]>
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> |
0d905bca | 42 | #include <linux/perf_counter.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 */ | |
84e9dabf | 296 | static DEFINE_PER_CPU(struct cfs_rq, init_tg_cfs_rq) ____cacheline_aligned_in_smp; |
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); | |
301 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
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 | ||
57310a98 PZ |
379 | #ifdef CONFIG_SMP |
380 | static int root_task_group_empty(void) | |
381 | { | |
382 | return 1; | |
383 | } | |
384 | #endif | |
385 | ||
6f505b16 | 386 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
387 | static inline struct task_group *task_group(struct task_struct *p) |
388 | { | |
389 | return NULL; | |
390 | } | |
29f59db3 | 391 | |
052f1dc7 | 392 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 393 | |
6aa645ea IM |
394 | /* CFS-related fields in a runqueue */ |
395 | struct cfs_rq { | |
396 | struct load_weight load; | |
397 | unsigned long nr_running; | |
398 | ||
6aa645ea | 399 | u64 exec_clock; |
e9acbff6 | 400 | u64 min_vruntime; |
6aa645ea IM |
401 | |
402 | struct rb_root tasks_timeline; | |
403 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
404 | |
405 | struct list_head tasks; | |
406 | struct list_head *balance_iterator; | |
407 | ||
408 | /* | |
409 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
410 | * It is set to NULL otherwise (i.e when none are currently running). |
411 | */ | |
4793241b | 412 | struct sched_entity *curr, *next, *last; |
ddc97297 | 413 | |
5ac5c4d6 | 414 | unsigned int nr_spread_over; |
ddc97297 | 415 | |
62160e3f | 416 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
417 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
418 | ||
41a2d6cf IM |
419 | /* |
420 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
421 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
422 | * (like users, containers etc.) | |
423 | * | |
424 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
425 | * list is used during load balance. | |
426 | */ | |
41a2d6cf IM |
427 | struct list_head leaf_cfs_rq_list; |
428 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
429 | |
430 | #ifdef CONFIG_SMP | |
c09595f6 | 431 | /* |
c8cba857 | 432 | * the part of load.weight contributed by tasks |
c09595f6 | 433 | */ |
c8cba857 | 434 | unsigned long task_weight; |
c09595f6 | 435 | |
c8cba857 PZ |
436 | /* |
437 | * h_load = weight * f(tg) | |
438 | * | |
439 | * Where f(tg) is the recursive weight fraction assigned to | |
440 | * this group. | |
441 | */ | |
442 | unsigned long h_load; | |
c09595f6 | 443 | |
c8cba857 PZ |
444 | /* |
445 | * this cpu's part of tg->shares | |
446 | */ | |
447 | unsigned long shares; | |
f1d239f7 PZ |
448 | |
449 | /* | |
450 | * load.weight at the time we set shares | |
451 | */ | |
452 | unsigned long rq_weight; | |
c09595f6 | 453 | #endif |
6aa645ea IM |
454 | #endif |
455 | }; | |
1da177e4 | 456 | |
6aa645ea IM |
457 | /* Real-Time classes' related field in a runqueue: */ |
458 | struct rt_rq { | |
459 | struct rt_prio_array active; | |
63489e45 | 460 | unsigned long rt_nr_running; |
052f1dc7 | 461 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
462 | struct { |
463 | int curr; /* highest queued rt task prio */ | |
398a153b | 464 | #ifdef CONFIG_SMP |
e864c499 | 465 | int next; /* next highest */ |
398a153b | 466 | #endif |
e864c499 | 467 | } highest_prio; |
6f505b16 | 468 | #endif |
fa85ae24 | 469 | #ifdef CONFIG_SMP |
73fe6aae | 470 | unsigned long rt_nr_migratory; |
a1ba4d8b | 471 | unsigned long rt_nr_total; |
a22d7fc1 | 472 | int overloaded; |
917b627d | 473 | struct plist_head pushable_tasks; |
fa85ae24 | 474 | #endif |
6f505b16 | 475 | int rt_throttled; |
fa85ae24 | 476 | u64 rt_time; |
ac086bc2 | 477 | u64 rt_runtime; |
ea736ed5 | 478 | /* Nests inside the rq lock: */ |
ac086bc2 | 479 | spinlock_t rt_runtime_lock; |
6f505b16 | 480 | |
052f1dc7 | 481 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
482 | unsigned long rt_nr_boosted; |
483 | ||
6f505b16 PZ |
484 | struct rq *rq; |
485 | struct list_head leaf_rt_rq_list; | |
486 | struct task_group *tg; | |
487 | struct sched_rt_entity *rt_se; | |
488 | #endif | |
6aa645ea IM |
489 | }; |
490 | ||
57d885fe GH |
491 | #ifdef CONFIG_SMP |
492 | ||
493 | /* | |
494 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
495 | * variables. Each exclusive cpuset essentially defines an island domain by |
496 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
497 | * exclusive cpuset is created, we also create and attach a new root-domain |
498 | * object. | |
499 | * | |
57d885fe GH |
500 | */ |
501 | struct root_domain { | |
502 | atomic_t refcount; | |
c6c4927b RR |
503 | cpumask_var_t span; |
504 | cpumask_var_t online; | |
637f5085 | 505 | |
0eab9146 | 506 | /* |
637f5085 GH |
507 | * The "RT overload" flag: it gets set if a CPU has more than |
508 | * one runnable RT task. | |
509 | */ | |
c6c4927b | 510 | cpumask_var_t rto_mask; |
0eab9146 | 511 | atomic_t rto_count; |
6e0534f2 GH |
512 | #ifdef CONFIG_SMP |
513 | struct cpupri cpupri; | |
514 | #endif | |
7a09b1a2 VS |
515 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
516 | /* | |
517 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
518 | * used when most cpus are idle in the system indicating overall very | |
519 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
520 | */ | |
521 | unsigned int sched_mc_preferred_wakeup_cpu; | |
522 | #endif | |
57d885fe GH |
523 | }; |
524 | ||
dc938520 GH |
525 | /* |
526 | * By default the system creates a single root-domain with all cpus as | |
527 | * members (mimicking the global state we have today). | |
528 | */ | |
57d885fe GH |
529 | static struct root_domain def_root_domain; |
530 | ||
531 | #endif | |
532 | ||
1da177e4 LT |
533 | /* |
534 | * This is the main, per-CPU runqueue data structure. | |
535 | * | |
536 | * Locking rule: those places that want to lock multiple runqueues | |
537 | * (such as the load balancing or the thread migration code), lock | |
538 | * acquire operations must be ordered by ascending &runqueue. | |
539 | */ | |
70b97a7f | 540 | struct rq { |
d8016491 IM |
541 | /* runqueue lock: */ |
542 | spinlock_t lock; | |
1da177e4 LT |
543 | |
544 | /* | |
545 | * nr_running and cpu_load should be in the same cacheline because | |
546 | * remote CPUs use both these fields when doing load calculation. | |
547 | */ | |
548 | unsigned long nr_running; | |
6aa645ea IM |
549 | #define CPU_LOAD_IDX_MAX 5 |
550 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 551 | #ifdef CONFIG_NO_HZ |
15934a37 | 552 | unsigned long last_tick_seen; |
46cb4b7c SS |
553 | unsigned char in_nohz_recently; |
554 | #endif | |
d8016491 IM |
555 | /* capture load from *all* tasks on this cpu: */ |
556 | struct load_weight load; | |
6aa645ea IM |
557 | unsigned long nr_load_updates; |
558 | u64 nr_switches; | |
23a185ca | 559 | u64 nr_migrations_in; |
6aa645ea IM |
560 | |
561 | struct cfs_rq cfs; | |
6f505b16 | 562 | struct rt_rq rt; |
6f505b16 | 563 | |
6aa645ea | 564 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
565 | /* list of leaf cfs_rq on this cpu: */ |
566 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
567 | #endif |
568 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 569 | struct list_head leaf_rt_rq_list; |
1da177e4 | 570 | #endif |
1da177e4 LT |
571 | |
572 | /* | |
573 | * This is part of a global counter where only the total sum | |
574 | * over all CPUs matters. A task can increase this counter on | |
575 | * one CPU and if it got migrated afterwards it may decrease | |
576 | * it on another CPU. Always updated under the runqueue lock: | |
577 | */ | |
578 | unsigned long nr_uninterruptible; | |
579 | ||
36c8b586 | 580 | struct task_struct *curr, *idle; |
c9819f45 | 581 | unsigned long next_balance; |
1da177e4 | 582 | struct mm_struct *prev_mm; |
6aa645ea | 583 | |
3e51f33f | 584 | u64 clock; |
6aa645ea | 585 | |
1da177e4 LT |
586 | atomic_t nr_iowait; |
587 | ||
588 | #ifdef CONFIG_SMP | |
0eab9146 | 589 | struct root_domain *rd; |
1da177e4 LT |
590 | struct sched_domain *sd; |
591 | ||
a0a522ce | 592 | unsigned char idle_at_tick; |
1da177e4 | 593 | /* For active balancing */ |
3f029d3c | 594 | int post_schedule; |
1da177e4 LT |
595 | int active_balance; |
596 | int push_cpu; | |
d8016491 IM |
597 | /* cpu of this runqueue: */ |
598 | int cpu; | |
1f11eb6a | 599 | int online; |
1da177e4 | 600 | |
a8a51d5e | 601 | unsigned long avg_load_per_task; |
1da177e4 | 602 | |
36c8b586 | 603 | struct task_struct *migration_thread; |
1da177e4 | 604 | struct list_head migration_queue; |
e9e9250b PZ |
605 | |
606 | u64 rt_avg; | |
607 | u64 age_stamp; | |
1da177e4 LT |
608 | #endif |
609 | ||
dce48a84 TG |
610 | /* calc_load related fields */ |
611 | unsigned long calc_load_update; | |
612 | long calc_load_active; | |
613 | ||
8f4d37ec | 614 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
615 | #ifdef CONFIG_SMP |
616 | int hrtick_csd_pending; | |
617 | struct call_single_data hrtick_csd; | |
618 | #endif | |
8f4d37ec PZ |
619 | struct hrtimer hrtick_timer; |
620 | #endif | |
621 | ||
1da177e4 LT |
622 | #ifdef CONFIG_SCHEDSTATS |
623 | /* latency stats */ | |
624 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
625 | unsigned long long rq_cpu_time; |
626 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
627 | |
628 | /* sys_sched_yield() stats */ | |
480b9434 | 629 | unsigned int yld_count; |
1da177e4 LT |
630 | |
631 | /* schedule() stats */ | |
480b9434 KC |
632 | unsigned int sched_switch; |
633 | unsigned int sched_count; | |
634 | unsigned int sched_goidle; | |
1da177e4 LT |
635 | |
636 | /* try_to_wake_up() stats */ | |
480b9434 KC |
637 | unsigned int ttwu_count; |
638 | unsigned int ttwu_local; | |
b8efb561 IM |
639 | |
640 | /* BKL stats */ | |
480b9434 | 641 | unsigned int bkl_count; |
1da177e4 LT |
642 | #endif |
643 | }; | |
644 | ||
f34e3b61 | 645 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 646 | |
15afe09b | 647 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 648 | { |
15afe09b | 649 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
650 | } |
651 | ||
0a2966b4 CL |
652 | static inline int cpu_of(struct rq *rq) |
653 | { | |
654 | #ifdef CONFIG_SMP | |
655 | return rq->cpu; | |
656 | #else | |
657 | return 0; | |
658 | #endif | |
659 | } | |
660 | ||
674311d5 NP |
661 | /* |
662 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 663 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
664 | * |
665 | * The domain tree of any CPU may only be accessed from within | |
666 | * preempt-disabled sections. | |
667 | */ | |
48f24c4d IM |
668 | #define for_each_domain(cpu, __sd) \ |
669 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
670 | |
671 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
672 | #define this_rq() (&__get_cpu_var(runqueues)) | |
673 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
674 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 675 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 676 | |
aa9c4c0f | 677 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
678 | { |
679 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
680 | } | |
681 | ||
bf5c91ba IM |
682 | /* |
683 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
684 | */ | |
685 | #ifdef CONFIG_SCHED_DEBUG | |
686 | # define const_debug __read_mostly | |
687 | #else | |
688 | # define const_debug static const | |
689 | #endif | |
690 | ||
017730c1 IM |
691 | /** |
692 | * runqueue_is_locked | |
693 | * | |
694 | * Returns true if the current cpu runqueue is locked. | |
695 | * This interface allows printk to be called with the runqueue lock | |
696 | * held and know whether or not it is OK to wake up the klogd. | |
697 | */ | |
698 | int runqueue_is_locked(void) | |
699 | { | |
700 | int cpu = get_cpu(); | |
701 | struct rq *rq = cpu_rq(cpu); | |
702 | int ret; | |
703 | ||
704 | ret = spin_is_locked(&rq->lock); | |
705 | put_cpu(); | |
706 | return ret; | |
707 | } | |
708 | ||
bf5c91ba IM |
709 | /* |
710 | * Debugging: various feature bits | |
711 | */ | |
f00b45c1 PZ |
712 | |
713 | #define SCHED_FEAT(name, enabled) \ | |
714 | __SCHED_FEAT_##name , | |
715 | ||
bf5c91ba | 716 | enum { |
f00b45c1 | 717 | #include "sched_features.h" |
bf5c91ba IM |
718 | }; |
719 | ||
f00b45c1 PZ |
720 | #undef SCHED_FEAT |
721 | ||
722 | #define SCHED_FEAT(name, enabled) \ | |
723 | (1UL << __SCHED_FEAT_##name) * enabled | | |
724 | ||
bf5c91ba | 725 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
726 | #include "sched_features.h" |
727 | 0; | |
728 | ||
729 | #undef SCHED_FEAT | |
730 | ||
731 | #ifdef CONFIG_SCHED_DEBUG | |
732 | #define SCHED_FEAT(name, enabled) \ | |
733 | #name , | |
734 | ||
983ed7a6 | 735 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
736 | #include "sched_features.h" |
737 | NULL | |
738 | }; | |
739 | ||
740 | #undef SCHED_FEAT | |
741 | ||
34f3a814 | 742 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 743 | { |
f00b45c1 PZ |
744 | int i; |
745 | ||
746 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
747 | if (!(sysctl_sched_features & (1UL << i))) |
748 | seq_puts(m, "NO_"); | |
749 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 750 | } |
34f3a814 | 751 | seq_puts(m, "\n"); |
f00b45c1 | 752 | |
34f3a814 | 753 | return 0; |
f00b45c1 PZ |
754 | } |
755 | ||
756 | static ssize_t | |
757 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
758 | size_t cnt, loff_t *ppos) | |
759 | { | |
760 | char buf[64]; | |
761 | char *cmp = buf; | |
762 | int neg = 0; | |
763 | int i; | |
764 | ||
765 | if (cnt > 63) | |
766 | cnt = 63; | |
767 | ||
768 | if (copy_from_user(&buf, ubuf, cnt)) | |
769 | return -EFAULT; | |
770 | ||
771 | buf[cnt] = 0; | |
772 | ||
c24b7c52 | 773 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
774 | neg = 1; |
775 | cmp += 3; | |
776 | } | |
777 | ||
778 | for (i = 0; sched_feat_names[i]; i++) { | |
779 | int len = strlen(sched_feat_names[i]); | |
780 | ||
781 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
782 | if (neg) | |
783 | sysctl_sched_features &= ~(1UL << i); | |
784 | else | |
785 | sysctl_sched_features |= (1UL << i); | |
786 | break; | |
787 | } | |
788 | } | |
789 | ||
790 | if (!sched_feat_names[i]) | |
791 | return -EINVAL; | |
792 | ||
793 | filp->f_pos += cnt; | |
794 | ||
795 | return cnt; | |
796 | } | |
797 | ||
34f3a814 LZ |
798 | static int sched_feat_open(struct inode *inode, struct file *filp) |
799 | { | |
800 | return single_open(filp, sched_feat_show, NULL); | |
801 | } | |
802 | ||
f00b45c1 | 803 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
804 | .open = sched_feat_open, |
805 | .write = sched_feat_write, | |
806 | .read = seq_read, | |
807 | .llseek = seq_lseek, | |
808 | .release = single_release, | |
f00b45c1 PZ |
809 | }; |
810 | ||
811 | static __init int sched_init_debug(void) | |
812 | { | |
f00b45c1 PZ |
813 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
814 | &sched_feat_fops); | |
815 | ||
816 | return 0; | |
817 | } | |
818 | late_initcall(sched_init_debug); | |
819 | ||
820 | #endif | |
821 | ||
822 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 823 | |
b82d9fdd PZ |
824 | /* |
825 | * Number of tasks to iterate in a single balance run. | |
826 | * Limited because this is done with IRQs disabled. | |
827 | */ | |
828 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
829 | ||
2398f2c6 PZ |
830 | /* |
831 | * ratelimit for updating the group shares. | |
55cd5340 | 832 | * default: 0.25ms |
2398f2c6 | 833 | */ |
55cd5340 | 834 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 835 | |
ffda12a1 PZ |
836 | /* |
837 | * Inject some fuzzyness into changing the per-cpu group shares | |
838 | * this avoids remote rq-locks at the expense of fairness. | |
839 | * default: 4 | |
840 | */ | |
841 | unsigned int sysctl_sched_shares_thresh = 4; | |
842 | ||
e9e9250b PZ |
843 | /* |
844 | * period over which we average the RT time consumption, measured | |
845 | * in ms. | |
846 | * | |
847 | * default: 1s | |
848 | */ | |
849 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
850 | ||
fa85ae24 | 851 | /* |
9f0c1e56 | 852 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
853 | * default: 1s |
854 | */ | |
9f0c1e56 | 855 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 856 | |
6892b75e IM |
857 | static __read_mostly int scheduler_running; |
858 | ||
9f0c1e56 PZ |
859 | /* |
860 | * part of the period that we allow rt tasks to run in us. | |
861 | * default: 0.95s | |
862 | */ | |
863 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 864 | |
d0b27fa7 PZ |
865 | static inline u64 global_rt_period(void) |
866 | { | |
867 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
868 | } | |
869 | ||
870 | static inline u64 global_rt_runtime(void) | |
871 | { | |
e26873bb | 872 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
873 | return RUNTIME_INF; |
874 | ||
875 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
876 | } | |
fa85ae24 | 877 | |
1da177e4 | 878 | #ifndef prepare_arch_switch |
4866cde0 NP |
879 | # define prepare_arch_switch(next) do { } while (0) |
880 | #endif | |
881 | #ifndef finish_arch_switch | |
882 | # define finish_arch_switch(prev) do { } while (0) | |
883 | #endif | |
884 | ||
051a1d1a DA |
885 | static inline int task_current(struct rq *rq, struct task_struct *p) |
886 | { | |
887 | return rq->curr == p; | |
888 | } | |
889 | ||
4866cde0 | 890 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 891 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 892 | { |
051a1d1a | 893 | return task_current(rq, p); |
4866cde0 NP |
894 | } |
895 | ||
70b97a7f | 896 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
897 | { |
898 | } | |
899 | ||
70b97a7f | 900 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 901 | { |
da04c035 IM |
902 | #ifdef CONFIG_DEBUG_SPINLOCK |
903 | /* this is a valid case when another task releases the spinlock */ | |
904 | rq->lock.owner = current; | |
905 | #endif | |
8a25d5de IM |
906 | /* |
907 | * If we are tracking spinlock dependencies then we have to | |
908 | * fix up the runqueue lock - which gets 'carried over' from | |
909 | * prev into current: | |
910 | */ | |
911 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
912 | ||
4866cde0 NP |
913 | spin_unlock_irq(&rq->lock); |
914 | } | |
915 | ||
916 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 917 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
918 | { |
919 | #ifdef CONFIG_SMP | |
920 | return p->oncpu; | |
921 | #else | |
051a1d1a | 922 | return task_current(rq, p); |
4866cde0 NP |
923 | #endif |
924 | } | |
925 | ||
70b97a7f | 926 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
927 | { |
928 | #ifdef CONFIG_SMP | |
929 | /* | |
930 | * We can optimise this out completely for !SMP, because the | |
931 | * SMP rebalancing from interrupt is the only thing that cares | |
932 | * here. | |
933 | */ | |
934 | next->oncpu = 1; | |
935 | #endif | |
936 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
937 | spin_unlock_irq(&rq->lock); | |
938 | #else | |
939 | spin_unlock(&rq->lock); | |
940 | #endif | |
941 | } | |
942 | ||
70b97a7f | 943 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
944 | { |
945 | #ifdef CONFIG_SMP | |
946 | /* | |
947 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
948 | * We must ensure this doesn't happen until the switch is completely | |
949 | * finished. | |
950 | */ | |
951 | smp_wmb(); | |
952 | prev->oncpu = 0; | |
953 | #endif | |
954 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
955 | local_irq_enable(); | |
1da177e4 | 956 | #endif |
4866cde0 NP |
957 | } |
958 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 959 | |
b29739f9 IM |
960 | /* |
961 | * __task_rq_lock - lock the runqueue a given task resides on. | |
962 | * Must be called interrupts disabled. | |
963 | */ | |
70b97a7f | 964 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
965 | __acquires(rq->lock) |
966 | { | |
3a5c359a AK |
967 | for (;;) { |
968 | struct rq *rq = task_rq(p); | |
969 | spin_lock(&rq->lock); | |
970 | if (likely(rq == task_rq(p))) | |
971 | return rq; | |
b29739f9 | 972 | spin_unlock(&rq->lock); |
b29739f9 | 973 | } |
b29739f9 IM |
974 | } |
975 | ||
1da177e4 LT |
976 | /* |
977 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 978 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
979 | * explicitly disabling preemption. |
980 | */ | |
70b97a7f | 981 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
982 | __acquires(rq->lock) |
983 | { | |
70b97a7f | 984 | struct rq *rq; |
1da177e4 | 985 | |
3a5c359a AK |
986 | for (;;) { |
987 | local_irq_save(*flags); | |
988 | rq = task_rq(p); | |
989 | spin_lock(&rq->lock); | |
990 | if (likely(rq == task_rq(p))) | |
991 | return rq; | |
1da177e4 | 992 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 993 | } |
1da177e4 LT |
994 | } |
995 | ||
ad474cac ON |
996 | void task_rq_unlock_wait(struct task_struct *p) |
997 | { | |
998 | struct rq *rq = task_rq(p); | |
999 | ||
1000 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1001 | spin_unlock_wait(&rq->lock); | |
1002 | } | |
1003 | ||
a9957449 | 1004 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1005 | __releases(rq->lock) |
1006 | { | |
1007 | spin_unlock(&rq->lock); | |
1008 | } | |
1009 | ||
70b97a7f | 1010 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1011 | __releases(rq->lock) |
1012 | { | |
1013 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1014 | } | |
1015 | ||
1da177e4 | 1016 | /* |
cc2a73b5 | 1017 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1018 | */ |
a9957449 | 1019 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1020 | __acquires(rq->lock) |
1021 | { | |
70b97a7f | 1022 | struct rq *rq; |
1da177e4 LT |
1023 | |
1024 | local_irq_disable(); | |
1025 | rq = this_rq(); | |
1026 | spin_lock(&rq->lock); | |
1027 | ||
1028 | return rq; | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | #ifdef CONFIG_SCHED_HRTICK |
1032 | /* | |
1033 | * Use HR-timers to deliver accurate preemption points. | |
1034 | * | |
1035 | * Its all a bit involved since we cannot program an hrt while holding the | |
1036 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1037 | * reschedule event. | |
1038 | * | |
1039 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1040 | * rq->lock. | |
1041 | */ | |
8f4d37ec PZ |
1042 | |
1043 | /* | |
1044 | * Use hrtick when: | |
1045 | * - enabled by features | |
1046 | * - hrtimer is actually high res | |
1047 | */ | |
1048 | static inline int hrtick_enabled(struct rq *rq) | |
1049 | { | |
1050 | if (!sched_feat(HRTICK)) | |
1051 | return 0; | |
ba42059f | 1052 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1053 | return 0; |
8f4d37ec PZ |
1054 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1055 | } | |
1056 | ||
8f4d37ec PZ |
1057 | static void hrtick_clear(struct rq *rq) |
1058 | { | |
1059 | if (hrtimer_active(&rq->hrtick_timer)) | |
1060 | hrtimer_cancel(&rq->hrtick_timer); | |
1061 | } | |
1062 | ||
8f4d37ec PZ |
1063 | /* |
1064 | * High-resolution timer tick. | |
1065 | * Runs from hardirq context with interrupts disabled. | |
1066 | */ | |
1067 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1068 | { | |
1069 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1070 | ||
1071 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1072 | ||
1073 | spin_lock(&rq->lock); | |
3e51f33f | 1074 | update_rq_clock(rq); |
8f4d37ec PZ |
1075 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1076 | spin_unlock(&rq->lock); | |
1077 | ||
1078 | return HRTIMER_NORESTART; | |
1079 | } | |
1080 | ||
95e904c7 | 1081 | #ifdef CONFIG_SMP |
31656519 PZ |
1082 | /* |
1083 | * called from hardirq (IPI) context | |
1084 | */ | |
1085 | static void __hrtick_start(void *arg) | |
b328ca18 | 1086 | { |
31656519 | 1087 | struct rq *rq = arg; |
b328ca18 | 1088 | |
31656519 PZ |
1089 | spin_lock(&rq->lock); |
1090 | hrtimer_restart(&rq->hrtick_timer); | |
1091 | rq->hrtick_csd_pending = 0; | |
1092 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1093 | } |
1094 | ||
31656519 PZ |
1095 | /* |
1096 | * Called to set the hrtick timer state. | |
1097 | * | |
1098 | * called with rq->lock held and irqs disabled | |
1099 | */ | |
1100 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1101 | { |
31656519 PZ |
1102 | struct hrtimer *timer = &rq->hrtick_timer; |
1103 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1104 | |
cc584b21 | 1105 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1106 | |
1107 | if (rq == this_rq()) { | |
1108 | hrtimer_restart(timer); | |
1109 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1110 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1111 | rq->hrtick_csd_pending = 1; |
1112 | } | |
b328ca18 PZ |
1113 | } |
1114 | ||
1115 | static int | |
1116 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1117 | { | |
1118 | int cpu = (int)(long)hcpu; | |
1119 | ||
1120 | switch (action) { | |
1121 | case CPU_UP_CANCELED: | |
1122 | case CPU_UP_CANCELED_FROZEN: | |
1123 | case CPU_DOWN_PREPARE: | |
1124 | case CPU_DOWN_PREPARE_FROZEN: | |
1125 | case CPU_DEAD: | |
1126 | case CPU_DEAD_FROZEN: | |
31656519 | 1127 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1128 | return NOTIFY_OK; |
1129 | } | |
1130 | ||
1131 | return NOTIFY_DONE; | |
1132 | } | |
1133 | ||
fa748203 | 1134 | static __init void init_hrtick(void) |
b328ca18 PZ |
1135 | { |
1136 | hotcpu_notifier(hotplug_hrtick, 0); | |
1137 | } | |
31656519 PZ |
1138 | #else |
1139 | /* | |
1140 | * Called to set the hrtick timer state. | |
1141 | * | |
1142 | * called with rq->lock held and irqs disabled | |
1143 | */ | |
1144 | static void hrtick_start(struct rq *rq, u64 delay) | |
1145 | { | |
7f1e2ca9 | 1146 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1147 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1148 | } |
b328ca18 | 1149 | |
006c75f1 | 1150 | static inline void init_hrtick(void) |
8f4d37ec | 1151 | { |
8f4d37ec | 1152 | } |
31656519 | 1153 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1154 | |
31656519 | 1155 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1156 | { |
31656519 PZ |
1157 | #ifdef CONFIG_SMP |
1158 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1159 | |
31656519 PZ |
1160 | rq->hrtick_csd.flags = 0; |
1161 | rq->hrtick_csd.func = __hrtick_start; | |
1162 | rq->hrtick_csd.info = rq; | |
1163 | #endif | |
8f4d37ec | 1164 | |
31656519 PZ |
1165 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1166 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1167 | } |
006c75f1 | 1168 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1169 | static inline void hrtick_clear(struct rq *rq) |
1170 | { | |
1171 | } | |
1172 | ||
8f4d37ec PZ |
1173 | static inline void init_rq_hrtick(struct rq *rq) |
1174 | { | |
1175 | } | |
1176 | ||
b328ca18 PZ |
1177 | static inline void init_hrtick(void) |
1178 | { | |
1179 | } | |
006c75f1 | 1180 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1181 | |
c24d20db IM |
1182 | /* |
1183 | * resched_task - mark a task 'to be rescheduled now'. | |
1184 | * | |
1185 | * On UP this means the setting of the need_resched flag, on SMP it | |
1186 | * might also involve a cross-CPU call to trigger the scheduler on | |
1187 | * the target CPU. | |
1188 | */ | |
1189 | #ifdef CONFIG_SMP | |
1190 | ||
1191 | #ifndef tsk_is_polling | |
1192 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1193 | #endif | |
1194 | ||
31656519 | 1195 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1196 | { |
1197 | int cpu; | |
1198 | ||
1199 | assert_spin_locked(&task_rq(p)->lock); | |
1200 | ||
5ed0cec0 | 1201 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1202 | return; |
1203 | ||
5ed0cec0 | 1204 | set_tsk_need_resched(p); |
c24d20db IM |
1205 | |
1206 | cpu = task_cpu(p); | |
1207 | if (cpu == smp_processor_id()) | |
1208 | return; | |
1209 | ||
1210 | /* NEED_RESCHED must be visible before we test polling */ | |
1211 | smp_mb(); | |
1212 | if (!tsk_is_polling(p)) | |
1213 | smp_send_reschedule(cpu); | |
1214 | } | |
1215 | ||
1216 | static void resched_cpu(int cpu) | |
1217 | { | |
1218 | struct rq *rq = cpu_rq(cpu); | |
1219 | unsigned long flags; | |
1220 | ||
1221 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1222 | return; | |
1223 | resched_task(cpu_curr(cpu)); | |
1224 | spin_unlock_irqrestore(&rq->lock, flags); | |
1225 | } | |
06d8308c TG |
1226 | |
1227 | #ifdef CONFIG_NO_HZ | |
1228 | /* | |
1229 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1230 | * idle CPU then this timer might expire before the next timer event | |
1231 | * which is scheduled to wake up that CPU. In case of a completely | |
1232 | * idle system the next event might even be infinite time into the | |
1233 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1234 | * leaves the inner idle loop so the newly added timer is taken into | |
1235 | * account when the CPU goes back to idle and evaluates the timer | |
1236 | * wheel for the next timer event. | |
1237 | */ | |
1238 | void wake_up_idle_cpu(int cpu) | |
1239 | { | |
1240 | struct rq *rq = cpu_rq(cpu); | |
1241 | ||
1242 | if (cpu == smp_processor_id()) | |
1243 | return; | |
1244 | ||
1245 | /* | |
1246 | * This is safe, as this function is called with the timer | |
1247 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1248 | * to idle and has not yet set rq->curr to idle then it will | |
1249 | * be serialized on the timer wheel base lock and take the new | |
1250 | * timer into account automatically. | |
1251 | */ | |
1252 | if (rq->curr != rq->idle) | |
1253 | return; | |
1254 | ||
1255 | /* | |
1256 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1257 | * lockless. The worst case is that the other CPU runs the | |
1258 | * idle task through an additional NOOP schedule() | |
1259 | */ | |
5ed0cec0 | 1260 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1261 | |
1262 | /* NEED_RESCHED must be visible before we test polling */ | |
1263 | smp_mb(); | |
1264 | if (!tsk_is_polling(rq->idle)) | |
1265 | smp_send_reschedule(cpu); | |
1266 | } | |
6d6bc0ad | 1267 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1268 | |
e9e9250b PZ |
1269 | static u64 sched_avg_period(void) |
1270 | { | |
1271 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1272 | } | |
1273 | ||
1274 | static void sched_avg_update(struct rq *rq) | |
1275 | { | |
1276 | s64 period = sched_avg_period(); | |
1277 | ||
1278 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1279 | rq->age_stamp += period; | |
1280 | rq->rt_avg /= 2; | |
1281 | } | |
1282 | } | |
1283 | ||
1284 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1285 | { | |
1286 | rq->rt_avg += rt_delta; | |
1287 | sched_avg_update(rq); | |
1288 | } | |
1289 | ||
6d6bc0ad | 1290 | #else /* !CONFIG_SMP */ |
31656519 | 1291 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1292 | { |
1293 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1294 | set_tsk_need_resched(p); |
c24d20db | 1295 | } |
e9e9250b PZ |
1296 | |
1297 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1298 | { | |
1299 | } | |
6d6bc0ad | 1300 | #endif /* CONFIG_SMP */ |
c24d20db | 1301 | |
45bf76df IM |
1302 | #if BITS_PER_LONG == 32 |
1303 | # define WMULT_CONST (~0UL) | |
1304 | #else | |
1305 | # define WMULT_CONST (1UL << 32) | |
1306 | #endif | |
1307 | ||
1308 | #define WMULT_SHIFT 32 | |
1309 | ||
194081eb IM |
1310 | /* |
1311 | * Shift right and round: | |
1312 | */ | |
cf2ab469 | 1313 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1314 | |
a7be37ac PZ |
1315 | /* |
1316 | * delta *= weight / lw | |
1317 | */ | |
cb1c4fc9 | 1318 | static unsigned long |
45bf76df IM |
1319 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1320 | struct load_weight *lw) | |
1321 | { | |
1322 | u64 tmp; | |
1323 | ||
7a232e03 LJ |
1324 | if (!lw->inv_weight) { |
1325 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1326 | lw->inv_weight = 1; | |
1327 | else | |
1328 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1329 | / (lw->weight+1); | |
1330 | } | |
45bf76df IM |
1331 | |
1332 | tmp = (u64)delta_exec * weight; | |
1333 | /* | |
1334 | * Check whether we'd overflow the 64-bit multiplication: | |
1335 | */ | |
194081eb | 1336 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1337 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1338 | WMULT_SHIFT/2); |
1339 | else | |
cf2ab469 | 1340 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1341 | |
ecf691da | 1342 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1343 | } |
1344 | ||
1091985b | 1345 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1346 | { |
1347 | lw->weight += inc; | |
e89996ae | 1348 | lw->inv_weight = 0; |
45bf76df IM |
1349 | } |
1350 | ||
1091985b | 1351 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1352 | { |
1353 | lw->weight -= dec; | |
e89996ae | 1354 | lw->inv_weight = 0; |
45bf76df IM |
1355 | } |
1356 | ||
2dd73a4f PW |
1357 | /* |
1358 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1359 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1360 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1361 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1362 | * scaled version of the new time slice allocation that they receive on time |
1363 | * slice expiry etc. | |
1364 | */ | |
1365 | ||
cce7ade8 PZ |
1366 | #define WEIGHT_IDLEPRIO 3 |
1367 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1368 | |
1369 | /* | |
1370 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1371 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1372 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1373 | * that remained on nice 0. | |
1374 | * | |
1375 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1376 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1377 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1378 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1379 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1380 | */ |
1381 | static const int prio_to_weight[40] = { | |
254753dc IM |
1382 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1383 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1384 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1385 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1386 | /* 0 */ 1024, 820, 655, 526, 423, | |
1387 | /* 5 */ 335, 272, 215, 172, 137, | |
1388 | /* 10 */ 110, 87, 70, 56, 45, | |
1389 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1390 | }; |
1391 | ||
5714d2de IM |
1392 | /* |
1393 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1394 | * | |
1395 | * In cases where the weight does not change often, we can use the | |
1396 | * precalculated inverse to speed up arithmetics by turning divisions | |
1397 | * into multiplications: | |
1398 | */ | |
dd41f596 | 1399 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1400 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1401 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1402 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1403 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1404 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1405 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1406 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1407 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1408 | }; |
2dd73a4f | 1409 | |
dd41f596 IM |
1410 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1411 | ||
1412 | /* | |
1413 | * runqueue iterator, to support SMP load-balancing between different | |
1414 | * scheduling classes, without having to expose their internal data | |
1415 | * structures to the load-balancing proper: | |
1416 | */ | |
1417 | struct rq_iterator { | |
1418 | void *arg; | |
1419 | struct task_struct *(*start)(void *); | |
1420 | struct task_struct *(*next)(void *); | |
1421 | }; | |
1422 | ||
e1d1484f PW |
1423 | #ifdef CONFIG_SMP |
1424 | static unsigned long | |
1425 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1426 | unsigned long max_load_move, struct sched_domain *sd, | |
1427 | enum cpu_idle_type idle, int *all_pinned, | |
1428 | int *this_best_prio, struct rq_iterator *iterator); | |
1429 | ||
1430 | static int | |
1431 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1432 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1433 | struct rq_iterator *iterator); | |
e1d1484f | 1434 | #endif |
dd41f596 | 1435 | |
ef12fefa BR |
1436 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1437 | enum cpuacct_stat_index { | |
1438 | CPUACCT_STAT_USER, /* ... user mode */ | |
1439 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1440 | ||
1441 | CPUACCT_STAT_NSTATS, | |
1442 | }; | |
1443 | ||
d842de87 SV |
1444 | #ifdef CONFIG_CGROUP_CPUACCT |
1445 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1446 | static void cpuacct_update_stats(struct task_struct *tsk, |
1447 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1448 | #else |
1449 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1450 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1451 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1452 | #endif |
1453 | ||
18d95a28 PZ |
1454 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1455 | { | |
1456 | update_load_add(&rq->load, load); | |
1457 | } | |
1458 | ||
1459 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1460 | { | |
1461 | update_load_sub(&rq->load, load); | |
1462 | } | |
1463 | ||
7940ca36 | 1464 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1465 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1466 | |
1467 | /* | |
1468 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1469 | * leaving it for the final time. | |
1470 | */ | |
eb755805 | 1471 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1472 | { |
1473 | struct task_group *parent, *child; | |
eb755805 | 1474 | int ret; |
c09595f6 PZ |
1475 | |
1476 | rcu_read_lock(); | |
1477 | parent = &root_task_group; | |
1478 | down: | |
eb755805 PZ |
1479 | ret = (*down)(parent, data); |
1480 | if (ret) | |
1481 | goto out_unlock; | |
c09595f6 PZ |
1482 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1483 | parent = child; | |
1484 | goto down; | |
1485 | ||
1486 | up: | |
1487 | continue; | |
1488 | } | |
eb755805 PZ |
1489 | ret = (*up)(parent, data); |
1490 | if (ret) | |
1491 | goto out_unlock; | |
c09595f6 PZ |
1492 | |
1493 | child = parent; | |
1494 | parent = parent->parent; | |
1495 | if (parent) | |
1496 | goto up; | |
eb755805 | 1497 | out_unlock: |
c09595f6 | 1498 | rcu_read_unlock(); |
eb755805 PZ |
1499 | |
1500 | return ret; | |
c09595f6 PZ |
1501 | } |
1502 | ||
eb755805 PZ |
1503 | static int tg_nop(struct task_group *tg, void *data) |
1504 | { | |
1505 | return 0; | |
c09595f6 | 1506 | } |
eb755805 PZ |
1507 | #endif |
1508 | ||
1509 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1510 | /* Used instead of source_load when we know the type == 0 */ |
1511 | static unsigned long weighted_cpuload(const int cpu) | |
1512 | { | |
1513 | return cpu_rq(cpu)->load.weight; | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * Return a low guess at the load of a migration-source cpu weighted | |
1518 | * according to the scheduling class and "nice" value. | |
1519 | * | |
1520 | * We want to under-estimate the load of migration sources, to | |
1521 | * balance conservatively. | |
1522 | */ | |
1523 | static unsigned long source_load(int cpu, int type) | |
1524 | { | |
1525 | struct rq *rq = cpu_rq(cpu); | |
1526 | unsigned long total = weighted_cpuload(cpu); | |
1527 | ||
1528 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1529 | return total; | |
1530 | ||
1531 | return min(rq->cpu_load[type-1], total); | |
1532 | } | |
1533 | ||
1534 | /* | |
1535 | * Return a high guess at the load of a migration-target cpu weighted | |
1536 | * according to the scheduling class and "nice" value. | |
1537 | */ | |
1538 | static unsigned long target_load(int cpu, int type) | |
1539 | { | |
1540 | struct rq *rq = cpu_rq(cpu); | |
1541 | unsigned long total = weighted_cpuload(cpu); | |
1542 | ||
1543 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1544 | return total; | |
1545 | ||
1546 | return max(rq->cpu_load[type-1], total); | |
1547 | } | |
1548 | ||
eb755805 PZ |
1549 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1550 | ||
1551 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1552 | { | |
1553 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1554 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1555 | |
4cd42620 SR |
1556 | if (nr_running) |
1557 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1558 | else |
1559 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1560 | |
1561 | return rq->avg_load_per_task; | |
1562 | } | |
1563 | ||
1564 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1565 | |
34d76c41 PZ |
1566 | struct update_shares_data { |
1567 | unsigned long rq_weight[NR_CPUS]; | |
1568 | }; | |
1569 | ||
1570 | static DEFINE_PER_CPU(struct update_shares_data, update_shares_data); | |
1571 | ||
c09595f6 PZ |
1572 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1573 | ||
1574 | /* | |
1575 | * Calculate and set the cpu's group shares. | |
1576 | */ | |
34d76c41 PZ |
1577 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1578 | unsigned long sd_shares, | |
1579 | unsigned long sd_rq_weight, | |
1580 | struct update_shares_data *usd) | |
18d95a28 | 1581 | { |
34d76c41 | 1582 | unsigned long shares, rq_weight; |
a5004278 | 1583 | int boost = 0; |
c09595f6 | 1584 | |
34d76c41 | 1585 | rq_weight = usd->rq_weight[cpu]; |
a5004278 PZ |
1586 | if (!rq_weight) { |
1587 | boost = 1; | |
1588 | rq_weight = NICE_0_LOAD; | |
1589 | } | |
c8cba857 | 1590 | |
c09595f6 | 1591 | /* |
a8af7246 PZ |
1592 | * \Sum_j shares_j * rq_weight_i |
1593 | * shares_i = ----------------------------- | |
1594 | * \Sum_j rq_weight_j | |
c09595f6 | 1595 | */ |
ec4e0e2f | 1596 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1597 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1598 | |
ffda12a1 PZ |
1599 | if (abs(shares - tg->se[cpu]->load.weight) > |
1600 | sysctl_sched_shares_thresh) { | |
1601 | struct rq *rq = cpu_rq(cpu); | |
1602 | unsigned long flags; | |
c09595f6 | 1603 | |
ffda12a1 | 1604 | spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1605 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1606 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1607 | __set_se_shares(tg->se[cpu], shares); |
1608 | spin_unlock_irqrestore(&rq->lock, flags); | |
1609 | } | |
18d95a28 | 1610 | } |
c09595f6 PZ |
1611 | |
1612 | /* | |
c8cba857 PZ |
1613 | * Re-compute the task group their per cpu shares over the given domain. |
1614 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1615 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1616 | */ |
eb755805 | 1617 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1618 | { |
34d76c41 PZ |
1619 | unsigned long weight, rq_weight = 0, shares = 0; |
1620 | struct update_shares_data *usd; | |
eb755805 | 1621 | struct sched_domain *sd = data; |
34d76c41 | 1622 | unsigned long flags; |
c8cba857 | 1623 | int i; |
c09595f6 | 1624 | |
34d76c41 PZ |
1625 | if (!tg->se[0]) |
1626 | return 0; | |
1627 | ||
1628 | local_irq_save(flags); | |
1629 | usd = &__get_cpu_var(update_shares_data); | |
1630 | ||
758b2cdc | 1631 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 PZ |
1632 | weight = tg->cfs_rq[i]->load.weight; |
1633 | usd->rq_weight[i] = weight; | |
1634 | ||
ec4e0e2f KC |
1635 | /* |
1636 | * If there are currently no tasks on the cpu pretend there | |
1637 | * is one of average load so that when a new task gets to | |
1638 | * run here it will not get delayed by group starvation. | |
1639 | */ | |
ec4e0e2f KC |
1640 | if (!weight) |
1641 | weight = NICE_0_LOAD; | |
1642 | ||
ec4e0e2f | 1643 | rq_weight += weight; |
c8cba857 | 1644 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1645 | } |
c09595f6 | 1646 | |
c8cba857 PZ |
1647 | if ((!shares && rq_weight) || shares > tg->shares) |
1648 | shares = tg->shares; | |
1649 | ||
1650 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1651 | shares = tg->shares; | |
c09595f6 | 1652 | |
758b2cdc | 1653 | for_each_cpu(i, sched_domain_span(sd)) |
34d76c41 PZ |
1654 | update_group_shares_cpu(tg, i, shares, rq_weight, usd); |
1655 | ||
1656 | local_irq_restore(flags); | |
eb755805 PZ |
1657 | |
1658 | return 0; | |
c09595f6 PZ |
1659 | } |
1660 | ||
1661 | /* | |
c8cba857 PZ |
1662 | * Compute the cpu's hierarchical load factor for each task group. |
1663 | * This needs to be done in a top-down fashion because the load of a child | |
1664 | * group is a fraction of its parents load. | |
c09595f6 | 1665 | */ |
eb755805 | 1666 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1667 | { |
c8cba857 | 1668 | unsigned long load; |
eb755805 | 1669 | long cpu = (long)data; |
c09595f6 | 1670 | |
c8cba857 PZ |
1671 | if (!tg->parent) { |
1672 | load = cpu_rq(cpu)->load.weight; | |
1673 | } else { | |
1674 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1675 | load *= tg->cfs_rq[cpu]->shares; | |
1676 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1677 | } | |
c09595f6 | 1678 | |
c8cba857 | 1679 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1680 | |
eb755805 | 1681 | return 0; |
c09595f6 PZ |
1682 | } |
1683 | ||
c8cba857 | 1684 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1685 | { |
e7097159 PZ |
1686 | s64 elapsed; |
1687 | u64 now; | |
1688 | ||
1689 | if (root_task_group_empty()) | |
1690 | return; | |
1691 | ||
1692 | now = cpu_clock(raw_smp_processor_id()); | |
1693 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1694 | |
1695 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1696 | sd->last_update = now; | |
eb755805 | 1697 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1698 | } |
4d8d595d PZ |
1699 | } |
1700 | ||
3e5459b4 PZ |
1701 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1702 | { | |
e7097159 PZ |
1703 | if (root_task_group_empty()) |
1704 | return; | |
1705 | ||
3e5459b4 PZ |
1706 | spin_unlock(&rq->lock); |
1707 | update_shares(sd); | |
1708 | spin_lock(&rq->lock); | |
1709 | } | |
1710 | ||
eb755805 | 1711 | static void update_h_load(long cpu) |
c09595f6 | 1712 | { |
e7097159 PZ |
1713 | if (root_task_group_empty()) |
1714 | return; | |
1715 | ||
eb755805 | 1716 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1717 | } |
1718 | ||
c09595f6 PZ |
1719 | #else |
1720 | ||
c8cba857 | 1721 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1722 | { |
1723 | } | |
1724 | ||
3e5459b4 PZ |
1725 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1726 | { | |
1727 | } | |
1728 | ||
18d95a28 PZ |
1729 | #endif |
1730 | ||
8f45e2b5 GH |
1731 | #ifdef CONFIG_PREEMPT |
1732 | ||
b78bb868 PZ |
1733 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1734 | ||
70574a99 | 1735 | /* |
8f45e2b5 GH |
1736 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1737 | * way at the expense of forcing extra atomic operations in all | |
1738 | * invocations. This assures that the double_lock is acquired using the | |
1739 | * same underlying policy as the spinlock_t on this architecture, which | |
1740 | * reduces latency compared to the unfair variant below. However, it | |
1741 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1742 | */ |
8f45e2b5 GH |
1743 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1744 | __releases(this_rq->lock) | |
1745 | __acquires(busiest->lock) | |
1746 | __acquires(this_rq->lock) | |
1747 | { | |
1748 | spin_unlock(&this_rq->lock); | |
1749 | double_rq_lock(this_rq, busiest); | |
1750 | ||
1751 | return 1; | |
1752 | } | |
1753 | ||
1754 | #else | |
1755 | /* | |
1756 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1757 | * latency by eliminating extra atomic operations when the locks are | |
1758 | * already in proper order on entry. This favors lower cpu-ids and will | |
1759 | * grant the double lock to lower cpus over higher ids under contention, | |
1760 | * regardless of entry order into the function. | |
1761 | */ | |
1762 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1763 | __releases(this_rq->lock) |
1764 | __acquires(busiest->lock) | |
1765 | __acquires(this_rq->lock) | |
1766 | { | |
1767 | int ret = 0; | |
1768 | ||
70574a99 AD |
1769 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1770 | if (busiest < this_rq) { | |
1771 | spin_unlock(&this_rq->lock); | |
1772 | spin_lock(&busiest->lock); | |
1773 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1774 | ret = 1; | |
1775 | } else | |
1776 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1777 | } | |
1778 | return ret; | |
1779 | } | |
1780 | ||
8f45e2b5 GH |
1781 | #endif /* CONFIG_PREEMPT */ |
1782 | ||
1783 | /* | |
1784 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1785 | */ | |
1786 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1787 | { | |
1788 | if (unlikely(!irqs_disabled())) { | |
1789 | /* printk() doesn't work good under rq->lock */ | |
1790 | spin_unlock(&this_rq->lock); | |
1791 | BUG_ON(1); | |
1792 | } | |
1793 | ||
1794 | return _double_lock_balance(this_rq, busiest); | |
1795 | } | |
1796 | ||
70574a99 AD |
1797 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1798 | __releases(busiest->lock) | |
1799 | { | |
1800 | spin_unlock(&busiest->lock); | |
1801 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1802 | } | |
18d95a28 PZ |
1803 | #endif |
1804 | ||
30432094 | 1805 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1806 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1807 | { | |
30432094 | 1808 | #ifdef CONFIG_SMP |
34e83e85 IM |
1809 | cfs_rq->shares = shares; |
1810 | #endif | |
1811 | } | |
30432094 | 1812 | #endif |
e7693a36 | 1813 | |
dce48a84 TG |
1814 | static void calc_load_account_active(struct rq *this_rq); |
1815 | ||
dd41f596 | 1816 | #include "sched_stats.h" |
dd41f596 | 1817 | #include "sched_idletask.c" |
5522d5d5 IM |
1818 | #include "sched_fair.c" |
1819 | #include "sched_rt.c" | |
dd41f596 IM |
1820 | #ifdef CONFIG_SCHED_DEBUG |
1821 | # include "sched_debug.c" | |
1822 | #endif | |
1823 | ||
1824 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1825 | #define for_each_class(class) \ |
1826 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1827 | |
c09595f6 | 1828 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1829 | { |
1830 | rq->nr_running++; | |
9c217245 IM |
1831 | } |
1832 | ||
c09595f6 | 1833 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1834 | { |
1835 | rq->nr_running--; | |
9c217245 IM |
1836 | } |
1837 | ||
45bf76df IM |
1838 | static void set_load_weight(struct task_struct *p) |
1839 | { | |
1840 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1841 | p->se.load.weight = prio_to_weight[0] * 2; |
1842 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1843 | return; | |
1844 | } | |
45bf76df | 1845 | |
dd41f596 IM |
1846 | /* |
1847 | * SCHED_IDLE tasks get minimal weight: | |
1848 | */ | |
1849 | if (p->policy == SCHED_IDLE) { | |
1850 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1851 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1852 | return; | |
1853 | } | |
71f8bd46 | 1854 | |
dd41f596 IM |
1855 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1856 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1857 | } |
1858 | ||
2087a1ad GH |
1859 | static void update_avg(u64 *avg, u64 sample) |
1860 | { | |
1861 | s64 diff = sample - *avg; | |
1862 | *avg += diff >> 3; | |
1863 | } | |
1864 | ||
8159f87e | 1865 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1866 | { |
831451ac PZ |
1867 | if (wakeup) |
1868 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1869 | ||
dd41f596 | 1870 | sched_info_queued(p); |
fd390f6a | 1871 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1872 | p->se.on_rq = 1; |
71f8bd46 IM |
1873 | } |
1874 | ||
69be72c1 | 1875 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1876 | { |
831451ac PZ |
1877 | if (sleep) { |
1878 | if (p->se.last_wakeup) { | |
1879 | update_avg(&p->se.avg_overlap, | |
1880 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1881 | p->se.last_wakeup = 0; | |
1882 | } else { | |
1883 | update_avg(&p->se.avg_wakeup, | |
1884 | sysctl_sched_wakeup_granularity); | |
1885 | } | |
2087a1ad GH |
1886 | } |
1887 | ||
46ac22ba | 1888 | sched_info_dequeued(p); |
f02231e5 | 1889 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1890 | p->se.on_rq = 0; |
71f8bd46 IM |
1891 | } |
1892 | ||
14531189 | 1893 | /* |
dd41f596 | 1894 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1895 | */ |
14531189 IM |
1896 | static inline int __normal_prio(struct task_struct *p) |
1897 | { | |
dd41f596 | 1898 | return p->static_prio; |
14531189 IM |
1899 | } |
1900 | ||
b29739f9 IM |
1901 | /* |
1902 | * Calculate the expected normal priority: i.e. priority | |
1903 | * without taking RT-inheritance into account. Might be | |
1904 | * boosted by interactivity modifiers. Changes upon fork, | |
1905 | * setprio syscalls, and whenever the interactivity | |
1906 | * estimator recalculates. | |
1907 | */ | |
36c8b586 | 1908 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1909 | { |
1910 | int prio; | |
1911 | ||
e05606d3 | 1912 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1913 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1914 | else | |
1915 | prio = __normal_prio(p); | |
1916 | return prio; | |
1917 | } | |
1918 | ||
1919 | /* | |
1920 | * Calculate the current priority, i.e. the priority | |
1921 | * taken into account by the scheduler. This value might | |
1922 | * be boosted by RT tasks, or might be boosted by | |
1923 | * interactivity modifiers. Will be RT if the task got | |
1924 | * RT-boosted. If not then it returns p->normal_prio. | |
1925 | */ | |
36c8b586 | 1926 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1927 | { |
1928 | p->normal_prio = normal_prio(p); | |
1929 | /* | |
1930 | * If we are RT tasks or we were boosted to RT priority, | |
1931 | * keep the priority unchanged. Otherwise, update priority | |
1932 | * to the normal priority: | |
1933 | */ | |
1934 | if (!rt_prio(p->prio)) | |
1935 | return p->normal_prio; | |
1936 | return p->prio; | |
1937 | } | |
1938 | ||
1da177e4 | 1939 | /* |
dd41f596 | 1940 | * activate_task - move a task to the runqueue. |
1da177e4 | 1941 | */ |
dd41f596 | 1942 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1943 | { |
d9514f6c | 1944 | if (task_contributes_to_load(p)) |
dd41f596 | 1945 | rq->nr_uninterruptible--; |
1da177e4 | 1946 | |
8159f87e | 1947 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1948 | inc_nr_running(rq); |
1da177e4 LT |
1949 | } |
1950 | ||
1da177e4 LT |
1951 | /* |
1952 | * deactivate_task - remove a task from the runqueue. | |
1953 | */ | |
2e1cb74a | 1954 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1955 | { |
d9514f6c | 1956 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1957 | rq->nr_uninterruptible++; |
1958 | ||
69be72c1 | 1959 | dequeue_task(rq, p, sleep); |
c09595f6 | 1960 | dec_nr_running(rq); |
1da177e4 LT |
1961 | } |
1962 | ||
1da177e4 LT |
1963 | /** |
1964 | * task_curr - is this task currently executing on a CPU? | |
1965 | * @p: the task in question. | |
1966 | */ | |
36c8b586 | 1967 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1968 | { |
1969 | return cpu_curr(task_cpu(p)) == p; | |
1970 | } | |
1971 | ||
dd41f596 IM |
1972 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1973 | { | |
6f505b16 | 1974 | set_task_rq(p, cpu); |
dd41f596 | 1975 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1976 | /* |
1977 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1978 | * successfuly executed on another CPU. We must ensure that updates of | |
1979 | * per-task data have been completed by this moment. | |
1980 | */ | |
1981 | smp_wmb(); | |
dd41f596 | 1982 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1983 | #endif |
2dd73a4f PW |
1984 | } |
1985 | ||
cb469845 SR |
1986 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1987 | const struct sched_class *prev_class, | |
1988 | int oldprio, int running) | |
1989 | { | |
1990 | if (prev_class != p->sched_class) { | |
1991 | if (prev_class->switched_from) | |
1992 | prev_class->switched_from(rq, p, running); | |
1993 | p->sched_class->switched_to(rq, p, running); | |
1994 | } else | |
1995 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1996 | } | |
1997 | ||
1da177e4 | 1998 | #ifdef CONFIG_SMP |
cc367732 IM |
1999 | /* |
2000 | * Is this task likely cache-hot: | |
2001 | */ | |
e7693a36 | 2002 | static int |
cc367732 IM |
2003 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2004 | { | |
2005 | s64 delta; | |
2006 | ||
f540a608 IM |
2007 | /* |
2008 | * Buddy candidates are cache hot: | |
2009 | */ | |
4793241b PZ |
2010 | if (sched_feat(CACHE_HOT_BUDDY) && |
2011 | (&p->se == cfs_rq_of(&p->se)->next || | |
2012 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2013 | return 1; |
2014 | ||
cc367732 IM |
2015 | if (p->sched_class != &fair_sched_class) |
2016 | return 0; | |
2017 | ||
6bc1665b IM |
2018 | if (sysctl_sched_migration_cost == -1) |
2019 | return 1; | |
2020 | if (sysctl_sched_migration_cost == 0) | |
2021 | return 0; | |
2022 | ||
cc367732 IM |
2023 | delta = now - p->se.exec_start; |
2024 | ||
2025 | return delta < (s64)sysctl_sched_migration_cost; | |
2026 | } | |
2027 | ||
2028 | ||
dd41f596 | 2029 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2030 | { |
dd41f596 IM |
2031 | int old_cpu = task_cpu(p); |
2032 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
2033 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
2034 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 2035 | u64 clock_offset; |
dd41f596 IM |
2036 | |
2037 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 2038 | |
de1d7286 | 2039 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2040 | |
6cfb0d5d IM |
2041 | #ifdef CONFIG_SCHEDSTATS |
2042 | if (p->se.wait_start) | |
2043 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
2044 | if (p->se.sleep_start) |
2045 | p->se.sleep_start -= clock_offset; | |
2046 | if (p->se.block_start) | |
2047 | p->se.block_start -= clock_offset; | |
6c594c21 | 2048 | #endif |
cc367732 | 2049 | if (old_cpu != new_cpu) { |
6c594c21 | 2050 | p->se.nr_migrations++; |
23a185ca | 2051 | new_rq->nr_migrations_in++; |
6c594c21 | 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 |
e5289d4a PZ |
2056 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
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 | */ | |
36c8b586 | 2309 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2310 | { |
cc367732 | 2311 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2312 | unsigned long flags; |
70b97a7f | 2313 | struct rq *rq; |
1da177e4 | 2314 | |
b85d0667 IM |
2315 | if (!sched_feat(SYNC_WAKEUPS)) |
2316 | sync = 0; | |
2317 | ||
2398f2c6 | 2318 | #ifdef CONFIG_SMP |
57310a98 | 2319 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2320 | struct sched_domain *sd; |
2321 | ||
2322 | this_cpu = raw_smp_processor_id(); | |
2323 | cpu = task_cpu(p); | |
2324 | ||
2325 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2326 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2327 | update_shares(sd); |
2328 | break; | |
2329 | } | |
2330 | } | |
2331 | } | |
2332 | #endif | |
2333 | ||
e9c84311 PZ |
2334 | this_cpu = get_cpu(); |
2335 | ||
04e2f174 | 2336 | smp_wmb(); |
1da177e4 | 2337 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2338 | update_rq_clock(rq); |
e9c84311 | 2339 | if (!(p->state & state)) |
1da177e4 LT |
2340 | goto out; |
2341 | ||
dd41f596 | 2342 | if (p->se.on_rq) |
1da177e4 LT |
2343 | goto out_running; |
2344 | ||
2345 | cpu = task_cpu(p); | |
cc367732 | 2346 | orig_cpu = cpu; |
1da177e4 LT |
2347 | |
2348 | #ifdef CONFIG_SMP | |
2349 | if (unlikely(task_running(rq, p))) | |
2350 | goto out_activate; | |
2351 | ||
e9c84311 PZ |
2352 | /* |
2353 | * In order to handle concurrent wakeups and release the rq->lock | |
2354 | * we put the task in TASK_WAKING state. | |
2355 | */ | |
2356 | p->state = TASK_WAKING; | |
2357 | task_rq_unlock(rq, &flags); | |
2358 | ||
5f3edc1b | 2359 | cpu = p->sched_class->select_task_rq(p, SD_BALANCE_WAKE, sync); |
e9c84311 | 2360 | if (cpu != orig_cpu) |
5d2f5a61 | 2361 | set_task_cpu(p, cpu); |
1da177e4 | 2362 | |
e9c84311 PZ |
2363 | rq = task_rq_lock(p, &flags); |
2364 | WARN_ON(p->state != TASK_WAKING); | |
2365 | cpu = task_cpu(p); | |
1da177e4 | 2366 | |
e7693a36 GH |
2367 | #ifdef CONFIG_SCHEDSTATS |
2368 | schedstat_inc(rq, ttwu_count); | |
2369 | if (cpu == this_cpu) | |
2370 | schedstat_inc(rq, ttwu_local); | |
2371 | else { | |
2372 | struct sched_domain *sd; | |
2373 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2374 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2375 | schedstat_inc(sd, ttwu_wake_remote); |
2376 | break; | |
2377 | } | |
2378 | } | |
2379 | } | |
6d6bc0ad | 2380 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2381 | |
1da177e4 LT |
2382 | out_activate: |
2383 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2384 | schedstat_inc(p, se.nr_wakeups); |
2385 | if (sync) | |
2386 | schedstat_inc(p, se.nr_wakeups_sync); | |
2387 | if (orig_cpu != cpu) | |
2388 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2389 | if (cpu == this_cpu) | |
2390 | schedstat_inc(p, se.nr_wakeups_local); | |
2391 | else | |
2392 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2393 | activate_task(rq, p, 1); |
1da177e4 LT |
2394 | success = 1; |
2395 | ||
831451ac PZ |
2396 | /* |
2397 | * Only attribute actual wakeups done by this task. | |
2398 | */ | |
2399 | if (!in_interrupt()) { | |
2400 | struct sched_entity *se = ¤t->se; | |
2401 | u64 sample = se->sum_exec_runtime; | |
2402 | ||
2403 | if (se->last_wakeup) | |
2404 | sample -= se->last_wakeup; | |
2405 | else | |
2406 | sample -= se->start_runtime; | |
2407 | update_avg(&se->avg_wakeup, sample); | |
2408 | ||
2409 | se->last_wakeup = se->sum_exec_runtime; | |
2410 | } | |
2411 | ||
1da177e4 | 2412 | out_running: |
468a15bb | 2413 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2414 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2415 | |
1da177e4 | 2416 | p->state = TASK_RUNNING; |
9a897c5a SR |
2417 | #ifdef CONFIG_SMP |
2418 | if (p->sched_class->task_wake_up) | |
2419 | p->sched_class->task_wake_up(rq, p); | |
2420 | #endif | |
1da177e4 LT |
2421 | out: |
2422 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2423 | put_cpu(); |
1da177e4 LT |
2424 | |
2425 | return success; | |
2426 | } | |
2427 | ||
50fa610a DH |
2428 | /** |
2429 | * wake_up_process - Wake up a specific process | |
2430 | * @p: The process to be woken up. | |
2431 | * | |
2432 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2433 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2434 | * running. | |
2435 | * | |
2436 | * It may be assumed that this function implies a write memory barrier before | |
2437 | * changing the task state if and only if any tasks are woken up. | |
2438 | */ | |
7ad5b3a5 | 2439 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2440 | { |
d9514f6c | 2441 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2442 | } |
1da177e4 LT |
2443 | EXPORT_SYMBOL(wake_up_process); |
2444 | ||
7ad5b3a5 | 2445 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2446 | { |
2447 | return try_to_wake_up(p, state, 0); | |
2448 | } | |
2449 | ||
1da177e4 LT |
2450 | /* |
2451 | * Perform scheduler related setup for a newly forked process p. | |
2452 | * p is forked by current. | |
dd41f596 IM |
2453 | * |
2454 | * __sched_fork() is basic setup used by init_idle() too: | |
2455 | */ | |
2456 | static void __sched_fork(struct task_struct *p) | |
2457 | { | |
dd41f596 IM |
2458 | p->se.exec_start = 0; |
2459 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2460 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2461 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2462 | p->se.last_wakeup = 0; |
2463 | p->se.avg_overlap = 0; | |
831451ac PZ |
2464 | p->se.start_runtime = 0; |
2465 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2466 | |
2467 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2468 | p->se.wait_start = 0; |
2469 | p->se.wait_max = 0; | |
2470 | p->se.wait_count = 0; | |
2471 | p->se.wait_sum = 0; | |
2472 | ||
2473 | p->se.sleep_start = 0; | |
2474 | p->se.sleep_max = 0; | |
2475 | p->se.sum_sleep_runtime = 0; | |
2476 | ||
2477 | p->se.block_start = 0; | |
2478 | p->se.block_max = 0; | |
2479 | p->se.exec_max = 0; | |
2480 | p->se.slice_max = 0; | |
2481 | ||
2482 | p->se.nr_migrations_cold = 0; | |
2483 | p->se.nr_failed_migrations_affine = 0; | |
2484 | p->se.nr_failed_migrations_running = 0; | |
2485 | p->se.nr_failed_migrations_hot = 0; | |
2486 | p->se.nr_forced_migrations = 0; | |
2487 | p->se.nr_forced2_migrations = 0; | |
2488 | ||
2489 | p->se.nr_wakeups = 0; | |
2490 | p->se.nr_wakeups_sync = 0; | |
2491 | p->se.nr_wakeups_migrate = 0; | |
2492 | p->se.nr_wakeups_local = 0; | |
2493 | p->se.nr_wakeups_remote = 0; | |
2494 | p->se.nr_wakeups_affine = 0; | |
2495 | p->se.nr_wakeups_affine_attempts = 0; | |
2496 | p->se.nr_wakeups_passive = 0; | |
2497 | p->se.nr_wakeups_idle = 0; | |
2498 | ||
6cfb0d5d | 2499 | #endif |
476d139c | 2500 | |
fa717060 | 2501 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2502 | p->se.on_rq = 0; |
4a55bd5e | 2503 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2504 | |
e107be36 AK |
2505 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2506 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2507 | #endif | |
2508 | ||
1da177e4 LT |
2509 | /* |
2510 | * We mark the process as running here, but have not actually | |
2511 | * inserted it onto the runqueue yet. This guarantees that | |
2512 | * nobody will actually run it, and a signal or other external | |
2513 | * event cannot wake it up and insert it on the runqueue either. | |
2514 | */ | |
2515 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * fork()/clone()-time setup: | |
2520 | */ | |
2521 | void sched_fork(struct task_struct *p, int clone_flags) | |
2522 | { | |
2523 | int cpu = get_cpu(); | |
2524 | ||
2525 | __sched_fork(p); | |
2526 | ||
b29739f9 | 2527 | /* |
b9dc29e7 | 2528 | * Make sure we do not leak PI boosting priority to the child. |
b29739f9 IM |
2529 | */ |
2530 | p->prio = current->normal_prio; | |
ca94c442 | 2531 | |
b9dc29e7 MG |
2532 | /* |
2533 | * Revert to default priority/policy on fork if requested. | |
2534 | */ | |
2535 | if (unlikely(p->sched_reset_on_fork)) { | |
2536 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) | |
2537 | p->policy = SCHED_NORMAL; | |
2538 | ||
2539 | if (p->normal_prio < DEFAULT_PRIO) | |
2540 | p->prio = DEFAULT_PRIO; | |
2541 | ||
6c697bdf MG |
2542 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2543 | p->static_prio = NICE_TO_PRIO(0); | |
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 | |
2ddbf952 HS |
2554 | if (!rt_prio(p->prio)) |
2555 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2556 | |
5f3edc1b PZ |
2557 | #ifdef CONFIG_SMP |
2558 | cpu = p->sched_class->select_task_rq(p, SD_BALANCE_FORK, 0); | |
2559 | #endif | |
2560 | set_task_cpu(p, cpu); | |
2561 | ||
52f17b6c | 2562 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2563 | if (likely(sched_info_on())) |
52f17b6c | 2564 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2565 | #endif |
d6077cb8 | 2566 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2567 | p->oncpu = 0; |
2568 | #endif | |
1da177e4 | 2569 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2570 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2571 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2572 | #endif |
917b627d GH |
2573 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2574 | ||
476d139c | 2575 | put_cpu(); |
1da177e4 LT |
2576 | } |
2577 | ||
2578 | /* | |
2579 | * wake_up_new_task - wake up a newly created task for the first time. | |
2580 | * | |
2581 | * This function will do some initial scheduler statistics housekeeping | |
2582 | * that must be done for every newly created context, then puts the task | |
2583 | * on the runqueue and wakes it. | |
2584 | */ | |
7ad5b3a5 | 2585 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2586 | { |
2587 | unsigned long flags; | |
dd41f596 | 2588 | struct rq *rq; |
1da177e4 LT |
2589 | |
2590 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2591 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2592 | update_rq_clock(rq); |
1da177e4 LT |
2593 | |
2594 | p->prio = effective_prio(p); | |
2595 | ||
b9dca1e0 | 2596 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2597 | activate_task(rq, p, 0); |
1da177e4 | 2598 | } else { |
1da177e4 | 2599 | /* |
dd41f596 IM |
2600 | * Let the scheduling class do new task startup |
2601 | * management (if any): | |
1da177e4 | 2602 | */ |
ee0827d8 | 2603 | p->sched_class->task_new(rq, p); |
c09595f6 | 2604 | inc_nr_running(rq); |
1da177e4 | 2605 | } |
c71dd42d | 2606 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2607 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2608 | #ifdef CONFIG_SMP |
2609 | if (p->sched_class->task_wake_up) | |
2610 | p->sched_class->task_wake_up(rq, p); | |
2611 | #endif | |
dd41f596 | 2612 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2613 | } |
2614 | ||
e107be36 AK |
2615 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2616 | ||
2617 | /** | |
80dd99b3 | 2618 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2619 | * @notifier: notifier struct to register |
e107be36 AK |
2620 | */ |
2621 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2622 | { | |
2623 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2624 | } | |
2625 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2626 | ||
2627 | /** | |
2628 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2629 | * @notifier: notifier struct to unregister |
e107be36 AK |
2630 | * |
2631 | * This is safe to call from within a preemption notifier. | |
2632 | */ | |
2633 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2634 | { | |
2635 | hlist_del(¬ifier->link); | |
2636 | } | |
2637 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2638 | ||
2639 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2640 | { | |
2641 | struct preempt_notifier *notifier; | |
2642 | struct hlist_node *node; | |
2643 | ||
2644 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2645 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2646 | } | |
2647 | ||
2648 | static void | |
2649 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2650 | struct task_struct *next) | |
2651 | { | |
2652 | struct preempt_notifier *notifier; | |
2653 | struct hlist_node *node; | |
2654 | ||
2655 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2656 | notifier->ops->sched_out(notifier, next); | |
2657 | } | |
2658 | ||
6d6bc0ad | 2659 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2660 | |
2661 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2662 | { | |
2663 | } | |
2664 | ||
2665 | static void | |
2666 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2667 | struct task_struct *next) | |
2668 | { | |
2669 | } | |
2670 | ||
6d6bc0ad | 2671 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2672 | |
4866cde0 NP |
2673 | /** |
2674 | * prepare_task_switch - prepare to switch tasks | |
2675 | * @rq: the runqueue preparing to switch | |
421cee29 | 2676 | * @prev: the current task that is being switched out |
4866cde0 NP |
2677 | * @next: the task we are going to switch to. |
2678 | * | |
2679 | * This is called with the rq lock held and interrupts off. It must | |
2680 | * be paired with a subsequent finish_task_switch after the context | |
2681 | * switch. | |
2682 | * | |
2683 | * prepare_task_switch sets up locking and calls architecture specific | |
2684 | * hooks. | |
2685 | */ | |
e107be36 AK |
2686 | static inline void |
2687 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2688 | struct task_struct *next) | |
4866cde0 | 2689 | { |
e107be36 | 2690 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2691 | prepare_lock_switch(rq, next); |
2692 | prepare_arch_switch(next); | |
2693 | } | |
2694 | ||
1da177e4 LT |
2695 | /** |
2696 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2697 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2698 | * @prev: the thread we just switched away from. |
2699 | * | |
4866cde0 NP |
2700 | * finish_task_switch must be called after the context switch, paired |
2701 | * with a prepare_task_switch call before the context switch. | |
2702 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2703 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2704 | * |
2705 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2706 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2707 | * with the lock held can cause deadlocks; see schedule() for |
2708 | * details.) | |
2709 | */ | |
a9957449 | 2710 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2711 | __releases(rq->lock) |
2712 | { | |
1da177e4 | 2713 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2714 | long prev_state; |
1da177e4 LT |
2715 | |
2716 | rq->prev_mm = NULL; | |
2717 | ||
2718 | /* | |
2719 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2720 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2721 | * schedule one last time. The schedule call will never return, and |
2722 | * the scheduled task must drop that reference. | |
c394cc9f | 2723 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2724 | * still held, otherwise prev could be scheduled on another cpu, die |
2725 | * there before we look at prev->state, and then the reference would | |
2726 | * be dropped twice. | |
2727 | * Manfred Spraul <manfred@colorfullife.com> | |
2728 | */ | |
55a101f8 | 2729 | prev_state = prev->state; |
4866cde0 | 2730 | finish_arch_switch(prev); |
0793a61d | 2731 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2732 | finish_lock_switch(rq, prev); |
e8fa1362 | 2733 | |
e107be36 | 2734 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2735 | if (mm) |
2736 | mmdrop(mm); | |
c394cc9f | 2737 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2738 | /* |
2739 | * Remove function-return probe instances associated with this | |
2740 | * task and put them back on the free list. | |
9761eea8 | 2741 | */ |
c6fd91f0 | 2742 | kprobe_flush_task(prev); |
1da177e4 | 2743 | put_task_struct(prev); |
c6fd91f0 | 2744 | } |
1da177e4 LT |
2745 | } |
2746 | ||
3f029d3c GH |
2747 | #ifdef CONFIG_SMP |
2748 | ||
2749 | /* assumes rq->lock is held */ | |
2750 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2751 | { | |
2752 | if (prev->sched_class->pre_schedule) | |
2753 | prev->sched_class->pre_schedule(rq, prev); | |
2754 | } | |
2755 | ||
2756 | /* rq->lock is NOT held, but preemption is disabled */ | |
2757 | static inline void post_schedule(struct rq *rq) | |
2758 | { | |
2759 | if (rq->post_schedule) { | |
2760 | unsigned long flags; | |
2761 | ||
2762 | spin_lock_irqsave(&rq->lock, flags); | |
2763 | if (rq->curr->sched_class->post_schedule) | |
2764 | rq->curr->sched_class->post_schedule(rq); | |
2765 | spin_unlock_irqrestore(&rq->lock, flags); | |
2766 | ||
2767 | rq->post_schedule = 0; | |
2768 | } | |
2769 | } | |
2770 | ||
2771 | #else | |
da19ab51 | 2772 | |
3f029d3c GH |
2773 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2774 | { | |
2775 | } | |
2776 | ||
2777 | static inline void post_schedule(struct rq *rq) | |
2778 | { | |
1da177e4 LT |
2779 | } |
2780 | ||
3f029d3c GH |
2781 | #endif |
2782 | ||
1da177e4 LT |
2783 | /** |
2784 | * schedule_tail - first thing a freshly forked thread must call. | |
2785 | * @prev: the thread we just switched away from. | |
2786 | */ | |
36c8b586 | 2787 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2788 | __releases(rq->lock) |
2789 | { | |
70b97a7f IM |
2790 | struct rq *rq = this_rq(); |
2791 | ||
4866cde0 | 2792 | finish_task_switch(rq, prev); |
da19ab51 | 2793 | |
3f029d3c GH |
2794 | /* |
2795 | * FIXME: do we need to worry about rq being invalidated by the | |
2796 | * task_switch? | |
2797 | */ | |
2798 | post_schedule(rq); | |
70b97a7f | 2799 | |
4866cde0 NP |
2800 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2801 | /* In this case, finish_task_switch does not reenable preemption */ | |
2802 | preempt_enable(); | |
2803 | #endif | |
1da177e4 | 2804 | if (current->set_child_tid) |
b488893a | 2805 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2806 | } |
2807 | ||
2808 | /* | |
2809 | * context_switch - switch to the new MM and the new | |
2810 | * thread's register state. | |
2811 | */ | |
dd41f596 | 2812 | static inline void |
70b97a7f | 2813 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2814 | struct task_struct *next) |
1da177e4 | 2815 | { |
dd41f596 | 2816 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2817 | |
e107be36 | 2818 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2819 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2820 | mm = next->mm; |
2821 | oldmm = prev->active_mm; | |
9226d125 ZA |
2822 | /* |
2823 | * For paravirt, this is coupled with an exit in switch_to to | |
2824 | * combine the page table reload and the switch backend into | |
2825 | * one hypercall. | |
2826 | */ | |
224101ed | 2827 | arch_start_context_switch(prev); |
9226d125 | 2828 | |
dd41f596 | 2829 | if (unlikely(!mm)) { |
1da177e4 LT |
2830 | next->active_mm = oldmm; |
2831 | atomic_inc(&oldmm->mm_count); | |
2832 | enter_lazy_tlb(oldmm, next); | |
2833 | } else | |
2834 | switch_mm(oldmm, mm, next); | |
2835 | ||
dd41f596 | 2836 | if (unlikely(!prev->mm)) { |
1da177e4 | 2837 | prev->active_mm = NULL; |
1da177e4 LT |
2838 | rq->prev_mm = oldmm; |
2839 | } | |
3a5f5e48 IM |
2840 | /* |
2841 | * Since the runqueue lock will be released by the next | |
2842 | * task (which is an invalid locking op but in the case | |
2843 | * of the scheduler it's an obvious special-case), so we | |
2844 | * do an early lockdep release here: | |
2845 | */ | |
2846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2847 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2848 | #endif |
1da177e4 LT |
2849 | |
2850 | /* Here we just switch the register state and the stack. */ | |
2851 | switch_to(prev, next, prev); | |
2852 | ||
dd41f596 IM |
2853 | barrier(); |
2854 | /* | |
2855 | * this_rq must be evaluated again because prev may have moved | |
2856 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2857 | * frame will be invalid. | |
2858 | */ | |
2859 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2860 | } |
2861 | ||
2862 | /* | |
2863 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2864 | * | |
2865 | * externally visible scheduler statistics: current number of runnable | |
2866 | * threads, current number of uninterruptible-sleeping threads, total | |
2867 | * number of context switches performed since bootup. | |
2868 | */ | |
2869 | unsigned long nr_running(void) | |
2870 | { | |
2871 | unsigned long i, sum = 0; | |
2872 | ||
2873 | for_each_online_cpu(i) | |
2874 | sum += cpu_rq(i)->nr_running; | |
2875 | ||
2876 | return sum; | |
2877 | } | |
2878 | ||
2879 | unsigned long nr_uninterruptible(void) | |
2880 | { | |
2881 | unsigned long i, sum = 0; | |
2882 | ||
0a945022 | 2883 | for_each_possible_cpu(i) |
1da177e4 LT |
2884 | sum += cpu_rq(i)->nr_uninterruptible; |
2885 | ||
2886 | /* | |
2887 | * Since we read the counters lockless, it might be slightly | |
2888 | * inaccurate. Do not allow it to go below zero though: | |
2889 | */ | |
2890 | if (unlikely((long)sum < 0)) | |
2891 | sum = 0; | |
2892 | ||
2893 | return sum; | |
2894 | } | |
2895 | ||
2896 | unsigned long long nr_context_switches(void) | |
2897 | { | |
cc94abfc SR |
2898 | int i; |
2899 | unsigned long long sum = 0; | |
1da177e4 | 2900 | |
0a945022 | 2901 | for_each_possible_cpu(i) |
1da177e4 LT |
2902 | sum += cpu_rq(i)->nr_switches; |
2903 | ||
2904 | return sum; | |
2905 | } | |
2906 | ||
2907 | unsigned long nr_iowait(void) | |
2908 | { | |
2909 | unsigned long i, sum = 0; | |
2910 | ||
0a945022 | 2911 | for_each_possible_cpu(i) |
1da177e4 LT |
2912 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2913 | ||
2914 | return sum; | |
2915 | } | |
2916 | ||
dce48a84 TG |
2917 | /* Variables and functions for calc_load */ |
2918 | static atomic_long_t calc_load_tasks; | |
2919 | static unsigned long calc_load_update; | |
2920 | unsigned long avenrun[3]; | |
2921 | EXPORT_SYMBOL(avenrun); | |
2922 | ||
2d02494f TG |
2923 | /** |
2924 | * get_avenrun - get the load average array | |
2925 | * @loads: pointer to dest load array | |
2926 | * @offset: offset to add | |
2927 | * @shift: shift count to shift the result left | |
2928 | * | |
2929 | * These values are estimates at best, so no need for locking. | |
2930 | */ | |
2931 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2932 | { | |
2933 | loads[0] = (avenrun[0] + offset) << shift; | |
2934 | loads[1] = (avenrun[1] + offset) << shift; | |
2935 | loads[2] = (avenrun[2] + offset) << shift; | |
2936 | } | |
2937 | ||
dce48a84 TG |
2938 | static unsigned long |
2939 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2940 | { |
dce48a84 TG |
2941 | load *= exp; |
2942 | load += active * (FIXED_1 - exp); | |
2943 | return load >> FSHIFT; | |
2944 | } | |
db1b1fef | 2945 | |
dce48a84 TG |
2946 | /* |
2947 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2948 | * CPUs have updated calc_load_tasks. | |
2949 | */ | |
2950 | void calc_global_load(void) | |
2951 | { | |
2952 | unsigned long upd = calc_load_update + 10; | |
2953 | long active; | |
2954 | ||
2955 | if (time_before(jiffies, upd)) | |
2956 | return; | |
db1b1fef | 2957 | |
dce48a84 TG |
2958 | active = atomic_long_read(&calc_load_tasks); |
2959 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2960 | |
dce48a84 TG |
2961 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2962 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2963 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2964 | ||
2965 | calc_load_update += LOAD_FREQ; | |
2966 | } | |
2967 | ||
2968 | /* | |
2969 | * Either called from update_cpu_load() or from a cpu going idle | |
2970 | */ | |
2971 | static void calc_load_account_active(struct rq *this_rq) | |
2972 | { | |
2973 | long nr_active, delta; | |
2974 | ||
2975 | nr_active = this_rq->nr_running; | |
2976 | nr_active += (long) this_rq->nr_uninterruptible; | |
2977 | ||
2978 | if (nr_active != this_rq->calc_load_active) { | |
2979 | delta = nr_active - this_rq->calc_load_active; | |
2980 | this_rq->calc_load_active = nr_active; | |
2981 | atomic_long_add(delta, &calc_load_tasks); | |
2982 | } | |
db1b1fef JS |
2983 | } |
2984 | ||
23a185ca PM |
2985 | /* |
2986 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
2987 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
2988 | */ | |
23a185ca PM |
2989 | u64 cpu_nr_migrations(int cpu) |
2990 | { | |
2991 | return cpu_rq(cpu)->nr_migrations_in; | |
2992 | } | |
2993 | ||
48f24c4d | 2994 | /* |
dd41f596 IM |
2995 | * Update rq->cpu_load[] statistics. This function is usually called every |
2996 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2997 | */ |
dd41f596 | 2998 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2999 | { |
495eca49 | 3000 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3001 | int i, scale; |
3002 | ||
3003 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3004 | |
3005 | /* Update our load: */ | |
3006 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3007 | unsigned long old_load, new_load; | |
3008 | ||
3009 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3010 | ||
3011 | old_load = this_rq->cpu_load[i]; | |
3012 | new_load = this_load; | |
a25707f3 IM |
3013 | /* |
3014 | * Round up the averaging division if load is increasing. This | |
3015 | * prevents us from getting stuck on 9 if the load is 10, for | |
3016 | * example. | |
3017 | */ | |
3018 | if (new_load > old_load) | |
3019 | new_load += scale-1; | |
dd41f596 IM |
3020 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3021 | } | |
dce48a84 TG |
3022 | |
3023 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3024 | this_rq->calc_load_update += LOAD_FREQ; | |
3025 | calc_load_account_active(this_rq); | |
3026 | } | |
48f24c4d IM |
3027 | } |
3028 | ||
dd41f596 IM |
3029 | #ifdef CONFIG_SMP |
3030 | ||
1da177e4 LT |
3031 | /* |
3032 | * double_rq_lock - safely lock two runqueues | |
3033 | * | |
3034 | * Note this does not disable interrupts like task_rq_lock, | |
3035 | * you need to do so manually before calling. | |
3036 | */ | |
70b97a7f | 3037 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3038 | __acquires(rq1->lock) |
3039 | __acquires(rq2->lock) | |
3040 | { | |
054b9108 | 3041 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3042 | if (rq1 == rq2) { |
3043 | spin_lock(&rq1->lock); | |
3044 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3045 | } else { | |
c96d145e | 3046 | if (rq1 < rq2) { |
1da177e4 | 3047 | spin_lock(&rq1->lock); |
5e710e37 | 3048 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3049 | } else { |
3050 | spin_lock(&rq2->lock); | |
5e710e37 | 3051 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3052 | } |
3053 | } | |
6e82a3be IM |
3054 | update_rq_clock(rq1); |
3055 | update_rq_clock(rq2); | |
1da177e4 LT |
3056 | } |
3057 | ||
3058 | /* | |
3059 | * double_rq_unlock - safely unlock two runqueues | |
3060 | * | |
3061 | * Note this does not restore interrupts like task_rq_unlock, | |
3062 | * you need to do so manually after calling. | |
3063 | */ | |
70b97a7f | 3064 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3065 | __releases(rq1->lock) |
3066 | __releases(rq2->lock) | |
3067 | { | |
3068 | spin_unlock(&rq1->lock); | |
3069 | if (rq1 != rq2) | |
3070 | spin_unlock(&rq2->lock); | |
3071 | else | |
3072 | __release(rq2->lock); | |
3073 | } | |
3074 | ||
1da177e4 LT |
3075 | /* |
3076 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3077 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3078 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3079 | * the cpu_allowed mask is restored. |
3080 | */ | |
36c8b586 | 3081 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3082 | { |
70b97a7f | 3083 | struct migration_req req; |
1da177e4 | 3084 | unsigned long flags; |
70b97a7f | 3085 | struct rq *rq; |
1da177e4 LT |
3086 | |
3087 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3088 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3089 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3090 | goto out; |
3091 | ||
3092 | /* force the process onto the specified CPU */ | |
3093 | if (migrate_task(p, dest_cpu, &req)) { | |
3094 | /* Need to wait for migration thread (might exit: take ref). */ | |
3095 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3096 | |
1da177e4 LT |
3097 | get_task_struct(mt); |
3098 | task_rq_unlock(rq, &flags); | |
3099 | wake_up_process(mt); | |
3100 | put_task_struct(mt); | |
3101 | wait_for_completion(&req.done); | |
36c8b586 | 3102 | |
1da177e4 LT |
3103 | return; |
3104 | } | |
3105 | out: | |
3106 | task_rq_unlock(rq, &flags); | |
3107 | } | |
3108 | ||
3109 | /* | |
476d139c NP |
3110 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3111 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3112 | */ |
3113 | void sched_exec(void) | |
3114 | { | |
1da177e4 | 3115 | int new_cpu, this_cpu = get_cpu(); |
5f3edc1b | 3116 | new_cpu = current->sched_class->select_task_rq(current, SD_BALANCE_EXEC, 0); |
1da177e4 | 3117 | put_cpu(); |
476d139c NP |
3118 | if (new_cpu != this_cpu) |
3119 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3120 | } |
3121 | ||
3122 | /* | |
3123 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3124 | * Both runqueues must be locked. | |
3125 | */ | |
dd41f596 IM |
3126 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3127 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3128 | { |
2e1cb74a | 3129 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3130 | set_task_cpu(p, this_cpu); |
dd41f596 | 3131 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3132 | /* |
3133 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3134 | * to be always true for them. | |
3135 | */ | |
15afe09b | 3136 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3137 | } |
3138 | ||
3139 | /* | |
3140 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3141 | */ | |
858119e1 | 3142 | static |
70b97a7f | 3143 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3144 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3145 | int *all_pinned) |
1da177e4 | 3146 | { |
708dc512 | 3147 | int tsk_cache_hot = 0; |
1da177e4 LT |
3148 | /* |
3149 | * We do not migrate tasks that are: | |
3150 | * 1) running (obviously), or | |
3151 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3152 | * 3) are cache-hot on their current CPU. | |
3153 | */ | |
96f874e2 | 3154 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3155 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3156 | return 0; |
cc367732 | 3157 | } |
81026794 NP |
3158 | *all_pinned = 0; |
3159 | ||
cc367732 IM |
3160 | if (task_running(rq, p)) { |
3161 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3162 | return 0; |
cc367732 | 3163 | } |
1da177e4 | 3164 | |
da84d961 IM |
3165 | /* |
3166 | * Aggressive migration if: | |
3167 | * 1) task is cache cold, or | |
3168 | * 2) too many balance attempts have failed. | |
3169 | */ | |
3170 | ||
708dc512 LH |
3171 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3172 | if (!tsk_cache_hot || | |
3173 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3174 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3175 | if (tsk_cache_hot) { |
da84d961 | 3176 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3177 | schedstat_inc(p, se.nr_forced_migrations); |
3178 | } | |
da84d961 IM |
3179 | #endif |
3180 | return 1; | |
3181 | } | |
3182 | ||
708dc512 | 3183 | if (tsk_cache_hot) { |
cc367732 | 3184 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3185 | return 0; |
cc367732 | 3186 | } |
1da177e4 LT |
3187 | return 1; |
3188 | } | |
3189 | ||
e1d1484f PW |
3190 | static unsigned long |
3191 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3192 | unsigned long max_load_move, struct sched_domain *sd, | |
3193 | enum cpu_idle_type idle, int *all_pinned, | |
3194 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3195 | { |
051c6764 | 3196 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3197 | struct task_struct *p; |
3198 | long rem_load_move = max_load_move; | |
1da177e4 | 3199 | |
e1d1484f | 3200 | if (max_load_move == 0) |
1da177e4 LT |
3201 | goto out; |
3202 | ||
81026794 NP |
3203 | pinned = 1; |
3204 | ||
1da177e4 | 3205 | /* |
dd41f596 | 3206 | * Start the load-balancing iterator: |
1da177e4 | 3207 | */ |
dd41f596 IM |
3208 | p = iterator->start(iterator->arg); |
3209 | next: | |
b82d9fdd | 3210 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3211 | goto out; |
051c6764 PZ |
3212 | |
3213 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3214 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3215 | p = iterator->next(iterator->arg); |
3216 | goto next; | |
1da177e4 LT |
3217 | } |
3218 | ||
dd41f596 | 3219 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3220 | pulled++; |
dd41f596 | 3221 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3222 | |
7e96fa58 GH |
3223 | #ifdef CONFIG_PREEMPT |
3224 | /* | |
3225 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3226 | * will stop after the first task is pulled to minimize the critical | |
3227 | * section. | |
3228 | */ | |
3229 | if (idle == CPU_NEWLY_IDLE) | |
3230 | goto out; | |
3231 | #endif | |
3232 | ||
2dd73a4f | 3233 | /* |
b82d9fdd | 3234 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3235 | */ |
e1d1484f | 3236 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3237 | if (p->prio < *this_best_prio) |
3238 | *this_best_prio = p->prio; | |
dd41f596 IM |
3239 | p = iterator->next(iterator->arg); |
3240 | goto next; | |
1da177e4 LT |
3241 | } |
3242 | out: | |
3243 | /* | |
e1d1484f | 3244 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3245 | * so we can safely collect pull_task() stats here rather than |
3246 | * inside pull_task(). | |
3247 | */ | |
3248 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3249 | |
3250 | if (all_pinned) | |
3251 | *all_pinned = pinned; | |
e1d1484f PW |
3252 | |
3253 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3254 | } |
3255 | ||
dd41f596 | 3256 | /* |
43010659 PW |
3257 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3258 | * this_rq, as part of a balancing operation within domain "sd". | |
3259 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3260 | * |
3261 | * Called with both runqueues locked. | |
3262 | */ | |
3263 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3264 | unsigned long max_load_move, |
dd41f596 IM |
3265 | struct sched_domain *sd, enum cpu_idle_type idle, |
3266 | int *all_pinned) | |
3267 | { | |
5522d5d5 | 3268 | const struct sched_class *class = sched_class_highest; |
43010659 | 3269 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3270 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3271 | |
3272 | do { | |
43010659 PW |
3273 | total_load_moved += |
3274 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3275 | max_load_move - total_load_moved, |
a4ac01c3 | 3276 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3277 | class = class->next; |
c4acb2c0 | 3278 | |
7e96fa58 GH |
3279 | #ifdef CONFIG_PREEMPT |
3280 | /* | |
3281 | * NEWIDLE balancing is a source of latency, so preemptible | |
3282 | * kernels will stop after the first task is pulled to minimize | |
3283 | * the critical section. | |
3284 | */ | |
c4acb2c0 GH |
3285 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3286 | break; | |
7e96fa58 | 3287 | #endif |
43010659 | 3288 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3289 | |
43010659 PW |
3290 | return total_load_moved > 0; |
3291 | } | |
3292 | ||
e1d1484f PW |
3293 | static int |
3294 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3295 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3296 | struct rq_iterator *iterator) | |
3297 | { | |
3298 | struct task_struct *p = iterator->start(iterator->arg); | |
3299 | int pinned = 0; | |
3300 | ||
3301 | while (p) { | |
3302 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3303 | pull_task(busiest, p, this_rq, this_cpu); | |
3304 | /* | |
3305 | * Right now, this is only the second place pull_task() | |
3306 | * is called, so we can safely collect pull_task() | |
3307 | * stats here rather than inside pull_task(). | |
3308 | */ | |
3309 | schedstat_inc(sd, lb_gained[idle]); | |
3310 | ||
3311 | return 1; | |
3312 | } | |
3313 | p = iterator->next(iterator->arg); | |
3314 | } | |
3315 | ||
3316 | return 0; | |
3317 | } | |
3318 | ||
43010659 PW |
3319 | /* |
3320 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3321 | * part of active balancing operations within "domain". | |
3322 | * Returns 1 if successful and 0 otherwise. | |
3323 | * | |
3324 | * Called with both runqueues locked. | |
3325 | */ | |
3326 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3327 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3328 | { | |
5522d5d5 | 3329 | const struct sched_class *class; |
43010659 | 3330 | |
cde7e5ca | 3331 | for_each_class(class) { |
e1d1484f | 3332 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3333 | return 1; |
cde7e5ca | 3334 | } |
43010659 PW |
3335 | |
3336 | return 0; | |
dd41f596 | 3337 | } |
67bb6c03 | 3338 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3339 | /* |
222d656d GS |
3340 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3341 | * during load balancing. | |
1da177e4 | 3342 | */ |
222d656d GS |
3343 | struct sd_lb_stats { |
3344 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3345 | struct sched_group *this; /* Local group in this sd */ | |
3346 | unsigned long total_load; /* Total load of all groups in sd */ | |
3347 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3348 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3349 | ||
3350 | /** Statistics of this group */ | |
3351 | unsigned long this_load; | |
3352 | unsigned long this_load_per_task; | |
3353 | unsigned long this_nr_running; | |
3354 | ||
3355 | /* Statistics of the busiest group */ | |
3356 | unsigned long max_load; | |
3357 | unsigned long busiest_load_per_task; | |
3358 | unsigned long busiest_nr_running; | |
3359 | ||
3360 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3361 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3362 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3363 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3364 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3365 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3366 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3367 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3368 | #endif |
222d656d | 3369 | }; |
1da177e4 | 3370 | |
d5ac537e | 3371 | /* |
381be78f GS |
3372 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3373 | */ | |
3374 | struct sg_lb_stats { | |
3375 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3376 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3377 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3378 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3379 | unsigned long group_capacity; | |
3380 | int group_imb; /* Is there an imbalance in the group ? */ | |
3381 | }; | |
408ed066 | 3382 | |
67bb6c03 GS |
3383 | /** |
3384 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3385 | * @group: The group whose first cpu is to be returned. | |
3386 | */ | |
3387 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3388 | { | |
3389 | return cpumask_first(sched_group_cpus(group)); | |
3390 | } | |
3391 | ||
3392 | /** | |
3393 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3394 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3395 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3396 | */ | |
3397 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3398 | enum cpu_idle_type idle) | |
3399 | { | |
3400 | int load_idx; | |
3401 | ||
3402 | switch (idle) { | |
3403 | case CPU_NOT_IDLE: | |
7897986b | 3404 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3405 | break; |
3406 | ||
3407 | case CPU_NEWLY_IDLE: | |
7897986b | 3408 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3409 | break; |
3410 | default: | |
7897986b | 3411 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3412 | break; |
3413 | } | |
1da177e4 | 3414 | |
67bb6c03 GS |
3415 | return load_idx; |
3416 | } | |
1da177e4 | 3417 | |
1da177e4 | 3418 | |
c071df18 GS |
3419 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3420 | /** | |
3421 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3422 | * the given sched_domain, during load balancing. | |
3423 | * | |
3424 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3425 | * @sds: Variable containing the statistics for sd. | |
3426 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3427 | */ | |
3428 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3429 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3430 | { | |
3431 | /* | |
3432 | * Busy processors will not participate in power savings | |
3433 | * balance. | |
3434 | */ | |
3435 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3436 | sds->power_savings_balance = 0; | |
3437 | else { | |
3438 | sds->power_savings_balance = 1; | |
3439 | sds->min_nr_running = ULONG_MAX; | |
3440 | sds->leader_nr_running = 0; | |
3441 | } | |
3442 | } | |
783609c6 | 3443 | |
c071df18 GS |
3444 | /** |
3445 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3446 | * sched_domain while performing load balancing. | |
3447 | * | |
3448 | * @group: sched_group belonging to the sched_domain under consideration. | |
3449 | * @sds: Variable containing the statistics of the sched_domain | |
3450 | * @local_group: Does group contain the CPU for which we're performing | |
3451 | * load balancing ? | |
3452 | * @sgs: Variable containing the statistics of the group. | |
3453 | */ | |
3454 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3455 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3456 | { | |
408ed066 | 3457 | |
c071df18 GS |
3458 | if (!sds->power_savings_balance) |
3459 | return; | |
1da177e4 | 3460 | |
c071df18 GS |
3461 | /* |
3462 | * If the local group is idle or completely loaded | |
3463 | * no need to do power savings balance at this domain | |
3464 | */ | |
3465 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3466 | !sds->this_nr_running)) | |
3467 | sds->power_savings_balance = 0; | |
2dd73a4f | 3468 | |
c071df18 GS |
3469 | /* |
3470 | * If a group is already running at full capacity or idle, | |
3471 | * don't include that group in power savings calculations | |
3472 | */ | |
3473 | if (!sds->power_savings_balance || | |
3474 | sgs->sum_nr_running >= sgs->group_capacity || | |
3475 | !sgs->sum_nr_running) | |
3476 | return; | |
5969fe06 | 3477 | |
c071df18 GS |
3478 | /* |
3479 | * Calculate the group which has the least non-idle load. | |
3480 | * This is the group from where we need to pick up the load | |
3481 | * for saving power | |
3482 | */ | |
3483 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3484 | (sgs->sum_nr_running == sds->min_nr_running && | |
3485 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3486 | sds->group_min = group; | |
3487 | sds->min_nr_running = sgs->sum_nr_running; | |
3488 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3489 | sgs->sum_nr_running; | |
3490 | } | |
783609c6 | 3491 | |
c071df18 GS |
3492 | /* |
3493 | * Calculate the group which is almost near its | |
3494 | * capacity but still has some space to pick up some load | |
3495 | * from other group and save more power | |
3496 | */ | |
d899a789 | 3497 | if (sgs->sum_nr_running + 1 > sgs->group_capacity) |
c071df18 | 3498 | return; |
1da177e4 | 3499 | |
c071df18 GS |
3500 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3501 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3502 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3503 | sds->group_leader = group; | |
3504 | sds->leader_nr_running = sgs->sum_nr_running; | |
3505 | } | |
3506 | } | |
408ed066 | 3507 | |
c071df18 | 3508 | /** |
d5ac537e | 3509 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3510 | * @sds: Variable containing the statistics of the sched_domain |
3511 | * under consideration. | |
3512 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3513 | * @imbalance: Variable to store the imbalance. | |
3514 | * | |
d5ac537e RD |
3515 | * Description: |
3516 | * Check if we have potential to perform some power-savings balance. | |
3517 | * If yes, set the busiest group to be the least loaded group in the | |
3518 | * sched_domain, so that it's CPUs can be put to idle. | |
3519 | * | |
c071df18 GS |
3520 | * Returns 1 if there is potential to perform power-savings balance. |
3521 | * Else returns 0. | |
3522 | */ | |
3523 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3524 | int this_cpu, unsigned long *imbalance) | |
3525 | { | |
3526 | if (!sds->power_savings_balance) | |
3527 | return 0; | |
1da177e4 | 3528 | |
c071df18 GS |
3529 | if (sds->this != sds->group_leader || |
3530 | sds->group_leader == sds->group_min) | |
3531 | return 0; | |
783609c6 | 3532 | |
c071df18 GS |
3533 | *imbalance = sds->min_load_per_task; |
3534 | sds->busiest = sds->group_min; | |
1da177e4 | 3535 | |
c071df18 GS |
3536 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3537 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3538 | group_first_cpu(sds->group_leader); | |
3539 | } | |
3540 | ||
3541 | return 1; | |
1da177e4 | 3542 | |
c071df18 GS |
3543 | } |
3544 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3545 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3546 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3547 | { | |
3548 | return; | |
3549 | } | |
408ed066 | 3550 | |
c071df18 GS |
3551 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3552 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3553 | { | |
3554 | return; | |
3555 | } | |
3556 | ||
3557 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3558 | int this_cpu, unsigned long *imbalance) | |
3559 | { | |
3560 | return 0; | |
3561 | } | |
3562 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3563 | ||
e9e9250b | 3564 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
ab29230e PZ |
3565 | { |
3566 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3567 | unsigned long smt_gain = sd->smt_gain; | |
3568 | ||
3569 | smt_gain /= weight; | |
3570 | ||
3571 | return smt_gain; | |
3572 | } | |
3573 | ||
e9e9250b PZ |
3574 | unsigned long scale_rt_power(int cpu) |
3575 | { | |
3576 | struct rq *rq = cpu_rq(cpu); | |
3577 | u64 total, available; | |
3578 | ||
3579 | sched_avg_update(rq); | |
3580 | ||
3581 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3582 | available = total - rq->rt_avg; | |
3583 | ||
3584 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3585 | total = SCHED_LOAD_SCALE; | |
3586 | ||
3587 | total >>= SCHED_LOAD_SHIFT; | |
3588 | ||
3589 | return div_u64(available, total); | |
3590 | } | |
3591 | ||
ab29230e PZ |
3592 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3593 | { | |
3594 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3595 | unsigned long power = SCHED_LOAD_SCALE; | |
3596 | struct sched_group *sdg = sd->groups; | |
ab29230e PZ |
3597 | |
3598 | /* here we could scale based on cpufreq */ | |
3599 | ||
3600 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
e9e9250b | 3601 | power *= arch_scale_smt_power(sd, cpu); |
ab29230e PZ |
3602 | power >>= SCHED_LOAD_SHIFT; |
3603 | } | |
3604 | ||
e9e9250b PZ |
3605 | power *= scale_rt_power(cpu); |
3606 | power >>= SCHED_LOAD_SHIFT; | |
3607 | ||
3608 | if (!power) | |
3609 | power = 1; | |
ab29230e | 3610 | |
18a3885f | 3611 | sdg->cpu_power = power; |
ab29230e PZ |
3612 | } |
3613 | ||
3614 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3615 | { |
3616 | struct sched_domain *child = sd->child; | |
3617 | struct sched_group *group, *sdg = sd->groups; | |
d7ea17a7 | 3618 | unsigned long power; |
cc9fba7d PZ |
3619 | |
3620 | if (!child) { | |
ab29230e | 3621 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3622 | return; |
3623 | } | |
3624 | ||
d7ea17a7 | 3625 | power = 0; |
cc9fba7d PZ |
3626 | |
3627 | group = child->groups; | |
3628 | do { | |
d7ea17a7 | 3629 | power += group->cpu_power; |
cc9fba7d PZ |
3630 | group = group->next; |
3631 | } while (group != child->groups); | |
d7ea17a7 IM |
3632 | |
3633 | sdg->cpu_power = power; | |
cc9fba7d | 3634 | } |
c071df18 | 3635 | |
1f8c553d GS |
3636 | /** |
3637 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3638 | * @group: sched_group whose statistics are to be updated. | |
3639 | * @this_cpu: Cpu for which load balance is currently performed. | |
3640 | * @idle: Idle status of this_cpu | |
3641 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3642 | * @sd_idle: Idle status of the sched_domain containing group. | |
3643 | * @local_group: Does group contain this_cpu. | |
3644 | * @cpus: Set of cpus considered for load balancing. | |
3645 | * @balance: Should we balance. | |
3646 | * @sgs: variable to hold the statistics for this group. | |
3647 | */ | |
cc9fba7d PZ |
3648 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3649 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3650 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3651 | int local_group, const struct cpumask *cpus, | |
3652 | int *balance, struct sg_lb_stats *sgs) | |
3653 | { | |
3654 | unsigned long load, max_cpu_load, min_cpu_load; | |
3655 | int i; | |
3656 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3657 | unsigned long sum_avg_load_per_task; | |
3658 | unsigned long avg_load_per_task; | |
3659 | ||
cc9fba7d | 3660 | if (local_group) { |
1f8c553d | 3661 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3662 | if (balance_cpu == this_cpu) |
ab29230e | 3663 | update_group_power(sd, this_cpu); |
cc9fba7d | 3664 | } |
1f8c553d GS |
3665 | |
3666 | /* Tally up the load of all CPUs in the group */ | |
3667 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3668 | max_cpu_load = 0; | |
3669 | min_cpu_load = ~0UL; | |
408ed066 | 3670 | |
1f8c553d GS |
3671 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3672 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3673 | |
1f8c553d GS |
3674 | if (*sd_idle && rq->nr_running) |
3675 | *sd_idle = 0; | |
5c45bf27 | 3676 | |
1f8c553d | 3677 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3678 | if (local_group) { |
1f8c553d GS |
3679 | if (idle_cpu(i) && !first_idle_cpu) { |
3680 | first_idle_cpu = 1; | |
3681 | balance_cpu = i; | |
3682 | } | |
3683 | ||
3684 | load = target_load(i, load_idx); | |
3685 | } else { | |
3686 | load = source_load(i, load_idx); | |
3687 | if (load > max_cpu_load) | |
3688 | max_cpu_load = load; | |
3689 | if (min_cpu_load > load) | |
3690 | min_cpu_load = load; | |
1da177e4 | 3691 | } |
5c45bf27 | 3692 | |
1f8c553d GS |
3693 | sgs->group_load += load; |
3694 | sgs->sum_nr_running += rq->nr_running; | |
3695 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3696 | |
1f8c553d GS |
3697 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3698 | } | |
5c45bf27 | 3699 | |
1f8c553d GS |
3700 | /* |
3701 | * First idle cpu or the first cpu(busiest) in this sched group | |
3702 | * is eligible for doing load balancing at this and above | |
3703 | * domains. In the newly idle case, we will allow all the cpu's | |
3704 | * to do the newly idle load balance. | |
3705 | */ | |
3706 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3707 | balance_cpu != this_cpu && balance) { | |
3708 | *balance = 0; | |
3709 | return; | |
3710 | } | |
5c45bf27 | 3711 | |
1f8c553d | 3712 | /* Adjust by relative CPU power of the group */ |
18a3885f | 3713 | sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power; |
5c45bf27 | 3714 | |
1f8c553d GS |
3715 | |
3716 | /* | |
3717 | * Consider the group unbalanced when the imbalance is larger | |
3718 | * than the average weight of two tasks. | |
3719 | * | |
3720 | * APZ: with cgroup the avg task weight can vary wildly and | |
3721 | * might not be a suitable number - should we keep a | |
3722 | * normalized nr_running number somewhere that negates | |
3723 | * the hierarchy? | |
3724 | */ | |
18a3885f PZ |
3725 | avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) / |
3726 | group->cpu_power; | |
1f8c553d GS |
3727 | |
3728 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3729 | sgs->group_imb = 1; | |
3730 | ||
bdb94aa5 | 3731 | sgs->group_capacity = |
18a3885f | 3732 | DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE); |
1f8c553d | 3733 | } |
dd41f596 | 3734 | |
37abe198 GS |
3735 | /** |
3736 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3737 | * @sd: sched_domain whose statistics are to be updated. | |
3738 | * @this_cpu: Cpu for which load balance is currently performed. | |
3739 | * @idle: Idle status of this_cpu | |
3740 | * @sd_idle: Idle status of the sched_domain containing group. | |
3741 | * @cpus: Set of cpus considered for load balancing. | |
3742 | * @balance: Should we balance. | |
3743 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3744 | */ |
37abe198 GS |
3745 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3746 | enum cpu_idle_type idle, int *sd_idle, | |
3747 | const struct cpumask *cpus, int *balance, | |
3748 | struct sd_lb_stats *sds) | |
1da177e4 | 3749 | { |
b5d978e0 | 3750 | struct sched_domain *child = sd->child; |
222d656d | 3751 | struct sched_group *group = sd->groups; |
37abe198 | 3752 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3753 | int load_idx, prefer_sibling = 0; |
3754 | ||
3755 | if (child && child->flags & SD_PREFER_SIBLING) | |
3756 | prefer_sibling = 1; | |
222d656d | 3757 | |
c071df18 | 3758 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3759 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3760 | |
3761 | do { | |
1da177e4 | 3762 | int local_group; |
1da177e4 | 3763 | |
758b2cdc RR |
3764 | local_group = cpumask_test_cpu(this_cpu, |
3765 | sched_group_cpus(group)); | |
381be78f | 3766 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3767 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3768 | local_group, cpus, balance, &sgs); |
1da177e4 | 3769 | |
37abe198 GS |
3770 | if (local_group && balance && !(*balance)) |
3771 | return; | |
783609c6 | 3772 | |
37abe198 | 3773 | sds->total_load += sgs.group_load; |
18a3885f | 3774 | sds->total_pwr += group->cpu_power; |
1da177e4 | 3775 | |
b5d978e0 PZ |
3776 | /* |
3777 | * In case the child domain prefers tasks go to siblings | |
3778 | * first, lower the group capacity to one so that we'll try | |
3779 | * and move all the excess tasks away. | |
3780 | */ | |
3781 | if (prefer_sibling) | |
bdb94aa5 | 3782 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
1da177e4 | 3783 | |
1da177e4 | 3784 | if (local_group) { |
37abe198 GS |
3785 | sds->this_load = sgs.avg_load; |
3786 | sds->this = group; | |
3787 | sds->this_nr_running = sgs.sum_nr_running; | |
3788 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3789 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3790 | (sgs.sum_nr_running > sgs.group_capacity || |
3791 | sgs.group_imb)) { | |
37abe198 GS |
3792 | sds->max_load = sgs.avg_load; |
3793 | sds->busiest = group; | |
3794 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3795 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3796 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3797 | } |
5c45bf27 | 3798 | |
c071df18 | 3799 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3800 | group = group->next; |
3801 | } while (group != sd->groups); | |
37abe198 | 3802 | } |
1da177e4 | 3803 | |
2e6f44ae GS |
3804 | /** |
3805 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3806 | * amongst the groups of a sched_domain, during |
3807 | * load balancing. | |
2e6f44ae GS |
3808 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3809 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3810 | * @imbalance: Variable to store the imbalance. | |
3811 | */ | |
3812 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3813 | int this_cpu, unsigned long *imbalance) | |
3814 | { | |
3815 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3816 | unsigned int imbn = 2; | |
3817 | ||
3818 | if (sds->this_nr_running) { | |
3819 | sds->this_load_per_task /= sds->this_nr_running; | |
3820 | if (sds->busiest_load_per_task > | |
3821 | sds->this_load_per_task) | |
3822 | imbn = 1; | |
3823 | } else | |
3824 | sds->this_load_per_task = | |
3825 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3826 | |
2e6f44ae GS |
3827 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3828 | sds->busiest_load_per_task * imbn) { | |
3829 | *imbalance = sds->busiest_load_per_task; | |
3830 | return; | |
3831 | } | |
908a7c1b | 3832 | |
1da177e4 | 3833 | /* |
2e6f44ae GS |
3834 | * OK, we don't have enough imbalance to justify moving tasks, |
3835 | * however we may be able to increase total CPU power used by | |
3836 | * moving them. | |
1da177e4 | 3837 | */ |
2dd73a4f | 3838 | |
18a3885f | 3839 | pwr_now += sds->busiest->cpu_power * |
2e6f44ae | 3840 | min(sds->busiest_load_per_task, sds->max_load); |
18a3885f | 3841 | pwr_now += sds->this->cpu_power * |
2e6f44ae GS |
3842 | min(sds->this_load_per_task, sds->this_load); |
3843 | pwr_now /= SCHED_LOAD_SCALE; | |
3844 | ||
3845 | /* Amount of load we'd subtract */ | |
18a3885f PZ |
3846 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3847 | sds->busiest->cpu_power; | |
2e6f44ae | 3848 | if (sds->max_load > tmp) |
18a3885f | 3849 | pwr_move += sds->busiest->cpu_power * |
2e6f44ae GS |
3850 | min(sds->busiest_load_per_task, sds->max_load - tmp); |
3851 | ||
3852 | /* Amount of load we'd add */ | |
18a3885f | 3853 | if (sds->max_load * sds->busiest->cpu_power < |
2e6f44ae | 3854 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) |
18a3885f PZ |
3855 | tmp = (sds->max_load * sds->busiest->cpu_power) / |
3856 | sds->this->cpu_power; | |
2e6f44ae | 3857 | else |
18a3885f PZ |
3858 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3859 | sds->this->cpu_power; | |
3860 | pwr_move += sds->this->cpu_power * | |
2e6f44ae GS |
3861 | min(sds->this_load_per_task, sds->this_load + tmp); |
3862 | pwr_move /= SCHED_LOAD_SCALE; | |
3863 | ||
3864 | /* Move if we gain throughput */ | |
3865 | if (pwr_move > pwr_now) | |
3866 | *imbalance = sds->busiest_load_per_task; | |
3867 | } | |
dbc523a3 GS |
3868 | |
3869 | /** | |
3870 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3871 | * groups of a given sched_domain during load balance. | |
3872 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3873 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3874 | * @imbalance: The variable to store the imbalance. | |
3875 | */ | |
3876 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3877 | unsigned long *imbalance) | |
3878 | { | |
3879 | unsigned long max_pull; | |
2dd73a4f PW |
3880 | /* |
3881 | * In the presence of smp nice balancing, certain scenarios can have | |
3882 | * max load less than avg load(as we skip the groups at or below | |
3883 | * its cpu_power, while calculating max_load..) | |
3884 | */ | |
dbc523a3 | 3885 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3886 | *imbalance = 0; |
dbc523a3 | 3887 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3888 | } |
0c117f1b SS |
3889 | |
3890 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3891 | max_pull = min(sds->max_load - sds->avg_load, |
3892 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3893 | |
1da177e4 | 3894 | /* How much load to actually move to equalise the imbalance */ |
18a3885f PZ |
3895 | *imbalance = min(max_pull * sds->busiest->cpu_power, |
3896 | (sds->avg_load - sds->this_load) * sds->this->cpu_power) | |
1da177e4 LT |
3897 | / SCHED_LOAD_SCALE; |
3898 | ||
2dd73a4f PW |
3899 | /* |
3900 | * if *imbalance is less than the average load per runnable task | |
3901 | * there is no gaurantee that any tasks will be moved so we'll have | |
3902 | * a think about bumping its value to force at least one task to be | |
3903 | * moved | |
3904 | */ | |
dbc523a3 GS |
3905 | if (*imbalance < sds->busiest_load_per_task) |
3906 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3907 | |
dbc523a3 | 3908 | } |
37abe198 | 3909 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3910 | |
b7bb4c9b GS |
3911 | /** |
3912 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3913 | * if there is an imbalance. If there isn't an imbalance, and | |
3914 | * the user has opted for power-savings, it returns a group whose | |
3915 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3916 | * such a group exists. | |
3917 | * | |
3918 | * Also calculates the amount of weighted load which should be moved | |
3919 | * to restore balance. | |
3920 | * | |
3921 | * @sd: The sched_domain whose busiest group is to be returned. | |
3922 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3923 | * @imbalance: Variable which stores amount of weighted load which should | |
3924 | * be moved to restore balance/put a group to idle. | |
3925 | * @idle: The idle status of this_cpu. | |
3926 | * @sd_idle: The idleness of sd | |
3927 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3928 | * @balance: Pointer to a variable indicating if this_cpu | |
3929 | * is the appropriate cpu to perform load balancing at this_level. | |
3930 | * | |
3931 | * Returns: - the busiest group if imbalance exists. | |
3932 | * - If no imbalance and user has opted for power-savings balance, | |
3933 | * return the least loaded group whose CPUs can be | |
3934 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3935 | */ |
3936 | static struct sched_group * | |
3937 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3938 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3939 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3940 | { | |
3941 | struct sd_lb_stats sds; | |
1da177e4 | 3942 | |
37abe198 | 3943 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3944 | |
37abe198 GS |
3945 | /* |
3946 | * Compute the various statistics relavent for load balancing at | |
3947 | * this level. | |
3948 | */ | |
3949 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3950 | balance, &sds); | |
3951 | ||
b7bb4c9b GS |
3952 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3953 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3954 | * at this level. | |
3955 | * 2) There is no busy sibling group to pull from. | |
3956 | * 3) This group is the busiest group. | |
3957 | * 4) This group is more busy than the avg busieness at this | |
3958 | * sched_domain. | |
3959 | * 5) The imbalance is within the specified limit. | |
3960 | * 6) Any rebalance would lead to ping-pong | |
3961 | */ | |
37abe198 GS |
3962 | if (balance && !(*balance)) |
3963 | goto ret; | |
1da177e4 | 3964 | |
b7bb4c9b GS |
3965 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3966 | goto out_balanced; | |
1da177e4 | 3967 | |
b7bb4c9b | 3968 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3969 | goto out_balanced; |
1da177e4 | 3970 | |
222d656d | 3971 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3972 | |
b7bb4c9b GS |
3973 | if (sds.this_load >= sds.avg_load) |
3974 | goto out_balanced; | |
3975 | ||
3976 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3977 | goto out_balanced; |
3978 | ||
222d656d GS |
3979 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3980 | if (sds.group_imb) | |
3981 | sds.busiest_load_per_task = | |
3982 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3983 | |
1da177e4 LT |
3984 | /* |
3985 | * We're trying to get all the cpus to the average_load, so we don't | |
3986 | * want to push ourselves above the average load, nor do we wish to | |
3987 | * reduce the max loaded cpu below the average load, as either of these | |
3988 | * actions would just result in more rebalancing later, and ping-pong | |
3989 | * tasks around. Thus we look for the minimum possible imbalance. | |
3990 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3991 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3992 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3993 | * appear as very large values with unsigned longs. |
3994 | */ | |
222d656d | 3995 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3996 | goto out_balanced; |
3997 | ||
dbc523a3 GS |
3998 | /* Looks like there is an imbalance. Compute it */ |
3999 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4000 | return sds.busiest; |
1da177e4 LT |
4001 | |
4002 | out_balanced: | |
c071df18 GS |
4003 | /* |
4004 | * There is no obvious imbalance. But check if we can do some balancing | |
4005 | * to save power. | |
4006 | */ | |
4007 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4008 | return sds.busiest; | |
783609c6 | 4009 | ret: |
1da177e4 LT |
4010 | *imbalance = 0; |
4011 | return NULL; | |
4012 | } | |
4013 | ||
bdb94aa5 PZ |
4014 | static struct sched_group *group_of(int cpu) |
4015 | { | |
4016 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
4017 | ||
4018 | if (!sd) | |
4019 | return NULL; | |
4020 | ||
4021 | return sd->groups; | |
4022 | } | |
4023 | ||
4024 | static unsigned long power_of(int cpu) | |
4025 | { | |
4026 | struct sched_group *group = group_of(cpu); | |
4027 | ||
4028 | if (!group) | |
4029 | return SCHED_LOAD_SCALE; | |
4030 | ||
18a3885f | 4031 | return group->cpu_power; |
bdb94aa5 PZ |
4032 | } |
4033 | ||
1da177e4 LT |
4034 | /* |
4035 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4036 | */ | |
70b97a7f | 4037 | static struct rq * |
d15bcfdb | 4038 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4039 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4040 | { |
70b97a7f | 4041 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4042 | unsigned long max_load = 0; |
1da177e4 LT |
4043 | int i; |
4044 | ||
758b2cdc | 4045 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4046 | unsigned long power = power_of(i); |
4047 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4048 | unsigned long wl; |
0a2966b4 | 4049 | |
96f874e2 | 4050 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4051 | continue; |
4052 | ||
48f24c4d | 4053 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4054 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4055 | wl /= power; | |
2dd73a4f | 4056 | |
bdb94aa5 | 4057 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4058 | continue; |
1da177e4 | 4059 | |
dd41f596 IM |
4060 | if (wl > max_load) { |
4061 | max_load = wl; | |
48f24c4d | 4062 | busiest = rq; |
1da177e4 LT |
4063 | } |
4064 | } | |
4065 | ||
4066 | return busiest; | |
4067 | } | |
4068 | ||
77391d71 NP |
4069 | /* |
4070 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4071 | * so long as it is large enough. | |
4072 | */ | |
4073 | #define MAX_PINNED_INTERVAL 512 | |
4074 | ||
df7c8e84 RR |
4075 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4076 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4077 | ||
1da177e4 LT |
4078 | /* |
4079 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4080 | * tasks if there is an imbalance. | |
1da177e4 | 4081 | */ |
70b97a7f | 4082 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4083 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4084 | int *balance) |
1da177e4 | 4085 | { |
43010659 | 4086 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4087 | struct sched_group *group; |
1da177e4 | 4088 | unsigned long imbalance; |
70b97a7f | 4089 | struct rq *busiest; |
fe2eea3f | 4090 | unsigned long flags; |
df7c8e84 | 4091 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4092 | |
96f874e2 | 4093 | cpumask_setall(cpus); |
7c16ec58 | 4094 | |
89c4710e SS |
4095 | /* |
4096 | * When power savings policy is enabled for the parent domain, idle | |
4097 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4098 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4099 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4100 | */ |
d15bcfdb | 4101 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4102 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4103 | sd_idle = 1; |
1da177e4 | 4104 | |
2d72376b | 4105 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4106 | |
0a2966b4 | 4107 | redo: |
c8cba857 | 4108 | update_shares(sd); |
0a2966b4 | 4109 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4110 | cpus, balance); |
783609c6 | 4111 | |
06066714 | 4112 | if (*balance == 0) |
783609c6 | 4113 | goto out_balanced; |
783609c6 | 4114 | |
1da177e4 LT |
4115 | if (!group) { |
4116 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4117 | goto out_balanced; | |
4118 | } | |
4119 | ||
7c16ec58 | 4120 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4121 | if (!busiest) { |
4122 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4123 | goto out_balanced; | |
4124 | } | |
4125 | ||
db935dbd | 4126 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4127 | |
4128 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4129 | ||
43010659 | 4130 | ld_moved = 0; |
1da177e4 LT |
4131 | if (busiest->nr_running > 1) { |
4132 | /* | |
4133 | * Attempt to move tasks. If find_busiest_group has found | |
4134 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4135 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4136 | * correctly treated as an imbalance. |
4137 | */ | |
fe2eea3f | 4138 | local_irq_save(flags); |
e17224bf | 4139 | double_rq_lock(this_rq, busiest); |
43010659 | 4140 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4141 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4142 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4143 | local_irq_restore(flags); |
81026794 | 4144 | |
46cb4b7c SS |
4145 | /* |
4146 | * some other cpu did the load balance for us. | |
4147 | */ | |
43010659 | 4148 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4149 | resched_cpu(this_cpu); |
4150 | ||
81026794 | 4151 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4152 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4153 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4154 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4155 | goto redo; |
81026794 | 4156 | goto out_balanced; |
0a2966b4 | 4157 | } |
1da177e4 | 4158 | } |
81026794 | 4159 | |
43010659 | 4160 | if (!ld_moved) { |
1da177e4 LT |
4161 | schedstat_inc(sd, lb_failed[idle]); |
4162 | sd->nr_balance_failed++; | |
4163 | ||
4164 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4165 | |
fe2eea3f | 4166 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4167 | |
4168 | /* don't kick the migration_thread, if the curr | |
4169 | * task on busiest cpu can't be moved to this_cpu | |
4170 | */ | |
96f874e2 RR |
4171 | if (!cpumask_test_cpu(this_cpu, |
4172 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4173 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4174 | all_pinned = 1; |
4175 | goto out_one_pinned; | |
4176 | } | |
4177 | ||
1da177e4 LT |
4178 | if (!busiest->active_balance) { |
4179 | busiest->active_balance = 1; | |
4180 | busiest->push_cpu = this_cpu; | |
81026794 | 4181 | active_balance = 1; |
1da177e4 | 4182 | } |
fe2eea3f | 4183 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4184 | if (active_balance) |
1da177e4 LT |
4185 | wake_up_process(busiest->migration_thread); |
4186 | ||
4187 | /* | |
4188 | * We've kicked active balancing, reset the failure | |
4189 | * counter. | |
4190 | */ | |
39507451 | 4191 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4192 | } |
81026794 | 4193 | } else |
1da177e4 LT |
4194 | sd->nr_balance_failed = 0; |
4195 | ||
81026794 | 4196 | if (likely(!active_balance)) { |
1da177e4 LT |
4197 | /* We were unbalanced, so reset the balancing interval */ |
4198 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4199 | } else { |
4200 | /* | |
4201 | * If we've begun active balancing, start to back off. This | |
4202 | * case may not be covered by the all_pinned logic if there | |
4203 | * is only 1 task on the busy runqueue (because we don't call | |
4204 | * move_tasks). | |
4205 | */ | |
4206 | if (sd->balance_interval < sd->max_interval) | |
4207 | sd->balance_interval *= 2; | |
1da177e4 LT |
4208 | } |
4209 | ||
43010659 | 4210 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4211 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4212 | ld_moved = -1; |
4213 | ||
4214 | goto out; | |
1da177e4 LT |
4215 | |
4216 | out_balanced: | |
1da177e4 LT |
4217 | schedstat_inc(sd, lb_balanced[idle]); |
4218 | ||
16cfb1c0 | 4219 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4220 | |
4221 | out_one_pinned: | |
1da177e4 | 4222 | /* tune up the balancing interval */ |
77391d71 NP |
4223 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4224 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4225 | sd->balance_interval *= 2; |
4226 | ||
48f24c4d | 4227 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4228 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4229 | ld_moved = -1; |
4230 | else | |
4231 | ld_moved = 0; | |
4232 | out: | |
c8cba857 PZ |
4233 | if (ld_moved) |
4234 | update_shares(sd); | |
c09595f6 | 4235 | return ld_moved; |
1da177e4 LT |
4236 | } |
4237 | ||
4238 | /* | |
4239 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4240 | * tasks if there is an imbalance. | |
4241 | * | |
d15bcfdb | 4242 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4243 | * this_rq is locked. |
4244 | */ | |
48f24c4d | 4245 | static int |
df7c8e84 | 4246 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4247 | { |
4248 | struct sched_group *group; | |
70b97a7f | 4249 | struct rq *busiest = NULL; |
1da177e4 | 4250 | unsigned long imbalance; |
43010659 | 4251 | int ld_moved = 0; |
5969fe06 | 4252 | int sd_idle = 0; |
969bb4e4 | 4253 | int all_pinned = 0; |
df7c8e84 | 4254 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4255 | |
96f874e2 | 4256 | cpumask_setall(cpus); |
5969fe06 | 4257 | |
89c4710e SS |
4258 | /* |
4259 | * When power savings policy is enabled for the parent domain, idle | |
4260 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4261 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4262 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4263 | */ |
4264 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4265 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4266 | sd_idle = 1; |
1da177e4 | 4267 | |
2d72376b | 4268 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4269 | redo: |
3e5459b4 | 4270 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4271 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4272 | &sd_idle, cpus, NULL); |
1da177e4 | 4273 | if (!group) { |
d15bcfdb | 4274 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4275 | goto out_balanced; |
1da177e4 LT |
4276 | } |
4277 | ||
7c16ec58 | 4278 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4279 | if (!busiest) { |
d15bcfdb | 4280 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4281 | goto out_balanced; |
1da177e4 LT |
4282 | } |
4283 | ||
db935dbd NP |
4284 | BUG_ON(busiest == this_rq); |
4285 | ||
d15bcfdb | 4286 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4287 | |
43010659 | 4288 | ld_moved = 0; |
d6d5cfaf NP |
4289 | if (busiest->nr_running > 1) { |
4290 | /* Attempt to move tasks */ | |
4291 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4292 | /* this_rq->clock is already updated */ |
4293 | update_rq_clock(busiest); | |
43010659 | 4294 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4295 | imbalance, sd, CPU_NEWLY_IDLE, |
4296 | &all_pinned); | |
1b12bbc7 | 4297 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4298 | |
969bb4e4 | 4299 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4300 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4301 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4302 | goto redo; |
4303 | } | |
d6d5cfaf NP |
4304 | } |
4305 | ||
43010659 | 4306 | if (!ld_moved) { |
36dffab6 | 4307 | int active_balance = 0; |
ad273b32 | 4308 | |
d15bcfdb | 4309 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4310 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4311 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4312 | return -1; |
ad273b32 VS |
4313 | |
4314 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4315 | return -1; | |
4316 | ||
4317 | if (sd->nr_balance_failed++ < 2) | |
4318 | return -1; | |
4319 | ||
4320 | /* | |
4321 | * The only task running in a non-idle cpu can be moved to this | |
4322 | * cpu in an attempt to completely freeup the other CPU | |
4323 | * package. The same method used to move task in load_balance() | |
4324 | * have been extended for load_balance_newidle() to speedup | |
4325 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4326 | * | |
4327 | * The package power saving logic comes from | |
4328 | * find_busiest_group(). If there are no imbalance, then | |
4329 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4330 | * f_b_g() will select a group from which a running task may be | |
4331 | * pulled to this cpu in order to make the other package idle. | |
4332 | * If there is no opportunity to make a package idle and if | |
4333 | * there are no imbalance, then f_b_g() will return NULL and no | |
4334 | * action will be taken in load_balance_newidle(). | |
4335 | * | |
4336 | * Under normal task pull operation due to imbalance, there | |
4337 | * will be more than one task in the source run queue and | |
4338 | * move_tasks() will succeed. ld_moved will be true and this | |
4339 | * active balance code will not be triggered. | |
4340 | */ | |
4341 | ||
4342 | /* Lock busiest in correct order while this_rq is held */ | |
4343 | double_lock_balance(this_rq, busiest); | |
4344 | ||
4345 | /* | |
4346 | * don't kick the migration_thread, if the curr | |
4347 | * task on busiest cpu can't be moved to this_cpu | |
4348 | */ | |
6ca09dfc | 4349 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4350 | double_unlock_balance(this_rq, busiest); |
4351 | all_pinned = 1; | |
4352 | return ld_moved; | |
4353 | } | |
4354 | ||
4355 | if (!busiest->active_balance) { | |
4356 | busiest->active_balance = 1; | |
4357 | busiest->push_cpu = this_cpu; | |
4358 | active_balance = 1; | |
4359 | } | |
4360 | ||
4361 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4362 | /* |
4363 | * Should not call ttwu while holding a rq->lock | |
4364 | */ | |
4365 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4366 | if (active_balance) |
4367 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4368 | spin_lock(&this_rq->lock); |
ad273b32 | 4369 | |
5969fe06 | 4370 | } else |
16cfb1c0 | 4371 | sd->nr_balance_failed = 0; |
1da177e4 | 4372 | |
3e5459b4 | 4373 | update_shares_locked(this_rq, sd); |
43010659 | 4374 | return ld_moved; |
16cfb1c0 NP |
4375 | |
4376 | out_balanced: | |
d15bcfdb | 4377 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4378 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4379 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4380 | return -1; |
16cfb1c0 | 4381 | sd->nr_balance_failed = 0; |
48f24c4d | 4382 | |
16cfb1c0 | 4383 | return 0; |
1da177e4 LT |
4384 | } |
4385 | ||
4386 | /* | |
4387 | * idle_balance is called by schedule() if this_cpu is about to become | |
4388 | * idle. Attempts to pull tasks from other CPUs. | |
4389 | */ | |
70b97a7f | 4390 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4391 | { |
4392 | struct sched_domain *sd; | |
efbe027e | 4393 | int pulled_task = 0; |
dd41f596 | 4394 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4395 | |
4396 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4397 | unsigned long interval; |
4398 | ||
4399 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4400 | continue; | |
4401 | ||
4402 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4403 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4404 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4405 | sd); |
92c4ca5c CL |
4406 | |
4407 | interval = msecs_to_jiffies(sd->balance_interval); | |
4408 | if (time_after(next_balance, sd->last_balance + interval)) | |
4409 | next_balance = sd->last_balance + interval; | |
4410 | if (pulled_task) | |
4411 | break; | |
1da177e4 | 4412 | } |
dd41f596 | 4413 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4414 | /* |
4415 | * We are going idle. next_balance may be set based on | |
4416 | * a busy processor. So reset next_balance. | |
4417 | */ | |
4418 | this_rq->next_balance = next_balance; | |
dd41f596 | 4419 | } |
1da177e4 LT |
4420 | } |
4421 | ||
4422 | /* | |
4423 | * active_load_balance is run by migration threads. It pushes running tasks | |
4424 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4425 | * running on each physical CPU where possible, and avoids physical / | |
4426 | * logical imbalances. | |
4427 | * | |
4428 | * Called with busiest_rq locked. | |
4429 | */ | |
70b97a7f | 4430 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4431 | { |
39507451 | 4432 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4433 | struct sched_domain *sd; |
4434 | struct rq *target_rq; | |
39507451 | 4435 | |
48f24c4d | 4436 | /* Is there any task to move? */ |
39507451 | 4437 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4438 | return; |
4439 | ||
4440 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4441 | |
4442 | /* | |
39507451 | 4443 | * This condition is "impossible", if it occurs |
41a2d6cf | 4444 | * we need to fix it. Originally reported by |
39507451 | 4445 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4446 | */ |
39507451 | 4447 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4448 | |
39507451 NP |
4449 | /* move a task from busiest_rq to target_rq */ |
4450 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4451 | update_rq_clock(busiest_rq); |
4452 | update_rq_clock(target_rq); | |
39507451 NP |
4453 | |
4454 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4455 | for_each_domain(target_cpu, sd) { |
39507451 | 4456 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4457 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4458 | break; |
c96d145e | 4459 | } |
39507451 | 4460 | |
48f24c4d | 4461 | if (likely(sd)) { |
2d72376b | 4462 | schedstat_inc(sd, alb_count); |
39507451 | 4463 | |
43010659 PW |
4464 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4465 | sd, CPU_IDLE)) | |
48f24c4d IM |
4466 | schedstat_inc(sd, alb_pushed); |
4467 | else | |
4468 | schedstat_inc(sd, alb_failed); | |
4469 | } | |
1b12bbc7 | 4470 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4471 | } |
4472 | ||
46cb4b7c SS |
4473 | #ifdef CONFIG_NO_HZ |
4474 | static struct { | |
4475 | atomic_t load_balancer; | |
7d1e6a9b | 4476 | cpumask_var_t cpu_mask; |
f711f609 | 4477 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4478 | } nohz ____cacheline_aligned = { |
4479 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4480 | }; |
4481 | ||
eea08f32 AB |
4482 | int get_nohz_load_balancer(void) |
4483 | { | |
4484 | return atomic_read(&nohz.load_balancer); | |
4485 | } | |
4486 | ||
f711f609 GS |
4487 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4488 | /** | |
4489 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4490 | * @cpu: The cpu whose lowest level of sched domain is to | |
4491 | * be returned. | |
4492 | * @flag: The flag to check for the lowest sched_domain | |
4493 | * for the given cpu. | |
4494 | * | |
4495 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4496 | */ | |
4497 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4498 | { | |
4499 | struct sched_domain *sd; | |
4500 | ||
4501 | for_each_domain(cpu, sd) | |
4502 | if (sd && (sd->flags & flag)) | |
4503 | break; | |
4504 | ||
4505 | return sd; | |
4506 | } | |
4507 | ||
4508 | /** | |
4509 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4510 | * @cpu: The cpu whose domains we're iterating over. | |
4511 | * @sd: variable holding the value of the power_savings_sd | |
4512 | * for cpu. | |
4513 | * @flag: The flag to filter the sched_domains to be iterated. | |
4514 | * | |
4515 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4516 | * set, starting from the lowest sched_domain to the highest. | |
4517 | */ | |
4518 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4519 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4520 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4521 | ||
4522 | /** | |
4523 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4524 | * @ilb_group: group to be checked for semi-idleness | |
4525 | * | |
4526 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4527 | * | |
4528 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4529 | * and atleast one non-idle CPU. This helper function checks if the given | |
4530 | * sched_group is semi-idle or not. | |
4531 | */ | |
4532 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4533 | { | |
4534 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4535 | sched_group_cpus(ilb_group)); | |
4536 | ||
4537 | /* | |
4538 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4539 | * and atleast one idle cpu. | |
4540 | */ | |
4541 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4542 | return 0; | |
4543 | ||
4544 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4545 | return 0; | |
4546 | ||
4547 | return 1; | |
4548 | } | |
4549 | /** | |
4550 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4551 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4552 | * | |
4553 | * Returns: Returns the id of the idle load balancer if it exists, | |
4554 | * Else, returns >= nr_cpu_ids. | |
4555 | * | |
4556 | * This algorithm picks the idle load balancer such that it belongs to a | |
4557 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4558 | * completely idle packages/cores just for the purpose of idle load balancing | |
4559 | * when there are other idle cpu's which are better suited for that job. | |
4560 | */ | |
4561 | static int find_new_ilb(int cpu) | |
4562 | { | |
4563 | struct sched_domain *sd; | |
4564 | struct sched_group *ilb_group; | |
4565 | ||
4566 | /* | |
4567 | * Have idle load balancer selection from semi-idle packages only | |
4568 | * when power-aware load balancing is enabled | |
4569 | */ | |
4570 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4571 | goto out_done; | |
4572 | ||
4573 | /* | |
4574 | * Optimize for the case when we have no idle CPUs or only one | |
4575 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4576 | */ | |
4577 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4578 | goto out_done; | |
4579 | ||
4580 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4581 | ilb_group = sd->groups; | |
4582 | ||
4583 | do { | |
4584 | if (is_semi_idle_group(ilb_group)) | |
4585 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4586 | ||
4587 | ilb_group = ilb_group->next; | |
4588 | ||
4589 | } while (ilb_group != sd->groups); | |
4590 | } | |
4591 | ||
4592 | out_done: | |
4593 | return cpumask_first(nohz.cpu_mask); | |
4594 | } | |
4595 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4596 | static inline int find_new_ilb(int call_cpu) | |
4597 | { | |
6e29ec57 | 4598 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4599 | } |
4600 | #endif | |
4601 | ||
7835b98b | 4602 | /* |
46cb4b7c SS |
4603 | * This routine will try to nominate the ilb (idle load balancing) |
4604 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4605 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4606 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4607 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4608 | * arrives... | |
4609 | * | |
4610 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4611 | * for idle load balancing. ilb owner will still be part of | |
4612 | * nohz.cpu_mask.. | |
7835b98b | 4613 | * |
46cb4b7c SS |
4614 | * While stopping the tick, this cpu will become the ilb owner if there |
4615 | * is no other owner. And will be the owner till that cpu becomes busy | |
4616 | * or if all cpus in the system stop their ticks at which point | |
4617 | * there is no need for ilb owner. | |
4618 | * | |
4619 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4620 | * next busy scheduler_tick() | |
4621 | */ | |
4622 | int select_nohz_load_balancer(int stop_tick) | |
4623 | { | |
4624 | int cpu = smp_processor_id(); | |
4625 | ||
4626 | if (stop_tick) { | |
46cb4b7c SS |
4627 | cpu_rq(cpu)->in_nohz_recently = 1; |
4628 | ||
483b4ee6 SS |
4629 | if (!cpu_active(cpu)) { |
4630 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4631 | return 0; | |
4632 | ||
4633 | /* | |
4634 | * If we are going offline and still the leader, | |
4635 | * give up! | |
4636 | */ | |
46cb4b7c SS |
4637 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4638 | BUG(); | |
483b4ee6 | 4639 | |
46cb4b7c SS |
4640 | return 0; |
4641 | } | |
4642 | ||
483b4ee6 SS |
4643 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4644 | ||
46cb4b7c | 4645 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4646 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4647 | if (atomic_read(&nohz.load_balancer) == cpu) |
4648 | atomic_set(&nohz.load_balancer, -1); | |
4649 | return 0; | |
4650 | } | |
4651 | ||
4652 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4653 | /* make me the ilb owner */ | |
4654 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4655 | return 1; | |
e790fb0b GS |
4656 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4657 | int new_ilb; | |
4658 | ||
4659 | if (!(sched_smt_power_savings || | |
4660 | sched_mc_power_savings)) | |
4661 | return 1; | |
4662 | /* | |
4663 | * Check to see if there is a more power-efficient | |
4664 | * ilb. | |
4665 | */ | |
4666 | new_ilb = find_new_ilb(cpu); | |
4667 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4668 | atomic_set(&nohz.load_balancer, -1); | |
4669 | resched_cpu(new_ilb); | |
4670 | return 0; | |
4671 | } | |
46cb4b7c | 4672 | return 1; |
e790fb0b | 4673 | } |
46cb4b7c | 4674 | } else { |
7d1e6a9b | 4675 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4676 | return 0; |
4677 | ||
7d1e6a9b | 4678 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4679 | |
4680 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4681 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4682 | BUG(); | |
4683 | } | |
4684 | return 0; | |
4685 | } | |
4686 | #endif | |
4687 | ||
4688 | static DEFINE_SPINLOCK(balancing); | |
4689 | ||
4690 | /* | |
7835b98b CL |
4691 | * It checks each scheduling domain to see if it is due to be balanced, |
4692 | * and initiates a balancing operation if so. | |
4693 | * | |
4694 | * Balancing parameters are set up in arch_init_sched_domains. | |
4695 | */ | |
a9957449 | 4696 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4697 | { |
46cb4b7c SS |
4698 | int balance = 1; |
4699 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4700 | unsigned long interval; |
4701 | struct sched_domain *sd; | |
46cb4b7c | 4702 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4703 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4704 | int update_next_balance = 0; |
d07355f5 | 4705 | int need_serialize; |
1da177e4 | 4706 | |
46cb4b7c | 4707 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4708 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4709 | continue; | |
4710 | ||
4711 | interval = sd->balance_interval; | |
d15bcfdb | 4712 | if (idle != CPU_IDLE) |
1da177e4 LT |
4713 | interval *= sd->busy_factor; |
4714 | ||
4715 | /* scale ms to jiffies */ | |
4716 | interval = msecs_to_jiffies(interval); | |
4717 | if (unlikely(!interval)) | |
4718 | interval = 1; | |
dd41f596 IM |
4719 | if (interval > HZ*NR_CPUS/10) |
4720 | interval = HZ*NR_CPUS/10; | |
4721 | ||
d07355f5 | 4722 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4723 | |
d07355f5 | 4724 | if (need_serialize) { |
08c183f3 CL |
4725 | if (!spin_trylock(&balancing)) |
4726 | goto out; | |
4727 | } | |
4728 | ||
c9819f45 | 4729 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4730 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4731 | /* |
4732 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4733 | * longer idle, or one of our SMT siblings is |
4734 | * not idle. | |
4735 | */ | |
d15bcfdb | 4736 | idle = CPU_NOT_IDLE; |
1da177e4 | 4737 | } |
1bd77f2d | 4738 | sd->last_balance = jiffies; |
1da177e4 | 4739 | } |
d07355f5 | 4740 | if (need_serialize) |
08c183f3 CL |
4741 | spin_unlock(&balancing); |
4742 | out: | |
f549da84 | 4743 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4744 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4745 | update_next_balance = 1; |
4746 | } | |
783609c6 SS |
4747 | |
4748 | /* | |
4749 | * Stop the load balance at this level. There is another | |
4750 | * CPU in our sched group which is doing load balancing more | |
4751 | * actively. | |
4752 | */ | |
4753 | if (!balance) | |
4754 | break; | |
1da177e4 | 4755 | } |
f549da84 SS |
4756 | |
4757 | /* | |
4758 | * next_balance will be updated only when there is a need. | |
4759 | * When the cpu is attached to null domain for ex, it will not be | |
4760 | * updated. | |
4761 | */ | |
4762 | if (likely(update_next_balance)) | |
4763 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4764 | } |
4765 | ||
4766 | /* | |
4767 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4768 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4769 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4770 | */ | |
4771 | static void run_rebalance_domains(struct softirq_action *h) | |
4772 | { | |
dd41f596 IM |
4773 | int this_cpu = smp_processor_id(); |
4774 | struct rq *this_rq = cpu_rq(this_cpu); | |
4775 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4776 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4777 | |
dd41f596 | 4778 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4779 | |
4780 | #ifdef CONFIG_NO_HZ | |
4781 | /* | |
4782 | * If this cpu is the owner for idle load balancing, then do the | |
4783 | * balancing on behalf of the other idle cpus whose ticks are | |
4784 | * stopped. | |
4785 | */ | |
dd41f596 IM |
4786 | if (this_rq->idle_at_tick && |
4787 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4788 | struct rq *rq; |
4789 | int balance_cpu; | |
4790 | ||
7d1e6a9b RR |
4791 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4792 | if (balance_cpu == this_cpu) | |
4793 | continue; | |
4794 | ||
46cb4b7c SS |
4795 | /* |
4796 | * If this cpu gets work to do, stop the load balancing | |
4797 | * work being done for other cpus. Next load | |
4798 | * balancing owner will pick it up. | |
4799 | */ | |
4800 | if (need_resched()) | |
4801 | break; | |
4802 | ||
de0cf899 | 4803 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4804 | |
4805 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4806 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4807 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4808 | } |
4809 | } | |
4810 | #endif | |
4811 | } | |
4812 | ||
8a0be9ef FW |
4813 | static inline int on_null_domain(int cpu) |
4814 | { | |
4815 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4816 | } | |
4817 | ||
46cb4b7c SS |
4818 | /* |
4819 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4820 | * | |
4821 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4822 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4823 | * if the whole system is idle. | |
4824 | */ | |
dd41f596 | 4825 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4826 | { |
46cb4b7c SS |
4827 | #ifdef CONFIG_NO_HZ |
4828 | /* | |
4829 | * If we were in the nohz mode recently and busy at the current | |
4830 | * scheduler tick, then check if we need to nominate new idle | |
4831 | * load balancer. | |
4832 | */ | |
4833 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4834 | rq->in_nohz_recently = 0; | |
4835 | ||
4836 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4837 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4838 | atomic_set(&nohz.load_balancer, -1); |
4839 | } | |
4840 | ||
4841 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4842 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4843 | |
434d53b0 | 4844 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4845 | resched_cpu(ilb); |
4846 | } | |
4847 | } | |
4848 | ||
4849 | /* | |
4850 | * If this cpu is idle and doing idle load balancing for all the | |
4851 | * cpus with ticks stopped, is it time for that to stop? | |
4852 | */ | |
4853 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4854 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4855 | resched_cpu(cpu); |
4856 | return; | |
4857 | } | |
4858 | ||
4859 | /* | |
4860 | * If this cpu is idle and the idle load balancing is done by | |
4861 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4862 | */ | |
4863 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4864 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4865 | return; |
4866 | #endif | |
8a0be9ef FW |
4867 | /* Don't need to rebalance while attached to NULL domain */ |
4868 | if (time_after_eq(jiffies, rq->next_balance) && | |
4869 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4870 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4871 | } |
dd41f596 IM |
4872 | |
4873 | #else /* CONFIG_SMP */ | |
4874 | ||
1da177e4 LT |
4875 | /* |
4876 | * on UP we do not need to balance between CPUs: | |
4877 | */ | |
70b97a7f | 4878 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4879 | { |
4880 | } | |
dd41f596 | 4881 | |
1da177e4 LT |
4882 | #endif |
4883 | ||
1da177e4 LT |
4884 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4885 | ||
4886 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4887 | ||
4888 | /* | |
c5f8d995 | 4889 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4890 | * @p in case that task is currently running. |
c5f8d995 HS |
4891 | * |
4892 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4893 | */ |
c5f8d995 HS |
4894 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4895 | { | |
4896 | u64 ns = 0; | |
4897 | ||
4898 | if (task_current(rq, p)) { | |
4899 | update_rq_clock(rq); | |
4900 | ns = rq->clock - p->se.exec_start; | |
4901 | if ((s64)ns < 0) | |
4902 | ns = 0; | |
4903 | } | |
4904 | ||
4905 | return ns; | |
4906 | } | |
4907 | ||
bb34d92f | 4908 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4909 | { |
1da177e4 | 4910 | unsigned long flags; |
41b86e9c | 4911 | struct rq *rq; |
bb34d92f | 4912 | u64 ns = 0; |
48f24c4d | 4913 | |
41b86e9c | 4914 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4915 | ns = do_task_delta_exec(p, rq); |
4916 | task_rq_unlock(rq, &flags); | |
1508487e | 4917 | |
c5f8d995 HS |
4918 | return ns; |
4919 | } | |
f06febc9 | 4920 | |
c5f8d995 HS |
4921 | /* |
4922 | * Return accounted runtime for the task. | |
4923 | * In case the task is currently running, return the runtime plus current's | |
4924 | * pending runtime that have not been accounted yet. | |
4925 | */ | |
4926 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4927 | { | |
4928 | unsigned long flags; | |
4929 | struct rq *rq; | |
4930 | u64 ns = 0; | |
4931 | ||
4932 | rq = task_rq_lock(p, &flags); | |
4933 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4934 | task_rq_unlock(rq, &flags); | |
4935 | ||
4936 | return ns; | |
4937 | } | |
48f24c4d | 4938 | |
c5f8d995 HS |
4939 | /* |
4940 | * Return sum_exec_runtime for the thread group. | |
4941 | * In case the task is currently running, return the sum plus current's | |
4942 | * pending runtime that have not been accounted yet. | |
4943 | * | |
4944 | * Note that the thread group might have other running tasks as well, | |
4945 | * so the return value not includes other pending runtime that other | |
4946 | * running tasks might have. | |
4947 | */ | |
4948 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4949 | { | |
4950 | struct task_cputime totals; | |
4951 | unsigned long flags; | |
4952 | struct rq *rq; | |
4953 | u64 ns; | |
4954 | ||
4955 | rq = task_rq_lock(p, &flags); | |
4956 | thread_group_cputime(p, &totals); | |
4957 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4958 | task_rq_unlock(rq, &flags); |
48f24c4d | 4959 | |
1da177e4 LT |
4960 | return ns; |
4961 | } | |
4962 | ||
1da177e4 LT |
4963 | /* |
4964 | * Account user cpu time to a process. | |
4965 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4966 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4967 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4968 | */ |
457533a7 MS |
4969 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4970 | cputime_t cputime_scaled) | |
1da177e4 LT |
4971 | { |
4972 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4973 | cputime64_t tmp; | |
4974 | ||
457533a7 | 4975 | /* Add user time to process. */ |
1da177e4 | 4976 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4977 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4978 | account_group_user_time(p, cputime); |
1da177e4 LT |
4979 | |
4980 | /* Add user time to cpustat. */ | |
4981 | tmp = cputime_to_cputime64(cputime); | |
4982 | if (TASK_NICE(p) > 0) | |
4983 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4984 | else | |
4985 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4986 | |
4987 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4988 | /* Account for user time used */ |
4989 | acct_update_integrals(p); | |
1da177e4 LT |
4990 | } |
4991 | ||
94886b84 LV |
4992 | /* |
4993 | * Account guest cpu time to a process. | |
4994 | * @p: the process that the cpu time gets accounted to | |
4995 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4996 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4997 | */ |
457533a7 MS |
4998 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4999 | cputime_t cputime_scaled) | |
94886b84 LV |
5000 | { |
5001 | cputime64_t tmp; | |
5002 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5003 | ||
5004 | tmp = cputime_to_cputime64(cputime); | |
5005 | ||
457533a7 | 5006 | /* Add guest time to process. */ |
94886b84 | 5007 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5008 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5009 | account_group_user_time(p, cputime); |
94886b84 LV |
5010 | p->gtime = cputime_add(p->gtime, cputime); |
5011 | ||
457533a7 | 5012 | /* Add guest time to cpustat. */ |
94886b84 LV |
5013 | cpustat->user = cputime64_add(cpustat->user, tmp); |
5014 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5015 | } | |
5016 | ||
1da177e4 LT |
5017 | /* |
5018 | * Account system cpu time to a process. | |
5019 | * @p: the process that the cpu time gets accounted to | |
5020 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5021 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5022 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5023 | */ |
5024 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5025 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5026 | { |
5027 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5028 | cputime64_t tmp; |
5029 | ||
983ed7a6 | 5030 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5031 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5032 | return; |
5033 | } | |
94886b84 | 5034 | |
457533a7 | 5035 | /* Add system time to process. */ |
1da177e4 | 5036 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5037 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5038 | account_group_system_time(p, cputime); |
1da177e4 LT |
5039 | |
5040 | /* Add system time to cpustat. */ | |
5041 | tmp = cputime_to_cputime64(cputime); | |
5042 | if (hardirq_count() - hardirq_offset) | |
5043 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5044 | else if (softirq_count()) | |
5045 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5046 | else |
79741dd3 MS |
5047 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5048 | ||
ef12fefa BR |
5049 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5050 | ||
1da177e4 LT |
5051 | /* Account for system time used */ |
5052 | acct_update_integrals(p); | |
1da177e4 LT |
5053 | } |
5054 | ||
c66f08be | 5055 | /* |
1da177e4 | 5056 | * Account for involuntary wait time. |
1da177e4 | 5057 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5058 | */ |
79741dd3 | 5059 | void account_steal_time(cputime_t cputime) |
c66f08be | 5060 | { |
79741dd3 MS |
5061 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5062 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5063 | ||
5064 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5065 | } |
5066 | ||
1da177e4 | 5067 | /* |
79741dd3 MS |
5068 | * Account for idle time. |
5069 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5070 | */ |
79741dd3 | 5071 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5072 | { |
5073 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5074 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5075 | struct rq *rq = this_rq(); |
1da177e4 | 5076 | |
79741dd3 MS |
5077 | if (atomic_read(&rq->nr_iowait) > 0) |
5078 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5079 | else | |
5080 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5081 | } |
5082 | ||
79741dd3 MS |
5083 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5084 | ||
5085 | /* | |
5086 | * Account a single tick of cpu time. | |
5087 | * @p: the process that the cpu time gets accounted to | |
5088 | * @user_tick: indicates if the tick is a user or a system tick | |
5089 | */ | |
5090 | void account_process_tick(struct task_struct *p, int user_tick) | |
5091 | { | |
5092 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5093 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5094 | struct rq *rq = this_rq(); | |
5095 | ||
5096 | if (user_tick) | |
5097 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5098 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5099 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5100 | one_jiffy_scaled); | |
5101 | else | |
5102 | account_idle_time(one_jiffy); | |
5103 | } | |
5104 | ||
5105 | /* | |
5106 | * Account multiple ticks of steal time. | |
5107 | * @p: the process from which the cpu time has been stolen | |
5108 | * @ticks: number of stolen ticks | |
5109 | */ | |
5110 | void account_steal_ticks(unsigned long ticks) | |
5111 | { | |
5112 | account_steal_time(jiffies_to_cputime(ticks)); | |
5113 | } | |
5114 | ||
5115 | /* | |
5116 | * Account multiple ticks of idle time. | |
5117 | * @ticks: number of stolen ticks | |
5118 | */ | |
5119 | void account_idle_ticks(unsigned long ticks) | |
5120 | { | |
5121 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5122 | } |
5123 | ||
79741dd3 MS |
5124 | #endif |
5125 | ||
49048622 BS |
5126 | /* |
5127 | * Use precise platform statistics if available: | |
5128 | */ | |
5129 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5130 | cputime_t task_utime(struct task_struct *p) | |
5131 | { | |
5132 | return p->utime; | |
5133 | } | |
5134 | ||
5135 | cputime_t task_stime(struct task_struct *p) | |
5136 | { | |
5137 | return p->stime; | |
5138 | } | |
5139 | #else | |
5140 | cputime_t task_utime(struct task_struct *p) | |
5141 | { | |
5142 | clock_t utime = cputime_to_clock_t(p->utime), | |
5143 | total = utime + cputime_to_clock_t(p->stime); | |
5144 | u64 temp; | |
5145 | ||
5146 | /* | |
5147 | * Use CFS's precise accounting: | |
5148 | */ | |
5149 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5150 | ||
5151 | if (total) { | |
5152 | temp *= utime; | |
5153 | do_div(temp, total); | |
5154 | } | |
5155 | utime = (clock_t)temp; | |
5156 | ||
5157 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5158 | return p->prev_utime; | |
5159 | } | |
5160 | ||
5161 | cputime_t task_stime(struct task_struct *p) | |
5162 | { | |
5163 | clock_t stime; | |
5164 | ||
5165 | /* | |
5166 | * Use CFS's precise accounting. (we subtract utime from | |
5167 | * the total, to make sure the total observed by userspace | |
5168 | * grows monotonically - apps rely on that): | |
5169 | */ | |
5170 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5171 | cputime_to_clock_t(task_utime(p)); | |
5172 | ||
5173 | if (stime >= 0) | |
5174 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5175 | ||
5176 | return p->prev_stime; | |
5177 | } | |
5178 | #endif | |
5179 | ||
5180 | inline cputime_t task_gtime(struct task_struct *p) | |
5181 | { | |
5182 | return p->gtime; | |
5183 | } | |
5184 | ||
7835b98b CL |
5185 | /* |
5186 | * This function gets called by the timer code, with HZ frequency. | |
5187 | * We call it with interrupts disabled. | |
5188 | * | |
5189 | * It also gets called by the fork code, when changing the parent's | |
5190 | * timeslices. | |
5191 | */ | |
5192 | void scheduler_tick(void) | |
5193 | { | |
7835b98b CL |
5194 | int cpu = smp_processor_id(); |
5195 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5196 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5197 | |
5198 | sched_clock_tick(); | |
dd41f596 IM |
5199 | |
5200 | spin_lock(&rq->lock); | |
3e51f33f | 5201 | update_rq_clock(rq); |
f1a438d8 | 5202 | update_cpu_load(rq); |
fa85ae24 | 5203 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5204 | spin_unlock(&rq->lock); |
7835b98b | 5205 | |
e220d2dc PZ |
5206 | perf_counter_task_tick(curr, cpu); |
5207 | ||
e418e1c2 | 5208 | #ifdef CONFIG_SMP |
dd41f596 IM |
5209 | rq->idle_at_tick = idle_cpu(cpu); |
5210 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5211 | #endif |
1da177e4 LT |
5212 | } |
5213 | ||
132380a0 | 5214 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5215 | { |
5216 | if (in_lock_functions(addr)) { | |
5217 | addr = CALLER_ADDR2; | |
5218 | if (in_lock_functions(addr)) | |
5219 | addr = CALLER_ADDR3; | |
5220 | } | |
5221 | return addr; | |
5222 | } | |
1da177e4 | 5223 | |
7e49fcce SR |
5224 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5225 | defined(CONFIG_PREEMPT_TRACER)) | |
5226 | ||
43627582 | 5227 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5228 | { |
6cd8a4bb | 5229 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5230 | /* |
5231 | * Underflow? | |
5232 | */ | |
9a11b49a IM |
5233 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5234 | return; | |
6cd8a4bb | 5235 | #endif |
1da177e4 | 5236 | preempt_count() += val; |
6cd8a4bb | 5237 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5238 | /* |
5239 | * Spinlock count overflowing soon? | |
5240 | */ | |
33859f7f MOS |
5241 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5242 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5243 | #endif |
5244 | if (preempt_count() == val) | |
5245 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5246 | } |
5247 | EXPORT_SYMBOL(add_preempt_count); | |
5248 | ||
43627582 | 5249 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5250 | { |
6cd8a4bb | 5251 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5252 | /* |
5253 | * Underflow? | |
5254 | */ | |
01e3eb82 | 5255 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5256 | return; |
1da177e4 LT |
5257 | /* |
5258 | * Is the spinlock portion underflowing? | |
5259 | */ | |
9a11b49a IM |
5260 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5261 | !(preempt_count() & PREEMPT_MASK))) | |
5262 | return; | |
6cd8a4bb | 5263 | #endif |
9a11b49a | 5264 | |
6cd8a4bb SR |
5265 | if (preempt_count() == val) |
5266 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5267 | preempt_count() -= val; |
5268 | } | |
5269 | EXPORT_SYMBOL(sub_preempt_count); | |
5270 | ||
5271 | #endif | |
5272 | ||
5273 | /* | |
dd41f596 | 5274 | * Print scheduling while atomic bug: |
1da177e4 | 5275 | */ |
dd41f596 | 5276 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5277 | { |
838225b4 SS |
5278 | struct pt_regs *regs = get_irq_regs(); |
5279 | ||
5280 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5281 | prev->comm, prev->pid, preempt_count()); | |
5282 | ||
dd41f596 | 5283 | debug_show_held_locks(prev); |
e21f5b15 | 5284 | print_modules(); |
dd41f596 IM |
5285 | if (irqs_disabled()) |
5286 | print_irqtrace_events(prev); | |
838225b4 SS |
5287 | |
5288 | if (regs) | |
5289 | show_regs(regs); | |
5290 | else | |
5291 | dump_stack(); | |
dd41f596 | 5292 | } |
1da177e4 | 5293 | |
dd41f596 IM |
5294 | /* |
5295 | * Various schedule()-time debugging checks and statistics: | |
5296 | */ | |
5297 | static inline void schedule_debug(struct task_struct *prev) | |
5298 | { | |
1da177e4 | 5299 | /* |
41a2d6cf | 5300 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5301 | * schedule() atomically, we ignore that path for now. |
5302 | * Otherwise, whine if we are scheduling when we should not be. | |
5303 | */ | |
3f33a7ce | 5304 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5305 | __schedule_bug(prev); |
5306 | ||
1da177e4 LT |
5307 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5308 | ||
2d72376b | 5309 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5310 | #ifdef CONFIG_SCHEDSTATS |
5311 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5312 | schedstat_inc(this_rq(), bkl_count); |
5313 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5314 | } |
5315 | #endif | |
dd41f596 IM |
5316 | } |
5317 | ||
df1c99d4 MG |
5318 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5319 | { | |
5320 | if (prev->state == TASK_RUNNING) { | |
5321 | u64 runtime = prev->se.sum_exec_runtime; | |
5322 | ||
5323 | runtime -= prev->se.prev_sum_exec_runtime; | |
5324 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5325 | ||
5326 | /* | |
5327 | * In order to avoid avg_overlap growing stale when we are | |
5328 | * indeed overlapping and hence not getting put to sleep, grow | |
5329 | * the avg_overlap on preemption. | |
5330 | * | |
5331 | * We use the average preemption runtime because that | |
5332 | * correlates to the amount of cache footprint a task can | |
5333 | * build up. | |
5334 | */ | |
5335 | update_avg(&prev->se.avg_overlap, runtime); | |
5336 | } | |
5337 | prev->sched_class->put_prev_task(rq, prev); | |
5338 | } | |
5339 | ||
dd41f596 IM |
5340 | /* |
5341 | * Pick up the highest-prio task: | |
5342 | */ | |
5343 | static inline struct task_struct * | |
b67802ea | 5344 | pick_next_task(struct rq *rq) |
dd41f596 | 5345 | { |
5522d5d5 | 5346 | const struct sched_class *class; |
dd41f596 | 5347 | struct task_struct *p; |
1da177e4 LT |
5348 | |
5349 | /* | |
dd41f596 IM |
5350 | * Optimization: we know that if all tasks are in |
5351 | * the fair class we can call that function directly: | |
1da177e4 | 5352 | */ |
dd41f596 | 5353 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5354 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5355 | if (likely(p)) |
5356 | return p; | |
1da177e4 LT |
5357 | } |
5358 | ||
dd41f596 IM |
5359 | class = sched_class_highest; |
5360 | for ( ; ; ) { | |
fb8d4724 | 5361 | p = class->pick_next_task(rq); |
dd41f596 IM |
5362 | if (p) |
5363 | return p; | |
5364 | /* | |
5365 | * Will never be NULL as the idle class always | |
5366 | * returns a non-NULL p: | |
5367 | */ | |
5368 | class = class->next; | |
5369 | } | |
5370 | } | |
1da177e4 | 5371 | |
dd41f596 IM |
5372 | /* |
5373 | * schedule() is the main scheduler function. | |
5374 | */ | |
ff743345 | 5375 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5376 | { |
5377 | struct task_struct *prev, *next; | |
67ca7bde | 5378 | unsigned long *switch_count; |
dd41f596 | 5379 | struct rq *rq; |
31656519 | 5380 | int cpu; |
dd41f596 | 5381 | |
ff743345 PZ |
5382 | need_resched: |
5383 | preempt_disable(); | |
dd41f596 IM |
5384 | cpu = smp_processor_id(); |
5385 | rq = cpu_rq(cpu); | |
d6714c22 | 5386 | rcu_sched_qs(cpu); |
dd41f596 IM |
5387 | prev = rq->curr; |
5388 | switch_count = &prev->nivcsw; | |
5389 | ||
5390 | release_kernel_lock(prev); | |
5391 | need_resched_nonpreemptible: | |
5392 | ||
5393 | schedule_debug(prev); | |
1da177e4 | 5394 | |
31656519 | 5395 | if (sched_feat(HRTICK)) |
f333fdc9 | 5396 | hrtick_clear(rq); |
8f4d37ec | 5397 | |
8cd162ce | 5398 | spin_lock_irq(&rq->lock); |
3e51f33f | 5399 | update_rq_clock(rq); |
1e819950 | 5400 | clear_tsk_need_resched(prev); |
1da177e4 | 5401 | |
1da177e4 | 5402 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5403 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5404 | prev->state = TASK_RUNNING; |
16882c1e | 5405 | else |
2e1cb74a | 5406 | deactivate_task(rq, prev, 1); |
dd41f596 | 5407 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5408 | } |
5409 | ||
3f029d3c | 5410 | pre_schedule(rq, prev); |
f65eda4f | 5411 | |
dd41f596 | 5412 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5413 | idle_balance(cpu, rq); |
1da177e4 | 5414 | |
df1c99d4 | 5415 | put_prev_task(rq, prev); |
b67802ea | 5416 | next = pick_next_task(rq); |
1da177e4 | 5417 | |
1da177e4 | 5418 | if (likely(prev != next)) { |
673a90a1 | 5419 | sched_info_switch(prev, next); |
564c2b21 | 5420 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5421 | |
1da177e4 LT |
5422 | rq->nr_switches++; |
5423 | rq->curr = next; | |
5424 | ++*switch_count; | |
5425 | ||
dd41f596 | 5426 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5427 | /* |
5428 | * the context switch might have flipped the stack from under | |
5429 | * us, hence refresh the local variables. | |
5430 | */ | |
5431 | cpu = smp_processor_id(); | |
5432 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5433 | } else |
5434 | spin_unlock_irq(&rq->lock); | |
5435 | ||
3f029d3c | 5436 | post_schedule(rq); |
1da177e4 | 5437 | |
8f4d37ec | 5438 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5439 | goto need_resched_nonpreemptible; |
8f4d37ec | 5440 | |
1da177e4 | 5441 | preempt_enable_no_resched(); |
ff743345 | 5442 | if (need_resched()) |
1da177e4 LT |
5443 | goto need_resched; |
5444 | } | |
1da177e4 LT |
5445 | EXPORT_SYMBOL(schedule); |
5446 | ||
0d66bf6d PZ |
5447 | #ifdef CONFIG_SMP |
5448 | /* | |
5449 | * Look out! "owner" is an entirely speculative pointer | |
5450 | * access and not reliable. | |
5451 | */ | |
5452 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5453 | { | |
5454 | unsigned int cpu; | |
5455 | struct rq *rq; | |
5456 | ||
5457 | if (!sched_feat(OWNER_SPIN)) | |
5458 | return 0; | |
5459 | ||
5460 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5461 | /* | |
5462 | * Need to access the cpu field knowing that | |
5463 | * DEBUG_PAGEALLOC could have unmapped it if | |
5464 | * the mutex owner just released it and exited. | |
5465 | */ | |
5466 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5467 | goto out; | |
5468 | #else | |
5469 | cpu = owner->cpu; | |
5470 | #endif | |
5471 | ||
5472 | /* | |
5473 | * Even if the access succeeded (likely case), | |
5474 | * the cpu field may no longer be valid. | |
5475 | */ | |
5476 | if (cpu >= nr_cpumask_bits) | |
5477 | goto out; | |
5478 | ||
5479 | /* | |
5480 | * We need to validate that we can do a | |
5481 | * get_cpu() and that we have the percpu area. | |
5482 | */ | |
5483 | if (!cpu_online(cpu)) | |
5484 | goto out; | |
5485 | ||
5486 | rq = cpu_rq(cpu); | |
5487 | ||
5488 | for (;;) { | |
5489 | /* | |
5490 | * Owner changed, break to re-assess state. | |
5491 | */ | |
5492 | if (lock->owner != owner) | |
5493 | break; | |
5494 | ||
5495 | /* | |
5496 | * Is that owner really running on that cpu? | |
5497 | */ | |
5498 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5499 | return 0; | |
5500 | ||
5501 | cpu_relax(); | |
5502 | } | |
5503 | out: | |
5504 | return 1; | |
5505 | } | |
5506 | #endif | |
5507 | ||
1da177e4 LT |
5508 | #ifdef CONFIG_PREEMPT |
5509 | /* | |
2ed6e34f | 5510 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5511 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5512 | * occur there and call schedule directly. |
5513 | */ | |
5514 | asmlinkage void __sched preempt_schedule(void) | |
5515 | { | |
5516 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5517 | |
1da177e4 LT |
5518 | /* |
5519 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5520 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5521 | */ |
beed33a8 | 5522 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5523 | return; |
5524 | ||
3a5c359a AK |
5525 | do { |
5526 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5527 | schedule(); |
3a5c359a | 5528 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5529 | |
3a5c359a AK |
5530 | /* |
5531 | * Check again in case we missed a preemption opportunity | |
5532 | * between schedule and now. | |
5533 | */ | |
5534 | barrier(); | |
5ed0cec0 | 5535 | } while (need_resched()); |
1da177e4 | 5536 | } |
1da177e4 LT |
5537 | EXPORT_SYMBOL(preempt_schedule); |
5538 | ||
5539 | /* | |
2ed6e34f | 5540 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5541 | * off of irq context. |
5542 | * Note, that this is called and return with irqs disabled. This will | |
5543 | * protect us against recursive calling from irq. | |
5544 | */ | |
5545 | asmlinkage void __sched preempt_schedule_irq(void) | |
5546 | { | |
5547 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5548 | |
2ed6e34f | 5549 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5550 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5551 | ||
3a5c359a AK |
5552 | do { |
5553 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5554 | local_irq_enable(); |
5555 | schedule(); | |
5556 | local_irq_disable(); | |
3a5c359a | 5557 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5558 | |
3a5c359a AK |
5559 | /* |
5560 | * Check again in case we missed a preemption opportunity | |
5561 | * between schedule and now. | |
5562 | */ | |
5563 | barrier(); | |
5ed0cec0 | 5564 | } while (need_resched()); |
1da177e4 LT |
5565 | } |
5566 | ||
5567 | #endif /* CONFIG_PREEMPT */ | |
5568 | ||
95cdf3b7 IM |
5569 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5570 | void *key) | |
1da177e4 | 5571 | { |
48f24c4d | 5572 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5573 | } |
1da177e4 LT |
5574 | EXPORT_SYMBOL(default_wake_function); |
5575 | ||
5576 | /* | |
41a2d6cf IM |
5577 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5578 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5579 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5580 | * | |
5581 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5582 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5583 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5584 | */ | |
78ddb08f | 5585 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5586 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5587 | { |
2e45874c | 5588 | wait_queue_t *curr, *next; |
1da177e4 | 5589 | |
2e45874c | 5590 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5591 | unsigned flags = curr->flags; |
5592 | ||
1da177e4 | 5593 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5594 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5595 | break; |
5596 | } | |
5597 | } | |
5598 | ||
5599 | /** | |
5600 | * __wake_up - wake up threads blocked on a waitqueue. | |
5601 | * @q: the waitqueue | |
5602 | * @mode: which threads | |
5603 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5604 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5605 | * |
5606 | * It may be assumed that this function implies a write memory barrier before | |
5607 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5608 | */ |
7ad5b3a5 | 5609 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5610 | int nr_exclusive, void *key) |
1da177e4 LT |
5611 | { |
5612 | unsigned long flags; | |
5613 | ||
5614 | spin_lock_irqsave(&q->lock, flags); | |
5615 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5616 | spin_unlock_irqrestore(&q->lock, flags); | |
5617 | } | |
1da177e4 LT |
5618 | EXPORT_SYMBOL(__wake_up); |
5619 | ||
5620 | /* | |
5621 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5622 | */ | |
7ad5b3a5 | 5623 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5624 | { |
5625 | __wake_up_common(q, mode, 1, 0, NULL); | |
5626 | } | |
5627 | ||
4ede816a DL |
5628 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5629 | { | |
5630 | __wake_up_common(q, mode, 1, 0, key); | |
5631 | } | |
5632 | ||
1da177e4 | 5633 | /** |
4ede816a | 5634 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5635 | * @q: the waitqueue |
5636 | * @mode: which threads | |
5637 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5638 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5639 | * |
5640 | * The sync wakeup differs that the waker knows that it will schedule | |
5641 | * away soon, so while the target thread will be woken up, it will not | |
5642 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5643 | * with each other. This can prevent needless bouncing between CPUs. | |
5644 | * | |
5645 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5646 | * |
5647 | * It may be assumed that this function implies a write memory barrier before | |
5648 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5649 | */ |
4ede816a DL |
5650 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5651 | int nr_exclusive, void *key) | |
1da177e4 LT |
5652 | { |
5653 | unsigned long flags; | |
5654 | int sync = 1; | |
5655 | ||
5656 | if (unlikely(!q)) | |
5657 | return; | |
5658 | ||
5659 | if (unlikely(!nr_exclusive)) | |
5660 | sync = 0; | |
5661 | ||
5662 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5663 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5664 | spin_unlock_irqrestore(&q->lock, flags); |
5665 | } | |
4ede816a DL |
5666 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5667 | ||
5668 | /* | |
5669 | * __wake_up_sync - see __wake_up_sync_key() | |
5670 | */ | |
5671 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5672 | { | |
5673 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5674 | } | |
1da177e4 LT |
5675 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5676 | ||
65eb3dc6 KD |
5677 | /** |
5678 | * complete: - signals a single thread waiting on this completion | |
5679 | * @x: holds the state of this particular completion | |
5680 | * | |
5681 | * This will wake up a single thread waiting on this completion. Threads will be | |
5682 | * awakened in the same order in which they were queued. | |
5683 | * | |
5684 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5685 | * |
5686 | * It may be assumed that this function implies a write memory barrier before | |
5687 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5688 | */ |
b15136e9 | 5689 | void complete(struct completion *x) |
1da177e4 LT |
5690 | { |
5691 | unsigned long flags; | |
5692 | ||
5693 | spin_lock_irqsave(&x->wait.lock, flags); | |
5694 | x->done++; | |
d9514f6c | 5695 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5696 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5697 | } | |
5698 | EXPORT_SYMBOL(complete); | |
5699 | ||
65eb3dc6 KD |
5700 | /** |
5701 | * complete_all: - signals all threads waiting on this completion | |
5702 | * @x: holds the state of this particular completion | |
5703 | * | |
5704 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5705 | * |
5706 | * It may be assumed that this function implies a write memory barrier before | |
5707 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5708 | */ |
b15136e9 | 5709 | void complete_all(struct completion *x) |
1da177e4 LT |
5710 | { |
5711 | unsigned long flags; | |
5712 | ||
5713 | spin_lock_irqsave(&x->wait.lock, flags); | |
5714 | x->done += UINT_MAX/2; | |
d9514f6c | 5715 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5716 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5717 | } | |
5718 | EXPORT_SYMBOL(complete_all); | |
5719 | ||
8cbbe86d AK |
5720 | static inline long __sched |
5721 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5722 | { |
1da177e4 LT |
5723 | if (!x->done) { |
5724 | DECLARE_WAITQUEUE(wait, current); | |
5725 | ||
5726 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5727 | __add_wait_queue_tail(&x->wait, &wait); | |
5728 | do { | |
94d3d824 | 5729 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5730 | timeout = -ERESTARTSYS; |
5731 | break; | |
8cbbe86d AK |
5732 | } |
5733 | __set_current_state(state); | |
1da177e4 LT |
5734 | spin_unlock_irq(&x->wait.lock); |
5735 | timeout = schedule_timeout(timeout); | |
5736 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5737 | } while (!x->done && timeout); |
1da177e4 | 5738 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5739 | if (!x->done) |
5740 | return timeout; | |
1da177e4 LT |
5741 | } |
5742 | x->done--; | |
ea71a546 | 5743 | return timeout ?: 1; |
1da177e4 | 5744 | } |
1da177e4 | 5745 | |
8cbbe86d AK |
5746 | static long __sched |
5747 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5748 | { |
1da177e4 LT |
5749 | might_sleep(); |
5750 | ||
5751 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5752 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5753 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5754 | return timeout; |
5755 | } | |
1da177e4 | 5756 | |
65eb3dc6 KD |
5757 | /** |
5758 | * wait_for_completion: - waits for completion of a task | |
5759 | * @x: holds the state of this particular completion | |
5760 | * | |
5761 | * This waits to be signaled for completion of a specific task. It is NOT | |
5762 | * interruptible and there is no timeout. | |
5763 | * | |
5764 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5765 | * and interrupt capability. Also see complete(). | |
5766 | */ | |
b15136e9 | 5767 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5768 | { |
5769 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5770 | } |
8cbbe86d | 5771 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5772 | |
65eb3dc6 KD |
5773 | /** |
5774 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5775 | * @x: holds the state of this particular completion | |
5776 | * @timeout: timeout value in jiffies | |
5777 | * | |
5778 | * This waits for either a completion of a specific task to be signaled or for a | |
5779 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5780 | * interruptible. | |
5781 | */ | |
b15136e9 | 5782 | unsigned long __sched |
8cbbe86d | 5783 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5784 | { |
8cbbe86d | 5785 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5786 | } |
8cbbe86d | 5787 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5788 | |
65eb3dc6 KD |
5789 | /** |
5790 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5791 | * @x: holds the state of this particular completion | |
5792 | * | |
5793 | * This waits for completion of a specific task to be signaled. It is | |
5794 | * interruptible. | |
5795 | */ | |
8cbbe86d | 5796 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5797 | { |
51e97990 AK |
5798 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5799 | if (t == -ERESTARTSYS) | |
5800 | return t; | |
5801 | return 0; | |
0fec171c | 5802 | } |
8cbbe86d | 5803 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5804 | |
65eb3dc6 KD |
5805 | /** |
5806 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5807 | * @x: holds the state of this particular completion | |
5808 | * @timeout: timeout value in jiffies | |
5809 | * | |
5810 | * This waits for either a completion of a specific task to be signaled or for a | |
5811 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5812 | */ | |
b15136e9 | 5813 | unsigned long __sched |
8cbbe86d AK |
5814 | wait_for_completion_interruptible_timeout(struct completion *x, |
5815 | unsigned long timeout) | |
0fec171c | 5816 | { |
8cbbe86d | 5817 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5818 | } |
8cbbe86d | 5819 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5820 | |
65eb3dc6 KD |
5821 | /** |
5822 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5823 | * @x: holds the state of this particular completion | |
5824 | * | |
5825 | * This waits to be signaled for completion of a specific task. It can be | |
5826 | * interrupted by a kill signal. | |
5827 | */ | |
009e577e MW |
5828 | int __sched wait_for_completion_killable(struct completion *x) |
5829 | { | |
5830 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5831 | if (t == -ERESTARTSYS) | |
5832 | return t; | |
5833 | return 0; | |
5834 | } | |
5835 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5836 | ||
be4de352 DC |
5837 | /** |
5838 | * try_wait_for_completion - try to decrement a completion without blocking | |
5839 | * @x: completion structure | |
5840 | * | |
5841 | * Returns: 0 if a decrement cannot be done without blocking | |
5842 | * 1 if a decrement succeeded. | |
5843 | * | |
5844 | * If a completion is being used as a counting completion, | |
5845 | * attempt to decrement the counter without blocking. This | |
5846 | * enables us to avoid waiting if the resource the completion | |
5847 | * is protecting is not available. | |
5848 | */ | |
5849 | bool try_wait_for_completion(struct completion *x) | |
5850 | { | |
5851 | int ret = 1; | |
5852 | ||
5853 | spin_lock_irq(&x->wait.lock); | |
5854 | if (!x->done) | |
5855 | ret = 0; | |
5856 | else | |
5857 | x->done--; | |
5858 | spin_unlock_irq(&x->wait.lock); | |
5859 | return ret; | |
5860 | } | |
5861 | EXPORT_SYMBOL(try_wait_for_completion); | |
5862 | ||
5863 | /** | |
5864 | * completion_done - Test to see if a completion has any waiters | |
5865 | * @x: completion structure | |
5866 | * | |
5867 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5868 | * 1 if there are no waiters. | |
5869 | * | |
5870 | */ | |
5871 | bool completion_done(struct completion *x) | |
5872 | { | |
5873 | int ret = 1; | |
5874 | ||
5875 | spin_lock_irq(&x->wait.lock); | |
5876 | if (!x->done) | |
5877 | ret = 0; | |
5878 | spin_unlock_irq(&x->wait.lock); | |
5879 | return ret; | |
5880 | } | |
5881 | EXPORT_SYMBOL(completion_done); | |
5882 | ||
8cbbe86d AK |
5883 | static long __sched |
5884 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5885 | { |
0fec171c IM |
5886 | unsigned long flags; |
5887 | wait_queue_t wait; | |
5888 | ||
5889 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5890 | |
8cbbe86d | 5891 | __set_current_state(state); |
1da177e4 | 5892 | |
8cbbe86d AK |
5893 | spin_lock_irqsave(&q->lock, flags); |
5894 | __add_wait_queue(q, &wait); | |
5895 | spin_unlock(&q->lock); | |
5896 | timeout = schedule_timeout(timeout); | |
5897 | spin_lock_irq(&q->lock); | |
5898 | __remove_wait_queue(q, &wait); | |
5899 | spin_unlock_irqrestore(&q->lock, flags); | |
5900 | ||
5901 | return timeout; | |
5902 | } | |
5903 | ||
5904 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5905 | { | |
5906 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5907 | } |
1da177e4 LT |
5908 | EXPORT_SYMBOL(interruptible_sleep_on); |
5909 | ||
0fec171c | 5910 | long __sched |
95cdf3b7 | 5911 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5912 | { |
8cbbe86d | 5913 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5914 | } |
1da177e4 LT |
5915 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5916 | ||
0fec171c | 5917 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5918 | { |
8cbbe86d | 5919 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5920 | } |
1da177e4 LT |
5921 | EXPORT_SYMBOL(sleep_on); |
5922 | ||
0fec171c | 5923 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5924 | { |
8cbbe86d | 5925 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5926 | } |
1da177e4 LT |
5927 | EXPORT_SYMBOL(sleep_on_timeout); |
5928 | ||
b29739f9 IM |
5929 | #ifdef CONFIG_RT_MUTEXES |
5930 | ||
5931 | /* | |
5932 | * rt_mutex_setprio - set the current priority of a task | |
5933 | * @p: task | |
5934 | * @prio: prio value (kernel-internal form) | |
5935 | * | |
5936 | * This function changes the 'effective' priority of a task. It does | |
5937 | * not touch ->normal_prio like __setscheduler(). | |
5938 | * | |
5939 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5940 | */ | |
36c8b586 | 5941 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5942 | { |
5943 | unsigned long flags; | |
83b699ed | 5944 | int oldprio, on_rq, running; |
70b97a7f | 5945 | struct rq *rq; |
cb469845 | 5946 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5947 | |
5948 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5949 | ||
5950 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5951 | update_rq_clock(rq); |
b29739f9 | 5952 | |
d5f9f942 | 5953 | oldprio = p->prio; |
dd41f596 | 5954 | on_rq = p->se.on_rq; |
051a1d1a | 5955 | running = task_current(rq, p); |
0e1f3483 | 5956 | if (on_rq) |
69be72c1 | 5957 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5958 | if (running) |
5959 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5960 | |
5961 | if (rt_prio(prio)) | |
5962 | p->sched_class = &rt_sched_class; | |
5963 | else | |
5964 | p->sched_class = &fair_sched_class; | |
5965 | ||
b29739f9 IM |
5966 | p->prio = prio; |
5967 | ||
0e1f3483 HS |
5968 | if (running) |
5969 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5970 | if (on_rq) { |
8159f87e | 5971 | enqueue_task(rq, p, 0); |
cb469845 SR |
5972 | |
5973 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5974 | } |
5975 | task_rq_unlock(rq, &flags); | |
5976 | } | |
5977 | ||
5978 | #endif | |
5979 | ||
36c8b586 | 5980 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5981 | { |
dd41f596 | 5982 | int old_prio, delta, on_rq; |
1da177e4 | 5983 | unsigned long flags; |
70b97a7f | 5984 | struct rq *rq; |
1da177e4 LT |
5985 | |
5986 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5987 | return; | |
5988 | /* | |
5989 | * We have to be careful, if called from sys_setpriority(), | |
5990 | * the task might be in the middle of scheduling on another CPU. | |
5991 | */ | |
5992 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5993 | update_rq_clock(rq); |
1da177e4 LT |
5994 | /* |
5995 | * The RT priorities are set via sched_setscheduler(), but we still | |
5996 | * allow the 'normal' nice value to be set - but as expected | |
5997 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5998 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5999 | */ |
e05606d3 | 6000 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6001 | p->static_prio = NICE_TO_PRIO(nice); |
6002 | goto out_unlock; | |
6003 | } | |
dd41f596 | 6004 | on_rq = p->se.on_rq; |
c09595f6 | 6005 | if (on_rq) |
69be72c1 | 6006 | dequeue_task(rq, p, 0); |
1da177e4 | 6007 | |
1da177e4 | 6008 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6009 | set_load_weight(p); |
b29739f9 IM |
6010 | old_prio = p->prio; |
6011 | p->prio = effective_prio(p); | |
6012 | delta = p->prio - old_prio; | |
1da177e4 | 6013 | |
dd41f596 | 6014 | if (on_rq) { |
8159f87e | 6015 | enqueue_task(rq, p, 0); |
1da177e4 | 6016 | /* |
d5f9f942 AM |
6017 | * If the task increased its priority or is running and |
6018 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6019 | */ |
d5f9f942 | 6020 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6021 | resched_task(rq->curr); |
6022 | } | |
6023 | out_unlock: | |
6024 | task_rq_unlock(rq, &flags); | |
6025 | } | |
1da177e4 LT |
6026 | EXPORT_SYMBOL(set_user_nice); |
6027 | ||
e43379f1 MM |
6028 | /* |
6029 | * can_nice - check if a task can reduce its nice value | |
6030 | * @p: task | |
6031 | * @nice: nice value | |
6032 | */ | |
36c8b586 | 6033 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6034 | { |
024f4747 MM |
6035 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6036 | int nice_rlim = 20 - nice; | |
48f24c4d | 6037 | |
e43379f1 MM |
6038 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6039 | capable(CAP_SYS_NICE)); | |
6040 | } | |
6041 | ||
1da177e4 LT |
6042 | #ifdef __ARCH_WANT_SYS_NICE |
6043 | ||
6044 | /* | |
6045 | * sys_nice - change the priority of the current process. | |
6046 | * @increment: priority increment | |
6047 | * | |
6048 | * sys_setpriority is a more generic, but much slower function that | |
6049 | * does similar things. | |
6050 | */ | |
5add95d4 | 6051 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6052 | { |
48f24c4d | 6053 | long nice, retval; |
1da177e4 LT |
6054 | |
6055 | /* | |
6056 | * Setpriority might change our priority at the same moment. | |
6057 | * We don't have to worry. Conceptually one call occurs first | |
6058 | * and we have a single winner. | |
6059 | */ | |
e43379f1 MM |
6060 | if (increment < -40) |
6061 | increment = -40; | |
1da177e4 LT |
6062 | if (increment > 40) |
6063 | increment = 40; | |
6064 | ||
2b8f836f | 6065 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6066 | if (nice < -20) |
6067 | nice = -20; | |
6068 | if (nice > 19) | |
6069 | nice = 19; | |
6070 | ||
e43379f1 MM |
6071 | if (increment < 0 && !can_nice(current, nice)) |
6072 | return -EPERM; | |
6073 | ||
1da177e4 LT |
6074 | retval = security_task_setnice(current, nice); |
6075 | if (retval) | |
6076 | return retval; | |
6077 | ||
6078 | set_user_nice(current, nice); | |
6079 | return 0; | |
6080 | } | |
6081 | ||
6082 | #endif | |
6083 | ||
6084 | /** | |
6085 | * task_prio - return the priority value of a given task. | |
6086 | * @p: the task in question. | |
6087 | * | |
6088 | * This is the priority value as seen by users in /proc. | |
6089 | * RT tasks are offset by -200. Normal tasks are centered | |
6090 | * around 0, value goes from -16 to +15. | |
6091 | */ | |
36c8b586 | 6092 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6093 | { |
6094 | return p->prio - MAX_RT_PRIO; | |
6095 | } | |
6096 | ||
6097 | /** | |
6098 | * task_nice - return the nice value of a given task. | |
6099 | * @p: the task in question. | |
6100 | */ | |
36c8b586 | 6101 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6102 | { |
6103 | return TASK_NICE(p); | |
6104 | } | |
150d8bed | 6105 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6106 | |
6107 | /** | |
6108 | * idle_cpu - is a given cpu idle currently? | |
6109 | * @cpu: the processor in question. | |
6110 | */ | |
6111 | int idle_cpu(int cpu) | |
6112 | { | |
6113 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6114 | } | |
6115 | ||
1da177e4 LT |
6116 | /** |
6117 | * idle_task - return the idle task for a given cpu. | |
6118 | * @cpu: the processor in question. | |
6119 | */ | |
36c8b586 | 6120 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6121 | { |
6122 | return cpu_rq(cpu)->idle; | |
6123 | } | |
6124 | ||
6125 | /** | |
6126 | * find_process_by_pid - find a process with a matching PID value. | |
6127 | * @pid: the pid in question. | |
6128 | */ | |
a9957449 | 6129 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6130 | { |
228ebcbe | 6131 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6132 | } |
6133 | ||
6134 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6135 | static void |
6136 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6137 | { |
dd41f596 | 6138 | BUG_ON(p->se.on_rq); |
48f24c4d | 6139 | |
1da177e4 | 6140 | p->policy = policy; |
dd41f596 IM |
6141 | switch (p->policy) { |
6142 | case SCHED_NORMAL: | |
6143 | case SCHED_BATCH: | |
6144 | case SCHED_IDLE: | |
6145 | p->sched_class = &fair_sched_class; | |
6146 | break; | |
6147 | case SCHED_FIFO: | |
6148 | case SCHED_RR: | |
6149 | p->sched_class = &rt_sched_class; | |
6150 | break; | |
6151 | } | |
6152 | ||
1da177e4 | 6153 | p->rt_priority = prio; |
b29739f9 IM |
6154 | p->normal_prio = normal_prio(p); |
6155 | /* we are holding p->pi_lock already */ | |
6156 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6157 | set_load_weight(p); |
1da177e4 LT |
6158 | } |
6159 | ||
c69e8d9c DH |
6160 | /* |
6161 | * check the target process has a UID that matches the current process's | |
6162 | */ | |
6163 | static bool check_same_owner(struct task_struct *p) | |
6164 | { | |
6165 | const struct cred *cred = current_cred(), *pcred; | |
6166 | bool match; | |
6167 | ||
6168 | rcu_read_lock(); | |
6169 | pcred = __task_cred(p); | |
6170 | match = (cred->euid == pcred->euid || | |
6171 | cred->euid == pcred->uid); | |
6172 | rcu_read_unlock(); | |
6173 | return match; | |
6174 | } | |
6175 | ||
961ccddd RR |
6176 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6177 | struct sched_param *param, bool user) | |
1da177e4 | 6178 | { |
83b699ed | 6179 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6180 | unsigned long flags; |
cb469845 | 6181 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6182 | struct rq *rq; |
ca94c442 | 6183 | int reset_on_fork; |
1da177e4 | 6184 | |
66e5393a SR |
6185 | /* may grab non-irq protected spin_locks */ |
6186 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6187 | recheck: |
6188 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6189 | if (policy < 0) { |
6190 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6191 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6192 | } else { |
6193 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6194 | policy &= ~SCHED_RESET_ON_FORK; | |
6195 | ||
6196 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6197 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6198 | policy != SCHED_IDLE) | |
6199 | return -EINVAL; | |
6200 | } | |
6201 | ||
1da177e4 LT |
6202 | /* |
6203 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6204 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6205 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6206 | */ |
6207 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6208 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6209 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6210 | return -EINVAL; |
e05606d3 | 6211 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6212 | return -EINVAL; |
6213 | ||
37e4ab3f OC |
6214 | /* |
6215 | * Allow unprivileged RT tasks to decrease priority: | |
6216 | */ | |
961ccddd | 6217 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6218 | if (rt_policy(policy)) { |
8dc3e909 | 6219 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6220 | |
6221 | if (!lock_task_sighand(p, &flags)) | |
6222 | return -ESRCH; | |
6223 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6224 | unlock_task_sighand(p, &flags); | |
6225 | ||
6226 | /* can't set/change the rt policy */ | |
6227 | if (policy != p->policy && !rlim_rtprio) | |
6228 | return -EPERM; | |
6229 | ||
6230 | /* can't increase priority */ | |
6231 | if (param->sched_priority > p->rt_priority && | |
6232 | param->sched_priority > rlim_rtprio) | |
6233 | return -EPERM; | |
6234 | } | |
dd41f596 IM |
6235 | /* |
6236 | * Like positive nice levels, dont allow tasks to | |
6237 | * move out of SCHED_IDLE either: | |
6238 | */ | |
6239 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6240 | return -EPERM; | |
5fe1d75f | 6241 | |
37e4ab3f | 6242 | /* can't change other user's priorities */ |
c69e8d9c | 6243 | if (!check_same_owner(p)) |
37e4ab3f | 6244 | return -EPERM; |
ca94c442 LP |
6245 | |
6246 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6247 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6248 | return -EPERM; | |
37e4ab3f | 6249 | } |
1da177e4 | 6250 | |
725aad24 | 6251 | if (user) { |
b68aa230 | 6252 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6253 | /* |
6254 | * Do not allow realtime tasks into groups that have no runtime | |
6255 | * assigned. | |
6256 | */ | |
9a7e0b18 PZ |
6257 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6258 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6259 | return -EPERM; |
b68aa230 PZ |
6260 | #endif |
6261 | ||
725aad24 JF |
6262 | retval = security_task_setscheduler(p, policy, param); |
6263 | if (retval) | |
6264 | return retval; | |
6265 | } | |
6266 | ||
b29739f9 IM |
6267 | /* |
6268 | * make sure no PI-waiters arrive (or leave) while we are | |
6269 | * changing the priority of the task: | |
6270 | */ | |
6271 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6272 | /* |
6273 | * To be able to change p->policy safely, the apropriate | |
6274 | * runqueue lock must be held. | |
6275 | */ | |
b29739f9 | 6276 | rq = __task_rq_lock(p); |
1da177e4 LT |
6277 | /* recheck policy now with rq lock held */ |
6278 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6279 | policy = oldpolicy = -1; | |
b29739f9 IM |
6280 | __task_rq_unlock(rq); |
6281 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6282 | goto recheck; |
6283 | } | |
2daa3577 | 6284 | update_rq_clock(rq); |
dd41f596 | 6285 | on_rq = p->se.on_rq; |
051a1d1a | 6286 | running = task_current(rq, p); |
0e1f3483 | 6287 | if (on_rq) |
2e1cb74a | 6288 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6289 | if (running) |
6290 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6291 | |
ca94c442 LP |
6292 | p->sched_reset_on_fork = reset_on_fork; |
6293 | ||
1da177e4 | 6294 | oldprio = p->prio; |
dd41f596 | 6295 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6296 | |
0e1f3483 HS |
6297 | if (running) |
6298 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6299 | if (on_rq) { |
6300 | activate_task(rq, p, 0); | |
cb469845 SR |
6301 | |
6302 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6303 | } |
b29739f9 IM |
6304 | __task_rq_unlock(rq); |
6305 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6306 | ||
95e02ca9 TG |
6307 | rt_mutex_adjust_pi(p); |
6308 | ||
1da177e4 LT |
6309 | return 0; |
6310 | } | |
961ccddd RR |
6311 | |
6312 | /** | |
6313 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6314 | * @p: the task in question. | |
6315 | * @policy: new policy. | |
6316 | * @param: structure containing the new RT priority. | |
6317 | * | |
6318 | * NOTE that the task may be already dead. | |
6319 | */ | |
6320 | int sched_setscheduler(struct task_struct *p, int policy, | |
6321 | struct sched_param *param) | |
6322 | { | |
6323 | return __sched_setscheduler(p, policy, param, true); | |
6324 | } | |
1da177e4 LT |
6325 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6326 | ||
961ccddd RR |
6327 | /** |
6328 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6329 | * @p: the task in question. | |
6330 | * @policy: new policy. | |
6331 | * @param: structure containing the new RT priority. | |
6332 | * | |
6333 | * Just like sched_setscheduler, only don't bother checking if the | |
6334 | * current context has permission. For example, this is needed in | |
6335 | * stop_machine(): we create temporary high priority worker threads, | |
6336 | * but our caller might not have that capability. | |
6337 | */ | |
6338 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6339 | struct sched_param *param) | |
6340 | { | |
6341 | return __sched_setscheduler(p, policy, param, false); | |
6342 | } | |
6343 | ||
95cdf3b7 IM |
6344 | static int |
6345 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6346 | { |
1da177e4 LT |
6347 | struct sched_param lparam; |
6348 | struct task_struct *p; | |
36c8b586 | 6349 | int retval; |
1da177e4 LT |
6350 | |
6351 | if (!param || pid < 0) | |
6352 | return -EINVAL; | |
6353 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6354 | return -EFAULT; | |
5fe1d75f ON |
6355 | |
6356 | rcu_read_lock(); | |
6357 | retval = -ESRCH; | |
1da177e4 | 6358 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6359 | if (p != NULL) |
6360 | retval = sched_setscheduler(p, policy, &lparam); | |
6361 | rcu_read_unlock(); | |
36c8b586 | 6362 | |
1da177e4 LT |
6363 | return retval; |
6364 | } | |
6365 | ||
6366 | /** | |
6367 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6368 | * @pid: the pid in question. | |
6369 | * @policy: new policy. | |
6370 | * @param: structure containing the new RT priority. | |
6371 | */ | |
5add95d4 HC |
6372 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6373 | struct sched_param __user *, param) | |
1da177e4 | 6374 | { |
c21761f1 JB |
6375 | /* negative values for policy are not valid */ |
6376 | if (policy < 0) | |
6377 | return -EINVAL; | |
6378 | ||
1da177e4 LT |
6379 | return do_sched_setscheduler(pid, policy, param); |
6380 | } | |
6381 | ||
6382 | /** | |
6383 | * sys_sched_setparam - set/change the RT priority of a thread | |
6384 | * @pid: the pid in question. | |
6385 | * @param: structure containing the new RT priority. | |
6386 | */ | |
5add95d4 | 6387 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6388 | { |
6389 | return do_sched_setscheduler(pid, -1, param); | |
6390 | } | |
6391 | ||
6392 | /** | |
6393 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6394 | * @pid: the pid in question. | |
6395 | */ | |
5add95d4 | 6396 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6397 | { |
36c8b586 | 6398 | struct task_struct *p; |
3a5c359a | 6399 | int retval; |
1da177e4 LT |
6400 | |
6401 | if (pid < 0) | |
3a5c359a | 6402 | return -EINVAL; |
1da177e4 LT |
6403 | |
6404 | retval = -ESRCH; | |
6405 | read_lock(&tasklist_lock); | |
6406 | p = find_process_by_pid(pid); | |
6407 | if (p) { | |
6408 | retval = security_task_getscheduler(p); | |
6409 | if (!retval) | |
ca94c442 LP |
6410 | retval = p->policy |
6411 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6412 | } |
6413 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6414 | return retval; |
6415 | } | |
6416 | ||
6417 | /** | |
ca94c442 | 6418 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6419 | * @pid: the pid in question. |
6420 | * @param: structure containing the RT priority. | |
6421 | */ | |
5add95d4 | 6422 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6423 | { |
6424 | struct sched_param lp; | |
36c8b586 | 6425 | struct task_struct *p; |
3a5c359a | 6426 | int retval; |
1da177e4 LT |
6427 | |
6428 | if (!param || pid < 0) | |
3a5c359a | 6429 | return -EINVAL; |
1da177e4 LT |
6430 | |
6431 | read_lock(&tasklist_lock); | |
6432 | p = find_process_by_pid(pid); | |
6433 | retval = -ESRCH; | |
6434 | if (!p) | |
6435 | goto out_unlock; | |
6436 | ||
6437 | retval = security_task_getscheduler(p); | |
6438 | if (retval) | |
6439 | goto out_unlock; | |
6440 | ||
6441 | lp.sched_priority = p->rt_priority; | |
6442 | read_unlock(&tasklist_lock); | |
6443 | ||
6444 | /* | |
6445 | * This one might sleep, we cannot do it with a spinlock held ... | |
6446 | */ | |
6447 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6448 | ||
1da177e4 LT |
6449 | return retval; |
6450 | ||
6451 | out_unlock: | |
6452 | read_unlock(&tasklist_lock); | |
6453 | return retval; | |
6454 | } | |
6455 | ||
96f874e2 | 6456 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6457 | { |
5a16f3d3 | 6458 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6459 | struct task_struct *p; |
6460 | int retval; | |
1da177e4 | 6461 | |
95402b38 | 6462 | get_online_cpus(); |
1da177e4 LT |
6463 | read_lock(&tasklist_lock); |
6464 | ||
6465 | p = find_process_by_pid(pid); | |
6466 | if (!p) { | |
6467 | read_unlock(&tasklist_lock); | |
95402b38 | 6468 | put_online_cpus(); |
1da177e4 LT |
6469 | return -ESRCH; |
6470 | } | |
6471 | ||
6472 | /* | |
6473 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6474 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6475 | * usage count and then drop tasklist_lock. |
6476 | */ | |
6477 | get_task_struct(p); | |
6478 | read_unlock(&tasklist_lock); | |
6479 | ||
5a16f3d3 RR |
6480 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6481 | retval = -ENOMEM; | |
6482 | goto out_put_task; | |
6483 | } | |
6484 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6485 | retval = -ENOMEM; | |
6486 | goto out_free_cpus_allowed; | |
6487 | } | |
1da177e4 | 6488 | retval = -EPERM; |
c69e8d9c | 6489 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6490 | goto out_unlock; |
6491 | ||
e7834f8f DQ |
6492 | retval = security_task_setscheduler(p, 0, NULL); |
6493 | if (retval) | |
6494 | goto out_unlock; | |
6495 | ||
5a16f3d3 RR |
6496 | cpuset_cpus_allowed(p, cpus_allowed); |
6497 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6498 | again: |
5a16f3d3 | 6499 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6500 | |
8707d8b8 | 6501 | if (!retval) { |
5a16f3d3 RR |
6502 | cpuset_cpus_allowed(p, cpus_allowed); |
6503 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6504 | /* |
6505 | * We must have raced with a concurrent cpuset | |
6506 | * update. Just reset the cpus_allowed to the | |
6507 | * cpuset's cpus_allowed | |
6508 | */ | |
5a16f3d3 | 6509 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6510 | goto again; |
6511 | } | |
6512 | } | |
1da177e4 | 6513 | out_unlock: |
5a16f3d3 RR |
6514 | free_cpumask_var(new_mask); |
6515 | out_free_cpus_allowed: | |
6516 | free_cpumask_var(cpus_allowed); | |
6517 | out_put_task: | |
1da177e4 | 6518 | put_task_struct(p); |
95402b38 | 6519 | put_online_cpus(); |
1da177e4 LT |
6520 | return retval; |
6521 | } | |
6522 | ||
6523 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6524 | struct cpumask *new_mask) |
1da177e4 | 6525 | { |
96f874e2 RR |
6526 | if (len < cpumask_size()) |
6527 | cpumask_clear(new_mask); | |
6528 | else if (len > cpumask_size()) | |
6529 | len = cpumask_size(); | |
6530 | ||
1da177e4 LT |
6531 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6532 | } | |
6533 | ||
6534 | /** | |
6535 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6536 | * @pid: pid of the process | |
6537 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6538 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6539 | */ | |
5add95d4 HC |
6540 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6541 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6542 | { |
5a16f3d3 | 6543 | cpumask_var_t new_mask; |
1da177e4 LT |
6544 | int retval; |
6545 | ||
5a16f3d3 RR |
6546 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6547 | return -ENOMEM; | |
1da177e4 | 6548 | |
5a16f3d3 RR |
6549 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6550 | if (retval == 0) | |
6551 | retval = sched_setaffinity(pid, new_mask); | |
6552 | free_cpumask_var(new_mask); | |
6553 | return retval; | |
1da177e4 LT |
6554 | } |
6555 | ||
96f874e2 | 6556 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6557 | { |
36c8b586 | 6558 | struct task_struct *p; |
1da177e4 | 6559 | int retval; |
1da177e4 | 6560 | |
95402b38 | 6561 | get_online_cpus(); |
1da177e4 LT |
6562 | read_lock(&tasklist_lock); |
6563 | ||
6564 | retval = -ESRCH; | |
6565 | p = find_process_by_pid(pid); | |
6566 | if (!p) | |
6567 | goto out_unlock; | |
6568 | ||
e7834f8f DQ |
6569 | retval = security_task_getscheduler(p); |
6570 | if (retval) | |
6571 | goto out_unlock; | |
6572 | ||
96f874e2 | 6573 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6574 | |
6575 | out_unlock: | |
6576 | read_unlock(&tasklist_lock); | |
95402b38 | 6577 | put_online_cpus(); |
1da177e4 | 6578 | |
9531b62f | 6579 | return retval; |
1da177e4 LT |
6580 | } |
6581 | ||
6582 | /** | |
6583 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6584 | * @pid: pid of the process | |
6585 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6586 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6587 | */ | |
5add95d4 HC |
6588 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6589 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6590 | { |
6591 | int ret; | |
f17c8607 | 6592 | cpumask_var_t mask; |
1da177e4 | 6593 | |
f17c8607 | 6594 | if (len < cpumask_size()) |
1da177e4 LT |
6595 | return -EINVAL; |
6596 | ||
f17c8607 RR |
6597 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6598 | return -ENOMEM; | |
1da177e4 | 6599 | |
f17c8607 RR |
6600 | ret = sched_getaffinity(pid, mask); |
6601 | if (ret == 0) { | |
6602 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6603 | ret = -EFAULT; | |
6604 | else | |
6605 | ret = cpumask_size(); | |
6606 | } | |
6607 | free_cpumask_var(mask); | |
1da177e4 | 6608 | |
f17c8607 | 6609 | return ret; |
1da177e4 LT |
6610 | } |
6611 | ||
6612 | /** | |
6613 | * sys_sched_yield - yield the current processor to other threads. | |
6614 | * | |
dd41f596 IM |
6615 | * This function yields the current CPU to other tasks. If there are no |
6616 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6617 | */ |
5add95d4 | 6618 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6619 | { |
70b97a7f | 6620 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6621 | |
2d72376b | 6622 | schedstat_inc(rq, yld_count); |
4530d7ab | 6623 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6624 | |
6625 | /* | |
6626 | * Since we are going to call schedule() anyway, there's | |
6627 | * no need to preempt or enable interrupts: | |
6628 | */ | |
6629 | __release(rq->lock); | |
8a25d5de | 6630 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6631 | _raw_spin_unlock(&rq->lock); |
6632 | preempt_enable_no_resched(); | |
6633 | ||
6634 | schedule(); | |
6635 | ||
6636 | return 0; | |
6637 | } | |
6638 | ||
d86ee480 PZ |
6639 | static inline int should_resched(void) |
6640 | { | |
6641 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6642 | } | |
6643 | ||
e7b38404 | 6644 | static void __cond_resched(void) |
1da177e4 | 6645 | { |
e7aaaa69 FW |
6646 | add_preempt_count(PREEMPT_ACTIVE); |
6647 | schedule(); | |
6648 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6649 | } |
6650 | ||
02b67cc3 | 6651 | int __sched _cond_resched(void) |
1da177e4 | 6652 | { |
d86ee480 | 6653 | if (should_resched()) { |
1da177e4 LT |
6654 | __cond_resched(); |
6655 | return 1; | |
6656 | } | |
6657 | return 0; | |
6658 | } | |
02b67cc3 | 6659 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6660 | |
6661 | /* | |
613afbf8 | 6662 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6663 | * call schedule, and on return reacquire the lock. |
6664 | * | |
41a2d6cf | 6665 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6666 | * operations here to prevent schedule() from being called twice (once via |
6667 | * spin_unlock(), once by hand). | |
6668 | */ | |
613afbf8 | 6669 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6670 | { |
d86ee480 | 6671 | int resched = should_resched(); |
6df3cecb JK |
6672 | int ret = 0; |
6673 | ||
f607c668 PZ |
6674 | lockdep_assert_held(lock); |
6675 | ||
95c354fe | 6676 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6677 | spin_unlock(lock); |
d86ee480 | 6678 | if (resched) |
95c354fe NP |
6679 | __cond_resched(); |
6680 | else | |
6681 | cpu_relax(); | |
6df3cecb | 6682 | ret = 1; |
1da177e4 | 6683 | spin_lock(lock); |
1da177e4 | 6684 | } |
6df3cecb | 6685 | return ret; |
1da177e4 | 6686 | } |
613afbf8 | 6687 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6688 | |
613afbf8 | 6689 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6690 | { |
6691 | BUG_ON(!in_softirq()); | |
6692 | ||
d86ee480 | 6693 | if (should_resched()) { |
98d82567 | 6694 | local_bh_enable(); |
1da177e4 LT |
6695 | __cond_resched(); |
6696 | local_bh_disable(); | |
6697 | return 1; | |
6698 | } | |
6699 | return 0; | |
6700 | } | |
613afbf8 | 6701 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6702 | |
1da177e4 LT |
6703 | /** |
6704 | * yield - yield the current processor to other threads. | |
6705 | * | |
72fd4a35 | 6706 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6707 | * thread runnable and calls sys_sched_yield(). |
6708 | */ | |
6709 | void __sched yield(void) | |
6710 | { | |
6711 | set_current_state(TASK_RUNNING); | |
6712 | sys_sched_yield(); | |
6713 | } | |
1da177e4 LT |
6714 | EXPORT_SYMBOL(yield); |
6715 | ||
6716 | /* | |
41a2d6cf | 6717 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6718 | * that process accounting knows that this is a task in IO wait state. |
6719 | * | |
6720 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6721 | * has set its backing_dev_info: the queue against which it should throttle) | |
6722 | */ | |
6723 | void __sched io_schedule(void) | |
6724 | { | |
54d35f29 | 6725 | struct rq *rq = raw_rq(); |
1da177e4 | 6726 | |
0ff92245 | 6727 | delayacct_blkio_start(); |
1da177e4 | 6728 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6729 | current->in_iowait = 1; |
1da177e4 | 6730 | schedule(); |
8f0dfc34 | 6731 | current->in_iowait = 0; |
1da177e4 | 6732 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6733 | delayacct_blkio_end(); |
1da177e4 | 6734 | } |
1da177e4 LT |
6735 | EXPORT_SYMBOL(io_schedule); |
6736 | ||
6737 | long __sched io_schedule_timeout(long timeout) | |
6738 | { | |
54d35f29 | 6739 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6740 | long ret; |
6741 | ||
0ff92245 | 6742 | delayacct_blkio_start(); |
1da177e4 | 6743 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6744 | current->in_iowait = 1; |
1da177e4 | 6745 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6746 | current->in_iowait = 0; |
1da177e4 | 6747 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6748 | delayacct_blkio_end(); |
1da177e4 LT |
6749 | return ret; |
6750 | } | |
6751 | ||
6752 | /** | |
6753 | * sys_sched_get_priority_max - return maximum RT priority. | |
6754 | * @policy: scheduling class. | |
6755 | * | |
6756 | * this syscall returns the maximum rt_priority that can be used | |
6757 | * by a given scheduling class. | |
6758 | */ | |
5add95d4 | 6759 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6760 | { |
6761 | int ret = -EINVAL; | |
6762 | ||
6763 | switch (policy) { | |
6764 | case SCHED_FIFO: | |
6765 | case SCHED_RR: | |
6766 | ret = MAX_USER_RT_PRIO-1; | |
6767 | break; | |
6768 | case SCHED_NORMAL: | |
b0a9499c | 6769 | case SCHED_BATCH: |
dd41f596 | 6770 | case SCHED_IDLE: |
1da177e4 LT |
6771 | ret = 0; |
6772 | break; | |
6773 | } | |
6774 | return ret; | |
6775 | } | |
6776 | ||
6777 | /** | |
6778 | * sys_sched_get_priority_min - return minimum RT priority. | |
6779 | * @policy: scheduling class. | |
6780 | * | |
6781 | * this syscall returns the minimum rt_priority that can be used | |
6782 | * by a given scheduling class. | |
6783 | */ | |
5add95d4 | 6784 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6785 | { |
6786 | int ret = -EINVAL; | |
6787 | ||
6788 | switch (policy) { | |
6789 | case SCHED_FIFO: | |
6790 | case SCHED_RR: | |
6791 | ret = 1; | |
6792 | break; | |
6793 | case SCHED_NORMAL: | |
b0a9499c | 6794 | case SCHED_BATCH: |
dd41f596 | 6795 | case SCHED_IDLE: |
1da177e4 LT |
6796 | ret = 0; |
6797 | } | |
6798 | return ret; | |
6799 | } | |
6800 | ||
6801 | /** | |
6802 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6803 | * @pid: pid of the process. | |
6804 | * @interval: userspace pointer to the timeslice value. | |
6805 | * | |
6806 | * this syscall writes the default timeslice value of a given process | |
6807 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6808 | */ | |
17da2bd9 | 6809 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6810 | struct timespec __user *, interval) |
1da177e4 | 6811 | { |
36c8b586 | 6812 | struct task_struct *p; |
a4ec24b4 | 6813 | unsigned int time_slice; |
3a5c359a | 6814 | int retval; |
1da177e4 | 6815 | struct timespec t; |
1da177e4 LT |
6816 | |
6817 | if (pid < 0) | |
3a5c359a | 6818 | return -EINVAL; |
1da177e4 LT |
6819 | |
6820 | retval = -ESRCH; | |
6821 | read_lock(&tasklist_lock); | |
6822 | p = find_process_by_pid(pid); | |
6823 | if (!p) | |
6824 | goto out_unlock; | |
6825 | ||
6826 | retval = security_task_getscheduler(p); | |
6827 | if (retval) | |
6828 | goto out_unlock; | |
6829 | ||
77034937 IM |
6830 | /* |
6831 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6832 | * tasks that are on an otherwise idle runqueue: | |
6833 | */ | |
6834 | time_slice = 0; | |
6835 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6836 | time_slice = DEF_TIMESLICE; |
1868f958 | 6837 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6838 | struct sched_entity *se = &p->se; |
6839 | unsigned long flags; | |
6840 | struct rq *rq; | |
6841 | ||
6842 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6843 | if (rq->cfs.load.weight) |
6844 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6845 | task_rq_unlock(rq, &flags); |
6846 | } | |
1da177e4 | 6847 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6848 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6849 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6850 | return retval; |
3a5c359a | 6851 | |
1da177e4 LT |
6852 | out_unlock: |
6853 | read_unlock(&tasklist_lock); | |
6854 | return retval; | |
6855 | } | |
6856 | ||
7c731e0a | 6857 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6858 | |
82a1fcb9 | 6859 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6860 | { |
1da177e4 | 6861 | unsigned long free = 0; |
36c8b586 | 6862 | unsigned state; |
1da177e4 | 6863 | |
1da177e4 | 6864 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6865 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6866 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6867 | #if BITS_PER_LONG == 32 |
1da177e4 | 6868 | if (state == TASK_RUNNING) |
cc4ea795 | 6869 | printk(KERN_CONT " running "); |
1da177e4 | 6870 | else |
cc4ea795 | 6871 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6872 | #else |
6873 | if (state == TASK_RUNNING) | |
cc4ea795 | 6874 | printk(KERN_CONT " running task "); |
1da177e4 | 6875 | else |
cc4ea795 | 6876 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6877 | #endif |
6878 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6879 | free = stack_not_used(p); |
1da177e4 | 6880 | #endif |
aa47b7e0 DR |
6881 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6882 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6883 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6884 | |
5fb5e6de | 6885 | show_stack(p, NULL); |
1da177e4 LT |
6886 | } |
6887 | ||
e59e2ae2 | 6888 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6889 | { |
36c8b586 | 6890 | struct task_struct *g, *p; |
1da177e4 | 6891 | |
4bd77321 IM |
6892 | #if BITS_PER_LONG == 32 |
6893 | printk(KERN_INFO | |
6894 | " task PC stack pid father\n"); | |
1da177e4 | 6895 | #else |
4bd77321 IM |
6896 | printk(KERN_INFO |
6897 | " task PC stack pid father\n"); | |
1da177e4 LT |
6898 | #endif |
6899 | read_lock(&tasklist_lock); | |
6900 | do_each_thread(g, p) { | |
6901 | /* | |
6902 | * reset the NMI-timeout, listing all files on a slow | |
6903 | * console might take alot of time: | |
6904 | */ | |
6905 | touch_nmi_watchdog(); | |
39bc89fd | 6906 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6907 | sched_show_task(p); |
1da177e4 LT |
6908 | } while_each_thread(g, p); |
6909 | ||
04c9167f JF |
6910 | touch_all_softlockup_watchdogs(); |
6911 | ||
dd41f596 IM |
6912 | #ifdef CONFIG_SCHED_DEBUG |
6913 | sysrq_sched_debug_show(); | |
6914 | #endif | |
1da177e4 | 6915 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6916 | /* |
6917 | * Only show locks if all tasks are dumped: | |
6918 | */ | |
6919 | if (state_filter == -1) | |
6920 | debug_show_all_locks(); | |
1da177e4 LT |
6921 | } |
6922 | ||
1df21055 IM |
6923 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6924 | { | |
dd41f596 | 6925 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6926 | } |
6927 | ||
f340c0d1 IM |
6928 | /** |
6929 | * init_idle - set up an idle thread for a given CPU | |
6930 | * @idle: task in question | |
6931 | * @cpu: cpu the idle task belongs to | |
6932 | * | |
6933 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6934 | * flag, to make booting more robust. | |
6935 | */ | |
5c1e1767 | 6936 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6937 | { |
70b97a7f | 6938 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6939 | unsigned long flags; |
6940 | ||
5cbd54ef IM |
6941 | spin_lock_irqsave(&rq->lock, flags); |
6942 | ||
dd41f596 IM |
6943 | __sched_fork(idle); |
6944 | idle->se.exec_start = sched_clock(); | |
6945 | ||
b29739f9 | 6946 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6947 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6948 | __set_task_cpu(idle, cpu); |
1da177e4 | 6949 | |
1da177e4 | 6950 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6951 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6952 | idle->oncpu = 1; | |
6953 | #endif | |
1da177e4 LT |
6954 | spin_unlock_irqrestore(&rq->lock, flags); |
6955 | ||
6956 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6957 | #if defined(CONFIG_PREEMPT) |
6958 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6959 | #else | |
a1261f54 | 6960 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6961 | #endif |
dd41f596 IM |
6962 | /* |
6963 | * The idle tasks have their own, simple scheduling class: | |
6964 | */ | |
6965 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6966 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6967 | } |
6968 | ||
6969 | /* | |
6970 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6971 | * indicates which cpus entered this state. This is used | |
6972 | * in the rcu update to wait only for active cpus. For system | |
6973 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6974 | * always be CPU_BITS_NONE. |
1da177e4 | 6975 | */ |
6a7b3dc3 | 6976 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6977 | |
19978ca6 IM |
6978 | /* |
6979 | * Increase the granularity value when there are more CPUs, | |
6980 | * because with more CPUs the 'effective latency' as visible | |
6981 | * to users decreases. But the relationship is not linear, | |
6982 | * so pick a second-best guess by going with the log2 of the | |
6983 | * number of CPUs. | |
6984 | * | |
6985 | * This idea comes from the SD scheduler of Con Kolivas: | |
6986 | */ | |
6987 | static inline void sched_init_granularity(void) | |
6988 | { | |
6989 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6990 | const unsigned long limit = 200000000; | |
6991 | ||
6992 | sysctl_sched_min_granularity *= factor; | |
6993 | if (sysctl_sched_min_granularity > limit) | |
6994 | sysctl_sched_min_granularity = limit; | |
6995 | ||
6996 | sysctl_sched_latency *= factor; | |
6997 | if (sysctl_sched_latency > limit) | |
6998 | sysctl_sched_latency = limit; | |
6999 | ||
7000 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
7001 | |
7002 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
7003 | } |
7004 | ||
1da177e4 LT |
7005 | #ifdef CONFIG_SMP |
7006 | /* | |
7007 | * This is how migration works: | |
7008 | * | |
70b97a7f | 7009 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7010 | * runqueue and wake up that CPU's migration thread. |
7011 | * 2) we down() the locked semaphore => thread blocks. | |
7012 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7013 | * thread off the CPU) | |
7014 | * 4) it gets the migration request and checks whether the migrated | |
7015 | * task is still in the wrong runqueue. | |
7016 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7017 | * it and puts it into the right queue. | |
7018 | * 6) migration thread up()s the semaphore. | |
7019 | * 7) we wake up and the migration is done. | |
7020 | */ | |
7021 | ||
7022 | /* | |
7023 | * Change a given task's CPU affinity. Migrate the thread to a | |
7024 | * proper CPU and schedule it away if the CPU it's executing on | |
7025 | * is removed from the allowed bitmask. | |
7026 | * | |
7027 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7028 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7029 | * call is not atomic; no spinlocks may be held. |
7030 | */ | |
96f874e2 | 7031 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7032 | { |
70b97a7f | 7033 | struct migration_req req; |
1da177e4 | 7034 | unsigned long flags; |
70b97a7f | 7035 | struct rq *rq; |
48f24c4d | 7036 | int ret = 0; |
1da177e4 LT |
7037 | |
7038 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7039 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7040 | ret = -EINVAL; |
7041 | goto out; | |
7042 | } | |
7043 | ||
9985b0ba | 7044 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7045 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7046 | ret = -EINVAL; |
7047 | goto out; | |
7048 | } | |
7049 | ||
73fe6aae | 7050 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7051 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7052 | else { |
96f874e2 RR |
7053 | cpumask_copy(&p->cpus_allowed, new_mask); |
7054 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7055 | } |
7056 | ||
1da177e4 | 7057 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7058 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7059 | goto out; |
7060 | ||
1e5ce4f4 | 7061 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 | 7062 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7063 | struct task_struct *mt = rq->migration_thread; |
7064 | ||
7065 | get_task_struct(mt); | |
1da177e4 LT |
7066 | task_rq_unlock(rq, &flags); |
7067 | wake_up_process(rq->migration_thread); | |
693525e3 | 7068 | put_task_struct(mt); |
1da177e4 LT |
7069 | wait_for_completion(&req.done); |
7070 | tlb_migrate_finish(p->mm); | |
7071 | return 0; | |
7072 | } | |
7073 | out: | |
7074 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7075 | |
1da177e4 LT |
7076 | return ret; |
7077 | } | |
cd8ba7cd | 7078 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7079 | |
7080 | /* | |
41a2d6cf | 7081 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7082 | * this because either it can't run here any more (set_cpus_allowed() |
7083 | * away from this CPU, or CPU going down), or because we're | |
7084 | * attempting to rebalance this task on exec (sched_exec). | |
7085 | * | |
7086 | * So we race with normal scheduler movements, but that's OK, as long | |
7087 | * as the task is no longer on this CPU. | |
efc30814 KK |
7088 | * |
7089 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7090 | */ |
efc30814 | 7091 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7092 | { |
70b97a7f | 7093 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7094 | int ret = 0, on_rq; |
1da177e4 | 7095 | |
e761b772 | 7096 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7097 | return ret; |
1da177e4 LT |
7098 | |
7099 | rq_src = cpu_rq(src_cpu); | |
7100 | rq_dest = cpu_rq(dest_cpu); | |
7101 | ||
7102 | double_rq_lock(rq_src, rq_dest); | |
7103 | /* Already moved. */ | |
7104 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7105 | goto done; |
1da177e4 | 7106 | /* Affinity changed (again). */ |
96f874e2 | 7107 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7108 | goto fail; |
1da177e4 | 7109 | |
dd41f596 | 7110 | on_rq = p->se.on_rq; |
6e82a3be | 7111 | if (on_rq) |
2e1cb74a | 7112 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7113 | |
1da177e4 | 7114 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7115 | if (on_rq) { |
7116 | activate_task(rq_dest, p, 0); | |
15afe09b | 7117 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7118 | } |
b1e38734 | 7119 | done: |
efc30814 | 7120 | ret = 1; |
b1e38734 | 7121 | fail: |
1da177e4 | 7122 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7123 | return ret; |
1da177e4 LT |
7124 | } |
7125 | ||
03b042bf PM |
7126 | #define RCU_MIGRATION_IDLE 0 |
7127 | #define RCU_MIGRATION_NEED_QS 1 | |
7128 | #define RCU_MIGRATION_GOT_QS 2 | |
7129 | #define RCU_MIGRATION_MUST_SYNC 3 | |
7130 | ||
1da177e4 LT |
7131 | /* |
7132 | * migration_thread - this is a highprio system thread that performs | |
7133 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7134 | * another runqueue. | |
7135 | */ | |
95cdf3b7 | 7136 | static int migration_thread(void *data) |
1da177e4 | 7137 | { |
03b042bf | 7138 | int badcpu; |
1da177e4 | 7139 | int cpu = (long)data; |
70b97a7f | 7140 | struct rq *rq; |
1da177e4 LT |
7141 | |
7142 | rq = cpu_rq(cpu); | |
7143 | BUG_ON(rq->migration_thread != current); | |
7144 | ||
7145 | set_current_state(TASK_INTERRUPTIBLE); | |
7146 | while (!kthread_should_stop()) { | |
70b97a7f | 7147 | struct migration_req *req; |
1da177e4 | 7148 | struct list_head *head; |
1da177e4 | 7149 | |
1da177e4 LT |
7150 | spin_lock_irq(&rq->lock); |
7151 | ||
7152 | if (cpu_is_offline(cpu)) { | |
7153 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7154 | break; |
1da177e4 LT |
7155 | } |
7156 | ||
7157 | if (rq->active_balance) { | |
7158 | active_load_balance(rq, cpu); | |
7159 | rq->active_balance = 0; | |
7160 | } | |
7161 | ||
7162 | head = &rq->migration_queue; | |
7163 | ||
7164 | if (list_empty(head)) { | |
7165 | spin_unlock_irq(&rq->lock); | |
7166 | schedule(); | |
7167 | set_current_state(TASK_INTERRUPTIBLE); | |
7168 | continue; | |
7169 | } | |
70b97a7f | 7170 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7171 | list_del_init(head->next); |
7172 | ||
03b042bf PM |
7173 | if (req->task != NULL) { |
7174 | spin_unlock(&rq->lock); | |
7175 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7176 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
7177 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
7178 | spin_unlock(&rq->lock); | |
7179 | } else { | |
7180 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
7181 | spin_unlock(&rq->lock); | |
7182 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); | |
7183 | } | |
674311d5 | 7184 | local_irq_enable(); |
1da177e4 LT |
7185 | |
7186 | complete(&req->done); | |
7187 | } | |
7188 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7189 | |
1da177e4 LT |
7190 | return 0; |
7191 | } | |
7192 | ||
7193 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7194 | |
7195 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7196 | { | |
7197 | int ret; | |
7198 | ||
7199 | local_irq_disable(); | |
7200 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7201 | local_irq_enable(); | |
7202 | return ret; | |
7203 | } | |
7204 | ||
054b9108 | 7205 | /* |
3a4fa0a2 | 7206 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7207 | */ |
48f24c4d | 7208 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7209 | { |
70b97a7f | 7210 | int dest_cpu; |
6ca09dfc | 7211 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7212 | |
7213 | again: | |
7214 | /* Look for allowed, online CPU in same node. */ | |
7215 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7216 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7217 | goto move; | |
7218 | ||
7219 | /* Any allowed, online CPU? */ | |
7220 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7221 | if (dest_cpu < nr_cpu_ids) | |
7222 | goto move; | |
7223 | ||
7224 | /* No more Mr. Nice Guy. */ | |
7225 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7226 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7227 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7228 | |
e76bd8d9 RR |
7229 | /* |
7230 | * Don't tell them about moving exiting tasks or | |
7231 | * kernel threads (both mm NULL), since they never | |
7232 | * leave kernel. | |
7233 | */ | |
7234 | if (p->mm && printk_ratelimit()) { | |
7235 | printk(KERN_INFO "process %d (%s) no " | |
7236 | "longer affine to cpu%d\n", | |
7237 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7238 | } |
e76bd8d9 RR |
7239 | } |
7240 | ||
7241 | move: | |
7242 | /* It can have affinity changed while we were choosing. */ | |
7243 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7244 | goto again; | |
1da177e4 LT |
7245 | } |
7246 | ||
7247 | /* | |
7248 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7249 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7250 | * for performance reasons the counter is not stricly tracking tasks to | |
7251 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7252 | * to keep the global sum constant after CPU-down: | |
7253 | */ | |
70b97a7f | 7254 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7255 | { |
1e5ce4f4 | 7256 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7257 | unsigned long flags; |
7258 | ||
7259 | local_irq_save(flags); | |
7260 | double_rq_lock(rq_src, rq_dest); | |
7261 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7262 | rq_src->nr_uninterruptible = 0; | |
7263 | double_rq_unlock(rq_src, rq_dest); | |
7264 | local_irq_restore(flags); | |
7265 | } | |
7266 | ||
7267 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7268 | static void migrate_live_tasks(int src_cpu) | |
7269 | { | |
48f24c4d | 7270 | struct task_struct *p, *t; |
1da177e4 | 7271 | |
f7b4cddc | 7272 | read_lock(&tasklist_lock); |
1da177e4 | 7273 | |
48f24c4d IM |
7274 | do_each_thread(t, p) { |
7275 | if (p == current) | |
1da177e4 LT |
7276 | continue; |
7277 | ||
48f24c4d IM |
7278 | if (task_cpu(p) == src_cpu) |
7279 | move_task_off_dead_cpu(src_cpu, p); | |
7280 | } while_each_thread(t, p); | |
1da177e4 | 7281 | |
f7b4cddc | 7282 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7283 | } |
7284 | ||
dd41f596 IM |
7285 | /* |
7286 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7287 | * It does so by boosting its priority to highest possible. |
7288 | * Used by CPU offline code. | |
1da177e4 LT |
7289 | */ |
7290 | void sched_idle_next(void) | |
7291 | { | |
48f24c4d | 7292 | int this_cpu = smp_processor_id(); |
70b97a7f | 7293 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7294 | struct task_struct *p = rq->idle; |
7295 | unsigned long flags; | |
7296 | ||
7297 | /* cpu has to be offline */ | |
48f24c4d | 7298 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7299 | |
48f24c4d IM |
7300 | /* |
7301 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7302 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7303 | */ |
7304 | spin_lock_irqsave(&rq->lock, flags); | |
7305 | ||
dd41f596 | 7306 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7307 | |
94bc9a7b DA |
7308 | update_rq_clock(rq); |
7309 | activate_task(rq, p, 0); | |
1da177e4 LT |
7310 | |
7311 | spin_unlock_irqrestore(&rq->lock, flags); | |
7312 | } | |
7313 | ||
48f24c4d IM |
7314 | /* |
7315 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7316 | * offline. |
7317 | */ | |
7318 | void idle_task_exit(void) | |
7319 | { | |
7320 | struct mm_struct *mm = current->active_mm; | |
7321 | ||
7322 | BUG_ON(cpu_online(smp_processor_id())); | |
7323 | ||
7324 | if (mm != &init_mm) | |
7325 | switch_mm(mm, &init_mm, current); | |
7326 | mmdrop(mm); | |
7327 | } | |
7328 | ||
054b9108 | 7329 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7330 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7331 | { |
70b97a7f | 7332 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7333 | |
7334 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7335 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7336 | |
7337 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7338 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7339 | |
48f24c4d | 7340 | get_task_struct(p); |
1da177e4 LT |
7341 | |
7342 | /* | |
7343 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7344 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7345 | * fine. |
7346 | */ | |
f7b4cddc | 7347 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7348 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7349 | spin_lock_irq(&rq->lock); |
1da177e4 | 7350 | |
48f24c4d | 7351 | put_task_struct(p); |
1da177e4 LT |
7352 | } |
7353 | ||
7354 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7355 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7356 | { | |
70b97a7f | 7357 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7358 | struct task_struct *next; |
48f24c4d | 7359 | |
dd41f596 IM |
7360 | for ( ; ; ) { |
7361 | if (!rq->nr_running) | |
7362 | break; | |
a8e504d2 | 7363 | update_rq_clock(rq); |
b67802ea | 7364 | next = pick_next_task(rq); |
dd41f596 IM |
7365 | if (!next) |
7366 | break; | |
79c53799 | 7367 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7368 | migrate_dead(dead_cpu, next); |
e692ab53 | 7369 | |
1da177e4 LT |
7370 | } |
7371 | } | |
dce48a84 TG |
7372 | |
7373 | /* | |
7374 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7375 | */ | |
7376 | static void calc_global_load_remove(struct rq *rq) | |
7377 | { | |
7378 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7379 | rq->calc_load_active = 0; |
dce48a84 | 7380 | } |
1da177e4 LT |
7381 | #endif /* CONFIG_HOTPLUG_CPU */ |
7382 | ||
e692ab53 NP |
7383 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7384 | ||
7385 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7386 | { |
7387 | .procname = "sched_domain", | |
c57baf1e | 7388 | .mode = 0555, |
e0361851 | 7389 | }, |
38605cae | 7390 | {0, }, |
e692ab53 NP |
7391 | }; |
7392 | ||
7393 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7394 | { |
c57baf1e | 7395 | .ctl_name = CTL_KERN, |
e0361851 | 7396 | .procname = "kernel", |
c57baf1e | 7397 | .mode = 0555, |
e0361851 AD |
7398 | .child = sd_ctl_dir, |
7399 | }, | |
38605cae | 7400 | {0, }, |
e692ab53 NP |
7401 | }; |
7402 | ||
7403 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7404 | { | |
7405 | struct ctl_table *entry = | |
5cf9f062 | 7406 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7407 | |
e692ab53 NP |
7408 | return entry; |
7409 | } | |
7410 | ||
6382bc90 MM |
7411 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7412 | { | |
cd790076 | 7413 | struct ctl_table *entry; |
6382bc90 | 7414 | |
cd790076 MM |
7415 | /* |
7416 | * In the intermediate directories, both the child directory and | |
7417 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7418 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7419 | * static strings and all have proc handlers. |
7420 | */ | |
7421 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7422 | if (entry->child) |
7423 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7424 | if (entry->proc_handler == NULL) |
7425 | kfree(entry->procname); | |
7426 | } | |
6382bc90 MM |
7427 | |
7428 | kfree(*tablep); | |
7429 | *tablep = NULL; | |
7430 | } | |
7431 | ||
e692ab53 | 7432 | static void |
e0361851 | 7433 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7434 | const char *procname, void *data, int maxlen, |
7435 | mode_t mode, proc_handler *proc_handler) | |
7436 | { | |
e692ab53 NP |
7437 | entry->procname = procname; |
7438 | entry->data = data; | |
7439 | entry->maxlen = maxlen; | |
7440 | entry->mode = mode; | |
7441 | entry->proc_handler = proc_handler; | |
7442 | } | |
7443 | ||
7444 | static struct ctl_table * | |
7445 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7446 | { | |
a5d8c348 | 7447 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7448 | |
ad1cdc1d MM |
7449 | if (table == NULL) |
7450 | return NULL; | |
7451 | ||
e0361851 | 7452 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7453 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7454 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7455 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7456 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7457 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7458 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7459 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7460 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7461 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7462 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7463 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7464 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7465 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7466 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7467 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7468 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7469 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7470 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7471 | &sd->cache_nice_tries, |
7472 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7473 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7474 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7475 | set_table_entry(&table[11], "name", sd->name, |
7476 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7477 | /* &table[12] is terminator */ | |
e692ab53 NP |
7478 | |
7479 | return table; | |
7480 | } | |
7481 | ||
9a4e7159 | 7482 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7483 | { |
7484 | struct ctl_table *entry, *table; | |
7485 | struct sched_domain *sd; | |
7486 | int domain_num = 0, i; | |
7487 | char buf[32]; | |
7488 | ||
7489 | for_each_domain(cpu, sd) | |
7490 | domain_num++; | |
7491 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7492 | if (table == NULL) |
7493 | return NULL; | |
e692ab53 NP |
7494 | |
7495 | i = 0; | |
7496 | for_each_domain(cpu, sd) { | |
7497 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7498 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7499 | entry->mode = 0555; |
e692ab53 NP |
7500 | entry->child = sd_alloc_ctl_domain_table(sd); |
7501 | entry++; | |
7502 | i++; | |
7503 | } | |
7504 | return table; | |
7505 | } | |
7506 | ||
7507 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7508 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7509 | { |
7510 | int i, cpu_num = num_online_cpus(); | |
7511 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7512 | char buf[32]; | |
7513 | ||
7378547f MM |
7514 | WARN_ON(sd_ctl_dir[0].child); |
7515 | sd_ctl_dir[0].child = entry; | |
7516 | ||
ad1cdc1d MM |
7517 | if (entry == NULL) |
7518 | return; | |
7519 | ||
97b6ea7b | 7520 | for_each_online_cpu(i) { |
e692ab53 | 7521 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7522 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7523 | entry->mode = 0555; |
e692ab53 | 7524 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7525 | entry++; |
e692ab53 | 7526 | } |
7378547f MM |
7527 | |
7528 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7529 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7530 | } | |
6382bc90 | 7531 | |
7378547f | 7532 | /* may be called multiple times per register */ |
6382bc90 MM |
7533 | static void unregister_sched_domain_sysctl(void) |
7534 | { | |
7378547f MM |
7535 | if (sd_sysctl_header) |
7536 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7537 | sd_sysctl_header = NULL; |
7378547f MM |
7538 | if (sd_ctl_dir[0].child) |
7539 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7540 | } |
e692ab53 | 7541 | #else |
6382bc90 MM |
7542 | static void register_sched_domain_sysctl(void) |
7543 | { | |
7544 | } | |
7545 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7546 | { |
7547 | } | |
7548 | #endif | |
7549 | ||
1f11eb6a GH |
7550 | static void set_rq_online(struct rq *rq) |
7551 | { | |
7552 | if (!rq->online) { | |
7553 | const struct sched_class *class; | |
7554 | ||
c6c4927b | 7555 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7556 | rq->online = 1; |
7557 | ||
7558 | for_each_class(class) { | |
7559 | if (class->rq_online) | |
7560 | class->rq_online(rq); | |
7561 | } | |
7562 | } | |
7563 | } | |
7564 | ||
7565 | static void set_rq_offline(struct rq *rq) | |
7566 | { | |
7567 | if (rq->online) { | |
7568 | const struct sched_class *class; | |
7569 | ||
7570 | for_each_class(class) { | |
7571 | if (class->rq_offline) | |
7572 | class->rq_offline(rq); | |
7573 | } | |
7574 | ||
c6c4927b | 7575 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7576 | rq->online = 0; |
7577 | } | |
7578 | } | |
7579 | ||
1da177e4 LT |
7580 | /* |
7581 | * migration_call - callback that gets triggered when a CPU is added. | |
7582 | * Here we can start up the necessary migration thread for the new CPU. | |
7583 | */ | |
48f24c4d IM |
7584 | static int __cpuinit |
7585 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7586 | { |
1da177e4 | 7587 | struct task_struct *p; |
48f24c4d | 7588 | int cpu = (long)hcpu; |
1da177e4 | 7589 | unsigned long flags; |
70b97a7f | 7590 | struct rq *rq; |
1da177e4 LT |
7591 | |
7592 | switch (action) { | |
5be9361c | 7593 | |
1da177e4 | 7594 | case CPU_UP_PREPARE: |
8bb78442 | 7595 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7596 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7597 | if (IS_ERR(p)) |
7598 | return NOTIFY_BAD; | |
1da177e4 LT |
7599 | kthread_bind(p, cpu); |
7600 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7601 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7602 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7603 | task_rq_unlock(rq, &flags); |
371cbb38 | 7604 | get_task_struct(p); |
1da177e4 | 7605 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7606 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7607 | break; |
48f24c4d | 7608 | |
1da177e4 | 7609 | case CPU_ONLINE: |
8bb78442 | 7610 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7611 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7612 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7613 | |
7614 | /* Update our root-domain */ | |
7615 | rq = cpu_rq(cpu); | |
7616 | spin_lock_irqsave(&rq->lock, flags); | |
7617 | if (rq->rd) { | |
c6c4927b | 7618 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7619 | |
7620 | set_rq_online(rq); | |
1f94ef59 GH |
7621 | } |
7622 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7623 | break; |
48f24c4d | 7624 | |
1da177e4 LT |
7625 | #ifdef CONFIG_HOTPLUG_CPU |
7626 | case CPU_UP_CANCELED: | |
8bb78442 | 7627 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7628 | if (!cpu_rq(cpu)->migration_thread) |
7629 | break; | |
41a2d6cf | 7630 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7631 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7632 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7633 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7634 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7635 | cpu_rq(cpu)->migration_thread = NULL; |
7636 | break; | |
48f24c4d | 7637 | |
1da177e4 | 7638 | case CPU_DEAD: |
8bb78442 | 7639 | case CPU_DEAD_FROZEN: |
470fd646 | 7640 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7641 | migrate_live_tasks(cpu); |
7642 | rq = cpu_rq(cpu); | |
7643 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7644 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7645 | rq->migration_thread = NULL; |
7646 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7647 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7648 | update_rq_clock(rq); |
2e1cb74a | 7649 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7650 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7651 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7652 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7653 | migrate_dead_tasks(cpu); |
d2da272a | 7654 | spin_unlock_irq(&rq->lock); |
470fd646 | 7655 | cpuset_unlock(); |
1da177e4 LT |
7656 | migrate_nr_uninterruptible(rq); |
7657 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7658 | calc_global_load_remove(rq); |
41a2d6cf IM |
7659 | /* |
7660 | * No need to migrate the tasks: it was best-effort if | |
7661 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7662 | * the requestors. | |
7663 | */ | |
1da177e4 LT |
7664 | spin_lock_irq(&rq->lock); |
7665 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7666 | struct migration_req *req; |
7667 | ||
1da177e4 | 7668 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7669 | struct migration_req, list); |
1da177e4 | 7670 | list_del_init(&req->list); |
9a2bd244 | 7671 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7672 | complete(&req->done); |
9a2bd244 | 7673 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7674 | } |
7675 | spin_unlock_irq(&rq->lock); | |
7676 | break; | |
57d885fe | 7677 | |
08f503b0 GH |
7678 | case CPU_DYING: |
7679 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7680 | /* Update our root-domain */ |
7681 | rq = cpu_rq(cpu); | |
7682 | spin_lock_irqsave(&rq->lock, flags); | |
7683 | if (rq->rd) { | |
c6c4927b | 7684 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7685 | set_rq_offline(rq); |
57d885fe GH |
7686 | } |
7687 | spin_unlock_irqrestore(&rq->lock, flags); | |
7688 | break; | |
1da177e4 LT |
7689 | #endif |
7690 | } | |
7691 | return NOTIFY_OK; | |
7692 | } | |
7693 | ||
f38b0820 PM |
7694 | /* |
7695 | * Register at high priority so that task migration (migrate_all_tasks) | |
7696 | * happens before everything else. This has to be lower priority than | |
7697 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7698 | */ |
26c2143b | 7699 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7700 | .notifier_call = migration_call, |
7701 | .priority = 10 | |
7702 | }; | |
7703 | ||
7babe8db | 7704 | static int __init migration_init(void) |
1da177e4 LT |
7705 | { |
7706 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7707 | int err; |
48f24c4d IM |
7708 | |
7709 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7710 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7711 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7712 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7713 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7714 | |
a004cd42 | 7715 | return 0; |
1da177e4 | 7716 | } |
7babe8db | 7717 | early_initcall(migration_init); |
1da177e4 LT |
7718 | #endif |
7719 | ||
7720 | #ifdef CONFIG_SMP | |
476f3534 | 7721 | |
3e9830dc | 7722 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7723 | |
7c16ec58 | 7724 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7725 | struct cpumask *groupmask) |
1da177e4 | 7726 | { |
4dcf6aff | 7727 | struct sched_group *group = sd->groups; |
434d53b0 | 7728 | char str[256]; |
1da177e4 | 7729 | |
968ea6d8 | 7730 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7731 | cpumask_clear(groupmask); |
4dcf6aff IM |
7732 | |
7733 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7734 | ||
7735 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7736 | printk("does not load-balance\n"); | |
7737 | if (sd->parent) | |
7738 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7739 | " has parent"); | |
7740 | return -1; | |
41c7ce9a NP |
7741 | } |
7742 | ||
eefd796a | 7743 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7744 | |
758b2cdc | 7745 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7746 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7747 | "CPU%d\n", cpu); | |
7748 | } | |
758b2cdc | 7749 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7750 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7751 | " CPU%d\n", cpu); | |
7752 | } | |
1da177e4 | 7753 | |
4dcf6aff | 7754 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7755 | do { |
4dcf6aff IM |
7756 | if (!group) { |
7757 | printk("\n"); | |
7758 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7759 | break; |
7760 | } | |
7761 | ||
18a3885f | 7762 | if (!group->cpu_power) { |
4dcf6aff IM |
7763 | printk(KERN_CONT "\n"); |
7764 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7765 | "set\n"); | |
7766 | break; | |
7767 | } | |
1da177e4 | 7768 | |
758b2cdc | 7769 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7770 | printk(KERN_CONT "\n"); |
7771 | printk(KERN_ERR "ERROR: empty group\n"); | |
7772 | break; | |
7773 | } | |
1da177e4 | 7774 | |
758b2cdc | 7775 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7776 | printk(KERN_CONT "\n"); |
7777 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7778 | break; | |
7779 | } | |
1da177e4 | 7780 | |
758b2cdc | 7781 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7782 | |
968ea6d8 | 7783 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7784 | |
7785 | printk(KERN_CONT " %s", str); | |
18a3885f PZ |
7786 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
7787 | printk(KERN_CONT " (cpu_power = %d)", | |
7788 | group->cpu_power); | |
381512cf | 7789 | } |
1da177e4 | 7790 | |
4dcf6aff IM |
7791 | group = group->next; |
7792 | } while (group != sd->groups); | |
7793 | printk(KERN_CONT "\n"); | |
1da177e4 | 7794 | |
758b2cdc | 7795 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7796 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7797 | |
758b2cdc RR |
7798 | if (sd->parent && |
7799 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7800 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7801 | "of domain->span\n"); | |
7802 | return 0; | |
7803 | } | |
1da177e4 | 7804 | |
4dcf6aff IM |
7805 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7806 | { | |
d5dd3db1 | 7807 | cpumask_var_t groupmask; |
4dcf6aff | 7808 | int level = 0; |
1da177e4 | 7809 | |
4dcf6aff IM |
7810 | if (!sd) { |
7811 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7812 | return; | |
7813 | } | |
1da177e4 | 7814 | |
4dcf6aff IM |
7815 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7816 | ||
d5dd3db1 | 7817 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7818 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7819 | return; | |
7820 | } | |
7821 | ||
4dcf6aff | 7822 | for (;;) { |
7c16ec58 | 7823 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7824 | break; |
1da177e4 LT |
7825 | level++; |
7826 | sd = sd->parent; | |
33859f7f | 7827 | if (!sd) |
4dcf6aff IM |
7828 | break; |
7829 | } | |
d5dd3db1 | 7830 | free_cpumask_var(groupmask); |
1da177e4 | 7831 | } |
6d6bc0ad | 7832 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7833 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7834 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7835 | |
1a20ff27 | 7836 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7837 | { |
758b2cdc | 7838 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7839 | return 1; |
7840 | ||
7841 | /* Following flags need at least 2 groups */ | |
7842 | if (sd->flags & (SD_LOAD_BALANCE | | |
7843 | SD_BALANCE_NEWIDLE | | |
7844 | SD_BALANCE_FORK | | |
89c4710e SS |
7845 | SD_BALANCE_EXEC | |
7846 | SD_SHARE_CPUPOWER | | |
7847 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7848 | if (sd->groups != sd->groups->next) |
7849 | return 0; | |
7850 | } | |
7851 | ||
7852 | /* Following flags don't use groups */ | |
7853 | if (sd->flags & (SD_WAKE_IDLE | | |
7854 | SD_WAKE_AFFINE | | |
7855 | SD_WAKE_BALANCE)) | |
7856 | return 0; | |
7857 | ||
7858 | return 1; | |
7859 | } | |
7860 | ||
48f24c4d IM |
7861 | static int |
7862 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7863 | { |
7864 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7865 | ||
7866 | if (sd_degenerate(parent)) | |
7867 | return 1; | |
7868 | ||
758b2cdc | 7869 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7870 | return 0; |
7871 | ||
7872 | /* Does parent contain flags not in child? */ | |
7873 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7874 | if (cflags & SD_WAKE_AFFINE) | |
7875 | pflags &= ~SD_WAKE_BALANCE; | |
7876 | /* Flags needing groups don't count if only 1 group in parent */ | |
7877 | if (parent->groups == parent->groups->next) { | |
7878 | pflags &= ~(SD_LOAD_BALANCE | | |
7879 | SD_BALANCE_NEWIDLE | | |
7880 | SD_BALANCE_FORK | | |
89c4710e SS |
7881 | SD_BALANCE_EXEC | |
7882 | SD_SHARE_CPUPOWER | | |
7883 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7884 | if (nr_node_ids == 1) |
7885 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7886 | } |
7887 | if (~cflags & pflags) | |
7888 | return 0; | |
7889 | ||
7890 | return 1; | |
7891 | } | |
7892 | ||
c6c4927b RR |
7893 | static void free_rootdomain(struct root_domain *rd) |
7894 | { | |
68e74568 RR |
7895 | cpupri_cleanup(&rd->cpupri); |
7896 | ||
c6c4927b RR |
7897 | free_cpumask_var(rd->rto_mask); |
7898 | free_cpumask_var(rd->online); | |
7899 | free_cpumask_var(rd->span); | |
7900 | kfree(rd); | |
7901 | } | |
7902 | ||
57d885fe GH |
7903 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7904 | { | |
a0490fa3 | 7905 | struct root_domain *old_rd = NULL; |
57d885fe | 7906 | unsigned long flags; |
57d885fe GH |
7907 | |
7908 | spin_lock_irqsave(&rq->lock, flags); | |
7909 | ||
7910 | if (rq->rd) { | |
a0490fa3 | 7911 | old_rd = rq->rd; |
57d885fe | 7912 | |
c6c4927b | 7913 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7914 | set_rq_offline(rq); |
57d885fe | 7915 | |
c6c4927b | 7916 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7917 | |
a0490fa3 IM |
7918 | /* |
7919 | * If we dont want to free the old_rt yet then | |
7920 | * set old_rd to NULL to skip the freeing later | |
7921 | * in this function: | |
7922 | */ | |
7923 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7924 | old_rd = NULL; | |
57d885fe GH |
7925 | } |
7926 | ||
7927 | atomic_inc(&rd->refcount); | |
7928 | rq->rd = rd; | |
7929 | ||
c6c4927b | 7930 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 7931 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 7932 | set_rq_online(rq); |
57d885fe GH |
7933 | |
7934 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7935 | |
7936 | if (old_rd) | |
7937 | free_rootdomain(old_rd); | |
57d885fe GH |
7938 | } |
7939 | ||
fd5e1b5d | 7940 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7941 | { |
36b7b6d4 PE |
7942 | gfp_t gfp = GFP_KERNEL; |
7943 | ||
57d885fe GH |
7944 | memset(rd, 0, sizeof(*rd)); |
7945 | ||
36b7b6d4 PE |
7946 | if (bootmem) |
7947 | gfp = GFP_NOWAIT; | |
c6c4927b | 7948 | |
36b7b6d4 | 7949 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7950 | goto out; |
36b7b6d4 | 7951 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7952 | goto free_span; |
36b7b6d4 | 7953 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7954 | goto free_online; |
6e0534f2 | 7955 | |
0fb53029 | 7956 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7957 | goto free_rto_mask; |
c6c4927b | 7958 | return 0; |
6e0534f2 | 7959 | |
68e74568 RR |
7960 | free_rto_mask: |
7961 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7962 | free_online: |
7963 | free_cpumask_var(rd->online); | |
7964 | free_span: | |
7965 | free_cpumask_var(rd->span); | |
0c910d28 | 7966 | out: |
c6c4927b | 7967 | return -ENOMEM; |
57d885fe GH |
7968 | } |
7969 | ||
7970 | static void init_defrootdomain(void) | |
7971 | { | |
c6c4927b RR |
7972 | init_rootdomain(&def_root_domain, true); |
7973 | ||
57d885fe GH |
7974 | atomic_set(&def_root_domain.refcount, 1); |
7975 | } | |
7976 | ||
dc938520 | 7977 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7978 | { |
7979 | struct root_domain *rd; | |
7980 | ||
7981 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7982 | if (!rd) | |
7983 | return NULL; | |
7984 | ||
c6c4927b RR |
7985 | if (init_rootdomain(rd, false) != 0) { |
7986 | kfree(rd); | |
7987 | return NULL; | |
7988 | } | |
57d885fe GH |
7989 | |
7990 | return rd; | |
7991 | } | |
7992 | ||
1da177e4 | 7993 | /* |
0eab9146 | 7994 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7995 | * hold the hotplug lock. |
7996 | */ | |
0eab9146 IM |
7997 | static void |
7998 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7999 | { |
70b97a7f | 8000 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8001 | struct sched_domain *tmp; |
8002 | ||
8003 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8004 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8005 | struct sched_domain *parent = tmp->parent; |
8006 | if (!parent) | |
8007 | break; | |
f29c9b1c | 8008 | |
1a848870 | 8009 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8010 | tmp->parent = parent->parent; |
1a848870 SS |
8011 | if (parent->parent) |
8012 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8013 | } else |
8014 | tmp = tmp->parent; | |
245af2c7 SS |
8015 | } |
8016 | ||
1a848870 | 8017 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8018 | sd = sd->parent; |
1a848870 SS |
8019 | if (sd) |
8020 | sd->child = NULL; | |
8021 | } | |
1da177e4 LT |
8022 | |
8023 | sched_domain_debug(sd, cpu); | |
8024 | ||
57d885fe | 8025 | rq_attach_root(rq, rd); |
674311d5 | 8026 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8027 | } |
8028 | ||
8029 | /* cpus with isolated domains */ | |
dcc30a35 | 8030 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8031 | |
8032 | /* Setup the mask of cpus configured for isolated domains */ | |
8033 | static int __init isolated_cpu_setup(char *str) | |
8034 | { | |
968ea6d8 | 8035 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8036 | return 1; |
8037 | } | |
8038 | ||
8927f494 | 8039 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8040 | |
8041 | /* | |
6711cab4 SS |
8042 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8043 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8044 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8045 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8046 | * |
8047 | * init_sched_build_groups will build a circular linked list of the groups | |
8048 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8049 | * and ->cpu_power to 0. | |
8050 | */ | |
a616058b | 8051 | static void |
96f874e2 RR |
8052 | init_sched_build_groups(const struct cpumask *span, |
8053 | const struct cpumask *cpu_map, | |
8054 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8055 | struct sched_group **sg, |
96f874e2 RR |
8056 | struct cpumask *tmpmask), |
8057 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8058 | { |
8059 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8060 | int i; |
8061 | ||
96f874e2 | 8062 | cpumask_clear(covered); |
7c16ec58 | 8063 | |
abcd083a | 8064 | for_each_cpu(i, span) { |
6711cab4 | 8065 | struct sched_group *sg; |
7c16ec58 | 8066 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8067 | int j; |
8068 | ||
758b2cdc | 8069 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8070 | continue; |
8071 | ||
758b2cdc | 8072 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 8073 | sg->cpu_power = 0; |
1da177e4 | 8074 | |
abcd083a | 8075 | for_each_cpu(j, span) { |
7c16ec58 | 8076 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8077 | continue; |
8078 | ||
96f874e2 | 8079 | cpumask_set_cpu(j, covered); |
758b2cdc | 8080 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8081 | } |
8082 | if (!first) | |
8083 | first = sg; | |
8084 | if (last) | |
8085 | last->next = sg; | |
8086 | last = sg; | |
8087 | } | |
8088 | last->next = first; | |
8089 | } | |
8090 | ||
9c1cfda2 | 8091 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8092 | |
9c1cfda2 | 8093 | #ifdef CONFIG_NUMA |
198e2f18 | 8094 | |
9c1cfda2 JH |
8095 | /** |
8096 | * find_next_best_node - find the next node to include in a sched_domain | |
8097 | * @node: node whose sched_domain we're building | |
8098 | * @used_nodes: nodes already in the sched_domain | |
8099 | * | |
41a2d6cf | 8100 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8101 | * finds the closest node not already in the @used_nodes map. |
8102 | * | |
8103 | * Should use nodemask_t. | |
8104 | */ | |
c5f59f08 | 8105 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8106 | { |
8107 | int i, n, val, min_val, best_node = 0; | |
8108 | ||
8109 | min_val = INT_MAX; | |
8110 | ||
076ac2af | 8111 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8112 | /* Start at @node */ |
076ac2af | 8113 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8114 | |
8115 | if (!nr_cpus_node(n)) | |
8116 | continue; | |
8117 | ||
8118 | /* Skip already used nodes */ | |
c5f59f08 | 8119 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8120 | continue; |
8121 | ||
8122 | /* Simple min distance search */ | |
8123 | val = node_distance(node, n); | |
8124 | ||
8125 | if (val < min_val) { | |
8126 | min_val = val; | |
8127 | best_node = n; | |
8128 | } | |
8129 | } | |
8130 | ||
c5f59f08 | 8131 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8132 | return best_node; |
8133 | } | |
8134 | ||
8135 | /** | |
8136 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8137 | * @node: node whose cpumask we're constructing | |
73486722 | 8138 | * @span: resulting cpumask |
9c1cfda2 | 8139 | * |
41a2d6cf | 8140 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8141 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8142 | * out optimally. | |
8143 | */ | |
96f874e2 | 8144 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8145 | { |
c5f59f08 | 8146 | nodemask_t used_nodes; |
48f24c4d | 8147 | int i; |
9c1cfda2 | 8148 | |
6ca09dfc | 8149 | cpumask_clear(span); |
c5f59f08 | 8150 | nodes_clear(used_nodes); |
9c1cfda2 | 8151 | |
6ca09dfc | 8152 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8153 | node_set(node, used_nodes); |
9c1cfda2 JH |
8154 | |
8155 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8156 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8157 | |
6ca09dfc | 8158 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8159 | } |
9c1cfda2 | 8160 | } |
6d6bc0ad | 8161 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8162 | |
5c45bf27 | 8163 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8164 | |
6c99e9ad RR |
8165 | /* |
8166 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8167 | * |
8168 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8169 | * and struct sched_domain. ) | |
6c99e9ad RR |
8170 | */ |
8171 | struct static_sched_group { | |
8172 | struct sched_group sg; | |
8173 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8174 | }; | |
8175 | ||
8176 | struct static_sched_domain { | |
8177 | struct sched_domain sd; | |
8178 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8179 | }; | |
8180 | ||
49a02c51 AH |
8181 | struct s_data { |
8182 | #ifdef CONFIG_NUMA | |
8183 | int sd_allnodes; | |
8184 | cpumask_var_t domainspan; | |
8185 | cpumask_var_t covered; | |
8186 | cpumask_var_t notcovered; | |
8187 | #endif | |
8188 | cpumask_var_t nodemask; | |
8189 | cpumask_var_t this_sibling_map; | |
8190 | cpumask_var_t this_core_map; | |
8191 | cpumask_var_t send_covered; | |
8192 | cpumask_var_t tmpmask; | |
8193 | struct sched_group **sched_group_nodes; | |
8194 | struct root_domain *rd; | |
8195 | }; | |
8196 | ||
2109b99e AH |
8197 | enum s_alloc { |
8198 | sa_sched_groups = 0, | |
8199 | sa_rootdomain, | |
8200 | sa_tmpmask, | |
8201 | sa_send_covered, | |
8202 | sa_this_core_map, | |
8203 | sa_this_sibling_map, | |
8204 | sa_nodemask, | |
8205 | sa_sched_group_nodes, | |
8206 | #ifdef CONFIG_NUMA | |
8207 | sa_notcovered, | |
8208 | sa_covered, | |
8209 | sa_domainspan, | |
8210 | #endif | |
8211 | sa_none, | |
8212 | }; | |
8213 | ||
9c1cfda2 | 8214 | /* |
48f24c4d | 8215 | * SMT sched-domains: |
9c1cfda2 | 8216 | */ |
1da177e4 | 8217 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8218 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8219 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8220 | |
41a2d6cf | 8221 | static int |
96f874e2 RR |
8222 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8223 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8224 | { |
6711cab4 | 8225 | if (sg) |
6c99e9ad | 8226 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8227 | return cpu; |
8228 | } | |
6d6bc0ad | 8229 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8230 | |
48f24c4d IM |
8231 | /* |
8232 | * multi-core sched-domains: | |
8233 | */ | |
1e9f28fa | 8234 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8235 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8236 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8237 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8238 | |
8239 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8240 | static int |
96f874e2 RR |
8241 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8242 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8243 | { |
6711cab4 | 8244 | int group; |
7c16ec58 | 8245 | |
c69fc56d | 8246 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8247 | group = cpumask_first(mask); |
6711cab4 | 8248 | if (sg) |
6c99e9ad | 8249 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8250 | return group; |
1e9f28fa SS |
8251 | } |
8252 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8253 | static int |
96f874e2 RR |
8254 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8255 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8256 | { |
6711cab4 | 8257 | if (sg) |
6c99e9ad | 8258 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8259 | return cpu; |
8260 | } | |
8261 | #endif | |
8262 | ||
6c99e9ad RR |
8263 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8264 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8265 | |
41a2d6cf | 8266 | static int |
96f874e2 RR |
8267 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8268 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8269 | { |
6711cab4 | 8270 | int group; |
48f24c4d | 8271 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8272 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8273 | group = cpumask_first(mask); |
1e9f28fa | 8274 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8275 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8276 | group = cpumask_first(mask); |
1da177e4 | 8277 | #else |
6711cab4 | 8278 | group = cpu; |
1da177e4 | 8279 | #endif |
6711cab4 | 8280 | if (sg) |
6c99e9ad | 8281 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8282 | return group; |
1da177e4 LT |
8283 | } |
8284 | ||
8285 | #ifdef CONFIG_NUMA | |
1da177e4 | 8286 | /* |
9c1cfda2 JH |
8287 | * The init_sched_build_groups can't handle what we want to do with node |
8288 | * groups, so roll our own. Now each node has its own list of groups which | |
8289 | * gets dynamically allocated. | |
1da177e4 | 8290 | */ |
62ea9ceb | 8291 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8292 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8293 | |
62ea9ceb | 8294 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8295 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8296 | |
96f874e2 RR |
8297 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8298 | struct sched_group **sg, | |
8299 | struct cpumask *nodemask) | |
9c1cfda2 | 8300 | { |
6711cab4 SS |
8301 | int group; |
8302 | ||
6ca09dfc | 8303 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8304 | group = cpumask_first(nodemask); |
6711cab4 SS |
8305 | |
8306 | if (sg) | |
6c99e9ad | 8307 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8308 | return group; |
1da177e4 | 8309 | } |
6711cab4 | 8310 | |
08069033 SS |
8311 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8312 | { | |
8313 | struct sched_group *sg = group_head; | |
8314 | int j; | |
8315 | ||
8316 | if (!sg) | |
8317 | return; | |
3a5c359a | 8318 | do { |
758b2cdc | 8319 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8320 | struct sched_domain *sd; |
08069033 | 8321 | |
6c99e9ad | 8322 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8323 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8324 | /* |
8325 | * Only add "power" once for each | |
8326 | * physical package. | |
8327 | */ | |
8328 | continue; | |
8329 | } | |
08069033 | 8330 | |
18a3885f | 8331 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
8332 | } |
8333 | sg = sg->next; | |
8334 | } while (sg != group_head); | |
08069033 | 8335 | } |
0601a88d AH |
8336 | |
8337 | static int build_numa_sched_groups(struct s_data *d, | |
8338 | const struct cpumask *cpu_map, int num) | |
8339 | { | |
8340 | struct sched_domain *sd; | |
8341 | struct sched_group *sg, *prev; | |
8342 | int n, j; | |
8343 | ||
8344 | cpumask_clear(d->covered); | |
8345 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8346 | if (cpumask_empty(d->nodemask)) { | |
8347 | d->sched_group_nodes[num] = NULL; | |
8348 | goto out; | |
8349 | } | |
8350 | ||
8351 | sched_domain_node_span(num, d->domainspan); | |
8352 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8353 | ||
8354 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8355 | GFP_KERNEL, num); | |
8356 | if (!sg) { | |
8357 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", | |
8358 | num); | |
8359 | return -ENOMEM; | |
8360 | } | |
8361 | d->sched_group_nodes[num] = sg; | |
8362 | ||
8363 | for_each_cpu(j, d->nodemask) { | |
8364 | sd = &per_cpu(node_domains, j).sd; | |
8365 | sd->groups = sg; | |
8366 | } | |
8367 | ||
18a3885f | 8368 | sg->cpu_power = 0; |
0601a88d AH |
8369 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
8370 | sg->next = sg; | |
8371 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8372 | ||
8373 | prev = sg; | |
8374 | for (j = 0; j < nr_node_ids; j++) { | |
8375 | n = (num + j) % nr_node_ids; | |
8376 | cpumask_complement(d->notcovered, d->covered); | |
8377 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8378 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8379 | if (cpumask_empty(d->tmpmask)) | |
8380 | break; | |
8381 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8382 | if (cpumask_empty(d->tmpmask)) | |
8383 | continue; | |
8384 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8385 | GFP_KERNEL, num); | |
8386 | if (!sg) { | |
8387 | printk(KERN_WARNING | |
8388 | "Can not alloc domain group for node %d\n", j); | |
8389 | return -ENOMEM; | |
8390 | } | |
18a3885f | 8391 | sg->cpu_power = 0; |
0601a88d AH |
8392 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
8393 | sg->next = prev->next; | |
8394 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8395 | prev->next = sg; | |
8396 | prev = sg; | |
8397 | } | |
8398 | out: | |
8399 | return 0; | |
8400 | } | |
6d6bc0ad | 8401 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8402 | |
a616058b | 8403 | #ifdef CONFIG_NUMA |
51888ca2 | 8404 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8405 | static void free_sched_groups(const struct cpumask *cpu_map, |
8406 | struct cpumask *nodemask) | |
51888ca2 | 8407 | { |
a616058b | 8408 | int cpu, i; |
51888ca2 | 8409 | |
abcd083a | 8410 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8411 | struct sched_group **sched_group_nodes |
8412 | = sched_group_nodes_bycpu[cpu]; | |
8413 | ||
51888ca2 SV |
8414 | if (!sched_group_nodes) |
8415 | continue; | |
8416 | ||
076ac2af | 8417 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8418 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8419 | ||
6ca09dfc | 8420 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8421 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8422 | continue; |
8423 | ||
8424 | if (sg == NULL) | |
8425 | continue; | |
8426 | sg = sg->next; | |
8427 | next_sg: | |
8428 | oldsg = sg; | |
8429 | sg = sg->next; | |
8430 | kfree(oldsg); | |
8431 | if (oldsg != sched_group_nodes[i]) | |
8432 | goto next_sg; | |
8433 | } | |
8434 | kfree(sched_group_nodes); | |
8435 | sched_group_nodes_bycpu[cpu] = NULL; | |
8436 | } | |
51888ca2 | 8437 | } |
6d6bc0ad | 8438 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8439 | static void free_sched_groups(const struct cpumask *cpu_map, |
8440 | struct cpumask *nodemask) | |
a616058b SS |
8441 | { |
8442 | } | |
6d6bc0ad | 8443 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8444 | |
89c4710e SS |
8445 | /* |
8446 | * Initialize sched groups cpu_power. | |
8447 | * | |
8448 | * cpu_power indicates the capacity of sched group, which is used while | |
8449 | * distributing the load between different sched groups in a sched domain. | |
8450 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8451 | * there are asymmetries in the topology. If there are asymmetries, group | |
8452 | * having more cpu_power will pickup more load compared to the group having | |
8453 | * less cpu_power. | |
89c4710e SS |
8454 | */ |
8455 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8456 | { | |
8457 | struct sched_domain *child; | |
8458 | struct sched_group *group; | |
f93e65c1 PZ |
8459 | long power; |
8460 | int weight; | |
89c4710e SS |
8461 | |
8462 | WARN_ON(!sd || !sd->groups); | |
8463 | ||
13318a71 | 8464 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8465 | return; |
8466 | ||
8467 | child = sd->child; | |
8468 | ||
18a3885f | 8469 | sd->groups->cpu_power = 0; |
5517d86b | 8470 | |
f93e65c1 PZ |
8471 | if (!child) { |
8472 | power = SCHED_LOAD_SCALE; | |
8473 | weight = cpumask_weight(sched_domain_span(sd)); | |
8474 | /* | |
8475 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8476 | * Usually multiple threads get a better yield out of |
8477 | * that one core than a single thread would have, | |
8478 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8479 | */ |
a52bfd73 PZ |
8480 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8481 | power *= sd->smt_gain; | |
f93e65c1 | 8482 | power /= weight; |
a52bfd73 PZ |
8483 | power >>= SCHED_LOAD_SHIFT; |
8484 | } | |
18a3885f | 8485 | sd->groups->cpu_power += power; |
89c4710e SS |
8486 | return; |
8487 | } | |
8488 | ||
89c4710e | 8489 | /* |
f93e65c1 | 8490 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8491 | */ |
8492 | group = child->groups; | |
8493 | do { | |
18a3885f | 8494 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
8495 | group = group->next; |
8496 | } while (group != child->groups); | |
8497 | } | |
8498 | ||
7c16ec58 MT |
8499 | /* |
8500 | * Initializers for schedule domains | |
8501 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8502 | */ | |
8503 | ||
a5d8c348 IM |
8504 | #ifdef CONFIG_SCHED_DEBUG |
8505 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8506 | #else | |
8507 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8508 | #endif | |
8509 | ||
7c16ec58 | 8510 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8511 | |
7c16ec58 MT |
8512 | #define SD_INIT_FUNC(type) \ |
8513 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8514 | { \ | |
8515 | memset(sd, 0, sizeof(*sd)); \ | |
8516 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8517 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8518 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8519 | } |
8520 | ||
8521 | SD_INIT_FUNC(CPU) | |
8522 | #ifdef CONFIG_NUMA | |
8523 | SD_INIT_FUNC(ALLNODES) | |
8524 | SD_INIT_FUNC(NODE) | |
8525 | #endif | |
8526 | #ifdef CONFIG_SCHED_SMT | |
8527 | SD_INIT_FUNC(SIBLING) | |
8528 | #endif | |
8529 | #ifdef CONFIG_SCHED_MC | |
8530 | SD_INIT_FUNC(MC) | |
8531 | #endif | |
8532 | ||
1d3504fc HS |
8533 | static int default_relax_domain_level = -1; |
8534 | ||
8535 | static int __init setup_relax_domain_level(char *str) | |
8536 | { | |
30e0e178 LZ |
8537 | unsigned long val; |
8538 | ||
8539 | val = simple_strtoul(str, NULL, 0); | |
8540 | if (val < SD_LV_MAX) | |
8541 | default_relax_domain_level = val; | |
8542 | ||
1d3504fc HS |
8543 | return 1; |
8544 | } | |
8545 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8546 | ||
8547 | static void set_domain_attribute(struct sched_domain *sd, | |
8548 | struct sched_domain_attr *attr) | |
8549 | { | |
8550 | int request; | |
8551 | ||
8552 | if (!attr || attr->relax_domain_level < 0) { | |
8553 | if (default_relax_domain_level < 0) | |
8554 | return; | |
8555 | else | |
8556 | request = default_relax_domain_level; | |
8557 | } else | |
8558 | request = attr->relax_domain_level; | |
8559 | if (request < sd->level) { | |
8560 | /* turn off idle balance on this domain */ | |
8561 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8562 | } else { | |
8563 | /* turn on idle balance on this domain */ | |
8564 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8565 | } | |
8566 | } | |
8567 | ||
2109b99e AH |
8568 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8569 | const struct cpumask *cpu_map) | |
8570 | { | |
8571 | switch (what) { | |
8572 | case sa_sched_groups: | |
8573 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8574 | d->sched_group_nodes = NULL; | |
8575 | case sa_rootdomain: | |
8576 | free_rootdomain(d->rd); /* fall through */ | |
8577 | case sa_tmpmask: | |
8578 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8579 | case sa_send_covered: | |
8580 | free_cpumask_var(d->send_covered); /* fall through */ | |
8581 | case sa_this_core_map: | |
8582 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8583 | case sa_this_sibling_map: | |
8584 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8585 | case sa_nodemask: | |
8586 | free_cpumask_var(d->nodemask); /* fall through */ | |
8587 | case sa_sched_group_nodes: | |
d1b55138 | 8588 | #ifdef CONFIG_NUMA |
2109b99e AH |
8589 | kfree(d->sched_group_nodes); /* fall through */ |
8590 | case sa_notcovered: | |
8591 | free_cpumask_var(d->notcovered); /* fall through */ | |
8592 | case sa_covered: | |
8593 | free_cpumask_var(d->covered); /* fall through */ | |
8594 | case sa_domainspan: | |
8595 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8596 | #endif |
2109b99e AH |
8597 | case sa_none: |
8598 | break; | |
8599 | } | |
8600 | } | |
3404c8d9 | 8601 | |
2109b99e AH |
8602 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8603 | const struct cpumask *cpu_map) | |
8604 | { | |
3404c8d9 | 8605 | #ifdef CONFIG_NUMA |
2109b99e AH |
8606 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8607 | return sa_none; | |
8608 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8609 | return sa_domainspan; | |
8610 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8611 | return sa_covered; | |
8612 | /* Allocate the per-node list of sched groups */ | |
8613 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8614 | sizeof(struct sched_group *), GFP_KERNEL); | |
8615 | if (!d->sched_group_nodes) { | |
d1b55138 | 8616 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 8617 | return sa_notcovered; |
d1b55138 | 8618 | } |
2109b99e | 8619 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8620 | #endif |
2109b99e AH |
8621 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8622 | return sa_sched_group_nodes; | |
8623 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8624 | return sa_nodemask; | |
8625 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8626 | return sa_this_sibling_map; | |
8627 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8628 | return sa_this_core_map; | |
8629 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8630 | return sa_send_covered; | |
8631 | d->rd = alloc_rootdomain(); | |
8632 | if (!d->rd) { | |
57d885fe | 8633 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 8634 | return sa_tmpmask; |
57d885fe | 8635 | } |
2109b99e AH |
8636 | return sa_rootdomain; |
8637 | } | |
57d885fe | 8638 | |
7f4588f3 AH |
8639 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8640 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8641 | { | |
8642 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8643 | #ifdef CONFIG_NUMA |
7f4588f3 | 8644 | struct sched_domain *parent; |
1da177e4 | 8645 | |
7f4588f3 AH |
8646 | d->sd_allnodes = 0; |
8647 | if (cpumask_weight(cpu_map) > | |
8648 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8649 | sd = &per_cpu(allnodes_domains, i).sd; | |
8650 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8651 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8652 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8653 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8654 | d->sd_allnodes = 1; | |
8655 | } | |
8656 | parent = sd; | |
8657 | ||
8658 | sd = &per_cpu(node_domains, i).sd; | |
8659 | SD_INIT(sd, NODE); | |
8660 | set_domain_attribute(sd, attr); | |
8661 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8662 | sd->parent = parent; | |
8663 | if (parent) | |
8664 | parent->child = sd; | |
8665 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8666 | #endif |
7f4588f3 AH |
8667 | return sd; |
8668 | } | |
1da177e4 | 8669 | |
87cce662 AH |
8670 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8671 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8672 | struct sched_domain *parent, int i) | |
8673 | { | |
8674 | struct sched_domain *sd; | |
8675 | sd = &per_cpu(phys_domains, i).sd; | |
8676 | SD_INIT(sd, CPU); | |
8677 | set_domain_attribute(sd, attr); | |
8678 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8679 | sd->parent = parent; | |
8680 | if (parent) | |
8681 | parent->child = sd; | |
8682 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8683 | return sd; | |
8684 | } | |
1da177e4 | 8685 | |
410c4081 AH |
8686 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8687 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8688 | struct sched_domain *parent, int i) | |
8689 | { | |
8690 | struct sched_domain *sd = parent; | |
1e9f28fa | 8691 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8692 | sd = &per_cpu(core_domains, i).sd; |
8693 | SD_INIT(sd, MC); | |
8694 | set_domain_attribute(sd, attr); | |
8695 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8696 | sd->parent = parent; | |
8697 | parent->child = sd; | |
8698 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8699 | #endif |
410c4081 AH |
8700 | return sd; |
8701 | } | |
1e9f28fa | 8702 | |
d8173535 AH |
8703 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8704 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8705 | struct sched_domain *parent, int i) | |
8706 | { | |
8707 | struct sched_domain *sd = parent; | |
1da177e4 | 8708 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8709 | sd = &per_cpu(cpu_domains, i).sd; |
8710 | SD_INIT(sd, SIBLING); | |
8711 | set_domain_attribute(sd, attr); | |
8712 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8713 | sd->parent = parent; | |
8714 | parent->child = sd; | |
8715 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8716 | #endif |
d8173535 AH |
8717 | return sd; |
8718 | } | |
1da177e4 | 8719 | |
0e8e85c9 AH |
8720 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8721 | const struct cpumask *cpu_map, int cpu) | |
8722 | { | |
8723 | switch (l) { | |
1da177e4 | 8724 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8725 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8726 | cpumask_and(d->this_sibling_map, cpu_map, | |
8727 | topology_thread_cpumask(cpu)); | |
8728 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8729 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8730 | &cpu_to_cpu_group, | |
8731 | d->send_covered, d->tmpmask); | |
8732 | break; | |
1da177e4 | 8733 | #endif |
1e9f28fa | 8734 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8735 | case SD_LV_MC: /* set up multi-core groups */ |
8736 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8737 | if (cpu == cpumask_first(d->this_core_map)) | |
8738 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8739 | &cpu_to_core_group, | |
8740 | d->send_covered, d->tmpmask); | |
8741 | break; | |
1e9f28fa | 8742 | #endif |
86548096 AH |
8743 | case SD_LV_CPU: /* set up physical groups */ |
8744 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8745 | if (!cpumask_empty(d->nodemask)) | |
8746 | init_sched_build_groups(d->nodemask, cpu_map, | |
8747 | &cpu_to_phys_group, | |
8748 | d->send_covered, d->tmpmask); | |
8749 | break; | |
1da177e4 | 8750 | #ifdef CONFIG_NUMA |
de616e36 AH |
8751 | case SD_LV_ALLNODES: |
8752 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8753 | d->send_covered, d->tmpmask); | |
8754 | break; | |
8755 | #endif | |
0e8e85c9 AH |
8756 | default: |
8757 | break; | |
7c16ec58 | 8758 | } |
0e8e85c9 | 8759 | } |
9c1cfda2 | 8760 | |
2109b99e AH |
8761 | /* |
8762 | * Build sched domains for a given set of cpus and attach the sched domains | |
8763 | * to the individual cpus | |
8764 | */ | |
8765 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8766 | struct sched_domain_attr *attr) | |
8767 | { | |
8768 | enum s_alloc alloc_state = sa_none; | |
8769 | struct s_data d; | |
294b0c96 | 8770 | struct sched_domain *sd; |
2109b99e | 8771 | int i; |
7c16ec58 | 8772 | #ifdef CONFIG_NUMA |
2109b99e | 8773 | d.sd_allnodes = 0; |
7c16ec58 | 8774 | #endif |
9c1cfda2 | 8775 | |
2109b99e AH |
8776 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8777 | if (alloc_state != sa_rootdomain) | |
8778 | goto error; | |
8779 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8780 | |
1da177e4 | 8781 | /* |
1a20ff27 | 8782 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8783 | */ |
abcd083a | 8784 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8785 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8786 | cpu_map); | |
9761eea8 | 8787 | |
7f4588f3 | 8788 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8789 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8790 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8791 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8792 | } |
9c1cfda2 | 8793 | |
abcd083a | 8794 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8795 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8796 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8797 | } |
9c1cfda2 | 8798 | |
1da177e4 | 8799 | /* Set up physical groups */ |
86548096 AH |
8800 | for (i = 0; i < nr_node_ids; i++) |
8801 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8802 | |
1da177e4 LT |
8803 | #ifdef CONFIG_NUMA |
8804 | /* Set up node groups */ | |
de616e36 AH |
8805 | if (d.sd_allnodes) |
8806 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8807 | |
0601a88d AH |
8808 | for (i = 0; i < nr_node_ids; i++) |
8809 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8810 | goto error; |
1da177e4 LT |
8811 | #endif |
8812 | ||
8813 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8814 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8815 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8816 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8817 | init_sched_groups_power(i, sd); |
5c45bf27 | 8818 | } |
1da177e4 | 8819 | #endif |
1e9f28fa | 8820 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8821 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8822 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8823 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8824 | } |
8825 | #endif | |
1e9f28fa | 8826 | |
abcd083a | 8827 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8828 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8829 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8830 | } |
8831 | ||
9c1cfda2 | 8832 | #ifdef CONFIG_NUMA |
076ac2af | 8833 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8834 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8835 | |
49a02c51 | 8836 | if (d.sd_allnodes) { |
6711cab4 | 8837 | struct sched_group *sg; |
f712c0c7 | 8838 | |
96f874e2 | 8839 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 8840 | d.tmpmask); |
f712c0c7 SS |
8841 | init_numa_sched_groups_power(sg); |
8842 | } | |
9c1cfda2 JH |
8843 | #endif |
8844 | ||
1da177e4 | 8845 | /* Attach the domains */ |
abcd083a | 8846 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8847 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8848 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8849 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8850 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8851 | #else |
6c99e9ad | 8852 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8853 | #endif |
49a02c51 | 8854 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 8855 | } |
51888ca2 | 8856 | |
2109b99e AH |
8857 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
8858 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
8859 | return 0; | |
51888ca2 | 8860 | |
51888ca2 | 8861 | error: |
2109b99e AH |
8862 | __free_domain_allocs(&d, alloc_state, cpu_map); |
8863 | return -ENOMEM; | |
1da177e4 | 8864 | } |
029190c5 | 8865 | |
96f874e2 | 8866 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8867 | { |
8868 | return __build_sched_domains(cpu_map, NULL); | |
8869 | } | |
8870 | ||
96f874e2 | 8871 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8872 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8873 | static struct sched_domain_attr *dattr_cur; |
8874 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8875 | |
8876 | /* | |
8877 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8878 | * cpumask) fails, then fallback to a single sched domain, |
8879 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8880 | */ |
4212823f | 8881 | static cpumask_var_t fallback_doms; |
029190c5 | 8882 | |
ee79d1bd HC |
8883 | /* |
8884 | * arch_update_cpu_topology lets virtualized architectures update the | |
8885 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8886 | * or 0 if it stayed the same. | |
8887 | */ | |
8888 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8889 | { |
ee79d1bd | 8890 | return 0; |
22e52b07 HC |
8891 | } |
8892 | ||
1a20ff27 | 8893 | /* |
41a2d6cf | 8894 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8895 | * For now this just excludes isolated cpus, but could be used to |
8896 | * exclude other special cases in the future. | |
1a20ff27 | 8897 | */ |
96f874e2 | 8898 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8899 | { |
7378547f MM |
8900 | int err; |
8901 | ||
22e52b07 | 8902 | arch_update_cpu_topology(); |
029190c5 | 8903 | ndoms_cur = 1; |
96f874e2 | 8904 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8905 | if (!doms_cur) |
4212823f | 8906 | doms_cur = fallback_doms; |
dcc30a35 | 8907 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8908 | dattr_cur = NULL; |
7378547f | 8909 | err = build_sched_domains(doms_cur); |
6382bc90 | 8910 | register_sched_domain_sysctl(); |
7378547f MM |
8911 | |
8912 | return err; | |
1a20ff27 DG |
8913 | } |
8914 | ||
96f874e2 RR |
8915 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8916 | struct cpumask *tmpmask) | |
1da177e4 | 8917 | { |
7c16ec58 | 8918 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8919 | } |
1da177e4 | 8920 | |
1a20ff27 DG |
8921 | /* |
8922 | * Detach sched domains from a group of cpus specified in cpu_map | |
8923 | * These cpus will now be attached to the NULL domain | |
8924 | */ | |
96f874e2 | 8925 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8926 | { |
96f874e2 RR |
8927 | /* Save because hotplug lock held. */ |
8928 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8929 | int i; |
8930 | ||
abcd083a | 8931 | for_each_cpu(i, cpu_map) |
57d885fe | 8932 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8933 | synchronize_sched(); |
96f874e2 | 8934 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8935 | } |
8936 | ||
1d3504fc HS |
8937 | /* handle null as "default" */ |
8938 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8939 | struct sched_domain_attr *new, int idx_new) | |
8940 | { | |
8941 | struct sched_domain_attr tmp; | |
8942 | ||
8943 | /* fast path */ | |
8944 | if (!new && !cur) | |
8945 | return 1; | |
8946 | ||
8947 | tmp = SD_ATTR_INIT; | |
8948 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8949 | new ? (new + idx_new) : &tmp, | |
8950 | sizeof(struct sched_domain_attr)); | |
8951 | } | |
8952 | ||
029190c5 PJ |
8953 | /* |
8954 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8955 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8956 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8957 | * It destroys each deleted domain and builds each new domain. | |
8958 | * | |
96f874e2 | 8959 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8960 | * The masks don't intersect (don't overlap.) We should setup one |
8961 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8962 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8963 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8964 | * it as it is. | |
8965 | * | |
41a2d6cf IM |
8966 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8967 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8968 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8969 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8970 | * the single partition 'fallback_doms', it also forces the domains | |
8971 | * to be rebuilt. | |
029190c5 | 8972 | * |
96f874e2 | 8973 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8974 | * ndoms_new == 0 is a special case for destroying existing domains, |
8975 | * and it will not create the default domain. | |
dfb512ec | 8976 | * |
029190c5 PJ |
8977 | * Call with hotplug lock held |
8978 | */ | |
96f874e2 RR |
8979 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8980 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8981 | struct sched_domain_attr *dattr_new) |
029190c5 | 8982 | { |
dfb512ec | 8983 | int i, j, n; |
d65bd5ec | 8984 | int new_topology; |
029190c5 | 8985 | |
712555ee | 8986 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8987 | |
7378547f MM |
8988 | /* always unregister in case we don't destroy any domains */ |
8989 | unregister_sched_domain_sysctl(); | |
8990 | ||
d65bd5ec HC |
8991 | /* Let architecture update cpu core mappings. */ |
8992 | new_topology = arch_update_cpu_topology(); | |
8993 | ||
dfb512ec | 8994 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8995 | |
8996 | /* Destroy deleted domains */ | |
8997 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8998 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8999 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 9000 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9001 | goto match1; |
9002 | } | |
9003 | /* no match - a current sched domain not in new doms_new[] */ | |
9004 | detach_destroy_domains(doms_cur + i); | |
9005 | match1: | |
9006 | ; | |
9007 | } | |
9008 | ||
e761b772 MK |
9009 | if (doms_new == NULL) { |
9010 | ndoms_cur = 0; | |
4212823f | 9011 | doms_new = fallback_doms; |
dcc30a35 | 9012 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 9013 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9014 | } |
9015 | ||
029190c5 PJ |
9016 | /* Build new domains */ |
9017 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9018 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 9019 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 9020 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9021 | goto match2; |
9022 | } | |
9023 | /* no match - add a new doms_new */ | |
1d3504fc HS |
9024 | __build_sched_domains(doms_new + i, |
9025 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
9026 | match2: |
9027 | ; | |
9028 | } | |
9029 | ||
9030 | /* Remember the new sched domains */ | |
4212823f | 9031 | if (doms_cur != fallback_doms) |
029190c5 | 9032 | kfree(doms_cur); |
1d3504fc | 9033 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9034 | doms_cur = doms_new; |
1d3504fc | 9035 | dattr_cur = dattr_new; |
029190c5 | 9036 | ndoms_cur = ndoms_new; |
7378547f MM |
9037 | |
9038 | register_sched_domain_sysctl(); | |
a1835615 | 9039 | |
712555ee | 9040 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9041 | } |
9042 | ||
5c45bf27 | 9043 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9044 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9045 | { |
95402b38 | 9046 | get_online_cpus(); |
dfb512ec MK |
9047 | |
9048 | /* Destroy domains first to force the rebuild */ | |
9049 | partition_sched_domains(0, NULL, NULL); | |
9050 | ||
e761b772 | 9051 | rebuild_sched_domains(); |
95402b38 | 9052 | put_online_cpus(); |
5c45bf27 SS |
9053 | } |
9054 | ||
9055 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9056 | { | |
afb8a9b7 | 9057 | unsigned int level = 0; |
5c45bf27 | 9058 | |
afb8a9b7 GS |
9059 | if (sscanf(buf, "%u", &level) != 1) |
9060 | return -EINVAL; | |
9061 | ||
9062 | /* | |
9063 | * level is always be positive so don't check for | |
9064 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9065 | * What happens on 0 or 1 byte write, | |
9066 | * need to check for count as well? | |
9067 | */ | |
9068 | ||
9069 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9070 | return -EINVAL; |
9071 | ||
9072 | if (smt) | |
afb8a9b7 | 9073 | sched_smt_power_savings = level; |
5c45bf27 | 9074 | else |
afb8a9b7 | 9075 | sched_mc_power_savings = level; |
5c45bf27 | 9076 | |
c70f22d2 | 9077 | arch_reinit_sched_domains(); |
5c45bf27 | 9078 | |
c70f22d2 | 9079 | return count; |
5c45bf27 SS |
9080 | } |
9081 | ||
5c45bf27 | 9082 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9083 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9084 | char *page) | |
5c45bf27 SS |
9085 | { |
9086 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9087 | } | |
f718cd4a | 9088 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9089 | const char *buf, size_t count) |
5c45bf27 SS |
9090 | { |
9091 | return sched_power_savings_store(buf, count, 0); | |
9092 | } | |
f718cd4a AK |
9093 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9094 | sched_mc_power_savings_show, | |
9095 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9096 | #endif |
9097 | ||
9098 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9099 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9100 | char *page) | |
5c45bf27 SS |
9101 | { |
9102 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9103 | } | |
f718cd4a | 9104 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9105 | const char *buf, size_t count) |
5c45bf27 SS |
9106 | { |
9107 | return sched_power_savings_store(buf, count, 1); | |
9108 | } | |
f718cd4a AK |
9109 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9110 | sched_smt_power_savings_show, | |
6707de00 AB |
9111 | sched_smt_power_savings_store); |
9112 | #endif | |
9113 | ||
39aac648 | 9114 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9115 | { |
9116 | int err = 0; | |
9117 | ||
9118 | #ifdef CONFIG_SCHED_SMT | |
9119 | if (smt_capable()) | |
9120 | err = sysfs_create_file(&cls->kset.kobj, | |
9121 | &attr_sched_smt_power_savings.attr); | |
9122 | #endif | |
9123 | #ifdef CONFIG_SCHED_MC | |
9124 | if (!err && mc_capable()) | |
9125 | err = sysfs_create_file(&cls->kset.kobj, | |
9126 | &attr_sched_mc_power_savings.attr); | |
9127 | #endif | |
9128 | return err; | |
9129 | } | |
6d6bc0ad | 9130 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9131 | |
e761b772 | 9132 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9133 | /* |
e761b772 MK |
9134 | * Add online and remove offline CPUs from the scheduler domains. |
9135 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9136 | */ |
9137 | static int update_sched_domains(struct notifier_block *nfb, | |
9138 | unsigned long action, void *hcpu) | |
e761b772 MK |
9139 | { |
9140 | switch (action) { | |
9141 | case CPU_ONLINE: | |
9142 | case CPU_ONLINE_FROZEN: | |
9143 | case CPU_DEAD: | |
9144 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9145 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9146 | return NOTIFY_OK; |
9147 | ||
9148 | default: | |
9149 | return NOTIFY_DONE; | |
9150 | } | |
9151 | } | |
9152 | #endif | |
9153 | ||
9154 | static int update_runtime(struct notifier_block *nfb, | |
9155 | unsigned long action, void *hcpu) | |
1da177e4 | 9156 | { |
7def2be1 PZ |
9157 | int cpu = (int)(long)hcpu; |
9158 | ||
1da177e4 | 9159 | switch (action) { |
1da177e4 | 9160 | case CPU_DOWN_PREPARE: |
8bb78442 | 9161 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9162 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9163 | return NOTIFY_OK; |
9164 | ||
1da177e4 | 9165 | case CPU_DOWN_FAILED: |
8bb78442 | 9166 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9167 | case CPU_ONLINE: |
8bb78442 | 9168 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9169 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9170 | return NOTIFY_OK; |
9171 | ||
1da177e4 LT |
9172 | default: |
9173 | return NOTIFY_DONE; | |
9174 | } | |
1da177e4 | 9175 | } |
1da177e4 LT |
9176 | |
9177 | void __init sched_init_smp(void) | |
9178 | { | |
dcc30a35 RR |
9179 | cpumask_var_t non_isolated_cpus; |
9180 | ||
9181 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 9182 | |
434d53b0 MT |
9183 | #if defined(CONFIG_NUMA) |
9184 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9185 | GFP_KERNEL); | |
9186 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9187 | #endif | |
95402b38 | 9188 | get_online_cpus(); |
712555ee | 9189 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9190 | arch_init_sched_domains(cpu_online_mask); |
9191 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9192 | if (cpumask_empty(non_isolated_cpus)) | |
9193 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9194 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9195 | put_online_cpus(); |
e761b772 MK |
9196 | |
9197 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9198 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9199 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9200 | #endif |
9201 | ||
9202 | /* RT runtime code needs to handle some hotplug events */ | |
9203 | hotcpu_notifier(update_runtime, 0); | |
9204 | ||
b328ca18 | 9205 | init_hrtick(); |
5c1e1767 NP |
9206 | |
9207 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9208 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9209 | BUG(); |
19978ca6 | 9210 | sched_init_granularity(); |
dcc30a35 | 9211 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9212 | |
9213 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9214 | init_sched_rt_class(); |
1da177e4 LT |
9215 | } |
9216 | #else | |
9217 | void __init sched_init_smp(void) | |
9218 | { | |
19978ca6 | 9219 | sched_init_granularity(); |
1da177e4 LT |
9220 | } |
9221 | #endif /* CONFIG_SMP */ | |
9222 | ||
cd1bb94b AB |
9223 | const_debug unsigned int sysctl_timer_migration = 1; |
9224 | ||
1da177e4 LT |
9225 | int in_sched_functions(unsigned long addr) |
9226 | { | |
1da177e4 LT |
9227 | return in_lock_functions(addr) || |
9228 | (addr >= (unsigned long)__sched_text_start | |
9229 | && addr < (unsigned long)__sched_text_end); | |
9230 | } | |
9231 | ||
a9957449 | 9232 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9233 | { |
9234 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9235 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9236 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9237 | cfs_rq->rq = rq; | |
9238 | #endif | |
67e9fb2a | 9239 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9240 | } |
9241 | ||
fa85ae24 PZ |
9242 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9243 | { | |
9244 | struct rt_prio_array *array; | |
9245 | int i; | |
9246 | ||
9247 | array = &rt_rq->active; | |
9248 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9249 | INIT_LIST_HEAD(array->queue + i); | |
9250 | __clear_bit(i, array->bitmap); | |
9251 | } | |
9252 | /* delimiter for bitsearch: */ | |
9253 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9254 | ||
052f1dc7 | 9255 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9256 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9257 | #ifdef CONFIG_SMP |
e864c499 | 9258 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9259 | #endif |
48d5e258 | 9260 | #endif |
fa85ae24 PZ |
9261 | #ifdef CONFIG_SMP |
9262 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9263 | rt_rq->overloaded = 0; |
c20b08e3 | 9264 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9265 | #endif |
9266 | ||
9267 | rt_rq->rt_time = 0; | |
9268 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9269 | rt_rq->rt_runtime = 0; |
9270 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9271 | |
052f1dc7 | 9272 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9273 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9274 | rt_rq->rq = rq; |
9275 | #endif | |
fa85ae24 PZ |
9276 | } |
9277 | ||
6f505b16 | 9278 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9279 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9280 | struct sched_entity *se, int cpu, int add, | |
9281 | struct sched_entity *parent) | |
6f505b16 | 9282 | { |
ec7dc8ac | 9283 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9284 | tg->cfs_rq[cpu] = cfs_rq; |
9285 | init_cfs_rq(cfs_rq, rq); | |
9286 | cfs_rq->tg = tg; | |
9287 | if (add) | |
9288 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9289 | ||
9290 | tg->se[cpu] = se; | |
354d60c2 DG |
9291 | /* se could be NULL for init_task_group */ |
9292 | if (!se) | |
9293 | return; | |
9294 | ||
ec7dc8ac DG |
9295 | if (!parent) |
9296 | se->cfs_rq = &rq->cfs; | |
9297 | else | |
9298 | se->cfs_rq = parent->my_q; | |
9299 | ||
6f505b16 PZ |
9300 | se->my_q = cfs_rq; |
9301 | se->load.weight = tg->shares; | |
e05510d0 | 9302 | se->load.inv_weight = 0; |
ec7dc8ac | 9303 | se->parent = parent; |
6f505b16 | 9304 | } |
052f1dc7 | 9305 | #endif |
6f505b16 | 9306 | |
052f1dc7 | 9307 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9308 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9309 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9310 | struct sched_rt_entity *parent) | |
6f505b16 | 9311 | { |
ec7dc8ac DG |
9312 | struct rq *rq = cpu_rq(cpu); |
9313 | ||
6f505b16 PZ |
9314 | tg->rt_rq[cpu] = rt_rq; |
9315 | init_rt_rq(rt_rq, rq); | |
9316 | rt_rq->tg = tg; | |
9317 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9318 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9319 | if (add) |
9320 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9321 | ||
9322 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9323 | if (!rt_se) |
9324 | return; | |
9325 | ||
ec7dc8ac DG |
9326 | if (!parent) |
9327 | rt_se->rt_rq = &rq->rt; | |
9328 | else | |
9329 | rt_se->rt_rq = parent->my_q; | |
9330 | ||
6f505b16 | 9331 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9332 | rt_se->parent = parent; |
6f505b16 PZ |
9333 | INIT_LIST_HEAD(&rt_se->run_list); |
9334 | } | |
9335 | #endif | |
9336 | ||
1da177e4 LT |
9337 | void __init sched_init(void) |
9338 | { | |
dd41f596 | 9339 | int i, j; |
434d53b0 MT |
9340 | unsigned long alloc_size = 0, ptr; |
9341 | ||
9342 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9343 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9344 | #endif | |
9345 | #ifdef CONFIG_RT_GROUP_SCHED | |
9346 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9347 | #endif |
9348 | #ifdef CONFIG_USER_SCHED | |
9349 | alloc_size *= 2; | |
df7c8e84 RR |
9350 | #endif |
9351 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9352 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9353 | #endif |
9354 | /* | |
9355 | * As sched_init() is called before page_alloc is setup, | |
9356 | * we use alloc_bootmem(). | |
9357 | */ | |
9358 | if (alloc_size) { | |
36b7b6d4 | 9359 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9360 | |
9361 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9362 | init_task_group.se = (struct sched_entity **)ptr; | |
9363 | ptr += nr_cpu_ids * sizeof(void **); | |
9364 | ||
9365 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9366 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9367 | |
9368 | #ifdef CONFIG_USER_SCHED | |
9369 | root_task_group.se = (struct sched_entity **)ptr; | |
9370 | ptr += nr_cpu_ids * sizeof(void **); | |
9371 | ||
9372 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9373 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9374 | #endif /* CONFIG_USER_SCHED */ |
9375 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9376 | #ifdef CONFIG_RT_GROUP_SCHED |
9377 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9378 | ptr += nr_cpu_ids * sizeof(void **); | |
9379 | ||
9380 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9381 | ptr += nr_cpu_ids * sizeof(void **); |
9382 | ||
9383 | #ifdef CONFIG_USER_SCHED | |
9384 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9385 | ptr += nr_cpu_ids * sizeof(void **); | |
9386 | ||
9387 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9388 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9389 | #endif /* CONFIG_USER_SCHED */ |
9390 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9391 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9392 | for_each_possible_cpu(i) { | |
9393 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9394 | ptr += cpumask_size(); | |
9395 | } | |
9396 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9397 | } |
dd41f596 | 9398 | |
57d885fe GH |
9399 | #ifdef CONFIG_SMP |
9400 | init_defrootdomain(); | |
9401 | #endif | |
9402 | ||
d0b27fa7 PZ |
9403 | init_rt_bandwidth(&def_rt_bandwidth, |
9404 | global_rt_period(), global_rt_runtime()); | |
9405 | ||
9406 | #ifdef CONFIG_RT_GROUP_SCHED | |
9407 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9408 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9409 | #ifdef CONFIG_USER_SCHED |
9410 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9411 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9412 | #endif /* CONFIG_USER_SCHED */ |
9413 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9414 | |
052f1dc7 | 9415 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9416 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9417 | INIT_LIST_HEAD(&init_task_group.children); |
9418 | ||
9419 | #ifdef CONFIG_USER_SCHED | |
9420 | INIT_LIST_HEAD(&root_task_group.children); | |
9421 | init_task_group.parent = &root_task_group; | |
9422 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9423 | #endif /* CONFIG_USER_SCHED */ |
9424 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9425 | |
0a945022 | 9426 | for_each_possible_cpu(i) { |
70b97a7f | 9427 | struct rq *rq; |
1da177e4 LT |
9428 | |
9429 | rq = cpu_rq(i); | |
9430 | spin_lock_init(&rq->lock); | |
7897986b | 9431 | rq->nr_running = 0; |
dce48a84 TG |
9432 | rq->calc_load_active = 0; |
9433 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9434 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9435 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9436 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9437 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9438 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9439 | #ifdef CONFIG_CGROUP_SCHED |
9440 | /* | |
9441 | * How much cpu bandwidth does init_task_group get? | |
9442 | * | |
9443 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9444 | * gets 100% of the cpu resources in the system. This overall | |
9445 | * system cpu resource is divided among the tasks of | |
9446 | * init_task_group and its child task-groups in a fair manner, | |
9447 | * based on each entity's (task or task-group's) weight | |
9448 | * (se->load.weight). | |
9449 | * | |
9450 | * In other words, if init_task_group has 10 tasks of weight | |
9451 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9452 | * then A0's share of the cpu resource is: | |
9453 | * | |
0d905bca | 9454 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9455 | * |
9456 | * We achieve this by letting init_task_group's tasks sit | |
9457 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9458 | */ | |
ec7dc8ac | 9459 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9460 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9461 | root_task_group.shares = NICE_0_LOAD; |
9462 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9463 | /* |
9464 | * In case of task-groups formed thr' the user id of tasks, | |
9465 | * init_task_group represents tasks belonging to root user. | |
9466 | * Hence it forms a sibling of all subsequent groups formed. | |
9467 | * In this case, init_task_group gets only a fraction of overall | |
9468 | * system cpu resource, based on the weight assigned to root | |
9469 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9470 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9471 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9472 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9473 | */ | |
ec7dc8ac | 9474 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9475 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9476 | &per_cpu(init_sched_entity, i), i, 1, |
9477 | root_task_group.se[i]); | |
6f505b16 | 9478 | |
052f1dc7 | 9479 | #endif |
354d60c2 DG |
9480 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9481 | ||
9482 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9483 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9484 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9485 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9486 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9487 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9488 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9489 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9490 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9491 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9492 | root_task_group.rt_se[i]); | |
354d60c2 | 9493 | #endif |
dd41f596 | 9494 | #endif |
1da177e4 | 9495 | |
dd41f596 IM |
9496 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9497 | rq->cpu_load[j] = 0; | |
1da177e4 | 9498 | #ifdef CONFIG_SMP |
41c7ce9a | 9499 | rq->sd = NULL; |
57d885fe | 9500 | rq->rd = NULL; |
3f029d3c | 9501 | rq->post_schedule = 0; |
1da177e4 | 9502 | rq->active_balance = 0; |
dd41f596 | 9503 | rq->next_balance = jiffies; |
1da177e4 | 9504 | rq->push_cpu = 0; |
0a2966b4 | 9505 | rq->cpu = i; |
1f11eb6a | 9506 | rq->online = 0; |
1da177e4 LT |
9507 | rq->migration_thread = NULL; |
9508 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9509 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9510 | #endif |
8f4d37ec | 9511 | init_rq_hrtick(rq); |
1da177e4 | 9512 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9513 | } |
9514 | ||
2dd73a4f | 9515 | set_load_weight(&init_task); |
b50f60ce | 9516 | |
e107be36 AK |
9517 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9518 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9519 | #endif | |
9520 | ||
c9819f45 | 9521 | #ifdef CONFIG_SMP |
962cf36c | 9522 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9523 | #endif |
9524 | ||
b50f60ce HC |
9525 | #ifdef CONFIG_RT_MUTEXES |
9526 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9527 | #endif | |
9528 | ||
1da177e4 LT |
9529 | /* |
9530 | * The boot idle thread does lazy MMU switching as well: | |
9531 | */ | |
9532 | atomic_inc(&init_mm.mm_count); | |
9533 | enter_lazy_tlb(&init_mm, current); | |
9534 | ||
9535 | /* | |
9536 | * Make us the idle thread. Technically, schedule() should not be | |
9537 | * called from this thread, however somewhere below it might be, | |
9538 | * but because we are the idle thread, we just pick up running again | |
9539 | * when this runqueue becomes "idle". | |
9540 | */ | |
9541 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9542 | |
9543 | calc_load_update = jiffies + LOAD_FREQ; | |
9544 | ||
dd41f596 IM |
9545 | /* |
9546 | * During early bootup we pretend to be a normal task: | |
9547 | */ | |
9548 | current->sched_class = &fair_sched_class; | |
6892b75e | 9549 | |
6a7b3dc3 | 9550 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9551 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9552 | #ifdef CONFIG_SMP |
7d1e6a9b | 9553 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9554 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9555 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9556 | #endif |
4bdddf8f | 9557 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9558 | #endif /* SMP */ |
6a7b3dc3 | 9559 | |
0d905bca IM |
9560 | perf_counter_init(); |
9561 | ||
6892b75e | 9562 | scheduler_running = 1; |
1da177e4 LT |
9563 | } |
9564 | ||
9565 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9566 | static inline int preempt_count_equals(int preempt_offset) |
9567 | { | |
9568 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9569 | ||
9570 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9571 | } | |
9572 | ||
9573 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9574 | { |
48f24c4d | 9575 | #ifdef in_atomic |
1da177e4 LT |
9576 | static unsigned long prev_jiffy; /* ratelimiting */ |
9577 | ||
e4aafea2 FW |
9578 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9579 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9580 | return; |
9581 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9582 | return; | |
9583 | prev_jiffy = jiffies; | |
9584 | ||
9585 | printk(KERN_ERR | |
9586 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9587 | file, line); | |
9588 | printk(KERN_ERR | |
9589 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9590 | in_atomic(), irqs_disabled(), | |
9591 | current->pid, current->comm); | |
9592 | ||
9593 | debug_show_held_locks(current); | |
9594 | if (irqs_disabled()) | |
9595 | print_irqtrace_events(current); | |
9596 | dump_stack(); | |
1da177e4 LT |
9597 | #endif |
9598 | } | |
9599 | EXPORT_SYMBOL(__might_sleep); | |
9600 | #endif | |
9601 | ||
9602 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9603 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9604 | { | |
9605 | int on_rq; | |
3e51f33f | 9606 | |
3a5e4dc1 AK |
9607 | update_rq_clock(rq); |
9608 | on_rq = p->se.on_rq; | |
9609 | if (on_rq) | |
9610 | deactivate_task(rq, p, 0); | |
9611 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9612 | if (on_rq) { | |
9613 | activate_task(rq, p, 0); | |
9614 | resched_task(rq->curr); | |
9615 | } | |
9616 | } | |
9617 | ||
1da177e4 LT |
9618 | void normalize_rt_tasks(void) |
9619 | { | |
a0f98a1c | 9620 | struct task_struct *g, *p; |
1da177e4 | 9621 | unsigned long flags; |
70b97a7f | 9622 | struct rq *rq; |
1da177e4 | 9623 | |
4cf5d77a | 9624 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9625 | do_each_thread(g, p) { |
178be793 IM |
9626 | /* |
9627 | * Only normalize user tasks: | |
9628 | */ | |
9629 | if (!p->mm) | |
9630 | continue; | |
9631 | ||
6cfb0d5d | 9632 | p->se.exec_start = 0; |
6cfb0d5d | 9633 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9634 | p->se.wait_start = 0; |
dd41f596 | 9635 | p->se.sleep_start = 0; |
dd41f596 | 9636 | p->se.block_start = 0; |
6cfb0d5d | 9637 | #endif |
dd41f596 IM |
9638 | |
9639 | if (!rt_task(p)) { | |
9640 | /* | |
9641 | * Renice negative nice level userspace | |
9642 | * tasks back to 0: | |
9643 | */ | |
9644 | if (TASK_NICE(p) < 0 && p->mm) | |
9645 | set_user_nice(p, 0); | |
1da177e4 | 9646 | continue; |
dd41f596 | 9647 | } |
1da177e4 | 9648 | |
4cf5d77a | 9649 | spin_lock(&p->pi_lock); |
b29739f9 | 9650 | rq = __task_rq_lock(p); |
1da177e4 | 9651 | |
178be793 | 9652 | normalize_task(rq, p); |
3a5e4dc1 | 9653 | |
b29739f9 | 9654 | __task_rq_unlock(rq); |
4cf5d77a | 9655 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9656 | } while_each_thread(g, p); |
9657 | ||
4cf5d77a | 9658 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9659 | } |
9660 | ||
9661 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9662 | |
9663 | #ifdef CONFIG_IA64 | |
9664 | /* | |
9665 | * These functions are only useful for the IA64 MCA handling. | |
9666 | * | |
9667 | * They can only be called when the whole system has been | |
9668 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9669 | * activity can take place. Using them for anything else would | |
9670 | * be a serious bug, and as a result, they aren't even visible | |
9671 | * under any other configuration. | |
9672 | */ | |
9673 | ||
9674 | /** | |
9675 | * curr_task - return the current task for a given cpu. | |
9676 | * @cpu: the processor in question. | |
9677 | * | |
9678 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9679 | */ | |
36c8b586 | 9680 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9681 | { |
9682 | return cpu_curr(cpu); | |
9683 | } | |
9684 | ||
9685 | /** | |
9686 | * set_curr_task - set the current task for a given cpu. | |
9687 | * @cpu: the processor in question. | |
9688 | * @p: the task pointer to set. | |
9689 | * | |
9690 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9691 | * are serviced on a separate stack. It allows the architecture to switch the |
9692 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9693 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9694 | * and caller must save the original value of the current task (see | |
9695 | * curr_task() above) and restore that value before reenabling interrupts and | |
9696 | * re-starting the system. | |
9697 | * | |
9698 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9699 | */ | |
36c8b586 | 9700 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9701 | { |
9702 | cpu_curr(cpu) = p; | |
9703 | } | |
9704 | ||
9705 | #endif | |
29f59db3 | 9706 | |
bccbe08a PZ |
9707 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9708 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9709 | { |
9710 | int i; | |
9711 | ||
9712 | for_each_possible_cpu(i) { | |
9713 | if (tg->cfs_rq) | |
9714 | kfree(tg->cfs_rq[i]); | |
9715 | if (tg->se) | |
9716 | kfree(tg->se[i]); | |
6f505b16 PZ |
9717 | } |
9718 | ||
9719 | kfree(tg->cfs_rq); | |
9720 | kfree(tg->se); | |
6f505b16 PZ |
9721 | } |
9722 | ||
ec7dc8ac DG |
9723 | static |
9724 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9725 | { |
29f59db3 | 9726 | struct cfs_rq *cfs_rq; |
eab17229 | 9727 | struct sched_entity *se; |
9b5b7751 | 9728 | struct rq *rq; |
29f59db3 SV |
9729 | int i; |
9730 | ||
434d53b0 | 9731 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9732 | if (!tg->cfs_rq) |
9733 | goto err; | |
434d53b0 | 9734 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9735 | if (!tg->se) |
9736 | goto err; | |
052f1dc7 PZ |
9737 | |
9738 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9739 | |
9740 | for_each_possible_cpu(i) { | |
9b5b7751 | 9741 | rq = cpu_rq(i); |
29f59db3 | 9742 | |
eab17229 LZ |
9743 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9744 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9745 | if (!cfs_rq) |
9746 | goto err; | |
9747 | ||
eab17229 LZ |
9748 | se = kzalloc_node(sizeof(struct sched_entity), |
9749 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9750 | if (!se) |
9751 | goto err; | |
9752 | ||
eab17229 | 9753 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9754 | } |
9755 | ||
9756 | return 1; | |
9757 | ||
9758 | err: | |
9759 | return 0; | |
9760 | } | |
9761 | ||
9762 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9763 | { | |
9764 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9765 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9766 | } | |
9767 | ||
9768 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9769 | { | |
9770 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9771 | } | |
6d6bc0ad | 9772 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9773 | static inline void free_fair_sched_group(struct task_group *tg) |
9774 | { | |
9775 | } | |
9776 | ||
ec7dc8ac DG |
9777 | static inline |
9778 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9779 | { |
9780 | return 1; | |
9781 | } | |
9782 | ||
9783 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9784 | { | |
9785 | } | |
9786 | ||
9787 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9788 | { | |
9789 | } | |
6d6bc0ad | 9790 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9791 | |
9792 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9793 | static void free_rt_sched_group(struct task_group *tg) |
9794 | { | |
9795 | int i; | |
9796 | ||
d0b27fa7 PZ |
9797 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9798 | ||
bccbe08a PZ |
9799 | for_each_possible_cpu(i) { |
9800 | if (tg->rt_rq) | |
9801 | kfree(tg->rt_rq[i]); | |
9802 | if (tg->rt_se) | |
9803 | kfree(tg->rt_se[i]); | |
9804 | } | |
9805 | ||
9806 | kfree(tg->rt_rq); | |
9807 | kfree(tg->rt_se); | |
9808 | } | |
9809 | ||
ec7dc8ac DG |
9810 | static |
9811 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9812 | { |
9813 | struct rt_rq *rt_rq; | |
eab17229 | 9814 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9815 | struct rq *rq; |
9816 | int i; | |
9817 | ||
434d53b0 | 9818 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9819 | if (!tg->rt_rq) |
9820 | goto err; | |
434d53b0 | 9821 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9822 | if (!tg->rt_se) |
9823 | goto err; | |
9824 | ||
d0b27fa7 PZ |
9825 | init_rt_bandwidth(&tg->rt_bandwidth, |
9826 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9827 | |
9828 | for_each_possible_cpu(i) { | |
9829 | rq = cpu_rq(i); | |
9830 | ||
eab17229 LZ |
9831 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9832 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9833 | if (!rt_rq) |
9834 | goto err; | |
29f59db3 | 9835 | |
eab17229 LZ |
9836 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9837 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9838 | if (!rt_se) |
9839 | goto err; | |
29f59db3 | 9840 | |
eab17229 | 9841 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9842 | } |
9843 | ||
bccbe08a PZ |
9844 | return 1; |
9845 | ||
9846 | err: | |
9847 | return 0; | |
9848 | } | |
9849 | ||
9850 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9851 | { | |
9852 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9853 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9854 | } | |
9855 | ||
9856 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9857 | { | |
9858 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9859 | } | |
6d6bc0ad | 9860 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9861 | static inline void free_rt_sched_group(struct task_group *tg) |
9862 | { | |
9863 | } | |
9864 | ||
ec7dc8ac DG |
9865 | static inline |
9866 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9867 | { |
9868 | return 1; | |
9869 | } | |
9870 | ||
9871 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9872 | { | |
9873 | } | |
9874 | ||
9875 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9876 | { | |
9877 | } | |
6d6bc0ad | 9878 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9879 | |
d0b27fa7 | 9880 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9881 | static void free_sched_group(struct task_group *tg) |
9882 | { | |
9883 | free_fair_sched_group(tg); | |
9884 | free_rt_sched_group(tg); | |
9885 | kfree(tg); | |
9886 | } | |
9887 | ||
9888 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9889 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9890 | { |
9891 | struct task_group *tg; | |
9892 | unsigned long flags; | |
9893 | int i; | |
9894 | ||
9895 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9896 | if (!tg) | |
9897 | return ERR_PTR(-ENOMEM); | |
9898 | ||
ec7dc8ac | 9899 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9900 | goto err; |
9901 | ||
ec7dc8ac | 9902 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9903 | goto err; |
9904 | ||
8ed36996 | 9905 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9906 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9907 | register_fair_sched_group(tg, i); |
9908 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9909 | } |
6f505b16 | 9910 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9911 | |
9912 | WARN_ON(!parent); /* root should already exist */ | |
9913 | ||
9914 | tg->parent = parent; | |
f473aa5e | 9915 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9916 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9917 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9918 | |
9b5b7751 | 9919 | return tg; |
29f59db3 SV |
9920 | |
9921 | err: | |
6f505b16 | 9922 | free_sched_group(tg); |
29f59db3 SV |
9923 | return ERR_PTR(-ENOMEM); |
9924 | } | |
9925 | ||
9b5b7751 | 9926 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9927 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9928 | { |
29f59db3 | 9929 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9930 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9931 | } |
9932 | ||
9b5b7751 | 9933 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9934 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9935 | { |
8ed36996 | 9936 | unsigned long flags; |
9b5b7751 | 9937 | int i; |
29f59db3 | 9938 | |
8ed36996 | 9939 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9940 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9941 | unregister_fair_sched_group(tg, i); |
9942 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9943 | } |
6f505b16 | 9944 | list_del_rcu(&tg->list); |
f473aa5e | 9945 | list_del_rcu(&tg->siblings); |
8ed36996 | 9946 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9947 | |
9b5b7751 | 9948 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9949 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9950 | } |
9951 | ||
9b5b7751 | 9952 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9953 | * The caller of this function should have put the task in its new group |
9954 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9955 | * reflect its new group. | |
9b5b7751 SV |
9956 | */ |
9957 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9958 | { |
9959 | int on_rq, running; | |
9960 | unsigned long flags; | |
9961 | struct rq *rq; | |
9962 | ||
9963 | rq = task_rq_lock(tsk, &flags); | |
9964 | ||
29f59db3 SV |
9965 | update_rq_clock(rq); |
9966 | ||
051a1d1a | 9967 | running = task_current(rq, tsk); |
29f59db3 SV |
9968 | on_rq = tsk->se.on_rq; |
9969 | ||
0e1f3483 | 9970 | if (on_rq) |
29f59db3 | 9971 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9972 | if (unlikely(running)) |
9973 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9974 | |
6f505b16 | 9975 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9976 | |
810b3817 PZ |
9977 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9978 | if (tsk->sched_class->moved_group) | |
9979 | tsk->sched_class->moved_group(tsk); | |
9980 | #endif | |
9981 | ||
0e1f3483 HS |
9982 | if (unlikely(running)) |
9983 | tsk->sched_class->set_curr_task(rq); | |
9984 | if (on_rq) | |
7074badb | 9985 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9986 | |
29f59db3 SV |
9987 | task_rq_unlock(rq, &flags); |
9988 | } | |
6d6bc0ad | 9989 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9990 | |
052f1dc7 | 9991 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9992 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9993 | { |
9994 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9995 | int on_rq; |
9996 | ||
29f59db3 | 9997 | on_rq = se->on_rq; |
62fb1851 | 9998 | if (on_rq) |
29f59db3 SV |
9999 | dequeue_entity(cfs_rq, se, 0); |
10000 | ||
10001 | se->load.weight = shares; | |
e05510d0 | 10002 | se->load.inv_weight = 0; |
29f59db3 | 10003 | |
62fb1851 | 10004 | if (on_rq) |
29f59db3 | 10005 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10006 | } |
62fb1851 | 10007 | |
c09595f6 PZ |
10008 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10009 | { | |
10010 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10011 | struct rq *rq = cfs_rq->rq; | |
10012 | unsigned long flags; | |
10013 | ||
10014 | spin_lock_irqsave(&rq->lock, flags); | |
10015 | __set_se_shares(se, shares); | |
10016 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
10017 | } |
10018 | ||
8ed36996 PZ |
10019 | static DEFINE_MUTEX(shares_mutex); |
10020 | ||
4cf86d77 | 10021 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10022 | { |
10023 | int i; | |
8ed36996 | 10024 | unsigned long flags; |
c61935fd | 10025 | |
ec7dc8ac DG |
10026 | /* |
10027 | * We can't change the weight of the root cgroup. | |
10028 | */ | |
10029 | if (!tg->se[0]) | |
10030 | return -EINVAL; | |
10031 | ||
18d95a28 PZ |
10032 | if (shares < MIN_SHARES) |
10033 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10034 | else if (shares > MAX_SHARES) |
10035 | shares = MAX_SHARES; | |
62fb1851 | 10036 | |
8ed36996 | 10037 | mutex_lock(&shares_mutex); |
9b5b7751 | 10038 | if (tg->shares == shares) |
5cb350ba | 10039 | goto done; |
29f59db3 | 10040 | |
8ed36996 | 10041 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10042 | for_each_possible_cpu(i) |
10043 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10044 | list_del_rcu(&tg->siblings); |
8ed36996 | 10045 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10046 | |
10047 | /* wait for any ongoing reference to this group to finish */ | |
10048 | synchronize_sched(); | |
10049 | ||
10050 | /* | |
10051 | * Now we are free to modify the group's share on each cpu | |
10052 | * w/o tripping rebalance_share or load_balance_fair. | |
10053 | */ | |
9b5b7751 | 10054 | tg->shares = shares; |
c09595f6 PZ |
10055 | for_each_possible_cpu(i) { |
10056 | /* | |
10057 | * force a rebalance | |
10058 | */ | |
10059 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10060 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10061 | } |
29f59db3 | 10062 | |
6b2d7700 SV |
10063 | /* |
10064 | * Enable load balance activity on this group, by inserting it back on | |
10065 | * each cpu's rq->leaf_cfs_rq_list. | |
10066 | */ | |
8ed36996 | 10067 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10068 | for_each_possible_cpu(i) |
10069 | register_fair_sched_group(tg, i); | |
f473aa5e | 10070 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10071 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10072 | done: |
8ed36996 | 10073 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10074 | return 0; |
29f59db3 SV |
10075 | } |
10076 | ||
5cb350ba DG |
10077 | unsigned long sched_group_shares(struct task_group *tg) |
10078 | { | |
10079 | return tg->shares; | |
10080 | } | |
052f1dc7 | 10081 | #endif |
5cb350ba | 10082 | |
052f1dc7 | 10083 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10084 | /* |
9f0c1e56 | 10085 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10086 | */ |
9f0c1e56 PZ |
10087 | static DEFINE_MUTEX(rt_constraints_mutex); |
10088 | ||
10089 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10090 | { | |
10091 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10092 | return 1ULL << 20; |
9f0c1e56 | 10093 | |
9a7e0b18 | 10094 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10095 | } |
10096 | ||
9a7e0b18 PZ |
10097 | /* Must be called with tasklist_lock held */ |
10098 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10099 | { |
9a7e0b18 | 10100 | struct task_struct *g, *p; |
b40b2e8e | 10101 | |
9a7e0b18 PZ |
10102 | do_each_thread(g, p) { |
10103 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10104 | return 1; | |
10105 | } while_each_thread(g, p); | |
b40b2e8e | 10106 | |
9a7e0b18 PZ |
10107 | return 0; |
10108 | } | |
b40b2e8e | 10109 | |
9a7e0b18 PZ |
10110 | struct rt_schedulable_data { |
10111 | struct task_group *tg; | |
10112 | u64 rt_period; | |
10113 | u64 rt_runtime; | |
10114 | }; | |
b40b2e8e | 10115 | |
9a7e0b18 PZ |
10116 | static int tg_schedulable(struct task_group *tg, void *data) |
10117 | { | |
10118 | struct rt_schedulable_data *d = data; | |
10119 | struct task_group *child; | |
10120 | unsigned long total, sum = 0; | |
10121 | u64 period, runtime; | |
b40b2e8e | 10122 | |
9a7e0b18 PZ |
10123 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10124 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10125 | |
9a7e0b18 PZ |
10126 | if (tg == d->tg) { |
10127 | period = d->rt_period; | |
10128 | runtime = d->rt_runtime; | |
b40b2e8e | 10129 | } |
b40b2e8e | 10130 | |
98a4826b PZ |
10131 | #ifdef CONFIG_USER_SCHED |
10132 | if (tg == &root_task_group) { | |
10133 | period = global_rt_period(); | |
10134 | runtime = global_rt_runtime(); | |
10135 | } | |
10136 | #endif | |
10137 | ||
4653f803 PZ |
10138 | /* |
10139 | * Cannot have more runtime than the period. | |
10140 | */ | |
10141 | if (runtime > period && runtime != RUNTIME_INF) | |
10142 | return -EINVAL; | |
6f505b16 | 10143 | |
4653f803 PZ |
10144 | /* |
10145 | * Ensure we don't starve existing RT tasks. | |
10146 | */ | |
9a7e0b18 PZ |
10147 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10148 | return -EBUSY; | |
6f505b16 | 10149 | |
9a7e0b18 | 10150 | total = to_ratio(period, runtime); |
6f505b16 | 10151 | |
4653f803 PZ |
10152 | /* |
10153 | * Nobody can have more than the global setting allows. | |
10154 | */ | |
10155 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10156 | return -EINVAL; | |
6f505b16 | 10157 | |
4653f803 PZ |
10158 | /* |
10159 | * The sum of our children's runtime should not exceed our own. | |
10160 | */ | |
9a7e0b18 PZ |
10161 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10162 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10163 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10164 | |
9a7e0b18 PZ |
10165 | if (child == d->tg) { |
10166 | period = d->rt_period; | |
10167 | runtime = d->rt_runtime; | |
10168 | } | |
6f505b16 | 10169 | |
9a7e0b18 | 10170 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10171 | } |
6f505b16 | 10172 | |
9a7e0b18 PZ |
10173 | if (sum > total) |
10174 | return -EINVAL; | |
10175 | ||
10176 | return 0; | |
6f505b16 PZ |
10177 | } |
10178 | ||
9a7e0b18 | 10179 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10180 | { |
9a7e0b18 PZ |
10181 | struct rt_schedulable_data data = { |
10182 | .tg = tg, | |
10183 | .rt_period = period, | |
10184 | .rt_runtime = runtime, | |
10185 | }; | |
10186 | ||
10187 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10188 | } |
10189 | ||
d0b27fa7 PZ |
10190 | static int tg_set_bandwidth(struct task_group *tg, |
10191 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10192 | { |
ac086bc2 | 10193 | int i, err = 0; |
9f0c1e56 | 10194 | |
9f0c1e56 | 10195 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10196 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10197 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10198 | if (err) | |
9f0c1e56 | 10199 | goto unlock; |
ac086bc2 PZ |
10200 | |
10201 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10202 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10203 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10204 | |
10205 | for_each_possible_cpu(i) { | |
10206 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10207 | ||
10208 | spin_lock(&rt_rq->rt_runtime_lock); | |
10209 | rt_rq->rt_runtime = rt_runtime; | |
10210 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10211 | } | |
10212 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10213 | unlock: |
521f1a24 | 10214 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10215 | mutex_unlock(&rt_constraints_mutex); |
10216 | ||
10217 | return err; | |
6f505b16 PZ |
10218 | } |
10219 | ||
d0b27fa7 PZ |
10220 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10221 | { | |
10222 | u64 rt_runtime, rt_period; | |
10223 | ||
10224 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10225 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10226 | if (rt_runtime_us < 0) | |
10227 | rt_runtime = RUNTIME_INF; | |
10228 | ||
10229 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10230 | } | |
10231 | ||
9f0c1e56 PZ |
10232 | long sched_group_rt_runtime(struct task_group *tg) |
10233 | { | |
10234 | u64 rt_runtime_us; | |
10235 | ||
d0b27fa7 | 10236 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10237 | return -1; |
10238 | ||
d0b27fa7 | 10239 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10240 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10241 | return rt_runtime_us; | |
10242 | } | |
d0b27fa7 PZ |
10243 | |
10244 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10245 | { | |
10246 | u64 rt_runtime, rt_period; | |
10247 | ||
10248 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10249 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10250 | ||
619b0488 R |
10251 | if (rt_period == 0) |
10252 | return -EINVAL; | |
10253 | ||
d0b27fa7 PZ |
10254 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10255 | } | |
10256 | ||
10257 | long sched_group_rt_period(struct task_group *tg) | |
10258 | { | |
10259 | u64 rt_period_us; | |
10260 | ||
10261 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10262 | do_div(rt_period_us, NSEC_PER_USEC); | |
10263 | return rt_period_us; | |
10264 | } | |
10265 | ||
10266 | static int sched_rt_global_constraints(void) | |
10267 | { | |
4653f803 | 10268 | u64 runtime, period; |
d0b27fa7 PZ |
10269 | int ret = 0; |
10270 | ||
ec5d4989 HS |
10271 | if (sysctl_sched_rt_period <= 0) |
10272 | return -EINVAL; | |
10273 | ||
4653f803 PZ |
10274 | runtime = global_rt_runtime(); |
10275 | period = global_rt_period(); | |
10276 | ||
10277 | /* | |
10278 | * Sanity check on the sysctl variables. | |
10279 | */ | |
10280 | if (runtime > period && runtime != RUNTIME_INF) | |
10281 | return -EINVAL; | |
10b612f4 | 10282 | |
d0b27fa7 | 10283 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10284 | read_lock(&tasklist_lock); |
4653f803 | 10285 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10286 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10287 | mutex_unlock(&rt_constraints_mutex); |
10288 | ||
10289 | return ret; | |
10290 | } | |
54e99124 DG |
10291 | |
10292 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10293 | { | |
10294 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10295 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10296 | return 0; | |
10297 | ||
10298 | return 1; | |
10299 | } | |
10300 | ||
6d6bc0ad | 10301 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10302 | static int sched_rt_global_constraints(void) |
10303 | { | |
ac086bc2 PZ |
10304 | unsigned long flags; |
10305 | int i; | |
10306 | ||
ec5d4989 HS |
10307 | if (sysctl_sched_rt_period <= 0) |
10308 | return -EINVAL; | |
10309 | ||
60aa605d PZ |
10310 | /* |
10311 | * There's always some RT tasks in the root group | |
10312 | * -- migration, kstopmachine etc.. | |
10313 | */ | |
10314 | if (sysctl_sched_rt_runtime == 0) | |
10315 | return -EBUSY; | |
10316 | ||
ac086bc2 PZ |
10317 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10318 | for_each_possible_cpu(i) { | |
10319 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10320 | ||
10321 | spin_lock(&rt_rq->rt_runtime_lock); | |
10322 | rt_rq->rt_runtime = global_rt_runtime(); | |
10323 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10324 | } | |
10325 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10326 | ||
d0b27fa7 PZ |
10327 | return 0; |
10328 | } | |
6d6bc0ad | 10329 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10330 | |
10331 | int sched_rt_handler(struct ctl_table *table, int write, | |
10332 | struct file *filp, void __user *buffer, size_t *lenp, | |
10333 | loff_t *ppos) | |
10334 | { | |
10335 | int ret; | |
10336 | int old_period, old_runtime; | |
10337 | static DEFINE_MUTEX(mutex); | |
10338 | ||
10339 | mutex_lock(&mutex); | |
10340 | old_period = sysctl_sched_rt_period; | |
10341 | old_runtime = sysctl_sched_rt_runtime; | |
10342 | ||
10343 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10344 | ||
10345 | if (!ret && write) { | |
10346 | ret = sched_rt_global_constraints(); | |
10347 | if (ret) { | |
10348 | sysctl_sched_rt_period = old_period; | |
10349 | sysctl_sched_rt_runtime = old_runtime; | |
10350 | } else { | |
10351 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10352 | def_rt_bandwidth.rt_period = | |
10353 | ns_to_ktime(global_rt_period()); | |
10354 | } | |
10355 | } | |
10356 | mutex_unlock(&mutex); | |
10357 | ||
10358 | return ret; | |
10359 | } | |
68318b8e | 10360 | |
052f1dc7 | 10361 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10362 | |
10363 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10364 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10365 | { |
2b01dfe3 PM |
10366 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10367 | struct task_group, css); | |
68318b8e SV |
10368 | } |
10369 | ||
10370 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10371 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10372 | { |
ec7dc8ac | 10373 | struct task_group *tg, *parent; |
68318b8e | 10374 | |
2b01dfe3 | 10375 | if (!cgrp->parent) { |
68318b8e | 10376 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10377 | return &init_task_group.css; |
10378 | } | |
10379 | ||
ec7dc8ac DG |
10380 | parent = cgroup_tg(cgrp->parent); |
10381 | tg = sched_create_group(parent); | |
68318b8e SV |
10382 | if (IS_ERR(tg)) |
10383 | return ERR_PTR(-ENOMEM); | |
10384 | ||
68318b8e SV |
10385 | return &tg->css; |
10386 | } | |
10387 | ||
41a2d6cf IM |
10388 | static void |
10389 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10390 | { |
2b01dfe3 | 10391 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10392 | |
10393 | sched_destroy_group(tg); | |
10394 | } | |
10395 | ||
41a2d6cf IM |
10396 | static int |
10397 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10398 | struct task_struct *tsk) | |
68318b8e | 10399 | { |
b68aa230 | 10400 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10401 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10402 | return -EINVAL; |
10403 | #else | |
68318b8e SV |
10404 | /* We don't support RT-tasks being in separate groups */ |
10405 | if (tsk->sched_class != &fair_sched_class) | |
10406 | return -EINVAL; | |
b68aa230 | 10407 | #endif |
68318b8e SV |
10408 | |
10409 | return 0; | |
10410 | } | |
10411 | ||
10412 | static void | |
2b01dfe3 | 10413 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10414 | struct cgroup *old_cont, struct task_struct *tsk) |
10415 | { | |
10416 | sched_move_task(tsk); | |
10417 | } | |
10418 | ||
052f1dc7 | 10419 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10420 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10421 | u64 shareval) |
68318b8e | 10422 | { |
2b01dfe3 | 10423 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10424 | } |
10425 | ||
f4c753b7 | 10426 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10427 | { |
2b01dfe3 | 10428 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10429 | |
10430 | return (u64) tg->shares; | |
10431 | } | |
6d6bc0ad | 10432 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10433 | |
052f1dc7 | 10434 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10435 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10436 | s64 val) |
6f505b16 | 10437 | { |
06ecb27c | 10438 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10439 | } |
10440 | ||
06ecb27c | 10441 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10442 | { |
06ecb27c | 10443 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10444 | } |
d0b27fa7 PZ |
10445 | |
10446 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10447 | u64 rt_period_us) | |
10448 | { | |
10449 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10450 | } | |
10451 | ||
10452 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10453 | { | |
10454 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10455 | } | |
6d6bc0ad | 10456 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10457 | |
fe5c7cc2 | 10458 | static struct cftype cpu_files[] = { |
052f1dc7 | 10459 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10460 | { |
10461 | .name = "shares", | |
f4c753b7 PM |
10462 | .read_u64 = cpu_shares_read_u64, |
10463 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10464 | }, |
052f1dc7 PZ |
10465 | #endif |
10466 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10467 | { |
9f0c1e56 | 10468 | .name = "rt_runtime_us", |
06ecb27c PM |
10469 | .read_s64 = cpu_rt_runtime_read, |
10470 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10471 | }, |
d0b27fa7 PZ |
10472 | { |
10473 | .name = "rt_period_us", | |
f4c753b7 PM |
10474 | .read_u64 = cpu_rt_period_read_uint, |
10475 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10476 | }, |
052f1dc7 | 10477 | #endif |
68318b8e SV |
10478 | }; |
10479 | ||
10480 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10481 | { | |
fe5c7cc2 | 10482 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10483 | } |
10484 | ||
10485 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10486 | .name = "cpu", |
10487 | .create = cpu_cgroup_create, | |
10488 | .destroy = cpu_cgroup_destroy, | |
10489 | .can_attach = cpu_cgroup_can_attach, | |
10490 | .attach = cpu_cgroup_attach, | |
10491 | .populate = cpu_cgroup_populate, | |
10492 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10493 | .early_init = 1, |
10494 | }; | |
10495 | ||
052f1dc7 | 10496 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10497 | |
10498 | #ifdef CONFIG_CGROUP_CPUACCT | |
10499 | ||
10500 | /* | |
10501 | * CPU accounting code for task groups. | |
10502 | * | |
10503 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10504 | * (balbir@in.ibm.com). | |
10505 | */ | |
10506 | ||
934352f2 | 10507 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10508 | struct cpuacct { |
10509 | struct cgroup_subsys_state css; | |
10510 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10511 | u64 *cpuusage; | |
ef12fefa | 10512 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10513 | struct cpuacct *parent; |
d842de87 SV |
10514 | }; |
10515 | ||
10516 | struct cgroup_subsys cpuacct_subsys; | |
10517 | ||
10518 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10519 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10520 | { |
32cd756a | 10521 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10522 | struct cpuacct, css); |
10523 | } | |
10524 | ||
10525 | /* return cpu accounting group to which this task belongs */ | |
10526 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10527 | { | |
10528 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10529 | struct cpuacct, css); | |
10530 | } | |
10531 | ||
10532 | /* create a new cpu accounting group */ | |
10533 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10534 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10535 | { |
10536 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10537 | int i; |
d842de87 SV |
10538 | |
10539 | if (!ca) | |
ef12fefa | 10540 | goto out; |
d842de87 SV |
10541 | |
10542 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10543 | if (!ca->cpuusage) |
10544 | goto out_free_ca; | |
10545 | ||
10546 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10547 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10548 | goto out_free_counters; | |
d842de87 | 10549 | |
934352f2 BR |
10550 | if (cgrp->parent) |
10551 | ca->parent = cgroup_ca(cgrp->parent); | |
10552 | ||
d842de87 | 10553 | return &ca->css; |
ef12fefa BR |
10554 | |
10555 | out_free_counters: | |
10556 | while (--i >= 0) | |
10557 | percpu_counter_destroy(&ca->cpustat[i]); | |
10558 | free_percpu(ca->cpuusage); | |
10559 | out_free_ca: | |
10560 | kfree(ca); | |
10561 | out: | |
10562 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10563 | } |
10564 | ||
10565 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10566 | static void |
32cd756a | 10567 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10568 | { |
32cd756a | 10569 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10570 | int i; |
d842de87 | 10571 | |
ef12fefa BR |
10572 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10573 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10574 | free_percpu(ca->cpuusage); |
10575 | kfree(ca); | |
10576 | } | |
10577 | ||
720f5498 KC |
10578 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10579 | { | |
b36128c8 | 10580 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10581 | u64 data; |
10582 | ||
10583 | #ifndef CONFIG_64BIT | |
10584 | /* | |
10585 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10586 | */ | |
10587 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10588 | data = *cpuusage; | |
10589 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10590 | #else | |
10591 | data = *cpuusage; | |
10592 | #endif | |
10593 | ||
10594 | return data; | |
10595 | } | |
10596 | ||
10597 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10598 | { | |
b36128c8 | 10599 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10600 | |
10601 | #ifndef CONFIG_64BIT | |
10602 | /* | |
10603 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10604 | */ | |
10605 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10606 | *cpuusage = val; | |
10607 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10608 | #else | |
10609 | *cpuusage = val; | |
10610 | #endif | |
10611 | } | |
10612 | ||
d842de87 | 10613 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10614 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10615 | { |
32cd756a | 10616 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10617 | u64 totalcpuusage = 0; |
10618 | int i; | |
10619 | ||
720f5498 KC |
10620 | for_each_present_cpu(i) |
10621 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10622 | |
10623 | return totalcpuusage; | |
10624 | } | |
10625 | ||
0297b803 DG |
10626 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10627 | u64 reset) | |
10628 | { | |
10629 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10630 | int err = 0; | |
10631 | int i; | |
10632 | ||
10633 | if (reset) { | |
10634 | err = -EINVAL; | |
10635 | goto out; | |
10636 | } | |
10637 | ||
720f5498 KC |
10638 | for_each_present_cpu(i) |
10639 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10640 | |
0297b803 DG |
10641 | out: |
10642 | return err; | |
10643 | } | |
10644 | ||
e9515c3c KC |
10645 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10646 | struct seq_file *m) | |
10647 | { | |
10648 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10649 | u64 percpu; | |
10650 | int i; | |
10651 | ||
10652 | for_each_present_cpu(i) { | |
10653 | percpu = cpuacct_cpuusage_read(ca, i); | |
10654 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10655 | } | |
10656 | seq_printf(m, "\n"); | |
10657 | return 0; | |
10658 | } | |
10659 | ||
ef12fefa BR |
10660 | static const char *cpuacct_stat_desc[] = { |
10661 | [CPUACCT_STAT_USER] = "user", | |
10662 | [CPUACCT_STAT_SYSTEM] = "system", | |
10663 | }; | |
10664 | ||
10665 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10666 | struct cgroup_map_cb *cb) | |
10667 | { | |
10668 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10669 | int i; | |
10670 | ||
10671 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10672 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10673 | val = cputime64_to_clock_t(val); | |
10674 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10675 | } | |
10676 | return 0; | |
10677 | } | |
10678 | ||
d842de87 SV |
10679 | static struct cftype files[] = { |
10680 | { | |
10681 | .name = "usage", | |
f4c753b7 PM |
10682 | .read_u64 = cpuusage_read, |
10683 | .write_u64 = cpuusage_write, | |
d842de87 | 10684 | }, |
e9515c3c KC |
10685 | { |
10686 | .name = "usage_percpu", | |
10687 | .read_seq_string = cpuacct_percpu_seq_read, | |
10688 | }, | |
ef12fefa BR |
10689 | { |
10690 | .name = "stat", | |
10691 | .read_map = cpuacct_stats_show, | |
10692 | }, | |
d842de87 SV |
10693 | }; |
10694 | ||
32cd756a | 10695 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10696 | { |
32cd756a | 10697 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10698 | } |
10699 | ||
10700 | /* | |
10701 | * charge this task's execution time to its accounting group. | |
10702 | * | |
10703 | * called with rq->lock held. | |
10704 | */ | |
10705 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10706 | { | |
10707 | struct cpuacct *ca; | |
934352f2 | 10708 | int cpu; |
d842de87 | 10709 | |
c40c6f85 | 10710 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10711 | return; |
10712 | ||
934352f2 | 10713 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10714 | |
10715 | rcu_read_lock(); | |
10716 | ||
d842de87 | 10717 | ca = task_ca(tsk); |
d842de87 | 10718 | |
934352f2 | 10719 | for (; ca; ca = ca->parent) { |
b36128c8 | 10720 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10721 | *cpuusage += cputime; |
10722 | } | |
a18b83b7 BR |
10723 | |
10724 | rcu_read_unlock(); | |
d842de87 SV |
10725 | } |
10726 | ||
ef12fefa BR |
10727 | /* |
10728 | * Charge the system/user time to the task's accounting group. | |
10729 | */ | |
10730 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10731 | enum cpuacct_stat_index idx, cputime_t val) | |
10732 | { | |
10733 | struct cpuacct *ca; | |
10734 | ||
10735 | if (unlikely(!cpuacct_subsys.active)) | |
10736 | return; | |
10737 | ||
10738 | rcu_read_lock(); | |
10739 | ca = task_ca(tsk); | |
10740 | ||
10741 | do { | |
10742 | percpu_counter_add(&ca->cpustat[idx], val); | |
10743 | ca = ca->parent; | |
10744 | } while (ca); | |
10745 | rcu_read_unlock(); | |
10746 | } | |
10747 | ||
d842de87 SV |
10748 | struct cgroup_subsys cpuacct_subsys = { |
10749 | .name = "cpuacct", | |
10750 | .create = cpuacct_create, | |
10751 | .destroy = cpuacct_destroy, | |
10752 | .populate = cpuacct_populate, | |
10753 | .subsys_id = cpuacct_subsys_id, | |
10754 | }; | |
10755 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
10756 | |
10757 | #ifndef CONFIG_SMP | |
10758 | ||
10759 | int rcu_expedited_torture_stats(char *page) | |
10760 | { | |
10761 | return 0; | |
10762 | } | |
10763 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10764 | ||
10765 | void synchronize_sched_expedited(void) | |
10766 | { | |
10767 | } | |
10768 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10769 | ||
10770 | #else /* #ifndef CONFIG_SMP */ | |
10771 | ||
10772 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
10773 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
10774 | ||
10775 | #define RCU_EXPEDITED_STATE_POST -2 | |
10776 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
10777 | ||
10778 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10779 | ||
10780 | int rcu_expedited_torture_stats(char *page) | |
10781 | { | |
10782 | int cnt = 0; | |
10783 | int cpu; | |
10784 | ||
10785 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
10786 | for_each_online_cpu(cpu) { | |
10787 | cnt += sprintf(&page[cnt], " %d:%d", | |
10788 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
10789 | } | |
10790 | cnt += sprintf(&page[cnt], "\n"); | |
10791 | return cnt; | |
10792 | } | |
10793 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10794 | ||
10795 | static long synchronize_sched_expedited_count; | |
10796 | ||
10797 | /* | |
10798 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
10799 | * approach to force grace period to end quickly. This consumes | |
10800 | * significant time on all CPUs, and is thus not recommended for | |
10801 | * any sort of common-case code. | |
10802 | * | |
10803 | * Note that it is illegal to call this function while holding any | |
10804 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
10805 | * observe this restriction will result in deadlock. | |
10806 | */ | |
10807 | void synchronize_sched_expedited(void) | |
10808 | { | |
10809 | int cpu; | |
10810 | unsigned long flags; | |
10811 | bool need_full_sync = 0; | |
10812 | struct rq *rq; | |
10813 | struct migration_req *req; | |
10814 | long snap; | |
10815 | int trycount = 0; | |
10816 | ||
10817 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
10818 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
10819 | get_online_cpus(); | |
10820 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
10821 | put_online_cpus(); | |
10822 | if (trycount++ < 10) | |
10823 | udelay(trycount * num_online_cpus()); | |
10824 | else { | |
10825 | synchronize_sched(); | |
10826 | return; | |
10827 | } | |
10828 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
10829 | smp_mb(); /* ensure test happens before caller kfree */ | |
10830 | return; | |
10831 | } | |
10832 | get_online_cpus(); | |
10833 | } | |
10834 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
10835 | for_each_online_cpu(cpu) { | |
10836 | rq = cpu_rq(cpu); | |
10837 | req = &per_cpu(rcu_migration_req, cpu); | |
10838 | init_completion(&req->done); | |
10839 | req->task = NULL; | |
10840 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
10841 | spin_lock_irqsave(&rq->lock, flags); | |
10842 | list_add(&req->list, &rq->migration_queue); | |
10843 | spin_unlock_irqrestore(&rq->lock, flags); | |
10844 | wake_up_process(rq->migration_thread); | |
10845 | } | |
10846 | for_each_online_cpu(cpu) { | |
10847 | rcu_expedited_state = cpu; | |
10848 | req = &per_cpu(rcu_migration_req, cpu); | |
10849 | rq = cpu_rq(cpu); | |
10850 | wait_for_completion(&req->done); | |
10851 | spin_lock_irqsave(&rq->lock, flags); | |
10852 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) | |
10853 | need_full_sync = 1; | |
10854 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
10855 | spin_unlock_irqrestore(&rq->lock, flags); | |
10856 | } | |
10857 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10858 | mutex_unlock(&rcu_sched_expedited_mutex); | |
10859 | put_online_cpus(); | |
10860 | if (need_full_sync) | |
10861 | synchronize_sched(); | |
10862 | } | |
10863 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10864 | ||
10865 | #endif /* #else #ifndef CONFIG_SMP */ |