start(void *arg) | |
b328ca18 | 1054 | { |
31656519 | 1055 | struct rq *rq = arg; |
b328ca18 | 1056 | |
31656519 PZ |
1057 | spin_lock(&rq->lock); |
1058 | hrtimer_restart(&rq->hrtick_timer); | |
1059 | rq->hrtick_csd_pending = 0; | |
1060 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1061 | } |
1062 | ||
31656519 PZ |
1063 | /* |
1064 | * Called to set the hrtick timer state. | |
1065 | * | |
1066 | * called with rq->lock held and irqs disabled | |
1067 | */ | |
1068 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1069 | { |
31656519 PZ |
1070 | struct hrtimer *timer = &rq->hrtick_timer; |
1071 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1072 | |
cc584b21 | 1073 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1074 | |
1075 | if (rq == this_rq()) { | |
1076 | hrtimer_restart(timer); | |
1077 | } else if (!rq->hrtick_csd_pending) { | |
1078 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1079 | rq->hrtick_csd_pending = 1; | |
1080 | } | |
b328ca18 PZ |
1081 | } |
1082 | ||
1083 | static int | |
1084 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1085 | { | |
1086 | int cpu = (int)(long)hcpu; | |
1087 | ||
1088 | switch (action) { | |
1089 | case CPU_UP_CANCELED: | |
1090 | case CPU_UP_CANCELED_FROZEN: | |
1091 | case CPU_DOWN_PREPARE: | |
1092 | case CPU_DOWN_PREPARE_FROZEN: | |
1093 | case CPU_DEAD: | |
1094 | case CPU_DEAD_FROZEN: | |
31656519 | 1095 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1096 | return NOTIFY_OK; |
1097 | } | |
1098 | ||
1099 | return NOTIFY_DONE; | |
1100 | } | |
1101 | ||
fa748203 | 1102 | static __init void init_hrtick(void) |
b328ca18 PZ |
1103 | { |
1104 | hotcpu_notifier(hotplug_hrtick, 0); | |
1105 | } | |
31656519 PZ |
1106 | #else |
1107 | /* | |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
1113 | { | |
1114 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1115 | } | |
b328ca18 | 1116 | |
006c75f1 | 1117 | static inline void init_hrtick(void) |
8f4d37ec | 1118 | { |
8f4d37ec | 1119 | } |
31656519 | 1120 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1121 | |
31656519 | 1122 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1123 | { |
31656519 PZ |
1124 | #ifdef CONFIG_SMP |
1125 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1126 | |
31656519 PZ |
1127 | rq->hrtick_csd.flags = 0; |
1128 | rq->hrtick_csd.func = __hrtick_start; | |
1129 | rq->hrtick_csd.info = rq; | |
1130 | #endif | |
8f4d37ec | 1131 | |
31656519 PZ |
1132 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1133 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1134 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1135 | } |
006c75f1 | 1136 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1137 | static inline void hrtick_clear(struct rq *rq) |
1138 | { | |
1139 | } | |
1140 | ||
8f4d37ec PZ |
1141 | static inline void init_rq_hrtick(struct rq *rq) |
1142 | { | |
1143 | } | |
1144 | ||
b328ca18 PZ |
1145 | static inline void init_hrtick(void) |
1146 | { | |
1147 | } | |
006c75f1 | 1148 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1149 | |
c24d20db IM |
1150 | /* |
1151 | * resched_task - mark a task 'to be rescheduled now'. | |
1152 | * | |
1153 | * On UP this means the setting of the need_resched flag, on SMP it | |
1154 | * might also involve a cross-CPU call to trigger the scheduler on | |
1155 | * the target CPU. | |
1156 | */ | |
1157 | #ifdef CONFIG_SMP | |
1158 | ||
1159 | #ifndef tsk_is_polling | |
1160 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1161 | #endif | |
1162 | ||
31656519 | 1163 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1164 | { |
1165 | int cpu; | |
1166 | ||
1167 | assert_spin_locked(&task_rq(p)->lock); | |
1168 | ||
31656519 | 1169 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1170 | return; |
1171 | ||
31656519 | 1172 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1173 | |
1174 | cpu = task_cpu(p); | |
1175 | if (cpu == smp_processor_id()) | |
1176 | return; | |
1177 | ||
1178 | /* NEED_RESCHED must be visible before we test polling */ | |
1179 | smp_mb(); | |
1180 | if (!tsk_is_polling(p)) | |
1181 | smp_send_reschedule(cpu); | |
1182 | } | |
1183 | ||
1184 | static void resched_cpu(int cpu) | |
1185 | { | |
1186 | struct rq *rq = cpu_rq(cpu); | |
1187 | unsigned long flags; | |
1188 | ||
1189 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1190 | return; | |
1191 | resched_task(cpu_curr(cpu)); | |
1192 | spin_unlock_irqrestore(&rq->lock, flags); | |
1193 | } | |
06d8308c TG |
1194 | |
1195 | #ifdef CONFIG_NO_HZ | |
1196 | /* | |
1197 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1198 | * idle CPU then this timer might expire before the next timer event | |
1199 | * which is scheduled to wake up that CPU. In case of a completely | |
1200 | * idle system the next event might even be infinite time into the | |
1201 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1202 | * leaves the inner idle loop so the newly added timer is taken into | |
1203 | * account when the CPU goes back to idle and evaluates the timer | |
1204 | * wheel for the next timer event. | |
1205 | */ | |
1206 | void wake_up_idle_cpu(int cpu) | |
1207 | { | |
1208 | struct rq *rq = cpu_rq(cpu); | |
1209 | ||
1210 | if (cpu == smp_processor_id()) | |
1211 | return; | |
1212 | ||
1213 | /* | |
1214 | * This is safe, as this function is called with the timer | |
1215 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1216 | * to idle and has not yet set rq->curr to idle then it will | |
1217 | * be serialized on the timer wheel base lock and take the new | |
1218 | * timer into account automatically. | |
1219 | */ | |
1220 | if (rq->curr != rq->idle) | |
1221 | return; | |
1222 | ||
1223 | /* | |
1224 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1225 | * lockless. The worst case is that the other CPU runs the | |
1226 | * idle task through an additional NOOP schedule() | |
1227 | */ | |
1228 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1229 | ||
1230 | /* NEED_RESCHED must be visible before we test polling */ | |
1231 | smp_mb(); | |
1232 | if (!tsk_is_polling(rq->idle)) | |
1233 | smp_send_reschedule(cpu); | |
1234 | } | |
6d6bc0ad | 1235 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1236 | |
6d6bc0ad | 1237 | #else /* !CONFIG_SMP */ |
31656519 | 1238 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1239 | { |
1240 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1241 | set_tsk_need_resched(p); |
c24d20db | 1242 | } |
6d6bc0ad | 1243 | #endif /* CONFIG_SMP */ |
c24d20db | 1244 | |
45bf76df IM |
1245 | #if BITS_PER_LONG == 32 |
1246 | # define WMULT_CONST (~0UL) | |
1247 | #else | |
1248 | # define WMULT_CONST (1UL << 32) | |
1249 | #endif | |
1250 | ||
1251 | #define WMULT_SHIFT 32 | |
1252 | ||
194081eb IM |
1253 | /* |
1254 | * Shift right and round: | |
1255 | */ | |
cf2ab469 | 1256 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1257 | |
a7be37ac PZ |
1258 | /* |
1259 | * delta *= weight / lw | |
1260 | */ | |
cb1c4fc9 | 1261 | static unsigned long |
45bf76df IM |
1262 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1263 | struct load_weight *lw) | |
1264 | { | |
1265 | u64 tmp; | |
1266 | ||
7a232e03 LJ |
1267 | if (!lw->inv_weight) { |
1268 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1269 | lw->inv_weight = 1; | |
1270 | else | |
1271 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1272 | / (lw->weight+1); | |
1273 | } | |
45bf76df IM |
1274 | |
1275 | tmp = (u64)delta_exec * weight; | |
1276 | /* | |
1277 | * Check whether we'd overflow the 64-bit multiplication: | |
1278 | */ | |
194081eb | 1279 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1280 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1281 | WMULT_SHIFT/2); |
1282 | else | |
cf2ab469 | 1283 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1284 | |
ecf691da | 1285 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1286 | } |
1287 | ||
1091985b | 1288 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1289 | { |
1290 | lw->weight += inc; | |
e89996ae | 1291 | lw->inv_weight = 0; |
45bf76df IM |
1292 | } |
1293 | ||
1091985b | 1294 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1295 | { |
1296 | lw->weight -= dec; | |
e89996ae | 1297 | lw->inv_weight = 0; |
45bf76df IM |
1298 | } |
1299 | ||
2dd73a4f PW |
1300 | /* |
1301 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1302 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1303 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1304 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1305 | * scaled version of the new time slice allocation that they receive on time |
1306 | * slice expiry etc. | |
1307 | */ | |
1308 | ||
dd41f596 IM |
1309 | #define WEIGHT_IDLEPRIO 2 |
1310 | #define WMULT_IDLEPRIO (1 << 31) | |
1311 | ||
1312 | /* | |
1313 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1314 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1315 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1316 | * that remained on nice 0. | |
1317 | * | |
1318 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1319 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1320 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1321 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1322 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1323 | */ |
1324 | static const int prio_to_weight[40] = { | |
254753dc IM |
1325 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1326 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1327 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1328 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1329 | /* 0 */ 1024, 820, 655, 526, 423, | |
1330 | /* 5 */ 335, 272, 215, 172, 137, | |
1331 | /* 10 */ 110, 87, 70, 56, 45, | |
1332 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1333 | }; |
1334 | ||
5714d2de IM |
1335 | /* |
1336 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1337 | * | |
1338 | * In cases where the weight does not change often, we can use the | |
1339 | * precalculated inverse to speed up arithmetics by turning divisions | |
1340 | * into multiplications: | |
1341 | */ | |
dd41f596 | 1342 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1343 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1344 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1345 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1346 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1347 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1348 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1349 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1350 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1351 | }; |
2dd73a4f | 1352 | |
dd41f596 IM |
1353 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1354 | ||
1355 | /* | |
1356 | * runqueue iterator, to support SMP load-balancing between different | |
1357 | * scheduling classes, without having to expose their internal data | |
1358 | * structures to the load-balancing proper: | |
1359 | */ | |
1360 | struct rq_iterator { | |
1361 | void *arg; | |
1362 | struct task_struct *(*start)(void *); | |
1363 | struct task_struct *(*next)(void *); | |
1364 | }; | |
1365 | ||
e1d1484f PW |
1366 | #ifdef CONFIG_SMP |
1367 | static unsigned long | |
1368 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1369 | unsigned long max_load_move, struct sched_domain *sd, | |
1370 | enum cpu_idle_type idle, int *all_pinned, | |
1371 | int *this_best_prio, struct rq_iterator *iterator); | |
1372 | ||
1373 | static int | |
1374 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1375 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1376 | struct rq_iterator *iterator); | |
e1d1484f | 1377 | #endif |
dd41f596 | 1378 | |
d842de87 SV |
1379 | #ifdef CONFIG_CGROUP_CPUACCT |
1380 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1381 | #else | |
1382 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1383 | #endif | |
1384 | ||
18d95a28 PZ |
1385 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1386 | { | |
1387 | update_load_add(&rq->load, load); | |
1388 | } | |
1389 | ||
1390 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1391 | { | |
1392 | update_load_sub(&rq->load, load); | |
1393 | } | |
1394 | ||
7940ca36 | 1395 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1396 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1397 | |
1398 | /* | |
1399 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1400 | * leaving it for the final time. | |
1401 | */ | |
eb755805 | 1402 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1403 | { |
1404 | struct task_group *parent, *child; | |
eb755805 | 1405 | int ret; |
c09595f6 PZ |
1406 | |
1407 | rcu_read_lock(); | |
1408 | parent = &root_task_group; | |
1409 | down: | |
eb755805 PZ |
1410 | ret = (*down)(parent, data); |
1411 | if (ret) | |
1412 | goto out_unlock; | |
c09595f6 PZ |
1413 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1414 | parent = child; | |
1415 | goto down; | |
1416 | ||
1417 | up: | |
1418 | continue; | |
1419 | } | |
eb755805 PZ |
1420 | ret = (*up)(parent, data); |
1421 | if (ret) | |
1422 | goto out_unlock; | |
c09595f6 PZ |
1423 | |
1424 | child = parent; | |
1425 | parent = parent->parent; | |
1426 | if (parent) | |
1427 | goto up; | |
eb755805 | 1428 | out_unlock: |
c09595f6 | 1429 | rcu_read_unlock(); |
eb755805 PZ |
1430 | |
1431 | return ret; | |
c09595f6 PZ |
1432 | } |
1433 | ||
eb755805 PZ |
1434 | static int tg_nop(struct task_group *tg, void *data) |
1435 | { | |
1436 | return 0; | |
c09595f6 | 1437 | } |
eb755805 PZ |
1438 | #endif |
1439 | ||
1440 | #ifdef CONFIG_SMP | |
1441 | static unsigned long source_load(int cpu, int type); | |
1442 | static unsigned long target_load(int cpu, int type); | |
1443 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1444 | ||
1445 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1446 | { | |
1447 | struct rq *rq = cpu_rq(cpu); | |
1448 | ||
1449 | if (rq->nr_running) | |
1450 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1451 | ||
1452 | return rq->avg_load_per_task; | |
1453 | } | |
1454 | ||
1455 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1456 | |
c09595f6 PZ |
1457 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1458 | ||
1459 | /* | |
1460 | * Calculate and set the cpu's group shares. | |
1461 | */ | |
1462 | static void | |
ffda12a1 PZ |
1463 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1464 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1465 | { |
c09595f6 PZ |
1466 | int boost = 0; |
1467 | unsigned long shares; | |
1468 | unsigned long rq_weight; | |
1469 | ||
c8cba857 | 1470 | if (!tg->se[cpu]) |
c09595f6 PZ |
1471 | return; |
1472 | ||
c8cba857 | 1473 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1474 | |
1475 | /* | |
1476 | * If there are currently no tasks on the cpu pretend there is one of | |
1477 | * average load so that when a new task gets to run here it will not | |
1478 | * get delayed by group starvation. | |
1479 | */ | |
1480 | if (!rq_weight) { | |
1481 | boost = 1; | |
1482 | rq_weight = NICE_0_LOAD; | |
1483 | } | |
1484 | ||
c8cba857 PZ |
1485 | if (unlikely(rq_weight > sd_rq_weight)) |
1486 | rq_weight = sd_rq_weight; | |
1487 | ||
c09595f6 PZ |
1488 | /* |
1489 | * \Sum shares * rq_weight | |
1490 | * shares = ----------------------- | |
1491 | * \Sum rq_weight | |
1492 | * | |
1493 | */ | |
c8cba857 | 1494 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
ffda12a1 | 1495 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1496 | |
ffda12a1 PZ |
1497 | if (abs(shares - tg->se[cpu]->load.weight) > |
1498 | sysctl_sched_shares_thresh) { | |
1499 | struct rq *rq = cpu_rq(cpu); | |
1500 | unsigned long flags; | |
c09595f6 | 1501 | |
ffda12a1 PZ |
1502 | spin_lock_irqsave(&rq->lock, flags); |
1503 | /* | |
1504 | * record the actual number of shares, not the boosted amount. | |
1505 | */ | |
1506 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; | |
1507 | tg->cfs_rq[cpu]->rq_weight = rq_weight; | |
c09595f6 | 1508 | |
ffda12a1 PZ |
1509 | __set_se_shares(tg->se[cpu], shares); |
1510 | spin_unlock_irqrestore(&rq->lock, flags); | |
1511 | } | |
18d95a28 | 1512 | } |
c09595f6 PZ |
1513 | |
1514 | /* | |
c8cba857 PZ |
1515 | * Re-compute the task group their per cpu shares over the given domain. |
1516 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1517 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1518 | */ |
eb755805 | 1519 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1520 | { |
c8cba857 PZ |
1521 | unsigned long rq_weight = 0; |
1522 | unsigned long shares = 0; | |
eb755805 | 1523 | struct sched_domain *sd = data; |
c8cba857 | 1524 | int i; |
c09595f6 | 1525 | |
c8cba857 PZ |
1526 | for_each_cpu_mask(i, sd->span) { |
1527 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1528 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1529 | } |
c09595f6 | 1530 | |
c8cba857 PZ |
1531 | if ((!shares && rq_weight) || shares > tg->shares) |
1532 | shares = tg->shares; | |
1533 | ||
1534 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1535 | shares = tg->shares; | |
c09595f6 | 1536 | |
cd80917e PZ |
1537 | if (!rq_weight) |
1538 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1539 | ||
ffda12a1 PZ |
1540 | for_each_cpu_mask(i, sd->span) |
1541 | update_group_shares_cpu(tg, i, shares, rq_weight); | |
eb755805 PZ |
1542 | |
1543 | return 0; | |
c09595f6 PZ |
1544 | } |
1545 | ||
1546 | /* | |
c8cba857 PZ |
1547 | * Compute the cpu's hierarchical load factor for each task group. |
1548 | * This needs to be done in a top-down fashion because the load of a child | |
1549 | * group is a fraction of its parents load. | |
c09595f6 | 1550 | */ |
eb755805 | 1551 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1552 | { |
c8cba857 | 1553 | unsigned long load; |
eb755805 | 1554 | long cpu = (long)data; |
c09595f6 | 1555 | |
c8cba857 PZ |
1556 | if (!tg->parent) { |
1557 | load = cpu_rq(cpu)->load.weight; | |
1558 | } else { | |
1559 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1560 | load *= tg->cfs_rq[cpu]->shares; | |
1561 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1562 | } | |
c09595f6 | 1563 | |
c8cba857 | 1564 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1565 | |
eb755805 | 1566 | return 0; |
c09595f6 PZ |
1567 | } |
1568 | ||
c8cba857 | 1569 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1570 | { |
2398f2c6 PZ |
1571 | u64 now = cpu_clock(raw_smp_processor_id()); |
1572 | s64 elapsed = now - sd->last_update; | |
1573 | ||
1574 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1575 | sd->last_update = now; | |
eb755805 | 1576 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1577 | } |
4d8d595d PZ |
1578 | } |
1579 | ||
3e5459b4 PZ |
1580 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1581 | { | |
1582 | spin_unlock(&rq->lock); | |
1583 | update_shares(sd); | |
1584 | spin_lock(&rq->lock); | |
1585 | } | |
1586 | ||
eb755805 | 1587 | static void update_h_load(long cpu) |
c09595f6 | 1588 | { |
eb755805 | 1589 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1590 | } |
1591 | ||
c09595f6 PZ |
1592 | #else |
1593 | ||
c8cba857 | 1594 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1595 | { |
1596 | } | |
1597 | ||
3e5459b4 PZ |
1598 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1599 | { | |
1600 | } | |
1601 | ||
18d95a28 PZ |
1602 | #endif |
1603 | ||
18d95a28 PZ |
1604 | #endif |
1605 | ||
30432094 | 1606 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1607 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1608 | { | |
30432094 | 1609 | #ifdef CONFIG_SMP |
34e83e85 IM |
1610 | cfs_rq->shares = shares; |
1611 | #endif | |
1612 | } | |
30432094 | 1613 | #endif |
e7693a36 | 1614 | |
dd41f596 | 1615 | #include "sched_stats.h" |
dd41f596 | 1616 | #include "sched_idletask.c" |
5522d5d5 IM |
1617 | #include "sched_fair.c" |
1618 | #include "sched_rt.c" | |
dd41f596 IM |
1619 | #ifdef CONFIG_SCHED_DEBUG |
1620 | # include "sched_debug.c" | |
1621 | #endif | |
1622 | ||
1623 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1624 | #define for_each_class(class) \ |
1625 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1626 | |
c09595f6 | 1627 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1628 | { |
1629 | rq->nr_running++; | |
9c217245 IM |
1630 | } |
1631 | ||
c09595f6 | 1632 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1633 | { |
1634 | rq->nr_running--; | |
9c217245 IM |
1635 | } |
1636 | ||
45bf76df IM |
1637 | static void set_load_weight(struct task_struct *p) |
1638 | { | |
1639 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1640 | p->se.load.weight = prio_to_weight[0] * 2; |
1641 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1642 | return; | |
1643 | } | |
45bf76df | 1644 | |
dd41f596 IM |
1645 | /* |
1646 | * SCHED_IDLE tasks get minimal weight: | |
1647 | */ | |
1648 | if (p->policy == SCHED_IDLE) { | |
1649 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1650 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1651 | return; | |
1652 | } | |
71f8bd46 | 1653 | |
dd41f596 IM |
1654 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1655 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1656 | } |
1657 | ||
2087a1ad GH |
1658 | static void update_avg(u64 *avg, u64 sample) |
1659 | { | |
1660 | s64 diff = sample - *avg; | |
1661 | *avg += diff >> 3; | |
1662 | } | |
1663 | ||
8159f87e | 1664 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1665 | { |
dd41f596 | 1666 | sched_info_queued(p); |
fd390f6a | 1667 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1668 | p->se.on_rq = 1; |
71f8bd46 IM |
1669 | } |
1670 | ||
69be72c1 | 1671 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1672 | { |
2087a1ad GH |
1673 | if (sleep && p->se.last_wakeup) { |
1674 | update_avg(&p->se.avg_overlap, | |
1675 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1676 | p->se.last_wakeup = 0; | |
1677 | } | |
1678 | ||
46ac22ba | 1679 | sched_info_dequeued(p); |
f02231e5 | 1680 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1681 | p->se.on_rq = 0; |
71f8bd46 IM |
1682 | } |
1683 | ||
14531189 | 1684 | /* |
dd41f596 | 1685 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1686 | */ |
14531189 IM |
1687 | static inline int __normal_prio(struct task_struct *p) |
1688 | { | |
dd41f596 | 1689 | return p->static_prio; |
14531189 IM |
1690 | } |
1691 | ||
b29739f9 IM |
1692 | /* |
1693 | * Calculate the expected normal priority: i.e. priority | |
1694 | * without taking RT-inheritance into account. Might be | |
1695 | * boosted by interactivity modifiers. Changes upon fork, | |
1696 | * setprio syscalls, and whenever the interactivity | |
1697 | * estimator recalculates. | |
1698 | */ | |
36c8b586 | 1699 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1700 | { |
1701 | int prio; | |
1702 | ||
e05606d3 | 1703 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1704 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1705 | else | |
1706 | prio = __normal_prio(p); | |
1707 | return prio; | |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * Calculate the current priority, i.e. the priority | |
1712 | * taken into account by the scheduler. This value might | |
1713 | * be boosted by RT tasks, or might be boosted by | |
1714 | * interactivity modifiers. Will be RT if the task got | |
1715 | * RT-boosted. If not then it returns p->normal_prio. | |
1716 | */ | |
36c8b586 | 1717 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1718 | { |
1719 | p->normal_prio = normal_prio(p); | |
1720 | /* | |
1721 | * If we are RT tasks or we were boosted to RT priority, | |
1722 | * keep the priority unchanged. Otherwise, update priority | |
1723 | * to the normal priority: | |
1724 | */ | |
1725 | if (!rt_prio(p->prio)) | |
1726 | return p->normal_prio; | |
1727 | return p->prio; | |
1728 | } | |
1729 | ||
1da177e4 | 1730 | /* |
dd41f596 | 1731 | * activate_task - move a task to the runqueue. |
1da177e4 | 1732 | */ |
dd41f596 | 1733 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1734 | { |
d9514f6c | 1735 | if (task_contributes_to_load(p)) |
dd41f596 | 1736 | rq->nr_uninterruptible--; |
1da177e4 | 1737 | |
8159f87e | 1738 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1739 | inc_nr_running(rq); |
1da177e4 LT |
1740 | } |
1741 | ||
1da177e4 LT |
1742 | /* |
1743 | * deactivate_task - remove a task from the runqueue. | |
1744 | */ | |
2e1cb74a | 1745 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1746 | { |
d9514f6c | 1747 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1748 | rq->nr_uninterruptible++; |
1749 | ||
69be72c1 | 1750 | dequeue_task(rq, p, sleep); |
c09595f6 | 1751 | dec_nr_running(rq); |
1da177e4 LT |
1752 | } |
1753 | ||
1da177e4 LT |
1754 | /** |
1755 | * task_curr - is this task currently executing on a CPU? | |
1756 | * @p: the task in question. | |
1757 | */ | |
36c8b586 | 1758 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1759 | { |
1760 | return cpu_curr(task_cpu(p)) == p; | |
1761 | } | |
1762 | ||
dd41f596 IM |
1763 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1764 | { | |
6f505b16 | 1765 | set_task_rq(p, cpu); |
dd41f596 | 1766 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1767 | /* |
1768 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1769 | * successfuly executed on another CPU. We must ensure that updates of | |
1770 | * per-task data have been completed by this moment. | |
1771 | */ | |
1772 | smp_wmb(); | |
dd41f596 | 1773 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1774 | #endif |
2dd73a4f PW |
1775 | } |
1776 | ||
cb469845 SR |
1777 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1778 | const struct sched_class *prev_class, | |
1779 | int oldprio, int running) | |
1780 | { | |
1781 | if (prev_class != p->sched_class) { | |
1782 | if (prev_class->switched_from) | |
1783 | prev_class->switched_from(rq, p, running); | |
1784 | p->sched_class->switched_to(rq, p, running); | |
1785 | } else | |
1786 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1787 | } | |
1788 | ||
1da177e4 | 1789 | #ifdef CONFIG_SMP |
c65cc870 | 1790 | |
e958b360 TG |
1791 | /* Used instead of source_load when we know the type == 0 */ |
1792 | static unsigned long weighted_cpuload(const int cpu) | |
1793 | { | |
1794 | return cpu_rq(cpu)->load.weight; | |
1795 | } | |
1796 | ||
cc367732 IM |
1797 | /* |
1798 | * Is this task likely cache-hot: | |
1799 | */ | |
e7693a36 | 1800 | static int |
cc367732 IM |
1801 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1802 | { | |
1803 | s64 delta; | |
1804 | ||
f540a608 IM |
1805 | /* |
1806 | * Buddy candidates are cache hot: | |
1807 | */ | |
4793241b PZ |
1808 | if (sched_feat(CACHE_HOT_BUDDY) && |
1809 | (&p->se == cfs_rq_of(&p->se)->next || | |
1810 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1811 | return 1; |
1812 | ||
cc367732 IM |
1813 | if (p->sched_class != &fair_sched_class) |
1814 | return 0; | |
1815 | ||
6bc1665b IM |
1816 | if (sysctl_sched_migration_cost == -1) |
1817 | return 1; | |
1818 | if (sysctl_sched_migration_cost == 0) | |
1819 | return 0; | |
1820 | ||
cc367732 IM |
1821 | delta = now - p->se.exec_start; |
1822 | ||
1823 | return delta < (s64)sysctl_sched_migration_cost; | |
1824 | } | |
1825 | ||
1826 | ||
dd41f596 | 1827 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1828 | { |
dd41f596 IM |
1829 | int old_cpu = task_cpu(p); |
1830 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1831 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1832 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1833 | u64 clock_offset; |
dd41f596 IM |
1834 | |
1835 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1836 | |
1837 | #ifdef CONFIG_SCHEDSTATS | |
1838 | if (p->se.wait_start) | |
1839 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1840 | if (p->se.sleep_start) |
1841 | p->se.sleep_start -= clock_offset; | |
1842 | if (p->se.block_start) | |
1843 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1844 | if (old_cpu != new_cpu) { |
1845 | schedstat_inc(p, se.nr_migrations); | |
1846 | if (task_hot(p, old_rq->clock, NULL)) | |
1847 | schedstat_inc(p, se.nr_forced2_migrations); | |
1848 | } | |
6cfb0d5d | 1849 | #endif |
2830cf8c SV |
1850 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1851 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1852 | |
1853 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1854 | } |
1855 | ||
70b97a7f | 1856 | struct migration_req { |
1da177e4 | 1857 | struct list_head list; |
1da177e4 | 1858 | |
36c8b586 | 1859 | struct task_struct *task; |
1da177e4 LT |
1860 | int dest_cpu; |
1861 | ||
1da177e4 | 1862 | struct completion done; |
70b97a7f | 1863 | }; |
1da177e4 LT |
1864 | |
1865 | /* | |
1866 | * The task's runqueue lock must be held. | |
1867 | * Returns true if you have to wait for migration thread. | |
1868 | */ | |
36c8b586 | 1869 | static int |
70b97a7f | 1870 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1871 | { |
70b97a7f | 1872 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1873 | |
1874 | /* | |
1875 | * If the task is not on a runqueue (and not running), then | |
1876 | * it is sufficient to simply update the task's cpu field. | |
1877 | */ | |
dd41f596 | 1878 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1879 | set_task_cpu(p, dest_cpu); |
1880 | return 0; | |
1881 | } | |
1882 | ||
1883 | init_completion(&req->done); | |
1da177e4 LT |
1884 | req->task = p; |
1885 | req->dest_cpu = dest_cpu; | |
1886 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1887 | |
1da177e4 LT |
1888 | return 1; |
1889 | } | |
1890 | ||
1891 | /* | |
1892 | * wait_task_inactive - wait for a thread to unschedule. | |
1893 | * | |
85ba2d86 RM |
1894 | * If @match_state is nonzero, it's the @p->state value just checked and |
1895 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1896 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1897 | * we return a positive number (its total switch count). If a second call | |
1898 | * a short while later returns the same number, the caller can be sure that | |
1899 | * @p has remained unscheduled the whole time. | |
1900 | * | |
1da177e4 LT |
1901 | * The caller must ensure that the task *will* unschedule sometime soon, |
1902 | * else this function might spin for a *long* time. This function can't | |
1903 | * be called with interrupts off, or it may introduce deadlock with | |
1904 | * smp_call_function() if an IPI is sent by the same process we are | |
1905 | * waiting to become inactive. | |
1906 | */ | |
85ba2d86 | 1907 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1908 | { |
1909 | unsigned long flags; | |
dd41f596 | 1910 | int running, on_rq; |
85ba2d86 | 1911 | unsigned long ncsw; |
70b97a7f | 1912 | struct rq *rq; |
1da177e4 | 1913 | |
3a5c359a AK |
1914 | for (;;) { |
1915 | /* | |
1916 | * We do the initial early heuristics without holding | |
1917 | * any task-queue locks at all. We'll only try to get | |
1918 | * the runqueue lock when things look like they will | |
1919 | * work out! | |
1920 | */ | |
1921 | rq = task_rq(p); | |
fa490cfd | 1922 | |
3a5c359a AK |
1923 | /* |
1924 | * If the task is actively running on another CPU | |
1925 | * still, just relax and busy-wait without holding | |
1926 | * any locks. | |
1927 | * | |
1928 | * NOTE! Since we don't hold any locks, it's not | |
1929 | * even sure that "rq" stays as the right runqueue! | |
1930 | * But we don't care, since "task_running()" will | |
1931 | * return false if the runqueue has changed and p | |
1932 | * is actually now running somewhere else! | |
1933 | */ | |
85ba2d86 RM |
1934 | while (task_running(rq, p)) { |
1935 | if (match_state && unlikely(p->state != match_state)) | |
1936 | return 0; | |
3a5c359a | 1937 | cpu_relax(); |
85ba2d86 | 1938 | } |
fa490cfd | 1939 | |
3a5c359a AK |
1940 | /* |
1941 | * Ok, time to look more closely! We need the rq | |
1942 | * lock now, to be *sure*. If we're wrong, we'll | |
1943 | * just go back and repeat. | |
1944 | */ | |
1945 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 1946 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
1947 | running = task_running(rq, p); |
1948 | on_rq = p->se.on_rq; | |
85ba2d86 | 1949 | ncsw = 0; |
f31e11d8 | 1950 | if (!match_state || p->state == match_state) |
93dcf55f | 1951 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1952 | task_rq_unlock(rq, &flags); |
fa490cfd | 1953 | |
85ba2d86 RM |
1954 | /* |
1955 | * If it changed from the expected state, bail out now. | |
1956 | */ | |
1957 | if (unlikely(!ncsw)) | |
1958 | break; | |
1959 | ||
3a5c359a AK |
1960 | /* |
1961 | * Was it really running after all now that we | |
1962 | * checked with the proper locks actually held? | |
1963 | * | |
1964 | * Oops. Go back and try again.. | |
1965 | */ | |
1966 | if (unlikely(running)) { | |
1967 | cpu_relax(); | |
1968 | continue; | |
1969 | } | |
fa490cfd | 1970 | |
3a5c359a AK |
1971 | /* |
1972 | * It's not enough that it's not actively running, | |
1973 | * it must be off the runqueue _entirely_, and not | |
1974 | * preempted! | |
1975 | * | |
1976 | * So if it wa still runnable (but just not actively | |
1977 | * running right now), it's preempted, and we should | |
1978 | * yield - it could be a while. | |
1979 | */ | |
1980 | if (unlikely(on_rq)) { | |
1981 | schedule_timeout_uninterruptible(1); | |
1982 | continue; | |
1983 | } | |
fa490cfd | 1984 | |
3a5c359a AK |
1985 | /* |
1986 | * Ahh, all good. It wasn't running, and it wasn't | |
1987 | * runnable, which means that it will never become | |
1988 | * running in the future either. We're all done! | |
1989 | */ | |
1990 | break; | |
1991 | } | |
85ba2d86 RM |
1992 | |
1993 | return ncsw; | |
1da177e4 LT |
1994 | } |
1995 | ||
1996 | /*** | |
1997 | * kick_process - kick a running thread to enter/exit the kernel | |
1998 | * @p: the to-be-kicked thread | |
1999 | * | |
2000 | * Cause a process which is running on another CPU to enter | |
2001 | * kernel-mode, without any delay. (to get signals handled.) | |
2002 | * | |
2003 | * NOTE: this function doesnt have to take the runqueue lock, | |
2004 | * because all it wants to ensure is that the remote task enters | |
2005 | * the kernel. If the IPI races and the task has been migrated | |
2006 | * to another CPU then no harm is done and the purpose has been | |
2007 | * achieved as well. | |
2008 | */ | |
36c8b586 | 2009 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2010 | { |
2011 | int cpu; | |
2012 | ||
2013 | preempt_disable(); | |
2014 | cpu = task_cpu(p); | |
2015 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2016 | smp_send_reschedule(cpu); | |
2017 | preempt_enable(); | |
2018 | } | |
2019 | ||
2020 | /* | |
2dd73a4f PW |
2021 | * Return a low guess at the load of a migration-source cpu weighted |
2022 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2023 | * |
2024 | * We want to under-estimate the load of migration sources, to | |
2025 | * balance conservatively. | |
2026 | */ | |
a9957449 | 2027 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2028 | { |
70b97a7f | 2029 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2030 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2031 | |
93b75217 | 2032 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2033 | return total; |
b910472d | 2034 | |
dd41f596 | 2035 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2036 | } |
2037 | ||
2038 | /* | |
2dd73a4f PW |
2039 | * Return a high guess at the load of a migration-target cpu weighted |
2040 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2041 | */ |
a9957449 | 2042 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2043 | { |
70b97a7f | 2044 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2045 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2046 | |
93b75217 | 2047 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2048 | return total; |
3b0bd9bc | 2049 | |
dd41f596 | 2050 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2051 | } |
2052 | ||
147cbb4b NP |
2053 | /* |
2054 | * find_idlest_group finds and returns the least busy CPU group within the | |
2055 | * domain. | |
2056 | */ | |
2057 | static struct sched_group * | |
2058 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2059 | { | |
2060 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2061 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2062 | int load_idx = sd->forkexec_idx; | |
2063 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2064 | ||
2065 | do { | |
2066 | unsigned long load, avg_load; | |
2067 | int local_group; | |
2068 | int i; | |
2069 | ||
da5a5522 BD |
2070 | /* Skip over this group if it has no CPUs allowed */ |
2071 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2072 | continue; |
da5a5522 | 2073 | |
147cbb4b | 2074 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2075 | |
2076 | /* Tally up the load of all CPUs in the group */ | |
2077 | avg_load = 0; | |
2078 | ||
363ab6f1 | 2079 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2080 | /* Bias balancing toward cpus of our domain */ |
2081 | if (local_group) | |
2082 | load = source_load(i, load_idx); | |
2083 | else | |
2084 | load = target_load(i, load_idx); | |
2085 | ||
2086 | avg_load += load; | |
2087 | } | |
2088 | ||
2089 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2090 | avg_load = sg_div_cpu_power(group, |
2091 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2092 | |
2093 | if (local_group) { | |
2094 | this_load = avg_load; | |
2095 | this = group; | |
2096 | } else if (avg_load < min_load) { | |
2097 | min_load = avg_load; | |
2098 | idlest = group; | |
2099 | } | |
3a5c359a | 2100 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2101 | |
2102 | if (!idlest || 100*this_load < imbalance*min_load) | |
2103 | return NULL; | |
2104 | return idlest; | |
2105 | } | |
2106 | ||
2107 | /* | |
0feaece9 | 2108 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2109 | */ |
95cdf3b7 | 2110 | static int |
7c16ec58 MT |
2111 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2112 | cpumask_t *tmp) | |
147cbb4b NP |
2113 | { |
2114 | unsigned long load, min_load = ULONG_MAX; | |
2115 | int idlest = -1; | |
2116 | int i; | |
2117 | ||
da5a5522 | 2118 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2119 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2120 | |
363ab6f1 | 2121 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2122 | load = weighted_cpuload(i); |
147cbb4b NP |
2123 | |
2124 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2125 | min_load = load; | |
2126 | idlest = i; | |
2127 | } | |
2128 | } | |
2129 | ||
2130 | return idlest; | |
2131 | } | |
2132 | ||
476d139c NP |
2133 | /* |
2134 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2135 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2136 | * SD_BALANCE_EXEC. | |
2137 | * | |
2138 | * Balance, ie. select the least loaded group. | |
2139 | * | |
2140 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2141 | * | |
2142 | * preempt must be disabled. | |
2143 | */ | |
2144 | static int sched_balance_self(int cpu, int flag) | |
2145 | { | |
2146 | struct task_struct *t = current; | |
2147 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2148 | |
c96d145e | 2149 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2150 | /* |
2151 | * If power savings logic is enabled for a domain, stop there. | |
2152 | */ | |
5c45bf27 SS |
2153 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2154 | break; | |
476d139c NP |
2155 | if (tmp->flags & flag) |
2156 | sd = tmp; | |
c96d145e | 2157 | } |
476d139c | 2158 | |
039a1c41 PZ |
2159 | if (sd) |
2160 | update_shares(sd); | |
2161 | ||
476d139c | 2162 | while (sd) { |
7c16ec58 | 2163 | cpumask_t span, tmpmask; |
476d139c | 2164 | struct sched_group *group; |
1a848870 SS |
2165 | int new_cpu, weight; |
2166 | ||
2167 | if (!(sd->flags & flag)) { | |
2168 | sd = sd->child; | |
2169 | continue; | |
2170 | } | |
476d139c NP |
2171 | |
2172 | span = sd->span; | |
2173 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2174 | if (!group) { |
2175 | sd = sd->child; | |
2176 | continue; | |
2177 | } | |
476d139c | 2178 | |
7c16ec58 | 2179 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2180 | if (new_cpu == -1 || new_cpu == cpu) { |
2181 | /* Now try balancing at a lower domain level of cpu */ | |
2182 | sd = sd->child; | |
2183 | continue; | |
2184 | } | |
476d139c | 2185 | |
1a848870 | 2186 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2187 | cpu = new_cpu; |
476d139c NP |
2188 | sd = NULL; |
2189 | weight = cpus_weight(span); | |
2190 | for_each_domain(cpu, tmp) { | |
2191 | if (weight <= cpus_weight(tmp->span)) | |
2192 | break; | |
2193 | if (tmp->flags & flag) | |
2194 | sd = tmp; | |
2195 | } | |
2196 | /* while loop will break here if sd == NULL */ | |
2197 | } | |
2198 | ||
2199 | return cpu; | |
2200 | } | |
2201 | ||
2202 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2203 | |
1da177e4 LT |
2204 | /*** |
2205 | * try_to_wake_up - wake up a thread | |
2206 | * @p: the to-be-woken-up thread | |
2207 | * @state: the mask of task states that can be woken | |
2208 | * @sync: do a synchronous wakeup? | |
2209 | * | |
2210 | * Put it on the run-queue if it's not already there. The "current" | |
2211 | * thread is always on the run-queue (except when the actual | |
2212 | * re-schedule is in progress), and as such you're allowed to do | |
2213 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2214 | * runnable without the overhead of this. | |
2215 | * | |
2216 | * returns failure only if the task is already active. | |
2217 | */ | |
36c8b586 | 2218 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2219 | { |
cc367732 | 2220 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2221 | unsigned long flags; |
2222 | long old_state; | |
70b97a7f | 2223 | struct rq *rq; |
1da177e4 | 2224 | |
b85d0667 IM |
2225 | if (!sched_feat(SYNC_WAKEUPS)) |
2226 | sync = 0; | |
2227 | ||
2398f2c6 PZ |
2228 | #ifdef CONFIG_SMP |
2229 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2230 | struct sched_domain *sd; | |
2231 | ||
2232 | this_cpu = raw_smp_processor_id(); | |
2233 | cpu = task_cpu(p); | |
2234 | ||
2235 | for_each_domain(this_cpu, sd) { | |
2236 | if (cpu_isset(cpu, sd->span)) { | |
2237 | update_shares(sd); | |
2238 | break; | |
2239 | } | |
2240 | } | |
2241 | } | |
2242 | #endif | |
2243 | ||
04e2f174 | 2244 | smp_wmb(); |
1da177e4 LT |
2245 | rq = task_rq_lock(p, &flags); |
2246 | old_state = p->state; | |
2247 | if (!(old_state & state)) | |
2248 | goto out; | |
2249 | ||
dd41f596 | 2250 | if (p->se.on_rq) |
1da177e4 LT |
2251 | goto out_running; |
2252 | ||
2253 | cpu = task_cpu(p); | |
cc367732 | 2254 | orig_cpu = cpu; |
1da177e4 LT |
2255 | this_cpu = smp_processor_id(); |
2256 | ||
2257 | #ifdef CONFIG_SMP | |
2258 | if (unlikely(task_running(rq, p))) | |
2259 | goto out_activate; | |
2260 | ||
5d2f5a61 DA |
2261 | cpu = p->sched_class->select_task_rq(p, sync); |
2262 | if (cpu != orig_cpu) { | |
2263 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2264 | task_rq_unlock(rq, &flags); |
2265 | /* might preempt at this point */ | |
2266 | rq = task_rq_lock(p, &flags); | |
2267 | old_state = p->state; | |
2268 | if (!(old_state & state)) | |
2269 | goto out; | |
dd41f596 | 2270 | if (p->se.on_rq) |
1da177e4 LT |
2271 | goto out_running; |
2272 | ||
2273 | this_cpu = smp_processor_id(); | |
2274 | cpu = task_cpu(p); | |
2275 | } | |
2276 | ||
e7693a36 GH |
2277 | #ifdef CONFIG_SCHEDSTATS |
2278 | schedstat_inc(rq, ttwu_count); | |
2279 | if (cpu == this_cpu) | |
2280 | schedstat_inc(rq, ttwu_local); | |
2281 | else { | |
2282 | struct sched_domain *sd; | |
2283 | for_each_domain(this_cpu, sd) { | |
2284 | if (cpu_isset(cpu, sd->span)) { | |
2285 | schedstat_inc(sd, ttwu_wake_remote); | |
2286 | break; | |
2287 | } | |
2288 | } | |
2289 | } | |
6d6bc0ad | 2290 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2291 | |
1da177e4 LT |
2292 | out_activate: |
2293 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2294 | schedstat_inc(p, se.nr_wakeups); |
2295 | if (sync) | |
2296 | schedstat_inc(p, se.nr_wakeups_sync); | |
2297 | if (orig_cpu != cpu) | |
2298 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2299 | if (cpu == this_cpu) | |
2300 | schedstat_inc(p, se.nr_wakeups_local); | |
2301 | else | |
2302 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2303 | update_rq_clock(rq); |
dd41f596 | 2304 | activate_task(rq, p, 1); |
1da177e4 LT |
2305 | success = 1; |
2306 | ||
2307 | out_running: | |
0a16b607 | 2308 | trace_sched_wakeup(rq, p); |
15afe09b | 2309 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2310 | |
1da177e4 | 2311 | p->state = TASK_RUNNING; |
9a897c5a SR |
2312 | #ifdef CONFIG_SMP |
2313 | if (p->sched_class->task_wake_up) | |
2314 | p->sched_class->task_wake_up(rq, p); | |
2315 | #endif | |
1da177e4 | 2316 | out: |
2087a1ad GH |
2317 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2318 | ||
1da177e4 LT |
2319 | task_rq_unlock(rq, &flags); |
2320 | ||
2321 | return success; | |
2322 | } | |
2323 | ||
7ad5b3a5 | 2324 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2325 | { |
d9514f6c | 2326 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2327 | } |
1da177e4 LT |
2328 | EXPORT_SYMBOL(wake_up_process); |
2329 | ||
7ad5b3a5 | 2330 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2331 | { |
2332 | return try_to_wake_up(p, state, 0); | |
2333 | } | |
2334 | ||
1da177e4 LT |
2335 | /* |
2336 | * Perform scheduler related setup for a newly forked process p. | |
2337 | * p is forked by current. | |
dd41f596 IM |
2338 | * |
2339 | * __sched_fork() is basic setup used by init_idle() too: | |
2340 | */ | |
2341 | static void __sched_fork(struct task_struct *p) | |
2342 | { | |
dd41f596 IM |
2343 | p->se.exec_start = 0; |
2344 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2345 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2346 | p->se.last_wakeup = 0; |
2347 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2348 | |
2349 | #ifdef CONFIG_SCHEDSTATS | |
2350 | p->se.wait_start = 0; | |
dd41f596 IM |
2351 | p->se.sum_sleep_runtime = 0; |
2352 | p->se.sleep_start = 0; | |
dd41f596 IM |
2353 | p->se.block_start = 0; |
2354 | p->se.sleep_max = 0; | |
2355 | p->se.block_max = 0; | |
2356 | p->se.exec_max = 0; | |
eba1ed4b | 2357 | p->se.slice_max = 0; |
dd41f596 | 2358 | p->se.wait_max = 0; |
6cfb0d5d | 2359 | #endif |
476d139c | 2360 | |
fa717060 | 2361 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2362 | p->se.on_rq = 0; |
4a55bd5e | 2363 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2364 | |
e107be36 AK |
2365 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2366 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2367 | #endif | |
2368 | ||
1da177e4 LT |
2369 | /* |
2370 | * We mark the process as running here, but have not actually | |
2371 | * inserted it onto the runqueue yet. This guarantees that | |
2372 | * nobody will actually run it, and a signal or other external | |
2373 | * event cannot wake it up and insert it on the runqueue either. | |
2374 | */ | |
2375 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2376 | } |
2377 | ||
2378 | /* | |
2379 | * fork()/clone()-time setup: | |
2380 | */ | |
2381 | void sched_fork(struct task_struct *p, int clone_flags) | |
2382 | { | |
2383 | int cpu = get_cpu(); | |
2384 | ||
2385 | __sched_fork(p); | |
2386 | ||
2387 | #ifdef CONFIG_SMP | |
2388 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2389 | #endif | |
02e4bac2 | 2390 | set_task_cpu(p, cpu); |
b29739f9 IM |
2391 | |
2392 | /* | |
2393 | * Make sure we do not leak PI boosting priority to the child: | |
2394 | */ | |
2395 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2396 | if (!rt_prio(p->prio)) |
2397 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2398 | |
52f17b6c | 2399 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2400 | if (likely(sched_info_on())) |
52f17b6c | 2401 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2402 | #endif |
d6077cb8 | 2403 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2404 | p->oncpu = 0; |
2405 | #endif | |
1da177e4 | 2406 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2407 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2408 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2409 | #endif |
476d139c | 2410 | put_cpu(); |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* | |
2414 | * wake_up_new_task - wake up a newly created task for the first time. | |
2415 | * | |
2416 | * This function will do some initial scheduler statistics housekeeping | |
2417 | * that must be done for every newly created context, then puts the task | |
2418 | * on the runqueue and wakes it. | |
2419 | */ | |
7ad5b3a5 | 2420 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2421 | { |
2422 | unsigned long flags; | |
dd41f596 | 2423 | struct rq *rq; |
1da177e4 LT |
2424 | |
2425 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2426 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2427 | update_rq_clock(rq); |
1da177e4 LT |
2428 | |
2429 | p->prio = effective_prio(p); | |
2430 | ||
b9dca1e0 | 2431 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2432 | activate_task(rq, p, 0); |
1da177e4 | 2433 | } else { |
1da177e4 | 2434 | /* |
dd41f596 IM |
2435 | * Let the scheduling class do new task startup |
2436 | * management (if any): | |
1da177e4 | 2437 | */ |
ee0827d8 | 2438 | p->sched_class->task_new(rq, p); |
c09595f6 | 2439 | inc_nr_running(rq); |
1da177e4 | 2440 | } |
0a16b607 | 2441 | trace_sched_wakeup_new(rq, p); |
15afe09b | 2442 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2443 | #ifdef CONFIG_SMP |
2444 | if (p->sched_class->task_wake_up) | |
2445 | p->sched_class->task_wake_up(rq, p); | |
2446 | #endif | |
dd41f596 | 2447 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2448 | } |
2449 | ||
e107be36 AK |
2450 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2451 | ||
2452 | /** | |
421cee29 RD |
2453 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2454 | * @notifier: notifier struct to register | |
e107be36 AK |
2455 | */ |
2456 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2457 | { | |
2458 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2459 | } | |
2460 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2461 | ||
2462 | /** | |
2463 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2464 | * @notifier: notifier struct to unregister |
e107be36 AK |
2465 | * |
2466 | * This is safe to call from within a preemption notifier. | |
2467 | */ | |
2468 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2469 | { | |
2470 | hlist_del(¬ifier->link); | |
2471 | } | |
2472 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2473 | ||
2474 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2475 | { | |
2476 | struct preempt_notifier *notifier; | |
2477 | struct hlist_node *node; | |
2478 | ||
2479 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2480 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2481 | } | |
2482 | ||
2483 | static void | |
2484 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2485 | struct task_struct *next) | |
2486 | { | |
2487 | struct preempt_notifier *notifier; | |
2488 | struct hlist_node *node; | |
2489 | ||
2490 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2491 | notifier->ops->sched_out(notifier, next); | |
2492 | } | |
2493 | ||
6d6bc0ad | 2494 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2495 | |
2496 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2497 | { | |
2498 | } | |
2499 | ||
2500 | static void | |
2501 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2502 | struct task_struct *next) | |
2503 | { | |
2504 | } | |
2505 | ||
6d6bc0ad | 2506 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2507 | |
4866cde0 NP |
2508 | /** |
2509 | * prepare_task_switch - prepare to switch tasks | |
2510 | * @rq: the runqueue preparing to switch | |
421cee29 | 2511 | * @prev: the current task that is being switched out |
4866cde0 NP |
2512 | * @next: the task we are going to switch to. |
2513 | * | |
2514 | * This is called with the rq lock held and interrupts off. It must | |
2515 | * be paired with a subsequent finish_task_switch after the context | |
2516 | * switch. | |
2517 | * | |
2518 | * prepare_task_switch sets up locking and calls architecture specific | |
2519 | * hooks. | |
2520 | */ | |
e107be36 AK |
2521 | static inline void |
2522 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2523 | struct task_struct *next) | |
4866cde0 | 2524 | { |
e107be36 | 2525 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2526 | prepare_lock_switch(rq, next); |
2527 | prepare_arch_switch(next); | |
2528 | } | |
2529 | ||
1da177e4 LT |
2530 | /** |
2531 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2532 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2533 | * @prev: the thread we just switched away from. |
2534 | * | |
4866cde0 NP |
2535 | * finish_task_switch must be called after the context switch, paired |
2536 | * with a prepare_task_switch call before the context switch. | |
2537 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2538 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2539 | * |
2540 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2541 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2542 | * with the lock held can cause deadlocks; see schedule() for |
2543 | * details.) | |
2544 | */ | |
a9957449 | 2545 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2546 | __releases(rq->lock) |
2547 | { | |
1da177e4 | 2548 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2549 | long prev_state; |
1da177e4 LT |
2550 | |
2551 | rq->prev_mm = NULL; | |
2552 | ||
2553 | /* | |
2554 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2555 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2556 | * schedule one last time. The schedule call will never return, and |
2557 | * the scheduled task must drop that reference. | |
c394cc9f | 2558 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2559 | * still held, otherwise prev could be scheduled on another cpu, die |
2560 | * there before we look at prev->state, and then the reference would | |
2561 | * be dropped twice. | |
2562 | * Manfred Spraul <manfred@colorfullife.com> | |
2563 | */ | |
55a101f8 | 2564 | prev_state = prev->state; |
4866cde0 NP |
2565 | finish_arch_switch(prev); |
2566 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2567 | #ifdef CONFIG_SMP |
2568 | if (current->sched_class->post_schedule) | |
2569 | current->sched_class->post_schedule(rq); | |
2570 | #endif | |
e8fa1362 | 2571 | |
e107be36 | 2572 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2573 | if (mm) |
2574 | mmdrop(mm); | |
c394cc9f | 2575 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2576 | /* |
2577 | * Remove function-return probe instances associated with this | |
2578 | * task and put them back on the free list. | |
9761eea8 | 2579 | */ |
c6fd91f0 | 2580 | kprobe_flush_task(prev); |
1da177e4 | 2581 | put_task_struct(prev); |
c6fd91f0 | 2582 | } |
1da177e4 LT |
2583 | } |
2584 | ||
2585 | /** | |
2586 | * schedule_tail - first thing a freshly forked thread must call. | |
2587 | * @prev: the thread we just switched away from. | |
2588 | */ | |
36c8b586 | 2589 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2590 | __releases(rq->lock) |
2591 | { | |
70b97a7f IM |
2592 | struct rq *rq = this_rq(); |
2593 | ||
4866cde0 NP |
2594 | finish_task_switch(rq, prev); |
2595 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2596 | /* In this case, finish_task_switch does not reenable preemption */ | |
2597 | preempt_enable(); | |
2598 | #endif | |
1da177e4 | 2599 | if (current->set_child_tid) |
b488893a | 2600 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2601 | } |
2602 | ||
2603 | /* | |
2604 | * context_switch - switch to the new MM and the new | |
2605 | * thread's register state. | |
2606 | */ | |
dd41f596 | 2607 | static inline void |
70b97a7f | 2608 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2609 | struct task_struct *next) |
1da177e4 | 2610 | { |
dd41f596 | 2611 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2612 | |
e107be36 | 2613 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2614 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2615 | mm = next->mm; |
2616 | oldmm = prev->active_mm; | |
9226d125 ZA |
2617 | /* |
2618 | * For paravirt, this is coupled with an exit in switch_to to | |
2619 | * combine the page table reload and the switch backend into | |
2620 | * one hypercall. | |
2621 | */ | |
2622 | arch_enter_lazy_cpu_mode(); | |
2623 | ||
dd41f596 | 2624 | if (unlikely(!mm)) { |
1da177e4 LT |
2625 | next->active_mm = oldmm; |
2626 | atomic_inc(&oldmm->mm_count); | |
2627 | enter_lazy_tlb(oldmm, next); | |
2628 | } else | |
2629 | switch_mm(oldmm, mm, next); | |
2630 | ||
dd41f596 | 2631 | if (unlikely(!prev->mm)) { |
1da177e4 | 2632 | prev->active_mm = NULL; |
1da177e4 LT |
2633 | rq->prev_mm = oldmm; |
2634 | } | |
3a5f5e48 IM |
2635 | /* |
2636 | * Since the runqueue lock will be released by the next | |
2637 | * task (which is an invalid locking op but in the case | |
2638 | * of the scheduler it's an obvious special-case), so we | |
2639 | * do an early lockdep release here: | |
2640 | */ | |
2641 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2642 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2643 | #endif |
1da177e4 LT |
2644 | |
2645 | /* Here we just switch the register state and the stack. */ | |
2646 | switch_to(prev, next, prev); | |
2647 | ||
dd41f596 IM |
2648 | barrier(); |
2649 | /* | |
2650 | * this_rq must be evaluated again because prev may have moved | |
2651 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2652 | * frame will be invalid. | |
2653 | */ | |
2654 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2655 | } |
2656 | ||
2657 | /* | |
2658 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2659 | * | |
2660 | * externally visible scheduler statistics: current number of runnable | |
2661 | * threads, current number of uninterruptible-sleeping threads, total | |
2662 | * number of context switches performed since bootup. | |
2663 | */ | |
2664 | unsigned long nr_running(void) | |
2665 | { | |
2666 | unsigned long i, sum = 0; | |
2667 | ||
2668 | for_each_online_cpu(i) | |
2669 | sum += cpu_rq(i)->nr_running; | |
2670 | ||
2671 | return sum; | |
2672 | } | |
2673 | ||
2674 | unsigned long nr_uninterruptible(void) | |
2675 | { | |
2676 | unsigned long i, sum = 0; | |
2677 | ||
0a945022 | 2678 | for_each_possible_cpu(i) |
1da177e4 LT |
2679 | sum += cpu_rq(i)->nr_uninterruptible; |
2680 | ||
2681 | /* | |
2682 | * Since we read the counters lockless, it might be slightly | |
2683 | * inaccurate. Do not allow it to go below zero though: | |
2684 | */ | |
2685 | if (unlikely((long)sum < 0)) | |
2686 | sum = 0; | |
2687 | ||
2688 | return sum; | |
2689 | } | |
2690 | ||
2691 | unsigned long long nr_context_switches(void) | |
2692 | { | |
cc94abfc SR |
2693 | int i; |
2694 | unsigned long long sum = 0; | |
1da177e4 | 2695 | |
0a945022 | 2696 | for_each_possible_cpu(i) |
1da177e4 LT |
2697 | sum += cpu_rq(i)->nr_switches; |
2698 | ||
2699 | return sum; | |
2700 | } | |
2701 | ||
2702 | unsigned long nr_iowait(void) | |
2703 | { | |
2704 | unsigned long i, sum = 0; | |
2705 | ||
0a945022 | 2706 | for_each_possible_cpu(i) |
1da177e4 LT |
2707 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2708 | ||
2709 | return sum; | |
2710 | } | |
2711 | ||
db1b1fef JS |
2712 | unsigned long nr_active(void) |
2713 | { | |
2714 | unsigned long i, running = 0, uninterruptible = 0; | |
2715 | ||
2716 | for_each_online_cpu(i) { | |
2717 | running += cpu_rq(i)->nr_running; | |
2718 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2719 | } | |
2720 | ||
2721 | if (unlikely((long)uninterruptible < 0)) | |
2722 | uninterruptible = 0; | |
2723 | ||
2724 | return running + uninterruptible; | |
2725 | } | |
2726 | ||
48f24c4d | 2727 | /* |
dd41f596 IM |
2728 | * Update rq->cpu_load[] statistics. This function is usually called every |
2729 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2730 | */ |
dd41f596 | 2731 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2732 | { |
495eca49 | 2733 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2734 | int i, scale; |
2735 | ||
2736 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2737 | |
2738 | /* Update our load: */ | |
2739 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2740 | unsigned long old_load, new_load; | |
2741 | ||
2742 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2743 | ||
2744 | old_load = this_rq->cpu_load[i]; | |
2745 | new_load = this_load; | |
a25707f3 IM |
2746 | /* |
2747 | * Round up the averaging division if load is increasing. This | |
2748 | * prevents us from getting stuck on 9 if the load is 10, for | |
2749 | * example. | |
2750 | */ | |
2751 | if (new_load > old_load) | |
2752 | new_load += scale-1; | |
dd41f596 IM |
2753 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2754 | } | |
48f24c4d IM |
2755 | } |
2756 | ||
dd41f596 IM |
2757 | #ifdef CONFIG_SMP |
2758 | ||
1da177e4 LT |
2759 | /* |
2760 | * double_rq_lock - safely lock two runqueues | |
2761 | * | |
2762 | * Note this does not disable interrupts like task_rq_lock, | |
2763 | * you need to do so manually before calling. | |
2764 | */ | |
70b97a7f | 2765 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2766 | __acquires(rq1->lock) |
2767 | __acquires(rq2->lock) | |
2768 | { | |
054b9108 | 2769 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2770 | if (rq1 == rq2) { |
2771 | spin_lock(&rq1->lock); | |
2772 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2773 | } else { | |
c96d145e | 2774 | if (rq1 < rq2) { |
1da177e4 | 2775 | spin_lock(&rq1->lock); |
5e710e37 | 2776 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2777 | } else { |
2778 | spin_lock(&rq2->lock); | |
5e710e37 | 2779 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2780 | } |
2781 | } | |
6e82a3be IM |
2782 | update_rq_clock(rq1); |
2783 | update_rq_clock(rq2); | |
1da177e4 LT |
2784 | } |
2785 | ||
2786 | /* | |
2787 | * double_rq_unlock - safely unlock two runqueues | |
2788 | * | |
2789 | * Note this does not restore interrupts like task_rq_unlock, | |
2790 | * you need to do so manually after calling. | |
2791 | */ | |
70b97a7f | 2792 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2793 | __releases(rq1->lock) |
2794 | __releases(rq2->lock) | |
2795 | { | |
2796 | spin_unlock(&rq1->lock); | |
2797 | if (rq1 != rq2) | |
2798 | spin_unlock(&rq2->lock); | |
2799 | else | |
2800 | __release(rq2->lock); | |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2805 | */ | |
e8fa1362 | 2806 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2807 | __releases(this_rq->lock) |
2808 | __acquires(busiest->lock) | |
2809 | __acquires(this_rq->lock) | |
2810 | { | |
e8fa1362 SR |
2811 | int ret = 0; |
2812 | ||
054b9108 KK |
2813 | if (unlikely(!irqs_disabled())) { |
2814 | /* printk() doesn't work good under rq->lock */ | |
2815 | spin_unlock(&this_rq->lock); | |
2816 | BUG_ON(1); | |
2817 | } | |
1da177e4 | 2818 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2819 | if (busiest < this_rq) { |
1da177e4 LT |
2820 | spin_unlock(&this_rq->lock); |
2821 | spin_lock(&busiest->lock); | |
5e710e37 | 2822 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2823 | ret = 1; |
1da177e4 | 2824 | } else |
5e710e37 | 2825 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2826 | } |
e8fa1362 | 2827 | return ret; |
1da177e4 LT |
2828 | } |
2829 | ||
1b12bbc7 PZ |
2830 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2831 | __releases(busiest->lock) | |
2832 | { | |
2833 | spin_unlock(&busiest->lock); | |
2834 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2835 | } | |
2836 | ||
1da177e4 LT |
2837 | /* |
2838 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2839 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2840 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2841 | * the cpu_allowed mask is restored. |
2842 | */ | |
36c8b586 | 2843 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2844 | { |
70b97a7f | 2845 | struct migration_req req; |
1da177e4 | 2846 | unsigned long flags; |
70b97a7f | 2847 | struct rq *rq; |
1da177e4 LT |
2848 | |
2849 | rq = task_rq_lock(p, &flags); | |
2850 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2851 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2852 | goto out; |
2853 | ||
0a16b607 | 2854 | trace_sched_migrate_task(rq, p, dest_cpu); |
1da177e4 LT |
2855 | /* force the process onto the specified CPU */ |
2856 | if (migrate_task(p, dest_cpu, &req)) { | |
2857 | /* Need to wait for migration thread (might exit: take ref). */ | |
2858 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2859 | |
1da177e4 LT |
2860 | get_task_struct(mt); |
2861 | task_rq_unlock(rq, &flags); | |
2862 | wake_up_process(mt); | |
2863 | put_task_struct(mt); | |
2864 | wait_for_completion(&req.done); | |
36c8b586 | 2865 | |
1da177e4 LT |
2866 | return; |
2867 | } | |
2868 | out: | |
2869 | task_rq_unlock(rq, &flags); | |
2870 | } | |
2871 | ||
2872 | /* | |
476d139c NP |
2873 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2874 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2875 | */ |
2876 | void sched_exec(void) | |
2877 | { | |
1da177e4 | 2878 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2879 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2880 | put_cpu(); |
476d139c NP |
2881 | if (new_cpu != this_cpu) |
2882 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2883 | } |
2884 | ||
2885 | /* | |
2886 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2887 | * Both runqueues must be locked. | |
2888 | */ | |
dd41f596 IM |
2889 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2890 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2891 | { |
2e1cb74a | 2892 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2893 | set_task_cpu(p, this_cpu); |
dd41f596 | 2894 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2895 | /* |
2896 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2897 | * to be always true for them. | |
2898 | */ | |
15afe09b | 2899 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2900 | } |
2901 | ||
2902 | /* | |
2903 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2904 | */ | |
858119e1 | 2905 | static |
70b97a7f | 2906 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2907 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2908 | int *all_pinned) |
1da177e4 LT |
2909 | { |
2910 | /* | |
2911 | * We do not migrate tasks that are: | |
2912 | * 1) running (obviously), or | |
2913 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2914 | * 3) are cache-hot on their current CPU. | |
2915 | */ | |
cc367732 IM |
2916 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2917 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2918 | return 0; |
cc367732 | 2919 | } |
81026794 NP |
2920 | *all_pinned = 0; |
2921 | ||
cc367732 IM |
2922 | if (task_running(rq, p)) { |
2923 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2924 | return 0; |
cc367732 | 2925 | } |
1da177e4 | 2926 | |
da84d961 IM |
2927 | /* |
2928 | * Aggressive migration if: | |
2929 | * 1) task is cache cold, or | |
2930 | * 2) too many balance attempts have failed. | |
2931 | */ | |
2932 | ||
6bc1665b IM |
2933 | if (!task_hot(p, rq->clock, sd) || |
2934 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2935 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2936 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2937 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2938 | schedstat_inc(p, se.nr_forced_migrations); |
2939 | } | |
da84d961 IM |
2940 | #endif |
2941 | return 1; | |
2942 | } | |
2943 | ||
cc367732 IM |
2944 | if (task_hot(p, rq->clock, sd)) { |
2945 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2946 | return 0; |
cc367732 | 2947 | } |
1da177e4 LT |
2948 | return 1; |
2949 | } | |
2950 | ||
e1d1484f PW |
2951 | static unsigned long |
2952 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2953 | unsigned long max_load_move, struct sched_domain *sd, | |
2954 | enum cpu_idle_type idle, int *all_pinned, | |
2955 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2956 | { |
051c6764 | 2957 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2958 | struct task_struct *p; |
2959 | long rem_load_move = max_load_move; | |
1da177e4 | 2960 | |
e1d1484f | 2961 | if (max_load_move == 0) |
1da177e4 LT |
2962 | goto out; |
2963 | ||
81026794 NP |
2964 | pinned = 1; |
2965 | ||
1da177e4 | 2966 | /* |
dd41f596 | 2967 | * Start the load-balancing iterator: |
1da177e4 | 2968 | */ |
dd41f596 IM |
2969 | p = iterator->start(iterator->arg); |
2970 | next: | |
b82d9fdd | 2971 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2972 | goto out; |
051c6764 PZ |
2973 | |
2974 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2975 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2976 | p = iterator->next(iterator->arg); |
2977 | goto next; | |
1da177e4 LT |
2978 | } |
2979 | ||
dd41f596 | 2980 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2981 | pulled++; |
dd41f596 | 2982 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2983 | |
2dd73a4f | 2984 | /* |
b82d9fdd | 2985 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2986 | */ |
e1d1484f | 2987 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2988 | if (p->prio < *this_best_prio) |
2989 | *this_best_prio = p->prio; | |
dd41f596 IM |
2990 | p = iterator->next(iterator->arg); |
2991 | goto next; | |
1da177e4 LT |
2992 | } |
2993 | out: | |
2994 | /* | |
e1d1484f | 2995 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2996 | * so we can safely collect pull_task() stats here rather than |
2997 | * inside pull_task(). | |
2998 | */ | |
2999 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3000 | |
3001 | if (all_pinned) | |
3002 | *all_pinned = pinned; | |
e1d1484f PW |
3003 | |
3004 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3005 | } |
3006 | ||
dd41f596 | 3007 | /* |
43010659 PW |
3008 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3009 | * this_rq, as part of a balancing operation within domain "sd". | |
3010 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3011 | * |
3012 | * Called with both runqueues locked. | |
3013 | */ | |
3014 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3015 | unsigned long max_load_move, |
dd41f596 IM |
3016 | struct sched_domain *sd, enum cpu_idle_type idle, |
3017 | int *all_pinned) | |
3018 | { | |
5522d5d5 | 3019 | const struct sched_class *class = sched_class_highest; |
43010659 | 3020 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3021 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3022 | |
3023 | do { | |
43010659 PW |
3024 | total_load_moved += |
3025 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3026 | max_load_move - total_load_moved, |
a4ac01c3 | 3027 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3028 | class = class->next; |
c4acb2c0 GH |
3029 | |
3030 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3031 | break; | |
3032 | ||
43010659 | 3033 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3034 | |
43010659 PW |
3035 | return total_load_moved > 0; |
3036 | } | |
3037 | ||
e1d1484f PW |
3038 | static int |
3039 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3040 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3041 | struct rq_iterator *iterator) | |
3042 | { | |
3043 | struct task_struct *p = iterator->start(iterator->arg); | |
3044 | int pinned = 0; | |
3045 | ||
3046 | while (p) { | |
3047 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3048 | pull_task(busiest, p, this_rq, this_cpu); | |
3049 | /* | |
3050 | * Right now, this is only the second place pull_task() | |
3051 | * is called, so we can safely collect pull_task() | |
3052 | * stats here rather than inside pull_task(). | |
3053 | */ | |
3054 | schedstat_inc(sd, lb_gained[idle]); | |
3055 | ||
3056 | return 1; | |
3057 | } | |
3058 | p = iterator->next(iterator->arg); | |
3059 | } | |
3060 | ||
3061 | return 0; | |
3062 | } | |
3063 | ||
43010659 PW |
3064 | /* |
3065 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3066 | * part of active balancing operations within "domain". | |
3067 | * Returns 1 if successful and 0 otherwise. | |
3068 | * | |
3069 | * Called with both runqueues locked. | |
3070 | */ | |
3071 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3072 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3073 | { | |
5522d5d5 | 3074 | const struct sched_class *class; |
43010659 PW |
3075 | |
3076 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3077 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3078 | return 1; |
3079 | ||
3080 | return 0; | |
dd41f596 IM |
3081 | } |
3082 | ||
1da177e4 LT |
3083 | /* |
3084 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3085 | * domain. It calculates and returns the amount of weighted load which |
3086 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3087 | */ |
3088 | static struct sched_group * | |
3089 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3090 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3091 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3092 | { |
3093 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3094 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3095 | unsigned long max_pull; |
2dd73a4f PW |
3096 | unsigned long busiest_load_per_task, busiest_nr_running; |
3097 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3098 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3099 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3100 | int power_savings_balance = 1; | |
3101 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3102 | unsigned long min_nr_running = ULONG_MAX; | |
3103 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3104 | #endif | |
1da177e4 LT |
3105 | |
3106 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3107 | busiest_load_per_task = busiest_nr_running = 0; |
3108 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3109 | |
d15bcfdb | 3110 | if (idle == CPU_NOT_IDLE) |
7897986b | 3111 | load_idx = sd->busy_idx; |
d15bcfdb | 3112 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3113 | load_idx = sd->newidle_idx; |
3114 | else | |
3115 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3116 | |
3117 | do { | |
908a7c1b | 3118 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3119 | int local_group; |
3120 | int i; | |
908a7c1b | 3121 | int __group_imb = 0; |
783609c6 | 3122 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3123 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3124 | unsigned long sum_avg_load_per_task; |
3125 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3126 | |
3127 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3128 | ||
783609c6 SS |
3129 | if (local_group) |
3130 | balance_cpu = first_cpu(group->cpumask); | |
3131 | ||
1da177e4 | 3132 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3133 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3134 | sum_avg_load_per_task = avg_load_per_task = 0; |
3135 | ||
908a7c1b KC |
3136 | max_cpu_load = 0; |
3137 | min_cpu_load = ~0UL; | |
1da177e4 | 3138 | |
363ab6f1 | 3139 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3140 | struct rq *rq; |
3141 | ||
3142 | if (!cpu_isset(i, *cpus)) | |
3143 | continue; | |
3144 | ||
3145 | rq = cpu_rq(i); | |
2dd73a4f | 3146 | |
9439aab8 | 3147 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3148 | *sd_idle = 0; |
3149 | ||
1da177e4 | 3150 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3151 | if (local_group) { |
3152 | if (idle_cpu(i) && !first_idle_cpu) { | |
3153 | first_idle_cpu = 1; | |
3154 | balance_cpu = i; | |
3155 | } | |
3156 | ||
a2000572 | 3157 | load = target_load(i, load_idx); |
908a7c1b | 3158 | } else { |
a2000572 | 3159 | load = source_load(i, load_idx); |
908a7c1b KC |
3160 | if (load > max_cpu_load) |
3161 | max_cpu_load = load; | |
3162 | if (min_cpu_load > load) | |
3163 | min_cpu_load = load; | |
3164 | } | |
1da177e4 LT |
3165 | |
3166 | avg_load += load; | |
2dd73a4f | 3167 | sum_nr_running += rq->nr_running; |
dd41f596 | 3168 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3169 | |
3170 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3171 | } |
3172 | ||
783609c6 SS |
3173 | /* |
3174 | * First idle cpu or the first cpu(busiest) in this sched group | |
3175 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3176 | * domains. In the newly idle case, we will allow all the cpu's |
3177 | * to do the newly idle load balance. | |
783609c6 | 3178 | */ |
9439aab8 SS |
3179 | if (idle != CPU_NEWLY_IDLE && local_group && |
3180 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3181 | *balance = 0; |
3182 | goto ret; | |
3183 | } | |
3184 | ||
1da177e4 | 3185 | total_load += avg_load; |
5517d86b | 3186 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3187 | |
3188 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3189 | avg_load = sg_div_cpu_power(group, |
3190 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3191 | |
408ed066 PZ |
3192 | |
3193 | /* | |
3194 | * Consider the group unbalanced when the imbalance is larger | |
3195 | * than the average weight of two tasks. | |
3196 | * | |
3197 | * APZ: with cgroup the avg task weight can vary wildly and | |
3198 | * might not be a suitable number - should we keep a | |
3199 | * normalized nr_running number somewhere that negates | |
3200 | * the hierarchy? | |
3201 | */ | |
3202 | avg_load_per_task = sg_div_cpu_power(group, | |
3203 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3204 | ||
3205 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3206 | __group_imb = 1; |
3207 | ||
5517d86b | 3208 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3209 | |
1da177e4 LT |
3210 | if (local_group) { |
3211 | this_load = avg_load; | |
3212 | this = group; | |
2dd73a4f PW |
3213 | this_nr_running = sum_nr_running; |
3214 | this_load_per_task = sum_weighted_load; | |
3215 | } else if (avg_load > max_load && | |
908a7c1b | 3216 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3217 | max_load = avg_load; |
3218 | busiest = group; | |
2dd73a4f PW |
3219 | busiest_nr_running = sum_nr_running; |
3220 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3221 | group_imb = __group_imb; |
1da177e4 | 3222 | } |
5c45bf27 SS |
3223 | |
3224 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3225 | /* | |
3226 | * Busy processors will not participate in power savings | |
3227 | * balance. | |
3228 | */ | |
dd41f596 IM |
3229 | if (idle == CPU_NOT_IDLE || |
3230 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3231 | goto group_next; | |
5c45bf27 SS |
3232 | |
3233 | /* | |
3234 | * If the local group is idle or completely loaded | |
3235 | * no need to do power savings balance at this domain | |
3236 | */ | |
3237 | if (local_group && (this_nr_running >= group_capacity || | |
3238 | !this_nr_running)) | |
3239 | power_savings_balance = 0; | |
3240 | ||
dd41f596 | 3241 | /* |
5c45bf27 SS |
3242 | * If a group is already running at full capacity or idle, |
3243 | * don't include that group in power savings calculations | |
dd41f596 IM |
3244 | */ |
3245 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3246 | || !sum_nr_running) |
dd41f596 | 3247 | goto group_next; |
5c45bf27 | 3248 | |
dd41f596 | 3249 | /* |
5c45bf27 | 3250 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3251 | * This is the group from where we need to pick up the load |
3252 | * for saving power | |
3253 | */ | |
3254 | if ((sum_nr_running < min_nr_running) || | |
3255 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3256 | first_cpu(group->cpumask) < |
3257 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3258 | group_min = group; |
3259 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3260 | min_load_per_task = sum_weighted_load / |
3261 | sum_nr_running; | |
dd41f596 | 3262 | } |
5c45bf27 | 3263 | |
dd41f596 | 3264 | /* |
5c45bf27 | 3265 | * Calculate the group which is almost near its |
dd41f596 IM |
3266 | * capacity but still has some space to pick up some load |
3267 | * from other group and save more power | |
3268 | */ | |
3269 | if (sum_nr_running <= group_capacity - 1) { | |
3270 | if (sum_nr_running > leader_nr_running || | |
3271 | (sum_nr_running == leader_nr_running && | |
3272 | first_cpu(group->cpumask) > | |
3273 | first_cpu(group_leader->cpumask))) { | |
3274 | group_leader = group; | |
3275 | leader_nr_running = sum_nr_running; | |
3276 | } | |
48f24c4d | 3277 | } |
5c45bf27 SS |
3278 | group_next: |
3279 | #endif | |
1da177e4 LT |
3280 | group = group->next; |
3281 | } while (group != sd->groups); | |
3282 | ||
2dd73a4f | 3283 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3284 | goto out_balanced; |
3285 | ||
3286 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3287 | ||
3288 | if (this_load >= avg_load || | |
3289 | 100*max_load <= sd->imbalance_pct*this_load) | |
3290 | goto out_balanced; | |
3291 | ||
2dd73a4f | 3292 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3293 | if (group_imb) |
3294 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3295 | ||
1da177e4 LT |
3296 | /* |
3297 | * We're trying to get all the cpus to the average_load, so we don't | |
3298 | * want to push ourselves above the average load, nor do we wish to | |
3299 | * reduce the max loaded cpu below the average load, as either of these | |
3300 | * actions would just result in more rebalancing later, and ping-pong | |
3301 | * tasks around. Thus we look for the minimum possible imbalance. | |
3302 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3303 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3304 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3305 | * appear as very large values with unsigned longs. |
3306 | */ | |
2dd73a4f PW |
3307 | if (max_load <= busiest_load_per_task) |
3308 | goto out_balanced; | |
3309 | ||
3310 | /* | |
3311 | * In the presence of smp nice balancing, certain scenarios can have | |
3312 | * max load less than avg load(as we skip the groups at or below | |
3313 | * its cpu_power, while calculating max_load..) | |
3314 | */ | |
3315 | if (max_load < avg_load) { | |
3316 | *imbalance = 0; | |
3317 | goto small_imbalance; | |
3318 | } | |
0c117f1b SS |
3319 | |
3320 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3321 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3322 | |
1da177e4 | 3323 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3324 | *imbalance = min(max_pull * busiest->__cpu_power, |
3325 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3326 | / SCHED_LOAD_SCALE; |
3327 | ||
2dd73a4f PW |
3328 | /* |
3329 | * if *imbalance is less than the average load per runnable task | |
3330 | * there is no gaurantee that any tasks will be moved so we'll have | |
3331 | * a think about bumping its value to force at least one task to be | |
3332 | * moved | |
3333 | */ | |
7fd0d2dd | 3334 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3335 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3336 | unsigned int imbn; |
3337 | ||
3338 | small_imbalance: | |
3339 | pwr_move = pwr_now = 0; | |
3340 | imbn = 2; | |
3341 | if (this_nr_running) { | |
3342 | this_load_per_task /= this_nr_running; | |
3343 | if (busiest_load_per_task > this_load_per_task) | |
3344 | imbn = 1; | |
3345 | } else | |
408ed066 | 3346 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3347 | |
01c8c57d | 3348 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3349 | busiest_load_per_task * imbn) { |
2dd73a4f | 3350 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3351 | return busiest; |
3352 | } | |
3353 | ||
3354 | /* | |
3355 | * OK, we don't have enough imbalance to justify moving tasks, | |
3356 | * however we may be able to increase total CPU power used by | |
3357 | * moving them. | |
3358 | */ | |
3359 | ||
5517d86b ED |
3360 | pwr_now += busiest->__cpu_power * |
3361 | min(busiest_load_per_task, max_load); | |
3362 | pwr_now += this->__cpu_power * | |
3363 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3364 | pwr_now /= SCHED_LOAD_SCALE; |
3365 | ||
3366 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3367 | tmp = sg_div_cpu_power(busiest, |
3368 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3369 | if (max_load > tmp) |
5517d86b | 3370 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3371 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3372 | |
3373 | /* Amount of load we'd add */ | |
5517d86b | 3374 | if (max_load * busiest->__cpu_power < |
33859f7f | 3375 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3376 | tmp = sg_div_cpu_power(this, |
3377 | max_load * busiest->__cpu_power); | |
1da177e4 | 3378 | else |
5517d86b ED |
3379 | tmp = sg_div_cpu_power(this, |
3380 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3381 | pwr_move += this->__cpu_power * | |
3382 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3383 | pwr_move /= SCHED_LOAD_SCALE; |
3384 | ||
3385 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3386 | if (pwr_move > pwr_now) |
3387 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3388 | } |
3389 | ||
1da177e4 LT |
3390 | return busiest; |
3391 | ||
3392 | out_balanced: | |
5c45bf27 | 3393 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3394 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3395 | goto ret; |
1da177e4 | 3396 | |
5c45bf27 SS |
3397 | if (this == group_leader && group_leader != group_min) { |
3398 | *imbalance = min_load_per_task; | |
3399 | return group_min; | |
3400 | } | |
5c45bf27 | 3401 | #endif |
783609c6 | 3402 | ret: |
1da177e4 LT |
3403 | *imbalance = 0; |
3404 | return NULL; | |
3405 | } | |
3406 | ||
3407 | /* | |
3408 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3409 | */ | |
70b97a7f | 3410 | static struct rq * |
d15bcfdb | 3411 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3412 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3413 | { |
70b97a7f | 3414 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3415 | unsigned long max_load = 0; |
1da177e4 LT |
3416 | int i; |
3417 | ||
363ab6f1 | 3418 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3419 | unsigned long wl; |
0a2966b4 CL |
3420 | |
3421 | if (!cpu_isset(i, *cpus)) | |
3422 | continue; | |
3423 | ||
48f24c4d | 3424 | rq = cpu_rq(i); |
dd41f596 | 3425 | wl = weighted_cpuload(i); |
2dd73a4f | 3426 | |
dd41f596 | 3427 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3428 | continue; |
1da177e4 | 3429 | |
dd41f596 IM |
3430 | if (wl > max_load) { |
3431 | max_load = wl; | |
48f24c4d | 3432 | busiest = rq; |
1da177e4 LT |
3433 | } |
3434 | } | |
3435 | ||
3436 | return busiest; | |
3437 | } | |
3438 | ||
77391d71 NP |
3439 | /* |
3440 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3441 | * so long as it is large enough. | |
3442 | */ | |
3443 | #define MAX_PINNED_INTERVAL 512 | |
3444 | ||
1da177e4 LT |
3445 | /* |
3446 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3447 | * tasks if there is an imbalance. | |
1da177e4 | 3448 | */ |
70b97a7f | 3449 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3450 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3451 | int *balance, cpumask_t *cpus) |
1da177e4 | 3452 | { |
43010659 | 3453 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3454 | struct sched_group *group; |
1da177e4 | 3455 | unsigned long imbalance; |
70b97a7f | 3456 | struct rq *busiest; |
fe2eea3f | 3457 | unsigned long flags; |
5969fe06 | 3458 | |
7c16ec58 MT |
3459 | cpus_setall(*cpus); |
3460 | ||
89c4710e SS |
3461 | /* |
3462 | * When power savings policy is enabled for the parent domain, idle | |
3463 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3464 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3465 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3466 | */ |
d15bcfdb | 3467 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3468 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3469 | sd_idle = 1; |
1da177e4 | 3470 | |
2d72376b | 3471 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3472 | |
0a2966b4 | 3473 | redo: |
c8cba857 | 3474 | update_shares(sd); |
0a2966b4 | 3475 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3476 | cpus, balance); |
783609c6 | 3477 | |
06066714 | 3478 | if (*balance == 0) |
783609c6 | 3479 | goto out_balanced; |
783609c6 | 3480 | |
1da177e4 LT |
3481 | if (!group) { |
3482 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3483 | goto out_balanced; | |
3484 | } | |
3485 | ||
7c16ec58 | 3486 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3487 | if (!busiest) { |
3488 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3489 | goto out_balanced; | |
3490 | } | |
3491 | ||
db935dbd | 3492 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3493 | |
3494 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3495 | ||
43010659 | 3496 | ld_moved = 0; |
1da177e4 LT |
3497 | if (busiest->nr_running > 1) { |
3498 | /* | |
3499 | * Attempt to move tasks. If find_busiest_group has found | |
3500 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3501 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3502 | * correctly treated as an imbalance. |
3503 | */ | |
fe2eea3f | 3504 | local_irq_save(flags); |
e17224bf | 3505 | double_rq_lock(this_rq, busiest); |
43010659 | 3506 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3507 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3508 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3509 | local_irq_restore(flags); |
81026794 | 3510 | |
46cb4b7c SS |
3511 | /* |
3512 | * some other cpu did the load balance for us. | |
3513 | */ | |
43010659 | 3514 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3515 | resched_cpu(this_cpu); |
3516 | ||
81026794 | 3517 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3518 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3519 | cpu_clear(cpu_of(busiest), *cpus); |
3520 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3521 | goto redo; |
81026794 | 3522 | goto out_balanced; |
0a2966b4 | 3523 | } |
1da177e4 | 3524 | } |
81026794 | 3525 | |
43010659 | 3526 | if (!ld_moved) { |
1da177e4 LT |
3527 | schedstat_inc(sd, lb_failed[idle]); |
3528 | sd->nr_balance_failed++; | |
3529 | ||
3530 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3531 | |
fe2eea3f | 3532 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3533 | |
3534 | /* don't kick the migration_thread, if the curr | |
3535 | * task on busiest cpu can't be moved to this_cpu | |
3536 | */ | |
3537 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3538 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3539 | all_pinned = 1; |
3540 | goto out_one_pinned; | |
3541 | } | |
3542 | ||
1da177e4 LT |
3543 | if (!busiest->active_balance) { |
3544 | busiest->active_balance = 1; | |
3545 | busiest->push_cpu = this_cpu; | |
81026794 | 3546 | active_balance = 1; |
1da177e4 | 3547 | } |
fe2eea3f | 3548 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3549 | if (active_balance) |
1da177e4 LT |
3550 | wake_up_process(busiest->migration_thread); |
3551 | ||
3552 | /* | |
3553 | * We've kicked active balancing, reset the failure | |
3554 | * counter. | |
3555 | */ | |
39507451 | 3556 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3557 | } |
81026794 | 3558 | } else |
1da177e4 LT |
3559 | sd->nr_balance_failed = 0; |
3560 | ||
81026794 | 3561 | if (likely(!active_balance)) { |
1da177e4 LT |
3562 | /* We were unbalanced, so reset the balancing interval */ |
3563 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3564 | } else { |
3565 | /* | |
3566 | * If we've begun active balancing, start to back off. This | |
3567 | * case may not be covered by the all_pinned logic if there | |
3568 | * is only 1 task on the busy runqueue (because we don't call | |
3569 | * move_tasks). | |
3570 | */ | |
3571 | if (sd->balance_interval < sd->max_interval) | |
3572 | sd->balance_interval *= 2; | |
1da177e4 LT |
3573 | } |
3574 | ||
43010659 | 3575 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3576 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3577 | ld_moved = -1; |
3578 | ||
3579 | goto out; | |
1da177e4 LT |
3580 | |
3581 | out_balanced: | |
1da177e4 LT |
3582 | schedstat_inc(sd, lb_balanced[idle]); |
3583 | ||
16cfb1c0 | 3584 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3585 | |
3586 | out_one_pinned: | |
1da177e4 | 3587 | /* tune up the balancing interval */ |
77391d71 NP |
3588 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3589 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3590 | sd->balance_interval *= 2; |
3591 | ||
48f24c4d | 3592 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3593 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3594 | ld_moved = -1; |
3595 | else | |
3596 | ld_moved = 0; | |
3597 | out: | |
c8cba857 PZ |
3598 | if (ld_moved) |
3599 | update_shares(sd); | |
c09595f6 | 3600 | return ld_moved; |
1da177e4 LT |
3601 | } |
3602 | ||
3603 | /* | |
3604 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3605 | * tasks if there is an imbalance. | |
3606 | * | |
d15bcfdb | 3607 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3608 | * this_rq is locked. |
3609 | */ | |
48f24c4d | 3610 | static int |
7c16ec58 MT |
3611 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3612 | cpumask_t *cpus) | |
1da177e4 LT |
3613 | { |
3614 | struct sched_group *group; | |
70b97a7f | 3615 | struct rq *busiest = NULL; |
1da177e4 | 3616 | unsigned long imbalance; |
43010659 | 3617 | int ld_moved = 0; |
5969fe06 | 3618 | int sd_idle = 0; |
969bb4e4 | 3619 | int all_pinned = 0; |
7c16ec58 MT |
3620 | |
3621 | cpus_setall(*cpus); | |
5969fe06 | 3622 | |
89c4710e SS |
3623 | /* |
3624 | * When power savings policy is enabled for the parent domain, idle | |
3625 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3626 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3627 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3628 | */ |
3629 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3630 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3631 | sd_idle = 1; |
1da177e4 | 3632 | |
2d72376b | 3633 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3634 | redo: |
3e5459b4 | 3635 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3636 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3637 | &sd_idle, cpus, NULL); |
1da177e4 | 3638 | if (!group) { |
d15bcfdb | 3639 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3640 | goto out_balanced; |
1da177e4 LT |
3641 | } |
3642 | ||
7c16ec58 | 3643 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3644 | if (!busiest) { |
d15bcfdb | 3645 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3646 | goto out_balanced; |
1da177e4 LT |
3647 | } |
3648 | ||
db935dbd NP |
3649 | BUG_ON(busiest == this_rq); |
3650 | ||
d15bcfdb | 3651 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3652 | |
43010659 | 3653 | ld_moved = 0; |
d6d5cfaf NP |
3654 | if (busiest->nr_running > 1) { |
3655 | /* Attempt to move tasks */ | |
3656 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3657 | /* this_rq->clock is already updated */ |
3658 | update_rq_clock(busiest); | |
43010659 | 3659 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3660 | imbalance, sd, CPU_NEWLY_IDLE, |
3661 | &all_pinned); | |
1b12bbc7 | 3662 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3663 | |
969bb4e4 | 3664 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3665 | cpu_clear(cpu_of(busiest), *cpus); |
3666 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3667 | goto redo; |
3668 | } | |
d6d5cfaf NP |
3669 | } |
3670 | ||
43010659 | 3671 | if (!ld_moved) { |
d15bcfdb | 3672 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3673 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3674 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3675 | return -1; |
3676 | } else | |
16cfb1c0 | 3677 | sd->nr_balance_failed = 0; |
1da177e4 | 3678 | |
3e5459b4 | 3679 | update_shares_locked(this_rq, sd); |
43010659 | 3680 | return ld_moved; |
16cfb1c0 NP |
3681 | |
3682 | out_balanced: | |
d15bcfdb | 3683 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3684 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3685 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3686 | return -1; |
16cfb1c0 | 3687 | sd->nr_balance_failed = 0; |
48f24c4d | 3688 | |
16cfb1c0 | 3689 | return 0; |
1da177e4 LT |
3690 | } |
3691 | ||
3692 | /* | |
3693 | * idle_balance is called by schedule() if this_cpu is about to become | |
3694 | * idle. Attempts to pull tasks from other CPUs. | |
3695 | */ | |
70b97a7f | 3696 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3697 | { |
3698 | struct sched_domain *sd; | |
dd41f596 IM |
3699 | int pulled_task = -1; |
3700 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3701 | cpumask_t tmpmask; |
1da177e4 LT |
3702 | |
3703 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3704 | unsigned long interval; |
3705 | ||
3706 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3707 | continue; | |
3708 | ||
3709 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3710 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3711 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3712 | sd, &tmpmask); | |
92c4ca5c CL |
3713 | |
3714 | interval = msecs_to_jiffies(sd->balance_interval); | |
3715 | if (time_after(next_balance, sd->last_balance + interval)) | |
3716 | next_balance = sd->last_balance + interval; | |
3717 | if (pulled_task) | |
3718 | break; | |
1da177e4 | 3719 | } |
dd41f596 | 3720 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3721 | /* |
3722 | * We are going idle. next_balance may be set based on | |
3723 | * a busy processor. So reset next_balance. | |
3724 | */ | |
3725 | this_rq->next_balance = next_balance; | |
dd41f596 | 3726 | } |
1da177e4 LT |
3727 | } |
3728 | ||
3729 | /* | |
3730 | * active_load_balance is run by migration threads. It pushes running tasks | |
3731 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3732 | * running on each physical CPU where possible, and avoids physical / | |
3733 | * logical imbalances. | |
3734 | * | |
3735 | * Called with busiest_rq locked. | |
3736 | */ | |
70b97a7f | 3737 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3738 | { |
39507451 | 3739 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3740 | struct sched_domain *sd; |
3741 | struct rq *target_rq; | |
39507451 | 3742 | |
48f24c4d | 3743 | /* Is there any task to move? */ |
39507451 | 3744 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3745 | return; |
3746 | ||
3747 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3748 | |
3749 | /* | |
39507451 | 3750 | * This condition is "impossible", if it occurs |
41a2d6cf | 3751 | * we need to fix it. Originally reported by |
39507451 | 3752 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3753 | */ |
39507451 | 3754 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3755 | |
39507451 NP |
3756 | /* move a task from busiest_rq to target_rq */ |
3757 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3758 | update_rq_clock(busiest_rq); |
3759 | update_rq_clock(target_rq); | |
39507451 NP |
3760 | |
3761 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3762 | for_each_domain(target_cpu, sd) { |
39507451 | 3763 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3764 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3765 | break; |
c96d145e | 3766 | } |
39507451 | 3767 | |
48f24c4d | 3768 | if (likely(sd)) { |
2d72376b | 3769 | schedstat_inc(sd, alb_count); |
39507451 | 3770 | |
43010659 PW |
3771 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3772 | sd, CPU_IDLE)) | |
48f24c4d IM |
3773 | schedstat_inc(sd, alb_pushed); |
3774 | else | |
3775 | schedstat_inc(sd, alb_failed); | |
3776 | } | |
1b12bbc7 | 3777 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3778 | } |
3779 | ||
46cb4b7c SS |
3780 | #ifdef CONFIG_NO_HZ |
3781 | static struct { | |
3782 | atomic_t load_balancer; | |
41a2d6cf | 3783 | cpumask_t cpu_mask; |
46cb4b7c SS |
3784 | } nohz ____cacheline_aligned = { |
3785 | .load_balancer = ATOMIC_INIT(-1), | |
3786 | .cpu_mask = CPU_MASK_NONE, | |
3787 | }; | |
3788 | ||
7835b98b | 3789 | /* |
46cb4b7c SS |
3790 | * This routine will try to nominate the ilb (idle load balancing) |
3791 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3792 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3793 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3794 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3795 | * arrives... | |
3796 | * | |
3797 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3798 | * for idle load balancing. ilb owner will still be part of | |
3799 | * nohz.cpu_mask.. | |
7835b98b | 3800 | * |
46cb4b7c SS |
3801 | * While stopping the tick, this cpu will become the ilb owner if there |
3802 | * is no other owner. And will be the owner till that cpu becomes busy | |
3803 | * or if all cpus in the system stop their ticks at which point | |
3804 | * there is no need for ilb owner. | |
3805 | * | |
3806 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3807 | * next busy scheduler_tick() | |
3808 | */ | |
3809 | int select_nohz_load_balancer(int stop_tick) | |
3810 | { | |
3811 | int cpu = smp_processor_id(); | |
3812 | ||
3813 | if (stop_tick) { | |
3814 | cpu_set(cpu, nohz.cpu_mask); | |
3815 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3816 | ||
3817 | /* | |
3818 | * If we are going offline and still the leader, give up! | |
3819 | */ | |
e761b772 | 3820 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3821 | atomic_read(&nohz.load_balancer) == cpu) { |
3822 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3823 | BUG(); | |
3824 | return 0; | |
3825 | } | |
3826 | ||
3827 | /* time for ilb owner also to sleep */ | |
3828 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3829 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3830 | atomic_set(&nohz.load_balancer, -1); | |
3831 | return 0; | |
3832 | } | |
3833 | ||
3834 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3835 | /* make me the ilb owner */ | |
3836 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3837 | return 1; | |
3838 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3839 | return 1; | |
3840 | } else { | |
3841 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3842 | return 0; | |
3843 | ||
3844 | cpu_clear(cpu, nohz.cpu_mask); | |
3845 | ||
3846 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3847 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3848 | BUG(); | |
3849 | } | |
3850 | return 0; | |
3851 | } | |
3852 | #endif | |
3853 | ||
3854 | static DEFINE_SPINLOCK(balancing); | |
3855 | ||
3856 | /* | |
7835b98b CL |
3857 | * It checks each scheduling domain to see if it is due to be balanced, |
3858 | * and initiates a balancing operation if so. | |
3859 | * | |
3860 | * Balancing parameters are set up in arch_init_sched_domains. | |
3861 | */ | |
a9957449 | 3862 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3863 | { |
46cb4b7c SS |
3864 | int balance = 1; |
3865 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3866 | unsigned long interval; |
3867 | struct sched_domain *sd; | |
46cb4b7c | 3868 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3869 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3870 | int update_next_balance = 0; |
d07355f5 | 3871 | int need_serialize; |
7c16ec58 | 3872 | cpumask_t tmp; |
1da177e4 | 3873 | |
46cb4b7c | 3874 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3875 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3876 | continue; | |
3877 | ||
3878 | interval = sd->balance_interval; | |
d15bcfdb | 3879 | if (idle != CPU_IDLE) |
1da177e4 LT |
3880 | interval *= sd->busy_factor; |
3881 | ||
3882 | /* scale ms to jiffies */ | |
3883 | interval = msecs_to_jiffies(interval); | |
3884 | if (unlikely(!interval)) | |
3885 | interval = 1; | |
dd41f596 IM |
3886 | if (interval > HZ*NR_CPUS/10) |
3887 | interval = HZ*NR_CPUS/10; | |
3888 | ||
d07355f5 | 3889 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3890 | |
d07355f5 | 3891 | if (need_serialize) { |
08c183f3 CL |
3892 | if (!spin_trylock(&balancing)) |
3893 | goto out; | |
3894 | } | |
3895 | ||
c9819f45 | 3896 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3897 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3898 | /* |
3899 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3900 | * longer idle, or one of our SMT siblings is |
3901 | * not idle. | |
3902 | */ | |
d15bcfdb | 3903 | idle = CPU_NOT_IDLE; |
1da177e4 | 3904 | } |
1bd77f2d | 3905 | sd->last_balance = jiffies; |
1da177e4 | 3906 | } |
d07355f5 | 3907 | if (need_serialize) |
08c183f3 CL |
3908 | spin_unlock(&balancing); |
3909 | out: | |
f549da84 | 3910 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3911 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3912 | update_next_balance = 1; |
3913 | } | |
783609c6 SS |
3914 | |
3915 | /* | |
3916 | * Stop the load balance at this level. There is another | |
3917 | * CPU in our sched group which is doing load balancing more | |
3918 | * actively. | |
3919 | */ | |
3920 | if (!balance) | |
3921 | break; | |
1da177e4 | 3922 | } |
f549da84 SS |
3923 | |
3924 | /* | |
3925 | * next_balance will be updated only when there is a need. | |
3926 | * When the cpu is attached to null domain for ex, it will not be | |
3927 | * updated. | |
3928 | */ | |
3929 | if (likely(update_next_balance)) | |
3930 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3931 | } |
3932 | ||
3933 | /* | |
3934 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3935 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3936 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3937 | */ | |
3938 | static void run_rebalance_domains(struct softirq_action *h) | |
3939 | { | |
dd41f596 IM |
3940 | int this_cpu = smp_processor_id(); |
3941 | struct rq *this_rq = cpu_rq(this_cpu); | |
3942 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3943 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3944 | |
dd41f596 | 3945 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3946 | |
3947 | #ifdef CONFIG_NO_HZ | |
3948 | /* | |
3949 | * If this cpu is the owner for idle load balancing, then do the | |
3950 | * balancing on behalf of the other idle cpus whose ticks are | |
3951 | * stopped. | |
3952 | */ | |
dd41f596 IM |
3953 | if (this_rq->idle_at_tick && |
3954 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3955 | cpumask_t cpus = nohz.cpu_mask; |
3956 | struct rq *rq; | |
3957 | int balance_cpu; | |
3958 | ||
dd41f596 | 3959 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3960 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3961 | /* |
3962 | * If this cpu gets work to do, stop the load balancing | |
3963 | * work being done for other cpus. Next load | |
3964 | * balancing owner will pick it up. | |
3965 | */ | |
3966 | if (need_resched()) | |
3967 | break; | |
3968 | ||
de0cf899 | 3969 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3970 | |
3971 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3972 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3973 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3974 | } |
3975 | } | |
3976 | #endif | |
3977 | } | |
3978 | ||
3979 | /* | |
3980 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3981 | * | |
3982 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3983 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3984 | * if the whole system is idle. | |
3985 | */ | |
dd41f596 | 3986 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3987 | { |
46cb4b7c SS |
3988 | #ifdef CONFIG_NO_HZ |
3989 | /* | |
3990 | * If we were in the nohz mode recently and busy at the current | |
3991 | * scheduler tick, then check if we need to nominate new idle | |
3992 | * load balancer. | |
3993 | */ | |
3994 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3995 | rq->in_nohz_recently = 0; | |
3996 | ||
3997 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3998 | cpu_clear(cpu, nohz.cpu_mask); | |
3999 | atomic_set(&nohz.load_balancer, -1); | |
4000 | } | |
4001 | ||
4002 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4003 | /* | |
4004 | * simple selection for now: Nominate the | |
4005 | * first cpu in the nohz list to be the next | |
4006 | * ilb owner. | |
4007 | * | |
4008 | * TBD: Traverse the sched domains and nominate | |
4009 | * the nearest cpu in the nohz.cpu_mask. | |
4010 | */ | |
4011 | int ilb = first_cpu(nohz.cpu_mask); | |
4012 | ||
434d53b0 | 4013 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4014 | resched_cpu(ilb); |
4015 | } | |
4016 | } | |
4017 | ||
4018 | /* | |
4019 | * If this cpu is idle and doing idle load balancing for all the | |
4020 | * cpus with ticks stopped, is it time for that to stop? | |
4021 | */ | |
4022 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4023 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4024 | resched_cpu(cpu); | |
4025 | return; | |
4026 | } | |
4027 | ||
4028 | /* | |
4029 | * If this cpu is idle and the idle load balancing is done by | |
4030 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4031 | */ | |
4032 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4033 | cpu_isset(cpu, nohz.cpu_mask)) | |
4034 | return; | |
4035 | #endif | |
4036 | if (time_after_eq(jiffies, rq->next_balance)) | |
4037 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4038 | } |
dd41f596 IM |
4039 | |
4040 | #else /* CONFIG_SMP */ | |
4041 | ||
1da177e4 LT |
4042 | /* |
4043 | * on UP we do not need to balance between CPUs: | |
4044 | */ | |
70b97a7f | 4045 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4046 | { |
4047 | } | |
dd41f596 | 4048 | |
1da177e4 LT |
4049 | #endif |
4050 | ||
1da177e4 LT |
4051 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4052 | ||
4053 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4054 | ||
4055 | /* | |
f06febc9 FM |
4056 | * Return any ns on the sched_clock that have not yet been banked in |
4057 | * @p in case that task is currently running. | |
1da177e4 | 4058 | */ |
bb34d92f | 4059 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4060 | { |
1da177e4 | 4061 | unsigned long flags; |
41b86e9c | 4062 | struct rq *rq; |
bb34d92f | 4063 | u64 ns = 0; |
48f24c4d | 4064 | |
41b86e9c | 4065 | rq = task_rq_lock(p, &flags); |
1508487e | 4066 | |
051a1d1a | 4067 | if (task_current(rq, p)) { |
f06febc9 FM |
4068 | u64 delta_exec; |
4069 | ||
a8e504d2 IM |
4070 | update_rq_clock(rq); |
4071 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4072 | if ((s64)delta_exec > 0) |
bb34d92f | 4073 | ns = delta_exec; |
41b86e9c | 4074 | } |
48f24c4d | 4075 | |
41b86e9c | 4076 | task_rq_unlock(rq, &flags); |
48f24c4d | 4077 | |
1da177e4 LT |
4078 | return ns; |
4079 | } | |
4080 | ||
1da177e4 LT |
4081 | /* |
4082 | * Account user cpu time to a process. | |
4083 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4084 | * @cputime: the cpu time spent in user space since the last update |
4085 | */ | |
4086 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4087 | { | |
4088 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4089 | cputime64_t tmp; | |
4090 | ||
4091 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4092 | account_group_user_time(p, cputime); |
1da177e4 LT |
4093 | |
4094 | /* Add user time to cpustat. */ | |
4095 | tmp = cputime_to_cputime64(cputime); | |
4096 | if (TASK_NICE(p) > 0) | |
4097 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4098 | else | |
4099 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4100 | /* Account for user time used */ |
4101 | acct_update_integrals(p); | |
1da177e4 LT |
4102 | } |
4103 | ||
94886b84 LV |
4104 | /* |
4105 | * Account guest cpu time to a process. | |
4106 | * @p: the process that the cpu time gets accounted to | |
4107 | * @cputime: the cpu time spent in virtual machine since the last update | |
4108 | */ | |
f7402e03 | 4109 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4110 | { |
4111 | cputime64_t tmp; | |
4112 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4113 | ||
4114 | tmp = cputime_to_cputime64(cputime); | |
4115 | ||
4116 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4117 | account_group_user_time(p, cputime); |
94886b84 LV |
4118 | p->gtime = cputime_add(p->gtime, cputime); |
4119 | ||
4120 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4121 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4122 | } | |
4123 | ||
c66f08be MN |
4124 | /* |
4125 | * Account scaled user cpu time to a process. | |
4126 | * @p: the process that the cpu time gets accounted to | |
4127 | * @cputime: the cpu time spent in user space since the last update | |
4128 | */ | |
4129 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4130 | { | |
4131 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4132 | } | |
4133 | ||
1da177e4 LT |
4134 | /* |
4135 | * Account system cpu time to a process. | |
4136 | * @p: the process that the cpu time gets accounted to | |
4137 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4138 | * @cputime: the cpu time spent in kernel space since the last update | |
4139 | */ | |
4140 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4141 | cputime_t cputime) | |
4142 | { | |
4143 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4144 | struct rq *rq = this_rq(); |
1da177e4 LT |
4145 | cputime64_t tmp; |
4146 | ||
983ed7a6 HH |
4147 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4148 | account_guest_time(p, cputime); | |
4149 | return; | |
4150 | } | |
94886b84 | 4151 | |
1da177e4 | 4152 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4153 | account_group_system_time(p, cputime); |
1da177e4 LT |
4154 | |
4155 | /* Add system time to cpustat. */ | |
4156 | tmp = cputime_to_cputime64(cputime); | |
4157 | if (hardirq_count() - hardirq_offset) | |
4158 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4159 | else if (softirq_count()) | |
4160 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4161 | else if (p != rq->idle) |
1da177e4 | 4162 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4163 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4164 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4165 | else | |
4166 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4167 | /* Account for system time used */ | |
4168 | acct_update_integrals(p); | |
1da177e4 LT |
4169 | } |
4170 | ||
c66f08be MN |
4171 | /* |
4172 | * Account scaled system cpu time to a process. | |
4173 | * @p: the process that the cpu time gets accounted to | |
4174 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4175 | * @cputime: the cpu time spent in kernel space since the last update | |
4176 | */ | |
4177 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4178 | { | |
4179 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4180 | } | |
4181 | ||
1da177e4 LT |
4182 | /* |
4183 | * Account for involuntary wait time. | |
4184 | * @p: the process from which the cpu time has been stolen | |
4185 | * @steal: the cpu time spent in involuntary wait | |
4186 | */ | |
4187 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4188 | { | |
4189 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4190 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4191 | struct rq *rq = this_rq(); |
1da177e4 LT |
4192 | |
4193 | if (p == rq->idle) { | |
4194 | p->stime = cputime_add(p->stime, steal); | |
f06febc9 | 4195 | account_group_system_time(p, steal); |
1da177e4 LT |
4196 | if (atomic_read(&rq->nr_iowait) > 0) |
4197 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4198 | else | |
4199 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4200 | } else |
1da177e4 LT |
4201 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4202 | } | |
4203 | ||
49048622 BS |
4204 | /* |
4205 | * Use precise platform statistics if available: | |
4206 | */ | |
4207 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4208 | cputime_t task_utime(struct task_struct *p) | |
4209 | { | |
4210 | return p->utime; | |
4211 | } | |
4212 | ||
4213 | cputime_t task_stime(struct task_struct *p) | |
4214 | { | |
4215 | return p->stime; | |
4216 | } | |
4217 | #else | |
4218 | cputime_t task_utime(struct task_struct *p) | |
4219 | { | |
4220 | clock_t utime = cputime_to_clock_t(p->utime), | |
4221 | total = utime + cputime_to_clock_t(p->stime); | |
4222 | u64 temp; | |
4223 | ||
4224 | /* | |
4225 | * Use CFS's precise accounting: | |
4226 | */ | |
4227 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4228 | ||
4229 | if (total) { | |
4230 | temp *= utime; | |
4231 | do_div(temp, total); | |
4232 | } | |
4233 | utime = (clock_t)temp; | |
4234 | ||
4235 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4236 | return p->prev_utime; | |
4237 | } | |
4238 | ||
4239 | cputime_t task_stime(struct task_struct *p) | |
4240 | { | |
4241 | clock_t stime; | |
4242 | ||
4243 | /* | |
4244 | * Use CFS's precise accounting. (we subtract utime from | |
4245 | * the total, to make sure the total observed by userspace | |
4246 | * grows monotonically - apps rely on that): | |
4247 | */ | |
4248 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4249 | cputime_to_clock_t(task_utime(p)); | |
4250 | ||
4251 | if (stime >= 0) | |
4252 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4253 | ||
4254 | return p->prev_stime; | |
4255 | } | |
4256 | #endif | |
4257 | ||
4258 | inline cputime_t task_gtime(struct task_struct *p) | |
4259 | { | |
4260 | return p->gtime; | |
4261 | } | |
4262 | ||
7835b98b CL |
4263 | /* |
4264 | * This function gets called by the timer code, with HZ frequency. | |
4265 | * We call it with interrupts disabled. | |
4266 | * | |
4267 | * It also gets called by the fork code, when changing the parent's | |
4268 | * timeslices. | |
4269 | */ | |
4270 | void scheduler_tick(void) | |
4271 | { | |
7835b98b CL |
4272 | int cpu = smp_processor_id(); |
4273 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4274 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4275 | |
4276 | sched_clock_tick(); | |
dd41f596 IM |
4277 | |
4278 | spin_lock(&rq->lock); | |
3e51f33f | 4279 | update_rq_clock(rq); |
f1a438d8 | 4280 | update_cpu_load(rq); |
fa85ae24 | 4281 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4282 | spin_unlock(&rq->lock); |
7835b98b | 4283 | |
e418e1c2 | 4284 | #ifdef CONFIG_SMP |
dd41f596 IM |
4285 | rq->idle_at_tick = idle_cpu(cpu); |
4286 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4287 | #endif |
1da177e4 LT |
4288 | } |
4289 | ||
6cd8a4bb SR |
4290 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4291 | defined(CONFIG_PREEMPT_TRACER)) | |
4292 | ||
4293 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4294 | { | |
4295 | if (in_lock_functions(addr)) { | |
4296 | addr = CALLER_ADDR2; | |
4297 | if (in_lock_functions(addr)) | |
4298 | addr = CALLER_ADDR3; | |
4299 | } | |
4300 | return addr; | |
4301 | } | |
1da177e4 | 4302 | |
43627582 | 4303 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4304 | { |
6cd8a4bb | 4305 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4306 | /* |
4307 | * Underflow? | |
4308 | */ | |
9a11b49a IM |
4309 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4310 | return; | |
6cd8a4bb | 4311 | #endif |
1da177e4 | 4312 | preempt_count() += val; |
6cd8a4bb | 4313 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4314 | /* |
4315 | * Spinlock count overflowing soon? | |
4316 | */ | |
33859f7f MOS |
4317 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4318 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4319 | #endif |
4320 | if (preempt_count() == val) | |
4321 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4322 | } |
4323 | EXPORT_SYMBOL(add_preempt_count); | |
4324 | ||
43627582 | 4325 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4326 | { |
6cd8a4bb | 4327 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4328 | /* |
4329 | * Underflow? | |
4330 | */ | |
9a11b49a IM |
4331 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4332 | return; | |
1da177e4 LT |
4333 | /* |
4334 | * Is the spinlock portion underflowing? | |
4335 | */ | |
9a11b49a IM |
4336 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4337 | !(preempt_count() & PREEMPT_MASK))) | |
4338 | return; | |
6cd8a4bb | 4339 | #endif |
9a11b49a | 4340 | |
6cd8a4bb SR |
4341 | if (preempt_count() == val) |
4342 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4343 | preempt_count() -= val; |
4344 | } | |
4345 | EXPORT_SYMBOL(sub_preempt_count); | |
4346 | ||
4347 | #endif | |
4348 | ||
4349 | /* | |
dd41f596 | 4350 | * Print scheduling while atomic bug: |
1da177e4 | 4351 | */ |
dd41f596 | 4352 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4353 | { |
838225b4 SS |
4354 | struct pt_regs *regs = get_irq_regs(); |
4355 | ||
4356 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4357 | prev->comm, prev->pid, preempt_count()); | |
4358 | ||
dd41f596 | 4359 | debug_show_held_locks(prev); |
e21f5b15 | 4360 | print_modules(); |
dd41f596 IM |
4361 | if (irqs_disabled()) |
4362 | print_irqtrace_events(prev); | |
838225b4 SS |
4363 | |
4364 | if (regs) | |
4365 | show_regs(regs); | |
4366 | else | |
4367 | dump_stack(); | |
dd41f596 | 4368 | } |
1da177e4 | 4369 | |
dd41f596 IM |
4370 | /* |
4371 | * Various schedule()-time debugging checks and statistics: | |
4372 | */ | |
4373 | static inline void schedule_debug(struct task_struct *prev) | |
4374 | { | |
1da177e4 | 4375 | /* |
41a2d6cf | 4376 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4377 | * schedule() atomically, we ignore that path for now. |
4378 | * Otherwise, whine if we are scheduling when we should not be. | |
4379 | */ | |
3f33a7ce | 4380 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4381 | __schedule_bug(prev); |
4382 | ||
1da177e4 LT |
4383 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4384 | ||
2d72376b | 4385 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4386 | #ifdef CONFIG_SCHEDSTATS |
4387 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4388 | schedstat_inc(this_rq(), bkl_count); |
4389 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4390 | } |
4391 | #endif | |
dd41f596 IM |
4392 | } |
4393 | ||
4394 | /* | |
4395 | * Pick up the highest-prio task: | |
4396 | */ | |
4397 | static inline struct task_struct * | |
ff95f3df | 4398 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4399 | { |
5522d5d5 | 4400 | const struct sched_class *class; |
dd41f596 | 4401 | struct task_struct *p; |
1da177e4 LT |
4402 | |
4403 | /* | |
dd41f596 IM |
4404 | * Optimization: we know that if all tasks are in |
4405 | * the fair class we can call that function directly: | |
1da177e4 | 4406 | */ |
dd41f596 | 4407 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4408 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4409 | if (likely(p)) |
4410 | return p; | |
1da177e4 LT |
4411 | } |
4412 | ||
dd41f596 IM |
4413 | class = sched_class_highest; |
4414 | for ( ; ; ) { | |
fb8d4724 | 4415 | p = class->pick_next_task(rq); |
dd41f596 IM |
4416 | if (p) |
4417 | return p; | |
4418 | /* | |
4419 | * Will never be NULL as the idle class always | |
4420 | * returns a non-NULL p: | |
4421 | */ | |
4422 | class = class->next; | |
4423 | } | |
4424 | } | |
1da177e4 | 4425 | |
dd41f596 IM |
4426 | /* |
4427 | * schedule() is the main scheduler function. | |
4428 | */ | |
4429 | asmlinkage void __sched schedule(void) | |
4430 | { | |
4431 | struct task_struct *prev, *next; | |
67ca7bde | 4432 | unsigned long *switch_count; |
dd41f596 | 4433 | struct rq *rq; |
31656519 | 4434 | int cpu; |
dd41f596 IM |
4435 | |
4436 | need_resched: | |
4437 | preempt_disable(); | |
4438 | cpu = smp_processor_id(); | |
4439 | rq = cpu_rq(cpu); | |
4440 | rcu_qsctr_inc(cpu); | |
4441 | prev = rq->curr; | |
4442 | switch_count = &prev->nivcsw; | |
4443 | ||
4444 | release_kernel_lock(prev); | |
4445 | need_resched_nonpreemptible: | |
4446 | ||
4447 | schedule_debug(prev); | |
1da177e4 | 4448 | |
31656519 | 4449 | if (sched_feat(HRTICK)) |
f333fdc9 | 4450 | hrtick_clear(rq); |
8f4d37ec | 4451 | |
8cd162ce | 4452 | spin_lock_irq(&rq->lock); |
3e51f33f | 4453 | update_rq_clock(rq); |
1e819950 | 4454 | clear_tsk_need_resched(prev); |
1da177e4 | 4455 | |
1da177e4 | 4456 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4457 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4458 | prev->state = TASK_RUNNING; |
16882c1e | 4459 | else |
2e1cb74a | 4460 | deactivate_task(rq, prev, 1); |
dd41f596 | 4461 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4462 | } |
4463 | ||
9a897c5a SR |
4464 | #ifdef CONFIG_SMP |
4465 | if (prev->sched_class->pre_schedule) | |
4466 | prev->sched_class->pre_schedule(rq, prev); | |
4467 | #endif | |
f65eda4f | 4468 | |
dd41f596 | 4469 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4470 | idle_balance(cpu, rq); |
1da177e4 | 4471 | |
31ee529c | 4472 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4473 | next = pick_next_task(rq, prev); |
1da177e4 | 4474 | |
1da177e4 | 4475 | if (likely(prev != next)) { |
673a90a1 DS |
4476 | sched_info_switch(prev, next); |
4477 | ||
1da177e4 LT |
4478 | rq->nr_switches++; |
4479 | rq->curr = next; | |
4480 | ++*switch_count; | |
4481 | ||
dd41f596 | 4482 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4483 | /* |
4484 | * the context switch might have flipped the stack from under | |
4485 | * us, hence refresh the local variables. | |
4486 | */ | |
4487 | cpu = smp_processor_id(); | |
4488 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4489 | } else |
4490 | spin_unlock_irq(&rq->lock); | |
4491 | ||
8f4d37ec | 4492 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4493 | goto need_resched_nonpreemptible; |
8f4d37ec | 4494 | |
1da177e4 LT |
4495 | preempt_enable_no_resched(); |
4496 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4497 | goto need_resched; | |
4498 | } | |
1da177e4 LT |
4499 | EXPORT_SYMBOL(schedule); |
4500 | ||
4501 | #ifdef CONFIG_PREEMPT | |
4502 | /* | |
2ed6e34f | 4503 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4504 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4505 | * occur there and call schedule directly. |
4506 | */ | |
4507 | asmlinkage void __sched preempt_schedule(void) | |
4508 | { | |
4509 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4510 | |
1da177e4 LT |
4511 | /* |
4512 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4513 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4514 | */ |
beed33a8 | 4515 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4516 | return; |
4517 | ||
3a5c359a AK |
4518 | do { |
4519 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4520 | schedule(); |
3a5c359a | 4521 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4522 | |
3a5c359a AK |
4523 | /* |
4524 | * Check again in case we missed a preemption opportunity | |
4525 | * between schedule and now. | |
4526 | */ | |
4527 | barrier(); | |
4528 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4529 | } |
1da177e4 LT |
4530 | EXPORT_SYMBOL(preempt_schedule); |
4531 | ||
4532 | /* | |
2ed6e34f | 4533 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4534 | * off of irq context. |
4535 | * Note, that this is called and return with irqs disabled. This will | |
4536 | * protect us against recursive calling from irq. | |
4537 | */ | |
4538 | asmlinkage void __sched preempt_schedule_irq(void) | |
4539 | { | |
4540 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4541 | |
2ed6e34f | 4542 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4543 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4544 | ||
3a5c359a AK |
4545 | do { |
4546 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4547 | local_irq_enable(); |
4548 | schedule(); | |
4549 | local_irq_disable(); | |
3a5c359a | 4550 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4551 | |
3a5c359a AK |
4552 | /* |
4553 | * Check again in case we missed a preemption opportunity | |
4554 | * between schedule and now. | |
4555 | */ | |
4556 | barrier(); | |
4557 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4558 | } |
4559 | ||
4560 | #endif /* CONFIG_PREEMPT */ | |
4561 | ||
95cdf3b7 IM |
4562 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4563 | void *key) | |
1da177e4 | 4564 | { |
48f24c4d | 4565 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4566 | } |
1da177e4 LT |
4567 | EXPORT_SYMBOL(default_wake_function); |
4568 | ||
4569 | /* | |
41a2d6cf IM |
4570 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4571 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4572 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4573 | * | |
4574 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4575 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4576 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4577 | */ | |
4578 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4579 | int nr_exclusive, int sync, void *key) | |
4580 | { | |
2e45874c | 4581 | wait_queue_t *curr, *next; |
1da177e4 | 4582 | |
2e45874c | 4583 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4584 | unsigned flags = curr->flags; |
4585 | ||
1da177e4 | 4586 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4587 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4588 | break; |
4589 | } | |
4590 | } | |
4591 | ||
4592 | /** | |
4593 | * __wake_up - wake up threads blocked on a waitqueue. | |
4594 | * @q: the waitqueue | |
4595 | * @mode: which threads | |
4596 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4597 | * @key: is directly passed to the wakeup function |
1da177e4 | 4598 | */ |
7ad5b3a5 | 4599 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4600 | int nr_exclusive, void *key) |
1da177e4 LT |
4601 | { |
4602 | unsigned long flags; | |
4603 | ||
4604 | spin_lock_irqsave(&q->lock, flags); | |
4605 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4606 | spin_unlock_irqrestore(&q->lock, flags); | |
4607 | } | |
1da177e4 LT |
4608 | EXPORT_SYMBOL(__wake_up); |
4609 | ||
4610 | /* | |
4611 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4612 | */ | |
7ad5b3a5 | 4613 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4614 | { |
4615 | __wake_up_common(q, mode, 1, 0, NULL); | |
4616 | } | |
4617 | ||
4618 | /** | |
67be2dd1 | 4619 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4620 | * @q: the waitqueue |
4621 | * @mode: which threads | |
4622 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4623 | * | |
4624 | * The sync wakeup differs that the waker knows that it will schedule | |
4625 | * away soon, so while the target thread will be woken up, it will not | |
4626 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4627 | * with each other. This can prevent needless bouncing between CPUs. | |
4628 | * | |
4629 | * On UP it can prevent extra preemption. | |
4630 | */ | |
7ad5b3a5 | 4631 | void |
95cdf3b7 | 4632 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4633 | { |
4634 | unsigned long flags; | |
4635 | int sync = 1; | |
4636 | ||
4637 | if (unlikely(!q)) | |
4638 | return; | |
4639 | ||
4640 | if (unlikely(!nr_exclusive)) | |
4641 | sync = 0; | |
4642 | ||
4643 | spin_lock_irqsave(&q->lock, flags); | |
4644 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4645 | spin_unlock_irqrestore(&q->lock, flags); | |
4646 | } | |
4647 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4648 | ||
65eb3dc6 KD |
4649 | /** |
4650 | * complete: - signals a single thread waiting on this completion | |
4651 | * @x: holds the state of this particular completion | |
4652 | * | |
4653 | * This will wake up a single thread waiting on this completion. Threads will be | |
4654 | * awakened in the same order in which they were queued. | |
4655 | * | |
4656 | * See also complete_all(), wait_for_completion() and related routines. | |
4657 | */ | |
b15136e9 | 4658 | void complete(struct completion *x) |
1da177e4 LT |
4659 | { |
4660 | unsigned long flags; | |
4661 | ||
4662 | spin_lock_irqsave(&x->wait.lock, flags); | |
4663 | x->done++; | |
d9514f6c | 4664 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4665 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4666 | } | |
4667 | EXPORT_SYMBOL(complete); | |
4668 | ||
65eb3dc6 KD |
4669 | /** |
4670 | * complete_all: - signals all threads waiting on this completion | |
4671 | * @x: holds the state of this particular completion | |
4672 | * | |
4673 | * This will wake up all threads waiting on this particular completion event. | |
4674 | */ | |
b15136e9 | 4675 | void complete_all(struct completion *x) |
1da177e4 LT |
4676 | { |
4677 | unsigned long flags; | |
4678 | ||
4679 | spin_lock_irqsave(&x->wait.lock, flags); | |
4680 | x->done += UINT_MAX/2; | |
d9514f6c | 4681 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4682 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4683 | } | |
4684 | EXPORT_SYMBOL(complete_all); | |
4685 | ||
8cbbe86d AK |
4686 | static inline long __sched |
4687 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4688 | { |
1da177e4 LT |
4689 | if (!x->done) { |
4690 | DECLARE_WAITQUEUE(wait, current); | |
4691 | ||
4692 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4693 | __add_wait_queue_tail(&x->wait, &wait); | |
4694 | do { | |
94d3d824 | 4695 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4696 | timeout = -ERESTARTSYS; |
4697 | break; | |
8cbbe86d AK |
4698 | } |
4699 | __set_current_state(state); | |
1da177e4 LT |
4700 | spin_unlock_irq(&x->wait.lock); |
4701 | timeout = schedule_timeout(timeout); | |
4702 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4703 | } while (!x->done && timeout); |
1da177e4 | 4704 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4705 | if (!x->done) |
4706 | return timeout; | |
1da177e4 LT |
4707 | } |
4708 | x->done--; | |
ea71a546 | 4709 | return timeout ?: 1; |
1da177e4 | 4710 | } |
1da177e4 | 4711 | |
8cbbe86d AK |
4712 | static long __sched |
4713 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4714 | { |
1da177e4 LT |
4715 | might_sleep(); |
4716 | ||
4717 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4718 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4719 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4720 | return timeout; |
4721 | } | |
1da177e4 | 4722 | |
65eb3dc6 KD |
4723 | /** |
4724 | * wait_for_completion: - waits for completion of a task | |
4725 | * @x: holds the state of this particular completion | |
4726 | * | |
4727 | * This waits to be signaled for completion of a specific task. It is NOT | |
4728 | * interruptible and there is no timeout. | |
4729 | * | |
4730 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4731 | * and interrupt capability. Also see complete(). | |
4732 | */ | |
b15136e9 | 4733 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4734 | { |
4735 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4736 | } |
8cbbe86d | 4737 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4738 | |
65eb3dc6 KD |
4739 | /** |
4740 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4741 | * @x: holds the state of this particular completion | |
4742 | * @timeout: timeout value in jiffies | |
4743 | * | |
4744 | * This waits for either a completion of a specific task to be signaled or for a | |
4745 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4746 | * interruptible. | |
4747 | */ | |
b15136e9 | 4748 | unsigned long __sched |
8cbbe86d | 4749 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4750 | { |
8cbbe86d | 4751 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4752 | } |
8cbbe86d | 4753 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4754 | |
65eb3dc6 KD |
4755 | /** |
4756 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4757 | * @x: holds the state of this particular completion | |
4758 | * | |
4759 | * This waits for completion of a specific task to be signaled. It is | |
4760 | * interruptible. | |
4761 | */ | |
8cbbe86d | 4762 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4763 | { |
51e97990 AK |
4764 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4765 | if (t == -ERESTARTSYS) | |
4766 | return t; | |
4767 | return 0; | |
0fec171c | 4768 | } |
8cbbe86d | 4769 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4770 | |
65eb3dc6 KD |
4771 | /** |
4772 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4773 | * @x: holds the state of this particular completion | |
4774 | * @timeout: timeout value in jiffies | |
4775 | * | |
4776 | * This waits for either a completion of a specific task to be signaled or for a | |
4777 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4778 | */ | |
b15136e9 | 4779 | unsigned long __sched |
8cbbe86d AK |
4780 | wait_for_completion_interruptible_timeout(struct completion *x, |
4781 | unsigned long timeout) | |
0fec171c | 4782 | { |
8cbbe86d | 4783 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4784 | } |
8cbbe86d | 4785 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4786 | |
65eb3dc6 KD |
4787 | /** |
4788 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4789 | * @x: holds the state of this particular completion | |
4790 | * | |
4791 | * This waits to be signaled for completion of a specific task. It can be | |
4792 | * interrupted by a kill signal. | |
4793 | */ | |
009e577e MW |
4794 | int __sched wait_for_completion_killable(struct completion *x) |
4795 | { | |
4796 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4797 | if (t == -ERESTARTSYS) | |
4798 | return t; | |
4799 | return 0; | |
4800 | } | |
4801 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4802 | ||
be4de352 DC |
4803 | /** |
4804 | * try_wait_for_completion - try to decrement a completion without blocking | |
4805 | * @x: completion structure | |
4806 | * | |
4807 | * Returns: 0 if a decrement cannot be done without blocking | |
4808 | * 1 if a decrement succeeded. | |
4809 | * | |
4810 | * If a completion is being used as a counting completion, | |
4811 | * attempt to decrement the counter without blocking. This | |
4812 | * enables us to avoid waiting if the resource the completion | |
4813 | * is protecting is not available. | |
4814 | */ | |
4815 | bool try_wait_for_completion(struct completion *x) | |
4816 | { | |
4817 | int ret = 1; | |
4818 | ||
4819 | spin_lock_irq(&x->wait.lock); | |
4820 | if (!x->done) | |
4821 | ret = 0; | |
4822 | else | |
4823 | x->done--; | |
4824 | spin_unlock_irq(&x->wait.lock); | |
4825 | return ret; | |
4826 | } | |
4827 | EXPORT_SYMBOL(try_wait_for_completion); | |
4828 | ||
4829 | /** | |
4830 | * completion_done - Test to see if a completion has any waiters | |
4831 | * @x: completion structure | |
4832 | * | |
4833 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4834 | * 1 if there are no waiters. | |
4835 | * | |
4836 | */ | |
4837 | bool completion_done(struct completion *x) | |
4838 | { | |
4839 | int ret = 1; | |
4840 | ||
4841 | spin_lock_irq(&x->wait.lock); | |
4842 | if (!x->done) | |
4843 | ret = 0; | |
4844 | spin_unlock_irq(&x->wait.lock); | |
4845 | return ret; | |
4846 | } | |
4847 | EXPORT_SYMBOL(completion_done); | |
4848 | ||
8cbbe86d AK |
4849 | static long __sched |
4850 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4851 | { |
0fec171c IM |
4852 | unsigned long flags; |
4853 | wait_queue_t wait; | |
4854 | ||
4855 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4856 | |
8cbbe86d | 4857 | __set_current_state(state); |
1da177e4 | 4858 | |
8cbbe86d AK |
4859 | spin_lock_irqsave(&q->lock, flags); |
4860 | __add_wait_queue(q, &wait); | |
4861 | spin_unlock(&q->lock); | |
4862 | timeout = schedule_timeout(timeout); | |
4863 | spin_lock_irq(&q->lock); | |
4864 | __remove_wait_queue(q, &wait); | |
4865 | spin_unlock_irqrestore(&q->lock, flags); | |
4866 | ||
4867 | return timeout; | |
4868 | } | |
4869 | ||
4870 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4871 | { | |
4872 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4873 | } |
1da177e4 LT |
4874 | EXPORT_SYMBOL(interruptible_sleep_on); |
4875 | ||
0fec171c | 4876 | long __sched |
95cdf3b7 | 4877 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4878 | { |
8cbbe86d | 4879 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4880 | } |
1da177e4 LT |
4881 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4882 | ||
0fec171c | 4883 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4884 | { |
8cbbe86d | 4885 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4886 | } |
1da177e4 LT |
4887 | EXPORT_SYMBOL(sleep_on); |
4888 | ||
0fec171c | 4889 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4890 | { |
8cbbe86d | 4891 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4892 | } |
1da177e4 LT |
4893 | EXPORT_SYMBOL(sleep_on_timeout); |
4894 | ||
b29739f9 IM |
4895 | #ifdef CONFIG_RT_MUTEXES |
4896 | ||
4897 | /* | |
4898 | * rt_mutex_setprio - set the current priority of a task | |
4899 | * @p: task | |
4900 | * @prio: prio value (kernel-internal form) | |
4901 | * | |
4902 | * This function changes the 'effective' priority of a task. It does | |
4903 | * not touch ->normal_prio like __setscheduler(). | |
4904 | * | |
4905 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4906 | */ | |
36c8b586 | 4907 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4908 | { |
4909 | unsigned long flags; | |
83b699ed | 4910 | int oldprio, on_rq, running; |
70b97a7f | 4911 | struct rq *rq; |
cb469845 | 4912 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4913 | |
4914 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4915 | ||
4916 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4917 | update_rq_clock(rq); |
b29739f9 | 4918 | |
d5f9f942 | 4919 | oldprio = p->prio; |
dd41f596 | 4920 | on_rq = p->se.on_rq; |
051a1d1a | 4921 | running = task_current(rq, p); |
0e1f3483 | 4922 | if (on_rq) |
69be72c1 | 4923 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4924 | if (running) |
4925 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4926 | |
4927 | if (rt_prio(prio)) | |
4928 | p->sched_class = &rt_sched_class; | |
4929 | else | |
4930 | p->sched_class = &fair_sched_class; | |
4931 | ||
b29739f9 IM |
4932 | p->prio = prio; |
4933 | ||
0e1f3483 HS |
4934 | if (running) |
4935 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4936 | if (on_rq) { |
8159f87e | 4937 | enqueue_task(rq, p, 0); |
cb469845 SR |
4938 | |
4939 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4940 | } |
4941 | task_rq_unlock(rq, &flags); | |
4942 | } | |
4943 | ||
4944 | #endif | |
4945 | ||
36c8b586 | 4946 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4947 | { |
dd41f596 | 4948 | int old_prio, delta, on_rq; |
1da177e4 | 4949 | unsigned long flags; |
70b97a7f | 4950 | struct rq *rq; |
1da177e4 LT |
4951 | |
4952 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4953 | return; | |
4954 | /* | |
4955 | * We have to be careful, if called from sys_setpriority(), | |
4956 | * the task might be in the middle of scheduling on another CPU. | |
4957 | */ | |
4958 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4959 | update_rq_clock(rq); |
1da177e4 LT |
4960 | /* |
4961 | * The RT priorities are set via sched_setscheduler(), but we still | |
4962 | * allow the 'normal' nice value to be set - but as expected | |
4963 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4964 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4965 | */ |
e05606d3 | 4966 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4967 | p->static_prio = NICE_TO_PRIO(nice); |
4968 | goto out_unlock; | |
4969 | } | |
dd41f596 | 4970 | on_rq = p->se.on_rq; |
c09595f6 | 4971 | if (on_rq) |
69be72c1 | 4972 | dequeue_task(rq, p, 0); |
1da177e4 | 4973 | |
1da177e4 | 4974 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4975 | set_load_weight(p); |
b29739f9 IM |
4976 | old_prio = p->prio; |
4977 | p->prio = effective_prio(p); | |
4978 | delta = p->prio - old_prio; | |
1da177e4 | 4979 | |
dd41f596 | 4980 | if (on_rq) { |
8159f87e | 4981 | enqueue_task(rq, p, 0); |
1da177e4 | 4982 | /* |
d5f9f942 AM |
4983 | * If the task increased its priority or is running and |
4984 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4985 | */ |
d5f9f942 | 4986 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4987 | resched_task(rq->curr); |
4988 | } | |
4989 | out_unlock: | |
4990 | task_rq_unlock(rq, &flags); | |
4991 | } | |
1da177e4 LT |
4992 | EXPORT_SYMBOL(set_user_nice); |
4993 | ||
e43379f1 MM |
4994 | /* |
4995 | * can_nice - check if a task can reduce its nice value | |
4996 | * @p: task | |
4997 | * @nice: nice value | |
4998 | */ | |
36c8b586 | 4999 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5000 | { |
024f4747 MM |
5001 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5002 | int nice_rlim = 20 - nice; | |
48f24c4d | 5003 | |
e43379f1 MM |
5004 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5005 | capable(CAP_SYS_NICE)); | |
5006 | } | |
5007 | ||
1da177e4 LT |
5008 | #ifdef __ARCH_WANT_SYS_NICE |
5009 | ||
5010 | /* | |
5011 | * sys_nice - change the priority of the current process. | |
5012 | * @increment: priority increment | |
5013 | * | |
5014 | * sys_setpriority is a more generic, but much slower function that | |
5015 | * does similar things. | |
5016 | */ | |
5017 | asmlinkage long sys_nice(int increment) | |
5018 | { | |
48f24c4d | 5019 | long nice, retval; |
1da177e4 LT |
5020 | |
5021 | /* | |
5022 | * Setpriority might change our priority at the same moment. | |
5023 | * We don't have to worry. Conceptually one call occurs first | |
5024 | * and we have a single winner. | |
5025 | */ | |
e43379f1 MM |
5026 | if (increment < -40) |
5027 | increment = -40; | |
1da177e4 LT |
5028 | if (increment > 40) |
5029 | increment = 40; | |
5030 | ||
5031 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5032 | if (nice < -20) | |
5033 | nice = -20; | |
5034 | if (nice > 19) | |
5035 | nice = 19; | |
5036 | ||
e43379f1 MM |
5037 | if (increment < 0 && !can_nice(current, nice)) |
5038 | return -EPERM; | |
5039 | ||
1da177e4 LT |
5040 | retval = security_task_setnice(current, nice); |
5041 | if (retval) | |
5042 | return retval; | |
5043 | ||
5044 | set_user_nice(current, nice); | |
5045 | return 0; | |
5046 | } | |
5047 | ||
5048 | #endif | |
5049 | ||
5050 | /** | |
5051 | * task_prio - return the priority value of a given task. | |
5052 | * @p: the task in question. | |
5053 | * | |
5054 | * This is the priority value as seen by users in /proc. | |
5055 | * RT tasks are offset by -200. Normal tasks are centered | |
5056 | * around 0, value goes from -16 to +15. | |
5057 | */ | |
36c8b586 | 5058 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5059 | { |
5060 | return p->prio - MAX_RT_PRIO; | |
5061 | } | |
5062 | ||
5063 | /** | |
5064 | * task_nice - return the nice value of a given task. | |
5065 | * @p: the task in question. | |
5066 | */ | |
36c8b586 | 5067 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5068 | { |
5069 | return TASK_NICE(p); | |
5070 | } | |
150d8bed | 5071 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5072 | |
5073 | /** | |
5074 | * idle_cpu - is a given cpu idle currently? | |
5075 | * @cpu: the processor in question. | |
5076 | */ | |
5077 | int idle_cpu(int cpu) | |
5078 | { | |
5079 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5080 | } | |
5081 | ||
1da177e4 LT |
5082 | /** |
5083 | * idle_task - return the idle task for a given cpu. | |
5084 | * @cpu: the processor in question. | |
5085 | */ | |
36c8b586 | 5086 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5087 | { |
5088 | return cpu_rq(cpu)->idle; | |
5089 | } | |
5090 | ||
5091 | /** | |
5092 | * find_process_by_pid - find a process with a matching PID value. | |
5093 | * @pid: the pid in question. | |
5094 | */ | |
a9957449 | 5095 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5096 | { |
228ebcbe | 5097 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5098 | } |
5099 | ||
5100 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5101 | static void |
5102 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5103 | { |
dd41f596 | 5104 | BUG_ON(p->se.on_rq); |
48f24c4d | 5105 | |
1da177e4 | 5106 | p->policy = policy; |
dd41f596 IM |
5107 | switch (p->policy) { |
5108 | case SCHED_NORMAL: | |
5109 | case SCHED_BATCH: | |
5110 | case SCHED_IDLE: | |
5111 | p->sched_class = &fair_sched_class; | |
5112 | break; | |
5113 | case SCHED_FIFO: | |
5114 | case SCHED_RR: | |
5115 | p->sched_class = &rt_sched_class; | |
5116 | break; | |
5117 | } | |
5118 | ||
1da177e4 | 5119 | p->rt_priority = prio; |
b29739f9 IM |
5120 | p->normal_prio = normal_prio(p); |
5121 | /* we are holding p->pi_lock already */ | |
5122 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5123 | set_load_weight(p); |
1da177e4 LT |
5124 | } |
5125 | ||
961ccddd RR |
5126 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5127 | struct sched_param *param, bool user) | |
1da177e4 | 5128 | { |
83b699ed | 5129 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5130 | unsigned long flags; |
cb469845 | 5131 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5132 | struct rq *rq; |
1da177e4 | 5133 | |
66e5393a SR |
5134 | /* may grab non-irq protected spin_locks */ |
5135 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5136 | recheck: |
5137 | /* double check policy once rq lock held */ | |
5138 | if (policy < 0) | |
5139 | policy = oldpolicy = p->policy; | |
5140 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5141 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5142 | policy != SCHED_IDLE) | |
b0a9499c | 5143 | return -EINVAL; |
1da177e4 LT |
5144 | /* |
5145 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5146 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5147 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5148 | */ |
5149 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5150 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5151 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5152 | return -EINVAL; |
e05606d3 | 5153 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5154 | return -EINVAL; |
5155 | ||
37e4ab3f OC |
5156 | /* |
5157 | * Allow unprivileged RT tasks to decrease priority: | |
5158 | */ | |
961ccddd | 5159 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5160 | if (rt_policy(policy)) { |
8dc3e909 | 5161 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5162 | |
5163 | if (!lock_task_sighand(p, &flags)) | |
5164 | return -ESRCH; | |
5165 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5166 | unlock_task_sighand(p, &flags); | |
5167 | ||
5168 | /* can't set/change the rt policy */ | |
5169 | if (policy != p->policy && !rlim_rtprio) | |
5170 | return -EPERM; | |
5171 | ||
5172 | /* can't increase priority */ | |
5173 | if (param->sched_priority > p->rt_priority && | |
5174 | param->sched_priority > rlim_rtprio) | |
5175 | return -EPERM; | |
5176 | } | |
dd41f596 IM |
5177 | /* |
5178 | * Like positive nice levels, dont allow tasks to | |
5179 | * move out of SCHED_IDLE either: | |
5180 | */ | |
5181 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5182 | return -EPERM; | |
5fe1d75f | 5183 | |
37e4ab3f OC |
5184 | /* can't change other user's priorities */ |
5185 | if ((current->euid != p->euid) && | |
5186 | (current->euid != p->uid)) | |
5187 | return -EPERM; | |
5188 | } | |
1da177e4 | 5189 | |
725aad24 | 5190 | if (user) { |
b68aa230 | 5191 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5192 | /* |
5193 | * Do not allow realtime tasks into groups that have no runtime | |
5194 | * assigned. | |
5195 | */ | |
9a7e0b18 PZ |
5196 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5197 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5198 | return -EPERM; |
b68aa230 PZ |
5199 | #endif |
5200 | ||
725aad24 JF |
5201 | retval = security_task_setscheduler(p, policy, param); |
5202 | if (retval) | |
5203 | return retval; | |
5204 | } | |
5205 | ||
b29739f9 IM |
5206 | /* |
5207 | * make sure no PI-waiters arrive (or leave) while we are | |
5208 | * changing the priority of the task: | |
5209 | */ | |
5210 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5211 | /* |
5212 | * To be able to change p->policy safely, the apropriate | |
5213 | * runqueue lock must be held. | |
5214 | */ | |
b29739f9 | 5215 | rq = __task_rq_lock(p); |
1da177e4 LT |
5216 | /* recheck policy now with rq lock held */ |
5217 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5218 | policy = oldpolicy = -1; | |
b29739f9 IM |
5219 | __task_rq_unlock(rq); |
5220 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5221 | goto recheck; |
5222 | } | |
2daa3577 | 5223 | update_rq_clock(rq); |
dd41f596 | 5224 | on_rq = p->se.on_rq; |
051a1d1a | 5225 | running = task_current(rq, p); |
0e1f3483 | 5226 | if (on_rq) |
2e1cb74a | 5227 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5228 | if (running) |
5229 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5230 | |
1da177e4 | 5231 | oldprio = p->prio; |
dd41f596 | 5232 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5233 | |
0e1f3483 HS |
5234 | if (running) |
5235 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5236 | if (on_rq) { |
5237 | activate_task(rq, p, 0); | |
cb469845 SR |
5238 | |
5239 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5240 | } |
b29739f9 IM |
5241 | __task_rq_unlock(rq); |
5242 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5243 | ||
95e02ca9 TG |
5244 | rt_mutex_adjust_pi(p); |
5245 | ||
1da177e4 LT |
5246 | return 0; |
5247 | } | |
961ccddd RR |
5248 | |
5249 | /** | |
5250 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5251 | * @p: the task in question. | |
5252 | * @policy: new policy. | |
5253 | * @param: structure containing the new RT priority. | |
5254 | * | |
5255 | * NOTE that the task may be already dead. | |
5256 | */ | |
5257 | int sched_setscheduler(struct task_struct *p, int policy, | |
5258 | struct sched_param *param) | |
5259 | { | |
5260 | return __sched_setscheduler(p, policy, param, true); | |
5261 | } | |
1da177e4 LT |
5262 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5263 | ||
961ccddd RR |
5264 | /** |
5265 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5266 | * @p: the task in question. | |
5267 | * @policy: new policy. | |
5268 | * @param: structure containing the new RT priority. | |
5269 | * | |
5270 | * Just like sched_setscheduler, only don't bother checking if the | |
5271 | * current context has permission. For example, this is needed in | |
5272 | * stop_machine(): we create temporary high priority worker threads, | |
5273 | * but our caller might not have that capability. | |
5274 | */ | |
5275 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5276 | struct sched_param *param) | |
5277 | { | |
5278 | return __sched_setscheduler(p, policy, param, false); | |
5279 | } | |
5280 | ||
95cdf3b7 IM |
5281 | static int |
5282 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5283 | { |
1da177e4 LT |
5284 | struct sched_param lparam; |
5285 | struct task_struct *p; | |
36c8b586 | 5286 | int retval; |
1da177e4 LT |
5287 | |
5288 | if (!param || pid < 0) | |
5289 | return -EINVAL; | |
5290 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5291 | return -EFAULT; | |
5fe1d75f ON |
5292 | |
5293 | rcu_read_lock(); | |
5294 | retval = -ESRCH; | |
1da177e4 | 5295 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5296 | if (p != NULL) |
5297 | retval = sched_setscheduler(p, policy, &lparam); | |
5298 | rcu_read_unlock(); | |
36c8b586 | 5299 | |
1da177e4 LT |
5300 | return retval; |
5301 | } | |
5302 | ||
5303 | /** | |
5304 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5305 | * @pid: the pid in question. | |
5306 | * @policy: new policy. | |
5307 | * @param: structure containing the new RT priority. | |
5308 | */ | |
41a2d6cf IM |
5309 | asmlinkage long |
5310 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5311 | { |
c21761f1 JB |
5312 | /* negative values for policy are not valid */ |
5313 | if (policy < 0) | |
5314 | return -EINVAL; | |
5315 | ||
1da177e4 LT |
5316 | return do_sched_setscheduler(pid, policy, param); |
5317 | } | |
5318 | ||
5319 | /** | |
5320 | * sys_sched_setparam - set/change the RT priority of a thread | |
5321 | * @pid: the pid in question. | |
5322 | * @param: structure containing the new RT priority. | |
5323 | */ | |
5324 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5325 | { | |
5326 | return do_sched_setscheduler(pid, -1, param); | |
5327 | } | |
5328 | ||
5329 | /** | |
5330 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5331 | * @pid: the pid in question. | |
5332 | */ | |
5333 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5334 | { | |
36c8b586 | 5335 | struct task_struct *p; |
3a5c359a | 5336 | int retval; |
1da177e4 LT |
5337 | |
5338 | if (pid < 0) | |
3a5c359a | 5339 | return -EINVAL; |
1da177e4 LT |
5340 | |
5341 | retval = -ESRCH; | |
5342 | read_lock(&tasklist_lock); | |
5343 | p = find_process_by_pid(pid); | |
5344 | if (p) { | |
5345 | retval = security_task_getscheduler(p); | |
5346 | if (!retval) | |
5347 | retval = p->policy; | |
5348 | } | |
5349 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5350 | return retval; |
5351 | } | |
5352 | ||
5353 | /** | |
5354 | * sys_sched_getscheduler - get the RT priority of a thread | |
5355 | * @pid: the pid in question. | |
5356 | * @param: structure containing the RT priority. | |
5357 | */ | |
5358 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5359 | { | |
5360 | struct sched_param lp; | |
36c8b586 | 5361 | struct task_struct *p; |
3a5c359a | 5362 | int retval; |
1da177e4 LT |
5363 | |
5364 | if (!param || pid < 0) | |
3a5c359a | 5365 | return -EINVAL; |
1da177e4 LT |
5366 | |
5367 | read_lock(&tasklist_lock); | |
5368 | p = find_process_by_pid(pid); | |
5369 | retval = -ESRCH; | |
5370 | if (!p) | |
5371 | goto out_unlock; | |
5372 | ||
5373 | retval = security_task_getscheduler(p); | |
5374 | if (retval) | |
5375 | goto out_unlock; | |
5376 | ||
5377 | lp.sched_priority = p->rt_priority; | |
5378 | read_unlock(&tasklist_lock); | |
5379 | ||
5380 | /* | |
5381 | * This one might sleep, we cannot do it with a spinlock held ... | |
5382 | */ | |
5383 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5384 | ||
1da177e4 LT |
5385 | return retval; |
5386 | ||
5387 | out_unlock: | |
5388 | read_unlock(&tasklist_lock); | |
5389 | return retval; | |
5390 | } | |
5391 | ||
b53e921b | 5392 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5393 | { |
1da177e4 | 5394 | cpumask_t cpus_allowed; |
b53e921b | 5395 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5396 | struct task_struct *p; |
5397 | int retval; | |
1da177e4 | 5398 | |
95402b38 | 5399 | get_online_cpus(); |
1da177e4 LT |
5400 | read_lock(&tasklist_lock); |
5401 | ||
5402 | p = find_process_by_pid(pid); | |
5403 | if (!p) { | |
5404 | read_unlock(&tasklist_lock); | |
95402b38 | 5405 | put_online_cpus(); |
1da177e4 LT |
5406 | return -ESRCH; |
5407 | } | |
5408 | ||
5409 | /* | |
5410 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5411 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5412 | * usage count and then drop tasklist_lock. |
5413 | */ | |
5414 | get_task_struct(p); | |
5415 | read_unlock(&tasklist_lock); | |
5416 | ||
5417 | retval = -EPERM; | |
5418 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5419 | !capable(CAP_SYS_NICE)) | |
5420 | goto out_unlock; | |
5421 | ||
e7834f8f DQ |
5422 | retval = security_task_setscheduler(p, 0, NULL); |
5423 | if (retval) | |
5424 | goto out_unlock; | |
5425 | ||
f9a86fcb | 5426 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5427 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5428 | again: |
7c16ec58 | 5429 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5430 | |
8707d8b8 | 5431 | if (!retval) { |
f9a86fcb | 5432 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5433 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5434 | /* | |
5435 | * We must have raced with a concurrent cpuset | |
5436 | * update. Just reset the cpus_allowed to the | |
5437 | * cpuset's cpus_allowed | |
5438 | */ | |
5439 | new_mask = cpus_allowed; | |
5440 | goto again; | |
5441 | } | |
5442 | } | |
1da177e4 LT |
5443 | out_unlock: |
5444 | put_task_struct(p); | |
95402b38 | 5445 | put_online_cpus(); |
1da177e4 LT |
5446 | return retval; |
5447 | } | |
5448 | ||
5449 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5450 | cpumask_t *new_mask) | |
5451 | { | |
5452 | if (len < sizeof(cpumask_t)) { | |
5453 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5454 | } else if (len > sizeof(cpumask_t)) { | |
5455 | len = sizeof(cpumask_t); | |
5456 | } | |
5457 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5458 | } | |
5459 | ||
5460 | /** | |
5461 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5462 | * @pid: pid of the process | |
5463 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5464 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5465 | */ | |
5466 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5467 | unsigned long __user *user_mask_ptr) | |
5468 | { | |
5469 | cpumask_t new_mask; | |
5470 | int retval; | |
5471 | ||
5472 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5473 | if (retval) | |
5474 | return retval; | |
5475 | ||
b53e921b | 5476 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5477 | } |
5478 | ||
1da177e4 LT |
5479 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5480 | { | |
36c8b586 | 5481 | struct task_struct *p; |
1da177e4 | 5482 | int retval; |
1da177e4 | 5483 | |
95402b38 | 5484 | get_online_cpus(); |
1da177e4 LT |
5485 | read_lock(&tasklist_lock); |
5486 | ||
5487 | retval = -ESRCH; | |
5488 | p = find_process_by_pid(pid); | |
5489 | if (!p) | |
5490 | goto out_unlock; | |
5491 | ||
e7834f8f DQ |
5492 | retval = security_task_getscheduler(p); |
5493 | if (retval) | |
5494 | goto out_unlock; | |
5495 | ||
2f7016d9 | 5496 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5497 | |
5498 | out_unlock: | |
5499 | read_unlock(&tasklist_lock); | |
95402b38 | 5500 | put_online_cpus(); |
1da177e4 | 5501 | |
9531b62f | 5502 | return retval; |
1da177e4 LT |
5503 | } |
5504 | ||
5505 | /** | |
5506 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5507 | * @pid: pid of the process | |
5508 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5509 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5510 | */ | |
5511 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5512 | unsigned long __user *user_mask_ptr) | |
5513 | { | |
5514 | int ret; | |
5515 | cpumask_t mask; | |
5516 | ||
5517 | if (len < sizeof(cpumask_t)) | |
5518 | return -EINVAL; | |
5519 | ||
5520 | ret = sched_getaffinity(pid, &mask); | |
5521 | if (ret < 0) | |
5522 | return ret; | |
5523 | ||
5524 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5525 | return -EFAULT; | |
5526 | ||
5527 | return sizeof(cpumask_t); | |
5528 | } | |
5529 | ||
5530 | /** | |
5531 | * sys_sched_yield - yield the current processor to other threads. | |
5532 | * | |
dd41f596 IM |
5533 | * This function yields the current CPU to other tasks. If there are no |
5534 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5535 | */ |
5536 | asmlinkage long sys_sched_yield(void) | |
5537 | { | |
70b97a7f | 5538 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5539 | |
2d72376b | 5540 | schedstat_inc(rq, yld_count); |
4530d7ab | 5541 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5542 | |
5543 | /* | |
5544 | * Since we are going to call schedule() anyway, there's | |
5545 | * no need to preempt or enable interrupts: | |
5546 | */ | |
5547 | __release(rq->lock); | |
8a25d5de | 5548 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5549 | _raw_spin_unlock(&rq->lock); |
5550 | preempt_enable_no_resched(); | |
5551 | ||
5552 | schedule(); | |
5553 | ||
5554 | return 0; | |
5555 | } | |
5556 | ||
e7b38404 | 5557 | static void __cond_resched(void) |
1da177e4 | 5558 | { |
8e0a43d8 IM |
5559 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5560 | __might_sleep(__FILE__, __LINE__); | |
5561 | #endif | |
5bbcfd90 IM |
5562 | /* |
5563 | * The BKS might be reacquired before we have dropped | |
5564 | * PREEMPT_ACTIVE, which could trigger a second | |
5565 | * cond_resched() call. | |
5566 | */ | |
1da177e4 LT |
5567 | do { |
5568 | add_preempt_count(PREEMPT_ACTIVE); | |
5569 | schedule(); | |
5570 | sub_preempt_count(PREEMPT_ACTIVE); | |
5571 | } while (need_resched()); | |
5572 | } | |
5573 | ||
02b67cc3 | 5574 | int __sched _cond_resched(void) |
1da177e4 | 5575 | { |
9414232f IM |
5576 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5577 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5578 | __cond_resched(); |
5579 | return 1; | |
5580 | } | |
5581 | return 0; | |
5582 | } | |
02b67cc3 | 5583 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5584 | |
5585 | /* | |
5586 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5587 | * call schedule, and on return reacquire the lock. | |
5588 | * | |
41a2d6cf | 5589 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5590 | * operations here to prevent schedule() from being called twice (once via |
5591 | * spin_unlock(), once by hand). | |
5592 | */ | |
95cdf3b7 | 5593 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5594 | { |
95c354fe | 5595 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5596 | int ret = 0; |
5597 | ||
95c354fe | 5598 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5599 | spin_unlock(lock); |
95c354fe NP |
5600 | if (resched && need_resched()) |
5601 | __cond_resched(); | |
5602 | else | |
5603 | cpu_relax(); | |
6df3cecb | 5604 | ret = 1; |
1da177e4 | 5605 | spin_lock(lock); |
1da177e4 | 5606 | } |
6df3cecb | 5607 | return ret; |
1da177e4 | 5608 | } |
1da177e4 LT |
5609 | EXPORT_SYMBOL(cond_resched_lock); |
5610 | ||
5611 | int __sched cond_resched_softirq(void) | |
5612 | { | |
5613 | BUG_ON(!in_softirq()); | |
5614 | ||
9414232f | 5615 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5616 | local_bh_enable(); |
1da177e4 LT |
5617 | __cond_resched(); |
5618 | local_bh_disable(); | |
5619 | return 1; | |
5620 | } | |
5621 | return 0; | |
5622 | } | |
1da177e4 LT |
5623 | EXPORT_SYMBOL(cond_resched_softirq); |
5624 | ||
1da177e4 LT |
5625 | /** |
5626 | * yield - yield the current processor to other threads. | |
5627 | * | |
72fd4a35 | 5628 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5629 | * thread runnable and calls sys_sched_yield(). |
5630 | */ | |
5631 | void __sched yield(void) | |
5632 | { | |
5633 | set_current_state(TASK_RUNNING); | |
5634 | sys_sched_yield(); | |
5635 | } | |
1da177e4 LT |
5636 | EXPORT_SYMBOL(yield); |
5637 | ||
5638 | /* | |
41a2d6cf | 5639 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5640 | * that process accounting knows that this is a task in IO wait state. |
5641 | * | |
5642 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5643 | * has set its backing_dev_info: the queue against which it should throttle) | |
5644 | */ | |
5645 | void __sched io_schedule(void) | |
5646 | { | |
70b97a7f | 5647 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5648 | |
0ff92245 | 5649 | delayacct_blkio_start(); |
1da177e4 LT |
5650 | atomic_inc(&rq->nr_iowait); |
5651 | schedule(); | |
5652 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5653 | delayacct_blkio_end(); |
1da177e4 | 5654 | } |
1da177e4 LT |
5655 | EXPORT_SYMBOL(io_schedule); |
5656 | ||
5657 | long __sched io_schedule_timeout(long timeout) | |
5658 | { | |
70b97a7f | 5659 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5660 | long ret; |
5661 | ||
0ff92245 | 5662 | delayacct_blkio_start(); |
1da177e4 LT |
5663 | atomic_inc(&rq->nr_iowait); |
5664 | ret = schedule_timeout(timeout); | |
5665 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5666 | delayacct_blkio_end(); |
1da177e4 LT |
5667 | return ret; |
5668 | } | |
5669 | ||
5670 | /** | |
5671 | * sys_sched_get_priority_max - return maximum RT priority. | |
5672 | * @policy: scheduling class. | |
5673 | * | |
5674 | * this syscall returns the maximum rt_priority that can be used | |
5675 | * by a given scheduling class. | |
5676 | */ | |
5677 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5678 | { | |
5679 | int ret = -EINVAL; | |
5680 | ||
5681 | switch (policy) { | |
5682 | case SCHED_FIFO: | |
5683 | case SCHED_RR: | |
5684 | ret = MAX_USER_RT_PRIO-1; | |
5685 | break; | |
5686 | case SCHED_NORMAL: | |
b0a9499c | 5687 | case SCHED_BATCH: |
dd41f596 | 5688 | case SCHED_IDLE: |
1da177e4 LT |
5689 | ret = 0; |
5690 | break; | |
5691 | } | |
5692 | return ret; | |
5693 | } | |
5694 | ||
5695 | /** | |
5696 | * sys_sched_get_priority_min - return minimum RT priority. | |
5697 | * @policy: scheduling class. | |
5698 | * | |
5699 | * this syscall returns the minimum rt_priority that can be used | |
5700 | * by a given scheduling class. | |
5701 | */ | |
5702 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5703 | { | |
5704 | int ret = -EINVAL; | |
5705 | ||
5706 | switch (policy) { | |
5707 | case SCHED_FIFO: | |
5708 | case SCHED_RR: | |
5709 | ret = 1; | |
5710 | break; | |
5711 | case SCHED_NORMAL: | |
b0a9499c | 5712 | case SCHED_BATCH: |
dd41f596 | 5713 | case SCHED_IDLE: |
1da177e4 LT |
5714 | ret = 0; |
5715 | } | |
5716 | return ret; | |
5717 | } | |
5718 | ||
5719 | /** | |
5720 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5721 | * @pid: pid of the process. | |
5722 | * @interval: userspace pointer to the timeslice value. | |
5723 | * | |
5724 | * this syscall writes the default timeslice value of a given process | |
5725 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5726 | */ | |
5727 | asmlinkage | |
5728 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5729 | { | |
36c8b586 | 5730 | struct task_struct *p; |
a4ec24b4 | 5731 | unsigned int time_slice; |
3a5c359a | 5732 | int retval; |
1da177e4 | 5733 | struct timespec t; |
1da177e4 LT |
5734 | |
5735 | if (pid < 0) | |
3a5c359a | 5736 | return -EINVAL; |
1da177e4 LT |
5737 | |
5738 | retval = -ESRCH; | |
5739 | read_lock(&tasklist_lock); | |
5740 | p = find_process_by_pid(pid); | |
5741 | if (!p) | |
5742 | goto out_unlock; | |
5743 | ||
5744 | retval = security_task_getscheduler(p); | |
5745 | if (retval) | |
5746 | goto out_unlock; | |
5747 | ||
77034937 IM |
5748 | /* |
5749 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5750 | * tasks that are on an otherwise idle runqueue: | |
5751 | */ | |
5752 | time_slice = 0; | |
5753 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5754 | time_slice = DEF_TIMESLICE; |
1868f958 | 5755 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5756 | struct sched_entity *se = &p->se; |
5757 | unsigned long flags; | |
5758 | struct rq *rq; | |
5759 | ||
5760 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5761 | if (rq->cfs.load.weight) |
5762 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5763 | task_rq_unlock(rq, &flags); |
5764 | } | |
1da177e4 | 5765 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5766 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5767 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5768 | return retval; |
3a5c359a | 5769 | |
1da177e4 LT |
5770 | out_unlock: |
5771 | read_unlock(&tasklist_lock); | |
5772 | return retval; | |
5773 | } | |
5774 | ||
7c731e0a | 5775 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5776 | |
82a1fcb9 | 5777 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5778 | { |
1da177e4 | 5779 | unsigned long free = 0; |
36c8b586 | 5780 | unsigned state; |
1da177e4 | 5781 | |
1da177e4 | 5782 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5783 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5784 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5785 | #if BITS_PER_LONG == 32 |
1da177e4 | 5786 | if (state == TASK_RUNNING) |
cc4ea795 | 5787 | printk(KERN_CONT " running "); |
1da177e4 | 5788 | else |
cc4ea795 | 5789 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5790 | #else |
5791 | if (state == TASK_RUNNING) | |
cc4ea795 | 5792 | printk(KERN_CONT " running task "); |
1da177e4 | 5793 | else |
cc4ea795 | 5794 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5795 | #endif |
5796 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5797 | { | |
10ebffde | 5798 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5799 | while (!*n) |
5800 | n++; | |
10ebffde | 5801 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5802 | } |
5803 | #endif | |
ba25f9dc | 5804 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5805 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5806 | |
5fb5e6de | 5807 | show_stack(p, NULL); |
1da177e4 LT |
5808 | } |
5809 | ||
e59e2ae2 | 5810 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5811 | { |
36c8b586 | 5812 | struct task_struct *g, *p; |
1da177e4 | 5813 | |
4bd77321 IM |
5814 | #if BITS_PER_LONG == 32 |
5815 | printk(KERN_INFO | |
5816 | " task PC stack pid father\n"); | |
1da177e4 | 5817 | #else |
4bd77321 IM |
5818 | printk(KERN_INFO |
5819 | " task PC stack pid father\n"); | |
1da177e4 LT |
5820 | #endif |
5821 | read_lock(&tasklist_lock); | |
5822 | do_each_thread(g, p) { | |
5823 | /* | |
5824 | * reset the NMI-timeout, listing all files on a slow | |
5825 | * console might take alot of time: | |
5826 | */ | |
5827 | touch_nmi_watchdog(); | |
39bc89fd | 5828 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5829 | sched_show_task(p); |
1da177e4 LT |
5830 | } while_each_thread(g, p); |
5831 | ||
04c9167f JF |
5832 | touch_all_softlockup_watchdogs(); |
5833 | ||
dd41f596 IM |
5834 | #ifdef CONFIG_SCHED_DEBUG |
5835 | sysrq_sched_debug_show(); | |
5836 | #endif | |
1da177e4 | 5837 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5838 | /* |
5839 | * Only show locks if all tasks are dumped: | |
5840 | */ | |
5841 | if (state_filter == -1) | |
5842 | debug_show_all_locks(); | |
1da177e4 LT |
5843 | } |
5844 | ||
1df21055 IM |
5845 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5846 | { | |
dd41f596 | 5847 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5848 | } |
5849 | ||
f340c0d1 IM |
5850 | /** |
5851 | * init_idle - set up an idle thread for a given CPU | |
5852 | * @idle: task in question | |
5853 | * @cpu: cpu the idle task belongs to | |
5854 | * | |
5855 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5856 | * flag, to make booting more robust. | |
5857 | */ | |
5c1e1767 | 5858 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5859 | { |
70b97a7f | 5860 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5861 | unsigned long flags; |
5862 | ||
dd41f596 IM |
5863 | __sched_fork(idle); |
5864 | idle->se.exec_start = sched_clock(); | |
5865 | ||
b29739f9 | 5866 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5867 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5868 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5869 | |
5870 | spin_lock_irqsave(&rq->lock, flags); | |
5871 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5872 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5873 | idle->oncpu = 1; | |
5874 | #endif | |
1da177e4 LT |
5875 | spin_unlock_irqrestore(&rq->lock, flags); |
5876 | ||
5877 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5878 | #if defined(CONFIG_PREEMPT) |
5879 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5880 | #else | |
a1261f54 | 5881 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5882 | #endif |
dd41f596 IM |
5883 | /* |
5884 | * The idle tasks have their own, simple scheduling class: | |
5885 | */ | |
5886 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5887 | } |
5888 | ||
5889 | /* | |
5890 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5891 | * indicates which cpus entered this state. This is used | |
5892 | * in the rcu update to wait only for active cpus. For system | |
5893 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5894 | * always be CPU_MASK_NONE. | |
5895 | */ | |
5896 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5897 | ||
19978ca6 IM |
5898 | /* |
5899 | * Increase the granularity value when there are more CPUs, | |
5900 | * because with more CPUs the 'effective latency' as visible | |
5901 | * to users decreases. But the relationship is not linear, | |
5902 | * so pick a second-best guess by going with the log2 of the | |
5903 | * number of CPUs. | |
5904 | * | |
5905 | * This idea comes from the SD scheduler of Con Kolivas: | |
5906 | */ | |
5907 | static inline void sched_init_granularity(void) | |
5908 | { | |
5909 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5910 | const unsigned long limit = 200000000; | |
5911 | ||
5912 | sysctl_sched_min_granularity *= factor; | |
5913 | if (sysctl_sched_min_granularity > limit) | |
5914 | sysctl_sched_min_granularity = limit; | |
5915 | ||
5916 | sysctl_sched_latency *= factor; | |
5917 | if (sysctl_sched_latency > limit) | |
5918 | sysctl_sched_latency = limit; | |
5919 | ||
5920 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5921 | |
5922 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5923 | } |
5924 | ||
1da177e4 LT |
5925 | #ifdef CONFIG_SMP |
5926 | /* | |
5927 | * This is how migration works: | |
5928 | * | |
70b97a7f | 5929 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5930 | * runqueue and wake up that CPU's migration thread. |
5931 | * 2) we down() the locked semaphore => thread blocks. | |
5932 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5933 | * thread off the CPU) | |
5934 | * 4) it gets the migration request and checks whether the migrated | |
5935 | * task is still in the wrong runqueue. | |
5936 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5937 | * it and puts it into the right queue. | |
5938 | * 6) migration thread up()s the semaphore. | |
5939 | * 7) we wake up and the migration is done. | |
5940 | */ | |
5941 | ||
5942 | /* | |
5943 | * Change a given task's CPU affinity. Migrate the thread to a | |
5944 | * proper CPU and schedule it away if the CPU it's executing on | |
5945 | * is removed from the allowed bitmask. | |
5946 | * | |
5947 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5948 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5949 | * call is not atomic; no spinlocks may be held. |
5950 | */ | |
cd8ba7cd | 5951 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5952 | { |
70b97a7f | 5953 | struct migration_req req; |
1da177e4 | 5954 | unsigned long flags; |
70b97a7f | 5955 | struct rq *rq; |
48f24c4d | 5956 | int ret = 0; |
1da177e4 LT |
5957 | |
5958 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5959 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5960 | ret = -EINVAL; |
5961 | goto out; | |
5962 | } | |
5963 | ||
9985b0ba DR |
5964 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5965 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5966 | ret = -EINVAL; | |
5967 | goto out; | |
5968 | } | |
5969 | ||
73fe6aae | 5970 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5971 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5972 | else { |
cd8ba7cd MT |
5973 | p->cpus_allowed = *new_mask; |
5974 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5975 | } |
5976 | ||
1da177e4 | 5977 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5978 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5979 | goto out; |
5980 | ||
cd8ba7cd | 5981 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5982 | /* Need help from migration thread: drop lock and wait. */ |
5983 | task_rq_unlock(rq, &flags); | |
5984 | wake_up_process(rq->migration_thread); | |
5985 | wait_for_completion(&req.done); | |
5986 | tlb_migrate_finish(p->mm); | |
5987 | return 0; | |
5988 | } | |
5989 | out: | |
5990 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5991 | |
1da177e4 LT |
5992 | return ret; |
5993 | } | |
cd8ba7cd | 5994 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5995 | |
5996 | /* | |
41a2d6cf | 5997 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5998 | * this because either it can't run here any more (set_cpus_allowed() |
5999 | * away from this CPU, or CPU going down), or because we're | |
6000 | * attempting to rebalance this task on exec (sched_exec). | |
6001 | * | |
6002 | * So we race with normal scheduler movements, but that's OK, as long | |
6003 | * as the task is no longer on this CPU. | |
efc30814 KK |
6004 | * |
6005 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6006 | */ |
efc30814 | 6007 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6008 | { |
70b97a7f | 6009 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6010 | int ret = 0, on_rq; |
1da177e4 | 6011 | |
e761b772 | 6012 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6013 | return ret; |
1da177e4 LT |
6014 | |
6015 | rq_src = cpu_rq(src_cpu); | |
6016 | rq_dest = cpu_rq(dest_cpu); | |
6017 | ||
6018 | double_rq_lock(rq_src, rq_dest); | |
6019 | /* Already moved. */ | |
6020 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6021 | goto done; |
1da177e4 LT |
6022 | /* Affinity changed (again). */ |
6023 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 6024 | goto fail; |
1da177e4 | 6025 | |
dd41f596 | 6026 | on_rq = p->se.on_rq; |
6e82a3be | 6027 | if (on_rq) |
2e1cb74a | 6028 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6029 | |
1da177e4 | 6030 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6031 | if (on_rq) { |
6032 | activate_task(rq_dest, p, 0); | |
15afe09b | 6033 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6034 | } |
b1e38734 | 6035 | done: |
efc30814 | 6036 | ret = 1; |
b1e38734 | 6037 | fail: |
1da177e4 | 6038 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6039 | return ret; |
1da177e4 LT |
6040 | } |
6041 | ||
6042 | /* | |
6043 | * migration_thread - this is a highprio system thread that performs | |
6044 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6045 | * another runqueue. | |
6046 | */ | |
95cdf3b7 | 6047 | static int migration_thread(void *data) |
1da177e4 | 6048 | { |
1da177e4 | 6049 | int cpu = (long)data; |
70b97a7f | 6050 | struct rq *rq; |
1da177e4 LT |
6051 | |
6052 | rq = cpu_rq(cpu); | |
6053 | BUG_ON(rq->migration_thread != current); | |
6054 | ||
6055 | set_current_state(TASK_INTERRUPTIBLE); | |
6056 | while (!kthread_should_stop()) { | |
70b97a7f | 6057 | struct migration_req *req; |
1da177e4 | 6058 | struct list_head *head; |
1da177e4 | 6059 | |
1da177e4 LT |
6060 | spin_lock_irq(&rq->lock); |
6061 | ||
6062 | if (cpu_is_offline(cpu)) { | |
6063 | spin_unlock_irq(&rq->lock); | |
6064 | goto wait_to_die; | |
6065 | } | |
6066 | ||
6067 | if (rq->active_balance) { | |
6068 | active_load_balance(rq, cpu); | |
6069 | rq->active_balance = 0; | |
6070 | } | |
6071 | ||
6072 | head = &rq->migration_queue; | |
6073 | ||
6074 | if (list_empty(head)) { | |
6075 | spin_unlock_irq(&rq->lock); | |
6076 | schedule(); | |
6077 | set_current_state(TASK_INTERRUPTIBLE); | |
6078 | continue; | |
6079 | } | |
70b97a7f | 6080 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6081 | list_del_init(head->next); |
6082 | ||
674311d5 NP |
6083 | spin_unlock(&rq->lock); |
6084 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6085 | local_irq_enable(); | |
1da177e4 LT |
6086 | |
6087 | complete(&req->done); | |
6088 | } | |
6089 | __set_current_state(TASK_RUNNING); | |
6090 | return 0; | |
6091 | ||
6092 | wait_to_die: | |
6093 | /* Wait for kthread_stop */ | |
6094 | set_current_state(TASK_INTERRUPTIBLE); | |
6095 | while (!kthread_should_stop()) { | |
6096 | schedule(); | |
6097 | set_current_state(TASK_INTERRUPTIBLE); | |
6098 | } | |
6099 | __set_current_state(TASK_RUNNING); | |
6100 | return 0; | |
6101 | } | |
6102 | ||
6103 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6104 | |
6105 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6106 | { | |
6107 | int ret; | |
6108 | ||
6109 | local_irq_disable(); | |
6110 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6111 | local_irq_enable(); | |
6112 | return ret; | |
6113 | } | |
6114 | ||
054b9108 | 6115 | /* |
3a4fa0a2 | 6116 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6117 | * NOTE: interrupts should be disabled by the caller |
6118 | */ | |
48f24c4d | 6119 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6120 | { |
efc30814 | 6121 | unsigned long flags; |
1da177e4 | 6122 | cpumask_t mask; |
70b97a7f IM |
6123 | struct rq *rq; |
6124 | int dest_cpu; | |
1da177e4 | 6125 | |
3a5c359a AK |
6126 | do { |
6127 | /* On same node? */ | |
6128 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6129 | cpus_and(mask, mask, p->cpus_allowed); | |
6130 | dest_cpu = any_online_cpu(mask); | |
6131 | ||
6132 | /* On any allowed CPU? */ | |
434d53b0 | 6133 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6134 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6135 | ||
6136 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6137 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6138 | cpumask_t cpus_allowed; |
6139 | ||
6140 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6141 | /* |
6142 | * Try to stay on the same cpuset, where the | |
6143 | * current cpuset may be a subset of all cpus. | |
6144 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6145 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6146 | * called within calls to cpuset_lock/cpuset_unlock. |
6147 | */ | |
3a5c359a | 6148 | rq = task_rq_lock(p, &flags); |
470fd646 | 6149 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6150 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6151 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6152 | |
3a5c359a AK |
6153 | /* |
6154 | * Don't tell them about moving exiting tasks or | |
6155 | * kernel threads (both mm NULL), since they never | |
6156 | * leave kernel. | |
6157 | */ | |
41a2d6cf | 6158 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6159 | printk(KERN_INFO "process %d (%s) no " |
6160 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6161 | task_pid_nr(p), p->comm, dead_cpu); |
6162 | } | |
3a5c359a | 6163 | } |
f7b4cddc | 6164 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6165 | } |
6166 | ||
6167 | /* | |
6168 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6169 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6170 | * for performance reasons the counter is not stricly tracking tasks to | |
6171 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6172 | * to keep the global sum constant after CPU-down: | |
6173 | */ | |
70b97a7f | 6174 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6175 | { |
7c16ec58 | 6176 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6177 | unsigned long flags; |
6178 | ||
6179 | local_irq_save(flags); | |
6180 | double_rq_lock(rq_src, rq_dest); | |
6181 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6182 | rq_src->nr_uninterruptible = 0; | |
6183 | double_rq_unlock(rq_src, rq_dest); | |
6184 | local_irq_restore(flags); | |
6185 | } | |
6186 | ||
6187 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6188 | static void migrate_live_tasks(int src_cpu) | |
6189 | { | |
48f24c4d | 6190 | struct task_struct *p, *t; |
1da177e4 | 6191 | |
f7b4cddc | 6192 | read_lock(&tasklist_lock); |
1da177e4 | 6193 | |
48f24c4d IM |
6194 | do_each_thread(t, p) { |
6195 | if (p == current) | |
1da177e4 LT |
6196 | continue; |
6197 | ||
48f24c4d IM |
6198 | if (task_cpu(p) == src_cpu) |
6199 | move_task_off_dead_cpu(src_cpu, p); | |
6200 | } while_each_thread(t, p); | |
1da177e4 | 6201 | |
f7b4cddc | 6202 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6203 | } |
6204 | ||
dd41f596 IM |
6205 | /* |
6206 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6207 | * It does so by boosting its priority to highest possible. |
6208 | * Used by CPU offline code. | |
1da177e4 LT |
6209 | */ |
6210 | void sched_idle_next(void) | |
6211 | { | |
48f24c4d | 6212 | int this_cpu = smp_processor_id(); |
70b97a7f | 6213 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6214 | struct task_struct *p = rq->idle; |
6215 | unsigned long flags; | |
6216 | ||
6217 | /* cpu has to be offline */ | |
48f24c4d | 6218 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6219 | |
48f24c4d IM |
6220 | /* |
6221 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6222 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6223 | */ |
6224 | spin_lock_irqsave(&rq->lock, flags); | |
6225 | ||
dd41f596 | 6226 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6227 | |
94bc9a7b DA |
6228 | update_rq_clock(rq); |
6229 | activate_task(rq, p, 0); | |
1da177e4 LT |
6230 | |
6231 | spin_unlock_irqrestore(&rq->lock, flags); | |
6232 | } | |
6233 | ||
48f24c4d IM |
6234 | /* |
6235 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6236 | * offline. |
6237 | */ | |
6238 | void idle_task_exit(void) | |
6239 | { | |
6240 | struct mm_struct *mm = current->active_mm; | |
6241 | ||
6242 | BUG_ON(cpu_online(smp_processor_id())); | |
6243 | ||
6244 | if (mm != &init_mm) | |
6245 | switch_mm(mm, &init_mm, current); | |
6246 | mmdrop(mm); | |
6247 | } | |
6248 | ||
054b9108 | 6249 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6250 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6251 | { |
70b97a7f | 6252 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6253 | |
6254 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6255 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6256 | |
6257 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6258 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6259 | |
48f24c4d | 6260 | get_task_struct(p); |
1da177e4 LT |
6261 | |
6262 | /* | |
6263 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6264 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6265 | * fine. |
6266 | */ | |
f7b4cddc | 6267 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6268 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6269 | spin_lock_irq(&rq->lock); |
1da177e4 | 6270 | |
48f24c4d | 6271 | put_task_struct(p); |
1da177e4 LT |
6272 | } |
6273 | ||
6274 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6275 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6276 | { | |
70b97a7f | 6277 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6278 | struct task_struct *next; |
48f24c4d | 6279 | |
dd41f596 IM |
6280 | for ( ; ; ) { |
6281 | if (!rq->nr_running) | |
6282 | break; | |
a8e504d2 | 6283 | update_rq_clock(rq); |
ff95f3df | 6284 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6285 | if (!next) |
6286 | break; | |
79c53799 | 6287 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6288 | migrate_dead(dead_cpu, next); |
e692ab53 | 6289 | |
1da177e4 LT |
6290 | } |
6291 | } | |
6292 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6293 | ||
e692ab53 NP |
6294 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6295 | ||
6296 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6297 | { |
6298 | .procname = "sched_domain", | |
c57baf1e | 6299 | .mode = 0555, |
e0361851 | 6300 | }, |
38605cae | 6301 | {0, }, |
e692ab53 NP |
6302 | }; |
6303 | ||
6304 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6305 | { |
c57baf1e | 6306 | .ctl_name = CTL_KERN, |
e0361851 | 6307 | .procname = "kernel", |
c57baf1e | 6308 | .mode = 0555, |
e0361851 AD |
6309 | .child = sd_ctl_dir, |
6310 | }, | |
38605cae | 6311 | {0, }, |
e692ab53 NP |
6312 | }; |
6313 | ||
6314 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6315 | { | |
6316 | struct ctl_table *entry = | |
5cf9f062 | 6317 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6318 | |
e692ab53 NP |
6319 | return entry; |
6320 | } | |
6321 | ||
6382bc90 MM |
6322 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6323 | { | |
cd790076 | 6324 | struct ctl_table *entry; |
6382bc90 | 6325 | |
cd790076 MM |
6326 | /* |
6327 | * In the intermediate directories, both the child directory and | |
6328 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6329 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6330 | * static strings and all have proc handlers. |
6331 | */ | |
6332 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6333 | if (entry->child) |
6334 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6335 | if (entry->proc_handler == NULL) |
6336 | kfree(entry->procname); | |
6337 | } | |
6382bc90 MM |
6338 | |
6339 | kfree(*tablep); | |
6340 | *tablep = NULL; | |
6341 | } | |
6342 | ||
e692ab53 | 6343 | static void |
e0361851 | 6344 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6345 | const char *procname, void *data, int maxlen, |
6346 | mode_t mode, proc_handler *proc_handler) | |
6347 | { | |
e692ab53 NP |
6348 | entry->procname = procname; |
6349 | entry->data = data; | |
6350 | entry->maxlen = maxlen; | |
6351 | entry->mode = mode; | |
6352 | entry->proc_handler = proc_handler; | |
6353 | } | |
6354 | ||
6355 | static struct ctl_table * | |
6356 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6357 | { | |
a5d8c348 | 6358 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6359 | |
ad1cdc1d MM |
6360 | if (table == NULL) |
6361 | return NULL; | |
6362 | ||
e0361851 | 6363 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6364 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6365 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6366 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6367 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6368 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6369 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6370 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6371 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6372 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6373 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6374 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6375 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6376 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6377 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6378 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6379 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6380 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6381 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6382 | &sd->cache_nice_tries, |
6383 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6384 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6385 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6386 | set_table_entry(&table[11], "name", sd->name, |
6387 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6388 | /* &table[12] is terminator */ | |
e692ab53 NP |
6389 | |
6390 | return table; | |
6391 | } | |
6392 | ||
9a4e7159 | 6393 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6394 | { |
6395 | struct ctl_table *entry, *table; | |
6396 | struct sched_domain *sd; | |
6397 | int domain_num = 0, i; | |
6398 | char buf[32]; | |
6399 | ||
6400 | for_each_domain(cpu, sd) | |
6401 | domain_num++; | |
6402 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6403 | if (table == NULL) |
6404 | return NULL; | |
e692ab53 NP |
6405 | |
6406 | i = 0; | |
6407 | for_each_domain(cpu, sd) { | |
6408 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6409 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6410 | entry->mode = 0555; |
e692ab53 NP |
6411 | entry->child = sd_alloc_ctl_domain_table(sd); |
6412 | entry++; | |
6413 | i++; | |
6414 | } | |
6415 | return table; | |
6416 | } | |
6417 | ||
6418 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6419 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6420 | { |
6421 | int i, cpu_num = num_online_cpus(); | |
6422 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6423 | char buf[32]; | |
6424 | ||
7378547f MM |
6425 | WARN_ON(sd_ctl_dir[0].child); |
6426 | sd_ctl_dir[0].child = entry; | |
6427 | ||
ad1cdc1d MM |
6428 | if (entry == NULL) |
6429 | return; | |
6430 | ||
97b6ea7b | 6431 | for_each_online_cpu(i) { |
e692ab53 | 6432 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6433 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6434 | entry->mode = 0555; |
e692ab53 | 6435 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6436 | entry++; |
e692ab53 | 6437 | } |
7378547f MM |
6438 | |
6439 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6440 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6441 | } | |
6382bc90 | 6442 | |
7378547f | 6443 | /* may be called multiple times per register */ |
6382bc90 MM |
6444 | static void unregister_sched_domain_sysctl(void) |
6445 | { | |
7378547f MM |
6446 | if (sd_sysctl_header) |
6447 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6448 | sd_sysctl_header = NULL; |
7378547f MM |
6449 | if (sd_ctl_dir[0].child) |
6450 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6451 | } |
e692ab53 | 6452 | #else |
6382bc90 MM |
6453 | static void register_sched_domain_sysctl(void) |
6454 | { | |
6455 | } | |
6456 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6457 | { |
6458 | } | |
6459 | #endif | |
6460 | ||
1f11eb6a GH |
6461 | static void set_rq_online(struct rq *rq) |
6462 | { | |
6463 | if (!rq->online) { | |
6464 | const struct sched_class *class; | |
6465 | ||
6466 | cpu_set(rq->cpu, rq->rd->online); | |
6467 | rq->online = 1; | |
6468 | ||
6469 | for_each_class(class) { | |
6470 | if (class->rq_online) | |
6471 | class->rq_online(rq); | |
6472 | } | |
6473 | } | |
6474 | } | |
6475 | ||
6476 | static void set_rq_offline(struct rq *rq) | |
6477 | { | |
6478 | if (rq->online) { | |
6479 | const struct sched_class *class; | |
6480 | ||
6481 | for_each_class(class) { | |
6482 | if (class->rq_offline) | |
6483 | class->rq_offline(rq); | |
6484 | } | |
6485 | ||
6486 | cpu_clear(rq->cpu, rq->rd->online); | |
6487 | rq->online = 0; | |
6488 | } | |
6489 | } | |
6490 | ||
1da177e4 LT |
6491 | /* |
6492 | * migration_call - callback that gets triggered when a CPU is added. | |
6493 | * Here we can start up the necessary migration thread for the new CPU. | |
6494 | */ | |
48f24c4d IM |
6495 | static int __cpuinit |
6496 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6497 | { |
1da177e4 | 6498 | struct task_struct *p; |
48f24c4d | 6499 | int cpu = (long)hcpu; |
1da177e4 | 6500 | unsigned long flags; |
70b97a7f | 6501 | struct rq *rq; |
1da177e4 LT |
6502 | |
6503 | switch (action) { | |
5be9361c | 6504 | |
1da177e4 | 6505 | case CPU_UP_PREPARE: |
8bb78442 | 6506 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6507 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6508 | if (IS_ERR(p)) |
6509 | return NOTIFY_BAD; | |
1da177e4 LT |
6510 | kthread_bind(p, cpu); |
6511 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6512 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6513 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6514 | task_rq_unlock(rq, &flags); |
6515 | cpu_rq(cpu)->migration_thread = p; | |
6516 | break; | |
48f24c4d | 6517 | |
1da177e4 | 6518 | case CPU_ONLINE: |
8bb78442 | 6519 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6520 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6521 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6522 | |
6523 | /* Update our root-domain */ | |
6524 | rq = cpu_rq(cpu); | |
6525 | spin_lock_irqsave(&rq->lock, flags); | |
6526 | if (rq->rd) { | |
6527 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6528 | |
6529 | set_rq_online(rq); | |
1f94ef59 GH |
6530 | } |
6531 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6532 | break; |
48f24c4d | 6533 | |
1da177e4 LT |
6534 | #ifdef CONFIG_HOTPLUG_CPU |
6535 | case CPU_UP_CANCELED: | |
8bb78442 | 6536 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6537 | if (!cpu_rq(cpu)->migration_thread) |
6538 | break; | |
41a2d6cf | 6539 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6540 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6541 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6542 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6543 | cpu_rq(cpu)->migration_thread = NULL; | |
6544 | break; | |
48f24c4d | 6545 | |
1da177e4 | 6546 | case CPU_DEAD: |
8bb78442 | 6547 | case CPU_DEAD_FROZEN: |
470fd646 | 6548 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6549 | migrate_live_tasks(cpu); |
6550 | rq = cpu_rq(cpu); | |
6551 | kthread_stop(rq->migration_thread); | |
6552 | rq->migration_thread = NULL; | |
6553 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6554 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6555 | update_rq_clock(rq); |
2e1cb74a | 6556 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6557 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6558 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6559 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6560 | migrate_dead_tasks(cpu); |
d2da272a | 6561 | spin_unlock_irq(&rq->lock); |
470fd646 | 6562 | cpuset_unlock(); |
1da177e4 LT |
6563 | migrate_nr_uninterruptible(rq); |
6564 | BUG_ON(rq->nr_running != 0); | |
6565 | ||
41a2d6cf IM |
6566 | /* |
6567 | * No need to migrate the tasks: it was best-effort if | |
6568 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6569 | * the requestors. | |
6570 | */ | |
1da177e4 LT |
6571 | spin_lock_irq(&rq->lock); |
6572 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6573 | struct migration_req *req; |
6574 | ||
1da177e4 | 6575 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6576 | struct migration_req, list); |
1da177e4 LT |
6577 | list_del_init(&req->list); |
6578 | complete(&req->done); | |
6579 | } | |
6580 | spin_unlock_irq(&rq->lock); | |
6581 | break; | |
57d885fe | 6582 | |
08f503b0 GH |
6583 | case CPU_DYING: |
6584 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6585 | /* Update our root-domain */ |
6586 | rq = cpu_rq(cpu); | |
6587 | spin_lock_irqsave(&rq->lock, flags); | |
6588 | if (rq->rd) { | |
6589 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6590 | set_rq_offline(rq); |
57d885fe GH |
6591 | } |
6592 | spin_unlock_irqrestore(&rq->lock, flags); | |
6593 | break; | |
1da177e4 LT |
6594 | #endif |
6595 | } | |
6596 | return NOTIFY_OK; | |
6597 | } | |
6598 | ||
6599 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6600 | * happens before everything else. | |
6601 | */ | |
26c2143b | 6602 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6603 | .notifier_call = migration_call, |
6604 | .priority = 10 | |
6605 | }; | |
6606 | ||
7babe8db | 6607 | static int __init migration_init(void) |
1da177e4 LT |
6608 | { |
6609 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6610 | int err; |
48f24c4d IM |
6611 | |
6612 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6613 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6614 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6615 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6616 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6617 | |
6618 | return err; | |
1da177e4 | 6619 | } |
7babe8db | 6620 | early_initcall(migration_init); |
1da177e4 LT |
6621 | #endif |
6622 | ||
6623 | #ifdef CONFIG_SMP | |
476f3534 | 6624 | |
3e9830dc | 6625 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6626 | |
099f98c8 GS |
6627 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6628 | { | |
6629 | switch (lvl) { | |
6630 | case SD_LV_NONE: | |
6631 | return "NONE"; | |
6632 | case SD_LV_SIBLING: | |
6633 | return "SIBLING"; | |
6634 | case SD_LV_MC: | |
6635 | return "MC"; | |
6636 | case SD_LV_CPU: | |
6637 | return "CPU"; | |
6638 | case SD_LV_NODE: | |
6639 | return "NODE"; | |
6640 | case SD_LV_ALLNODES: | |
6641 | return "ALLNODES"; | |
6642 | case SD_LV_MAX: | |
6643 | return "MAX"; | |
6644 | ||
6645 | } | |
6646 | return "MAX"; | |
6647 | } | |
6648 | ||
7c16ec58 MT |
6649 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6650 | cpumask_t *groupmask) | |
1da177e4 | 6651 | { |
4dcf6aff | 6652 | struct sched_group *group = sd->groups; |
434d53b0 | 6653 | char str[256]; |
1da177e4 | 6654 | |
434d53b0 | 6655 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6656 | cpus_clear(*groupmask); |
4dcf6aff IM |
6657 | |
6658 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6659 | ||
6660 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6661 | printk("does not load-balance\n"); | |
6662 | if (sd->parent) | |
6663 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6664 | " has parent"); | |
6665 | return -1; | |
41c7ce9a NP |
6666 | } |
6667 | ||
099f98c8 GS |
6668 | printk(KERN_CONT "span %s level %s\n", |
6669 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6670 | |
6671 | if (!cpu_isset(cpu, sd->span)) { | |
6672 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6673 | "CPU%d\n", cpu); | |
6674 | } | |
6675 | if (!cpu_isset(cpu, group->cpumask)) { | |
6676 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6677 | " CPU%d\n", cpu); | |
6678 | } | |
1da177e4 | 6679 | |
4dcf6aff | 6680 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6681 | do { |
4dcf6aff IM |
6682 | if (!group) { |
6683 | printk("\n"); | |
6684 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6685 | break; |
6686 | } | |
6687 | ||
4dcf6aff IM |
6688 | if (!group->__cpu_power) { |
6689 | printk(KERN_CONT "\n"); | |
6690 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6691 | "set\n"); | |
6692 | break; | |
6693 | } | |
1da177e4 | 6694 | |
4dcf6aff IM |
6695 | if (!cpus_weight(group->cpumask)) { |
6696 | printk(KERN_CONT "\n"); | |
6697 | printk(KERN_ERR "ERROR: empty group\n"); | |
6698 | break; | |
6699 | } | |
1da177e4 | 6700 | |
7c16ec58 | 6701 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6702 | printk(KERN_CONT "\n"); |
6703 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6704 | break; | |
6705 | } | |
1da177e4 | 6706 | |
7c16ec58 | 6707 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6708 | |
434d53b0 | 6709 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6710 | printk(KERN_CONT " %s", str); |
1da177e4 | 6711 | |
4dcf6aff IM |
6712 | group = group->next; |
6713 | } while (group != sd->groups); | |
6714 | printk(KERN_CONT "\n"); | |
1da177e4 | 6715 | |
7c16ec58 | 6716 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6717 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6718 | |
7c16ec58 | 6719 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6720 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6721 | "of domain->span\n"); | |
6722 | return 0; | |
6723 | } | |
1da177e4 | 6724 | |
4dcf6aff IM |
6725 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6726 | { | |
7c16ec58 | 6727 | cpumask_t *groupmask; |
4dcf6aff | 6728 | int level = 0; |
1da177e4 | 6729 | |
4dcf6aff IM |
6730 | if (!sd) { |
6731 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6732 | return; | |
6733 | } | |
1da177e4 | 6734 | |
4dcf6aff IM |
6735 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6736 | ||
7c16ec58 MT |
6737 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6738 | if (!groupmask) { | |
6739 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6740 | return; | |
6741 | } | |
6742 | ||
4dcf6aff | 6743 | for (;;) { |
7c16ec58 | 6744 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6745 | break; |
1da177e4 LT |
6746 | level++; |
6747 | sd = sd->parent; | |
33859f7f | 6748 | if (!sd) |
4dcf6aff IM |
6749 | break; |
6750 | } | |
7c16ec58 | 6751 | kfree(groupmask); |
1da177e4 | 6752 | } |
6d6bc0ad | 6753 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6754 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6755 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6756 | |
1a20ff27 | 6757 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6758 | { |
6759 | if (cpus_weight(sd->span) == 1) | |
6760 | return 1; | |
6761 | ||
6762 | /* Following flags need at least 2 groups */ | |
6763 | if (sd->flags & (SD_LOAD_BALANCE | | |
6764 | SD_BALANCE_NEWIDLE | | |
6765 | SD_BALANCE_FORK | | |
89c4710e SS |
6766 | SD_BALANCE_EXEC | |
6767 | SD_SHARE_CPUPOWER | | |
6768 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6769 | if (sd->groups != sd->groups->next) |
6770 | return 0; | |
6771 | } | |
6772 | ||
6773 | /* Following flags don't use groups */ | |
6774 | if (sd->flags & (SD_WAKE_IDLE | | |
6775 | SD_WAKE_AFFINE | | |
6776 | SD_WAKE_BALANCE)) | |
6777 | return 0; | |
6778 | ||
6779 | return 1; | |
6780 | } | |
6781 | ||
48f24c4d IM |
6782 | static int |
6783 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6784 | { |
6785 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6786 | ||
6787 | if (sd_degenerate(parent)) | |
6788 | return 1; | |
6789 | ||
6790 | if (!cpus_equal(sd->span, parent->span)) | |
6791 | return 0; | |
6792 | ||
6793 | /* Does parent contain flags not in child? */ | |
6794 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6795 | if (cflags & SD_WAKE_AFFINE) | |
6796 | pflags &= ~SD_WAKE_BALANCE; | |
6797 | /* Flags needing groups don't count if only 1 group in parent */ | |
6798 | if (parent->groups == parent->groups->next) { | |
6799 | pflags &= ~(SD_LOAD_BALANCE | | |
6800 | SD_BALANCE_NEWIDLE | | |
6801 | SD_BALANCE_FORK | | |
89c4710e SS |
6802 | SD_BALANCE_EXEC | |
6803 | SD_SHARE_CPUPOWER | | |
6804 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6805 | } |
6806 | if (~cflags & pflags) | |
6807 | return 0; | |
6808 | ||
6809 | return 1; | |
6810 | } | |
6811 | ||
57d885fe GH |
6812 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6813 | { | |
6814 | unsigned long flags; | |
57d885fe GH |
6815 | |
6816 | spin_lock_irqsave(&rq->lock, flags); | |
6817 | ||
6818 | if (rq->rd) { | |
6819 | struct root_domain *old_rd = rq->rd; | |
6820 | ||
1f11eb6a GH |
6821 | if (cpu_isset(rq->cpu, old_rd->online)) |
6822 | set_rq_offline(rq); | |
57d885fe | 6823 | |
dc938520 | 6824 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6825 | |
57d885fe GH |
6826 | if (atomic_dec_and_test(&old_rd->refcount)) |
6827 | kfree(old_rd); | |
6828 | } | |
6829 | ||
6830 | atomic_inc(&rd->refcount); | |
6831 | rq->rd = rd; | |
6832 | ||
dc938520 | 6833 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6834 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6835 | set_rq_online(rq); |
57d885fe GH |
6836 | |
6837 | spin_unlock_irqrestore(&rq->lock, flags); | |
6838 | } | |
6839 | ||
dc938520 | 6840 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6841 | { |
6842 | memset(rd, 0, sizeof(*rd)); | |
6843 | ||
dc938520 GH |
6844 | cpus_clear(rd->span); |
6845 | cpus_clear(rd->online); | |
6e0534f2 GH |
6846 | |
6847 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6848 | } |
6849 | ||
6850 | static void init_defrootdomain(void) | |
6851 | { | |
dc938520 | 6852 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6853 | atomic_set(&def_root_domain.refcount, 1); |
6854 | } | |
6855 | ||
dc938520 | 6856 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6857 | { |
6858 | struct root_domain *rd; | |
6859 | ||
6860 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6861 | if (!rd) | |
6862 | return NULL; | |
6863 | ||
dc938520 | 6864 | init_rootdomain(rd); |
57d885fe GH |
6865 | |
6866 | return rd; | |
6867 | } | |
6868 | ||
1da177e4 | 6869 | /* |
0eab9146 | 6870 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6871 | * hold the hotplug lock. |
6872 | */ | |
0eab9146 IM |
6873 | static void |
6874 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6875 | { |
70b97a7f | 6876 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6877 | struct sched_domain *tmp; |
6878 | ||
6879 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6880 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6881 | struct sched_domain *parent = tmp->parent; | |
6882 | if (!parent) | |
6883 | break; | |
1a848870 | 6884 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6885 | tmp->parent = parent->parent; |
1a848870 SS |
6886 | if (parent->parent) |
6887 | parent->parent->child = tmp; | |
6888 | } | |
245af2c7 SS |
6889 | } |
6890 | ||
1a848870 | 6891 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6892 | sd = sd->parent; |
1a848870 SS |
6893 | if (sd) |
6894 | sd->child = NULL; | |
6895 | } | |
1da177e4 LT |
6896 | |
6897 | sched_domain_debug(sd, cpu); | |
6898 | ||
57d885fe | 6899 | rq_attach_root(rq, rd); |
674311d5 | 6900 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6901 | } |
6902 | ||
6903 | /* cpus with isolated domains */ | |
67af63a6 | 6904 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6905 | |
6906 | /* Setup the mask of cpus configured for isolated domains */ | |
6907 | static int __init isolated_cpu_setup(char *str) | |
6908 | { | |
13b40c1e MT |
6909 | static int __initdata ints[NR_CPUS]; |
6910 | int i; | |
1da177e4 LT |
6911 | |
6912 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6913 | cpus_clear(cpu_isolated_map); | |
6914 | for (i = 1; i <= ints[0]; i++) | |
6915 | if (ints[i] < NR_CPUS) | |
6916 | cpu_set(ints[i], cpu_isolated_map); | |
6917 | return 1; | |
6918 | } | |
6919 | ||
8927f494 | 6920 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6921 | |
6922 | /* | |
6711cab4 SS |
6923 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6924 | * to a function which identifies what group(along with sched group) a CPU | |
6925 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6926 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6927 | * |
6928 | * init_sched_build_groups will build a circular linked list of the groups | |
6929 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6930 | * and ->cpu_power to 0. | |
6931 | */ | |
a616058b | 6932 | static void |
7c16ec58 | 6933 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6934 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6935 | struct sched_group **sg, |
6936 | cpumask_t *tmpmask), | |
6937 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6938 | { |
6939 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6940 | int i; |
6941 | ||
7c16ec58 MT |
6942 | cpus_clear(*covered); |
6943 | ||
363ab6f1 | 6944 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6945 | struct sched_group *sg; |
7c16ec58 | 6946 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6947 | int j; |
6948 | ||
7c16ec58 | 6949 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6950 | continue; |
6951 | ||
7c16ec58 | 6952 | cpus_clear(sg->cpumask); |
5517d86b | 6953 | sg->__cpu_power = 0; |
1da177e4 | 6954 | |
363ab6f1 | 6955 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6956 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6957 | continue; |
6958 | ||
7c16ec58 | 6959 | cpu_set(j, *covered); |
1da177e4 LT |
6960 | cpu_set(j, sg->cpumask); |
6961 | } | |
6962 | if (!first) | |
6963 | first = sg; | |
6964 | if (last) | |
6965 | last->next = sg; | |
6966 | last = sg; | |
6967 | } | |
6968 | last->next = first; | |
6969 | } | |
6970 | ||
9c1cfda2 | 6971 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6972 | |
9c1cfda2 | 6973 | #ifdef CONFIG_NUMA |
198e2f18 | 6974 | |
9c1cfda2 JH |
6975 | /** |
6976 | * find_next_best_node - find the next node to include in a sched_domain | |
6977 | * @node: node whose sched_domain we're building | |
6978 | * @used_nodes: nodes already in the sched_domain | |
6979 | * | |
41a2d6cf | 6980 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6981 | * finds the closest node not already in the @used_nodes map. |
6982 | * | |
6983 | * Should use nodemask_t. | |
6984 | */ | |
c5f59f08 | 6985 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6986 | { |
6987 | int i, n, val, min_val, best_node = 0; | |
6988 | ||
6989 | min_val = INT_MAX; | |
6990 | ||
076ac2af | 6991 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6992 | /* Start at @node */ |
076ac2af | 6993 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6994 | |
6995 | if (!nr_cpus_node(n)) | |
6996 | continue; | |
6997 | ||
6998 | /* Skip already used nodes */ | |
c5f59f08 | 6999 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7000 | continue; |
7001 | ||
7002 | /* Simple min distance search */ | |
7003 | val = node_distance(node, n); | |
7004 | ||
7005 | if (val < min_val) { | |
7006 | min_val = val; | |
7007 | best_node = n; | |
7008 | } | |
7009 | } | |
7010 | ||
c5f59f08 | 7011 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7012 | return best_node; |
7013 | } | |
7014 | ||
7015 | /** | |
7016 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7017 | * @node: node whose cpumask we're constructing | |
73486722 | 7018 | * @span: resulting cpumask |
9c1cfda2 | 7019 | * |
41a2d6cf | 7020 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7021 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7022 | * out optimally. | |
7023 | */ | |
4bdbaad3 | 7024 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 7025 | { |
c5f59f08 | 7026 | nodemask_t used_nodes; |
c5f59f08 | 7027 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7028 | int i; |
9c1cfda2 | 7029 | |
4bdbaad3 | 7030 | cpus_clear(*span); |
c5f59f08 | 7031 | nodes_clear(used_nodes); |
9c1cfda2 | 7032 | |
4bdbaad3 | 7033 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7034 | node_set(node, used_nodes); |
9c1cfda2 JH |
7035 | |
7036 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7037 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7038 | |
c5f59f08 | 7039 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7040 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7041 | } |
9c1cfda2 | 7042 | } |
6d6bc0ad | 7043 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7044 | |
5c45bf27 | 7045 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7046 | |
9c1cfda2 | 7047 | /* |
48f24c4d | 7048 | * SMT sched-domains: |
9c1cfda2 | 7049 | */ |
1da177e4 LT |
7050 | #ifdef CONFIG_SCHED_SMT |
7051 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7052 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7053 | |
41a2d6cf | 7054 | static int |
7c16ec58 MT |
7055 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7056 | cpumask_t *unused) | |
1da177e4 | 7057 | { |
6711cab4 SS |
7058 | if (sg) |
7059 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7060 | return cpu; |
7061 | } | |
6d6bc0ad | 7062 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7063 | |
48f24c4d IM |
7064 | /* |
7065 | * multi-core sched-domains: | |
7066 | */ | |
1e9f28fa SS |
7067 | #ifdef CONFIG_SCHED_MC |
7068 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7069 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7070 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7071 | |
7072 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7073 | static int |
7c16ec58 MT |
7074 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7075 | cpumask_t *mask) | |
1e9f28fa | 7076 | { |
6711cab4 | 7077 | int group; |
7c16ec58 MT |
7078 | |
7079 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7080 | cpus_and(*mask, *mask, *cpu_map); | |
7081 | group = first_cpu(*mask); | |
6711cab4 SS |
7082 | if (sg) |
7083 | *sg = &per_cpu(sched_group_core, group); | |
7084 | return group; | |
1e9f28fa SS |
7085 | } |
7086 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7087 | static int |
7c16ec58 MT |
7088 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7089 | cpumask_t *unused) | |
1e9f28fa | 7090 | { |
6711cab4 SS |
7091 | if (sg) |
7092 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7093 | return cpu; |
7094 | } | |
7095 | #endif | |
7096 | ||
1da177e4 | 7097 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7098 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7099 | |
41a2d6cf | 7100 | static int |
7c16ec58 MT |
7101 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7102 | cpumask_t *mask) | |
1da177e4 | 7103 | { |
6711cab4 | 7104 | int group; |
48f24c4d | 7105 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7106 | *mask = cpu_coregroup_map(cpu); |
7107 | cpus_and(*mask, *mask, *cpu_map); | |
7108 | group = first_cpu(*mask); | |
1e9f28fa | 7109 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7110 | *mask = per_cpu(cpu_sibling_map, cpu); |
7111 | cpus_and(*mask, *mask, *cpu_map); | |
7112 | group = first_cpu(*mask); | |
1da177e4 | 7113 | #else |
6711cab4 | 7114 | group = cpu; |
1da177e4 | 7115 | #endif |
6711cab4 SS |
7116 | if (sg) |
7117 | *sg = &per_cpu(sched_group_phys, group); | |
7118 | return group; | |
1da177e4 LT |
7119 | } |
7120 | ||
7121 | #ifdef CONFIG_NUMA | |
1da177e4 | 7122 | /* |
9c1cfda2 JH |
7123 | * The init_sched_build_groups can't handle what we want to do with node |
7124 | * groups, so roll our own. Now each node has its own list of groups which | |
7125 | * gets dynamically allocated. | |
1da177e4 | 7126 | */ |
9c1cfda2 | 7127 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7128 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7129 | |
9c1cfda2 | 7130 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7131 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7132 | |
6711cab4 | 7133 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7134 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7135 | { |
6711cab4 SS |
7136 | int group; |
7137 | ||
7c16ec58 MT |
7138 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7139 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7140 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7141 | |
7142 | if (sg) | |
7143 | *sg = &per_cpu(sched_group_allnodes, group); | |
7144 | return group; | |
1da177e4 | 7145 | } |
6711cab4 | 7146 | |
08069033 SS |
7147 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7148 | { | |
7149 | struct sched_group *sg = group_head; | |
7150 | int j; | |
7151 | ||
7152 | if (!sg) | |
7153 | return; | |
3a5c359a | 7154 | do { |
363ab6f1 | 7155 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7156 | struct sched_domain *sd; |
08069033 | 7157 | |
3a5c359a AK |
7158 | sd = &per_cpu(phys_domains, j); |
7159 | if (j != first_cpu(sd->groups->cpumask)) { | |
7160 | /* | |
7161 | * Only add "power" once for each | |
7162 | * physical package. | |
7163 | */ | |
7164 | continue; | |
7165 | } | |
08069033 | 7166 | |
3a5c359a AK |
7167 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7168 | } | |
7169 | sg = sg->next; | |
7170 | } while (sg != group_head); | |
08069033 | 7171 | } |
6d6bc0ad | 7172 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7173 | |
a616058b | 7174 | #ifdef CONFIG_NUMA |
51888ca2 | 7175 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7176 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7177 | { |
a616058b | 7178 | int cpu, i; |
51888ca2 | 7179 | |
363ab6f1 | 7180 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7181 | struct sched_group **sched_group_nodes |
7182 | = sched_group_nodes_bycpu[cpu]; | |
7183 | ||
51888ca2 SV |
7184 | if (!sched_group_nodes) |
7185 | continue; | |
7186 | ||
076ac2af | 7187 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7188 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7189 | ||
7c16ec58 MT |
7190 | *nodemask = node_to_cpumask(i); |
7191 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7192 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7193 | continue; |
7194 | ||
7195 | if (sg == NULL) | |
7196 | continue; | |
7197 | sg = sg->next; | |
7198 | next_sg: | |
7199 | oldsg = sg; | |
7200 | sg = sg->next; | |
7201 | kfree(oldsg); | |
7202 | if (oldsg != sched_group_nodes[i]) | |
7203 | goto next_sg; | |
7204 | } | |
7205 | kfree(sched_group_nodes); | |
7206 | sched_group_nodes_bycpu[cpu] = NULL; | |
7207 | } | |
51888ca2 | 7208 | } |
6d6bc0ad | 7209 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7210 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7211 | { |
7212 | } | |
6d6bc0ad | 7213 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7214 | |
89c4710e SS |
7215 | /* |
7216 | * Initialize sched groups cpu_power. | |
7217 | * | |
7218 | * cpu_power indicates the capacity of sched group, which is used while | |
7219 | * distributing the load between different sched groups in a sched domain. | |
7220 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7221 | * there are asymmetries in the topology. If there are asymmetries, group | |
7222 | * having more cpu_power will pickup more load compared to the group having | |
7223 | * less cpu_power. | |
7224 | * | |
7225 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7226 | * the maximum number of tasks a group can handle in the presence of other idle | |
7227 | * or lightly loaded groups in the same sched domain. | |
7228 | */ | |
7229 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7230 | { | |
7231 | struct sched_domain *child; | |
7232 | struct sched_group *group; | |
7233 | ||
7234 | WARN_ON(!sd || !sd->groups); | |
7235 | ||
7236 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7237 | return; | |
7238 | ||
7239 | child = sd->child; | |
7240 | ||
5517d86b ED |
7241 | sd->groups->__cpu_power = 0; |
7242 | ||
89c4710e SS |
7243 | /* |
7244 | * For perf policy, if the groups in child domain share resources | |
7245 | * (for example cores sharing some portions of the cache hierarchy | |
7246 | * or SMT), then set this domain groups cpu_power such that each group | |
7247 | * can handle only one task, when there are other idle groups in the | |
7248 | * same sched domain. | |
7249 | */ | |
7250 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7251 | (child->flags & | |
7252 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7253 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7254 | return; |
7255 | } | |
7256 | ||
89c4710e SS |
7257 | /* |
7258 | * add cpu_power of each child group to this groups cpu_power | |
7259 | */ | |
7260 | group = child->groups; | |
7261 | do { | |
5517d86b | 7262 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7263 | group = group->next; |
7264 | } while (group != child->groups); | |
7265 | } | |
7266 | ||
7c16ec58 MT |
7267 | /* |
7268 | * Initializers for schedule domains | |
7269 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7270 | */ | |
7271 | ||
a5d8c348 IM |
7272 | #ifdef CONFIG_SCHED_DEBUG |
7273 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7274 | #else | |
7275 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7276 | #endif | |
7277 | ||
7c16ec58 | 7278 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7279 | |
7c16ec58 MT |
7280 | #define SD_INIT_FUNC(type) \ |
7281 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7282 | { \ | |
7283 | memset(sd, 0, sizeof(*sd)); \ | |
7284 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7285 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7286 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7287 | } |
7288 | ||
7289 | SD_INIT_FUNC(CPU) | |
7290 | #ifdef CONFIG_NUMA | |
7291 | SD_INIT_FUNC(ALLNODES) | |
7292 | SD_INIT_FUNC(NODE) | |
7293 | #endif | |
7294 | #ifdef CONFIG_SCHED_SMT | |
7295 | SD_INIT_FUNC(SIBLING) | |
7296 | #endif | |
7297 | #ifdef CONFIG_SCHED_MC | |
7298 | SD_INIT_FUNC(MC) | |
7299 | #endif | |
7300 | ||
7301 | /* | |
7302 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7303 | * space as the usage in build_sched_domains() dictates. Used only | |
7304 | * if the amount of space is significant. | |
7305 | */ | |
7306 | struct allmasks { | |
7307 | cpumask_t tmpmask; /* make this one first */ | |
7308 | union { | |
7309 | cpumask_t nodemask; | |
7310 | cpumask_t this_sibling_map; | |
7311 | cpumask_t this_core_map; | |
7312 | }; | |
7313 | cpumask_t send_covered; | |
7314 | ||
7315 | #ifdef CONFIG_NUMA | |
7316 | cpumask_t domainspan; | |
7317 | cpumask_t covered; | |
7318 | cpumask_t notcovered; | |
7319 | #endif | |
7320 | }; | |
7321 | ||
7322 | #if NR_CPUS > 128 | |
7323 | #define SCHED_CPUMASK_ALLOC 1 | |
7324 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7325 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7326 | #else | |
7327 | #define SCHED_CPUMASK_ALLOC 0 | |
7328 | #define SCHED_CPUMASK_FREE(v) | |
7329 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7330 | #endif | |
7331 | ||
7332 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7333 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7334 | ||
1d3504fc HS |
7335 | static int default_relax_domain_level = -1; |
7336 | ||
7337 | static int __init setup_relax_domain_level(char *str) | |
7338 | { | |
30e0e178 LZ |
7339 | unsigned long val; |
7340 | ||
7341 | val = simple_strtoul(str, NULL, 0); | |
7342 | if (val < SD_LV_MAX) | |
7343 | default_relax_domain_level = val; | |
7344 | ||
1d3504fc HS |
7345 | return 1; |
7346 | } | |
7347 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7348 | ||
7349 | static void set_domain_attribute(struct sched_domain *sd, | |
7350 | struct sched_domain_attr *attr) | |
7351 | { | |
7352 | int request; | |
7353 | ||
7354 | if (!attr || attr->relax_domain_level < 0) { | |
7355 | if (default_relax_domain_level < 0) | |
7356 | return; | |
7357 | else | |
7358 | request = default_relax_domain_level; | |
7359 | } else | |
7360 | request = attr->relax_domain_level; | |
7361 | if (request < sd->level) { | |
7362 | /* turn off idle balance on this domain */ | |
7363 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7364 | } else { | |
7365 | /* turn on idle balance on this domain */ | |
7366 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7367 | } | |
7368 | } | |
7369 | ||
1da177e4 | 7370 | /* |
1a20ff27 DG |
7371 | * Build sched domains for a given set of cpus and attach the sched domains |
7372 | * to the individual cpus | |
1da177e4 | 7373 | */ |
1d3504fc HS |
7374 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7375 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7376 | { |
7377 | int i; | |
57d885fe | 7378 | struct root_domain *rd; |
7c16ec58 MT |
7379 | SCHED_CPUMASK_DECLARE(allmasks); |
7380 | cpumask_t *tmpmask; | |
d1b55138 JH |
7381 | #ifdef CONFIG_NUMA |
7382 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7383 | int sd_allnodes = 0; |
d1b55138 JH |
7384 | |
7385 | /* | |
7386 | * Allocate the per-node list of sched groups | |
7387 | */ | |
076ac2af | 7388 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7389 | GFP_KERNEL); |
d1b55138 JH |
7390 | if (!sched_group_nodes) { |
7391 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7392 | return -ENOMEM; |
d1b55138 | 7393 | } |
d1b55138 | 7394 | #endif |
1da177e4 | 7395 | |
dc938520 | 7396 | rd = alloc_rootdomain(); |
57d885fe GH |
7397 | if (!rd) { |
7398 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7399 | #ifdef CONFIG_NUMA |
7400 | kfree(sched_group_nodes); | |
7401 | #endif | |
57d885fe GH |
7402 | return -ENOMEM; |
7403 | } | |
7404 | ||
7c16ec58 MT |
7405 | #if SCHED_CPUMASK_ALLOC |
7406 | /* get space for all scratch cpumask variables */ | |
7407 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7408 | if (!allmasks) { | |
7409 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7410 | kfree(rd); | |
7411 | #ifdef CONFIG_NUMA | |
7412 | kfree(sched_group_nodes); | |
7413 | #endif | |
7414 | return -ENOMEM; | |
7415 | } | |
7416 | #endif | |
7417 | tmpmask = (cpumask_t *)allmasks; | |
7418 | ||
7419 | ||
7420 | #ifdef CONFIG_NUMA | |
7421 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7422 | #endif | |
7423 | ||
1da177e4 | 7424 | /* |
1a20ff27 | 7425 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7426 | */ |
363ab6f1 | 7427 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7428 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7429 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7430 | |
7c16ec58 MT |
7431 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7432 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7433 | |
7434 | #ifdef CONFIG_NUMA | |
dd41f596 | 7435 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7436 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7437 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7438 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7439 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7440 | sd->span = *cpu_map; |
7c16ec58 | 7441 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7442 | p = sd; |
6711cab4 | 7443 | sd_allnodes = 1; |
9c1cfda2 JH |
7444 | } else |
7445 | p = NULL; | |
7446 | ||
1da177e4 | 7447 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7448 | SD_INIT(sd, NODE); |
1d3504fc | 7449 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7450 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7451 | sd->parent = p; |
1a848870 SS |
7452 | if (p) |
7453 | p->child = sd; | |
9c1cfda2 | 7454 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7455 | #endif |
7456 | ||
7457 | p = sd; | |
7458 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7459 | SD_INIT(sd, CPU); |
1d3504fc | 7460 | set_domain_attribute(sd, attr); |
7c16ec58 | 7461 | sd->span = *nodemask; |
1da177e4 | 7462 | sd->parent = p; |
1a848870 SS |
7463 | if (p) |
7464 | p->child = sd; | |
7c16ec58 | 7465 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7466 | |
1e9f28fa SS |
7467 | #ifdef CONFIG_SCHED_MC |
7468 | p = sd; | |
7469 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7470 | SD_INIT(sd, MC); |
1d3504fc | 7471 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7472 | sd->span = cpu_coregroup_map(i); |
7473 | cpus_and(sd->span, sd->span, *cpu_map); | |
7474 | sd->parent = p; | |
1a848870 | 7475 | p->child = sd; |
7c16ec58 | 7476 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7477 | #endif |
7478 | ||
1da177e4 LT |
7479 | #ifdef CONFIG_SCHED_SMT |
7480 | p = sd; | |
7481 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7482 | SD_INIT(sd, SIBLING); |
1d3504fc | 7483 | set_domain_attribute(sd, attr); |
d5a7430d | 7484 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7485 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7486 | sd->parent = p; |
1a848870 | 7487 | p->child = sd; |
7c16ec58 | 7488 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7489 | #endif |
7490 | } | |
7491 | ||
7492 | #ifdef CONFIG_SCHED_SMT | |
7493 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7494 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7495 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7496 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7497 | ||
7498 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7499 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7500 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7501 | continue; |
7502 | ||
dd41f596 | 7503 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7504 | &cpu_to_cpu_group, |
7505 | send_covered, tmpmask); | |
1da177e4 LT |
7506 | } |
7507 | #endif | |
7508 | ||
1e9f28fa SS |
7509 | #ifdef CONFIG_SCHED_MC |
7510 | /* Set up multi-core groups */ | |
363ab6f1 | 7511 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7512 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7513 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7514 | ||
7515 | *this_core_map = cpu_coregroup_map(i); | |
7516 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7517 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7518 | continue; |
7c16ec58 | 7519 | |
dd41f596 | 7520 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7521 | &cpu_to_core_group, |
7522 | send_covered, tmpmask); | |
1e9f28fa SS |
7523 | } |
7524 | #endif | |
7525 | ||
1da177e4 | 7526 | /* Set up physical groups */ |
076ac2af | 7527 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7528 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7529 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7530 | |
7c16ec58 MT |
7531 | *nodemask = node_to_cpumask(i); |
7532 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7533 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7534 | continue; |
7535 | ||
7c16ec58 MT |
7536 | init_sched_build_groups(nodemask, cpu_map, |
7537 | &cpu_to_phys_group, | |
7538 | send_covered, tmpmask); | |
1da177e4 LT |
7539 | } |
7540 | ||
7541 | #ifdef CONFIG_NUMA | |
7542 | /* Set up node groups */ | |
7c16ec58 MT |
7543 | if (sd_allnodes) { |
7544 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7545 | ||
7546 | init_sched_build_groups(cpu_map, cpu_map, | |
7547 | &cpu_to_allnodes_group, | |
7548 | send_covered, tmpmask); | |
7549 | } | |
9c1cfda2 | 7550 | |
076ac2af | 7551 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7552 | /* Set up node groups */ |
7553 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7554 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7555 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7556 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7557 | int j; |
7558 | ||
7c16ec58 MT |
7559 | *nodemask = node_to_cpumask(i); |
7560 | cpus_clear(*covered); | |
7561 | ||
7562 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7563 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7564 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7565 | continue; |
d1b55138 | 7566 | } |
9c1cfda2 | 7567 | |
4bdbaad3 | 7568 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7569 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7570 | |
15f0b676 | 7571 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7572 | if (!sg) { |
7573 | printk(KERN_WARNING "Can not alloc domain group for " | |
7574 | "node %d\n", i); | |
7575 | goto error; | |
7576 | } | |
9c1cfda2 | 7577 | sched_group_nodes[i] = sg; |
363ab6f1 | 7578 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7579 | struct sched_domain *sd; |
9761eea8 | 7580 | |
9c1cfda2 JH |
7581 | sd = &per_cpu(node_domains, j); |
7582 | sd->groups = sg; | |
9c1cfda2 | 7583 | } |
5517d86b | 7584 | sg->__cpu_power = 0; |
7c16ec58 | 7585 | sg->cpumask = *nodemask; |
51888ca2 | 7586 | sg->next = sg; |
7c16ec58 | 7587 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7588 | prev = sg; |
7589 | ||
076ac2af | 7590 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7591 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7592 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7593 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7594 | |
7c16ec58 MT |
7595 | cpus_complement(*notcovered, *covered); |
7596 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7597 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7598 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7599 | break; |
7600 | ||
7c16ec58 MT |
7601 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7602 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7603 | continue; |
7604 | ||
15f0b676 SV |
7605 | sg = kmalloc_node(sizeof(struct sched_group), |
7606 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7607 | if (!sg) { |
7608 | printk(KERN_WARNING | |
7609 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7610 | goto error; |
9c1cfda2 | 7611 | } |
5517d86b | 7612 | sg->__cpu_power = 0; |
7c16ec58 | 7613 | sg->cpumask = *tmpmask; |
51888ca2 | 7614 | sg->next = prev->next; |
7c16ec58 | 7615 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7616 | prev->next = sg; |
7617 | prev = sg; | |
7618 | } | |
9c1cfda2 | 7619 | } |
1da177e4 LT |
7620 | #endif |
7621 | ||
7622 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7623 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7624 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7625 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7626 | ||
89c4710e | 7627 | init_sched_groups_power(i, sd); |
5c45bf27 | 7628 | } |
1da177e4 | 7629 | #endif |
1e9f28fa | 7630 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7631 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7632 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7633 | ||
89c4710e | 7634 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7635 | } |
7636 | #endif | |
1e9f28fa | 7637 | |
363ab6f1 | 7638 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7639 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7640 | ||
89c4710e | 7641 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7642 | } |
7643 | ||
9c1cfda2 | 7644 | #ifdef CONFIG_NUMA |
076ac2af | 7645 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7646 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7647 | |
6711cab4 SS |
7648 | if (sd_allnodes) { |
7649 | struct sched_group *sg; | |
f712c0c7 | 7650 | |
7c16ec58 MT |
7651 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7652 | tmpmask); | |
f712c0c7 SS |
7653 | init_numa_sched_groups_power(sg); |
7654 | } | |
9c1cfda2 JH |
7655 | #endif |
7656 | ||
1da177e4 | 7657 | /* Attach the domains */ |
363ab6f1 | 7658 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7659 | struct sched_domain *sd; |
7660 | #ifdef CONFIG_SCHED_SMT | |
7661 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7662 | #elif defined(CONFIG_SCHED_MC) |
7663 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7664 | #else |
7665 | sd = &per_cpu(phys_domains, i); | |
7666 | #endif | |
57d885fe | 7667 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7668 | } |
51888ca2 | 7669 | |
7c16ec58 | 7670 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7671 | return 0; |
7672 | ||
a616058b | 7673 | #ifdef CONFIG_NUMA |
51888ca2 | 7674 | error: |
7c16ec58 MT |
7675 | free_sched_groups(cpu_map, tmpmask); |
7676 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7677 | return -ENOMEM; |
a616058b | 7678 | #endif |
1da177e4 | 7679 | } |
029190c5 | 7680 | |
1d3504fc HS |
7681 | static int build_sched_domains(const cpumask_t *cpu_map) |
7682 | { | |
7683 | return __build_sched_domains(cpu_map, NULL); | |
7684 | } | |
7685 | ||
029190c5 PJ |
7686 | static cpumask_t *doms_cur; /* current sched domains */ |
7687 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7688 | static struct sched_domain_attr *dattr_cur; |
7689 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7690 | |
7691 | /* | |
7692 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7693 | * cpumask_t) fails, then fallback to a single sched domain, | |
7694 | * as determined by the single cpumask_t fallback_doms. | |
7695 | */ | |
7696 | static cpumask_t fallback_doms; | |
7697 | ||
22e52b07 HC |
7698 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7699 | { | |
7700 | } | |
7701 | ||
1a20ff27 | 7702 | /* |
41a2d6cf | 7703 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7704 | * For now this just excludes isolated cpus, but could be used to |
7705 | * exclude other special cases in the future. | |
1a20ff27 | 7706 | */ |
51888ca2 | 7707 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7708 | { |
7378547f MM |
7709 | int err; |
7710 | ||
22e52b07 | 7711 | arch_update_cpu_topology(); |
029190c5 PJ |
7712 | ndoms_cur = 1; |
7713 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7714 | if (!doms_cur) | |
7715 | doms_cur = &fallback_doms; | |
7716 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7717 | dattr_cur = NULL; |
7378547f | 7718 | err = build_sched_domains(doms_cur); |
6382bc90 | 7719 | register_sched_domain_sysctl(); |
7378547f MM |
7720 | |
7721 | return err; | |
1a20ff27 DG |
7722 | } |
7723 | ||
7c16ec58 MT |
7724 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7725 | cpumask_t *tmpmask) | |
1da177e4 | 7726 | { |
7c16ec58 | 7727 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7728 | } |
1da177e4 | 7729 | |
1a20ff27 DG |
7730 | /* |
7731 | * Detach sched domains from a group of cpus specified in cpu_map | |
7732 | * These cpus will now be attached to the NULL domain | |
7733 | */ | |
858119e1 | 7734 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7735 | { |
7c16ec58 | 7736 | cpumask_t tmpmask; |
1a20ff27 DG |
7737 | int i; |
7738 | ||
6382bc90 MM |
7739 | unregister_sched_domain_sysctl(); |
7740 | ||
363ab6f1 | 7741 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7742 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7743 | synchronize_sched(); |
7c16ec58 | 7744 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7745 | } |
7746 | ||
1d3504fc HS |
7747 | /* handle null as "default" */ |
7748 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7749 | struct sched_domain_attr *new, int idx_new) | |
7750 | { | |
7751 | struct sched_domain_attr tmp; | |
7752 | ||
7753 | /* fast path */ | |
7754 | if (!new && !cur) | |
7755 | return 1; | |
7756 | ||
7757 | tmp = SD_ATTR_INIT; | |
7758 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7759 | new ? (new + idx_new) : &tmp, | |
7760 | sizeof(struct sched_domain_attr)); | |
7761 | } | |
7762 | ||
029190c5 PJ |
7763 | /* |
7764 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7765 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7766 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7767 | * It destroys each deleted domain and builds each new domain. | |
7768 | * | |
7769 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7770 | * The masks don't intersect (don't overlap.) We should setup one |
7771 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7772 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7773 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7774 | * it as it is. | |
7775 | * | |
41a2d6cf IM |
7776 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7777 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7778 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7779 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7780 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 | 7781 | * |
dfb512ec MK |
7782 | * If doms_new==NULL it will be replaced with cpu_online_map. |
7783 | * ndoms_new==0 is a special case for destroying existing domains. | |
7784 | * It will not create the default domain. | |
7785 | * | |
029190c5 PJ |
7786 | * Call with hotplug lock held |
7787 | */ | |
1d3504fc HS |
7788 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7789 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7790 | { |
dfb512ec | 7791 | int i, j, n; |
029190c5 | 7792 | |
712555ee | 7793 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7794 | |
7378547f MM |
7795 | /* always unregister in case we don't destroy any domains */ |
7796 | unregister_sched_domain_sysctl(); | |
7797 | ||
dfb512ec | 7798 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7799 | |
7800 | /* Destroy deleted domains */ | |
7801 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7802 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7803 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7804 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7805 | goto match1; |
7806 | } | |
7807 | /* no match - a current sched domain not in new doms_new[] */ | |
7808 | detach_destroy_domains(doms_cur + i); | |
7809 | match1: | |
7810 | ; | |
7811 | } | |
7812 | ||
e761b772 MK |
7813 | if (doms_new == NULL) { |
7814 | ndoms_cur = 0; | |
e761b772 MK |
7815 | doms_new = &fallback_doms; |
7816 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7817 | dattr_new = NULL; | |
7818 | } | |
7819 | ||
029190c5 PJ |
7820 | /* Build new domains */ |
7821 | for (i = 0; i < ndoms_new; i++) { | |
7822 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7823 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7824 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7825 | goto match2; |
7826 | } | |
7827 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7828 | __build_sched_domains(doms_new + i, |
7829 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7830 | match2: |
7831 | ; | |
7832 | } | |
7833 | ||
7834 | /* Remember the new sched domains */ | |
7835 | if (doms_cur != &fallback_doms) | |
7836 | kfree(doms_cur); | |
1d3504fc | 7837 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7838 | doms_cur = doms_new; |
1d3504fc | 7839 | dattr_cur = dattr_new; |
029190c5 | 7840 | ndoms_cur = ndoms_new; |
7378547f MM |
7841 | |
7842 | register_sched_domain_sysctl(); | |
a1835615 | 7843 | |
712555ee | 7844 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7845 | } |
7846 | ||
5c45bf27 | 7847 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7848 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7849 | { |
95402b38 | 7850 | get_online_cpus(); |
dfb512ec MK |
7851 | |
7852 | /* Destroy domains first to force the rebuild */ | |
7853 | partition_sched_domains(0, NULL, NULL); | |
7854 | ||
e761b772 | 7855 | rebuild_sched_domains(); |
95402b38 | 7856 | put_online_cpus(); |
dfb512ec | 7857 | |
e761b772 | 7858 | return 0; |
5c45bf27 SS |
7859 | } |
7860 | ||
7861 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7862 | { | |
7863 | int ret; | |
7864 | ||
7865 | if (buf[0] != '0' && buf[0] != '1') | |
7866 | return -EINVAL; | |
7867 | ||
7868 | if (smt) | |
7869 | sched_smt_power_savings = (buf[0] == '1'); | |
7870 | else | |
7871 | sched_mc_power_savings = (buf[0] == '1'); | |
7872 | ||
7873 | ret = arch_reinit_sched_domains(); | |
7874 | ||
7875 | return ret ? ret : count; | |
7876 | } | |
7877 | ||
5c45bf27 | 7878 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7879 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7880 | char *page) | |
5c45bf27 SS |
7881 | { |
7882 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7883 | } | |
f718cd4a | 7884 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7885 | const char *buf, size_t count) |
5c45bf27 SS |
7886 | { |
7887 | return sched_power_savings_store(buf, count, 0); | |
7888 | } | |
f718cd4a AK |
7889 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7890 | sched_mc_power_savings_show, | |
7891 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7892 | #endif |
7893 | ||
7894 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7895 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7896 | char *page) | |
5c45bf27 SS |
7897 | { |
7898 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7899 | } | |
f718cd4a | 7900 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7901 | const char *buf, size_t count) |
5c45bf27 SS |
7902 | { |
7903 | return sched_power_savings_store(buf, count, 1); | |
7904 | } | |
f718cd4a AK |
7905 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7906 | sched_smt_power_savings_show, | |
6707de00 AB |
7907 | sched_smt_power_savings_store); |
7908 | #endif | |
7909 | ||
7910 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7911 | { | |
7912 | int err = 0; | |
7913 | ||
7914 | #ifdef CONFIG_SCHED_SMT | |
7915 | if (smt_capable()) | |
7916 | err = sysfs_create_file(&cls->kset.kobj, | |
7917 | &attr_sched_smt_power_savings.attr); | |
7918 | #endif | |
7919 | #ifdef CONFIG_SCHED_MC | |
7920 | if (!err && mc_capable()) | |
7921 | err = sysfs_create_file(&cls->kset.kobj, | |
7922 | &attr_sched_mc_power_savings.attr); | |
7923 | #endif | |
7924 | return err; | |
7925 | } | |
6d6bc0ad | 7926 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7927 | |
e761b772 | 7928 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7929 | /* |
e761b772 MK |
7930 | * Add online and remove offline CPUs from the scheduler domains. |
7931 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7932 | */ |
7933 | static int update_sched_domains(struct notifier_block *nfb, | |
7934 | unsigned long action, void *hcpu) | |
e761b772 MK |
7935 | { |
7936 | switch (action) { | |
7937 | case CPU_ONLINE: | |
7938 | case CPU_ONLINE_FROZEN: | |
7939 | case CPU_DEAD: | |
7940 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7941 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7942 | return NOTIFY_OK; |
7943 | ||
7944 | default: | |
7945 | return NOTIFY_DONE; | |
7946 | } | |
7947 | } | |
7948 | #endif | |
7949 | ||
7950 | static int update_runtime(struct notifier_block *nfb, | |
7951 | unsigned long action, void *hcpu) | |
1da177e4 | 7952 | { |
7def2be1 PZ |
7953 | int cpu = (int)(long)hcpu; |
7954 | ||
1da177e4 | 7955 | switch (action) { |
1da177e4 | 7956 | case CPU_DOWN_PREPARE: |
8bb78442 | 7957 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7958 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7959 | return NOTIFY_OK; |
7960 | ||
1da177e4 | 7961 | case CPU_DOWN_FAILED: |
8bb78442 | 7962 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7963 | case CPU_ONLINE: |
8bb78442 | 7964 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7965 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7966 | return NOTIFY_OK; |
7967 | ||
1da177e4 LT |
7968 | default: |
7969 | return NOTIFY_DONE; | |
7970 | } | |
1da177e4 | 7971 | } |
1da177e4 LT |
7972 | |
7973 | void __init sched_init_smp(void) | |
7974 | { | |
5c1e1767 NP |
7975 | cpumask_t non_isolated_cpus; |
7976 | ||
434d53b0 MT |
7977 | #if defined(CONFIG_NUMA) |
7978 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7979 | GFP_KERNEL); | |
7980 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7981 | #endif | |
95402b38 | 7982 | get_online_cpus(); |
712555ee | 7983 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7984 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7985 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7986 | if (cpus_empty(non_isolated_cpus)) |
7987 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7988 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7989 | put_online_cpus(); |
e761b772 MK |
7990 | |
7991 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7992 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7993 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7994 | #endif |
7995 | ||
7996 | /* RT runtime code needs to handle some hotplug events */ | |
7997 | hotcpu_notifier(update_runtime, 0); | |
7998 | ||
b328ca18 | 7999 | init_hrtick(); |
5c1e1767 NP |
8000 | |
8001 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 8002 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 8003 | BUG(); |
19978ca6 | 8004 | sched_init_granularity(); |
1da177e4 LT |
8005 | } |
8006 | #else | |
8007 | void __init sched_init_smp(void) | |
8008 | { | |
19978ca6 | 8009 | sched_init_granularity(); |
1da177e4 LT |
8010 | } |
8011 | #endif /* CONFIG_SMP */ | |
8012 | ||
8013 | int in_sched_functions(unsigned long addr) | |
8014 | { | |
1da177e4 LT |
8015 | return in_lock_functions(addr) || |
8016 | (addr >= (unsigned long)__sched_text_start | |
8017 | && addr < (unsigned long)__sched_text_end); | |
8018 | } | |
8019 | ||
a9957449 | 8020 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8021 | { |
8022 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8023 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8024 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8025 | cfs_rq->rq = rq; | |
8026 | #endif | |
67e9fb2a | 8027 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8028 | } |
8029 | ||
fa85ae24 PZ |
8030 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8031 | { | |
8032 | struct rt_prio_array *array; | |
8033 | int i; | |
8034 | ||
8035 | array = &rt_rq->active; | |
8036 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8037 | INIT_LIST_HEAD(array->queue + i); | |
8038 | __clear_bit(i, array->bitmap); | |
8039 | } | |
8040 | /* delimiter for bitsearch: */ | |
8041 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8042 | ||
052f1dc7 | 8043 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8044 | rt_rq->highest_prio = MAX_RT_PRIO; |
8045 | #endif | |
fa85ae24 PZ |
8046 | #ifdef CONFIG_SMP |
8047 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8048 | rt_rq->overloaded = 0; |
8049 | #endif | |
8050 | ||
8051 | rt_rq->rt_time = 0; | |
8052 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8053 | rt_rq->rt_runtime = 0; |
8054 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8055 | |
052f1dc7 | 8056 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8057 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8058 | rt_rq->rq = rq; |
8059 | #endif | |
fa85ae24 PZ |
8060 | } |
8061 | ||
6f505b16 | 8062 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8063 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8064 | struct sched_entity *se, int cpu, int add, | |
8065 | struct sched_entity *parent) | |
6f505b16 | 8066 | { |
ec7dc8ac | 8067 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8068 | tg->cfs_rq[cpu] = cfs_rq; |
8069 | init_cfs_rq(cfs_rq, rq); | |
8070 | cfs_rq->tg = tg; | |
8071 | if (add) | |
8072 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8073 | ||
8074 | tg->se[cpu] = se; | |
354d60c2 DG |
8075 | /* se could be NULL for init_task_group */ |
8076 | if (!se) | |
8077 | return; | |
8078 | ||
ec7dc8ac DG |
8079 | if (!parent) |
8080 | se->cfs_rq = &rq->cfs; | |
8081 | else | |
8082 | se->cfs_rq = parent->my_q; | |
8083 | ||
6f505b16 PZ |
8084 | se->my_q = cfs_rq; |
8085 | se->load.weight = tg->shares; | |
e05510d0 | 8086 | se->load.inv_weight = 0; |
ec7dc8ac | 8087 | se->parent = parent; |
6f505b16 | 8088 | } |
052f1dc7 | 8089 | #endif |
6f505b16 | 8090 | |
052f1dc7 | 8091 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8092 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8093 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8094 | struct sched_rt_entity *parent) | |
6f505b16 | 8095 | { |
ec7dc8ac DG |
8096 | struct rq *rq = cpu_rq(cpu); |
8097 | ||
6f505b16 PZ |
8098 | tg->rt_rq[cpu] = rt_rq; |
8099 | init_rt_rq(rt_rq, rq); | |
8100 | rt_rq->tg = tg; | |
8101 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8102 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8103 | if (add) |
8104 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8105 | ||
8106 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8107 | if (!rt_se) |
8108 | return; | |
8109 | ||
ec7dc8ac DG |
8110 | if (!parent) |
8111 | rt_se->rt_rq = &rq->rt; | |
8112 | else | |
8113 | rt_se->rt_rq = parent->my_q; | |
8114 | ||
6f505b16 | 8115 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8116 | rt_se->parent = parent; |
6f505b16 PZ |
8117 | INIT_LIST_HEAD(&rt_se->run_list); |
8118 | } | |
8119 | #endif | |
8120 | ||
1da177e4 LT |
8121 | void __init sched_init(void) |
8122 | { | |
dd41f596 | 8123 | int i, j; |
434d53b0 MT |
8124 | unsigned long alloc_size = 0, ptr; |
8125 | ||
8126 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8127 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8128 | #endif | |
8129 | #ifdef CONFIG_RT_GROUP_SCHED | |
8130 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8131 | #endif |
8132 | #ifdef CONFIG_USER_SCHED | |
8133 | alloc_size *= 2; | |
434d53b0 MT |
8134 | #endif |
8135 | /* | |
8136 | * As sched_init() is called before page_alloc is setup, | |
8137 | * we use alloc_bootmem(). | |
8138 | */ | |
8139 | if (alloc_size) { | |
5a9d3225 | 8140 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8141 | |
8142 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8143 | init_task_group.se = (struct sched_entity **)ptr; | |
8144 | ptr += nr_cpu_ids * sizeof(void **); | |
8145 | ||
8146 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8147 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8148 | |
8149 | #ifdef CONFIG_USER_SCHED | |
8150 | root_task_group.se = (struct sched_entity **)ptr; | |
8151 | ptr += nr_cpu_ids * sizeof(void **); | |
8152 | ||
8153 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8154 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8155 | #endif /* CONFIG_USER_SCHED */ |
8156 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8157 | #ifdef CONFIG_RT_GROUP_SCHED |
8158 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8159 | ptr += nr_cpu_ids * sizeof(void **); | |
8160 | ||
8161 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8162 | ptr += nr_cpu_ids * sizeof(void **); |
8163 | ||
8164 | #ifdef CONFIG_USER_SCHED | |
8165 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8166 | ptr += nr_cpu_ids * sizeof(void **); | |
8167 | ||
8168 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8169 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8170 | #endif /* CONFIG_USER_SCHED */ |
8171 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8172 | } |
dd41f596 | 8173 | |
57d885fe GH |
8174 | #ifdef CONFIG_SMP |
8175 | init_defrootdomain(); | |
8176 | #endif | |
8177 | ||
d0b27fa7 PZ |
8178 | init_rt_bandwidth(&def_rt_bandwidth, |
8179 | global_rt_period(), global_rt_runtime()); | |
8180 | ||
8181 | #ifdef CONFIG_RT_GROUP_SCHED | |
8182 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8183 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8184 | #ifdef CONFIG_USER_SCHED |
8185 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8186 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8187 | #endif /* CONFIG_USER_SCHED */ |
8188 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8189 | |
052f1dc7 | 8190 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8191 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8192 | INIT_LIST_HEAD(&init_task_group.children); |
8193 | ||
8194 | #ifdef CONFIG_USER_SCHED | |
8195 | INIT_LIST_HEAD(&root_task_group.children); | |
8196 | init_task_group.parent = &root_task_group; | |
8197 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8198 | #endif /* CONFIG_USER_SCHED */ |
8199 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8200 | |
0a945022 | 8201 | for_each_possible_cpu(i) { |
70b97a7f | 8202 | struct rq *rq; |
1da177e4 LT |
8203 | |
8204 | rq = cpu_rq(i); | |
8205 | spin_lock_init(&rq->lock); | |
7897986b | 8206 | rq->nr_running = 0; |
dd41f596 | 8207 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8208 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8209 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8210 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8211 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8212 | #ifdef CONFIG_CGROUP_SCHED |
8213 | /* | |
8214 | * How much cpu bandwidth does init_task_group get? | |
8215 | * | |
8216 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8217 | * gets 100% of the cpu resources in the system. This overall | |
8218 | * system cpu resource is divided among the tasks of | |
8219 | * init_task_group and its child task-groups in a fair manner, | |
8220 | * based on each entity's (task or task-group's) weight | |
8221 | * (se->load.weight). | |
8222 | * | |
8223 | * In other words, if init_task_group has 10 tasks of weight | |
8224 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8225 | * then A0's share of the cpu resource is: | |
8226 | * | |
8227 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8228 | * | |
8229 | * We achieve this by letting init_task_group's tasks sit | |
8230 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8231 | */ | |
ec7dc8ac | 8232 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8233 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8234 | root_task_group.shares = NICE_0_LOAD; |
8235 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8236 | /* |
8237 | * In case of task-groups formed thr' the user id of tasks, | |
8238 | * init_task_group represents tasks belonging to root user. | |
8239 | * Hence it forms a sibling of all subsequent groups formed. | |
8240 | * In this case, init_task_group gets only a fraction of overall | |
8241 | * system cpu resource, based on the weight assigned to root | |
8242 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8243 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8244 | * (init_cfs_rq) and having one entity represent this group of | |
8245 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8246 | */ | |
ec7dc8ac | 8247 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8248 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8249 | &per_cpu(init_sched_entity, i), i, 1, |
8250 | root_task_group.se[i]); | |
6f505b16 | 8251 | |
052f1dc7 | 8252 | #endif |
354d60c2 DG |
8253 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8254 | ||
8255 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8256 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8257 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8258 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8259 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8260 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8261 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8262 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8263 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8264 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8265 | root_task_group.rt_se[i]); | |
354d60c2 | 8266 | #endif |
dd41f596 | 8267 | #endif |
1da177e4 | 8268 | |
dd41f596 IM |
8269 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8270 | rq->cpu_load[j] = 0; | |
1da177e4 | 8271 | #ifdef CONFIG_SMP |
41c7ce9a | 8272 | rq->sd = NULL; |
57d885fe | 8273 | rq->rd = NULL; |
1da177e4 | 8274 | rq->active_balance = 0; |
dd41f596 | 8275 | rq->next_balance = jiffies; |
1da177e4 | 8276 | rq->push_cpu = 0; |
0a2966b4 | 8277 | rq->cpu = i; |
1f11eb6a | 8278 | rq->online = 0; |
1da177e4 LT |
8279 | rq->migration_thread = NULL; |
8280 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8281 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8282 | #endif |
8f4d37ec | 8283 | init_rq_hrtick(rq); |
1da177e4 | 8284 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8285 | } |
8286 | ||
2dd73a4f | 8287 | set_load_weight(&init_task); |
b50f60ce | 8288 | |
e107be36 AK |
8289 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8290 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8291 | #endif | |
8292 | ||
c9819f45 | 8293 | #ifdef CONFIG_SMP |
962cf36c | 8294 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8295 | #endif |
8296 | ||
b50f60ce HC |
8297 | #ifdef CONFIG_RT_MUTEXES |
8298 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8299 | #endif | |
8300 | ||
1da177e4 LT |
8301 | /* |
8302 | * The boot idle thread does lazy MMU switching as well: | |
8303 | */ | |
8304 | atomic_inc(&init_mm.mm_count); | |
8305 | enter_lazy_tlb(&init_mm, current); | |
8306 | ||
8307 | /* | |
8308 | * Make us the idle thread. Technically, schedule() should not be | |
8309 | * called from this thread, however somewhere below it might be, | |
8310 | * but because we are the idle thread, we just pick up running again | |
8311 | * when this runqueue becomes "idle". | |
8312 | */ | |
8313 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8314 | /* |
8315 | * During early bootup we pretend to be a normal task: | |
8316 | */ | |
8317 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8318 | |
8319 | scheduler_running = 1; | |
1da177e4 LT |
8320 | } |
8321 | ||
8322 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8323 | void __might_sleep(char *file, int line) | |
8324 | { | |
48f24c4d | 8325 | #ifdef in_atomic |
1da177e4 LT |
8326 | static unsigned long prev_jiffy; /* ratelimiting */ |
8327 | ||
aef745fc IM |
8328 | if ((!in_atomic() && !irqs_disabled()) || |
8329 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8330 | return; | |
8331 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8332 | return; | |
8333 | prev_jiffy = jiffies; | |
8334 | ||
8335 | printk(KERN_ERR | |
8336 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8337 | file, line); | |
8338 | printk(KERN_ERR | |
8339 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8340 | in_atomic(), irqs_disabled(), | |
8341 | current->pid, current->comm); | |
8342 | ||
8343 | debug_show_held_locks(current); | |
8344 | if (irqs_disabled()) | |
8345 | print_irqtrace_events(current); | |
8346 | dump_stack(); | |
1da177e4 LT |
8347 | #endif |
8348 | } | |
8349 | EXPORT_SYMBOL(__might_sleep); | |
8350 | #endif | |
8351 | ||
8352 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8353 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8354 | { | |
8355 | int on_rq; | |
3e51f33f | 8356 | |
3a5e4dc1 AK |
8357 | update_rq_clock(rq); |
8358 | on_rq = p->se.on_rq; | |
8359 | if (on_rq) | |
8360 | deactivate_task(rq, p, 0); | |
8361 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8362 | if (on_rq) { | |
8363 | activate_task(rq, p, 0); | |
8364 | resched_task(rq->curr); | |
8365 | } | |
8366 | } | |
8367 | ||
1da177e4 LT |
8368 | void normalize_rt_tasks(void) |
8369 | { | |
a0f98a1c | 8370 | struct task_struct *g, *p; |
1da177e4 | 8371 | unsigned long flags; |
70b97a7f | 8372 | struct rq *rq; |
1da177e4 | 8373 | |
4cf5d77a | 8374 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8375 | do_each_thread(g, p) { |
178be793 IM |
8376 | /* |
8377 | * Only normalize user tasks: | |
8378 | */ | |
8379 | if (!p->mm) | |
8380 | continue; | |
8381 | ||
6cfb0d5d | 8382 | p->se.exec_start = 0; |
6cfb0d5d | 8383 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8384 | p->se.wait_start = 0; |
dd41f596 | 8385 | p->se.sleep_start = 0; |
dd41f596 | 8386 | p->se.block_start = 0; |
6cfb0d5d | 8387 | #endif |
dd41f596 IM |
8388 | |
8389 | if (!rt_task(p)) { | |
8390 | /* | |
8391 | * Renice negative nice level userspace | |
8392 | * tasks back to 0: | |
8393 | */ | |
8394 | if (TASK_NICE(p) < 0 && p->mm) | |
8395 | set_user_nice(p, 0); | |
1da177e4 | 8396 | continue; |
dd41f596 | 8397 | } |
1da177e4 | 8398 | |
4cf5d77a | 8399 | spin_lock(&p->pi_lock); |
b29739f9 | 8400 | rq = __task_rq_lock(p); |
1da177e4 | 8401 | |
178be793 | 8402 | normalize_task(rq, p); |
3a5e4dc1 | 8403 | |
b29739f9 | 8404 | __task_rq_unlock(rq); |
4cf5d77a | 8405 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8406 | } while_each_thread(g, p); |
8407 | ||
4cf5d77a | 8408 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8409 | } |
8410 | ||
8411 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8412 | |
8413 | #ifdef CONFIG_IA64 | |
8414 | /* | |
8415 | * These functions are only useful for the IA64 MCA handling. | |
8416 | * | |
8417 | * They can only be called when the whole system has been | |
8418 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8419 | * activity can take place. Using them for anything else would | |
8420 | * be a serious bug, and as a result, they aren't even visible | |
8421 | * under any other configuration. | |
8422 | */ | |
8423 | ||
8424 | /** | |
8425 | * curr_task - return the current task for a given cpu. | |
8426 | * @cpu: the processor in question. | |
8427 | * | |
8428 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8429 | */ | |
36c8b586 | 8430 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8431 | { |
8432 | return cpu_curr(cpu); | |
8433 | } | |
8434 | ||
8435 | /** | |
8436 | * set_curr_task - set the current task for a given cpu. | |
8437 | * @cpu: the processor in question. | |
8438 | * @p: the task pointer to set. | |
8439 | * | |
8440 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8441 | * are serviced on a separate stack. It allows the architecture to switch the |
8442 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8443 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8444 | * and caller must save the original value of the current task (see | |
8445 | * curr_task() above) and restore that value before reenabling interrupts and | |
8446 | * re-starting the system. | |
8447 | * | |
8448 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8449 | */ | |
36c8b586 | 8450 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8451 | { |
8452 | cpu_curr(cpu) = p; | |
8453 | } | |
8454 | ||
8455 | #endif | |
29f59db3 | 8456 | |
bccbe08a PZ |
8457 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8458 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8459 | { |
8460 | int i; | |
8461 | ||
8462 | for_each_possible_cpu(i) { | |
8463 | if (tg->cfs_rq) | |
8464 | kfree(tg->cfs_rq[i]); | |
8465 | if (tg->se) | |
8466 | kfree(tg->se[i]); | |
6f505b16 PZ |
8467 | } |
8468 | ||
8469 | kfree(tg->cfs_rq); | |
8470 | kfree(tg->se); | |
6f505b16 PZ |
8471 | } |
8472 | ||
ec7dc8ac DG |
8473 | static |
8474 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8475 | { |
29f59db3 | 8476 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8477 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8478 | struct rq *rq; |
29f59db3 SV |
8479 | int i; |
8480 | ||
434d53b0 | 8481 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8482 | if (!tg->cfs_rq) |
8483 | goto err; | |
434d53b0 | 8484 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8485 | if (!tg->se) |
8486 | goto err; | |
052f1dc7 PZ |
8487 | |
8488 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8489 | |
8490 | for_each_possible_cpu(i) { | |
9b5b7751 | 8491 | rq = cpu_rq(i); |
29f59db3 | 8492 | |
6f505b16 PZ |
8493 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8494 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8495 | if (!cfs_rq) |
8496 | goto err; | |
8497 | ||
6f505b16 PZ |
8498 | se = kmalloc_node(sizeof(struct sched_entity), |
8499 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8500 | if (!se) |
8501 | goto err; | |
8502 | ||
ec7dc8ac DG |
8503 | parent_se = parent ? parent->se[i] : NULL; |
8504 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8505 | } |
8506 | ||
8507 | return 1; | |
8508 | ||
8509 | err: | |
8510 | return 0; | |
8511 | } | |
8512 | ||
8513 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8514 | { | |
8515 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8516 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8517 | } | |
8518 | ||
8519 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8520 | { | |
8521 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8522 | } | |
6d6bc0ad | 8523 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8524 | static inline void free_fair_sched_group(struct task_group *tg) |
8525 | { | |
8526 | } | |
8527 | ||
ec7dc8ac DG |
8528 | static inline |
8529 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8530 | { |
8531 | return 1; | |
8532 | } | |
8533 | ||
8534 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8535 | { | |
8536 | } | |
8537 | ||
8538 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8539 | { | |
8540 | } | |
6d6bc0ad | 8541 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8542 | |
8543 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8544 | static void free_rt_sched_group(struct task_group *tg) |
8545 | { | |
8546 | int i; | |
8547 | ||
d0b27fa7 PZ |
8548 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8549 | ||
bccbe08a PZ |
8550 | for_each_possible_cpu(i) { |
8551 | if (tg->rt_rq) | |
8552 | kfree(tg->rt_rq[i]); | |
8553 | if (tg->rt_se) | |
8554 | kfree(tg->rt_se[i]); | |
8555 | } | |
8556 | ||
8557 | kfree(tg->rt_rq); | |
8558 | kfree(tg->rt_se); | |
8559 | } | |
8560 | ||
ec7dc8ac DG |
8561 | static |
8562 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8563 | { |
8564 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8565 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8566 | struct rq *rq; |
8567 | int i; | |
8568 | ||
434d53b0 | 8569 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8570 | if (!tg->rt_rq) |
8571 | goto err; | |
434d53b0 | 8572 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8573 | if (!tg->rt_se) |
8574 | goto err; | |
8575 | ||
d0b27fa7 PZ |
8576 | init_rt_bandwidth(&tg->rt_bandwidth, |
8577 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8578 | |
8579 | for_each_possible_cpu(i) { | |
8580 | rq = cpu_rq(i); | |
8581 | ||
6f505b16 PZ |
8582 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8583 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8584 | if (!rt_rq) | |
8585 | goto err; | |
29f59db3 | 8586 | |
6f505b16 PZ |
8587 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8588 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8589 | if (!rt_se) | |
8590 | goto err; | |
29f59db3 | 8591 | |
ec7dc8ac DG |
8592 | parent_se = parent ? parent->rt_se[i] : NULL; |
8593 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8594 | } |
8595 | ||
bccbe08a PZ |
8596 | return 1; |
8597 | ||
8598 | err: | |
8599 | return 0; | |
8600 | } | |
8601 | ||
8602 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8603 | { | |
8604 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8605 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8606 | } | |
8607 | ||
8608 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8609 | { | |
8610 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8611 | } | |
6d6bc0ad | 8612 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8613 | static inline void free_rt_sched_group(struct task_group *tg) |
8614 | { | |
8615 | } | |
8616 | ||
ec7dc8ac DG |
8617 | static inline |
8618 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8619 | { |
8620 | return 1; | |
8621 | } | |
8622 | ||
8623 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8624 | { | |
8625 | } | |
8626 | ||
8627 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8628 | { | |
8629 | } | |
6d6bc0ad | 8630 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8631 | |
d0b27fa7 | 8632 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8633 | static void free_sched_group(struct task_group *tg) |
8634 | { | |
8635 | free_fair_sched_group(tg); | |
8636 | free_rt_sched_group(tg); | |
8637 | kfree(tg); | |
8638 | } | |
8639 | ||
8640 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8641 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8642 | { |
8643 | struct task_group *tg; | |
8644 | unsigned long flags; | |
8645 | int i; | |
8646 | ||
8647 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8648 | if (!tg) | |
8649 | return ERR_PTR(-ENOMEM); | |
8650 | ||
ec7dc8ac | 8651 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8652 | goto err; |
8653 | ||
ec7dc8ac | 8654 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8655 | goto err; |
8656 | ||
8ed36996 | 8657 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8658 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8659 | register_fair_sched_group(tg, i); |
8660 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8661 | } |
6f505b16 | 8662 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8663 | |
8664 | WARN_ON(!parent); /* root should already exist */ | |
8665 | ||
8666 | tg->parent = parent; | |
f473aa5e | 8667 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8668 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8669 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8670 | |
9b5b7751 | 8671 | return tg; |
29f59db3 SV |
8672 | |
8673 | err: | |
6f505b16 | 8674 | free_sched_group(tg); |
29f59db3 SV |
8675 | return ERR_PTR(-ENOMEM); |
8676 | } | |
8677 | ||
9b5b7751 | 8678 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8679 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8680 | { |
29f59db3 | 8681 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8682 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8683 | } |
8684 | ||
9b5b7751 | 8685 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8686 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8687 | { |
8ed36996 | 8688 | unsigned long flags; |
9b5b7751 | 8689 | int i; |
29f59db3 | 8690 | |
8ed36996 | 8691 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8692 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8693 | unregister_fair_sched_group(tg, i); |
8694 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8695 | } |
6f505b16 | 8696 | list_del_rcu(&tg->list); |
f473aa5e | 8697 | list_del_rcu(&tg->siblings); |
8ed36996 | 8698 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8699 | |
9b5b7751 | 8700 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8701 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8702 | } |
8703 | ||
9b5b7751 | 8704 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8705 | * The caller of this function should have put the task in its new group |
8706 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8707 | * reflect its new group. | |
9b5b7751 SV |
8708 | */ |
8709 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8710 | { |
8711 | int on_rq, running; | |
8712 | unsigned long flags; | |
8713 | struct rq *rq; | |
8714 | ||
8715 | rq = task_rq_lock(tsk, &flags); | |
8716 | ||
29f59db3 SV |
8717 | update_rq_clock(rq); |
8718 | ||
051a1d1a | 8719 | running = task_current(rq, tsk); |
29f59db3 SV |
8720 | on_rq = tsk->se.on_rq; |
8721 | ||
0e1f3483 | 8722 | if (on_rq) |
29f59db3 | 8723 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8724 | if (unlikely(running)) |
8725 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8726 | |
6f505b16 | 8727 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8728 | |
810b3817 PZ |
8729 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8730 | if (tsk->sched_class->moved_group) | |
8731 | tsk->sched_class->moved_group(tsk); | |
8732 | #endif | |
8733 | ||
0e1f3483 HS |
8734 | if (unlikely(running)) |
8735 | tsk->sched_class->set_curr_task(rq); | |
8736 | if (on_rq) | |
7074badb | 8737 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8738 | |
29f59db3 SV |
8739 | task_rq_unlock(rq, &flags); |
8740 | } | |
6d6bc0ad | 8741 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8742 | |
052f1dc7 | 8743 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8744 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8745 | { |
8746 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8747 | int on_rq; |
8748 | ||
29f59db3 | 8749 | on_rq = se->on_rq; |
62fb1851 | 8750 | if (on_rq) |
29f59db3 SV |
8751 | dequeue_entity(cfs_rq, se, 0); |
8752 | ||
8753 | se->load.weight = shares; | |
e05510d0 | 8754 | se->load.inv_weight = 0; |
29f59db3 | 8755 | |
62fb1851 | 8756 | if (on_rq) |
29f59db3 | 8757 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8758 | } |
62fb1851 | 8759 | |
c09595f6 PZ |
8760 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8761 | { | |
8762 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8763 | struct rq *rq = cfs_rq->rq; | |
8764 | unsigned long flags; | |
8765 | ||
8766 | spin_lock_irqsave(&rq->lock, flags); | |
8767 | __set_se_shares(se, shares); | |
8768 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8769 | } |
8770 | ||
8ed36996 PZ |
8771 | static DEFINE_MUTEX(shares_mutex); |
8772 | ||
4cf86d77 | 8773 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8774 | { |
8775 | int i; | |
8ed36996 | 8776 | unsigned long flags; |
c61935fd | 8777 | |
ec7dc8ac DG |
8778 | /* |
8779 | * We can't change the weight of the root cgroup. | |
8780 | */ | |
8781 | if (!tg->se[0]) | |
8782 | return -EINVAL; | |
8783 | ||
18d95a28 PZ |
8784 | if (shares < MIN_SHARES) |
8785 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8786 | else if (shares > MAX_SHARES) |
8787 | shares = MAX_SHARES; | |
62fb1851 | 8788 | |
8ed36996 | 8789 | mutex_lock(&shares_mutex); |
9b5b7751 | 8790 | if (tg->shares == shares) |
5cb350ba | 8791 | goto done; |
29f59db3 | 8792 | |
8ed36996 | 8793 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8794 | for_each_possible_cpu(i) |
8795 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8796 | list_del_rcu(&tg->siblings); |
8ed36996 | 8797 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8798 | |
8799 | /* wait for any ongoing reference to this group to finish */ | |
8800 | synchronize_sched(); | |
8801 | ||
8802 | /* | |
8803 | * Now we are free to modify the group's share on each cpu | |
8804 | * w/o tripping rebalance_share or load_balance_fair. | |
8805 | */ | |
9b5b7751 | 8806 | tg->shares = shares; |
c09595f6 PZ |
8807 | for_each_possible_cpu(i) { |
8808 | /* | |
8809 | * force a rebalance | |
8810 | */ | |
8811 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8812 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8813 | } |
29f59db3 | 8814 | |
6b2d7700 SV |
8815 | /* |
8816 | * Enable load balance activity on this group, by inserting it back on | |
8817 | * each cpu's rq->leaf_cfs_rq_list. | |
8818 | */ | |
8ed36996 | 8819 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8820 | for_each_possible_cpu(i) |
8821 | register_fair_sched_group(tg, i); | |
f473aa5e | 8822 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8823 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8824 | done: |
8ed36996 | 8825 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8826 | return 0; |
29f59db3 SV |
8827 | } |
8828 | ||
5cb350ba DG |
8829 | unsigned long sched_group_shares(struct task_group *tg) |
8830 | { | |
8831 | return tg->shares; | |
8832 | } | |
052f1dc7 | 8833 | #endif |
5cb350ba | 8834 | |
052f1dc7 | 8835 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8836 | /* |
9f0c1e56 | 8837 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8838 | */ |
9f0c1e56 PZ |
8839 | static DEFINE_MUTEX(rt_constraints_mutex); |
8840 | ||
8841 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8842 | { | |
8843 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8844 | return 1ULL << 20; |
9f0c1e56 | 8845 | |
9a7e0b18 | 8846 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8847 | } |
8848 | ||
9a7e0b18 PZ |
8849 | /* Must be called with tasklist_lock held */ |
8850 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8851 | { |
9a7e0b18 | 8852 | struct task_struct *g, *p; |
b40b2e8e | 8853 | |
9a7e0b18 PZ |
8854 | do_each_thread(g, p) { |
8855 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8856 | return 1; | |
8857 | } while_each_thread(g, p); | |
b40b2e8e | 8858 | |
9a7e0b18 PZ |
8859 | return 0; |
8860 | } | |
b40b2e8e | 8861 | |
9a7e0b18 PZ |
8862 | struct rt_schedulable_data { |
8863 | struct task_group *tg; | |
8864 | u64 rt_period; | |
8865 | u64 rt_runtime; | |
8866 | }; | |
b40b2e8e | 8867 | |
9a7e0b18 PZ |
8868 | static int tg_schedulable(struct task_group *tg, void *data) |
8869 | { | |
8870 | struct rt_schedulable_data *d = data; | |
8871 | struct task_group *child; | |
8872 | unsigned long total, sum = 0; | |
8873 | u64 period, runtime; | |
b40b2e8e | 8874 | |
9a7e0b18 PZ |
8875 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8876 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8877 | |
9a7e0b18 PZ |
8878 | if (tg == d->tg) { |
8879 | period = d->rt_period; | |
8880 | runtime = d->rt_runtime; | |
b40b2e8e | 8881 | } |
b40b2e8e | 8882 | |
4653f803 PZ |
8883 | /* |
8884 | * Cannot have more runtime than the period. | |
8885 | */ | |
8886 | if (runtime > period && runtime != RUNTIME_INF) | |
8887 | return -EINVAL; | |
6f505b16 | 8888 | |
4653f803 PZ |
8889 | /* |
8890 | * Ensure we don't starve existing RT tasks. | |
8891 | */ | |
9a7e0b18 PZ |
8892 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8893 | return -EBUSY; | |
6f505b16 | 8894 | |
9a7e0b18 | 8895 | total = to_ratio(period, runtime); |
6f505b16 | 8896 | |
4653f803 PZ |
8897 | /* |
8898 | * Nobody can have more than the global setting allows. | |
8899 | */ | |
8900 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8901 | return -EINVAL; | |
6f505b16 | 8902 | |
4653f803 PZ |
8903 | /* |
8904 | * The sum of our children's runtime should not exceed our own. | |
8905 | */ | |
9a7e0b18 PZ |
8906 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8907 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8908 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8909 | |
9a7e0b18 PZ |
8910 | if (child == d->tg) { |
8911 | period = d->rt_period; | |
8912 | runtime = d->rt_runtime; | |
8913 | } | |
6f505b16 | 8914 | |
9a7e0b18 | 8915 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8916 | } |
6f505b16 | 8917 | |
9a7e0b18 PZ |
8918 | if (sum > total) |
8919 | return -EINVAL; | |
8920 | ||
8921 | return 0; | |
6f505b16 PZ |
8922 | } |
8923 | ||
9a7e0b18 | 8924 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8925 | { |
9a7e0b18 PZ |
8926 | struct rt_schedulable_data data = { |
8927 | .tg = tg, | |
8928 | .rt_period = period, | |
8929 | .rt_runtime = runtime, | |
8930 | }; | |
8931 | ||
8932 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8933 | } |
8934 | ||
d0b27fa7 PZ |
8935 | static int tg_set_bandwidth(struct task_group *tg, |
8936 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8937 | { |
ac086bc2 | 8938 | int i, err = 0; |
9f0c1e56 | 8939 | |
9f0c1e56 | 8940 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8941 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8942 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8943 | if (err) | |
9f0c1e56 | 8944 | goto unlock; |
ac086bc2 PZ |
8945 | |
8946 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8947 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8948 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8949 | |
8950 | for_each_possible_cpu(i) { | |
8951 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8952 | ||
8953 | spin_lock(&rt_rq->rt_runtime_lock); | |
8954 | rt_rq->rt_runtime = rt_runtime; | |
8955 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8956 | } | |
8957 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8958 | unlock: |
521f1a24 | 8959 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8960 | mutex_unlock(&rt_constraints_mutex); |
8961 | ||
8962 | return err; | |
6f505b16 PZ |
8963 | } |
8964 | ||
d0b27fa7 PZ |
8965 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8966 | { | |
8967 | u64 rt_runtime, rt_period; | |
8968 | ||
8969 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8970 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8971 | if (rt_runtime_us < 0) | |
8972 | rt_runtime = RUNTIME_INF; | |
8973 | ||
8974 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8975 | } | |
8976 | ||
9f0c1e56 PZ |
8977 | long sched_group_rt_runtime(struct task_group *tg) |
8978 | { | |
8979 | u64 rt_runtime_us; | |
8980 | ||
d0b27fa7 | 8981 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8982 | return -1; |
8983 | ||
d0b27fa7 | 8984 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8985 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8986 | return rt_runtime_us; | |
8987 | } | |
d0b27fa7 PZ |
8988 | |
8989 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8990 | { | |
8991 | u64 rt_runtime, rt_period; | |
8992 | ||
8993 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8994 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8995 | ||
619b0488 R |
8996 | if (rt_period == 0) |
8997 | return -EINVAL; | |
8998 | ||
d0b27fa7 PZ |
8999 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9000 | } | |
9001 | ||
9002 | long sched_group_rt_period(struct task_group *tg) | |
9003 | { | |
9004 | u64 rt_period_us; | |
9005 | ||
9006 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9007 | do_div(rt_period_us, NSEC_PER_USEC); | |
9008 | return rt_period_us; | |
9009 | } | |
9010 | ||
9011 | static int sched_rt_global_constraints(void) | |
9012 | { | |
4653f803 | 9013 | u64 runtime, period; |
d0b27fa7 PZ |
9014 | int ret = 0; |
9015 | ||
ec5d4989 HS |
9016 | if (sysctl_sched_rt_period <= 0) |
9017 | return -EINVAL; | |
9018 | ||
4653f803 PZ |
9019 | runtime = global_rt_runtime(); |
9020 | period = global_rt_period(); | |
9021 | ||
9022 | /* | |
9023 | * Sanity check on the sysctl variables. | |
9024 | */ | |
9025 | if (runtime > period && runtime != RUNTIME_INF) | |
9026 | return -EINVAL; | |
10b612f4 | 9027 | |
d0b27fa7 | 9028 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9029 | read_lock(&tasklist_lock); |
4653f803 | 9030 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9031 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9032 | mutex_unlock(&rt_constraints_mutex); |
9033 | ||
9034 | return ret; | |
9035 | } | |
6d6bc0ad | 9036 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9037 | static int sched_rt_global_constraints(void) |
9038 | { | |
ac086bc2 PZ |
9039 | unsigned long flags; |
9040 | int i; | |
9041 | ||
ec5d4989 HS |
9042 | if (sysctl_sched_rt_period <= 0) |
9043 | return -EINVAL; | |
9044 | ||
ac086bc2 PZ |
9045 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9046 | for_each_possible_cpu(i) { | |
9047 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9048 | ||
9049 | spin_lock(&rt_rq->rt_runtime_lock); | |
9050 | rt_rq->rt_runtime = global_rt_runtime(); | |
9051 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9052 | } | |
9053 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9054 | ||
d0b27fa7 PZ |
9055 | return 0; |
9056 | } | |
6d6bc0ad | 9057 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9058 | |
9059 | int sched_rt_handler(struct ctl_table *table, int write, | |
9060 | struct file *filp, void __user *buffer, size_t *lenp, | |
9061 | loff_t *ppos) | |
9062 | { | |
9063 | int ret; | |
9064 | int old_period, old_runtime; | |
9065 | static DEFINE_MUTEX(mutex); | |
9066 | ||
9067 | mutex_lock(&mutex); | |
9068 | old_period = sysctl_sched_rt_period; | |
9069 | old_runtime = sysctl_sched_rt_runtime; | |
9070 | ||
9071 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9072 | ||
9073 | if (!ret && write) { | |
9074 | ret = sched_rt_global_constraints(); | |
9075 | if (ret) { | |
9076 | sysctl_sched_rt_period = old_period; | |
9077 | sysctl_sched_rt_runtime = old_runtime; | |
9078 | } else { | |
9079 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9080 | def_rt_bandwidth.rt_period = | |
9081 | ns_to_ktime(global_rt_period()); | |
9082 | } | |
9083 | } | |
9084 | mutex_unlock(&mutex); | |
9085 | ||
9086 | return ret; | |
9087 | } | |
68318b8e | 9088 | |
052f1dc7 | 9089 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9090 | |
9091 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9092 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9093 | { |
2b01dfe3 PM |
9094 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9095 | struct task_group, css); | |
68318b8e SV |
9096 | } |
9097 | ||
9098 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9099 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9100 | { |
ec7dc8ac | 9101 | struct task_group *tg, *parent; |
68318b8e | 9102 | |
2b01dfe3 | 9103 | if (!cgrp->parent) { |
68318b8e | 9104 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9105 | return &init_task_group.css; |
9106 | } | |
9107 | ||
ec7dc8ac DG |
9108 | parent = cgroup_tg(cgrp->parent); |
9109 | tg = sched_create_group(parent); | |
68318b8e SV |
9110 | if (IS_ERR(tg)) |
9111 | return ERR_PTR(-ENOMEM); | |
9112 | ||
68318b8e SV |
9113 | return &tg->css; |
9114 | } | |
9115 | ||
41a2d6cf IM |
9116 | static void |
9117 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9118 | { |
2b01dfe3 | 9119 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9120 | |
9121 | sched_destroy_group(tg); | |
9122 | } | |
9123 | ||
41a2d6cf IM |
9124 | static int |
9125 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9126 | struct task_struct *tsk) | |
68318b8e | 9127 | { |
b68aa230 PZ |
9128 | #ifdef CONFIG_RT_GROUP_SCHED |
9129 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9130 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9131 | return -EINVAL; |
9132 | #else | |
68318b8e SV |
9133 | /* We don't support RT-tasks being in separate groups */ |
9134 | if (tsk->sched_class != &fair_sched_class) | |
9135 | return -EINVAL; | |
b68aa230 | 9136 | #endif |
68318b8e SV |
9137 | |
9138 | return 0; | |
9139 | } | |
9140 | ||
9141 | static void | |
2b01dfe3 | 9142 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9143 | struct cgroup *old_cont, struct task_struct *tsk) |
9144 | { | |
9145 | sched_move_task(tsk); | |
9146 | } | |
9147 | ||
052f1dc7 | 9148 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9149 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9150 | u64 shareval) |
68318b8e | 9151 | { |
2b01dfe3 | 9152 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9153 | } |
9154 | ||
f4c753b7 | 9155 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9156 | { |
2b01dfe3 | 9157 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9158 | |
9159 | return (u64) tg->shares; | |
9160 | } | |
6d6bc0ad | 9161 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9162 | |
052f1dc7 | 9163 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9164 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9165 | s64 val) |
6f505b16 | 9166 | { |
06ecb27c | 9167 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9168 | } |
9169 | ||
06ecb27c | 9170 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9171 | { |
06ecb27c | 9172 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9173 | } |
d0b27fa7 PZ |
9174 | |
9175 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9176 | u64 rt_period_us) | |
9177 | { | |
9178 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9179 | } | |
9180 | ||
9181 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9182 | { | |
9183 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9184 | } | |
6d6bc0ad | 9185 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9186 | |
fe5c7cc2 | 9187 | static struct cftype cpu_files[] = { |
052f1dc7 | 9188 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9189 | { |
9190 | .name = "shares", | |
f4c753b7 PM |
9191 | .read_u64 = cpu_shares_read_u64, |
9192 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9193 | }, |
052f1dc7 PZ |
9194 | #endif |
9195 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9196 | { |
9f0c1e56 | 9197 | .name = "rt_runtime_us", |
06ecb27c PM |
9198 | .read_s64 = cpu_rt_runtime_read, |
9199 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9200 | }, |
d0b27fa7 PZ |
9201 | { |
9202 | .name = "rt_period_us", | |
f4c753b7 PM |
9203 | .read_u64 = cpu_rt_period_read_uint, |
9204 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9205 | }, |
052f1dc7 | 9206 | #endif |
68318b8e SV |
9207 | }; |
9208 | ||
9209 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9210 | { | |
fe5c7cc2 | 9211 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9212 | } |
9213 | ||
9214 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9215 | .name = "cpu", |
9216 | .create = cpu_cgroup_create, | |
9217 | .destroy = cpu_cgroup_destroy, | |
9218 | .can_attach = cpu_cgroup_can_attach, | |
9219 | .attach = cpu_cgroup_attach, | |
9220 | .populate = cpu_cgroup_populate, | |
9221 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9222 | .early_init = 1, |
9223 | }; | |
9224 | ||
052f1dc7 | 9225 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9226 | |
9227 | #ifdef CONFIG_CGROUP_CPUACCT | |
9228 | ||
9229 | /* | |
9230 | * CPU accounting code for task groups. | |
9231 | * | |
9232 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9233 | * (balbir@in.ibm.com). | |
9234 | */ | |
9235 | ||
9236 | /* track cpu usage of a group of tasks */ | |
9237 | struct cpuacct { | |
9238 | struct cgroup_subsys_state css; | |
9239 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9240 | u64 *cpuusage; | |
9241 | }; | |
9242 | ||
9243 | struct cgroup_subsys cpuacct_subsys; | |
9244 | ||
9245 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9246 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9247 | { |
32cd756a | 9248 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9249 | struct cpuacct, css); |
9250 | } | |
9251 | ||
9252 | /* return cpu accounting group to which this task belongs */ | |
9253 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9254 | { | |
9255 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9256 | struct cpuacct, css); | |
9257 | } | |
9258 | ||
9259 | /* create a new cpu accounting group */ | |
9260 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9261 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9262 | { |
9263 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9264 | ||
9265 | if (!ca) | |
9266 | return ERR_PTR(-ENOMEM); | |
9267 | ||
9268 | ca->cpuusage = alloc_percpu(u64); | |
9269 | if (!ca->cpuusage) { | |
9270 | kfree(ca); | |
9271 | return ERR_PTR(-ENOMEM); | |
9272 | } | |
9273 | ||
9274 | return &ca->css; | |
9275 | } | |
9276 | ||
9277 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9278 | static void |
32cd756a | 9279 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9280 | { |
32cd756a | 9281 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9282 | |
9283 | free_percpu(ca->cpuusage); | |
9284 | kfree(ca); | |
9285 | } | |
9286 | ||
9287 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9288 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9289 | { |
32cd756a | 9290 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9291 | u64 totalcpuusage = 0; |
9292 | int i; | |
9293 | ||
9294 | for_each_possible_cpu(i) { | |
9295 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9296 | ||
9297 | /* | |
9298 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9299 | * platforms. | |
9300 | */ | |
9301 | spin_lock_irq(&cpu_rq(i)->lock); | |
9302 | totalcpuusage += *cpuusage; | |
9303 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9304 | } | |
9305 | ||
9306 | return totalcpuusage; | |
9307 | } | |
9308 | ||
0297b803 DG |
9309 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9310 | u64 reset) | |
9311 | { | |
9312 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9313 | int err = 0; | |
9314 | int i; | |
9315 | ||
9316 | if (reset) { | |
9317 | err = -EINVAL; | |
9318 | goto out; | |
9319 | } | |
9320 | ||
9321 | for_each_possible_cpu(i) { | |
9322 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9323 | ||
9324 | spin_lock_irq(&cpu_rq(i)->lock); | |
9325 | *cpuusage = 0; | |
9326 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9327 | } | |
9328 | out: | |
9329 | return err; | |
9330 | } | |
9331 | ||
d842de87 SV |
9332 | static struct cftype files[] = { |
9333 | { | |
9334 | .name = "usage", | |
f4c753b7 PM |
9335 | .read_u64 = cpuusage_read, |
9336 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9337 | }, |
9338 | }; | |
9339 | ||
32cd756a | 9340 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9341 | { |
32cd756a | 9342 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9343 | } |
9344 | ||
9345 | /* | |
9346 | * charge this task's execution time to its accounting group. | |
9347 | * | |
9348 | * called with rq->lock held. | |
9349 | */ | |
9350 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9351 | { | |
9352 | struct cpuacct *ca; | |
9353 | ||
9354 | if (!cpuacct_subsys.active) | |
9355 | return; | |
9356 | ||
9357 | ca = task_ca(tsk); | |
9358 | if (ca) { | |
9359 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9360 | ||
9361 | *cpuusage += cputime; | |
9362 | } | |
9363 | } | |
9364 | ||
9365 | struct cgroup_subsys cpuacct_subsys = { | |
9366 | .name = "cpuacct", | |
9367 | .create = cpuacct_create, | |
9368 | .destroy = cpuacct_destroy, | |
9369 | .populate = cpuacct_populate, | |
9370 | .subsys_id = cpuacct_subsys_id, | |
9371 | }; | |
9372 | #endif /* CONFIG_CGROUP_CPUACCT */ |