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 | ||
7e066fb8 MD |
121 | DEFINE_TRACE(sched_wait_task); |
122 | DEFINE_TRACE(sched_wakeup); | |
123 | DEFINE_TRACE(sched_wakeup_new); | |
124 | DEFINE_TRACE(sched_switch); | |
125 | DEFINE_TRACE(sched_migrate_task); | |
126 | ||
5517d86b | 127 | #ifdef CONFIG_SMP |
fd2ab30b SN |
128 | |
129 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
130 | ||
5517d86b ED |
131 | /* |
132 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
133 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
134 | */ | |
135 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
136 | { | |
137 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Each time a sched group cpu_power is changed, | |
142 | * we must compute its reciprocal value | |
143 | */ | |
144 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
145 | { | |
146 | sg->__cpu_power += val; | |
147 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
148 | } | |
149 | #endif | |
150 | ||
e05606d3 IM |
151 | static inline int rt_policy(int policy) |
152 | { | |
3f33a7ce | 153 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
154 | return 1; |
155 | return 0; | |
156 | } | |
157 | ||
158 | static inline int task_has_rt_policy(struct task_struct *p) | |
159 | { | |
160 | return rt_policy(p->policy); | |
161 | } | |
162 | ||
1da177e4 | 163 | /* |
6aa645ea | 164 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 165 | */ |
6aa645ea IM |
166 | struct rt_prio_array { |
167 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
168 | struct list_head queue[MAX_RT_PRIO]; | |
169 | }; | |
170 | ||
d0b27fa7 | 171 | struct rt_bandwidth { |
ea736ed5 IM |
172 | /* nests inside the rq lock: */ |
173 | spinlock_t rt_runtime_lock; | |
174 | ktime_t rt_period; | |
175 | u64 rt_runtime; | |
176 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
177 | }; |
178 | ||
179 | static struct rt_bandwidth def_rt_bandwidth; | |
180 | ||
181 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
182 | ||
183 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
184 | { | |
185 | struct rt_bandwidth *rt_b = | |
186 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
187 | ktime_t now; | |
188 | int overrun; | |
189 | int idle = 0; | |
190 | ||
191 | for (;;) { | |
192 | now = hrtimer_cb_get_time(timer); | |
193 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
194 | ||
195 | if (!overrun) | |
196 | break; | |
197 | ||
198 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
199 | } | |
200 | ||
201 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
202 | } | |
203 | ||
204 | static | |
205 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
206 | { | |
207 | rt_b->rt_period = ns_to_ktime(period); | |
208 | rt_b->rt_runtime = runtime; | |
209 | ||
ac086bc2 PZ |
210 | spin_lock_init(&rt_b->rt_runtime_lock); |
211 | ||
d0b27fa7 PZ |
212 | hrtimer_init(&rt_b->rt_period_timer, |
213 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
214 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
215 | } |
216 | ||
c8bfff6d KH |
217 | static inline int rt_bandwidth_enabled(void) |
218 | { | |
219 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
220 | } |
221 | ||
222 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
223 | { | |
224 | ktime_t now; | |
225 | ||
cac64d00 | 226 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
227 | return; |
228 | ||
229 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
230 | return; | |
231 | ||
232 | spin_lock(&rt_b->rt_runtime_lock); | |
233 | for (;;) { | |
234 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
235 | break; | |
236 | ||
237 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
238 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
239 | hrtimer_start_expires(&rt_b->rt_period_timer, |
240 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
241 | } |
242 | spin_unlock(&rt_b->rt_runtime_lock); | |
243 | } | |
244 | ||
245 | #ifdef CONFIG_RT_GROUP_SCHED | |
246 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
247 | { | |
248 | hrtimer_cancel(&rt_b->rt_period_timer); | |
249 | } | |
250 | #endif | |
251 | ||
712555ee HC |
252 | /* |
253 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
254 | * detach_destroy_domains and partition_sched_domains. | |
255 | */ | |
256 | static DEFINE_MUTEX(sched_domains_mutex); | |
257 | ||
052f1dc7 | 258 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 259 | |
68318b8e SV |
260 | #include <linux/cgroup.h> |
261 | ||
29f59db3 SV |
262 | struct cfs_rq; |
263 | ||
6f505b16 PZ |
264 | static LIST_HEAD(task_groups); |
265 | ||
29f59db3 | 266 | /* task group related information */ |
4cf86d77 | 267 | struct task_group { |
052f1dc7 | 268 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
269 | struct cgroup_subsys_state css; |
270 | #endif | |
052f1dc7 | 271 | |
6c415b92 AB |
272 | #ifdef CONFIG_USER_SCHED |
273 | uid_t uid; | |
274 | #endif | |
275 | ||
052f1dc7 | 276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
277 | /* schedulable entities of this group on each cpu */ |
278 | struct sched_entity **se; | |
279 | /* runqueue "owned" by this group on each cpu */ | |
280 | struct cfs_rq **cfs_rq; | |
281 | unsigned long shares; | |
052f1dc7 PZ |
282 | #endif |
283 | ||
284 | #ifdef CONFIG_RT_GROUP_SCHED | |
285 | struct sched_rt_entity **rt_se; | |
286 | struct rt_rq **rt_rq; | |
287 | ||
d0b27fa7 | 288 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 289 | #endif |
6b2d7700 | 290 | |
ae8393e5 | 291 | struct rcu_head rcu; |
6f505b16 | 292 | struct list_head list; |
f473aa5e PZ |
293 | |
294 | struct task_group *parent; | |
295 | struct list_head siblings; | |
296 | struct list_head children; | |
29f59db3 SV |
297 | }; |
298 | ||
354d60c2 | 299 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 300 | |
6c415b92 AB |
301 | /* Helper function to pass uid information to create_sched_user() */ |
302 | void set_tg_uid(struct user_struct *user) | |
303 | { | |
304 | user->tg->uid = user->uid; | |
305 | } | |
306 | ||
eff766a6 PZ |
307 | /* |
308 | * Root task group. | |
309 | * Every UID task group (including init_task_group aka UID-0) will | |
310 | * be a child to this group. | |
311 | */ | |
312 | struct task_group root_task_group; | |
313 | ||
052f1dc7 | 314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
315 | /* Default task group's sched entity on each cpu */ |
316 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
317 | /* Default task group's cfs_rq on each cpu */ | |
318 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 319 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
320 | |
321 | #ifdef CONFIG_RT_GROUP_SCHED | |
322 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
323 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 325 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 326 | #define root_task_group init_task_group |
9a7e0b18 | 327 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 328 | |
8ed36996 | 329 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
330 | * a task group's cpu shares. |
331 | */ | |
8ed36996 | 332 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 333 | |
57310a98 PZ |
334 | #ifdef CONFIG_SMP |
335 | static int root_task_group_empty(void) | |
336 | { | |
337 | return list_empty(&root_task_group.children); | |
338 | } | |
339 | #endif | |
340 | ||
052f1dc7 | 341 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
342 | #ifdef CONFIG_USER_SCHED |
343 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 344 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 345 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 346 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 347 | |
cb4ad1ff | 348 | /* |
2e084786 LJ |
349 | * A weight of 0 or 1 can cause arithmetics problems. |
350 | * A weight of a cfs_rq is the sum of weights of which entities | |
351 | * are queued on this cfs_rq, so a weight of a entity should not be | |
352 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
353 | * (The default weight is 1024 - so there's no practical |
354 | * limitation from this.) | |
355 | */ | |
18d95a28 | 356 | #define MIN_SHARES 2 |
2e084786 | 357 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 358 | |
052f1dc7 PZ |
359 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
360 | #endif | |
361 | ||
29f59db3 | 362 | /* Default task group. |
3a252015 | 363 | * Every task in system belong to this group at bootup. |
29f59db3 | 364 | */ |
434d53b0 | 365 | struct task_group init_task_group; |
29f59db3 SV |
366 | |
367 | /* return group to which a task belongs */ | |
4cf86d77 | 368 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 369 | { |
4cf86d77 | 370 | struct task_group *tg; |
9b5b7751 | 371 | |
052f1dc7 | 372 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
373 | rcu_read_lock(); |
374 | tg = __task_cred(p)->user->tg; | |
375 | rcu_read_unlock(); | |
052f1dc7 | 376 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
377 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
378 | struct task_group, css); | |
24e377a8 | 379 | #else |
41a2d6cf | 380 | tg = &init_task_group; |
24e377a8 | 381 | #endif |
9b5b7751 | 382 | return tg; |
29f59db3 SV |
383 | } |
384 | ||
385 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 386 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 387 | { |
052f1dc7 | 388 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
389 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
390 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 391 | #endif |
6f505b16 | 392 | |
052f1dc7 | 393 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
394 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
395 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 396 | #endif |
29f59db3 SV |
397 | } |
398 | ||
399 | #else | |
400 | ||
57310a98 PZ |
401 | #ifdef CONFIG_SMP |
402 | static int root_task_group_empty(void) | |
403 | { | |
404 | return 1; | |
405 | } | |
406 | #endif | |
407 | ||
6f505b16 | 408 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
409 | static inline struct task_group *task_group(struct task_struct *p) |
410 | { | |
411 | return NULL; | |
412 | } | |
29f59db3 | 413 | |
052f1dc7 | 414 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 415 | |
6aa645ea IM |
416 | /* CFS-related fields in a runqueue */ |
417 | struct cfs_rq { | |
418 | struct load_weight load; | |
419 | unsigned long nr_running; | |
420 | ||
6aa645ea | 421 | u64 exec_clock; |
e9acbff6 | 422 | u64 min_vruntime; |
6aa645ea IM |
423 | |
424 | struct rb_root tasks_timeline; | |
425 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
426 | |
427 | struct list_head tasks; | |
428 | struct list_head *balance_iterator; | |
429 | ||
430 | /* | |
431 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
432 | * It is set to NULL otherwise (i.e when none are currently running). |
433 | */ | |
4793241b | 434 | struct sched_entity *curr, *next, *last; |
ddc97297 | 435 | |
5ac5c4d6 | 436 | unsigned int nr_spread_over; |
ddc97297 | 437 | |
62160e3f | 438 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
439 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
440 | ||
41a2d6cf IM |
441 | /* |
442 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
443 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
444 | * (like users, containers etc.) | |
445 | * | |
446 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
447 | * list is used during load balance. | |
448 | */ | |
41a2d6cf IM |
449 | struct list_head leaf_cfs_rq_list; |
450 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
451 | |
452 | #ifdef CONFIG_SMP | |
c09595f6 | 453 | /* |
c8cba857 | 454 | * the part of load.weight contributed by tasks |
c09595f6 | 455 | */ |
c8cba857 | 456 | unsigned long task_weight; |
c09595f6 | 457 | |
c8cba857 PZ |
458 | /* |
459 | * h_load = weight * f(tg) | |
460 | * | |
461 | * Where f(tg) is the recursive weight fraction assigned to | |
462 | * this group. | |
463 | */ | |
464 | unsigned long h_load; | |
c09595f6 | 465 | |
c8cba857 PZ |
466 | /* |
467 | * this cpu's part of tg->shares | |
468 | */ | |
469 | unsigned long shares; | |
f1d239f7 PZ |
470 | |
471 | /* | |
472 | * load.weight at the time we set shares | |
473 | */ | |
474 | unsigned long rq_weight; | |
c09595f6 | 475 | #endif |
6aa645ea IM |
476 | #endif |
477 | }; | |
1da177e4 | 478 | |
6aa645ea IM |
479 | /* Real-Time classes' related field in a runqueue: */ |
480 | struct rt_rq { | |
481 | struct rt_prio_array active; | |
63489e45 | 482 | unsigned long rt_nr_running; |
052f1dc7 | 483 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
484 | struct { |
485 | int curr; /* highest queued rt task prio */ | |
398a153b | 486 | #ifdef CONFIG_SMP |
e864c499 | 487 | int next; /* next highest */ |
398a153b | 488 | #endif |
e864c499 | 489 | } highest_prio; |
6f505b16 | 490 | #endif |
fa85ae24 | 491 | #ifdef CONFIG_SMP |
73fe6aae | 492 | unsigned long rt_nr_migratory; |
a22d7fc1 | 493 | int overloaded; |
917b627d | 494 | struct plist_head pushable_tasks; |
fa85ae24 | 495 | #endif |
6f505b16 | 496 | int rt_throttled; |
fa85ae24 | 497 | u64 rt_time; |
ac086bc2 | 498 | u64 rt_runtime; |
ea736ed5 | 499 | /* Nests inside the rq lock: */ |
ac086bc2 | 500 | spinlock_t rt_runtime_lock; |
6f505b16 | 501 | |
052f1dc7 | 502 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
503 | unsigned long rt_nr_boosted; |
504 | ||
6f505b16 PZ |
505 | struct rq *rq; |
506 | struct list_head leaf_rt_rq_list; | |
507 | struct task_group *tg; | |
508 | struct sched_rt_entity *rt_se; | |
509 | #endif | |
6aa645ea IM |
510 | }; |
511 | ||
57d885fe GH |
512 | #ifdef CONFIG_SMP |
513 | ||
514 | /* | |
515 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
516 | * variables. Each exclusive cpuset essentially defines an island domain by |
517 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
518 | * exclusive cpuset is created, we also create and attach a new root-domain |
519 | * object. | |
520 | * | |
57d885fe GH |
521 | */ |
522 | struct root_domain { | |
523 | atomic_t refcount; | |
c6c4927b RR |
524 | cpumask_var_t span; |
525 | cpumask_var_t online; | |
637f5085 | 526 | |
0eab9146 | 527 | /* |
637f5085 GH |
528 | * The "RT overload" flag: it gets set if a CPU has more than |
529 | * one runnable RT task. | |
530 | */ | |
c6c4927b | 531 | cpumask_var_t rto_mask; |
0eab9146 | 532 | atomic_t rto_count; |
6e0534f2 GH |
533 | #ifdef CONFIG_SMP |
534 | struct cpupri cpupri; | |
535 | #endif | |
7a09b1a2 VS |
536 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
537 | /* | |
538 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
539 | * used when most cpus are idle in the system indicating overall very | |
540 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
541 | */ | |
542 | unsigned int sched_mc_preferred_wakeup_cpu; | |
543 | #endif | |
57d885fe GH |
544 | }; |
545 | ||
dc938520 GH |
546 | /* |
547 | * By default the system creates a single root-domain with all cpus as | |
548 | * members (mimicking the global state we have today). | |
549 | */ | |
57d885fe GH |
550 | static struct root_domain def_root_domain; |
551 | ||
552 | #endif | |
553 | ||
1da177e4 LT |
554 | /* |
555 | * This is the main, per-CPU runqueue data structure. | |
556 | * | |
557 | * Locking rule: those places that want to lock multiple runqueues | |
558 | * (such as the load balancing or the thread migration code), lock | |
559 | * acquire operations must be ordered by ascending &runqueue. | |
560 | */ | |
70b97a7f | 561 | struct rq { |
d8016491 IM |
562 | /* runqueue lock: */ |
563 | spinlock_t lock; | |
1da177e4 LT |
564 | |
565 | /* | |
566 | * nr_running and cpu_load should be in the same cacheline because | |
567 | * remote CPUs use both these fields when doing load calculation. | |
568 | */ | |
569 | unsigned long nr_running; | |
6aa645ea IM |
570 | #define CPU_LOAD_IDX_MAX 5 |
571 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 572 | #ifdef CONFIG_NO_HZ |
15934a37 | 573 | unsigned long last_tick_seen; |
46cb4b7c SS |
574 | unsigned char in_nohz_recently; |
575 | #endif | |
d8016491 IM |
576 | /* capture load from *all* tasks on this cpu: */ |
577 | struct load_weight load; | |
6aa645ea IM |
578 | unsigned long nr_load_updates; |
579 | u64 nr_switches; | |
580 | ||
581 | struct cfs_rq cfs; | |
6f505b16 | 582 | struct rt_rq rt; |
6f505b16 | 583 | |
6aa645ea | 584 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
585 | /* list of leaf cfs_rq on this cpu: */ |
586 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
587 | #endif |
588 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 589 | struct list_head leaf_rt_rq_list; |
1da177e4 | 590 | #endif |
1da177e4 LT |
591 | |
592 | /* | |
593 | * This is part of a global counter where only the total sum | |
594 | * over all CPUs matters. A task can increase this counter on | |
595 | * one CPU and if it got migrated afterwards it may decrease | |
596 | * it on another CPU. Always updated under the runqueue lock: | |
597 | */ | |
598 | unsigned long nr_uninterruptible; | |
599 | ||
36c8b586 | 600 | struct task_struct *curr, *idle; |
c9819f45 | 601 | unsigned long next_balance; |
1da177e4 | 602 | struct mm_struct *prev_mm; |
6aa645ea | 603 | |
3e51f33f | 604 | u64 clock; |
6aa645ea | 605 | |
1da177e4 LT |
606 | atomic_t nr_iowait; |
607 | ||
608 | #ifdef CONFIG_SMP | |
0eab9146 | 609 | struct root_domain *rd; |
1da177e4 LT |
610 | struct sched_domain *sd; |
611 | ||
a0a522ce | 612 | unsigned char idle_at_tick; |
1da177e4 LT |
613 | /* For active balancing */ |
614 | int active_balance; | |
615 | int push_cpu; | |
d8016491 IM |
616 | /* cpu of this runqueue: */ |
617 | int cpu; | |
1f11eb6a | 618 | int online; |
1da177e4 | 619 | |
a8a51d5e | 620 | unsigned long avg_load_per_task; |
1da177e4 | 621 | |
36c8b586 | 622 | struct task_struct *migration_thread; |
1da177e4 LT |
623 | struct list_head migration_queue; |
624 | #endif | |
625 | ||
8f4d37ec | 626 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
627 | #ifdef CONFIG_SMP |
628 | int hrtick_csd_pending; | |
629 | struct call_single_data hrtick_csd; | |
630 | #endif | |
8f4d37ec PZ |
631 | struct hrtimer hrtick_timer; |
632 | #endif | |
633 | ||
1da177e4 LT |
634 | #ifdef CONFIG_SCHEDSTATS |
635 | /* latency stats */ | |
636 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
637 | unsigned long long rq_cpu_time; |
638 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
639 | |
640 | /* sys_sched_yield() stats */ | |
480b9434 | 641 | unsigned int yld_count; |
1da177e4 LT |
642 | |
643 | /* schedule() stats */ | |
480b9434 KC |
644 | unsigned int sched_switch; |
645 | unsigned int sched_count; | |
646 | unsigned int sched_goidle; | |
1da177e4 LT |
647 | |
648 | /* try_to_wake_up() stats */ | |
480b9434 KC |
649 | unsigned int ttwu_count; |
650 | unsigned int ttwu_local; | |
b8efb561 IM |
651 | |
652 | /* BKL stats */ | |
480b9434 | 653 | unsigned int bkl_count; |
1da177e4 LT |
654 | #endif |
655 | }; | |
656 | ||
f34e3b61 | 657 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 658 | |
15afe09b | 659 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 660 | { |
15afe09b | 661 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
662 | } |
663 | ||
0a2966b4 CL |
664 | static inline int cpu_of(struct rq *rq) |
665 | { | |
666 | #ifdef CONFIG_SMP | |
667 | return rq->cpu; | |
668 | #else | |
669 | return 0; | |
670 | #endif | |
671 | } | |
672 | ||
674311d5 NP |
673 | /* |
674 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 675 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
676 | * |
677 | * The domain tree of any CPU may only be accessed from within | |
678 | * preempt-disabled sections. | |
679 | */ | |
48f24c4d IM |
680 | #define for_each_domain(cpu, __sd) \ |
681 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
682 | |
683 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
684 | #define this_rq() (&__get_cpu_var(runqueues)) | |
685 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
686 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
687 | ||
3e51f33f PZ |
688 | static inline void update_rq_clock(struct rq *rq) |
689 | { | |
690 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
691 | } | |
692 | ||
bf5c91ba IM |
693 | /* |
694 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
695 | */ | |
696 | #ifdef CONFIG_SCHED_DEBUG | |
697 | # define const_debug __read_mostly | |
698 | #else | |
699 | # define const_debug static const | |
700 | #endif | |
701 | ||
017730c1 IM |
702 | /** |
703 | * runqueue_is_locked | |
704 | * | |
705 | * Returns true if the current cpu runqueue is locked. | |
706 | * This interface allows printk to be called with the runqueue lock | |
707 | * held and know whether or not it is OK to wake up the klogd. | |
708 | */ | |
709 | int runqueue_is_locked(void) | |
710 | { | |
711 | int cpu = get_cpu(); | |
712 | struct rq *rq = cpu_rq(cpu); | |
713 | int ret; | |
714 | ||
715 | ret = spin_is_locked(&rq->lock); | |
716 | put_cpu(); | |
717 | return ret; | |
718 | } | |
719 | ||
bf5c91ba IM |
720 | /* |
721 | * Debugging: various feature bits | |
722 | */ | |
f00b45c1 PZ |
723 | |
724 | #define SCHED_FEAT(name, enabled) \ | |
725 | __SCHED_FEAT_##name , | |
726 | ||
bf5c91ba | 727 | enum { |
f00b45c1 | 728 | #include "sched_features.h" |
bf5c91ba IM |
729 | }; |
730 | ||
f00b45c1 PZ |
731 | #undef SCHED_FEAT |
732 | ||
733 | #define SCHED_FEAT(name, enabled) \ | |
734 | (1UL << __SCHED_FEAT_##name) * enabled | | |
735 | ||
bf5c91ba | 736 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
737 | #include "sched_features.h" |
738 | 0; | |
739 | ||
740 | #undef SCHED_FEAT | |
741 | ||
742 | #ifdef CONFIG_SCHED_DEBUG | |
743 | #define SCHED_FEAT(name, enabled) \ | |
744 | #name , | |
745 | ||
983ed7a6 | 746 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
747 | #include "sched_features.h" |
748 | NULL | |
749 | }; | |
750 | ||
751 | #undef SCHED_FEAT | |
752 | ||
34f3a814 | 753 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 754 | { |
f00b45c1 PZ |
755 | int i; |
756 | ||
757 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
758 | if (!(sysctl_sched_features & (1UL << i))) |
759 | seq_puts(m, "NO_"); | |
760 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 761 | } |
34f3a814 | 762 | seq_puts(m, "\n"); |
f00b45c1 | 763 | |
34f3a814 | 764 | return 0; |
f00b45c1 PZ |
765 | } |
766 | ||
767 | static ssize_t | |
768 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
769 | size_t cnt, loff_t *ppos) | |
770 | { | |
771 | char buf[64]; | |
772 | char *cmp = buf; | |
773 | int neg = 0; | |
774 | int i; | |
775 | ||
776 | if (cnt > 63) | |
777 | cnt = 63; | |
778 | ||
779 | if (copy_from_user(&buf, ubuf, cnt)) | |
780 | return -EFAULT; | |
781 | ||
782 | buf[cnt] = 0; | |
783 | ||
c24b7c52 | 784 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
785 | neg = 1; |
786 | cmp += 3; | |
787 | } | |
788 | ||
789 | for (i = 0; sched_feat_names[i]; i++) { | |
790 | int len = strlen(sched_feat_names[i]); | |
791 | ||
792 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
793 | if (neg) | |
794 | sysctl_sched_features &= ~(1UL << i); | |
795 | else | |
796 | sysctl_sched_features |= (1UL << i); | |
797 | break; | |
798 | } | |
799 | } | |
800 | ||
801 | if (!sched_feat_names[i]) | |
802 | return -EINVAL; | |
803 | ||
804 | filp->f_pos += cnt; | |
805 | ||
806 | return cnt; | |
807 | } | |
808 | ||
34f3a814 LZ |
809 | static int sched_feat_open(struct inode *inode, struct file *filp) |
810 | { | |
811 | return single_open(filp, sched_feat_show, NULL); | |
812 | } | |
813 | ||
f00b45c1 | 814 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
815 | .open = sched_feat_open, |
816 | .write = sched_feat_write, | |
817 | .read = seq_read, | |
818 | .llseek = seq_lseek, | |
819 | .release = single_release, | |
f00b45c1 PZ |
820 | }; |
821 | ||
822 | static __init int sched_init_debug(void) | |
823 | { | |
f00b45c1 PZ |
824 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
825 | &sched_feat_fops); | |
826 | ||
827 | return 0; | |
828 | } | |
829 | late_initcall(sched_init_debug); | |
830 | ||
831 | #endif | |
832 | ||
833 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 834 | |
b82d9fdd PZ |
835 | /* |
836 | * Number of tasks to iterate in a single balance run. | |
837 | * Limited because this is done with IRQs disabled. | |
838 | */ | |
839 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
840 | ||
2398f2c6 PZ |
841 | /* |
842 | * ratelimit for updating the group shares. | |
55cd5340 | 843 | * default: 0.25ms |
2398f2c6 | 844 | */ |
55cd5340 | 845 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 846 | |
ffda12a1 PZ |
847 | /* |
848 | * Inject some fuzzyness into changing the per-cpu group shares | |
849 | * this avoids remote rq-locks at the expense of fairness. | |
850 | * default: 4 | |
851 | */ | |
852 | unsigned int sysctl_sched_shares_thresh = 4; | |
853 | ||
fa85ae24 | 854 | /* |
9f0c1e56 | 855 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
856 | * default: 1s |
857 | */ | |
9f0c1e56 | 858 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 859 | |
6892b75e IM |
860 | static __read_mostly int scheduler_running; |
861 | ||
9f0c1e56 PZ |
862 | /* |
863 | * part of the period that we allow rt tasks to run in us. | |
864 | * default: 0.95s | |
865 | */ | |
866 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 867 | |
d0b27fa7 PZ |
868 | static inline u64 global_rt_period(void) |
869 | { | |
870 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
871 | } | |
872 | ||
873 | static inline u64 global_rt_runtime(void) | |
874 | { | |
e26873bb | 875 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
876 | return RUNTIME_INF; |
877 | ||
878 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
879 | } | |
fa85ae24 | 880 | |
1da177e4 | 881 | #ifndef prepare_arch_switch |
4866cde0 NP |
882 | # define prepare_arch_switch(next) do { } while (0) |
883 | #endif | |
884 | #ifndef finish_arch_switch | |
885 | # define finish_arch_switch(prev) do { } while (0) | |
886 | #endif | |
887 | ||
051a1d1a DA |
888 | static inline int task_current(struct rq *rq, struct task_struct *p) |
889 | { | |
890 | return rq->curr == p; | |
891 | } | |
892 | ||
4866cde0 | 893 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 894 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 895 | { |
051a1d1a | 896 | return task_current(rq, p); |
4866cde0 NP |
897 | } |
898 | ||
70b97a7f | 899 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
900 | { |
901 | } | |
902 | ||
70b97a7f | 903 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 904 | { |
da04c035 IM |
905 | #ifdef CONFIG_DEBUG_SPINLOCK |
906 | /* this is a valid case when another task releases the spinlock */ | |
907 | rq->lock.owner = current; | |
908 | #endif | |
8a25d5de IM |
909 | /* |
910 | * If we are tracking spinlock dependencies then we have to | |
911 | * fix up the runqueue lock - which gets 'carried over' from | |
912 | * prev into current: | |
913 | */ | |
914 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
915 | ||
4866cde0 NP |
916 | spin_unlock_irq(&rq->lock); |
917 | } | |
918 | ||
919 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 920 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
921 | { |
922 | #ifdef CONFIG_SMP | |
923 | return p->oncpu; | |
924 | #else | |
051a1d1a | 925 | return task_current(rq, p); |
4866cde0 NP |
926 | #endif |
927 | } | |
928 | ||
70b97a7f | 929 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
930 | { |
931 | #ifdef CONFIG_SMP | |
932 | /* | |
933 | * We can optimise this out completely for !SMP, because the | |
934 | * SMP rebalancing from interrupt is the only thing that cares | |
935 | * here. | |
936 | */ | |
937 | next->oncpu = 1; | |
938 | #endif | |
939 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
940 | spin_unlock_irq(&rq->lock); | |
941 | #else | |
942 | spin_unlock(&rq->lock); | |
943 | #endif | |
944 | } | |
945 | ||
70b97a7f | 946 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
947 | { |
948 | #ifdef CONFIG_SMP | |
949 | /* | |
950 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
951 | * We must ensure this doesn't happen until the switch is completely | |
952 | * finished. | |
953 | */ | |
954 | smp_wmb(); | |
955 | prev->oncpu = 0; | |
956 | #endif | |
957 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
958 | local_irq_enable(); | |
1da177e4 | 959 | #endif |
4866cde0 NP |
960 | } |
961 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 962 | |
b29739f9 IM |
963 | /* |
964 | * __task_rq_lock - lock the runqueue a given task resides on. | |
965 | * Must be called interrupts disabled. | |
966 | */ | |
70b97a7f | 967 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
968 | __acquires(rq->lock) |
969 | { | |
3a5c359a AK |
970 | for (;;) { |
971 | struct rq *rq = task_rq(p); | |
972 | spin_lock(&rq->lock); | |
973 | if (likely(rq == task_rq(p))) | |
974 | return rq; | |
b29739f9 | 975 | spin_unlock(&rq->lock); |
b29739f9 | 976 | } |
b29739f9 IM |
977 | } |
978 | ||
1da177e4 LT |
979 | /* |
980 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 981 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
982 | * explicitly disabling preemption. |
983 | */ | |
70b97a7f | 984 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
985 | __acquires(rq->lock) |
986 | { | |
70b97a7f | 987 | struct rq *rq; |
1da177e4 | 988 | |
3a5c359a AK |
989 | for (;;) { |
990 | local_irq_save(*flags); | |
991 | rq = task_rq(p); | |
992 | spin_lock(&rq->lock); | |
993 | if (likely(rq == task_rq(p))) | |
994 | return rq; | |
1da177e4 | 995 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 996 | } |
1da177e4 LT |
997 | } |
998 | ||
ad474cac ON |
999 | void task_rq_unlock_wait(struct task_struct *p) |
1000 | { | |
1001 | struct rq *rq = task_rq(p); | |
1002 | ||
1003 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1004 | spin_unlock_wait(&rq->lock); | |
1005 | } | |
1006 | ||
a9957449 | 1007 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1008 | __releases(rq->lock) |
1009 | { | |
1010 | spin_unlock(&rq->lock); | |
1011 | } | |
1012 | ||
70b97a7f | 1013 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1014 | __releases(rq->lock) |
1015 | { | |
1016 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1017 | } | |
1018 | ||
1da177e4 | 1019 | /* |
cc2a73b5 | 1020 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1021 | */ |
a9957449 | 1022 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1023 | __acquires(rq->lock) |
1024 | { | |
70b97a7f | 1025 | struct rq *rq; |
1da177e4 LT |
1026 | |
1027 | local_irq_disable(); | |
1028 | rq = this_rq(); | |
1029 | spin_lock(&rq->lock); | |
1030 | ||
1031 | return rq; | |
1032 | } | |
1033 | ||
8f4d37ec PZ |
1034 | #ifdef CONFIG_SCHED_HRTICK |
1035 | /* | |
1036 | * Use HR-timers to deliver accurate preemption points. | |
1037 | * | |
1038 | * Its all a bit involved since we cannot program an hrt while holding the | |
1039 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1040 | * reschedule event. | |
1041 | * | |
1042 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1043 | * rq->lock. | |
1044 | */ | |
8f4d37ec PZ |
1045 | |
1046 | /* | |
1047 | * Use hrtick when: | |
1048 | * - enabled by features | |
1049 | * - hrtimer is actually high res | |
1050 | */ | |
1051 | static inline int hrtick_enabled(struct rq *rq) | |
1052 | { | |
1053 | if (!sched_feat(HRTICK)) | |
1054 | return 0; | |
ba42059f | 1055 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1056 | return 0; |
8f4d37ec PZ |
1057 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1058 | } | |
1059 | ||
8f4d37ec PZ |
1060 | static void hrtick_clear(struct rq *rq) |
1061 | { | |
1062 | if (hrtimer_active(&rq->hrtick_timer)) | |
1063 | hrtimer_cancel(&rq->hrtick_timer); | |
1064 | } | |
1065 | ||
8f4d37ec PZ |
1066 | /* |
1067 | * High-resolution timer tick. | |
1068 | * Runs from hardirq context with interrupts disabled. | |
1069 | */ | |
1070 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1071 | { | |
1072 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1073 | ||
1074 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1075 | ||
1076 | spin_lock(&rq->lock); | |
3e51f33f | 1077 | update_rq_clock(rq); |
8f4d37ec PZ |
1078 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1079 | spin_unlock(&rq->lock); | |
1080 | ||
1081 | return HRTIMER_NORESTART; | |
1082 | } | |
1083 | ||
95e904c7 | 1084 | #ifdef CONFIG_SMP |
31656519 PZ |
1085 | /* |
1086 | * called from hardirq (IPI) context | |
1087 | */ | |
1088 | static void __hrtick_start(void *arg) | |
b328ca18 | 1089 | { |
31656519 | 1090 | struct rq *rq = arg; |
b328ca18 | 1091 | |
31656519 PZ |
1092 | spin_lock(&rq->lock); |
1093 | hrtimer_restart(&rq->hrtick_timer); | |
1094 | rq->hrtick_csd_pending = 0; | |
1095 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1096 | } |
1097 | ||
31656519 PZ |
1098 | /* |
1099 | * Called to set the hrtick timer state. | |
1100 | * | |
1101 | * called with rq->lock held and irqs disabled | |
1102 | */ | |
1103 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1104 | { |
31656519 PZ |
1105 | struct hrtimer *timer = &rq->hrtick_timer; |
1106 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1107 | |
cc584b21 | 1108 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1109 | |
1110 | if (rq == this_rq()) { | |
1111 | hrtimer_restart(timer); | |
1112 | } else if (!rq->hrtick_csd_pending) { | |
1113 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1114 | rq->hrtick_csd_pending = 1; | |
1115 | } | |
b328ca18 PZ |
1116 | } |
1117 | ||
1118 | static int | |
1119 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1120 | { | |
1121 | int cpu = (int)(long)hcpu; | |
1122 | ||
1123 | switch (action) { | |
1124 | case CPU_UP_CANCELED: | |
1125 | case CPU_UP_CANCELED_FROZEN: | |
1126 | case CPU_DOWN_PREPARE: | |
1127 | case CPU_DOWN_PREPARE_FROZEN: | |
1128 | case CPU_DEAD: | |
1129 | case CPU_DEAD_FROZEN: | |
31656519 | 1130 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1131 | return NOTIFY_OK; |
1132 | } | |
1133 | ||
1134 | return NOTIFY_DONE; | |
1135 | } | |
1136 | ||
fa748203 | 1137 | static __init void init_hrtick(void) |
b328ca18 PZ |
1138 | { |
1139 | hotcpu_notifier(hotplug_hrtick, 0); | |
1140 | } | |
31656519 PZ |
1141 | #else |
1142 | /* | |
1143 | * Called to set the hrtick timer state. | |
1144 | * | |
1145 | * called with rq->lock held and irqs disabled | |
1146 | */ | |
1147 | static void hrtick_start(struct rq *rq, u64 delay) | |
1148 | { | |
1149 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1150 | } | |
b328ca18 | 1151 | |
006c75f1 | 1152 | static inline void init_hrtick(void) |
8f4d37ec | 1153 | { |
8f4d37ec | 1154 | } |
31656519 | 1155 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1156 | |
31656519 | 1157 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1158 | { |
31656519 PZ |
1159 | #ifdef CONFIG_SMP |
1160 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1161 | |
31656519 PZ |
1162 | rq->hrtick_csd.flags = 0; |
1163 | rq->hrtick_csd.func = __hrtick_start; | |
1164 | rq->hrtick_csd.info = rq; | |
1165 | #endif | |
8f4d37ec | 1166 | |
31656519 PZ |
1167 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1168 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1169 | } |
006c75f1 | 1170 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1171 | static inline void hrtick_clear(struct rq *rq) |
1172 | { | |
1173 | } | |
1174 | ||
8f4d37ec PZ |
1175 | static inline void init_rq_hrtick(struct rq *rq) |
1176 | { | |
1177 | } | |
1178 | ||
b328ca18 PZ |
1179 | static inline void init_hrtick(void) |
1180 | { | |
1181 | } | |
006c75f1 | 1182 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1183 | |
c24d20db IM |
1184 | /* |
1185 | * resched_task - mark a task 'to be rescheduled now'. | |
1186 | * | |
1187 | * On UP this means the setting of the need_resched flag, on SMP it | |
1188 | * might also involve a cross-CPU call to trigger the scheduler on | |
1189 | * the target CPU. | |
1190 | */ | |
1191 | #ifdef CONFIG_SMP | |
1192 | ||
1193 | #ifndef tsk_is_polling | |
1194 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1195 | #endif | |
1196 | ||
31656519 | 1197 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1198 | { |
1199 | int cpu; | |
1200 | ||
1201 | assert_spin_locked(&task_rq(p)->lock); | |
1202 | ||
5ed0cec0 | 1203 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1204 | return; |
1205 | ||
5ed0cec0 | 1206 | set_tsk_need_resched(p); |
c24d20db IM |
1207 | |
1208 | cpu = task_cpu(p); | |
1209 | if (cpu == smp_processor_id()) | |
1210 | return; | |
1211 | ||
1212 | /* NEED_RESCHED must be visible before we test polling */ | |
1213 | smp_mb(); | |
1214 | if (!tsk_is_polling(p)) | |
1215 | smp_send_reschedule(cpu); | |
1216 | } | |
1217 | ||
1218 | static void resched_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | unsigned long flags; | |
1222 | ||
1223 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1224 | return; | |
1225 | resched_task(cpu_curr(cpu)); | |
1226 | spin_unlock_irqrestore(&rq->lock, flags); | |
1227 | } | |
06d8308c TG |
1228 | |
1229 | #ifdef CONFIG_NO_HZ | |
1230 | /* | |
1231 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1232 | * idle CPU then this timer might expire before the next timer event | |
1233 | * which is scheduled to wake up that CPU. In case of a completely | |
1234 | * idle system the next event might even be infinite time into the | |
1235 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1236 | * leaves the inner idle loop so the newly added timer is taken into | |
1237 | * account when the CPU goes back to idle and evaluates the timer | |
1238 | * wheel for the next timer event. | |
1239 | */ | |
1240 | void wake_up_idle_cpu(int cpu) | |
1241 | { | |
1242 | struct rq *rq = cpu_rq(cpu); | |
1243 | ||
1244 | if (cpu == smp_processor_id()) | |
1245 | return; | |
1246 | ||
1247 | /* | |
1248 | * This is safe, as this function is called with the timer | |
1249 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1250 | * to idle and has not yet set rq->curr to idle then it will | |
1251 | * be serialized on the timer wheel base lock and take the new | |
1252 | * timer into account automatically. | |
1253 | */ | |
1254 | if (rq->curr != rq->idle) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1259 | * lockless. The worst case is that the other CPU runs the | |
1260 | * idle task through an additional NOOP schedule() | |
1261 | */ | |
5ed0cec0 | 1262 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1263 | |
1264 | /* NEED_RESCHED must be visible before we test polling */ | |
1265 | smp_mb(); | |
1266 | if (!tsk_is_polling(rq->idle)) | |
1267 | smp_send_reschedule(cpu); | |
1268 | } | |
6d6bc0ad | 1269 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1270 | |
6d6bc0ad | 1271 | #else /* !CONFIG_SMP */ |
31656519 | 1272 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1273 | { |
1274 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1275 | set_tsk_need_resched(p); |
c24d20db | 1276 | } |
6d6bc0ad | 1277 | #endif /* CONFIG_SMP */ |
c24d20db | 1278 | |
45bf76df IM |
1279 | #if BITS_PER_LONG == 32 |
1280 | # define WMULT_CONST (~0UL) | |
1281 | #else | |
1282 | # define WMULT_CONST (1UL << 32) | |
1283 | #endif | |
1284 | ||
1285 | #define WMULT_SHIFT 32 | |
1286 | ||
194081eb IM |
1287 | /* |
1288 | * Shift right and round: | |
1289 | */ | |
cf2ab469 | 1290 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1291 | |
a7be37ac PZ |
1292 | /* |
1293 | * delta *= weight / lw | |
1294 | */ | |
cb1c4fc9 | 1295 | static unsigned long |
45bf76df IM |
1296 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1297 | struct load_weight *lw) | |
1298 | { | |
1299 | u64 tmp; | |
1300 | ||
7a232e03 LJ |
1301 | if (!lw->inv_weight) { |
1302 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1303 | lw->inv_weight = 1; | |
1304 | else | |
1305 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1306 | / (lw->weight+1); | |
1307 | } | |
45bf76df IM |
1308 | |
1309 | tmp = (u64)delta_exec * weight; | |
1310 | /* | |
1311 | * Check whether we'd overflow the 64-bit multiplication: | |
1312 | */ | |
194081eb | 1313 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1314 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1315 | WMULT_SHIFT/2); |
1316 | else | |
cf2ab469 | 1317 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1318 | |
ecf691da | 1319 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1320 | } |
1321 | ||
1091985b | 1322 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1323 | { |
1324 | lw->weight += inc; | |
e89996ae | 1325 | lw->inv_weight = 0; |
45bf76df IM |
1326 | } |
1327 | ||
1091985b | 1328 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1329 | { |
1330 | lw->weight -= dec; | |
e89996ae | 1331 | lw->inv_weight = 0; |
45bf76df IM |
1332 | } |
1333 | ||
2dd73a4f PW |
1334 | /* |
1335 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1336 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1337 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1338 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1339 | * scaled version of the new time slice allocation that they receive on time |
1340 | * slice expiry etc. | |
1341 | */ | |
1342 | ||
cce7ade8 PZ |
1343 | #define WEIGHT_IDLEPRIO 3 |
1344 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1345 | |
1346 | /* | |
1347 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1348 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1349 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1350 | * that remained on nice 0. | |
1351 | * | |
1352 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1353 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1354 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1355 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1356 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1357 | */ |
1358 | static const int prio_to_weight[40] = { | |
254753dc IM |
1359 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1360 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1361 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1362 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1363 | /* 0 */ 1024, 820, 655, 526, 423, | |
1364 | /* 5 */ 335, 272, 215, 172, 137, | |
1365 | /* 10 */ 110, 87, 70, 56, 45, | |
1366 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1367 | }; |
1368 | ||
5714d2de IM |
1369 | /* |
1370 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1371 | * | |
1372 | * In cases where the weight does not change often, we can use the | |
1373 | * precalculated inverse to speed up arithmetics by turning divisions | |
1374 | * into multiplications: | |
1375 | */ | |
dd41f596 | 1376 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1377 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1378 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1379 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1380 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1381 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1382 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1383 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1384 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1385 | }; |
2dd73a4f | 1386 | |
dd41f596 IM |
1387 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1388 | ||
1389 | /* | |
1390 | * runqueue iterator, to support SMP load-balancing between different | |
1391 | * scheduling classes, without having to expose their internal data | |
1392 | * structures to the load-balancing proper: | |
1393 | */ | |
1394 | struct rq_iterator { | |
1395 | void *arg; | |
1396 | struct task_struct *(*start)(void *); | |
1397 | struct task_struct *(*next)(void *); | |
1398 | }; | |
1399 | ||
e1d1484f PW |
1400 | #ifdef CONFIG_SMP |
1401 | static unsigned long | |
1402 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1403 | unsigned long max_load_move, struct sched_domain *sd, | |
1404 | enum cpu_idle_type idle, int *all_pinned, | |
1405 | int *this_best_prio, struct rq_iterator *iterator); | |
1406 | ||
1407 | static int | |
1408 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1409 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1410 | struct rq_iterator *iterator); | |
e1d1484f | 1411 | #endif |
dd41f596 | 1412 | |
d842de87 SV |
1413 | #ifdef CONFIG_CGROUP_CPUACCT |
1414 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1415 | #else | |
1416 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1417 | #endif | |
1418 | ||
18d95a28 PZ |
1419 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1420 | { | |
1421 | update_load_add(&rq->load, load); | |
1422 | } | |
1423 | ||
1424 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1425 | { | |
1426 | update_load_sub(&rq->load, load); | |
1427 | } | |
1428 | ||
7940ca36 | 1429 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1430 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1431 | |
1432 | /* | |
1433 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1434 | * leaving it for the final time. | |
1435 | */ | |
eb755805 | 1436 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1437 | { |
1438 | struct task_group *parent, *child; | |
eb755805 | 1439 | int ret; |
c09595f6 PZ |
1440 | |
1441 | rcu_read_lock(); | |
1442 | parent = &root_task_group; | |
1443 | down: | |
eb755805 PZ |
1444 | ret = (*down)(parent, data); |
1445 | if (ret) | |
1446 | goto out_unlock; | |
c09595f6 PZ |
1447 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1448 | parent = child; | |
1449 | goto down; | |
1450 | ||
1451 | up: | |
1452 | continue; | |
1453 | } | |
eb755805 PZ |
1454 | ret = (*up)(parent, data); |
1455 | if (ret) | |
1456 | goto out_unlock; | |
c09595f6 PZ |
1457 | |
1458 | child = parent; | |
1459 | parent = parent->parent; | |
1460 | if (parent) | |
1461 | goto up; | |
eb755805 | 1462 | out_unlock: |
c09595f6 | 1463 | rcu_read_unlock(); |
eb755805 PZ |
1464 | |
1465 | return ret; | |
c09595f6 PZ |
1466 | } |
1467 | ||
eb755805 PZ |
1468 | static int tg_nop(struct task_group *tg, void *data) |
1469 | { | |
1470 | return 0; | |
c09595f6 | 1471 | } |
eb755805 PZ |
1472 | #endif |
1473 | ||
1474 | #ifdef CONFIG_SMP | |
1475 | static unsigned long source_load(int cpu, int type); | |
1476 | static unsigned long target_load(int cpu, int type); | |
1477 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1478 | ||
1479 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1480 | { | |
1481 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1482 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1483 | |
4cd42620 SR |
1484 | if (nr_running) |
1485 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1486 | else |
1487 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1488 | |
1489 | return rq->avg_load_per_task; | |
1490 | } | |
1491 | ||
1492 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1493 | |
c09595f6 PZ |
1494 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1495 | ||
1496 | /* | |
1497 | * Calculate and set the cpu's group shares. | |
1498 | */ | |
1499 | static void | |
ffda12a1 PZ |
1500 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1501 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1502 | { |
c09595f6 PZ |
1503 | unsigned long shares; |
1504 | unsigned long rq_weight; | |
1505 | ||
c8cba857 | 1506 | if (!tg->se[cpu]) |
c09595f6 PZ |
1507 | return; |
1508 | ||
ec4e0e2f | 1509 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1510 | |
c09595f6 PZ |
1511 | /* |
1512 | * \Sum shares * rq_weight | |
1513 | * shares = ----------------------- | |
1514 | * \Sum rq_weight | |
1515 | * | |
1516 | */ | |
ec4e0e2f | 1517 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1518 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1519 | |
ffda12a1 PZ |
1520 | if (abs(shares - tg->se[cpu]->load.weight) > |
1521 | sysctl_sched_shares_thresh) { | |
1522 | struct rq *rq = cpu_rq(cpu); | |
1523 | unsigned long flags; | |
c09595f6 | 1524 | |
ffda12a1 | 1525 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1526 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1527 | |
ffda12a1 PZ |
1528 | __set_se_shares(tg->se[cpu], shares); |
1529 | spin_unlock_irqrestore(&rq->lock, flags); | |
1530 | } | |
18d95a28 | 1531 | } |
c09595f6 PZ |
1532 | |
1533 | /* | |
c8cba857 PZ |
1534 | * Re-compute the task group their per cpu shares over the given domain. |
1535 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1536 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1537 | */ |
eb755805 | 1538 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1539 | { |
ec4e0e2f | 1540 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1541 | unsigned long shares = 0; |
eb755805 | 1542 | struct sched_domain *sd = data; |
c8cba857 | 1543 | int i; |
c09595f6 | 1544 | |
758b2cdc | 1545 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1546 | /* |
1547 | * If there are currently no tasks on the cpu pretend there | |
1548 | * is one of average load so that when a new task gets to | |
1549 | * run here it will not get delayed by group starvation. | |
1550 | */ | |
1551 | weight = tg->cfs_rq[i]->load.weight; | |
1552 | if (!weight) | |
1553 | weight = NICE_0_LOAD; | |
1554 | ||
1555 | tg->cfs_rq[i]->rq_weight = weight; | |
1556 | rq_weight += weight; | |
c8cba857 | 1557 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1558 | } |
c09595f6 | 1559 | |
c8cba857 PZ |
1560 | if ((!shares && rq_weight) || shares > tg->shares) |
1561 | shares = tg->shares; | |
1562 | ||
1563 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1564 | shares = tg->shares; | |
c09595f6 | 1565 | |
758b2cdc | 1566 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1567 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1568 | |
1569 | return 0; | |
c09595f6 PZ |
1570 | } |
1571 | ||
1572 | /* | |
c8cba857 PZ |
1573 | * Compute the cpu's hierarchical load factor for each task group. |
1574 | * This needs to be done in a top-down fashion because the load of a child | |
1575 | * group is a fraction of its parents load. | |
c09595f6 | 1576 | */ |
eb755805 | 1577 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1578 | { |
c8cba857 | 1579 | unsigned long load; |
eb755805 | 1580 | long cpu = (long)data; |
c09595f6 | 1581 | |
c8cba857 PZ |
1582 | if (!tg->parent) { |
1583 | load = cpu_rq(cpu)->load.weight; | |
1584 | } else { | |
1585 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1586 | load *= tg->cfs_rq[cpu]->shares; | |
1587 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1588 | } | |
c09595f6 | 1589 | |
c8cba857 | 1590 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1591 | |
eb755805 | 1592 | return 0; |
c09595f6 PZ |
1593 | } |
1594 | ||
c8cba857 | 1595 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1596 | { |
2398f2c6 PZ |
1597 | u64 now = cpu_clock(raw_smp_processor_id()); |
1598 | s64 elapsed = now - sd->last_update; | |
1599 | ||
1600 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1601 | sd->last_update = now; | |
eb755805 | 1602 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1603 | } |
4d8d595d PZ |
1604 | } |
1605 | ||
3e5459b4 PZ |
1606 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1607 | { | |
1608 | spin_unlock(&rq->lock); | |
1609 | update_shares(sd); | |
1610 | spin_lock(&rq->lock); | |
1611 | } | |
1612 | ||
eb755805 | 1613 | static void update_h_load(long cpu) |
c09595f6 | 1614 | { |
eb755805 | 1615 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1616 | } |
1617 | ||
c09595f6 PZ |
1618 | #else |
1619 | ||
c8cba857 | 1620 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1621 | { |
1622 | } | |
1623 | ||
3e5459b4 PZ |
1624 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1625 | { | |
1626 | } | |
1627 | ||
18d95a28 PZ |
1628 | #endif |
1629 | ||
8f45e2b5 GH |
1630 | #ifdef CONFIG_PREEMPT |
1631 | ||
70574a99 | 1632 | /* |
8f45e2b5 GH |
1633 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1634 | * way at the expense of forcing extra atomic operations in all | |
1635 | * invocations. This assures that the double_lock is acquired using the | |
1636 | * same underlying policy as the spinlock_t on this architecture, which | |
1637 | * reduces latency compared to the unfair variant below. However, it | |
1638 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1639 | */ |
8f45e2b5 GH |
1640 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1641 | __releases(this_rq->lock) | |
1642 | __acquires(busiest->lock) | |
1643 | __acquires(this_rq->lock) | |
1644 | { | |
1645 | spin_unlock(&this_rq->lock); | |
1646 | double_rq_lock(this_rq, busiest); | |
1647 | ||
1648 | return 1; | |
1649 | } | |
1650 | ||
1651 | #else | |
1652 | /* | |
1653 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1654 | * latency by eliminating extra atomic operations when the locks are | |
1655 | * already in proper order on entry. This favors lower cpu-ids and will | |
1656 | * grant the double lock to lower cpus over higher ids under contention, | |
1657 | * regardless of entry order into the function. | |
1658 | */ | |
1659 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1660 | __releases(this_rq->lock) |
1661 | __acquires(busiest->lock) | |
1662 | __acquires(this_rq->lock) | |
1663 | { | |
1664 | int ret = 0; | |
1665 | ||
70574a99 AD |
1666 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1667 | if (busiest < this_rq) { | |
1668 | spin_unlock(&this_rq->lock); | |
1669 | spin_lock(&busiest->lock); | |
1670 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1671 | ret = 1; | |
1672 | } else | |
1673 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1674 | } | |
1675 | return ret; | |
1676 | } | |
1677 | ||
8f45e2b5 GH |
1678 | #endif /* CONFIG_PREEMPT */ |
1679 | ||
1680 | /* | |
1681 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1682 | */ | |
1683 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1684 | { | |
1685 | if (unlikely(!irqs_disabled())) { | |
1686 | /* printk() doesn't work good under rq->lock */ | |
1687 | spin_unlock(&this_rq->lock); | |
1688 | BUG_ON(1); | |
1689 | } | |
1690 | ||
1691 | return _double_lock_balance(this_rq, busiest); | |
1692 | } | |
1693 | ||
70574a99 AD |
1694 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1695 | __releases(busiest->lock) | |
1696 | { | |
1697 | spin_unlock(&busiest->lock); | |
1698 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1699 | } | |
18d95a28 PZ |
1700 | #endif |
1701 | ||
30432094 | 1702 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1703 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1704 | { | |
30432094 | 1705 | #ifdef CONFIG_SMP |
34e83e85 IM |
1706 | cfs_rq->shares = shares; |
1707 | #endif | |
1708 | } | |
30432094 | 1709 | #endif |
e7693a36 | 1710 | |
dd41f596 | 1711 | #include "sched_stats.h" |
dd41f596 | 1712 | #include "sched_idletask.c" |
5522d5d5 IM |
1713 | #include "sched_fair.c" |
1714 | #include "sched_rt.c" | |
dd41f596 IM |
1715 | #ifdef CONFIG_SCHED_DEBUG |
1716 | # include "sched_debug.c" | |
1717 | #endif | |
1718 | ||
1719 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1720 | #define for_each_class(class) \ |
1721 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1722 | |
c09595f6 | 1723 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1724 | { |
1725 | rq->nr_running++; | |
9c217245 IM |
1726 | } |
1727 | ||
c09595f6 | 1728 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1729 | { |
1730 | rq->nr_running--; | |
9c217245 IM |
1731 | } |
1732 | ||
45bf76df IM |
1733 | static void set_load_weight(struct task_struct *p) |
1734 | { | |
1735 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1736 | p->se.load.weight = prio_to_weight[0] * 2; |
1737 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1738 | return; | |
1739 | } | |
45bf76df | 1740 | |
dd41f596 IM |
1741 | /* |
1742 | * SCHED_IDLE tasks get minimal weight: | |
1743 | */ | |
1744 | if (p->policy == SCHED_IDLE) { | |
1745 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1746 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1747 | return; | |
1748 | } | |
71f8bd46 | 1749 | |
dd41f596 IM |
1750 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1751 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1752 | } |
1753 | ||
2087a1ad GH |
1754 | static void update_avg(u64 *avg, u64 sample) |
1755 | { | |
1756 | s64 diff = sample - *avg; | |
1757 | *avg += diff >> 3; | |
1758 | } | |
1759 | ||
8159f87e | 1760 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1761 | { |
831451ac PZ |
1762 | if (wakeup) |
1763 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1764 | ||
dd41f596 | 1765 | sched_info_queued(p); |
fd390f6a | 1766 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1767 | p->se.on_rq = 1; |
71f8bd46 IM |
1768 | } |
1769 | ||
69be72c1 | 1770 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1771 | { |
831451ac PZ |
1772 | if (sleep) { |
1773 | if (p->se.last_wakeup) { | |
1774 | update_avg(&p->se.avg_overlap, | |
1775 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1776 | p->se.last_wakeup = 0; | |
1777 | } else { | |
1778 | update_avg(&p->se.avg_wakeup, | |
1779 | sysctl_sched_wakeup_granularity); | |
1780 | } | |
2087a1ad GH |
1781 | } |
1782 | ||
46ac22ba | 1783 | sched_info_dequeued(p); |
f02231e5 | 1784 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1785 | p->se.on_rq = 0; |
71f8bd46 IM |
1786 | } |
1787 | ||
14531189 | 1788 | /* |
dd41f596 | 1789 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1790 | */ |
14531189 IM |
1791 | static inline int __normal_prio(struct task_struct *p) |
1792 | { | |
dd41f596 | 1793 | return p->static_prio; |
14531189 IM |
1794 | } |
1795 | ||
b29739f9 IM |
1796 | /* |
1797 | * Calculate the expected normal priority: i.e. priority | |
1798 | * without taking RT-inheritance into account. Might be | |
1799 | * boosted by interactivity modifiers. Changes upon fork, | |
1800 | * setprio syscalls, and whenever the interactivity | |
1801 | * estimator recalculates. | |
1802 | */ | |
36c8b586 | 1803 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1804 | { |
1805 | int prio; | |
1806 | ||
e05606d3 | 1807 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1808 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1809 | else | |
1810 | prio = __normal_prio(p); | |
1811 | return prio; | |
1812 | } | |
1813 | ||
1814 | /* | |
1815 | * Calculate the current priority, i.e. the priority | |
1816 | * taken into account by the scheduler. This value might | |
1817 | * be boosted by RT tasks, or might be boosted by | |
1818 | * interactivity modifiers. Will be RT if the task got | |
1819 | * RT-boosted. If not then it returns p->normal_prio. | |
1820 | */ | |
36c8b586 | 1821 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1822 | { |
1823 | p->normal_prio = normal_prio(p); | |
1824 | /* | |
1825 | * If we are RT tasks or we were boosted to RT priority, | |
1826 | * keep the priority unchanged. Otherwise, update priority | |
1827 | * to the normal priority: | |
1828 | */ | |
1829 | if (!rt_prio(p->prio)) | |
1830 | return p->normal_prio; | |
1831 | return p->prio; | |
1832 | } | |
1833 | ||
1da177e4 | 1834 | /* |
dd41f596 | 1835 | * activate_task - move a task to the runqueue. |
1da177e4 | 1836 | */ |
dd41f596 | 1837 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1838 | { |
d9514f6c | 1839 | if (task_contributes_to_load(p)) |
dd41f596 | 1840 | rq->nr_uninterruptible--; |
1da177e4 | 1841 | |
8159f87e | 1842 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1843 | inc_nr_running(rq); |
1da177e4 LT |
1844 | } |
1845 | ||
1da177e4 LT |
1846 | /* |
1847 | * deactivate_task - remove a task from the runqueue. | |
1848 | */ | |
2e1cb74a | 1849 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1850 | { |
d9514f6c | 1851 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1852 | rq->nr_uninterruptible++; |
1853 | ||
69be72c1 | 1854 | dequeue_task(rq, p, sleep); |
c09595f6 | 1855 | dec_nr_running(rq); |
1da177e4 LT |
1856 | } |
1857 | ||
1da177e4 LT |
1858 | /** |
1859 | * task_curr - is this task currently executing on a CPU? | |
1860 | * @p: the task in question. | |
1861 | */ | |
36c8b586 | 1862 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1863 | { |
1864 | return cpu_curr(task_cpu(p)) == p; | |
1865 | } | |
1866 | ||
dd41f596 IM |
1867 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1868 | { | |
6f505b16 | 1869 | set_task_rq(p, cpu); |
dd41f596 | 1870 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1871 | /* |
1872 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1873 | * successfuly executed on another CPU. We must ensure that updates of | |
1874 | * per-task data have been completed by this moment. | |
1875 | */ | |
1876 | smp_wmb(); | |
dd41f596 | 1877 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1878 | #endif |
2dd73a4f PW |
1879 | } |
1880 | ||
cb469845 SR |
1881 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1882 | const struct sched_class *prev_class, | |
1883 | int oldprio, int running) | |
1884 | { | |
1885 | if (prev_class != p->sched_class) { | |
1886 | if (prev_class->switched_from) | |
1887 | prev_class->switched_from(rq, p, running); | |
1888 | p->sched_class->switched_to(rq, p, running); | |
1889 | } else | |
1890 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1891 | } | |
1892 | ||
1da177e4 | 1893 | #ifdef CONFIG_SMP |
c65cc870 | 1894 | |
e958b360 TG |
1895 | /* Used instead of source_load when we know the type == 0 */ |
1896 | static unsigned long weighted_cpuload(const int cpu) | |
1897 | { | |
1898 | return cpu_rq(cpu)->load.weight; | |
1899 | } | |
1900 | ||
cc367732 IM |
1901 | /* |
1902 | * Is this task likely cache-hot: | |
1903 | */ | |
e7693a36 | 1904 | static int |
cc367732 IM |
1905 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1906 | { | |
1907 | s64 delta; | |
1908 | ||
f540a608 IM |
1909 | /* |
1910 | * Buddy candidates are cache hot: | |
1911 | */ | |
4793241b PZ |
1912 | if (sched_feat(CACHE_HOT_BUDDY) && |
1913 | (&p->se == cfs_rq_of(&p->se)->next || | |
1914 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1915 | return 1; |
1916 | ||
cc367732 IM |
1917 | if (p->sched_class != &fair_sched_class) |
1918 | return 0; | |
1919 | ||
6bc1665b IM |
1920 | if (sysctl_sched_migration_cost == -1) |
1921 | return 1; | |
1922 | if (sysctl_sched_migration_cost == 0) | |
1923 | return 0; | |
1924 | ||
cc367732 IM |
1925 | delta = now - p->se.exec_start; |
1926 | ||
1927 | return delta < (s64)sysctl_sched_migration_cost; | |
1928 | } | |
1929 | ||
1930 | ||
dd41f596 | 1931 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1932 | { |
dd41f596 IM |
1933 | int old_cpu = task_cpu(p); |
1934 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1935 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1936 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1937 | u64 clock_offset; |
dd41f596 IM |
1938 | |
1939 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1940 | |
cbc34ed1 PZ |
1941 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1942 | ||
6cfb0d5d IM |
1943 | #ifdef CONFIG_SCHEDSTATS |
1944 | if (p->se.wait_start) | |
1945 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1946 | if (p->se.sleep_start) |
1947 | p->se.sleep_start -= clock_offset; | |
1948 | if (p->se.block_start) | |
1949 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1950 | if (old_cpu != new_cpu) { |
1951 | schedstat_inc(p, se.nr_migrations); | |
1952 | if (task_hot(p, old_rq->clock, NULL)) | |
1953 | schedstat_inc(p, se.nr_forced2_migrations); | |
1954 | } | |
6cfb0d5d | 1955 | #endif |
2830cf8c SV |
1956 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1957 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1958 | |
1959 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1960 | } |
1961 | ||
70b97a7f | 1962 | struct migration_req { |
1da177e4 | 1963 | struct list_head list; |
1da177e4 | 1964 | |
36c8b586 | 1965 | struct task_struct *task; |
1da177e4 LT |
1966 | int dest_cpu; |
1967 | ||
1da177e4 | 1968 | struct completion done; |
70b97a7f | 1969 | }; |
1da177e4 LT |
1970 | |
1971 | /* | |
1972 | * The task's runqueue lock must be held. | |
1973 | * Returns true if you have to wait for migration thread. | |
1974 | */ | |
36c8b586 | 1975 | static int |
70b97a7f | 1976 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1977 | { |
70b97a7f | 1978 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1979 | |
1980 | /* | |
1981 | * If the task is not on a runqueue (and not running), then | |
1982 | * it is sufficient to simply update the task's cpu field. | |
1983 | */ | |
dd41f596 | 1984 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1985 | set_task_cpu(p, dest_cpu); |
1986 | return 0; | |
1987 | } | |
1988 | ||
1989 | init_completion(&req->done); | |
1da177e4 LT |
1990 | req->task = p; |
1991 | req->dest_cpu = dest_cpu; | |
1992 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1993 | |
1da177e4 LT |
1994 | return 1; |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * wait_task_inactive - wait for a thread to unschedule. | |
1999 | * | |
85ba2d86 RM |
2000 | * If @match_state is nonzero, it's the @p->state value just checked and |
2001 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2002 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2003 | * we return a positive number (its total switch count). If a second call | |
2004 | * a short while later returns the same number, the caller can be sure that | |
2005 | * @p has remained unscheduled the whole time. | |
2006 | * | |
1da177e4 LT |
2007 | * The caller must ensure that the task *will* unschedule sometime soon, |
2008 | * else this function might spin for a *long* time. This function can't | |
2009 | * be called with interrupts off, or it may introduce deadlock with | |
2010 | * smp_call_function() if an IPI is sent by the same process we are | |
2011 | * waiting to become inactive. | |
2012 | */ | |
85ba2d86 | 2013 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2014 | { |
2015 | unsigned long flags; | |
dd41f596 | 2016 | int running, on_rq; |
85ba2d86 | 2017 | unsigned long ncsw; |
70b97a7f | 2018 | struct rq *rq; |
1da177e4 | 2019 | |
3a5c359a AK |
2020 | for (;;) { |
2021 | /* | |
2022 | * We do the initial early heuristics without holding | |
2023 | * any task-queue locks at all. We'll only try to get | |
2024 | * the runqueue lock when things look like they will | |
2025 | * work out! | |
2026 | */ | |
2027 | rq = task_rq(p); | |
fa490cfd | 2028 | |
3a5c359a AK |
2029 | /* |
2030 | * If the task is actively running on another CPU | |
2031 | * still, just relax and busy-wait without holding | |
2032 | * any locks. | |
2033 | * | |
2034 | * NOTE! Since we don't hold any locks, it's not | |
2035 | * even sure that "rq" stays as the right runqueue! | |
2036 | * But we don't care, since "task_running()" will | |
2037 | * return false if the runqueue has changed and p | |
2038 | * is actually now running somewhere else! | |
2039 | */ | |
85ba2d86 RM |
2040 | while (task_running(rq, p)) { |
2041 | if (match_state && unlikely(p->state != match_state)) | |
2042 | return 0; | |
3a5c359a | 2043 | cpu_relax(); |
85ba2d86 | 2044 | } |
fa490cfd | 2045 | |
3a5c359a AK |
2046 | /* |
2047 | * Ok, time to look more closely! We need the rq | |
2048 | * lock now, to be *sure*. If we're wrong, we'll | |
2049 | * just go back and repeat. | |
2050 | */ | |
2051 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2052 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2053 | running = task_running(rq, p); |
2054 | on_rq = p->se.on_rq; | |
85ba2d86 | 2055 | ncsw = 0; |
f31e11d8 | 2056 | if (!match_state || p->state == match_state) |
93dcf55f | 2057 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2058 | task_rq_unlock(rq, &flags); |
fa490cfd | 2059 | |
85ba2d86 RM |
2060 | /* |
2061 | * If it changed from the expected state, bail out now. | |
2062 | */ | |
2063 | if (unlikely(!ncsw)) | |
2064 | break; | |
2065 | ||
3a5c359a AK |
2066 | /* |
2067 | * Was it really running after all now that we | |
2068 | * checked with the proper locks actually held? | |
2069 | * | |
2070 | * Oops. Go back and try again.. | |
2071 | */ | |
2072 | if (unlikely(running)) { | |
2073 | cpu_relax(); | |
2074 | continue; | |
2075 | } | |
fa490cfd | 2076 | |
3a5c359a AK |
2077 | /* |
2078 | * It's not enough that it's not actively running, | |
2079 | * it must be off the runqueue _entirely_, and not | |
2080 | * preempted! | |
2081 | * | |
80dd99b3 | 2082 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2083 | * running right now), it's preempted, and we should |
2084 | * yield - it could be a while. | |
2085 | */ | |
2086 | if (unlikely(on_rq)) { | |
2087 | schedule_timeout_uninterruptible(1); | |
2088 | continue; | |
2089 | } | |
fa490cfd | 2090 | |
3a5c359a AK |
2091 | /* |
2092 | * Ahh, all good. It wasn't running, and it wasn't | |
2093 | * runnable, which means that it will never become | |
2094 | * running in the future either. We're all done! | |
2095 | */ | |
2096 | break; | |
2097 | } | |
85ba2d86 RM |
2098 | |
2099 | return ncsw; | |
1da177e4 LT |
2100 | } |
2101 | ||
2102 | /*** | |
2103 | * kick_process - kick a running thread to enter/exit the kernel | |
2104 | * @p: the to-be-kicked thread | |
2105 | * | |
2106 | * Cause a process which is running on another CPU to enter | |
2107 | * kernel-mode, without any delay. (to get signals handled.) | |
2108 | * | |
2109 | * NOTE: this function doesnt have to take the runqueue lock, | |
2110 | * because all it wants to ensure is that the remote task enters | |
2111 | * the kernel. If the IPI races and the task has been migrated | |
2112 | * to another CPU then no harm is done and the purpose has been | |
2113 | * achieved as well. | |
2114 | */ | |
36c8b586 | 2115 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2116 | { |
2117 | int cpu; | |
2118 | ||
2119 | preempt_disable(); | |
2120 | cpu = task_cpu(p); | |
2121 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2122 | smp_send_reschedule(cpu); | |
2123 | preempt_enable(); | |
2124 | } | |
2125 | ||
2126 | /* | |
2dd73a4f PW |
2127 | * Return a low guess at the load of a migration-source cpu weighted |
2128 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2129 | * |
2130 | * We want to under-estimate the load of migration sources, to | |
2131 | * balance conservatively. | |
2132 | */ | |
a9957449 | 2133 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2134 | { |
70b97a7f | 2135 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2136 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2137 | |
93b75217 | 2138 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2139 | return total; |
b910472d | 2140 | |
dd41f596 | 2141 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2142 | } |
2143 | ||
2144 | /* | |
2dd73a4f PW |
2145 | * Return a high guess at the load of a migration-target cpu weighted |
2146 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2147 | */ |
a9957449 | 2148 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2149 | { |
70b97a7f | 2150 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2151 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2152 | |
93b75217 | 2153 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2154 | return total; |
3b0bd9bc | 2155 | |
dd41f596 | 2156 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2157 | } |
2158 | ||
147cbb4b NP |
2159 | /* |
2160 | * find_idlest_group finds and returns the least busy CPU group within the | |
2161 | * domain. | |
2162 | */ | |
2163 | static struct sched_group * | |
2164 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2165 | { | |
2166 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2167 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2168 | int load_idx = sd->forkexec_idx; | |
2169 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2170 | ||
2171 | do { | |
2172 | unsigned long load, avg_load; | |
2173 | int local_group; | |
2174 | int i; | |
2175 | ||
da5a5522 | 2176 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2177 | if (!cpumask_intersects(sched_group_cpus(group), |
2178 | &p->cpus_allowed)) | |
3a5c359a | 2179 | continue; |
da5a5522 | 2180 | |
758b2cdc RR |
2181 | local_group = cpumask_test_cpu(this_cpu, |
2182 | sched_group_cpus(group)); | |
147cbb4b NP |
2183 | |
2184 | /* Tally up the load of all CPUs in the group */ | |
2185 | avg_load = 0; | |
2186 | ||
758b2cdc | 2187 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2188 | /* Bias balancing toward cpus of our domain */ |
2189 | if (local_group) | |
2190 | load = source_load(i, load_idx); | |
2191 | else | |
2192 | load = target_load(i, load_idx); | |
2193 | ||
2194 | avg_load += load; | |
2195 | } | |
2196 | ||
2197 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2198 | avg_load = sg_div_cpu_power(group, |
2199 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2200 | |
2201 | if (local_group) { | |
2202 | this_load = avg_load; | |
2203 | this = group; | |
2204 | } else if (avg_load < min_load) { | |
2205 | min_load = avg_load; | |
2206 | idlest = group; | |
2207 | } | |
3a5c359a | 2208 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2209 | |
2210 | if (!idlest || 100*this_load < imbalance*min_load) | |
2211 | return NULL; | |
2212 | return idlest; | |
2213 | } | |
2214 | ||
2215 | /* | |
0feaece9 | 2216 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2217 | */ |
95cdf3b7 | 2218 | static int |
758b2cdc | 2219 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2220 | { |
2221 | unsigned long load, min_load = ULONG_MAX; | |
2222 | int idlest = -1; | |
2223 | int i; | |
2224 | ||
da5a5522 | 2225 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2226 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2227 | load = weighted_cpuload(i); |
147cbb4b NP |
2228 | |
2229 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2230 | min_load = load; | |
2231 | idlest = i; | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | return idlest; | |
2236 | } | |
2237 | ||
476d139c NP |
2238 | /* |
2239 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2240 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2241 | * SD_BALANCE_EXEC. | |
2242 | * | |
2243 | * Balance, ie. select the least loaded group. | |
2244 | * | |
2245 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2246 | * | |
2247 | * preempt must be disabled. | |
2248 | */ | |
2249 | static int sched_balance_self(int cpu, int flag) | |
2250 | { | |
2251 | struct task_struct *t = current; | |
2252 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2253 | |
c96d145e | 2254 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2255 | /* |
2256 | * If power savings logic is enabled for a domain, stop there. | |
2257 | */ | |
5c45bf27 SS |
2258 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2259 | break; | |
476d139c NP |
2260 | if (tmp->flags & flag) |
2261 | sd = tmp; | |
c96d145e | 2262 | } |
476d139c | 2263 | |
039a1c41 PZ |
2264 | if (sd) |
2265 | update_shares(sd); | |
2266 | ||
476d139c | 2267 | while (sd) { |
476d139c | 2268 | struct sched_group *group; |
1a848870 SS |
2269 | int new_cpu, weight; |
2270 | ||
2271 | if (!(sd->flags & flag)) { | |
2272 | sd = sd->child; | |
2273 | continue; | |
2274 | } | |
476d139c | 2275 | |
476d139c | 2276 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2277 | if (!group) { |
2278 | sd = sd->child; | |
2279 | continue; | |
2280 | } | |
476d139c | 2281 | |
758b2cdc | 2282 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2283 | if (new_cpu == -1 || new_cpu == cpu) { |
2284 | /* Now try balancing at a lower domain level of cpu */ | |
2285 | sd = sd->child; | |
2286 | continue; | |
2287 | } | |
476d139c | 2288 | |
1a848870 | 2289 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2290 | cpu = new_cpu; |
758b2cdc | 2291 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2292 | sd = NULL; |
476d139c | 2293 | for_each_domain(cpu, tmp) { |
758b2cdc | 2294 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2295 | break; |
2296 | if (tmp->flags & flag) | |
2297 | sd = tmp; | |
2298 | } | |
2299 | /* while loop will break here if sd == NULL */ | |
2300 | } | |
2301 | ||
2302 | return cpu; | |
2303 | } | |
2304 | ||
2305 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2306 | |
1da177e4 LT |
2307 | /*** |
2308 | * try_to_wake_up - wake up a thread | |
2309 | * @p: the to-be-woken-up thread | |
2310 | * @state: the mask of task states that can be woken | |
2311 | * @sync: do a synchronous wakeup? | |
2312 | * | |
2313 | * Put it on the run-queue if it's not already there. The "current" | |
2314 | * thread is always on the run-queue (except when the actual | |
2315 | * re-schedule is in progress), and as such you're allowed to do | |
2316 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2317 | * runnable without the overhead of this. | |
2318 | * | |
2319 | * returns failure only if the task is already active. | |
2320 | */ | |
36c8b586 | 2321 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2322 | { |
cc367732 | 2323 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2324 | unsigned long flags; |
2325 | long old_state; | |
70b97a7f | 2326 | struct rq *rq; |
1da177e4 | 2327 | |
b85d0667 IM |
2328 | if (!sched_feat(SYNC_WAKEUPS)) |
2329 | sync = 0; | |
2330 | ||
2398f2c6 | 2331 | #ifdef CONFIG_SMP |
57310a98 | 2332 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2333 | struct sched_domain *sd; |
2334 | ||
2335 | this_cpu = raw_smp_processor_id(); | |
2336 | cpu = task_cpu(p); | |
2337 | ||
2338 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2339 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2340 | update_shares(sd); |
2341 | break; | |
2342 | } | |
2343 | } | |
2344 | } | |
2345 | #endif | |
2346 | ||
04e2f174 | 2347 | smp_wmb(); |
1da177e4 | 2348 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2349 | update_rq_clock(rq); |
1da177e4 LT |
2350 | old_state = p->state; |
2351 | if (!(old_state & state)) | |
2352 | goto out; | |
2353 | ||
dd41f596 | 2354 | if (p->se.on_rq) |
1da177e4 LT |
2355 | goto out_running; |
2356 | ||
2357 | cpu = task_cpu(p); | |
cc367732 | 2358 | orig_cpu = cpu; |
1da177e4 LT |
2359 | this_cpu = smp_processor_id(); |
2360 | ||
2361 | #ifdef CONFIG_SMP | |
2362 | if (unlikely(task_running(rq, p))) | |
2363 | goto out_activate; | |
2364 | ||
5d2f5a61 DA |
2365 | cpu = p->sched_class->select_task_rq(p, sync); |
2366 | if (cpu != orig_cpu) { | |
2367 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2368 | task_rq_unlock(rq, &flags); |
2369 | /* might preempt at this point */ | |
2370 | rq = task_rq_lock(p, &flags); | |
2371 | old_state = p->state; | |
2372 | if (!(old_state & state)) | |
2373 | goto out; | |
dd41f596 | 2374 | if (p->se.on_rq) |
1da177e4 LT |
2375 | goto out_running; |
2376 | ||
2377 | this_cpu = smp_processor_id(); | |
2378 | cpu = task_cpu(p); | |
2379 | } | |
2380 | ||
e7693a36 GH |
2381 | #ifdef CONFIG_SCHEDSTATS |
2382 | schedstat_inc(rq, ttwu_count); | |
2383 | if (cpu == this_cpu) | |
2384 | schedstat_inc(rq, ttwu_local); | |
2385 | else { | |
2386 | struct sched_domain *sd; | |
2387 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2388 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2389 | schedstat_inc(sd, ttwu_wake_remote); |
2390 | break; | |
2391 | } | |
2392 | } | |
2393 | } | |
6d6bc0ad | 2394 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2395 | |
1da177e4 LT |
2396 | out_activate: |
2397 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2398 | schedstat_inc(p, se.nr_wakeups); |
2399 | if (sync) | |
2400 | schedstat_inc(p, se.nr_wakeups_sync); | |
2401 | if (orig_cpu != cpu) | |
2402 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2403 | if (cpu == this_cpu) | |
2404 | schedstat_inc(p, se.nr_wakeups_local); | |
2405 | else | |
2406 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2407 | activate_task(rq, p, 1); |
1da177e4 LT |
2408 | success = 1; |
2409 | ||
831451ac PZ |
2410 | /* |
2411 | * Only attribute actual wakeups done by this task. | |
2412 | */ | |
2413 | if (!in_interrupt()) { | |
2414 | struct sched_entity *se = ¤t->se; | |
2415 | u64 sample = se->sum_exec_runtime; | |
2416 | ||
2417 | if (se->last_wakeup) | |
2418 | sample -= se->last_wakeup; | |
2419 | else | |
2420 | sample -= se->start_runtime; | |
2421 | update_avg(&se->avg_wakeup, sample); | |
2422 | ||
2423 | se->last_wakeup = se->sum_exec_runtime; | |
2424 | } | |
2425 | ||
1da177e4 | 2426 | out_running: |
468a15bb | 2427 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2428 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2429 | |
1da177e4 | 2430 | p->state = TASK_RUNNING; |
9a897c5a SR |
2431 | #ifdef CONFIG_SMP |
2432 | if (p->sched_class->task_wake_up) | |
2433 | p->sched_class->task_wake_up(rq, p); | |
2434 | #endif | |
1da177e4 LT |
2435 | out: |
2436 | task_rq_unlock(rq, &flags); | |
2437 | ||
2438 | return success; | |
2439 | } | |
2440 | ||
7ad5b3a5 | 2441 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2442 | { |
d9514f6c | 2443 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2444 | } |
1da177e4 LT |
2445 | EXPORT_SYMBOL(wake_up_process); |
2446 | ||
7ad5b3a5 | 2447 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2448 | { |
2449 | return try_to_wake_up(p, state, 0); | |
2450 | } | |
2451 | ||
1da177e4 LT |
2452 | /* |
2453 | * Perform scheduler related setup for a newly forked process p. | |
2454 | * p is forked by current. | |
dd41f596 IM |
2455 | * |
2456 | * __sched_fork() is basic setup used by init_idle() too: | |
2457 | */ | |
2458 | static void __sched_fork(struct task_struct *p) | |
2459 | { | |
dd41f596 IM |
2460 | p->se.exec_start = 0; |
2461 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2462 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2463 | p->se.last_wakeup = 0; |
2464 | p->se.avg_overlap = 0; | |
831451ac PZ |
2465 | p->se.start_runtime = 0; |
2466 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2467 | |
2468 | #ifdef CONFIG_SCHEDSTATS | |
2469 | p->se.wait_start = 0; | |
dd41f596 IM |
2470 | p->se.sum_sleep_runtime = 0; |
2471 | p->se.sleep_start = 0; | |
dd41f596 IM |
2472 | p->se.block_start = 0; |
2473 | p->se.sleep_max = 0; | |
2474 | p->se.block_max = 0; | |
2475 | p->se.exec_max = 0; | |
eba1ed4b | 2476 | p->se.slice_max = 0; |
dd41f596 | 2477 | p->se.wait_max = 0; |
6cfb0d5d | 2478 | #endif |
476d139c | 2479 | |
fa717060 | 2480 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2481 | p->se.on_rq = 0; |
4a55bd5e | 2482 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2483 | |
e107be36 AK |
2484 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2485 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2486 | #endif | |
2487 | ||
1da177e4 LT |
2488 | /* |
2489 | * We mark the process as running here, but have not actually | |
2490 | * inserted it onto the runqueue yet. This guarantees that | |
2491 | * nobody will actually run it, and a signal or other external | |
2492 | * event cannot wake it up and insert it on the runqueue either. | |
2493 | */ | |
2494 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2495 | } |
2496 | ||
2497 | /* | |
2498 | * fork()/clone()-time setup: | |
2499 | */ | |
2500 | void sched_fork(struct task_struct *p, int clone_flags) | |
2501 | { | |
2502 | int cpu = get_cpu(); | |
2503 | ||
2504 | __sched_fork(p); | |
2505 | ||
2506 | #ifdef CONFIG_SMP | |
2507 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2508 | #endif | |
02e4bac2 | 2509 | set_task_cpu(p, cpu); |
b29739f9 IM |
2510 | |
2511 | /* | |
2512 | * Make sure we do not leak PI boosting priority to the child: | |
2513 | */ | |
2514 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2515 | if (!rt_prio(p->prio)) |
2516 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2517 | |
52f17b6c | 2518 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2519 | if (likely(sched_info_on())) |
52f17b6c | 2520 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2521 | #endif |
d6077cb8 | 2522 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2523 | p->oncpu = 0; |
2524 | #endif | |
1da177e4 | 2525 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2526 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2527 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2528 | #endif |
917b627d GH |
2529 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2530 | ||
476d139c | 2531 | put_cpu(); |
1da177e4 LT |
2532 | } |
2533 | ||
2534 | /* | |
2535 | * wake_up_new_task - wake up a newly created task for the first time. | |
2536 | * | |
2537 | * This function will do some initial scheduler statistics housekeeping | |
2538 | * that must be done for every newly created context, then puts the task | |
2539 | * on the runqueue and wakes it. | |
2540 | */ | |
7ad5b3a5 | 2541 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2542 | { |
2543 | unsigned long flags; | |
dd41f596 | 2544 | struct rq *rq; |
1da177e4 LT |
2545 | |
2546 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2547 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2548 | update_rq_clock(rq); |
1da177e4 LT |
2549 | |
2550 | p->prio = effective_prio(p); | |
2551 | ||
b9dca1e0 | 2552 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2553 | activate_task(rq, p, 0); |
1da177e4 | 2554 | } else { |
1da177e4 | 2555 | /* |
dd41f596 IM |
2556 | * Let the scheduling class do new task startup |
2557 | * management (if any): | |
1da177e4 | 2558 | */ |
ee0827d8 | 2559 | p->sched_class->task_new(rq, p); |
c09595f6 | 2560 | inc_nr_running(rq); |
1da177e4 | 2561 | } |
c71dd42d | 2562 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2563 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2564 | #ifdef CONFIG_SMP |
2565 | if (p->sched_class->task_wake_up) | |
2566 | p->sched_class->task_wake_up(rq, p); | |
2567 | #endif | |
dd41f596 | 2568 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2569 | } |
2570 | ||
e107be36 AK |
2571 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2572 | ||
2573 | /** | |
80dd99b3 | 2574 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2575 | * @notifier: notifier struct to register |
e107be36 AK |
2576 | */ |
2577 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2578 | { | |
2579 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2580 | } | |
2581 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2582 | ||
2583 | /** | |
2584 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2585 | * @notifier: notifier struct to unregister |
e107be36 AK |
2586 | * |
2587 | * This is safe to call from within a preemption notifier. | |
2588 | */ | |
2589 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2590 | { | |
2591 | hlist_del(¬ifier->link); | |
2592 | } | |
2593 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2594 | ||
2595 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2596 | { | |
2597 | struct preempt_notifier *notifier; | |
2598 | struct hlist_node *node; | |
2599 | ||
2600 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2601 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2602 | } | |
2603 | ||
2604 | static void | |
2605 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2606 | struct task_struct *next) | |
2607 | { | |
2608 | struct preempt_notifier *notifier; | |
2609 | struct hlist_node *node; | |
2610 | ||
2611 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2612 | notifier->ops->sched_out(notifier, next); | |
2613 | } | |
2614 | ||
6d6bc0ad | 2615 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2616 | |
2617 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2618 | { | |
2619 | } | |
2620 | ||
2621 | static void | |
2622 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2623 | struct task_struct *next) | |
2624 | { | |
2625 | } | |
2626 | ||
6d6bc0ad | 2627 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2628 | |
4866cde0 NP |
2629 | /** |
2630 | * prepare_task_switch - prepare to switch tasks | |
2631 | * @rq: the runqueue preparing to switch | |
421cee29 | 2632 | * @prev: the current task that is being switched out |
4866cde0 NP |
2633 | * @next: the task we are going to switch to. |
2634 | * | |
2635 | * This is called with the rq lock held and interrupts off. It must | |
2636 | * be paired with a subsequent finish_task_switch after the context | |
2637 | * switch. | |
2638 | * | |
2639 | * prepare_task_switch sets up locking and calls architecture specific | |
2640 | * hooks. | |
2641 | */ | |
e107be36 AK |
2642 | static inline void |
2643 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2644 | struct task_struct *next) | |
4866cde0 | 2645 | { |
e107be36 | 2646 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2647 | prepare_lock_switch(rq, next); |
2648 | prepare_arch_switch(next); | |
2649 | } | |
2650 | ||
1da177e4 LT |
2651 | /** |
2652 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2653 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2654 | * @prev: the thread we just switched away from. |
2655 | * | |
4866cde0 NP |
2656 | * finish_task_switch must be called after the context switch, paired |
2657 | * with a prepare_task_switch call before the context switch. | |
2658 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2659 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2660 | * |
2661 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2662 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2663 | * with the lock held can cause deadlocks; see schedule() for |
2664 | * details.) | |
2665 | */ | |
a9957449 | 2666 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2667 | __releases(rq->lock) |
2668 | { | |
1da177e4 | 2669 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2670 | long prev_state; |
967fc046 GH |
2671 | #ifdef CONFIG_SMP |
2672 | int post_schedule = 0; | |
2673 | ||
2674 | if (current->sched_class->needs_post_schedule) | |
2675 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2676 | #endif | |
1da177e4 LT |
2677 | |
2678 | rq->prev_mm = NULL; | |
2679 | ||
2680 | /* | |
2681 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2682 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2683 | * schedule one last time. The schedule call will never return, and |
2684 | * the scheduled task must drop that reference. | |
c394cc9f | 2685 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2686 | * still held, otherwise prev could be scheduled on another cpu, die |
2687 | * there before we look at prev->state, and then the reference would | |
2688 | * be dropped twice. | |
2689 | * Manfred Spraul <manfred@colorfullife.com> | |
2690 | */ | |
55a101f8 | 2691 | prev_state = prev->state; |
4866cde0 NP |
2692 | finish_arch_switch(prev); |
2693 | finish_lock_switch(rq, prev); | |
9a897c5a | 2694 | #ifdef CONFIG_SMP |
967fc046 | 2695 | if (post_schedule) |
9a897c5a SR |
2696 | current->sched_class->post_schedule(rq); |
2697 | #endif | |
e8fa1362 | 2698 | |
e107be36 | 2699 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2700 | if (mm) |
2701 | mmdrop(mm); | |
c394cc9f | 2702 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2703 | /* |
2704 | * Remove function-return probe instances associated with this | |
2705 | * task and put them back on the free list. | |
9761eea8 | 2706 | */ |
c6fd91f0 | 2707 | kprobe_flush_task(prev); |
1da177e4 | 2708 | put_task_struct(prev); |
c6fd91f0 | 2709 | } |
1da177e4 LT |
2710 | } |
2711 | ||
2712 | /** | |
2713 | * schedule_tail - first thing a freshly forked thread must call. | |
2714 | * @prev: the thread we just switched away from. | |
2715 | */ | |
36c8b586 | 2716 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2717 | __releases(rq->lock) |
2718 | { | |
70b97a7f IM |
2719 | struct rq *rq = this_rq(); |
2720 | ||
4866cde0 NP |
2721 | finish_task_switch(rq, prev); |
2722 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2723 | /* In this case, finish_task_switch does not reenable preemption */ | |
2724 | preempt_enable(); | |
2725 | #endif | |
1da177e4 | 2726 | if (current->set_child_tid) |
b488893a | 2727 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2728 | } |
2729 | ||
2730 | /* | |
2731 | * context_switch - switch to the new MM and the new | |
2732 | * thread's register state. | |
2733 | */ | |
dd41f596 | 2734 | static inline void |
70b97a7f | 2735 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2736 | struct task_struct *next) |
1da177e4 | 2737 | { |
dd41f596 | 2738 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2739 | |
e107be36 | 2740 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2741 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2742 | mm = next->mm; |
2743 | oldmm = prev->active_mm; | |
9226d125 ZA |
2744 | /* |
2745 | * For paravirt, this is coupled with an exit in switch_to to | |
2746 | * combine the page table reload and the switch backend into | |
2747 | * one hypercall. | |
2748 | */ | |
2749 | arch_enter_lazy_cpu_mode(); | |
2750 | ||
dd41f596 | 2751 | if (unlikely(!mm)) { |
1da177e4 LT |
2752 | next->active_mm = oldmm; |
2753 | atomic_inc(&oldmm->mm_count); | |
2754 | enter_lazy_tlb(oldmm, next); | |
2755 | } else | |
2756 | switch_mm(oldmm, mm, next); | |
2757 | ||
dd41f596 | 2758 | if (unlikely(!prev->mm)) { |
1da177e4 | 2759 | prev->active_mm = NULL; |
1da177e4 LT |
2760 | rq->prev_mm = oldmm; |
2761 | } | |
3a5f5e48 IM |
2762 | /* |
2763 | * Since the runqueue lock will be released by the next | |
2764 | * task (which is an invalid locking op but in the case | |
2765 | * of the scheduler it's an obvious special-case), so we | |
2766 | * do an early lockdep release here: | |
2767 | */ | |
2768 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2769 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2770 | #endif |
1da177e4 LT |
2771 | |
2772 | /* Here we just switch the register state and the stack. */ | |
2773 | switch_to(prev, next, prev); | |
2774 | ||
dd41f596 IM |
2775 | barrier(); |
2776 | /* | |
2777 | * this_rq must be evaluated again because prev may have moved | |
2778 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2779 | * frame will be invalid. | |
2780 | */ | |
2781 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2782 | } |
2783 | ||
2784 | /* | |
2785 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2786 | * | |
2787 | * externally visible scheduler statistics: current number of runnable | |
2788 | * threads, current number of uninterruptible-sleeping threads, total | |
2789 | * number of context switches performed since bootup. | |
2790 | */ | |
2791 | unsigned long nr_running(void) | |
2792 | { | |
2793 | unsigned long i, sum = 0; | |
2794 | ||
2795 | for_each_online_cpu(i) | |
2796 | sum += cpu_rq(i)->nr_running; | |
2797 | ||
2798 | return sum; | |
2799 | } | |
2800 | ||
2801 | unsigned long nr_uninterruptible(void) | |
2802 | { | |
2803 | unsigned long i, sum = 0; | |
2804 | ||
0a945022 | 2805 | for_each_possible_cpu(i) |
1da177e4 LT |
2806 | sum += cpu_rq(i)->nr_uninterruptible; |
2807 | ||
2808 | /* | |
2809 | * Since we read the counters lockless, it might be slightly | |
2810 | * inaccurate. Do not allow it to go below zero though: | |
2811 | */ | |
2812 | if (unlikely((long)sum < 0)) | |
2813 | sum = 0; | |
2814 | ||
2815 | return sum; | |
2816 | } | |
2817 | ||
2818 | unsigned long long nr_context_switches(void) | |
2819 | { | |
cc94abfc SR |
2820 | int i; |
2821 | unsigned long long sum = 0; | |
1da177e4 | 2822 | |
0a945022 | 2823 | for_each_possible_cpu(i) |
1da177e4 LT |
2824 | sum += cpu_rq(i)->nr_switches; |
2825 | ||
2826 | return sum; | |
2827 | } | |
2828 | ||
2829 | unsigned long nr_iowait(void) | |
2830 | { | |
2831 | unsigned long i, sum = 0; | |
2832 | ||
0a945022 | 2833 | for_each_possible_cpu(i) |
1da177e4 LT |
2834 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2835 | ||
2836 | return sum; | |
2837 | } | |
2838 | ||
db1b1fef JS |
2839 | unsigned long nr_active(void) |
2840 | { | |
2841 | unsigned long i, running = 0, uninterruptible = 0; | |
2842 | ||
2843 | for_each_online_cpu(i) { | |
2844 | running += cpu_rq(i)->nr_running; | |
2845 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2846 | } | |
2847 | ||
2848 | if (unlikely((long)uninterruptible < 0)) | |
2849 | uninterruptible = 0; | |
2850 | ||
2851 | return running + uninterruptible; | |
2852 | } | |
2853 | ||
48f24c4d | 2854 | /* |
dd41f596 IM |
2855 | * Update rq->cpu_load[] statistics. This function is usually called every |
2856 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2857 | */ |
dd41f596 | 2858 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2859 | { |
495eca49 | 2860 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2861 | int i, scale; |
2862 | ||
2863 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2864 | |
2865 | /* Update our load: */ | |
2866 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2867 | unsigned long old_load, new_load; | |
2868 | ||
2869 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2870 | ||
2871 | old_load = this_rq->cpu_load[i]; | |
2872 | new_load = this_load; | |
a25707f3 IM |
2873 | /* |
2874 | * Round up the averaging division if load is increasing. This | |
2875 | * prevents us from getting stuck on 9 if the load is 10, for | |
2876 | * example. | |
2877 | */ | |
2878 | if (new_load > old_load) | |
2879 | new_load += scale-1; | |
dd41f596 IM |
2880 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2881 | } | |
48f24c4d IM |
2882 | } |
2883 | ||
dd41f596 IM |
2884 | #ifdef CONFIG_SMP |
2885 | ||
1da177e4 LT |
2886 | /* |
2887 | * double_rq_lock - safely lock two runqueues | |
2888 | * | |
2889 | * Note this does not disable interrupts like task_rq_lock, | |
2890 | * you need to do so manually before calling. | |
2891 | */ | |
70b97a7f | 2892 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2893 | __acquires(rq1->lock) |
2894 | __acquires(rq2->lock) | |
2895 | { | |
054b9108 | 2896 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2897 | if (rq1 == rq2) { |
2898 | spin_lock(&rq1->lock); | |
2899 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2900 | } else { | |
c96d145e | 2901 | if (rq1 < rq2) { |
1da177e4 | 2902 | spin_lock(&rq1->lock); |
5e710e37 | 2903 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2904 | } else { |
2905 | spin_lock(&rq2->lock); | |
5e710e37 | 2906 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2907 | } |
2908 | } | |
6e82a3be IM |
2909 | update_rq_clock(rq1); |
2910 | update_rq_clock(rq2); | |
1da177e4 LT |
2911 | } |
2912 | ||
2913 | /* | |
2914 | * double_rq_unlock - safely unlock two runqueues | |
2915 | * | |
2916 | * Note this does not restore interrupts like task_rq_unlock, | |
2917 | * you need to do so manually after calling. | |
2918 | */ | |
70b97a7f | 2919 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2920 | __releases(rq1->lock) |
2921 | __releases(rq2->lock) | |
2922 | { | |
2923 | spin_unlock(&rq1->lock); | |
2924 | if (rq1 != rq2) | |
2925 | spin_unlock(&rq2->lock); | |
2926 | else | |
2927 | __release(rq2->lock); | |
2928 | } | |
2929 | ||
1da177e4 LT |
2930 | /* |
2931 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2932 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2933 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2934 | * the cpu_allowed mask is restored. |
2935 | */ | |
36c8b586 | 2936 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2937 | { |
70b97a7f | 2938 | struct migration_req req; |
1da177e4 | 2939 | unsigned long flags; |
70b97a7f | 2940 | struct rq *rq; |
1da177e4 LT |
2941 | |
2942 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2943 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2944 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2945 | goto out; |
2946 | ||
2947 | /* force the process onto the specified CPU */ | |
2948 | if (migrate_task(p, dest_cpu, &req)) { | |
2949 | /* Need to wait for migration thread (might exit: take ref). */ | |
2950 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2951 | |
1da177e4 LT |
2952 | get_task_struct(mt); |
2953 | task_rq_unlock(rq, &flags); | |
2954 | wake_up_process(mt); | |
2955 | put_task_struct(mt); | |
2956 | wait_for_completion(&req.done); | |
36c8b586 | 2957 | |
1da177e4 LT |
2958 | return; |
2959 | } | |
2960 | out: | |
2961 | task_rq_unlock(rq, &flags); | |
2962 | } | |
2963 | ||
2964 | /* | |
476d139c NP |
2965 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2966 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2967 | */ |
2968 | void sched_exec(void) | |
2969 | { | |
1da177e4 | 2970 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2971 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2972 | put_cpu(); |
476d139c NP |
2973 | if (new_cpu != this_cpu) |
2974 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2975 | } |
2976 | ||
2977 | /* | |
2978 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2979 | * Both runqueues must be locked. | |
2980 | */ | |
dd41f596 IM |
2981 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2982 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2983 | { |
2e1cb74a | 2984 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2985 | set_task_cpu(p, this_cpu); |
dd41f596 | 2986 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2987 | /* |
2988 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2989 | * to be always true for them. | |
2990 | */ | |
15afe09b | 2991 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2992 | } |
2993 | ||
2994 | /* | |
2995 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2996 | */ | |
858119e1 | 2997 | static |
70b97a7f | 2998 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2999 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3000 | int *all_pinned) |
1da177e4 | 3001 | { |
708dc512 | 3002 | int tsk_cache_hot = 0; |
1da177e4 LT |
3003 | /* |
3004 | * We do not migrate tasks that are: | |
3005 | * 1) running (obviously), or | |
3006 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3007 | * 3) are cache-hot on their current CPU. | |
3008 | */ | |
96f874e2 | 3009 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3010 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3011 | return 0; |
cc367732 | 3012 | } |
81026794 NP |
3013 | *all_pinned = 0; |
3014 | ||
cc367732 IM |
3015 | if (task_running(rq, p)) { |
3016 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3017 | return 0; |
cc367732 | 3018 | } |
1da177e4 | 3019 | |
da84d961 IM |
3020 | /* |
3021 | * Aggressive migration if: | |
3022 | * 1) task is cache cold, or | |
3023 | * 2) too many balance attempts have failed. | |
3024 | */ | |
3025 | ||
708dc512 LH |
3026 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3027 | if (!tsk_cache_hot || | |
3028 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3029 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3030 | if (tsk_cache_hot) { |
da84d961 | 3031 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3032 | schedstat_inc(p, se.nr_forced_migrations); |
3033 | } | |
da84d961 IM |
3034 | #endif |
3035 | return 1; | |
3036 | } | |
3037 | ||
708dc512 | 3038 | if (tsk_cache_hot) { |
cc367732 | 3039 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3040 | return 0; |
cc367732 | 3041 | } |
1da177e4 LT |
3042 | return 1; |
3043 | } | |
3044 | ||
e1d1484f PW |
3045 | static unsigned long |
3046 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3047 | unsigned long max_load_move, struct sched_domain *sd, | |
3048 | enum cpu_idle_type idle, int *all_pinned, | |
3049 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3050 | { |
051c6764 | 3051 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3052 | struct task_struct *p; |
3053 | long rem_load_move = max_load_move; | |
1da177e4 | 3054 | |
e1d1484f | 3055 | if (max_load_move == 0) |
1da177e4 LT |
3056 | goto out; |
3057 | ||
81026794 NP |
3058 | pinned = 1; |
3059 | ||
1da177e4 | 3060 | /* |
dd41f596 | 3061 | * Start the load-balancing iterator: |
1da177e4 | 3062 | */ |
dd41f596 IM |
3063 | p = iterator->start(iterator->arg); |
3064 | next: | |
b82d9fdd | 3065 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3066 | goto out; |
051c6764 PZ |
3067 | |
3068 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3069 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3070 | p = iterator->next(iterator->arg); |
3071 | goto next; | |
1da177e4 LT |
3072 | } |
3073 | ||
dd41f596 | 3074 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3075 | pulled++; |
dd41f596 | 3076 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3077 | |
7e96fa58 GH |
3078 | #ifdef CONFIG_PREEMPT |
3079 | /* | |
3080 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3081 | * will stop after the first task is pulled to minimize the critical | |
3082 | * section. | |
3083 | */ | |
3084 | if (idle == CPU_NEWLY_IDLE) | |
3085 | goto out; | |
3086 | #endif | |
3087 | ||
2dd73a4f | 3088 | /* |
b82d9fdd | 3089 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3090 | */ |
e1d1484f | 3091 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3092 | if (p->prio < *this_best_prio) |
3093 | *this_best_prio = p->prio; | |
dd41f596 IM |
3094 | p = iterator->next(iterator->arg); |
3095 | goto next; | |
1da177e4 LT |
3096 | } |
3097 | out: | |
3098 | /* | |
e1d1484f | 3099 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3100 | * so we can safely collect pull_task() stats here rather than |
3101 | * inside pull_task(). | |
3102 | */ | |
3103 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3104 | |
3105 | if (all_pinned) | |
3106 | *all_pinned = pinned; | |
e1d1484f PW |
3107 | |
3108 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3109 | } |
3110 | ||
dd41f596 | 3111 | /* |
43010659 PW |
3112 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3113 | * this_rq, as part of a balancing operation within domain "sd". | |
3114 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3115 | * |
3116 | * Called with both runqueues locked. | |
3117 | */ | |
3118 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3119 | unsigned long max_load_move, |
dd41f596 IM |
3120 | struct sched_domain *sd, enum cpu_idle_type idle, |
3121 | int *all_pinned) | |
3122 | { | |
5522d5d5 | 3123 | const struct sched_class *class = sched_class_highest; |
43010659 | 3124 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3125 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3126 | |
3127 | do { | |
43010659 PW |
3128 | total_load_moved += |
3129 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3130 | max_load_move - total_load_moved, |
a4ac01c3 | 3131 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3132 | class = class->next; |
c4acb2c0 | 3133 | |
7e96fa58 GH |
3134 | #ifdef CONFIG_PREEMPT |
3135 | /* | |
3136 | * NEWIDLE balancing is a source of latency, so preemptible | |
3137 | * kernels will stop after the first task is pulled to minimize | |
3138 | * the critical section. | |
3139 | */ | |
c4acb2c0 GH |
3140 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3141 | break; | |
7e96fa58 | 3142 | #endif |
43010659 | 3143 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3144 | |
43010659 PW |
3145 | return total_load_moved > 0; |
3146 | } | |
3147 | ||
e1d1484f PW |
3148 | static int |
3149 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3150 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3151 | struct rq_iterator *iterator) | |
3152 | { | |
3153 | struct task_struct *p = iterator->start(iterator->arg); | |
3154 | int pinned = 0; | |
3155 | ||
3156 | while (p) { | |
3157 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3158 | pull_task(busiest, p, this_rq, this_cpu); | |
3159 | /* | |
3160 | * Right now, this is only the second place pull_task() | |
3161 | * is called, so we can safely collect pull_task() | |
3162 | * stats here rather than inside pull_task(). | |
3163 | */ | |
3164 | schedstat_inc(sd, lb_gained[idle]); | |
3165 | ||
3166 | return 1; | |
3167 | } | |
3168 | p = iterator->next(iterator->arg); | |
3169 | } | |
3170 | ||
3171 | return 0; | |
3172 | } | |
3173 | ||
43010659 PW |
3174 | /* |
3175 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3176 | * part of active balancing operations within "domain". | |
3177 | * Returns 1 if successful and 0 otherwise. | |
3178 | * | |
3179 | * Called with both runqueues locked. | |
3180 | */ | |
3181 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3182 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3183 | { | |
5522d5d5 | 3184 | const struct sched_class *class; |
43010659 PW |
3185 | |
3186 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3187 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3188 | return 1; |
3189 | ||
3190 | return 0; | |
dd41f596 | 3191 | } |
67bb6c03 | 3192 | /********** Helpers for find_busiest_group ************************/ |
222d656d GS |
3193 | /** |
3194 | * sd_lb_stats - Structure to store the statistics of a sched_domain | |
3195 | * during load balancing. | |
3196 | */ | |
3197 | struct sd_lb_stats { | |
3198 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3199 | struct sched_group *this; /* Local group in this sd */ | |
3200 | unsigned long total_load; /* Total load of all groups in sd */ | |
3201 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3202 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3203 | ||
3204 | /** Statistics of this group */ | |
3205 | unsigned long this_load; | |
3206 | unsigned long this_load_per_task; | |
3207 | unsigned long this_nr_running; | |
3208 | ||
3209 | /* Statistics of the busiest group */ | |
3210 | unsigned long max_load; | |
3211 | unsigned long busiest_load_per_task; | |
3212 | unsigned long busiest_nr_running; | |
3213 | ||
3214 | int group_imb; /* Is there imbalance in this sd */ | |
3215 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3216 | int power_savings_balance; /* Is powersave balance needed for this sd */ | |
3217 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3218 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3219 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3220 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3221 | unsigned long min_nr_running; /* Nr running of group_min */ | |
3222 | #endif | |
3223 | }; | |
67bb6c03 | 3224 | |
381be78f GS |
3225 | /** |
3226 | * sg_lb_stats - stats of a sched_group required for load_balancing | |
3227 | */ | |
3228 | struct sg_lb_stats { | |
3229 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3230 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3231 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3232 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3233 | unsigned long group_capacity; | |
3234 | int group_imb; /* Is there an imbalance in the group ? */ | |
3235 | }; | |
3236 | ||
67bb6c03 GS |
3237 | /** |
3238 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3239 | * @group: The group whose first cpu is to be returned. | |
3240 | */ | |
3241 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3242 | { | |
3243 | return cpumask_first(sched_group_cpus(group)); | |
3244 | } | |
3245 | ||
3246 | /** | |
3247 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3248 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3249 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3250 | */ | |
3251 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3252 | enum cpu_idle_type idle) | |
3253 | { | |
3254 | int load_idx; | |
3255 | ||
3256 | switch (idle) { | |
3257 | case CPU_NOT_IDLE: | |
3258 | load_idx = sd->busy_idx; | |
3259 | break; | |
3260 | ||
3261 | case CPU_NEWLY_IDLE: | |
3262 | load_idx = sd->newidle_idx; | |
3263 | break; | |
3264 | default: | |
3265 | load_idx = sd->idle_idx; | |
3266 | break; | |
3267 | } | |
3268 | ||
3269 | return load_idx; | |
3270 | } | |
1f8c553d GS |
3271 | |
3272 | ||
c071df18 GS |
3273 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3274 | /** | |
3275 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3276 | * the given sched_domain, during load balancing. | |
3277 | * | |
3278 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3279 | * @sds: Variable containing the statistics for sd. | |
3280 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3281 | */ | |
3282 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3283 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3284 | { | |
3285 | /* | |
3286 | * Busy processors will not participate in power savings | |
3287 | * balance. | |
3288 | */ | |
3289 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3290 | sds->power_savings_balance = 0; | |
3291 | else { | |
3292 | sds->power_savings_balance = 1; | |
3293 | sds->min_nr_running = ULONG_MAX; | |
3294 | sds->leader_nr_running = 0; | |
3295 | } | |
3296 | } | |
3297 | ||
3298 | /** | |
3299 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3300 | * sched_domain while performing load balancing. | |
3301 | * | |
3302 | * @group: sched_group belonging to the sched_domain under consideration. | |
3303 | * @sds: Variable containing the statistics of the sched_domain | |
3304 | * @local_group: Does group contain the CPU for which we're performing | |
3305 | * load balancing ? | |
3306 | * @sgs: Variable containing the statistics of the group. | |
3307 | */ | |
3308 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3309 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3310 | { | |
3311 | ||
3312 | if (!sds->power_savings_balance) | |
3313 | return; | |
3314 | ||
3315 | /* | |
3316 | * If the local group is idle or completely loaded | |
3317 | * no need to do power savings balance at this domain | |
3318 | */ | |
3319 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3320 | !sds->this_nr_running)) | |
3321 | sds->power_savings_balance = 0; | |
3322 | ||
3323 | /* | |
3324 | * If a group is already running at full capacity or idle, | |
3325 | * don't include that group in power savings calculations | |
3326 | */ | |
3327 | if (!sds->power_savings_balance || | |
3328 | sgs->sum_nr_running >= sgs->group_capacity || | |
3329 | !sgs->sum_nr_running) | |
3330 | return; | |
3331 | ||
3332 | /* | |
3333 | * Calculate the group which has the least non-idle load. | |
3334 | * This is the group from where we need to pick up the load | |
3335 | * for saving power | |
3336 | */ | |
3337 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3338 | (sgs->sum_nr_running == sds->min_nr_running && | |
3339 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3340 | sds->group_min = group; | |
3341 | sds->min_nr_running = sgs->sum_nr_running; | |
3342 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3343 | sgs->sum_nr_running; | |
3344 | } | |
3345 | ||
3346 | /* | |
3347 | * Calculate the group which is almost near its | |
3348 | * capacity but still has some space to pick up some load | |
3349 | * from other group and save more power | |
3350 | */ | |
3351 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3352 | return; | |
3353 | ||
3354 | if (sgs->sum_nr_running > sds->leader_nr_running || | |
3355 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3356 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3357 | sds->group_leader = group; | |
3358 | sds->leader_nr_running = sgs->sum_nr_running; | |
3359 | } | |
3360 | } | |
3361 | ||
3362 | /** | |
3363 | * check_power_save_busiest_group - Check if we have potential to perform | |
3364 | * some power-savings balance. If yes, set the busiest group to be | |
3365 | * the least loaded group in the sched_domain, so that it's CPUs can | |
3366 | * be put to idle. | |
3367 | * | |
3368 | * @sds: Variable containing the statistics of the sched_domain | |
3369 | * under consideration. | |
3370 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3371 | * @imbalance: Variable to store the imbalance. | |
3372 | * | |
3373 | * Returns 1 if there is potential to perform power-savings balance. | |
3374 | * Else returns 0. | |
3375 | */ | |
3376 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3377 | int this_cpu, unsigned long *imbalance) | |
3378 | { | |
3379 | if (!sds->power_savings_balance) | |
3380 | return 0; | |
3381 | ||
3382 | if (sds->this != sds->group_leader || | |
3383 | sds->group_leader == sds->group_min) | |
3384 | return 0; | |
3385 | ||
3386 | *imbalance = sds->min_load_per_task; | |
3387 | sds->busiest = sds->group_min; | |
3388 | ||
3389 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { | |
3390 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3391 | group_first_cpu(sds->group_leader); | |
3392 | } | |
3393 | ||
3394 | return 1; | |
3395 | ||
3396 | } | |
3397 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3398 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3399 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3400 | { | |
3401 | return; | |
3402 | } | |
3403 | ||
3404 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3405 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3406 | { | |
3407 | return; | |
3408 | } | |
3409 | ||
3410 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3411 | int this_cpu, unsigned long *imbalance) | |
3412 | { | |
3413 | return 0; | |
3414 | } | |
3415 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3416 | ||
3417 | ||
1f8c553d GS |
3418 | /** |
3419 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3420 | * @group: sched_group whose statistics are to be updated. | |
3421 | * @this_cpu: Cpu for which load balance is currently performed. | |
3422 | * @idle: Idle status of this_cpu | |
3423 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3424 | * @sd_idle: Idle status of the sched_domain containing group. | |
3425 | * @local_group: Does group contain this_cpu. | |
3426 | * @cpus: Set of cpus considered for load balancing. | |
3427 | * @balance: Should we balance. | |
3428 | * @sgs: variable to hold the statistics for this group. | |
3429 | */ | |
3430 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3431 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3432 | int local_group, const struct cpumask *cpus, | |
3433 | int *balance, struct sg_lb_stats *sgs) | |
3434 | { | |
3435 | unsigned long load, max_cpu_load, min_cpu_load; | |
3436 | int i; | |
3437 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3438 | unsigned long sum_avg_load_per_task; | |
3439 | unsigned long avg_load_per_task; | |
3440 | ||
3441 | if (local_group) | |
3442 | balance_cpu = group_first_cpu(group); | |
3443 | ||
3444 | /* Tally up the load of all CPUs in the group */ | |
3445 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3446 | max_cpu_load = 0; | |
3447 | min_cpu_load = ~0UL; | |
3448 | ||
3449 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { | |
3450 | struct rq *rq = cpu_rq(i); | |
3451 | ||
3452 | if (*sd_idle && rq->nr_running) | |
3453 | *sd_idle = 0; | |
3454 | ||
3455 | /* Bias balancing toward cpus of our domain */ | |
3456 | if (local_group) { | |
3457 | if (idle_cpu(i) && !first_idle_cpu) { | |
3458 | first_idle_cpu = 1; | |
3459 | balance_cpu = i; | |
3460 | } | |
3461 | ||
3462 | load = target_load(i, load_idx); | |
3463 | } else { | |
3464 | load = source_load(i, load_idx); | |
3465 | if (load > max_cpu_load) | |
3466 | max_cpu_load = load; | |
3467 | if (min_cpu_load > load) | |
3468 | min_cpu_load = load; | |
3469 | } | |
3470 | ||
3471 | sgs->group_load += load; | |
3472 | sgs->sum_nr_running += rq->nr_running; | |
3473 | sgs->sum_weighted_load += weighted_cpuload(i); | |
3474 | ||
3475 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
3476 | } | |
3477 | ||
3478 | /* | |
3479 | * First idle cpu or the first cpu(busiest) in this sched group | |
3480 | * is eligible for doing load balancing at this and above | |
3481 | * domains. In the newly idle case, we will allow all the cpu's | |
3482 | * to do the newly idle load balance. | |
3483 | */ | |
3484 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3485 | balance_cpu != this_cpu && balance) { | |
3486 | *balance = 0; | |
3487 | return; | |
3488 | } | |
3489 | ||
3490 | /* Adjust by relative CPU power of the group */ | |
3491 | sgs->avg_load = sg_div_cpu_power(group, | |
3492 | sgs->group_load * SCHED_LOAD_SCALE); | |
3493 | ||
3494 | ||
3495 | /* | |
3496 | * Consider the group unbalanced when the imbalance is larger | |
3497 | * than the average weight of two tasks. | |
3498 | * | |
3499 | * APZ: with cgroup the avg task weight can vary wildly and | |
3500 | * might not be a suitable number - should we keep a | |
3501 | * normalized nr_running number somewhere that negates | |
3502 | * the hierarchy? | |
3503 | */ | |
3504 | avg_load_per_task = sg_div_cpu_power(group, | |
3505 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3506 | ||
3507 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3508 | sgs->group_imb = 1; | |
3509 | ||
3510 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3511 | ||
3512 | } | |
dd41f596 | 3513 | |
37abe198 GS |
3514 | /** |
3515 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3516 | * @sd: sched_domain whose statistics are to be updated. | |
3517 | * @this_cpu: Cpu for which load balance is currently performed. | |
3518 | * @idle: Idle status of this_cpu | |
3519 | * @sd_idle: Idle status of the sched_domain containing group. | |
3520 | * @cpus: Set of cpus considered for load balancing. | |
3521 | * @balance: Should we balance. | |
3522 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3523 | */ |
37abe198 GS |
3524 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3525 | enum cpu_idle_type idle, int *sd_idle, | |
3526 | const struct cpumask *cpus, int *balance, | |
3527 | struct sd_lb_stats *sds) | |
1da177e4 | 3528 | { |
222d656d | 3529 | struct sched_group *group = sd->groups; |
37abe198 | 3530 | struct sg_lb_stats sgs; |
222d656d GS |
3531 | int load_idx; |
3532 | ||
c071df18 | 3533 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3534 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3535 | |
3536 | do { | |
1da177e4 | 3537 | int local_group; |
1da177e4 | 3538 | |
758b2cdc RR |
3539 | local_group = cpumask_test_cpu(this_cpu, |
3540 | sched_group_cpus(group)); | |
381be78f | 3541 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3542 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3543 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3544 | |
37abe198 GS |
3545 | if (local_group && balance && !(*balance)) |
3546 | return; | |
783609c6 | 3547 | |
37abe198 GS |
3548 | sds->total_load += sgs.group_load; |
3549 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3550 | |
1da177e4 | 3551 | if (local_group) { |
37abe198 GS |
3552 | sds->this_load = sgs.avg_load; |
3553 | sds->this = group; | |
3554 | sds->this_nr_running = sgs.sum_nr_running; | |
3555 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3556 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3557 | (sgs.sum_nr_running > sgs.group_capacity || |
3558 | sgs.group_imb)) { | |
37abe198 GS |
3559 | sds->max_load = sgs.avg_load; |
3560 | sds->busiest = group; | |
3561 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3562 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3563 | sds->group_imb = sgs.group_imb; | |
1da177e4 | 3564 | } |
5c45bf27 | 3565 | |
c071df18 | 3566 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3567 | group = group->next; |
3568 | } while (group != sd->groups); | |
3569 | ||
37abe198 | 3570 | } |
2e6f44ae GS |
3571 | |
3572 | /** | |
3573 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3574 | * amongst the groups of a sched_domain, during |
3575 | * load balancing. | |
2e6f44ae GS |
3576 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3577 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3578 | * @imbalance: Variable to store the imbalance. | |
3579 | */ | |
3580 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3581 | int this_cpu, unsigned long *imbalance) | |
3582 | { | |
3583 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3584 | unsigned int imbn = 2; | |
3585 | ||
3586 | if (sds->this_nr_running) { | |
3587 | sds->this_load_per_task /= sds->this_nr_running; | |
3588 | if (sds->busiest_load_per_task > | |
3589 | sds->this_load_per_task) | |
3590 | imbn = 1; | |
3591 | } else | |
3592 | sds->this_load_per_task = | |
3593 | cpu_avg_load_per_task(this_cpu); | |
3594 | ||
3595 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= | |
3596 | sds->busiest_load_per_task * imbn) { | |
3597 | *imbalance = sds->busiest_load_per_task; | |
3598 | return; | |
3599 | } | |
3600 | ||
3601 | /* | |
3602 | * OK, we don't have enough imbalance to justify moving tasks, | |
3603 | * however we may be able to increase total CPU power used by | |
3604 | * moving them. | |
3605 | */ | |
3606 | ||
3607 | pwr_now += sds->busiest->__cpu_power * | |
3608 | min(sds->busiest_load_per_task, sds->max_load); | |
3609 | pwr_now += sds->this->__cpu_power * | |
3610 | min(sds->this_load_per_task, sds->this_load); | |
3611 | pwr_now /= SCHED_LOAD_SCALE; | |
3612 | ||
3613 | /* Amount of load we'd subtract */ | |
3614 | tmp = sg_div_cpu_power(sds->busiest, | |
3615 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3616 | if (sds->max_load > tmp) | |
3617 | pwr_move += sds->busiest->__cpu_power * | |
3618 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3619 | ||
3620 | /* Amount of load we'd add */ | |
3621 | if (sds->max_load * sds->busiest->__cpu_power < | |
3622 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3623 | tmp = sg_div_cpu_power(sds->this, | |
3624 | sds->max_load * sds->busiest->__cpu_power); | |
3625 | else | |
3626 | tmp = sg_div_cpu_power(sds->this, | |
3627 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3628 | pwr_move += sds->this->__cpu_power * | |
3629 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3630 | pwr_move /= SCHED_LOAD_SCALE; | |
3631 | ||
3632 | /* Move if we gain throughput */ | |
3633 | if (pwr_move > pwr_now) | |
3634 | *imbalance = sds->busiest_load_per_task; | |
3635 | } | |
dbc523a3 GS |
3636 | |
3637 | /** | |
3638 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3639 | * groups of a given sched_domain during load balance. | |
3640 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3641 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3642 | * @imbalance: The variable to store the imbalance. | |
3643 | */ | |
3644 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3645 | unsigned long *imbalance) | |
3646 | { | |
3647 | unsigned long max_pull; | |
3648 | /* | |
3649 | * In the presence of smp nice balancing, certain scenarios can have | |
3650 | * max load less than avg load(as we skip the groups at or below | |
3651 | * its cpu_power, while calculating max_load..) | |
3652 | */ | |
3653 | if (sds->max_load < sds->avg_load) { | |
3654 | *imbalance = 0; | |
3655 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
3656 | } | |
3657 | ||
3658 | /* Don't want to pull so many tasks that a group would go idle */ | |
3659 | max_pull = min(sds->max_load - sds->avg_load, | |
3660 | sds->max_load - sds->busiest_load_per_task); | |
3661 | ||
3662 | /* How much load to actually move to equalise the imbalance */ | |
3663 | *imbalance = min(max_pull * sds->busiest->__cpu_power, | |
3664 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
3665 | / SCHED_LOAD_SCALE; | |
3666 | ||
3667 | /* | |
3668 | * if *imbalance is less than the average load per runnable task | |
3669 | * there is no gaurantee that any tasks will be moved so we'll have | |
3670 | * a think about bumping its value to force at least one task to be | |
3671 | * moved | |
3672 | */ | |
3673 | if (*imbalance < sds->busiest_load_per_task) | |
3674 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
3675 | ||
3676 | } | |
37abe198 GS |
3677 | /******* find_busiest_group() helpers end here *********************/ |
3678 | ||
3679 | /* | |
3680 | * find_busiest_group finds and returns the busiest CPU group within the | |
3681 | * domain. It calculates and returns the amount of weighted load which | |
3682 | * should be moved to restore balance via the imbalance parameter. | |
3683 | */ | |
3684 | static struct sched_group * | |
3685 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3686 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3687 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3688 | { | |
3689 | struct sd_lb_stats sds; | |
37abe198 GS |
3690 | |
3691 | memset(&sds, 0, sizeof(sds)); | |
3692 | ||
3693 | /* | |
3694 | * Compute the various statistics relavent for load balancing at | |
3695 | * this level. | |
3696 | */ | |
3697 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3698 | balance, &sds); | |
3699 | ||
3700 | if (balance && !(*balance)) | |
3701 | goto ret; | |
3702 | ||
222d656d GS |
3703 | if (!sds.busiest || sds.this_load >= sds.max_load |
3704 | || sds.busiest_nr_running == 0) | |
1da177e4 LT |
3705 | goto out_balanced; |
3706 | ||
222d656d | 3707 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3708 | |
222d656d GS |
3709 | if (sds.this_load >= sds.avg_load || |
3710 | 100*sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3711 | goto out_balanced; |
3712 | ||
222d656d GS |
3713 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3714 | if (sds.group_imb) | |
3715 | sds.busiest_load_per_task = | |
3716 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3717 | |
1da177e4 LT |
3718 | /* |
3719 | * We're trying to get all the cpus to the average_load, so we don't | |
3720 | * want to push ourselves above the average load, nor do we wish to | |
3721 | * reduce the max loaded cpu below the average load, as either of these | |
3722 | * actions would just result in more rebalancing later, and ping-pong | |
3723 | * tasks around. Thus we look for the minimum possible imbalance. | |
3724 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3725 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3726 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3727 | * appear as very large values with unsigned longs. |
3728 | */ | |
222d656d | 3729 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3730 | goto out_balanced; |
3731 | ||
dbc523a3 GS |
3732 | /* Looks like there is an imbalance. Compute it */ |
3733 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3734 | return sds.busiest; |
1da177e4 LT |
3735 | |
3736 | out_balanced: | |
c071df18 GS |
3737 | /* |
3738 | * There is no obvious imbalance. But check if we can do some balancing | |
3739 | * to save power. | |
3740 | */ | |
3741 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3742 | return sds.busiest; | |
783609c6 | 3743 | ret: |
1da177e4 LT |
3744 | *imbalance = 0; |
3745 | return NULL; | |
3746 | } | |
3747 | ||
3748 | /* | |
3749 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3750 | */ | |
70b97a7f | 3751 | static struct rq * |
d15bcfdb | 3752 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3753 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3754 | { |
70b97a7f | 3755 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3756 | unsigned long max_load = 0; |
1da177e4 LT |
3757 | int i; |
3758 | ||
758b2cdc | 3759 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3760 | unsigned long wl; |
0a2966b4 | 3761 | |
96f874e2 | 3762 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3763 | continue; |
3764 | ||
48f24c4d | 3765 | rq = cpu_rq(i); |
dd41f596 | 3766 | wl = weighted_cpuload(i); |
2dd73a4f | 3767 | |
dd41f596 | 3768 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3769 | continue; |
1da177e4 | 3770 | |
dd41f596 IM |
3771 | if (wl > max_load) { |
3772 | max_load = wl; | |
48f24c4d | 3773 | busiest = rq; |
1da177e4 LT |
3774 | } |
3775 | } | |
3776 | ||
3777 | return busiest; | |
3778 | } | |
3779 | ||
77391d71 NP |
3780 | /* |
3781 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3782 | * so long as it is large enough. | |
3783 | */ | |
3784 | #define MAX_PINNED_INTERVAL 512 | |
3785 | ||
1da177e4 LT |
3786 | /* |
3787 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3788 | * tasks if there is an imbalance. | |
1da177e4 | 3789 | */ |
70b97a7f | 3790 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3791 | struct sched_domain *sd, enum cpu_idle_type idle, |
96f874e2 | 3792 | int *balance, struct cpumask *cpus) |
1da177e4 | 3793 | { |
43010659 | 3794 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3795 | struct sched_group *group; |
1da177e4 | 3796 | unsigned long imbalance; |
70b97a7f | 3797 | struct rq *busiest; |
fe2eea3f | 3798 | unsigned long flags; |
5969fe06 | 3799 | |
96f874e2 | 3800 | cpumask_setall(cpus); |
7c16ec58 | 3801 | |
89c4710e SS |
3802 | /* |
3803 | * When power savings policy is enabled for the parent domain, idle | |
3804 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3805 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3806 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3807 | */ |
d15bcfdb | 3808 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3809 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3810 | sd_idle = 1; |
1da177e4 | 3811 | |
2d72376b | 3812 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3813 | |
0a2966b4 | 3814 | redo: |
c8cba857 | 3815 | update_shares(sd); |
0a2966b4 | 3816 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3817 | cpus, balance); |
783609c6 | 3818 | |
06066714 | 3819 | if (*balance == 0) |
783609c6 | 3820 | goto out_balanced; |
783609c6 | 3821 | |
1da177e4 LT |
3822 | if (!group) { |
3823 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3824 | goto out_balanced; | |
3825 | } | |
3826 | ||
7c16ec58 | 3827 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3828 | if (!busiest) { |
3829 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3830 | goto out_balanced; | |
3831 | } | |
3832 | ||
db935dbd | 3833 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3834 | |
3835 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3836 | ||
43010659 | 3837 | ld_moved = 0; |
1da177e4 LT |
3838 | if (busiest->nr_running > 1) { |
3839 | /* | |
3840 | * Attempt to move tasks. If find_busiest_group has found | |
3841 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3842 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3843 | * correctly treated as an imbalance. |
3844 | */ | |
fe2eea3f | 3845 | local_irq_save(flags); |
e17224bf | 3846 | double_rq_lock(this_rq, busiest); |
43010659 | 3847 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3848 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3849 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3850 | local_irq_restore(flags); |
81026794 | 3851 | |
46cb4b7c SS |
3852 | /* |
3853 | * some other cpu did the load balance for us. | |
3854 | */ | |
43010659 | 3855 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3856 | resched_cpu(this_cpu); |
3857 | ||
81026794 | 3858 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3859 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3860 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3861 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3862 | goto redo; |
81026794 | 3863 | goto out_balanced; |
0a2966b4 | 3864 | } |
1da177e4 | 3865 | } |
81026794 | 3866 | |
43010659 | 3867 | if (!ld_moved) { |
1da177e4 LT |
3868 | schedstat_inc(sd, lb_failed[idle]); |
3869 | sd->nr_balance_failed++; | |
3870 | ||
3871 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3872 | |
fe2eea3f | 3873 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3874 | |
3875 | /* don't kick the migration_thread, if the curr | |
3876 | * task on busiest cpu can't be moved to this_cpu | |
3877 | */ | |
96f874e2 RR |
3878 | if (!cpumask_test_cpu(this_cpu, |
3879 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3880 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3881 | all_pinned = 1; |
3882 | goto out_one_pinned; | |
3883 | } | |
3884 | ||
1da177e4 LT |
3885 | if (!busiest->active_balance) { |
3886 | busiest->active_balance = 1; | |
3887 | busiest->push_cpu = this_cpu; | |
81026794 | 3888 | active_balance = 1; |
1da177e4 | 3889 | } |
fe2eea3f | 3890 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3891 | if (active_balance) |
1da177e4 LT |
3892 | wake_up_process(busiest->migration_thread); |
3893 | ||
3894 | /* | |
3895 | * We've kicked active balancing, reset the failure | |
3896 | * counter. | |
3897 | */ | |
39507451 | 3898 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3899 | } |
81026794 | 3900 | } else |
1da177e4 LT |
3901 | sd->nr_balance_failed = 0; |
3902 | ||
81026794 | 3903 | if (likely(!active_balance)) { |
1da177e4 LT |
3904 | /* We were unbalanced, so reset the balancing interval */ |
3905 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3906 | } else { |
3907 | /* | |
3908 | * If we've begun active balancing, start to back off. This | |
3909 | * case may not be covered by the all_pinned logic if there | |
3910 | * is only 1 task on the busy runqueue (because we don't call | |
3911 | * move_tasks). | |
3912 | */ | |
3913 | if (sd->balance_interval < sd->max_interval) | |
3914 | sd->balance_interval *= 2; | |
1da177e4 LT |
3915 | } |
3916 | ||
43010659 | 3917 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3918 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3919 | ld_moved = -1; |
3920 | ||
3921 | goto out; | |
1da177e4 LT |
3922 | |
3923 | out_balanced: | |
1da177e4 LT |
3924 | schedstat_inc(sd, lb_balanced[idle]); |
3925 | ||
16cfb1c0 | 3926 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3927 | |
3928 | out_one_pinned: | |
1da177e4 | 3929 | /* tune up the balancing interval */ |
77391d71 NP |
3930 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3931 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3932 | sd->balance_interval *= 2; |
3933 | ||
48f24c4d | 3934 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3935 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3936 | ld_moved = -1; |
3937 | else | |
3938 | ld_moved = 0; | |
3939 | out: | |
c8cba857 PZ |
3940 | if (ld_moved) |
3941 | update_shares(sd); | |
c09595f6 | 3942 | return ld_moved; |
1da177e4 LT |
3943 | } |
3944 | ||
3945 | /* | |
3946 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3947 | * tasks if there is an imbalance. | |
3948 | * | |
d15bcfdb | 3949 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3950 | * this_rq is locked. |
3951 | */ | |
48f24c4d | 3952 | static int |
7c16ec58 | 3953 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
96f874e2 | 3954 | struct cpumask *cpus) |
1da177e4 LT |
3955 | { |
3956 | struct sched_group *group; | |
70b97a7f | 3957 | struct rq *busiest = NULL; |
1da177e4 | 3958 | unsigned long imbalance; |
43010659 | 3959 | int ld_moved = 0; |
5969fe06 | 3960 | int sd_idle = 0; |
969bb4e4 | 3961 | int all_pinned = 0; |
7c16ec58 | 3962 | |
96f874e2 | 3963 | cpumask_setall(cpus); |
5969fe06 | 3964 | |
89c4710e SS |
3965 | /* |
3966 | * When power savings policy is enabled for the parent domain, idle | |
3967 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3968 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3969 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3970 | */ |
3971 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3972 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3973 | sd_idle = 1; |
1da177e4 | 3974 | |
2d72376b | 3975 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3976 | redo: |
3e5459b4 | 3977 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3978 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3979 | &sd_idle, cpus, NULL); |
1da177e4 | 3980 | if (!group) { |
d15bcfdb | 3981 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3982 | goto out_balanced; |
1da177e4 LT |
3983 | } |
3984 | ||
7c16ec58 | 3985 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3986 | if (!busiest) { |
d15bcfdb | 3987 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3988 | goto out_balanced; |
1da177e4 LT |
3989 | } |
3990 | ||
db935dbd NP |
3991 | BUG_ON(busiest == this_rq); |
3992 | ||
d15bcfdb | 3993 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3994 | |
43010659 | 3995 | ld_moved = 0; |
d6d5cfaf NP |
3996 | if (busiest->nr_running > 1) { |
3997 | /* Attempt to move tasks */ | |
3998 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3999 | /* this_rq->clock is already updated */ |
4000 | update_rq_clock(busiest); | |
43010659 | 4001 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4002 | imbalance, sd, CPU_NEWLY_IDLE, |
4003 | &all_pinned); | |
1b12bbc7 | 4004 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4005 | |
969bb4e4 | 4006 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4007 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4008 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4009 | goto redo; |
4010 | } | |
d6d5cfaf NP |
4011 | } |
4012 | ||
43010659 | 4013 | if (!ld_moved) { |
36dffab6 | 4014 | int active_balance = 0; |
ad273b32 | 4015 | |
d15bcfdb | 4016 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4017 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4018 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4019 | return -1; |
ad273b32 VS |
4020 | |
4021 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4022 | return -1; | |
4023 | ||
4024 | if (sd->nr_balance_failed++ < 2) | |
4025 | return -1; | |
4026 | ||
4027 | /* | |
4028 | * The only task running in a non-idle cpu can be moved to this | |
4029 | * cpu in an attempt to completely freeup the other CPU | |
4030 | * package. The same method used to move task in load_balance() | |
4031 | * have been extended for load_balance_newidle() to speedup | |
4032 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4033 | * | |
4034 | * The package power saving logic comes from | |
4035 | * find_busiest_group(). If there are no imbalance, then | |
4036 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4037 | * f_b_g() will select a group from which a running task may be | |
4038 | * pulled to this cpu in order to make the other package idle. | |
4039 | * If there is no opportunity to make a package idle and if | |
4040 | * there are no imbalance, then f_b_g() will return NULL and no | |
4041 | * action will be taken in load_balance_newidle(). | |
4042 | * | |
4043 | * Under normal task pull operation due to imbalance, there | |
4044 | * will be more than one task in the source run queue and | |
4045 | * move_tasks() will succeed. ld_moved will be true and this | |
4046 | * active balance code will not be triggered. | |
4047 | */ | |
4048 | ||
4049 | /* Lock busiest in correct order while this_rq is held */ | |
4050 | double_lock_balance(this_rq, busiest); | |
4051 | ||
4052 | /* | |
4053 | * don't kick the migration_thread, if the curr | |
4054 | * task on busiest cpu can't be moved to this_cpu | |
4055 | */ | |
6ca09dfc | 4056 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4057 | double_unlock_balance(this_rq, busiest); |
4058 | all_pinned = 1; | |
4059 | return ld_moved; | |
4060 | } | |
4061 | ||
4062 | if (!busiest->active_balance) { | |
4063 | busiest->active_balance = 1; | |
4064 | busiest->push_cpu = this_cpu; | |
4065 | active_balance = 1; | |
4066 | } | |
4067 | ||
4068 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4069 | /* |
4070 | * Should not call ttwu while holding a rq->lock | |
4071 | */ | |
4072 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4073 | if (active_balance) |
4074 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4075 | spin_lock(&this_rq->lock); |
ad273b32 | 4076 | |
5969fe06 | 4077 | } else |
16cfb1c0 | 4078 | sd->nr_balance_failed = 0; |
1da177e4 | 4079 | |
3e5459b4 | 4080 | update_shares_locked(this_rq, sd); |
43010659 | 4081 | return ld_moved; |
16cfb1c0 NP |
4082 | |
4083 | out_balanced: | |
d15bcfdb | 4084 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4085 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4086 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4087 | return -1; |
16cfb1c0 | 4088 | sd->nr_balance_failed = 0; |
48f24c4d | 4089 | |
16cfb1c0 | 4090 | return 0; |
1da177e4 LT |
4091 | } |
4092 | ||
4093 | /* | |
4094 | * idle_balance is called by schedule() if this_cpu is about to become | |
4095 | * idle. Attempts to pull tasks from other CPUs. | |
4096 | */ | |
70b97a7f | 4097 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4098 | { |
4099 | struct sched_domain *sd; | |
efbe027e | 4100 | int pulled_task = 0; |
dd41f596 | 4101 | unsigned long next_balance = jiffies + HZ; |
4d2732c6 RR |
4102 | cpumask_var_t tmpmask; |
4103 | ||
4104 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | |
4105 | return; | |
1da177e4 LT |
4106 | |
4107 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4108 | unsigned long interval; |
4109 | ||
4110 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4111 | continue; | |
4112 | ||
4113 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4114 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4115 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4d2732c6 | 4116 | sd, tmpmask); |
92c4ca5c CL |
4117 | |
4118 | interval = msecs_to_jiffies(sd->balance_interval); | |
4119 | if (time_after(next_balance, sd->last_balance + interval)) | |
4120 | next_balance = sd->last_balance + interval; | |
4121 | if (pulled_task) | |
4122 | break; | |
1da177e4 | 4123 | } |
dd41f596 | 4124 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4125 | /* |
4126 | * We are going idle. next_balance may be set based on | |
4127 | * a busy processor. So reset next_balance. | |
4128 | */ | |
4129 | this_rq->next_balance = next_balance; | |
dd41f596 | 4130 | } |
4d2732c6 | 4131 | free_cpumask_var(tmpmask); |
1da177e4 LT |
4132 | } |
4133 | ||
4134 | /* | |
4135 | * active_load_balance is run by migration threads. It pushes running tasks | |
4136 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4137 | * running on each physical CPU where possible, and avoids physical / | |
4138 | * logical imbalances. | |
4139 | * | |
4140 | * Called with busiest_rq locked. | |
4141 | */ | |
70b97a7f | 4142 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4143 | { |
39507451 | 4144 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4145 | struct sched_domain *sd; |
4146 | struct rq *target_rq; | |
39507451 | 4147 | |
48f24c4d | 4148 | /* Is there any task to move? */ |
39507451 | 4149 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4150 | return; |
4151 | ||
4152 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4153 | |
4154 | /* | |
39507451 | 4155 | * This condition is "impossible", if it occurs |
41a2d6cf | 4156 | * we need to fix it. Originally reported by |
39507451 | 4157 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4158 | */ |
39507451 | 4159 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4160 | |
39507451 NP |
4161 | /* move a task from busiest_rq to target_rq */ |
4162 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4163 | update_rq_clock(busiest_rq); |
4164 | update_rq_clock(target_rq); | |
39507451 NP |
4165 | |
4166 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4167 | for_each_domain(target_cpu, sd) { |
39507451 | 4168 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4169 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4170 | break; |
c96d145e | 4171 | } |
39507451 | 4172 | |
48f24c4d | 4173 | if (likely(sd)) { |
2d72376b | 4174 | schedstat_inc(sd, alb_count); |
39507451 | 4175 | |
43010659 PW |
4176 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4177 | sd, CPU_IDLE)) | |
48f24c4d IM |
4178 | schedstat_inc(sd, alb_pushed); |
4179 | else | |
4180 | schedstat_inc(sd, alb_failed); | |
4181 | } | |
1b12bbc7 | 4182 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4183 | } |
4184 | ||
46cb4b7c SS |
4185 | #ifdef CONFIG_NO_HZ |
4186 | static struct { | |
4187 | atomic_t load_balancer; | |
7d1e6a9b | 4188 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
4189 | } nohz ____cacheline_aligned = { |
4190 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4191 | }; |
4192 | ||
7835b98b | 4193 | /* |
46cb4b7c SS |
4194 | * This routine will try to nominate the ilb (idle load balancing) |
4195 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4196 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4197 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4198 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4199 | * arrives... | |
4200 | * | |
4201 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4202 | * for idle load balancing. ilb owner will still be part of | |
4203 | * nohz.cpu_mask.. | |
7835b98b | 4204 | * |
46cb4b7c SS |
4205 | * While stopping the tick, this cpu will become the ilb owner if there |
4206 | * is no other owner. And will be the owner till that cpu becomes busy | |
4207 | * or if all cpus in the system stop their ticks at which point | |
4208 | * there is no need for ilb owner. | |
4209 | * | |
4210 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4211 | * next busy scheduler_tick() | |
4212 | */ | |
4213 | int select_nohz_load_balancer(int stop_tick) | |
4214 | { | |
4215 | int cpu = smp_processor_id(); | |
4216 | ||
4217 | if (stop_tick) { | |
46cb4b7c SS |
4218 | cpu_rq(cpu)->in_nohz_recently = 1; |
4219 | ||
483b4ee6 SS |
4220 | if (!cpu_active(cpu)) { |
4221 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4222 | return 0; | |
4223 | ||
4224 | /* | |
4225 | * If we are going offline and still the leader, | |
4226 | * give up! | |
4227 | */ | |
46cb4b7c SS |
4228 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4229 | BUG(); | |
483b4ee6 | 4230 | |
46cb4b7c SS |
4231 | return 0; |
4232 | } | |
4233 | ||
483b4ee6 SS |
4234 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4235 | ||
46cb4b7c | 4236 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4237 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4238 | if (atomic_read(&nohz.load_balancer) == cpu) |
4239 | atomic_set(&nohz.load_balancer, -1); | |
4240 | return 0; | |
4241 | } | |
4242 | ||
4243 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4244 | /* make me the ilb owner */ | |
4245 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4246 | return 1; | |
4247 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
4248 | return 1; | |
4249 | } else { | |
7d1e6a9b | 4250 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4251 | return 0; |
4252 | ||
7d1e6a9b | 4253 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4254 | |
4255 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4256 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4257 | BUG(); | |
4258 | } | |
4259 | return 0; | |
4260 | } | |
4261 | #endif | |
4262 | ||
4263 | static DEFINE_SPINLOCK(balancing); | |
4264 | ||
4265 | /* | |
7835b98b CL |
4266 | * It checks each scheduling domain to see if it is due to be balanced, |
4267 | * and initiates a balancing operation if so. | |
4268 | * | |
4269 | * Balancing parameters are set up in arch_init_sched_domains. | |
4270 | */ | |
a9957449 | 4271 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4272 | { |
46cb4b7c SS |
4273 | int balance = 1; |
4274 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4275 | unsigned long interval; |
4276 | struct sched_domain *sd; | |
46cb4b7c | 4277 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4278 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4279 | int update_next_balance = 0; |
d07355f5 | 4280 | int need_serialize; |
a0e90245 RR |
4281 | cpumask_var_t tmp; |
4282 | ||
4283 | /* Fails alloc? Rebalancing probably not a priority right now. */ | |
4284 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | |
4285 | return; | |
1da177e4 | 4286 | |
46cb4b7c | 4287 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4288 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4289 | continue; | |
4290 | ||
4291 | interval = sd->balance_interval; | |
d15bcfdb | 4292 | if (idle != CPU_IDLE) |
1da177e4 LT |
4293 | interval *= sd->busy_factor; |
4294 | ||
4295 | /* scale ms to jiffies */ | |
4296 | interval = msecs_to_jiffies(interval); | |
4297 | if (unlikely(!interval)) | |
4298 | interval = 1; | |
dd41f596 IM |
4299 | if (interval > HZ*NR_CPUS/10) |
4300 | interval = HZ*NR_CPUS/10; | |
4301 | ||
d07355f5 | 4302 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4303 | |
d07355f5 | 4304 | if (need_serialize) { |
08c183f3 CL |
4305 | if (!spin_trylock(&balancing)) |
4306 | goto out; | |
4307 | } | |
4308 | ||
c9819f45 | 4309 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
a0e90245 | 4310 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { |
fa3b6ddc SS |
4311 | /* |
4312 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4313 | * longer idle, or one of our SMT siblings is |
4314 | * not idle. | |
4315 | */ | |
d15bcfdb | 4316 | idle = CPU_NOT_IDLE; |
1da177e4 | 4317 | } |
1bd77f2d | 4318 | sd->last_balance = jiffies; |
1da177e4 | 4319 | } |
d07355f5 | 4320 | if (need_serialize) |
08c183f3 CL |
4321 | spin_unlock(&balancing); |
4322 | out: | |
f549da84 | 4323 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4324 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4325 | update_next_balance = 1; |
4326 | } | |
783609c6 SS |
4327 | |
4328 | /* | |
4329 | * Stop the load balance at this level. There is another | |
4330 | * CPU in our sched group which is doing load balancing more | |
4331 | * actively. | |
4332 | */ | |
4333 | if (!balance) | |
4334 | break; | |
1da177e4 | 4335 | } |
f549da84 SS |
4336 | |
4337 | /* | |
4338 | * next_balance will be updated only when there is a need. | |
4339 | * When the cpu is attached to null domain for ex, it will not be | |
4340 | * updated. | |
4341 | */ | |
4342 | if (likely(update_next_balance)) | |
4343 | rq->next_balance = next_balance; | |
a0e90245 RR |
4344 | |
4345 | free_cpumask_var(tmp); | |
46cb4b7c SS |
4346 | } |
4347 | ||
4348 | /* | |
4349 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4350 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4351 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4352 | */ | |
4353 | static void run_rebalance_domains(struct softirq_action *h) | |
4354 | { | |
dd41f596 IM |
4355 | int this_cpu = smp_processor_id(); |
4356 | struct rq *this_rq = cpu_rq(this_cpu); | |
4357 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4358 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4359 | |
dd41f596 | 4360 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4361 | |
4362 | #ifdef CONFIG_NO_HZ | |
4363 | /* | |
4364 | * If this cpu is the owner for idle load balancing, then do the | |
4365 | * balancing on behalf of the other idle cpus whose ticks are | |
4366 | * stopped. | |
4367 | */ | |
dd41f596 IM |
4368 | if (this_rq->idle_at_tick && |
4369 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4370 | struct rq *rq; |
4371 | int balance_cpu; | |
4372 | ||
7d1e6a9b RR |
4373 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4374 | if (balance_cpu == this_cpu) | |
4375 | continue; | |
4376 | ||
46cb4b7c SS |
4377 | /* |
4378 | * If this cpu gets work to do, stop the load balancing | |
4379 | * work being done for other cpus. Next load | |
4380 | * balancing owner will pick it up. | |
4381 | */ | |
4382 | if (need_resched()) | |
4383 | break; | |
4384 | ||
de0cf899 | 4385 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4386 | |
4387 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4388 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4389 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4390 | } |
4391 | } | |
4392 | #endif | |
4393 | } | |
4394 | ||
8a0be9ef FW |
4395 | static inline int on_null_domain(int cpu) |
4396 | { | |
4397 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4398 | } | |
4399 | ||
46cb4b7c SS |
4400 | /* |
4401 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4402 | * | |
4403 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4404 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4405 | * if the whole system is idle. | |
4406 | */ | |
dd41f596 | 4407 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4408 | { |
46cb4b7c SS |
4409 | #ifdef CONFIG_NO_HZ |
4410 | /* | |
4411 | * If we were in the nohz mode recently and busy at the current | |
4412 | * scheduler tick, then check if we need to nominate new idle | |
4413 | * load balancer. | |
4414 | */ | |
4415 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4416 | rq->in_nohz_recently = 0; | |
4417 | ||
4418 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4419 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4420 | atomic_set(&nohz.load_balancer, -1); |
4421 | } | |
4422 | ||
4423 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4424 | /* | |
4425 | * simple selection for now: Nominate the | |
4426 | * first cpu in the nohz list to be the next | |
4427 | * ilb owner. | |
4428 | * | |
4429 | * TBD: Traverse the sched domains and nominate | |
4430 | * the nearest cpu in the nohz.cpu_mask. | |
4431 | */ | |
7d1e6a9b | 4432 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4433 | |
434d53b0 | 4434 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4435 | resched_cpu(ilb); |
4436 | } | |
4437 | } | |
4438 | ||
4439 | /* | |
4440 | * If this cpu is idle and doing idle load balancing for all the | |
4441 | * cpus with ticks stopped, is it time for that to stop? | |
4442 | */ | |
4443 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4444 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4445 | resched_cpu(cpu); |
4446 | return; | |
4447 | } | |
4448 | ||
4449 | /* | |
4450 | * If this cpu is idle and the idle load balancing is done by | |
4451 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4452 | */ | |
4453 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4454 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4455 | return; |
4456 | #endif | |
8a0be9ef FW |
4457 | /* Don't need to rebalance while attached to NULL domain */ |
4458 | if (time_after_eq(jiffies, rq->next_balance) && | |
4459 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4460 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4461 | } |
dd41f596 IM |
4462 | |
4463 | #else /* CONFIG_SMP */ | |
4464 | ||
1da177e4 LT |
4465 | /* |
4466 | * on UP we do not need to balance between CPUs: | |
4467 | */ | |
70b97a7f | 4468 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4469 | { |
4470 | } | |
dd41f596 | 4471 | |
1da177e4 LT |
4472 | #endif |
4473 | ||
1da177e4 LT |
4474 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4475 | ||
4476 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4477 | ||
4478 | /* | |
f06febc9 FM |
4479 | * Return any ns on the sched_clock that have not yet been banked in |
4480 | * @p in case that task is currently running. | |
1da177e4 | 4481 | */ |
bb34d92f | 4482 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4483 | { |
1da177e4 | 4484 | unsigned long flags; |
41b86e9c | 4485 | struct rq *rq; |
bb34d92f | 4486 | u64 ns = 0; |
48f24c4d | 4487 | |
41b86e9c | 4488 | rq = task_rq_lock(p, &flags); |
1508487e | 4489 | |
051a1d1a | 4490 | if (task_current(rq, p)) { |
f06febc9 FM |
4491 | u64 delta_exec; |
4492 | ||
a8e504d2 IM |
4493 | update_rq_clock(rq); |
4494 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4495 | if ((s64)delta_exec > 0) |
bb34d92f | 4496 | ns = delta_exec; |
41b86e9c | 4497 | } |
48f24c4d | 4498 | |
41b86e9c | 4499 | task_rq_unlock(rq, &flags); |
48f24c4d | 4500 | |
1da177e4 LT |
4501 | return ns; |
4502 | } | |
4503 | ||
1da177e4 LT |
4504 | /* |
4505 | * Account user cpu time to a process. | |
4506 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4507 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4508 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4509 | */ |
457533a7 MS |
4510 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4511 | cputime_t cputime_scaled) | |
1da177e4 LT |
4512 | { |
4513 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4514 | cputime64_t tmp; | |
4515 | ||
457533a7 | 4516 | /* Add user time to process. */ |
1da177e4 | 4517 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4518 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4519 | account_group_user_time(p, cputime); |
1da177e4 LT |
4520 | |
4521 | /* Add user time to cpustat. */ | |
4522 | tmp = cputime_to_cputime64(cputime); | |
4523 | if (TASK_NICE(p) > 0) | |
4524 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4525 | else | |
4526 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4527 | /* Account for user time used */ |
4528 | acct_update_integrals(p); | |
1da177e4 LT |
4529 | } |
4530 | ||
94886b84 LV |
4531 | /* |
4532 | * Account guest cpu time to a process. | |
4533 | * @p: the process that the cpu time gets accounted to | |
4534 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4535 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4536 | */ |
457533a7 MS |
4537 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4538 | cputime_t cputime_scaled) | |
94886b84 LV |
4539 | { |
4540 | cputime64_t tmp; | |
4541 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4542 | ||
4543 | tmp = cputime_to_cputime64(cputime); | |
4544 | ||
457533a7 | 4545 | /* Add guest time to process. */ |
94886b84 | 4546 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4547 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4548 | account_group_user_time(p, cputime); |
94886b84 LV |
4549 | p->gtime = cputime_add(p->gtime, cputime); |
4550 | ||
457533a7 | 4551 | /* Add guest time to cpustat. */ |
94886b84 LV |
4552 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4553 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4554 | } | |
4555 | ||
1da177e4 LT |
4556 | /* |
4557 | * Account system cpu time to a process. | |
4558 | * @p: the process that the cpu time gets accounted to | |
4559 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4560 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4561 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4562 | */ |
4563 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4564 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4565 | { |
4566 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4567 | cputime64_t tmp; |
4568 | ||
983ed7a6 | 4569 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4570 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4571 | return; |
4572 | } | |
94886b84 | 4573 | |
457533a7 | 4574 | /* Add system time to process. */ |
1da177e4 | 4575 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4576 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4577 | account_group_system_time(p, cputime); |
1da177e4 LT |
4578 | |
4579 | /* Add system time to cpustat. */ | |
4580 | tmp = cputime_to_cputime64(cputime); | |
4581 | if (hardirq_count() - hardirq_offset) | |
4582 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4583 | else if (softirq_count()) | |
4584 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4585 | else |
79741dd3 MS |
4586 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4587 | ||
1da177e4 LT |
4588 | /* Account for system time used */ |
4589 | acct_update_integrals(p); | |
1da177e4 LT |
4590 | } |
4591 | ||
c66f08be | 4592 | /* |
1da177e4 | 4593 | * Account for involuntary wait time. |
1da177e4 | 4594 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4595 | */ |
79741dd3 | 4596 | void account_steal_time(cputime_t cputime) |
c66f08be | 4597 | { |
79741dd3 MS |
4598 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4599 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4600 | ||
4601 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4602 | } |
4603 | ||
1da177e4 | 4604 | /* |
79741dd3 MS |
4605 | * Account for idle time. |
4606 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4607 | */ |
79741dd3 | 4608 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4609 | { |
4610 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4611 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4612 | struct rq *rq = this_rq(); |
1da177e4 | 4613 | |
79741dd3 MS |
4614 | if (atomic_read(&rq->nr_iowait) > 0) |
4615 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4616 | else | |
4617 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4618 | } |
4619 | ||
79741dd3 MS |
4620 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4621 | ||
4622 | /* | |
4623 | * Account a single tick of cpu time. | |
4624 | * @p: the process that the cpu time gets accounted to | |
4625 | * @user_tick: indicates if the tick is a user or a system tick | |
4626 | */ | |
4627 | void account_process_tick(struct task_struct *p, int user_tick) | |
4628 | { | |
4629 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4630 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4631 | struct rq *rq = this_rq(); | |
4632 | ||
4633 | if (user_tick) | |
4634 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
4635 | else if (p != rq->idle) | |
4636 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, | |
4637 | one_jiffy_scaled); | |
4638 | else | |
4639 | account_idle_time(one_jiffy); | |
4640 | } | |
4641 | ||
4642 | /* | |
4643 | * Account multiple ticks of steal time. | |
4644 | * @p: the process from which the cpu time has been stolen | |
4645 | * @ticks: number of stolen ticks | |
4646 | */ | |
4647 | void account_steal_ticks(unsigned long ticks) | |
4648 | { | |
4649 | account_steal_time(jiffies_to_cputime(ticks)); | |
4650 | } | |
4651 | ||
4652 | /* | |
4653 | * Account multiple ticks of idle time. | |
4654 | * @ticks: number of stolen ticks | |
4655 | */ | |
4656 | void account_idle_ticks(unsigned long ticks) | |
4657 | { | |
4658 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4659 | } |
4660 | ||
79741dd3 MS |
4661 | #endif |
4662 | ||
49048622 BS |
4663 | /* |
4664 | * Use precise platform statistics if available: | |
4665 | */ | |
4666 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4667 | cputime_t task_utime(struct task_struct *p) | |
4668 | { | |
4669 | return p->utime; | |
4670 | } | |
4671 | ||
4672 | cputime_t task_stime(struct task_struct *p) | |
4673 | { | |
4674 | return p->stime; | |
4675 | } | |
4676 | #else | |
4677 | cputime_t task_utime(struct task_struct *p) | |
4678 | { | |
4679 | clock_t utime = cputime_to_clock_t(p->utime), | |
4680 | total = utime + cputime_to_clock_t(p->stime); | |
4681 | u64 temp; | |
4682 | ||
4683 | /* | |
4684 | * Use CFS's precise accounting: | |
4685 | */ | |
4686 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4687 | ||
4688 | if (total) { | |
4689 | temp *= utime; | |
4690 | do_div(temp, total); | |
4691 | } | |
4692 | utime = (clock_t)temp; | |
4693 | ||
4694 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4695 | return p->prev_utime; | |
4696 | } | |
4697 | ||
4698 | cputime_t task_stime(struct task_struct *p) | |
4699 | { | |
4700 | clock_t stime; | |
4701 | ||
4702 | /* | |
4703 | * Use CFS's precise accounting. (we subtract utime from | |
4704 | * the total, to make sure the total observed by userspace | |
4705 | * grows monotonically - apps rely on that): | |
4706 | */ | |
4707 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4708 | cputime_to_clock_t(task_utime(p)); | |
4709 | ||
4710 | if (stime >= 0) | |
4711 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4712 | ||
4713 | return p->prev_stime; | |
4714 | } | |
4715 | #endif | |
4716 | ||
4717 | inline cputime_t task_gtime(struct task_struct *p) | |
4718 | { | |
4719 | return p->gtime; | |
4720 | } | |
4721 | ||
7835b98b CL |
4722 | /* |
4723 | * This function gets called by the timer code, with HZ frequency. | |
4724 | * We call it with interrupts disabled. | |
4725 | * | |
4726 | * It also gets called by the fork code, when changing the parent's | |
4727 | * timeslices. | |
4728 | */ | |
4729 | void scheduler_tick(void) | |
4730 | { | |
7835b98b CL |
4731 | int cpu = smp_processor_id(); |
4732 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4733 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4734 | |
4735 | sched_clock_tick(); | |
dd41f596 IM |
4736 | |
4737 | spin_lock(&rq->lock); | |
3e51f33f | 4738 | update_rq_clock(rq); |
f1a438d8 | 4739 | update_cpu_load(rq); |
fa85ae24 | 4740 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4741 | spin_unlock(&rq->lock); |
7835b98b | 4742 | |
e418e1c2 | 4743 | #ifdef CONFIG_SMP |
dd41f596 IM |
4744 | rq->idle_at_tick = idle_cpu(cpu); |
4745 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4746 | #endif |
1da177e4 LT |
4747 | } |
4748 | ||
6cd8a4bb SR |
4749 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4750 | defined(CONFIG_PREEMPT_TRACER)) | |
4751 | ||
4752 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4753 | { | |
4754 | if (in_lock_functions(addr)) { | |
4755 | addr = CALLER_ADDR2; | |
4756 | if (in_lock_functions(addr)) | |
4757 | addr = CALLER_ADDR3; | |
4758 | } | |
4759 | return addr; | |
4760 | } | |
1da177e4 | 4761 | |
43627582 | 4762 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4763 | { |
6cd8a4bb | 4764 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4765 | /* |
4766 | * Underflow? | |
4767 | */ | |
9a11b49a IM |
4768 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4769 | return; | |
6cd8a4bb | 4770 | #endif |
1da177e4 | 4771 | preempt_count() += val; |
6cd8a4bb | 4772 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4773 | /* |
4774 | * Spinlock count overflowing soon? | |
4775 | */ | |
33859f7f MOS |
4776 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4777 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4778 | #endif |
4779 | if (preempt_count() == val) | |
4780 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4781 | } |
4782 | EXPORT_SYMBOL(add_preempt_count); | |
4783 | ||
43627582 | 4784 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4785 | { |
6cd8a4bb | 4786 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4787 | /* |
4788 | * Underflow? | |
4789 | */ | |
01e3eb82 | 4790 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4791 | return; |
1da177e4 LT |
4792 | /* |
4793 | * Is the spinlock portion underflowing? | |
4794 | */ | |
9a11b49a IM |
4795 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4796 | !(preempt_count() & PREEMPT_MASK))) | |
4797 | return; | |
6cd8a4bb | 4798 | #endif |
9a11b49a | 4799 | |
6cd8a4bb SR |
4800 | if (preempt_count() == val) |
4801 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4802 | preempt_count() -= val; |
4803 | } | |
4804 | EXPORT_SYMBOL(sub_preempt_count); | |
4805 | ||
4806 | #endif | |
4807 | ||
4808 | /* | |
dd41f596 | 4809 | * Print scheduling while atomic bug: |
1da177e4 | 4810 | */ |
dd41f596 | 4811 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4812 | { |
838225b4 SS |
4813 | struct pt_regs *regs = get_irq_regs(); |
4814 | ||
4815 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4816 | prev->comm, prev->pid, preempt_count()); | |
4817 | ||
dd41f596 | 4818 | debug_show_held_locks(prev); |
e21f5b15 | 4819 | print_modules(); |
dd41f596 IM |
4820 | if (irqs_disabled()) |
4821 | print_irqtrace_events(prev); | |
838225b4 SS |
4822 | |
4823 | if (regs) | |
4824 | show_regs(regs); | |
4825 | else | |
4826 | dump_stack(); | |
dd41f596 | 4827 | } |
1da177e4 | 4828 | |
dd41f596 IM |
4829 | /* |
4830 | * Various schedule()-time debugging checks and statistics: | |
4831 | */ | |
4832 | static inline void schedule_debug(struct task_struct *prev) | |
4833 | { | |
1da177e4 | 4834 | /* |
41a2d6cf | 4835 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4836 | * schedule() atomically, we ignore that path for now. |
4837 | * Otherwise, whine if we are scheduling when we should not be. | |
4838 | */ | |
3f33a7ce | 4839 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4840 | __schedule_bug(prev); |
4841 | ||
1da177e4 LT |
4842 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4843 | ||
2d72376b | 4844 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4845 | #ifdef CONFIG_SCHEDSTATS |
4846 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4847 | schedstat_inc(this_rq(), bkl_count); |
4848 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4849 | } |
4850 | #endif | |
dd41f596 IM |
4851 | } |
4852 | ||
df1c99d4 MG |
4853 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
4854 | { | |
4855 | if (prev->state == TASK_RUNNING) { | |
4856 | u64 runtime = prev->se.sum_exec_runtime; | |
4857 | ||
4858 | runtime -= prev->se.prev_sum_exec_runtime; | |
4859 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
4860 | ||
4861 | /* | |
4862 | * In order to avoid avg_overlap growing stale when we are | |
4863 | * indeed overlapping and hence not getting put to sleep, grow | |
4864 | * the avg_overlap on preemption. | |
4865 | * | |
4866 | * We use the average preemption runtime because that | |
4867 | * correlates to the amount of cache footprint a task can | |
4868 | * build up. | |
4869 | */ | |
4870 | update_avg(&prev->se.avg_overlap, runtime); | |
4871 | } | |
4872 | prev->sched_class->put_prev_task(rq, prev); | |
4873 | } | |
4874 | ||
dd41f596 IM |
4875 | /* |
4876 | * Pick up the highest-prio task: | |
4877 | */ | |
4878 | static inline struct task_struct * | |
b67802ea | 4879 | pick_next_task(struct rq *rq) |
dd41f596 | 4880 | { |
5522d5d5 | 4881 | const struct sched_class *class; |
dd41f596 | 4882 | struct task_struct *p; |
1da177e4 LT |
4883 | |
4884 | /* | |
dd41f596 IM |
4885 | * Optimization: we know that if all tasks are in |
4886 | * the fair class we can call that function directly: | |
1da177e4 | 4887 | */ |
dd41f596 | 4888 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4889 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4890 | if (likely(p)) |
4891 | return p; | |
1da177e4 LT |
4892 | } |
4893 | ||
dd41f596 IM |
4894 | class = sched_class_highest; |
4895 | for ( ; ; ) { | |
fb8d4724 | 4896 | p = class->pick_next_task(rq); |
dd41f596 IM |
4897 | if (p) |
4898 | return p; | |
4899 | /* | |
4900 | * Will never be NULL as the idle class always | |
4901 | * returns a non-NULL p: | |
4902 | */ | |
4903 | class = class->next; | |
4904 | } | |
4905 | } | |
1da177e4 | 4906 | |
dd41f596 IM |
4907 | /* |
4908 | * schedule() is the main scheduler function. | |
4909 | */ | |
4910 | asmlinkage void __sched schedule(void) | |
4911 | { | |
4912 | struct task_struct *prev, *next; | |
67ca7bde | 4913 | unsigned long *switch_count; |
dd41f596 | 4914 | struct rq *rq; |
31656519 | 4915 | int cpu; |
dd41f596 IM |
4916 | |
4917 | need_resched: | |
4918 | preempt_disable(); | |
4919 | cpu = smp_processor_id(); | |
4920 | rq = cpu_rq(cpu); | |
4921 | rcu_qsctr_inc(cpu); | |
4922 | prev = rq->curr; | |
4923 | switch_count = &prev->nivcsw; | |
4924 | ||
4925 | release_kernel_lock(prev); | |
4926 | need_resched_nonpreemptible: | |
4927 | ||
4928 | schedule_debug(prev); | |
1da177e4 | 4929 | |
31656519 | 4930 | if (sched_feat(HRTICK)) |
f333fdc9 | 4931 | hrtick_clear(rq); |
8f4d37ec | 4932 | |
8cd162ce | 4933 | spin_lock_irq(&rq->lock); |
3e51f33f | 4934 | update_rq_clock(rq); |
1e819950 | 4935 | clear_tsk_need_resched(prev); |
1da177e4 | 4936 | |
1da177e4 | 4937 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4938 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4939 | prev->state = TASK_RUNNING; |
16882c1e | 4940 | else |
2e1cb74a | 4941 | deactivate_task(rq, prev, 1); |
dd41f596 | 4942 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4943 | } |
4944 | ||
9a897c5a SR |
4945 | #ifdef CONFIG_SMP |
4946 | if (prev->sched_class->pre_schedule) | |
4947 | prev->sched_class->pre_schedule(rq, prev); | |
4948 | #endif | |
f65eda4f | 4949 | |
dd41f596 | 4950 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4951 | idle_balance(cpu, rq); |
1da177e4 | 4952 | |
df1c99d4 | 4953 | put_prev_task(rq, prev); |
b67802ea | 4954 | next = pick_next_task(rq); |
1da177e4 | 4955 | |
1da177e4 | 4956 | if (likely(prev != next)) { |
673a90a1 DS |
4957 | sched_info_switch(prev, next); |
4958 | ||
1da177e4 LT |
4959 | rq->nr_switches++; |
4960 | rq->curr = next; | |
4961 | ++*switch_count; | |
4962 | ||
dd41f596 | 4963 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4964 | /* |
4965 | * the context switch might have flipped the stack from under | |
4966 | * us, hence refresh the local variables. | |
4967 | */ | |
4968 | cpu = smp_processor_id(); | |
4969 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4970 | } else |
4971 | spin_unlock_irq(&rq->lock); | |
4972 | ||
8f4d37ec | 4973 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4974 | goto need_resched_nonpreemptible; |
8f4d37ec | 4975 | |
1da177e4 LT |
4976 | preempt_enable_no_resched(); |
4977 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4978 | goto need_resched; | |
4979 | } | |
1da177e4 LT |
4980 | EXPORT_SYMBOL(schedule); |
4981 | ||
4982 | #ifdef CONFIG_PREEMPT | |
4983 | /* | |
2ed6e34f | 4984 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4985 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4986 | * occur there and call schedule directly. |
4987 | */ | |
4988 | asmlinkage void __sched preempt_schedule(void) | |
4989 | { | |
4990 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4991 | |
1da177e4 LT |
4992 | /* |
4993 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4994 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4995 | */ |
beed33a8 | 4996 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4997 | return; |
4998 | ||
3a5c359a AK |
4999 | do { |
5000 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5001 | schedule(); |
3a5c359a | 5002 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5003 | |
3a5c359a AK |
5004 | /* |
5005 | * Check again in case we missed a preemption opportunity | |
5006 | * between schedule and now. | |
5007 | */ | |
5008 | barrier(); | |
5ed0cec0 | 5009 | } while (need_resched()); |
1da177e4 | 5010 | } |
1da177e4 LT |
5011 | EXPORT_SYMBOL(preempt_schedule); |
5012 | ||
5013 | /* | |
2ed6e34f | 5014 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5015 | * off of irq context. |
5016 | * Note, that this is called and return with irqs disabled. This will | |
5017 | * protect us against recursive calling from irq. | |
5018 | */ | |
5019 | asmlinkage void __sched preempt_schedule_irq(void) | |
5020 | { | |
5021 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5022 | |
2ed6e34f | 5023 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5024 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5025 | ||
3a5c359a AK |
5026 | do { |
5027 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5028 | local_irq_enable(); |
5029 | schedule(); | |
5030 | local_irq_disable(); | |
3a5c359a | 5031 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5032 | |
3a5c359a AK |
5033 | /* |
5034 | * Check again in case we missed a preemption opportunity | |
5035 | * between schedule and now. | |
5036 | */ | |
5037 | barrier(); | |
5ed0cec0 | 5038 | } while (need_resched()); |
1da177e4 LT |
5039 | } |
5040 | ||
5041 | #endif /* CONFIG_PREEMPT */ | |
5042 | ||
95cdf3b7 IM |
5043 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5044 | void *key) | |
1da177e4 | 5045 | { |
48f24c4d | 5046 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5047 | } |
1da177e4 LT |
5048 | EXPORT_SYMBOL(default_wake_function); |
5049 | ||
5050 | /* | |
41a2d6cf IM |
5051 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5052 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5053 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5054 | * | |
5055 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5056 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5057 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5058 | */ | |
777c6c5f JW |
5059 | void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
5060 | int nr_exclusive, int sync, void *key) | |
1da177e4 | 5061 | { |
2e45874c | 5062 | wait_queue_t *curr, *next; |
1da177e4 | 5063 | |
2e45874c | 5064 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5065 | unsigned flags = curr->flags; |
5066 | ||
1da177e4 | 5067 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5068 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5069 | break; |
5070 | } | |
5071 | } | |
5072 | ||
5073 | /** | |
5074 | * __wake_up - wake up threads blocked on a waitqueue. | |
5075 | * @q: the waitqueue | |
5076 | * @mode: which threads | |
5077 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5078 | * @key: is directly passed to the wakeup function |
1da177e4 | 5079 | */ |
7ad5b3a5 | 5080 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5081 | int nr_exclusive, void *key) |
1da177e4 LT |
5082 | { |
5083 | unsigned long flags; | |
5084 | ||
5085 | spin_lock_irqsave(&q->lock, flags); | |
5086 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5087 | spin_unlock_irqrestore(&q->lock, flags); | |
5088 | } | |
1da177e4 LT |
5089 | EXPORT_SYMBOL(__wake_up); |
5090 | ||
5091 | /* | |
5092 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5093 | */ | |
7ad5b3a5 | 5094 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5095 | { |
5096 | __wake_up_common(q, mode, 1, 0, NULL); | |
5097 | } | |
5098 | ||
5099 | /** | |
67be2dd1 | 5100 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5101 | * @q: the waitqueue |
5102 | * @mode: which threads | |
5103 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
5104 | * | |
5105 | * The sync wakeup differs that the waker knows that it will schedule | |
5106 | * away soon, so while the target thread will be woken up, it will not | |
5107 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5108 | * with each other. This can prevent needless bouncing between CPUs. | |
5109 | * | |
5110 | * On UP it can prevent extra preemption. | |
5111 | */ | |
7ad5b3a5 | 5112 | void |
95cdf3b7 | 5113 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
5114 | { |
5115 | unsigned long flags; | |
5116 | int sync = 1; | |
5117 | ||
5118 | if (unlikely(!q)) | |
5119 | return; | |
5120 | ||
5121 | if (unlikely(!nr_exclusive)) | |
5122 | sync = 0; | |
5123 | ||
5124 | spin_lock_irqsave(&q->lock, flags); | |
5125 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
5126 | spin_unlock_irqrestore(&q->lock, flags); | |
5127 | } | |
5128 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
5129 | ||
65eb3dc6 KD |
5130 | /** |
5131 | * complete: - signals a single thread waiting on this completion | |
5132 | * @x: holds the state of this particular completion | |
5133 | * | |
5134 | * This will wake up a single thread waiting on this completion. Threads will be | |
5135 | * awakened in the same order in which they were queued. | |
5136 | * | |
5137 | * See also complete_all(), wait_for_completion() and related routines. | |
5138 | */ | |
b15136e9 | 5139 | void complete(struct completion *x) |
1da177e4 LT |
5140 | { |
5141 | unsigned long flags; | |
5142 | ||
5143 | spin_lock_irqsave(&x->wait.lock, flags); | |
5144 | x->done++; | |
d9514f6c | 5145 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5146 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5147 | } | |
5148 | EXPORT_SYMBOL(complete); | |
5149 | ||
65eb3dc6 KD |
5150 | /** |
5151 | * complete_all: - signals all threads waiting on this completion | |
5152 | * @x: holds the state of this particular completion | |
5153 | * | |
5154 | * This will wake up all threads waiting on this particular completion event. | |
5155 | */ | |
b15136e9 | 5156 | void complete_all(struct completion *x) |
1da177e4 LT |
5157 | { |
5158 | unsigned long flags; | |
5159 | ||
5160 | spin_lock_irqsave(&x->wait.lock, flags); | |
5161 | x->done += UINT_MAX/2; | |
d9514f6c | 5162 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5163 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5164 | } | |
5165 | EXPORT_SYMBOL(complete_all); | |
5166 | ||
8cbbe86d AK |
5167 | static inline long __sched |
5168 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5169 | { |
1da177e4 LT |
5170 | if (!x->done) { |
5171 | DECLARE_WAITQUEUE(wait, current); | |
5172 | ||
5173 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5174 | __add_wait_queue_tail(&x->wait, &wait); | |
5175 | do { | |
94d3d824 | 5176 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5177 | timeout = -ERESTARTSYS; |
5178 | break; | |
8cbbe86d AK |
5179 | } |
5180 | __set_current_state(state); | |
1da177e4 LT |
5181 | spin_unlock_irq(&x->wait.lock); |
5182 | timeout = schedule_timeout(timeout); | |
5183 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5184 | } while (!x->done && timeout); |
1da177e4 | 5185 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5186 | if (!x->done) |
5187 | return timeout; | |
1da177e4 LT |
5188 | } |
5189 | x->done--; | |
ea71a546 | 5190 | return timeout ?: 1; |
1da177e4 | 5191 | } |
1da177e4 | 5192 | |
8cbbe86d AK |
5193 | static long __sched |
5194 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5195 | { |
1da177e4 LT |
5196 | might_sleep(); |
5197 | ||
5198 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5199 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5200 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5201 | return timeout; |
5202 | } | |
1da177e4 | 5203 | |
65eb3dc6 KD |
5204 | /** |
5205 | * wait_for_completion: - waits for completion of a task | |
5206 | * @x: holds the state of this particular completion | |
5207 | * | |
5208 | * This waits to be signaled for completion of a specific task. It is NOT | |
5209 | * interruptible and there is no timeout. | |
5210 | * | |
5211 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5212 | * and interrupt capability. Also see complete(). | |
5213 | */ | |
b15136e9 | 5214 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5215 | { |
5216 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5217 | } |
8cbbe86d | 5218 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5219 | |
65eb3dc6 KD |
5220 | /** |
5221 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5222 | * @x: holds the state of this particular completion | |
5223 | * @timeout: timeout value in jiffies | |
5224 | * | |
5225 | * This waits for either a completion of a specific task to be signaled or for a | |
5226 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5227 | * interruptible. | |
5228 | */ | |
b15136e9 | 5229 | unsigned long __sched |
8cbbe86d | 5230 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5231 | { |
8cbbe86d | 5232 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5233 | } |
8cbbe86d | 5234 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5235 | |
65eb3dc6 KD |
5236 | /** |
5237 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5238 | * @x: holds the state of this particular completion | |
5239 | * | |
5240 | * This waits for completion of a specific task to be signaled. It is | |
5241 | * interruptible. | |
5242 | */ | |
8cbbe86d | 5243 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5244 | { |
51e97990 AK |
5245 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5246 | if (t == -ERESTARTSYS) | |
5247 | return t; | |
5248 | return 0; | |
0fec171c | 5249 | } |
8cbbe86d | 5250 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5251 | |
65eb3dc6 KD |
5252 | /** |
5253 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5254 | * @x: holds the state of this particular completion | |
5255 | * @timeout: timeout value in jiffies | |
5256 | * | |
5257 | * This waits for either a completion of a specific task to be signaled or for a | |
5258 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5259 | */ | |
b15136e9 | 5260 | unsigned long __sched |
8cbbe86d AK |
5261 | wait_for_completion_interruptible_timeout(struct completion *x, |
5262 | unsigned long timeout) | |
0fec171c | 5263 | { |
8cbbe86d | 5264 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5265 | } |
8cbbe86d | 5266 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5267 | |
65eb3dc6 KD |
5268 | /** |
5269 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5270 | * @x: holds the state of this particular completion | |
5271 | * | |
5272 | * This waits to be signaled for completion of a specific task. It can be | |
5273 | * interrupted by a kill signal. | |
5274 | */ | |
009e577e MW |
5275 | int __sched wait_for_completion_killable(struct completion *x) |
5276 | { | |
5277 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5278 | if (t == -ERESTARTSYS) | |
5279 | return t; | |
5280 | return 0; | |
5281 | } | |
5282 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5283 | ||
be4de352 DC |
5284 | /** |
5285 | * try_wait_for_completion - try to decrement a completion without blocking | |
5286 | * @x: completion structure | |
5287 | * | |
5288 | * Returns: 0 if a decrement cannot be done without blocking | |
5289 | * 1 if a decrement succeeded. | |
5290 | * | |
5291 | * If a completion is being used as a counting completion, | |
5292 | * attempt to decrement the counter without blocking. This | |
5293 | * enables us to avoid waiting if the resource the completion | |
5294 | * is protecting is not available. | |
5295 | */ | |
5296 | bool try_wait_for_completion(struct completion *x) | |
5297 | { | |
5298 | int ret = 1; | |
5299 | ||
5300 | spin_lock_irq(&x->wait.lock); | |
5301 | if (!x->done) | |
5302 | ret = 0; | |
5303 | else | |
5304 | x->done--; | |
5305 | spin_unlock_irq(&x->wait.lock); | |
5306 | return ret; | |
5307 | } | |
5308 | EXPORT_SYMBOL(try_wait_for_completion); | |
5309 | ||
5310 | /** | |
5311 | * completion_done - Test to see if a completion has any waiters | |
5312 | * @x: completion structure | |
5313 | * | |
5314 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5315 | * 1 if there are no waiters. | |
5316 | * | |
5317 | */ | |
5318 | bool completion_done(struct completion *x) | |
5319 | { | |
5320 | int ret = 1; | |
5321 | ||
5322 | spin_lock_irq(&x->wait.lock); | |
5323 | if (!x->done) | |
5324 | ret = 0; | |
5325 | spin_unlock_irq(&x->wait.lock); | |
5326 | return ret; | |
5327 | } | |
5328 | EXPORT_SYMBOL(completion_done); | |
5329 | ||
8cbbe86d AK |
5330 | static long __sched |
5331 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5332 | { |
0fec171c IM |
5333 | unsigned long flags; |
5334 | wait_queue_t wait; | |
5335 | ||
5336 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5337 | |
8cbbe86d | 5338 | __set_current_state(state); |
1da177e4 | 5339 | |
8cbbe86d AK |
5340 | spin_lock_irqsave(&q->lock, flags); |
5341 | __add_wait_queue(q, &wait); | |
5342 | spin_unlock(&q->lock); | |
5343 | timeout = schedule_timeout(timeout); | |
5344 | spin_lock_irq(&q->lock); | |
5345 | __remove_wait_queue(q, &wait); | |
5346 | spin_unlock_irqrestore(&q->lock, flags); | |
5347 | ||
5348 | return timeout; | |
5349 | } | |
5350 | ||
5351 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5352 | { | |
5353 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5354 | } |
1da177e4 LT |
5355 | EXPORT_SYMBOL(interruptible_sleep_on); |
5356 | ||
0fec171c | 5357 | long __sched |
95cdf3b7 | 5358 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5359 | { |
8cbbe86d | 5360 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5361 | } |
1da177e4 LT |
5362 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5363 | ||
0fec171c | 5364 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5365 | { |
8cbbe86d | 5366 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5367 | } |
1da177e4 LT |
5368 | EXPORT_SYMBOL(sleep_on); |
5369 | ||
0fec171c | 5370 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5371 | { |
8cbbe86d | 5372 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5373 | } |
1da177e4 LT |
5374 | EXPORT_SYMBOL(sleep_on_timeout); |
5375 | ||
b29739f9 IM |
5376 | #ifdef CONFIG_RT_MUTEXES |
5377 | ||
5378 | /* | |
5379 | * rt_mutex_setprio - set the current priority of a task | |
5380 | * @p: task | |
5381 | * @prio: prio value (kernel-internal form) | |
5382 | * | |
5383 | * This function changes the 'effective' priority of a task. It does | |
5384 | * not touch ->normal_prio like __setscheduler(). | |
5385 | * | |
5386 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5387 | */ | |
36c8b586 | 5388 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5389 | { |
5390 | unsigned long flags; | |
83b699ed | 5391 | int oldprio, on_rq, running; |
70b97a7f | 5392 | struct rq *rq; |
cb469845 | 5393 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5394 | |
5395 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5396 | ||
5397 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5398 | update_rq_clock(rq); |
b29739f9 | 5399 | |
d5f9f942 | 5400 | oldprio = p->prio; |
dd41f596 | 5401 | on_rq = p->se.on_rq; |
051a1d1a | 5402 | running = task_current(rq, p); |
0e1f3483 | 5403 | if (on_rq) |
69be72c1 | 5404 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5405 | if (running) |
5406 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5407 | |
5408 | if (rt_prio(prio)) | |
5409 | p->sched_class = &rt_sched_class; | |
5410 | else | |
5411 | p->sched_class = &fair_sched_class; | |
5412 | ||
b29739f9 IM |
5413 | p->prio = prio; |
5414 | ||
0e1f3483 HS |
5415 | if (running) |
5416 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5417 | if (on_rq) { |
8159f87e | 5418 | enqueue_task(rq, p, 0); |
cb469845 SR |
5419 | |
5420 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5421 | } |
5422 | task_rq_unlock(rq, &flags); | |
5423 | } | |
5424 | ||
5425 | #endif | |
5426 | ||
36c8b586 | 5427 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5428 | { |
dd41f596 | 5429 | int old_prio, delta, on_rq; |
1da177e4 | 5430 | unsigned long flags; |
70b97a7f | 5431 | struct rq *rq; |
1da177e4 LT |
5432 | |
5433 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5434 | return; | |
5435 | /* | |
5436 | * We have to be careful, if called from sys_setpriority(), | |
5437 | * the task might be in the middle of scheduling on another CPU. | |
5438 | */ | |
5439 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5440 | update_rq_clock(rq); |
1da177e4 LT |
5441 | /* |
5442 | * The RT priorities are set via sched_setscheduler(), but we still | |
5443 | * allow the 'normal' nice value to be set - but as expected | |
5444 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5445 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5446 | */ |
e05606d3 | 5447 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5448 | p->static_prio = NICE_TO_PRIO(nice); |
5449 | goto out_unlock; | |
5450 | } | |
dd41f596 | 5451 | on_rq = p->se.on_rq; |
c09595f6 | 5452 | if (on_rq) |
69be72c1 | 5453 | dequeue_task(rq, p, 0); |
1da177e4 | 5454 | |
1da177e4 | 5455 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5456 | set_load_weight(p); |
b29739f9 IM |
5457 | old_prio = p->prio; |
5458 | p->prio = effective_prio(p); | |
5459 | delta = p->prio - old_prio; | |
1da177e4 | 5460 | |
dd41f596 | 5461 | if (on_rq) { |
8159f87e | 5462 | enqueue_task(rq, p, 0); |
1da177e4 | 5463 | /* |
d5f9f942 AM |
5464 | * If the task increased its priority or is running and |
5465 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5466 | */ |
d5f9f942 | 5467 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5468 | resched_task(rq->curr); |
5469 | } | |
5470 | out_unlock: | |
5471 | task_rq_unlock(rq, &flags); | |
5472 | } | |
1da177e4 LT |
5473 | EXPORT_SYMBOL(set_user_nice); |
5474 | ||
e43379f1 MM |
5475 | /* |
5476 | * can_nice - check if a task can reduce its nice value | |
5477 | * @p: task | |
5478 | * @nice: nice value | |
5479 | */ | |
36c8b586 | 5480 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5481 | { |
024f4747 MM |
5482 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5483 | int nice_rlim = 20 - nice; | |
48f24c4d | 5484 | |
e43379f1 MM |
5485 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5486 | capable(CAP_SYS_NICE)); | |
5487 | } | |
5488 | ||
1da177e4 LT |
5489 | #ifdef __ARCH_WANT_SYS_NICE |
5490 | ||
5491 | /* | |
5492 | * sys_nice - change the priority of the current process. | |
5493 | * @increment: priority increment | |
5494 | * | |
5495 | * sys_setpriority is a more generic, but much slower function that | |
5496 | * does similar things. | |
5497 | */ | |
5add95d4 | 5498 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5499 | { |
48f24c4d | 5500 | long nice, retval; |
1da177e4 LT |
5501 | |
5502 | /* | |
5503 | * Setpriority might change our priority at the same moment. | |
5504 | * We don't have to worry. Conceptually one call occurs first | |
5505 | * and we have a single winner. | |
5506 | */ | |
e43379f1 MM |
5507 | if (increment < -40) |
5508 | increment = -40; | |
1da177e4 LT |
5509 | if (increment > 40) |
5510 | increment = 40; | |
5511 | ||
2b8f836f | 5512 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5513 | if (nice < -20) |
5514 | nice = -20; | |
5515 | if (nice > 19) | |
5516 | nice = 19; | |
5517 | ||
e43379f1 MM |
5518 | if (increment < 0 && !can_nice(current, nice)) |
5519 | return -EPERM; | |
5520 | ||
1da177e4 LT |
5521 | retval = security_task_setnice(current, nice); |
5522 | if (retval) | |
5523 | return retval; | |
5524 | ||
5525 | set_user_nice(current, nice); | |
5526 | return 0; | |
5527 | } | |
5528 | ||
5529 | #endif | |
5530 | ||
5531 | /** | |
5532 | * task_prio - return the priority value of a given task. | |
5533 | * @p: the task in question. | |
5534 | * | |
5535 | * This is the priority value as seen by users in /proc. | |
5536 | * RT tasks are offset by -200. Normal tasks are centered | |
5537 | * around 0, value goes from -16 to +15. | |
5538 | */ | |
36c8b586 | 5539 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5540 | { |
5541 | return p->prio - MAX_RT_PRIO; | |
5542 | } | |
5543 | ||
5544 | /** | |
5545 | * task_nice - return the nice value of a given task. | |
5546 | * @p: the task in question. | |
5547 | */ | |
36c8b586 | 5548 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5549 | { |
5550 | return TASK_NICE(p); | |
5551 | } | |
150d8bed | 5552 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5553 | |
5554 | /** | |
5555 | * idle_cpu - is a given cpu idle currently? | |
5556 | * @cpu: the processor in question. | |
5557 | */ | |
5558 | int idle_cpu(int cpu) | |
5559 | { | |
5560 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5561 | } | |
5562 | ||
1da177e4 LT |
5563 | /** |
5564 | * idle_task - return the idle task for a given cpu. | |
5565 | * @cpu: the processor in question. | |
5566 | */ | |
36c8b586 | 5567 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5568 | { |
5569 | return cpu_rq(cpu)->idle; | |
5570 | } | |
5571 | ||
5572 | /** | |
5573 | * find_process_by_pid - find a process with a matching PID value. | |
5574 | * @pid: the pid in question. | |
5575 | */ | |
a9957449 | 5576 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5577 | { |
228ebcbe | 5578 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5579 | } |
5580 | ||
5581 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5582 | static void |
5583 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5584 | { |
dd41f596 | 5585 | BUG_ON(p->se.on_rq); |
48f24c4d | 5586 | |
1da177e4 | 5587 | p->policy = policy; |
dd41f596 IM |
5588 | switch (p->policy) { |
5589 | case SCHED_NORMAL: | |
5590 | case SCHED_BATCH: | |
5591 | case SCHED_IDLE: | |
5592 | p->sched_class = &fair_sched_class; | |
5593 | break; | |
5594 | case SCHED_FIFO: | |
5595 | case SCHED_RR: | |
5596 | p->sched_class = &rt_sched_class; | |
5597 | break; | |
5598 | } | |
5599 | ||
1da177e4 | 5600 | p->rt_priority = prio; |
b29739f9 IM |
5601 | p->normal_prio = normal_prio(p); |
5602 | /* we are holding p->pi_lock already */ | |
5603 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5604 | set_load_weight(p); |
1da177e4 LT |
5605 | } |
5606 | ||
c69e8d9c DH |
5607 | /* |
5608 | * check the target process has a UID that matches the current process's | |
5609 | */ | |
5610 | static bool check_same_owner(struct task_struct *p) | |
5611 | { | |
5612 | const struct cred *cred = current_cred(), *pcred; | |
5613 | bool match; | |
5614 | ||
5615 | rcu_read_lock(); | |
5616 | pcred = __task_cred(p); | |
5617 | match = (cred->euid == pcred->euid || | |
5618 | cred->euid == pcred->uid); | |
5619 | rcu_read_unlock(); | |
5620 | return match; | |
5621 | } | |
5622 | ||
961ccddd RR |
5623 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5624 | struct sched_param *param, bool user) | |
1da177e4 | 5625 | { |
83b699ed | 5626 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5627 | unsigned long flags; |
cb469845 | 5628 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5629 | struct rq *rq; |
1da177e4 | 5630 | |
66e5393a SR |
5631 | /* may grab non-irq protected spin_locks */ |
5632 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5633 | recheck: |
5634 | /* double check policy once rq lock held */ | |
5635 | if (policy < 0) | |
5636 | policy = oldpolicy = p->policy; | |
5637 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5638 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5639 | policy != SCHED_IDLE) | |
b0a9499c | 5640 | return -EINVAL; |
1da177e4 LT |
5641 | /* |
5642 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5643 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5644 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5645 | */ |
5646 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5647 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5648 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5649 | return -EINVAL; |
e05606d3 | 5650 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5651 | return -EINVAL; |
5652 | ||
37e4ab3f OC |
5653 | /* |
5654 | * Allow unprivileged RT tasks to decrease priority: | |
5655 | */ | |
961ccddd | 5656 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5657 | if (rt_policy(policy)) { |
8dc3e909 | 5658 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5659 | |
5660 | if (!lock_task_sighand(p, &flags)) | |
5661 | return -ESRCH; | |
5662 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5663 | unlock_task_sighand(p, &flags); | |
5664 | ||
5665 | /* can't set/change the rt policy */ | |
5666 | if (policy != p->policy && !rlim_rtprio) | |
5667 | return -EPERM; | |
5668 | ||
5669 | /* can't increase priority */ | |
5670 | if (param->sched_priority > p->rt_priority && | |
5671 | param->sched_priority > rlim_rtprio) | |
5672 | return -EPERM; | |
5673 | } | |
dd41f596 IM |
5674 | /* |
5675 | * Like positive nice levels, dont allow tasks to | |
5676 | * move out of SCHED_IDLE either: | |
5677 | */ | |
5678 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5679 | return -EPERM; | |
5fe1d75f | 5680 | |
37e4ab3f | 5681 | /* can't change other user's priorities */ |
c69e8d9c | 5682 | if (!check_same_owner(p)) |
37e4ab3f OC |
5683 | return -EPERM; |
5684 | } | |
1da177e4 | 5685 | |
725aad24 | 5686 | if (user) { |
b68aa230 | 5687 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5688 | /* |
5689 | * Do not allow realtime tasks into groups that have no runtime | |
5690 | * assigned. | |
5691 | */ | |
9a7e0b18 PZ |
5692 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5693 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5694 | return -EPERM; |
b68aa230 PZ |
5695 | #endif |
5696 | ||
725aad24 JF |
5697 | retval = security_task_setscheduler(p, policy, param); |
5698 | if (retval) | |
5699 | return retval; | |
5700 | } | |
5701 | ||
b29739f9 IM |
5702 | /* |
5703 | * make sure no PI-waiters arrive (or leave) while we are | |
5704 | * changing the priority of the task: | |
5705 | */ | |
5706 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5707 | /* |
5708 | * To be able to change p->policy safely, the apropriate | |
5709 | * runqueue lock must be held. | |
5710 | */ | |
b29739f9 | 5711 | rq = __task_rq_lock(p); |
1da177e4 LT |
5712 | /* recheck policy now with rq lock held */ |
5713 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5714 | policy = oldpolicy = -1; | |
b29739f9 IM |
5715 | __task_rq_unlock(rq); |
5716 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5717 | goto recheck; |
5718 | } | |
2daa3577 | 5719 | update_rq_clock(rq); |
dd41f596 | 5720 | on_rq = p->se.on_rq; |
051a1d1a | 5721 | running = task_current(rq, p); |
0e1f3483 | 5722 | if (on_rq) |
2e1cb74a | 5723 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5724 | if (running) |
5725 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5726 | |
1da177e4 | 5727 | oldprio = p->prio; |
dd41f596 | 5728 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5729 | |
0e1f3483 HS |
5730 | if (running) |
5731 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5732 | if (on_rq) { |
5733 | activate_task(rq, p, 0); | |
cb469845 SR |
5734 | |
5735 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5736 | } |
b29739f9 IM |
5737 | __task_rq_unlock(rq); |
5738 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5739 | ||
95e02ca9 TG |
5740 | rt_mutex_adjust_pi(p); |
5741 | ||
1da177e4 LT |
5742 | return 0; |
5743 | } | |
961ccddd RR |
5744 | |
5745 | /** | |
5746 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5747 | * @p: the task in question. | |
5748 | * @policy: new policy. | |
5749 | * @param: structure containing the new RT priority. | |
5750 | * | |
5751 | * NOTE that the task may be already dead. | |
5752 | */ | |
5753 | int sched_setscheduler(struct task_struct *p, int policy, | |
5754 | struct sched_param *param) | |
5755 | { | |
5756 | return __sched_setscheduler(p, policy, param, true); | |
5757 | } | |
1da177e4 LT |
5758 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5759 | ||
961ccddd RR |
5760 | /** |
5761 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5762 | * @p: the task in question. | |
5763 | * @policy: new policy. | |
5764 | * @param: structure containing the new RT priority. | |
5765 | * | |
5766 | * Just like sched_setscheduler, only don't bother checking if the | |
5767 | * current context has permission. For example, this is needed in | |
5768 | * stop_machine(): we create temporary high priority worker threads, | |
5769 | * but our caller might not have that capability. | |
5770 | */ | |
5771 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5772 | struct sched_param *param) | |
5773 | { | |
5774 | return __sched_setscheduler(p, policy, param, false); | |
5775 | } | |
5776 | ||
95cdf3b7 IM |
5777 | static int |
5778 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5779 | { |
1da177e4 LT |
5780 | struct sched_param lparam; |
5781 | struct task_struct *p; | |
36c8b586 | 5782 | int retval; |
1da177e4 LT |
5783 | |
5784 | if (!param || pid < 0) | |
5785 | return -EINVAL; | |
5786 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5787 | return -EFAULT; | |
5fe1d75f ON |
5788 | |
5789 | rcu_read_lock(); | |
5790 | retval = -ESRCH; | |
1da177e4 | 5791 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5792 | if (p != NULL) |
5793 | retval = sched_setscheduler(p, policy, &lparam); | |
5794 | rcu_read_unlock(); | |
36c8b586 | 5795 | |
1da177e4 LT |
5796 | return retval; |
5797 | } | |
5798 | ||
5799 | /** | |
5800 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5801 | * @pid: the pid in question. | |
5802 | * @policy: new policy. | |
5803 | * @param: structure containing the new RT priority. | |
5804 | */ | |
5add95d4 HC |
5805 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5806 | struct sched_param __user *, param) | |
1da177e4 | 5807 | { |
c21761f1 JB |
5808 | /* negative values for policy are not valid */ |
5809 | if (policy < 0) | |
5810 | return -EINVAL; | |
5811 | ||
1da177e4 LT |
5812 | return do_sched_setscheduler(pid, policy, param); |
5813 | } | |
5814 | ||
5815 | /** | |
5816 | * sys_sched_setparam - set/change the RT priority of a thread | |
5817 | * @pid: the pid in question. | |
5818 | * @param: structure containing the new RT priority. | |
5819 | */ | |
5add95d4 | 5820 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5821 | { |
5822 | return do_sched_setscheduler(pid, -1, param); | |
5823 | } | |
5824 | ||
5825 | /** | |
5826 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5827 | * @pid: the pid in question. | |
5828 | */ | |
5add95d4 | 5829 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5830 | { |
36c8b586 | 5831 | struct task_struct *p; |
3a5c359a | 5832 | int retval; |
1da177e4 LT |
5833 | |
5834 | if (pid < 0) | |
3a5c359a | 5835 | return -EINVAL; |
1da177e4 LT |
5836 | |
5837 | retval = -ESRCH; | |
5838 | read_lock(&tasklist_lock); | |
5839 | p = find_process_by_pid(pid); | |
5840 | if (p) { | |
5841 | retval = security_task_getscheduler(p); | |
5842 | if (!retval) | |
5843 | retval = p->policy; | |
5844 | } | |
5845 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5846 | return retval; |
5847 | } | |
5848 | ||
5849 | /** | |
5850 | * sys_sched_getscheduler - get the RT priority of a thread | |
5851 | * @pid: the pid in question. | |
5852 | * @param: structure containing the RT priority. | |
5853 | */ | |
5add95d4 | 5854 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5855 | { |
5856 | struct sched_param lp; | |
36c8b586 | 5857 | struct task_struct *p; |
3a5c359a | 5858 | int retval; |
1da177e4 LT |
5859 | |
5860 | if (!param || pid < 0) | |
3a5c359a | 5861 | return -EINVAL; |
1da177e4 LT |
5862 | |
5863 | read_lock(&tasklist_lock); | |
5864 | p = find_process_by_pid(pid); | |
5865 | retval = -ESRCH; | |
5866 | if (!p) | |
5867 | goto out_unlock; | |
5868 | ||
5869 | retval = security_task_getscheduler(p); | |
5870 | if (retval) | |
5871 | goto out_unlock; | |
5872 | ||
5873 | lp.sched_priority = p->rt_priority; | |
5874 | read_unlock(&tasklist_lock); | |
5875 | ||
5876 | /* | |
5877 | * This one might sleep, we cannot do it with a spinlock held ... | |
5878 | */ | |
5879 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5880 | ||
1da177e4 LT |
5881 | return retval; |
5882 | ||
5883 | out_unlock: | |
5884 | read_unlock(&tasklist_lock); | |
5885 | return retval; | |
5886 | } | |
5887 | ||
96f874e2 | 5888 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5889 | { |
5a16f3d3 | 5890 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5891 | struct task_struct *p; |
5892 | int retval; | |
1da177e4 | 5893 | |
95402b38 | 5894 | get_online_cpus(); |
1da177e4 LT |
5895 | read_lock(&tasklist_lock); |
5896 | ||
5897 | p = find_process_by_pid(pid); | |
5898 | if (!p) { | |
5899 | read_unlock(&tasklist_lock); | |
95402b38 | 5900 | put_online_cpus(); |
1da177e4 LT |
5901 | return -ESRCH; |
5902 | } | |
5903 | ||
5904 | /* | |
5905 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5906 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5907 | * usage count and then drop tasklist_lock. |
5908 | */ | |
5909 | get_task_struct(p); | |
5910 | read_unlock(&tasklist_lock); | |
5911 | ||
5a16f3d3 RR |
5912 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5913 | retval = -ENOMEM; | |
5914 | goto out_put_task; | |
5915 | } | |
5916 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5917 | retval = -ENOMEM; | |
5918 | goto out_free_cpus_allowed; | |
5919 | } | |
1da177e4 | 5920 | retval = -EPERM; |
c69e8d9c | 5921 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5922 | goto out_unlock; |
5923 | ||
e7834f8f DQ |
5924 | retval = security_task_setscheduler(p, 0, NULL); |
5925 | if (retval) | |
5926 | goto out_unlock; | |
5927 | ||
5a16f3d3 RR |
5928 | cpuset_cpus_allowed(p, cpus_allowed); |
5929 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 5930 | again: |
5a16f3d3 | 5931 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5932 | |
8707d8b8 | 5933 | if (!retval) { |
5a16f3d3 RR |
5934 | cpuset_cpus_allowed(p, cpus_allowed); |
5935 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5936 | /* |
5937 | * We must have raced with a concurrent cpuset | |
5938 | * update. Just reset the cpus_allowed to the | |
5939 | * cpuset's cpus_allowed | |
5940 | */ | |
5a16f3d3 | 5941 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5942 | goto again; |
5943 | } | |
5944 | } | |
1da177e4 | 5945 | out_unlock: |
5a16f3d3 RR |
5946 | free_cpumask_var(new_mask); |
5947 | out_free_cpus_allowed: | |
5948 | free_cpumask_var(cpus_allowed); | |
5949 | out_put_task: | |
1da177e4 | 5950 | put_task_struct(p); |
95402b38 | 5951 | put_online_cpus(); |
1da177e4 LT |
5952 | return retval; |
5953 | } | |
5954 | ||
5955 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5956 | struct cpumask *new_mask) |
1da177e4 | 5957 | { |
96f874e2 RR |
5958 | if (len < cpumask_size()) |
5959 | cpumask_clear(new_mask); | |
5960 | else if (len > cpumask_size()) | |
5961 | len = cpumask_size(); | |
5962 | ||
1da177e4 LT |
5963 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5964 | } | |
5965 | ||
5966 | /** | |
5967 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5968 | * @pid: pid of the process | |
5969 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5970 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5971 | */ | |
5add95d4 HC |
5972 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5973 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5974 | { |
5a16f3d3 | 5975 | cpumask_var_t new_mask; |
1da177e4 LT |
5976 | int retval; |
5977 | ||
5a16f3d3 RR |
5978 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5979 | return -ENOMEM; | |
1da177e4 | 5980 | |
5a16f3d3 RR |
5981 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5982 | if (retval == 0) | |
5983 | retval = sched_setaffinity(pid, new_mask); | |
5984 | free_cpumask_var(new_mask); | |
5985 | return retval; | |
1da177e4 LT |
5986 | } |
5987 | ||
96f874e2 | 5988 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5989 | { |
36c8b586 | 5990 | struct task_struct *p; |
1da177e4 | 5991 | int retval; |
1da177e4 | 5992 | |
95402b38 | 5993 | get_online_cpus(); |
1da177e4 LT |
5994 | read_lock(&tasklist_lock); |
5995 | ||
5996 | retval = -ESRCH; | |
5997 | p = find_process_by_pid(pid); | |
5998 | if (!p) | |
5999 | goto out_unlock; | |
6000 | ||
e7834f8f DQ |
6001 | retval = security_task_getscheduler(p); |
6002 | if (retval) | |
6003 | goto out_unlock; | |
6004 | ||
96f874e2 | 6005 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6006 | |
6007 | out_unlock: | |
6008 | read_unlock(&tasklist_lock); | |
95402b38 | 6009 | put_online_cpus(); |
1da177e4 | 6010 | |
9531b62f | 6011 | return retval; |
1da177e4 LT |
6012 | } |
6013 | ||
6014 | /** | |
6015 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6016 | * @pid: pid of the process | |
6017 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6018 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6019 | */ | |
5add95d4 HC |
6020 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6021 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6022 | { |
6023 | int ret; | |
f17c8607 | 6024 | cpumask_var_t mask; |
1da177e4 | 6025 | |
f17c8607 | 6026 | if (len < cpumask_size()) |
1da177e4 LT |
6027 | return -EINVAL; |
6028 | ||
f17c8607 RR |
6029 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6030 | return -ENOMEM; | |
1da177e4 | 6031 | |
f17c8607 RR |
6032 | ret = sched_getaffinity(pid, mask); |
6033 | if (ret == 0) { | |
6034 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6035 | ret = -EFAULT; | |
6036 | else | |
6037 | ret = cpumask_size(); | |
6038 | } | |
6039 | free_cpumask_var(mask); | |
1da177e4 | 6040 | |
f17c8607 | 6041 | return ret; |
1da177e4 LT |
6042 | } |
6043 | ||
6044 | /** | |
6045 | * sys_sched_yield - yield the current processor to other threads. | |
6046 | * | |
dd41f596 IM |
6047 | * This function yields the current CPU to other tasks. If there are no |
6048 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6049 | */ |
5add95d4 | 6050 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6051 | { |
70b97a7f | 6052 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6053 | |
2d72376b | 6054 | schedstat_inc(rq, yld_count); |
4530d7ab | 6055 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6056 | |
6057 | /* | |
6058 | * Since we are going to call schedule() anyway, there's | |
6059 | * no need to preempt or enable interrupts: | |
6060 | */ | |
6061 | __release(rq->lock); | |
8a25d5de | 6062 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6063 | _raw_spin_unlock(&rq->lock); |
6064 | preempt_enable_no_resched(); | |
6065 | ||
6066 | schedule(); | |
6067 | ||
6068 | return 0; | |
6069 | } | |
6070 | ||
e7b38404 | 6071 | static void __cond_resched(void) |
1da177e4 | 6072 | { |
8e0a43d8 IM |
6073 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6074 | __might_sleep(__FILE__, __LINE__); | |
6075 | #endif | |
5bbcfd90 IM |
6076 | /* |
6077 | * The BKS might be reacquired before we have dropped | |
6078 | * PREEMPT_ACTIVE, which could trigger a second | |
6079 | * cond_resched() call. | |
6080 | */ | |
1da177e4 LT |
6081 | do { |
6082 | add_preempt_count(PREEMPT_ACTIVE); | |
6083 | schedule(); | |
6084 | sub_preempt_count(PREEMPT_ACTIVE); | |
6085 | } while (need_resched()); | |
6086 | } | |
6087 | ||
02b67cc3 | 6088 | int __sched _cond_resched(void) |
1da177e4 | 6089 | { |
9414232f IM |
6090 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6091 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6092 | __cond_resched(); |
6093 | return 1; | |
6094 | } | |
6095 | return 0; | |
6096 | } | |
02b67cc3 | 6097 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6098 | |
6099 | /* | |
6100 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6101 | * call schedule, and on return reacquire the lock. | |
6102 | * | |
41a2d6cf | 6103 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6104 | * operations here to prevent schedule() from being called twice (once via |
6105 | * spin_unlock(), once by hand). | |
6106 | */ | |
95cdf3b7 | 6107 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6108 | { |
95c354fe | 6109 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6110 | int ret = 0; |
6111 | ||
95c354fe | 6112 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6113 | spin_unlock(lock); |
95c354fe NP |
6114 | if (resched && need_resched()) |
6115 | __cond_resched(); | |
6116 | else | |
6117 | cpu_relax(); | |
6df3cecb | 6118 | ret = 1; |
1da177e4 | 6119 | spin_lock(lock); |
1da177e4 | 6120 | } |
6df3cecb | 6121 | return ret; |
1da177e4 | 6122 | } |
1da177e4 LT |
6123 | EXPORT_SYMBOL(cond_resched_lock); |
6124 | ||
6125 | int __sched cond_resched_softirq(void) | |
6126 | { | |
6127 | BUG_ON(!in_softirq()); | |
6128 | ||
9414232f | 6129 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6130 | local_bh_enable(); |
1da177e4 LT |
6131 | __cond_resched(); |
6132 | local_bh_disable(); | |
6133 | return 1; | |
6134 | } | |
6135 | return 0; | |
6136 | } | |
1da177e4 LT |
6137 | EXPORT_SYMBOL(cond_resched_softirq); |
6138 | ||
1da177e4 LT |
6139 | /** |
6140 | * yield - yield the current processor to other threads. | |
6141 | * | |
72fd4a35 | 6142 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6143 | * thread runnable and calls sys_sched_yield(). |
6144 | */ | |
6145 | void __sched yield(void) | |
6146 | { | |
6147 | set_current_state(TASK_RUNNING); | |
6148 | sys_sched_yield(); | |
6149 | } | |
1da177e4 LT |
6150 | EXPORT_SYMBOL(yield); |
6151 | ||
6152 | /* | |
41a2d6cf | 6153 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6154 | * that process accounting knows that this is a task in IO wait state. |
6155 | * | |
6156 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6157 | * has set its backing_dev_info: the queue against which it should throttle) | |
6158 | */ | |
6159 | void __sched io_schedule(void) | |
6160 | { | |
70b97a7f | 6161 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6162 | |
0ff92245 | 6163 | delayacct_blkio_start(); |
1da177e4 LT |
6164 | atomic_inc(&rq->nr_iowait); |
6165 | schedule(); | |
6166 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6167 | delayacct_blkio_end(); |
1da177e4 | 6168 | } |
1da177e4 LT |
6169 | EXPORT_SYMBOL(io_schedule); |
6170 | ||
6171 | long __sched io_schedule_timeout(long timeout) | |
6172 | { | |
70b97a7f | 6173 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6174 | long ret; |
6175 | ||
0ff92245 | 6176 | delayacct_blkio_start(); |
1da177e4 LT |
6177 | atomic_inc(&rq->nr_iowait); |
6178 | ret = schedule_timeout(timeout); | |
6179 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6180 | delayacct_blkio_end(); |
1da177e4 LT |
6181 | return ret; |
6182 | } | |
6183 | ||
6184 | /** | |
6185 | * sys_sched_get_priority_max - return maximum RT priority. | |
6186 | * @policy: scheduling class. | |
6187 | * | |
6188 | * this syscall returns the maximum rt_priority that can be used | |
6189 | * by a given scheduling class. | |
6190 | */ | |
5add95d4 | 6191 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6192 | { |
6193 | int ret = -EINVAL; | |
6194 | ||
6195 | switch (policy) { | |
6196 | case SCHED_FIFO: | |
6197 | case SCHED_RR: | |
6198 | ret = MAX_USER_RT_PRIO-1; | |
6199 | break; | |
6200 | case SCHED_NORMAL: | |
b0a9499c | 6201 | case SCHED_BATCH: |
dd41f596 | 6202 | case SCHED_IDLE: |
1da177e4 LT |
6203 | ret = 0; |
6204 | break; | |
6205 | } | |
6206 | return ret; | |
6207 | } | |
6208 | ||
6209 | /** | |
6210 | * sys_sched_get_priority_min - return minimum RT priority. | |
6211 | * @policy: scheduling class. | |
6212 | * | |
6213 | * this syscall returns the minimum rt_priority that can be used | |
6214 | * by a given scheduling class. | |
6215 | */ | |
5add95d4 | 6216 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6217 | { |
6218 | int ret = -EINVAL; | |
6219 | ||
6220 | switch (policy) { | |
6221 | case SCHED_FIFO: | |
6222 | case SCHED_RR: | |
6223 | ret = 1; | |
6224 | break; | |
6225 | case SCHED_NORMAL: | |
b0a9499c | 6226 | case SCHED_BATCH: |
dd41f596 | 6227 | case SCHED_IDLE: |
1da177e4 LT |
6228 | ret = 0; |
6229 | } | |
6230 | return ret; | |
6231 | } | |
6232 | ||
6233 | /** | |
6234 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6235 | * @pid: pid of the process. | |
6236 | * @interval: userspace pointer to the timeslice value. | |
6237 | * | |
6238 | * this syscall writes the default timeslice value of a given process | |
6239 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6240 | */ | |
17da2bd9 | 6241 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6242 | struct timespec __user *, interval) |
1da177e4 | 6243 | { |
36c8b586 | 6244 | struct task_struct *p; |
a4ec24b4 | 6245 | unsigned int time_slice; |
3a5c359a | 6246 | int retval; |
1da177e4 | 6247 | struct timespec t; |
1da177e4 LT |
6248 | |
6249 | if (pid < 0) | |
3a5c359a | 6250 | return -EINVAL; |
1da177e4 LT |
6251 | |
6252 | retval = -ESRCH; | |
6253 | read_lock(&tasklist_lock); | |
6254 | p = find_process_by_pid(pid); | |
6255 | if (!p) | |
6256 | goto out_unlock; | |
6257 | ||
6258 | retval = security_task_getscheduler(p); | |
6259 | if (retval) | |
6260 | goto out_unlock; | |
6261 | ||
77034937 IM |
6262 | /* |
6263 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6264 | * tasks that are on an otherwise idle runqueue: | |
6265 | */ | |
6266 | time_slice = 0; | |
6267 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6268 | time_slice = DEF_TIMESLICE; |
1868f958 | 6269 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6270 | struct sched_entity *se = &p->se; |
6271 | unsigned long flags; | |
6272 | struct rq *rq; | |
6273 | ||
6274 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6275 | if (rq->cfs.load.weight) |
6276 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6277 | task_rq_unlock(rq, &flags); |
6278 | } | |
1da177e4 | 6279 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6280 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6281 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6282 | return retval; |
3a5c359a | 6283 | |
1da177e4 LT |
6284 | out_unlock: |
6285 | read_unlock(&tasklist_lock); | |
6286 | return retval; | |
6287 | } | |
6288 | ||
7c731e0a | 6289 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6290 | |
82a1fcb9 | 6291 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6292 | { |
1da177e4 | 6293 | unsigned long free = 0; |
36c8b586 | 6294 | unsigned state; |
1da177e4 | 6295 | |
1da177e4 | 6296 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6297 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6298 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6299 | #if BITS_PER_LONG == 32 |
1da177e4 | 6300 | if (state == TASK_RUNNING) |
cc4ea795 | 6301 | printk(KERN_CONT " running "); |
1da177e4 | 6302 | else |
cc4ea795 | 6303 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6304 | #else |
6305 | if (state == TASK_RUNNING) | |
cc4ea795 | 6306 | printk(KERN_CONT " running task "); |
1da177e4 | 6307 | else |
cc4ea795 | 6308 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6309 | #endif |
6310 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
6311 | { | |
10ebffde | 6312 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
6313 | while (!*n) |
6314 | n++; | |
10ebffde | 6315 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
6316 | } |
6317 | #endif | |
ba25f9dc | 6318 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 6319 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 6320 | |
5fb5e6de | 6321 | show_stack(p, NULL); |
1da177e4 LT |
6322 | } |
6323 | ||
e59e2ae2 | 6324 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6325 | { |
36c8b586 | 6326 | struct task_struct *g, *p; |
1da177e4 | 6327 | |
4bd77321 IM |
6328 | #if BITS_PER_LONG == 32 |
6329 | printk(KERN_INFO | |
6330 | " task PC stack pid father\n"); | |
1da177e4 | 6331 | #else |
4bd77321 IM |
6332 | printk(KERN_INFO |
6333 | " task PC stack pid father\n"); | |
1da177e4 LT |
6334 | #endif |
6335 | read_lock(&tasklist_lock); | |
6336 | do_each_thread(g, p) { | |
6337 | /* | |
6338 | * reset the NMI-timeout, listing all files on a slow | |
6339 | * console might take alot of time: | |
6340 | */ | |
6341 | touch_nmi_watchdog(); | |
39bc89fd | 6342 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6343 | sched_show_task(p); |
1da177e4 LT |
6344 | } while_each_thread(g, p); |
6345 | ||
04c9167f JF |
6346 | touch_all_softlockup_watchdogs(); |
6347 | ||
dd41f596 IM |
6348 | #ifdef CONFIG_SCHED_DEBUG |
6349 | sysrq_sched_debug_show(); | |
6350 | #endif | |
1da177e4 | 6351 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6352 | /* |
6353 | * Only show locks if all tasks are dumped: | |
6354 | */ | |
6355 | if (state_filter == -1) | |
6356 | debug_show_all_locks(); | |
1da177e4 LT |
6357 | } |
6358 | ||
1df21055 IM |
6359 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6360 | { | |
dd41f596 | 6361 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6362 | } |
6363 | ||
f340c0d1 IM |
6364 | /** |
6365 | * init_idle - set up an idle thread for a given CPU | |
6366 | * @idle: task in question | |
6367 | * @cpu: cpu the idle task belongs to | |
6368 | * | |
6369 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6370 | * flag, to make booting more robust. | |
6371 | */ | |
5c1e1767 | 6372 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6373 | { |
70b97a7f | 6374 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6375 | unsigned long flags; |
6376 | ||
5cbd54ef IM |
6377 | spin_lock_irqsave(&rq->lock, flags); |
6378 | ||
dd41f596 IM |
6379 | __sched_fork(idle); |
6380 | idle->se.exec_start = sched_clock(); | |
6381 | ||
b29739f9 | 6382 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6383 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6384 | __set_task_cpu(idle, cpu); |
1da177e4 | 6385 | |
1da177e4 | 6386 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6387 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6388 | idle->oncpu = 1; | |
6389 | #endif | |
1da177e4 LT |
6390 | spin_unlock_irqrestore(&rq->lock, flags); |
6391 | ||
6392 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6393 | #if defined(CONFIG_PREEMPT) |
6394 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6395 | #else | |
a1261f54 | 6396 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6397 | #endif |
dd41f596 IM |
6398 | /* |
6399 | * The idle tasks have their own, simple scheduling class: | |
6400 | */ | |
6401 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6402 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6403 | } |
6404 | ||
6405 | /* | |
6406 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6407 | * indicates which cpus entered this state. This is used | |
6408 | * in the rcu update to wait only for active cpus. For system | |
6409 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6410 | * always be CPU_BITS_NONE. |
1da177e4 | 6411 | */ |
6a7b3dc3 | 6412 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6413 | |
19978ca6 IM |
6414 | /* |
6415 | * Increase the granularity value when there are more CPUs, | |
6416 | * because with more CPUs the 'effective latency' as visible | |
6417 | * to users decreases. But the relationship is not linear, | |
6418 | * so pick a second-best guess by going with the log2 of the | |
6419 | * number of CPUs. | |
6420 | * | |
6421 | * This idea comes from the SD scheduler of Con Kolivas: | |
6422 | */ | |
6423 | static inline void sched_init_granularity(void) | |
6424 | { | |
6425 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6426 | const unsigned long limit = 200000000; | |
6427 | ||
6428 | sysctl_sched_min_granularity *= factor; | |
6429 | if (sysctl_sched_min_granularity > limit) | |
6430 | sysctl_sched_min_granularity = limit; | |
6431 | ||
6432 | sysctl_sched_latency *= factor; | |
6433 | if (sysctl_sched_latency > limit) | |
6434 | sysctl_sched_latency = limit; | |
6435 | ||
6436 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6437 | |
6438 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6439 | } |
6440 | ||
1da177e4 LT |
6441 | #ifdef CONFIG_SMP |
6442 | /* | |
6443 | * This is how migration works: | |
6444 | * | |
70b97a7f | 6445 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6446 | * runqueue and wake up that CPU's migration thread. |
6447 | * 2) we down() the locked semaphore => thread blocks. | |
6448 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6449 | * thread off the CPU) | |
6450 | * 4) it gets the migration request and checks whether the migrated | |
6451 | * task is still in the wrong runqueue. | |
6452 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6453 | * it and puts it into the right queue. | |
6454 | * 6) migration thread up()s the semaphore. | |
6455 | * 7) we wake up and the migration is done. | |
6456 | */ | |
6457 | ||
6458 | /* | |
6459 | * Change a given task's CPU affinity. Migrate the thread to a | |
6460 | * proper CPU and schedule it away if the CPU it's executing on | |
6461 | * is removed from the allowed bitmask. | |
6462 | * | |
6463 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6464 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6465 | * call is not atomic; no spinlocks may be held. |
6466 | */ | |
96f874e2 | 6467 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6468 | { |
70b97a7f | 6469 | struct migration_req req; |
1da177e4 | 6470 | unsigned long flags; |
70b97a7f | 6471 | struct rq *rq; |
48f24c4d | 6472 | int ret = 0; |
1da177e4 LT |
6473 | |
6474 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6475 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6476 | ret = -EINVAL; |
6477 | goto out; | |
6478 | } | |
6479 | ||
9985b0ba | 6480 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6481 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6482 | ret = -EINVAL; |
6483 | goto out; | |
6484 | } | |
6485 | ||
73fe6aae | 6486 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6487 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6488 | else { |
96f874e2 RR |
6489 | cpumask_copy(&p->cpus_allowed, new_mask); |
6490 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6491 | } |
6492 | ||
1da177e4 | 6493 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6494 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6495 | goto out; |
6496 | ||
1e5ce4f4 | 6497 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6498 | /* Need help from migration thread: drop lock and wait. */ |
6499 | task_rq_unlock(rq, &flags); | |
6500 | wake_up_process(rq->migration_thread); | |
6501 | wait_for_completion(&req.done); | |
6502 | tlb_migrate_finish(p->mm); | |
6503 | return 0; | |
6504 | } | |
6505 | out: | |
6506 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6507 | |
1da177e4 LT |
6508 | return ret; |
6509 | } | |
cd8ba7cd | 6510 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6511 | |
6512 | /* | |
41a2d6cf | 6513 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6514 | * this because either it can't run here any more (set_cpus_allowed() |
6515 | * away from this CPU, or CPU going down), or because we're | |
6516 | * attempting to rebalance this task on exec (sched_exec). | |
6517 | * | |
6518 | * So we race with normal scheduler movements, but that's OK, as long | |
6519 | * as the task is no longer on this CPU. | |
efc30814 KK |
6520 | * |
6521 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6522 | */ |
efc30814 | 6523 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6524 | { |
70b97a7f | 6525 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6526 | int ret = 0, on_rq; |
1da177e4 | 6527 | |
e761b772 | 6528 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6529 | return ret; |
1da177e4 LT |
6530 | |
6531 | rq_src = cpu_rq(src_cpu); | |
6532 | rq_dest = cpu_rq(dest_cpu); | |
6533 | ||
6534 | double_rq_lock(rq_src, rq_dest); | |
6535 | /* Already moved. */ | |
6536 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6537 | goto done; |
1da177e4 | 6538 | /* Affinity changed (again). */ |
96f874e2 | 6539 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6540 | goto fail; |
1da177e4 | 6541 | |
dd41f596 | 6542 | on_rq = p->se.on_rq; |
6e82a3be | 6543 | if (on_rq) |
2e1cb74a | 6544 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6545 | |
1da177e4 | 6546 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6547 | if (on_rq) { |
6548 | activate_task(rq_dest, p, 0); | |
15afe09b | 6549 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6550 | } |
b1e38734 | 6551 | done: |
efc30814 | 6552 | ret = 1; |
b1e38734 | 6553 | fail: |
1da177e4 | 6554 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6555 | return ret; |
1da177e4 LT |
6556 | } |
6557 | ||
6558 | /* | |
6559 | * migration_thread - this is a highprio system thread that performs | |
6560 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6561 | * another runqueue. | |
6562 | */ | |
95cdf3b7 | 6563 | static int migration_thread(void *data) |
1da177e4 | 6564 | { |
1da177e4 | 6565 | int cpu = (long)data; |
70b97a7f | 6566 | struct rq *rq; |
1da177e4 LT |
6567 | |
6568 | rq = cpu_rq(cpu); | |
6569 | BUG_ON(rq->migration_thread != current); | |
6570 | ||
6571 | set_current_state(TASK_INTERRUPTIBLE); | |
6572 | while (!kthread_should_stop()) { | |
70b97a7f | 6573 | struct migration_req *req; |
1da177e4 | 6574 | struct list_head *head; |
1da177e4 | 6575 | |
1da177e4 LT |
6576 | spin_lock_irq(&rq->lock); |
6577 | ||
6578 | if (cpu_is_offline(cpu)) { | |
6579 | spin_unlock_irq(&rq->lock); | |
6580 | goto wait_to_die; | |
6581 | } | |
6582 | ||
6583 | if (rq->active_balance) { | |
6584 | active_load_balance(rq, cpu); | |
6585 | rq->active_balance = 0; | |
6586 | } | |
6587 | ||
6588 | head = &rq->migration_queue; | |
6589 | ||
6590 | if (list_empty(head)) { | |
6591 | spin_unlock_irq(&rq->lock); | |
6592 | schedule(); | |
6593 | set_current_state(TASK_INTERRUPTIBLE); | |
6594 | continue; | |
6595 | } | |
70b97a7f | 6596 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6597 | list_del_init(head->next); |
6598 | ||
674311d5 NP |
6599 | spin_unlock(&rq->lock); |
6600 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6601 | local_irq_enable(); | |
1da177e4 LT |
6602 | |
6603 | complete(&req->done); | |
6604 | } | |
6605 | __set_current_state(TASK_RUNNING); | |
6606 | return 0; | |
6607 | ||
6608 | wait_to_die: | |
6609 | /* Wait for kthread_stop */ | |
6610 | set_current_state(TASK_INTERRUPTIBLE); | |
6611 | while (!kthread_should_stop()) { | |
6612 | schedule(); | |
6613 | set_current_state(TASK_INTERRUPTIBLE); | |
6614 | } | |
6615 | __set_current_state(TASK_RUNNING); | |
6616 | return 0; | |
6617 | } | |
6618 | ||
6619 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6620 | |
6621 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6622 | { | |
6623 | int ret; | |
6624 | ||
6625 | local_irq_disable(); | |
6626 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6627 | local_irq_enable(); | |
6628 | return ret; | |
6629 | } | |
6630 | ||
054b9108 | 6631 | /* |
3a4fa0a2 | 6632 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6633 | */ |
48f24c4d | 6634 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6635 | { |
70b97a7f | 6636 | int dest_cpu; |
6ca09dfc | 6637 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
6638 | |
6639 | again: | |
6640 | /* Look for allowed, online CPU in same node. */ | |
6641 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6642 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6643 | goto move; | |
6644 | ||
6645 | /* Any allowed, online CPU? */ | |
6646 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6647 | if (dest_cpu < nr_cpu_ids) | |
6648 | goto move; | |
6649 | ||
6650 | /* No more Mr. Nice Guy. */ | |
6651 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6652 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6653 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6654 | |
e76bd8d9 RR |
6655 | /* |
6656 | * Don't tell them about moving exiting tasks or | |
6657 | * kernel threads (both mm NULL), since they never | |
6658 | * leave kernel. | |
6659 | */ | |
6660 | if (p->mm && printk_ratelimit()) { | |
6661 | printk(KERN_INFO "process %d (%s) no " | |
6662 | "longer affine to cpu%d\n", | |
6663 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6664 | } |
e76bd8d9 RR |
6665 | } |
6666 | ||
6667 | move: | |
6668 | /* It can have affinity changed while we were choosing. */ | |
6669 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6670 | goto again; | |
1da177e4 LT |
6671 | } |
6672 | ||
6673 | /* | |
6674 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6675 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6676 | * for performance reasons the counter is not stricly tracking tasks to | |
6677 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6678 | * to keep the global sum constant after CPU-down: | |
6679 | */ | |
70b97a7f | 6680 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6681 | { |
1e5ce4f4 | 6682 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6683 | unsigned long flags; |
6684 | ||
6685 | local_irq_save(flags); | |
6686 | double_rq_lock(rq_src, rq_dest); | |
6687 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6688 | rq_src->nr_uninterruptible = 0; | |
6689 | double_rq_unlock(rq_src, rq_dest); | |
6690 | local_irq_restore(flags); | |
6691 | } | |
6692 | ||
6693 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6694 | static void migrate_live_tasks(int src_cpu) | |
6695 | { | |
48f24c4d | 6696 | struct task_struct *p, *t; |
1da177e4 | 6697 | |
f7b4cddc | 6698 | read_lock(&tasklist_lock); |
1da177e4 | 6699 | |
48f24c4d IM |
6700 | do_each_thread(t, p) { |
6701 | if (p == current) | |
1da177e4 LT |
6702 | continue; |
6703 | ||
48f24c4d IM |
6704 | if (task_cpu(p) == src_cpu) |
6705 | move_task_off_dead_cpu(src_cpu, p); | |
6706 | } while_each_thread(t, p); | |
1da177e4 | 6707 | |
f7b4cddc | 6708 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6709 | } |
6710 | ||
dd41f596 IM |
6711 | /* |
6712 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6713 | * It does so by boosting its priority to highest possible. |
6714 | * Used by CPU offline code. | |
1da177e4 LT |
6715 | */ |
6716 | void sched_idle_next(void) | |
6717 | { | |
48f24c4d | 6718 | int this_cpu = smp_processor_id(); |
70b97a7f | 6719 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6720 | struct task_struct *p = rq->idle; |
6721 | unsigned long flags; | |
6722 | ||
6723 | /* cpu has to be offline */ | |
48f24c4d | 6724 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6725 | |
48f24c4d IM |
6726 | /* |
6727 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6728 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6729 | */ |
6730 | spin_lock_irqsave(&rq->lock, flags); | |
6731 | ||
dd41f596 | 6732 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6733 | |
94bc9a7b DA |
6734 | update_rq_clock(rq); |
6735 | activate_task(rq, p, 0); | |
1da177e4 LT |
6736 | |
6737 | spin_unlock_irqrestore(&rq->lock, flags); | |
6738 | } | |
6739 | ||
48f24c4d IM |
6740 | /* |
6741 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6742 | * offline. |
6743 | */ | |
6744 | void idle_task_exit(void) | |
6745 | { | |
6746 | struct mm_struct *mm = current->active_mm; | |
6747 | ||
6748 | BUG_ON(cpu_online(smp_processor_id())); | |
6749 | ||
6750 | if (mm != &init_mm) | |
6751 | switch_mm(mm, &init_mm, current); | |
6752 | mmdrop(mm); | |
6753 | } | |
6754 | ||
054b9108 | 6755 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6756 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6757 | { |
70b97a7f | 6758 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6759 | |
6760 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6761 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6762 | |
6763 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6764 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6765 | |
48f24c4d | 6766 | get_task_struct(p); |
1da177e4 LT |
6767 | |
6768 | /* | |
6769 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6770 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6771 | * fine. |
6772 | */ | |
f7b4cddc | 6773 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6774 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6775 | spin_lock_irq(&rq->lock); |
1da177e4 | 6776 | |
48f24c4d | 6777 | put_task_struct(p); |
1da177e4 LT |
6778 | } |
6779 | ||
6780 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6781 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6782 | { | |
70b97a7f | 6783 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6784 | struct task_struct *next; |
48f24c4d | 6785 | |
dd41f596 IM |
6786 | for ( ; ; ) { |
6787 | if (!rq->nr_running) | |
6788 | break; | |
a8e504d2 | 6789 | update_rq_clock(rq); |
b67802ea | 6790 | next = pick_next_task(rq); |
dd41f596 IM |
6791 | if (!next) |
6792 | break; | |
79c53799 | 6793 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6794 | migrate_dead(dead_cpu, next); |
e692ab53 | 6795 | |
1da177e4 LT |
6796 | } |
6797 | } | |
6798 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6799 | ||
e692ab53 NP |
6800 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6801 | ||
6802 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6803 | { |
6804 | .procname = "sched_domain", | |
c57baf1e | 6805 | .mode = 0555, |
e0361851 | 6806 | }, |
38605cae | 6807 | {0, }, |
e692ab53 NP |
6808 | }; |
6809 | ||
6810 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6811 | { |
c57baf1e | 6812 | .ctl_name = CTL_KERN, |
e0361851 | 6813 | .procname = "kernel", |
c57baf1e | 6814 | .mode = 0555, |
e0361851 AD |
6815 | .child = sd_ctl_dir, |
6816 | }, | |
38605cae | 6817 | {0, }, |
e692ab53 NP |
6818 | }; |
6819 | ||
6820 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6821 | { | |
6822 | struct ctl_table *entry = | |
5cf9f062 | 6823 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6824 | |
e692ab53 NP |
6825 | return entry; |
6826 | } | |
6827 | ||
6382bc90 MM |
6828 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6829 | { | |
cd790076 | 6830 | struct ctl_table *entry; |
6382bc90 | 6831 | |
cd790076 MM |
6832 | /* |
6833 | * In the intermediate directories, both the child directory and | |
6834 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6835 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6836 | * static strings and all have proc handlers. |
6837 | */ | |
6838 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6839 | if (entry->child) |
6840 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6841 | if (entry->proc_handler == NULL) |
6842 | kfree(entry->procname); | |
6843 | } | |
6382bc90 MM |
6844 | |
6845 | kfree(*tablep); | |
6846 | *tablep = NULL; | |
6847 | } | |
6848 | ||
e692ab53 | 6849 | static void |
e0361851 | 6850 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6851 | const char *procname, void *data, int maxlen, |
6852 | mode_t mode, proc_handler *proc_handler) | |
6853 | { | |
e692ab53 NP |
6854 | entry->procname = procname; |
6855 | entry->data = data; | |
6856 | entry->maxlen = maxlen; | |
6857 | entry->mode = mode; | |
6858 | entry->proc_handler = proc_handler; | |
6859 | } | |
6860 | ||
6861 | static struct ctl_table * | |
6862 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6863 | { | |
a5d8c348 | 6864 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6865 | |
ad1cdc1d MM |
6866 | if (table == NULL) |
6867 | return NULL; | |
6868 | ||
e0361851 | 6869 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6870 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6871 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6872 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6873 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6874 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6875 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6876 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6877 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6878 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6879 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6880 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6881 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6882 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6883 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6884 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6885 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6886 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6887 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6888 | &sd->cache_nice_tries, |
6889 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6890 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6891 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6892 | set_table_entry(&table[11], "name", sd->name, |
6893 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6894 | /* &table[12] is terminator */ | |
e692ab53 NP |
6895 | |
6896 | return table; | |
6897 | } | |
6898 | ||
9a4e7159 | 6899 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6900 | { |
6901 | struct ctl_table *entry, *table; | |
6902 | struct sched_domain *sd; | |
6903 | int domain_num = 0, i; | |
6904 | char buf[32]; | |
6905 | ||
6906 | for_each_domain(cpu, sd) | |
6907 | domain_num++; | |
6908 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6909 | if (table == NULL) |
6910 | return NULL; | |
e692ab53 NP |
6911 | |
6912 | i = 0; | |
6913 | for_each_domain(cpu, sd) { | |
6914 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6915 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6916 | entry->mode = 0555; |
e692ab53 NP |
6917 | entry->child = sd_alloc_ctl_domain_table(sd); |
6918 | entry++; | |
6919 | i++; | |
6920 | } | |
6921 | return table; | |
6922 | } | |
6923 | ||
6924 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6925 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6926 | { |
6927 | int i, cpu_num = num_online_cpus(); | |
6928 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6929 | char buf[32]; | |
6930 | ||
7378547f MM |
6931 | WARN_ON(sd_ctl_dir[0].child); |
6932 | sd_ctl_dir[0].child = entry; | |
6933 | ||
ad1cdc1d MM |
6934 | if (entry == NULL) |
6935 | return; | |
6936 | ||
97b6ea7b | 6937 | for_each_online_cpu(i) { |
e692ab53 | 6938 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6939 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6940 | entry->mode = 0555; |
e692ab53 | 6941 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6942 | entry++; |
e692ab53 | 6943 | } |
7378547f MM |
6944 | |
6945 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6946 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6947 | } | |
6382bc90 | 6948 | |
7378547f | 6949 | /* may be called multiple times per register */ |
6382bc90 MM |
6950 | static void unregister_sched_domain_sysctl(void) |
6951 | { | |
7378547f MM |
6952 | if (sd_sysctl_header) |
6953 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6954 | sd_sysctl_header = NULL; |
7378547f MM |
6955 | if (sd_ctl_dir[0].child) |
6956 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6957 | } |
e692ab53 | 6958 | #else |
6382bc90 MM |
6959 | static void register_sched_domain_sysctl(void) |
6960 | { | |
6961 | } | |
6962 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6963 | { |
6964 | } | |
6965 | #endif | |
6966 | ||
1f11eb6a GH |
6967 | static void set_rq_online(struct rq *rq) |
6968 | { | |
6969 | if (!rq->online) { | |
6970 | const struct sched_class *class; | |
6971 | ||
c6c4927b | 6972 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6973 | rq->online = 1; |
6974 | ||
6975 | for_each_class(class) { | |
6976 | if (class->rq_online) | |
6977 | class->rq_online(rq); | |
6978 | } | |
6979 | } | |
6980 | } | |
6981 | ||
6982 | static void set_rq_offline(struct rq *rq) | |
6983 | { | |
6984 | if (rq->online) { | |
6985 | const struct sched_class *class; | |
6986 | ||
6987 | for_each_class(class) { | |
6988 | if (class->rq_offline) | |
6989 | class->rq_offline(rq); | |
6990 | } | |
6991 | ||
c6c4927b | 6992 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6993 | rq->online = 0; |
6994 | } | |
6995 | } | |
6996 | ||
1da177e4 LT |
6997 | /* |
6998 | * migration_call - callback that gets triggered when a CPU is added. | |
6999 | * Here we can start up the necessary migration thread for the new CPU. | |
7000 | */ | |
48f24c4d IM |
7001 | static int __cpuinit |
7002 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7003 | { |
1da177e4 | 7004 | struct task_struct *p; |
48f24c4d | 7005 | int cpu = (long)hcpu; |
1da177e4 | 7006 | unsigned long flags; |
70b97a7f | 7007 | struct rq *rq; |
1da177e4 LT |
7008 | |
7009 | switch (action) { | |
5be9361c | 7010 | |
1da177e4 | 7011 | case CPU_UP_PREPARE: |
8bb78442 | 7012 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7013 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7014 | if (IS_ERR(p)) |
7015 | return NOTIFY_BAD; | |
1da177e4 LT |
7016 | kthread_bind(p, cpu); |
7017 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7018 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7019 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7020 | task_rq_unlock(rq, &flags); |
7021 | cpu_rq(cpu)->migration_thread = p; | |
7022 | break; | |
48f24c4d | 7023 | |
1da177e4 | 7024 | case CPU_ONLINE: |
8bb78442 | 7025 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7026 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7027 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7028 | |
7029 | /* Update our root-domain */ | |
7030 | rq = cpu_rq(cpu); | |
7031 | spin_lock_irqsave(&rq->lock, flags); | |
7032 | if (rq->rd) { | |
c6c4927b | 7033 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7034 | |
7035 | set_rq_online(rq); | |
1f94ef59 GH |
7036 | } |
7037 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7038 | break; |
48f24c4d | 7039 | |
1da177e4 LT |
7040 | #ifdef CONFIG_HOTPLUG_CPU |
7041 | case CPU_UP_CANCELED: | |
8bb78442 | 7042 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7043 | if (!cpu_rq(cpu)->migration_thread) |
7044 | break; | |
41a2d6cf | 7045 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7046 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7047 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7048 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7049 | cpu_rq(cpu)->migration_thread = NULL; | |
7050 | break; | |
48f24c4d | 7051 | |
1da177e4 | 7052 | case CPU_DEAD: |
8bb78442 | 7053 | case CPU_DEAD_FROZEN: |
470fd646 | 7054 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7055 | migrate_live_tasks(cpu); |
7056 | rq = cpu_rq(cpu); | |
7057 | kthread_stop(rq->migration_thread); | |
7058 | rq->migration_thread = NULL; | |
7059 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7060 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7061 | update_rq_clock(rq); |
2e1cb74a | 7062 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7063 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7064 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7065 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7066 | migrate_dead_tasks(cpu); |
d2da272a | 7067 | spin_unlock_irq(&rq->lock); |
470fd646 | 7068 | cpuset_unlock(); |
1da177e4 LT |
7069 | migrate_nr_uninterruptible(rq); |
7070 | BUG_ON(rq->nr_running != 0); | |
7071 | ||
41a2d6cf IM |
7072 | /* |
7073 | * No need to migrate the tasks: it was best-effort if | |
7074 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7075 | * the requestors. | |
7076 | */ | |
1da177e4 LT |
7077 | spin_lock_irq(&rq->lock); |
7078 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7079 | struct migration_req *req; |
7080 | ||
1da177e4 | 7081 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7082 | struct migration_req, list); |
1da177e4 | 7083 | list_del_init(&req->list); |
9a2bd244 | 7084 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7085 | complete(&req->done); |
9a2bd244 | 7086 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7087 | } |
7088 | spin_unlock_irq(&rq->lock); | |
7089 | break; | |
57d885fe | 7090 | |
08f503b0 GH |
7091 | case CPU_DYING: |
7092 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7093 | /* Update our root-domain */ |
7094 | rq = cpu_rq(cpu); | |
7095 | spin_lock_irqsave(&rq->lock, flags); | |
7096 | if (rq->rd) { | |
c6c4927b | 7097 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7098 | set_rq_offline(rq); |
57d885fe GH |
7099 | } |
7100 | spin_unlock_irqrestore(&rq->lock, flags); | |
7101 | break; | |
1da177e4 LT |
7102 | #endif |
7103 | } | |
7104 | return NOTIFY_OK; | |
7105 | } | |
7106 | ||
7107 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
7108 | * happens before everything else. | |
7109 | */ | |
26c2143b | 7110 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7111 | .notifier_call = migration_call, |
7112 | .priority = 10 | |
7113 | }; | |
7114 | ||
7babe8db | 7115 | static int __init migration_init(void) |
1da177e4 LT |
7116 | { |
7117 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7118 | int err; |
48f24c4d IM |
7119 | |
7120 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7121 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7122 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7123 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7124 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7125 | |
7126 | return err; | |
1da177e4 | 7127 | } |
7babe8db | 7128 | early_initcall(migration_init); |
1da177e4 LT |
7129 | #endif |
7130 | ||
7131 | #ifdef CONFIG_SMP | |
476f3534 | 7132 | |
3e9830dc | 7133 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7134 | |
7c16ec58 | 7135 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7136 | struct cpumask *groupmask) |
1da177e4 | 7137 | { |
4dcf6aff | 7138 | struct sched_group *group = sd->groups; |
434d53b0 | 7139 | char str[256]; |
1da177e4 | 7140 | |
968ea6d8 | 7141 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7142 | cpumask_clear(groupmask); |
4dcf6aff IM |
7143 | |
7144 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7145 | ||
7146 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7147 | printk("does not load-balance\n"); | |
7148 | if (sd->parent) | |
7149 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7150 | " has parent"); | |
7151 | return -1; | |
41c7ce9a NP |
7152 | } |
7153 | ||
eefd796a | 7154 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7155 | |
758b2cdc | 7156 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7157 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7158 | "CPU%d\n", cpu); | |
7159 | } | |
758b2cdc | 7160 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7161 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7162 | " CPU%d\n", cpu); | |
7163 | } | |
1da177e4 | 7164 | |
4dcf6aff | 7165 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7166 | do { |
4dcf6aff IM |
7167 | if (!group) { |
7168 | printk("\n"); | |
7169 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7170 | break; |
7171 | } | |
7172 | ||
4dcf6aff IM |
7173 | if (!group->__cpu_power) { |
7174 | printk(KERN_CONT "\n"); | |
7175 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7176 | "set\n"); | |
7177 | break; | |
7178 | } | |
1da177e4 | 7179 | |
758b2cdc | 7180 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7181 | printk(KERN_CONT "\n"); |
7182 | printk(KERN_ERR "ERROR: empty group\n"); | |
7183 | break; | |
7184 | } | |
1da177e4 | 7185 | |
758b2cdc | 7186 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7187 | printk(KERN_CONT "\n"); |
7188 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7189 | break; | |
7190 | } | |
1da177e4 | 7191 | |
758b2cdc | 7192 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7193 | |
968ea6d8 | 7194 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
4dcf6aff | 7195 | printk(KERN_CONT " %s", str); |
1da177e4 | 7196 | |
4dcf6aff IM |
7197 | group = group->next; |
7198 | } while (group != sd->groups); | |
7199 | printk(KERN_CONT "\n"); | |
1da177e4 | 7200 | |
758b2cdc | 7201 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7202 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7203 | |
758b2cdc RR |
7204 | if (sd->parent && |
7205 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7206 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7207 | "of domain->span\n"); | |
7208 | return 0; | |
7209 | } | |
1da177e4 | 7210 | |
4dcf6aff IM |
7211 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7212 | { | |
d5dd3db1 | 7213 | cpumask_var_t groupmask; |
4dcf6aff | 7214 | int level = 0; |
1da177e4 | 7215 | |
4dcf6aff IM |
7216 | if (!sd) { |
7217 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7218 | return; | |
7219 | } | |
1da177e4 | 7220 | |
4dcf6aff IM |
7221 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7222 | ||
d5dd3db1 | 7223 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7224 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7225 | return; | |
7226 | } | |
7227 | ||
4dcf6aff | 7228 | for (;;) { |
7c16ec58 | 7229 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7230 | break; |
1da177e4 LT |
7231 | level++; |
7232 | sd = sd->parent; | |
33859f7f | 7233 | if (!sd) |
4dcf6aff IM |
7234 | break; |
7235 | } | |
d5dd3db1 | 7236 | free_cpumask_var(groupmask); |
1da177e4 | 7237 | } |
6d6bc0ad | 7238 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7239 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7240 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7241 | |
1a20ff27 | 7242 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7243 | { |
758b2cdc | 7244 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7245 | return 1; |
7246 | ||
7247 | /* Following flags need at least 2 groups */ | |
7248 | if (sd->flags & (SD_LOAD_BALANCE | | |
7249 | SD_BALANCE_NEWIDLE | | |
7250 | SD_BALANCE_FORK | | |
89c4710e SS |
7251 | SD_BALANCE_EXEC | |
7252 | SD_SHARE_CPUPOWER | | |
7253 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7254 | if (sd->groups != sd->groups->next) |
7255 | return 0; | |
7256 | } | |
7257 | ||
7258 | /* Following flags don't use groups */ | |
7259 | if (sd->flags & (SD_WAKE_IDLE | | |
7260 | SD_WAKE_AFFINE | | |
7261 | SD_WAKE_BALANCE)) | |
7262 | return 0; | |
7263 | ||
7264 | return 1; | |
7265 | } | |
7266 | ||
48f24c4d IM |
7267 | static int |
7268 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7269 | { |
7270 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7271 | ||
7272 | if (sd_degenerate(parent)) | |
7273 | return 1; | |
7274 | ||
758b2cdc | 7275 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7276 | return 0; |
7277 | ||
7278 | /* Does parent contain flags not in child? */ | |
7279 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7280 | if (cflags & SD_WAKE_AFFINE) | |
7281 | pflags &= ~SD_WAKE_BALANCE; | |
7282 | /* Flags needing groups don't count if only 1 group in parent */ | |
7283 | if (parent->groups == parent->groups->next) { | |
7284 | pflags &= ~(SD_LOAD_BALANCE | | |
7285 | SD_BALANCE_NEWIDLE | | |
7286 | SD_BALANCE_FORK | | |
89c4710e SS |
7287 | SD_BALANCE_EXEC | |
7288 | SD_SHARE_CPUPOWER | | |
7289 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7290 | if (nr_node_ids == 1) |
7291 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7292 | } |
7293 | if (~cflags & pflags) | |
7294 | return 0; | |
7295 | ||
7296 | return 1; | |
7297 | } | |
7298 | ||
c6c4927b RR |
7299 | static void free_rootdomain(struct root_domain *rd) |
7300 | { | |
68e74568 RR |
7301 | cpupri_cleanup(&rd->cpupri); |
7302 | ||
c6c4927b RR |
7303 | free_cpumask_var(rd->rto_mask); |
7304 | free_cpumask_var(rd->online); | |
7305 | free_cpumask_var(rd->span); | |
7306 | kfree(rd); | |
7307 | } | |
7308 | ||
57d885fe GH |
7309 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7310 | { | |
a0490fa3 | 7311 | struct root_domain *old_rd = NULL; |
57d885fe | 7312 | unsigned long flags; |
57d885fe GH |
7313 | |
7314 | spin_lock_irqsave(&rq->lock, flags); | |
7315 | ||
7316 | if (rq->rd) { | |
a0490fa3 | 7317 | old_rd = rq->rd; |
57d885fe | 7318 | |
c6c4927b | 7319 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7320 | set_rq_offline(rq); |
57d885fe | 7321 | |
c6c4927b | 7322 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7323 | |
a0490fa3 IM |
7324 | /* |
7325 | * If we dont want to free the old_rt yet then | |
7326 | * set old_rd to NULL to skip the freeing later | |
7327 | * in this function: | |
7328 | */ | |
7329 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7330 | old_rd = NULL; | |
57d885fe GH |
7331 | } |
7332 | ||
7333 | atomic_inc(&rd->refcount); | |
7334 | rq->rd = rd; | |
7335 | ||
c6c4927b RR |
7336 | cpumask_set_cpu(rq->cpu, rd->span); |
7337 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7338 | set_rq_online(rq); |
57d885fe GH |
7339 | |
7340 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7341 | |
7342 | if (old_rd) | |
7343 | free_rootdomain(old_rd); | |
57d885fe GH |
7344 | } |
7345 | ||
db2f59c8 | 7346 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7347 | { |
7348 | memset(rd, 0, sizeof(*rd)); | |
7349 | ||
c6c4927b RR |
7350 | if (bootmem) { |
7351 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7352 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7353 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7354 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7355 | return 0; |
7356 | } | |
7357 | ||
7358 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7359 | goto out; |
c6c4927b RR |
7360 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7361 | goto free_span; | |
7362 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7363 | goto free_online; | |
6e0534f2 | 7364 | |
68e74568 RR |
7365 | if (cpupri_init(&rd->cpupri, false) != 0) |
7366 | goto free_rto_mask; | |
c6c4927b | 7367 | return 0; |
6e0534f2 | 7368 | |
68e74568 RR |
7369 | free_rto_mask: |
7370 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7371 | free_online: |
7372 | free_cpumask_var(rd->online); | |
7373 | free_span: | |
7374 | free_cpumask_var(rd->span); | |
0c910d28 | 7375 | out: |
c6c4927b | 7376 | return -ENOMEM; |
57d885fe GH |
7377 | } |
7378 | ||
7379 | static void init_defrootdomain(void) | |
7380 | { | |
c6c4927b RR |
7381 | init_rootdomain(&def_root_domain, true); |
7382 | ||
57d885fe GH |
7383 | atomic_set(&def_root_domain.refcount, 1); |
7384 | } | |
7385 | ||
dc938520 | 7386 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7387 | { |
7388 | struct root_domain *rd; | |
7389 | ||
7390 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7391 | if (!rd) | |
7392 | return NULL; | |
7393 | ||
c6c4927b RR |
7394 | if (init_rootdomain(rd, false) != 0) { |
7395 | kfree(rd); | |
7396 | return NULL; | |
7397 | } | |
57d885fe GH |
7398 | |
7399 | return rd; | |
7400 | } | |
7401 | ||
1da177e4 | 7402 | /* |
0eab9146 | 7403 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7404 | * hold the hotplug lock. |
7405 | */ | |
0eab9146 IM |
7406 | static void |
7407 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7408 | { |
70b97a7f | 7409 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7410 | struct sched_domain *tmp; |
7411 | ||
7412 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7413 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7414 | struct sched_domain *parent = tmp->parent; |
7415 | if (!parent) | |
7416 | break; | |
f29c9b1c | 7417 | |
1a848870 | 7418 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7419 | tmp->parent = parent->parent; |
1a848870 SS |
7420 | if (parent->parent) |
7421 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7422 | } else |
7423 | tmp = tmp->parent; | |
245af2c7 SS |
7424 | } |
7425 | ||
1a848870 | 7426 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7427 | sd = sd->parent; |
1a848870 SS |
7428 | if (sd) |
7429 | sd->child = NULL; | |
7430 | } | |
1da177e4 LT |
7431 | |
7432 | sched_domain_debug(sd, cpu); | |
7433 | ||
57d885fe | 7434 | rq_attach_root(rq, rd); |
674311d5 | 7435 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7436 | } |
7437 | ||
7438 | /* cpus with isolated domains */ | |
dcc30a35 | 7439 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7440 | |
7441 | /* Setup the mask of cpus configured for isolated domains */ | |
7442 | static int __init isolated_cpu_setup(char *str) | |
7443 | { | |
968ea6d8 | 7444 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7445 | return 1; |
7446 | } | |
7447 | ||
8927f494 | 7448 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7449 | |
7450 | /* | |
6711cab4 SS |
7451 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7452 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7453 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7454 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7455 | * |
7456 | * init_sched_build_groups will build a circular linked list of the groups | |
7457 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7458 | * and ->cpu_power to 0. | |
7459 | */ | |
a616058b | 7460 | static void |
96f874e2 RR |
7461 | init_sched_build_groups(const struct cpumask *span, |
7462 | const struct cpumask *cpu_map, | |
7463 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7464 | struct sched_group **sg, |
96f874e2 RR |
7465 | struct cpumask *tmpmask), |
7466 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7467 | { |
7468 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7469 | int i; |
7470 | ||
96f874e2 | 7471 | cpumask_clear(covered); |
7c16ec58 | 7472 | |
abcd083a | 7473 | for_each_cpu(i, span) { |
6711cab4 | 7474 | struct sched_group *sg; |
7c16ec58 | 7475 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7476 | int j; |
7477 | ||
758b2cdc | 7478 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7479 | continue; |
7480 | ||
758b2cdc | 7481 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7482 | sg->__cpu_power = 0; |
1da177e4 | 7483 | |
abcd083a | 7484 | for_each_cpu(j, span) { |
7c16ec58 | 7485 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7486 | continue; |
7487 | ||
96f874e2 | 7488 | cpumask_set_cpu(j, covered); |
758b2cdc | 7489 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7490 | } |
7491 | if (!first) | |
7492 | first = sg; | |
7493 | if (last) | |
7494 | last->next = sg; | |
7495 | last = sg; | |
7496 | } | |
7497 | last->next = first; | |
7498 | } | |
7499 | ||
9c1cfda2 | 7500 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7501 | |
9c1cfda2 | 7502 | #ifdef CONFIG_NUMA |
198e2f18 | 7503 | |
9c1cfda2 JH |
7504 | /** |
7505 | * find_next_best_node - find the next node to include in a sched_domain | |
7506 | * @node: node whose sched_domain we're building | |
7507 | * @used_nodes: nodes already in the sched_domain | |
7508 | * | |
41a2d6cf | 7509 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7510 | * finds the closest node not already in the @used_nodes map. |
7511 | * | |
7512 | * Should use nodemask_t. | |
7513 | */ | |
c5f59f08 | 7514 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7515 | { |
7516 | int i, n, val, min_val, best_node = 0; | |
7517 | ||
7518 | min_val = INT_MAX; | |
7519 | ||
076ac2af | 7520 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7521 | /* Start at @node */ |
076ac2af | 7522 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7523 | |
7524 | if (!nr_cpus_node(n)) | |
7525 | continue; | |
7526 | ||
7527 | /* Skip already used nodes */ | |
c5f59f08 | 7528 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7529 | continue; |
7530 | ||
7531 | /* Simple min distance search */ | |
7532 | val = node_distance(node, n); | |
7533 | ||
7534 | if (val < min_val) { | |
7535 | min_val = val; | |
7536 | best_node = n; | |
7537 | } | |
7538 | } | |
7539 | ||
c5f59f08 | 7540 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7541 | return best_node; |
7542 | } | |
7543 | ||
7544 | /** | |
7545 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7546 | * @node: node whose cpumask we're constructing | |
73486722 | 7547 | * @span: resulting cpumask |
9c1cfda2 | 7548 | * |
41a2d6cf | 7549 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7550 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7551 | * out optimally. | |
7552 | */ | |
96f874e2 | 7553 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7554 | { |
c5f59f08 | 7555 | nodemask_t used_nodes; |
48f24c4d | 7556 | int i; |
9c1cfda2 | 7557 | |
6ca09dfc | 7558 | cpumask_clear(span); |
c5f59f08 | 7559 | nodes_clear(used_nodes); |
9c1cfda2 | 7560 | |
6ca09dfc | 7561 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7562 | node_set(node, used_nodes); |
9c1cfda2 JH |
7563 | |
7564 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7565 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7566 | |
6ca09dfc | 7567 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7568 | } |
9c1cfda2 | 7569 | } |
6d6bc0ad | 7570 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7571 | |
5c45bf27 | 7572 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7573 | |
6c99e9ad RR |
7574 | /* |
7575 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7576 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7577 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7578 | */ | |
7579 | struct static_sched_group { | |
7580 | struct sched_group sg; | |
7581 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7582 | }; | |
7583 | ||
7584 | struct static_sched_domain { | |
7585 | struct sched_domain sd; | |
7586 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7587 | }; | |
7588 | ||
9c1cfda2 | 7589 | /* |
48f24c4d | 7590 | * SMT sched-domains: |
9c1cfda2 | 7591 | */ |
1da177e4 | 7592 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7593 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7594 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7595 | |
41a2d6cf | 7596 | static int |
96f874e2 RR |
7597 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7598 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7599 | { |
6711cab4 | 7600 | if (sg) |
6c99e9ad | 7601 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7602 | return cpu; |
7603 | } | |
6d6bc0ad | 7604 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7605 | |
48f24c4d IM |
7606 | /* |
7607 | * multi-core sched-domains: | |
7608 | */ | |
1e9f28fa | 7609 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7610 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7611 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7612 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7613 | |
7614 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7615 | static int |
96f874e2 RR |
7616 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7617 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7618 | { |
6711cab4 | 7619 | int group; |
7c16ec58 | 7620 | |
96f874e2 RR |
7621 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7622 | group = cpumask_first(mask); | |
6711cab4 | 7623 | if (sg) |
6c99e9ad | 7624 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7625 | return group; |
1e9f28fa SS |
7626 | } |
7627 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7628 | static int |
96f874e2 RR |
7629 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7630 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7631 | { |
6711cab4 | 7632 | if (sg) |
6c99e9ad | 7633 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7634 | return cpu; |
7635 | } | |
7636 | #endif | |
7637 | ||
6c99e9ad RR |
7638 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7639 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7640 | |
41a2d6cf | 7641 | static int |
96f874e2 RR |
7642 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7643 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7644 | { |
6711cab4 | 7645 | int group; |
48f24c4d | 7646 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7647 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7648 | group = cpumask_first(mask); |
1e9f28fa | 7649 | #elif defined(CONFIG_SCHED_SMT) |
96f874e2 RR |
7650 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7651 | group = cpumask_first(mask); | |
1da177e4 | 7652 | #else |
6711cab4 | 7653 | group = cpu; |
1da177e4 | 7654 | #endif |
6711cab4 | 7655 | if (sg) |
6c99e9ad | 7656 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7657 | return group; |
1da177e4 LT |
7658 | } |
7659 | ||
7660 | #ifdef CONFIG_NUMA | |
1da177e4 | 7661 | /* |
9c1cfda2 JH |
7662 | * The init_sched_build_groups can't handle what we want to do with node |
7663 | * groups, so roll our own. Now each node has its own list of groups which | |
7664 | * gets dynamically allocated. | |
1da177e4 | 7665 | */ |
62ea9ceb | 7666 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7667 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7668 | |
62ea9ceb | 7669 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7670 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7671 | |
96f874e2 RR |
7672 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7673 | struct sched_group **sg, | |
7674 | struct cpumask *nodemask) | |
9c1cfda2 | 7675 | { |
6711cab4 SS |
7676 | int group; |
7677 | ||
6ca09dfc | 7678 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7679 | group = cpumask_first(nodemask); |
6711cab4 SS |
7680 | |
7681 | if (sg) | |
6c99e9ad | 7682 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7683 | return group; |
1da177e4 | 7684 | } |
6711cab4 | 7685 | |
08069033 SS |
7686 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7687 | { | |
7688 | struct sched_group *sg = group_head; | |
7689 | int j; | |
7690 | ||
7691 | if (!sg) | |
7692 | return; | |
3a5c359a | 7693 | do { |
758b2cdc | 7694 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7695 | struct sched_domain *sd; |
08069033 | 7696 | |
6c99e9ad | 7697 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7698 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7699 | /* |
7700 | * Only add "power" once for each | |
7701 | * physical package. | |
7702 | */ | |
7703 | continue; | |
7704 | } | |
08069033 | 7705 | |
3a5c359a AK |
7706 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7707 | } | |
7708 | sg = sg->next; | |
7709 | } while (sg != group_head); | |
08069033 | 7710 | } |
6d6bc0ad | 7711 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7712 | |
a616058b | 7713 | #ifdef CONFIG_NUMA |
51888ca2 | 7714 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7715 | static void free_sched_groups(const struct cpumask *cpu_map, |
7716 | struct cpumask *nodemask) | |
51888ca2 | 7717 | { |
a616058b | 7718 | int cpu, i; |
51888ca2 | 7719 | |
abcd083a | 7720 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7721 | struct sched_group **sched_group_nodes |
7722 | = sched_group_nodes_bycpu[cpu]; | |
7723 | ||
51888ca2 SV |
7724 | if (!sched_group_nodes) |
7725 | continue; | |
7726 | ||
076ac2af | 7727 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7728 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7729 | ||
6ca09dfc | 7730 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7731 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7732 | continue; |
7733 | ||
7734 | if (sg == NULL) | |
7735 | continue; | |
7736 | sg = sg->next; | |
7737 | next_sg: | |
7738 | oldsg = sg; | |
7739 | sg = sg->next; | |
7740 | kfree(oldsg); | |
7741 | if (oldsg != sched_group_nodes[i]) | |
7742 | goto next_sg; | |
7743 | } | |
7744 | kfree(sched_group_nodes); | |
7745 | sched_group_nodes_bycpu[cpu] = NULL; | |
7746 | } | |
51888ca2 | 7747 | } |
6d6bc0ad | 7748 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7749 | static void free_sched_groups(const struct cpumask *cpu_map, |
7750 | struct cpumask *nodemask) | |
a616058b SS |
7751 | { |
7752 | } | |
6d6bc0ad | 7753 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7754 | |
89c4710e SS |
7755 | /* |
7756 | * Initialize sched groups cpu_power. | |
7757 | * | |
7758 | * cpu_power indicates the capacity of sched group, which is used while | |
7759 | * distributing the load between different sched groups in a sched domain. | |
7760 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7761 | * there are asymmetries in the topology. If there are asymmetries, group | |
7762 | * having more cpu_power will pickup more load compared to the group having | |
7763 | * less cpu_power. | |
7764 | * | |
7765 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7766 | * the maximum number of tasks a group can handle in the presence of other idle | |
7767 | * or lightly loaded groups in the same sched domain. | |
7768 | */ | |
7769 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7770 | { | |
7771 | struct sched_domain *child; | |
7772 | struct sched_group *group; | |
7773 | ||
7774 | WARN_ON(!sd || !sd->groups); | |
7775 | ||
758b2cdc | 7776 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
7777 | return; |
7778 | ||
7779 | child = sd->child; | |
7780 | ||
5517d86b ED |
7781 | sd->groups->__cpu_power = 0; |
7782 | ||
89c4710e SS |
7783 | /* |
7784 | * For perf policy, if the groups in child domain share resources | |
7785 | * (for example cores sharing some portions of the cache hierarchy | |
7786 | * or SMT), then set this domain groups cpu_power such that each group | |
7787 | * can handle only one task, when there are other idle groups in the | |
7788 | * same sched domain. | |
7789 | */ | |
7790 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7791 | (child->flags & | |
7792 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7793 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7794 | return; |
7795 | } | |
7796 | ||
89c4710e SS |
7797 | /* |
7798 | * add cpu_power of each child group to this groups cpu_power | |
7799 | */ | |
7800 | group = child->groups; | |
7801 | do { | |
5517d86b | 7802 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7803 | group = group->next; |
7804 | } while (group != child->groups); | |
7805 | } | |
7806 | ||
7c16ec58 MT |
7807 | /* |
7808 | * Initializers for schedule domains | |
7809 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7810 | */ | |
7811 | ||
a5d8c348 IM |
7812 | #ifdef CONFIG_SCHED_DEBUG |
7813 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7814 | #else | |
7815 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7816 | #endif | |
7817 | ||
7c16ec58 | 7818 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7819 | |
7c16ec58 MT |
7820 | #define SD_INIT_FUNC(type) \ |
7821 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7822 | { \ | |
7823 | memset(sd, 0, sizeof(*sd)); \ | |
7824 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7825 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7826 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7827 | } |
7828 | ||
7829 | SD_INIT_FUNC(CPU) | |
7830 | #ifdef CONFIG_NUMA | |
7831 | SD_INIT_FUNC(ALLNODES) | |
7832 | SD_INIT_FUNC(NODE) | |
7833 | #endif | |
7834 | #ifdef CONFIG_SCHED_SMT | |
7835 | SD_INIT_FUNC(SIBLING) | |
7836 | #endif | |
7837 | #ifdef CONFIG_SCHED_MC | |
7838 | SD_INIT_FUNC(MC) | |
7839 | #endif | |
7840 | ||
1d3504fc HS |
7841 | static int default_relax_domain_level = -1; |
7842 | ||
7843 | static int __init setup_relax_domain_level(char *str) | |
7844 | { | |
30e0e178 LZ |
7845 | unsigned long val; |
7846 | ||
7847 | val = simple_strtoul(str, NULL, 0); | |
7848 | if (val < SD_LV_MAX) | |
7849 | default_relax_domain_level = val; | |
7850 | ||
1d3504fc HS |
7851 | return 1; |
7852 | } | |
7853 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7854 | ||
7855 | static void set_domain_attribute(struct sched_domain *sd, | |
7856 | struct sched_domain_attr *attr) | |
7857 | { | |
7858 | int request; | |
7859 | ||
7860 | if (!attr || attr->relax_domain_level < 0) { | |
7861 | if (default_relax_domain_level < 0) | |
7862 | return; | |
7863 | else | |
7864 | request = default_relax_domain_level; | |
7865 | } else | |
7866 | request = attr->relax_domain_level; | |
7867 | if (request < sd->level) { | |
7868 | /* turn off idle balance on this domain */ | |
7869 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7870 | } else { | |
7871 | /* turn on idle balance on this domain */ | |
7872 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7873 | } | |
7874 | } | |
7875 | ||
1da177e4 | 7876 | /* |
1a20ff27 DG |
7877 | * Build sched domains for a given set of cpus and attach the sched domains |
7878 | * to the individual cpus | |
1da177e4 | 7879 | */ |
96f874e2 | 7880 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 7881 | struct sched_domain_attr *attr) |
1da177e4 | 7882 | { |
3404c8d9 | 7883 | int i, err = -ENOMEM; |
57d885fe | 7884 | struct root_domain *rd; |
3404c8d9 RR |
7885 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
7886 | tmpmask; | |
d1b55138 | 7887 | #ifdef CONFIG_NUMA |
3404c8d9 | 7888 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 7889 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 7890 | int sd_allnodes = 0; |
d1b55138 | 7891 | |
3404c8d9 RR |
7892 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
7893 | goto out; | |
7894 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
7895 | goto free_domainspan; | |
7896 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
7897 | goto free_covered; | |
7898 | #endif | |
7899 | ||
7900 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
7901 | goto free_notcovered; | |
7902 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
7903 | goto free_nodemask; | |
7904 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
7905 | goto free_this_sibling_map; | |
7906 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
7907 | goto free_this_core_map; | |
7908 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
7909 | goto free_send_covered; | |
7910 | ||
7911 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
7912 | /* |
7913 | * Allocate the per-node list of sched groups | |
7914 | */ | |
076ac2af | 7915 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7916 | GFP_KERNEL); |
d1b55138 JH |
7917 | if (!sched_group_nodes) { |
7918 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 7919 | goto free_tmpmask; |
d1b55138 | 7920 | } |
d1b55138 | 7921 | #endif |
1da177e4 | 7922 | |
dc938520 | 7923 | rd = alloc_rootdomain(); |
57d885fe GH |
7924 | if (!rd) { |
7925 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 7926 | goto free_sched_groups; |
57d885fe GH |
7927 | } |
7928 | ||
7c16ec58 | 7929 | #ifdef CONFIG_NUMA |
96f874e2 | 7930 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
7931 | #endif |
7932 | ||
1da177e4 | 7933 | /* |
1a20ff27 | 7934 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7935 | */ |
abcd083a | 7936 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7937 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 7938 | |
6ca09dfc | 7939 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
7940 | |
7941 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
7942 | if (cpumask_weight(cpu_map) > |
7943 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 7944 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 7945 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7946 | set_domain_attribute(sd, attr); |
758b2cdc | 7947 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 7948 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7949 | p = sd; |
6711cab4 | 7950 | sd_allnodes = 1; |
9c1cfda2 JH |
7951 | } else |
7952 | p = NULL; | |
7953 | ||
62ea9ceb | 7954 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 7955 | SD_INIT(sd, NODE); |
1d3504fc | 7956 | set_domain_attribute(sd, attr); |
758b2cdc | 7957 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 7958 | sd->parent = p; |
1a848870 SS |
7959 | if (p) |
7960 | p->child = sd; | |
758b2cdc RR |
7961 | cpumask_and(sched_domain_span(sd), |
7962 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
7963 | #endif |
7964 | ||
7965 | p = sd; | |
6c99e9ad | 7966 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 7967 | SD_INIT(sd, CPU); |
1d3504fc | 7968 | set_domain_attribute(sd, attr); |
758b2cdc | 7969 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 7970 | sd->parent = p; |
1a848870 SS |
7971 | if (p) |
7972 | p->child = sd; | |
7c16ec58 | 7973 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7974 | |
1e9f28fa SS |
7975 | #ifdef CONFIG_SCHED_MC |
7976 | p = sd; | |
6c99e9ad | 7977 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 7978 | SD_INIT(sd, MC); |
1d3504fc | 7979 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
7980 | cpumask_and(sched_domain_span(sd), cpu_map, |
7981 | cpu_coregroup_mask(i)); | |
1e9f28fa | 7982 | sd->parent = p; |
1a848870 | 7983 | p->child = sd; |
7c16ec58 | 7984 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7985 | #endif |
7986 | ||
1da177e4 LT |
7987 | #ifdef CONFIG_SCHED_SMT |
7988 | p = sd; | |
6c99e9ad | 7989 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 7990 | SD_INIT(sd, SIBLING); |
1d3504fc | 7991 | set_domain_attribute(sd, attr); |
758b2cdc RR |
7992 | cpumask_and(sched_domain_span(sd), |
7993 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
1da177e4 | 7994 | sd->parent = p; |
1a848870 | 7995 | p->child = sd; |
7c16ec58 | 7996 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7997 | #endif |
7998 | } | |
7999 | ||
8000 | #ifdef CONFIG_SCHED_SMT | |
8001 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8002 | for_each_cpu(i, cpu_map) { |
96f874e2 RR |
8003 | cpumask_and(this_sibling_map, |
8004 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
8005 | if (i != cpumask_first(this_sibling_map)) | |
1da177e4 LT |
8006 | continue; |
8007 | ||
dd41f596 | 8008 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8009 | &cpu_to_cpu_group, |
8010 | send_covered, tmpmask); | |
1da177e4 LT |
8011 | } |
8012 | #endif | |
8013 | ||
1e9f28fa SS |
8014 | #ifdef CONFIG_SCHED_MC |
8015 | /* Set up multi-core groups */ | |
abcd083a | 8016 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8017 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8018 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8019 | continue; |
7c16ec58 | 8020 | |
dd41f596 | 8021 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8022 | &cpu_to_core_group, |
8023 | send_covered, tmpmask); | |
1e9f28fa SS |
8024 | } |
8025 | #endif | |
8026 | ||
1da177e4 | 8027 | /* Set up physical groups */ |
076ac2af | 8028 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8029 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8030 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8031 | continue; |
8032 | ||
7c16ec58 MT |
8033 | init_sched_build_groups(nodemask, cpu_map, |
8034 | &cpu_to_phys_group, | |
8035 | send_covered, tmpmask); | |
1da177e4 LT |
8036 | } |
8037 | ||
8038 | #ifdef CONFIG_NUMA | |
8039 | /* Set up node groups */ | |
7c16ec58 | 8040 | if (sd_allnodes) { |
7c16ec58 MT |
8041 | init_sched_build_groups(cpu_map, cpu_map, |
8042 | &cpu_to_allnodes_group, | |
8043 | send_covered, tmpmask); | |
8044 | } | |
9c1cfda2 | 8045 | |
076ac2af | 8046 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8047 | /* Set up node groups */ |
8048 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8049 | int j; |
8050 | ||
96f874e2 | 8051 | cpumask_clear(covered); |
6ca09dfc | 8052 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8053 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8054 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8055 | continue; |
d1b55138 | 8056 | } |
9c1cfda2 | 8057 | |
4bdbaad3 | 8058 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8059 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8060 | |
6c99e9ad RR |
8061 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8062 | GFP_KERNEL, i); | |
51888ca2 SV |
8063 | if (!sg) { |
8064 | printk(KERN_WARNING "Can not alloc domain group for " | |
8065 | "node %d\n", i); | |
8066 | goto error; | |
8067 | } | |
9c1cfda2 | 8068 | sched_group_nodes[i] = sg; |
abcd083a | 8069 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8070 | struct sched_domain *sd; |
9761eea8 | 8071 | |
62ea9ceb | 8072 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8073 | sd->groups = sg; |
9c1cfda2 | 8074 | } |
5517d86b | 8075 | sg->__cpu_power = 0; |
758b2cdc | 8076 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8077 | sg->next = sg; |
96f874e2 | 8078 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8079 | prev = sg; |
8080 | ||
076ac2af | 8081 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8082 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8083 | |
96f874e2 RR |
8084 | cpumask_complement(notcovered, covered); |
8085 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8086 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8087 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8088 | break; |
8089 | ||
6ca09dfc | 8090 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8091 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8092 | continue; |
8093 | ||
6c99e9ad RR |
8094 | sg = kmalloc_node(sizeof(struct sched_group) + |
8095 | cpumask_size(), | |
15f0b676 | 8096 | GFP_KERNEL, i); |
9c1cfda2 JH |
8097 | if (!sg) { |
8098 | printk(KERN_WARNING | |
8099 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8100 | goto error; |
9c1cfda2 | 8101 | } |
5517d86b | 8102 | sg->__cpu_power = 0; |
758b2cdc | 8103 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8104 | sg->next = prev->next; |
96f874e2 | 8105 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8106 | prev->next = sg; |
8107 | prev = sg; | |
8108 | } | |
9c1cfda2 | 8109 | } |
1da177e4 LT |
8110 | #endif |
8111 | ||
8112 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8113 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8114 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8115 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8116 | |
89c4710e | 8117 | init_sched_groups_power(i, sd); |
5c45bf27 | 8118 | } |
1da177e4 | 8119 | #endif |
1e9f28fa | 8120 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8121 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8122 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8123 | |
89c4710e | 8124 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8125 | } |
8126 | #endif | |
1e9f28fa | 8127 | |
abcd083a | 8128 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8129 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8130 | |
89c4710e | 8131 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8132 | } |
8133 | ||
9c1cfda2 | 8134 | #ifdef CONFIG_NUMA |
076ac2af | 8135 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8136 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8137 | |
6711cab4 SS |
8138 | if (sd_allnodes) { |
8139 | struct sched_group *sg; | |
f712c0c7 | 8140 | |
96f874e2 | 8141 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8142 | tmpmask); |
f712c0c7 SS |
8143 | init_numa_sched_groups_power(sg); |
8144 | } | |
9c1cfda2 JH |
8145 | #endif |
8146 | ||
1da177e4 | 8147 | /* Attach the domains */ |
abcd083a | 8148 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8149 | struct sched_domain *sd; |
8150 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8151 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8152 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8153 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8154 | #else |
6c99e9ad | 8155 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8156 | #endif |
57d885fe | 8157 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8158 | } |
51888ca2 | 8159 | |
3404c8d9 RR |
8160 | err = 0; |
8161 | ||
8162 | free_tmpmask: | |
8163 | free_cpumask_var(tmpmask); | |
8164 | free_send_covered: | |
8165 | free_cpumask_var(send_covered); | |
8166 | free_this_core_map: | |
8167 | free_cpumask_var(this_core_map); | |
8168 | free_this_sibling_map: | |
8169 | free_cpumask_var(this_sibling_map); | |
8170 | free_nodemask: | |
8171 | free_cpumask_var(nodemask); | |
8172 | free_notcovered: | |
8173 | #ifdef CONFIG_NUMA | |
8174 | free_cpumask_var(notcovered); | |
8175 | free_covered: | |
8176 | free_cpumask_var(covered); | |
8177 | free_domainspan: | |
8178 | free_cpumask_var(domainspan); | |
8179 | out: | |
8180 | #endif | |
8181 | return err; | |
8182 | ||
8183 | free_sched_groups: | |
8184 | #ifdef CONFIG_NUMA | |
8185 | kfree(sched_group_nodes); | |
8186 | #endif | |
8187 | goto free_tmpmask; | |
51888ca2 | 8188 | |
a616058b | 8189 | #ifdef CONFIG_NUMA |
51888ca2 | 8190 | error: |
7c16ec58 | 8191 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8192 | free_rootdomain(rd); |
3404c8d9 | 8193 | goto free_tmpmask; |
a616058b | 8194 | #endif |
1da177e4 | 8195 | } |
029190c5 | 8196 | |
96f874e2 | 8197 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8198 | { |
8199 | return __build_sched_domains(cpu_map, NULL); | |
8200 | } | |
8201 | ||
96f874e2 | 8202 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8203 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8204 | static struct sched_domain_attr *dattr_cur; |
8205 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8206 | |
8207 | /* | |
8208 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8209 | * cpumask) fails, then fallback to a single sched domain, |
8210 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8211 | */ |
4212823f | 8212 | static cpumask_var_t fallback_doms; |
029190c5 | 8213 | |
ee79d1bd HC |
8214 | /* |
8215 | * arch_update_cpu_topology lets virtualized architectures update the | |
8216 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8217 | * or 0 if it stayed the same. | |
8218 | */ | |
8219 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8220 | { |
ee79d1bd | 8221 | return 0; |
22e52b07 HC |
8222 | } |
8223 | ||
1a20ff27 | 8224 | /* |
41a2d6cf | 8225 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8226 | * For now this just excludes isolated cpus, but could be used to |
8227 | * exclude other special cases in the future. | |
1a20ff27 | 8228 | */ |
96f874e2 | 8229 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8230 | { |
7378547f MM |
8231 | int err; |
8232 | ||
22e52b07 | 8233 | arch_update_cpu_topology(); |
029190c5 | 8234 | ndoms_cur = 1; |
96f874e2 | 8235 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8236 | if (!doms_cur) |
4212823f | 8237 | doms_cur = fallback_doms; |
dcc30a35 | 8238 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8239 | dattr_cur = NULL; |
7378547f | 8240 | err = build_sched_domains(doms_cur); |
6382bc90 | 8241 | register_sched_domain_sysctl(); |
7378547f MM |
8242 | |
8243 | return err; | |
1a20ff27 DG |
8244 | } |
8245 | ||
96f874e2 RR |
8246 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8247 | struct cpumask *tmpmask) | |
1da177e4 | 8248 | { |
7c16ec58 | 8249 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8250 | } |
1da177e4 | 8251 | |
1a20ff27 DG |
8252 | /* |
8253 | * Detach sched domains from a group of cpus specified in cpu_map | |
8254 | * These cpus will now be attached to the NULL domain | |
8255 | */ | |
96f874e2 | 8256 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8257 | { |
96f874e2 RR |
8258 | /* Save because hotplug lock held. */ |
8259 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8260 | int i; |
8261 | ||
abcd083a | 8262 | for_each_cpu(i, cpu_map) |
57d885fe | 8263 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8264 | synchronize_sched(); |
96f874e2 | 8265 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8266 | } |
8267 | ||
1d3504fc HS |
8268 | /* handle null as "default" */ |
8269 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8270 | struct sched_domain_attr *new, int idx_new) | |
8271 | { | |
8272 | struct sched_domain_attr tmp; | |
8273 | ||
8274 | /* fast path */ | |
8275 | if (!new && !cur) | |
8276 | return 1; | |
8277 | ||
8278 | tmp = SD_ATTR_INIT; | |
8279 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8280 | new ? (new + idx_new) : &tmp, | |
8281 | sizeof(struct sched_domain_attr)); | |
8282 | } | |
8283 | ||
029190c5 PJ |
8284 | /* |
8285 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8286 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8287 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8288 | * It destroys each deleted domain and builds each new domain. | |
8289 | * | |
96f874e2 | 8290 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8291 | * The masks don't intersect (don't overlap.) We should setup one |
8292 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8293 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8294 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8295 | * it as it is. | |
8296 | * | |
41a2d6cf IM |
8297 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8298 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8299 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8300 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8301 | * the single partition 'fallback_doms', it also forces the domains | |
8302 | * to be rebuilt. | |
029190c5 | 8303 | * |
96f874e2 | 8304 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8305 | * ndoms_new == 0 is a special case for destroying existing domains, |
8306 | * and it will not create the default domain. | |
dfb512ec | 8307 | * |
029190c5 PJ |
8308 | * Call with hotplug lock held |
8309 | */ | |
96f874e2 RR |
8310 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8311 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8312 | struct sched_domain_attr *dattr_new) |
029190c5 | 8313 | { |
dfb512ec | 8314 | int i, j, n; |
d65bd5ec | 8315 | int new_topology; |
029190c5 | 8316 | |
712555ee | 8317 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8318 | |
7378547f MM |
8319 | /* always unregister in case we don't destroy any domains */ |
8320 | unregister_sched_domain_sysctl(); | |
8321 | ||
d65bd5ec HC |
8322 | /* Let architecture update cpu core mappings. */ |
8323 | new_topology = arch_update_cpu_topology(); | |
8324 | ||
dfb512ec | 8325 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8326 | |
8327 | /* Destroy deleted domains */ | |
8328 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8329 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8330 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8331 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8332 | goto match1; |
8333 | } | |
8334 | /* no match - a current sched domain not in new doms_new[] */ | |
8335 | detach_destroy_domains(doms_cur + i); | |
8336 | match1: | |
8337 | ; | |
8338 | } | |
8339 | ||
e761b772 MK |
8340 | if (doms_new == NULL) { |
8341 | ndoms_cur = 0; | |
4212823f | 8342 | doms_new = fallback_doms; |
dcc30a35 | 8343 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8344 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8345 | } |
8346 | ||
029190c5 PJ |
8347 | /* Build new domains */ |
8348 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8349 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8350 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8351 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8352 | goto match2; |
8353 | } | |
8354 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8355 | __build_sched_domains(doms_new + i, |
8356 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8357 | match2: |
8358 | ; | |
8359 | } | |
8360 | ||
8361 | /* Remember the new sched domains */ | |
4212823f | 8362 | if (doms_cur != fallback_doms) |
029190c5 | 8363 | kfree(doms_cur); |
1d3504fc | 8364 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8365 | doms_cur = doms_new; |
1d3504fc | 8366 | dattr_cur = dattr_new; |
029190c5 | 8367 | ndoms_cur = ndoms_new; |
7378547f MM |
8368 | |
8369 | register_sched_domain_sysctl(); | |
a1835615 | 8370 | |
712555ee | 8371 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8372 | } |
8373 | ||
5c45bf27 | 8374 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8375 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8376 | { |
95402b38 | 8377 | get_online_cpus(); |
dfb512ec MK |
8378 | |
8379 | /* Destroy domains first to force the rebuild */ | |
8380 | partition_sched_domains(0, NULL, NULL); | |
8381 | ||
e761b772 | 8382 | rebuild_sched_domains(); |
95402b38 | 8383 | put_online_cpus(); |
5c45bf27 SS |
8384 | } |
8385 | ||
8386 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8387 | { | |
afb8a9b7 | 8388 | unsigned int level = 0; |
5c45bf27 | 8389 | |
afb8a9b7 GS |
8390 | if (sscanf(buf, "%u", &level) != 1) |
8391 | return -EINVAL; | |
8392 | ||
8393 | /* | |
8394 | * level is always be positive so don't check for | |
8395 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8396 | * What happens on 0 or 1 byte write, | |
8397 | * need to check for count as well? | |
8398 | */ | |
8399 | ||
8400 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8401 | return -EINVAL; |
8402 | ||
8403 | if (smt) | |
afb8a9b7 | 8404 | sched_smt_power_savings = level; |
5c45bf27 | 8405 | else |
afb8a9b7 | 8406 | sched_mc_power_savings = level; |
5c45bf27 | 8407 | |
c70f22d2 | 8408 | arch_reinit_sched_domains(); |
5c45bf27 | 8409 | |
c70f22d2 | 8410 | return count; |
5c45bf27 SS |
8411 | } |
8412 | ||
5c45bf27 | 8413 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8414 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8415 | char *page) | |
5c45bf27 SS |
8416 | { |
8417 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8418 | } | |
f718cd4a | 8419 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8420 | const char *buf, size_t count) |
5c45bf27 SS |
8421 | { |
8422 | return sched_power_savings_store(buf, count, 0); | |
8423 | } | |
f718cd4a AK |
8424 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8425 | sched_mc_power_savings_show, | |
8426 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8427 | #endif |
8428 | ||
8429 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8430 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8431 | char *page) | |
5c45bf27 SS |
8432 | { |
8433 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8434 | } | |
f718cd4a | 8435 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8436 | const char *buf, size_t count) |
5c45bf27 SS |
8437 | { |
8438 | return sched_power_savings_store(buf, count, 1); | |
8439 | } | |
f718cd4a AK |
8440 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8441 | sched_smt_power_savings_show, | |
6707de00 AB |
8442 | sched_smt_power_savings_store); |
8443 | #endif | |
8444 | ||
39aac648 | 8445 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8446 | { |
8447 | int err = 0; | |
8448 | ||
8449 | #ifdef CONFIG_SCHED_SMT | |
8450 | if (smt_capable()) | |
8451 | err = sysfs_create_file(&cls->kset.kobj, | |
8452 | &attr_sched_smt_power_savings.attr); | |
8453 | #endif | |
8454 | #ifdef CONFIG_SCHED_MC | |
8455 | if (!err && mc_capable()) | |
8456 | err = sysfs_create_file(&cls->kset.kobj, | |
8457 | &attr_sched_mc_power_savings.attr); | |
8458 | #endif | |
8459 | return err; | |
8460 | } | |
6d6bc0ad | 8461 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8462 | |
e761b772 | 8463 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8464 | /* |
e761b772 MK |
8465 | * Add online and remove offline CPUs from the scheduler domains. |
8466 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8467 | */ |
8468 | static int update_sched_domains(struct notifier_block *nfb, | |
8469 | unsigned long action, void *hcpu) | |
e761b772 MK |
8470 | { |
8471 | switch (action) { | |
8472 | case CPU_ONLINE: | |
8473 | case CPU_ONLINE_FROZEN: | |
8474 | case CPU_DEAD: | |
8475 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8476 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8477 | return NOTIFY_OK; |
8478 | ||
8479 | default: | |
8480 | return NOTIFY_DONE; | |
8481 | } | |
8482 | } | |
8483 | #endif | |
8484 | ||
8485 | static int update_runtime(struct notifier_block *nfb, | |
8486 | unsigned long action, void *hcpu) | |
1da177e4 | 8487 | { |
7def2be1 PZ |
8488 | int cpu = (int)(long)hcpu; |
8489 | ||
1da177e4 | 8490 | switch (action) { |
1da177e4 | 8491 | case CPU_DOWN_PREPARE: |
8bb78442 | 8492 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8493 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8494 | return NOTIFY_OK; |
8495 | ||
1da177e4 | 8496 | case CPU_DOWN_FAILED: |
8bb78442 | 8497 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8498 | case CPU_ONLINE: |
8bb78442 | 8499 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8500 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8501 | return NOTIFY_OK; |
8502 | ||
1da177e4 LT |
8503 | default: |
8504 | return NOTIFY_DONE; | |
8505 | } | |
1da177e4 | 8506 | } |
1da177e4 LT |
8507 | |
8508 | void __init sched_init_smp(void) | |
8509 | { | |
dcc30a35 RR |
8510 | cpumask_var_t non_isolated_cpus; |
8511 | ||
8512 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8513 | |
434d53b0 MT |
8514 | #if defined(CONFIG_NUMA) |
8515 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8516 | GFP_KERNEL); | |
8517 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8518 | #endif | |
95402b38 | 8519 | get_online_cpus(); |
712555ee | 8520 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8521 | arch_init_sched_domains(cpu_online_mask); |
8522 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8523 | if (cpumask_empty(non_isolated_cpus)) | |
8524 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8525 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8526 | put_online_cpus(); |
e761b772 MK |
8527 | |
8528 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8529 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8530 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8531 | #endif |
8532 | ||
8533 | /* RT runtime code needs to handle some hotplug events */ | |
8534 | hotcpu_notifier(update_runtime, 0); | |
8535 | ||
b328ca18 | 8536 | init_hrtick(); |
5c1e1767 NP |
8537 | |
8538 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8539 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8540 | BUG(); |
19978ca6 | 8541 | sched_init_granularity(); |
dcc30a35 | 8542 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8543 | |
8544 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8545 | init_sched_rt_class(); |
1da177e4 LT |
8546 | } |
8547 | #else | |
8548 | void __init sched_init_smp(void) | |
8549 | { | |
19978ca6 | 8550 | sched_init_granularity(); |
1da177e4 LT |
8551 | } |
8552 | #endif /* CONFIG_SMP */ | |
8553 | ||
8554 | int in_sched_functions(unsigned long addr) | |
8555 | { | |
1da177e4 LT |
8556 | return in_lock_functions(addr) || |
8557 | (addr >= (unsigned long)__sched_text_start | |
8558 | && addr < (unsigned long)__sched_text_end); | |
8559 | } | |
8560 | ||
a9957449 | 8561 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8562 | { |
8563 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8564 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8565 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8566 | cfs_rq->rq = rq; | |
8567 | #endif | |
67e9fb2a | 8568 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8569 | } |
8570 | ||
fa85ae24 PZ |
8571 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8572 | { | |
8573 | struct rt_prio_array *array; | |
8574 | int i; | |
8575 | ||
8576 | array = &rt_rq->active; | |
8577 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8578 | INIT_LIST_HEAD(array->queue + i); | |
8579 | __clear_bit(i, array->bitmap); | |
8580 | } | |
8581 | /* delimiter for bitsearch: */ | |
8582 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8583 | ||
052f1dc7 | 8584 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8585 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8586 | #ifdef CONFIG_SMP |
e864c499 | 8587 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8588 | #endif |
48d5e258 | 8589 | #endif |
fa85ae24 PZ |
8590 | #ifdef CONFIG_SMP |
8591 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8592 | rt_rq->overloaded = 0; |
917b627d | 8593 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8594 | #endif |
8595 | ||
8596 | rt_rq->rt_time = 0; | |
8597 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8598 | rt_rq->rt_runtime = 0; |
8599 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8600 | |
052f1dc7 | 8601 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8602 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8603 | rt_rq->rq = rq; |
8604 | #endif | |
fa85ae24 PZ |
8605 | } |
8606 | ||
6f505b16 | 8607 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8608 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8609 | struct sched_entity *se, int cpu, int add, | |
8610 | struct sched_entity *parent) | |
6f505b16 | 8611 | { |
ec7dc8ac | 8612 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8613 | tg->cfs_rq[cpu] = cfs_rq; |
8614 | init_cfs_rq(cfs_rq, rq); | |
8615 | cfs_rq->tg = tg; | |
8616 | if (add) | |
8617 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8618 | ||
8619 | tg->se[cpu] = se; | |
354d60c2 DG |
8620 | /* se could be NULL for init_task_group */ |
8621 | if (!se) | |
8622 | return; | |
8623 | ||
ec7dc8ac DG |
8624 | if (!parent) |
8625 | se->cfs_rq = &rq->cfs; | |
8626 | else | |
8627 | se->cfs_rq = parent->my_q; | |
8628 | ||
6f505b16 PZ |
8629 | se->my_q = cfs_rq; |
8630 | se->load.weight = tg->shares; | |
e05510d0 | 8631 | se->load.inv_weight = 0; |
ec7dc8ac | 8632 | se->parent = parent; |
6f505b16 | 8633 | } |
052f1dc7 | 8634 | #endif |
6f505b16 | 8635 | |
052f1dc7 | 8636 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8637 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8638 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8639 | struct sched_rt_entity *parent) | |
6f505b16 | 8640 | { |
ec7dc8ac DG |
8641 | struct rq *rq = cpu_rq(cpu); |
8642 | ||
6f505b16 PZ |
8643 | tg->rt_rq[cpu] = rt_rq; |
8644 | init_rt_rq(rt_rq, rq); | |
8645 | rt_rq->tg = tg; | |
8646 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8647 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8648 | if (add) |
8649 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8650 | ||
8651 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8652 | if (!rt_se) |
8653 | return; | |
8654 | ||
ec7dc8ac DG |
8655 | if (!parent) |
8656 | rt_se->rt_rq = &rq->rt; | |
8657 | else | |
8658 | rt_se->rt_rq = parent->my_q; | |
8659 | ||
6f505b16 | 8660 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8661 | rt_se->parent = parent; |
6f505b16 PZ |
8662 | INIT_LIST_HEAD(&rt_se->run_list); |
8663 | } | |
8664 | #endif | |
8665 | ||
1da177e4 LT |
8666 | void __init sched_init(void) |
8667 | { | |
dd41f596 | 8668 | int i, j; |
434d53b0 MT |
8669 | unsigned long alloc_size = 0, ptr; |
8670 | ||
8671 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8672 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8673 | #endif | |
8674 | #ifdef CONFIG_RT_GROUP_SCHED | |
8675 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8676 | #endif |
8677 | #ifdef CONFIG_USER_SCHED | |
8678 | alloc_size *= 2; | |
434d53b0 MT |
8679 | #endif |
8680 | /* | |
8681 | * As sched_init() is called before page_alloc is setup, | |
8682 | * we use alloc_bootmem(). | |
8683 | */ | |
8684 | if (alloc_size) { | |
5a9d3225 | 8685 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8686 | |
8687 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8688 | init_task_group.se = (struct sched_entity **)ptr; | |
8689 | ptr += nr_cpu_ids * sizeof(void **); | |
8690 | ||
8691 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8692 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8693 | |
8694 | #ifdef CONFIG_USER_SCHED | |
8695 | root_task_group.se = (struct sched_entity **)ptr; | |
8696 | ptr += nr_cpu_ids * sizeof(void **); | |
8697 | ||
8698 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8699 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8700 | #endif /* CONFIG_USER_SCHED */ |
8701 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8702 | #ifdef CONFIG_RT_GROUP_SCHED |
8703 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8704 | ptr += nr_cpu_ids * sizeof(void **); | |
8705 | ||
8706 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8707 | ptr += nr_cpu_ids * sizeof(void **); |
8708 | ||
8709 | #ifdef CONFIG_USER_SCHED | |
8710 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8711 | ptr += nr_cpu_ids * sizeof(void **); | |
8712 | ||
8713 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8714 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8715 | #endif /* CONFIG_USER_SCHED */ |
8716 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8717 | } |
dd41f596 | 8718 | |
57d885fe GH |
8719 | #ifdef CONFIG_SMP |
8720 | init_defrootdomain(); | |
8721 | #endif | |
8722 | ||
d0b27fa7 PZ |
8723 | init_rt_bandwidth(&def_rt_bandwidth, |
8724 | global_rt_period(), global_rt_runtime()); | |
8725 | ||
8726 | #ifdef CONFIG_RT_GROUP_SCHED | |
8727 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8728 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8729 | #ifdef CONFIG_USER_SCHED |
8730 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8731 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8732 | #endif /* CONFIG_USER_SCHED */ |
8733 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8734 | |
052f1dc7 | 8735 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8736 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8737 | INIT_LIST_HEAD(&init_task_group.children); |
8738 | ||
8739 | #ifdef CONFIG_USER_SCHED | |
8740 | INIT_LIST_HEAD(&root_task_group.children); | |
8741 | init_task_group.parent = &root_task_group; | |
8742 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8743 | #endif /* CONFIG_USER_SCHED */ |
8744 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8745 | |
0a945022 | 8746 | for_each_possible_cpu(i) { |
70b97a7f | 8747 | struct rq *rq; |
1da177e4 LT |
8748 | |
8749 | rq = cpu_rq(i); | |
8750 | spin_lock_init(&rq->lock); | |
7897986b | 8751 | rq->nr_running = 0; |
dd41f596 | 8752 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8753 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8754 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8755 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8756 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8757 | #ifdef CONFIG_CGROUP_SCHED |
8758 | /* | |
8759 | * How much cpu bandwidth does init_task_group get? | |
8760 | * | |
8761 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8762 | * gets 100% of the cpu resources in the system. This overall | |
8763 | * system cpu resource is divided among the tasks of | |
8764 | * init_task_group and its child task-groups in a fair manner, | |
8765 | * based on each entity's (task or task-group's) weight | |
8766 | * (se->load.weight). | |
8767 | * | |
8768 | * In other words, if init_task_group has 10 tasks of weight | |
8769 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8770 | * then A0's share of the cpu resource is: | |
8771 | * | |
8772 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8773 | * | |
8774 | * We achieve this by letting init_task_group's tasks sit | |
8775 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8776 | */ | |
ec7dc8ac | 8777 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8778 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8779 | root_task_group.shares = NICE_0_LOAD; |
8780 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8781 | /* |
8782 | * In case of task-groups formed thr' the user id of tasks, | |
8783 | * init_task_group represents tasks belonging to root user. | |
8784 | * Hence it forms a sibling of all subsequent groups formed. | |
8785 | * In this case, init_task_group gets only a fraction of overall | |
8786 | * system cpu resource, based on the weight assigned to root | |
8787 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8788 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8789 | * (init_cfs_rq) and having one entity represent this group of | |
8790 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8791 | */ | |
ec7dc8ac | 8792 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8793 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8794 | &per_cpu(init_sched_entity, i), i, 1, |
8795 | root_task_group.se[i]); | |
6f505b16 | 8796 | |
052f1dc7 | 8797 | #endif |
354d60c2 DG |
8798 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8799 | ||
8800 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8801 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8802 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8803 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8804 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8805 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8806 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8807 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8808 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8809 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8810 | root_task_group.rt_se[i]); | |
354d60c2 | 8811 | #endif |
dd41f596 | 8812 | #endif |
1da177e4 | 8813 | |
dd41f596 IM |
8814 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8815 | rq->cpu_load[j] = 0; | |
1da177e4 | 8816 | #ifdef CONFIG_SMP |
41c7ce9a | 8817 | rq->sd = NULL; |
57d885fe | 8818 | rq->rd = NULL; |
1da177e4 | 8819 | rq->active_balance = 0; |
dd41f596 | 8820 | rq->next_balance = jiffies; |
1da177e4 | 8821 | rq->push_cpu = 0; |
0a2966b4 | 8822 | rq->cpu = i; |
1f11eb6a | 8823 | rq->online = 0; |
1da177e4 LT |
8824 | rq->migration_thread = NULL; |
8825 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8826 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8827 | #endif |
8f4d37ec | 8828 | init_rq_hrtick(rq); |
1da177e4 | 8829 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8830 | } |
8831 | ||
2dd73a4f | 8832 | set_load_weight(&init_task); |
b50f60ce | 8833 | |
e107be36 AK |
8834 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8835 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8836 | #endif | |
8837 | ||
c9819f45 | 8838 | #ifdef CONFIG_SMP |
962cf36c | 8839 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8840 | #endif |
8841 | ||
b50f60ce HC |
8842 | #ifdef CONFIG_RT_MUTEXES |
8843 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8844 | #endif | |
8845 | ||
1da177e4 LT |
8846 | /* |
8847 | * The boot idle thread does lazy MMU switching as well: | |
8848 | */ | |
8849 | atomic_inc(&init_mm.mm_count); | |
8850 | enter_lazy_tlb(&init_mm, current); | |
8851 | ||
8852 | /* | |
8853 | * Make us the idle thread. Technically, schedule() should not be | |
8854 | * called from this thread, however somewhere below it might be, | |
8855 | * but because we are the idle thread, we just pick up running again | |
8856 | * when this runqueue becomes "idle". | |
8857 | */ | |
8858 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8859 | /* |
8860 | * During early bootup we pretend to be a normal task: | |
8861 | */ | |
8862 | current->sched_class = &fair_sched_class; | |
6892b75e | 8863 | |
6a7b3dc3 RR |
8864 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
8865 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 8866 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
8867 | #ifdef CONFIG_NO_HZ |
8868 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
8869 | #endif | |
dcc30a35 | 8870 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 8871 | #endif /* SMP */ |
6a7b3dc3 | 8872 | |
6892b75e | 8873 | scheduler_running = 1; |
1da177e4 LT |
8874 | } |
8875 | ||
8876 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8877 | void __might_sleep(char *file, int line) | |
8878 | { | |
48f24c4d | 8879 | #ifdef in_atomic |
1da177e4 LT |
8880 | static unsigned long prev_jiffy; /* ratelimiting */ |
8881 | ||
aef745fc IM |
8882 | if ((!in_atomic() && !irqs_disabled()) || |
8883 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8884 | return; | |
8885 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8886 | return; | |
8887 | prev_jiffy = jiffies; | |
8888 | ||
8889 | printk(KERN_ERR | |
8890 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8891 | file, line); | |
8892 | printk(KERN_ERR | |
8893 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8894 | in_atomic(), irqs_disabled(), | |
8895 | current->pid, current->comm); | |
8896 | ||
8897 | debug_show_held_locks(current); | |
8898 | if (irqs_disabled()) | |
8899 | print_irqtrace_events(current); | |
8900 | dump_stack(); | |
1da177e4 LT |
8901 | #endif |
8902 | } | |
8903 | EXPORT_SYMBOL(__might_sleep); | |
8904 | #endif | |
8905 | ||
8906 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8907 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8908 | { | |
8909 | int on_rq; | |
3e51f33f | 8910 | |
3a5e4dc1 AK |
8911 | update_rq_clock(rq); |
8912 | on_rq = p->se.on_rq; | |
8913 | if (on_rq) | |
8914 | deactivate_task(rq, p, 0); | |
8915 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8916 | if (on_rq) { | |
8917 | activate_task(rq, p, 0); | |
8918 | resched_task(rq->curr); | |
8919 | } | |
8920 | } | |
8921 | ||
1da177e4 LT |
8922 | void normalize_rt_tasks(void) |
8923 | { | |
a0f98a1c | 8924 | struct task_struct *g, *p; |
1da177e4 | 8925 | unsigned long flags; |
70b97a7f | 8926 | struct rq *rq; |
1da177e4 | 8927 | |
4cf5d77a | 8928 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8929 | do_each_thread(g, p) { |
178be793 IM |
8930 | /* |
8931 | * Only normalize user tasks: | |
8932 | */ | |
8933 | if (!p->mm) | |
8934 | continue; | |
8935 | ||
6cfb0d5d | 8936 | p->se.exec_start = 0; |
6cfb0d5d | 8937 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8938 | p->se.wait_start = 0; |
dd41f596 | 8939 | p->se.sleep_start = 0; |
dd41f596 | 8940 | p->se.block_start = 0; |
6cfb0d5d | 8941 | #endif |
dd41f596 IM |
8942 | |
8943 | if (!rt_task(p)) { | |
8944 | /* | |
8945 | * Renice negative nice level userspace | |
8946 | * tasks back to 0: | |
8947 | */ | |
8948 | if (TASK_NICE(p) < 0 && p->mm) | |
8949 | set_user_nice(p, 0); | |
1da177e4 | 8950 | continue; |
dd41f596 | 8951 | } |
1da177e4 | 8952 | |
4cf5d77a | 8953 | spin_lock(&p->pi_lock); |
b29739f9 | 8954 | rq = __task_rq_lock(p); |
1da177e4 | 8955 | |
178be793 | 8956 | normalize_task(rq, p); |
3a5e4dc1 | 8957 | |
b29739f9 | 8958 | __task_rq_unlock(rq); |
4cf5d77a | 8959 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8960 | } while_each_thread(g, p); |
8961 | ||
4cf5d77a | 8962 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8963 | } |
8964 | ||
8965 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8966 | |
8967 | #ifdef CONFIG_IA64 | |
8968 | /* | |
8969 | * These functions are only useful for the IA64 MCA handling. | |
8970 | * | |
8971 | * They can only be called when the whole system has been | |
8972 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8973 | * activity can take place. Using them for anything else would | |
8974 | * be a serious bug, and as a result, they aren't even visible | |
8975 | * under any other configuration. | |
8976 | */ | |
8977 | ||
8978 | /** | |
8979 | * curr_task - return the current task for a given cpu. | |
8980 | * @cpu: the processor in question. | |
8981 | * | |
8982 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8983 | */ | |
36c8b586 | 8984 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8985 | { |
8986 | return cpu_curr(cpu); | |
8987 | } | |
8988 | ||
8989 | /** | |
8990 | * set_curr_task - set the current task for a given cpu. | |
8991 | * @cpu: the processor in question. | |
8992 | * @p: the task pointer to set. | |
8993 | * | |
8994 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8995 | * are serviced on a separate stack. It allows the architecture to switch the |
8996 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8997 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8998 | * and caller must save the original value of the current task (see | |
8999 | * curr_task() above) and restore that value before reenabling interrupts and | |
9000 | * re-starting the system. | |
9001 | * | |
9002 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9003 | */ | |
36c8b586 | 9004 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9005 | { |
9006 | cpu_curr(cpu) = p; | |
9007 | } | |
9008 | ||
9009 | #endif | |
29f59db3 | 9010 | |
bccbe08a PZ |
9011 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9012 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9013 | { |
9014 | int i; | |
9015 | ||
9016 | for_each_possible_cpu(i) { | |
9017 | if (tg->cfs_rq) | |
9018 | kfree(tg->cfs_rq[i]); | |
9019 | if (tg->se) | |
9020 | kfree(tg->se[i]); | |
6f505b16 PZ |
9021 | } |
9022 | ||
9023 | kfree(tg->cfs_rq); | |
9024 | kfree(tg->se); | |
6f505b16 PZ |
9025 | } |
9026 | ||
ec7dc8ac DG |
9027 | static |
9028 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9029 | { |
29f59db3 | 9030 | struct cfs_rq *cfs_rq; |
eab17229 | 9031 | struct sched_entity *se; |
9b5b7751 | 9032 | struct rq *rq; |
29f59db3 SV |
9033 | int i; |
9034 | ||
434d53b0 | 9035 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9036 | if (!tg->cfs_rq) |
9037 | goto err; | |
434d53b0 | 9038 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9039 | if (!tg->se) |
9040 | goto err; | |
052f1dc7 PZ |
9041 | |
9042 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9043 | |
9044 | for_each_possible_cpu(i) { | |
9b5b7751 | 9045 | rq = cpu_rq(i); |
29f59db3 | 9046 | |
eab17229 LZ |
9047 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9048 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9049 | if (!cfs_rq) |
9050 | goto err; | |
9051 | ||
eab17229 LZ |
9052 | se = kzalloc_node(sizeof(struct sched_entity), |
9053 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9054 | if (!se) |
9055 | goto err; | |
9056 | ||
eab17229 | 9057 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9058 | } |
9059 | ||
9060 | return 1; | |
9061 | ||
9062 | err: | |
9063 | return 0; | |
9064 | } | |
9065 | ||
9066 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9067 | { | |
9068 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9069 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9070 | } | |
9071 | ||
9072 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9073 | { | |
9074 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9075 | } | |
6d6bc0ad | 9076 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9077 | static inline void free_fair_sched_group(struct task_group *tg) |
9078 | { | |
9079 | } | |
9080 | ||
ec7dc8ac DG |
9081 | static inline |
9082 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9083 | { |
9084 | return 1; | |
9085 | } | |
9086 | ||
9087 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9088 | { | |
9089 | } | |
9090 | ||
9091 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9092 | { | |
9093 | } | |
6d6bc0ad | 9094 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9095 | |
9096 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9097 | static void free_rt_sched_group(struct task_group *tg) |
9098 | { | |
9099 | int i; | |
9100 | ||
d0b27fa7 PZ |
9101 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9102 | ||
bccbe08a PZ |
9103 | for_each_possible_cpu(i) { |
9104 | if (tg->rt_rq) | |
9105 | kfree(tg->rt_rq[i]); | |
9106 | if (tg->rt_se) | |
9107 | kfree(tg->rt_se[i]); | |
9108 | } | |
9109 | ||
9110 | kfree(tg->rt_rq); | |
9111 | kfree(tg->rt_se); | |
9112 | } | |
9113 | ||
ec7dc8ac DG |
9114 | static |
9115 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9116 | { |
9117 | struct rt_rq *rt_rq; | |
eab17229 | 9118 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9119 | struct rq *rq; |
9120 | int i; | |
9121 | ||
434d53b0 | 9122 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9123 | if (!tg->rt_rq) |
9124 | goto err; | |
434d53b0 | 9125 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9126 | if (!tg->rt_se) |
9127 | goto err; | |
9128 | ||
d0b27fa7 PZ |
9129 | init_rt_bandwidth(&tg->rt_bandwidth, |
9130 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9131 | |
9132 | for_each_possible_cpu(i) { | |
9133 | rq = cpu_rq(i); | |
9134 | ||
eab17229 LZ |
9135 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9136 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9137 | if (!rt_rq) |
9138 | goto err; | |
29f59db3 | 9139 | |
eab17229 LZ |
9140 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9141 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9142 | if (!rt_se) |
9143 | goto err; | |
29f59db3 | 9144 | |
eab17229 | 9145 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9146 | } |
9147 | ||
bccbe08a PZ |
9148 | return 1; |
9149 | ||
9150 | err: | |
9151 | return 0; | |
9152 | } | |
9153 | ||
9154 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9155 | { | |
9156 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9157 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9158 | } | |
9159 | ||
9160 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9161 | { | |
9162 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9163 | } | |
6d6bc0ad | 9164 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9165 | static inline void free_rt_sched_group(struct task_group *tg) |
9166 | { | |
9167 | } | |
9168 | ||
ec7dc8ac DG |
9169 | static inline |
9170 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9171 | { |
9172 | return 1; | |
9173 | } | |
9174 | ||
9175 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9176 | { | |
9177 | } | |
9178 | ||
9179 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9180 | { | |
9181 | } | |
6d6bc0ad | 9182 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9183 | |
d0b27fa7 | 9184 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9185 | static void free_sched_group(struct task_group *tg) |
9186 | { | |
9187 | free_fair_sched_group(tg); | |
9188 | free_rt_sched_group(tg); | |
9189 | kfree(tg); | |
9190 | } | |
9191 | ||
9192 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9193 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9194 | { |
9195 | struct task_group *tg; | |
9196 | unsigned long flags; | |
9197 | int i; | |
9198 | ||
9199 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9200 | if (!tg) | |
9201 | return ERR_PTR(-ENOMEM); | |
9202 | ||
ec7dc8ac | 9203 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9204 | goto err; |
9205 | ||
ec7dc8ac | 9206 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9207 | goto err; |
9208 | ||
8ed36996 | 9209 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9210 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9211 | register_fair_sched_group(tg, i); |
9212 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9213 | } |
6f505b16 | 9214 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9215 | |
9216 | WARN_ON(!parent); /* root should already exist */ | |
9217 | ||
9218 | tg->parent = parent; | |
f473aa5e | 9219 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9220 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9221 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9222 | |
9b5b7751 | 9223 | return tg; |
29f59db3 SV |
9224 | |
9225 | err: | |
6f505b16 | 9226 | free_sched_group(tg); |
29f59db3 SV |
9227 | return ERR_PTR(-ENOMEM); |
9228 | } | |
9229 | ||
9b5b7751 | 9230 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9231 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9232 | { |
29f59db3 | 9233 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9234 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9235 | } |
9236 | ||
9b5b7751 | 9237 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9238 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9239 | { |
8ed36996 | 9240 | unsigned long flags; |
9b5b7751 | 9241 | int i; |
29f59db3 | 9242 | |
8ed36996 | 9243 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9244 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9245 | unregister_fair_sched_group(tg, i); |
9246 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9247 | } |
6f505b16 | 9248 | list_del_rcu(&tg->list); |
f473aa5e | 9249 | list_del_rcu(&tg->siblings); |
8ed36996 | 9250 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9251 | |
9b5b7751 | 9252 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9253 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9254 | } |
9255 | ||
9b5b7751 | 9256 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9257 | * The caller of this function should have put the task in its new group |
9258 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9259 | * reflect its new group. | |
9b5b7751 SV |
9260 | */ |
9261 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9262 | { |
9263 | int on_rq, running; | |
9264 | unsigned long flags; | |
9265 | struct rq *rq; | |
9266 | ||
9267 | rq = task_rq_lock(tsk, &flags); | |
9268 | ||
29f59db3 SV |
9269 | update_rq_clock(rq); |
9270 | ||
051a1d1a | 9271 | running = task_current(rq, tsk); |
29f59db3 SV |
9272 | on_rq = tsk->se.on_rq; |
9273 | ||
0e1f3483 | 9274 | if (on_rq) |
29f59db3 | 9275 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9276 | if (unlikely(running)) |
9277 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9278 | |
6f505b16 | 9279 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9280 | |
810b3817 PZ |
9281 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9282 | if (tsk->sched_class->moved_group) | |
9283 | tsk->sched_class->moved_group(tsk); | |
9284 | #endif | |
9285 | ||
0e1f3483 HS |
9286 | if (unlikely(running)) |
9287 | tsk->sched_class->set_curr_task(rq); | |
9288 | if (on_rq) | |
7074badb | 9289 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9290 | |
29f59db3 SV |
9291 | task_rq_unlock(rq, &flags); |
9292 | } | |
6d6bc0ad | 9293 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9294 | |
052f1dc7 | 9295 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9296 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9297 | { |
9298 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9299 | int on_rq; |
9300 | ||
29f59db3 | 9301 | on_rq = se->on_rq; |
62fb1851 | 9302 | if (on_rq) |
29f59db3 SV |
9303 | dequeue_entity(cfs_rq, se, 0); |
9304 | ||
9305 | se->load.weight = shares; | |
e05510d0 | 9306 | se->load.inv_weight = 0; |
29f59db3 | 9307 | |
62fb1851 | 9308 | if (on_rq) |
29f59db3 | 9309 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9310 | } |
62fb1851 | 9311 | |
c09595f6 PZ |
9312 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9313 | { | |
9314 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9315 | struct rq *rq = cfs_rq->rq; | |
9316 | unsigned long flags; | |
9317 | ||
9318 | spin_lock_irqsave(&rq->lock, flags); | |
9319 | __set_se_shares(se, shares); | |
9320 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9321 | } |
9322 | ||
8ed36996 PZ |
9323 | static DEFINE_MUTEX(shares_mutex); |
9324 | ||
4cf86d77 | 9325 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9326 | { |
9327 | int i; | |
8ed36996 | 9328 | unsigned long flags; |
c61935fd | 9329 | |
ec7dc8ac DG |
9330 | /* |
9331 | * We can't change the weight of the root cgroup. | |
9332 | */ | |
9333 | if (!tg->se[0]) | |
9334 | return -EINVAL; | |
9335 | ||
18d95a28 PZ |
9336 | if (shares < MIN_SHARES) |
9337 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9338 | else if (shares > MAX_SHARES) |
9339 | shares = MAX_SHARES; | |
62fb1851 | 9340 | |
8ed36996 | 9341 | mutex_lock(&shares_mutex); |
9b5b7751 | 9342 | if (tg->shares == shares) |
5cb350ba | 9343 | goto done; |
29f59db3 | 9344 | |
8ed36996 | 9345 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9346 | for_each_possible_cpu(i) |
9347 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9348 | list_del_rcu(&tg->siblings); |
8ed36996 | 9349 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9350 | |
9351 | /* wait for any ongoing reference to this group to finish */ | |
9352 | synchronize_sched(); | |
9353 | ||
9354 | /* | |
9355 | * Now we are free to modify the group's share on each cpu | |
9356 | * w/o tripping rebalance_share or load_balance_fair. | |
9357 | */ | |
9b5b7751 | 9358 | tg->shares = shares; |
c09595f6 PZ |
9359 | for_each_possible_cpu(i) { |
9360 | /* | |
9361 | * force a rebalance | |
9362 | */ | |
9363 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9364 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9365 | } |
29f59db3 | 9366 | |
6b2d7700 SV |
9367 | /* |
9368 | * Enable load balance activity on this group, by inserting it back on | |
9369 | * each cpu's rq->leaf_cfs_rq_list. | |
9370 | */ | |
8ed36996 | 9371 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9372 | for_each_possible_cpu(i) |
9373 | register_fair_sched_group(tg, i); | |
f473aa5e | 9374 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9375 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9376 | done: |
8ed36996 | 9377 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9378 | return 0; |
29f59db3 SV |
9379 | } |
9380 | ||
5cb350ba DG |
9381 | unsigned long sched_group_shares(struct task_group *tg) |
9382 | { | |
9383 | return tg->shares; | |
9384 | } | |
052f1dc7 | 9385 | #endif |
5cb350ba | 9386 | |
052f1dc7 | 9387 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9388 | /* |
9f0c1e56 | 9389 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9390 | */ |
9f0c1e56 PZ |
9391 | static DEFINE_MUTEX(rt_constraints_mutex); |
9392 | ||
9393 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9394 | { | |
9395 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9396 | return 1ULL << 20; |
9f0c1e56 | 9397 | |
9a7e0b18 | 9398 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9399 | } |
9400 | ||
9a7e0b18 PZ |
9401 | /* Must be called with tasklist_lock held */ |
9402 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9403 | { |
9a7e0b18 | 9404 | struct task_struct *g, *p; |
b40b2e8e | 9405 | |
9a7e0b18 PZ |
9406 | do_each_thread(g, p) { |
9407 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9408 | return 1; | |
9409 | } while_each_thread(g, p); | |
b40b2e8e | 9410 | |
9a7e0b18 PZ |
9411 | return 0; |
9412 | } | |
b40b2e8e | 9413 | |
9a7e0b18 PZ |
9414 | struct rt_schedulable_data { |
9415 | struct task_group *tg; | |
9416 | u64 rt_period; | |
9417 | u64 rt_runtime; | |
9418 | }; | |
b40b2e8e | 9419 | |
9a7e0b18 PZ |
9420 | static int tg_schedulable(struct task_group *tg, void *data) |
9421 | { | |
9422 | struct rt_schedulable_data *d = data; | |
9423 | struct task_group *child; | |
9424 | unsigned long total, sum = 0; | |
9425 | u64 period, runtime; | |
b40b2e8e | 9426 | |
9a7e0b18 PZ |
9427 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9428 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9429 | |
9a7e0b18 PZ |
9430 | if (tg == d->tg) { |
9431 | period = d->rt_period; | |
9432 | runtime = d->rt_runtime; | |
b40b2e8e | 9433 | } |
b40b2e8e | 9434 | |
98a4826b PZ |
9435 | #ifdef CONFIG_USER_SCHED |
9436 | if (tg == &root_task_group) { | |
9437 | period = global_rt_period(); | |
9438 | runtime = global_rt_runtime(); | |
9439 | } | |
9440 | #endif | |
9441 | ||
4653f803 PZ |
9442 | /* |
9443 | * Cannot have more runtime than the period. | |
9444 | */ | |
9445 | if (runtime > period && runtime != RUNTIME_INF) | |
9446 | return -EINVAL; | |
6f505b16 | 9447 | |
4653f803 PZ |
9448 | /* |
9449 | * Ensure we don't starve existing RT tasks. | |
9450 | */ | |
9a7e0b18 PZ |
9451 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9452 | return -EBUSY; | |
6f505b16 | 9453 | |
9a7e0b18 | 9454 | total = to_ratio(period, runtime); |
6f505b16 | 9455 | |
4653f803 PZ |
9456 | /* |
9457 | * Nobody can have more than the global setting allows. | |
9458 | */ | |
9459 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9460 | return -EINVAL; | |
6f505b16 | 9461 | |
4653f803 PZ |
9462 | /* |
9463 | * The sum of our children's runtime should not exceed our own. | |
9464 | */ | |
9a7e0b18 PZ |
9465 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9466 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9467 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9468 | |
9a7e0b18 PZ |
9469 | if (child == d->tg) { |
9470 | period = d->rt_period; | |
9471 | runtime = d->rt_runtime; | |
9472 | } | |
6f505b16 | 9473 | |
9a7e0b18 | 9474 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9475 | } |
6f505b16 | 9476 | |
9a7e0b18 PZ |
9477 | if (sum > total) |
9478 | return -EINVAL; | |
9479 | ||
9480 | return 0; | |
6f505b16 PZ |
9481 | } |
9482 | ||
9a7e0b18 | 9483 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9484 | { |
9a7e0b18 PZ |
9485 | struct rt_schedulable_data data = { |
9486 | .tg = tg, | |
9487 | .rt_period = period, | |
9488 | .rt_runtime = runtime, | |
9489 | }; | |
9490 | ||
9491 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9492 | } |
9493 | ||
d0b27fa7 PZ |
9494 | static int tg_set_bandwidth(struct task_group *tg, |
9495 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9496 | { |
ac086bc2 | 9497 | int i, err = 0; |
9f0c1e56 | 9498 | |
9f0c1e56 | 9499 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9500 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9501 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9502 | if (err) | |
9f0c1e56 | 9503 | goto unlock; |
ac086bc2 PZ |
9504 | |
9505 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9506 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9507 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9508 | |
9509 | for_each_possible_cpu(i) { | |
9510 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9511 | ||
9512 | spin_lock(&rt_rq->rt_runtime_lock); | |
9513 | rt_rq->rt_runtime = rt_runtime; | |
9514 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9515 | } | |
9516 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9517 | unlock: |
521f1a24 | 9518 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9519 | mutex_unlock(&rt_constraints_mutex); |
9520 | ||
9521 | return err; | |
6f505b16 PZ |
9522 | } |
9523 | ||
d0b27fa7 PZ |
9524 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9525 | { | |
9526 | u64 rt_runtime, rt_period; | |
9527 | ||
9528 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9529 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9530 | if (rt_runtime_us < 0) | |
9531 | rt_runtime = RUNTIME_INF; | |
9532 | ||
9533 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9534 | } | |
9535 | ||
9f0c1e56 PZ |
9536 | long sched_group_rt_runtime(struct task_group *tg) |
9537 | { | |
9538 | u64 rt_runtime_us; | |
9539 | ||
d0b27fa7 | 9540 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9541 | return -1; |
9542 | ||
d0b27fa7 | 9543 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9544 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9545 | return rt_runtime_us; | |
9546 | } | |
d0b27fa7 PZ |
9547 | |
9548 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9549 | { | |
9550 | u64 rt_runtime, rt_period; | |
9551 | ||
9552 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9553 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9554 | ||
619b0488 R |
9555 | if (rt_period == 0) |
9556 | return -EINVAL; | |
9557 | ||
d0b27fa7 PZ |
9558 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9559 | } | |
9560 | ||
9561 | long sched_group_rt_period(struct task_group *tg) | |
9562 | { | |
9563 | u64 rt_period_us; | |
9564 | ||
9565 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9566 | do_div(rt_period_us, NSEC_PER_USEC); | |
9567 | return rt_period_us; | |
9568 | } | |
9569 | ||
9570 | static int sched_rt_global_constraints(void) | |
9571 | { | |
4653f803 | 9572 | u64 runtime, period; |
d0b27fa7 PZ |
9573 | int ret = 0; |
9574 | ||
ec5d4989 HS |
9575 | if (sysctl_sched_rt_period <= 0) |
9576 | return -EINVAL; | |
9577 | ||
4653f803 PZ |
9578 | runtime = global_rt_runtime(); |
9579 | period = global_rt_period(); | |
9580 | ||
9581 | /* | |
9582 | * Sanity check on the sysctl variables. | |
9583 | */ | |
9584 | if (runtime > period && runtime != RUNTIME_INF) | |
9585 | return -EINVAL; | |
10b612f4 | 9586 | |
d0b27fa7 | 9587 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9588 | read_lock(&tasklist_lock); |
4653f803 | 9589 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9590 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9591 | mutex_unlock(&rt_constraints_mutex); |
9592 | ||
9593 | return ret; | |
9594 | } | |
54e99124 DG |
9595 | |
9596 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9597 | { | |
9598 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9599 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9600 | return 0; | |
9601 | ||
9602 | return 1; | |
9603 | } | |
9604 | ||
6d6bc0ad | 9605 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9606 | static int sched_rt_global_constraints(void) |
9607 | { | |
ac086bc2 PZ |
9608 | unsigned long flags; |
9609 | int i; | |
9610 | ||
ec5d4989 HS |
9611 | if (sysctl_sched_rt_period <= 0) |
9612 | return -EINVAL; | |
9613 | ||
ac086bc2 PZ |
9614 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9615 | for_each_possible_cpu(i) { | |
9616 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9617 | ||
9618 | spin_lock(&rt_rq->rt_runtime_lock); | |
9619 | rt_rq->rt_runtime = global_rt_runtime(); | |
9620 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9621 | } | |
9622 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9623 | ||
d0b27fa7 PZ |
9624 | return 0; |
9625 | } | |
6d6bc0ad | 9626 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9627 | |
9628 | int sched_rt_handler(struct ctl_table *table, int write, | |
9629 | struct file *filp, void __user *buffer, size_t *lenp, | |
9630 | loff_t *ppos) | |
9631 | { | |
9632 | int ret; | |
9633 | int old_period, old_runtime; | |
9634 | static DEFINE_MUTEX(mutex); | |
9635 | ||
9636 | mutex_lock(&mutex); | |
9637 | old_period = sysctl_sched_rt_period; | |
9638 | old_runtime = sysctl_sched_rt_runtime; | |
9639 | ||
9640 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9641 | ||
9642 | if (!ret && write) { | |
9643 | ret = sched_rt_global_constraints(); | |
9644 | if (ret) { | |
9645 | sysctl_sched_rt_period = old_period; | |
9646 | sysctl_sched_rt_runtime = old_runtime; | |
9647 | } else { | |
9648 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9649 | def_rt_bandwidth.rt_period = | |
9650 | ns_to_ktime(global_rt_period()); | |
9651 | } | |
9652 | } | |
9653 | mutex_unlock(&mutex); | |
9654 | ||
9655 | return ret; | |
9656 | } | |
68318b8e | 9657 | |
052f1dc7 | 9658 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9659 | |
9660 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9661 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9662 | { |
2b01dfe3 PM |
9663 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9664 | struct task_group, css); | |
68318b8e SV |
9665 | } |
9666 | ||
9667 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9668 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9669 | { |
ec7dc8ac | 9670 | struct task_group *tg, *parent; |
68318b8e | 9671 | |
2b01dfe3 | 9672 | if (!cgrp->parent) { |
68318b8e | 9673 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9674 | return &init_task_group.css; |
9675 | } | |
9676 | ||
ec7dc8ac DG |
9677 | parent = cgroup_tg(cgrp->parent); |
9678 | tg = sched_create_group(parent); | |
68318b8e SV |
9679 | if (IS_ERR(tg)) |
9680 | return ERR_PTR(-ENOMEM); | |
9681 | ||
68318b8e SV |
9682 | return &tg->css; |
9683 | } | |
9684 | ||
41a2d6cf IM |
9685 | static void |
9686 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9687 | { |
2b01dfe3 | 9688 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9689 | |
9690 | sched_destroy_group(tg); | |
9691 | } | |
9692 | ||
41a2d6cf IM |
9693 | static int |
9694 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9695 | struct task_struct *tsk) | |
68318b8e | 9696 | { |
b68aa230 | 9697 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9698 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9699 | return -EINVAL; |
9700 | #else | |
68318b8e SV |
9701 | /* We don't support RT-tasks being in separate groups */ |
9702 | if (tsk->sched_class != &fair_sched_class) | |
9703 | return -EINVAL; | |
b68aa230 | 9704 | #endif |
68318b8e SV |
9705 | |
9706 | return 0; | |
9707 | } | |
9708 | ||
9709 | static void | |
2b01dfe3 | 9710 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9711 | struct cgroup *old_cont, struct task_struct *tsk) |
9712 | { | |
9713 | sched_move_task(tsk); | |
9714 | } | |
9715 | ||
052f1dc7 | 9716 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9717 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9718 | u64 shareval) |
68318b8e | 9719 | { |
2b01dfe3 | 9720 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9721 | } |
9722 | ||
f4c753b7 | 9723 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9724 | { |
2b01dfe3 | 9725 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9726 | |
9727 | return (u64) tg->shares; | |
9728 | } | |
6d6bc0ad | 9729 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9730 | |
052f1dc7 | 9731 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9732 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9733 | s64 val) |
6f505b16 | 9734 | { |
06ecb27c | 9735 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9736 | } |
9737 | ||
06ecb27c | 9738 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9739 | { |
06ecb27c | 9740 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9741 | } |
d0b27fa7 PZ |
9742 | |
9743 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9744 | u64 rt_period_us) | |
9745 | { | |
9746 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9747 | } | |
9748 | ||
9749 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9750 | { | |
9751 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9752 | } | |
6d6bc0ad | 9753 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9754 | |
fe5c7cc2 | 9755 | static struct cftype cpu_files[] = { |
052f1dc7 | 9756 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9757 | { |
9758 | .name = "shares", | |
f4c753b7 PM |
9759 | .read_u64 = cpu_shares_read_u64, |
9760 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9761 | }, |
052f1dc7 PZ |
9762 | #endif |
9763 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9764 | { |
9f0c1e56 | 9765 | .name = "rt_runtime_us", |
06ecb27c PM |
9766 | .read_s64 = cpu_rt_runtime_read, |
9767 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9768 | }, |
d0b27fa7 PZ |
9769 | { |
9770 | .name = "rt_period_us", | |
f4c753b7 PM |
9771 | .read_u64 = cpu_rt_period_read_uint, |
9772 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9773 | }, |
052f1dc7 | 9774 | #endif |
68318b8e SV |
9775 | }; |
9776 | ||
9777 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9778 | { | |
fe5c7cc2 | 9779 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9780 | } |
9781 | ||
9782 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9783 | .name = "cpu", |
9784 | .create = cpu_cgroup_create, | |
9785 | .destroy = cpu_cgroup_destroy, | |
9786 | .can_attach = cpu_cgroup_can_attach, | |
9787 | .attach = cpu_cgroup_attach, | |
9788 | .populate = cpu_cgroup_populate, | |
9789 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9790 | .early_init = 1, |
9791 | }; | |
9792 | ||
052f1dc7 | 9793 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9794 | |
9795 | #ifdef CONFIG_CGROUP_CPUACCT | |
9796 | ||
9797 | /* | |
9798 | * CPU accounting code for task groups. | |
9799 | * | |
9800 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9801 | * (balbir@in.ibm.com). | |
9802 | */ | |
9803 | ||
934352f2 | 9804 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9805 | struct cpuacct { |
9806 | struct cgroup_subsys_state css; | |
9807 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9808 | u64 *cpuusage; | |
934352f2 | 9809 | struct cpuacct *parent; |
d842de87 SV |
9810 | }; |
9811 | ||
9812 | struct cgroup_subsys cpuacct_subsys; | |
9813 | ||
9814 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9815 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9816 | { |
32cd756a | 9817 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9818 | struct cpuacct, css); |
9819 | } | |
9820 | ||
9821 | /* return cpu accounting group to which this task belongs */ | |
9822 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9823 | { | |
9824 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9825 | struct cpuacct, css); | |
9826 | } | |
9827 | ||
9828 | /* create a new cpu accounting group */ | |
9829 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9830 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9831 | { |
9832 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9833 | ||
9834 | if (!ca) | |
9835 | return ERR_PTR(-ENOMEM); | |
9836 | ||
9837 | ca->cpuusage = alloc_percpu(u64); | |
9838 | if (!ca->cpuusage) { | |
9839 | kfree(ca); | |
9840 | return ERR_PTR(-ENOMEM); | |
9841 | } | |
9842 | ||
934352f2 BR |
9843 | if (cgrp->parent) |
9844 | ca->parent = cgroup_ca(cgrp->parent); | |
9845 | ||
d842de87 SV |
9846 | return &ca->css; |
9847 | } | |
9848 | ||
9849 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9850 | static void |
32cd756a | 9851 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9852 | { |
32cd756a | 9853 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9854 | |
9855 | free_percpu(ca->cpuusage); | |
9856 | kfree(ca); | |
9857 | } | |
9858 | ||
720f5498 KC |
9859 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9860 | { | |
9861 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9862 | u64 data; | |
9863 | ||
9864 | #ifndef CONFIG_64BIT | |
9865 | /* | |
9866 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9867 | */ | |
9868 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9869 | data = *cpuusage; | |
9870 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9871 | #else | |
9872 | data = *cpuusage; | |
9873 | #endif | |
9874 | ||
9875 | return data; | |
9876 | } | |
9877 | ||
9878 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9879 | { | |
9880 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9881 | ||
9882 | #ifndef CONFIG_64BIT | |
9883 | /* | |
9884 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9885 | */ | |
9886 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9887 | *cpuusage = val; | |
9888 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9889 | #else | |
9890 | *cpuusage = val; | |
9891 | #endif | |
9892 | } | |
9893 | ||
d842de87 | 9894 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9895 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9896 | { |
32cd756a | 9897 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9898 | u64 totalcpuusage = 0; |
9899 | int i; | |
9900 | ||
720f5498 KC |
9901 | for_each_present_cpu(i) |
9902 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9903 | |
9904 | return totalcpuusage; | |
9905 | } | |
9906 | ||
0297b803 DG |
9907 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9908 | u64 reset) | |
9909 | { | |
9910 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9911 | int err = 0; | |
9912 | int i; | |
9913 | ||
9914 | if (reset) { | |
9915 | err = -EINVAL; | |
9916 | goto out; | |
9917 | } | |
9918 | ||
720f5498 KC |
9919 | for_each_present_cpu(i) |
9920 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9921 | |
0297b803 DG |
9922 | out: |
9923 | return err; | |
9924 | } | |
9925 | ||
e9515c3c KC |
9926 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9927 | struct seq_file *m) | |
9928 | { | |
9929 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9930 | u64 percpu; | |
9931 | int i; | |
9932 | ||
9933 | for_each_present_cpu(i) { | |
9934 | percpu = cpuacct_cpuusage_read(ca, i); | |
9935 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9936 | } | |
9937 | seq_printf(m, "\n"); | |
9938 | return 0; | |
9939 | } | |
9940 | ||
d842de87 SV |
9941 | static struct cftype files[] = { |
9942 | { | |
9943 | .name = "usage", | |
f4c753b7 PM |
9944 | .read_u64 = cpuusage_read, |
9945 | .write_u64 = cpuusage_write, | |
d842de87 | 9946 | }, |
e9515c3c KC |
9947 | { |
9948 | .name = "usage_percpu", | |
9949 | .read_seq_string = cpuacct_percpu_seq_read, | |
9950 | }, | |
9951 | ||
d842de87 SV |
9952 | }; |
9953 | ||
32cd756a | 9954 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9955 | { |
32cd756a | 9956 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9957 | } |
9958 | ||
9959 | /* | |
9960 | * charge this task's execution time to its accounting group. | |
9961 | * | |
9962 | * called with rq->lock held. | |
9963 | */ | |
9964 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9965 | { | |
9966 | struct cpuacct *ca; | |
934352f2 | 9967 | int cpu; |
d842de87 | 9968 | |
c40c6f85 | 9969 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9970 | return; |
9971 | ||
934352f2 | 9972 | cpu = task_cpu(tsk); |
d842de87 | 9973 | ca = task_ca(tsk); |
d842de87 | 9974 | |
934352f2 BR |
9975 | for (; ca; ca = ca->parent) { |
9976 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
d842de87 SV |
9977 | *cpuusage += cputime; |
9978 | } | |
9979 | } | |
9980 | ||
9981 | struct cgroup_subsys cpuacct_subsys = { | |
9982 | .name = "cpuacct", | |
9983 | .create = cpuacct_create, | |
9984 | .destroy = cpuacct_destroy, | |
9985 | .populate = cpuacct_populate, | |
9986 | .subsys_id = cpuacct_subsys_id, | |
9987 | }; | |
9988 | #endif /* CONFIG_CGROUP_CPUACCT */ |