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> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.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> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
a8d154b0 | 80 | #define CREATE_TRACE_POINTS |
ad8d75ff | 81 | #include <trace/events/sched.h> |
a8d154b0 | 82 | |
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
d0b27fa7 PZ |
117 | /* |
118 | * single value that denotes runtime == period, ie unlimited time. | |
119 | */ | |
120 | #define RUNTIME_INF ((u64)~0ULL) | |
121 | ||
e05606d3 IM |
122 | static inline int rt_policy(int policy) |
123 | { | |
3f33a7ce | 124 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
125 | return 1; |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int task_has_rt_policy(struct task_struct *p) | |
130 | { | |
131 | return rt_policy(p->policy); | |
132 | } | |
133 | ||
1da177e4 | 134 | /* |
6aa645ea | 135 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 136 | */ |
6aa645ea IM |
137 | struct rt_prio_array { |
138 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
139 | struct list_head queue[MAX_RT_PRIO]; | |
140 | }; | |
141 | ||
d0b27fa7 | 142 | struct rt_bandwidth { |
ea736ed5 IM |
143 | /* nests inside the rq lock: */ |
144 | spinlock_t rt_runtime_lock; | |
145 | ktime_t rt_period; | |
146 | u64 rt_runtime; | |
147 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
148 | }; |
149 | ||
150 | static struct rt_bandwidth def_rt_bandwidth; | |
151 | ||
152 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
153 | ||
154 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
155 | { | |
156 | struct rt_bandwidth *rt_b = | |
157 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
158 | ktime_t now; | |
159 | int overrun; | |
160 | int idle = 0; | |
161 | ||
162 | for (;;) { | |
163 | now = hrtimer_cb_get_time(timer); | |
164 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
165 | ||
166 | if (!overrun) | |
167 | break; | |
168 | ||
169 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
170 | } | |
171 | ||
172 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
173 | } | |
174 | ||
175 | static | |
176 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
177 | { | |
178 | rt_b->rt_period = ns_to_ktime(period); | |
179 | rt_b->rt_runtime = runtime; | |
180 | ||
ac086bc2 PZ |
181 | spin_lock_init(&rt_b->rt_runtime_lock); |
182 | ||
d0b27fa7 PZ |
183 | hrtimer_init(&rt_b->rt_period_timer, |
184 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
185 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
186 | } |
187 | ||
c8bfff6d KH |
188 | static inline int rt_bandwidth_enabled(void) |
189 | { | |
190 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
191 | } |
192 | ||
193 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
194 | { | |
195 | ktime_t now; | |
196 | ||
cac64d00 | 197 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
198 | return; |
199 | ||
200 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
201 | return; | |
202 | ||
203 | spin_lock(&rt_b->rt_runtime_lock); | |
204 | for (;;) { | |
7f1e2ca9 PZ |
205 | unsigned long delta; |
206 | ktime_t soft, hard; | |
207 | ||
d0b27fa7 PZ |
208 | if (hrtimer_active(&rt_b->rt_period_timer)) |
209 | break; | |
210 | ||
211 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
212 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
213 | |
214 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
215 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
216 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
217 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 218 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 PZ |
219 | } |
220 | spin_unlock(&rt_b->rt_runtime_lock); | |
221 | } | |
222 | ||
223 | #ifdef CONFIG_RT_GROUP_SCHED | |
224 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
225 | { | |
226 | hrtimer_cancel(&rt_b->rt_period_timer); | |
227 | } | |
228 | #endif | |
229 | ||
712555ee HC |
230 | /* |
231 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
232 | * detach_destroy_domains and partition_sched_domains. | |
233 | */ | |
234 | static DEFINE_MUTEX(sched_domains_mutex); | |
235 | ||
052f1dc7 | 236 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 237 | |
68318b8e SV |
238 | #include <linux/cgroup.h> |
239 | ||
29f59db3 SV |
240 | struct cfs_rq; |
241 | ||
6f505b16 PZ |
242 | static LIST_HEAD(task_groups); |
243 | ||
29f59db3 | 244 | /* task group related information */ |
4cf86d77 | 245 | struct task_group { |
052f1dc7 | 246 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
247 | struct cgroup_subsys_state css; |
248 | #endif | |
052f1dc7 | 249 | |
6c415b92 AB |
250 | #ifdef CONFIG_USER_SCHED |
251 | uid_t uid; | |
252 | #endif | |
253 | ||
052f1dc7 | 254 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
255 | /* schedulable entities of this group on each cpu */ |
256 | struct sched_entity **se; | |
257 | /* runqueue "owned" by this group on each cpu */ | |
258 | struct cfs_rq **cfs_rq; | |
259 | unsigned long shares; | |
052f1dc7 PZ |
260 | #endif |
261 | ||
262 | #ifdef CONFIG_RT_GROUP_SCHED | |
263 | struct sched_rt_entity **rt_se; | |
264 | struct rt_rq **rt_rq; | |
265 | ||
d0b27fa7 | 266 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 267 | #endif |
6b2d7700 | 268 | |
ae8393e5 | 269 | struct rcu_head rcu; |
6f505b16 | 270 | struct list_head list; |
f473aa5e PZ |
271 | |
272 | struct task_group *parent; | |
273 | struct list_head siblings; | |
274 | struct list_head children; | |
29f59db3 SV |
275 | }; |
276 | ||
354d60c2 | 277 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 278 | |
6c415b92 AB |
279 | /* Helper function to pass uid information to create_sched_user() */ |
280 | void set_tg_uid(struct user_struct *user) | |
281 | { | |
282 | user->tg->uid = user->uid; | |
283 | } | |
284 | ||
eff766a6 PZ |
285 | /* |
286 | * Root task group. | |
84e9dabf AS |
287 | * Every UID task group (including init_task_group aka UID-0) will |
288 | * be a child to this group. | |
eff766a6 PZ |
289 | */ |
290 | struct task_group root_task_group; | |
291 | ||
052f1dc7 | 292 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
293 | /* Default task group's sched entity on each cpu */ |
294 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
295 | /* Default task group's cfs_rq on each cpu */ | |
ada3fa15 | 296 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct cfs_rq, init_tg_cfs_rq); |
6d6bc0ad | 297 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
298 | |
299 | #ifdef CONFIG_RT_GROUP_SCHED | |
300 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
b9bf3121 | 301 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rt_rq, init_rt_rq); |
6d6bc0ad | 302 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 303 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 304 | #define root_task_group init_task_group |
9a7e0b18 | 305 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 306 | |
8ed36996 | 307 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
308 | * a task group's cpu shares. |
309 | */ | |
8ed36996 | 310 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 311 | |
57310a98 PZ |
312 | #ifdef CONFIG_SMP |
313 | static int root_task_group_empty(void) | |
314 | { | |
315 | return list_empty(&root_task_group.children); | |
316 | } | |
317 | #endif | |
318 | ||
052f1dc7 | 319 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
320 | #ifdef CONFIG_USER_SCHED |
321 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 322 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 323 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 324 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 325 | |
cb4ad1ff | 326 | /* |
2e084786 LJ |
327 | * A weight of 0 or 1 can cause arithmetics problems. |
328 | * A weight of a cfs_rq is the sum of weights of which entities | |
329 | * are queued on this cfs_rq, so a weight of a entity should not be | |
330 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
331 | * (The default weight is 1024 - so there's no practical |
332 | * limitation from this.) | |
333 | */ | |
18d95a28 | 334 | #define MIN_SHARES 2 |
2e084786 | 335 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 336 | |
052f1dc7 PZ |
337 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
338 | #endif | |
339 | ||
29f59db3 | 340 | /* Default task group. |
3a252015 | 341 | * Every task in system belong to this group at bootup. |
29f59db3 | 342 | */ |
434d53b0 | 343 | struct task_group init_task_group; |
29f59db3 SV |
344 | |
345 | /* return group to which a task belongs */ | |
4cf86d77 | 346 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 347 | { |
4cf86d77 | 348 | struct task_group *tg; |
9b5b7751 | 349 | |
052f1dc7 | 350 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
351 | rcu_read_lock(); |
352 | tg = __task_cred(p)->user->tg; | |
353 | rcu_read_unlock(); | |
052f1dc7 | 354 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
355 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
356 | struct task_group, css); | |
24e377a8 | 357 | #else |
41a2d6cf | 358 | tg = &init_task_group; |
24e377a8 | 359 | #endif |
9b5b7751 | 360 | return tg; |
29f59db3 SV |
361 | } |
362 | ||
363 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 364 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 365 | { |
052f1dc7 | 366 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
367 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
368 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 369 | #endif |
6f505b16 | 370 | |
052f1dc7 | 371 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
372 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
373 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 374 | #endif |
29f59db3 SV |
375 | } |
376 | ||
377 | #else | |
378 | ||
6f505b16 | 379 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
380 | static inline struct task_group *task_group(struct task_struct *p) |
381 | { | |
382 | return NULL; | |
383 | } | |
29f59db3 | 384 | |
052f1dc7 | 385 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 386 | |
6aa645ea IM |
387 | /* CFS-related fields in a runqueue */ |
388 | struct cfs_rq { | |
389 | struct load_weight load; | |
390 | unsigned long nr_running; | |
391 | ||
6aa645ea | 392 | u64 exec_clock; |
e9acbff6 | 393 | u64 min_vruntime; |
6aa645ea IM |
394 | |
395 | struct rb_root tasks_timeline; | |
396 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
397 | |
398 | struct list_head tasks; | |
399 | struct list_head *balance_iterator; | |
400 | ||
401 | /* | |
402 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
403 | * It is set to NULL otherwise (i.e when none are currently running). |
404 | */ | |
4793241b | 405 | struct sched_entity *curr, *next, *last; |
ddc97297 | 406 | |
5ac5c4d6 | 407 | unsigned int nr_spread_over; |
ddc97297 | 408 | |
62160e3f | 409 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
410 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
411 | ||
41a2d6cf IM |
412 | /* |
413 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
414 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
415 | * (like users, containers etc.) | |
416 | * | |
417 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
418 | * list is used during load balance. | |
419 | */ | |
41a2d6cf IM |
420 | struct list_head leaf_cfs_rq_list; |
421 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
422 | |
423 | #ifdef CONFIG_SMP | |
c09595f6 | 424 | /* |
c8cba857 | 425 | * the part of load.weight contributed by tasks |
c09595f6 | 426 | */ |
c8cba857 | 427 | unsigned long task_weight; |
c09595f6 | 428 | |
c8cba857 PZ |
429 | /* |
430 | * h_load = weight * f(tg) | |
431 | * | |
432 | * Where f(tg) is the recursive weight fraction assigned to | |
433 | * this group. | |
434 | */ | |
435 | unsigned long h_load; | |
c09595f6 | 436 | |
c8cba857 PZ |
437 | /* |
438 | * this cpu's part of tg->shares | |
439 | */ | |
440 | unsigned long shares; | |
f1d239f7 PZ |
441 | |
442 | /* | |
443 | * load.weight at the time we set shares | |
444 | */ | |
445 | unsigned long rq_weight; | |
c09595f6 | 446 | #endif |
6aa645ea IM |
447 | #endif |
448 | }; | |
1da177e4 | 449 | |
6aa645ea IM |
450 | /* Real-Time classes' related field in a runqueue: */ |
451 | struct rt_rq { | |
452 | struct rt_prio_array active; | |
63489e45 | 453 | unsigned long rt_nr_running; |
052f1dc7 | 454 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
455 | struct { |
456 | int curr; /* highest queued rt task prio */ | |
398a153b | 457 | #ifdef CONFIG_SMP |
e864c499 | 458 | int next; /* next highest */ |
398a153b | 459 | #endif |
e864c499 | 460 | } highest_prio; |
6f505b16 | 461 | #endif |
fa85ae24 | 462 | #ifdef CONFIG_SMP |
73fe6aae | 463 | unsigned long rt_nr_migratory; |
a1ba4d8b | 464 | unsigned long rt_nr_total; |
a22d7fc1 | 465 | int overloaded; |
917b627d | 466 | struct plist_head pushable_tasks; |
fa85ae24 | 467 | #endif |
6f505b16 | 468 | int rt_throttled; |
fa85ae24 | 469 | u64 rt_time; |
ac086bc2 | 470 | u64 rt_runtime; |
ea736ed5 | 471 | /* Nests inside the rq lock: */ |
ac086bc2 | 472 | spinlock_t rt_runtime_lock; |
6f505b16 | 473 | |
052f1dc7 | 474 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
475 | unsigned long rt_nr_boosted; |
476 | ||
6f505b16 PZ |
477 | struct rq *rq; |
478 | struct list_head leaf_rt_rq_list; | |
479 | struct task_group *tg; | |
480 | struct sched_rt_entity *rt_se; | |
481 | #endif | |
6aa645ea IM |
482 | }; |
483 | ||
57d885fe GH |
484 | #ifdef CONFIG_SMP |
485 | ||
486 | /* | |
487 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
488 | * variables. Each exclusive cpuset essentially defines an island domain by |
489 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
490 | * exclusive cpuset is created, we also create and attach a new root-domain |
491 | * object. | |
492 | * | |
57d885fe GH |
493 | */ |
494 | struct root_domain { | |
495 | atomic_t refcount; | |
c6c4927b RR |
496 | cpumask_var_t span; |
497 | cpumask_var_t online; | |
637f5085 | 498 | |
0eab9146 | 499 | /* |
637f5085 GH |
500 | * The "RT overload" flag: it gets set if a CPU has more than |
501 | * one runnable RT task. | |
502 | */ | |
c6c4927b | 503 | cpumask_var_t rto_mask; |
0eab9146 | 504 | atomic_t rto_count; |
6e0534f2 GH |
505 | #ifdef CONFIG_SMP |
506 | struct cpupri cpupri; | |
507 | #endif | |
57d885fe GH |
508 | }; |
509 | ||
dc938520 GH |
510 | /* |
511 | * By default the system creates a single root-domain with all cpus as | |
512 | * members (mimicking the global state we have today). | |
513 | */ | |
57d885fe GH |
514 | static struct root_domain def_root_domain; |
515 | ||
516 | #endif | |
517 | ||
1da177e4 LT |
518 | /* |
519 | * This is the main, per-CPU runqueue data structure. | |
520 | * | |
521 | * Locking rule: those places that want to lock multiple runqueues | |
522 | * (such as the load balancing or the thread migration code), lock | |
523 | * acquire operations must be ordered by ascending &runqueue. | |
524 | */ | |
70b97a7f | 525 | struct rq { |
d8016491 IM |
526 | /* runqueue lock: */ |
527 | spinlock_t lock; | |
1da177e4 LT |
528 | |
529 | /* | |
530 | * nr_running and cpu_load should be in the same cacheline because | |
531 | * remote CPUs use both these fields when doing load calculation. | |
532 | */ | |
533 | unsigned long nr_running; | |
6aa645ea IM |
534 | #define CPU_LOAD_IDX_MAX 5 |
535 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c SS |
536 | #ifdef CONFIG_NO_HZ |
537 | unsigned char in_nohz_recently; | |
538 | #endif | |
d8016491 IM |
539 | /* capture load from *all* tasks on this cpu: */ |
540 | struct load_weight load; | |
6aa645ea IM |
541 | unsigned long nr_load_updates; |
542 | u64 nr_switches; | |
543 | ||
544 | struct cfs_rq cfs; | |
6f505b16 | 545 | struct rt_rq rt; |
6f505b16 | 546 | |
6aa645ea | 547 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
548 | /* list of leaf cfs_rq on this cpu: */ |
549 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
550 | #endif |
551 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 552 | struct list_head leaf_rt_rq_list; |
1da177e4 | 553 | #endif |
1da177e4 LT |
554 | |
555 | /* | |
556 | * This is part of a global counter where only the total sum | |
557 | * over all CPUs matters. A task can increase this counter on | |
558 | * one CPU and if it got migrated afterwards it may decrease | |
559 | * it on another CPU. Always updated under the runqueue lock: | |
560 | */ | |
561 | unsigned long nr_uninterruptible; | |
562 | ||
36c8b586 | 563 | struct task_struct *curr, *idle; |
c9819f45 | 564 | unsigned long next_balance; |
1da177e4 | 565 | struct mm_struct *prev_mm; |
6aa645ea | 566 | |
3e51f33f | 567 | u64 clock; |
6aa645ea | 568 | |
1da177e4 LT |
569 | atomic_t nr_iowait; |
570 | ||
571 | #ifdef CONFIG_SMP | |
0eab9146 | 572 | struct root_domain *rd; |
1da177e4 LT |
573 | struct sched_domain *sd; |
574 | ||
a0a522ce | 575 | unsigned char idle_at_tick; |
1da177e4 | 576 | /* For active balancing */ |
3f029d3c | 577 | int post_schedule; |
1da177e4 LT |
578 | int active_balance; |
579 | int push_cpu; | |
d8016491 IM |
580 | /* cpu of this runqueue: */ |
581 | int cpu; | |
1f11eb6a | 582 | int online; |
1da177e4 | 583 | |
a8a51d5e | 584 | unsigned long avg_load_per_task; |
1da177e4 | 585 | |
36c8b586 | 586 | struct task_struct *migration_thread; |
1da177e4 | 587 | struct list_head migration_queue; |
e9e9250b PZ |
588 | |
589 | u64 rt_avg; | |
590 | u64 age_stamp; | |
1b9508f6 MG |
591 | u64 idle_stamp; |
592 | u64 avg_idle; | |
1da177e4 LT |
593 | #endif |
594 | ||
dce48a84 TG |
595 | /* calc_load related fields */ |
596 | unsigned long calc_load_update; | |
597 | long calc_load_active; | |
598 | ||
8f4d37ec | 599 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
600 | #ifdef CONFIG_SMP |
601 | int hrtick_csd_pending; | |
602 | struct call_single_data hrtick_csd; | |
603 | #endif | |
8f4d37ec PZ |
604 | struct hrtimer hrtick_timer; |
605 | #endif | |
606 | ||
1da177e4 LT |
607 | #ifdef CONFIG_SCHEDSTATS |
608 | /* latency stats */ | |
609 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
610 | unsigned long long rq_cpu_time; |
611 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
612 | |
613 | /* sys_sched_yield() stats */ | |
480b9434 | 614 | unsigned int yld_count; |
1da177e4 LT |
615 | |
616 | /* schedule() stats */ | |
480b9434 KC |
617 | unsigned int sched_switch; |
618 | unsigned int sched_count; | |
619 | unsigned int sched_goidle; | |
1da177e4 LT |
620 | |
621 | /* try_to_wake_up() stats */ | |
480b9434 KC |
622 | unsigned int ttwu_count; |
623 | unsigned int ttwu_local; | |
b8efb561 IM |
624 | |
625 | /* BKL stats */ | |
480b9434 | 626 | unsigned int bkl_count; |
1da177e4 LT |
627 | #endif |
628 | }; | |
629 | ||
f34e3b61 | 630 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 631 | |
7d478721 PZ |
632 | static inline |
633 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 634 | { |
7d478721 | 635 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
dd41f596 IM |
636 | } |
637 | ||
0a2966b4 CL |
638 | static inline int cpu_of(struct rq *rq) |
639 | { | |
640 | #ifdef CONFIG_SMP | |
641 | return rq->cpu; | |
642 | #else | |
643 | return 0; | |
644 | #endif | |
645 | } | |
646 | ||
674311d5 NP |
647 | /* |
648 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 649 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
650 | * |
651 | * The domain tree of any CPU may only be accessed from within | |
652 | * preempt-disabled sections. | |
653 | */ | |
48f24c4d IM |
654 | #define for_each_domain(cpu, __sd) \ |
655 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
656 | |
657 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
658 | #define this_rq() (&__get_cpu_var(runqueues)) | |
659 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
660 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 661 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 662 | |
aa9c4c0f | 663 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
664 | { |
665 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
666 | } | |
667 | ||
bf5c91ba IM |
668 | /* |
669 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
670 | */ | |
671 | #ifdef CONFIG_SCHED_DEBUG | |
672 | # define const_debug __read_mostly | |
673 | #else | |
674 | # define const_debug static const | |
675 | #endif | |
676 | ||
017730c1 IM |
677 | /** |
678 | * runqueue_is_locked | |
e17b38bf | 679 | * @cpu: the processor in question. |
017730c1 IM |
680 | * |
681 | * Returns true if the current cpu runqueue is locked. | |
682 | * This interface allows printk to be called with the runqueue lock | |
683 | * held and know whether or not it is OK to wake up the klogd. | |
684 | */ | |
89f19f04 | 685 | int runqueue_is_locked(int cpu) |
017730c1 | 686 | { |
89f19f04 | 687 | return spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
688 | } |
689 | ||
bf5c91ba IM |
690 | /* |
691 | * Debugging: various feature bits | |
692 | */ | |
f00b45c1 PZ |
693 | |
694 | #define SCHED_FEAT(name, enabled) \ | |
695 | __SCHED_FEAT_##name , | |
696 | ||
bf5c91ba | 697 | enum { |
f00b45c1 | 698 | #include "sched_features.h" |
bf5c91ba IM |
699 | }; |
700 | ||
f00b45c1 PZ |
701 | #undef SCHED_FEAT |
702 | ||
703 | #define SCHED_FEAT(name, enabled) \ | |
704 | (1UL << __SCHED_FEAT_##name) * enabled | | |
705 | ||
bf5c91ba | 706 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
707 | #include "sched_features.h" |
708 | 0; | |
709 | ||
710 | #undef SCHED_FEAT | |
711 | ||
712 | #ifdef CONFIG_SCHED_DEBUG | |
713 | #define SCHED_FEAT(name, enabled) \ | |
714 | #name , | |
715 | ||
983ed7a6 | 716 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
717 | #include "sched_features.h" |
718 | NULL | |
719 | }; | |
720 | ||
721 | #undef SCHED_FEAT | |
722 | ||
34f3a814 | 723 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 724 | { |
f00b45c1 PZ |
725 | int i; |
726 | ||
727 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
728 | if (!(sysctl_sched_features & (1UL << i))) |
729 | seq_puts(m, "NO_"); | |
730 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 731 | } |
34f3a814 | 732 | seq_puts(m, "\n"); |
f00b45c1 | 733 | |
34f3a814 | 734 | return 0; |
f00b45c1 PZ |
735 | } |
736 | ||
737 | static ssize_t | |
738 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
739 | size_t cnt, loff_t *ppos) | |
740 | { | |
741 | char buf[64]; | |
742 | char *cmp = buf; | |
743 | int neg = 0; | |
744 | int i; | |
745 | ||
746 | if (cnt > 63) | |
747 | cnt = 63; | |
748 | ||
749 | if (copy_from_user(&buf, ubuf, cnt)) | |
750 | return -EFAULT; | |
751 | ||
752 | buf[cnt] = 0; | |
753 | ||
c24b7c52 | 754 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
755 | neg = 1; |
756 | cmp += 3; | |
757 | } | |
758 | ||
759 | for (i = 0; sched_feat_names[i]; i++) { | |
760 | int len = strlen(sched_feat_names[i]); | |
761 | ||
762 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
763 | if (neg) | |
764 | sysctl_sched_features &= ~(1UL << i); | |
765 | else | |
766 | sysctl_sched_features |= (1UL << i); | |
767 | break; | |
768 | } | |
769 | } | |
770 | ||
771 | if (!sched_feat_names[i]) | |
772 | return -EINVAL; | |
773 | ||
42994724 | 774 | *ppos += cnt; |
f00b45c1 PZ |
775 | |
776 | return cnt; | |
777 | } | |
778 | ||
34f3a814 LZ |
779 | static int sched_feat_open(struct inode *inode, struct file *filp) |
780 | { | |
781 | return single_open(filp, sched_feat_show, NULL); | |
782 | } | |
783 | ||
828c0950 | 784 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
785 | .open = sched_feat_open, |
786 | .write = sched_feat_write, | |
787 | .read = seq_read, | |
788 | .llseek = seq_lseek, | |
789 | .release = single_release, | |
f00b45c1 PZ |
790 | }; |
791 | ||
792 | static __init int sched_init_debug(void) | |
793 | { | |
f00b45c1 PZ |
794 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
795 | &sched_feat_fops); | |
796 | ||
797 | return 0; | |
798 | } | |
799 | late_initcall(sched_init_debug); | |
800 | ||
801 | #endif | |
802 | ||
803 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 804 | |
b82d9fdd PZ |
805 | /* |
806 | * Number of tasks to iterate in a single balance run. | |
807 | * Limited because this is done with IRQs disabled. | |
808 | */ | |
809 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
810 | ||
2398f2c6 PZ |
811 | /* |
812 | * ratelimit for updating the group shares. | |
55cd5340 | 813 | * default: 0.25ms |
2398f2c6 | 814 | */ |
55cd5340 | 815 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 816 | |
ffda12a1 PZ |
817 | /* |
818 | * Inject some fuzzyness into changing the per-cpu group shares | |
819 | * this avoids remote rq-locks at the expense of fairness. | |
820 | * default: 4 | |
821 | */ | |
822 | unsigned int sysctl_sched_shares_thresh = 4; | |
823 | ||
e9e9250b PZ |
824 | /* |
825 | * period over which we average the RT time consumption, measured | |
826 | * in ms. | |
827 | * | |
828 | * default: 1s | |
829 | */ | |
830 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
831 | ||
fa85ae24 | 832 | /* |
9f0c1e56 | 833 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
834 | * default: 1s |
835 | */ | |
9f0c1e56 | 836 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 837 | |
6892b75e IM |
838 | static __read_mostly int scheduler_running; |
839 | ||
9f0c1e56 PZ |
840 | /* |
841 | * part of the period that we allow rt tasks to run in us. | |
842 | * default: 0.95s | |
843 | */ | |
844 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 845 | |
d0b27fa7 PZ |
846 | static inline u64 global_rt_period(void) |
847 | { | |
848 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
849 | } | |
850 | ||
851 | static inline u64 global_rt_runtime(void) | |
852 | { | |
e26873bb | 853 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
854 | return RUNTIME_INF; |
855 | ||
856 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
857 | } | |
fa85ae24 | 858 | |
1da177e4 | 859 | #ifndef prepare_arch_switch |
4866cde0 NP |
860 | # define prepare_arch_switch(next) do { } while (0) |
861 | #endif | |
862 | #ifndef finish_arch_switch | |
863 | # define finish_arch_switch(prev) do { } while (0) | |
864 | #endif | |
865 | ||
051a1d1a DA |
866 | static inline int task_current(struct rq *rq, struct task_struct *p) |
867 | { | |
868 | return rq->curr == p; | |
869 | } | |
870 | ||
4866cde0 | 871 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 872 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 873 | { |
051a1d1a | 874 | return task_current(rq, p); |
4866cde0 NP |
875 | } |
876 | ||
70b97a7f | 877 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
878 | { |
879 | } | |
880 | ||
70b97a7f | 881 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 882 | { |
da04c035 IM |
883 | #ifdef CONFIG_DEBUG_SPINLOCK |
884 | /* this is a valid case when another task releases the spinlock */ | |
885 | rq->lock.owner = current; | |
886 | #endif | |
8a25d5de IM |
887 | /* |
888 | * If we are tracking spinlock dependencies then we have to | |
889 | * fix up the runqueue lock - which gets 'carried over' from | |
890 | * prev into current: | |
891 | */ | |
892 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
893 | ||
4866cde0 NP |
894 | spin_unlock_irq(&rq->lock); |
895 | } | |
896 | ||
897 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 898 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
899 | { |
900 | #ifdef CONFIG_SMP | |
901 | return p->oncpu; | |
902 | #else | |
051a1d1a | 903 | return task_current(rq, p); |
4866cde0 NP |
904 | #endif |
905 | } | |
906 | ||
70b97a7f | 907 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
908 | { |
909 | #ifdef CONFIG_SMP | |
910 | /* | |
911 | * We can optimise this out completely for !SMP, because the | |
912 | * SMP rebalancing from interrupt is the only thing that cares | |
913 | * here. | |
914 | */ | |
915 | next->oncpu = 1; | |
916 | #endif | |
917 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
918 | spin_unlock_irq(&rq->lock); | |
919 | #else | |
920 | spin_unlock(&rq->lock); | |
921 | #endif | |
922 | } | |
923 | ||
70b97a7f | 924 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
925 | { |
926 | #ifdef CONFIG_SMP | |
927 | /* | |
928 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
929 | * We must ensure this doesn't happen until the switch is completely | |
930 | * finished. | |
931 | */ | |
932 | smp_wmb(); | |
933 | prev->oncpu = 0; | |
934 | #endif | |
935 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
936 | local_irq_enable(); | |
1da177e4 | 937 | #endif |
4866cde0 NP |
938 | } |
939 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 940 | |
b29739f9 IM |
941 | /* |
942 | * __task_rq_lock - lock the runqueue a given task resides on. | |
943 | * Must be called interrupts disabled. | |
944 | */ | |
70b97a7f | 945 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
946 | __acquires(rq->lock) |
947 | { | |
3a5c359a AK |
948 | for (;;) { |
949 | struct rq *rq = task_rq(p); | |
950 | spin_lock(&rq->lock); | |
951 | if (likely(rq == task_rq(p))) | |
952 | return rq; | |
b29739f9 | 953 | spin_unlock(&rq->lock); |
b29739f9 | 954 | } |
b29739f9 IM |
955 | } |
956 | ||
1da177e4 LT |
957 | /* |
958 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 959 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
960 | * explicitly disabling preemption. |
961 | */ | |
70b97a7f | 962 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
963 | __acquires(rq->lock) |
964 | { | |
70b97a7f | 965 | struct rq *rq; |
1da177e4 | 966 | |
3a5c359a AK |
967 | for (;;) { |
968 | local_irq_save(*flags); | |
969 | rq = task_rq(p); | |
970 | spin_lock(&rq->lock); | |
971 | if (likely(rq == task_rq(p))) | |
972 | return rq; | |
1da177e4 | 973 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 974 | } |
1da177e4 LT |
975 | } |
976 | ||
ad474cac ON |
977 | void task_rq_unlock_wait(struct task_struct *p) |
978 | { | |
979 | struct rq *rq = task_rq(p); | |
980 | ||
981 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
982 | spin_unlock_wait(&rq->lock); | |
983 | } | |
984 | ||
a9957449 | 985 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
986 | __releases(rq->lock) |
987 | { | |
988 | spin_unlock(&rq->lock); | |
989 | } | |
990 | ||
70b97a7f | 991 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
992 | __releases(rq->lock) |
993 | { | |
994 | spin_unlock_irqrestore(&rq->lock, *flags); | |
995 | } | |
996 | ||
1da177e4 | 997 | /* |
cc2a73b5 | 998 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 999 | */ |
a9957449 | 1000 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1001 | __acquires(rq->lock) |
1002 | { | |
70b97a7f | 1003 | struct rq *rq; |
1da177e4 LT |
1004 | |
1005 | local_irq_disable(); | |
1006 | rq = this_rq(); | |
1007 | spin_lock(&rq->lock); | |
1008 | ||
1009 | return rq; | |
1010 | } | |
1011 | ||
8f4d37ec PZ |
1012 | #ifdef CONFIG_SCHED_HRTICK |
1013 | /* | |
1014 | * Use HR-timers to deliver accurate preemption points. | |
1015 | * | |
1016 | * Its all a bit involved since we cannot program an hrt while holding the | |
1017 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1018 | * reschedule event. | |
1019 | * | |
1020 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1021 | * rq->lock. | |
1022 | */ | |
8f4d37ec PZ |
1023 | |
1024 | /* | |
1025 | * Use hrtick when: | |
1026 | * - enabled by features | |
1027 | * - hrtimer is actually high res | |
1028 | */ | |
1029 | static inline int hrtick_enabled(struct rq *rq) | |
1030 | { | |
1031 | if (!sched_feat(HRTICK)) | |
1032 | return 0; | |
ba42059f | 1033 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1034 | return 0; |
8f4d37ec PZ |
1035 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1036 | } | |
1037 | ||
8f4d37ec PZ |
1038 | static void hrtick_clear(struct rq *rq) |
1039 | { | |
1040 | if (hrtimer_active(&rq->hrtick_timer)) | |
1041 | hrtimer_cancel(&rq->hrtick_timer); | |
1042 | } | |
1043 | ||
8f4d37ec PZ |
1044 | /* |
1045 | * High-resolution timer tick. | |
1046 | * Runs from hardirq context with interrupts disabled. | |
1047 | */ | |
1048 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1049 | { | |
1050 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1051 | ||
1052 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1053 | ||
1054 | spin_lock(&rq->lock); | |
3e51f33f | 1055 | update_rq_clock(rq); |
8f4d37ec PZ |
1056 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1057 | spin_unlock(&rq->lock); | |
1058 | ||
1059 | return HRTIMER_NORESTART; | |
1060 | } | |
1061 | ||
95e904c7 | 1062 | #ifdef CONFIG_SMP |
31656519 PZ |
1063 | /* |
1064 | * called from hardirq (IPI) context | |
1065 | */ | |
1066 | static void __hrtick_start(void *arg) | |
b328ca18 | 1067 | { |
31656519 | 1068 | struct rq *rq = arg; |
b328ca18 | 1069 | |
31656519 PZ |
1070 | spin_lock(&rq->lock); |
1071 | hrtimer_restart(&rq->hrtick_timer); | |
1072 | rq->hrtick_csd_pending = 0; | |
1073 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1074 | } |
1075 | ||
31656519 PZ |
1076 | /* |
1077 | * Called to set the hrtick timer state. | |
1078 | * | |
1079 | * called with rq->lock held and irqs disabled | |
1080 | */ | |
1081 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1082 | { |
31656519 PZ |
1083 | struct hrtimer *timer = &rq->hrtick_timer; |
1084 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1085 | |
cc584b21 | 1086 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1087 | |
1088 | if (rq == this_rq()) { | |
1089 | hrtimer_restart(timer); | |
1090 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1091 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1092 | rq->hrtick_csd_pending = 1; |
1093 | } | |
b328ca18 PZ |
1094 | } |
1095 | ||
1096 | static int | |
1097 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1098 | { | |
1099 | int cpu = (int)(long)hcpu; | |
1100 | ||
1101 | switch (action) { | |
1102 | case CPU_UP_CANCELED: | |
1103 | case CPU_UP_CANCELED_FROZEN: | |
1104 | case CPU_DOWN_PREPARE: | |
1105 | case CPU_DOWN_PREPARE_FROZEN: | |
1106 | case CPU_DEAD: | |
1107 | case CPU_DEAD_FROZEN: | |
31656519 | 1108 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1109 | return NOTIFY_OK; |
1110 | } | |
1111 | ||
1112 | return NOTIFY_DONE; | |
1113 | } | |
1114 | ||
fa748203 | 1115 | static __init void init_hrtick(void) |
b328ca18 PZ |
1116 | { |
1117 | hotcpu_notifier(hotplug_hrtick, 0); | |
1118 | } | |
31656519 PZ |
1119 | #else |
1120 | /* | |
1121 | * Called to set the hrtick timer state. | |
1122 | * | |
1123 | * called with rq->lock held and irqs disabled | |
1124 | */ | |
1125 | static void hrtick_start(struct rq *rq, u64 delay) | |
1126 | { | |
7f1e2ca9 | 1127 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1128 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1129 | } |
b328ca18 | 1130 | |
006c75f1 | 1131 | static inline void init_hrtick(void) |
8f4d37ec | 1132 | { |
8f4d37ec | 1133 | } |
31656519 | 1134 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1135 | |
31656519 | 1136 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1137 | { |
31656519 PZ |
1138 | #ifdef CONFIG_SMP |
1139 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1140 | |
31656519 PZ |
1141 | rq->hrtick_csd.flags = 0; |
1142 | rq->hrtick_csd.func = __hrtick_start; | |
1143 | rq->hrtick_csd.info = rq; | |
1144 | #endif | |
8f4d37ec | 1145 | |
31656519 PZ |
1146 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1147 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1148 | } |
006c75f1 | 1149 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1150 | static inline void hrtick_clear(struct rq *rq) |
1151 | { | |
1152 | } | |
1153 | ||
8f4d37ec PZ |
1154 | static inline void init_rq_hrtick(struct rq *rq) |
1155 | { | |
1156 | } | |
1157 | ||
b328ca18 PZ |
1158 | static inline void init_hrtick(void) |
1159 | { | |
1160 | } | |
006c75f1 | 1161 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1162 | |
c24d20db IM |
1163 | /* |
1164 | * resched_task - mark a task 'to be rescheduled now'. | |
1165 | * | |
1166 | * On UP this means the setting of the need_resched flag, on SMP it | |
1167 | * might also involve a cross-CPU call to trigger the scheduler on | |
1168 | * the target CPU. | |
1169 | */ | |
1170 | #ifdef CONFIG_SMP | |
1171 | ||
1172 | #ifndef tsk_is_polling | |
1173 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1174 | #endif | |
1175 | ||
31656519 | 1176 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1177 | { |
1178 | int cpu; | |
1179 | ||
1180 | assert_spin_locked(&task_rq(p)->lock); | |
1181 | ||
5ed0cec0 | 1182 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1183 | return; |
1184 | ||
5ed0cec0 | 1185 | set_tsk_need_resched(p); |
c24d20db IM |
1186 | |
1187 | cpu = task_cpu(p); | |
1188 | if (cpu == smp_processor_id()) | |
1189 | return; | |
1190 | ||
1191 | /* NEED_RESCHED must be visible before we test polling */ | |
1192 | smp_mb(); | |
1193 | if (!tsk_is_polling(p)) | |
1194 | smp_send_reschedule(cpu); | |
1195 | } | |
1196 | ||
1197 | static void resched_cpu(int cpu) | |
1198 | { | |
1199 | struct rq *rq = cpu_rq(cpu); | |
1200 | unsigned long flags; | |
1201 | ||
1202 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1203 | return; | |
1204 | resched_task(cpu_curr(cpu)); | |
1205 | spin_unlock_irqrestore(&rq->lock, flags); | |
1206 | } | |
06d8308c TG |
1207 | |
1208 | #ifdef CONFIG_NO_HZ | |
1209 | /* | |
1210 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1211 | * idle CPU then this timer might expire before the next timer event | |
1212 | * which is scheduled to wake up that CPU. In case of a completely | |
1213 | * idle system the next event might even be infinite time into the | |
1214 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1215 | * leaves the inner idle loop so the newly added timer is taken into | |
1216 | * account when the CPU goes back to idle and evaluates the timer | |
1217 | * wheel for the next timer event. | |
1218 | */ | |
1219 | void wake_up_idle_cpu(int cpu) | |
1220 | { | |
1221 | struct rq *rq = cpu_rq(cpu); | |
1222 | ||
1223 | if (cpu == smp_processor_id()) | |
1224 | return; | |
1225 | ||
1226 | /* | |
1227 | * This is safe, as this function is called with the timer | |
1228 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1229 | * to idle and has not yet set rq->curr to idle then it will | |
1230 | * be serialized on the timer wheel base lock and take the new | |
1231 | * timer into account automatically. | |
1232 | */ | |
1233 | if (rq->curr != rq->idle) | |
1234 | return; | |
1235 | ||
1236 | /* | |
1237 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1238 | * lockless. The worst case is that the other CPU runs the | |
1239 | * idle task through an additional NOOP schedule() | |
1240 | */ | |
5ed0cec0 | 1241 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1242 | |
1243 | /* NEED_RESCHED must be visible before we test polling */ | |
1244 | smp_mb(); | |
1245 | if (!tsk_is_polling(rq->idle)) | |
1246 | smp_send_reschedule(cpu); | |
1247 | } | |
6d6bc0ad | 1248 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1249 | |
e9e9250b PZ |
1250 | static u64 sched_avg_period(void) |
1251 | { | |
1252 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1253 | } | |
1254 | ||
1255 | static void sched_avg_update(struct rq *rq) | |
1256 | { | |
1257 | s64 period = sched_avg_period(); | |
1258 | ||
1259 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1260 | rq->age_stamp += period; | |
1261 | rq->rt_avg /= 2; | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1266 | { | |
1267 | rq->rt_avg += rt_delta; | |
1268 | sched_avg_update(rq); | |
1269 | } | |
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 | } |
e9e9250b PZ |
1277 | |
1278 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1279 | { | |
1280 | } | |
6d6bc0ad | 1281 | #endif /* CONFIG_SMP */ |
c24d20db | 1282 | |
45bf76df IM |
1283 | #if BITS_PER_LONG == 32 |
1284 | # define WMULT_CONST (~0UL) | |
1285 | #else | |
1286 | # define WMULT_CONST (1UL << 32) | |
1287 | #endif | |
1288 | ||
1289 | #define WMULT_SHIFT 32 | |
1290 | ||
194081eb IM |
1291 | /* |
1292 | * Shift right and round: | |
1293 | */ | |
cf2ab469 | 1294 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1295 | |
a7be37ac PZ |
1296 | /* |
1297 | * delta *= weight / lw | |
1298 | */ | |
cb1c4fc9 | 1299 | static unsigned long |
45bf76df IM |
1300 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1301 | struct load_weight *lw) | |
1302 | { | |
1303 | u64 tmp; | |
1304 | ||
7a232e03 LJ |
1305 | if (!lw->inv_weight) { |
1306 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1307 | lw->inv_weight = 1; | |
1308 | else | |
1309 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1310 | / (lw->weight+1); | |
1311 | } | |
45bf76df IM |
1312 | |
1313 | tmp = (u64)delta_exec * weight; | |
1314 | /* | |
1315 | * Check whether we'd overflow the 64-bit multiplication: | |
1316 | */ | |
194081eb | 1317 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1318 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1319 | WMULT_SHIFT/2); |
1320 | else | |
cf2ab469 | 1321 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1322 | |
ecf691da | 1323 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1324 | } |
1325 | ||
1091985b | 1326 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1327 | { |
1328 | lw->weight += inc; | |
e89996ae | 1329 | lw->inv_weight = 0; |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1333 | { |
1334 | lw->weight -= dec; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
2dd73a4f PW |
1338 | /* |
1339 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1340 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1341 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1342 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1343 | * scaled version of the new time slice allocation that they receive on time |
1344 | * slice expiry etc. | |
1345 | */ | |
1346 | ||
cce7ade8 PZ |
1347 | #define WEIGHT_IDLEPRIO 3 |
1348 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1349 | |
1350 | /* | |
1351 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1352 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1353 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1354 | * that remained on nice 0. | |
1355 | * | |
1356 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1357 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1358 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1359 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1360 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1361 | */ |
1362 | static const int prio_to_weight[40] = { | |
254753dc IM |
1363 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1364 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1365 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1366 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1367 | /* 0 */ 1024, 820, 655, 526, 423, | |
1368 | /* 5 */ 335, 272, 215, 172, 137, | |
1369 | /* 10 */ 110, 87, 70, 56, 45, | |
1370 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1371 | }; |
1372 | ||
5714d2de IM |
1373 | /* |
1374 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1375 | * | |
1376 | * In cases where the weight does not change often, we can use the | |
1377 | * precalculated inverse to speed up arithmetics by turning divisions | |
1378 | * into multiplications: | |
1379 | */ | |
dd41f596 | 1380 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1381 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1382 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1383 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1384 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1385 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1386 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1387 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1388 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1389 | }; |
2dd73a4f | 1390 | |
dd41f596 IM |
1391 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1392 | ||
1393 | /* | |
1394 | * runqueue iterator, to support SMP load-balancing between different | |
1395 | * scheduling classes, without having to expose their internal data | |
1396 | * structures to the load-balancing proper: | |
1397 | */ | |
1398 | struct rq_iterator { | |
1399 | void *arg; | |
1400 | struct task_struct *(*start)(void *); | |
1401 | struct task_struct *(*next)(void *); | |
1402 | }; | |
1403 | ||
e1d1484f PW |
1404 | #ifdef CONFIG_SMP |
1405 | static unsigned long | |
1406 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1407 | unsigned long max_load_move, struct sched_domain *sd, | |
1408 | enum cpu_idle_type idle, int *all_pinned, | |
1409 | int *this_best_prio, struct rq_iterator *iterator); | |
1410 | ||
1411 | static int | |
1412 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1413 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1414 | struct rq_iterator *iterator); | |
e1d1484f | 1415 | #endif |
dd41f596 | 1416 | |
ef12fefa BR |
1417 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1418 | enum cpuacct_stat_index { | |
1419 | CPUACCT_STAT_USER, /* ... user mode */ | |
1420 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1421 | ||
1422 | CPUACCT_STAT_NSTATS, | |
1423 | }; | |
1424 | ||
d842de87 SV |
1425 | #ifdef CONFIG_CGROUP_CPUACCT |
1426 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1427 | static void cpuacct_update_stats(struct task_struct *tsk, |
1428 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1429 | #else |
1430 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1431 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1432 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1433 | #endif |
1434 | ||
18d95a28 PZ |
1435 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1436 | { | |
1437 | update_load_add(&rq->load, load); | |
1438 | } | |
1439 | ||
1440 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1441 | { | |
1442 | update_load_sub(&rq->load, load); | |
1443 | } | |
1444 | ||
7940ca36 | 1445 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1446 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1447 | |
1448 | /* | |
1449 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1450 | * leaving it for the final time. | |
1451 | */ | |
eb755805 | 1452 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1453 | { |
1454 | struct task_group *parent, *child; | |
eb755805 | 1455 | int ret; |
c09595f6 PZ |
1456 | |
1457 | rcu_read_lock(); | |
1458 | parent = &root_task_group; | |
1459 | down: | |
eb755805 PZ |
1460 | ret = (*down)(parent, data); |
1461 | if (ret) | |
1462 | goto out_unlock; | |
c09595f6 PZ |
1463 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1464 | parent = child; | |
1465 | goto down; | |
1466 | ||
1467 | up: | |
1468 | continue; | |
1469 | } | |
eb755805 PZ |
1470 | ret = (*up)(parent, data); |
1471 | if (ret) | |
1472 | goto out_unlock; | |
c09595f6 PZ |
1473 | |
1474 | child = parent; | |
1475 | parent = parent->parent; | |
1476 | if (parent) | |
1477 | goto up; | |
eb755805 | 1478 | out_unlock: |
c09595f6 | 1479 | rcu_read_unlock(); |
eb755805 PZ |
1480 | |
1481 | return ret; | |
c09595f6 PZ |
1482 | } |
1483 | ||
eb755805 PZ |
1484 | static int tg_nop(struct task_group *tg, void *data) |
1485 | { | |
1486 | return 0; | |
c09595f6 | 1487 | } |
eb755805 PZ |
1488 | #endif |
1489 | ||
1490 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1491 | /* Used instead of source_load when we know the type == 0 */ |
1492 | static unsigned long weighted_cpuload(const int cpu) | |
1493 | { | |
1494 | return cpu_rq(cpu)->load.weight; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Return a low guess at the load of a migration-source cpu weighted | |
1499 | * according to the scheduling class and "nice" value. | |
1500 | * | |
1501 | * We want to under-estimate the load of migration sources, to | |
1502 | * balance conservatively. | |
1503 | */ | |
1504 | static unsigned long source_load(int cpu, int type) | |
1505 | { | |
1506 | struct rq *rq = cpu_rq(cpu); | |
1507 | unsigned long total = weighted_cpuload(cpu); | |
1508 | ||
1509 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1510 | return total; | |
1511 | ||
1512 | return min(rq->cpu_load[type-1], total); | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * Return a high guess at the load of a migration-target cpu weighted | |
1517 | * according to the scheduling class and "nice" value. | |
1518 | */ | |
1519 | static unsigned long target_load(int cpu, int type) | |
1520 | { | |
1521 | struct rq *rq = cpu_rq(cpu); | |
1522 | unsigned long total = weighted_cpuload(cpu); | |
1523 | ||
1524 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1525 | return total; | |
1526 | ||
1527 | return max(rq->cpu_load[type-1], total); | |
1528 | } | |
1529 | ||
ae154be1 PZ |
1530 | static struct sched_group *group_of(int cpu) |
1531 | { | |
1532 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
1533 | ||
1534 | if (!sd) | |
1535 | return NULL; | |
1536 | ||
1537 | return sd->groups; | |
1538 | } | |
1539 | ||
1540 | static unsigned long power_of(int cpu) | |
1541 | { | |
1542 | struct sched_group *group = group_of(cpu); | |
1543 | ||
1544 | if (!group) | |
1545 | return SCHED_LOAD_SCALE; | |
1546 | ||
1547 | return group->cpu_power; | |
1548 | } | |
1549 | ||
eb755805 PZ |
1550 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1551 | ||
1552 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1553 | { | |
1554 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1555 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1556 | |
4cd42620 SR |
1557 | if (nr_running) |
1558 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1559 | else |
1560 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1561 | |
1562 | return rq->avg_load_per_task; | |
1563 | } | |
1564 | ||
1565 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1566 | |
34d76c41 PZ |
1567 | struct update_shares_data { |
1568 | unsigned long rq_weight[NR_CPUS]; | |
1569 | }; | |
1570 | ||
1571 | static DEFINE_PER_CPU(struct update_shares_data, update_shares_data); | |
1572 | ||
c09595f6 PZ |
1573 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1574 | ||
1575 | /* | |
1576 | * Calculate and set the cpu's group shares. | |
1577 | */ | |
34d76c41 PZ |
1578 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1579 | unsigned long sd_shares, | |
1580 | unsigned long sd_rq_weight, | |
1581 | struct update_shares_data *usd) | |
18d95a28 | 1582 | { |
34d76c41 | 1583 | unsigned long shares, rq_weight; |
a5004278 | 1584 | int boost = 0; |
c09595f6 | 1585 | |
34d76c41 | 1586 | rq_weight = usd->rq_weight[cpu]; |
a5004278 PZ |
1587 | if (!rq_weight) { |
1588 | boost = 1; | |
1589 | rq_weight = NICE_0_LOAD; | |
1590 | } | |
c8cba857 | 1591 | |
c09595f6 | 1592 | /* |
a8af7246 PZ |
1593 | * \Sum_j shares_j * rq_weight_i |
1594 | * shares_i = ----------------------------- | |
1595 | * \Sum_j rq_weight_j | |
c09595f6 | 1596 | */ |
ec4e0e2f | 1597 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1598 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1599 | |
ffda12a1 PZ |
1600 | if (abs(shares - tg->se[cpu]->load.weight) > |
1601 | sysctl_sched_shares_thresh) { | |
1602 | struct rq *rq = cpu_rq(cpu); | |
1603 | unsigned long flags; | |
c09595f6 | 1604 | |
ffda12a1 | 1605 | spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1606 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1607 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1608 | __set_se_shares(tg->se[cpu], shares); |
1609 | spin_unlock_irqrestore(&rq->lock, flags); | |
1610 | } | |
18d95a28 | 1611 | } |
c09595f6 PZ |
1612 | |
1613 | /* | |
c8cba857 PZ |
1614 | * Re-compute the task group their per cpu shares over the given domain. |
1615 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1616 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1617 | */ |
eb755805 | 1618 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1619 | { |
34d76c41 PZ |
1620 | unsigned long weight, rq_weight = 0, shares = 0; |
1621 | struct update_shares_data *usd; | |
eb755805 | 1622 | struct sched_domain *sd = data; |
34d76c41 | 1623 | unsigned long flags; |
c8cba857 | 1624 | int i; |
c09595f6 | 1625 | |
34d76c41 PZ |
1626 | if (!tg->se[0]) |
1627 | return 0; | |
1628 | ||
1629 | local_irq_save(flags); | |
1630 | usd = &__get_cpu_var(update_shares_data); | |
1631 | ||
758b2cdc | 1632 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 PZ |
1633 | weight = tg->cfs_rq[i]->load.weight; |
1634 | usd->rq_weight[i] = weight; | |
1635 | ||
ec4e0e2f KC |
1636 | /* |
1637 | * If there are currently no tasks on the cpu pretend there | |
1638 | * is one of average load so that when a new task gets to | |
1639 | * run here it will not get delayed by group starvation. | |
1640 | */ | |
ec4e0e2f KC |
1641 | if (!weight) |
1642 | weight = NICE_0_LOAD; | |
1643 | ||
ec4e0e2f | 1644 | rq_weight += weight; |
c8cba857 | 1645 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1646 | } |
c09595f6 | 1647 | |
c8cba857 PZ |
1648 | if ((!shares && rq_weight) || shares > tg->shares) |
1649 | shares = tg->shares; | |
1650 | ||
1651 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1652 | shares = tg->shares; | |
c09595f6 | 1653 | |
758b2cdc | 1654 | for_each_cpu(i, sched_domain_span(sd)) |
34d76c41 PZ |
1655 | update_group_shares_cpu(tg, i, shares, rq_weight, usd); |
1656 | ||
1657 | local_irq_restore(flags); | |
eb755805 PZ |
1658 | |
1659 | return 0; | |
c09595f6 PZ |
1660 | } |
1661 | ||
1662 | /* | |
c8cba857 PZ |
1663 | * Compute the cpu's hierarchical load factor for each task group. |
1664 | * This needs to be done in a top-down fashion because the load of a child | |
1665 | * group is a fraction of its parents load. | |
c09595f6 | 1666 | */ |
eb755805 | 1667 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1668 | { |
c8cba857 | 1669 | unsigned long load; |
eb755805 | 1670 | long cpu = (long)data; |
c09595f6 | 1671 | |
c8cba857 PZ |
1672 | if (!tg->parent) { |
1673 | load = cpu_rq(cpu)->load.weight; | |
1674 | } else { | |
1675 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1676 | load *= tg->cfs_rq[cpu]->shares; | |
1677 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1678 | } | |
c09595f6 | 1679 | |
c8cba857 | 1680 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1681 | |
eb755805 | 1682 | return 0; |
c09595f6 PZ |
1683 | } |
1684 | ||
c8cba857 | 1685 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1686 | { |
e7097159 PZ |
1687 | s64 elapsed; |
1688 | u64 now; | |
1689 | ||
1690 | if (root_task_group_empty()) | |
1691 | return; | |
1692 | ||
1693 | now = cpu_clock(raw_smp_processor_id()); | |
1694 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1695 | |
1696 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1697 | sd->last_update = now; | |
eb755805 | 1698 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1699 | } |
4d8d595d PZ |
1700 | } |
1701 | ||
3e5459b4 PZ |
1702 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1703 | { | |
e7097159 PZ |
1704 | if (root_task_group_empty()) |
1705 | return; | |
1706 | ||
3e5459b4 PZ |
1707 | spin_unlock(&rq->lock); |
1708 | update_shares(sd); | |
1709 | spin_lock(&rq->lock); | |
1710 | } | |
1711 | ||
eb755805 | 1712 | static void update_h_load(long cpu) |
c09595f6 | 1713 | { |
e7097159 PZ |
1714 | if (root_task_group_empty()) |
1715 | return; | |
1716 | ||
eb755805 | 1717 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1718 | } |
1719 | ||
c09595f6 PZ |
1720 | #else |
1721 | ||
c8cba857 | 1722 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1723 | { |
1724 | } | |
1725 | ||
3e5459b4 PZ |
1726 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1727 | { | |
1728 | } | |
1729 | ||
18d95a28 PZ |
1730 | #endif |
1731 | ||
8f45e2b5 GH |
1732 | #ifdef CONFIG_PREEMPT |
1733 | ||
b78bb868 PZ |
1734 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1735 | ||
70574a99 | 1736 | /* |
8f45e2b5 GH |
1737 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1738 | * way at the expense of forcing extra atomic operations in all | |
1739 | * invocations. This assures that the double_lock is acquired using the | |
1740 | * same underlying policy as the spinlock_t on this architecture, which | |
1741 | * reduces latency compared to the unfair variant below. However, it | |
1742 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1743 | */ |
8f45e2b5 GH |
1744 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1745 | __releases(this_rq->lock) | |
1746 | __acquires(busiest->lock) | |
1747 | __acquires(this_rq->lock) | |
1748 | { | |
1749 | spin_unlock(&this_rq->lock); | |
1750 | double_rq_lock(this_rq, busiest); | |
1751 | ||
1752 | return 1; | |
1753 | } | |
1754 | ||
1755 | #else | |
1756 | /* | |
1757 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1758 | * latency by eliminating extra atomic operations when the locks are | |
1759 | * already in proper order on entry. This favors lower cpu-ids and will | |
1760 | * grant the double lock to lower cpus over higher ids under contention, | |
1761 | * regardless of entry order into the function. | |
1762 | */ | |
1763 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1764 | __releases(this_rq->lock) |
1765 | __acquires(busiest->lock) | |
1766 | __acquires(this_rq->lock) | |
1767 | { | |
1768 | int ret = 0; | |
1769 | ||
70574a99 AD |
1770 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1771 | if (busiest < this_rq) { | |
1772 | spin_unlock(&this_rq->lock); | |
1773 | spin_lock(&busiest->lock); | |
1774 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1775 | ret = 1; | |
1776 | } else | |
1777 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1778 | } | |
1779 | return ret; | |
1780 | } | |
1781 | ||
8f45e2b5 GH |
1782 | #endif /* CONFIG_PREEMPT */ |
1783 | ||
1784 | /* | |
1785 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1786 | */ | |
1787 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1788 | { | |
1789 | if (unlikely(!irqs_disabled())) { | |
1790 | /* printk() doesn't work good under rq->lock */ | |
1791 | spin_unlock(&this_rq->lock); | |
1792 | BUG_ON(1); | |
1793 | } | |
1794 | ||
1795 | return _double_lock_balance(this_rq, busiest); | |
1796 | } | |
1797 | ||
70574a99 AD |
1798 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1799 | __releases(busiest->lock) | |
1800 | { | |
1801 | spin_unlock(&busiest->lock); | |
1802 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1803 | } | |
18d95a28 PZ |
1804 | #endif |
1805 | ||
30432094 | 1806 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1807 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1808 | { | |
30432094 | 1809 | #ifdef CONFIG_SMP |
34e83e85 IM |
1810 | cfs_rq->shares = shares; |
1811 | #endif | |
1812 | } | |
30432094 | 1813 | #endif |
e7693a36 | 1814 | |
dce48a84 TG |
1815 | static void calc_load_account_active(struct rq *this_rq); |
1816 | ||
dd41f596 | 1817 | #include "sched_stats.h" |
dd41f596 | 1818 | #include "sched_idletask.c" |
5522d5d5 IM |
1819 | #include "sched_fair.c" |
1820 | #include "sched_rt.c" | |
dd41f596 IM |
1821 | #ifdef CONFIG_SCHED_DEBUG |
1822 | # include "sched_debug.c" | |
1823 | #endif | |
1824 | ||
1825 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1826 | #define for_each_class(class) \ |
1827 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1828 | |
c09595f6 | 1829 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1830 | { |
1831 | rq->nr_running++; | |
9c217245 IM |
1832 | } |
1833 | ||
c09595f6 | 1834 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1835 | { |
1836 | rq->nr_running--; | |
9c217245 IM |
1837 | } |
1838 | ||
45bf76df IM |
1839 | static void set_load_weight(struct task_struct *p) |
1840 | { | |
1841 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1842 | p->se.load.weight = prio_to_weight[0] * 2; |
1843 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1844 | return; | |
1845 | } | |
45bf76df | 1846 | |
dd41f596 IM |
1847 | /* |
1848 | * SCHED_IDLE tasks get minimal weight: | |
1849 | */ | |
1850 | if (p->policy == SCHED_IDLE) { | |
1851 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1852 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1853 | return; | |
1854 | } | |
71f8bd46 | 1855 | |
dd41f596 IM |
1856 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1857 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1858 | } |
1859 | ||
2087a1ad GH |
1860 | static void update_avg(u64 *avg, u64 sample) |
1861 | { | |
1862 | s64 diff = sample - *avg; | |
1863 | *avg += diff >> 3; | |
1864 | } | |
1865 | ||
8159f87e | 1866 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1867 | { |
831451ac PZ |
1868 | if (wakeup) |
1869 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1870 | ||
dd41f596 | 1871 | sched_info_queued(p); |
fd390f6a | 1872 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1873 | p->se.on_rq = 1; |
71f8bd46 IM |
1874 | } |
1875 | ||
69be72c1 | 1876 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1877 | { |
831451ac PZ |
1878 | if (sleep) { |
1879 | if (p->se.last_wakeup) { | |
1880 | update_avg(&p->se.avg_overlap, | |
1881 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1882 | p->se.last_wakeup = 0; | |
1883 | } else { | |
1884 | update_avg(&p->se.avg_wakeup, | |
1885 | sysctl_sched_wakeup_granularity); | |
1886 | } | |
2087a1ad GH |
1887 | } |
1888 | ||
46ac22ba | 1889 | sched_info_dequeued(p); |
f02231e5 | 1890 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1891 | p->se.on_rq = 0; |
71f8bd46 IM |
1892 | } |
1893 | ||
14531189 | 1894 | /* |
dd41f596 | 1895 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1896 | */ |
14531189 IM |
1897 | static inline int __normal_prio(struct task_struct *p) |
1898 | { | |
dd41f596 | 1899 | return p->static_prio; |
14531189 IM |
1900 | } |
1901 | ||
b29739f9 IM |
1902 | /* |
1903 | * Calculate the expected normal priority: i.e. priority | |
1904 | * without taking RT-inheritance into account. Might be | |
1905 | * boosted by interactivity modifiers. Changes upon fork, | |
1906 | * setprio syscalls, and whenever the interactivity | |
1907 | * estimator recalculates. | |
1908 | */ | |
36c8b586 | 1909 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1910 | { |
1911 | int prio; | |
1912 | ||
e05606d3 | 1913 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1914 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1915 | else | |
1916 | prio = __normal_prio(p); | |
1917 | return prio; | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * Calculate the current priority, i.e. the priority | |
1922 | * taken into account by the scheduler. This value might | |
1923 | * be boosted by RT tasks, or might be boosted by | |
1924 | * interactivity modifiers. Will be RT if the task got | |
1925 | * RT-boosted. If not then it returns p->normal_prio. | |
1926 | */ | |
36c8b586 | 1927 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1928 | { |
1929 | p->normal_prio = normal_prio(p); | |
1930 | /* | |
1931 | * If we are RT tasks or we were boosted to RT priority, | |
1932 | * keep the priority unchanged. Otherwise, update priority | |
1933 | * to the normal priority: | |
1934 | */ | |
1935 | if (!rt_prio(p->prio)) | |
1936 | return p->normal_prio; | |
1937 | return p->prio; | |
1938 | } | |
1939 | ||
1da177e4 | 1940 | /* |
dd41f596 | 1941 | * activate_task - move a task to the runqueue. |
1da177e4 | 1942 | */ |
dd41f596 | 1943 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1944 | { |
d9514f6c | 1945 | if (task_contributes_to_load(p)) |
dd41f596 | 1946 | rq->nr_uninterruptible--; |
1da177e4 | 1947 | |
8159f87e | 1948 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1949 | inc_nr_running(rq); |
1da177e4 LT |
1950 | } |
1951 | ||
1da177e4 LT |
1952 | /* |
1953 | * deactivate_task - remove a task from the runqueue. | |
1954 | */ | |
2e1cb74a | 1955 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1956 | { |
d9514f6c | 1957 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1958 | rq->nr_uninterruptible++; |
1959 | ||
69be72c1 | 1960 | dequeue_task(rq, p, sleep); |
c09595f6 | 1961 | dec_nr_running(rq); |
1da177e4 LT |
1962 | } |
1963 | ||
1da177e4 LT |
1964 | /** |
1965 | * task_curr - is this task currently executing on a CPU? | |
1966 | * @p: the task in question. | |
1967 | */ | |
36c8b586 | 1968 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1969 | { |
1970 | return cpu_curr(task_cpu(p)) == p; | |
1971 | } | |
1972 | ||
dd41f596 IM |
1973 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1974 | { | |
6f505b16 | 1975 | set_task_rq(p, cpu); |
dd41f596 | 1976 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1977 | /* |
1978 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1979 | * successfuly executed on another CPU. We must ensure that updates of | |
1980 | * per-task data have been completed by this moment. | |
1981 | */ | |
1982 | smp_wmb(); | |
dd41f596 | 1983 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1984 | #endif |
2dd73a4f PW |
1985 | } |
1986 | ||
cb469845 SR |
1987 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1988 | const struct sched_class *prev_class, | |
1989 | int oldprio, int running) | |
1990 | { | |
1991 | if (prev_class != p->sched_class) { | |
1992 | if (prev_class->switched_from) | |
1993 | prev_class->switched_from(rq, p, running); | |
1994 | p->sched_class->switched_to(rq, p, running); | |
1995 | } else | |
1996 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1997 | } | |
1998 | ||
1da177e4 | 1999 | #ifdef CONFIG_SMP |
cc367732 IM |
2000 | /* |
2001 | * Is this task likely cache-hot: | |
2002 | */ | |
e7693a36 | 2003 | static int |
cc367732 IM |
2004 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2005 | { | |
2006 | s64 delta; | |
2007 | ||
f540a608 IM |
2008 | /* |
2009 | * Buddy candidates are cache hot: | |
2010 | */ | |
4793241b PZ |
2011 | if (sched_feat(CACHE_HOT_BUDDY) && |
2012 | (&p->se == cfs_rq_of(&p->se)->next || | |
2013 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2014 | return 1; |
2015 | ||
cc367732 IM |
2016 | if (p->sched_class != &fair_sched_class) |
2017 | return 0; | |
2018 | ||
6bc1665b IM |
2019 | if (sysctl_sched_migration_cost == -1) |
2020 | return 1; | |
2021 | if (sysctl_sched_migration_cost == 0) | |
2022 | return 0; | |
2023 | ||
cc367732 IM |
2024 | delta = now - p->se.exec_start; |
2025 | ||
2026 | return delta < (s64)sysctl_sched_migration_cost; | |
2027 | } | |
2028 | ||
2029 | ||
dd41f596 | 2030 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2031 | { |
dd41f596 IM |
2032 | int old_cpu = task_cpu(p); |
2033 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
2034 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
2035 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 2036 | u64 clock_offset; |
dd41f596 IM |
2037 | |
2038 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 2039 | |
de1d7286 | 2040 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2041 | |
6cfb0d5d IM |
2042 | #ifdef CONFIG_SCHEDSTATS |
2043 | if (p->se.wait_start) | |
2044 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
2045 | if (p->se.sleep_start) |
2046 | p->se.sleep_start -= clock_offset; | |
2047 | if (p->se.block_start) | |
2048 | p->se.block_start -= clock_offset; | |
6c594c21 | 2049 | #endif |
cc367732 | 2050 | if (old_cpu != new_cpu) { |
6c594c21 IM |
2051 | p->se.nr_migrations++; |
2052 | #ifdef CONFIG_SCHEDSTATS | |
cc367732 IM |
2053 | if (task_hot(p, old_rq->clock, NULL)) |
2054 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 2055 | #endif |
cdd6c482 | 2056 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
e5289d4a | 2057 | 1, 1, NULL, 0); |
6c594c21 | 2058 | } |
2830cf8c SV |
2059 | p->se.vruntime -= old_cfsrq->min_vruntime - |
2060 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
2061 | |
2062 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2063 | } |
2064 | ||
70b97a7f | 2065 | struct migration_req { |
1da177e4 | 2066 | struct list_head list; |
1da177e4 | 2067 | |
36c8b586 | 2068 | struct task_struct *task; |
1da177e4 LT |
2069 | int dest_cpu; |
2070 | ||
1da177e4 | 2071 | struct completion done; |
70b97a7f | 2072 | }; |
1da177e4 LT |
2073 | |
2074 | /* | |
2075 | * The task's runqueue lock must be held. | |
2076 | * Returns true if you have to wait for migration thread. | |
2077 | */ | |
36c8b586 | 2078 | static int |
70b97a7f | 2079 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2080 | { |
70b97a7f | 2081 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2082 | |
2083 | /* | |
2084 | * If the task is not on a runqueue (and not running), then | |
2085 | * it is sufficient to simply update the task's cpu field. | |
2086 | */ | |
dd41f596 | 2087 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2088 | set_task_cpu(p, dest_cpu); |
2089 | return 0; | |
2090 | } | |
2091 | ||
2092 | init_completion(&req->done); | |
1da177e4 LT |
2093 | req->task = p; |
2094 | req->dest_cpu = dest_cpu; | |
2095 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2096 | |
1da177e4 LT |
2097 | return 1; |
2098 | } | |
2099 | ||
a26b89f0 MM |
2100 | /* |
2101 | * wait_task_context_switch - wait for a thread to complete at least one | |
2102 | * context switch. | |
2103 | * | |
2104 | * @p must not be current. | |
2105 | */ | |
2106 | void wait_task_context_switch(struct task_struct *p) | |
2107 | { | |
2108 | unsigned long nvcsw, nivcsw, flags; | |
2109 | int running; | |
2110 | struct rq *rq; | |
2111 | ||
2112 | nvcsw = p->nvcsw; | |
2113 | nivcsw = p->nivcsw; | |
2114 | for (;;) { | |
2115 | /* | |
2116 | * The runqueue is assigned before the actual context | |
2117 | * switch. We need to take the runqueue lock. | |
2118 | * | |
2119 | * We could check initially without the lock but it is | |
2120 | * very likely that we need to take the lock in every | |
2121 | * iteration. | |
2122 | */ | |
2123 | rq = task_rq_lock(p, &flags); | |
2124 | running = task_running(rq, p); | |
2125 | task_rq_unlock(rq, &flags); | |
2126 | ||
2127 | if (likely(!running)) | |
2128 | break; | |
2129 | /* | |
2130 | * The switch count is incremented before the actual | |
2131 | * context switch. We thus wait for two switches to be | |
2132 | * sure at least one completed. | |
2133 | */ | |
2134 | if ((p->nvcsw - nvcsw) > 1) | |
2135 | break; | |
2136 | if ((p->nivcsw - nivcsw) > 1) | |
2137 | break; | |
2138 | ||
2139 | cpu_relax(); | |
2140 | } | |
2141 | } | |
2142 | ||
1da177e4 LT |
2143 | /* |
2144 | * wait_task_inactive - wait for a thread to unschedule. | |
2145 | * | |
85ba2d86 RM |
2146 | * If @match_state is nonzero, it's the @p->state value just checked and |
2147 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2148 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2149 | * we return a positive number (its total switch count). If a second call | |
2150 | * a short while later returns the same number, the caller can be sure that | |
2151 | * @p has remained unscheduled the whole time. | |
2152 | * | |
1da177e4 LT |
2153 | * The caller must ensure that the task *will* unschedule sometime soon, |
2154 | * else this function might spin for a *long* time. This function can't | |
2155 | * be called with interrupts off, or it may introduce deadlock with | |
2156 | * smp_call_function() if an IPI is sent by the same process we are | |
2157 | * waiting to become inactive. | |
2158 | */ | |
85ba2d86 | 2159 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2160 | { |
2161 | unsigned long flags; | |
dd41f596 | 2162 | int running, on_rq; |
85ba2d86 | 2163 | unsigned long ncsw; |
70b97a7f | 2164 | struct rq *rq; |
1da177e4 | 2165 | |
3a5c359a AK |
2166 | for (;;) { |
2167 | /* | |
2168 | * We do the initial early heuristics without holding | |
2169 | * any task-queue locks at all. We'll only try to get | |
2170 | * the runqueue lock when things look like they will | |
2171 | * work out! | |
2172 | */ | |
2173 | rq = task_rq(p); | |
fa490cfd | 2174 | |
3a5c359a AK |
2175 | /* |
2176 | * If the task is actively running on another CPU | |
2177 | * still, just relax and busy-wait without holding | |
2178 | * any locks. | |
2179 | * | |
2180 | * NOTE! Since we don't hold any locks, it's not | |
2181 | * even sure that "rq" stays as the right runqueue! | |
2182 | * But we don't care, since "task_running()" will | |
2183 | * return false if the runqueue has changed and p | |
2184 | * is actually now running somewhere else! | |
2185 | */ | |
85ba2d86 RM |
2186 | while (task_running(rq, p)) { |
2187 | if (match_state && unlikely(p->state != match_state)) | |
2188 | return 0; | |
3a5c359a | 2189 | cpu_relax(); |
85ba2d86 | 2190 | } |
fa490cfd | 2191 | |
3a5c359a AK |
2192 | /* |
2193 | * Ok, time to look more closely! We need the rq | |
2194 | * lock now, to be *sure*. If we're wrong, we'll | |
2195 | * just go back and repeat. | |
2196 | */ | |
2197 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2198 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2199 | running = task_running(rq, p); |
2200 | on_rq = p->se.on_rq; | |
85ba2d86 | 2201 | ncsw = 0; |
f31e11d8 | 2202 | if (!match_state || p->state == match_state) |
93dcf55f | 2203 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2204 | task_rq_unlock(rq, &flags); |
fa490cfd | 2205 | |
85ba2d86 RM |
2206 | /* |
2207 | * If it changed from the expected state, bail out now. | |
2208 | */ | |
2209 | if (unlikely(!ncsw)) | |
2210 | break; | |
2211 | ||
3a5c359a AK |
2212 | /* |
2213 | * Was it really running after all now that we | |
2214 | * checked with the proper locks actually held? | |
2215 | * | |
2216 | * Oops. Go back and try again.. | |
2217 | */ | |
2218 | if (unlikely(running)) { | |
2219 | cpu_relax(); | |
2220 | continue; | |
2221 | } | |
fa490cfd | 2222 | |
3a5c359a AK |
2223 | /* |
2224 | * It's not enough that it's not actively running, | |
2225 | * it must be off the runqueue _entirely_, and not | |
2226 | * preempted! | |
2227 | * | |
80dd99b3 | 2228 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2229 | * running right now), it's preempted, and we should |
2230 | * yield - it could be a while. | |
2231 | */ | |
2232 | if (unlikely(on_rq)) { | |
2233 | schedule_timeout_uninterruptible(1); | |
2234 | continue; | |
2235 | } | |
fa490cfd | 2236 | |
3a5c359a AK |
2237 | /* |
2238 | * Ahh, all good. It wasn't running, and it wasn't | |
2239 | * runnable, which means that it will never become | |
2240 | * running in the future either. We're all done! | |
2241 | */ | |
2242 | break; | |
2243 | } | |
85ba2d86 RM |
2244 | |
2245 | return ncsw; | |
1da177e4 LT |
2246 | } |
2247 | ||
2248 | /*** | |
2249 | * kick_process - kick a running thread to enter/exit the kernel | |
2250 | * @p: the to-be-kicked thread | |
2251 | * | |
2252 | * Cause a process which is running on another CPU to enter | |
2253 | * kernel-mode, without any delay. (to get signals handled.) | |
2254 | * | |
2255 | * NOTE: this function doesnt have to take the runqueue lock, | |
2256 | * because all it wants to ensure is that the remote task enters | |
2257 | * the kernel. If the IPI races and the task has been migrated | |
2258 | * to another CPU then no harm is done and the purpose has been | |
2259 | * achieved as well. | |
2260 | */ | |
36c8b586 | 2261 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2262 | { |
2263 | int cpu; | |
2264 | ||
2265 | preempt_disable(); | |
2266 | cpu = task_cpu(p); | |
2267 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2268 | smp_send_reschedule(cpu); | |
2269 | preempt_enable(); | |
2270 | } | |
b43e3521 | 2271 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2272 | #endif /* CONFIG_SMP */ |
1da177e4 | 2273 | |
0793a61d TG |
2274 | /** |
2275 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2276 | * @p: the task to evaluate | |
2277 | * @func: the function to be called | |
2278 | * @info: the function call argument | |
2279 | * | |
2280 | * Calls the function @func when the task is currently running. This might | |
2281 | * be on the current CPU, which just calls the function directly | |
2282 | */ | |
2283 | void task_oncpu_function_call(struct task_struct *p, | |
2284 | void (*func) (void *info), void *info) | |
2285 | { | |
2286 | int cpu; | |
2287 | ||
2288 | preempt_disable(); | |
2289 | cpu = task_cpu(p); | |
2290 | if (task_curr(p)) | |
2291 | smp_call_function_single(cpu, func, info, 1); | |
2292 | preempt_enable(); | |
2293 | } | |
2294 | ||
1da177e4 LT |
2295 | /*** |
2296 | * try_to_wake_up - wake up a thread | |
2297 | * @p: the to-be-woken-up thread | |
2298 | * @state: the mask of task states that can be woken | |
2299 | * @sync: do a synchronous wakeup? | |
2300 | * | |
2301 | * Put it on the run-queue if it's not already there. The "current" | |
2302 | * thread is always on the run-queue (except when the actual | |
2303 | * re-schedule is in progress), and as such you're allowed to do | |
2304 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2305 | * runnable without the overhead of this. | |
2306 | * | |
2307 | * returns failure only if the task is already active. | |
2308 | */ | |
7d478721 PZ |
2309 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2310 | int wake_flags) | |
1da177e4 | 2311 | { |
cc367732 | 2312 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2313 | unsigned long flags; |
f5dc3753 | 2314 | struct rq *rq, *orig_rq; |
1da177e4 | 2315 | |
b85d0667 | 2316 | if (!sched_feat(SYNC_WAKEUPS)) |
7d478721 | 2317 | wake_flags &= ~WF_SYNC; |
2398f2c6 | 2318 | |
e9c84311 | 2319 | this_cpu = get_cpu(); |
2398f2c6 | 2320 | |
04e2f174 | 2321 | smp_wmb(); |
f5dc3753 | 2322 | rq = orig_rq = task_rq_lock(p, &flags); |
03e89e45 | 2323 | update_rq_clock(rq); |
e9c84311 | 2324 | if (!(p->state & state)) |
1da177e4 LT |
2325 | goto out; |
2326 | ||
dd41f596 | 2327 | if (p->se.on_rq) |
1da177e4 LT |
2328 | goto out_running; |
2329 | ||
2330 | cpu = task_cpu(p); | |
cc367732 | 2331 | orig_cpu = cpu; |
1da177e4 LT |
2332 | |
2333 | #ifdef CONFIG_SMP | |
2334 | if (unlikely(task_running(rq, p))) | |
2335 | goto out_activate; | |
2336 | ||
e9c84311 PZ |
2337 | /* |
2338 | * In order to handle concurrent wakeups and release the rq->lock | |
2339 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2340 | * |
2341 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2342 | */ |
eb24073b IM |
2343 | if (task_contributes_to_load(p)) |
2344 | rq->nr_uninterruptible--; | |
e9c84311 PZ |
2345 | p->state = TASK_WAKING; |
2346 | task_rq_unlock(rq, &flags); | |
2347 | ||
7d478721 | 2348 | cpu = p->sched_class->select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
e9c84311 | 2349 | if (cpu != orig_cpu) |
5d2f5a61 | 2350 | set_task_cpu(p, cpu); |
1da177e4 | 2351 | |
e9c84311 | 2352 | rq = task_rq_lock(p, &flags); |
f5dc3753 MG |
2353 | |
2354 | if (rq != orig_rq) | |
2355 | update_rq_clock(rq); | |
2356 | ||
e9c84311 PZ |
2357 | WARN_ON(p->state != TASK_WAKING); |
2358 | cpu = task_cpu(p); | |
1da177e4 | 2359 | |
e7693a36 GH |
2360 | #ifdef CONFIG_SCHEDSTATS |
2361 | schedstat_inc(rq, ttwu_count); | |
2362 | if (cpu == this_cpu) | |
2363 | schedstat_inc(rq, ttwu_local); | |
2364 | else { | |
2365 | struct sched_domain *sd; | |
2366 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2367 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2368 | schedstat_inc(sd, ttwu_wake_remote); |
2369 | break; | |
2370 | } | |
2371 | } | |
2372 | } | |
6d6bc0ad | 2373 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2374 | |
1da177e4 LT |
2375 | out_activate: |
2376 | #endif /* CONFIG_SMP */ | |
cc367732 | 2377 | schedstat_inc(p, se.nr_wakeups); |
7d478721 | 2378 | if (wake_flags & WF_SYNC) |
cc367732 IM |
2379 | schedstat_inc(p, se.nr_wakeups_sync); |
2380 | if (orig_cpu != cpu) | |
2381 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2382 | if (cpu == this_cpu) | |
2383 | schedstat_inc(p, se.nr_wakeups_local); | |
2384 | else | |
2385 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2386 | activate_task(rq, p, 1); |
1da177e4 LT |
2387 | success = 1; |
2388 | ||
831451ac PZ |
2389 | /* |
2390 | * Only attribute actual wakeups done by this task. | |
2391 | */ | |
2392 | if (!in_interrupt()) { | |
2393 | struct sched_entity *se = ¤t->se; | |
2394 | u64 sample = se->sum_exec_runtime; | |
2395 | ||
2396 | if (se->last_wakeup) | |
2397 | sample -= se->last_wakeup; | |
2398 | else | |
2399 | sample -= se->start_runtime; | |
2400 | update_avg(&se->avg_wakeup, sample); | |
2401 | ||
2402 | se->last_wakeup = se->sum_exec_runtime; | |
2403 | } | |
2404 | ||
1da177e4 | 2405 | out_running: |
468a15bb | 2406 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2407 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2408 | |
1da177e4 | 2409 | p->state = TASK_RUNNING; |
9a897c5a SR |
2410 | #ifdef CONFIG_SMP |
2411 | if (p->sched_class->task_wake_up) | |
2412 | p->sched_class->task_wake_up(rq, p); | |
eae0c9df MG |
2413 | |
2414 | if (unlikely(rq->idle_stamp)) { | |
2415 | u64 delta = rq->clock - rq->idle_stamp; | |
2416 | u64 max = 2*sysctl_sched_migration_cost; | |
2417 | ||
2418 | if (delta > max) | |
2419 | rq->avg_idle = max; | |
2420 | else | |
2421 | update_avg(&rq->avg_idle, delta); | |
2422 | rq->idle_stamp = 0; | |
2423 | } | |
9a897c5a | 2424 | #endif |
1da177e4 LT |
2425 | out: |
2426 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2427 | put_cpu(); |
1da177e4 LT |
2428 | |
2429 | return success; | |
2430 | } | |
2431 | ||
50fa610a DH |
2432 | /** |
2433 | * wake_up_process - Wake up a specific process | |
2434 | * @p: The process to be woken up. | |
2435 | * | |
2436 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2437 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2438 | * running. | |
2439 | * | |
2440 | * It may be assumed that this function implies a write memory barrier before | |
2441 | * changing the task state if and only if any tasks are woken up. | |
2442 | */ | |
7ad5b3a5 | 2443 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2444 | { |
d9514f6c | 2445 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2446 | } |
1da177e4 LT |
2447 | EXPORT_SYMBOL(wake_up_process); |
2448 | ||
7ad5b3a5 | 2449 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2450 | { |
2451 | return try_to_wake_up(p, state, 0); | |
2452 | } | |
2453 | ||
1da177e4 LT |
2454 | /* |
2455 | * Perform scheduler related setup for a newly forked process p. | |
2456 | * p is forked by current. | |
dd41f596 IM |
2457 | * |
2458 | * __sched_fork() is basic setup used by init_idle() too: | |
2459 | */ | |
2460 | static void __sched_fork(struct task_struct *p) | |
2461 | { | |
dd41f596 IM |
2462 | p->se.exec_start = 0; |
2463 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2464 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2465 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2466 | p->se.last_wakeup = 0; |
2467 | p->se.avg_overlap = 0; | |
831451ac PZ |
2468 | p->se.start_runtime = 0; |
2469 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
ad4b78bb | 2470 | p->se.avg_running = 0; |
6cfb0d5d IM |
2471 | |
2472 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2473 | p->se.wait_start = 0; |
2474 | p->se.wait_max = 0; | |
2475 | p->se.wait_count = 0; | |
2476 | p->se.wait_sum = 0; | |
2477 | ||
2478 | p->se.sleep_start = 0; | |
2479 | p->se.sleep_max = 0; | |
2480 | p->se.sum_sleep_runtime = 0; | |
2481 | ||
2482 | p->se.block_start = 0; | |
2483 | p->se.block_max = 0; | |
2484 | p->se.exec_max = 0; | |
2485 | p->se.slice_max = 0; | |
2486 | ||
2487 | p->se.nr_migrations_cold = 0; | |
2488 | p->se.nr_failed_migrations_affine = 0; | |
2489 | p->se.nr_failed_migrations_running = 0; | |
2490 | p->se.nr_failed_migrations_hot = 0; | |
2491 | p->se.nr_forced_migrations = 0; | |
2492 | p->se.nr_forced2_migrations = 0; | |
2493 | ||
2494 | p->se.nr_wakeups = 0; | |
2495 | p->se.nr_wakeups_sync = 0; | |
2496 | p->se.nr_wakeups_migrate = 0; | |
2497 | p->se.nr_wakeups_local = 0; | |
2498 | p->se.nr_wakeups_remote = 0; | |
2499 | p->se.nr_wakeups_affine = 0; | |
2500 | p->se.nr_wakeups_affine_attempts = 0; | |
2501 | p->se.nr_wakeups_passive = 0; | |
2502 | p->se.nr_wakeups_idle = 0; | |
2503 | ||
6cfb0d5d | 2504 | #endif |
476d139c | 2505 | |
fa717060 | 2506 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2507 | p->se.on_rq = 0; |
4a55bd5e | 2508 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2509 | |
e107be36 AK |
2510 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2511 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2512 | #endif | |
2513 | ||
1da177e4 LT |
2514 | /* |
2515 | * We mark the process as running here, but have not actually | |
2516 | * inserted it onto the runqueue yet. This guarantees that | |
2517 | * nobody will actually run it, and a signal or other external | |
2518 | * event cannot wake it up and insert it on the runqueue either. | |
2519 | */ | |
2520 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2521 | } |
2522 | ||
2523 | /* | |
2524 | * fork()/clone()-time setup: | |
2525 | */ | |
2526 | void sched_fork(struct task_struct *p, int clone_flags) | |
2527 | { | |
2528 | int cpu = get_cpu(); | |
2529 | ||
2530 | __sched_fork(p); | |
2531 | ||
b9dc29e7 MG |
2532 | /* |
2533 | * Revert to default priority/policy on fork if requested. | |
2534 | */ | |
2535 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2536 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2537 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2538 | p->normal_prio = p->static_prio; |
2539 | } | |
b9dc29e7 | 2540 | |
6c697bdf MG |
2541 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2542 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2543 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2544 | set_load_weight(p); |
2545 | } | |
2546 | ||
b9dc29e7 MG |
2547 | /* |
2548 | * We don't need the reset flag anymore after the fork. It has | |
2549 | * fulfilled its duty: | |
2550 | */ | |
2551 | p->sched_reset_on_fork = 0; | |
2552 | } | |
ca94c442 | 2553 | |
f83f9ac2 PW |
2554 | /* |
2555 | * Make sure we do not leak PI boosting priority to the child. | |
2556 | */ | |
2557 | p->prio = current->normal_prio; | |
2558 | ||
2ddbf952 HS |
2559 | if (!rt_prio(p->prio)) |
2560 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2561 | |
5f3edc1b PZ |
2562 | #ifdef CONFIG_SMP |
2563 | cpu = p->sched_class->select_task_rq(p, SD_BALANCE_FORK, 0); | |
2564 | #endif | |
2565 | set_task_cpu(p, cpu); | |
2566 | ||
52f17b6c | 2567 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2568 | if (likely(sched_info_on())) |
52f17b6c | 2569 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2570 | #endif |
d6077cb8 | 2571 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2572 | p->oncpu = 0; |
2573 | #endif | |
1da177e4 | 2574 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2575 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2576 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2577 | #endif |
917b627d GH |
2578 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2579 | ||
476d139c | 2580 | put_cpu(); |
1da177e4 LT |
2581 | } |
2582 | ||
2583 | /* | |
2584 | * wake_up_new_task - wake up a newly created task for the first time. | |
2585 | * | |
2586 | * This function will do some initial scheduler statistics housekeeping | |
2587 | * that must be done for every newly created context, then puts the task | |
2588 | * on the runqueue and wakes it. | |
2589 | */ | |
7ad5b3a5 | 2590 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2591 | { |
2592 | unsigned long flags; | |
dd41f596 | 2593 | struct rq *rq; |
1da177e4 LT |
2594 | |
2595 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2596 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2597 | update_rq_clock(rq); |
1da177e4 | 2598 | |
b9dca1e0 | 2599 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2600 | activate_task(rq, p, 0); |
1da177e4 | 2601 | } else { |
1da177e4 | 2602 | /* |
dd41f596 IM |
2603 | * Let the scheduling class do new task startup |
2604 | * management (if any): | |
1da177e4 | 2605 | */ |
ee0827d8 | 2606 | p->sched_class->task_new(rq, p); |
c09595f6 | 2607 | inc_nr_running(rq); |
1da177e4 | 2608 | } |
c71dd42d | 2609 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2610 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a SR |
2611 | #ifdef CONFIG_SMP |
2612 | if (p->sched_class->task_wake_up) | |
2613 | p->sched_class->task_wake_up(rq, p); | |
2614 | #endif | |
dd41f596 | 2615 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2616 | } |
2617 | ||
e107be36 AK |
2618 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2619 | ||
2620 | /** | |
80dd99b3 | 2621 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2622 | * @notifier: notifier struct to register |
e107be36 AK |
2623 | */ |
2624 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2625 | { | |
2626 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2627 | } | |
2628 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2629 | ||
2630 | /** | |
2631 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2632 | * @notifier: notifier struct to unregister |
e107be36 AK |
2633 | * |
2634 | * This is safe to call from within a preemption notifier. | |
2635 | */ | |
2636 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2637 | { | |
2638 | hlist_del(¬ifier->link); | |
2639 | } | |
2640 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2641 | ||
2642 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2643 | { | |
2644 | struct preempt_notifier *notifier; | |
2645 | struct hlist_node *node; | |
2646 | ||
2647 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2648 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2649 | } | |
2650 | ||
2651 | static void | |
2652 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2653 | struct task_struct *next) | |
2654 | { | |
2655 | struct preempt_notifier *notifier; | |
2656 | struct hlist_node *node; | |
2657 | ||
2658 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2659 | notifier->ops->sched_out(notifier, next); | |
2660 | } | |
2661 | ||
6d6bc0ad | 2662 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2663 | |
2664 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2665 | { | |
2666 | } | |
2667 | ||
2668 | static void | |
2669 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2670 | struct task_struct *next) | |
2671 | { | |
2672 | } | |
2673 | ||
6d6bc0ad | 2674 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2675 | |
4866cde0 NP |
2676 | /** |
2677 | * prepare_task_switch - prepare to switch tasks | |
2678 | * @rq: the runqueue preparing to switch | |
421cee29 | 2679 | * @prev: the current task that is being switched out |
4866cde0 NP |
2680 | * @next: the task we are going to switch to. |
2681 | * | |
2682 | * This is called with the rq lock held and interrupts off. It must | |
2683 | * be paired with a subsequent finish_task_switch after the context | |
2684 | * switch. | |
2685 | * | |
2686 | * prepare_task_switch sets up locking and calls architecture specific | |
2687 | * hooks. | |
2688 | */ | |
e107be36 AK |
2689 | static inline void |
2690 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2691 | struct task_struct *next) | |
4866cde0 | 2692 | { |
e107be36 | 2693 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2694 | prepare_lock_switch(rq, next); |
2695 | prepare_arch_switch(next); | |
2696 | } | |
2697 | ||
1da177e4 LT |
2698 | /** |
2699 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2700 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2701 | * @prev: the thread we just switched away from. |
2702 | * | |
4866cde0 NP |
2703 | * finish_task_switch must be called after the context switch, paired |
2704 | * with a prepare_task_switch call before the context switch. | |
2705 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2706 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2707 | * |
2708 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2709 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2710 | * with the lock held can cause deadlocks; see schedule() for |
2711 | * details.) | |
2712 | */ | |
a9957449 | 2713 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2714 | __releases(rq->lock) |
2715 | { | |
1da177e4 | 2716 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2717 | long prev_state; |
1da177e4 LT |
2718 | |
2719 | rq->prev_mm = NULL; | |
2720 | ||
2721 | /* | |
2722 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2723 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2724 | * schedule one last time. The schedule call will never return, and |
2725 | * the scheduled task must drop that reference. | |
c394cc9f | 2726 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2727 | * still held, otherwise prev could be scheduled on another cpu, die |
2728 | * there before we look at prev->state, and then the reference would | |
2729 | * be dropped twice. | |
2730 | * Manfred Spraul <manfred@colorfullife.com> | |
2731 | */ | |
55a101f8 | 2732 | prev_state = prev->state; |
4866cde0 | 2733 | finish_arch_switch(prev); |
cdd6c482 | 2734 | perf_event_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2735 | finish_lock_switch(rq, prev); |
e8fa1362 | 2736 | |
e107be36 | 2737 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2738 | if (mm) |
2739 | mmdrop(mm); | |
c394cc9f | 2740 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2741 | /* |
2742 | * Remove function-return probe instances associated with this | |
2743 | * task and put them back on the free list. | |
9761eea8 | 2744 | */ |
c6fd91f0 | 2745 | kprobe_flush_task(prev); |
1da177e4 | 2746 | put_task_struct(prev); |
c6fd91f0 | 2747 | } |
1da177e4 LT |
2748 | } |
2749 | ||
3f029d3c GH |
2750 | #ifdef CONFIG_SMP |
2751 | ||
2752 | /* assumes rq->lock is held */ | |
2753 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2754 | { | |
2755 | if (prev->sched_class->pre_schedule) | |
2756 | prev->sched_class->pre_schedule(rq, prev); | |
2757 | } | |
2758 | ||
2759 | /* rq->lock is NOT held, but preemption is disabled */ | |
2760 | static inline void post_schedule(struct rq *rq) | |
2761 | { | |
2762 | if (rq->post_schedule) { | |
2763 | unsigned long flags; | |
2764 | ||
2765 | spin_lock_irqsave(&rq->lock, flags); | |
2766 | if (rq->curr->sched_class->post_schedule) | |
2767 | rq->curr->sched_class->post_schedule(rq); | |
2768 | spin_unlock_irqrestore(&rq->lock, flags); | |
2769 | ||
2770 | rq->post_schedule = 0; | |
2771 | } | |
2772 | } | |
2773 | ||
2774 | #else | |
da19ab51 | 2775 | |
3f029d3c GH |
2776 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2777 | { | |
2778 | } | |
2779 | ||
2780 | static inline void post_schedule(struct rq *rq) | |
2781 | { | |
1da177e4 LT |
2782 | } |
2783 | ||
3f029d3c GH |
2784 | #endif |
2785 | ||
1da177e4 LT |
2786 | /** |
2787 | * schedule_tail - first thing a freshly forked thread must call. | |
2788 | * @prev: the thread we just switched away from. | |
2789 | */ | |
36c8b586 | 2790 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2791 | __releases(rq->lock) |
2792 | { | |
70b97a7f IM |
2793 | struct rq *rq = this_rq(); |
2794 | ||
4866cde0 | 2795 | finish_task_switch(rq, prev); |
da19ab51 | 2796 | |
3f029d3c GH |
2797 | /* |
2798 | * FIXME: do we need to worry about rq being invalidated by the | |
2799 | * task_switch? | |
2800 | */ | |
2801 | post_schedule(rq); | |
70b97a7f | 2802 | |
4866cde0 NP |
2803 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2804 | /* In this case, finish_task_switch does not reenable preemption */ | |
2805 | preempt_enable(); | |
2806 | #endif | |
1da177e4 | 2807 | if (current->set_child_tid) |
b488893a | 2808 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2809 | } |
2810 | ||
2811 | /* | |
2812 | * context_switch - switch to the new MM and the new | |
2813 | * thread's register state. | |
2814 | */ | |
dd41f596 | 2815 | static inline void |
70b97a7f | 2816 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2817 | struct task_struct *next) |
1da177e4 | 2818 | { |
dd41f596 | 2819 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2820 | |
e107be36 | 2821 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2822 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2823 | mm = next->mm; |
2824 | oldmm = prev->active_mm; | |
9226d125 ZA |
2825 | /* |
2826 | * For paravirt, this is coupled with an exit in switch_to to | |
2827 | * combine the page table reload and the switch backend into | |
2828 | * one hypercall. | |
2829 | */ | |
224101ed | 2830 | arch_start_context_switch(prev); |
9226d125 | 2831 | |
710390d9 | 2832 | if (likely(!mm)) { |
1da177e4 LT |
2833 | next->active_mm = oldmm; |
2834 | atomic_inc(&oldmm->mm_count); | |
2835 | enter_lazy_tlb(oldmm, next); | |
2836 | } else | |
2837 | switch_mm(oldmm, mm, next); | |
2838 | ||
710390d9 | 2839 | if (likely(!prev->mm)) { |
1da177e4 | 2840 | prev->active_mm = NULL; |
1da177e4 LT |
2841 | rq->prev_mm = oldmm; |
2842 | } | |
3a5f5e48 IM |
2843 | /* |
2844 | * Since the runqueue lock will be released by the next | |
2845 | * task (which is an invalid locking op but in the case | |
2846 | * of the scheduler it's an obvious special-case), so we | |
2847 | * do an early lockdep release here: | |
2848 | */ | |
2849 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2850 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2851 | #endif |
1da177e4 LT |
2852 | |
2853 | /* Here we just switch the register state and the stack. */ | |
2854 | switch_to(prev, next, prev); | |
2855 | ||
dd41f596 IM |
2856 | barrier(); |
2857 | /* | |
2858 | * this_rq must be evaluated again because prev may have moved | |
2859 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2860 | * frame will be invalid. | |
2861 | */ | |
2862 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /* | |
2866 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2867 | * | |
2868 | * externally visible scheduler statistics: current number of runnable | |
2869 | * threads, current number of uninterruptible-sleeping threads, total | |
2870 | * number of context switches performed since bootup. | |
2871 | */ | |
2872 | unsigned long nr_running(void) | |
2873 | { | |
2874 | unsigned long i, sum = 0; | |
2875 | ||
2876 | for_each_online_cpu(i) | |
2877 | sum += cpu_rq(i)->nr_running; | |
2878 | ||
2879 | return sum; | |
2880 | } | |
2881 | ||
2882 | unsigned long nr_uninterruptible(void) | |
2883 | { | |
2884 | unsigned long i, sum = 0; | |
2885 | ||
0a945022 | 2886 | for_each_possible_cpu(i) |
1da177e4 LT |
2887 | sum += cpu_rq(i)->nr_uninterruptible; |
2888 | ||
2889 | /* | |
2890 | * Since we read the counters lockless, it might be slightly | |
2891 | * inaccurate. Do not allow it to go below zero though: | |
2892 | */ | |
2893 | if (unlikely((long)sum < 0)) | |
2894 | sum = 0; | |
2895 | ||
2896 | return sum; | |
2897 | } | |
2898 | ||
2899 | unsigned long long nr_context_switches(void) | |
2900 | { | |
cc94abfc SR |
2901 | int i; |
2902 | unsigned long long sum = 0; | |
1da177e4 | 2903 | |
0a945022 | 2904 | for_each_possible_cpu(i) |
1da177e4 LT |
2905 | sum += cpu_rq(i)->nr_switches; |
2906 | ||
2907 | return sum; | |
2908 | } | |
2909 | ||
2910 | unsigned long nr_iowait(void) | |
2911 | { | |
2912 | unsigned long i, sum = 0; | |
2913 | ||
0a945022 | 2914 | for_each_possible_cpu(i) |
1da177e4 LT |
2915 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2916 | ||
2917 | return sum; | |
2918 | } | |
2919 | ||
69d25870 AV |
2920 | unsigned long nr_iowait_cpu(void) |
2921 | { | |
2922 | struct rq *this = this_rq(); | |
2923 | return atomic_read(&this->nr_iowait); | |
2924 | } | |
2925 | ||
2926 | unsigned long this_cpu_load(void) | |
2927 | { | |
2928 | struct rq *this = this_rq(); | |
2929 | return this->cpu_load[0]; | |
2930 | } | |
2931 | ||
2932 | ||
dce48a84 TG |
2933 | /* Variables and functions for calc_load */ |
2934 | static atomic_long_t calc_load_tasks; | |
2935 | static unsigned long calc_load_update; | |
2936 | unsigned long avenrun[3]; | |
2937 | EXPORT_SYMBOL(avenrun); | |
2938 | ||
2d02494f TG |
2939 | /** |
2940 | * get_avenrun - get the load average array | |
2941 | * @loads: pointer to dest load array | |
2942 | * @offset: offset to add | |
2943 | * @shift: shift count to shift the result left | |
2944 | * | |
2945 | * These values are estimates at best, so no need for locking. | |
2946 | */ | |
2947 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2948 | { | |
2949 | loads[0] = (avenrun[0] + offset) << shift; | |
2950 | loads[1] = (avenrun[1] + offset) << shift; | |
2951 | loads[2] = (avenrun[2] + offset) << shift; | |
2952 | } | |
2953 | ||
dce48a84 TG |
2954 | static unsigned long |
2955 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2956 | { |
dce48a84 TG |
2957 | load *= exp; |
2958 | load += active * (FIXED_1 - exp); | |
2959 | return load >> FSHIFT; | |
2960 | } | |
db1b1fef | 2961 | |
dce48a84 TG |
2962 | /* |
2963 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2964 | * CPUs have updated calc_load_tasks. | |
2965 | */ | |
2966 | void calc_global_load(void) | |
2967 | { | |
2968 | unsigned long upd = calc_load_update + 10; | |
2969 | long active; | |
2970 | ||
2971 | if (time_before(jiffies, upd)) | |
2972 | return; | |
db1b1fef | 2973 | |
dce48a84 TG |
2974 | active = atomic_long_read(&calc_load_tasks); |
2975 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2976 | |
dce48a84 TG |
2977 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2978 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2979 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2980 | ||
2981 | calc_load_update += LOAD_FREQ; | |
2982 | } | |
2983 | ||
2984 | /* | |
2985 | * Either called from update_cpu_load() or from a cpu going idle | |
2986 | */ | |
2987 | static void calc_load_account_active(struct rq *this_rq) | |
2988 | { | |
2989 | long nr_active, delta; | |
2990 | ||
2991 | nr_active = this_rq->nr_running; | |
2992 | nr_active += (long) this_rq->nr_uninterruptible; | |
2993 | ||
2994 | if (nr_active != this_rq->calc_load_active) { | |
2995 | delta = nr_active - this_rq->calc_load_active; | |
2996 | this_rq->calc_load_active = nr_active; | |
2997 | atomic_long_add(delta, &calc_load_tasks); | |
2998 | } | |
db1b1fef JS |
2999 | } |
3000 | ||
48f24c4d | 3001 | /* |
dd41f596 IM |
3002 | * Update rq->cpu_load[] statistics. This function is usually called every |
3003 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3004 | */ |
dd41f596 | 3005 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3006 | { |
495eca49 | 3007 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3008 | int i, scale; |
3009 | ||
3010 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3011 | |
3012 | /* Update our load: */ | |
3013 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3014 | unsigned long old_load, new_load; | |
3015 | ||
3016 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3017 | ||
3018 | old_load = this_rq->cpu_load[i]; | |
3019 | new_load = this_load; | |
a25707f3 IM |
3020 | /* |
3021 | * Round up the averaging division if load is increasing. This | |
3022 | * prevents us from getting stuck on 9 if the load is 10, for | |
3023 | * example. | |
3024 | */ | |
3025 | if (new_load > old_load) | |
3026 | new_load += scale-1; | |
dd41f596 IM |
3027 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3028 | } | |
dce48a84 TG |
3029 | |
3030 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3031 | this_rq->calc_load_update += LOAD_FREQ; | |
3032 | calc_load_account_active(this_rq); | |
3033 | } | |
48f24c4d IM |
3034 | } |
3035 | ||
dd41f596 IM |
3036 | #ifdef CONFIG_SMP |
3037 | ||
1da177e4 LT |
3038 | /* |
3039 | * double_rq_lock - safely lock two runqueues | |
3040 | * | |
3041 | * Note this does not disable interrupts like task_rq_lock, | |
3042 | * you need to do so manually before calling. | |
3043 | */ | |
70b97a7f | 3044 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3045 | __acquires(rq1->lock) |
3046 | __acquires(rq2->lock) | |
3047 | { | |
054b9108 | 3048 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3049 | if (rq1 == rq2) { |
3050 | spin_lock(&rq1->lock); | |
3051 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3052 | } else { | |
c96d145e | 3053 | if (rq1 < rq2) { |
1da177e4 | 3054 | spin_lock(&rq1->lock); |
5e710e37 | 3055 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3056 | } else { |
3057 | spin_lock(&rq2->lock); | |
5e710e37 | 3058 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3059 | } |
3060 | } | |
6e82a3be IM |
3061 | update_rq_clock(rq1); |
3062 | update_rq_clock(rq2); | |
1da177e4 LT |
3063 | } |
3064 | ||
3065 | /* | |
3066 | * double_rq_unlock - safely unlock two runqueues | |
3067 | * | |
3068 | * Note this does not restore interrupts like task_rq_unlock, | |
3069 | * you need to do so manually after calling. | |
3070 | */ | |
70b97a7f | 3071 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3072 | __releases(rq1->lock) |
3073 | __releases(rq2->lock) | |
3074 | { | |
3075 | spin_unlock(&rq1->lock); | |
3076 | if (rq1 != rq2) | |
3077 | spin_unlock(&rq2->lock); | |
3078 | else | |
3079 | __release(rq2->lock); | |
3080 | } | |
3081 | ||
1da177e4 LT |
3082 | /* |
3083 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3084 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3085 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3086 | * the cpu_allowed mask is restored. |
3087 | */ | |
36c8b586 | 3088 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3089 | { |
70b97a7f | 3090 | struct migration_req req; |
1da177e4 | 3091 | unsigned long flags; |
70b97a7f | 3092 | struct rq *rq; |
1da177e4 LT |
3093 | |
3094 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3095 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3096 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3097 | goto out; |
3098 | ||
3099 | /* force the process onto the specified CPU */ | |
3100 | if (migrate_task(p, dest_cpu, &req)) { | |
3101 | /* Need to wait for migration thread (might exit: take ref). */ | |
3102 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3103 | |
1da177e4 LT |
3104 | get_task_struct(mt); |
3105 | task_rq_unlock(rq, &flags); | |
3106 | wake_up_process(mt); | |
3107 | put_task_struct(mt); | |
3108 | wait_for_completion(&req.done); | |
36c8b586 | 3109 | |
1da177e4 LT |
3110 | return; |
3111 | } | |
3112 | out: | |
3113 | task_rq_unlock(rq, &flags); | |
3114 | } | |
3115 | ||
3116 | /* | |
476d139c NP |
3117 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3118 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3119 | */ |
3120 | void sched_exec(void) | |
3121 | { | |
1da177e4 | 3122 | int new_cpu, this_cpu = get_cpu(); |
5f3edc1b | 3123 | new_cpu = current->sched_class->select_task_rq(current, SD_BALANCE_EXEC, 0); |
1da177e4 | 3124 | put_cpu(); |
476d139c NP |
3125 | if (new_cpu != this_cpu) |
3126 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3127 | } |
3128 | ||
3129 | /* | |
3130 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3131 | * Both runqueues must be locked. | |
3132 | */ | |
dd41f596 IM |
3133 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3134 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3135 | { |
2e1cb74a | 3136 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3137 | set_task_cpu(p, this_cpu); |
dd41f596 | 3138 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3139 | /* |
3140 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3141 | * to be always true for them. | |
3142 | */ | |
15afe09b | 3143 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3144 | } |
3145 | ||
3146 | /* | |
3147 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3148 | */ | |
858119e1 | 3149 | static |
70b97a7f | 3150 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3151 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3152 | int *all_pinned) |
1da177e4 | 3153 | { |
708dc512 | 3154 | int tsk_cache_hot = 0; |
1da177e4 LT |
3155 | /* |
3156 | * We do not migrate tasks that are: | |
3157 | * 1) running (obviously), or | |
3158 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3159 | * 3) are cache-hot on their current CPU. | |
3160 | */ | |
96f874e2 | 3161 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3162 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3163 | return 0; |
cc367732 | 3164 | } |
81026794 NP |
3165 | *all_pinned = 0; |
3166 | ||
cc367732 IM |
3167 | if (task_running(rq, p)) { |
3168 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3169 | return 0; |
cc367732 | 3170 | } |
1da177e4 | 3171 | |
da84d961 IM |
3172 | /* |
3173 | * Aggressive migration if: | |
3174 | * 1) task is cache cold, or | |
3175 | * 2) too many balance attempts have failed. | |
3176 | */ | |
3177 | ||
708dc512 LH |
3178 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3179 | if (!tsk_cache_hot || | |
3180 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3181 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3182 | if (tsk_cache_hot) { |
da84d961 | 3183 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3184 | schedstat_inc(p, se.nr_forced_migrations); |
3185 | } | |
da84d961 IM |
3186 | #endif |
3187 | return 1; | |
3188 | } | |
3189 | ||
708dc512 | 3190 | if (tsk_cache_hot) { |
cc367732 | 3191 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3192 | return 0; |
cc367732 | 3193 | } |
1da177e4 LT |
3194 | return 1; |
3195 | } | |
3196 | ||
e1d1484f PW |
3197 | static unsigned long |
3198 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3199 | unsigned long max_load_move, struct sched_domain *sd, | |
3200 | enum cpu_idle_type idle, int *all_pinned, | |
3201 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3202 | { |
051c6764 | 3203 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3204 | struct task_struct *p; |
3205 | long rem_load_move = max_load_move; | |
1da177e4 | 3206 | |
e1d1484f | 3207 | if (max_load_move == 0) |
1da177e4 LT |
3208 | goto out; |
3209 | ||
81026794 NP |
3210 | pinned = 1; |
3211 | ||
1da177e4 | 3212 | /* |
dd41f596 | 3213 | * Start the load-balancing iterator: |
1da177e4 | 3214 | */ |
dd41f596 IM |
3215 | p = iterator->start(iterator->arg); |
3216 | next: | |
b82d9fdd | 3217 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3218 | goto out; |
051c6764 PZ |
3219 | |
3220 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3221 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3222 | p = iterator->next(iterator->arg); |
3223 | goto next; | |
1da177e4 LT |
3224 | } |
3225 | ||
dd41f596 | 3226 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3227 | pulled++; |
dd41f596 | 3228 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3229 | |
7e96fa58 GH |
3230 | #ifdef CONFIG_PREEMPT |
3231 | /* | |
3232 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3233 | * will stop after the first task is pulled to minimize the critical | |
3234 | * section. | |
3235 | */ | |
3236 | if (idle == CPU_NEWLY_IDLE) | |
3237 | goto out; | |
3238 | #endif | |
3239 | ||
2dd73a4f | 3240 | /* |
b82d9fdd | 3241 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3242 | */ |
e1d1484f | 3243 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3244 | if (p->prio < *this_best_prio) |
3245 | *this_best_prio = p->prio; | |
dd41f596 IM |
3246 | p = iterator->next(iterator->arg); |
3247 | goto next; | |
1da177e4 LT |
3248 | } |
3249 | out: | |
3250 | /* | |
e1d1484f | 3251 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3252 | * so we can safely collect pull_task() stats here rather than |
3253 | * inside pull_task(). | |
3254 | */ | |
3255 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3256 | |
3257 | if (all_pinned) | |
3258 | *all_pinned = pinned; | |
e1d1484f PW |
3259 | |
3260 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3261 | } |
3262 | ||
dd41f596 | 3263 | /* |
43010659 PW |
3264 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3265 | * this_rq, as part of a balancing operation within domain "sd". | |
3266 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3267 | * |
3268 | * Called with both runqueues locked. | |
3269 | */ | |
3270 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3271 | unsigned long max_load_move, |
dd41f596 IM |
3272 | struct sched_domain *sd, enum cpu_idle_type idle, |
3273 | int *all_pinned) | |
3274 | { | |
5522d5d5 | 3275 | const struct sched_class *class = sched_class_highest; |
43010659 | 3276 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3277 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3278 | |
3279 | do { | |
43010659 PW |
3280 | total_load_moved += |
3281 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3282 | max_load_move - total_load_moved, |
a4ac01c3 | 3283 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3284 | class = class->next; |
c4acb2c0 | 3285 | |
7e96fa58 GH |
3286 | #ifdef CONFIG_PREEMPT |
3287 | /* | |
3288 | * NEWIDLE balancing is a source of latency, so preemptible | |
3289 | * kernels will stop after the first task is pulled to minimize | |
3290 | * the critical section. | |
3291 | */ | |
c4acb2c0 GH |
3292 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3293 | break; | |
7e96fa58 | 3294 | #endif |
43010659 | 3295 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3296 | |
43010659 PW |
3297 | return total_load_moved > 0; |
3298 | } | |
3299 | ||
e1d1484f PW |
3300 | static int |
3301 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3302 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3303 | struct rq_iterator *iterator) | |
3304 | { | |
3305 | struct task_struct *p = iterator->start(iterator->arg); | |
3306 | int pinned = 0; | |
3307 | ||
3308 | while (p) { | |
3309 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3310 | pull_task(busiest, p, this_rq, this_cpu); | |
3311 | /* | |
3312 | * Right now, this is only the second place pull_task() | |
3313 | * is called, so we can safely collect pull_task() | |
3314 | * stats here rather than inside pull_task(). | |
3315 | */ | |
3316 | schedstat_inc(sd, lb_gained[idle]); | |
3317 | ||
3318 | return 1; | |
3319 | } | |
3320 | p = iterator->next(iterator->arg); | |
3321 | } | |
3322 | ||
3323 | return 0; | |
3324 | } | |
3325 | ||
43010659 PW |
3326 | /* |
3327 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3328 | * part of active balancing operations within "domain". | |
3329 | * Returns 1 if successful and 0 otherwise. | |
3330 | * | |
3331 | * Called with both runqueues locked. | |
3332 | */ | |
3333 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3334 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3335 | { | |
5522d5d5 | 3336 | const struct sched_class *class; |
43010659 | 3337 | |
cde7e5ca | 3338 | for_each_class(class) { |
e1d1484f | 3339 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3340 | return 1; |
cde7e5ca | 3341 | } |
43010659 PW |
3342 | |
3343 | return 0; | |
dd41f596 | 3344 | } |
67bb6c03 | 3345 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3346 | /* |
222d656d GS |
3347 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3348 | * during load balancing. | |
1da177e4 | 3349 | */ |
222d656d GS |
3350 | struct sd_lb_stats { |
3351 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3352 | struct sched_group *this; /* Local group in this sd */ | |
3353 | unsigned long total_load; /* Total load of all groups in sd */ | |
3354 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3355 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3356 | ||
3357 | /** Statistics of this group */ | |
3358 | unsigned long this_load; | |
3359 | unsigned long this_load_per_task; | |
3360 | unsigned long this_nr_running; | |
3361 | ||
3362 | /* Statistics of the busiest group */ | |
3363 | unsigned long max_load; | |
3364 | unsigned long busiest_load_per_task; | |
3365 | unsigned long busiest_nr_running; | |
3366 | ||
3367 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3368 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3369 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3370 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3371 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3372 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3373 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3374 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3375 | #endif |
222d656d | 3376 | }; |
1da177e4 | 3377 | |
d5ac537e | 3378 | /* |
381be78f GS |
3379 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3380 | */ | |
3381 | struct sg_lb_stats { | |
3382 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3383 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3384 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3385 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3386 | unsigned long group_capacity; | |
3387 | int group_imb; /* Is there an imbalance in the group ? */ | |
3388 | }; | |
408ed066 | 3389 | |
67bb6c03 GS |
3390 | /** |
3391 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3392 | * @group: The group whose first cpu is to be returned. | |
3393 | */ | |
3394 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3395 | { | |
3396 | return cpumask_first(sched_group_cpus(group)); | |
3397 | } | |
3398 | ||
3399 | /** | |
3400 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3401 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3402 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3403 | */ | |
3404 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3405 | enum cpu_idle_type idle) | |
3406 | { | |
3407 | int load_idx; | |
3408 | ||
3409 | switch (idle) { | |
3410 | case CPU_NOT_IDLE: | |
7897986b | 3411 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3412 | break; |
3413 | ||
3414 | case CPU_NEWLY_IDLE: | |
7897986b | 3415 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3416 | break; |
3417 | default: | |
7897986b | 3418 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3419 | break; |
3420 | } | |
1da177e4 | 3421 | |
67bb6c03 GS |
3422 | return load_idx; |
3423 | } | |
1da177e4 | 3424 | |
1da177e4 | 3425 | |
c071df18 GS |
3426 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3427 | /** | |
3428 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3429 | * the given sched_domain, during load balancing. | |
3430 | * | |
3431 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3432 | * @sds: Variable containing the statistics for sd. | |
3433 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3434 | */ | |
3435 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3436 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3437 | { | |
3438 | /* | |
3439 | * Busy processors will not participate in power savings | |
3440 | * balance. | |
3441 | */ | |
3442 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3443 | sds->power_savings_balance = 0; | |
3444 | else { | |
3445 | sds->power_savings_balance = 1; | |
3446 | sds->min_nr_running = ULONG_MAX; | |
3447 | sds->leader_nr_running = 0; | |
3448 | } | |
3449 | } | |
783609c6 | 3450 | |
c071df18 GS |
3451 | /** |
3452 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3453 | * sched_domain while performing load balancing. | |
3454 | * | |
3455 | * @group: sched_group belonging to the sched_domain under consideration. | |
3456 | * @sds: Variable containing the statistics of the sched_domain | |
3457 | * @local_group: Does group contain the CPU for which we're performing | |
3458 | * load balancing ? | |
3459 | * @sgs: Variable containing the statistics of the group. | |
3460 | */ | |
3461 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3462 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3463 | { | |
408ed066 | 3464 | |
c071df18 GS |
3465 | if (!sds->power_savings_balance) |
3466 | return; | |
1da177e4 | 3467 | |
c071df18 GS |
3468 | /* |
3469 | * If the local group is idle or completely loaded | |
3470 | * no need to do power savings balance at this domain | |
3471 | */ | |
3472 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3473 | !sds->this_nr_running)) | |
3474 | sds->power_savings_balance = 0; | |
2dd73a4f | 3475 | |
c071df18 GS |
3476 | /* |
3477 | * If a group is already running at full capacity or idle, | |
3478 | * don't include that group in power savings calculations | |
3479 | */ | |
3480 | if (!sds->power_savings_balance || | |
3481 | sgs->sum_nr_running >= sgs->group_capacity || | |
3482 | !sgs->sum_nr_running) | |
3483 | return; | |
5969fe06 | 3484 | |
c071df18 GS |
3485 | /* |
3486 | * Calculate the group which has the least non-idle load. | |
3487 | * This is the group from where we need to pick up the load | |
3488 | * for saving power | |
3489 | */ | |
3490 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3491 | (sgs->sum_nr_running == sds->min_nr_running && | |
3492 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3493 | sds->group_min = group; | |
3494 | sds->min_nr_running = sgs->sum_nr_running; | |
3495 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3496 | sgs->sum_nr_running; | |
3497 | } | |
783609c6 | 3498 | |
c071df18 GS |
3499 | /* |
3500 | * Calculate the group which is almost near its | |
3501 | * capacity but still has some space to pick up some load | |
3502 | * from other group and save more power | |
3503 | */ | |
d899a789 | 3504 | if (sgs->sum_nr_running + 1 > sgs->group_capacity) |
c071df18 | 3505 | return; |
1da177e4 | 3506 | |
c071df18 GS |
3507 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3508 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3509 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3510 | sds->group_leader = group; | |
3511 | sds->leader_nr_running = sgs->sum_nr_running; | |
3512 | } | |
3513 | } | |
408ed066 | 3514 | |
c071df18 | 3515 | /** |
d5ac537e | 3516 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3517 | * @sds: Variable containing the statistics of the sched_domain |
3518 | * under consideration. | |
3519 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3520 | * @imbalance: Variable to store the imbalance. | |
3521 | * | |
d5ac537e RD |
3522 | * Description: |
3523 | * Check if we have potential to perform some power-savings balance. | |
3524 | * If yes, set the busiest group to be the least loaded group in the | |
3525 | * sched_domain, so that it's CPUs can be put to idle. | |
3526 | * | |
c071df18 GS |
3527 | * Returns 1 if there is potential to perform power-savings balance. |
3528 | * Else returns 0. | |
3529 | */ | |
3530 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3531 | int this_cpu, unsigned long *imbalance) | |
3532 | { | |
3533 | if (!sds->power_savings_balance) | |
3534 | return 0; | |
1da177e4 | 3535 | |
c071df18 GS |
3536 | if (sds->this != sds->group_leader || |
3537 | sds->group_leader == sds->group_min) | |
3538 | return 0; | |
783609c6 | 3539 | |
c071df18 GS |
3540 | *imbalance = sds->min_load_per_task; |
3541 | sds->busiest = sds->group_min; | |
1da177e4 | 3542 | |
c071df18 | 3543 | return 1; |
1da177e4 | 3544 | |
c071df18 GS |
3545 | } |
3546 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3547 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3548 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3549 | { | |
3550 | return; | |
3551 | } | |
408ed066 | 3552 | |
c071df18 GS |
3553 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3554 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3555 | { | |
3556 | return; | |
3557 | } | |
3558 | ||
3559 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3560 | int this_cpu, unsigned long *imbalance) | |
3561 | { | |
3562 | return 0; | |
3563 | } | |
3564 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3565 | ||
d6a59aa3 PZ |
3566 | |
3567 | unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu) | |
3568 | { | |
3569 | return SCHED_LOAD_SCALE; | |
3570 | } | |
3571 | ||
3572 | unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu) | |
3573 | { | |
3574 | return default_scale_freq_power(sd, cpu); | |
3575 | } | |
3576 | ||
3577 | unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu) | |
ab29230e PZ |
3578 | { |
3579 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3580 | unsigned long smt_gain = sd->smt_gain; | |
3581 | ||
3582 | smt_gain /= weight; | |
3583 | ||
3584 | return smt_gain; | |
3585 | } | |
3586 | ||
d6a59aa3 PZ |
3587 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
3588 | { | |
3589 | return default_scale_smt_power(sd, cpu); | |
3590 | } | |
3591 | ||
e9e9250b PZ |
3592 | unsigned long scale_rt_power(int cpu) |
3593 | { | |
3594 | struct rq *rq = cpu_rq(cpu); | |
3595 | u64 total, available; | |
3596 | ||
3597 | sched_avg_update(rq); | |
3598 | ||
3599 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3600 | available = total - rq->rt_avg; | |
3601 | ||
3602 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3603 | total = SCHED_LOAD_SCALE; | |
3604 | ||
3605 | total >>= SCHED_LOAD_SHIFT; | |
3606 | ||
3607 | return div_u64(available, total); | |
3608 | } | |
3609 | ||
ab29230e PZ |
3610 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3611 | { | |
3612 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3613 | unsigned long power = SCHED_LOAD_SCALE; | |
3614 | struct sched_group *sdg = sd->groups; | |
ab29230e | 3615 | |
8e6598af PZ |
3616 | if (sched_feat(ARCH_POWER)) |
3617 | power *= arch_scale_freq_power(sd, cpu); | |
3618 | else | |
3619 | power *= default_scale_freq_power(sd, cpu); | |
3620 | ||
d6a59aa3 | 3621 | power >>= SCHED_LOAD_SHIFT; |
ab29230e PZ |
3622 | |
3623 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
8e6598af PZ |
3624 | if (sched_feat(ARCH_POWER)) |
3625 | power *= arch_scale_smt_power(sd, cpu); | |
3626 | else | |
3627 | power *= default_scale_smt_power(sd, cpu); | |
3628 | ||
ab29230e PZ |
3629 | power >>= SCHED_LOAD_SHIFT; |
3630 | } | |
3631 | ||
e9e9250b PZ |
3632 | power *= scale_rt_power(cpu); |
3633 | power >>= SCHED_LOAD_SHIFT; | |
3634 | ||
3635 | if (!power) | |
3636 | power = 1; | |
ab29230e | 3637 | |
18a3885f | 3638 | sdg->cpu_power = power; |
ab29230e PZ |
3639 | } |
3640 | ||
3641 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3642 | { |
3643 | struct sched_domain *child = sd->child; | |
3644 | struct sched_group *group, *sdg = sd->groups; | |
d7ea17a7 | 3645 | unsigned long power; |
cc9fba7d PZ |
3646 | |
3647 | if (!child) { | |
ab29230e | 3648 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3649 | return; |
3650 | } | |
3651 | ||
d7ea17a7 | 3652 | power = 0; |
cc9fba7d PZ |
3653 | |
3654 | group = child->groups; | |
3655 | do { | |
d7ea17a7 | 3656 | power += group->cpu_power; |
cc9fba7d PZ |
3657 | group = group->next; |
3658 | } while (group != child->groups); | |
d7ea17a7 IM |
3659 | |
3660 | sdg->cpu_power = power; | |
cc9fba7d | 3661 | } |
c071df18 | 3662 | |
1f8c553d GS |
3663 | /** |
3664 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
e17b38bf | 3665 | * @sd: The sched_domain whose statistics are to be updated. |
1f8c553d GS |
3666 | * @group: sched_group whose statistics are to be updated. |
3667 | * @this_cpu: Cpu for which load balance is currently performed. | |
3668 | * @idle: Idle status of this_cpu | |
3669 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3670 | * @sd_idle: Idle status of the sched_domain containing group. | |
3671 | * @local_group: Does group contain this_cpu. | |
3672 | * @cpus: Set of cpus considered for load balancing. | |
3673 | * @balance: Should we balance. | |
3674 | * @sgs: variable to hold the statistics for this group. | |
3675 | */ | |
cc9fba7d PZ |
3676 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3677 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3678 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3679 | int local_group, const struct cpumask *cpus, | |
3680 | int *balance, struct sg_lb_stats *sgs) | |
3681 | { | |
3682 | unsigned long load, max_cpu_load, min_cpu_load; | |
3683 | int i; | |
3684 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3685 | unsigned long sum_avg_load_per_task; | |
3686 | unsigned long avg_load_per_task; | |
3687 | ||
cc9fba7d | 3688 | if (local_group) { |
1f8c553d | 3689 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3690 | if (balance_cpu == this_cpu) |
ab29230e | 3691 | update_group_power(sd, this_cpu); |
cc9fba7d | 3692 | } |
1f8c553d GS |
3693 | |
3694 | /* Tally up the load of all CPUs in the group */ | |
3695 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3696 | max_cpu_load = 0; | |
3697 | min_cpu_load = ~0UL; | |
408ed066 | 3698 | |
1f8c553d GS |
3699 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3700 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3701 | |
1f8c553d GS |
3702 | if (*sd_idle && rq->nr_running) |
3703 | *sd_idle = 0; | |
5c45bf27 | 3704 | |
1f8c553d | 3705 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3706 | if (local_group) { |
1f8c553d GS |
3707 | if (idle_cpu(i) && !first_idle_cpu) { |
3708 | first_idle_cpu = 1; | |
3709 | balance_cpu = i; | |
3710 | } | |
3711 | ||
3712 | load = target_load(i, load_idx); | |
3713 | } else { | |
3714 | load = source_load(i, load_idx); | |
3715 | if (load > max_cpu_load) | |
3716 | max_cpu_load = load; | |
3717 | if (min_cpu_load > load) | |
3718 | min_cpu_load = load; | |
1da177e4 | 3719 | } |
5c45bf27 | 3720 | |
1f8c553d GS |
3721 | sgs->group_load += load; |
3722 | sgs->sum_nr_running += rq->nr_running; | |
3723 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3724 | |
1f8c553d GS |
3725 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3726 | } | |
5c45bf27 | 3727 | |
1f8c553d GS |
3728 | /* |
3729 | * First idle cpu or the first cpu(busiest) in this sched group | |
3730 | * is eligible for doing load balancing at this and above | |
3731 | * domains. In the newly idle case, we will allow all the cpu's | |
3732 | * to do the newly idle load balance. | |
3733 | */ | |
3734 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3735 | balance_cpu != this_cpu && balance) { | |
3736 | *balance = 0; | |
3737 | return; | |
3738 | } | |
5c45bf27 | 3739 | |
1f8c553d | 3740 | /* Adjust by relative CPU power of the group */ |
18a3885f | 3741 | sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power; |
5c45bf27 | 3742 | |
1f8c553d GS |
3743 | |
3744 | /* | |
3745 | * Consider the group unbalanced when the imbalance is larger | |
3746 | * than the average weight of two tasks. | |
3747 | * | |
3748 | * APZ: with cgroup the avg task weight can vary wildly and | |
3749 | * might not be a suitable number - should we keep a | |
3750 | * normalized nr_running number somewhere that negates | |
3751 | * the hierarchy? | |
3752 | */ | |
18a3885f PZ |
3753 | avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) / |
3754 | group->cpu_power; | |
1f8c553d GS |
3755 | |
3756 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3757 | sgs->group_imb = 1; | |
3758 | ||
bdb94aa5 | 3759 | sgs->group_capacity = |
18a3885f | 3760 | DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE); |
1f8c553d | 3761 | } |
dd41f596 | 3762 | |
37abe198 GS |
3763 | /** |
3764 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3765 | * @sd: sched_domain whose statistics are to be updated. | |
3766 | * @this_cpu: Cpu for which load balance is currently performed. | |
3767 | * @idle: Idle status of this_cpu | |
3768 | * @sd_idle: Idle status of the sched_domain containing group. | |
3769 | * @cpus: Set of cpus considered for load balancing. | |
3770 | * @balance: Should we balance. | |
3771 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3772 | */ |
37abe198 GS |
3773 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3774 | enum cpu_idle_type idle, int *sd_idle, | |
3775 | const struct cpumask *cpus, int *balance, | |
3776 | struct sd_lb_stats *sds) | |
1da177e4 | 3777 | { |
b5d978e0 | 3778 | struct sched_domain *child = sd->child; |
222d656d | 3779 | struct sched_group *group = sd->groups; |
37abe198 | 3780 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3781 | int load_idx, prefer_sibling = 0; |
3782 | ||
3783 | if (child && child->flags & SD_PREFER_SIBLING) | |
3784 | prefer_sibling = 1; | |
222d656d | 3785 | |
c071df18 | 3786 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3787 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3788 | |
3789 | do { | |
1da177e4 | 3790 | int local_group; |
1da177e4 | 3791 | |
758b2cdc RR |
3792 | local_group = cpumask_test_cpu(this_cpu, |
3793 | sched_group_cpus(group)); | |
381be78f | 3794 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3795 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3796 | local_group, cpus, balance, &sgs); |
1da177e4 | 3797 | |
37abe198 GS |
3798 | if (local_group && balance && !(*balance)) |
3799 | return; | |
783609c6 | 3800 | |
37abe198 | 3801 | sds->total_load += sgs.group_load; |
18a3885f | 3802 | sds->total_pwr += group->cpu_power; |
1da177e4 | 3803 | |
b5d978e0 PZ |
3804 | /* |
3805 | * In case the child domain prefers tasks go to siblings | |
3806 | * first, lower the group capacity to one so that we'll try | |
3807 | * and move all the excess tasks away. | |
3808 | */ | |
3809 | if (prefer_sibling) | |
bdb94aa5 | 3810 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
1da177e4 | 3811 | |
1da177e4 | 3812 | if (local_group) { |
37abe198 GS |
3813 | sds->this_load = sgs.avg_load; |
3814 | sds->this = group; | |
3815 | sds->this_nr_running = sgs.sum_nr_running; | |
3816 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3817 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3818 | (sgs.sum_nr_running > sgs.group_capacity || |
3819 | sgs.group_imb)) { | |
37abe198 GS |
3820 | sds->max_load = sgs.avg_load; |
3821 | sds->busiest = group; | |
3822 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3823 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3824 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3825 | } |
5c45bf27 | 3826 | |
c071df18 | 3827 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3828 | group = group->next; |
3829 | } while (group != sd->groups); | |
37abe198 | 3830 | } |
1da177e4 | 3831 | |
2e6f44ae GS |
3832 | /** |
3833 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3834 | * amongst the groups of a sched_domain, during |
3835 | * load balancing. | |
2e6f44ae GS |
3836 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3837 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3838 | * @imbalance: Variable to store the imbalance. | |
3839 | */ | |
3840 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3841 | int this_cpu, unsigned long *imbalance) | |
3842 | { | |
3843 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3844 | unsigned int imbn = 2; | |
3845 | ||
3846 | if (sds->this_nr_running) { | |
3847 | sds->this_load_per_task /= sds->this_nr_running; | |
3848 | if (sds->busiest_load_per_task > | |
3849 | sds->this_load_per_task) | |
3850 | imbn = 1; | |
3851 | } else | |
3852 | sds->this_load_per_task = | |
3853 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3854 | |
2e6f44ae GS |
3855 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3856 | sds->busiest_load_per_task * imbn) { | |
3857 | *imbalance = sds->busiest_load_per_task; | |
3858 | return; | |
3859 | } | |
908a7c1b | 3860 | |
1da177e4 | 3861 | /* |
2e6f44ae GS |
3862 | * OK, we don't have enough imbalance to justify moving tasks, |
3863 | * however we may be able to increase total CPU power used by | |
3864 | * moving them. | |
1da177e4 | 3865 | */ |
2dd73a4f | 3866 | |
18a3885f | 3867 | pwr_now += sds->busiest->cpu_power * |
2e6f44ae | 3868 | min(sds->busiest_load_per_task, sds->max_load); |
18a3885f | 3869 | pwr_now += sds->this->cpu_power * |
2e6f44ae GS |
3870 | min(sds->this_load_per_task, sds->this_load); |
3871 | pwr_now /= SCHED_LOAD_SCALE; | |
3872 | ||
3873 | /* Amount of load we'd subtract */ | |
18a3885f PZ |
3874 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3875 | sds->busiest->cpu_power; | |
2e6f44ae | 3876 | if (sds->max_load > tmp) |
18a3885f | 3877 | pwr_move += sds->busiest->cpu_power * |
2e6f44ae GS |
3878 | min(sds->busiest_load_per_task, sds->max_load - tmp); |
3879 | ||
3880 | /* Amount of load we'd add */ | |
18a3885f | 3881 | if (sds->max_load * sds->busiest->cpu_power < |
2e6f44ae | 3882 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) |
18a3885f PZ |
3883 | tmp = (sds->max_load * sds->busiest->cpu_power) / |
3884 | sds->this->cpu_power; | |
2e6f44ae | 3885 | else |
18a3885f PZ |
3886 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3887 | sds->this->cpu_power; | |
3888 | pwr_move += sds->this->cpu_power * | |
2e6f44ae GS |
3889 | min(sds->this_load_per_task, sds->this_load + tmp); |
3890 | pwr_move /= SCHED_LOAD_SCALE; | |
3891 | ||
3892 | /* Move if we gain throughput */ | |
3893 | if (pwr_move > pwr_now) | |
3894 | *imbalance = sds->busiest_load_per_task; | |
3895 | } | |
dbc523a3 GS |
3896 | |
3897 | /** | |
3898 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3899 | * groups of a given sched_domain during load balance. | |
3900 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3901 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3902 | * @imbalance: The variable to store the imbalance. | |
3903 | */ | |
3904 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3905 | unsigned long *imbalance) | |
3906 | { | |
3907 | unsigned long max_pull; | |
2dd73a4f PW |
3908 | /* |
3909 | * In the presence of smp nice balancing, certain scenarios can have | |
3910 | * max load less than avg load(as we skip the groups at or below | |
3911 | * its cpu_power, while calculating max_load..) | |
3912 | */ | |
dbc523a3 | 3913 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3914 | *imbalance = 0; |
dbc523a3 | 3915 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3916 | } |
0c117f1b SS |
3917 | |
3918 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3919 | max_pull = min(sds->max_load - sds->avg_load, |
3920 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3921 | |
1da177e4 | 3922 | /* How much load to actually move to equalise the imbalance */ |
18a3885f PZ |
3923 | *imbalance = min(max_pull * sds->busiest->cpu_power, |
3924 | (sds->avg_load - sds->this_load) * sds->this->cpu_power) | |
1da177e4 LT |
3925 | / SCHED_LOAD_SCALE; |
3926 | ||
2dd73a4f PW |
3927 | /* |
3928 | * if *imbalance is less than the average load per runnable task | |
3929 | * there is no gaurantee that any tasks will be moved so we'll have | |
3930 | * a think about bumping its value to force at least one task to be | |
3931 | * moved | |
3932 | */ | |
dbc523a3 GS |
3933 | if (*imbalance < sds->busiest_load_per_task) |
3934 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3935 | |
dbc523a3 | 3936 | } |
37abe198 | 3937 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3938 | |
b7bb4c9b GS |
3939 | /** |
3940 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3941 | * if there is an imbalance. If there isn't an imbalance, and | |
3942 | * the user has opted for power-savings, it returns a group whose | |
3943 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3944 | * such a group exists. | |
3945 | * | |
3946 | * Also calculates the amount of weighted load which should be moved | |
3947 | * to restore balance. | |
3948 | * | |
3949 | * @sd: The sched_domain whose busiest group is to be returned. | |
3950 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3951 | * @imbalance: Variable which stores amount of weighted load which should | |
3952 | * be moved to restore balance/put a group to idle. | |
3953 | * @idle: The idle status of this_cpu. | |
3954 | * @sd_idle: The idleness of sd | |
3955 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3956 | * @balance: Pointer to a variable indicating if this_cpu | |
3957 | * is the appropriate cpu to perform load balancing at this_level. | |
3958 | * | |
3959 | * Returns: - the busiest group if imbalance exists. | |
3960 | * - If no imbalance and user has opted for power-savings balance, | |
3961 | * return the least loaded group whose CPUs can be | |
3962 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3963 | */ |
3964 | static struct sched_group * | |
3965 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3966 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3967 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3968 | { | |
3969 | struct sd_lb_stats sds; | |
1da177e4 | 3970 | |
37abe198 | 3971 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3972 | |
37abe198 GS |
3973 | /* |
3974 | * Compute the various statistics relavent for load balancing at | |
3975 | * this level. | |
3976 | */ | |
3977 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3978 | balance, &sds); | |
3979 | ||
b7bb4c9b GS |
3980 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3981 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3982 | * at this level. | |
3983 | * 2) There is no busy sibling group to pull from. | |
3984 | * 3) This group is the busiest group. | |
3985 | * 4) This group is more busy than the avg busieness at this | |
3986 | * sched_domain. | |
3987 | * 5) The imbalance is within the specified limit. | |
3988 | * 6) Any rebalance would lead to ping-pong | |
3989 | */ | |
37abe198 GS |
3990 | if (balance && !(*balance)) |
3991 | goto ret; | |
1da177e4 | 3992 | |
b7bb4c9b GS |
3993 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3994 | goto out_balanced; | |
1da177e4 | 3995 | |
b7bb4c9b | 3996 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3997 | goto out_balanced; |
1da177e4 | 3998 | |
222d656d | 3999 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4000 | |
b7bb4c9b GS |
4001 | if (sds.this_load >= sds.avg_load) |
4002 | goto out_balanced; | |
4003 | ||
4004 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4005 | goto out_balanced; |
4006 | ||
222d656d GS |
4007 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4008 | if (sds.group_imb) | |
4009 | sds.busiest_load_per_task = | |
4010 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4011 | |
1da177e4 LT |
4012 | /* |
4013 | * We're trying to get all the cpus to the average_load, so we don't | |
4014 | * want to push ourselves above the average load, nor do we wish to | |
4015 | * reduce the max loaded cpu below the average load, as either of these | |
4016 | * actions would just result in more rebalancing later, and ping-pong | |
4017 | * tasks around. Thus we look for the minimum possible imbalance. | |
4018 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4019 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4020 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4021 | * appear as very large values with unsigned longs. |
4022 | */ | |
222d656d | 4023 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4024 | goto out_balanced; |
4025 | ||
dbc523a3 GS |
4026 | /* Looks like there is an imbalance. Compute it */ |
4027 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4028 | return sds.busiest; |
1da177e4 LT |
4029 | |
4030 | out_balanced: | |
c071df18 GS |
4031 | /* |
4032 | * There is no obvious imbalance. But check if we can do some balancing | |
4033 | * to save power. | |
4034 | */ | |
4035 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4036 | return sds.busiest; | |
783609c6 | 4037 | ret: |
1da177e4 LT |
4038 | *imbalance = 0; |
4039 | return NULL; | |
4040 | } | |
4041 | ||
4042 | /* | |
4043 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4044 | */ | |
70b97a7f | 4045 | static struct rq * |
d15bcfdb | 4046 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4047 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4048 | { |
70b97a7f | 4049 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4050 | unsigned long max_load = 0; |
1da177e4 LT |
4051 | int i; |
4052 | ||
758b2cdc | 4053 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4054 | unsigned long power = power_of(i); |
4055 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4056 | unsigned long wl; |
0a2966b4 | 4057 | |
96f874e2 | 4058 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4059 | continue; |
4060 | ||
48f24c4d | 4061 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4062 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4063 | wl /= power; | |
2dd73a4f | 4064 | |
bdb94aa5 | 4065 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4066 | continue; |
1da177e4 | 4067 | |
dd41f596 IM |
4068 | if (wl > max_load) { |
4069 | max_load = wl; | |
48f24c4d | 4070 | busiest = rq; |
1da177e4 LT |
4071 | } |
4072 | } | |
4073 | ||
4074 | return busiest; | |
4075 | } | |
4076 | ||
77391d71 NP |
4077 | /* |
4078 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4079 | * so long as it is large enough. | |
4080 | */ | |
4081 | #define MAX_PINNED_INTERVAL 512 | |
4082 | ||
df7c8e84 RR |
4083 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4084 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4085 | ||
1da177e4 LT |
4086 | /* |
4087 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4088 | * tasks if there is an imbalance. | |
1da177e4 | 4089 | */ |
70b97a7f | 4090 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4091 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4092 | int *balance) |
1da177e4 | 4093 | { |
43010659 | 4094 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4095 | struct sched_group *group; |
1da177e4 | 4096 | unsigned long imbalance; |
70b97a7f | 4097 | struct rq *busiest; |
fe2eea3f | 4098 | unsigned long flags; |
df7c8e84 | 4099 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4100 | |
eae0c9df | 4101 | cpumask_copy(cpus, cpu_online_mask); |
7c16ec58 | 4102 | |
89c4710e SS |
4103 | /* |
4104 | * When power savings policy is enabled for the parent domain, idle | |
4105 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4106 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4107 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4108 | */ |
d15bcfdb | 4109 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4110 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4111 | sd_idle = 1; |
1da177e4 | 4112 | |
2d72376b | 4113 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4114 | |
0a2966b4 | 4115 | redo: |
c8cba857 | 4116 | update_shares(sd); |
0a2966b4 | 4117 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4118 | cpus, balance); |
783609c6 | 4119 | |
06066714 | 4120 | if (*balance == 0) |
783609c6 | 4121 | goto out_balanced; |
783609c6 | 4122 | |
1da177e4 LT |
4123 | if (!group) { |
4124 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4125 | goto out_balanced; | |
4126 | } | |
4127 | ||
7c16ec58 | 4128 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4129 | if (!busiest) { |
4130 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4131 | goto out_balanced; | |
4132 | } | |
4133 | ||
db935dbd | 4134 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4135 | |
4136 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4137 | ||
43010659 | 4138 | ld_moved = 0; |
1da177e4 LT |
4139 | if (busiest->nr_running > 1) { |
4140 | /* | |
4141 | * Attempt to move tasks. If find_busiest_group has found | |
4142 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4143 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4144 | * correctly treated as an imbalance. |
4145 | */ | |
fe2eea3f | 4146 | local_irq_save(flags); |
e17224bf | 4147 | double_rq_lock(this_rq, busiest); |
43010659 | 4148 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4149 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4150 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4151 | local_irq_restore(flags); |
81026794 | 4152 | |
46cb4b7c SS |
4153 | /* |
4154 | * some other cpu did the load balance for us. | |
4155 | */ | |
43010659 | 4156 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4157 | resched_cpu(this_cpu); |
4158 | ||
81026794 | 4159 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4160 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4161 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4162 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4163 | goto redo; |
81026794 | 4164 | goto out_balanced; |
0a2966b4 | 4165 | } |
1da177e4 | 4166 | } |
81026794 | 4167 | |
43010659 | 4168 | if (!ld_moved) { |
1da177e4 LT |
4169 | schedstat_inc(sd, lb_failed[idle]); |
4170 | sd->nr_balance_failed++; | |
4171 | ||
4172 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4173 | |
fe2eea3f | 4174 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4175 | |
4176 | /* don't kick the migration_thread, if the curr | |
4177 | * task on busiest cpu can't be moved to this_cpu | |
4178 | */ | |
96f874e2 RR |
4179 | if (!cpumask_test_cpu(this_cpu, |
4180 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4181 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4182 | all_pinned = 1; |
4183 | goto out_one_pinned; | |
4184 | } | |
4185 | ||
1da177e4 LT |
4186 | if (!busiest->active_balance) { |
4187 | busiest->active_balance = 1; | |
4188 | busiest->push_cpu = this_cpu; | |
81026794 | 4189 | active_balance = 1; |
1da177e4 | 4190 | } |
fe2eea3f | 4191 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4192 | if (active_balance) |
1da177e4 LT |
4193 | wake_up_process(busiest->migration_thread); |
4194 | ||
4195 | /* | |
4196 | * We've kicked active balancing, reset the failure | |
4197 | * counter. | |
4198 | */ | |
39507451 | 4199 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4200 | } |
81026794 | 4201 | } else |
1da177e4 LT |
4202 | sd->nr_balance_failed = 0; |
4203 | ||
81026794 | 4204 | if (likely(!active_balance)) { |
1da177e4 LT |
4205 | /* We were unbalanced, so reset the balancing interval */ |
4206 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4207 | } else { |
4208 | /* | |
4209 | * If we've begun active balancing, start to back off. This | |
4210 | * case may not be covered by the all_pinned logic if there | |
4211 | * is only 1 task on the busy runqueue (because we don't call | |
4212 | * move_tasks). | |
4213 | */ | |
4214 | if (sd->balance_interval < sd->max_interval) | |
4215 | sd->balance_interval *= 2; | |
1da177e4 LT |
4216 | } |
4217 | ||
43010659 | 4218 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4219 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4220 | ld_moved = -1; |
4221 | ||
4222 | goto out; | |
1da177e4 LT |
4223 | |
4224 | out_balanced: | |
1da177e4 LT |
4225 | schedstat_inc(sd, lb_balanced[idle]); |
4226 | ||
16cfb1c0 | 4227 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4228 | |
4229 | out_one_pinned: | |
1da177e4 | 4230 | /* tune up the balancing interval */ |
77391d71 NP |
4231 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4232 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4233 | sd->balance_interval *= 2; |
4234 | ||
48f24c4d | 4235 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4236 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4237 | ld_moved = -1; |
4238 | else | |
4239 | ld_moved = 0; | |
4240 | out: | |
c8cba857 PZ |
4241 | if (ld_moved) |
4242 | update_shares(sd); | |
c09595f6 | 4243 | return ld_moved; |
1da177e4 LT |
4244 | } |
4245 | ||
4246 | /* | |
4247 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4248 | * tasks if there is an imbalance. | |
4249 | * | |
d15bcfdb | 4250 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4251 | * this_rq is locked. |
4252 | */ | |
48f24c4d | 4253 | static int |
df7c8e84 | 4254 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4255 | { |
4256 | struct sched_group *group; | |
70b97a7f | 4257 | struct rq *busiest = NULL; |
1da177e4 | 4258 | unsigned long imbalance; |
43010659 | 4259 | int ld_moved = 0; |
5969fe06 | 4260 | int sd_idle = 0; |
969bb4e4 | 4261 | int all_pinned = 0; |
df7c8e84 | 4262 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4263 | |
eae0c9df | 4264 | cpumask_copy(cpus, cpu_online_mask); |
5969fe06 | 4265 | |
89c4710e SS |
4266 | /* |
4267 | * When power savings policy is enabled for the parent domain, idle | |
4268 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4269 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4270 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4271 | */ |
4272 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4273 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4274 | sd_idle = 1; |
1da177e4 | 4275 | |
2d72376b | 4276 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4277 | redo: |
3e5459b4 | 4278 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4279 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4280 | &sd_idle, cpus, NULL); |
1da177e4 | 4281 | if (!group) { |
d15bcfdb | 4282 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4283 | goto out_balanced; |
1da177e4 LT |
4284 | } |
4285 | ||
7c16ec58 | 4286 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4287 | if (!busiest) { |
d15bcfdb | 4288 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4289 | goto out_balanced; |
1da177e4 LT |
4290 | } |
4291 | ||
db935dbd NP |
4292 | BUG_ON(busiest == this_rq); |
4293 | ||
d15bcfdb | 4294 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4295 | |
43010659 | 4296 | ld_moved = 0; |
d6d5cfaf NP |
4297 | if (busiest->nr_running > 1) { |
4298 | /* Attempt to move tasks */ | |
4299 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4300 | /* this_rq->clock is already updated */ |
4301 | update_rq_clock(busiest); | |
43010659 | 4302 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4303 | imbalance, sd, CPU_NEWLY_IDLE, |
4304 | &all_pinned); | |
1b12bbc7 | 4305 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4306 | |
969bb4e4 | 4307 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4308 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4309 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4310 | goto redo; |
4311 | } | |
d6d5cfaf NP |
4312 | } |
4313 | ||
43010659 | 4314 | if (!ld_moved) { |
36dffab6 | 4315 | int active_balance = 0; |
ad273b32 | 4316 | |
d15bcfdb | 4317 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4318 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4319 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4320 | return -1; |
ad273b32 VS |
4321 | |
4322 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4323 | return -1; | |
4324 | ||
4325 | if (sd->nr_balance_failed++ < 2) | |
4326 | return -1; | |
4327 | ||
4328 | /* | |
4329 | * The only task running in a non-idle cpu can be moved to this | |
4330 | * cpu in an attempt to completely freeup the other CPU | |
4331 | * package. The same method used to move task in load_balance() | |
4332 | * have been extended for load_balance_newidle() to speedup | |
4333 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4334 | * | |
4335 | * The package power saving logic comes from | |
4336 | * find_busiest_group(). If there are no imbalance, then | |
4337 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4338 | * f_b_g() will select a group from which a running task may be | |
4339 | * pulled to this cpu in order to make the other package idle. | |
4340 | * If there is no opportunity to make a package idle and if | |
4341 | * there are no imbalance, then f_b_g() will return NULL and no | |
4342 | * action will be taken in load_balance_newidle(). | |
4343 | * | |
4344 | * Under normal task pull operation due to imbalance, there | |
4345 | * will be more than one task in the source run queue and | |
4346 | * move_tasks() will succeed. ld_moved will be true and this | |
4347 | * active balance code will not be triggered. | |
4348 | */ | |
4349 | ||
4350 | /* Lock busiest in correct order while this_rq is held */ | |
4351 | double_lock_balance(this_rq, busiest); | |
4352 | ||
4353 | /* | |
4354 | * don't kick the migration_thread, if the curr | |
4355 | * task on busiest cpu can't be moved to this_cpu | |
4356 | */ | |
6ca09dfc | 4357 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4358 | double_unlock_balance(this_rq, busiest); |
4359 | all_pinned = 1; | |
4360 | return ld_moved; | |
4361 | } | |
4362 | ||
4363 | if (!busiest->active_balance) { | |
4364 | busiest->active_balance = 1; | |
4365 | busiest->push_cpu = this_cpu; | |
4366 | active_balance = 1; | |
4367 | } | |
4368 | ||
4369 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4370 | /* |
4371 | * Should not call ttwu while holding a rq->lock | |
4372 | */ | |
4373 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4374 | if (active_balance) |
4375 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4376 | spin_lock(&this_rq->lock); |
ad273b32 | 4377 | |
5969fe06 | 4378 | } else |
16cfb1c0 | 4379 | sd->nr_balance_failed = 0; |
1da177e4 | 4380 | |
3e5459b4 | 4381 | update_shares_locked(this_rq, sd); |
43010659 | 4382 | return ld_moved; |
16cfb1c0 NP |
4383 | |
4384 | out_balanced: | |
d15bcfdb | 4385 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4386 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4387 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4388 | return -1; |
16cfb1c0 | 4389 | sd->nr_balance_failed = 0; |
48f24c4d | 4390 | |
16cfb1c0 | 4391 | return 0; |
1da177e4 LT |
4392 | } |
4393 | ||
4394 | /* | |
4395 | * idle_balance is called by schedule() if this_cpu is about to become | |
4396 | * idle. Attempts to pull tasks from other CPUs. | |
4397 | */ | |
70b97a7f | 4398 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4399 | { |
4400 | struct sched_domain *sd; | |
efbe027e | 4401 | int pulled_task = 0; |
dd41f596 | 4402 | unsigned long next_balance = jiffies + HZ; |
1da177e4 | 4403 | |
1b9508f6 MG |
4404 | this_rq->idle_stamp = this_rq->clock; |
4405 | ||
4406 | if (this_rq->avg_idle < sysctl_sched_migration_cost) | |
4407 | return; | |
4408 | ||
1da177e4 | 4409 | for_each_domain(this_cpu, sd) { |
92c4ca5c CL |
4410 | unsigned long interval; |
4411 | ||
4412 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4413 | continue; | |
4414 | ||
4415 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4416 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4417 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4418 | sd); |
92c4ca5c CL |
4419 | |
4420 | interval = msecs_to_jiffies(sd->balance_interval); | |
4421 | if (time_after(next_balance, sd->last_balance + interval)) | |
4422 | next_balance = sd->last_balance + interval; | |
1b9508f6 MG |
4423 | if (pulled_task) { |
4424 | this_rq->idle_stamp = 0; | |
92c4ca5c | 4425 | break; |
1b9508f6 | 4426 | } |
1da177e4 | 4427 | } |
dd41f596 | 4428 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4429 | /* |
4430 | * We are going idle. next_balance may be set based on | |
4431 | * a busy processor. So reset next_balance. | |
4432 | */ | |
4433 | this_rq->next_balance = next_balance; | |
dd41f596 | 4434 | } |
1da177e4 LT |
4435 | } |
4436 | ||
4437 | /* | |
4438 | * active_load_balance is run by migration threads. It pushes running tasks | |
4439 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4440 | * running on each physical CPU where possible, and avoids physical / | |
4441 | * logical imbalances. | |
4442 | * | |
4443 | * Called with busiest_rq locked. | |
4444 | */ | |
70b97a7f | 4445 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4446 | { |
39507451 | 4447 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4448 | struct sched_domain *sd; |
4449 | struct rq *target_rq; | |
39507451 | 4450 | |
48f24c4d | 4451 | /* Is there any task to move? */ |
39507451 | 4452 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4453 | return; |
4454 | ||
4455 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4456 | |
4457 | /* | |
39507451 | 4458 | * This condition is "impossible", if it occurs |
41a2d6cf | 4459 | * we need to fix it. Originally reported by |
39507451 | 4460 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4461 | */ |
39507451 | 4462 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4463 | |
39507451 NP |
4464 | /* move a task from busiest_rq to target_rq */ |
4465 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4466 | update_rq_clock(busiest_rq); |
4467 | update_rq_clock(target_rq); | |
39507451 NP |
4468 | |
4469 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4470 | for_each_domain(target_cpu, sd) { |
39507451 | 4471 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4472 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4473 | break; |
c96d145e | 4474 | } |
39507451 | 4475 | |
48f24c4d | 4476 | if (likely(sd)) { |
2d72376b | 4477 | schedstat_inc(sd, alb_count); |
39507451 | 4478 | |
43010659 PW |
4479 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4480 | sd, CPU_IDLE)) | |
48f24c4d IM |
4481 | schedstat_inc(sd, alb_pushed); |
4482 | else | |
4483 | schedstat_inc(sd, alb_failed); | |
4484 | } | |
1b12bbc7 | 4485 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4486 | } |
4487 | ||
46cb4b7c SS |
4488 | #ifdef CONFIG_NO_HZ |
4489 | static struct { | |
4490 | atomic_t load_balancer; | |
7d1e6a9b | 4491 | cpumask_var_t cpu_mask; |
f711f609 | 4492 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4493 | } nohz ____cacheline_aligned = { |
4494 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4495 | }; |
4496 | ||
eea08f32 AB |
4497 | int get_nohz_load_balancer(void) |
4498 | { | |
4499 | return atomic_read(&nohz.load_balancer); | |
4500 | } | |
4501 | ||
f711f609 GS |
4502 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4503 | /** | |
4504 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4505 | * @cpu: The cpu whose lowest level of sched domain is to | |
4506 | * be returned. | |
4507 | * @flag: The flag to check for the lowest sched_domain | |
4508 | * for the given cpu. | |
4509 | * | |
4510 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4511 | */ | |
4512 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4513 | { | |
4514 | struct sched_domain *sd; | |
4515 | ||
4516 | for_each_domain(cpu, sd) | |
4517 | if (sd && (sd->flags & flag)) | |
4518 | break; | |
4519 | ||
4520 | return sd; | |
4521 | } | |
4522 | ||
4523 | /** | |
4524 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4525 | * @cpu: The cpu whose domains we're iterating over. | |
4526 | * @sd: variable holding the value of the power_savings_sd | |
4527 | * for cpu. | |
4528 | * @flag: The flag to filter the sched_domains to be iterated. | |
4529 | * | |
4530 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4531 | * set, starting from the lowest sched_domain to the highest. | |
4532 | */ | |
4533 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4534 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4535 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4536 | ||
4537 | /** | |
4538 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4539 | * @ilb_group: group to be checked for semi-idleness | |
4540 | * | |
4541 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4542 | * | |
4543 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4544 | * and atleast one non-idle CPU. This helper function checks if the given | |
4545 | * sched_group is semi-idle or not. | |
4546 | */ | |
4547 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4548 | { | |
4549 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4550 | sched_group_cpus(ilb_group)); | |
4551 | ||
4552 | /* | |
4553 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4554 | * and atleast one idle cpu. | |
4555 | */ | |
4556 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4557 | return 0; | |
4558 | ||
4559 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4560 | return 0; | |
4561 | ||
4562 | return 1; | |
4563 | } | |
4564 | /** | |
4565 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4566 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4567 | * | |
4568 | * Returns: Returns the id of the idle load balancer if it exists, | |
4569 | * Else, returns >= nr_cpu_ids. | |
4570 | * | |
4571 | * This algorithm picks the idle load balancer such that it belongs to a | |
4572 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4573 | * completely idle packages/cores just for the purpose of idle load balancing | |
4574 | * when there are other idle cpu's which are better suited for that job. | |
4575 | */ | |
4576 | static int find_new_ilb(int cpu) | |
4577 | { | |
4578 | struct sched_domain *sd; | |
4579 | struct sched_group *ilb_group; | |
4580 | ||
4581 | /* | |
4582 | * Have idle load balancer selection from semi-idle packages only | |
4583 | * when power-aware load balancing is enabled | |
4584 | */ | |
4585 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4586 | goto out_done; | |
4587 | ||
4588 | /* | |
4589 | * Optimize for the case when we have no idle CPUs or only one | |
4590 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4591 | */ | |
4592 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4593 | goto out_done; | |
4594 | ||
4595 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4596 | ilb_group = sd->groups; | |
4597 | ||
4598 | do { | |
4599 | if (is_semi_idle_group(ilb_group)) | |
4600 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4601 | ||
4602 | ilb_group = ilb_group->next; | |
4603 | ||
4604 | } while (ilb_group != sd->groups); | |
4605 | } | |
4606 | ||
4607 | out_done: | |
4608 | return cpumask_first(nohz.cpu_mask); | |
4609 | } | |
4610 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4611 | static inline int find_new_ilb(int call_cpu) | |
4612 | { | |
6e29ec57 | 4613 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4614 | } |
4615 | #endif | |
4616 | ||
7835b98b | 4617 | /* |
46cb4b7c SS |
4618 | * This routine will try to nominate the ilb (idle load balancing) |
4619 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4620 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4621 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4622 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4623 | * arrives... | |
4624 | * | |
4625 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4626 | * for idle load balancing. ilb owner will still be part of | |
4627 | * nohz.cpu_mask.. | |
7835b98b | 4628 | * |
46cb4b7c SS |
4629 | * While stopping the tick, this cpu will become the ilb owner if there |
4630 | * is no other owner. And will be the owner till that cpu becomes busy | |
4631 | * or if all cpus in the system stop their ticks at which point | |
4632 | * there is no need for ilb owner. | |
4633 | * | |
4634 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4635 | * next busy scheduler_tick() | |
4636 | */ | |
4637 | int select_nohz_load_balancer(int stop_tick) | |
4638 | { | |
4639 | int cpu = smp_processor_id(); | |
4640 | ||
4641 | if (stop_tick) { | |
46cb4b7c SS |
4642 | cpu_rq(cpu)->in_nohz_recently = 1; |
4643 | ||
483b4ee6 SS |
4644 | if (!cpu_active(cpu)) { |
4645 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4646 | return 0; | |
4647 | ||
4648 | /* | |
4649 | * If we are going offline and still the leader, | |
4650 | * give up! | |
4651 | */ | |
46cb4b7c SS |
4652 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4653 | BUG(); | |
483b4ee6 | 4654 | |
46cb4b7c SS |
4655 | return 0; |
4656 | } | |
4657 | ||
483b4ee6 SS |
4658 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4659 | ||
46cb4b7c | 4660 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4661 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4662 | if (atomic_read(&nohz.load_balancer) == cpu) |
4663 | atomic_set(&nohz.load_balancer, -1); | |
4664 | return 0; | |
4665 | } | |
4666 | ||
4667 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4668 | /* make me the ilb owner */ | |
4669 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4670 | return 1; | |
e790fb0b GS |
4671 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4672 | int new_ilb; | |
4673 | ||
4674 | if (!(sched_smt_power_savings || | |
4675 | sched_mc_power_savings)) | |
4676 | return 1; | |
4677 | /* | |
4678 | * Check to see if there is a more power-efficient | |
4679 | * ilb. | |
4680 | */ | |
4681 | new_ilb = find_new_ilb(cpu); | |
4682 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4683 | atomic_set(&nohz.load_balancer, -1); | |
4684 | resched_cpu(new_ilb); | |
4685 | return 0; | |
4686 | } | |
46cb4b7c | 4687 | return 1; |
e790fb0b | 4688 | } |
46cb4b7c | 4689 | } else { |
7d1e6a9b | 4690 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4691 | return 0; |
4692 | ||
7d1e6a9b | 4693 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4694 | |
4695 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4696 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4697 | BUG(); | |
4698 | } | |
4699 | return 0; | |
4700 | } | |
4701 | #endif | |
4702 | ||
4703 | static DEFINE_SPINLOCK(balancing); | |
4704 | ||
4705 | /* | |
7835b98b CL |
4706 | * It checks each scheduling domain to see if it is due to be balanced, |
4707 | * and initiates a balancing operation if so. | |
4708 | * | |
4709 | * Balancing parameters are set up in arch_init_sched_domains. | |
4710 | */ | |
a9957449 | 4711 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4712 | { |
46cb4b7c SS |
4713 | int balance = 1; |
4714 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4715 | unsigned long interval; |
4716 | struct sched_domain *sd; | |
46cb4b7c | 4717 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4718 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4719 | int update_next_balance = 0; |
d07355f5 | 4720 | int need_serialize; |
1da177e4 | 4721 | |
46cb4b7c | 4722 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4723 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4724 | continue; | |
4725 | ||
4726 | interval = sd->balance_interval; | |
d15bcfdb | 4727 | if (idle != CPU_IDLE) |
1da177e4 LT |
4728 | interval *= sd->busy_factor; |
4729 | ||
4730 | /* scale ms to jiffies */ | |
4731 | interval = msecs_to_jiffies(interval); | |
4732 | if (unlikely(!interval)) | |
4733 | interval = 1; | |
dd41f596 IM |
4734 | if (interval > HZ*NR_CPUS/10) |
4735 | interval = HZ*NR_CPUS/10; | |
4736 | ||
d07355f5 | 4737 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4738 | |
d07355f5 | 4739 | if (need_serialize) { |
08c183f3 CL |
4740 | if (!spin_trylock(&balancing)) |
4741 | goto out; | |
4742 | } | |
4743 | ||
c9819f45 | 4744 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4745 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4746 | /* |
4747 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4748 | * longer idle, or one of our SMT siblings is |
4749 | * not idle. | |
4750 | */ | |
d15bcfdb | 4751 | idle = CPU_NOT_IDLE; |
1da177e4 | 4752 | } |
1bd77f2d | 4753 | sd->last_balance = jiffies; |
1da177e4 | 4754 | } |
d07355f5 | 4755 | if (need_serialize) |
08c183f3 CL |
4756 | spin_unlock(&balancing); |
4757 | out: | |
f549da84 | 4758 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4759 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4760 | update_next_balance = 1; |
4761 | } | |
783609c6 SS |
4762 | |
4763 | /* | |
4764 | * Stop the load balance at this level. There is another | |
4765 | * CPU in our sched group which is doing load balancing more | |
4766 | * actively. | |
4767 | */ | |
4768 | if (!balance) | |
4769 | break; | |
1da177e4 | 4770 | } |
f549da84 SS |
4771 | |
4772 | /* | |
4773 | * next_balance will be updated only when there is a need. | |
4774 | * When the cpu is attached to null domain for ex, it will not be | |
4775 | * updated. | |
4776 | */ | |
4777 | if (likely(update_next_balance)) | |
4778 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4779 | } |
4780 | ||
4781 | /* | |
4782 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4783 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4784 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4785 | */ | |
4786 | static void run_rebalance_domains(struct softirq_action *h) | |
4787 | { | |
dd41f596 IM |
4788 | int this_cpu = smp_processor_id(); |
4789 | struct rq *this_rq = cpu_rq(this_cpu); | |
4790 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4791 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4792 | |
dd41f596 | 4793 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4794 | |
4795 | #ifdef CONFIG_NO_HZ | |
4796 | /* | |
4797 | * If this cpu is the owner for idle load balancing, then do the | |
4798 | * balancing on behalf of the other idle cpus whose ticks are | |
4799 | * stopped. | |
4800 | */ | |
dd41f596 IM |
4801 | if (this_rq->idle_at_tick && |
4802 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4803 | struct rq *rq; |
4804 | int balance_cpu; | |
4805 | ||
7d1e6a9b RR |
4806 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4807 | if (balance_cpu == this_cpu) | |
4808 | continue; | |
4809 | ||
46cb4b7c SS |
4810 | /* |
4811 | * If this cpu gets work to do, stop the load balancing | |
4812 | * work being done for other cpus. Next load | |
4813 | * balancing owner will pick it up. | |
4814 | */ | |
4815 | if (need_resched()) | |
4816 | break; | |
4817 | ||
de0cf899 | 4818 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4819 | |
4820 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4821 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4822 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4823 | } |
4824 | } | |
4825 | #endif | |
4826 | } | |
4827 | ||
8a0be9ef FW |
4828 | static inline int on_null_domain(int cpu) |
4829 | { | |
4830 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4831 | } | |
4832 | ||
46cb4b7c SS |
4833 | /* |
4834 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4835 | * | |
4836 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4837 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4838 | * if the whole system is idle. | |
4839 | */ | |
dd41f596 | 4840 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4841 | { |
46cb4b7c SS |
4842 | #ifdef CONFIG_NO_HZ |
4843 | /* | |
4844 | * If we were in the nohz mode recently and busy at the current | |
4845 | * scheduler tick, then check if we need to nominate new idle | |
4846 | * load balancer. | |
4847 | */ | |
4848 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4849 | rq->in_nohz_recently = 0; | |
4850 | ||
4851 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4852 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4853 | atomic_set(&nohz.load_balancer, -1); |
4854 | } | |
4855 | ||
4856 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4857 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4858 | |
434d53b0 | 4859 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4860 | resched_cpu(ilb); |
4861 | } | |
4862 | } | |
4863 | ||
4864 | /* | |
4865 | * If this cpu is idle and doing idle load balancing for all the | |
4866 | * cpus with ticks stopped, is it time for that to stop? | |
4867 | */ | |
4868 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4869 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4870 | resched_cpu(cpu); |
4871 | return; | |
4872 | } | |
4873 | ||
4874 | /* | |
4875 | * If this cpu is idle and the idle load balancing is done by | |
4876 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4877 | */ | |
4878 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4879 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4880 | return; |
4881 | #endif | |
8a0be9ef FW |
4882 | /* Don't need to rebalance while attached to NULL domain */ |
4883 | if (time_after_eq(jiffies, rq->next_balance) && | |
4884 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4885 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4886 | } |
dd41f596 IM |
4887 | |
4888 | #else /* CONFIG_SMP */ | |
4889 | ||
1da177e4 LT |
4890 | /* |
4891 | * on UP we do not need to balance between CPUs: | |
4892 | */ | |
70b97a7f | 4893 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4894 | { |
4895 | } | |
dd41f596 | 4896 | |
1da177e4 LT |
4897 | #endif |
4898 | ||
1da177e4 LT |
4899 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4900 | ||
4901 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4902 | ||
4903 | /* | |
c5f8d995 | 4904 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4905 | * @p in case that task is currently running. |
c5f8d995 HS |
4906 | * |
4907 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4908 | */ |
c5f8d995 HS |
4909 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4910 | { | |
4911 | u64 ns = 0; | |
4912 | ||
4913 | if (task_current(rq, p)) { | |
4914 | update_rq_clock(rq); | |
4915 | ns = rq->clock - p->se.exec_start; | |
4916 | if ((s64)ns < 0) | |
4917 | ns = 0; | |
4918 | } | |
4919 | ||
4920 | return ns; | |
4921 | } | |
4922 | ||
bb34d92f | 4923 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4924 | { |
1da177e4 | 4925 | unsigned long flags; |
41b86e9c | 4926 | struct rq *rq; |
bb34d92f | 4927 | u64 ns = 0; |
48f24c4d | 4928 | |
41b86e9c | 4929 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4930 | ns = do_task_delta_exec(p, rq); |
4931 | task_rq_unlock(rq, &flags); | |
1508487e | 4932 | |
c5f8d995 HS |
4933 | return ns; |
4934 | } | |
f06febc9 | 4935 | |
c5f8d995 HS |
4936 | /* |
4937 | * Return accounted runtime for the task. | |
4938 | * In case the task is currently running, return the runtime plus current's | |
4939 | * pending runtime that have not been accounted yet. | |
4940 | */ | |
4941 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4942 | { | |
4943 | unsigned long flags; | |
4944 | struct rq *rq; | |
4945 | u64 ns = 0; | |
4946 | ||
4947 | rq = task_rq_lock(p, &flags); | |
4948 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4949 | task_rq_unlock(rq, &flags); | |
4950 | ||
4951 | return ns; | |
4952 | } | |
48f24c4d | 4953 | |
c5f8d995 HS |
4954 | /* |
4955 | * Return sum_exec_runtime for the thread group. | |
4956 | * In case the task is currently running, return the sum plus current's | |
4957 | * pending runtime that have not been accounted yet. | |
4958 | * | |
4959 | * Note that the thread group might have other running tasks as well, | |
4960 | * so the return value not includes other pending runtime that other | |
4961 | * running tasks might have. | |
4962 | */ | |
4963 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4964 | { | |
4965 | struct task_cputime totals; | |
4966 | unsigned long flags; | |
4967 | struct rq *rq; | |
4968 | u64 ns; | |
4969 | ||
4970 | rq = task_rq_lock(p, &flags); | |
4971 | thread_group_cputime(p, &totals); | |
4972 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4973 | task_rq_unlock(rq, &flags); |
48f24c4d | 4974 | |
1da177e4 LT |
4975 | return ns; |
4976 | } | |
4977 | ||
1da177e4 LT |
4978 | /* |
4979 | * Account user cpu time to a process. | |
4980 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4981 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4982 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4983 | */ |
457533a7 MS |
4984 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4985 | cputime_t cputime_scaled) | |
1da177e4 LT |
4986 | { |
4987 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4988 | cputime64_t tmp; | |
4989 | ||
457533a7 | 4990 | /* Add user time to process. */ |
1da177e4 | 4991 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4992 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4993 | account_group_user_time(p, cputime); |
1da177e4 LT |
4994 | |
4995 | /* Add user time to cpustat. */ | |
4996 | tmp = cputime_to_cputime64(cputime); | |
4997 | if (TASK_NICE(p) > 0) | |
4998 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4999 | else | |
5000 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5001 | |
5002 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5003 | /* Account for user time used */ |
5004 | acct_update_integrals(p); | |
1da177e4 LT |
5005 | } |
5006 | ||
94886b84 LV |
5007 | /* |
5008 | * Account guest cpu time to a process. | |
5009 | * @p: the process that the cpu time gets accounted to | |
5010 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5011 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5012 | */ |
457533a7 MS |
5013 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5014 | cputime_t cputime_scaled) | |
94886b84 LV |
5015 | { |
5016 | cputime64_t tmp; | |
5017 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5018 | ||
5019 | tmp = cputime_to_cputime64(cputime); | |
5020 | ||
457533a7 | 5021 | /* Add guest time to process. */ |
94886b84 | 5022 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5023 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5024 | account_group_user_time(p, cputime); |
94886b84 LV |
5025 | p->gtime = cputime_add(p->gtime, cputime); |
5026 | ||
457533a7 | 5027 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
5028 | if (TASK_NICE(p) > 0) { |
5029 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5030 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
5031 | } else { | |
5032 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
5033 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5034 | } | |
94886b84 LV |
5035 | } |
5036 | ||
1da177e4 LT |
5037 | /* |
5038 | * Account system cpu time to a process. | |
5039 | * @p: the process that the cpu time gets accounted to | |
5040 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5041 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5042 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5043 | */ |
5044 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5045 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5046 | { |
5047 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5048 | cputime64_t tmp; |
5049 | ||
983ed7a6 | 5050 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5051 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5052 | return; |
5053 | } | |
94886b84 | 5054 | |
457533a7 | 5055 | /* Add system time to process. */ |
1da177e4 | 5056 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5057 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5058 | account_group_system_time(p, cputime); |
1da177e4 LT |
5059 | |
5060 | /* Add system time to cpustat. */ | |
5061 | tmp = cputime_to_cputime64(cputime); | |
5062 | if (hardirq_count() - hardirq_offset) | |
5063 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5064 | else if (softirq_count()) | |
5065 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5066 | else |
79741dd3 MS |
5067 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5068 | ||
ef12fefa BR |
5069 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5070 | ||
1da177e4 LT |
5071 | /* Account for system time used */ |
5072 | acct_update_integrals(p); | |
1da177e4 LT |
5073 | } |
5074 | ||
c66f08be | 5075 | /* |
1da177e4 | 5076 | * Account for involuntary wait time. |
1da177e4 | 5077 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5078 | */ |
79741dd3 | 5079 | void account_steal_time(cputime_t cputime) |
c66f08be | 5080 | { |
79741dd3 MS |
5081 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5082 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5083 | ||
5084 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5085 | } |
5086 | ||
1da177e4 | 5087 | /* |
79741dd3 MS |
5088 | * Account for idle time. |
5089 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5090 | */ |
79741dd3 | 5091 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5092 | { |
5093 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5094 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5095 | struct rq *rq = this_rq(); |
1da177e4 | 5096 | |
79741dd3 MS |
5097 | if (atomic_read(&rq->nr_iowait) > 0) |
5098 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5099 | else | |
5100 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5101 | } |
5102 | ||
79741dd3 MS |
5103 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5104 | ||
5105 | /* | |
5106 | * Account a single tick of cpu time. | |
5107 | * @p: the process that the cpu time gets accounted to | |
5108 | * @user_tick: indicates if the tick is a user or a system tick | |
5109 | */ | |
5110 | void account_process_tick(struct task_struct *p, int user_tick) | |
5111 | { | |
a42548a1 | 5112 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
5113 | struct rq *rq = this_rq(); |
5114 | ||
5115 | if (user_tick) | |
a42548a1 | 5116 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 5117 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 5118 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
5119 | one_jiffy_scaled); |
5120 | else | |
a42548a1 | 5121 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
5122 | } |
5123 | ||
5124 | /* | |
5125 | * Account multiple ticks of steal time. | |
5126 | * @p: the process from which the cpu time has been stolen | |
5127 | * @ticks: number of stolen ticks | |
5128 | */ | |
5129 | void account_steal_ticks(unsigned long ticks) | |
5130 | { | |
5131 | account_steal_time(jiffies_to_cputime(ticks)); | |
5132 | } | |
5133 | ||
5134 | /* | |
5135 | * Account multiple ticks of idle time. | |
5136 | * @ticks: number of stolen ticks | |
5137 | */ | |
5138 | void account_idle_ticks(unsigned long ticks) | |
5139 | { | |
5140 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5141 | } |
5142 | ||
79741dd3 MS |
5143 | #endif |
5144 | ||
49048622 BS |
5145 | /* |
5146 | * Use precise platform statistics if available: | |
5147 | */ | |
5148 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5149 | cputime_t task_utime(struct task_struct *p) | |
5150 | { | |
5151 | return p->utime; | |
5152 | } | |
5153 | ||
5154 | cputime_t task_stime(struct task_struct *p) | |
5155 | { | |
5156 | return p->stime; | |
5157 | } | |
5158 | #else | |
761b1d26 HS |
5159 | |
5160 | #ifndef nsecs_to_cputime | |
5161 | # define nsecs_to_cputime(__nsecs) \ | |
5162 | msecs_to_cputime(div_u64((__nsecs), NSEC_PER_MSEC)) | |
5163 | #endif | |
5164 | ||
49048622 BS |
5165 | cputime_t task_utime(struct task_struct *p) |
5166 | { | |
761b1d26 | 5167 | cputime_t utime = p->utime, total = utime + p->stime; |
49048622 BS |
5168 | u64 temp; |
5169 | ||
5170 | /* | |
5171 | * Use CFS's precise accounting: | |
5172 | */ | |
761b1d26 | 5173 | temp = (u64)nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
5174 | |
5175 | if (total) { | |
5176 | temp *= utime; | |
5177 | do_div(temp, total); | |
5178 | } | |
761b1d26 | 5179 | utime = (cputime_t)temp; |
49048622 | 5180 | |
761b1d26 | 5181 | p->prev_utime = max(p->prev_utime, utime); |
49048622 BS |
5182 | return p->prev_utime; |
5183 | } | |
5184 | ||
5185 | cputime_t task_stime(struct task_struct *p) | |
5186 | { | |
761b1d26 | 5187 | cputime_t stime; |
49048622 BS |
5188 | |
5189 | /* | |
5190 | * Use CFS's precise accounting. (we subtract utime from | |
5191 | * the total, to make sure the total observed by userspace | |
5192 | * grows monotonically - apps rely on that): | |
5193 | */ | |
761b1d26 | 5194 | stime = nsecs_to_cputime(p->se.sum_exec_runtime) - task_utime(p); |
49048622 BS |
5195 | |
5196 | if (stime >= 0) | |
761b1d26 | 5197 | p->prev_stime = max(p->prev_stime, stime); |
49048622 BS |
5198 | |
5199 | return p->prev_stime; | |
5200 | } | |
5201 | #endif | |
5202 | ||
5203 | inline cputime_t task_gtime(struct task_struct *p) | |
5204 | { | |
5205 | return p->gtime; | |
5206 | } | |
5207 | ||
7835b98b CL |
5208 | /* |
5209 | * This function gets called by the timer code, with HZ frequency. | |
5210 | * We call it with interrupts disabled. | |
5211 | * | |
5212 | * It also gets called by the fork code, when changing the parent's | |
5213 | * timeslices. | |
5214 | */ | |
5215 | void scheduler_tick(void) | |
5216 | { | |
7835b98b CL |
5217 | int cpu = smp_processor_id(); |
5218 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5219 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5220 | |
5221 | sched_clock_tick(); | |
dd41f596 IM |
5222 | |
5223 | spin_lock(&rq->lock); | |
3e51f33f | 5224 | update_rq_clock(rq); |
f1a438d8 | 5225 | update_cpu_load(rq); |
fa85ae24 | 5226 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5227 | spin_unlock(&rq->lock); |
7835b98b | 5228 | |
cdd6c482 | 5229 | perf_event_task_tick(curr, cpu); |
e220d2dc | 5230 | |
e418e1c2 | 5231 | #ifdef CONFIG_SMP |
dd41f596 IM |
5232 | rq->idle_at_tick = idle_cpu(cpu); |
5233 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5234 | #endif |
1da177e4 LT |
5235 | } |
5236 | ||
132380a0 | 5237 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5238 | { |
5239 | if (in_lock_functions(addr)) { | |
5240 | addr = CALLER_ADDR2; | |
5241 | if (in_lock_functions(addr)) | |
5242 | addr = CALLER_ADDR3; | |
5243 | } | |
5244 | return addr; | |
5245 | } | |
1da177e4 | 5246 | |
7e49fcce SR |
5247 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5248 | defined(CONFIG_PREEMPT_TRACER)) | |
5249 | ||
43627582 | 5250 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5251 | { |
6cd8a4bb | 5252 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5253 | /* |
5254 | * Underflow? | |
5255 | */ | |
9a11b49a IM |
5256 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5257 | return; | |
6cd8a4bb | 5258 | #endif |
1da177e4 | 5259 | preempt_count() += val; |
6cd8a4bb | 5260 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5261 | /* |
5262 | * Spinlock count overflowing soon? | |
5263 | */ | |
33859f7f MOS |
5264 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5265 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5266 | #endif |
5267 | if (preempt_count() == val) | |
5268 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5269 | } |
5270 | EXPORT_SYMBOL(add_preempt_count); | |
5271 | ||
43627582 | 5272 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5273 | { |
6cd8a4bb | 5274 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5275 | /* |
5276 | * Underflow? | |
5277 | */ | |
01e3eb82 | 5278 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5279 | return; |
1da177e4 LT |
5280 | /* |
5281 | * Is the spinlock portion underflowing? | |
5282 | */ | |
9a11b49a IM |
5283 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5284 | !(preempt_count() & PREEMPT_MASK))) | |
5285 | return; | |
6cd8a4bb | 5286 | #endif |
9a11b49a | 5287 | |
6cd8a4bb SR |
5288 | if (preempt_count() == val) |
5289 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5290 | preempt_count() -= val; |
5291 | } | |
5292 | EXPORT_SYMBOL(sub_preempt_count); | |
5293 | ||
5294 | #endif | |
5295 | ||
5296 | /* | |
dd41f596 | 5297 | * Print scheduling while atomic bug: |
1da177e4 | 5298 | */ |
dd41f596 | 5299 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5300 | { |
838225b4 SS |
5301 | struct pt_regs *regs = get_irq_regs(); |
5302 | ||
5303 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5304 | prev->comm, prev->pid, preempt_count()); | |
5305 | ||
dd41f596 | 5306 | debug_show_held_locks(prev); |
e21f5b15 | 5307 | print_modules(); |
dd41f596 IM |
5308 | if (irqs_disabled()) |
5309 | print_irqtrace_events(prev); | |
838225b4 SS |
5310 | |
5311 | if (regs) | |
5312 | show_regs(regs); | |
5313 | else | |
5314 | dump_stack(); | |
dd41f596 | 5315 | } |
1da177e4 | 5316 | |
dd41f596 IM |
5317 | /* |
5318 | * Various schedule()-time debugging checks and statistics: | |
5319 | */ | |
5320 | static inline void schedule_debug(struct task_struct *prev) | |
5321 | { | |
1da177e4 | 5322 | /* |
41a2d6cf | 5323 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5324 | * schedule() atomically, we ignore that path for now. |
5325 | * Otherwise, whine if we are scheduling when we should not be. | |
5326 | */ | |
3f33a7ce | 5327 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5328 | __schedule_bug(prev); |
5329 | ||
1da177e4 LT |
5330 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5331 | ||
2d72376b | 5332 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5333 | #ifdef CONFIG_SCHEDSTATS |
5334 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5335 | schedstat_inc(this_rq(), bkl_count); |
5336 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5337 | } |
5338 | #endif | |
dd41f596 IM |
5339 | } |
5340 | ||
ad4b78bb | 5341 | static void put_prev_task(struct rq *rq, struct task_struct *p) |
df1c99d4 | 5342 | { |
ad4b78bb | 5343 | u64 runtime = p->se.sum_exec_runtime - p->se.prev_sum_exec_runtime; |
df1c99d4 | 5344 | |
ad4b78bb | 5345 | update_avg(&p->se.avg_running, runtime); |
df1c99d4 | 5346 | |
ad4b78bb | 5347 | if (p->state == TASK_RUNNING) { |
df1c99d4 MG |
5348 | /* |
5349 | * In order to avoid avg_overlap growing stale when we are | |
5350 | * indeed overlapping and hence not getting put to sleep, grow | |
5351 | * the avg_overlap on preemption. | |
5352 | * | |
5353 | * We use the average preemption runtime because that | |
5354 | * correlates to the amount of cache footprint a task can | |
5355 | * build up. | |
5356 | */ | |
ad4b78bb PZ |
5357 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); |
5358 | update_avg(&p->se.avg_overlap, runtime); | |
5359 | } else { | |
5360 | update_avg(&p->se.avg_running, 0); | |
df1c99d4 | 5361 | } |
ad4b78bb | 5362 | p->sched_class->put_prev_task(rq, p); |
df1c99d4 MG |
5363 | } |
5364 | ||
dd41f596 IM |
5365 | /* |
5366 | * Pick up the highest-prio task: | |
5367 | */ | |
5368 | static inline struct task_struct * | |
b67802ea | 5369 | pick_next_task(struct rq *rq) |
dd41f596 | 5370 | { |
5522d5d5 | 5371 | const struct sched_class *class; |
dd41f596 | 5372 | struct task_struct *p; |
1da177e4 LT |
5373 | |
5374 | /* | |
dd41f596 IM |
5375 | * Optimization: we know that if all tasks are in |
5376 | * the fair class we can call that function directly: | |
1da177e4 | 5377 | */ |
dd41f596 | 5378 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5379 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5380 | if (likely(p)) |
5381 | return p; | |
1da177e4 LT |
5382 | } |
5383 | ||
dd41f596 IM |
5384 | class = sched_class_highest; |
5385 | for ( ; ; ) { | |
fb8d4724 | 5386 | p = class->pick_next_task(rq); |
dd41f596 IM |
5387 | if (p) |
5388 | return p; | |
5389 | /* | |
5390 | * Will never be NULL as the idle class always | |
5391 | * returns a non-NULL p: | |
5392 | */ | |
5393 | class = class->next; | |
5394 | } | |
5395 | } | |
1da177e4 | 5396 | |
dd41f596 IM |
5397 | /* |
5398 | * schedule() is the main scheduler function. | |
5399 | */ | |
ff743345 | 5400 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5401 | { |
5402 | struct task_struct *prev, *next; | |
67ca7bde | 5403 | unsigned long *switch_count; |
dd41f596 | 5404 | struct rq *rq; |
31656519 | 5405 | int cpu; |
dd41f596 | 5406 | |
ff743345 PZ |
5407 | need_resched: |
5408 | preempt_disable(); | |
dd41f596 IM |
5409 | cpu = smp_processor_id(); |
5410 | rq = cpu_rq(cpu); | |
d6714c22 | 5411 | rcu_sched_qs(cpu); |
dd41f596 IM |
5412 | prev = rq->curr; |
5413 | switch_count = &prev->nivcsw; | |
5414 | ||
5415 | release_kernel_lock(prev); | |
5416 | need_resched_nonpreemptible: | |
5417 | ||
5418 | schedule_debug(prev); | |
1da177e4 | 5419 | |
31656519 | 5420 | if (sched_feat(HRTICK)) |
f333fdc9 | 5421 | hrtick_clear(rq); |
8f4d37ec | 5422 | |
8cd162ce | 5423 | spin_lock_irq(&rq->lock); |
3e51f33f | 5424 | update_rq_clock(rq); |
1e819950 | 5425 | clear_tsk_need_resched(prev); |
1da177e4 | 5426 | |
1da177e4 | 5427 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5428 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5429 | prev->state = TASK_RUNNING; |
16882c1e | 5430 | else |
2e1cb74a | 5431 | deactivate_task(rq, prev, 1); |
dd41f596 | 5432 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5433 | } |
5434 | ||
3f029d3c | 5435 | pre_schedule(rq, prev); |
f65eda4f | 5436 | |
dd41f596 | 5437 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5438 | idle_balance(cpu, rq); |
1da177e4 | 5439 | |
df1c99d4 | 5440 | put_prev_task(rq, prev); |
b67802ea | 5441 | next = pick_next_task(rq); |
1da177e4 | 5442 | |
1da177e4 | 5443 | if (likely(prev != next)) { |
673a90a1 | 5444 | sched_info_switch(prev, next); |
cdd6c482 | 5445 | perf_event_task_sched_out(prev, next, cpu); |
673a90a1 | 5446 | |
1da177e4 LT |
5447 | rq->nr_switches++; |
5448 | rq->curr = next; | |
5449 | ++*switch_count; | |
5450 | ||
dd41f596 | 5451 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5452 | /* |
5453 | * the context switch might have flipped the stack from under | |
5454 | * us, hence refresh the local variables. | |
5455 | */ | |
5456 | cpu = smp_processor_id(); | |
5457 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5458 | } else |
5459 | spin_unlock_irq(&rq->lock); | |
5460 | ||
3f029d3c | 5461 | post_schedule(rq); |
1da177e4 | 5462 | |
8f4d37ec | 5463 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5464 | goto need_resched_nonpreemptible; |
8f4d37ec | 5465 | |
1da177e4 | 5466 | preempt_enable_no_resched(); |
ff743345 | 5467 | if (need_resched()) |
1da177e4 LT |
5468 | goto need_resched; |
5469 | } | |
1da177e4 LT |
5470 | EXPORT_SYMBOL(schedule); |
5471 | ||
0d66bf6d PZ |
5472 | #ifdef CONFIG_SMP |
5473 | /* | |
5474 | * Look out! "owner" is an entirely speculative pointer | |
5475 | * access and not reliable. | |
5476 | */ | |
5477 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5478 | { | |
5479 | unsigned int cpu; | |
5480 | struct rq *rq; | |
5481 | ||
5482 | if (!sched_feat(OWNER_SPIN)) | |
5483 | return 0; | |
5484 | ||
5485 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5486 | /* | |
5487 | * Need to access the cpu field knowing that | |
5488 | * DEBUG_PAGEALLOC could have unmapped it if | |
5489 | * the mutex owner just released it and exited. | |
5490 | */ | |
5491 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5492 | goto out; | |
5493 | #else | |
5494 | cpu = owner->cpu; | |
5495 | #endif | |
5496 | ||
5497 | /* | |
5498 | * Even if the access succeeded (likely case), | |
5499 | * the cpu field may no longer be valid. | |
5500 | */ | |
5501 | if (cpu >= nr_cpumask_bits) | |
5502 | goto out; | |
5503 | ||
5504 | /* | |
5505 | * We need to validate that we can do a | |
5506 | * get_cpu() and that we have the percpu area. | |
5507 | */ | |
5508 | if (!cpu_online(cpu)) | |
5509 | goto out; | |
5510 | ||
5511 | rq = cpu_rq(cpu); | |
5512 | ||
5513 | for (;;) { | |
5514 | /* | |
5515 | * Owner changed, break to re-assess state. | |
5516 | */ | |
5517 | if (lock->owner != owner) | |
5518 | break; | |
5519 | ||
5520 | /* | |
5521 | * Is that owner really running on that cpu? | |
5522 | */ | |
5523 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5524 | return 0; | |
5525 | ||
5526 | cpu_relax(); | |
5527 | } | |
5528 | out: | |
5529 | return 1; | |
5530 | } | |
5531 | #endif | |
5532 | ||
1da177e4 LT |
5533 | #ifdef CONFIG_PREEMPT |
5534 | /* | |
2ed6e34f | 5535 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5536 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5537 | * occur there and call schedule directly. |
5538 | */ | |
5539 | asmlinkage void __sched preempt_schedule(void) | |
5540 | { | |
5541 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5542 | |
1da177e4 LT |
5543 | /* |
5544 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5545 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5546 | */ |
beed33a8 | 5547 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5548 | return; |
5549 | ||
3a5c359a AK |
5550 | do { |
5551 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5552 | schedule(); |
3a5c359a | 5553 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5554 | |
3a5c359a AK |
5555 | /* |
5556 | * Check again in case we missed a preemption opportunity | |
5557 | * between schedule and now. | |
5558 | */ | |
5559 | barrier(); | |
5ed0cec0 | 5560 | } while (need_resched()); |
1da177e4 | 5561 | } |
1da177e4 LT |
5562 | EXPORT_SYMBOL(preempt_schedule); |
5563 | ||
5564 | /* | |
2ed6e34f | 5565 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5566 | * off of irq context. |
5567 | * Note, that this is called and return with irqs disabled. This will | |
5568 | * protect us against recursive calling from irq. | |
5569 | */ | |
5570 | asmlinkage void __sched preempt_schedule_irq(void) | |
5571 | { | |
5572 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5573 | |
2ed6e34f | 5574 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5575 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5576 | ||
3a5c359a AK |
5577 | do { |
5578 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5579 | local_irq_enable(); |
5580 | schedule(); | |
5581 | local_irq_disable(); | |
3a5c359a | 5582 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5583 | |
3a5c359a AK |
5584 | /* |
5585 | * Check again in case we missed a preemption opportunity | |
5586 | * between schedule and now. | |
5587 | */ | |
5588 | barrier(); | |
5ed0cec0 | 5589 | } while (need_resched()); |
1da177e4 LT |
5590 | } |
5591 | ||
5592 | #endif /* CONFIG_PREEMPT */ | |
5593 | ||
63859d4f | 5594 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5595 | void *key) |
1da177e4 | 5596 | { |
63859d4f | 5597 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5598 | } |
1da177e4 LT |
5599 | EXPORT_SYMBOL(default_wake_function); |
5600 | ||
5601 | /* | |
41a2d6cf IM |
5602 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5603 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5604 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5605 | * | |
5606 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5607 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5608 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5609 | */ | |
78ddb08f | 5610 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 5611 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 5612 | { |
2e45874c | 5613 | wait_queue_t *curr, *next; |
1da177e4 | 5614 | |
2e45874c | 5615 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5616 | unsigned flags = curr->flags; |
5617 | ||
63859d4f | 5618 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 5619 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5620 | break; |
5621 | } | |
5622 | } | |
5623 | ||
5624 | /** | |
5625 | * __wake_up - wake up threads blocked on a waitqueue. | |
5626 | * @q: the waitqueue | |
5627 | * @mode: which threads | |
5628 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5629 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5630 | * |
5631 | * It may be assumed that this function implies a write memory barrier before | |
5632 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5633 | */ |
7ad5b3a5 | 5634 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5635 | int nr_exclusive, void *key) |
1da177e4 LT |
5636 | { |
5637 | unsigned long flags; | |
5638 | ||
5639 | spin_lock_irqsave(&q->lock, flags); | |
5640 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5641 | spin_unlock_irqrestore(&q->lock, flags); | |
5642 | } | |
1da177e4 LT |
5643 | EXPORT_SYMBOL(__wake_up); |
5644 | ||
5645 | /* | |
5646 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5647 | */ | |
7ad5b3a5 | 5648 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5649 | { |
5650 | __wake_up_common(q, mode, 1, 0, NULL); | |
5651 | } | |
5652 | ||
4ede816a DL |
5653 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5654 | { | |
5655 | __wake_up_common(q, mode, 1, 0, key); | |
5656 | } | |
5657 | ||
1da177e4 | 5658 | /** |
4ede816a | 5659 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5660 | * @q: the waitqueue |
5661 | * @mode: which threads | |
5662 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5663 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5664 | * |
5665 | * The sync wakeup differs that the waker knows that it will schedule | |
5666 | * away soon, so while the target thread will be woken up, it will not | |
5667 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5668 | * with each other. This can prevent needless bouncing between CPUs. | |
5669 | * | |
5670 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5671 | * |
5672 | * It may be assumed that this function implies a write memory barrier before | |
5673 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5674 | */ |
4ede816a DL |
5675 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5676 | int nr_exclusive, void *key) | |
1da177e4 LT |
5677 | { |
5678 | unsigned long flags; | |
7d478721 | 5679 | int wake_flags = WF_SYNC; |
1da177e4 LT |
5680 | |
5681 | if (unlikely(!q)) | |
5682 | return; | |
5683 | ||
5684 | if (unlikely(!nr_exclusive)) | |
7d478721 | 5685 | wake_flags = 0; |
1da177e4 LT |
5686 | |
5687 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 5688 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
5689 | spin_unlock_irqrestore(&q->lock, flags); |
5690 | } | |
4ede816a DL |
5691 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5692 | ||
5693 | /* | |
5694 | * __wake_up_sync - see __wake_up_sync_key() | |
5695 | */ | |
5696 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5697 | { | |
5698 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5699 | } | |
1da177e4 LT |
5700 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5701 | ||
65eb3dc6 KD |
5702 | /** |
5703 | * complete: - signals a single thread waiting on this completion | |
5704 | * @x: holds the state of this particular completion | |
5705 | * | |
5706 | * This will wake up a single thread waiting on this completion. Threads will be | |
5707 | * awakened in the same order in which they were queued. | |
5708 | * | |
5709 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5710 | * |
5711 | * It may be assumed that this function implies a write memory barrier before | |
5712 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5713 | */ |
b15136e9 | 5714 | void complete(struct completion *x) |
1da177e4 LT |
5715 | { |
5716 | unsigned long flags; | |
5717 | ||
5718 | spin_lock_irqsave(&x->wait.lock, flags); | |
5719 | x->done++; | |
d9514f6c | 5720 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5721 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5722 | } | |
5723 | EXPORT_SYMBOL(complete); | |
5724 | ||
65eb3dc6 KD |
5725 | /** |
5726 | * complete_all: - signals all threads waiting on this completion | |
5727 | * @x: holds the state of this particular completion | |
5728 | * | |
5729 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5730 | * |
5731 | * It may be assumed that this function implies a write memory barrier before | |
5732 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5733 | */ |
b15136e9 | 5734 | void complete_all(struct completion *x) |
1da177e4 LT |
5735 | { |
5736 | unsigned long flags; | |
5737 | ||
5738 | spin_lock_irqsave(&x->wait.lock, flags); | |
5739 | x->done += UINT_MAX/2; | |
d9514f6c | 5740 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5741 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5742 | } | |
5743 | EXPORT_SYMBOL(complete_all); | |
5744 | ||
8cbbe86d AK |
5745 | static inline long __sched |
5746 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5747 | { |
1da177e4 LT |
5748 | if (!x->done) { |
5749 | DECLARE_WAITQUEUE(wait, current); | |
5750 | ||
5751 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5752 | __add_wait_queue_tail(&x->wait, &wait); | |
5753 | do { | |
94d3d824 | 5754 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5755 | timeout = -ERESTARTSYS; |
5756 | break; | |
8cbbe86d AK |
5757 | } |
5758 | __set_current_state(state); | |
1da177e4 LT |
5759 | spin_unlock_irq(&x->wait.lock); |
5760 | timeout = schedule_timeout(timeout); | |
5761 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5762 | } while (!x->done && timeout); |
1da177e4 | 5763 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5764 | if (!x->done) |
5765 | return timeout; | |
1da177e4 LT |
5766 | } |
5767 | x->done--; | |
ea71a546 | 5768 | return timeout ?: 1; |
1da177e4 | 5769 | } |
1da177e4 | 5770 | |
8cbbe86d AK |
5771 | static long __sched |
5772 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5773 | { |
1da177e4 LT |
5774 | might_sleep(); |
5775 | ||
5776 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5777 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5778 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5779 | return timeout; |
5780 | } | |
1da177e4 | 5781 | |
65eb3dc6 KD |
5782 | /** |
5783 | * wait_for_completion: - waits for completion of a task | |
5784 | * @x: holds the state of this particular completion | |
5785 | * | |
5786 | * This waits to be signaled for completion of a specific task. It is NOT | |
5787 | * interruptible and there is no timeout. | |
5788 | * | |
5789 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5790 | * and interrupt capability. Also see complete(). | |
5791 | */ | |
b15136e9 | 5792 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5793 | { |
5794 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5795 | } |
8cbbe86d | 5796 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5797 | |
65eb3dc6 KD |
5798 | /** |
5799 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5800 | * @x: holds the state of this particular completion | |
5801 | * @timeout: timeout value in jiffies | |
5802 | * | |
5803 | * This waits for either a completion of a specific task to be signaled or for a | |
5804 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5805 | * interruptible. | |
5806 | */ | |
b15136e9 | 5807 | unsigned long __sched |
8cbbe86d | 5808 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5809 | { |
8cbbe86d | 5810 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5811 | } |
8cbbe86d | 5812 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5813 | |
65eb3dc6 KD |
5814 | /** |
5815 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5816 | * @x: holds the state of this particular completion | |
5817 | * | |
5818 | * This waits for completion of a specific task to be signaled. It is | |
5819 | * interruptible. | |
5820 | */ | |
8cbbe86d | 5821 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5822 | { |
51e97990 AK |
5823 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5824 | if (t == -ERESTARTSYS) | |
5825 | return t; | |
5826 | return 0; | |
0fec171c | 5827 | } |
8cbbe86d | 5828 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5829 | |
65eb3dc6 KD |
5830 | /** |
5831 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5832 | * @x: holds the state of this particular completion | |
5833 | * @timeout: timeout value in jiffies | |
5834 | * | |
5835 | * This waits for either a completion of a specific task to be signaled or for a | |
5836 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5837 | */ | |
b15136e9 | 5838 | unsigned long __sched |
8cbbe86d AK |
5839 | wait_for_completion_interruptible_timeout(struct completion *x, |
5840 | unsigned long timeout) | |
0fec171c | 5841 | { |
8cbbe86d | 5842 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5843 | } |
8cbbe86d | 5844 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5845 | |
65eb3dc6 KD |
5846 | /** |
5847 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5848 | * @x: holds the state of this particular completion | |
5849 | * | |
5850 | * This waits to be signaled for completion of a specific task. It can be | |
5851 | * interrupted by a kill signal. | |
5852 | */ | |
009e577e MW |
5853 | int __sched wait_for_completion_killable(struct completion *x) |
5854 | { | |
5855 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5856 | if (t == -ERESTARTSYS) | |
5857 | return t; | |
5858 | return 0; | |
5859 | } | |
5860 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5861 | ||
be4de352 DC |
5862 | /** |
5863 | * try_wait_for_completion - try to decrement a completion without blocking | |
5864 | * @x: completion structure | |
5865 | * | |
5866 | * Returns: 0 if a decrement cannot be done without blocking | |
5867 | * 1 if a decrement succeeded. | |
5868 | * | |
5869 | * If a completion is being used as a counting completion, | |
5870 | * attempt to decrement the counter without blocking. This | |
5871 | * enables us to avoid waiting if the resource the completion | |
5872 | * is protecting is not available. | |
5873 | */ | |
5874 | bool try_wait_for_completion(struct completion *x) | |
5875 | { | |
5876 | int ret = 1; | |
5877 | ||
5878 | spin_lock_irq(&x->wait.lock); | |
5879 | if (!x->done) | |
5880 | ret = 0; | |
5881 | else | |
5882 | x->done--; | |
5883 | spin_unlock_irq(&x->wait.lock); | |
5884 | return ret; | |
5885 | } | |
5886 | EXPORT_SYMBOL(try_wait_for_completion); | |
5887 | ||
5888 | /** | |
5889 | * completion_done - Test to see if a completion has any waiters | |
5890 | * @x: completion structure | |
5891 | * | |
5892 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5893 | * 1 if there are no waiters. | |
5894 | * | |
5895 | */ | |
5896 | bool completion_done(struct completion *x) | |
5897 | { | |
5898 | int ret = 1; | |
5899 | ||
5900 | spin_lock_irq(&x->wait.lock); | |
5901 | if (!x->done) | |
5902 | ret = 0; | |
5903 | spin_unlock_irq(&x->wait.lock); | |
5904 | return ret; | |
5905 | } | |
5906 | EXPORT_SYMBOL(completion_done); | |
5907 | ||
8cbbe86d AK |
5908 | static long __sched |
5909 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5910 | { |
0fec171c IM |
5911 | unsigned long flags; |
5912 | wait_queue_t wait; | |
5913 | ||
5914 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5915 | |
8cbbe86d | 5916 | __set_current_state(state); |
1da177e4 | 5917 | |
8cbbe86d AK |
5918 | spin_lock_irqsave(&q->lock, flags); |
5919 | __add_wait_queue(q, &wait); | |
5920 | spin_unlock(&q->lock); | |
5921 | timeout = schedule_timeout(timeout); | |
5922 | spin_lock_irq(&q->lock); | |
5923 | __remove_wait_queue(q, &wait); | |
5924 | spin_unlock_irqrestore(&q->lock, flags); | |
5925 | ||
5926 | return timeout; | |
5927 | } | |
5928 | ||
5929 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5930 | { | |
5931 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5932 | } |
1da177e4 LT |
5933 | EXPORT_SYMBOL(interruptible_sleep_on); |
5934 | ||
0fec171c | 5935 | long __sched |
95cdf3b7 | 5936 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5937 | { |
8cbbe86d | 5938 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5939 | } |
1da177e4 LT |
5940 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5941 | ||
0fec171c | 5942 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5943 | { |
8cbbe86d | 5944 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5945 | } |
1da177e4 LT |
5946 | EXPORT_SYMBOL(sleep_on); |
5947 | ||
0fec171c | 5948 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5949 | { |
8cbbe86d | 5950 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5951 | } |
1da177e4 LT |
5952 | EXPORT_SYMBOL(sleep_on_timeout); |
5953 | ||
b29739f9 IM |
5954 | #ifdef CONFIG_RT_MUTEXES |
5955 | ||
5956 | /* | |
5957 | * rt_mutex_setprio - set the current priority of a task | |
5958 | * @p: task | |
5959 | * @prio: prio value (kernel-internal form) | |
5960 | * | |
5961 | * This function changes the 'effective' priority of a task. It does | |
5962 | * not touch ->normal_prio like __setscheduler(). | |
5963 | * | |
5964 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5965 | */ | |
36c8b586 | 5966 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5967 | { |
5968 | unsigned long flags; | |
83b699ed | 5969 | int oldprio, on_rq, running; |
70b97a7f | 5970 | struct rq *rq; |
cb469845 | 5971 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5972 | |
5973 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5974 | ||
5975 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5976 | update_rq_clock(rq); |
b29739f9 | 5977 | |
d5f9f942 | 5978 | oldprio = p->prio; |
dd41f596 | 5979 | on_rq = p->se.on_rq; |
051a1d1a | 5980 | running = task_current(rq, p); |
0e1f3483 | 5981 | if (on_rq) |
69be72c1 | 5982 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5983 | if (running) |
5984 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5985 | |
5986 | if (rt_prio(prio)) | |
5987 | p->sched_class = &rt_sched_class; | |
5988 | else | |
5989 | p->sched_class = &fair_sched_class; | |
5990 | ||
b29739f9 IM |
5991 | p->prio = prio; |
5992 | ||
0e1f3483 HS |
5993 | if (running) |
5994 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5995 | if (on_rq) { |
8159f87e | 5996 | enqueue_task(rq, p, 0); |
cb469845 SR |
5997 | |
5998 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5999 | } |
6000 | task_rq_unlock(rq, &flags); | |
6001 | } | |
6002 | ||
6003 | #endif | |
6004 | ||
36c8b586 | 6005 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6006 | { |
dd41f596 | 6007 | int old_prio, delta, on_rq; |
1da177e4 | 6008 | unsigned long flags; |
70b97a7f | 6009 | struct rq *rq; |
1da177e4 LT |
6010 | |
6011 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6012 | return; | |
6013 | /* | |
6014 | * We have to be careful, if called from sys_setpriority(), | |
6015 | * the task might be in the middle of scheduling on another CPU. | |
6016 | */ | |
6017 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6018 | update_rq_clock(rq); |
1da177e4 LT |
6019 | /* |
6020 | * The RT priorities are set via sched_setscheduler(), but we still | |
6021 | * allow the 'normal' nice value to be set - but as expected | |
6022 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6023 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6024 | */ |
e05606d3 | 6025 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6026 | p->static_prio = NICE_TO_PRIO(nice); |
6027 | goto out_unlock; | |
6028 | } | |
dd41f596 | 6029 | on_rq = p->se.on_rq; |
c09595f6 | 6030 | if (on_rq) |
69be72c1 | 6031 | dequeue_task(rq, p, 0); |
1da177e4 | 6032 | |
1da177e4 | 6033 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6034 | set_load_weight(p); |
b29739f9 IM |
6035 | old_prio = p->prio; |
6036 | p->prio = effective_prio(p); | |
6037 | delta = p->prio - old_prio; | |
1da177e4 | 6038 | |
dd41f596 | 6039 | if (on_rq) { |
8159f87e | 6040 | enqueue_task(rq, p, 0); |
1da177e4 | 6041 | /* |
d5f9f942 AM |
6042 | * If the task increased its priority or is running and |
6043 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6044 | */ |
d5f9f942 | 6045 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6046 | resched_task(rq->curr); |
6047 | } | |
6048 | out_unlock: | |
6049 | task_rq_unlock(rq, &flags); | |
6050 | } | |
1da177e4 LT |
6051 | EXPORT_SYMBOL(set_user_nice); |
6052 | ||
e43379f1 MM |
6053 | /* |
6054 | * can_nice - check if a task can reduce its nice value | |
6055 | * @p: task | |
6056 | * @nice: nice value | |
6057 | */ | |
36c8b586 | 6058 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6059 | { |
024f4747 MM |
6060 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6061 | int nice_rlim = 20 - nice; | |
48f24c4d | 6062 | |
e43379f1 MM |
6063 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6064 | capable(CAP_SYS_NICE)); | |
6065 | } | |
6066 | ||
1da177e4 LT |
6067 | #ifdef __ARCH_WANT_SYS_NICE |
6068 | ||
6069 | /* | |
6070 | * sys_nice - change the priority of the current process. | |
6071 | * @increment: priority increment | |
6072 | * | |
6073 | * sys_setpriority is a more generic, but much slower function that | |
6074 | * does similar things. | |
6075 | */ | |
5add95d4 | 6076 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6077 | { |
48f24c4d | 6078 | long nice, retval; |
1da177e4 LT |
6079 | |
6080 | /* | |
6081 | * Setpriority might change our priority at the same moment. | |
6082 | * We don't have to worry. Conceptually one call occurs first | |
6083 | * and we have a single winner. | |
6084 | */ | |
e43379f1 MM |
6085 | if (increment < -40) |
6086 | increment = -40; | |
1da177e4 LT |
6087 | if (increment > 40) |
6088 | increment = 40; | |
6089 | ||
2b8f836f | 6090 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6091 | if (nice < -20) |
6092 | nice = -20; | |
6093 | if (nice > 19) | |
6094 | nice = 19; | |
6095 | ||
e43379f1 MM |
6096 | if (increment < 0 && !can_nice(current, nice)) |
6097 | return -EPERM; | |
6098 | ||
1da177e4 LT |
6099 | retval = security_task_setnice(current, nice); |
6100 | if (retval) | |
6101 | return retval; | |
6102 | ||
6103 | set_user_nice(current, nice); | |
6104 | return 0; | |
6105 | } | |
6106 | ||
6107 | #endif | |
6108 | ||
6109 | /** | |
6110 | * task_prio - return the priority value of a given task. | |
6111 | * @p: the task in question. | |
6112 | * | |
6113 | * This is the priority value as seen by users in /proc. | |
6114 | * RT tasks are offset by -200. Normal tasks are centered | |
6115 | * around 0, value goes from -16 to +15. | |
6116 | */ | |
36c8b586 | 6117 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6118 | { |
6119 | return p->prio - MAX_RT_PRIO; | |
6120 | } | |
6121 | ||
6122 | /** | |
6123 | * task_nice - return the nice value of a given task. | |
6124 | * @p: the task in question. | |
6125 | */ | |
36c8b586 | 6126 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6127 | { |
6128 | return TASK_NICE(p); | |
6129 | } | |
150d8bed | 6130 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6131 | |
6132 | /** | |
6133 | * idle_cpu - is a given cpu idle currently? | |
6134 | * @cpu: the processor in question. | |
6135 | */ | |
6136 | int idle_cpu(int cpu) | |
6137 | { | |
6138 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6139 | } | |
6140 | ||
1da177e4 LT |
6141 | /** |
6142 | * idle_task - return the idle task for a given cpu. | |
6143 | * @cpu: the processor in question. | |
6144 | */ | |
36c8b586 | 6145 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6146 | { |
6147 | return cpu_rq(cpu)->idle; | |
6148 | } | |
6149 | ||
6150 | /** | |
6151 | * find_process_by_pid - find a process with a matching PID value. | |
6152 | * @pid: the pid in question. | |
6153 | */ | |
a9957449 | 6154 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6155 | { |
228ebcbe | 6156 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6157 | } |
6158 | ||
6159 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6160 | static void |
6161 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6162 | { |
dd41f596 | 6163 | BUG_ON(p->se.on_rq); |
48f24c4d | 6164 | |
1da177e4 LT |
6165 | p->policy = policy; |
6166 | p->rt_priority = prio; | |
b29739f9 IM |
6167 | p->normal_prio = normal_prio(p); |
6168 | /* we are holding p->pi_lock already */ | |
6169 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
6170 | if (rt_prio(p->prio)) |
6171 | p->sched_class = &rt_sched_class; | |
6172 | else | |
6173 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 6174 | set_load_weight(p); |
1da177e4 LT |
6175 | } |
6176 | ||
c69e8d9c DH |
6177 | /* |
6178 | * check the target process has a UID that matches the current process's | |
6179 | */ | |
6180 | static bool check_same_owner(struct task_struct *p) | |
6181 | { | |
6182 | const struct cred *cred = current_cred(), *pcred; | |
6183 | bool match; | |
6184 | ||
6185 | rcu_read_lock(); | |
6186 | pcred = __task_cred(p); | |
6187 | match = (cred->euid == pcred->euid || | |
6188 | cred->euid == pcred->uid); | |
6189 | rcu_read_unlock(); | |
6190 | return match; | |
6191 | } | |
6192 | ||
961ccddd RR |
6193 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6194 | struct sched_param *param, bool user) | |
1da177e4 | 6195 | { |
83b699ed | 6196 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6197 | unsigned long flags; |
cb469845 | 6198 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6199 | struct rq *rq; |
ca94c442 | 6200 | int reset_on_fork; |
1da177e4 | 6201 | |
66e5393a SR |
6202 | /* may grab non-irq protected spin_locks */ |
6203 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6204 | recheck: |
6205 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6206 | if (policy < 0) { |
6207 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6208 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6209 | } else { |
6210 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6211 | policy &= ~SCHED_RESET_ON_FORK; | |
6212 | ||
6213 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6214 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6215 | policy != SCHED_IDLE) | |
6216 | return -EINVAL; | |
6217 | } | |
6218 | ||
1da177e4 LT |
6219 | /* |
6220 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6221 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6222 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6223 | */ |
6224 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6225 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6226 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6227 | return -EINVAL; |
e05606d3 | 6228 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6229 | return -EINVAL; |
6230 | ||
37e4ab3f OC |
6231 | /* |
6232 | * Allow unprivileged RT tasks to decrease priority: | |
6233 | */ | |
961ccddd | 6234 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6235 | if (rt_policy(policy)) { |
8dc3e909 | 6236 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6237 | |
6238 | if (!lock_task_sighand(p, &flags)) | |
6239 | return -ESRCH; | |
6240 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6241 | unlock_task_sighand(p, &flags); | |
6242 | ||
6243 | /* can't set/change the rt policy */ | |
6244 | if (policy != p->policy && !rlim_rtprio) | |
6245 | return -EPERM; | |
6246 | ||
6247 | /* can't increase priority */ | |
6248 | if (param->sched_priority > p->rt_priority && | |
6249 | param->sched_priority > rlim_rtprio) | |
6250 | return -EPERM; | |
6251 | } | |
dd41f596 IM |
6252 | /* |
6253 | * Like positive nice levels, dont allow tasks to | |
6254 | * move out of SCHED_IDLE either: | |
6255 | */ | |
6256 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6257 | return -EPERM; | |
5fe1d75f | 6258 | |
37e4ab3f | 6259 | /* can't change other user's priorities */ |
c69e8d9c | 6260 | if (!check_same_owner(p)) |
37e4ab3f | 6261 | return -EPERM; |
ca94c442 LP |
6262 | |
6263 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6264 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6265 | return -EPERM; | |
37e4ab3f | 6266 | } |
1da177e4 | 6267 | |
725aad24 | 6268 | if (user) { |
b68aa230 | 6269 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6270 | /* |
6271 | * Do not allow realtime tasks into groups that have no runtime | |
6272 | * assigned. | |
6273 | */ | |
9a7e0b18 PZ |
6274 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6275 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6276 | return -EPERM; |
b68aa230 PZ |
6277 | #endif |
6278 | ||
725aad24 JF |
6279 | retval = security_task_setscheduler(p, policy, param); |
6280 | if (retval) | |
6281 | return retval; | |
6282 | } | |
6283 | ||
b29739f9 IM |
6284 | /* |
6285 | * make sure no PI-waiters arrive (or leave) while we are | |
6286 | * changing the priority of the task: | |
6287 | */ | |
6288 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6289 | /* |
6290 | * To be able to change p->policy safely, the apropriate | |
6291 | * runqueue lock must be held. | |
6292 | */ | |
b29739f9 | 6293 | rq = __task_rq_lock(p); |
1da177e4 LT |
6294 | /* recheck policy now with rq lock held */ |
6295 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6296 | policy = oldpolicy = -1; | |
b29739f9 IM |
6297 | __task_rq_unlock(rq); |
6298 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6299 | goto recheck; |
6300 | } | |
2daa3577 | 6301 | update_rq_clock(rq); |
dd41f596 | 6302 | on_rq = p->se.on_rq; |
051a1d1a | 6303 | running = task_current(rq, p); |
0e1f3483 | 6304 | if (on_rq) |
2e1cb74a | 6305 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6306 | if (running) |
6307 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6308 | |
ca94c442 LP |
6309 | p->sched_reset_on_fork = reset_on_fork; |
6310 | ||
1da177e4 | 6311 | oldprio = p->prio; |
dd41f596 | 6312 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6313 | |
0e1f3483 HS |
6314 | if (running) |
6315 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6316 | if (on_rq) { |
6317 | activate_task(rq, p, 0); | |
cb469845 SR |
6318 | |
6319 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6320 | } |
b29739f9 IM |
6321 | __task_rq_unlock(rq); |
6322 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6323 | ||
95e02ca9 TG |
6324 | rt_mutex_adjust_pi(p); |
6325 | ||
1da177e4 LT |
6326 | return 0; |
6327 | } | |
961ccddd RR |
6328 | |
6329 | /** | |
6330 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6331 | * @p: the task in question. | |
6332 | * @policy: new policy. | |
6333 | * @param: structure containing the new RT priority. | |
6334 | * | |
6335 | * NOTE that the task may be already dead. | |
6336 | */ | |
6337 | int sched_setscheduler(struct task_struct *p, int policy, | |
6338 | struct sched_param *param) | |
6339 | { | |
6340 | return __sched_setscheduler(p, policy, param, true); | |
6341 | } | |
1da177e4 LT |
6342 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6343 | ||
961ccddd RR |
6344 | /** |
6345 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6346 | * @p: the task in question. | |
6347 | * @policy: new policy. | |
6348 | * @param: structure containing the new RT priority. | |
6349 | * | |
6350 | * Just like sched_setscheduler, only don't bother checking if the | |
6351 | * current context has permission. For example, this is needed in | |
6352 | * stop_machine(): we create temporary high priority worker threads, | |
6353 | * but our caller might not have that capability. | |
6354 | */ | |
6355 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6356 | struct sched_param *param) | |
6357 | { | |
6358 | return __sched_setscheduler(p, policy, param, false); | |
6359 | } | |
6360 | ||
95cdf3b7 IM |
6361 | static int |
6362 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6363 | { |
1da177e4 LT |
6364 | struct sched_param lparam; |
6365 | struct task_struct *p; | |
36c8b586 | 6366 | int retval; |
1da177e4 LT |
6367 | |
6368 | if (!param || pid < 0) | |
6369 | return -EINVAL; | |
6370 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6371 | return -EFAULT; | |
5fe1d75f ON |
6372 | |
6373 | rcu_read_lock(); | |
6374 | retval = -ESRCH; | |
1da177e4 | 6375 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6376 | if (p != NULL) |
6377 | retval = sched_setscheduler(p, policy, &lparam); | |
6378 | rcu_read_unlock(); | |
36c8b586 | 6379 | |
1da177e4 LT |
6380 | return retval; |
6381 | } | |
6382 | ||
6383 | /** | |
6384 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6385 | * @pid: the pid in question. | |
6386 | * @policy: new policy. | |
6387 | * @param: structure containing the new RT priority. | |
6388 | */ | |
5add95d4 HC |
6389 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6390 | struct sched_param __user *, param) | |
1da177e4 | 6391 | { |
c21761f1 JB |
6392 | /* negative values for policy are not valid */ |
6393 | if (policy < 0) | |
6394 | return -EINVAL; | |
6395 | ||
1da177e4 LT |
6396 | return do_sched_setscheduler(pid, policy, param); |
6397 | } | |
6398 | ||
6399 | /** | |
6400 | * sys_sched_setparam - set/change the RT priority of a thread | |
6401 | * @pid: the pid in question. | |
6402 | * @param: structure containing the new RT priority. | |
6403 | */ | |
5add95d4 | 6404 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6405 | { |
6406 | return do_sched_setscheduler(pid, -1, param); | |
6407 | } | |
6408 | ||
6409 | /** | |
6410 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6411 | * @pid: the pid in question. | |
6412 | */ | |
5add95d4 | 6413 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6414 | { |
36c8b586 | 6415 | struct task_struct *p; |
3a5c359a | 6416 | int retval; |
1da177e4 LT |
6417 | |
6418 | if (pid < 0) | |
3a5c359a | 6419 | return -EINVAL; |
1da177e4 LT |
6420 | |
6421 | retval = -ESRCH; | |
6422 | read_lock(&tasklist_lock); | |
6423 | p = find_process_by_pid(pid); | |
6424 | if (p) { | |
6425 | retval = security_task_getscheduler(p); | |
6426 | if (!retval) | |
ca94c442 LP |
6427 | retval = p->policy |
6428 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6429 | } |
6430 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6431 | return retval; |
6432 | } | |
6433 | ||
6434 | /** | |
ca94c442 | 6435 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6436 | * @pid: the pid in question. |
6437 | * @param: structure containing the RT priority. | |
6438 | */ | |
5add95d4 | 6439 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6440 | { |
6441 | struct sched_param lp; | |
36c8b586 | 6442 | struct task_struct *p; |
3a5c359a | 6443 | int retval; |
1da177e4 LT |
6444 | |
6445 | if (!param || pid < 0) | |
3a5c359a | 6446 | return -EINVAL; |
1da177e4 LT |
6447 | |
6448 | read_lock(&tasklist_lock); | |
6449 | p = find_process_by_pid(pid); | |
6450 | retval = -ESRCH; | |
6451 | if (!p) | |
6452 | goto out_unlock; | |
6453 | ||
6454 | retval = security_task_getscheduler(p); | |
6455 | if (retval) | |
6456 | goto out_unlock; | |
6457 | ||
6458 | lp.sched_priority = p->rt_priority; | |
6459 | read_unlock(&tasklist_lock); | |
6460 | ||
6461 | /* | |
6462 | * This one might sleep, we cannot do it with a spinlock held ... | |
6463 | */ | |
6464 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6465 | ||
1da177e4 LT |
6466 | return retval; |
6467 | ||
6468 | out_unlock: | |
6469 | read_unlock(&tasklist_lock); | |
6470 | return retval; | |
6471 | } | |
6472 | ||
96f874e2 | 6473 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6474 | { |
5a16f3d3 | 6475 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6476 | struct task_struct *p; |
6477 | int retval; | |
1da177e4 | 6478 | |
95402b38 | 6479 | get_online_cpus(); |
1da177e4 LT |
6480 | read_lock(&tasklist_lock); |
6481 | ||
6482 | p = find_process_by_pid(pid); | |
6483 | if (!p) { | |
6484 | read_unlock(&tasklist_lock); | |
95402b38 | 6485 | put_online_cpus(); |
1da177e4 LT |
6486 | return -ESRCH; |
6487 | } | |
6488 | ||
6489 | /* | |
6490 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6491 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6492 | * usage count and then drop tasklist_lock. |
6493 | */ | |
6494 | get_task_struct(p); | |
6495 | read_unlock(&tasklist_lock); | |
6496 | ||
5a16f3d3 RR |
6497 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6498 | retval = -ENOMEM; | |
6499 | goto out_put_task; | |
6500 | } | |
6501 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6502 | retval = -ENOMEM; | |
6503 | goto out_free_cpus_allowed; | |
6504 | } | |
1da177e4 | 6505 | retval = -EPERM; |
c69e8d9c | 6506 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6507 | goto out_unlock; |
6508 | ||
e7834f8f DQ |
6509 | retval = security_task_setscheduler(p, 0, NULL); |
6510 | if (retval) | |
6511 | goto out_unlock; | |
6512 | ||
5a16f3d3 RR |
6513 | cpuset_cpus_allowed(p, cpus_allowed); |
6514 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6515 | again: |
5a16f3d3 | 6516 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6517 | |
8707d8b8 | 6518 | if (!retval) { |
5a16f3d3 RR |
6519 | cpuset_cpus_allowed(p, cpus_allowed); |
6520 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6521 | /* |
6522 | * We must have raced with a concurrent cpuset | |
6523 | * update. Just reset the cpus_allowed to the | |
6524 | * cpuset's cpus_allowed | |
6525 | */ | |
5a16f3d3 | 6526 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6527 | goto again; |
6528 | } | |
6529 | } | |
1da177e4 | 6530 | out_unlock: |
5a16f3d3 RR |
6531 | free_cpumask_var(new_mask); |
6532 | out_free_cpus_allowed: | |
6533 | free_cpumask_var(cpus_allowed); | |
6534 | out_put_task: | |
1da177e4 | 6535 | put_task_struct(p); |
95402b38 | 6536 | put_online_cpus(); |
1da177e4 LT |
6537 | return retval; |
6538 | } | |
6539 | ||
6540 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6541 | struct cpumask *new_mask) |
1da177e4 | 6542 | { |
96f874e2 RR |
6543 | if (len < cpumask_size()) |
6544 | cpumask_clear(new_mask); | |
6545 | else if (len > cpumask_size()) | |
6546 | len = cpumask_size(); | |
6547 | ||
1da177e4 LT |
6548 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6549 | } | |
6550 | ||
6551 | /** | |
6552 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6553 | * @pid: pid of the process | |
6554 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6555 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6556 | */ | |
5add95d4 HC |
6557 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6558 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6559 | { |
5a16f3d3 | 6560 | cpumask_var_t new_mask; |
1da177e4 LT |
6561 | int retval; |
6562 | ||
5a16f3d3 RR |
6563 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6564 | return -ENOMEM; | |
1da177e4 | 6565 | |
5a16f3d3 RR |
6566 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6567 | if (retval == 0) | |
6568 | retval = sched_setaffinity(pid, new_mask); | |
6569 | free_cpumask_var(new_mask); | |
6570 | return retval; | |
1da177e4 LT |
6571 | } |
6572 | ||
96f874e2 | 6573 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6574 | { |
36c8b586 | 6575 | struct task_struct *p; |
1da177e4 | 6576 | int retval; |
1da177e4 | 6577 | |
95402b38 | 6578 | get_online_cpus(); |
1da177e4 LT |
6579 | read_lock(&tasklist_lock); |
6580 | ||
6581 | retval = -ESRCH; | |
6582 | p = find_process_by_pid(pid); | |
6583 | if (!p) | |
6584 | goto out_unlock; | |
6585 | ||
e7834f8f DQ |
6586 | retval = security_task_getscheduler(p); |
6587 | if (retval) | |
6588 | goto out_unlock; | |
6589 | ||
96f874e2 | 6590 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6591 | |
6592 | out_unlock: | |
6593 | read_unlock(&tasklist_lock); | |
95402b38 | 6594 | put_online_cpus(); |
1da177e4 | 6595 | |
9531b62f | 6596 | return retval; |
1da177e4 LT |
6597 | } |
6598 | ||
6599 | /** | |
6600 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6601 | * @pid: pid of the process | |
6602 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6603 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6604 | */ | |
5add95d4 HC |
6605 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6606 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6607 | { |
6608 | int ret; | |
f17c8607 | 6609 | cpumask_var_t mask; |
1da177e4 | 6610 | |
f17c8607 | 6611 | if (len < cpumask_size()) |
1da177e4 LT |
6612 | return -EINVAL; |
6613 | ||
f17c8607 RR |
6614 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6615 | return -ENOMEM; | |
1da177e4 | 6616 | |
f17c8607 RR |
6617 | ret = sched_getaffinity(pid, mask); |
6618 | if (ret == 0) { | |
6619 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6620 | ret = -EFAULT; | |
6621 | else | |
6622 | ret = cpumask_size(); | |
6623 | } | |
6624 | free_cpumask_var(mask); | |
1da177e4 | 6625 | |
f17c8607 | 6626 | return ret; |
1da177e4 LT |
6627 | } |
6628 | ||
6629 | /** | |
6630 | * sys_sched_yield - yield the current processor to other threads. | |
6631 | * | |
dd41f596 IM |
6632 | * This function yields the current CPU to other tasks. If there are no |
6633 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6634 | */ |
5add95d4 | 6635 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6636 | { |
70b97a7f | 6637 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6638 | |
2d72376b | 6639 | schedstat_inc(rq, yld_count); |
4530d7ab | 6640 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6641 | |
6642 | /* | |
6643 | * Since we are going to call schedule() anyway, there's | |
6644 | * no need to preempt or enable interrupts: | |
6645 | */ | |
6646 | __release(rq->lock); | |
8a25d5de | 6647 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6648 | _raw_spin_unlock(&rq->lock); |
6649 | preempt_enable_no_resched(); | |
6650 | ||
6651 | schedule(); | |
6652 | ||
6653 | return 0; | |
6654 | } | |
6655 | ||
d86ee480 PZ |
6656 | static inline int should_resched(void) |
6657 | { | |
6658 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6659 | } | |
6660 | ||
e7b38404 | 6661 | static void __cond_resched(void) |
1da177e4 | 6662 | { |
e7aaaa69 FW |
6663 | add_preempt_count(PREEMPT_ACTIVE); |
6664 | schedule(); | |
6665 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6666 | } |
6667 | ||
02b67cc3 | 6668 | int __sched _cond_resched(void) |
1da177e4 | 6669 | { |
d86ee480 | 6670 | if (should_resched()) { |
1da177e4 LT |
6671 | __cond_resched(); |
6672 | return 1; | |
6673 | } | |
6674 | return 0; | |
6675 | } | |
02b67cc3 | 6676 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6677 | |
6678 | /* | |
613afbf8 | 6679 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6680 | * call schedule, and on return reacquire the lock. |
6681 | * | |
41a2d6cf | 6682 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6683 | * operations here to prevent schedule() from being called twice (once via |
6684 | * spin_unlock(), once by hand). | |
6685 | */ | |
613afbf8 | 6686 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6687 | { |
d86ee480 | 6688 | int resched = should_resched(); |
6df3cecb JK |
6689 | int ret = 0; |
6690 | ||
f607c668 PZ |
6691 | lockdep_assert_held(lock); |
6692 | ||
95c354fe | 6693 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6694 | spin_unlock(lock); |
d86ee480 | 6695 | if (resched) |
95c354fe NP |
6696 | __cond_resched(); |
6697 | else | |
6698 | cpu_relax(); | |
6df3cecb | 6699 | ret = 1; |
1da177e4 | 6700 | spin_lock(lock); |
1da177e4 | 6701 | } |
6df3cecb | 6702 | return ret; |
1da177e4 | 6703 | } |
613afbf8 | 6704 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6705 | |
613afbf8 | 6706 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6707 | { |
6708 | BUG_ON(!in_softirq()); | |
6709 | ||
d86ee480 | 6710 | if (should_resched()) { |
98d82567 | 6711 | local_bh_enable(); |
1da177e4 LT |
6712 | __cond_resched(); |
6713 | local_bh_disable(); | |
6714 | return 1; | |
6715 | } | |
6716 | return 0; | |
6717 | } | |
613afbf8 | 6718 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6719 | |
1da177e4 LT |
6720 | /** |
6721 | * yield - yield the current processor to other threads. | |
6722 | * | |
72fd4a35 | 6723 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6724 | * thread runnable and calls sys_sched_yield(). |
6725 | */ | |
6726 | void __sched yield(void) | |
6727 | { | |
6728 | set_current_state(TASK_RUNNING); | |
6729 | sys_sched_yield(); | |
6730 | } | |
1da177e4 LT |
6731 | EXPORT_SYMBOL(yield); |
6732 | ||
6733 | /* | |
41a2d6cf | 6734 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6735 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
6736 | */ |
6737 | void __sched io_schedule(void) | |
6738 | { | |
54d35f29 | 6739 | struct rq *rq = raw_rq(); |
1da177e4 | 6740 | |
0ff92245 | 6741 | delayacct_blkio_start(); |
1da177e4 | 6742 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6743 | current->in_iowait = 1; |
1da177e4 | 6744 | schedule(); |
8f0dfc34 | 6745 | current->in_iowait = 0; |
1da177e4 | 6746 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6747 | delayacct_blkio_end(); |
1da177e4 | 6748 | } |
1da177e4 LT |
6749 | EXPORT_SYMBOL(io_schedule); |
6750 | ||
6751 | long __sched io_schedule_timeout(long timeout) | |
6752 | { | |
54d35f29 | 6753 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6754 | long ret; |
6755 | ||
0ff92245 | 6756 | delayacct_blkio_start(); |
1da177e4 | 6757 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6758 | current->in_iowait = 1; |
1da177e4 | 6759 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6760 | current->in_iowait = 0; |
1da177e4 | 6761 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6762 | delayacct_blkio_end(); |
1da177e4 LT |
6763 | return ret; |
6764 | } | |
6765 | ||
6766 | /** | |
6767 | * sys_sched_get_priority_max - return maximum RT priority. | |
6768 | * @policy: scheduling class. | |
6769 | * | |
6770 | * this syscall returns the maximum rt_priority that can be used | |
6771 | * by a given scheduling class. | |
6772 | */ | |
5add95d4 | 6773 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6774 | { |
6775 | int ret = -EINVAL; | |
6776 | ||
6777 | switch (policy) { | |
6778 | case SCHED_FIFO: | |
6779 | case SCHED_RR: | |
6780 | ret = MAX_USER_RT_PRIO-1; | |
6781 | break; | |
6782 | case SCHED_NORMAL: | |
b0a9499c | 6783 | case SCHED_BATCH: |
dd41f596 | 6784 | case SCHED_IDLE: |
1da177e4 LT |
6785 | ret = 0; |
6786 | break; | |
6787 | } | |
6788 | return ret; | |
6789 | } | |
6790 | ||
6791 | /** | |
6792 | * sys_sched_get_priority_min - return minimum RT priority. | |
6793 | * @policy: scheduling class. | |
6794 | * | |
6795 | * this syscall returns the minimum rt_priority that can be used | |
6796 | * by a given scheduling class. | |
6797 | */ | |
5add95d4 | 6798 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6799 | { |
6800 | int ret = -EINVAL; | |
6801 | ||
6802 | switch (policy) { | |
6803 | case SCHED_FIFO: | |
6804 | case SCHED_RR: | |
6805 | ret = 1; | |
6806 | break; | |
6807 | case SCHED_NORMAL: | |
b0a9499c | 6808 | case SCHED_BATCH: |
dd41f596 | 6809 | case SCHED_IDLE: |
1da177e4 LT |
6810 | ret = 0; |
6811 | } | |
6812 | return ret; | |
6813 | } | |
6814 | ||
6815 | /** | |
6816 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6817 | * @pid: pid of the process. | |
6818 | * @interval: userspace pointer to the timeslice value. | |
6819 | * | |
6820 | * this syscall writes the default timeslice value of a given process | |
6821 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6822 | */ | |
17da2bd9 | 6823 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6824 | struct timespec __user *, interval) |
1da177e4 | 6825 | { |
36c8b586 | 6826 | struct task_struct *p; |
a4ec24b4 | 6827 | unsigned int time_slice; |
3a5c359a | 6828 | int retval; |
1da177e4 | 6829 | struct timespec t; |
1da177e4 LT |
6830 | |
6831 | if (pid < 0) | |
3a5c359a | 6832 | return -EINVAL; |
1da177e4 LT |
6833 | |
6834 | retval = -ESRCH; | |
6835 | read_lock(&tasklist_lock); | |
6836 | p = find_process_by_pid(pid); | |
6837 | if (!p) | |
6838 | goto out_unlock; | |
6839 | ||
6840 | retval = security_task_getscheduler(p); | |
6841 | if (retval) | |
6842 | goto out_unlock; | |
6843 | ||
0d721cea | 6844 | time_slice = p->sched_class->get_rr_interval(p); |
a4ec24b4 | 6845 | |
1da177e4 | 6846 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6847 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6848 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6849 | return retval; |
3a5c359a | 6850 | |
1da177e4 LT |
6851 | out_unlock: |
6852 | read_unlock(&tasklist_lock); | |
6853 | return retval; | |
6854 | } | |
6855 | ||
7c731e0a | 6856 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6857 | |
82a1fcb9 | 6858 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6859 | { |
1da177e4 | 6860 | unsigned long free = 0; |
36c8b586 | 6861 | unsigned state; |
1da177e4 | 6862 | |
1da177e4 | 6863 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6864 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6865 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6866 | #if BITS_PER_LONG == 32 |
1da177e4 | 6867 | if (state == TASK_RUNNING) |
cc4ea795 | 6868 | printk(KERN_CONT " running "); |
1da177e4 | 6869 | else |
cc4ea795 | 6870 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6871 | #else |
6872 | if (state == TASK_RUNNING) | |
cc4ea795 | 6873 | printk(KERN_CONT " running task "); |
1da177e4 | 6874 | else |
cc4ea795 | 6875 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6876 | #endif |
6877 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6878 | free = stack_not_used(p); |
1da177e4 | 6879 | #endif |
aa47b7e0 DR |
6880 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6881 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6882 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6883 | |
5fb5e6de | 6884 | show_stack(p, NULL); |
1da177e4 LT |
6885 | } |
6886 | ||
e59e2ae2 | 6887 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6888 | { |
36c8b586 | 6889 | struct task_struct *g, *p; |
1da177e4 | 6890 | |
4bd77321 IM |
6891 | #if BITS_PER_LONG == 32 |
6892 | printk(KERN_INFO | |
6893 | " task PC stack pid father\n"); | |
1da177e4 | 6894 | #else |
4bd77321 IM |
6895 | printk(KERN_INFO |
6896 | " task PC stack pid father\n"); | |
1da177e4 LT |
6897 | #endif |
6898 | read_lock(&tasklist_lock); | |
6899 | do_each_thread(g, p) { | |
6900 | /* | |
6901 | * reset the NMI-timeout, listing all files on a slow | |
6902 | * console might take alot of time: | |
6903 | */ | |
6904 | touch_nmi_watchdog(); | |
39bc89fd | 6905 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6906 | sched_show_task(p); |
1da177e4 LT |
6907 | } while_each_thread(g, p); |
6908 | ||
04c9167f JF |
6909 | touch_all_softlockup_watchdogs(); |
6910 | ||
dd41f596 IM |
6911 | #ifdef CONFIG_SCHED_DEBUG |
6912 | sysrq_sched_debug_show(); | |
6913 | #endif | |
1da177e4 | 6914 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6915 | /* |
6916 | * Only show locks if all tasks are dumped: | |
6917 | */ | |
6918 | if (state_filter == -1) | |
6919 | debug_show_all_locks(); | |
1da177e4 LT |
6920 | } |
6921 | ||
1df21055 IM |
6922 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6923 | { | |
dd41f596 | 6924 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6925 | } |
6926 | ||
f340c0d1 IM |
6927 | /** |
6928 | * init_idle - set up an idle thread for a given CPU | |
6929 | * @idle: task in question | |
6930 | * @cpu: cpu the idle task belongs to | |
6931 | * | |
6932 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6933 | * flag, to make booting more robust. | |
6934 | */ | |
5c1e1767 | 6935 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6936 | { |
70b97a7f | 6937 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6938 | unsigned long flags; |
6939 | ||
5cbd54ef IM |
6940 | spin_lock_irqsave(&rq->lock, flags); |
6941 | ||
dd41f596 IM |
6942 | __sched_fork(idle); |
6943 | idle->se.exec_start = sched_clock(); | |
6944 | ||
b29739f9 | 6945 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6946 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6947 | __set_task_cpu(idle, cpu); |
1da177e4 | 6948 | |
1da177e4 | 6949 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6950 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6951 | idle->oncpu = 1; | |
6952 | #endif | |
1da177e4 LT |
6953 | spin_unlock_irqrestore(&rq->lock, flags); |
6954 | ||
6955 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6956 | #if defined(CONFIG_PREEMPT) |
6957 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6958 | #else | |
a1261f54 | 6959 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6960 | #endif |
dd41f596 IM |
6961 | /* |
6962 | * The idle tasks have their own, simple scheduling class: | |
6963 | */ | |
6964 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6965 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6966 | } |
6967 | ||
6968 | /* | |
6969 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6970 | * indicates which cpus entered this state. This is used | |
6971 | * in the rcu update to wait only for active cpus. For system | |
6972 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6973 | * always be CPU_BITS_NONE. |
1da177e4 | 6974 | */ |
6a7b3dc3 | 6975 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6976 | |
19978ca6 IM |
6977 | /* |
6978 | * Increase the granularity value when there are more CPUs, | |
6979 | * because with more CPUs the 'effective latency' as visible | |
6980 | * to users decreases. But the relationship is not linear, | |
6981 | * so pick a second-best guess by going with the log2 of the | |
6982 | * number of CPUs. | |
6983 | * | |
6984 | * This idea comes from the SD scheduler of Con Kolivas: | |
6985 | */ | |
6986 | static inline void sched_init_granularity(void) | |
6987 | { | |
6988 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6989 | const unsigned long limit = 200000000; | |
6990 | ||
6991 | sysctl_sched_min_granularity *= factor; | |
6992 | if (sysctl_sched_min_granularity > limit) | |
6993 | sysctl_sched_min_granularity = limit; | |
6994 | ||
6995 | sysctl_sched_latency *= factor; | |
6996 | if (sysctl_sched_latency > limit) | |
6997 | sysctl_sched_latency = limit; | |
6998 | ||
6999 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
7000 | |
7001 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
7002 | } |
7003 | ||
1da177e4 LT |
7004 | #ifdef CONFIG_SMP |
7005 | /* | |
7006 | * This is how migration works: | |
7007 | * | |
70b97a7f | 7008 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7009 | * runqueue and wake up that CPU's migration thread. |
7010 | * 2) we down() the locked semaphore => thread blocks. | |
7011 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7012 | * thread off the CPU) | |
7013 | * 4) it gets the migration request and checks whether the migrated | |
7014 | * task is still in the wrong runqueue. | |
7015 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7016 | * it and puts it into the right queue. | |
7017 | * 6) migration thread up()s the semaphore. | |
7018 | * 7) we wake up and the migration is done. | |
7019 | */ | |
7020 | ||
7021 | /* | |
7022 | * Change a given task's CPU affinity. Migrate the thread to a | |
7023 | * proper CPU and schedule it away if the CPU it's executing on | |
7024 | * is removed from the allowed bitmask. | |
7025 | * | |
7026 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7027 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7028 | * call is not atomic; no spinlocks may be held. |
7029 | */ | |
96f874e2 | 7030 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7031 | { |
70b97a7f | 7032 | struct migration_req req; |
1da177e4 | 7033 | unsigned long flags; |
70b97a7f | 7034 | struct rq *rq; |
48f24c4d | 7035 | int ret = 0; |
1da177e4 LT |
7036 | |
7037 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7038 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7039 | ret = -EINVAL; |
7040 | goto out; | |
7041 | } | |
7042 | ||
9985b0ba | 7043 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7044 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7045 | ret = -EINVAL; |
7046 | goto out; | |
7047 | } | |
7048 | ||
73fe6aae | 7049 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7050 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7051 | else { |
96f874e2 RR |
7052 | cpumask_copy(&p->cpus_allowed, new_mask); |
7053 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7054 | } |
7055 | ||
1da177e4 | 7056 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7057 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7058 | goto out; |
7059 | ||
1e5ce4f4 | 7060 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 | 7061 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7062 | struct task_struct *mt = rq->migration_thread; |
7063 | ||
7064 | get_task_struct(mt); | |
1da177e4 LT |
7065 | task_rq_unlock(rq, &flags); |
7066 | wake_up_process(rq->migration_thread); | |
693525e3 | 7067 | put_task_struct(mt); |
1da177e4 LT |
7068 | wait_for_completion(&req.done); |
7069 | tlb_migrate_finish(p->mm); | |
7070 | return 0; | |
7071 | } | |
7072 | out: | |
7073 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7074 | |
1da177e4 LT |
7075 | return ret; |
7076 | } | |
cd8ba7cd | 7077 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7078 | |
7079 | /* | |
41a2d6cf | 7080 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7081 | * this because either it can't run here any more (set_cpus_allowed() |
7082 | * away from this CPU, or CPU going down), or because we're | |
7083 | * attempting to rebalance this task on exec (sched_exec). | |
7084 | * | |
7085 | * So we race with normal scheduler movements, but that's OK, as long | |
7086 | * as the task is no longer on this CPU. | |
efc30814 KK |
7087 | * |
7088 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7089 | */ |
efc30814 | 7090 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7091 | { |
70b97a7f | 7092 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7093 | int ret = 0, on_rq; |
1da177e4 | 7094 | |
e761b772 | 7095 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7096 | return ret; |
1da177e4 LT |
7097 | |
7098 | rq_src = cpu_rq(src_cpu); | |
7099 | rq_dest = cpu_rq(dest_cpu); | |
7100 | ||
7101 | double_rq_lock(rq_src, rq_dest); | |
7102 | /* Already moved. */ | |
7103 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7104 | goto done; |
1da177e4 | 7105 | /* Affinity changed (again). */ |
96f874e2 | 7106 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7107 | goto fail; |
1da177e4 | 7108 | |
dd41f596 | 7109 | on_rq = p->se.on_rq; |
6e82a3be | 7110 | if (on_rq) |
2e1cb74a | 7111 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7112 | |
1da177e4 | 7113 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7114 | if (on_rq) { |
7115 | activate_task(rq_dest, p, 0); | |
15afe09b | 7116 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7117 | } |
b1e38734 | 7118 | done: |
efc30814 | 7119 | ret = 1; |
b1e38734 | 7120 | fail: |
1da177e4 | 7121 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7122 | return ret; |
1da177e4 LT |
7123 | } |
7124 | ||
03b042bf PM |
7125 | #define RCU_MIGRATION_IDLE 0 |
7126 | #define RCU_MIGRATION_NEED_QS 1 | |
7127 | #define RCU_MIGRATION_GOT_QS 2 | |
7128 | #define RCU_MIGRATION_MUST_SYNC 3 | |
7129 | ||
1da177e4 LT |
7130 | /* |
7131 | * migration_thread - this is a highprio system thread that performs | |
7132 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7133 | * another runqueue. | |
7134 | */ | |
95cdf3b7 | 7135 | static int migration_thread(void *data) |
1da177e4 | 7136 | { |
03b042bf | 7137 | int badcpu; |
1da177e4 | 7138 | int cpu = (long)data; |
70b97a7f | 7139 | struct rq *rq; |
1da177e4 LT |
7140 | |
7141 | rq = cpu_rq(cpu); | |
7142 | BUG_ON(rq->migration_thread != current); | |
7143 | ||
7144 | set_current_state(TASK_INTERRUPTIBLE); | |
7145 | while (!kthread_should_stop()) { | |
70b97a7f | 7146 | struct migration_req *req; |
1da177e4 | 7147 | struct list_head *head; |
1da177e4 | 7148 | |
1da177e4 LT |
7149 | spin_lock_irq(&rq->lock); |
7150 | ||
7151 | if (cpu_is_offline(cpu)) { | |
7152 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7153 | break; |
1da177e4 LT |
7154 | } |
7155 | ||
7156 | if (rq->active_balance) { | |
7157 | active_load_balance(rq, cpu); | |
7158 | rq->active_balance = 0; | |
7159 | } | |
7160 | ||
7161 | head = &rq->migration_queue; | |
7162 | ||
7163 | if (list_empty(head)) { | |
7164 | spin_unlock_irq(&rq->lock); | |
7165 | schedule(); | |
7166 | set_current_state(TASK_INTERRUPTIBLE); | |
7167 | continue; | |
7168 | } | |
70b97a7f | 7169 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7170 | list_del_init(head->next); |
7171 | ||
03b042bf PM |
7172 | if (req->task != NULL) { |
7173 | spin_unlock(&rq->lock); | |
7174 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7175 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
7176 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
7177 | spin_unlock(&rq->lock); | |
7178 | } else { | |
7179 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
7180 | spin_unlock(&rq->lock); | |
7181 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); | |
7182 | } | |
674311d5 | 7183 | local_irq_enable(); |
1da177e4 LT |
7184 | |
7185 | complete(&req->done); | |
7186 | } | |
7187 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7188 | |
1da177e4 LT |
7189 | return 0; |
7190 | } | |
7191 | ||
7192 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7193 | |
7194 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7195 | { | |
7196 | int ret; | |
7197 | ||
7198 | local_irq_disable(); | |
7199 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7200 | local_irq_enable(); | |
7201 | return ret; | |
7202 | } | |
7203 | ||
054b9108 | 7204 | /* |
3a4fa0a2 | 7205 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7206 | */ |
48f24c4d | 7207 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7208 | { |
70b97a7f | 7209 | int dest_cpu; |
6ca09dfc | 7210 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7211 | |
7212 | again: | |
7213 | /* Look for allowed, online CPU in same node. */ | |
7214 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7215 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7216 | goto move; | |
7217 | ||
7218 | /* Any allowed, online CPU? */ | |
7219 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7220 | if (dest_cpu < nr_cpu_ids) | |
7221 | goto move; | |
7222 | ||
7223 | /* No more Mr. Nice Guy. */ | |
7224 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7225 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7226 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7227 | |
e76bd8d9 RR |
7228 | /* |
7229 | * Don't tell them about moving exiting tasks or | |
7230 | * kernel threads (both mm NULL), since they never | |
7231 | * leave kernel. | |
7232 | */ | |
7233 | if (p->mm && printk_ratelimit()) { | |
7234 | printk(KERN_INFO "process %d (%s) no " | |
7235 | "longer affine to cpu%d\n", | |
7236 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7237 | } |
e76bd8d9 RR |
7238 | } |
7239 | ||
7240 | move: | |
7241 | /* It can have affinity changed while we were choosing. */ | |
7242 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7243 | goto again; | |
1da177e4 LT |
7244 | } |
7245 | ||
7246 | /* | |
7247 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7248 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7249 | * for performance reasons the counter is not stricly tracking tasks to | |
7250 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7251 | * to keep the global sum constant after CPU-down: | |
7252 | */ | |
70b97a7f | 7253 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7254 | { |
1e5ce4f4 | 7255 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7256 | unsigned long flags; |
7257 | ||
7258 | local_irq_save(flags); | |
7259 | double_rq_lock(rq_src, rq_dest); | |
7260 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7261 | rq_src->nr_uninterruptible = 0; | |
7262 | double_rq_unlock(rq_src, rq_dest); | |
7263 | local_irq_restore(flags); | |
7264 | } | |
7265 | ||
7266 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7267 | static void migrate_live_tasks(int src_cpu) | |
7268 | { | |
48f24c4d | 7269 | struct task_struct *p, *t; |
1da177e4 | 7270 | |
f7b4cddc | 7271 | read_lock(&tasklist_lock); |
1da177e4 | 7272 | |
48f24c4d IM |
7273 | do_each_thread(t, p) { |
7274 | if (p == current) | |
1da177e4 LT |
7275 | continue; |
7276 | ||
48f24c4d IM |
7277 | if (task_cpu(p) == src_cpu) |
7278 | move_task_off_dead_cpu(src_cpu, p); | |
7279 | } while_each_thread(t, p); | |
1da177e4 | 7280 | |
f7b4cddc | 7281 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7282 | } |
7283 | ||
dd41f596 IM |
7284 | /* |
7285 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7286 | * It does so by boosting its priority to highest possible. |
7287 | * Used by CPU offline code. | |
1da177e4 LT |
7288 | */ |
7289 | void sched_idle_next(void) | |
7290 | { | |
48f24c4d | 7291 | int this_cpu = smp_processor_id(); |
70b97a7f | 7292 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7293 | struct task_struct *p = rq->idle; |
7294 | unsigned long flags; | |
7295 | ||
7296 | /* cpu has to be offline */ | |
48f24c4d | 7297 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7298 | |
48f24c4d IM |
7299 | /* |
7300 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7301 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7302 | */ |
7303 | spin_lock_irqsave(&rq->lock, flags); | |
7304 | ||
dd41f596 | 7305 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7306 | |
94bc9a7b DA |
7307 | update_rq_clock(rq); |
7308 | activate_task(rq, p, 0); | |
1da177e4 LT |
7309 | |
7310 | spin_unlock_irqrestore(&rq->lock, flags); | |
7311 | } | |
7312 | ||
48f24c4d IM |
7313 | /* |
7314 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7315 | * offline. |
7316 | */ | |
7317 | void idle_task_exit(void) | |
7318 | { | |
7319 | struct mm_struct *mm = current->active_mm; | |
7320 | ||
7321 | BUG_ON(cpu_online(smp_processor_id())); | |
7322 | ||
7323 | if (mm != &init_mm) | |
7324 | switch_mm(mm, &init_mm, current); | |
7325 | mmdrop(mm); | |
7326 | } | |
7327 | ||
054b9108 | 7328 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7329 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7330 | { |
70b97a7f | 7331 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7332 | |
7333 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7334 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7335 | |
7336 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7337 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7338 | |
48f24c4d | 7339 | get_task_struct(p); |
1da177e4 LT |
7340 | |
7341 | /* | |
7342 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7343 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7344 | * fine. |
7345 | */ | |
f7b4cddc | 7346 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7347 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7348 | spin_lock_irq(&rq->lock); |
1da177e4 | 7349 | |
48f24c4d | 7350 | put_task_struct(p); |
1da177e4 LT |
7351 | } |
7352 | ||
7353 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7354 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7355 | { | |
70b97a7f | 7356 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7357 | struct task_struct *next; |
48f24c4d | 7358 | |
dd41f596 IM |
7359 | for ( ; ; ) { |
7360 | if (!rq->nr_running) | |
7361 | break; | |
a8e504d2 | 7362 | update_rq_clock(rq); |
b67802ea | 7363 | next = pick_next_task(rq); |
dd41f596 IM |
7364 | if (!next) |
7365 | break; | |
79c53799 | 7366 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7367 | migrate_dead(dead_cpu, next); |
e692ab53 | 7368 | |
1da177e4 LT |
7369 | } |
7370 | } | |
dce48a84 TG |
7371 | |
7372 | /* | |
7373 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7374 | */ | |
7375 | static void calc_global_load_remove(struct rq *rq) | |
7376 | { | |
7377 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7378 | rq->calc_load_active = 0; |
dce48a84 | 7379 | } |
1da177e4 LT |
7380 | #endif /* CONFIG_HOTPLUG_CPU */ |
7381 | ||
e692ab53 NP |
7382 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7383 | ||
7384 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7385 | { |
7386 | .procname = "sched_domain", | |
c57baf1e | 7387 | .mode = 0555, |
e0361851 | 7388 | }, |
38605cae | 7389 | {0, }, |
e692ab53 NP |
7390 | }; |
7391 | ||
7392 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7393 | { |
c57baf1e | 7394 | .ctl_name = CTL_KERN, |
e0361851 | 7395 | .procname = "kernel", |
c57baf1e | 7396 | .mode = 0555, |
e0361851 AD |
7397 | .child = sd_ctl_dir, |
7398 | }, | |
38605cae | 7399 | {0, }, |
e692ab53 NP |
7400 | }; |
7401 | ||
7402 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7403 | { | |
7404 | struct ctl_table *entry = | |
5cf9f062 | 7405 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7406 | |
e692ab53 NP |
7407 | return entry; |
7408 | } | |
7409 | ||
6382bc90 MM |
7410 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7411 | { | |
cd790076 | 7412 | struct ctl_table *entry; |
6382bc90 | 7413 | |
cd790076 MM |
7414 | /* |
7415 | * In the intermediate directories, both the child directory and | |
7416 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7417 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7418 | * static strings and all have proc handlers. |
7419 | */ | |
7420 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7421 | if (entry->child) |
7422 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7423 | if (entry->proc_handler == NULL) |
7424 | kfree(entry->procname); | |
7425 | } | |
6382bc90 MM |
7426 | |
7427 | kfree(*tablep); | |
7428 | *tablep = NULL; | |
7429 | } | |
7430 | ||
e692ab53 | 7431 | static void |
e0361851 | 7432 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7433 | const char *procname, void *data, int maxlen, |
7434 | mode_t mode, proc_handler *proc_handler) | |
7435 | { | |
e692ab53 NP |
7436 | entry->procname = procname; |
7437 | entry->data = data; | |
7438 | entry->maxlen = maxlen; | |
7439 | entry->mode = mode; | |
7440 | entry->proc_handler = proc_handler; | |
7441 | } | |
7442 | ||
7443 | static struct ctl_table * | |
7444 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7445 | { | |
a5d8c348 | 7446 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7447 | |
ad1cdc1d MM |
7448 | if (table == NULL) |
7449 | return NULL; | |
7450 | ||
e0361851 | 7451 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7452 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7453 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7454 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7455 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7456 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7457 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7458 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7459 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7460 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7461 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7462 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7463 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7464 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7465 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7466 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7467 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7468 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7469 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7470 | &sd->cache_nice_tries, |
7471 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7472 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7473 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7474 | set_table_entry(&table[11], "name", sd->name, |
7475 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7476 | /* &table[12] is terminator */ | |
e692ab53 NP |
7477 | |
7478 | return table; | |
7479 | } | |
7480 | ||
9a4e7159 | 7481 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7482 | { |
7483 | struct ctl_table *entry, *table; | |
7484 | struct sched_domain *sd; | |
7485 | int domain_num = 0, i; | |
7486 | char buf[32]; | |
7487 | ||
7488 | for_each_domain(cpu, sd) | |
7489 | domain_num++; | |
7490 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7491 | if (table == NULL) |
7492 | return NULL; | |
e692ab53 NP |
7493 | |
7494 | i = 0; | |
7495 | for_each_domain(cpu, sd) { | |
7496 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7497 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7498 | entry->mode = 0555; |
e692ab53 NP |
7499 | entry->child = sd_alloc_ctl_domain_table(sd); |
7500 | entry++; | |
7501 | i++; | |
7502 | } | |
7503 | return table; | |
7504 | } | |
7505 | ||
7506 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7507 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7508 | { |
7509 | int i, cpu_num = num_online_cpus(); | |
7510 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7511 | char buf[32]; | |
7512 | ||
7378547f MM |
7513 | WARN_ON(sd_ctl_dir[0].child); |
7514 | sd_ctl_dir[0].child = entry; | |
7515 | ||
ad1cdc1d MM |
7516 | if (entry == NULL) |
7517 | return; | |
7518 | ||
97b6ea7b | 7519 | for_each_online_cpu(i) { |
e692ab53 | 7520 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7521 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7522 | entry->mode = 0555; |
e692ab53 | 7523 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7524 | entry++; |
e692ab53 | 7525 | } |
7378547f MM |
7526 | |
7527 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7528 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7529 | } | |
6382bc90 | 7530 | |
7378547f | 7531 | /* may be called multiple times per register */ |
6382bc90 MM |
7532 | static void unregister_sched_domain_sysctl(void) |
7533 | { | |
7378547f MM |
7534 | if (sd_sysctl_header) |
7535 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7536 | sd_sysctl_header = NULL; |
7378547f MM |
7537 | if (sd_ctl_dir[0].child) |
7538 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7539 | } |
e692ab53 | 7540 | #else |
6382bc90 MM |
7541 | static void register_sched_domain_sysctl(void) |
7542 | { | |
7543 | } | |
7544 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7545 | { |
7546 | } | |
7547 | #endif | |
7548 | ||
1f11eb6a GH |
7549 | static void set_rq_online(struct rq *rq) |
7550 | { | |
7551 | if (!rq->online) { | |
7552 | const struct sched_class *class; | |
7553 | ||
c6c4927b | 7554 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7555 | rq->online = 1; |
7556 | ||
7557 | for_each_class(class) { | |
7558 | if (class->rq_online) | |
7559 | class->rq_online(rq); | |
7560 | } | |
7561 | } | |
7562 | } | |
7563 | ||
7564 | static void set_rq_offline(struct rq *rq) | |
7565 | { | |
7566 | if (rq->online) { | |
7567 | const struct sched_class *class; | |
7568 | ||
7569 | for_each_class(class) { | |
7570 | if (class->rq_offline) | |
7571 | class->rq_offline(rq); | |
7572 | } | |
7573 | ||
c6c4927b | 7574 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7575 | rq->online = 0; |
7576 | } | |
7577 | } | |
7578 | ||
1da177e4 LT |
7579 | /* |
7580 | * migration_call - callback that gets triggered when a CPU is added. | |
7581 | * Here we can start up the necessary migration thread for the new CPU. | |
7582 | */ | |
48f24c4d IM |
7583 | static int __cpuinit |
7584 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7585 | { |
1da177e4 | 7586 | struct task_struct *p; |
48f24c4d | 7587 | int cpu = (long)hcpu; |
1da177e4 | 7588 | unsigned long flags; |
70b97a7f | 7589 | struct rq *rq; |
1da177e4 LT |
7590 | |
7591 | switch (action) { | |
5be9361c | 7592 | |
1da177e4 | 7593 | case CPU_UP_PREPARE: |
8bb78442 | 7594 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7595 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7596 | if (IS_ERR(p)) |
7597 | return NOTIFY_BAD; | |
1da177e4 LT |
7598 | kthread_bind(p, cpu); |
7599 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7600 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7601 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7602 | task_rq_unlock(rq, &flags); |
371cbb38 | 7603 | get_task_struct(p); |
1da177e4 | 7604 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7605 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7606 | break; |
48f24c4d | 7607 | |
1da177e4 | 7608 | case CPU_ONLINE: |
8bb78442 | 7609 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7610 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7611 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7612 | |
7613 | /* Update our root-domain */ | |
7614 | rq = cpu_rq(cpu); | |
7615 | spin_lock_irqsave(&rq->lock, flags); | |
7616 | if (rq->rd) { | |
c6c4927b | 7617 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7618 | |
7619 | set_rq_online(rq); | |
1f94ef59 GH |
7620 | } |
7621 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7622 | break; |
48f24c4d | 7623 | |
1da177e4 LT |
7624 | #ifdef CONFIG_HOTPLUG_CPU |
7625 | case CPU_UP_CANCELED: | |
8bb78442 | 7626 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7627 | if (!cpu_rq(cpu)->migration_thread) |
7628 | break; | |
41a2d6cf | 7629 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7630 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7631 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7632 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7633 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7634 | cpu_rq(cpu)->migration_thread = NULL; |
7635 | break; | |
48f24c4d | 7636 | |
1da177e4 | 7637 | case CPU_DEAD: |
8bb78442 | 7638 | case CPU_DEAD_FROZEN: |
470fd646 | 7639 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7640 | migrate_live_tasks(cpu); |
7641 | rq = cpu_rq(cpu); | |
7642 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7643 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7644 | rq->migration_thread = NULL; |
7645 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7646 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7647 | update_rq_clock(rq); |
2e1cb74a | 7648 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7649 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7650 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7651 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7652 | migrate_dead_tasks(cpu); |
d2da272a | 7653 | spin_unlock_irq(&rq->lock); |
470fd646 | 7654 | cpuset_unlock(); |
1da177e4 LT |
7655 | migrate_nr_uninterruptible(rq); |
7656 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7657 | calc_global_load_remove(rq); |
41a2d6cf IM |
7658 | /* |
7659 | * No need to migrate the tasks: it was best-effort if | |
7660 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7661 | * the requestors. | |
7662 | */ | |
1da177e4 LT |
7663 | spin_lock_irq(&rq->lock); |
7664 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7665 | struct migration_req *req; |
7666 | ||
1da177e4 | 7667 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7668 | struct migration_req, list); |
1da177e4 | 7669 | list_del_init(&req->list); |
9a2bd244 | 7670 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7671 | complete(&req->done); |
9a2bd244 | 7672 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7673 | } |
7674 | spin_unlock_irq(&rq->lock); | |
7675 | break; | |
57d885fe | 7676 | |
08f503b0 GH |
7677 | case CPU_DYING: |
7678 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7679 | /* Update our root-domain */ |
7680 | rq = cpu_rq(cpu); | |
7681 | spin_lock_irqsave(&rq->lock, flags); | |
7682 | if (rq->rd) { | |
c6c4927b | 7683 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7684 | set_rq_offline(rq); |
57d885fe GH |
7685 | } |
7686 | spin_unlock_irqrestore(&rq->lock, flags); | |
7687 | break; | |
1da177e4 LT |
7688 | #endif |
7689 | } | |
7690 | return NOTIFY_OK; | |
7691 | } | |
7692 | ||
f38b0820 PM |
7693 | /* |
7694 | * Register at high priority so that task migration (migrate_all_tasks) | |
7695 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 7696 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 7697 | */ |
26c2143b | 7698 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7699 | .notifier_call = migration_call, |
7700 | .priority = 10 | |
7701 | }; | |
7702 | ||
7babe8db | 7703 | static int __init migration_init(void) |
1da177e4 LT |
7704 | { |
7705 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7706 | int err; |
48f24c4d IM |
7707 | |
7708 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7709 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7710 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7711 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7712 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7713 | |
a004cd42 | 7714 | return 0; |
1da177e4 | 7715 | } |
7babe8db | 7716 | early_initcall(migration_init); |
1da177e4 LT |
7717 | #endif |
7718 | ||
7719 | #ifdef CONFIG_SMP | |
476f3534 | 7720 | |
3e9830dc | 7721 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7722 | |
7c16ec58 | 7723 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7724 | struct cpumask *groupmask) |
1da177e4 | 7725 | { |
4dcf6aff | 7726 | struct sched_group *group = sd->groups; |
434d53b0 | 7727 | char str[256]; |
1da177e4 | 7728 | |
968ea6d8 | 7729 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7730 | cpumask_clear(groupmask); |
4dcf6aff IM |
7731 | |
7732 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7733 | ||
7734 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7735 | printk("does not load-balance\n"); | |
7736 | if (sd->parent) | |
7737 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7738 | " has parent"); | |
7739 | return -1; | |
41c7ce9a NP |
7740 | } |
7741 | ||
eefd796a | 7742 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7743 | |
758b2cdc | 7744 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7745 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7746 | "CPU%d\n", cpu); | |
7747 | } | |
758b2cdc | 7748 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7749 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7750 | " CPU%d\n", cpu); | |
7751 | } | |
1da177e4 | 7752 | |
4dcf6aff | 7753 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7754 | do { |
4dcf6aff IM |
7755 | if (!group) { |
7756 | printk("\n"); | |
7757 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7758 | break; |
7759 | } | |
7760 | ||
18a3885f | 7761 | if (!group->cpu_power) { |
4dcf6aff IM |
7762 | printk(KERN_CONT "\n"); |
7763 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7764 | "set\n"); | |
7765 | break; | |
7766 | } | |
1da177e4 | 7767 | |
758b2cdc | 7768 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7769 | printk(KERN_CONT "\n"); |
7770 | printk(KERN_ERR "ERROR: empty group\n"); | |
7771 | break; | |
7772 | } | |
1da177e4 | 7773 | |
758b2cdc | 7774 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7775 | printk(KERN_CONT "\n"); |
7776 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7777 | break; | |
7778 | } | |
1da177e4 | 7779 | |
758b2cdc | 7780 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7781 | |
968ea6d8 | 7782 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7783 | |
7784 | printk(KERN_CONT " %s", str); | |
18a3885f PZ |
7785 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
7786 | printk(KERN_CONT " (cpu_power = %d)", | |
7787 | group->cpu_power); | |
381512cf | 7788 | } |
1da177e4 | 7789 | |
4dcf6aff IM |
7790 | group = group->next; |
7791 | } while (group != sd->groups); | |
7792 | printk(KERN_CONT "\n"); | |
1da177e4 | 7793 | |
758b2cdc | 7794 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7795 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7796 | |
758b2cdc RR |
7797 | if (sd->parent && |
7798 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7799 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7800 | "of domain->span\n"); | |
7801 | return 0; | |
7802 | } | |
1da177e4 | 7803 | |
4dcf6aff IM |
7804 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7805 | { | |
d5dd3db1 | 7806 | cpumask_var_t groupmask; |
4dcf6aff | 7807 | int level = 0; |
1da177e4 | 7808 | |
4dcf6aff IM |
7809 | if (!sd) { |
7810 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7811 | return; | |
7812 | } | |
1da177e4 | 7813 | |
4dcf6aff IM |
7814 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7815 | ||
d5dd3db1 | 7816 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7817 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7818 | return; | |
7819 | } | |
7820 | ||
4dcf6aff | 7821 | for (;;) { |
7c16ec58 | 7822 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7823 | break; |
1da177e4 LT |
7824 | level++; |
7825 | sd = sd->parent; | |
33859f7f | 7826 | if (!sd) |
4dcf6aff IM |
7827 | break; |
7828 | } | |
d5dd3db1 | 7829 | free_cpumask_var(groupmask); |
1da177e4 | 7830 | } |
6d6bc0ad | 7831 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7832 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7833 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7834 | |
1a20ff27 | 7835 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7836 | { |
758b2cdc | 7837 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7838 | return 1; |
7839 | ||
7840 | /* Following flags need at least 2 groups */ | |
7841 | if (sd->flags & (SD_LOAD_BALANCE | | |
7842 | SD_BALANCE_NEWIDLE | | |
7843 | SD_BALANCE_FORK | | |
89c4710e SS |
7844 | SD_BALANCE_EXEC | |
7845 | SD_SHARE_CPUPOWER | | |
7846 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7847 | if (sd->groups != sd->groups->next) |
7848 | return 0; | |
7849 | } | |
7850 | ||
7851 | /* Following flags don't use groups */ | |
c88d5910 | 7852 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
7853 | return 0; |
7854 | ||
7855 | return 1; | |
7856 | } | |
7857 | ||
48f24c4d IM |
7858 | static int |
7859 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7860 | { |
7861 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7862 | ||
7863 | if (sd_degenerate(parent)) | |
7864 | return 1; | |
7865 | ||
758b2cdc | 7866 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7867 | return 0; |
7868 | ||
245af2c7 SS |
7869 | /* Flags needing groups don't count if only 1 group in parent */ |
7870 | if (parent->groups == parent->groups->next) { | |
7871 | pflags &= ~(SD_LOAD_BALANCE | | |
7872 | SD_BALANCE_NEWIDLE | | |
7873 | SD_BALANCE_FORK | | |
89c4710e SS |
7874 | SD_BALANCE_EXEC | |
7875 | SD_SHARE_CPUPOWER | | |
7876 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7877 | if (nr_node_ids == 1) |
7878 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7879 | } |
7880 | if (~cflags & pflags) | |
7881 | return 0; | |
7882 | ||
7883 | return 1; | |
7884 | } | |
7885 | ||
c6c4927b RR |
7886 | static void free_rootdomain(struct root_domain *rd) |
7887 | { | |
68e74568 RR |
7888 | cpupri_cleanup(&rd->cpupri); |
7889 | ||
c6c4927b RR |
7890 | free_cpumask_var(rd->rto_mask); |
7891 | free_cpumask_var(rd->online); | |
7892 | free_cpumask_var(rd->span); | |
7893 | kfree(rd); | |
7894 | } | |
7895 | ||
57d885fe GH |
7896 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7897 | { | |
a0490fa3 | 7898 | struct root_domain *old_rd = NULL; |
57d885fe | 7899 | unsigned long flags; |
57d885fe GH |
7900 | |
7901 | spin_lock_irqsave(&rq->lock, flags); | |
7902 | ||
7903 | if (rq->rd) { | |
a0490fa3 | 7904 | old_rd = rq->rd; |
57d885fe | 7905 | |
c6c4927b | 7906 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7907 | set_rq_offline(rq); |
57d885fe | 7908 | |
c6c4927b | 7909 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7910 | |
a0490fa3 IM |
7911 | /* |
7912 | * If we dont want to free the old_rt yet then | |
7913 | * set old_rd to NULL to skip the freeing later | |
7914 | * in this function: | |
7915 | */ | |
7916 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7917 | old_rd = NULL; | |
57d885fe GH |
7918 | } |
7919 | ||
7920 | atomic_inc(&rd->refcount); | |
7921 | rq->rd = rd; | |
7922 | ||
c6c4927b | 7923 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 7924 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 7925 | set_rq_online(rq); |
57d885fe GH |
7926 | |
7927 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7928 | |
7929 | if (old_rd) | |
7930 | free_rootdomain(old_rd); | |
57d885fe GH |
7931 | } |
7932 | ||
fd5e1b5d | 7933 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7934 | { |
36b7b6d4 PE |
7935 | gfp_t gfp = GFP_KERNEL; |
7936 | ||
57d885fe GH |
7937 | memset(rd, 0, sizeof(*rd)); |
7938 | ||
36b7b6d4 PE |
7939 | if (bootmem) |
7940 | gfp = GFP_NOWAIT; | |
c6c4927b | 7941 | |
36b7b6d4 | 7942 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7943 | goto out; |
36b7b6d4 | 7944 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7945 | goto free_span; |
36b7b6d4 | 7946 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7947 | goto free_online; |
6e0534f2 | 7948 | |
0fb53029 | 7949 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7950 | goto free_rto_mask; |
c6c4927b | 7951 | return 0; |
6e0534f2 | 7952 | |
68e74568 RR |
7953 | free_rto_mask: |
7954 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7955 | free_online: |
7956 | free_cpumask_var(rd->online); | |
7957 | free_span: | |
7958 | free_cpumask_var(rd->span); | |
0c910d28 | 7959 | out: |
c6c4927b | 7960 | return -ENOMEM; |
57d885fe GH |
7961 | } |
7962 | ||
7963 | static void init_defrootdomain(void) | |
7964 | { | |
c6c4927b RR |
7965 | init_rootdomain(&def_root_domain, true); |
7966 | ||
57d885fe GH |
7967 | atomic_set(&def_root_domain.refcount, 1); |
7968 | } | |
7969 | ||
dc938520 | 7970 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7971 | { |
7972 | struct root_domain *rd; | |
7973 | ||
7974 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7975 | if (!rd) | |
7976 | return NULL; | |
7977 | ||
c6c4927b RR |
7978 | if (init_rootdomain(rd, false) != 0) { |
7979 | kfree(rd); | |
7980 | return NULL; | |
7981 | } | |
57d885fe GH |
7982 | |
7983 | return rd; | |
7984 | } | |
7985 | ||
1da177e4 | 7986 | /* |
0eab9146 | 7987 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7988 | * hold the hotplug lock. |
7989 | */ | |
0eab9146 IM |
7990 | static void |
7991 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7992 | { |
70b97a7f | 7993 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7994 | struct sched_domain *tmp; |
7995 | ||
7996 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7997 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7998 | struct sched_domain *parent = tmp->parent; |
7999 | if (!parent) | |
8000 | break; | |
f29c9b1c | 8001 | |
1a848870 | 8002 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8003 | tmp->parent = parent->parent; |
1a848870 SS |
8004 | if (parent->parent) |
8005 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8006 | } else |
8007 | tmp = tmp->parent; | |
245af2c7 SS |
8008 | } |
8009 | ||
1a848870 | 8010 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8011 | sd = sd->parent; |
1a848870 SS |
8012 | if (sd) |
8013 | sd->child = NULL; | |
8014 | } | |
1da177e4 LT |
8015 | |
8016 | sched_domain_debug(sd, cpu); | |
8017 | ||
57d885fe | 8018 | rq_attach_root(rq, rd); |
674311d5 | 8019 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8020 | } |
8021 | ||
8022 | /* cpus with isolated domains */ | |
dcc30a35 | 8023 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8024 | |
8025 | /* Setup the mask of cpus configured for isolated domains */ | |
8026 | static int __init isolated_cpu_setup(char *str) | |
8027 | { | |
968ea6d8 | 8028 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8029 | return 1; |
8030 | } | |
8031 | ||
8927f494 | 8032 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8033 | |
8034 | /* | |
6711cab4 SS |
8035 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8036 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8037 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8038 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8039 | * |
8040 | * init_sched_build_groups will build a circular linked list of the groups | |
8041 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8042 | * and ->cpu_power to 0. | |
8043 | */ | |
a616058b | 8044 | static void |
96f874e2 RR |
8045 | init_sched_build_groups(const struct cpumask *span, |
8046 | const struct cpumask *cpu_map, | |
8047 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8048 | struct sched_group **sg, |
96f874e2 RR |
8049 | struct cpumask *tmpmask), |
8050 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8051 | { |
8052 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8053 | int i; |
8054 | ||
96f874e2 | 8055 | cpumask_clear(covered); |
7c16ec58 | 8056 | |
abcd083a | 8057 | for_each_cpu(i, span) { |
6711cab4 | 8058 | struct sched_group *sg; |
7c16ec58 | 8059 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8060 | int j; |
8061 | ||
758b2cdc | 8062 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8063 | continue; |
8064 | ||
758b2cdc | 8065 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 8066 | sg->cpu_power = 0; |
1da177e4 | 8067 | |
abcd083a | 8068 | for_each_cpu(j, span) { |
7c16ec58 | 8069 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8070 | continue; |
8071 | ||
96f874e2 | 8072 | cpumask_set_cpu(j, covered); |
758b2cdc | 8073 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8074 | } |
8075 | if (!first) | |
8076 | first = sg; | |
8077 | if (last) | |
8078 | last->next = sg; | |
8079 | last = sg; | |
8080 | } | |
8081 | last->next = first; | |
8082 | } | |
8083 | ||
9c1cfda2 | 8084 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8085 | |
9c1cfda2 | 8086 | #ifdef CONFIG_NUMA |
198e2f18 | 8087 | |
9c1cfda2 JH |
8088 | /** |
8089 | * find_next_best_node - find the next node to include in a sched_domain | |
8090 | * @node: node whose sched_domain we're building | |
8091 | * @used_nodes: nodes already in the sched_domain | |
8092 | * | |
41a2d6cf | 8093 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8094 | * finds the closest node not already in the @used_nodes map. |
8095 | * | |
8096 | * Should use nodemask_t. | |
8097 | */ | |
c5f59f08 | 8098 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8099 | { |
8100 | int i, n, val, min_val, best_node = 0; | |
8101 | ||
8102 | min_val = INT_MAX; | |
8103 | ||
076ac2af | 8104 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8105 | /* Start at @node */ |
076ac2af | 8106 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8107 | |
8108 | if (!nr_cpus_node(n)) | |
8109 | continue; | |
8110 | ||
8111 | /* Skip already used nodes */ | |
c5f59f08 | 8112 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8113 | continue; |
8114 | ||
8115 | /* Simple min distance search */ | |
8116 | val = node_distance(node, n); | |
8117 | ||
8118 | if (val < min_val) { | |
8119 | min_val = val; | |
8120 | best_node = n; | |
8121 | } | |
8122 | } | |
8123 | ||
c5f59f08 | 8124 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8125 | return best_node; |
8126 | } | |
8127 | ||
8128 | /** | |
8129 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8130 | * @node: node whose cpumask we're constructing | |
73486722 | 8131 | * @span: resulting cpumask |
9c1cfda2 | 8132 | * |
41a2d6cf | 8133 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8134 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8135 | * out optimally. | |
8136 | */ | |
96f874e2 | 8137 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8138 | { |
c5f59f08 | 8139 | nodemask_t used_nodes; |
48f24c4d | 8140 | int i; |
9c1cfda2 | 8141 | |
6ca09dfc | 8142 | cpumask_clear(span); |
c5f59f08 | 8143 | nodes_clear(used_nodes); |
9c1cfda2 | 8144 | |
6ca09dfc | 8145 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8146 | node_set(node, used_nodes); |
9c1cfda2 JH |
8147 | |
8148 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8149 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8150 | |
6ca09dfc | 8151 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8152 | } |
9c1cfda2 | 8153 | } |
6d6bc0ad | 8154 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8155 | |
5c45bf27 | 8156 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8157 | |
6c99e9ad RR |
8158 | /* |
8159 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8160 | * |
8161 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8162 | * and struct sched_domain. ) | |
6c99e9ad RR |
8163 | */ |
8164 | struct static_sched_group { | |
8165 | struct sched_group sg; | |
8166 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8167 | }; | |
8168 | ||
8169 | struct static_sched_domain { | |
8170 | struct sched_domain sd; | |
8171 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8172 | }; | |
8173 | ||
49a02c51 AH |
8174 | struct s_data { |
8175 | #ifdef CONFIG_NUMA | |
8176 | int sd_allnodes; | |
8177 | cpumask_var_t domainspan; | |
8178 | cpumask_var_t covered; | |
8179 | cpumask_var_t notcovered; | |
8180 | #endif | |
8181 | cpumask_var_t nodemask; | |
8182 | cpumask_var_t this_sibling_map; | |
8183 | cpumask_var_t this_core_map; | |
8184 | cpumask_var_t send_covered; | |
8185 | cpumask_var_t tmpmask; | |
8186 | struct sched_group **sched_group_nodes; | |
8187 | struct root_domain *rd; | |
8188 | }; | |
8189 | ||
2109b99e AH |
8190 | enum s_alloc { |
8191 | sa_sched_groups = 0, | |
8192 | sa_rootdomain, | |
8193 | sa_tmpmask, | |
8194 | sa_send_covered, | |
8195 | sa_this_core_map, | |
8196 | sa_this_sibling_map, | |
8197 | sa_nodemask, | |
8198 | sa_sched_group_nodes, | |
8199 | #ifdef CONFIG_NUMA | |
8200 | sa_notcovered, | |
8201 | sa_covered, | |
8202 | sa_domainspan, | |
8203 | #endif | |
8204 | sa_none, | |
8205 | }; | |
8206 | ||
9c1cfda2 | 8207 | /* |
48f24c4d | 8208 | * SMT sched-domains: |
9c1cfda2 | 8209 | */ |
1da177e4 | 8210 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8211 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8212 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8213 | |
41a2d6cf | 8214 | static int |
96f874e2 RR |
8215 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8216 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8217 | { |
6711cab4 | 8218 | if (sg) |
6c99e9ad | 8219 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8220 | return cpu; |
8221 | } | |
6d6bc0ad | 8222 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8223 | |
48f24c4d IM |
8224 | /* |
8225 | * multi-core sched-domains: | |
8226 | */ | |
1e9f28fa | 8227 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8228 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8229 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8230 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8231 | |
8232 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8233 | static int |
96f874e2 RR |
8234 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8235 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8236 | { |
6711cab4 | 8237 | int group; |
7c16ec58 | 8238 | |
c69fc56d | 8239 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8240 | group = cpumask_first(mask); |
6711cab4 | 8241 | if (sg) |
6c99e9ad | 8242 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8243 | return group; |
1e9f28fa SS |
8244 | } |
8245 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8246 | static int |
96f874e2 RR |
8247 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8248 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8249 | { |
6711cab4 | 8250 | if (sg) |
6c99e9ad | 8251 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8252 | return cpu; |
8253 | } | |
8254 | #endif | |
8255 | ||
6c99e9ad RR |
8256 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8257 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8258 | |
41a2d6cf | 8259 | static int |
96f874e2 RR |
8260 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8261 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8262 | { |
6711cab4 | 8263 | int group; |
48f24c4d | 8264 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8265 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8266 | group = cpumask_first(mask); |
1e9f28fa | 8267 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8268 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8269 | group = cpumask_first(mask); |
1da177e4 | 8270 | #else |
6711cab4 | 8271 | group = cpu; |
1da177e4 | 8272 | #endif |
6711cab4 | 8273 | if (sg) |
6c99e9ad | 8274 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8275 | return group; |
1da177e4 LT |
8276 | } |
8277 | ||
8278 | #ifdef CONFIG_NUMA | |
1da177e4 | 8279 | /* |
9c1cfda2 JH |
8280 | * The init_sched_build_groups can't handle what we want to do with node |
8281 | * groups, so roll our own. Now each node has its own list of groups which | |
8282 | * gets dynamically allocated. | |
1da177e4 | 8283 | */ |
62ea9ceb | 8284 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8285 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8286 | |
62ea9ceb | 8287 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8288 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8289 | |
96f874e2 RR |
8290 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8291 | struct sched_group **sg, | |
8292 | struct cpumask *nodemask) | |
9c1cfda2 | 8293 | { |
6711cab4 SS |
8294 | int group; |
8295 | ||
6ca09dfc | 8296 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8297 | group = cpumask_first(nodemask); |
6711cab4 SS |
8298 | |
8299 | if (sg) | |
6c99e9ad | 8300 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8301 | return group; |
1da177e4 | 8302 | } |
6711cab4 | 8303 | |
08069033 SS |
8304 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8305 | { | |
8306 | struct sched_group *sg = group_head; | |
8307 | int j; | |
8308 | ||
8309 | if (!sg) | |
8310 | return; | |
3a5c359a | 8311 | do { |
758b2cdc | 8312 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8313 | struct sched_domain *sd; |
08069033 | 8314 | |
6c99e9ad | 8315 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8316 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8317 | /* |
8318 | * Only add "power" once for each | |
8319 | * physical package. | |
8320 | */ | |
8321 | continue; | |
8322 | } | |
08069033 | 8323 | |
18a3885f | 8324 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
8325 | } |
8326 | sg = sg->next; | |
8327 | } while (sg != group_head); | |
08069033 | 8328 | } |
0601a88d AH |
8329 | |
8330 | static int build_numa_sched_groups(struct s_data *d, | |
8331 | const struct cpumask *cpu_map, int num) | |
8332 | { | |
8333 | struct sched_domain *sd; | |
8334 | struct sched_group *sg, *prev; | |
8335 | int n, j; | |
8336 | ||
8337 | cpumask_clear(d->covered); | |
8338 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8339 | if (cpumask_empty(d->nodemask)) { | |
8340 | d->sched_group_nodes[num] = NULL; | |
8341 | goto out; | |
8342 | } | |
8343 | ||
8344 | sched_domain_node_span(num, d->domainspan); | |
8345 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8346 | ||
8347 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8348 | GFP_KERNEL, num); | |
8349 | if (!sg) { | |
8350 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", | |
8351 | num); | |
8352 | return -ENOMEM; | |
8353 | } | |
8354 | d->sched_group_nodes[num] = sg; | |
8355 | ||
8356 | for_each_cpu(j, d->nodemask) { | |
8357 | sd = &per_cpu(node_domains, j).sd; | |
8358 | sd->groups = sg; | |
8359 | } | |
8360 | ||
18a3885f | 8361 | sg->cpu_power = 0; |
0601a88d AH |
8362 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
8363 | sg->next = sg; | |
8364 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8365 | ||
8366 | prev = sg; | |
8367 | for (j = 0; j < nr_node_ids; j++) { | |
8368 | n = (num + j) % nr_node_ids; | |
8369 | cpumask_complement(d->notcovered, d->covered); | |
8370 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8371 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8372 | if (cpumask_empty(d->tmpmask)) | |
8373 | break; | |
8374 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8375 | if (cpumask_empty(d->tmpmask)) | |
8376 | continue; | |
8377 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8378 | GFP_KERNEL, num); | |
8379 | if (!sg) { | |
8380 | printk(KERN_WARNING | |
8381 | "Can not alloc domain group for node %d\n", j); | |
8382 | return -ENOMEM; | |
8383 | } | |
18a3885f | 8384 | sg->cpu_power = 0; |
0601a88d AH |
8385 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
8386 | sg->next = prev->next; | |
8387 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8388 | prev->next = sg; | |
8389 | prev = sg; | |
8390 | } | |
8391 | out: | |
8392 | return 0; | |
8393 | } | |
6d6bc0ad | 8394 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8395 | |
a616058b | 8396 | #ifdef CONFIG_NUMA |
51888ca2 | 8397 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8398 | static void free_sched_groups(const struct cpumask *cpu_map, |
8399 | struct cpumask *nodemask) | |
51888ca2 | 8400 | { |
a616058b | 8401 | int cpu, i; |
51888ca2 | 8402 | |
abcd083a | 8403 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8404 | struct sched_group **sched_group_nodes |
8405 | = sched_group_nodes_bycpu[cpu]; | |
8406 | ||
51888ca2 SV |
8407 | if (!sched_group_nodes) |
8408 | continue; | |
8409 | ||
076ac2af | 8410 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8411 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8412 | ||
6ca09dfc | 8413 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8414 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8415 | continue; |
8416 | ||
8417 | if (sg == NULL) | |
8418 | continue; | |
8419 | sg = sg->next; | |
8420 | next_sg: | |
8421 | oldsg = sg; | |
8422 | sg = sg->next; | |
8423 | kfree(oldsg); | |
8424 | if (oldsg != sched_group_nodes[i]) | |
8425 | goto next_sg; | |
8426 | } | |
8427 | kfree(sched_group_nodes); | |
8428 | sched_group_nodes_bycpu[cpu] = NULL; | |
8429 | } | |
51888ca2 | 8430 | } |
6d6bc0ad | 8431 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8432 | static void free_sched_groups(const struct cpumask *cpu_map, |
8433 | struct cpumask *nodemask) | |
a616058b SS |
8434 | { |
8435 | } | |
6d6bc0ad | 8436 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8437 | |
89c4710e SS |
8438 | /* |
8439 | * Initialize sched groups cpu_power. | |
8440 | * | |
8441 | * cpu_power indicates the capacity of sched group, which is used while | |
8442 | * distributing the load between different sched groups in a sched domain. | |
8443 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8444 | * there are asymmetries in the topology. If there are asymmetries, group | |
8445 | * having more cpu_power will pickup more load compared to the group having | |
8446 | * less cpu_power. | |
89c4710e SS |
8447 | */ |
8448 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8449 | { | |
8450 | struct sched_domain *child; | |
8451 | struct sched_group *group; | |
f93e65c1 PZ |
8452 | long power; |
8453 | int weight; | |
89c4710e SS |
8454 | |
8455 | WARN_ON(!sd || !sd->groups); | |
8456 | ||
13318a71 | 8457 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8458 | return; |
8459 | ||
8460 | child = sd->child; | |
8461 | ||
18a3885f | 8462 | sd->groups->cpu_power = 0; |
5517d86b | 8463 | |
f93e65c1 PZ |
8464 | if (!child) { |
8465 | power = SCHED_LOAD_SCALE; | |
8466 | weight = cpumask_weight(sched_domain_span(sd)); | |
8467 | /* | |
8468 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8469 | * Usually multiple threads get a better yield out of |
8470 | * that one core than a single thread would have, | |
8471 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8472 | */ |
a52bfd73 PZ |
8473 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8474 | power *= sd->smt_gain; | |
f93e65c1 | 8475 | power /= weight; |
a52bfd73 PZ |
8476 | power >>= SCHED_LOAD_SHIFT; |
8477 | } | |
18a3885f | 8478 | sd->groups->cpu_power += power; |
89c4710e SS |
8479 | return; |
8480 | } | |
8481 | ||
89c4710e | 8482 | /* |
f93e65c1 | 8483 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8484 | */ |
8485 | group = child->groups; | |
8486 | do { | |
18a3885f | 8487 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
8488 | group = group->next; |
8489 | } while (group != child->groups); | |
8490 | } | |
8491 | ||
7c16ec58 MT |
8492 | /* |
8493 | * Initializers for schedule domains | |
8494 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8495 | */ | |
8496 | ||
a5d8c348 IM |
8497 | #ifdef CONFIG_SCHED_DEBUG |
8498 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8499 | #else | |
8500 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8501 | #endif | |
8502 | ||
7c16ec58 | 8503 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8504 | |
7c16ec58 MT |
8505 | #define SD_INIT_FUNC(type) \ |
8506 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8507 | { \ | |
8508 | memset(sd, 0, sizeof(*sd)); \ | |
8509 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8510 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8511 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8512 | } |
8513 | ||
8514 | SD_INIT_FUNC(CPU) | |
8515 | #ifdef CONFIG_NUMA | |
8516 | SD_INIT_FUNC(ALLNODES) | |
8517 | SD_INIT_FUNC(NODE) | |
8518 | #endif | |
8519 | #ifdef CONFIG_SCHED_SMT | |
8520 | SD_INIT_FUNC(SIBLING) | |
8521 | #endif | |
8522 | #ifdef CONFIG_SCHED_MC | |
8523 | SD_INIT_FUNC(MC) | |
8524 | #endif | |
8525 | ||
1d3504fc HS |
8526 | static int default_relax_domain_level = -1; |
8527 | ||
8528 | static int __init setup_relax_domain_level(char *str) | |
8529 | { | |
30e0e178 LZ |
8530 | unsigned long val; |
8531 | ||
8532 | val = simple_strtoul(str, NULL, 0); | |
8533 | if (val < SD_LV_MAX) | |
8534 | default_relax_domain_level = val; | |
8535 | ||
1d3504fc HS |
8536 | return 1; |
8537 | } | |
8538 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8539 | ||
8540 | static void set_domain_attribute(struct sched_domain *sd, | |
8541 | struct sched_domain_attr *attr) | |
8542 | { | |
8543 | int request; | |
8544 | ||
8545 | if (!attr || attr->relax_domain_level < 0) { | |
8546 | if (default_relax_domain_level < 0) | |
8547 | return; | |
8548 | else | |
8549 | request = default_relax_domain_level; | |
8550 | } else | |
8551 | request = attr->relax_domain_level; | |
8552 | if (request < sd->level) { | |
8553 | /* turn off idle balance on this domain */ | |
c88d5910 | 8554 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8555 | } else { |
8556 | /* turn on idle balance on this domain */ | |
c88d5910 | 8557 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8558 | } |
8559 | } | |
8560 | ||
2109b99e AH |
8561 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8562 | const struct cpumask *cpu_map) | |
8563 | { | |
8564 | switch (what) { | |
8565 | case sa_sched_groups: | |
8566 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8567 | d->sched_group_nodes = NULL; | |
8568 | case sa_rootdomain: | |
8569 | free_rootdomain(d->rd); /* fall through */ | |
8570 | case sa_tmpmask: | |
8571 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8572 | case sa_send_covered: | |
8573 | free_cpumask_var(d->send_covered); /* fall through */ | |
8574 | case sa_this_core_map: | |
8575 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8576 | case sa_this_sibling_map: | |
8577 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8578 | case sa_nodemask: | |
8579 | free_cpumask_var(d->nodemask); /* fall through */ | |
8580 | case sa_sched_group_nodes: | |
d1b55138 | 8581 | #ifdef CONFIG_NUMA |
2109b99e AH |
8582 | kfree(d->sched_group_nodes); /* fall through */ |
8583 | case sa_notcovered: | |
8584 | free_cpumask_var(d->notcovered); /* fall through */ | |
8585 | case sa_covered: | |
8586 | free_cpumask_var(d->covered); /* fall through */ | |
8587 | case sa_domainspan: | |
8588 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8589 | #endif |
2109b99e AH |
8590 | case sa_none: |
8591 | break; | |
8592 | } | |
8593 | } | |
3404c8d9 | 8594 | |
2109b99e AH |
8595 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8596 | const struct cpumask *cpu_map) | |
8597 | { | |
3404c8d9 | 8598 | #ifdef CONFIG_NUMA |
2109b99e AH |
8599 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8600 | return sa_none; | |
8601 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8602 | return sa_domainspan; | |
8603 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8604 | return sa_covered; | |
8605 | /* Allocate the per-node list of sched groups */ | |
8606 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8607 | sizeof(struct sched_group *), GFP_KERNEL); | |
8608 | if (!d->sched_group_nodes) { | |
d1b55138 | 8609 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 8610 | return sa_notcovered; |
d1b55138 | 8611 | } |
2109b99e | 8612 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8613 | #endif |
2109b99e AH |
8614 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8615 | return sa_sched_group_nodes; | |
8616 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8617 | return sa_nodemask; | |
8618 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8619 | return sa_this_sibling_map; | |
8620 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8621 | return sa_this_core_map; | |
8622 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8623 | return sa_send_covered; | |
8624 | d->rd = alloc_rootdomain(); | |
8625 | if (!d->rd) { | |
57d885fe | 8626 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 8627 | return sa_tmpmask; |
57d885fe | 8628 | } |
2109b99e AH |
8629 | return sa_rootdomain; |
8630 | } | |
57d885fe | 8631 | |
7f4588f3 AH |
8632 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8633 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8634 | { | |
8635 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8636 | #ifdef CONFIG_NUMA |
7f4588f3 | 8637 | struct sched_domain *parent; |
1da177e4 | 8638 | |
7f4588f3 AH |
8639 | d->sd_allnodes = 0; |
8640 | if (cpumask_weight(cpu_map) > | |
8641 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8642 | sd = &per_cpu(allnodes_domains, i).sd; | |
8643 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8644 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8645 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8646 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8647 | d->sd_allnodes = 1; | |
8648 | } | |
8649 | parent = sd; | |
8650 | ||
8651 | sd = &per_cpu(node_domains, i).sd; | |
8652 | SD_INIT(sd, NODE); | |
8653 | set_domain_attribute(sd, attr); | |
8654 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8655 | sd->parent = parent; | |
8656 | if (parent) | |
8657 | parent->child = sd; | |
8658 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8659 | #endif |
7f4588f3 AH |
8660 | return sd; |
8661 | } | |
1da177e4 | 8662 | |
87cce662 AH |
8663 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8664 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8665 | struct sched_domain *parent, int i) | |
8666 | { | |
8667 | struct sched_domain *sd; | |
8668 | sd = &per_cpu(phys_domains, i).sd; | |
8669 | SD_INIT(sd, CPU); | |
8670 | set_domain_attribute(sd, attr); | |
8671 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8672 | sd->parent = parent; | |
8673 | if (parent) | |
8674 | parent->child = sd; | |
8675 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8676 | return sd; | |
8677 | } | |
1da177e4 | 8678 | |
410c4081 AH |
8679 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8680 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8681 | struct sched_domain *parent, int i) | |
8682 | { | |
8683 | struct sched_domain *sd = parent; | |
1e9f28fa | 8684 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8685 | sd = &per_cpu(core_domains, i).sd; |
8686 | SD_INIT(sd, MC); | |
8687 | set_domain_attribute(sd, attr); | |
8688 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8689 | sd->parent = parent; | |
8690 | parent->child = sd; | |
8691 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8692 | #endif |
410c4081 AH |
8693 | return sd; |
8694 | } | |
1e9f28fa | 8695 | |
d8173535 AH |
8696 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8697 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8698 | struct sched_domain *parent, int i) | |
8699 | { | |
8700 | struct sched_domain *sd = parent; | |
1da177e4 | 8701 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8702 | sd = &per_cpu(cpu_domains, i).sd; |
8703 | SD_INIT(sd, SIBLING); | |
8704 | set_domain_attribute(sd, attr); | |
8705 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8706 | sd->parent = parent; | |
8707 | parent->child = sd; | |
8708 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8709 | #endif |
d8173535 AH |
8710 | return sd; |
8711 | } | |
1da177e4 | 8712 | |
0e8e85c9 AH |
8713 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8714 | const struct cpumask *cpu_map, int cpu) | |
8715 | { | |
8716 | switch (l) { | |
1da177e4 | 8717 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8718 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8719 | cpumask_and(d->this_sibling_map, cpu_map, | |
8720 | topology_thread_cpumask(cpu)); | |
8721 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8722 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8723 | &cpu_to_cpu_group, | |
8724 | d->send_covered, d->tmpmask); | |
8725 | break; | |
1da177e4 | 8726 | #endif |
1e9f28fa | 8727 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8728 | case SD_LV_MC: /* set up multi-core groups */ |
8729 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8730 | if (cpu == cpumask_first(d->this_core_map)) | |
8731 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8732 | &cpu_to_core_group, | |
8733 | d->send_covered, d->tmpmask); | |
8734 | break; | |
1e9f28fa | 8735 | #endif |
86548096 AH |
8736 | case SD_LV_CPU: /* set up physical groups */ |
8737 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8738 | if (!cpumask_empty(d->nodemask)) | |
8739 | init_sched_build_groups(d->nodemask, cpu_map, | |
8740 | &cpu_to_phys_group, | |
8741 | d->send_covered, d->tmpmask); | |
8742 | break; | |
1da177e4 | 8743 | #ifdef CONFIG_NUMA |
de616e36 AH |
8744 | case SD_LV_ALLNODES: |
8745 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8746 | d->send_covered, d->tmpmask); | |
8747 | break; | |
8748 | #endif | |
0e8e85c9 AH |
8749 | default: |
8750 | break; | |
7c16ec58 | 8751 | } |
0e8e85c9 | 8752 | } |
9c1cfda2 | 8753 | |
2109b99e AH |
8754 | /* |
8755 | * Build sched domains for a given set of cpus and attach the sched domains | |
8756 | * to the individual cpus | |
8757 | */ | |
8758 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8759 | struct sched_domain_attr *attr) | |
8760 | { | |
8761 | enum s_alloc alloc_state = sa_none; | |
8762 | struct s_data d; | |
294b0c96 | 8763 | struct sched_domain *sd; |
2109b99e | 8764 | int i; |
7c16ec58 | 8765 | #ifdef CONFIG_NUMA |
2109b99e | 8766 | d.sd_allnodes = 0; |
7c16ec58 | 8767 | #endif |
9c1cfda2 | 8768 | |
2109b99e AH |
8769 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8770 | if (alloc_state != sa_rootdomain) | |
8771 | goto error; | |
8772 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8773 | |
1da177e4 | 8774 | /* |
1a20ff27 | 8775 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8776 | */ |
abcd083a | 8777 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8778 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8779 | cpu_map); | |
9761eea8 | 8780 | |
7f4588f3 | 8781 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8782 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8783 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8784 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8785 | } |
9c1cfda2 | 8786 | |
abcd083a | 8787 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8788 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8789 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8790 | } |
9c1cfda2 | 8791 | |
1da177e4 | 8792 | /* Set up physical groups */ |
86548096 AH |
8793 | for (i = 0; i < nr_node_ids; i++) |
8794 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8795 | |
1da177e4 LT |
8796 | #ifdef CONFIG_NUMA |
8797 | /* Set up node groups */ | |
de616e36 AH |
8798 | if (d.sd_allnodes) |
8799 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8800 | |
0601a88d AH |
8801 | for (i = 0; i < nr_node_ids; i++) |
8802 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8803 | goto error; |
1da177e4 LT |
8804 | #endif |
8805 | ||
8806 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8807 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8808 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8809 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8810 | init_sched_groups_power(i, sd); |
5c45bf27 | 8811 | } |
1da177e4 | 8812 | #endif |
1e9f28fa | 8813 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8814 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8815 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8816 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8817 | } |
8818 | #endif | |
1e9f28fa | 8819 | |
abcd083a | 8820 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8821 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8822 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8823 | } |
8824 | ||
9c1cfda2 | 8825 | #ifdef CONFIG_NUMA |
076ac2af | 8826 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8827 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8828 | |
49a02c51 | 8829 | if (d.sd_allnodes) { |
6711cab4 | 8830 | struct sched_group *sg; |
f712c0c7 | 8831 | |
96f874e2 | 8832 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 8833 | d.tmpmask); |
f712c0c7 SS |
8834 | init_numa_sched_groups_power(sg); |
8835 | } | |
9c1cfda2 JH |
8836 | #endif |
8837 | ||
1da177e4 | 8838 | /* Attach the domains */ |
abcd083a | 8839 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8840 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8841 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8842 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8843 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8844 | #else |
6c99e9ad | 8845 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8846 | #endif |
49a02c51 | 8847 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 8848 | } |
51888ca2 | 8849 | |
2109b99e AH |
8850 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
8851 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
8852 | return 0; | |
51888ca2 | 8853 | |
51888ca2 | 8854 | error: |
2109b99e AH |
8855 | __free_domain_allocs(&d, alloc_state, cpu_map); |
8856 | return -ENOMEM; | |
1da177e4 | 8857 | } |
029190c5 | 8858 | |
96f874e2 | 8859 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8860 | { |
8861 | return __build_sched_domains(cpu_map, NULL); | |
8862 | } | |
8863 | ||
acc3f5d7 | 8864 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 8865 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8866 | static struct sched_domain_attr *dattr_cur; |
8867 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8868 | |
8869 | /* | |
8870 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8871 | * cpumask) fails, then fallback to a single sched domain, |
8872 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8873 | */ |
4212823f | 8874 | static cpumask_var_t fallback_doms; |
029190c5 | 8875 | |
ee79d1bd HC |
8876 | /* |
8877 | * arch_update_cpu_topology lets virtualized architectures update the | |
8878 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8879 | * or 0 if it stayed the same. | |
8880 | */ | |
8881 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8882 | { |
ee79d1bd | 8883 | return 0; |
22e52b07 HC |
8884 | } |
8885 | ||
acc3f5d7 RR |
8886 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
8887 | { | |
8888 | int i; | |
8889 | cpumask_var_t *doms; | |
8890 | ||
8891 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
8892 | if (!doms) | |
8893 | return NULL; | |
8894 | for (i = 0; i < ndoms; i++) { | |
8895 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
8896 | free_sched_domains(doms, i); | |
8897 | return NULL; | |
8898 | } | |
8899 | } | |
8900 | return doms; | |
8901 | } | |
8902 | ||
8903 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
8904 | { | |
8905 | unsigned int i; | |
8906 | for (i = 0; i < ndoms; i++) | |
8907 | free_cpumask_var(doms[i]); | |
8908 | kfree(doms); | |
8909 | } | |
8910 | ||
1a20ff27 | 8911 | /* |
41a2d6cf | 8912 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8913 | * For now this just excludes isolated cpus, but could be used to |
8914 | * exclude other special cases in the future. | |
1a20ff27 | 8915 | */ |
96f874e2 | 8916 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8917 | { |
7378547f MM |
8918 | int err; |
8919 | ||
22e52b07 | 8920 | arch_update_cpu_topology(); |
029190c5 | 8921 | ndoms_cur = 1; |
acc3f5d7 | 8922 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 8923 | if (!doms_cur) |
acc3f5d7 RR |
8924 | doms_cur = &fallback_doms; |
8925 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 8926 | dattr_cur = NULL; |
acc3f5d7 | 8927 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 8928 | register_sched_domain_sysctl(); |
7378547f MM |
8929 | |
8930 | return err; | |
1a20ff27 DG |
8931 | } |
8932 | ||
96f874e2 RR |
8933 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8934 | struct cpumask *tmpmask) | |
1da177e4 | 8935 | { |
7c16ec58 | 8936 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8937 | } |
1da177e4 | 8938 | |
1a20ff27 DG |
8939 | /* |
8940 | * Detach sched domains from a group of cpus specified in cpu_map | |
8941 | * These cpus will now be attached to the NULL domain | |
8942 | */ | |
96f874e2 | 8943 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8944 | { |
96f874e2 RR |
8945 | /* Save because hotplug lock held. */ |
8946 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8947 | int i; |
8948 | ||
abcd083a | 8949 | for_each_cpu(i, cpu_map) |
57d885fe | 8950 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8951 | synchronize_sched(); |
96f874e2 | 8952 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8953 | } |
8954 | ||
1d3504fc HS |
8955 | /* handle null as "default" */ |
8956 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8957 | struct sched_domain_attr *new, int idx_new) | |
8958 | { | |
8959 | struct sched_domain_attr tmp; | |
8960 | ||
8961 | /* fast path */ | |
8962 | if (!new && !cur) | |
8963 | return 1; | |
8964 | ||
8965 | tmp = SD_ATTR_INIT; | |
8966 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8967 | new ? (new + idx_new) : &tmp, | |
8968 | sizeof(struct sched_domain_attr)); | |
8969 | } | |
8970 | ||
029190c5 PJ |
8971 | /* |
8972 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8973 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8974 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8975 | * It destroys each deleted domain and builds each new domain. | |
8976 | * | |
acc3f5d7 | 8977 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
8978 | * The masks don't intersect (don't overlap.) We should setup one |
8979 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8980 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8981 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8982 | * it as it is. | |
8983 | * | |
acc3f5d7 RR |
8984 | * The passed in 'doms_new' should be allocated using |
8985 | * alloc_sched_domains. This routine takes ownership of it and will | |
8986 | * free_sched_domains it when done with it. If the caller failed the | |
8987 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
8988 | * and partition_sched_domains() will fallback to the single partition | |
8989 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 8990 | * |
96f874e2 | 8991 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8992 | * ndoms_new == 0 is a special case for destroying existing domains, |
8993 | * and it will not create the default domain. | |
dfb512ec | 8994 | * |
029190c5 PJ |
8995 | * Call with hotplug lock held |
8996 | */ | |
acc3f5d7 | 8997 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 8998 | struct sched_domain_attr *dattr_new) |
029190c5 | 8999 | { |
dfb512ec | 9000 | int i, j, n; |
d65bd5ec | 9001 | int new_topology; |
029190c5 | 9002 | |
712555ee | 9003 | mutex_lock(&sched_domains_mutex); |
a1835615 | 9004 | |
7378547f MM |
9005 | /* always unregister in case we don't destroy any domains */ |
9006 | unregister_sched_domain_sysctl(); | |
9007 | ||
d65bd5ec HC |
9008 | /* Let architecture update cpu core mappings. */ |
9009 | new_topology = arch_update_cpu_topology(); | |
9010 | ||
dfb512ec | 9011 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
9012 | |
9013 | /* Destroy deleted domains */ | |
9014 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 9015 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 9016 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 9017 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9018 | goto match1; |
9019 | } | |
9020 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 9021 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
9022 | match1: |
9023 | ; | |
9024 | } | |
9025 | ||
e761b772 MK |
9026 | if (doms_new == NULL) { |
9027 | ndoms_cur = 0; | |
acc3f5d7 RR |
9028 | doms_new = &fallback_doms; |
9029 | cpumask_andnot(doms_new[0], cpu_online_mask, cpu_isolated_map); | |
faa2f98f | 9030 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9031 | } |
9032 | ||
029190c5 PJ |
9033 | /* Build new domains */ |
9034 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9035 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 9036 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 9037 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9038 | goto match2; |
9039 | } | |
9040 | /* no match - add a new doms_new */ | |
acc3f5d7 | 9041 | __build_sched_domains(doms_new[i], |
1d3504fc | 9042 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
9043 | match2: |
9044 | ; | |
9045 | } | |
9046 | ||
9047 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
9048 | if (doms_cur != &fallback_doms) |
9049 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 9050 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9051 | doms_cur = doms_new; |
1d3504fc | 9052 | dattr_cur = dattr_new; |
029190c5 | 9053 | ndoms_cur = ndoms_new; |
7378547f MM |
9054 | |
9055 | register_sched_domain_sysctl(); | |
a1835615 | 9056 | |
712555ee | 9057 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9058 | } |
9059 | ||
5c45bf27 | 9060 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9061 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9062 | { |
95402b38 | 9063 | get_online_cpus(); |
dfb512ec MK |
9064 | |
9065 | /* Destroy domains first to force the rebuild */ | |
9066 | partition_sched_domains(0, NULL, NULL); | |
9067 | ||
e761b772 | 9068 | rebuild_sched_domains(); |
95402b38 | 9069 | put_online_cpus(); |
5c45bf27 SS |
9070 | } |
9071 | ||
9072 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9073 | { | |
afb8a9b7 | 9074 | unsigned int level = 0; |
5c45bf27 | 9075 | |
afb8a9b7 GS |
9076 | if (sscanf(buf, "%u", &level) != 1) |
9077 | return -EINVAL; | |
9078 | ||
9079 | /* | |
9080 | * level is always be positive so don't check for | |
9081 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9082 | * What happens on 0 or 1 byte write, | |
9083 | * need to check for count as well? | |
9084 | */ | |
9085 | ||
9086 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9087 | return -EINVAL; |
9088 | ||
9089 | if (smt) | |
afb8a9b7 | 9090 | sched_smt_power_savings = level; |
5c45bf27 | 9091 | else |
afb8a9b7 | 9092 | sched_mc_power_savings = level; |
5c45bf27 | 9093 | |
c70f22d2 | 9094 | arch_reinit_sched_domains(); |
5c45bf27 | 9095 | |
c70f22d2 | 9096 | return count; |
5c45bf27 SS |
9097 | } |
9098 | ||
5c45bf27 | 9099 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9100 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9101 | char *page) | |
5c45bf27 SS |
9102 | { |
9103 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9104 | } | |
f718cd4a | 9105 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9106 | const char *buf, size_t count) |
5c45bf27 SS |
9107 | { |
9108 | return sched_power_savings_store(buf, count, 0); | |
9109 | } | |
f718cd4a AK |
9110 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9111 | sched_mc_power_savings_show, | |
9112 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9113 | #endif |
9114 | ||
9115 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9116 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9117 | char *page) | |
5c45bf27 SS |
9118 | { |
9119 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9120 | } | |
f718cd4a | 9121 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9122 | const char *buf, size_t count) |
5c45bf27 SS |
9123 | { |
9124 | return sched_power_savings_store(buf, count, 1); | |
9125 | } | |
f718cd4a AK |
9126 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9127 | sched_smt_power_savings_show, | |
6707de00 AB |
9128 | sched_smt_power_savings_store); |
9129 | #endif | |
9130 | ||
39aac648 | 9131 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9132 | { |
9133 | int err = 0; | |
9134 | ||
9135 | #ifdef CONFIG_SCHED_SMT | |
9136 | if (smt_capable()) | |
9137 | err = sysfs_create_file(&cls->kset.kobj, | |
9138 | &attr_sched_smt_power_savings.attr); | |
9139 | #endif | |
9140 | #ifdef CONFIG_SCHED_MC | |
9141 | if (!err && mc_capable()) | |
9142 | err = sysfs_create_file(&cls->kset.kobj, | |
9143 | &attr_sched_mc_power_savings.attr); | |
9144 | #endif | |
9145 | return err; | |
9146 | } | |
6d6bc0ad | 9147 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9148 | |
e761b772 | 9149 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9150 | /* |
e761b772 MK |
9151 | * Add online and remove offline CPUs from the scheduler domains. |
9152 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9153 | */ |
9154 | static int update_sched_domains(struct notifier_block *nfb, | |
9155 | unsigned long action, void *hcpu) | |
e761b772 MK |
9156 | { |
9157 | switch (action) { | |
9158 | case CPU_ONLINE: | |
9159 | case CPU_ONLINE_FROZEN: | |
9160 | case CPU_DEAD: | |
9161 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9162 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9163 | return NOTIFY_OK; |
9164 | ||
9165 | default: | |
9166 | return NOTIFY_DONE; | |
9167 | } | |
9168 | } | |
9169 | #endif | |
9170 | ||
9171 | static int update_runtime(struct notifier_block *nfb, | |
9172 | unsigned long action, void *hcpu) | |
1da177e4 | 9173 | { |
7def2be1 PZ |
9174 | int cpu = (int)(long)hcpu; |
9175 | ||
1da177e4 | 9176 | switch (action) { |
1da177e4 | 9177 | case CPU_DOWN_PREPARE: |
8bb78442 | 9178 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9179 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9180 | return NOTIFY_OK; |
9181 | ||
1da177e4 | 9182 | case CPU_DOWN_FAILED: |
8bb78442 | 9183 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9184 | case CPU_ONLINE: |
8bb78442 | 9185 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9186 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9187 | return NOTIFY_OK; |
9188 | ||
1da177e4 LT |
9189 | default: |
9190 | return NOTIFY_DONE; | |
9191 | } | |
1da177e4 | 9192 | } |
1da177e4 LT |
9193 | |
9194 | void __init sched_init_smp(void) | |
9195 | { | |
dcc30a35 RR |
9196 | cpumask_var_t non_isolated_cpus; |
9197 | ||
9198 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 9199 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 9200 | |
434d53b0 MT |
9201 | #if defined(CONFIG_NUMA) |
9202 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9203 | GFP_KERNEL); | |
9204 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9205 | #endif | |
95402b38 | 9206 | get_online_cpus(); |
712555ee | 9207 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9208 | arch_init_sched_domains(cpu_online_mask); |
9209 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9210 | if (cpumask_empty(non_isolated_cpus)) | |
9211 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9212 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9213 | put_online_cpus(); |
e761b772 MK |
9214 | |
9215 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9216 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9217 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9218 | #endif |
9219 | ||
9220 | /* RT runtime code needs to handle some hotplug events */ | |
9221 | hotcpu_notifier(update_runtime, 0); | |
9222 | ||
b328ca18 | 9223 | init_hrtick(); |
5c1e1767 NP |
9224 | |
9225 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9226 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9227 | BUG(); |
19978ca6 | 9228 | sched_init_granularity(); |
dcc30a35 | 9229 | free_cpumask_var(non_isolated_cpus); |
4212823f | 9230 | |
0e3900e6 | 9231 | init_sched_rt_class(); |
1da177e4 LT |
9232 | } |
9233 | #else | |
9234 | void __init sched_init_smp(void) | |
9235 | { | |
19978ca6 | 9236 | sched_init_granularity(); |
1da177e4 LT |
9237 | } |
9238 | #endif /* CONFIG_SMP */ | |
9239 | ||
cd1bb94b AB |
9240 | const_debug unsigned int sysctl_timer_migration = 1; |
9241 | ||
1da177e4 LT |
9242 | int in_sched_functions(unsigned long addr) |
9243 | { | |
1da177e4 LT |
9244 | return in_lock_functions(addr) || |
9245 | (addr >= (unsigned long)__sched_text_start | |
9246 | && addr < (unsigned long)__sched_text_end); | |
9247 | } | |
9248 | ||
a9957449 | 9249 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9250 | { |
9251 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9252 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9253 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9254 | cfs_rq->rq = rq; | |
9255 | #endif | |
67e9fb2a | 9256 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9257 | } |
9258 | ||
fa85ae24 PZ |
9259 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9260 | { | |
9261 | struct rt_prio_array *array; | |
9262 | int i; | |
9263 | ||
9264 | array = &rt_rq->active; | |
9265 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9266 | INIT_LIST_HEAD(array->queue + i); | |
9267 | __clear_bit(i, array->bitmap); | |
9268 | } | |
9269 | /* delimiter for bitsearch: */ | |
9270 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9271 | ||
052f1dc7 | 9272 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9273 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9274 | #ifdef CONFIG_SMP |
e864c499 | 9275 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9276 | #endif |
48d5e258 | 9277 | #endif |
fa85ae24 PZ |
9278 | #ifdef CONFIG_SMP |
9279 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9280 | rt_rq->overloaded = 0; |
c20b08e3 | 9281 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9282 | #endif |
9283 | ||
9284 | rt_rq->rt_time = 0; | |
9285 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9286 | rt_rq->rt_runtime = 0; |
9287 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9288 | |
052f1dc7 | 9289 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9290 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9291 | rt_rq->rq = rq; |
9292 | #endif | |
fa85ae24 PZ |
9293 | } |
9294 | ||
6f505b16 | 9295 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9296 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9297 | struct sched_entity *se, int cpu, int add, | |
9298 | struct sched_entity *parent) | |
6f505b16 | 9299 | { |
ec7dc8ac | 9300 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9301 | tg->cfs_rq[cpu] = cfs_rq; |
9302 | init_cfs_rq(cfs_rq, rq); | |
9303 | cfs_rq->tg = tg; | |
9304 | if (add) | |
9305 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9306 | ||
9307 | tg->se[cpu] = se; | |
354d60c2 DG |
9308 | /* se could be NULL for init_task_group */ |
9309 | if (!se) | |
9310 | return; | |
9311 | ||
ec7dc8ac DG |
9312 | if (!parent) |
9313 | se->cfs_rq = &rq->cfs; | |
9314 | else | |
9315 | se->cfs_rq = parent->my_q; | |
9316 | ||
6f505b16 PZ |
9317 | se->my_q = cfs_rq; |
9318 | se->load.weight = tg->shares; | |
e05510d0 | 9319 | se->load.inv_weight = 0; |
ec7dc8ac | 9320 | se->parent = parent; |
6f505b16 | 9321 | } |
052f1dc7 | 9322 | #endif |
6f505b16 | 9323 | |
052f1dc7 | 9324 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9325 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9326 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9327 | struct sched_rt_entity *parent) | |
6f505b16 | 9328 | { |
ec7dc8ac DG |
9329 | struct rq *rq = cpu_rq(cpu); |
9330 | ||
6f505b16 PZ |
9331 | tg->rt_rq[cpu] = rt_rq; |
9332 | init_rt_rq(rt_rq, rq); | |
9333 | rt_rq->tg = tg; | |
9334 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9335 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9336 | if (add) |
9337 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9338 | ||
9339 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9340 | if (!rt_se) |
9341 | return; | |
9342 | ||
ec7dc8ac DG |
9343 | if (!parent) |
9344 | rt_se->rt_rq = &rq->rt; | |
9345 | else | |
9346 | rt_se->rt_rq = parent->my_q; | |
9347 | ||
6f505b16 | 9348 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9349 | rt_se->parent = parent; |
6f505b16 PZ |
9350 | INIT_LIST_HEAD(&rt_se->run_list); |
9351 | } | |
9352 | #endif | |
9353 | ||
1da177e4 LT |
9354 | void __init sched_init(void) |
9355 | { | |
dd41f596 | 9356 | int i, j; |
434d53b0 MT |
9357 | unsigned long alloc_size = 0, ptr; |
9358 | ||
9359 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9360 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9361 | #endif | |
9362 | #ifdef CONFIG_RT_GROUP_SCHED | |
9363 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9364 | #endif |
9365 | #ifdef CONFIG_USER_SCHED | |
9366 | alloc_size *= 2; | |
df7c8e84 RR |
9367 | #endif |
9368 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9369 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 9370 | #endif |
434d53b0 | 9371 | if (alloc_size) { |
36b7b6d4 | 9372 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9373 | |
9374 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9375 | init_task_group.se = (struct sched_entity **)ptr; | |
9376 | ptr += nr_cpu_ids * sizeof(void **); | |
9377 | ||
9378 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9379 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9380 | |
9381 | #ifdef CONFIG_USER_SCHED | |
9382 | root_task_group.se = (struct sched_entity **)ptr; | |
9383 | ptr += nr_cpu_ids * sizeof(void **); | |
9384 | ||
9385 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9386 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9387 | #endif /* CONFIG_USER_SCHED */ |
9388 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9389 | #ifdef CONFIG_RT_GROUP_SCHED |
9390 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9391 | ptr += nr_cpu_ids * sizeof(void **); | |
9392 | ||
9393 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9394 | ptr += nr_cpu_ids * sizeof(void **); |
9395 | ||
9396 | #ifdef CONFIG_USER_SCHED | |
9397 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9398 | ptr += nr_cpu_ids * sizeof(void **); | |
9399 | ||
9400 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9401 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9402 | #endif /* CONFIG_USER_SCHED */ |
9403 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9404 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9405 | for_each_possible_cpu(i) { | |
9406 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9407 | ptr += cpumask_size(); | |
9408 | } | |
9409 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9410 | } |
dd41f596 | 9411 | |
57d885fe GH |
9412 | #ifdef CONFIG_SMP |
9413 | init_defrootdomain(); | |
9414 | #endif | |
9415 | ||
d0b27fa7 PZ |
9416 | init_rt_bandwidth(&def_rt_bandwidth, |
9417 | global_rt_period(), global_rt_runtime()); | |
9418 | ||
9419 | #ifdef CONFIG_RT_GROUP_SCHED | |
9420 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9421 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9422 | #ifdef CONFIG_USER_SCHED |
9423 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9424 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9425 | #endif /* CONFIG_USER_SCHED */ |
9426 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9427 | |
052f1dc7 | 9428 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9429 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9430 | INIT_LIST_HEAD(&init_task_group.children); |
9431 | ||
9432 | #ifdef CONFIG_USER_SCHED | |
9433 | INIT_LIST_HEAD(&root_task_group.children); | |
9434 | init_task_group.parent = &root_task_group; | |
9435 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9436 | #endif /* CONFIG_USER_SCHED */ |
9437 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9438 | |
0a945022 | 9439 | for_each_possible_cpu(i) { |
70b97a7f | 9440 | struct rq *rq; |
1da177e4 LT |
9441 | |
9442 | rq = cpu_rq(i); | |
9443 | spin_lock_init(&rq->lock); | |
7897986b | 9444 | rq->nr_running = 0; |
dce48a84 TG |
9445 | rq->calc_load_active = 0; |
9446 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9447 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9448 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9449 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9450 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9451 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9452 | #ifdef CONFIG_CGROUP_SCHED |
9453 | /* | |
9454 | * How much cpu bandwidth does init_task_group get? | |
9455 | * | |
9456 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9457 | * gets 100% of the cpu resources in the system. This overall | |
9458 | * system cpu resource is divided among the tasks of | |
9459 | * init_task_group and its child task-groups in a fair manner, | |
9460 | * based on each entity's (task or task-group's) weight | |
9461 | * (se->load.weight). | |
9462 | * | |
9463 | * In other words, if init_task_group has 10 tasks of weight | |
9464 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9465 | * then A0's share of the cpu resource is: | |
9466 | * | |
0d905bca | 9467 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9468 | * |
9469 | * We achieve this by letting init_task_group's tasks sit | |
9470 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9471 | */ | |
ec7dc8ac | 9472 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9473 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9474 | root_task_group.shares = NICE_0_LOAD; |
9475 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9476 | /* |
9477 | * In case of task-groups formed thr' the user id of tasks, | |
9478 | * init_task_group represents tasks belonging to root user. | |
9479 | * Hence it forms a sibling of all subsequent groups formed. | |
9480 | * In this case, init_task_group gets only a fraction of overall | |
9481 | * system cpu resource, based on the weight assigned to root | |
9482 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9483 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9484 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9485 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9486 | */ | |
ec7dc8ac | 9487 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9488 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9489 | &per_cpu(init_sched_entity, i), i, 1, |
9490 | root_task_group.se[i]); | |
6f505b16 | 9491 | |
052f1dc7 | 9492 | #endif |
354d60c2 DG |
9493 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9494 | ||
9495 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9496 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9497 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9498 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9499 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9500 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9501 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9502 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9503 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9504 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9505 | root_task_group.rt_se[i]); | |
354d60c2 | 9506 | #endif |
dd41f596 | 9507 | #endif |
1da177e4 | 9508 | |
dd41f596 IM |
9509 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9510 | rq->cpu_load[j] = 0; | |
1da177e4 | 9511 | #ifdef CONFIG_SMP |
41c7ce9a | 9512 | rq->sd = NULL; |
57d885fe | 9513 | rq->rd = NULL; |
3f029d3c | 9514 | rq->post_schedule = 0; |
1da177e4 | 9515 | rq->active_balance = 0; |
dd41f596 | 9516 | rq->next_balance = jiffies; |
1da177e4 | 9517 | rq->push_cpu = 0; |
0a2966b4 | 9518 | rq->cpu = i; |
1f11eb6a | 9519 | rq->online = 0; |
1da177e4 | 9520 | rq->migration_thread = NULL; |
eae0c9df MG |
9521 | rq->idle_stamp = 0; |
9522 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 9523 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 9524 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9525 | #endif |
8f4d37ec | 9526 | init_rq_hrtick(rq); |
1da177e4 | 9527 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9528 | } |
9529 | ||
2dd73a4f | 9530 | set_load_weight(&init_task); |
b50f60ce | 9531 | |
e107be36 AK |
9532 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9533 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9534 | #endif | |
9535 | ||
c9819f45 | 9536 | #ifdef CONFIG_SMP |
962cf36c | 9537 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9538 | #endif |
9539 | ||
b50f60ce HC |
9540 | #ifdef CONFIG_RT_MUTEXES |
9541 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9542 | #endif | |
9543 | ||
1da177e4 LT |
9544 | /* |
9545 | * The boot idle thread does lazy MMU switching as well: | |
9546 | */ | |
9547 | atomic_inc(&init_mm.mm_count); | |
9548 | enter_lazy_tlb(&init_mm, current); | |
9549 | ||
9550 | /* | |
9551 | * Make us the idle thread. Technically, schedule() should not be | |
9552 | * called from this thread, however somewhere below it might be, | |
9553 | * but because we are the idle thread, we just pick up running again | |
9554 | * when this runqueue becomes "idle". | |
9555 | */ | |
9556 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9557 | |
9558 | calc_load_update = jiffies + LOAD_FREQ; | |
9559 | ||
dd41f596 IM |
9560 | /* |
9561 | * During early bootup we pretend to be a normal task: | |
9562 | */ | |
9563 | current->sched_class = &fair_sched_class; | |
6892b75e | 9564 | |
6a7b3dc3 | 9565 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9566 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9567 | #ifdef CONFIG_SMP |
7d1e6a9b | 9568 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9569 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9570 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9571 | #endif |
4bdddf8f | 9572 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9573 | #endif /* SMP */ |
6a7b3dc3 | 9574 | |
cdd6c482 | 9575 | perf_event_init(); |
0d905bca | 9576 | |
6892b75e | 9577 | scheduler_running = 1; |
1da177e4 LT |
9578 | } |
9579 | ||
9580 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9581 | static inline int preempt_count_equals(int preempt_offset) |
9582 | { | |
9583 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9584 | ||
9585 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9586 | } | |
9587 | ||
9588 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9589 | { |
48f24c4d | 9590 | #ifdef in_atomic |
1da177e4 LT |
9591 | static unsigned long prev_jiffy; /* ratelimiting */ |
9592 | ||
e4aafea2 FW |
9593 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9594 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9595 | return; |
9596 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9597 | return; | |
9598 | prev_jiffy = jiffies; | |
9599 | ||
9600 | printk(KERN_ERR | |
9601 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9602 | file, line); | |
9603 | printk(KERN_ERR | |
9604 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9605 | in_atomic(), irqs_disabled(), | |
9606 | current->pid, current->comm); | |
9607 | ||
9608 | debug_show_held_locks(current); | |
9609 | if (irqs_disabled()) | |
9610 | print_irqtrace_events(current); | |
9611 | dump_stack(); | |
1da177e4 LT |
9612 | #endif |
9613 | } | |
9614 | EXPORT_SYMBOL(__might_sleep); | |
9615 | #endif | |
9616 | ||
9617 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9618 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9619 | { | |
9620 | int on_rq; | |
3e51f33f | 9621 | |
3a5e4dc1 AK |
9622 | update_rq_clock(rq); |
9623 | on_rq = p->se.on_rq; | |
9624 | if (on_rq) | |
9625 | deactivate_task(rq, p, 0); | |
9626 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9627 | if (on_rq) { | |
9628 | activate_task(rq, p, 0); | |
9629 | resched_task(rq->curr); | |
9630 | } | |
9631 | } | |
9632 | ||
1da177e4 LT |
9633 | void normalize_rt_tasks(void) |
9634 | { | |
a0f98a1c | 9635 | struct task_struct *g, *p; |
1da177e4 | 9636 | unsigned long flags; |
70b97a7f | 9637 | struct rq *rq; |
1da177e4 | 9638 | |
4cf5d77a | 9639 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9640 | do_each_thread(g, p) { |
178be793 IM |
9641 | /* |
9642 | * Only normalize user tasks: | |
9643 | */ | |
9644 | if (!p->mm) | |
9645 | continue; | |
9646 | ||
6cfb0d5d | 9647 | p->se.exec_start = 0; |
6cfb0d5d | 9648 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9649 | p->se.wait_start = 0; |
dd41f596 | 9650 | p->se.sleep_start = 0; |
dd41f596 | 9651 | p->se.block_start = 0; |
6cfb0d5d | 9652 | #endif |
dd41f596 IM |
9653 | |
9654 | if (!rt_task(p)) { | |
9655 | /* | |
9656 | * Renice negative nice level userspace | |
9657 | * tasks back to 0: | |
9658 | */ | |
9659 | if (TASK_NICE(p) < 0 && p->mm) | |
9660 | set_user_nice(p, 0); | |
1da177e4 | 9661 | continue; |
dd41f596 | 9662 | } |
1da177e4 | 9663 | |
4cf5d77a | 9664 | spin_lock(&p->pi_lock); |
b29739f9 | 9665 | rq = __task_rq_lock(p); |
1da177e4 | 9666 | |
178be793 | 9667 | normalize_task(rq, p); |
3a5e4dc1 | 9668 | |
b29739f9 | 9669 | __task_rq_unlock(rq); |
4cf5d77a | 9670 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9671 | } while_each_thread(g, p); |
9672 | ||
4cf5d77a | 9673 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9674 | } |
9675 | ||
9676 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9677 | |
9678 | #ifdef CONFIG_IA64 | |
9679 | /* | |
9680 | * These functions are only useful for the IA64 MCA handling. | |
9681 | * | |
9682 | * They can only be called when the whole system has been | |
9683 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9684 | * activity can take place. Using them for anything else would | |
9685 | * be a serious bug, and as a result, they aren't even visible | |
9686 | * under any other configuration. | |
9687 | */ | |
9688 | ||
9689 | /** | |
9690 | * curr_task - return the current task for a given cpu. | |
9691 | * @cpu: the processor in question. | |
9692 | * | |
9693 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9694 | */ | |
36c8b586 | 9695 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9696 | { |
9697 | return cpu_curr(cpu); | |
9698 | } | |
9699 | ||
9700 | /** | |
9701 | * set_curr_task - set the current task for a given cpu. | |
9702 | * @cpu: the processor in question. | |
9703 | * @p: the task pointer to set. | |
9704 | * | |
9705 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9706 | * are serviced on a separate stack. It allows the architecture to switch the |
9707 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9708 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9709 | * and caller must save the original value of the current task (see | |
9710 | * curr_task() above) and restore that value before reenabling interrupts and | |
9711 | * re-starting the system. | |
9712 | * | |
9713 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9714 | */ | |
36c8b586 | 9715 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9716 | { |
9717 | cpu_curr(cpu) = p; | |
9718 | } | |
9719 | ||
9720 | #endif | |
29f59db3 | 9721 | |
bccbe08a PZ |
9722 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9723 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9724 | { |
9725 | int i; | |
9726 | ||
9727 | for_each_possible_cpu(i) { | |
9728 | if (tg->cfs_rq) | |
9729 | kfree(tg->cfs_rq[i]); | |
9730 | if (tg->se) | |
9731 | kfree(tg->se[i]); | |
6f505b16 PZ |
9732 | } |
9733 | ||
9734 | kfree(tg->cfs_rq); | |
9735 | kfree(tg->se); | |
6f505b16 PZ |
9736 | } |
9737 | ||
ec7dc8ac DG |
9738 | static |
9739 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9740 | { |
29f59db3 | 9741 | struct cfs_rq *cfs_rq; |
eab17229 | 9742 | struct sched_entity *se; |
9b5b7751 | 9743 | struct rq *rq; |
29f59db3 SV |
9744 | int i; |
9745 | ||
434d53b0 | 9746 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9747 | if (!tg->cfs_rq) |
9748 | goto err; | |
434d53b0 | 9749 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9750 | if (!tg->se) |
9751 | goto err; | |
052f1dc7 PZ |
9752 | |
9753 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9754 | |
9755 | for_each_possible_cpu(i) { | |
9b5b7751 | 9756 | rq = cpu_rq(i); |
29f59db3 | 9757 | |
eab17229 LZ |
9758 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9759 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9760 | if (!cfs_rq) |
9761 | goto err; | |
9762 | ||
eab17229 LZ |
9763 | se = kzalloc_node(sizeof(struct sched_entity), |
9764 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9765 | if (!se) |
9766 | goto err; | |
9767 | ||
eab17229 | 9768 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9769 | } |
9770 | ||
9771 | return 1; | |
9772 | ||
9773 | err: | |
9774 | return 0; | |
9775 | } | |
9776 | ||
9777 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9778 | { | |
9779 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9780 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9781 | } | |
9782 | ||
9783 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9784 | { | |
9785 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9786 | } | |
6d6bc0ad | 9787 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9788 | static inline void free_fair_sched_group(struct task_group *tg) |
9789 | { | |
9790 | } | |
9791 | ||
ec7dc8ac DG |
9792 | static inline |
9793 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9794 | { |
9795 | return 1; | |
9796 | } | |
9797 | ||
9798 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9799 | { | |
9800 | } | |
9801 | ||
9802 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9803 | { | |
9804 | } | |
6d6bc0ad | 9805 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9806 | |
9807 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9808 | static void free_rt_sched_group(struct task_group *tg) |
9809 | { | |
9810 | int i; | |
9811 | ||
d0b27fa7 PZ |
9812 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9813 | ||
bccbe08a PZ |
9814 | for_each_possible_cpu(i) { |
9815 | if (tg->rt_rq) | |
9816 | kfree(tg->rt_rq[i]); | |
9817 | if (tg->rt_se) | |
9818 | kfree(tg->rt_se[i]); | |
9819 | } | |
9820 | ||
9821 | kfree(tg->rt_rq); | |
9822 | kfree(tg->rt_se); | |
9823 | } | |
9824 | ||
ec7dc8ac DG |
9825 | static |
9826 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9827 | { |
9828 | struct rt_rq *rt_rq; | |
eab17229 | 9829 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9830 | struct rq *rq; |
9831 | int i; | |
9832 | ||
434d53b0 | 9833 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9834 | if (!tg->rt_rq) |
9835 | goto err; | |
434d53b0 | 9836 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9837 | if (!tg->rt_se) |
9838 | goto err; | |
9839 | ||
d0b27fa7 PZ |
9840 | init_rt_bandwidth(&tg->rt_bandwidth, |
9841 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9842 | |
9843 | for_each_possible_cpu(i) { | |
9844 | rq = cpu_rq(i); | |
9845 | ||
eab17229 LZ |
9846 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9847 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9848 | if (!rt_rq) |
9849 | goto err; | |
29f59db3 | 9850 | |
eab17229 LZ |
9851 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9852 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9853 | if (!rt_se) |
9854 | goto err; | |
29f59db3 | 9855 | |
eab17229 | 9856 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9857 | } |
9858 | ||
bccbe08a PZ |
9859 | return 1; |
9860 | ||
9861 | err: | |
9862 | return 0; | |
9863 | } | |
9864 | ||
9865 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9866 | { | |
9867 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9868 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9869 | } | |
9870 | ||
9871 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9872 | { | |
9873 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9874 | } | |
6d6bc0ad | 9875 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9876 | static inline void free_rt_sched_group(struct task_group *tg) |
9877 | { | |
9878 | } | |
9879 | ||
ec7dc8ac DG |
9880 | static inline |
9881 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9882 | { |
9883 | return 1; | |
9884 | } | |
9885 | ||
9886 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9887 | { | |
9888 | } | |
9889 | ||
9890 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9891 | { | |
9892 | } | |
6d6bc0ad | 9893 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9894 | |
d0b27fa7 | 9895 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9896 | static void free_sched_group(struct task_group *tg) |
9897 | { | |
9898 | free_fair_sched_group(tg); | |
9899 | free_rt_sched_group(tg); | |
9900 | kfree(tg); | |
9901 | } | |
9902 | ||
9903 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9904 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9905 | { |
9906 | struct task_group *tg; | |
9907 | unsigned long flags; | |
9908 | int i; | |
9909 | ||
9910 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9911 | if (!tg) | |
9912 | return ERR_PTR(-ENOMEM); | |
9913 | ||
ec7dc8ac | 9914 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9915 | goto err; |
9916 | ||
ec7dc8ac | 9917 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9918 | goto err; |
9919 | ||
8ed36996 | 9920 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9921 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9922 | register_fair_sched_group(tg, i); |
9923 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9924 | } |
6f505b16 | 9925 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9926 | |
9927 | WARN_ON(!parent); /* root should already exist */ | |
9928 | ||
9929 | tg->parent = parent; | |
f473aa5e | 9930 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9931 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9932 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9933 | |
9b5b7751 | 9934 | return tg; |
29f59db3 SV |
9935 | |
9936 | err: | |
6f505b16 | 9937 | free_sched_group(tg); |
29f59db3 SV |
9938 | return ERR_PTR(-ENOMEM); |
9939 | } | |
9940 | ||
9b5b7751 | 9941 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9942 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9943 | { |
29f59db3 | 9944 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9945 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9946 | } |
9947 | ||
9b5b7751 | 9948 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9949 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9950 | { |
8ed36996 | 9951 | unsigned long flags; |
9b5b7751 | 9952 | int i; |
29f59db3 | 9953 | |
8ed36996 | 9954 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9955 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9956 | unregister_fair_sched_group(tg, i); |
9957 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9958 | } |
6f505b16 | 9959 | list_del_rcu(&tg->list); |
f473aa5e | 9960 | list_del_rcu(&tg->siblings); |
8ed36996 | 9961 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9962 | |
9b5b7751 | 9963 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9964 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9965 | } |
9966 | ||
9b5b7751 | 9967 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9968 | * The caller of this function should have put the task in its new group |
9969 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9970 | * reflect its new group. | |
9b5b7751 SV |
9971 | */ |
9972 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9973 | { |
9974 | int on_rq, running; | |
9975 | unsigned long flags; | |
9976 | struct rq *rq; | |
9977 | ||
9978 | rq = task_rq_lock(tsk, &flags); | |
9979 | ||
29f59db3 SV |
9980 | update_rq_clock(rq); |
9981 | ||
051a1d1a | 9982 | running = task_current(rq, tsk); |
29f59db3 SV |
9983 | on_rq = tsk->se.on_rq; |
9984 | ||
0e1f3483 | 9985 | if (on_rq) |
29f59db3 | 9986 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9987 | if (unlikely(running)) |
9988 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9989 | |
6f505b16 | 9990 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9991 | |
810b3817 PZ |
9992 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9993 | if (tsk->sched_class->moved_group) | |
9994 | tsk->sched_class->moved_group(tsk); | |
9995 | #endif | |
9996 | ||
0e1f3483 HS |
9997 | if (unlikely(running)) |
9998 | tsk->sched_class->set_curr_task(rq); | |
9999 | if (on_rq) | |
7074badb | 10000 | enqueue_task(rq, tsk, 0); |
29f59db3 | 10001 | |
29f59db3 SV |
10002 | task_rq_unlock(rq, &flags); |
10003 | } | |
6d6bc0ad | 10004 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 10005 | |
052f1dc7 | 10006 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 10007 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
10008 | { |
10009 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
10010 | int on_rq; |
10011 | ||
29f59db3 | 10012 | on_rq = se->on_rq; |
62fb1851 | 10013 | if (on_rq) |
29f59db3 SV |
10014 | dequeue_entity(cfs_rq, se, 0); |
10015 | ||
10016 | se->load.weight = shares; | |
e05510d0 | 10017 | se->load.inv_weight = 0; |
29f59db3 | 10018 | |
62fb1851 | 10019 | if (on_rq) |
29f59db3 | 10020 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10021 | } |
62fb1851 | 10022 | |
c09595f6 PZ |
10023 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10024 | { | |
10025 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10026 | struct rq *rq = cfs_rq->rq; | |
10027 | unsigned long flags; | |
10028 | ||
10029 | spin_lock_irqsave(&rq->lock, flags); | |
10030 | __set_se_shares(se, shares); | |
10031 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
10032 | } |
10033 | ||
8ed36996 PZ |
10034 | static DEFINE_MUTEX(shares_mutex); |
10035 | ||
4cf86d77 | 10036 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10037 | { |
10038 | int i; | |
8ed36996 | 10039 | unsigned long flags; |
c61935fd | 10040 | |
ec7dc8ac DG |
10041 | /* |
10042 | * We can't change the weight of the root cgroup. | |
10043 | */ | |
10044 | if (!tg->se[0]) | |
10045 | return -EINVAL; | |
10046 | ||
18d95a28 PZ |
10047 | if (shares < MIN_SHARES) |
10048 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10049 | else if (shares > MAX_SHARES) |
10050 | shares = MAX_SHARES; | |
62fb1851 | 10051 | |
8ed36996 | 10052 | mutex_lock(&shares_mutex); |
9b5b7751 | 10053 | if (tg->shares == shares) |
5cb350ba | 10054 | goto done; |
29f59db3 | 10055 | |
8ed36996 | 10056 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10057 | for_each_possible_cpu(i) |
10058 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10059 | list_del_rcu(&tg->siblings); |
8ed36996 | 10060 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10061 | |
10062 | /* wait for any ongoing reference to this group to finish */ | |
10063 | synchronize_sched(); | |
10064 | ||
10065 | /* | |
10066 | * Now we are free to modify the group's share on each cpu | |
10067 | * w/o tripping rebalance_share or load_balance_fair. | |
10068 | */ | |
9b5b7751 | 10069 | tg->shares = shares; |
c09595f6 PZ |
10070 | for_each_possible_cpu(i) { |
10071 | /* | |
10072 | * force a rebalance | |
10073 | */ | |
10074 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10075 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10076 | } |
29f59db3 | 10077 | |
6b2d7700 SV |
10078 | /* |
10079 | * Enable load balance activity on this group, by inserting it back on | |
10080 | * each cpu's rq->leaf_cfs_rq_list. | |
10081 | */ | |
8ed36996 | 10082 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10083 | for_each_possible_cpu(i) |
10084 | register_fair_sched_group(tg, i); | |
f473aa5e | 10085 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10086 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10087 | done: |
8ed36996 | 10088 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10089 | return 0; |
29f59db3 SV |
10090 | } |
10091 | ||
5cb350ba DG |
10092 | unsigned long sched_group_shares(struct task_group *tg) |
10093 | { | |
10094 | return tg->shares; | |
10095 | } | |
052f1dc7 | 10096 | #endif |
5cb350ba | 10097 | |
052f1dc7 | 10098 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10099 | /* |
9f0c1e56 | 10100 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10101 | */ |
9f0c1e56 PZ |
10102 | static DEFINE_MUTEX(rt_constraints_mutex); |
10103 | ||
10104 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10105 | { | |
10106 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10107 | return 1ULL << 20; |
9f0c1e56 | 10108 | |
9a7e0b18 | 10109 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10110 | } |
10111 | ||
9a7e0b18 PZ |
10112 | /* Must be called with tasklist_lock held */ |
10113 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10114 | { |
9a7e0b18 | 10115 | struct task_struct *g, *p; |
b40b2e8e | 10116 | |
9a7e0b18 PZ |
10117 | do_each_thread(g, p) { |
10118 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10119 | return 1; | |
10120 | } while_each_thread(g, p); | |
b40b2e8e | 10121 | |
9a7e0b18 PZ |
10122 | return 0; |
10123 | } | |
b40b2e8e | 10124 | |
9a7e0b18 PZ |
10125 | struct rt_schedulable_data { |
10126 | struct task_group *tg; | |
10127 | u64 rt_period; | |
10128 | u64 rt_runtime; | |
10129 | }; | |
b40b2e8e | 10130 | |
9a7e0b18 PZ |
10131 | static int tg_schedulable(struct task_group *tg, void *data) |
10132 | { | |
10133 | struct rt_schedulable_data *d = data; | |
10134 | struct task_group *child; | |
10135 | unsigned long total, sum = 0; | |
10136 | u64 period, runtime; | |
b40b2e8e | 10137 | |
9a7e0b18 PZ |
10138 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10139 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10140 | |
9a7e0b18 PZ |
10141 | if (tg == d->tg) { |
10142 | period = d->rt_period; | |
10143 | runtime = d->rt_runtime; | |
b40b2e8e | 10144 | } |
b40b2e8e | 10145 | |
98a4826b PZ |
10146 | #ifdef CONFIG_USER_SCHED |
10147 | if (tg == &root_task_group) { | |
10148 | period = global_rt_period(); | |
10149 | runtime = global_rt_runtime(); | |
10150 | } | |
10151 | #endif | |
10152 | ||
4653f803 PZ |
10153 | /* |
10154 | * Cannot have more runtime than the period. | |
10155 | */ | |
10156 | if (runtime > period && runtime != RUNTIME_INF) | |
10157 | return -EINVAL; | |
6f505b16 | 10158 | |
4653f803 PZ |
10159 | /* |
10160 | * Ensure we don't starve existing RT tasks. | |
10161 | */ | |
9a7e0b18 PZ |
10162 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10163 | return -EBUSY; | |
6f505b16 | 10164 | |
9a7e0b18 | 10165 | total = to_ratio(period, runtime); |
6f505b16 | 10166 | |
4653f803 PZ |
10167 | /* |
10168 | * Nobody can have more than the global setting allows. | |
10169 | */ | |
10170 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10171 | return -EINVAL; | |
6f505b16 | 10172 | |
4653f803 PZ |
10173 | /* |
10174 | * The sum of our children's runtime should not exceed our own. | |
10175 | */ | |
9a7e0b18 PZ |
10176 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10177 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10178 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10179 | |
9a7e0b18 PZ |
10180 | if (child == d->tg) { |
10181 | period = d->rt_period; | |
10182 | runtime = d->rt_runtime; | |
10183 | } | |
6f505b16 | 10184 | |
9a7e0b18 | 10185 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10186 | } |
6f505b16 | 10187 | |
9a7e0b18 PZ |
10188 | if (sum > total) |
10189 | return -EINVAL; | |
10190 | ||
10191 | return 0; | |
6f505b16 PZ |
10192 | } |
10193 | ||
9a7e0b18 | 10194 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10195 | { |
9a7e0b18 PZ |
10196 | struct rt_schedulable_data data = { |
10197 | .tg = tg, | |
10198 | .rt_period = period, | |
10199 | .rt_runtime = runtime, | |
10200 | }; | |
10201 | ||
10202 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10203 | } |
10204 | ||
d0b27fa7 PZ |
10205 | static int tg_set_bandwidth(struct task_group *tg, |
10206 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10207 | { |
ac086bc2 | 10208 | int i, err = 0; |
9f0c1e56 | 10209 | |
9f0c1e56 | 10210 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10211 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10212 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10213 | if (err) | |
9f0c1e56 | 10214 | goto unlock; |
ac086bc2 PZ |
10215 | |
10216 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10217 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10218 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10219 | |
10220 | for_each_possible_cpu(i) { | |
10221 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10222 | ||
10223 | spin_lock(&rt_rq->rt_runtime_lock); | |
10224 | rt_rq->rt_runtime = rt_runtime; | |
10225 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10226 | } | |
10227 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10228 | unlock: |
521f1a24 | 10229 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10230 | mutex_unlock(&rt_constraints_mutex); |
10231 | ||
10232 | return err; | |
6f505b16 PZ |
10233 | } |
10234 | ||
d0b27fa7 PZ |
10235 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10236 | { | |
10237 | u64 rt_runtime, rt_period; | |
10238 | ||
10239 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10240 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10241 | if (rt_runtime_us < 0) | |
10242 | rt_runtime = RUNTIME_INF; | |
10243 | ||
10244 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10245 | } | |
10246 | ||
9f0c1e56 PZ |
10247 | long sched_group_rt_runtime(struct task_group *tg) |
10248 | { | |
10249 | u64 rt_runtime_us; | |
10250 | ||
d0b27fa7 | 10251 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10252 | return -1; |
10253 | ||
d0b27fa7 | 10254 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10255 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10256 | return rt_runtime_us; | |
10257 | } | |
d0b27fa7 PZ |
10258 | |
10259 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10260 | { | |
10261 | u64 rt_runtime, rt_period; | |
10262 | ||
10263 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10264 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10265 | ||
619b0488 R |
10266 | if (rt_period == 0) |
10267 | return -EINVAL; | |
10268 | ||
d0b27fa7 PZ |
10269 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10270 | } | |
10271 | ||
10272 | long sched_group_rt_period(struct task_group *tg) | |
10273 | { | |
10274 | u64 rt_period_us; | |
10275 | ||
10276 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10277 | do_div(rt_period_us, NSEC_PER_USEC); | |
10278 | return rt_period_us; | |
10279 | } | |
10280 | ||
10281 | static int sched_rt_global_constraints(void) | |
10282 | { | |
4653f803 | 10283 | u64 runtime, period; |
d0b27fa7 PZ |
10284 | int ret = 0; |
10285 | ||
ec5d4989 HS |
10286 | if (sysctl_sched_rt_period <= 0) |
10287 | return -EINVAL; | |
10288 | ||
4653f803 PZ |
10289 | runtime = global_rt_runtime(); |
10290 | period = global_rt_period(); | |
10291 | ||
10292 | /* | |
10293 | * Sanity check on the sysctl variables. | |
10294 | */ | |
10295 | if (runtime > period && runtime != RUNTIME_INF) | |
10296 | return -EINVAL; | |
10b612f4 | 10297 | |
d0b27fa7 | 10298 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10299 | read_lock(&tasklist_lock); |
4653f803 | 10300 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10301 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10302 | mutex_unlock(&rt_constraints_mutex); |
10303 | ||
10304 | return ret; | |
10305 | } | |
54e99124 DG |
10306 | |
10307 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10308 | { | |
10309 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10310 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10311 | return 0; | |
10312 | ||
10313 | return 1; | |
10314 | } | |
10315 | ||
6d6bc0ad | 10316 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10317 | static int sched_rt_global_constraints(void) |
10318 | { | |
ac086bc2 PZ |
10319 | unsigned long flags; |
10320 | int i; | |
10321 | ||
ec5d4989 HS |
10322 | if (sysctl_sched_rt_period <= 0) |
10323 | return -EINVAL; | |
10324 | ||
60aa605d PZ |
10325 | /* |
10326 | * There's always some RT tasks in the root group | |
10327 | * -- migration, kstopmachine etc.. | |
10328 | */ | |
10329 | if (sysctl_sched_rt_runtime == 0) | |
10330 | return -EBUSY; | |
10331 | ||
ac086bc2 PZ |
10332 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10333 | for_each_possible_cpu(i) { | |
10334 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10335 | ||
10336 | spin_lock(&rt_rq->rt_runtime_lock); | |
10337 | rt_rq->rt_runtime = global_rt_runtime(); | |
10338 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10339 | } | |
10340 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10341 | ||
d0b27fa7 PZ |
10342 | return 0; |
10343 | } | |
6d6bc0ad | 10344 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10345 | |
10346 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 10347 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
10348 | loff_t *ppos) |
10349 | { | |
10350 | int ret; | |
10351 | int old_period, old_runtime; | |
10352 | static DEFINE_MUTEX(mutex); | |
10353 | ||
10354 | mutex_lock(&mutex); | |
10355 | old_period = sysctl_sched_rt_period; | |
10356 | old_runtime = sysctl_sched_rt_runtime; | |
10357 | ||
8d65af78 | 10358 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
10359 | |
10360 | if (!ret && write) { | |
10361 | ret = sched_rt_global_constraints(); | |
10362 | if (ret) { | |
10363 | sysctl_sched_rt_period = old_period; | |
10364 | sysctl_sched_rt_runtime = old_runtime; | |
10365 | } else { | |
10366 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10367 | def_rt_bandwidth.rt_period = | |
10368 | ns_to_ktime(global_rt_period()); | |
10369 | } | |
10370 | } | |
10371 | mutex_unlock(&mutex); | |
10372 | ||
10373 | return ret; | |
10374 | } | |
68318b8e | 10375 | |
052f1dc7 | 10376 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10377 | |
10378 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10379 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10380 | { |
2b01dfe3 PM |
10381 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10382 | struct task_group, css); | |
68318b8e SV |
10383 | } |
10384 | ||
10385 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10386 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10387 | { |
ec7dc8ac | 10388 | struct task_group *tg, *parent; |
68318b8e | 10389 | |
2b01dfe3 | 10390 | if (!cgrp->parent) { |
68318b8e | 10391 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10392 | return &init_task_group.css; |
10393 | } | |
10394 | ||
ec7dc8ac DG |
10395 | parent = cgroup_tg(cgrp->parent); |
10396 | tg = sched_create_group(parent); | |
68318b8e SV |
10397 | if (IS_ERR(tg)) |
10398 | return ERR_PTR(-ENOMEM); | |
10399 | ||
68318b8e SV |
10400 | return &tg->css; |
10401 | } | |
10402 | ||
41a2d6cf IM |
10403 | static void |
10404 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10405 | { |
2b01dfe3 | 10406 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10407 | |
10408 | sched_destroy_group(tg); | |
10409 | } | |
10410 | ||
41a2d6cf | 10411 | static int |
be367d09 | 10412 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 10413 | { |
b68aa230 | 10414 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10415 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10416 | return -EINVAL; |
10417 | #else | |
68318b8e SV |
10418 | /* We don't support RT-tasks being in separate groups */ |
10419 | if (tsk->sched_class != &fair_sched_class) | |
10420 | return -EINVAL; | |
b68aa230 | 10421 | #endif |
be367d09 BB |
10422 | return 0; |
10423 | } | |
68318b8e | 10424 | |
be367d09 BB |
10425 | static int |
10426 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10427 | struct task_struct *tsk, bool threadgroup) | |
10428 | { | |
10429 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
10430 | if (retval) | |
10431 | return retval; | |
10432 | if (threadgroup) { | |
10433 | struct task_struct *c; | |
10434 | rcu_read_lock(); | |
10435 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10436 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
10437 | if (retval) { | |
10438 | rcu_read_unlock(); | |
10439 | return retval; | |
10440 | } | |
10441 | } | |
10442 | rcu_read_unlock(); | |
10443 | } | |
68318b8e SV |
10444 | return 0; |
10445 | } | |
10446 | ||
10447 | static void | |
2b01dfe3 | 10448 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
10449 | struct cgroup *old_cont, struct task_struct *tsk, |
10450 | bool threadgroup) | |
68318b8e SV |
10451 | { |
10452 | sched_move_task(tsk); | |
be367d09 BB |
10453 | if (threadgroup) { |
10454 | struct task_struct *c; | |
10455 | rcu_read_lock(); | |
10456 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10457 | sched_move_task(c); | |
10458 | } | |
10459 | rcu_read_unlock(); | |
10460 | } | |
68318b8e SV |
10461 | } |
10462 | ||
052f1dc7 | 10463 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10464 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10465 | u64 shareval) |
68318b8e | 10466 | { |
2b01dfe3 | 10467 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10468 | } |
10469 | ||
f4c753b7 | 10470 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10471 | { |
2b01dfe3 | 10472 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10473 | |
10474 | return (u64) tg->shares; | |
10475 | } | |
6d6bc0ad | 10476 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10477 | |
052f1dc7 | 10478 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10479 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10480 | s64 val) |
6f505b16 | 10481 | { |
06ecb27c | 10482 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10483 | } |
10484 | ||
06ecb27c | 10485 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10486 | { |
06ecb27c | 10487 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10488 | } |
d0b27fa7 PZ |
10489 | |
10490 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10491 | u64 rt_period_us) | |
10492 | { | |
10493 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10494 | } | |
10495 | ||
10496 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10497 | { | |
10498 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10499 | } | |
6d6bc0ad | 10500 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10501 | |
fe5c7cc2 | 10502 | static struct cftype cpu_files[] = { |
052f1dc7 | 10503 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10504 | { |
10505 | .name = "shares", | |
f4c753b7 PM |
10506 | .read_u64 = cpu_shares_read_u64, |
10507 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10508 | }, |
052f1dc7 PZ |
10509 | #endif |
10510 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10511 | { |
9f0c1e56 | 10512 | .name = "rt_runtime_us", |
06ecb27c PM |
10513 | .read_s64 = cpu_rt_runtime_read, |
10514 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10515 | }, |
d0b27fa7 PZ |
10516 | { |
10517 | .name = "rt_period_us", | |
f4c753b7 PM |
10518 | .read_u64 = cpu_rt_period_read_uint, |
10519 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10520 | }, |
052f1dc7 | 10521 | #endif |
68318b8e SV |
10522 | }; |
10523 | ||
10524 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10525 | { | |
fe5c7cc2 | 10526 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10527 | } |
10528 | ||
10529 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10530 | .name = "cpu", |
10531 | .create = cpu_cgroup_create, | |
10532 | .destroy = cpu_cgroup_destroy, | |
10533 | .can_attach = cpu_cgroup_can_attach, | |
10534 | .attach = cpu_cgroup_attach, | |
10535 | .populate = cpu_cgroup_populate, | |
10536 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10537 | .early_init = 1, |
10538 | }; | |
10539 | ||
052f1dc7 | 10540 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10541 | |
10542 | #ifdef CONFIG_CGROUP_CPUACCT | |
10543 | ||
10544 | /* | |
10545 | * CPU accounting code for task groups. | |
10546 | * | |
10547 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10548 | * (balbir@in.ibm.com). | |
10549 | */ | |
10550 | ||
934352f2 | 10551 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10552 | struct cpuacct { |
10553 | struct cgroup_subsys_state css; | |
10554 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10555 | u64 *cpuusage; | |
ef12fefa | 10556 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10557 | struct cpuacct *parent; |
d842de87 SV |
10558 | }; |
10559 | ||
10560 | struct cgroup_subsys cpuacct_subsys; | |
10561 | ||
10562 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10563 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10564 | { |
32cd756a | 10565 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10566 | struct cpuacct, css); |
10567 | } | |
10568 | ||
10569 | /* return cpu accounting group to which this task belongs */ | |
10570 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10571 | { | |
10572 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10573 | struct cpuacct, css); | |
10574 | } | |
10575 | ||
10576 | /* create a new cpu accounting group */ | |
10577 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10578 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10579 | { |
10580 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10581 | int i; |
d842de87 SV |
10582 | |
10583 | if (!ca) | |
ef12fefa | 10584 | goto out; |
d842de87 SV |
10585 | |
10586 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10587 | if (!ca->cpuusage) |
10588 | goto out_free_ca; | |
10589 | ||
10590 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10591 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10592 | goto out_free_counters; | |
d842de87 | 10593 | |
934352f2 BR |
10594 | if (cgrp->parent) |
10595 | ca->parent = cgroup_ca(cgrp->parent); | |
10596 | ||
d842de87 | 10597 | return &ca->css; |
ef12fefa BR |
10598 | |
10599 | out_free_counters: | |
10600 | while (--i >= 0) | |
10601 | percpu_counter_destroy(&ca->cpustat[i]); | |
10602 | free_percpu(ca->cpuusage); | |
10603 | out_free_ca: | |
10604 | kfree(ca); | |
10605 | out: | |
10606 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10607 | } |
10608 | ||
10609 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10610 | static void |
32cd756a | 10611 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10612 | { |
32cd756a | 10613 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10614 | int i; |
d842de87 | 10615 | |
ef12fefa BR |
10616 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10617 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10618 | free_percpu(ca->cpuusage); |
10619 | kfree(ca); | |
10620 | } | |
10621 | ||
720f5498 KC |
10622 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10623 | { | |
b36128c8 | 10624 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10625 | u64 data; |
10626 | ||
10627 | #ifndef CONFIG_64BIT | |
10628 | /* | |
10629 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10630 | */ | |
10631 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10632 | data = *cpuusage; | |
10633 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10634 | #else | |
10635 | data = *cpuusage; | |
10636 | #endif | |
10637 | ||
10638 | return data; | |
10639 | } | |
10640 | ||
10641 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10642 | { | |
b36128c8 | 10643 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10644 | |
10645 | #ifndef CONFIG_64BIT | |
10646 | /* | |
10647 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10648 | */ | |
10649 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10650 | *cpuusage = val; | |
10651 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10652 | #else | |
10653 | *cpuusage = val; | |
10654 | #endif | |
10655 | } | |
10656 | ||
d842de87 | 10657 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10658 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10659 | { |
32cd756a | 10660 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10661 | u64 totalcpuusage = 0; |
10662 | int i; | |
10663 | ||
720f5498 KC |
10664 | for_each_present_cpu(i) |
10665 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10666 | |
10667 | return totalcpuusage; | |
10668 | } | |
10669 | ||
0297b803 DG |
10670 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10671 | u64 reset) | |
10672 | { | |
10673 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10674 | int err = 0; | |
10675 | int i; | |
10676 | ||
10677 | if (reset) { | |
10678 | err = -EINVAL; | |
10679 | goto out; | |
10680 | } | |
10681 | ||
720f5498 KC |
10682 | for_each_present_cpu(i) |
10683 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10684 | |
0297b803 DG |
10685 | out: |
10686 | return err; | |
10687 | } | |
10688 | ||
e9515c3c KC |
10689 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10690 | struct seq_file *m) | |
10691 | { | |
10692 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10693 | u64 percpu; | |
10694 | int i; | |
10695 | ||
10696 | for_each_present_cpu(i) { | |
10697 | percpu = cpuacct_cpuusage_read(ca, i); | |
10698 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10699 | } | |
10700 | seq_printf(m, "\n"); | |
10701 | return 0; | |
10702 | } | |
10703 | ||
ef12fefa BR |
10704 | static const char *cpuacct_stat_desc[] = { |
10705 | [CPUACCT_STAT_USER] = "user", | |
10706 | [CPUACCT_STAT_SYSTEM] = "system", | |
10707 | }; | |
10708 | ||
10709 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10710 | struct cgroup_map_cb *cb) | |
10711 | { | |
10712 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10713 | int i; | |
10714 | ||
10715 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10716 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10717 | val = cputime64_to_clock_t(val); | |
10718 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10719 | } | |
10720 | return 0; | |
10721 | } | |
10722 | ||
d842de87 SV |
10723 | static struct cftype files[] = { |
10724 | { | |
10725 | .name = "usage", | |
f4c753b7 PM |
10726 | .read_u64 = cpuusage_read, |
10727 | .write_u64 = cpuusage_write, | |
d842de87 | 10728 | }, |
e9515c3c KC |
10729 | { |
10730 | .name = "usage_percpu", | |
10731 | .read_seq_string = cpuacct_percpu_seq_read, | |
10732 | }, | |
ef12fefa BR |
10733 | { |
10734 | .name = "stat", | |
10735 | .read_map = cpuacct_stats_show, | |
10736 | }, | |
d842de87 SV |
10737 | }; |
10738 | ||
32cd756a | 10739 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10740 | { |
32cd756a | 10741 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10742 | } |
10743 | ||
10744 | /* | |
10745 | * charge this task's execution time to its accounting group. | |
10746 | * | |
10747 | * called with rq->lock held. | |
10748 | */ | |
10749 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10750 | { | |
10751 | struct cpuacct *ca; | |
934352f2 | 10752 | int cpu; |
d842de87 | 10753 | |
c40c6f85 | 10754 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10755 | return; |
10756 | ||
934352f2 | 10757 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10758 | |
10759 | rcu_read_lock(); | |
10760 | ||
d842de87 | 10761 | ca = task_ca(tsk); |
d842de87 | 10762 | |
934352f2 | 10763 | for (; ca; ca = ca->parent) { |
b36128c8 | 10764 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10765 | *cpuusage += cputime; |
10766 | } | |
a18b83b7 BR |
10767 | |
10768 | rcu_read_unlock(); | |
d842de87 SV |
10769 | } |
10770 | ||
ef12fefa BR |
10771 | /* |
10772 | * Charge the system/user time to the task's accounting group. | |
10773 | */ | |
10774 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10775 | enum cpuacct_stat_index idx, cputime_t val) | |
10776 | { | |
10777 | struct cpuacct *ca; | |
10778 | ||
10779 | if (unlikely(!cpuacct_subsys.active)) | |
10780 | return; | |
10781 | ||
10782 | rcu_read_lock(); | |
10783 | ca = task_ca(tsk); | |
10784 | ||
10785 | do { | |
10786 | percpu_counter_add(&ca->cpustat[idx], val); | |
10787 | ca = ca->parent; | |
10788 | } while (ca); | |
10789 | rcu_read_unlock(); | |
10790 | } | |
10791 | ||
d842de87 SV |
10792 | struct cgroup_subsys cpuacct_subsys = { |
10793 | .name = "cpuacct", | |
10794 | .create = cpuacct_create, | |
10795 | .destroy = cpuacct_destroy, | |
10796 | .populate = cpuacct_populate, | |
10797 | .subsys_id = cpuacct_subsys_id, | |
10798 | }; | |
10799 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
10800 | |
10801 | #ifndef CONFIG_SMP | |
10802 | ||
10803 | int rcu_expedited_torture_stats(char *page) | |
10804 | { | |
10805 | return 0; | |
10806 | } | |
10807 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10808 | ||
10809 | void synchronize_sched_expedited(void) | |
10810 | { | |
10811 | } | |
10812 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10813 | ||
10814 | #else /* #ifndef CONFIG_SMP */ | |
10815 | ||
10816 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
10817 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
10818 | ||
10819 | #define RCU_EXPEDITED_STATE_POST -2 | |
10820 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
10821 | ||
10822 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10823 | ||
10824 | int rcu_expedited_torture_stats(char *page) | |
10825 | { | |
10826 | int cnt = 0; | |
10827 | int cpu; | |
10828 | ||
10829 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
10830 | for_each_online_cpu(cpu) { | |
10831 | cnt += sprintf(&page[cnt], " %d:%d", | |
10832 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
10833 | } | |
10834 | cnt += sprintf(&page[cnt], "\n"); | |
10835 | return cnt; | |
10836 | } | |
10837 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10838 | ||
10839 | static long synchronize_sched_expedited_count; | |
10840 | ||
10841 | /* | |
10842 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
10843 | * approach to force grace period to end quickly. This consumes | |
10844 | * significant time on all CPUs, and is thus not recommended for | |
10845 | * any sort of common-case code. | |
10846 | * | |
10847 | * Note that it is illegal to call this function while holding any | |
10848 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
10849 | * observe this restriction will result in deadlock. | |
10850 | */ | |
10851 | void synchronize_sched_expedited(void) | |
10852 | { | |
10853 | int cpu; | |
10854 | unsigned long flags; | |
10855 | bool need_full_sync = 0; | |
10856 | struct rq *rq; | |
10857 | struct migration_req *req; | |
10858 | long snap; | |
10859 | int trycount = 0; | |
10860 | ||
10861 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
10862 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
10863 | get_online_cpus(); | |
10864 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
10865 | put_online_cpus(); | |
10866 | if (trycount++ < 10) | |
10867 | udelay(trycount * num_online_cpus()); | |
10868 | else { | |
10869 | synchronize_sched(); | |
10870 | return; | |
10871 | } | |
10872 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
10873 | smp_mb(); /* ensure test happens before caller kfree */ | |
10874 | return; | |
10875 | } | |
10876 | get_online_cpus(); | |
10877 | } | |
10878 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
10879 | for_each_online_cpu(cpu) { | |
10880 | rq = cpu_rq(cpu); | |
10881 | req = &per_cpu(rcu_migration_req, cpu); | |
10882 | init_completion(&req->done); | |
10883 | req->task = NULL; | |
10884 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
10885 | spin_lock_irqsave(&rq->lock, flags); | |
10886 | list_add(&req->list, &rq->migration_queue); | |
10887 | spin_unlock_irqrestore(&rq->lock, flags); | |
10888 | wake_up_process(rq->migration_thread); | |
10889 | } | |
10890 | for_each_online_cpu(cpu) { | |
10891 | rcu_expedited_state = cpu; | |
10892 | req = &per_cpu(rcu_migration_req, cpu); | |
10893 | rq = cpu_rq(cpu); | |
10894 | wait_for_completion(&req->done); | |
10895 | spin_lock_irqsave(&rq->lock, flags); | |
10896 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) | |
10897 | need_full_sync = 1; | |
10898 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
10899 | spin_unlock_irqrestore(&rq->lock, flags); | |
10900 | } | |
10901 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10902 | mutex_unlock(&rcu_sched_expedited_mutex); | |
10903 | put_online_cpus(); | |
10904 | if (need_full_sync) | |
10905 | synchronize_sched(); | |
10906 | } | |
10907 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10908 | ||
10909 | #endif /* #else #ifndef CONFIG_SMP */ |