]>
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 | ||
663997d4 JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
1da177e4 LT |
31 | #include <linux/mm.h> |
32 | #include <linux/module.h> | |
33 | #include <linux/nmi.h> | |
34 | #include <linux/init.h> | |
dff06c15 | 35 | #include <linux/uaccess.h> |
1da177e4 LT |
36 | #include <linux/highmem.h> |
37 | #include <linux/smp_lock.h> | |
38 | #include <asm/mmu_context.h> | |
39 | #include <linux/interrupt.h> | |
c59ede7b | 40 | #include <linux/capability.h> |
1da177e4 LT |
41 | #include <linux/completion.h> |
42 | #include <linux/kernel_stat.h> | |
9a11b49a | 43 | #include <linux/debug_locks.h> |
cdd6c482 | 44 | #include <linux/perf_event.h> |
1da177e4 LT |
45 | #include <linux/security.h> |
46 | #include <linux/notifier.h> | |
47 | #include <linux/profile.h> | |
7dfb7103 | 48 | #include <linux/freezer.h> |
198e2f18 | 49 | #include <linux/vmalloc.h> |
1da177e4 LT |
50 | #include <linux/blkdev.h> |
51 | #include <linux/delay.h> | |
b488893a | 52 | #include <linux/pid_namespace.h> |
1da177e4 LT |
53 | #include <linux/smp.h> |
54 | #include <linux/threads.h> | |
55 | #include <linux/timer.h> | |
56 | #include <linux/rcupdate.h> | |
57 | #include <linux/cpu.h> | |
58 | #include <linux/cpuset.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/kthread.h> | |
b5aadf7f | 61 | #include <linux/proc_fs.h> |
1da177e4 | 62 | #include <linux/seq_file.h> |
e692ab53 | 63 | #include <linux/sysctl.h> |
1da177e4 LT |
64 | #include <linux/syscalls.h> |
65 | #include <linux/times.h> | |
8f0ab514 | 66 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 67 | #include <linux/kprobes.h> |
0ff92245 | 68 | #include <linux/delayacct.h> |
dff06c15 | 69 | #include <linux/unistd.h> |
f5ff8422 | 70 | #include <linux/pagemap.h> |
8f4d37ec | 71 | #include <linux/hrtimer.h> |
30914a58 | 72 | #include <linux/tick.h> |
f00b45c1 PZ |
73 | #include <linux/debugfs.h> |
74 | #include <linux/ctype.h> | |
6cd8a4bb | 75 | #include <linux/ftrace.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 IM |
145 | /* nests inside the rq lock: */ |
146 | spinlock_t rt_runtime_lock; | |
147 | ktime_t rt_period; | |
148 | u64 rt_runtime; | |
149 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
150 | }; |
151 | ||
152 | static struct rt_bandwidth def_rt_bandwidth; | |
153 | ||
154 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
155 | ||
156 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
157 | { | |
158 | struct rt_bandwidth *rt_b = | |
159 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
160 | ktime_t now; | |
161 | int overrun; | |
162 | int idle = 0; | |
163 | ||
164 | for (;;) { | |
165 | now = hrtimer_cb_get_time(timer); | |
166 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
167 | ||
168 | if (!overrun) | |
169 | break; | |
170 | ||
171 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
172 | } | |
173 | ||
174 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
175 | } | |
176 | ||
177 | static | |
178 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
179 | { | |
180 | rt_b->rt_period = ns_to_ktime(period); | |
181 | rt_b->rt_runtime = runtime; | |
182 | ||
ac086bc2 PZ |
183 | spin_lock_init(&rt_b->rt_runtime_lock); |
184 | ||
d0b27fa7 PZ |
185 | hrtimer_init(&rt_b->rt_period_timer, |
186 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
187 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
188 | } |
189 | ||
c8bfff6d KH |
190 | static inline int rt_bandwidth_enabled(void) |
191 | { | |
192 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
193 | } |
194 | ||
195 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
196 | { | |
197 | ktime_t now; | |
198 | ||
cac64d00 | 199 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
200 | return; |
201 | ||
202 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
203 | return; | |
204 | ||
205 | spin_lock(&rt_b->rt_runtime_lock); | |
206 | for (;;) { | |
7f1e2ca9 PZ |
207 | unsigned long delta; |
208 | ktime_t soft, hard; | |
209 | ||
d0b27fa7 PZ |
210 | if (hrtimer_active(&rt_b->rt_period_timer)) |
211 | break; | |
212 | ||
213 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
214 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
215 | |
216 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
217 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
218 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
219 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 220 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 PZ |
221 | } |
222 | spin_unlock(&rt_b->rt_runtime_lock); | |
223 | } | |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
052f1dc7 | 238 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
052f1dc7 | 248 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
249 | struct cgroup_subsys_state css; |
250 | #endif | |
052f1dc7 | 251 | |
6c415b92 AB |
252 | #ifdef CONFIG_USER_SCHED |
253 | uid_t uid; | |
254 | #endif | |
255 | ||
052f1dc7 | 256 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
257 | /* schedulable entities of this group on each cpu */ |
258 | struct sched_entity **se; | |
259 | /* runqueue "owned" by this group on each cpu */ | |
260 | struct cfs_rq **cfs_rq; | |
261 | unsigned long shares; | |
052f1dc7 PZ |
262 | #endif |
263 | ||
264 | #ifdef CONFIG_RT_GROUP_SCHED | |
265 | struct sched_rt_entity **rt_se; | |
266 | struct rt_rq **rt_rq; | |
267 | ||
d0b27fa7 | 268 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 269 | #endif |
6b2d7700 | 270 | |
ae8393e5 | 271 | struct rcu_head rcu; |
6f505b16 | 272 | struct list_head list; |
f473aa5e PZ |
273 | |
274 | struct task_group *parent; | |
275 | struct list_head siblings; | |
276 | struct list_head children; | |
29f59db3 SV |
277 | }; |
278 | ||
354d60c2 | 279 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 280 | |
6c415b92 AB |
281 | /* Helper function to pass uid information to create_sched_user() */ |
282 | void set_tg_uid(struct user_struct *user) | |
283 | { | |
284 | user->tg->uid = user->uid; | |
285 | } | |
286 | ||
eff766a6 PZ |
287 | /* |
288 | * Root task group. | |
84e9dabf AS |
289 | * Every UID task group (including init_task_group aka UID-0) will |
290 | * be a child to this group. | |
eff766a6 PZ |
291 | */ |
292 | struct task_group root_task_group; | |
293 | ||
052f1dc7 | 294 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
295 | /* Default task group's sched entity on each cpu */ |
296 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
297 | /* Default task group's cfs_rq on each cpu */ | |
ada3fa15 | 298 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct cfs_rq, init_tg_cfs_rq); |
6d6bc0ad | 299 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
300 | |
301 | #ifdef CONFIG_RT_GROUP_SCHED | |
302 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
b9bf3121 | 303 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rt_rq, init_rt_rq); |
6d6bc0ad | 304 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 305 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 306 | #define root_task_group init_task_group |
9a7e0b18 | 307 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 308 | |
8ed36996 | 309 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
310 | * a task group's cpu shares. |
311 | */ | |
8ed36996 | 312 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 313 | |
e9036b36 CG |
314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
315 | ||
57310a98 PZ |
316 | #ifdef CONFIG_SMP |
317 | static int root_task_group_empty(void) | |
318 | { | |
319 | return list_empty(&root_task_group.children); | |
320 | } | |
321 | #endif | |
322 | ||
052f1dc7 PZ |
323 | #ifdef CONFIG_USER_SCHED |
324 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 325 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 326 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 327 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 328 | |
cb4ad1ff | 329 | /* |
2e084786 LJ |
330 | * A weight of 0 or 1 can cause arithmetics problems. |
331 | * A weight of a cfs_rq is the sum of weights of which entities | |
332 | * are queued on this cfs_rq, so a weight of a entity should not be | |
333 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
334 | * (The default weight is 1024 - so there's no practical |
335 | * limitation from this.) | |
336 | */ | |
18d95a28 | 337 | #define MIN_SHARES 2 |
2e084786 | 338 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 339 | |
052f1dc7 PZ |
340 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
341 | #endif | |
342 | ||
29f59db3 | 343 | /* Default task group. |
3a252015 | 344 | * Every task in system belong to this group at bootup. |
29f59db3 | 345 | */ |
434d53b0 | 346 | struct task_group init_task_group; |
29f59db3 SV |
347 | |
348 | /* return group to which a task belongs */ | |
4cf86d77 | 349 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 350 | { |
4cf86d77 | 351 | struct task_group *tg; |
9b5b7751 | 352 | |
052f1dc7 | 353 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
354 | rcu_read_lock(); |
355 | tg = __task_cred(p)->user->tg; | |
356 | rcu_read_unlock(); | |
052f1dc7 | 357 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
358 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
359 | struct task_group, css); | |
24e377a8 | 360 | #else |
41a2d6cf | 361 | tg = &init_task_group; |
24e377a8 | 362 | #endif |
9b5b7751 | 363 | return tg; |
29f59db3 SV |
364 | } |
365 | ||
366 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 367 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 368 | { |
052f1dc7 | 369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
370 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
371 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 372 | #endif |
6f505b16 | 373 | |
052f1dc7 | 374 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
375 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
376 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 377 | #endif |
29f59db3 SV |
378 | } |
379 | ||
380 | #else | |
381 | ||
6f505b16 | 382 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
383 | static inline struct task_group *task_group(struct task_struct *p) |
384 | { | |
385 | return NULL; | |
386 | } | |
29f59db3 | 387 | |
052f1dc7 | 388 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 389 | |
6aa645ea IM |
390 | /* CFS-related fields in a runqueue */ |
391 | struct cfs_rq { | |
392 | struct load_weight load; | |
393 | unsigned long nr_running; | |
394 | ||
6aa645ea | 395 | u64 exec_clock; |
e9acbff6 | 396 | u64 min_vruntime; |
6aa645ea IM |
397 | |
398 | struct rb_root tasks_timeline; | |
399 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
400 | |
401 | struct list_head tasks; | |
402 | struct list_head *balance_iterator; | |
403 | ||
404 | /* | |
405 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
406 | * It is set to NULL otherwise (i.e when none are currently running). |
407 | */ | |
4793241b | 408 | struct sched_entity *curr, *next, *last; |
ddc97297 | 409 | |
5ac5c4d6 | 410 | unsigned int nr_spread_over; |
ddc97297 | 411 | |
62160e3f | 412 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
413 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
414 | ||
41a2d6cf IM |
415 | /* |
416 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
417 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
418 | * (like users, containers etc.) | |
419 | * | |
420 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
421 | * list is used during load balance. | |
422 | */ | |
41a2d6cf IM |
423 | struct list_head leaf_cfs_rq_list; |
424 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
425 | |
426 | #ifdef CONFIG_SMP | |
c09595f6 | 427 | /* |
c8cba857 | 428 | * the part of load.weight contributed by tasks |
c09595f6 | 429 | */ |
c8cba857 | 430 | unsigned long task_weight; |
c09595f6 | 431 | |
c8cba857 PZ |
432 | /* |
433 | * h_load = weight * f(tg) | |
434 | * | |
435 | * Where f(tg) is the recursive weight fraction assigned to | |
436 | * this group. | |
437 | */ | |
438 | unsigned long h_load; | |
c09595f6 | 439 | |
c8cba857 PZ |
440 | /* |
441 | * this cpu's part of tg->shares | |
442 | */ | |
443 | unsigned long shares; | |
f1d239f7 PZ |
444 | |
445 | /* | |
446 | * load.weight at the time we set shares | |
447 | */ | |
448 | unsigned long rq_weight; | |
c09595f6 | 449 | #endif |
6aa645ea IM |
450 | #endif |
451 | }; | |
1da177e4 | 452 | |
6aa645ea IM |
453 | /* Real-Time classes' related field in a runqueue: */ |
454 | struct rt_rq { | |
455 | struct rt_prio_array active; | |
63489e45 | 456 | unsigned long rt_nr_running; |
052f1dc7 | 457 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
458 | struct { |
459 | int curr; /* highest queued rt task prio */ | |
398a153b | 460 | #ifdef CONFIG_SMP |
e864c499 | 461 | int next; /* next highest */ |
398a153b | 462 | #endif |
e864c499 | 463 | } highest_prio; |
6f505b16 | 464 | #endif |
fa85ae24 | 465 | #ifdef CONFIG_SMP |
73fe6aae | 466 | unsigned long rt_nr_migratory; |
a1ba4d8b | 467 | unsigned long rt_nr_total; |
a22d7fc1 | 468 | int overloaded; |
917b627d | 469 | struct plist_head pushable_tasks; |
fa85ae24 | 470 | #endif |
6f505b16 | 471 | int rt_throttled; |
fa85ae24 | 472 | u64 rt_time; |
ac086bc2 | 473 | u64 rt_runtime; |
ea736ed5 | 474 | /* Nests inside the rq lock: */ |
ac086bc2 | 475 | spinlock_t rt_runtime_lock; |
6f505b16 | 476 | |
052f1dc7 | 477 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
478 | unsigned long rt_nr_boosted; |
479 | ||
6f505b16 PZ |
480 | struct rq *rq; |
481 | struct list_head leaf_rt_rq_list; | |
482 | struct task_group *tg; | |
483 | struct sched_rt_entity *rt_se; | |
484 | #endif | |
6aa645ea IM |
485 | }; |
486 | ||
57d885fe GH |
487 | #ifdef CONFIG_SMP |
488 | ||
489 | /* | |
490 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
491 | * variables. Each exclusive cpuset essentially defines an island domain by |
492 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
493 | * exclusive cpuset is created, we also create and attach a new root-domain |
494 | * object. | |
495 | * | |
57d885fe GH |
496 | */ |
497 | struct root_domain { | |
498 | atomic_t refcount; | |
c6c4927b RR |
499 | cpumask_var_t span; |
500 | cpumask_var_t online; | |
637f5085 | 501 | |
0eab9146 | 502 | /* |
637f5085 GH |
503 | * The "RT overload" flag: it gets set if a CPU has more than |
504 | * one runnable RT task. | |
505 | */ | |
c6c4927b | 506 | cpumask_var_t rto_mask; |
0eab9146 | 507 | atomic_t rto_count; |
6e0534f2 GH |
508 | #ifdef CONFIG_SMP |
509 | struct cpupri cpupri; | |
510 | #endif | |
57d885fe GH |
511 | }; |
512 | ||
dc938520 GH |
513 | /* |
514 | * By default the system creates a single root-domain with all cpus as | |
515 | * members (mimicking the global state we have today). | |
516 | */ | |
57d885fe GH |
517 | static struct root_domain def_root_domain; |
518 | ||
519 | #endif | |
520 | ||
1da177e4 LT |
521 | /* |
522 | * This is the main, per-CPU runqueue data structure. | |
523 | * | |
524 | * Locking rule: those places that want to lock multiple runqueues | |
525 | * (such as the load balancing or the thread migration code), lock | |
526 | * acquire operations must be ordered by ascending &runqueue. | |
527 | */ | |
70b97a7f | 528 | struct rq { |
d8016491 IM |
529 | /* runqueue lock: */ |
530 | spinlock_t lock; | |
1da177e4 LT |
531 | |
532 | /* | |
533 | * nr_running and cpu_load should be in the same cacheline because | |
534 | * remote CPUs use both these fields when doing load calculation. | |
535 | */ | |
536 | unsigned long nr_running; | |
6aa645ea IM |
537 | #define CPU_LOAD_IDX_MAX 5 |
538 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c SS |
539 | #ifdef CONFIG_NO_HZ |
540 | unsigned char in_nohz_recently; | |
541 | #endif | |
d8016491 IM |
542 | /* capture load from *all* tasks on this cpu: */ |
543 | struct load_weight load; | |
6aa645ea IM |
544 | unsigned long nr_load_updates; |
545 | u64 nr_switches; | |
546 | ||
547 | struct cfs_rq cfs; | |
6f505b16 | 548 | struct rt_rq rt; |
6f505b16 | 549 | |
6aa645ea | 550 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
551 | /* list of leaf cfs_rq on this cpu: */ |
552 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
553 | #endif |
554 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 555 | struct list_head leaf_rt_rq_list; |
1da177e4 | 556 | #endif |
1da177e4 LT |
557 | |
558 | /* | |
559 | * This is part of a global counter where only the total sum | |
560 | * over all CPUs matters. A task can increase this counter on | |
561 | * one CPU and if it got migrated afterwards it may decrease | |
562 | * it on another CPU. Always updated under the runqueue lock: | |
563 | */ | |
564 | unsigned long nr_uninterruptible; | |
565 | ||
36c8b586 | 566 | struct task_struct *curr, *idle; |
c9819f45 | 567 | unsigned long next_balance; |
1da177e4 | 568 | struct mm_struct *prev_mm; |
6aa645ea | 569 | |
3e51f33f | 570 | u64 clock; |
6aa645ea | 571 | |
1da177e4 LT |
572 | atomic_t nr_iowait; |
573 | ||
574 | #ifdef CONFIG_SMP | |
0eab9146 | 575 | struct root_domain *rd; |
1da177e4 LT |
576 | struct sched_domain *sd; |
577 | ||
a0a522ce | 578 | unsigned char idle_at_tick; |
1da177e4 | 579 | /* For active balancing */ |
3f029d3c | 580 | int post_schedule; |
1da177e4 LT |
581 | int active_balance; |
582 | int push_cpu; | |
d8016491 IM |
583 | /* cpu of this runqueue: */ |
584 | int cpu; | |
1f11eb6a | 585 | int online; |
1da177e4 | 586 | |
a8a51d5e | 587 | unsigned long avg_load_per_task; |
1da177e4 | 588 | |
36c8b586 | 589 | struct task_struct *migration_thread; |
1da177e4 | 590 | struct list_head migration_queue; |
e9e9250b PZ |
591 | |
592 | u64 rt_avg; | |
593 | u64 age_stamp; | |
1b9508f6 MG |
594 | u64 idle_stamp; |
595 | u64 avg_idle; | |
1da177e4 LT |
596 | #endif |
597 | ||
dce48a84 TG |
598 | /* calc_load related fields */ |
599 | unsigned long calc_load_update; | |
600 | long calc_load_active; | |
601 | ||
8f4d37ec | 602 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
603 | #ifdef CONFIG_SMP |
604 | int hrtick_csd_pending; | |
605 | struct call_single_data hrtick_csd; | |
606 | #endif | |
8f4d37ec PZ |
607 | struct hrtimer hrtick_timer; |
608 | #endif | |
609 | ||
1da177e4 LT |
610 | #ifdef CONFIG_SCHEDSTATS |
611 | /* latency stats */ | |
612 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
613 | unsigned long long rq_cpu_time; |
614 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
615 | |
616 | /* sys_sched_yield() stats */ | |
480b9434 | 617 | unsigned int yld_count; |
1da177e4 LT |
618 | |
619 | /* schedule() stats */ | |
480b9434 KC |
620 | unsigned int sched_switch; |
621 | unsigned int sched_count; | |
622 | unsigned int sched_goidle; | |
1da177e4 LT |
623 | |
624 | /* try_to_wake_up() stats */ | |
480b9434 KC |
625 | unsigned int ttwu_count; |
626 | unsigned int ttwu_local; | |
b8efb561 IM |
627 | |
628 | /* BKL stats */ | |
480b9434 | 629 | unsigned int bkl_count; |
1da177e4 LT |
630 | #endif |
631 | }; | |
632 | ||
f34e3b61 | 633 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 634 | |
7d478721 PZ |
635 | static inline |
636 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 637 | { |
7d478721 | 638 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
dd41f596 IM |
639 | } |
640 | ||
0a2966b4 CL |
641 | static inline int cpu_of(struct rq *rq) |
642 | { | |
643 | #ifdef CONFIG_SMP | |
644 | return rq->cpu; | |
645 | #else | |
646 | return 0; | |
647 | #endif | |
648 | } | |
649 | ||
674311d5 NP |
650 | /* |
651 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 652 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
653 | * |
654 | * The domain tree of any CPU may only be accessed from within | |
655 | * preempt-disabled sections. | |
656 | */ | |
48f24c4d IM |
657 | #define for_each_domain(cpu, __sd) \ |
658 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
659 | |
660 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
661 | #define this_rq() (&__get_cpu_var(runqueues)) | |
662 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
663 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 664 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 665 | |
aa9c4c0f | 666 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
667 | { |
668 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
669 | } | |
670 | ||
bf5c91ba IM |
671 | /* |
672 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
673 | */ | |
674 | #ifdef CONFIG_SCHED_DEBUG | |
675 | # define const_debug __read_mostly | |
676 | #else | |
677 | # define const_debug static const | |
678 | #endif | |
679 | ||
017730c1 IM |
680 | /** |
681 | * runqueue_is_locked | |
e17b38bf | 682 | * @cpu: the processor in question. |
017730c1 IM |
683 | * |
684 | * Returns true if the current cpu runqueue is locked. | |
685 | * This interface allows printk to be called with the runqueue lock | |
686 | * held and know whether or not it is OK to wake up the klogd. | |
687 | */ | |
89f19f04 | 688 | int runqueue_is_locked(int cpu) |
017730c1 | 689 | { |
89f19f04 | 690 | return spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
691 | } |
692 | ||
bf5c91ba IM |
693 | /* |
694 | * Debugging: various feature bits | |
695 | */ | |
f00b45c1 PZ |
696 | |
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | __SCHED_FEAT_##name , | |
699 | ||
bf5c91ba | 700 | enum { |
f00b45c1 | 701 | #include "sched_features.h" |
bf5c91ba IM |
702 | }; |
703 | ||
f00b45c1 PZ |
704 | #undef SCHED_FEAT |
705 | ||
706 | #define SCHED_FEAT(name, enabled) \ | |
707 | (1UL << __SCHED_FEAT_##name) * enabled | | |
708 | ||
bf5c91ba | 709 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
710 | #include "sched_features.h" |
711 | 0; | |
712 | ||
713 | #undef SCHED_FEAT | |
714 | ||
715 | #ifdef CONFIG_SCHED_DEBUG | |
716 | #define SCHED_FEAT(name, enabled) \ | |
717 | #name , | |
718 | ||
983ed7a6 | 719 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
720 | #include "sched_features.h" |
721 | NULL | |
722 | }; | |
723 | ||
724 | #undef SCHED_FEAT | |
725 | ||
34f3a814 | 726 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 727 | { |
f00b45c1 PZ |
728 | int i; |
729 | ||
730 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
731 | if (!(sysctl_sched_features & (1UL << i))) |
732 | seq_puts(m, "NO_"); | |
733 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 734 | } |
34f3a814 | 735 | seq_puts(m, "\n"); |
f00b45c1 | 736 | |
34f3a814 | 737 | return 0; |
f00b45c1 PZ |
738 | } |
739 | ||
740 | static ssize_t | |
741 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
742 | size_t cnt, loff_t *ppos) | |
743 | { | |
744 | char buf[64]; | |
745 | char *cmp = buf; | |
746 | int neg = 0; | |
747 | int i; | |
748 | ||
749 | if (cnt > 63) | |
750 | cnt = 63; | |
751 | ||
752 | if (copy_from_user(&buf, ubuf, cnt)) | |
753 | return -EFAULT; | |
754 | ||
755 | buf[cnt] = 0; | |
756 | ||
c24b7c52 | 757 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
758 | neg = 1; |
759 | cmp += 3; | |
760 | } | |
761 | ||
762 | for (i = 0; sched_feat_names[i]; i++) { | |
763 | int len = strlen(sched_feat_names[i]); | |
764 | ||
765 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
766 | if (neg) | |
767 | sysctl_sched_features &= ~(1UL << i); | |
768 | else | |
769 | sysctl_sched_features |= (1UL << i); | |
770 | break; | |
771 | } | |
772 | } | |
773 | ||
774 | if (!sched_feat_names[i]) | |
775 | return -EINVAL; | |
776 | ||
42994724 | 777 | *ppos += cnt; |
f00b45c1 PZ |
778 | |
779 | return cnt; | |
780 | } | |
781 | ||
34f3a814 LZ |
782 | static int sched_feat_open(struct inode *inode, struct file *filp) |
783 | { | |
784 | return single_open(filp, sched_feat_show, NULL); | |
785 | } | |
786 | ||
828c0950 | 787 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
788 | .open = sched_feat_open, |
789 | .write = sched_feat_write, | |
790 | .read = seq_read, | |
791 | .llseek = seq_lseek, | |
792 | .release = single_release, | |
f00b45c1 PZ |
793 | }; |
794 | ||
795 | static __init int sched_init_debug(void) | |
796 | { | |
f00b45c1 PZ |
797 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
798 | &sched_feat_fops); | |
799 | ||
800 | return 0; | |
801 | } | |
802 | late_initcall(sched_init_debug); | |
803 | ||
804 | #endif | |
805 | ||
806 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 807 | |
b82d9fdd PZ |
808 | /* |
809 | * Number of tasks to iterate in a single balance run. | |
810 | * Limited because this is done with IRQs disabled. | |
811 | */ | |
812 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
813 | ||
2398f2c6 PZ |
814 | /* |
815 | * ratelimit for updating the group shares. | |
55cd5340 | 816 | * default: 0.25ms |
2398f2c6 | 817 | */ |
55cd5340 | 818 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 819 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 820 | |
ffda12a1 PZ |
821 | /* |
822 | * Inject some fuzzyness into changing the per-cpu group shares | |
823 | * this avoids remote rq-locks at the expense of fairness. | |
824 | * default: 4 | |
825 | */ | |
826 | unsigned int sysctl_sched_shares_thresh = 4; | |
827 | ||
e9e9250b PZ |
828 | /* |
829 | * period over which we average the RT time consumption, measured | |
830 | * in ms. | |
831 | * | |
832 | * default: 1s | |
833 | */ | |
834 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
835 | ||
fa85ae24 | 836 | /* |
9f0c1e56 | 837 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
838 | * default: 1s |
839 | */ | |
9f0c1e56 | 840 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 841 | |
6892b75e IM |
842 | static __read_mostly int scheduler_running; |
843 | ||
9f0c1e56 PZ |
844 | /* |
845 | * part of the period that we allow rt tasks to run in us. | |
846 | * default: 0.95s | |
847 | */ | |
848 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 849 | |
d0b27fa7 PZ |
850 | static inline u64 global_rt_period(void) |
851 | { | |
852 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
853 | } | |
854 | ||
855 | static inline u64 global_rt_runtime(void) | |
856 | { | |
e26873bb | 857 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
858 | return RUNTIME_INF; |
859 | ||
860 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
861 | } | |
fa85ae24 | 862 | |
1da177e4 | 863 | #ifndef prepare_arch_switch |
4866cde0 NP |
864 | # define prepare_arch_switch(next) do { } while (0) |
865 | #endif | |
866 | #ifndef finish_arch_switch | |
867 | # define finish_arch_switch(prev) do { } while (0) | |
868 | #endif | |
869 | ||
051a1d1a DA |
870 | static inline int task_current(struct rq *rq, struct task_struct *p) |
871 | { | |
872 | return rq->curr == p; | |
873 | } | |
874 | ||
4866cde0 | 875 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 876 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 877 | { |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | } |
880 | ||
70b97a7f | 881 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
882 | { |
883 | } | |
884 | ||
70b97a7f | 885 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 886 | { |
da04c035 IM |
887 | #ifdef CONFIG_DEBUG_SPINLOCK |
888 | /* this is a valid case when another task releases the spinlock */ | |
889 | rq->lock.owner = current; | |
890 | #endif | |
8a25d5de IM |
891 | /* |
892 | * If we are tracking spinlock dependencies then we have to | |
893 | * fix up the runqueue lock - which gets 'carried over' from | |
894 | * prev into current: | |
895 | */ | |
896 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
897 | ||
4866cde0 NP |
898 | spin_unlock_irq(&rq->lock); |
899 | } | |
900 | ||
901 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 902 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
903 | { |
904 | #ifdef CONFIG_SMP | |
905 | return p->oncpu; | |
906 | #else | |
051a1d1a | 907 | return task_current(rq, p); |
4866cde0 NP |
908 | #endif |
909 | } | |
910 | ||
70b97a7f | 911 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
912 | { |
913 | #ifdef CONFIG_SMP | |
914 | /* | |
915 | * We can optimise this out completely for !SMP, because the | |
916 | * SMP rebalancing from interrupt is the only thing that cares | |
917 | * here. | |
918 | */ | |
919 | next->oncpu = 1; | |
920 | #endif | |
921 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
922 | spin_unlock_irq(&rq->lock); | |
923 | #else | |
924 | spin_unlock(&rq->lock); | |
925 | #endif | |
926 | } | |
927 | ||
70b97a7f | 928 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
929 | { |
930 | #ifdef CONFIG_SMP | |
931 | /* | |
932 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
933 | * We must ensure this doesn't happen until the switch is completely | |
934 | * finished. | |
935 | */ | |
936 | smp_wmb(); | |
937 | prev->oncpu = 0; | |
938 | #endif | |
939 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
940 | local_irq_enable(); | |
1da177e4 | 941 | #endif |
4866cde0 NP |
942 | } |
943 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 944 | |
b29739f9 IM |
945 | /* |
946 | * __task_rq_lock - lock the runqueue a given task resides on. | |
947 | * Must be called interrupts disabled. | |
948 | */ | |
70b97a7f | 949 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
950 | __acquires(rq->lock) |
951 | { | |
3a5c359a AK |
952 | for (;;) { |
953 | struct rq *rq = task_rq(p); | |
954 | spin_lock(&rq->lock); | |
955 | if (likely(rq == task_rq(p))) | |
956 | return rq; | |
b29739f9 | 957 | spin_unlock(&rq->lock); |
b29739f9 | 958 | } |
b29739f9 IM |
959 | } |
960 | ||
1da177e4 LT |
961 | /* |
962 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 963 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
964 | * explicitly disabling preemption. |
965 | */ | |
70b97a7f | 966 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
967 | __acquires(rq->lock) |
968 | { | |
70b97a7f | 969 | struct rq *rq; |
1da177e4 | 970 | |
3a5c359a AK |
971 | for (;;) { |
972 | local_irq_save(*flags); | |
973 | rq = task_rq(p); | |
974 | spin_lock(&rq->lock); | |
975 | if (likely(rq == task_rq(p))) | |
976 | return rq; | |
1da177e4 | 977 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 978 | } |
1da177e4 LT |
979 | } |
980 | ||
ad474cac ON |
981 | void task_rq_unlock_wait(struct task_struct *p) |
982 | { | |
983 | struct rq *rq = task_rq(p); | |
984 | ||
985 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
986 | spin_unlock_wait(&rq->lock); | |
987 | } | |
988 | ||
a9957449 | 989 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
990 | __releases(rq->lock) |
991 | { | |
992 | spin_unlock(&rq->lock); | |
993 | } | |
994 | ||
70b97a7f | 995 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
996 | __releases(rq->lock) |
997 | { | |
998 | spin_unlock_irqrestore(&rq->lock, *flags); | |
999 | } | |
1000 | ||
1da177e4 | 1001 | /* |
cc2a73b5 | 1002 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1003 | */ |
a9957449 | 1004 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1005 | __acquires(rq->lock) |
1006 | { | |
70b97a7f | 1007 | struct rq *rq; |
1da177e4 LT |
1008 | |
1009 | local_irq_disable(); | |
1010 | rq = this_rq(); | |
1011 | spin_lock(&rq->lock); | |
1012 | ||
1013 | return rq; | |
1014 | } | |
1015 | ||
8f4d37ec PZ |
1016 | #ifdef CONFIG_SCHED_HRTICK |
1017 | /* | |
1018 | * Use HR-timers to deliver accurate preemption points. | |
1019 | * | |
1020 | * Its all a bit involved since we cannot program an hrt while holding the | |
1021 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1022 | * reschedule event. | |
1023 | * | |
1024 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1025 | * rq->lock. | |
1026 | */ | |
8f4d37ec PZ |
1027 | |
1028 | /* | |
1029 | * Use hrtick when: | |
1030 | * - enabled by features | |
1031 | * - hrtimer is actually high res | |
1032 | */ | |
1033 | static inline int hrtick_enabled(struct rq *rq) | |
1034 | { | |
1035 | if (!sched_feat(HRTICK)) | |
1036 | return 0; | |
ba42059f | 1037 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1038 | return 0; |
8f4d37ec PZ |
1039 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1040 | } | |
1041 | ||
8f4d37ec PZ |
1042 | static void hrtick_clear(struct rq *rq) |
1043 | { | |
1044 | if (hrtimer_active(&rq->hrtick_timer)) | |
1045 | hrtimer_cancel(&rq->hrtick_timer); | |
1046 | } | |
1047 | ||
8f4d37ec PZ |
1048 | /* |
1049 | * High-resolution timer tick. | |
1050 | * Runs from hardirq context with interrupts disabled. | |
1051 | */ | |
1052 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1053 | { | |
1054 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1055 | ||
1056 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1057 | ||
1058 | spin_lock(&rq->lock); | |
3e51f33f | 1059 | update_rq_clock(rq); |
8f4d37ec PZ |
1060 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1061 | spin_unlock(&rq->lock); | |
1062 | ||
1063 | return HRTIMER_NORESTART; | |
1064 | } | |
1065 | ||
95e904c7 | 1066 | #ifdef CONFIG_SMP |
31656519 PZ |
1067 | /* |
1068 | * called from hardirq (IPI) context | |
1069 | */ | |
1070 | static void __hrtick_start(void *arg) | |
b328ca18 | 1071 | { |
31656519 | 1072 | struct rq *rq = arg; |
b328ca18 | 1073 | |
31656519 PZ |
1074 | spin_lock(&rq->lock); |
1075 | hrtimer_restart(&rq->hrtick_timer); | |
1076 | rq->hrtick_csd_pending = 0; | |
1077 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1078 | } |
1079 | ||
31656519 PZ |
1080 | /* |
1081 | * Called to set the hrtick timer state. | |
1082 | * | |
1083 | * called with rq->lock held and irqs disabled | |
1084 | */ | |
1085 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1086 | { |
31656519 PZ |
1087 | struct hrtimer *timer = &rq->hrtick_timer; |
1088 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1089 | |
cc584b21 | 1090 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1091 | |
1092 | if (rq == this_rq()) { | |
1093 | hrtimer_restart(timer); | |
1094 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1095 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1096 | rq->hrtick_csd_pending = 1; |
1097 | } | |
b328ca18 PZ |
1098 | } |
1099 | ||
1100 | static int | |
1101 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1102 | { | |
1103 | int cpu = (int)(long)hcpu; | |
1104 | ||
1105 | switch (action) { | |
1106 | case CPU_UP_CANCELED: | |
1107 | case CPU_UP_CANCELED_FROZEN: | |
1108 | case CPU_DOWN_PREPARE: | |
1109 | case CPU_DOWN_PREPARE_FROZEN: | |
1110 | case CPU_DEAD: | |
1111 | case CPU_DEAD_FROZEN: | |
31656519 | 1112 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1113 | return NOTIFY_OK; |
1114 | } | |
1115 | ||
1116 | return NOTIFY_DONE; | |
1117 | } | |
1118 | ||
fa748203 | 1119 | static __init void init_hrtick(void) |
b328ca18 PZ |
1120 | { |
1121 | hotcpu_notifier(hotplug_hrtick, 0); | |
1122 | } | |
31656519 PZ |
1123 | #else |
1124 | /* | |
1125 | * Called to set the hrtick timer state. | |
1126 | * | |
1127 | * called with rq->lock held and irqs disabled | |
1128 | */ | |
1129 | static void hrtick_start(struct rq *rq, u64 delay) | |
1130 | { | |
7f1e2ca9 | 1131 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1132 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1133 | } |
b328ca18 | 1134 | |
006c75f1 | 1135 | static inline void init_hrtick(void) |
8f4d37ec | 1136 | { |
8f4d37ec | 1137 | } |
31656519 | 1138 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1139 | |
31656519 | 1140 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1141 | { |
31656519 PZ |
1142 | #ifdef CONFIG_SMP |
1143 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1144 | |
31656519 PZ |
1145 | rq->hrtick_csd.flags = 0; |
1146 | rq->hrtick_csd.func = __hrtick_start; | |
1147 | rq->hrtick_csd.info = rq; | |
1148 | #endif | |
8f4d37ec | 1149 | |
31656519 PZ |
1150 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1151 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1152 | } |
006c75f1 | 1153 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1154 | static inline void hrtick_clear(struct rq *rq) |
1155 | { | |
1156 | } | |
1157 | ||
8f4d37ec PZ |
1158 | static inline void init_rq_hrtick(struct rq *rq) |
1159 | { | |
1160 | } | |
1161 | ||
b328ca18 PZ |
1162 | static inline void init_hrtick(void) |
1163 | { | |
1164 | } | |
006c75f1 | 1165 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1166 | |
c24d20db IM |
1167 | /* |
1168 | * resched_task - mark a task 'to be rescheduled now'. | |
1169 | * | |
1170 | * On UP this means the setting of the need_resched flag, on SMP it | |
1171 | * might also involve a cross-CPU call to trigger the scheduler on | |
1172 | * the target CPU. | |
1173 | */ | |
1174 | #ifdef CONFIG_SMP | |
1175 | ||
1176 | #ifndef tsk_is_polling | |
1177 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1178 | #endif | |
1179 | ||
31656519 | 1180 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1181 | { |
1182 | int cpu; | |
1183 | ||
1184 | assert_spin_locked(&task_rq(p)->lock); | |
1185 | ||
5ed0cec0 | 1186 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1187 | return; |
1188 | ||
5ed0cec0 | 1189 | set_tsk_need_resched(p); |
c24d20db IM |
1190 | |
1191 | cpu = task_cpu(p); | |
1192 | if (cpu == smp_processor_id()) | |
1193 | return; | |
1194 | ||
1195 | /* NEED_RESCHED must be visible before we test polling */ | |
1196 | smp_mb(); | |
1197 | if (!tsk_is_polling(p)) | |
1198 | smp_send_reschedule(cpu); | |
1199 | } | |
1200 | ||
1201 | static void resched_cpu(int cpu) | |
1202 | { | |
1203 | struct rq *rq = cpu_rq(cpu); | |
1204 | unsigned long flags; | |
1205 | ||
1206 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1207 | return; | |
1208 | resched_task(cpu_curr(cpu)); | |
1209 | spin_unlock_irqrestore(&rq->lock, flags); | |
1210 | } | |
06d8308c TG |
1211 | |
1212 | #ifdef CONFIG_NO_HZ | |
1213 | /* | |
1214 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1215 | * idle CPU then this timer might expire before the next timer event | |
1216 | * which is scheduled to wake up that CPU. In case of a completely | |
1217 | * idle system the next event might even be infinite time into the | |
1218 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1219 | * leaves the inner idle loop so the newly added timer is taken into | |
1220 | * account when the CPU goes back to idle and evaluates the timer | |
1221 | * wheel for the next timer event. | |
1222 | */ | |
1223 | void wake_up_idle_cpu(int cpu) | |
1224 | { | |
1225 | struct rq *rq = cpu_rq(cpu); | |
1226 | ||
1227 | if (cpu == smp_processor_id()) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * This is safe, as this function is called with the timer | |
1232 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1233 | * to idle and has not yet set rq->curr to idle then it will | |
1234 | * be serialized on the timer wheel base lock and take the new | |
1235 | * timer into account automatically. | |
1236 | */ | |
1237 | if (rq->curr != rq->idle) | |
1238 | return; | |
1239 | ||
1240 | /* | |
1241 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1242 | * lockless. The worst case is that the other CPU runs the | |
1243 | * idle task through an additional NOOP schedule() | |
1244 | */ | |
5ed0cec0 | 1245 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1246 | |
1247 | /* NEED_RESCHED must be visible before we test polling */ | |
1248 | smp_mb(); | |
1249 | if (!tsk_is_polling(rq->idle)) | |
1250 | smp_send_reschedule(cpu); | |
1251 | } | |
6d6bc0ad | 1252 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1253 | |
e9e9250b PZ |
1254 | static u64 sched_avg_period(void) |
1255 | { | |
1256 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1257 | } | |
1258 | ||
1259 | static void sched_avg_update(struct rq *rq) | |
1260 | { | |
1261 | s64 period = sched_avg_period(); | |
1262 | ||
1263 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1264 | rq->age_stamp += period; | |
1265 | rq->rt_avg /= 2; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1270 | { | |
1271 | rq->rt_avg += rt_delta; | |
1272 | sched_avg_update(rq); | |
1273 | } | |
1274 | ||
6d6bc0ad | 1275 | #else /* !CONFIG_SMP */ |
31656519 | 1276 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1277 | { |
1278 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1279 | set_tsk_need_resched(p); |
c24d20db | 1280 | } |
e9e9250b PZ |
1281 | |
1282 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1283 | { | |
1284 | } | |
6d6bc0ad | 1285 | #endif /* CONFIG_SMP */ |
c24d20db | 1286 | |
45bf76df IM |
1287 | #if BITS_PER_LONG == 32 |
1288 | # define WMULT_CONST (~0UL) | |
1289 | #else | |
1290 | # define WMULT_CONST (1UL << 32) | |
1291 | #endif | |
1292 | ||
1293 | #define WMULT_SHIFT 32 | |
1294 | ||
194081eb IM |
1295 | /* |
1296 | * Shift right and round: | |
1297 | */ | |
cf2ab469 | 1298 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1299 | |
a7be37ac PZ |
1300 | /* |
1301 | * delta *= weight / lw | |
1302 | */ | |
cb1c4fc9 | 1303 | static unsigned long |
45bf76df IM |
1304 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1305 | struct load_weight *lw) | |
1306 | { | |
1307 | u64 tmp; | |
1308 | ||
7a232e03 LJ |
1309 | if (!lw->inv_weight) { |
1310 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1311 | lw->inv_weight = 1; | |
1312 | else | |
1313 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1314 | / (lw->weight+1); | |
1315 | } | |
45bf76df IM |
1316 | |
1317 | tmp = (u64)delta_exec * weight; | |
1318 | /* | |
1319 | * Check whether we'd overflow the 64-bit multiplication: | |
1320 | */ | |
194081eb | 1321 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1322 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1323 | WMULT_SHIFT/2); |
1324 | else | |
cf2ab469 | 1325 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1326 | |
ecf691da | 1327 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1331 | { |
1332 | lw->weight += inc; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1337 | { |
1338 | lw->weight -= dec; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
2dd73a4f PW |
1342 | /* |
1343 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1344 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1345 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1346 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1347 | * scaled version of the new time slice allocation that they receive on time |
1348 | * slice expiry etc. | |
1349 | */ | |
1350 | ||
cce7ade8 PZ |
1351 | #define WEIGHT_IDLEPRIO 3 |
1352 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1353 | |
1354 | /* | |
1355 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1356 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1357 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1358 | * that remained on nice 0. | |
1359 | * | |
1360 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1361 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1362 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1363 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1364 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1365 | */ |
1366 | static const int prio_to_weight[40] = { | |
254753dc IM |
1367 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1368 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1369 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1370 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1371 | /* 0 */ 1024, 820, 655, 526, 423, | |
1372 | /* 5 */ 335, 272, 215, 172, 137, | |
1373 | /* 10 */ 110, 87, 70, 56, 45, | |
1374 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1375 | }; |
1376 | ||
5714d2de IM |
1377 | /* |
1378 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1379 | * | |
1380 | * In cases where the weight does not change often, we can use the | |
1381 | * precalculated inverse to speed up arithmetics by turning divisions | |
1382 | * into multiplications: | |
1383 | */ | |
dd41f596 | 1384 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1385 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1386 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1387 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1388 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1389 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1390 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1391 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1392 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1393 | }; |
2dd73a4f | 1394 | |
dd41f596 IM |
1395 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1396 | ||
1397 | /* | |
1398 | * runqueue iterator, to support SMP load-balancing between different | |
1399 | * scheduling classes, without having to expose their internal data | |
1400 | * structures to the load-balancing proper: | |
1401 | */ | |
1402 | struct rq_iterator { | |
1403 | void *arg; | |
1404 | struct task_struct *(*start)(void *); | |
1405 | struct task_struct *(*next)(void *); | |
1406 | }; | |
1407 | ||
e1d1484f PW |
1408 | #ifdef CONFIG_SMP |
1409 | static unsigned long | |
1410 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1411 | unsigned long max_load_move, struct sched_domain *sd, | |
1412 | enum cpu_idle_type idle, int *all_pinned, | |
1413 | int *this_best_prio, struct rq_iterator *iterator); | |
1414 | ||
1415 | static int | |
1416 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1417 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1418 | struct rq_iterator *iterator); | |
e1d1484f | 1419 | #endif |
dd41f596 | 1420 | |
ef12fefa BR |
1421 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1422 | enum cpuacct_stat_index { | |
1423 | CPUACCT_STAT_USER, /* ... user mode */ | |
1424 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1425 | ||
1426 | CPUACCT_STAT_NSTATS, | |
1427 | }; | |
1428 | ||
d842de87 SV |
1429 | #ifdef CONFIG_CGROUP_CPUACCT |
1430 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1431 | static void cpuacct_update_stats(struct task_struct *tsk, |
1432 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1433 | #else |
1434 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1435 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1437 | #endif |
1438 | ||
18d95a28 PZ |
1439 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1440 | { | |
1441 | update_load_add(&rq->load, load); | |
1442 | } | |
1443 | ||
1444 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1445 | { | |
1446 | update_load_sub(&rq->load, load); | |
1447 | } | |
1448 | ||
7940ca36 | 1449 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1450 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1451 | |
1452 | /* | |
1453 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1454 | * leaving it for the final time. | |
1455 | */ | |
eb755805 | 1456 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1457 | { |
1458 | struct task_group *parent, *child; | |
eb755805 | 1459 | int ret; |
c09595f6 PZ |
1460 | |
1461 | rcu_read_lock(); | |
1462 | parent = &root_task_group; | |
1463 | down: | |
eb755805 PZ |
1464 | ret = (*down)(parent, data); |
1465 | if (ret) | |
1466 | goto out_unlock; | |
c09595f6 PZ |
1467 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1468 | parent = child; | |
1469 | goto down; | |
1470 | ||
1471 | up: | |
1472 | continue; | |
1473 | } | |
eb755805 PZ |
1474 | ret = (*up)(parent, data); |
1475 | if (ret) | |
1476 | goto out_unlock; | |
c09595f6 PZ |
1477 | |
1478 | child = parent; | |
1479 | parent = parent->parent; | |
1480 | if (parent) | |
1481 | goto up; | |
eb755805 | 1482 | out_unlock: |
c09595f6 | 1483 | rcu_read_unlock(); |
eb755805 PZ |
1484 | |
1485 | return ret; | |
c09595f6 PZ |
1486 | } |
1487 | ||
eb755805 PZ |
1488 | static int tg_nop(struct task_group *tg, void *data) |
1489 | { | |
1490 | return 0; | |
c09595f6 | 1491 | } |
eb755805 PZ |
1492 | #endif |
1493 | ||
1494 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1495 | /* Used instead of source_load when we know the type == 0 */ |
1496 | static unsigned long weighted_cpuload(const int cpu) | |
1497 | { | |
1498 | return cpu_rq(cpu)->load.weight; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * Return a low guess at the load of a migration-source cpu weighted | |
1503 | * according to the scheduling class and "nice" value. | |
1504 | * | |
1505 | * We want to under-estimate the load of migration sources, to | |
1506 | * balance conservatively. | |
1507 | */ | |
1508 | static unsigned long source_load(int cpu, int type) | |
1509 | { | |
1510 | struct rq *rq = cpu_rq(cpu); | |
1511 | unsigned long total = weighted_cpuload(cpu); | |
1512 | ||
1513 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1514 | return total; | |
1515 | ||
1516 | return min(rq->cpu_load[type-1], total); | |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Return a high guess at the load of a migration-target cpu weighted | |
1521 | * according to the scheduling class and "nice" value. | |
1522 | */ | |
1523 | static unsigned long target_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 max(rq->cpu_load[type-1], total); | |
1532 | } | |
1533 | ||
ae154be1 PZ |
1534 | static struct sched_group *group_of(int cpu) |
1535 | { | |
1536 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
1537 | ||
1538 | if (!sd) | |
1539 | return NULL; | |
1540 | ||
1541 | return sd->groups; | |
1542 | } | |
1543 | ||
1544 | static unsigned long power_of(int cpu) | |
1545 | { | |
1546 | struct sched_group *group = group_of(cpu); | |
1547 | ||
1548 | if (!group) | |
1549 | return SCHED_LOAD_SCALE; | |
1550 | ||
1551 | return group->cpu_power; | |
1552 | } | |
1553 | ||
eb755805 PZ |
1554 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1555 | ||
1556 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1557 | { | |
1558 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1559 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1560 | |
4cd42620 SR |
1561 | if (nr_running) |
1562 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1563 | else |
1564 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1565 | |
1566 | return rq->avg_load_per_task; | |
1567 | } | |
1568 | ||
1569 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1570 | |
4a6cc4bd | 1571 | static __read_mostly unsigned long *update_shares_data; |
34d76c41 | 1572 | |
c09595f6 PZ |
1573 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1574 | ||
1575 | /* | |
1576 | * Calculate and set the cpu's group shares. | |
1577 | */ | |
34d76c41 PZ |
1578 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1579 | unsigned long sd_shares, | |
1580 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1581 | unsigned long *usd_rq_weight) |
18d95a28 | 1582 | { |
34d76c41 | 1583 | unsigned long shares, rq_weight; |
a5004278 | 1584 | int boost = 0; |
c09595f6 | 1585 | |
4a6cc4bd | 1586 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1587 | if (!rq_weight) { |
1588 | boost = 1; | |
1589 | rq_weight = NICE_0_LOAD; | |
1590 | } | |
c8cba857 | 1591 | |
c09595f6 | 1592 | /* |
a8af7246 PZ |
1593 | * \Sum_j shares_j * rq_weight_i |
1594 | * shares_i = ----------------------------- | |
1595 | * \Sum_j rq_weight_j | |
c09595f6 | 1596 | */ |
ec4e0e2f | 1597 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1598 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1599 | |
ffda12a1 PZ |
1600 | if (abs(shares - tg->se[cpu]->load.weight) > |
1601 | sysctl_sched_shares_thresh) { | |
1602 | struct rq *rq = cpu_rq(cpu); | |
1603 | unsigned long flags; | |
c09595f6 | 1604 | |
ffda12a1 | 1605 | spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1606 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1607 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1608 | __set_se_shares(tg->se[cpu], shares); |
1609 | spin_unlock_irqrestore(&rq->lock, flags); | |
1610 | } | |
18d95a28 | 1611 | } |
c09595f6 PZ |
1612 | |
1613 | /* | |
c8cba857 PZ |
1614 | * Re-compute the task group their per cpu shares over the given domain. |
1615 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1616 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1617 | */ |
eb755805 | 1618 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1619 | { |
cd8ad40d | 1620 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1621 | unsigned long *usd_rq_weight; |
eb755805 | 1622 | struct sched_domain *sd = data; |
34d76c41 | 1623 | unsigned long flags; |
c8cba857 | 1624 | int i; |
c09595f6 | 1625 | |
34d76c41 PZ |
1626 | if (!tg->se[0]) |
1627 | return 0; | |
1628 | ||
1629 | local_irq_save(flags); | |
4a6cc4bd | 1630 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1631 | |
758b2cdc | 1632 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1633 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1634 | usd_rq_weight[i] = weight; |
34d76c41 | 1635 | |
cd8ad40d | 1636 | rq_weight += weight; |
ec4e0e2f KC |
1637 | /* |
1638 | * If there are currently no tasks on the cpu pretend there | |
1639 | * is one of average load so that when a new task gets to | |
1640 | * run here it will not get delayed by group starvation. | |
1641 | */ | |
ec4e0e2f KC |
1642 | if (!weight) |
1643 | weight = NICE_0_LOAD; | |
1644 | ||
cd8ad40d | 1645 | sum_weight += weight; |
c8cba857 | 1646 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1647 | } |
c09595f6 | 1648 | |
cd8ad40d PZ |
1649 | if (!rq_weight) |
1650 | rq_weight = sum_weight; | |
1651 | ||
c8cba857 PZ |
1652 | if ((!shares && rq_weight) || shares > tg->shares) |
1653 | shares = tg->shares; | |
1654 | ||
1655 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1656 | shares = tg->shares; | |
c09595f6 | 1657 | |
758b2cdc | 1658 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1659 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1660 | |
1661 | local_irq_restore(flags); | |
eb755805 PZ |
1662 | |
1663 | return 0; | |
c09595f6 PZ |
1664 | } |
1665 | ||
1666 | /* | |
c8cba857 PZ |
1667 | * Compute the cpu's hierarchical load factor for each task group. |
1668 | * This needs to be done in a top-down fashion because the load of a child | |
1669 | * group is a fraction of its parents load. | |
c09595f6 | 1670 | */ |
eb755805 | 1671 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1672 | { |
c8cba857 | 1673 | unsigned long load; |
eb755805 | 1674 | long cpu = (long)data; |
c09595f6 | 1675 | |
c8cba857 PZ |
1676 | if (!tg->parent) { |
1677 | load = cpu_rq(cpu)->load.weight; | |
1678 | } else { | |
1679 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1680 | load *= tg->cfs_rq[cpu]->shares; | |
1681 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1682 | } | |
c09595f6 | 1683 | |
c8cba857 | 1684 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1685 | |
eb755805 | 1686 | return 0; |
c09595f6 PZ |
1687 | } |
1688 | ||
c8cba857 | 1689 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1690 | { |
e7097159 PZ |
1691 | s64 elapsed; |
1692 | u64 now; | |
1693 | ||
1694 | if (root_task_group_empty()) | |
1695 | return; | |
1696 | ||
1697 | now = cpu_clock(raw_smp_processor_id()); | |
1698 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1699 | |
1700 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1701 | sd->last_update = now; | |
eb755805 | 1702 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1703 | } |
4d8d595d PZ |
1704 | } |
1705 | ||
3e5459b4 PZ |
1706 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1707 | { | |
e7097159 PZ |
1708 | if (root_task_group_empty()) |
1709 | return; | |
1710 | ||
3e5459b4 PZ |
1711 | spin_unlock(&rq->lock); |
1712 | update_shares(sd); | |
1713 | spin_lock(&rq->lock); | |
1714 | } | |
1715 | ||
eb755805 | 1716 | static void update_h_load(long cpu) |
c09595f6 | 1717 | { |
e7097159 PZ |
1718 | if (root_task_group_empty()) |
1719 | return; | |
1720 | ||
eb755805 | 1721 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1722 | } |
1723 | ||
c09595f6 PZ |
1724 | #else |
1725 | ||
c8cba857 | 1726 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1727 | { |
1728 | } | |
1729 | ||
3e5459b4 PZ |
1730 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1731 | { | |
1732 | } | |
1733 | ||
18d95a28 PZ |
1734 | #endif |
1735 | ||
8f45e2b5 GH |
1736 | #ifdef CONFIG_PREEMPT |
1737 | ||
b78bb868 PZ |
1738 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1739 | ||
70574a99 | 1740 | /* |
8f45e2b5 GH |
1741 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1742 | * way at the expense of forcing extra atomic operations in all | |
1743 | * invocations. This assures that the double_lock is acquired using the | |
1744 | * same underlying policy as the spinlock_t on this architecture, which | |
1745 | * reduces latency compared to the unfair variant below. However, it | |
1746 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1747 | */ |
8f45e2b5 GH |
1748 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1749 | __releases(this_rq->lock) | |
1750 | __acquires(busiest->lock) | |
1751 | __acquires(this_rq->lock) | |
1752 | { | |
1753 | spin_unlock(&this_rq->lock); | |
1754 | double_rq_lock(this_rq, busiest); | |
1755 | ||
1756 | return 1; | |
1757 | } | |
1758 | ||
1759 | #else | |
1760 | /* | |
1761 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1762 | * latency by eliminating extra atomic operations when the locks are | |
1763 | * already in proper order on entry. This favors lower cpu-ids and will | |
1764 | * grant the double lock to lower cpus over higher ids under contention, | |
1765 | * regardless of entry order into the function. | |
1766 | */ | |
1767 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1768 | __releases(this_rq->lock) |
1769 | __acquires(busiest->lock) | |
1770 | __acquires(this_rq->lock) | |
1771 | { | |
1772 | int ret = 0; | |
1773 | ||
70574a99 AD |
1774 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1775 | if (busiest < this_rq) { | |
1776 | spin_unlock(&this_rq->lock); | |
1777 | spin_lock(&busiest->lock); | |
1778 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1779 | ret = 1; | |
1780 | } else | |
1781 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1782 | } | |
1783 | return ret; | |
1784 | } | |
1785 | ||
8f45e2b5 GH |
1786 | #endif /* CONFIG_PREEMPT */ |
1787 | ||
1788 | /* | |
1789 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1790 | */ | |
1791 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1792 | { | |
1793 | if (unlikely(!irqs_disabled())) { | |
1794 | /* printk() doesn't work good under rq->lock */ | |
1795 | spin_unlock(&this_rq->lock); | |
1796 | BUG_ON(1); | |
1797 | } | |
1798 | ||
1799 | return _double_lock_balance(this_rq, busiest); | |
1800 | } | |
1801 | ||
70574a99 AD |
1802 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1803 | __releases(busiest->lock) | |
1804 | { | |
1805 | spin_unlock(&busiest->lock); | |
1806 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1807 | } | |
18d95a28 PZ |
1808 | #endif |
1809 | ||
30432094 | 1810 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1811 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1812 | { | |
30432094 | 1813 | #ifdef CONFIG_SMP |
34e83e85 IM |
1814 | cfs_rq->shares = shares; |
1815 | #endif | |
1816 | } | |
30432094 | 1817 | #endif |
e7693a36 | 1818 | |
dce48a84 | 1819 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1820 | static void update_sysctl(void); |
acb4a848 | 1821 | static int get_update_sysctl_factor(void); |
dce48a84 | 1822 | |
cd29fe6f PZ |
1823 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1824 | { | |
1825 | set_task_rq(p, cpu); | |
1826 | #ifdef CONFIG_SMP | |
1827 | /* | |
1828 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1829 | * successfuly executed on another CPU. We must ensure that updates of | |
1830 | * per-task data have been completed by this moment. | |
1831 | */ | |
1832 | smp_wmb(); | |
1833 | task_thread_info(p)->cpu = cpu; | |
1834 | #endif | |
1835 | } | |
dce48a84 | 1836 | |
dd41f596 | 1837 | #include "sched_stats.h" |
dd41f596 | 1838 | #include "sched_idletask.c" |
5522d5d5 IM |
1839 | #include "sched_fair.c" |
1840 | #include "sched_rt.c" | |
dd41f596 IM |
1841 | #ifdef CONFIG_SCHED_DEBUG |
1842 | # include "sched_debug.c" | |
1843 | #endif | |
1844 | ||
1845 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1846 | #define for_each_class(class) \ |
1847 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1848 | |
c09595f6 | 1849 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1850 | { |
1851 | rq->nr_running++; | |
9c217245 IM |
1852 | } |
1853 | ||
c09595f6 | 1854 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1855 | { |
1856 | rq->nr_running--; | |
9c217245 IM |
1857 | } |
1858 | ||
45bf76df IM |
1859 | static void set_load_weight(struct task_struct *p) |
1860 | { | |
1861 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1862 | p->se.load.weight = prio_to_weight[0] * 2; |
1863 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1864 | return; | |
1865 | } | |
45bf76df | 1866 | |
dd41f596 IM |
1867 | /* |
1868 | * SCHED_IDLE tasks get minimal weight: | |
1869 | */ | |
1870 | if (p->policy == SCHED_IDLE) { | |
1871 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1872 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1873 | return; | |
1874 | } | |
71f8bd46 | 1875 | |
dd41f596 IM |
1876 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1877 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1878 | } |
1879 | ||
2087a1ad GH |
1880 | static void update_avg(u64 *avg, u64 sample) |
1881 | { | |
1882 | s64 diff = sample - *avg; | |
1883 | *avg += diff >> 3; | |
1884 | } | |
1885 | ||
8159f87e | 1886 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1887 | { |
831451ac PZ |
1888 | if (wakeup) |
1889 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1890 | ||
dd41f596 | 1891 | sched_info_queued(p); |
fd390f6a | 1892 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1893 | p->se.on_rq = 1; |
71f8bd46 IM |
1894 | } |
1895 | ||
69be72c1 | 1896 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1897 | { |
831451ac PZ |
1898 | if (sleep) { |
1899 | if (p->se.last_wakeup) { | |
1900 | update_avg(&p->se.avg_overlap, | |
1901 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1902 | p->se.last_wakeup = 0; | |
1903 | } else { | |
1904 | update_avg(&p->se.avg_wakeup, | |
1905 | sysctl_sched_wakeup_granularity); | |
1906 | } | |
2087a1ad GH |
1907 | } |
1908 | ||
46ac22ba | 1909 | sched_info_dequeued(p); |
f02231e5 | 1910 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1911 | p->se.on_rq = 0; |
71f8bd46 IM |
1912 | } |
1913 | ||
14531189 | 1914 | /* |
dd41f596 | 1915 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1916 | */ |
14531189 IM |
1917 | static inline int __normal_prio(struct task_struct *p) |
1918 | { | |
dd41f596 | 1919 | return p->static_prio; |
14531189 IM |
1920 | } |
1921 | ||
b29739f9 IM |
1922 | /* |
1923 | * Calculate the expected normal priority: i.e. priority | |
1924 | * without taking RT-inheritance into account. Might be | |
1925 | * boosted by interactivity modifiers. Changes upon fork, | |
1926 | * setprio syscalls, and whenever the interactivity | |
1927 | * estimator recalculates. | |
1928 | */ | |
36c8b586 | 1929 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1930 | { |
1931 | int prio; | |
1932 | ||
e05606d3 | 1933 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1934 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1935 | else | |
1936 | prio = __normal_prio(p); | |
1937 | return prio; | |
1938 | } | |
1939 | ||
1940 | /* | |
1941 | * Calculate the current priority, i.e. the priority | |
1942 | * taken into account by the scheduler. This value might | |
1943 | * be boosted by RT tasks, or might be boosted by | |
1944 | * interactivity modifiers. Will be RT if the task got | |
1945 | * RT-boosted. If not then it returns p->normal_prio. | |
1946 | */ | |
36c8b586 | 1947 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1948 | { |
1949 | p->normal_prio = normal_prio(p); | |
1950 | /* | |
1951 | * If we are RT tasks or we were boosted to RT priority, | |
1952 | * keep the priority unchanged. Otherwise, update priority | |
1953 | * to the normal priority: | |
1954 | */ | |
1955 | if (!rt_prio(p->prio)) | |
1956 | return p->normal_prio; | |
1957 | return p->prio; | |
1958 | } | |
1959 | ||
1da177e4 | 1960 | /* |
dd41f596 | 1961 | * activate_task - move a task to the runqueue. |
1da177e4 | 1962 | */ |
dd41f596 | 1963 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1964 | { |
d9514f6c | 1965 | if (task_contributes_to_load(p)) |
dd41f596 | 1966 | rq->nr_uninterruptible--; |
1da177e4 | 1967 | |
8159f87e | 1968 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1969 | inc_nr_running(rq); |
1da177e4 LT |
1970 | } |
1971 | ||
1da177e4 LT |
1972 | /* |
1973 | * deactivate_task - remove a task from the runqueue. | |
1974 | */ | |
2e1cb74a | 1975 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1976 | { |
d9514f6c | 1977 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1978 | rq->nr_uninterruptible++; |
1979 | ||
69be72c1 | 1980 | dequeue_task(rq, p, sleep); |
c09595f6 | 1981 | dec_nr_running(rq); |
1da177e4 LT |
1982 | } |
1983 | ||
1da177e4 LT |
1984 | /** |
1985 | * task_curr - is this task currently executing on a CPU? | |
1986 | * @p: the task in question. | |
1987 | */ | |
36c8b586 | 1988 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1989 | { |
1990 | return cpu_curr(task_cpu(p)) == p; | |
1991 | } | |
1992 | ||
cb469845 SR |
1993 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1994 | const struct sched_class *prev_class, | |
1995 | int oldprio, int running) | |
1996 | { | |
1997 | if (prev_class != p->sched_class) { | |
1998 | if (prev_class->switched_from) | |
1999 | prev_class->switched_from(rq, p, running); | |
2000 | p->sched_class->switched_to(rq, p, running); | |
2001 | } else | |
2002 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2003 | } | |
2004 | ||
b84ff7d6 MG |
2005 | /** |
2006 | * kthread_bind - bind a just-created kthread to a cpu. | |
968c8645 | 2007 | * @p: thread created by kthread_create(). |
b84ff7d6 MG |
2008 | * @cpu: cpu (might not be online, must be possible) for @k to run on. |
2009 | * | |
2010 | * Description: This function is equivalent to set_cpus_allowed(), | |
2011 | * except that @cpu doesn't need to be online, and the thread must be | |
2012 | * stopped (i.e., just returned from kthread_create()). | |
2013 | * | |
2014 | * Function lives here instead of kthread.c because it messes with | |
2015 | * scheduler internals which require locking. | |
2016 | */ | |
2017 | void kthread_bind(struct task_struct *p, unsigned int cpu) | |
2018 | { | |
2019 | struct rq *rq = cpu_rq(cpu); | |
2020 | unsigned long flags; | |
2021 | ||
2022 | /* Must have done schedule() in kthread() before we set_task_cpu */ | |
2023 | if (!wait_task_inactive(p, TASK_UNINTERRUPTIBLE)) { | |
2024 | WARN_ON(1); | |
2025 | return; | |
2026 | } | |
2027 | ||
2028 | spin_lock_irqsave(&rq->lock, flags); | |
055a0086 | 2029 | update_rq_clock(rq); |
b84ff7d6 MG |
2030 | set_task_cpu(p, cpu); |
2031 | p->cpus_allowed = cpumask_of_cpu(cpu); | |
2032 | p->rt.nr_cpus_allowed = 1; | |
2033 | p->flags |= PF_THREAD_BOUND; | |
2034 | spin_unlock_irqrestore(&rq->lock, flags); | |
2035 | } | |
2036 | EXPORT_SYMBOL(kthread_bind); | |
2037 | ||
1da177e4 | 2038 | #ifdef CONFIG_SMP |
cc367732 IM |
2039 | /* |
2040 | * Is this task likely cache-hot: | |
2041 | */ | |
e7693a36 | 2042 | static int |
cc367732 IM |
2043 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2044 | { | |
2045 | s64 delta; | |
2046 | ||
f540a608 IM |
2047 | /* |
2048 | * Buddy candidates are cache hot: | |
2049 | */ | |
f685ceac | 2050 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2051 | (&p->se == cfs_rq_of(&p->se)->next || |
2052 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2053 | return 1; |
2054 | ||
cc367732 IM |
2055 | if (p->sched_class != &fair_sched_class) |
2056 | return 0; | |
2057 | ||
6bc1665b IM |
2058 | if (sysctl_sched_migration_cost == -1) |
2059 | return 1; | |
2060 | if (sysctl_sched_migration_cost == 0) | |
2061 | return 0; | |
2062 | ||
cc367732 IM |
2063 | delta = now - p->se.exec_start; |
2064 | ||
2065 | return delta < (s64)sysctl_sched_migration_cost; | |
2066 | } | |
2067 | ||
2068 | ||
dd41f596 | 2069 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2070 | { |
dd41f596 | 2071 | int old_cpu = task_cpu(p); |
2830cf8c SV |
2072 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
2073 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
6cfb0d5d | 2074 | |
de1d7286 | 2075 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2076 | |
cc367732 | 2077 | if (old_cpu != new_cpu) { |
6c594c21 | 2078 | p->se.nr_migrations++; |
cdd6c482 | 2079 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
e5289d4a | 2080 | 1, 1, NULL, 0); |
6c594c21 | 2081 | } |
2830cf8c SV |
2082 | p->se.vruntime -= old_cfsrq->min_vruntime - |
2083 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
2084 | |
2085 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2086 | } |
2087 | ||
70b97a7f | 2088 | struct migration_req { |
1da177e4 | 2089 | struct list_head list; |
1da177e4 | 2090 | |
36c8b586 | 2091 | struct task_struct *task; |
1da177e4 LT |
2092 | int dest_cpu; |
2093 | ||
1da177e4 | 2094 | struct completion done; |
70b97a7f | 2095 | }; |
1da177e4 LT |
2096 | |
2097 | /* | |
2098 | * The task's runqueue lock must be held. | |
2099 | * Returns true if you have to wait for migration thread. | |
2100 | */ | |
36c8b586 | 2101 | static int |
70b97a7f | 2102 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2103 | { |
70b97a7f | 2104 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2105 | |
2106 | /* | |
2107 | * If the task is not on a runqueue (and not running), then | |
2108 | * it is sufficient to simply update the task's cpu field. | |
2109 | */ | |
dd41f596 | 2110 | if (!p->se.on_rq && !task_running(rq, p)) { |
055a0086 | 2111 | update_rq_clock(rq); |
1da177e4 LT |
2112 | set_task_cpu(p, dest_cpu); |
2113 | return 0; | |
2114 | } | |
2115 | ||
2116 | init_completion(&req->done); | |
1da177e4 LT |
2117 | req->task = p; |
2118 | req->dest_cpu = dest_cpu; | |
2119 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2120 | |
1da177e4 LT |
2121 | return 1; |
2122 | } | |
2123 | ||
a26b89f0 MM |
2124 | /* |
2125 | * wait_task_context_switch - wait for a thread to complete at least one | |
2126 | * context switch. | |
2127 | * | |
2128 | * @p must not be current. | |
2129 | */ | |
2130 | void wait_task_context_switch(struct task_struct *p) | |
2131 | { | |
2132 | unsigned long nvcsw, nivcsw, flags; | |
2133 | int running; | |
2134 | struct rq *rq; | |
2135 | ||
2136 | nvcsw = p->nvcsw; | |
2137 | nivcsw = p->nivcsw; | |
2138 | for (;;) { | |
2139 | /* | |
2140 | * The runqueue is assigned before the actual context | |
2141 | * switch. We need to take the runqueue lock. | |
2142 | * | |
2143 | * We could check initially without the lock but it is | |
2144 | * very likely that we need to take the lock in every | |
2145 | * iteration. | |
2146 | */ | |
2147 | rq = task_rq_lock(p, &flags); | |
2148 | running = task_running(rq, p); | |
2149 | task_rq_unlock(rq, &flags); | |
2150 | ||
2151 | if (likely(!running)) | |
2152 | break; | |
2153 | /* | |
2154 | * The switch count is incremented before the actual | |
2155 | * context switch. We thus wait for two switches to be | |
2156 | * sure at least one completed. | |
2157 | */ | |
2158 | if ((p->nvcsw - nvcsw) > 1) | |
2159 | break; | |
2160 | if ((p->nivcsw - nivcsw) > 1) | |
2161 | break; | |
2162 | ||
2163 | cpu_relax(); | |
2164 | } | |
2165 | } | |
2166 | ||
1da177e4 LT |
2167 | /* |
2168 | * wait_task_inactive - wait for a thread to unschedule. | |
2169 | * | |
85ba2d86 RM |
2170 | * If @match_state is nonzero, it's the @p->state value just checked and |
2171 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2172 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2173 | * we return a positive number (its total switch count). If a second call | |
2174 | * a short while later returns the same number, the caller can be sure that | |
2175 | * @p has remained unscheduled the whole time. | |
2176 | * | |
1da177e4 LT |
2177 | * The caller must ensure that the task *will* unschedule sometime soon, |
2178 | * else this function might spin for a *long* time. This function can't | |
2179 | * be called with interrupts off, or it may introduce deadlock with | |
2180 | * smp_call_function() if an IPI is sent by the same process we are | |
2181 | * waiting to become inactive. | |
2182 | */ | |
85ba2d86 | 2183 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2184 | { |
2185 | unsigned long flags; | |
dd41f596 | 2186 | int running, on_rq; |
85ba2d86 | 2187 | unsigned long ncsw; |
70b97a7f | 2188 | struct rq *rq; |
1da177e4 | 2189 | |
3a5c359a AK |
2190 | for (;;) { |
2191 | /* | |
2192 | * We do the initial early heuristics without holding | |
2193 | * any task-queue locks at all. We'll only try to get | |
2194 | * the runqueue lock when things look like they will | |
2195 | * work out! | |
2196 | */ | |
2197 | rq = task_rq(p); | |
fa490cfd | 2198 | |
3a5c359a AK |
2199 | /* |
2200 | * If the task is actively running on another CPU | |
2201 | * still, just relax and busy-wait without holding | |
2202 | * any locks. | |
2203 | * | |
2204 | * NOTE! Since we don't hold any locks, it's not | |
2205 | * even sure that "rq" stays as the right runqueue! | |
2206 | * But we don't care, since "task_running()" will | |
2207 | * return false if the runqueue has changed and p | |
2208 | * is actually now running somewhere else! | |
2209 | */ | |
85ba2d86 RM |
2210 | while (task_running(rq, p)) { |
2211 | if (match_state && unlikely(p->state != match_state)) | |
2212 | return 0; | |
3a5c359a | 2213 | cpu_relax(); |
85ba2d86 | 2214 | } |
fa490cfd | 2215 | |
3a5c359a AK |
2216 | /* |
2217 | * Ok, time to look more closely! We need the rq | |
2218 | * lock now, to be *sure*. If we're wrong, we'll | |
2219 | * just go back and repeat. | |
2220 | */ | |
2221 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2222 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2223 | running = task_running(rq, p); |
2224 | on_rq = p->se.on_rq; | |
85ba2d86 | 2225 | ncsw = 0; |
f31e11d8 | 2226 | if (!match_state || p->state == match_state) |
93dcf55f | 2227 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2228 | task_rq_unlock(rq, &flags); |
fa490cfd | 2229 | |
85ba2d86 RM |
2230 | /* |
2231 | * If it changed from the expected state, bail out now. | |
2232 | */ | |
2233 | if (unlikely(!ncsw)) | |
2234 | break; | |
2235 | ||
3a5c359a AK |
2236 | /* |
2237 | * Was it really running after all now that we | |
2238 | * checked with the proper locks actually held? | |
2239 | * | |
2240 | * Oops. Go back and try again.. | |
2241 | */ | |
2242 | if (unlikely(running)) { | |
2243 | cpu_relax(); | |
2244 | continue; | |
2245 | } | |
fa490cfd | 2246 | |
3a5c359a AK |
2247 | /* |
2248 | * It's not enough that it's not actively running, | |
2249 | * it must be off the runqueue _entirely_, and not | |
2250 | * preempted! | |
2251 | * | |
80dd99b3 | 2252 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2253 | * running right now), it's preempted, and we should |
2254 | * yield - it could be a while. | |
2255 | */ | |
2256 | if (unlikely(on_rq)) { | |
2257 | schedule_timeout_uninterruptible(1); | |
2258 | continue; | |
2259 | } | |
fa490cfd | 2260 | |
3a5c359a AK |
2261 | /* |
2262 | * Ahh, all good. It wasn't running, and it wasn't | |
2263 | * runnable, which means that it will never become | |
2264 | * running in the future either. We're all done! | |
2265 | */ | |
2266 | break; | |
2267 | } | |
85ba2d86 RM |
2268 | |
2269 | return ncsw; | |
1da177e4 LT |
2270 | } |
2271 | ||
2272 | /*** | |
2273 | * kick_process - kick a running thread to enter/exit the kernel | |
2274 | * @p: the to-be-kicked thread | |
2275 | * | |
2276 | * Cause a process which is running on another CPU to enter | |
2277 | * kernel-mode, without any delay. (to get signals handled.) | |
2278 | * | |
2279 | * NOTE: this function doesnt have to take the runqueue lock, | |
2280 | * because all it wants to ensure is that the remote task enters | |
2281 | * the kernel. If the IPI races and the task has been migrated | |
2282 | * to another CPU then no harm is done and the purpose has been | |
2283 | * achieved as well. | |
2284 | */ | |
36c8b586 | 2285 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2286 | { |
2287 | int cpu; | |
2288 | ||
2289 | preempt_disable(); | |
2290 | cpu = task_cpu(p); | |
2291 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2292 | smp_send_reschedule(cpu); | |
2293 | preempt_enable(); | |
2294 | } | |
b43e3521 | 2295 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2296 | #endif /* CONFIG_SMP */ |
1da177e4 | 2297 | |
0793a61d TG |
2298 | /** |
2299 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2300 | * @p: the task to evaluate | |
2301 | * @func: the function to be called | |
2302 | * @info: the function call argument | |
2303 | * | |
2304 | * Calls the function @func when the task is currently running. This might | |
2305 | * be on the current CPU, which just calls the function directly | |
2306 | */ | |
2307 | void task_oncpu_function_call(struct task_struct *p, | |
2308 | void (*func) (void *info), void *info) | |
2309 | { | |
2310 | int cpu; | |
2311 | ||
2312 | preempt_disable(); | |
2313 | cpu = task_cpu(p); | |
2314 | if (task_curr(p)) | |
2315 | smp_call_function_single(cpu, func, info, 1); | |
2316 | preempt_enable(); | |
2317 | } | |
2318 | ||
970b13ba PZ |
2319 | #ifdef CONFIG_SMP |
2320 | static inline | |
2321 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2322 | { | |
2323 | return p->sched_class->select_task_rq(p, sd_flags, wake_flags); | |
2324 | } | |
2325 | #endif | |
2326 | ||
1da177e4 LT |
2327 | /*** |
2328 | * try_to_wake_up - wake up a thread | |
2329 | * @p: the to-be-woken-up thread | |
2330 | * @state: the mask of task states that can be woken | |
2331 | * @sync: do a synchronous wakeup? | |
2332 | * | |
2333 | * Put it on the run-queue if it's not already there. The "current" | |
2334 | * thread is always on the run-queue (except when the actual | |
2335 | * re-schedule is in progress), and as such you're allowed to do | |
2336 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2337 | * runnable without the overhead of this. | |
2338 | * | |
2339 | * returns failure only if the task is already active. | |
2340 | */ | |
7d478721 PZ |
2341 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2342 | int wake_flags) | |
1da177e4 | 2343 | { |
cc367732 | 2344 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2345 | unsigned long flags; |
f5dc3753 | 2346 | struct rq *rq, *orig_rq; |
1da177e4 | 2347 | |
b85d0667 | 2348 | if (!sched_feat(SYNC_WAKEUPS)) |
7d478721 | 2349 | wake_flags &= ~WF_SYNC; |
2398f2c6 | 2350 | |
e9c84311 | 2351 | this_cpu = get_cpu(); |
2398f2c6 | 2352 | |
04e2f174 | 2353 | smp_wmb(); |
f5dc3753 | 2354 | rq = orig_rq = task_rq_lock(p, &flags); |
03e89e45 | 2355 | update_rq_clock(rq); |
e9c84311 | 2356 | if (!(p->state & state)) |
1da177e4 LT |
2357 | goto out; |
2358 | ||
dd41f596 | 2359 | if (p->se.on_rq) |
1da177e4 LT |
2360 | goto out_running; |
2361 | ||
2362 | cpu = task_cpu(p); | |
cc367732 | 2363 | orig_cpu = cpu; |
1da177e4 LT |
2364 | |
2365 | #ifdef CONFIG_SMP | |
2366 | if (unlikely(task_running(rq, p))) | |
2367 | goto out_activate; | |
2368 | ||
e9c84311 PZ |
2369 | /* |
2370 | * In order to handle concurrent wakeups and release the rq->lock | |
2371 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2372 | * |
2373 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2374 | */ |
eb24073b IM |
2375 | if (task_contributes_to_load(p)) |
2376 | rq->nr_uninterruptible--; | |
e9c84311 | 2377 | p->state = TASK_WAKING; |
ab19cb23 | 2378 | __task_rq_unlock(rq); |
e9c84311 | 2379 | |
970b13ba | 2380 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
ab19cb23 | 2381 | if (cpu != orig_cpu) |
5d2f5a61 | 2382 | set_task_cpu(p, cpu); |
ab19cb23 PZ |
2383 | |
2384 | rq = __task_rq_lock(p); | |
2385 | update_rq_clock(rq); | |
f5dc3753 | 2386 | |
e9c84311 PZ |
2387 | WARN_ON(p->state != TASK_WAKING); |
2388 | cpu = task_cpu(p); | |
1da177e4 | 2389 | |
e7693a36 GH |
2390 | #ifdef CONFIG_SCHEDSTATS |
2391 | schedstat_inc(rq, ttwu_count); | |
2392 | if (cpu == this_cpu) | |
2393 | schedstat_inc(rq, ttwu_local); | |
2394 | else { | |
2395 | struct sched_domain *sd; | |
2396 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2397 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2398 | schedstat_inc(sd, ttwu_wake_remote); |
2399 | break; | |
2400 | } | |
2401 | } | |
2402 | } | |
6d6bc0ad | 2403 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2404 | |
1da177e4 LT |
2405 | out_activate: |
2406 | #endif /* CONFIG_SMP */ | |
cc367732 | 2407 | schedstat_inc(p, se.nr_wakeups); |
7d478721 | 2408 | if (wake_flags & WF_SYNC) |
cc367732 IM |
2409 | schedstat_inc(p, se.nr_wakeups_sync); |
2410 | if (orig_cpu != cpu) | |
2411 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2412 | if (cpu == this_cpu) | |
2413 | schedstat_inc(p, se.nr_wakeups_local); | |
2414 | else | |
2415 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2416 | activate_task(rq, p, 1); |
1da177e4 LT |
2417 | success = 1; |
2418 | ||
831451ac PZ |
2419 | /* |
2420 | * Only attribute actual wakeups done by this task. | |
2421 | */ | |
2422 | if (!in_interrupt()) { | |
2423 | struct sched_entity *se = ¤t->se; | |
2424 | u64 sample = se->sum_exec_runtime; | |
2425 | ||
2426 | if (se->last_wakeup) | |
2427 | sample -= se->last_wakeup; | |
2428 | else | |
2429 | sample -= se->start_runtime; | |
2430 | update_avg(&se->avg_wakeup, sample); | |
2431 | ||
2432 | se->last_wakeup = se->sum_exec_runtime; | |
2433 | } | |
2434 | ||
1da177e4 | 2435 | out_running: |
468a15bb | 2436 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2437 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2438 | |
1da177e4 | 2439 | p->state = TASK_RUNNING; |
9a897c5a SR |
2440 | #ifdef CONFIG_SMP |
2441 | if (p->sched_class->task_wake_up) | |
2442 | p->sched_class->task_wake_up(rq, p); | |
eae0c9df MG |
2443 | |
2444 | if (unlikely(rq->idle_stamp)) { | |
2445 | u64 delta = rq->clock - rq->idle_stamp; | |
2446 | u64 max = 2*sysctl_sched_migration_cost; | |
2447 | ||
2448 | if (delta > max) | |
2449 | rq->avg_idle = max; | |
2450 | else | |
2451 | update_avg(&rq->avg_idle, delta); | |
2452 | rq->idle_stamp = 0; | |
2453 | } | |
9a897c5a | 2454 | #endif |
1da177e4 LT |
2455 | out: |
2456 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2457 | put_cpu(); |
1da177e4 LT |
2458 | |
2459 | return success; | |
2460 | } | |
2461 | ||
50fa610a DH |
2462 | /** |
2463 | * wake_up_process - Wake up a specific process | |
2464 | * @p: The process to be woken up. | |
2465 | * | |
2466 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2467 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2468 | * running. | |
2469 | * | |
2470 | * It may be assumed that this function implies a write memory barrier before | |
2471 | * changing the task state if and only if any tasks are woken up. | |
2472 | */ | |
7ad5b3a5 | 2473 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2474 | { |
d9514f6c | 2475 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2476 | } |
1da177e4 LT |
2477 | EXPORT_SYMBOL(wake_up_process); |
2478 | ||
7ad5b3a5 | 2479 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2480 | { |
2481 | return try_to_wake_up(p, state, 0); | |
2482 | } | |
2483 | ||
1da177e4 LT |
2484 | /* |
2485 | * Perform scheduler related setup for a newly forked process p. | |
2486 | * p is forked by current. | |
dd41f596 IM |
2487 | * |
2488 | * __sched_fork() is basic setup used by init_idle() too: | |
2489 | */ | |
2490 | static void __sched_fork(struct task_struct *p) | |
2491 | { | |
dd41f596 IM |
2492 | p->se.exec_start = 0; |
2493 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2494 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2495 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2496 | p->se.last_wakeup = 0; |
2497 | p->se.avg_overlap = 0; | |
831451ac PZ |
2498 | p->se.start_runtime = 0; |
2499 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2500 | |
2501 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2502 | p->se.wait_start = 0; |
2503 | p->se.wait_max = 0; | |
2504 | p->se.wait_count = 0; | |
2505 | p->se.wait_sum = 0; | |
2506 | ||
2507 | p->se.sleep_start = 0; | |
2508 | p->se.sleep_max = 0; | |
2509 | p->se.sum_sleep_runtime = 0; | |
2510 | ||
2511 | p->se.block_start = 0; | |
2512 | p->se.block_max = 0; | |
2513 | p->se.exec_max = 0; | |
2514 | p->se.slice_max = 0; | |
2515 | ||
2516 | p->se.nr_migrations_cold = 0; | |
2517 | p->se.nr_failed_migrations_affine = 0; | |
2518 | p->se.nr_failed_migrations_running = 0; | |
2519 | p->se.nr_failed_migrations_hot = 0; | |
2520 | p->se.nr_forced_migrations = 0; | |
7793527b LDM |
2521 | |
2522 | p->se.nr_wakeups = 0; | |
2523 | p->se.nr_wakeups_sync = 0; | |
2524 | p->se.nr_wakeups_migrate = 0; | |
2525 | p->se.nr_wakeups_local = 0; | |
2526 | p->se.nr_wakeups_remote = 0; | |
2527 | p->se.nr_wakeups_affine = 0; | |
2528 | p->se.nr_wakeups_affine_attempts = 0; | |
2529 | p->se.nr_wakeups_passive = 0; | |
2530 | p->se.nr_wakeups_idle = 0; | |
2531 | ||
6cfb0d5d | 2532 | #endif |
476d139c | 2533 | |
fa717060 | 2534 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2535 | p->se.on_rq = 0; |
4a55bd5e | 2536 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2537 | |
e107be36 AK |
2538 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2539 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2540 | #endif | |
2541 | ||
1da177e4 LT |
2542 | /* |
2543 | * We mark the process as running here, but have not actually | |
2544 | * inserted it onto the runqueue yet. This guarantees that | |
2545 | * nobody will actually run it, and a signal or other external | |
2546 | * event cannot wake it up and insert it on the runqueue either. | |
2547 | */ | |
2548 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2549 | } |
2550 | ||
2551 | /* | |
2552 | * fork()/clone()-time setup: | |
2553 | */ | |
2554 | void sched_fork(struct task_struct *p, int clone_flags) | |
2555 | { | |
2556 | int cpu = get_cpu(); | |
2557 | ||
2558 | __sched_fork(p); | |
2559 | ||
b9dc29e7 MG |
2560 | /* |
2561 | * Revert to default priority/policy on fork if requested. | |
2562 | */ | |
2563 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2564 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2565 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2566 | p->normal_prio = p->static_prio; |
2567 | } | |
b9dc29e7 | 2568 | |
6c697bdf MG |
2569 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2570 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2571 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2572 | set_load_weight(p); |
2573 | } | |
2574 | ||
b9dc29e7 MG |
2575 | /* |
2576 | * We don't need the reset flag anymore after the fork. It has | |
2577 | * fulfilled its duty: | |
2578 | */ | |
2579 | p->sched_reset_on_fork = 0; | |
2580 | } | |
ca94c442 | 2581 | |
f83f9ac2 PW |
2582 | /* |
2583 | * Make sure we do not leak PI boosting priority to the child. | |
2584 | */ | |
2585 | p->prio = current->normal_prio; | |
2586 | ||
2ddbf952 HS |
2587 | if (!rt_prio(p->prio)) |
2588 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2589 | |
cd29fe6f PZ |
2590 | if (p->sched_class->task_fork) |
2591 | p->sched_class->task_fork(p); | |
2592 | ||
5f3edc1b | 2593 | #ifdef CONFIG_SMP |
970b13ba | 2594 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); |
5f3edc1b PZ |
2595 | #endif |
2596 | set_task_cpu(p, cpu); | |
2597 | ||
52f17b6c | 2598 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2599 | if (likely(sched_info_on())) |
52f17b6c | 2600 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2601 | #endif |
d6077cb8 | 2602 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2603 | p->oncpu = 0; |
2604 | #endif | |
1da177e4 | 2605 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2606 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2607 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2608 | #endif |
917b627d GH |
2609 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2610 | ||
476d139c | 2611 | put_cpu(); |
1da177e4 LT |
2612 | } |
2613 | ||
2614 | /* | |
2615 | * wake_up_new_task - wake up a newly created task for the first time. | |
2616 | * | |
2617 | * This function will do some initial scheduler statistics housekeeping | |
2618 | * that must be done for every newly created context, then puts the task | |
2619 | * on the runqueue and wakes it. | |
2620 | */ | |
7ad5b3a5 | 2621 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2622 | { |
2623 | unsigned long flags; | |
dd41f596 | 2624 | struct rq *rq; |
1da177e4 LT |
2625 | |
2626 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2627 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2628 | update_rq_clock(rq); |
cd29fe6f | 2629 | activate_task(rq, p, 0); |
c71dd42d | 2630 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2631 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a SR |
2632 | #ifdef CONFIG_SMP |
2633 | if (p->sched_class->task_wake_up) | |
2634 | p->sched_class->task_wake_up(rq, p); | |
2635 | #endif | |
dd41f596 | 2636 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2637 | } |
2638 | ||
e107be36 AK |
2639 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2640 | ||
2641 | /** | |
80dd99b3 | 2642 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2643 | * @notifier: notifier struct to register |
e107be36 AK |
2644 | */ |
2645 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2646 | { | |
2647 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2648 | } | |
2649 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2650 | ||
2651 | /** | |
2652 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2653 | * @notifier: notifier struct to unregister |
e107be36 AK |
2654 | * |
2655 | * This is safe to call from within a preemption notifier. | |
2656 | */ | |
2657 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2658 | { | |
2659 | hlist_del(¬ifier->link); | |
2660 | } | |
2661 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2662 | ||
2663 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2664 | { | |
2665 | struct preempt_notifier *notifier; | |
2666 | struct hlist_node *node; | |
2667 | ||
2668 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2669 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2670 | } | |
2671 | ||
2672 | static void | |
2673 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2674 | struct task_struct *next) | |
2675 | { | |
2676 | struct preempt_notifier *notifier; | |
2677 | struct hlist_node *node; | |
2678 | ||
2679 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2680 | notifier->ops->sched_out(notifier, next); | |
2681 | } | |
2682 | ||
6d6bc0ad | 2683 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2684 | |
2685 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2686 | { | |
2687 | } | |
2688 | ||
2689 | static void | |
2690 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2691 | struct task_struct *next) | |
2692 | { | |
2693 | } | |
2694 | ||
6d6bc0ad | 2695 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2696 | |
4866cde0 NP |
2697 | /** |
2698 | * prepare_task_switch - prepare to switch tasks | |
2699 | * @rq: the runqueue preparing to switch | |
421cee29 | 2700 | * @prev: the current task that is being switched out |
4866cde0 NP |
2701 | * @next: the task we are going to switch to. |
2702 | * | |
2703 | * This is called with the rq lock held and interrupts off. It must | |
2704 | * be paired with a subsequent finish_task_switch after the context | |
2705 | * switch. | |
2706 | * | |
2707 | * prepare_task_switch sets up locking and calls architecture specific | |
2708 | * hooks. | |
2709 | */ | |
e107be36 AK |
2710 | static inline void |
2711 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2712 | struct task_struct *next) | |
4866cde0 | 2713 | { |
e107be36 | 2714 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2715 | prepare_lock_switch(rq, next); |
2716 | prepare_arch_switch(next); | |
2717 | } | |
2718 | ||
1da177e4 LT |
2719 | /** |
2720 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2721 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2722 | * @prev: the thread we just switched away from. |
2723 | * | |
4866cde0 NP |
2724 | * finish_task_switch must be called after the context switch, paired |
2725 | * with a prepare_task_switch call before the context switch. | |
2726 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2727 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2728 | * |
2729 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2730 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2731 | * with the lock held can cause deadlocks; see schedule() for |
2732 | * details.) | |
2733 | */ | |
a9957449 | 2734 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2735 | __releases(rq->lock) |
2736 | { | |
1da177e4 | 2737 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2738 | long prev_state; |
1da177e4 LT |
2739 | |
2740 | rq->prev_mm = NULL; | |
2741 | ||
2742 | /* | |
2743 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2744 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2745 | * schedule one last time. The schedule call will never return, and |
2746 | * the scheduled task must drop that reference. | |
c394cc9f | 2747 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2748 | * still held, otherwise prev could be scheduled on another cpu, die |
2749 | * there before we look at prev->state, and then the reference would | |
2750 | * be dropped twice. | |
2751 | * Manfred Spraul <manfred@colorfullife.com> | |
2752 | */ | |
55a101f8 | 2753 | prev_state = prev->state; |
4866cde0 | 2754 | finish_arch_switch(prev); |
cdd6c482 | 2755 | perf_event_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2756 | finish_lock_switch(rq, prev); |
e8fa1362 | 2757 | |
e107be36 | 2758 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2759 | if (mm) |
2760 | mmdrop(mm); | |
c394cc9f | 2761 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2762 | /* |
2763 | * Remove function-return probe instances associated with this | |
2764 | * task and put them back on the free list. | |
9761eea8 | 2765 | */ |
c6fd91f0 | 2766 | kprobe_flush_task(prev); |
1da177e4 | 2767 | put_task_struct(prev); |
c6fd91f0 | 2768 | } |
1da177e4 LT |
2769 | } |
2770 | ||
3f029d3c GH |
2771 | #ifdef CONFIG_SMP |
2772 | ||
2773 | /* assumes rq->lock is held */ | |
2774 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2775 | { | |
2776 | if (prev->sched_class->pre_schedule) | |
2777 | prev->sched_class->pre_schedule(rq, prev); | |
2778 | } | |
2779 | ||
2780 | /* rq->lock is NOT held, but preemption is disabled */ | |
2781 | static inline void post_schedule(struct rq *rq) | |
2782 | { | |
2783 | if (rq->post_schedule) { | |
2784 | unsigned long flags; | |
2785 | ||
2786 | spin_lock_irqsave(&rq->lock, flags); | |
2787 | if (rq->curr->sched_class->post_schedule) | |
2788 | rq->curr->sched_class->post_schedule(rq); | |
2789 | spin_unlock_irqrestore(&rq->lock, flags); | |
2790 | ||
2791 | rq->post_schedule = 0; | |
2792 | } | |
2793 | } | |
2794 | ||
2795 | #else | |
da19ab51 | 2796 | |
3f029d3c GH |
2797 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2798 | { | |
2799 | } | |
2800 | ||
2801 | static inline void post_schedule(struct rq *rq) | |
2802 | { | |
1da177e4 LT |
2803 | } |
2804 | ||
3f029d3c GH |
2805 | #endif |
2806 | ||
1da177e4 LT |
2807 | /** |
2808 | * schedule_tail - first thing a freshly forked thread must call. | |
2809 | * @prev: the thread we just switched away from. | |
2810 | */ | |
36c8b586 | 2811 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2812 | __releases(rq->lock) |
2813 | { | |
70b97a7f IM |
2814 | struct rq *rq = this_rq(); |
2815 | ||
4866cde0 | 2816 | finish_task_switch(rq, prev); |
da19ab51 | 2817 | |
3f029d3c GH |
2818 | /* |
2819 | * FIXME: do we need to worry about rq being invalidated by the | |
2820 | * task_switch? | |
2821 | */ | |
2822 | post_schedule(rq); | |
70b97a7f | 2823 | |
4866cde0 NP |
2824 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2825 | /* In this case, finish_task_switch does not reenable preemption */ | |
2826 | preempt_enable(); | |
2827 | #endif | |
1da177e4 | 2828 | if (current->set_child_tid) |
b488893a | 2829 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2830 | } |
2831 | ||
2832 | /* | |
2833 | * context_switch - switch to the new MM and the new | |
2834 | * thread's register state. | |
2835 | */ | |
dd41f596 | 2836 | static inline void |
70b97a7f | 2837 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2838 | struct task_struct *next) |
1da177e4 | 2839 | { |
dd41f596 | 2840 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2841 | |
e107be36 | 2842 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2843 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2844 | mm = next->mm; |
2845 | oldmm = prev->active_mm; | |
9226d125 ZA |
2846 | /* |
2847 | * For paravirt, this is coupled with an exit in switch_to to | |
2848 | * combine the page table reload and the switch backend into | |
2849 | * one hypercall. | |
2850 | */ | |
224101ed | 2851 | arch_start_context_switch(prev); |
9226d125 | 2852 | |
710390d9 | 2853 | if (likely(!mm)) { |
1da177e4 LT |
2854 | next->active_mm = oldmm; |
2855 | atomic_inc(&oldmm->mm_count); | |
2856 | enter_lazy_tlb(oldmm, next); | |
2857 | } else | |
2858 | switch_mm(oldmm, mm, next); | |
2859 | ||
710390d9 | 2860 | if (likely(!prev->mm)) { |
1da177e4 | 2861 | prev->active_mm = NULL; |
1da177e4 LT |
2862 | rq->prev_mm = oldmm; |
2863 | } | |
3a5f5e48 IM |
2864 | /* |
2865 | * Since the runqueue lock will be released by the next | |
2866 | * task (which is an invalid locking op but in the case | |
2867 | * of the scheduler it's an obvious special-case), so we | |
2868 | * do an early lockdep release here: | |
2869 | */ | |
2870 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2871 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2872 | #endif |
1da177e4 LT |
2873 | |
2874 | /* Here we just switch the register state and the stack. */ | |
2875 | switch_to(prev, next, prev); | |
2876 | ||
dd41f596 IM |
2877 | barrier(); |
2878 | /* | |
2879 | * this_rq must be evaluated again because prev may have moved | |
2880 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2881 | * frame will be invalid. | |
2882 | */ | |
2883 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2884 | } |
2885 | ||
2886 | /* | |
2887 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2888 | * | |
2889 | * externally visible scheduler statistics: current number of runnable | |
2890 | * threads, current number of uninterruptible-sleeping threads, total | |
2891 | * number of context switches performed since bootup. | |
2892 | */ | |
2893 | unsigned long nr_running(void) | |
2894 | { | |
2895 | unsigned long i, sum = 0; | |
2896 | ||
2897 | for_each_online_cpu(i) | |
2898 | sum += cpu_rq(i)->nr_running; | |
2899 | ||
2900 | return sum; | |
2901 | } | |
2902 | ||
2903 | unsigned long nr_uninterruptible(void) | |
2904 | { | |
2905 | unsigned long i, sum = 0; | |
2906 | ||
0a945022 | 2907 | for_each_possible_cpu(i) |
1da177e4 LT |
2908 | sum += cpu_rq(i)->nr_uninterruptible; |
2909 | ||
2910 | /* | |
2911 | * Since we read the counters lockless, it might be slightly | |
2912 | * inaccurate. Do not allow it to go below zero though: | |
2913 | */ | |
2914 | if (unlikely((long)sum < 0)) | |
2915 | sum = 0; | |
2916 | ||
2917 | return sum; | |
2918 | } | |
2919 | ||
2920 | unsigned long long nr_context_switches(void) | |
2921 | { | |
cc94abfc SR |
2922 | int i; |
2923 | unsigned long long sum = 0; | |
1da177e4 | 2924 | |
0a945022 | 2925 | for_each_possible_cpu(i) |
1da177e4 LT |
2926 | sum += cpu_rq(i)->nr_switches; |
2927 | ||
2928 | return sum; | |
2929 | } | |
2930 | ||
2931 | unsigned long nr_iowait(void) | |
2932 | { | |
2933 | unsigned long i, sum = 0; | |
2934 | ||
0a945022 | 2935 | for_each_possible_cpu(i) |
1da177e4 LT |
2936 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2937 | ||
2938 | return sum; | |
2939 | } | |
2940 | ||
69d25870 AV |
2941 | unsigned long nr_iowait_cpu(void) |
2942 | { | |
2943 | struct rq *this = this_rq(); | |
2944 | return atomic_read(&this->nr_iowait); | |
2945 | } | |
2946 | ||
2947 | unsigned long this_cpu_load(void) | |
2948 | { | |
2949 | struct rq *this = this_rq(); | |
2950 | return this->cpu_load[0]; | |
2951 | } | |
2952 | ||
2953 | ||
dce48a84 TG |
2954 | /* Variables and functions for calc_load */ |
2955 | static atomic_long_t calc_load_tasks; | |
2956 | static unsigned long calc_load_update; | |
2957 | unsigned long avenrun[3]; | |
2958 | EXPORT_SYMBOL(avenrun); | |
2959 | ||
2d02494f TG |
2960 | /** |
2961 | * get_avenrun - get the load average array | |
2962 | * @loads: pointer to dest load array | |
2963 | * @offset: offset to add | |
2964 | * @shift: shift count to shift the result left | |
2965 | * | |
2966 | * These values are estimates at best, so no need for locking. | |
2967 | */ | |
2968 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2969 | { | |
2970 | loads[0] = (avenrun[0] + offset) << shift; | |
2971 | loads[1] = (avenrun[1] + offset) << shift; | |
2972 | loads[2] = (avenrun[2] + offset) << shift; | |
2973 | } | |
2974 | ||
dce48a84 TG |
2975 | static unsigned long |
2976 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2977 | { |
dce48a84 TG |
2978 | load *= exp; |
2979 | load += active * (FIXED_1 - exp); | |
2980 | return load >> FSHIFT; | |
2981 | } | |
db1b1fef | 2982 | |
dce48a84 TG |
2983 | /* |
2984 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2985 | * CPUs have updated calc_load_tasks. | |
2986 | */ | |
2987 | void calc_global_load(void) | |
2988 | { | |
2989 | unsigned long upd = calc_load_update + 10; | |
2990 | long active; | |
2991 | ||
2992 | if (time_before(jiffies, upd)) | |
2993 | return; | |
db1b1fef | 2994 | |
dce48a84 TG |
2995 | active = atomic_long_read(&calc_load_tasks); |
2996 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2997 | |
dce48a84 TG |
2998 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2999 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3000 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3001 | ||
3002 | calc_load_update += LOAD_FREQ; | |
3003 | } | |
3004 | ||
3005 | /* | |
3006 | * Either called from update_cpu_load() or from a cpu going idle | |
3007 | */ | |
3008 | static void calc_load_account_active(struct rq *this_rq) | |
3009 | { | |
3010 | long nr_active, delta; | |
3011 | ||
3012 | nr_active = this_rq->nr_running; | |
3013 | nr_active += (long) this_rq->nr_uninterruptible; | |
3014 | ||
3015 | if (nr_active != this_rq->calc_load_active) { | |
3016 | delta = nr_active - this_rq->calc_load_active; | |
3017 | this_rq->calc_load_active = nr_active; | |
3018 | atomic_long_add(delta, &calc_load_tasks); | |
3019 | } | |
db1b1fef JS |
3020 | } |
3021 | ||
48f24c4d | 3022 | /* |
dd41f596 IM |
3023 | * Update rq->cpu_load[] statistics. This function is usually called every |
3024 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3025 | */ |
dd41f596 | 3026 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3027 | { |
495eca49 | 3028 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3029 | int i, scale; |
3030 | ||
3031 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3032 | |
3033 | /* Update our load: */ | |
3034 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3035 | unsigned long old_load, new_load; | |
3036 | ||
3037 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3038 | ||
3039 | old_load = this_rq->cpu_load[i]; | |
3040 | new_load = this_load; | |
a25707f3 IM |
3041 | /* |
3042 | * Round up the averaging division if load is increasing. This | |
3043 | * prevents us from getting stuck on 9 if the load is 10, for | |
3044 | * example. | |
3045 | */ | |
3046 | if (new_load > old_load) | |
3047 | new_load += scale-1; | |
dd41f596 IM |
3048 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3049 | } | |
dce48a84 TG |
3050 | |
3051 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3052 | this_rq->calc_load_update += LOAD_FREQ; | |
3053 | calc_load_account_active(this_rq); | |
3054 | } | |
48f24c4d IM |
3055 | } |
3056 | ||
dd41f596 IM |
3057 | #ifdef CONFIG_SMP |
3058 | ||
1da177e4 LT |
3059 | /* |
3060 | * double_rq_lock - safely lock two runqueues | |
3061 | * | |
3062 | * Note this does not disable interrupts like task_rq_lock, | |
3063 | * you need to do so manually before calling. | |
3064 | */ | |
70b97a7f | 3065 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3066 | __acquires(rq1->lock) |
3067 | __acquires(rq2->lock) | |
3068 | { | |
054b9108 | 3069 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3070 | if (rq1 == rq2) { |
3071 | spin_lock(&rq1->lock); | |
3072 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3073 | } else { | |
c96d145e | 3074 | if (rq1 < rq2) { |
1da177e4 | 3075 | spin_lock(&rq1->lock); |
5e710e37 | 3076 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3077 | } else { |
3078 | spin_lock(&rq2->lock); | |
5e710e37 | 3079 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3080 | } |
3081 | } | |
6e82a3be IM |
3082 | update_rq_clock(rq1); |
3083 | update_rq_clock(rq2); | |
1da177e4 LT |
3084 | } |
3085 | ||
3086 | /* | |
3087 | * double_rq_unlock - safely unlock two runqueues | |
3088 | * | |
3089 | * Note this does not restore interrupts like task_rq_unlock, | |
3090 | * you need to do so manually after calling. | |
3091 | */ | |
70b97a7f | 3092 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3093 | __releases(rq1->lock) |
3094 | __releases(rq2->lock) | |
3095 | { | |
3096 | spin_unlock(&rq1->lock); | |
3097 | if (rq1 != rq2) | |
3098 | spin_unlock(&rq2->lock); | |
3099 | else | |
3100 | __release(rq2->lock); | |
3101 | } | |
3102 | ||
1da177e4 LT |
3103 | /* |
3104 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3105 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3106 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3107 | * the cpu_allowed mask is restored. |
3108 | */ | |
36c8b586 | 3109 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3110 | { |
70b97a7f | 3111 | struct migration_req req; |
1da177e4 | 3112 | unsigned long flags; |
70b97a7f | 3113 | struct rq *rq; |
1da177e4 LT |
3114 | |
3115 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3116 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3117 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3118 | goto out; |
3119 | ||
3120 | /* force the process onto the specified CPU */ | |
3121 | if (migrate_task(p, dest_cpu, &req)) { | |
3122 | /* Need to wait for migration thread (might exit: take ref). */ | |
3123 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3124 | |
1da177e4 LT |
3125 | get_task_struct(mt); |
3126 | task_rq_unlock(rq, &flags); | |
3127 | wake_up_process(mt); | |
3128 | put_task_struct(mt); | |
3129 | wait_for_completion(&req.done); | |
36c8b586 | 3130 | |
1da177e4 LT |
3131 | return; |
3132 | } | |
3133 | out: | |
3134 | task_rq_unlock(rq, &flags); | |
3135 | } | |
3136 | ||
3137 | /* | |
476d139c NP |
3138 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3139 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3140 | */ |
3141 | void sched_exec(void) | |
3142 | { | |
1da177e4 | 3143 | int new_cpu, this_cpu = get_cpu(); |
970b13ba | 3144 | new_cpu = select_task_rq(current, SD_BALANCE_EXEC, 0); |
1da177e4 | 3145 | put_cpu(); |
476d139c NP |
3146 | if (new_cpu != this_cpu) |
3147 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3148 | } |
3149 | ||
3150 | /* | |
3151 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3152 | * Both runqueues must be locked. | |
3153 | */ | |
dd41f596 IM |
3154 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3155 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3156 | { |
2e1cb74a | 3157 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3158 | set_task_cpu(p, this_cpu); |
dd41f596 | 3159 | activate_task(this_rq, p, 0); |
15afe09b | 3160 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3161 | } |
3162 | ||
3163 | /* | |
3164 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3165 | */ | |
858119e1 | 3166 | static |
70b97a7f | 3167 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3168 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3169 | int *all_pinned) |
1da177e4 | 3170 | { |
708dc512 | 3171 | int tsk_cache_hot = 0; |
1da177e4 LT |
3172 | /* |
3173 | * We do not migrate tasks that are: | |
3174 | * 1) running (obviously), or | |
3175 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3176 | * 3) are cache-hot on their current CPU. | |
3177 | */ | |
96f874e2 | 3178 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3179 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3180 | return 0; |
cc367732 | 3181 | } |
81026794 NP |
3182 | *all_pinned = 0; |
3183 | ||
cc367732 IM |
3184 | if (task_running(rq, p)) { |
3185 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3186 | return 0; |
cc367732 | 3187 | } |
1da177e4 | 3188 | |
da84d961 IM |
3189 | /* |
3190 | * Aggressive migration if: | |
3191 | * 1) task is cache cold, or | |
3192 | * 2) too many balance attempts have failed. | |
3193 | */ | |
3194 | ||
708dc512 LH |
3195 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3196 | if (!tsk_cache_hot || | |
3197 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3198 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3199 | if (tsk_cache_hot) { |
da84d961 | 3200 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3201 | schedstat_inc(p, se.nr_forced_migrations); |
3202 | } | |
da84d961 IM |
3203 | #endif |
3204 | return 1; | |
3205 | } | |
3206 | ||
708dc512 | 3207 | if (tsk_cache_hot) { |
cc367732 | 3208 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3209 | return 0; |
cc367732 | 3210 | } |
1da177e4 LT |
3211 | return 1; |
3212 | } | |
3213 | ||
e1d1484f PW |
3214 | static unsigned long |
3215 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3216 | unsigned long max_load_move, struct sched_domain *sd, | |
3217 | enum cpu_idle_type idle, int *all_pinned, | |
3218 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3219 | { |
051c6764 | 3220 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3221 | struct task_struct *p; |
3222 | long rem_load_move = max_load_move; | |
1da177e4 | 3223 | |
e1d1484f | 3224 | if (max_load_move == 0) |
1da177e4 LT |
3225 | goto out; |
3226 | ||
81026794 NP |
3227 | pinned = 1; |
3228 | ||
1da177e4 | 3229 | /* |
dd41f596 | 3230 | * Start the load-balancing iterator: |
1da177e4 | 3231 | */ |
dd41f596 IM |
3232 | p = iterator->start(iterator->arg); |
3233 | next: | |
b82d9fdd | 3234 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3235 | goto out; |
051c6764 PZ |
3236 | |
3237 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3238 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3239 | p = iterator->next(iterator->arg); |
3240 | goto next; | |
1da177e4 LT |
3241 | } |
3242 | ||
dd41f596 | 3243 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3244 | pulled++; |
dd41f596 | 3245 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3246 | |
7e96fa58 GH |
3247 | #ifdef CONFIG_PREEMPT |
3248 | /* | |
3249 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3250 | * will stop after the first task is pulled to minimize the critical | |
3251 | * section. | |
3252 | */ | |
3253 | if (idle == CPU_NEWLY_IDLE) | |
3254 | goto out; | |
3255 | #endif | |
3256 | ||
2dd73a4f | 3257 | /* |
b82d9fdd | 3258 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3259 | */ |
e1d1484f | 3260 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3261 | if (p->prio < *this_best_prio) |
3262 | *this_best_prio = p->prio; | |
dd41f596 IM |
3263 | p = iterator->next(iterator->arg); |
3264 | goto next; | |
1da177e4 LT |
3265 | } |
3266 | out: | |
3267 | /* | |
e1d1484f | 3268 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3269 | * so we can safely collect pull_task() stats here rather than |
3270 | * inside pull_task(). | |
3271 | */ | |
3272 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3273 | |
3274 | if (all_pinned) | |
3275 | *all_pinned = pinned; | |
e1d1484f PW |
3276 | |
3277 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3278 | } |
3279 | ||
dd41f596 | 3280 | /* |
43010659 PW |
3281 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3282 | * this_rq, as part of a balancing operation within domain "sd". | |
3283 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3284 | * |
3285 | * Called with both runqueues locked. | |
3286 | */ | |
3287 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3288 | unsigned long max_load_move, |
dd41f596 IM |
3289 | struct sched_domain *sd, enum cpu_idle_type idle, |
3290 | int *all_pinned) | |
3291 | { | |
5522d5d5 | 3292 | const struct sched_class *class = sched_class_highest; |
43010659 | 3293 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3294 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3295 | |
3296 | do { | |
43010659 PW |
3297 | total_load_moved += |
3298 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3299 | max_load_move - total_load_moved, |
a4ac01c3 | 3300 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3301 | class = class->next; |
c4acb2c0 | 3302 | |
7e96fa58 GH |
3303 | #ifdef CONFIG_PREEMPT |
3304 | /* | |
3305 | * NEWIDLE balancing is a source of latency, so preemptible | |
3306 | * kernels will stop after the first task is pulled to minimize | |
3307 | * the critical section. | |
3308 | */ | |
c4acb2c0 GH |
3309 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3310 | break; | |
7e96fa58 | 3311 | #endif |
43010659 | 3312 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3313 | |
43010659 PW |
3314 | return total_load_moved > 0; |
3315 | } | |
3316 | ||
e1d1484f PW |
3317 | static int |
3318 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3319 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3320 | struct rq_iterator *iterator) | |
3321 | { | |
3322 | struct task_struct *p = iterator->start(iterator->arg); | |
3323 | int pinned = 0; | |
3324 | ||
3325 | while (p) { | |
3326 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3327 | pull_task(busiest, p, this_rq, this_cpu); | |
3328 | /* | |
3329 | * Right now, this is only the second place pull_task() | |
3330 | * is called, so we can safely collect pull_task() | |
3331 | * stats here rather than inside pull_task(). | |
3332 | */ | |
3333 | schedstat_inc(sd, lb_gained[idle]); | |
3334 | ||
3335 | return 1; | |
3336 | } | |
3337 | p = iterator->next(iterator->arg); | |
3338 | } | |
3339 | ||
3340 | return 0; | |
3341 | } | |
3342 | ||
43010659 PW |
3343 | /* |
3344 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3345 | * part of active balancing operations within "domain". | |
3346 | * Returns 1 if successful and 0 otherwise. | |
3347 | * | |
3348 | * Called with both runqueues locked. | |
3349 | */ | |
3350 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3351 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3352 | { | |
5522d5d5 | 3353 | const struct sched_class *class; |
43010659 | 3354 | |
cde7e5ca | 3355 | for_each_class(class) { |
e1d1484f | 3356 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3357 | return 1; |
cde7e5ca | 3358 | } |
43010659 PW |
3359 | |
3360 | return 0; | |
dd41f596 | 3361 | } |
67bb6c03 | 3362 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3363 | /* |
222d656d GS |
3364 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3365 | * during load balancing. | |
1da177e4 | 3366 | */ |
222d656d GS |
3367 | struct sd_lb_stats { |
3368 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3369 | struct sched_group *this; /* Local group in this sd */ | |
3370 | unsigned long total_load; /* Total load of all groups in sd */ | |
3371 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3372 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3373 | ||
3374 | /** Statistics of this group */ | |
3375 | unsigned long this_load; | |
3376 | unsigned long this_load_per_task; | |
3377 | unsigned long this_nr_running; | |
3378 | ||
3379 | /* Statistics of the busiest group */ | |
3380 | unsigned long max_load; | |
3381 | unsigned long busiest_load_per_task; | |
3382 | unsigned long busiest_nr_running; | |
3383 | ||
3384 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3385 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3386 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3387 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3388 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3389 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3390 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3391 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3392 | #endif |
222d656d | 3393 | }; |
1da177e4 | 3394 | |
d5ac537e | 3395 | /* |
381be78f GS |
3396 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3397 | */ | |
3398 | struct sg_lb_stats { | |
3399 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3400 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3401 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3402 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3403 | unsigned long group_capacity; | |
3404 | int group_imb; /* Is there an imbalance in the group ? */ | |
3405 | }; | |
408ed066 | 3406 | |
67bb6c03 GS |
3407 | /** |
3408 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3409 | * @group: The group whose first cpu is to be returned. | |
3410 | */ | |
3411 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3412 | { | |
3413 | return cpumask_first(sched_group_cpus(group)); | |
3414 | } | |
3415 | ||
3416 | /** | |
3417 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3418 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3419 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3420 | */ | |
3421 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3422 | enum cpu_idle_type idle) | |
3423 | { | |
3424 | int load_idx; | |
3425 | ||
3426 | switch (idle) { | |
3427 | case CPU_NOT_IDLE: | |
7897986b | 3428 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3429 | break; |
3430 | ||
3431 | case CPU_NEWLY_IDLE: | |
7897986b | 3432 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3433 | break; |
3434 | default: | |
7897986b | 3435 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3436 | break; |
3437 | } | |
1da177e4 | 3438 | |
67bb6c03 GS |
3439 | return load_idx; |
3440 | } | |
1da177e4 | 3441 | |
1da177e4 | 3442 | |
c071df18 GS |
3443 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3444 | /** | |
3445 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3446 | * the given sched_domain, during load balancing. | |
3447 | * | |
3448 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3449 | * @sds: Variable containing the statistics for sd. | |
3450 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3451 | */ | |
3452 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3453 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3454 | { | |
3455 | /* | |
3456 | * Busy processors will not participate in power savings | |
3457 | * balance. | |
3458 | */ | |
3459 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3460 | sds->power_savings_balance = 0; | |
3461 | else { | |
3462 | sds->power_savings_balance = 1; | |
3463 | sds->min_nr_running = ULONG_MAX; | |
3464 | sds->leader_nr_running = 0; | |
3465 | } | |
3466 | } | |
783609c6 | 3467 | |
c071df18 GS |
3468 | /** |
3469 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3470 | * sched_domain while performing load balancing. | |
3471 | * | |
3472 | * @group: sched_group belonging to the sched_domain under consideration. | |
3473 | * @sds: Variable containing the statistics of the sched_domain | |
3474 | * @local_group: Does group contain the CPU for which we're performing | |
3475 | * load balancing ? | |
3476 | * @sgs: Variable containing the statistics of the group. | |
3477 | */ | |
3478 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3479 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3480 | { | |
408ed066 | 3481 | |
c071df18 GS |
3482 | if (!sds->power_savings_balance) |
3483 | return; | |
1da177e4 | 3484 | |
c071df18 GS |
3485 | /* |
3486 | * If the local group is idle or completely loaded | |
3487 | * no need to do power savings balance at this domain | |
3488 | */ | |
3489 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3490 | !sds->this_nr_running)) | |
3491 | sds->power_savings_balance = 0; | |
2dd73a4f | 3492 | |
c071df18 GS |
3493 | /* |
3494 | * If a group is already running at full capacity or idle, | |
3495 | * don't include that group in power savings calculations | |
3496 | */ | |
3497 | if (!sds->power_savings_balance || | |
3498 | sgs->sum_nr_running >= sgs->group_capacity || | |
3499 | !sgs->sum_nr_running) | |
3500 | return; | |
5969fe06 | 3501 | |
c071df18 GS |
3502 | /* |
3503 | * Calculate the group which has the least non-idle load. | |
3504 | * This is the group from where we need to pick up the load | |
3505 | * for saving power | |
3506 | */ | |
3507 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3508 | (sgs->sum_nr_running == sds->min_nr_running && | |
3509 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3510 | sds->group_min = group; | |
3511 | sds->min_nr_running = sgs->sum_nr_running; | |
3512 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3513 | sgs->sum_nr_running; | |
3514 | } | |
783609c6 | 3515 | |
c071df18 GS |
3516 | /* |
3517 | * Calculate the group which is almost near its | |
3518 | * capacity but still has some space to pick up some load | |
3519 | * from other group and save more power | |
3520 | */ | |
d899a789 | 3521 | if (sgs->sum_nr_running + 1 > sgs->group_capacity) |
c071df18 | 3522 | return; |
1da177e4 | 3523 | |
c071df18 GS |
3524 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3525 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3526 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3527 | sds->group_leader = group; | |
3528 | sds->leader_nr_running = sgs->sum_nr_running; | |
3529 | } | |
3530 | } | |
408ed066 | 3531 | |
c071df18 | 3532 | /** |
d5ac537e | 3533 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3534 | * @sds: Variable containing the statistics of the sched_domain |
3535 | * under consideration. | |
3536 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3537 | * @imbalance: Variable to store the imbalance. | |
3538 | * | |
d5ac537e RD |
3539 | * Description: |
3540 | * Check if we have potential to perform some power-savings balance. | |
3541 | * If yes, set the busiest group to be the least loaded group in the | |
3542 | * sched_domain, so that it's CPUs can be put to idle. | |
3543 | * | |
c071df18 GS |
3544 | * Returns 1 if there is potential to perform power-savings balance. |
3545 | * Else returns 0. | |
3546 | */ | |
3547 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3548 | int this_cpu, unsigned long *imbalance) | |
3549 | { | |
3550 | if (!sds->power_savings_balance) | |
3551 | return 0; | |
1da177e4 | 3552 | |
c071df18 GS |
3553 | if (sds->this != sds->group_leader || |
3554 | sds->group_leader == sds->group_min) | |
3555 | return 0; | |
783609c6 | 3556 | |
c071df18 GS |
3557 | *imbalance = sds->min_load_per_task; |
3558 | sds->busiest = sds->group_min; | |
1da177e4 | 3559 | |
c071df18 | 3560 | return 1; |
1da177e4 | 3561 | |
c071df18 GS |
3562 | } |
3563 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3564 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3565 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3566 | { | |
3567 | return; | |
3568 | } | |
408ed066 | 3569 | |
c071df18 GS |
3570 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3571 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3572 | { | |
3573 | return; | |
3574 | } | |
3575 | ||
3576 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3577 | int this_cpu, unsigned long *imbalance) | |
3578 | { | |
3579 | return 0; | |
3580 | } | |
3581 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3582 | ||
d6a59aa3 PZ |
3583 | |
3584 | unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu) | |
3585 | { | |
3586 | return SCHED_LOAD_SCALE; | |
3587 | } | |
3588 | ||
3589 | unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu) | |
3590 | { | |
3591 | return default_scale_freq_power(sd, cpu); | |
3592 | } | |
3593 | ||
3594 | unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu) | |
ab29230e PZ |
3595 | { |
3596 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3597 | unsigned long smt_gain = sd->smt_gain; | |
3598 | ||
3599 | smt_gain /= weight; | |
3600 | ||
3601 | return smt_gain; | |
3602 | } | |
3603 | ||
d6a59aa3 PZ |
3604 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
3605 | { | |
3606 | return default_scale_smt_power(sd, cpu); | |
3607 | } | |
3608 | ||
e9e9250b PZ |
3609 | unsigned long scale_rt_power(int cpu) |
3610 | { | |
3611 | struct rq *rq = cpu_rq(cpu); | |
3612 | u64 total, available; | |
3613 | ||
3614 | sched_avg_update(rq); | |
3615 | ||
3616 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3617 | available = total - rq->rt_avg; | |
3618 | ||
3619 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3620 | total = SCHED_LOAD_SCALE; | |
3621 | ||
3622 | total >>= SCHED_LOAD_SHIFT; | |
3623 | ||
3624 | return div_u64(available, total); | |
3625 | } | |
3626 | ||
ab29230e PZ |
3627 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3628 | { | |
3629 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3630 | unsigned long power = SCHED_LOAD_SCALE; | |
3631 | struct sched_group *sdg = sd->groups; | |
ab29230e | 3632 | |
8e6598af PZ |
3633 | if (sched_feat(ARCH_POWER)) |
3634 | power *= arch_scale_freq_power(sd, cpu); | |
3635 | else | |
3636 | power *= default_scale_freq_power(sd, cpu); | |
3637 | ||
d6a59aa3 | 3638 | power >>= SCHED_LOAD_SHIFT; |
ab29230e PZ |
3639 | |
3640 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
8e6598af PZ |
3641 | if (sched_feat(ARCH_POWER)) |
3642 | power *= arch_scale_smt_power(sd, cpu); | |
3643 | else | |
3644 | power *= default_scale_smt_power(sd, cpu); | |
3645 | ||
ab29230e PZ |
3646 | power >>= SCHED_LOAD_SHIFT; |
3647 | } | |
3648 | ||
e9e9250b PZ |
3649 | power *= scale_rt_power(cpu); |
3650 | power >>= SCHED_LOAD_SHIFT; | |
3651 | ||
3652 | if (!power) | |
3653 | power = 1; | |
ab29230e | 3654 | |
18a3885f | 3655 | sdg->cpu_power = power; |
ab29230e PZ |
3656 | } |
3657 | ||
3658 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3659 | { |
3660 | struct sched_domain *child = sd->child; | |
3661 | struct sched_group *group, *sdg = sd->groups; | |
d7ea17a7 | 3662 | unsigned long power; |
cc9fba7d PZ |
3663 | |
3664 | if (!child) { | |
ab29230e | 3665 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3666 | return; |
3667 | } | |
3668 | ||
d7ea17a7 | 3669 | power = 0; |
cc9fba7d PZ |
3670 | |
3671 | group = child->groups; | |
3672 | do { | |
d7ea17a7 | 3673 | power += group->cpu_power; |
cc9fba7d PZ |
3674 | group = group->next; |
3675 | } while (group != child->groups); | |
d7ea17a7 IM |
3676 | |
3677 | sdg->cpu_power = power; | |
cc9fba7d | 3678 | } |
c071df18 | 3679 | |
1f8c553d GS |
3680 | /** |
3681 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
e17b38bf | 3682 | * @sd: The sched_domain whose statistics are to be updated. |
1f8c553d GS |
3683 | * @group: sched_group whose statistics are to be updated. |
3684 | * @this_cpu: Cpu for which load balance is currently performed. | |
3685 | * @idle: Idle status of this_cpu | |
3686 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3687 | * @sd_idle: Idle status of the sched_domain containing group. | |
3688 | * @local_group: Does group contain this_cpu. | |
3689 | * @cpus: Set of cpus considered for load balancing. | |
3690 | * @balance: Should we balance. | |
3691 | * @sgs: variable to hold the statistics for this group. | |
3692 | */ | |
cc9fba7d PZ |
3693 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3694 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3695 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3696 | int local_group, const struct cpumask *cpus, | |
3697 | int *balance, struct sg_lb_stats *sgs) | |
3698 | { | |
3699 | unsigned long load, max_cpu_load, min_cpu_load; | |
3700 | int i; | |
3701 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3702 | unsigned long sum_avg_load_per_task; | |
3703 | unsigned long avg_load_per_task; | |
3704 | ||
cc9fba7d | 3705 | if (local_group) { |
1f8c553d | 3706 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3707 | if (balance_cpu == this_cpu) |
ab29230e | 3708 | update_group_power(sd, this_cpu); |
cc9fba7d | 3709 | } |
1f8c553d GS |
3710 | |
3711 | /* Tally up the load of all CPUs in the group */ | |
3712 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3713 | max_cpu_load = 0; | |
3714 | min_cpu_load = ~0UL; | |
408ed066 | 3715 | |
1f8c553d GS |
3716 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3717 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3718 | |
1f8c553d GS |
3719 | if (*sd_idle && rq->nr_running) |
3720 | *sd_idle = 0; | |
5c45bf27 | 3721 | |
1f8c553d | 3722 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3723 | if (local_group) { |
1f8c553d GS |
3724 | if (idle_cpu(i) && !first_idle_cpu) { |
3725 | first_idle_cpu = 1; | |
3726 | balance_cpu = i; | |
3727 | } | |
3728 | ||
3729 | load = target_load(i, load_idx); | |
3730 | } else { | |
3731 | load = source_load(i, load_idx); | |
3732 | if (load > max_cpu_load) | |
3733 | max_cpu_load = load; | |
3734 | if (min_cpu_load > load) | |
3735 | min_cpu_load = load; | |
1da177e4 | 3736 | } |
5c45bf27 | 3737 | |
1f8c553d GS |
3738 | sgs->group_load += load; |
3739 | sgs->sum_nr_running += rq->nr_running; | |
3740 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3741 | |
1f8c553d GS |
3742 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3743 | } | |
5c45bf27 | 3744 | |
1f8c553d GS |
3745 | /* |
3746 | * First idle cpu or the first cpu(busiest) in this sched group | |
3747 | * is eligible for doing load balancing at this and above | |
3748 | * domains. In the newly idle case, we will allow all the cpu's | |
3749 | * to do the newly idle load balance. | |
3750 | */ | |
3751 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3752 | balance_cpu != this_cpu && balance) { | |
3753 | *balance = 0; | |
3754 | return; | |
3755 | } | |
5c45bf27 | 3756 | |
1f8c553d | 3757 | /* Adjust by relative CPU power of the group */ |
18a3885f | 3758 | sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power; |
5c45bf27 | 3759 | |
1f8c553d GS |
3760 | |
3761 | /* | |
3762 | * Consider the group unbalanced when the imbalance is larger | |
3763 | * than the average weight of two tasks. | |
3764 | * | |
3765 | * APZ: with cgroup the avg task weight can vary wildly and | |
3766 | * might not be a suitable number - should we keep a | |
3767 | * normalized nr_running number somewhere that negates | |
3768 | * the hierarchy? | |
3769 | */ | |
18a3885f PZ |
3770 | avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) / |
3771 | group->cpu_power; | |
1f8c553d GS |
3772 | |
3773 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3774 | sgs->group_imb = 1; | |
3775 | ||
bdb94aa5 | 3776 | sgs->group_capacity = |
18a3885f | 3777 | DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE); |
1f8c553d | 3778 | } |
dd41f596 | 3779 | |
37abe198 GS |
3780 | /** |
3781 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3782 | * @sd: sched_domain whose statistics are to be updated. | |
3783 | * @this_cpu: Cpu for which load balance is currently performed. | |
3784 | * @idle: Idle status of this_cpu | |
3785 | * @sd_idle: Idle status of the sched_domain containing group. | |
3786 | * @cpus: Set of cpus considered for load balancing. | |
3787 | * @balance: Should we balance. | |
3788 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3789 | */ |
37abe198 GS |
3790 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3791 | enum cpu_idle_type idle, int *sd_idle, | |
3792 | const struct cpumask *cpus, int *balance, | |
3793 | struct sd_lb_stats *sds) | |
1da177e4 | 3794 | { |
b5d978e0 | 3795 | struct sched_domain *child = sd->child; |
222d656d | 3796 | struct sched_group *group = sd->groups; |
37abe198 | 3797 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3798 | int load_idx, prefer_sibling = 0; |
3799 | ||
3800 | if (child && child->flags & SD_PREFER_SIBLING) | |
3801 | prefer_sibling = 1; | |
222d656d | 3802 | |
c071df18 | 3803 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3804 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3805 | |
3806 | do { | |
1da177e4 | 3807 | int local_group; |
1da177e4 | 3808 | |
758b2cdc RR |
3809 | local_group = cpumask_test_cpu(this_cpu, |
3810 | sched_group_cpus(group)); | |
381be78f | 3811 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3812 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3813 | local_group, cpus, balance, &sgs); |
1da177e4 | 3814 | |
37abe198 GS |
3815 | if (local_group && balance && !(*balance)) |
3816 | return; | |
783609c6 | 3817 | |
37abe198 | 3818 | sds->total_load += sgs.group_load; |
18a3885f | 3819 | sds->total_pwr += group->cpu_power; |
1da177e4 | 3820 | |
b5d978e0 PZ |
3821 | /* |
3822 | * In case the child domain prefers tasks go to siblings | |
3823 | * first, lower the group capacity to one so that we'll try | |
3824 | * and move all the excess tasks away. | |
3825 | */ | |
3826 | if (prefer_sibling) | |
bdb94aa5 | 3827 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
1da177e4 | 3828 | |
1da177e4 | 3829 | if (local_group) { |
37abe198 GS |
3830 | sds->this_load = sgs.avg_load; |
3831 | sds->this = group; | |
3832 | sds->this_nr_running = sgs.sum_nr_running; | |
3833 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3834 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3835 | (sgs.sum_nr_running > sgs.group_capacity || |
3836 | sgs.group_imb)) { | |
37abe198 GS |
3837 | sds->max_load = sgs.avg_load; |
3838 | sds->busiest = group; | |
3839 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3840 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3841 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3842 | } |
5c45bf27 | 3843 | |
c071df18 | 3844 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3845 | group = group->next; |
3846 | } while (group != sd->groups); | |
37abe198 | 3847 | } |
1da177e4 | 3848 | |
2e6f44ae GS |
3849 | /** |
3850 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3851 | * amongst the groups of a sched_domain, during |
3852 | * load balancing. | |
2e6f44ae GS |
3853 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3854 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3855 | * @imbalance: Variable to store the imbalance. | |
3856 | */ | |
3857 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3858 | int this_cpu, unsigned long *imbalance) | |
3859 | { | |
3860 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3861 | unsigned int imbn = 2; | |
3862 | ||
3863 | if (sds->this_nr_running) { | |
3864 | sds->this_load_per_task /= sds->this_nr_running; | |
3865 | if (sds->busiest_load_per_task > | |
3866 | sds->this_load_per_task) | |
3867 | imbn = 1; | |
3868 | } else | |
3869 | sds->this_load_per_task = | |
3870 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3871 | |
2e6f44ae GS |
3872 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3873 | sds->busiest_load_per_task * imbn) { | |
3874 | *imbalance = sds->busiest_load_per_task; | |
3875 | return; | |
3876 | } | |
908a7c1b | 3877 | |
1da177e4 | 3878 | /* |
2e6f44ae GS |
3879 | * OK, we don't have enough imbalance to justify moving tasks, |
3880 | * however we may be able to increase total CPU power used by | |
3881 | * moving them. | |
1da177e4 | 3882 | */ |
2dd73a4f | 3883 | |
18a3885f | 3884 | pwr_now += sds->busiest->cpu_power * |
2e6f44ae | 3885 | min(sds->busiest_load_per_task, sds->max_load); |
18a3885f | 3886 | pwr_now += sds->this->cpu_power * |
2e6f44ae GS |
3887 | min(sds->this_load_per_task, sds->this_load); |
3888 | pwr_now /= SCHED_LOAD_SCALE; | |
3889 | ||
3890 | /* Amount of load we'd subtract */ | |
18a3885f PZ |
3891 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3892 | sds->busiest->cpu_power; | |
2e6f44ae | 3893 | if (sds->max_load > tmp) |
18a3885f | 3894 | pwr_move += sds->busiest->cpu_power * |
2e6f44ae GS |
3895 | min(sds->busiest_load_per_task, sds->max_load - tmp); |
3896 | ||
3897 | /* Amount of load we'd add */ | |
18a3885f | 3898 | if (sds->max_load * sds->busiest->cpu_power < |
2e6f44ae | 3899 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) |
18a3885f PZ |
3900 | tmp = (sds->max_load * sds->busiest->cpu_power) / |
3901 | sds->this->cpu_power; | |
2e6f44ae | 3902 | else |
18a3885f PZ |
3903 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3904 | sds->this->cpu_power; | |
3905 | pwr_move += sds->this->cpu_power * | |
2e6f44ae GS |
3906 | min(sds->this_load_per_task, sds->this_load + tmp); |
3907 | pwr_move /= SCHED_LOAD_SCALE; | |
3908 | ||
3909 | /* Move if we gain throughput */ | |
3910 | if (pwr_move > pwr_now) | |
3911 | *imbalance = sds->busiest_load_per_task; | |
3912 | } | |
dbc523a3 GS |
3913 | |
3914 | /** | |
3915 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3916 | * groups of a given sched_domain during load balance. | |
3917 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3918 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3919 | * @imbalance: The variable to store the imbalance. | |
3920 | */ | |
3921 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3922 | unsigned long *imbalance) | |
3923 | { | |
3924 | unsigned long max_pull; | |
2dd73a4f PW |
3925 | /* |
3926 | * In the presence of smp nice balancing, certain scenarios can have | |
3927 | * max load less than avg load(as we skip the groups at or below | |
3928 | * its cpu_power, while calculating max_load..) | |
3929 | */ | |
dbc523a3 | 3930 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3931 | *imbalance = 0; |
dbc523a3 | 3932 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3933 | } |
0c117f1b SS |
3934 | |
3935 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3936 | max_pull = min(sds->max_load - sds->avg_load, |
3937 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3938 | |
1da177e4 | 3939 | /* How much load to actually move to equalise the imbalance */ |
18a3885f PZ |
3940 | *imbalance = min(max_pull * sds->busiest->cpu_power, |
3941 | (sds->avg_load - sds->this_load) * sds->this->cpu_power) | |
1da177e4 LT |
3942 | / SCHED_LOAD_SCALE; |
3943 | ||
2dd73a4f PW |
3944 | /* |
3945 | * if *imbalance is less than the average load per runnable task | |
3946 | * there is no gaurantee that any tasks will be moved so we'll have | |
3947 | * a think about bumping its value to force at least one task to be | |
3948 | * moved | |
3949 | */ | |
dbc523a3 GS |
3950 | if (*imbalance < sds->busiest_load_per_task) |
3951 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3952 | |
dbc523a3 | 3953 | } |
37abe198 | 3954 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3955 | |
b7bb4c9b GS |
3956 | /** |
3957 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3958 | * if there is an imbalance. If there isn't an imbalance, and | |
3959 | * the user has opted for power-savings, it returns a group whose | |
3960 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3961 | * such a group exists. | |
3962 | * | |
3963 | * Also calculates the amount of weighted load which should be moved | |
3964 | * to restore balance. | |
3965 | * | |
3966 | * @sd: The sched_domain whose busiest group is to be returned. | |
3967 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3968 | * @imbalance: Variable which stores amount of weighted load which should | |
3969 | * be moved to restore balance/put a group to idle. | |
3970 | * @idle: The idle status of this_cpu. | |
3971 | * @sd_idle: The idleness of sd | |
3972 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3973 | * @balance: Pointer to a variable indicating if this_cpu | |
3974 | * is the appropriate cpu to perform load balancing at this_level. | |
3975 | * | |
3976 | * Returns: - the busiest group if imbalance exists. | |
3977 | * - If no imbalance and user has opted for power-savings balance, | |
3978 | * return the least loaded group whose CPUs can be | |
3979 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3980 | */ |
3981 | static struct sched_group * | |
3982 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3983 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3984 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3985 | { | |
3986 | struct sd_lb_stats sds; | |
1da177e4 | 3987 | |
37abe198 | 3988 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3989 | |
37abe198 GS |
3990 | /* |
3991 | * Compute the various statistics relavent for load balancing at | |
3992 | * this level. | |
3993 | */ | |
3994 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3995 | balance, &sds); | |
3996 | ||
b7bb4c9b GS |
3997 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3998 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3999 | * at this level. | |
4000 | * 2) There is no busy sibling group to pull from. | |
4001 | * 3) This group is the busiest group. | |
4002 | * 4) This group is more busy than the avg busieness at this | |
4003 | * sched_domain. | |
4004 | * 5) The imbalance is within the specified limit. | |
4005 | * 6) Any rebalance would lead to ping-pong | |
4006 | */ | |
37abe198 GS |
4007 | if (balance && !(*balance)) |
4008 | goto ret; | |
1da177e4 | 4009 | |
b7bb4c9b GS |
4010 | if (!sds.busiest || sds.busiest_nr_running == 0) |
4011 | goto out_balanced; | |
1da177e4 | 4012 | |
b7bb4c9b | 4013 | if (sds.this_load >= sds.max_load) |
1da177e4 | 4014 | goto out_balanced; |
1da177e4 | 4015 | |
222d656d | 4016 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4017 | |
b7bb4c9b GS |
4018 | if (sds.this_load >= sds.avg_load) |
4019 | goto out_balanced; | |
4020 | ||
4021 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4022 | goto out_balanced; |
4023 | ||
222d656d GS |
4024 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4025 | if (sds.group_imb) | |
4026 | sds.busiest_load_per_task = | |
4027 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4028 | |
1da177e4 LT |
4029 | /* |
4030 | * We're trying to get all the cpus to the average_load, so we don't | |
4031 | * want to push ourselves above the average load, nor do we wish to | |
4032 | * reduce the max loaded cpu below the average load, as either of these | |
4033 | * actions would just result in more rebalancing later, and ping-pong | |
4034 | * tasks around. Thus we look for the minimum possible imbalance. | |
4035 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4036 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4037 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4038 | * appear as very large values with unsigned longs. |
4039 | */ | |
222d656d | 4040 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4041 | goto out_balanced; |
4042 | ||
dbc523a3 GS |
4043 | /* Looks like there is an imbalance. Compute it */ |
4044 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4045 | return sds.busiest; |
1da177e4 LT |
4046 | |
4047 | out_balanced: | |
c071df18 GS |
4048 | /* |
4049 | * There is no obvious imbalance. But check if we can do some balancing | |
4050 | * to save power. | |
4051 | */ | |
4052 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4053 | return sds.busiest; | |
783609c6 | 4054 | ret: |
1da177e4 LT |
4055 | *imbalance = 0; |
4056 | return NULL; | |
4057 | } | |
4058 | ||
4059 | /* | |
4060 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4061 | */ | |
70b97a7f | 4062 | static struct rq * |
d15bcfdb | 4063 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4064 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4065 | { |
70b97a7f | 4066 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4067 | unsigned long max_load = 0; |
1da177e4 LT |
4068 | int i; |
4069 | ||
758b2cdc | 4070 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4071 | unsigned long power = power_of(i); |
4072 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4073 | unsigned long wl; |
0a2966b4 | 4074 | |
96f874e2 | 4075 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4076 | continue; |
4077 | ||
48f24c4d | 4078 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4079 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4080 | wl /= power; | |
2dd73a4f | 4081 | |
bdb94aa5 | 4082 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4083 | continue; |
1da177e4 | 4084 | |
dd41f596 IM |
4085 | if (wl > max_load) { |
4086 | max_load = wl; | |
48f24c4d | 4087 | busiest = rq; |
1da177e4 LT |
4088 | } |
4089 | } | |
4090 | ||
4091 | return busiest; | |
4092 | } | |
4093 | ||
77391d71 NP |
4094 | /* |
4095 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4096 | * so long as it is large enough. | |
4097 | */ | |
4098 | #define MAX_PINNED_INTERVAL 512 | |
4099 | ||
df7c8e84 RR |
4100 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4101 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4102 | ||
1da177e4 LT |
4103 | /* |
4104 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4105 | * tasks if there is an imbalance. | |
1da177e4 | 4106 | */ |
70b97a7f | 4107 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4108 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4109 | int *balance) |
1da177e4 | 4110 | { |
43010659 | 4111 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4112 | struct sched_group *group; |
1da177e4 | 4113 | unsigned long imbalance; |
70b97a7f | 4114 | struct rq *busiest; |
fe2eea3f | 4115 | unsigned long flags; |
df7c8e84 | 4116 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4117 | |
6ad4c188 | 4118 | cpumask_copy(cpus, cpu_active_mask); |
7c16ec58 | 4119 | |
89c4710e SS |
4120 | /* |
4121 | * When power savings policy is enabled for the parent domain, idle | |
4122 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4123 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4124 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4125 | */ |
d15bcfdb | 4126 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4127 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4128 | sd_idle = 1; |
1da177e4 | 4129 | |
2d72376b | 4130 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4131 | |
0a2966b4 | 4132 | redo: |
c8cba857 | 4133 | update_shares(sd); |
0a2966b4 | 4134 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4135 | cpus, balance); |
783609c6 | 4136 | |
06066714 | 4137 | if (*balance == 0) |
783609c6 | 4138 | goto out_balanced; |
783609c6 | 4139 | |
1da177e4 LT |
4140 | if (!group) { |
4141 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4142 | goto out_balanced; | |
4143 | } | |
4144 | ||
7c16ec58 | 4145 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4146 | if (!busiest) { |
4147 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4148 | goto out_balanced; | |
4149 | } | |
4150 | ||
db935dbd | 4151 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4152 | |
4153 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4154 | ||
43010659 | 4155 | ld_moved = 0; |
1da177e4 LT |
4156 | if (busiest->nr_running > 1) { |
4157 | /* | |
4158 | * Attempt to move tasks. If find_busiest_group has found | |
4159 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4160 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4161 | * correctly treated as an imbalance. |
4162 | */ | |
fe2eea3f | 4163 | local_irq_save(flags); |
e17224bf | 4164 | double_rq_lock(this_rq, busiest); |
43010659 | 4165 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4166 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4167 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4168 | local_irq_restore(flags); |
81026794 | 4169 | |
46cb4b7c SS |
4170 | /* |
4171 | * some other cpu did the load balance for us. | |
4172 | */ | |
43010659 | 4173 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4174 | resched_cpu(this_cpu); |
4175 | ||
81026794 | 4176 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4177 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4178 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4179 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4180 | goto redo; |
81026794 | 4181 | goto out_balanced; |
0a2966b4 | 4182 | } |
1da177e4 | 4183 | } |
81026794 | 4184 | |
43010659 | 4185 | if (!ld_moved) { |
1da177e4 LT |
4186 | schedstat_inc(sd, lb_failed[idle]); |
4187 | sd->nr_balance_failed++; | |
4188 | ||
4189 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4190 | |
fe2eea3f | 4191 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4192 | |
4193 | /* don't kick the migration_thread, if the curr | |
4194 | * task on busiest cpu can't be moved to this_cpu | |
4195 | */ | |
96f874e2 RR |
4196 | if (!cpumask_test_cpu(this_cpu, |
4197 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4198 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4199 | all_pinned = 1; |
4200 | goto out_one_pinned; | |
4201 | } | |
4202 | ||
1da177e4 LT |
4203 | if (!busiest->active_balance) { |
4204 | busiest->active_balance = 1; | |
4205 | busiest->push_cpu = this_cpu; | |
81026794 | 4206 | active_balance = 1; |
1da177e4 | 4207 | } |
fe2eea3f | 4208 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4209 | if (active_balance) |
1da177e4 LT |
4210 | wake_up_process(busiest->migration_thread); |
4211 | ||
4212 | /* | |
4213 | * We've kicked active balancing, reset the failure | |
4214 | * counter. | |
4215 | */ | |
39507451 | 4216 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4217 | } |
81026794 | 4218 | } else |
1da177e4 LT |
4219 | sd->nr_balance_failed = 0; |
4220 | ||
81026794 | 4221 | if (likely(!active_balance)) { |
1da177e4 LT |
4222 | /* We were unbalanced, so reset the balancing interval */ |
4223 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4224 | } else { |
4225 | /* | |
4226 | * If we've begun active balancing, start to back off. This | |
4227 | * case may not be covered by the all_pinned logic if there | |
4228 | * is only 1 task on the busy runqueue (because we don't call | |
4229 | * move_tasks). | |
4230 | */ | |
4231 | if (sd->balance_interval < sd->max_interval) | |
4232 | sd->balance_interval *= 2; | |
1da177e4 LT |
4233 | } |
4234 | ||
43010659 | 4235 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4236 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4237 | ld_moved = -1; |
4238 | ||
4239 | goto out; | |
1da177e4 LT |
4240 | |
4241 | out_balanced: | |
1da177e4 LT |
4242 | schedstat_inc(sd, lb_balanced[idle]); |
4243 | ||
16cfb1c0 | 4244 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4245 | |
4246 | out_one_pinned: | |
1da177e4 | 4247 | /* tune up the balancing interval */ |
77391d71 NP |
4248 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4249 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4250 | sd->balance_interval *= 2; |
4251 | ||
48f24c4d | 4252 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4253 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4254 | ld_moved = -1; |
4255 | else | |
4256 | ld_moved = 0; | |
4257 | out: | |
c8cba857 PZ |
4258 | if (ld_moved) |
4259 | update_shares(sd); | |
c09595f6 | 4260 | return ld_moved; |
1da177e4 LT |
4261 | } |
4262 | ||
4263 | /* | |
4264 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4265 | * tasks if there is an imbalance. | |
4266 | * | |
d15bcfdb | 4267 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4268 | * this_rq is locked. |
4269 | */ | |
48f24c4d | 4270 | static int |
df7c8e84 | 4271 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4272 | { |
4273 | struct sched_group *group; | |
70b97a7f | 4274 | struct rq *busiest = NULL; |
1da177e4 | 4275 | unsigned long imbalance; |
43010659 | 4276 | int ld_moved = 0; |
5969fe06 | 4277 | int sd_idle = 0; |
969bb4e4 | 4278 | int all_pinned = 0; |
df7c8e84 | 4279 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4280 | |
6ad4c188 | 4281 | cpumask_copy(cpus, cpu_active_mask); |
5969fe06 | 4282 | |
89c4710e SS |
4283 | /* |
4284 | * When power savings policy is enabled for the parent domain, idle | |
4285 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4286 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4287 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4288 | */ |
4289 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4290 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4291 | sd_idle = 1; |
1da177e4 | 4292 | |
2d72376b | 4293 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4294 | redo: |
3e5459b4 | 4295 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4296 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4297 | &sd_idle, cpus, NULL); |
1da177e4 | 4298 | if (!group) { |
d15bcfdb | 4299 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4300 | goto out_balanced; |
1da177e4 LT |
4301 | } |
4302 | ||
7c16ec58 | 4303 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4304 | if (!busiest) { |
d15bcfdb | 4305 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4306 | goto out_balanced; |
1da177e4 LT |
4307 | } |
4308 | ||
db935dbd NP |
4309 | BUG_ON(busiest == this_rq); |
4310 | ||
d15bcfdb | 4311 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4312 | |
43010659 | 4313 | ld_moved = 0; |
d6d5cfaf NP |
4314 | if (busiest->nr_running > 1) { |
4315 | /* Attempt to move tasks */ | |
4316 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4317 | /* this_rq->clock is already updated */ |
4318 | update_rq_clock(busiest); | |
43010659 | 4319 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4320 | imbalance, sd, CPU_NEWLY_IDLE, |
4321 | &all_pinned); | |
1b12bbc7 | 4322 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4323 | |
969bb4e4 | 4324 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4325 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4326 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4327 | goto redo; |
4328 | } | |
d6d5cfaf NP |
4329 | } |
4330 | ||
43010659 | 4331 | if (!ld_moved) { |
36dffab6 | 4332 | int active_balance = 0; |
ad273b32 | 4333 | |
d15bcfdb | 4334 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4335 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4336 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4337 | return -1; |
ad273b32 VS |
4338 | |
4339 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4340 | return -1; | |
4341 | ||
4342 | if (sd->nr_balance_failed++ < 2) | |
4343 | return -1; | |
4344 | ||
4345 | /* | |
4346 | * The only task running in a non-idle cpu can be moved to this | |
4347 | * cpu in an attempt to completely freeup the other CPU | |
4348 | * package. The same method used to move task in load_balance() | |
4349 | * have been extended for load_balance_newidle() to speedup | |
4350 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4351 | * | |
4352 | * The package power saving logic comes from | |
4353 | * find_busiest_group(). If there are no imbalance, then | |
4354 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4355 | * f_b_g() will select a group from which a running task may be | |
4356 | * pulled to this cpu in order to make the other package idle. | |
4357 | * If there is no opportunity to make a package idle and if | |
4358 | * there are no imbalance, then f_b_g() will return NULL and no | |
4359 | * action will be taken in load_balance_newidle(). | |
4360 | * | |
4361 | * Under normal task pull operation due to imbalance, there | |
4362 | * will be more than one task in the source run queue and | |
4363 | * move_tasks() will succeed. ld_moved will be true and this | |
4364 | * active balance code will not be triggered. | |
4365 | */ | |
4366 | ||
4367 | /* Lock busiest in correct order while this_rq is held */ | |
4368 | double_lock_balance(this_rq, busiest); | |
4369 | ||
4370 | /* | |
4371 | * don't kick the migration_thread, if the curr | |
4372 | * task on busiest cpu can't be moved to this_cpu | |
4373 | */ | |
6ca09dfc | 4374 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4375 | double_unlock_balance(this_rq, busiest); |
4376 | all_pinned = 1; | |
4377 | return ld_moved; | |
4378 | } | |
4379 | ||
4380 | if (!busiest->active_balance) { | |
4381 | busiest->active_balance = 1; | |
4382 | busiest->push_cpu = this_cpu; | |
4383 | active_balance = 1; | |
4384 | } | |
4385 | ||
4386 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4387 | /* |
4388 | * Should not call ttwu while holding a rq->lock | |
4389 | */ | |
4390 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4391 | if (active_balance) |
4392 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4393 | spin_lock(&this_rq->lock); |
ad273b32 | 4394 | |
5969fe06 | 4395 | } else |
16cfb1c0 | 4396 | sd->nr_balance_failed = 0; |
1da177e4 | 4397 | |
3e5459b4 | 4398 | update_shares_locked(this_rq, sd); |
43010659 | 4399 | return ld_moved; |
16cfb1c0 NP |
4400 | |
4401 | out_balanced: | |
d15bcfdb | 4402 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4403 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4404 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4405 | return -1; |
16cfb1c0 | 4406 | sd->nr_balance_failed = 0; |
48f24c4d | 4407 | |
16cfb1c0 | 4408 | return 0; |
1da177e4 LT |
4409 | } |
4410 | ||
4411 | /* | |
4412 | * idle_balance is called by schedule() if this_cpu is about to become | |
4413 | * idle. Attempts to pull tasks from other CPUs. | |
4414 | */ | |
70b97a7f | 4415 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4416 | { |
4417 | struct sched_domain *sd; | |
efbe027e | 4418 | int pulled_task = 0; |
dd41f596 | 4419 | unsigned long next_balance = jiffies + HZ; |
1da177e4 | 4420 | |
1b9508f6 MG |
4421 | this_rq->idle_stamp = this_rq->clock; |
4422 | ||
4423 | if (this_rq->avg_idle < sysctl_sched_migration_cost) | |
4424 | return; | |
4425 | ||
1da177e4 | 4426 | for_each_domain(this_cpu, sd) { |
92c4ca5c CL |
4427 | unsigned long interval; |
4428 | ||
4429 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4430 | continue; | |
4431 | ||
4432 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4433 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4434 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4435 | sd); |
92c4ca5c CL |
4436 | |
4437 | interval = msecs_to_jiffies(sd->balance_interval); | |
4438 | if (time_after(next_balance, sd->last_balance + interval)) | |
4439 | next_balance = sd->last_balance + interval; | |
1b9508f6 MG |
4440 | if (pulled_task) { |
4441 | this_rq->idle_stamp = 0; | |
92c4ca5c | 4442 | break; |
1b9508f6 | 4443 | } |
1da177e4 | 4444 | } |
dd41f596 | 4445 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4446 | /* |
4447 | * We are going idle. next_balance may be set based on | |
4448 | * a busy processor. So reset next_balance. | |
4449 | */ | |
4450 | this_rq->next_balance = next_balance; | |
dd41f596 | 4451 | } |
1da177e4 LT |
4452 | } |
4453 | ||
4454 | /* | |
4455 | * active_load_balance is run by migration threads. It pushes running tasks | |
4456 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4457 | * running on each physical CPU where possible, and avoids physical / | |
4458 | * logical imbalances. | |
4459 | * | |
4460 | * Called with busiest_rq locked. | |
4461 | */ | |
70b97a7f | 4462 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4463 | { |
39507451 | 4464 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4465 | struct sched_domain *sd; |
4466 | struct rq *target_rq; | |
39507451 | 4467 | |
48f24c4d | 4468 | /* Is there any task to move? */ |
39507451 | 4469 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4470 | return; |
4471 | ||
4472 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4473 | |
4474 | /* | |
39507451 | 4475 | * This condition is "impossible", if it occurs |
41a2d6cf | 4476 | * we need to fix it. Originally reported by |
39507451 | 4477 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4478 | */ |
39507451 | 4479 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4480 | |
39507451 NP |
4481 | /* move a task from busiest_rq to target_rq */ |
4482 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4483 | update_rq_clock(busiest_rq); |
4484 | update_rq_clock(target_rq); | |
39507451 NP |
4485 | |
4486 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4487 | for_each_domain(target_cpu, sd) { |
39507451 | 4488 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4489 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4490 | break; |
c96d145e | 4491 | } |
39507451 | 4492 | |
48f24c4d | 4493 | if (likely(sd)) { |
2d72376b | 4494 | schedstat_inc(sd, alb_count); |
39507451 | 4495 | |
43010659 PW |
4496 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4497 | sd, CPU_IDLE)) | |
48f24c4d IM |
4498 | schedstat_inc(sd, alb_pushed); |
4499 | else | |
4500 | schedstat_inc(sd, alb_failed); | |
4501 | } | |
1b12bbc7 | 4502 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4503 | } |
4504 | ||
46cb4b7c SS |
4505 | #ifdef CONFIG_NO_HZ |
4506 | static struct { | |
4507 | atomic_t load_balancer; | |
7d1e6a9b | 4508 | cpumask_var_t cpu_mask; |
f711f609 | 4509 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4510 | } nohz ____cacheline_aligned = { |
4511 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4512 | }; |
4513 | ||
eea08f32 AB |
4514 | int get_nohz_load_balancer(void) |
4515 | { | |
4516 | return atomic_read(&nohz.load_balancer); | |
4517 | } | |
4518 | ||
f711f609 GS |
4519 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4520 | /** | |
4521 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4522 | * @cpu: The cpu whose lowest level of sched domain is to | |
4523 | * be returned. | |
4524 | * @flag: The flag to check for the lowest sched_domain | |
4525 | * for the given cpu. | |
4526 | * | |
4527 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4528 | */ | |
4529 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4530 | { | |
4531 | struct sched_domain *sd; | |
4532 | ||
4533 | for_each_domain(cpu, sd) | |
4534 | if (sd && (sd->flags & flag)) | |
4535 | break; | |
4536 | ||
4537 | return sd; | |
4538 | } | |
4539 | ||
4540 | /** | |
4541 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4542 | * @cpu: The cpu whose domains we're iterating over. | |
4543 | * @sd: variable holding the value of the power_savings_sd | |
4544 | * for cpu. | |
4545 | * @flag: The flag to filter the sched_domains to be iterated. | |
4546 | * | |
4547 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4548 | * set, starting from the lowest sched_domain to the highest. | |
4549 | */ | |
4550 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4551 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4552 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4553 | ||
4554 | /** | |
4555 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4556 | * @ilb_group: group to be checked for semi-idleness | |
4557 | * | |
4558 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4559 | * | |
4560 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4561 | * and atleast one non-idle CPU. This helper function checks if the given | |
4562 | * sched_group is semi-idle or not. | |
4563 | */ | |
4564 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4565 | { | |
4566 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4567 | sched_group_cpus(ilb_group)); | |
4568 | ||
4569 | /* | |
4570 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4571 | * and atleast one idle cpu. | |
4572 | */ | |
4573 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4574 | return 0; | |
4575 | ||
4576 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4577 | return 0; | |
4578 | ||
4579 | return 1; | |
4580 | } | |
4581 | /** | |
4582 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4583 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4584 | * | |
4585 | * Returns: Returns the id of the idle load balancer if it exists, | |
4586 | * Else, returns >= nr_cpu_ids. | |
4587 | * | |
4588 | * This algorithm picks the idle load balancer such that it belongs to a | |
4589 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4590 | * completely idle packages/cores just for the purpose of idle load balancing | |
4591 | * when there are other idle cpu's which are better suited for that job. | |
4592 | */ | |
4593 | static int find_new_ilb(int cpu) | |
4594 | { | |
4595 | struct sched_domain *sd; | |
4596 | struct sched_group *ilb_group; | |
4597 | ||
4598 | /* | |
4599 | * Have idle load balancer selection from semi-idle packages only | |
4600 | * when power-aware load balancing is enabled | |
4601 | */ | |
4602 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4603 | goto out_done; | |
4604 | ||
4605 | /* | |
4606 | * Optimize for the case when we have no idle CPUs or only one | |
4607 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4608 | */ | |
4609 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4610 | goto out_done; | |
4611 | ||
4612 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4613 | ilb_group = sd->groups; | |
4614 | ||
4615 | do { | |
4616 | if (is_semi_idle_group(ilb_group)) | |
4617 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4618 | ||
4619 | ilb_group = ilb_group->next; | |
4620 | ||
4621 | } while (ilb_group != sd->groups); | |
4622 | } | |
4623 | ||
4624 | out_done: | |
4625 | return cpumask_first(nohz.cpu_mask); | |
4626 | } | |
4627 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4628 | static inline int find_new_ilb(int call_cpu) | |
4629 | { | |
6e29ec57 | 4630 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4631 | } |
4632 | #endif | |
4633 | ||
7835b98b | 4634 | /* |
46cb4b7c SS |
4635 | * This routine will try to nominate the ilb (idle load balancing) |
4636 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4637 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4638 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4639 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4640 | * arrives... | |
4641 | * | |
4642 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4643 | * for idle load balancing. ilb owner will still be part of | |
4644 | * nohz.cpu_mask.. | |
7835b98b | 4645 | * |
46cb4b7c SS |
4646 | * While stopping the tick, this cpu will become the ilb owner if there |
4647 | * is no other owner. And will be the owner till that cpu becomes busy | |
4648 | * or if all cpus in the system stop their ticks at which point | |
4649 | * there is no need for ilb owner. | |
4650 | * | |
4651 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4652 | * next busy scheduler_tick() | |
4653 | */ | |
4654 | int select_nohz_load_balancer(int stop_tick) | |
4655 | { | |
4656 | int cpu = smp_processor_id(); | |
4657 | ||
4658 | if (stop_tick) { | |
46cb4b7c SS |
4659 | cpu_rq(cpu)->in_nohz_recently = 1; |
4660 | ||
483b4ee6 SS |
4661 | if (!cpu_active(cpu)) { |
4662 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4663 | return 0; | |
4664 | ||
4665 | /* | |
4666 | * If we are going offline and still the leader, | |
4667 | * give up! | |
4668 | */ | |
46cb4b7c SS |
4669 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4670 | BUG(); | |
483b4ee6 | 4671 | |
46cb4b7c SS |
4672 | return 0; |
4673 | } | |
4674 | ||
483b4ee6 SS |
4675 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4676 | ||
46cb4b7c | 4677 | /* time for ilb owner also to sleep */ |
6ad4c188 | 4678 | if (cpumask_weight(nohz.cpu_mask) == num_active_cpus()) { |
46cb4b7c SS |
4679 | if (atomic_read(&nohz.load_balancer) == cpu) |
4680 | atomic_set(&nohz.load_balancer, -1); | |
4681 | return 0; | |
4682 | } | |
4683 | ||
4684 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4685 | /* make me the ilb owner */ | |
4686 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4687 | return 1; | |
e790fb0b GS |
4688 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4689 | int new_ilb; | |
4690 | ||
4691 | if (!(sched_smt_power_savings || | |
4692 | sched_mc_power_savings)) | |
4693 | return 1; | |
4694 | /* | |
4695 | * Check to see if there is a more power-efficient | |
4696 | * ilb. | |
4697 | */ | |
4698 | new_ilb = find_new_ilb(cpu); | |
4699 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4700 | atomic_set(&nohz.load_balancer, -1); | |
4701 | resched_cpu(new_ilb); | |
4702 | return 0; | |
4703 | } | |
46cb4b7c | 4704 | return 1; |
e790fb0b | 4705 | } |
46cb4b7c | 4706 | } else { |
7d1e6a9b | 4707 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4708 | return 0; |
4709 | ||
7d1e6a9b | 4710 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4711 | |
4712 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4713 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4714 | BUG(); | |
4715 | } | |
4716 | return 0; | |
4717 | } | |
4718 | #endif | |
4719 | ||
4720 | static DEFINE_SPINLOCK(balancing); | |
4721 | ||
4722 | /* | |
7835b98b CL |
4723 | * It checks each scheduling domain to see if it is due to be balanced, |
4724 | * and initiates a balancing operation if so. | |
4725 | * | |
4726 | * Balancing parameters are set up in arch_init_sched_domains. | |
4727 | */ | |
a9957449 | 4728 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4729 | { |
46cb4b7c SS |
4730 | int balance = 1; |
4731 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4732 | unsigned long interval; |
4733 | struct sched_domain *sd; | |
46cb4b7c | 4734 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4735 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4736 | int update_next_balance = 0; |
d07355f5 | 4737 | int need_serialize; |
1da177e4 | 4738 | |
46cb4b7c | 4739 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4740 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4741 | continue; | |
4742 | ||
4743 | interval = sd->balance_interval; | |
d15bcfdb | 4744 | if (idle != CPU_IDLE) |
1da177e4 LT |
4745 | interval *= sd->busy_factor; |
4746 | ||
4747 | /* scale ms to jiffies */ | |
4748 | interval = msecs_to_jiffies(interval); | |
4749 | if (unlikely(!interval)) | |
4750 | interval = 1; | |
dd41f596 IM |
4751 | if (interval > HZ*NR_CPUS/10) |
4752 | interval = HZ*NR_CPUS/10; | |
4753 | ||
d07355f5 | 4754 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4755 | |
d07355f5 | 4756 | if (need_serialize) { |
08c183f3 CL |
4757 | if (!spin_trylock(&balancing)) |
4758 | goto out; | |
4759 | } | |
4760 | ||
c9819f45 | 4761 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4762 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4763 | /* |
4764 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4765 | * longer idle, or one of our SMT siblings is |
4766 | * not idle. | |
4767 | */ | |
d15bcfdb | 4768 | idle = CPU_NOT_IDLE; |
1da177e4 | 4769 | } |
1bd77f2d | 4770 | sd->last_balance = jiffies; |
1da177e4 | 4771 | } |
d07355f5 | 4772 | if (need_serialize) |
08c183f3 CL |
4773 | spin_unlock(&balancing); |
4774 | out: | |
f549da84 | 4775 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4776 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4777 | update_next_balance = 1; |
4778 | } | |
783609c6 SS |
4779 | |
4780 | /* | |
4781 | * Stop the load balance at this level. There is another | |
4782 | * CPU in our sched group which is doing load balancing more | |
4783 | * actively. | |
4784 | */ | |
4785 | if (!balance) | |
4786 | break; | |
1da177e4 | 4787 | } |
f549da84 SS |
4788 | |
4789 | /* | |
4790 | * next_balance will be updated only when there is a need. | |
4791 | * When the cpu is attached to null domain for ex, it will not be | |
4792 | * updated. | |
4793 | */ | |
4794 | if (likely(update_next_balance)) | |
4795 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4796 | } |
4797 | ||
4798 | /* | |
4799 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4800 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4801 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4802 | */ | |
4803 | static void run_rebalance_domains(struct softirq_action *h) | |
4804 | { | |
dd41f596 IM |
4805 | int this_cpu = smp_processor_id(); |
4806 | struct rq *this_rq = cpu_rq(this_cpu); | |
4807 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4808 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4809 | |
dd41f596 | 4810 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4811 | |
4812 | #ifdef CONFIG_NO_HZ | |
4813 | /* | |
4814 | * If this cpu is the owner for idle load balancing, then do the | |
4815 | * balancing on behalf of the other idle cpus whose ticks are | |
4816 | * stopped. | |
4817 | */ | |
dd41f596 IM |
4818 | if (this_rq->idle_at_tick && |
4819 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4820 | struct rq *rq; |
4821 | int balance_cpu; | |
4822 | ||
7d1e6a9b RR |
4823 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4824 | if (balance_cpu == this_cpu) | |
4825 | continue; | |
4826 | ||
46cb4b7c SS |
4827 | /* |
4828 | * If this cpu gets work to do, stop the load balancing | |
4829 | * work being done for other cpus. Next load | |
4830 | * balancing owner will pick it up. | |
4831 | */ | |
4832 | if (need_resched()) | |
4833 | break; | |
4834 | ||
de0cf899 | 4835 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4836 | |
4837 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4838 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4839 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4840 | } |
4841 | } | |
4842 | #endif | |
4843 | } | |
4844 | ||
8a0be9ef FW |
4845 | static inline int on_null_domain(int cpu) |
4846 | { | |
4847 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4848 | } | |
4849 | ||
46cb4b7c SS |
4850 | /* |
4851 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4852 | * | |
4853 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4854 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4855 | * if the whole system is idle. | |
4856 | */ | |
dd41f596 | 4857 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4858 | { |
46cb4b7c SS |
4859 | #ifdef CONFIG_NO_HZ |
4860 | /* | |
4861 | * If we were in the nohz mode recently and busy at the current | |
4862 | * scheduler tick, then check if we need to nominate new idle | |
4863 | * load balancer. | |
4864 | */ | |
4865 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4866 | rq->in_nohz_recently = 0; | |
4867 | ||
4868 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4869 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4870 | atomic_set(&nohz.load_balancer, -1); |
4871 | } | |
4872 | ||
4873 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4874 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4875 | |
434d53b0 | 4876 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4877 | resched_cpu(ilb); |
4878 | } | |
4879 | } | |
4880 | ||
4881 | /* | |
4882 | * If this cpu is idle and doing idle load balancing for all the | |
4883 | * cpus with ticks stopped, is it time for that to stop? | |
4884 | */ | |
4885 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4886 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4887 | resched_cpu(cpu); |
4888 | return; | |
4889 | } | |
4890 | ||
4891 | /* | |
4892 | * If this cpu is idle and the idle load balancing is done by | |
4893 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4894 | */ | |
4895 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4896 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4897 | return; |
4898 | #endif | |
8a0be9ef FW |
4899 | /* Don't need to rebalance while attached to NULL domain */ |
4900 | if (time_after_eq(jiffies, rq->next_balance) && | |
4901 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4902 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4903 | } |
dd41f596 IM |
4904 | |
4905 | #else /* CONFIG_SMP */ | |
4906 | ||
1da177e4 LT |
4907 | /* |
4908 | * on UP we do not need to balance between CPUs: | |
4909 | */ | |
70b97a7f | 4910 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4911 | { |
4912 | } | |
dd41f596 | 4913 | |
1da177e4 LT |
4914 | #endif |
4915 | ||
1da177e4 LT |
4916 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4917 | ||
4918 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4919 | ||
4920 | /* | |
c5f8d995 | 4921 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4922 | * @p in case that task is currently running. |
c5f8d995 HS |
4923 | * |
4924 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4925 | */ |
c5f8d995 HS |
4926 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4927 | { | |
4928 | u64 ns = 0; | |
4929 | ||
4930 | if (task_current(rq, p)) { | |
4931 | update_rq_clock(rq); | |
4932 | ns = rq->clock - p->se.exec_start; | |
4933 | if ((s64)ns < 0) | |
4934 | ns = 0; | |
4935 | } | |
4936 | ||
4937 | return ns; | |
4938 | } | |
4939 | ||
bb34d92f | 4940 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4941 | { |
1da177e4 | 4942 | unsigned long flags; |
41b86e9c | 4943 | struct rq *rq; |
bb34d92f | 4944 | u64 ns = 0; |
48f24c4d | 4945 | |
41b86e9c | 4946 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4947 | ns = do_task_delta_exec(p, rq); |
4948 | task_rq_unlock(rq, &flags); | |
1508487e | 4949 | |
c5f8d995 HS |
4950 | return ns; |
4951 | } | |
f06febc9 | 4952 | |
c5f8d995 HS |
4953 | /* |
4954 | * Return accounted runtime for the task. | |
4955 | * In case the task is currently running, return the runtime plus current's | |
4956 | * pending runtime that have not been accounted yet. | |
4957 | */ | |
4958 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4959 | { | |
4960 | unsigned long flags; | |
4961 | struct rq *rq; | |
4962 | u64 ns = 0; | |
4963 | ||
4964 | rq = task_rq_lock(p, &flags); | |
4965 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4966 | task_rq_unlock(rq, &flags); | |
4967 | ||
4968 | return ns; | |
4969 | } | |
48f24c4d | 4970 | |
c5f8d995 HS |
4971 | /* |
4972 | * Return sum_exec_runtime for the thread group. | |
4973 | * In case the task is currently running, return the sum plus current's | |
4974 | * pending runtime that have not been accounted yet. | |
4975 | * | |
4976 | * Note that the thread group might have other running tasks as well, | |
4977 | * so the return value not includes other pending runtime that other | |
4978 | * running tasks might have. | |
4979 | */ | |
4980 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4981 | { | |
4982 | struct task_cputime totals; | |
4983 | unsigned long flags; | |
4984 | struct rq *rq; | |
4985 | u64 ns; | |
4986 | ||
4987 | rq = task_rq_lock(p, &flags); | |
4988 | thread_group_cputime(p, &totals); | |
4989 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4990 | task_rq_unlock(rq, &flags); |
48f24c4d | 4991 | |
1da177e4 LT |
4992 | return ns; |
4993 | } | |
4994 | ||
1da177e4 LT |
4995 | /* |
4996 | * Account user cpu time to a process. | |
4997 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4998 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4999 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 5000 | */ |
457533a7 MS |
5001 | void account_user_time(struct task_struct *p, cputime_t cputime, |
5002 | cputime_t cputime_scaled) | |
1da177e4 LT |
5003 | { |
5004 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5005 | cputime64_t tmp; | |
5006 | ||
457533a7 | 5007 | /* Add user time to process. */ |
1da177e4 | 5008 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5009 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5010 | account_group_user_time(p, cputime); |
1da177e4 LT |
5011 | |
5012 | /* Add user time to cpustat. */ | |
5013 | tmp = cputime_to_cputime64(cputime); | |
5014 | if (TASK_NICE(p) > 0) | |
5015 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5016 | else | |
5017 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5018 | |
5019 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5020 | /* Account for user time used */ |
5021 | acct_update_integrals(p); | |
1da177e4 LT |
5022 | } |
5023 | ||
94886b84 LV |
5024 | /* |
5025 | * Account guest cpu time to a process. | |
5026 | * @p: the process that the cpu time gets accounted to | |
5027 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5028 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5029 | */ |
457533a7 MS |
5030 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5031 | cputime_t cputime_scaled) | |
94886b84 LV |
5032 | { |
5033 | cputime64_t tmp; | |
5034 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5035 | ||
5036 | tmp = cputime_to_cputime64(cputime); | |
5037 | ||
457533a7 | 5038 | /* Add guest time to process. */ |
94886b84 | 5039 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5040 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5041 | account_group_user_time(p, cputime); |
94886b84 LV |
5042 | p->gtime = cputime_add(p->gtime, cputime); |
5043 | ||
457533a7 | 5044 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
5045 | if (TASK_NICE(p) > 0) { |
5046 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5047 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
5048 | } else { | |
5049 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
5050 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5051 | } | |
94886b84 LV |
5052 | } |
5053 | ||
1da177e4 LT |
5054 | /* |
5055 | * Account system cpu time to a process. | |
5056 | * @p: the process that the cpu time gets accounted to | |
5057 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5058 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5059 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5060 | */ |
5061 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5062 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5063 | { |
5064 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5065 | cputime64_t tmp; |
5066 | ||
983ed7a6 | 5067 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5068 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5069 | return; |
5070 | } | |
94886b84 | 5071 | |
457533a7 | 5072 | /* Add system time to process. */ |
1da177e4 | 5073 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5074 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5075 | account_group_system_time(p, cputime); |
1da177e4 LT |
5076 | |
5077 | /* Add system time to cpustat. */ | |
5078 | tmp = cputime_to_cputime64(cputime); | |
5079 | if (hardirq_count() - hardirq_offset) | |
5080 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5081 | else if (softirq_count()) | |
5082 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5083 | else |
79741dd3 MS |
5084 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5085 | ||
ef12fefa BR |
5086 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5087 | ||
1da177e4 LT |
5088 | /* Account for system time used */ |
5089 | acct_update_integrals(p); | |
1da177e4 LT |
5090 | } |
5091 | ||
c66f08be | 5092 | /* |
1da177e4 | 5093 | * Account for involuntary wait time. |
1da177e4 | 5094 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5095 | */ |
79741dd3 | 5096 | void account_steal_time(cputime_t cputime) |
c66f08be | 5097 | { |
79741dd3 MS |
5098 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5099 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5100 | ||
5101 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5102 | } |
5103 | ||
1da177e4 | 5104 | /* |
79741dd3 MS |
5105 | * Account for idle time. |
5106 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5107 | */ |
79741dd3 | 5108 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5109 | { |
5110 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5111 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5112 | struct rq *rq = this_rq(); |
1da177e4 | 5113 | |
79741dd3 MS |
5114 | if (atomic_read(&rq->nr_iowait) > 0) |
5115 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5116 | else | |
5117 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5118 | } |
5119 | ||
79741dd3 MS |
5120 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5121 | ||
5122 | /* | |
5123 | * Account a single tick of cpu time. | |
5124 | * @p: the process that the cpu time gets accounted to | |
5125 | * @user_tick: indicates if the tick is a user or a system tick | |
5126 | */ | |
5127 | void account_process_tick(struct task_struct *p, int user_tick) | |
5128 | { | |
a42548a1 | 5129 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
5130 | struct rq *rq = this_rq(); |
5131 | ||
5132 | if (user_tick) | |
a42548a1 | 5133 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 5134 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 5135 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
5136 | one_jiffy_scaled); |
5137 | else | |
a42548a1 | 5138 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
5139 | } |
5140 | ||
5141 | /* | |
5142 | * Account multiple ticks of steal time. | |
5143 | * @p: the process from which the cpu time has been stolen | |
5144 | * @ticks: number of stolen ticks | |
5145 | */ | |
5146 | void account_steal_ticks(unsigned long ticks) | |
5147 | { | |
5148 | account_steal_time(jiffies_to_cputime(ticks)); | |
5149 | } | |
5150 | ||
5151 | /* | |
5152 | * Account multiple ticks of idle time. | |
5153 | * @ticks: number of stolen ticks | |
5154 | */ | |
5155 | void account_idle_ticks(unsigned long ticks) | |
5156 | { | |
5157 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5158 | } |
5159 | ||
79741dd3 MS |
5160 | #endif |
5161 | ||
49048622 BS |
5162 | /* |
5163 | * Use precise platform statistics if available: | |
5164 | */ | |
5165 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 5166 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5167 | { |
d99ca3b9 HS |
5168 | *ut = p->utime; |
5169 | *st = p->stime; | |
49048622 BS |
5170 | } |
5171 | ||
0cf55e1e | 5172 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5173 | { |
0cf55e1e HS |
5174 | struct task_cputime cputime; |
5175 | ||
5176 | thread_group_cputime(p, &cputime); | |
5177 | ||
5178 | *ut = cputime.utime; | |
5179 | *st = cputime.stime; | |
49048622 BS |
5180 | } |
5181 | #else | |
761b1d26 HS |
5182 | |
5183 | #ifndef nsecs_to_cputime | |
b7b20df9 | 5184 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
5185 | #endif |
5186 | ||
d180c5bc | 5187 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5188 | { |
d99ca3b9 | 5189 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
5190 | |
5191 | /* | |
5192 | * Use CFS's precise accounting: | |
5193 | */ | |
d180c5bc | 5194 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
5195 | |
5196 | if (total) { | |
d180c5bc HS |
5197 | u64 temp; |
5198 | ||
5199 | temp = (u64)(rtime * utime); | |
49048622 | 5200 | do_div(temp, total); |
d180c5bc HS |
5201 | utime = (cputime_t)temp; |
5202 | } else | |
5203 | utime = rtime; | |
49048622 | 5204 | |
d180c5bc HS |
5205 | /* |
5206 | * Compare with previous values, to keep monotonicity: | |
5207 | */ | |
761b1d26 | 5208 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 5209 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 5210 | |
d99ca3b9 HS |
5211 | *ut = p->prev_utime; |
5212 | *st = p->prev_stime; | |
49048622 BS |
5213 | } |
5214 | ||
0cf55e1e HS |
5215 | /* |
5216 | * Must be called with siglock held. | |
5217 | */ | |
5218 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 5219 | { |
0cf55e1e HS |
5220 | struct signal_struct *sig = p->signal; |
5221 | struct task_cputime cputime; | |
5222 | cputime_t rtime, utime, total; | |
49048622 | 5223 | |
0cf55e1e | 5224 | thread_group_cputime(p, &cputime); |
49048622 | 5225 | |
0cf55e1e HS |
5226 | total = cputime_add(cputime.utime, cputime.stime); |
5227 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 5228 | |
0cf55e1e HS |
5229 | if (total) { |
5230 | u64 temp; | |
49048622 | 5231 | |
0cf55e1e HS |
5232 | temp = (u64)(rtime * cputime.utime); |
5233 | do_div(temp, total); | |
5234 | utime = (cputime_t)temp; | |
5235 | } else | |
5236 | utime = rtime; | |
5237 | ||
5238 | sig->prev_utime = max(sig->prev_utime, utime); | |
5239 | sig->prev_stime = max(sig->prev_stime, | |
5240 | cputime_sub(rtime, sig->prev_utime)); | |
5241 | ||
5242 | *ut = sig->prev_utime; | |
5243 | *st = sig->prev_stime; | |
49048622 | 5244 | } |
49048622 | 5245 | #endif |
49048622 | 5246 | |
7835b98b CL |
5247 | /* |
5248 | * This function gets called by the timer code, with HZ frequency. | |
5249 | * We call it with interrupts disabled. | |
5250 | * | |
5251 | * It also gets called by the fork code, when changing the parent's | |
5252 | * timeslices. | |
5253 | */ | |
5254 | void scheduler_tick(void) | |
5255 | { | |
7835b98b CL |
5256 | int cpu = smp_processor_id(); |
5257 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5258 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5259 | |
5260 | sched_clock_tick(); | |
dd41f596 IM |
5261 | |
5262 | spin_lock(&rq->lock); | |
3e51f33f | 5263 | update_rq_clock(rq); |
f1a438d8 | 5264 | update_cpu_load(rq); |
fa85ae24 | 5265 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5266 | spin_unlock(&rq->lock); |
7835b98b | 5267 | |
cdd6c482 | 5268 | perf_event_task_tick(curr, cpu); |
e220d2dc | 5269 | |
e418e1c2 | 5270 | #ifdef CONFIG_SMP |
dd41f596 IM |
5271 | rq->idle_at_tick = idle_cpu(cpu); |
5272 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5273 | #endif |
1da177e4 LT |
5274 | } |
5275 | ||
132380a0 | 5276 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5277 | { |
5278 | if (in_lock_functions(addr)) { | |
5279 | addr = CALLER_ADDR2; | |
5280 | if (in_lock_functions(addr)) | |
5281 | addr = CALLER_ADDR3; | |
5282 | } | |
5283 | return addr; | |
5284 | } | |
1da177e4 | 5285 | |
7e49fcce SR |
5286 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5287 | defined(CONFIG_PREEMPT_TRACER)) | |
5288 | ||
43627582 | 5289 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5290 | { |
6cd8a4bb | 5291 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5292 | /* |
5293 | * Underflow? | |
5294 | */ | |
9a11b49a IM |
5295 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5296 | return; | |
6cd8a4bb | 5297 | #endif |
1da177e4 | 5298 | preempt_count() += val; |
6cd8a4bb | 5299 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5300 | /* |
5301 | * Spinlock count overflowing soon? | |
5302 | */ | |
33859f7f MOS |
5303 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5304 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5305 | #endif |
5306 | if (preempt_count() == val) | |
5307 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5308 | } |
5309 | EXPORT_SYMBOL(add_preempt_count); | |
5310 | ||
43627582 | 5311 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5312 | { |
6cd8a4bb | 5313 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5314 | /* |
5315 | * Underflow? | |
5316 | */ | |
01e3eb82 | 5317 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5318 | return; |
1da177e4 LT |
5319 | /* |
5320 | * Is the spinlock portion underflowing? | |
5321 | */ | |
9a11b49a IM |
5322 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5323 | !(preempt_count() & PREEMPT_MASK))) | |
5324 | return; | |
6cd8a4bb | 5325 | #endif |
9a11b49a | 5326 | |
6cd8a4bb SR |
5327 | if (preempt_count() == val) |
5328 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5329 | preempt_count() -= val; |
5330 | } | |
5331 | EXPORT_SYMBOL(sub_preempt_count); | |
5332 | ||
5333 | #endif | |
5334 | ||
5335 | /* | |
dd41f596 | 5336 | * Print scheduling while atomic bug: |
1da177e4 | 5337 | */ |
dd41f596 | 5338 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5339 | { |
838225b4 SS |
5340 | struct pt_regs *regs = get_irq_regs(); |
5341 | ||
663997d4 JP |
5342 | pr_err("BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5343 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5344 | |
dd41f596 | 5345 | debug_show_held_locks(prev); |
e21f5b15 | 5346 | print_modules(); |
dd41f596 IM |
5347 | if (irqs_disabled()) |
5348 | print_irqtrace_events(prev); | |
838225b4 SS |
5349 | |
5350 | if (regs) | |
5351 | show_regs(regs); | |
5352 | else | |
5353 | dump_stack(); | |
dd41f596 | 5354 | } |
1da177e4 | 5355 | |
dd41f596 IM |
5356 | /* |
5357 | * Various schedule()-time debugging checks and statistics: | |
5358 | */ | |
5359 | static inline void schedule_debug(struct task_struct *prev) | |
5360 | { | |
1da177e4 | 5361 | /* |
41a2d6cf | 5362 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5363 | * schedule() atomically, we ignore that path for now. |
5364 | * Otherwise, whine if we are scheduling when we should not be. | |
5365 | */ | |
3f33a7ce | 5366 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5367 | __schedule_bug(prev); |
5368 | ||
1da177e4 LT |
5369 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5370 | ||
2d72376b | 5371 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5372 | #ifdef CONFIG_SCHEDSTATS |
5373 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5374 | schedstat_inc(this_rq(), bkl_count); |
5375 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5376 | } |
5377 | #endif | |
dd41f596 IM |
5378 | } |
5379 | ||
6cecd084 | 5380 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 5381 | { |
6cecd084 PZ |
5382 | if (prev->state == TASK_RUNNING) { |
5383 | u64 runtime = prev->se.sum_exec_runtime; | |
df1c99d4 | 5384 | |
6cecd084 PZ |
5385 | runtime -= prev->se.prev_sum_exec_runtime; |
5386 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
df1c99d4 MG |
5387 | |
5388 | /* | |
5389 | * In order to avoid avg_overlap growing stale when we are | |
5390 | * indeed overlapping and hence not getting put to sleep, grow | |
5391 | * the avg_overlap on preemption. | |
5392 | * | |
5393 | * We use the average preemption runtime because that | |
5394 | * correlates to the amount of cache footprint a task can | |
5395 | * build up. | |
5396 | */ | |
6cecd084 | 5397 | update_avg(&prev->se.avg_overlap, runtime); |
df1c99d4 | 5398 | } |
6cecd084 | 5399 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
5400 | } |
5401 | ||
dd41f596 IM |
5402 | /* |
5403 | * Pick up the highest-prio task: | |
5404 | */ | |
5405 | static inline struct task_struct * | |
b67802ea | 5406 | pick_next_task(struct rq *rq) |
dd41f596 | 5407 | { |
5522d5d5 | 5408 | const struct sched_class *class; |
dd41f596 | 5409 | struct task_struct *p; |
1da177e4 LT |
5410 | |
5411 | /* | |
dd41f596 IM |
5412 | * Optimization: we know that if all tasks are in |
5413 | * the fair class we can call that function directly: | |
1da177e4 | 5414 | */ |
dd41f596 | 5415 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5416 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5417 | if (likely(p)) |
5418 | return p; | |
1da177e4 LT |
5419 | } |
5420 | ||
dd41f596 IM |
5421 | class = sched_class_highest; |
5422 | for ( ; ; ) { | |
fb8d4724 | 5423 | p = class->pick_next_task(rq); |
dd41f596 IM |
5424 | if (p) |
5425 | return p; | |
5426 | /* | |
5427 | * Will never be NULL as the idle class always | |
5428 | * returns a non-NULL p: | |
5429 | */ | |
5430 | class = class->next; | |
5431 | } | |
5432 | } | |
1da177e4 | 5433 | |
dd41f596 IM |
5434 | /* |
5435 | * schedule() is the main scheduler function. | |
5436 | */ | |
ff743345 | 5437 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5438 | { |
5439 | struct task_struct *prev, *next; | |
67ca7bde | 5440 | unsigned long *switch_count; |
dd41f596 | 5441 | struct rq *rq; |
31656519 | 5442 | int cpu; |
dd41f596 | 5443 | |
ff743345 PZ |
5444 | need_resched: |
5445 | preempt_disable(); | |
dd41f596 IM |
5446 | cpu = smp_processor_id(); |
5447 | rq = cpu_rq(cpu); | |
d6714c22 | 5448 | rcu_sched_qs(cpu); |
dd41f596 IM |
5449 | prev = rq->curr; |
5450 | switch_count = &prev->nivcsw; | |
5451 | ||
5452 | release_kernel_lock(prev); | |
5453 | need_resched_nonpreemptible: | |
5454 | ||
5455 | schedule_debug(prev); | |
1da177e4 | 5456 | |
31656519 | 5457 | if (sched_feat(HRTICK)) |
f333fdc9 | 5458 | hrtick_clear(rq); |
8f4d37ec | 5459 | |
8cd162ce | 5460 | spin_lock_irq(&rq->lock); |
3e51f33f | 5461 | update_rq_clock(rq); |
1e819950 | 5462 | clear_tsk_need_resched(prev); |
1da177e4 | 5463 | |
1da177e4 | 5464 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5465 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5466 | prev->state = TASK_RUNNING; |
16882c1e | 5467 | else |
2e1cb74a | 5468 | deactivate_task(rq, prev, 1); |
dd41f596 | 5469 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5470 | } |
5471 | ||
3f029d3c | 5472 | pre_schedule(rq, prev); |
f65eda4f | 5473 | |
dd41f596 | 5474 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5475 | idle_balance(cpu, rq); |
1da177e4 | 5476 | |
df1c99d4 | 5477 | put_prev_task(rq, prev); |
b67802ea | 5478 | next = pick_next_task(rq); |
1da177e4 | 5479 | |
1da177e4 | 5480 | if (likely(prev != next)) { |
673a90a1 | 5481 | sched_info_switch(prev, next); |
cdd6c482 | 5482 | perf_event_task_sched_out(prev, next, cpu); |
673a90a1 | 5483 | |
1da177e4 LT |
5484 | rq->nr_switches++; |
5485 | rq->curr = next; | |
5486 | ++*switch_count; | |
5487 | ||
dd41f596 | 5488 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5489 | /* |
5490 | * the context switch might have flipped the stack from under | |
5491 | * us, hence refresh the local variables. | |
5492 | */ | |
5493 | cpu = smp_processor_id(); | |
5494 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5495 | } else |
5496 | spin_unlock_irq(&rq->lock); | |
5497 | ||
3f029d3c | 5498 | post_schedule(rq); |
1da177e4 | 5499 | |
8f4d37ec | 5500 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5501 | goto need_resched_nonpreemptible; |
8f4d37ec | 5502 | |
1da177e4 | 5503 | preempt_enable_no_resched(); |
ff743345 | 5504 | if (need_resched()) |
1da177e4 LT |
5505 | goto need_resched; |
5506 | } | |
1da177e4 LT |
5507 | EXPORT_SYMBOL(schedule); |
5508 | ||
c08f7829 | 5509 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
5510 | /* |
5511 | * Look out! "owner" is an entirely speculative pointer | |
5512 | * access and not reliable. | |
5513 | */ | |
5514 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5515 | { | |
5516 | unsigned int cpu; | |
5517 | struct rq *rq; | |
5518 | ||
5519 | if (!sched_feat(OWNER_SPIN)) | |
5520 | return 0; | |
5521 | ||
5522 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5523 | /* | |
5524 | * Need to access the cpu field knowing that | |
5525 | * DEBUG_PAGEALLOC could have unmapped it if | |
5526 | * the mutex owner just released it and exited. | |
5527 | */ | |
5528 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5529 | goto out; | |
5530 | #else | |
5531 | cpu = owner->cpu; | |
5532 | #endif | |
5533 | ||
5534 | /* | |
5535 | * Even if the access succeeded (likely case), | |
5536 | * the cpu field may no longer be valid. | |
5537 | */ | |
5538 | if (cpu >= nr_cpumask_bits) | |
5539 | goto out; | |
5540 | ||
5541 | /* | |
5542 | * We need to validate that we can do a | |
5543 | * get_cpu() and that we have the percpu area. | |
5544 | */ | |
5545 | if (!cpu_online(cpu)) | |
5546 | goto out; | |
5547 | ||
5548 | rq = cpu_rq(cpu); | |
5549 | ||
5550 | for (;;) { | |
5551 | /* | |
5552 | * Owner changed, break to re-assess state. | |
5553 | */ | |
5554 | if (lock->owner != owner) | |
5555 | break; | |
5556 | ||
5557 | /* | |
5558 | * Is that owner really running on that cpu? | |
5559 | */ | |
5560 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5561 | return 0; | |
5562 | ||
5563 | cpu_relax(); | |
5564 | } | |
5565 | out: | |
5566 | return 1; | |
5567 | } | |
5568 | #endif | |
5569 | ||
1da177e4 LT |
5570 | #ifdef CONFIG_PREEMPT |
5571 | /* | |
2ed6e34f | 5572 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5573 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5574 | * occur there and call schedule directly. |
5575 | */ | |
5576 | asmlinkage void __sched preempt_schedule(void) | |
5577 | { | |
5578 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5579 | |
1da177e4 LT |
5580 | /* |
5581 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5582 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5583 | */ |
beed33a8 | 5584 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5585 | return; |
5586 | ||
3a5c359a AK |
5587 | do { |
5588 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5589 | schedule(); |
3a5c359a | 5590 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5591 | |
3a5c359a AK |
5592 | /* |
5593 | * Check again in case we missed a preemption opportunity | |
5594 | * between schedule and now. | |
5595 | */ | |
5596 | barrier(); | |
5ed0cec0 | 5597 | } while (need_resched()); |
1da177e4 | 5598 | } |
1da177e4 LT |
5599 | EXPORT_SYMBOL(preempt_schedule); |
5600 | ||
5601 | /* | |
2ed6e34f | 5602 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5603 | * off of irq context. |
5604 | * Note, that this is called and return with irqs disabled. This will | |
5605 | * protect us against recursive calling from irq. | |
5606 | */ | |
5607 | asmlinkage void __sched preempt_schedule_irq(void) | |
5608 | { | |
5609 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5610 | |
2ed6e34f | 5611 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5612 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5613 | ||
3a5c359a AK |
5614 | do { |
5615 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5616 | local_irq_enable(); |
5617 | schedule(); | |
5618 | local_irq_disable(); | |
3a5c359a | 5619 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5620 | |
3a5c359a AK |
5621 | /* |
5622 | * Check again in case we missed a preemption opportunity | |
5623 | * between schedule and now. | |
5624 | */ | |
5625 | barrier(); | |
5ed0cec0 | 5626 | } while (need_resched()); |
1da177e4 LT |
5627 | } |
5628 | ||
5629 | #endif /* CONFIG_PREEMPT */ | |
5630 | ||
63859d4f | 5631 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5632 | void *key) |
1da177e4 | 5633 | { |
63859d4f | 5634 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5635 | } |
1da177e4 LT |
5636 | EXPORT_SYMBOL(default_wake_function); |
5637 | ||
5638 | /* | |
41a2d6cf IM |
5639 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5640 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5641 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5642 | * | |
5643 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5644 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5645 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5646 | */ | |
78ddb08f | 5647 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 5648 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 5649 | { |
2e45874c | 5650 | wait_queue_t *curr, *next; |
1da177e4 | 5651 | |
2e45874c | 5652 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5653 | unsigned flags = curr->flags; |
5654 | ||
63859d4f | 5655 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 5656 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5657 | break; |
5658 | } | |
5659 | } | |
5660 | ||
5661 | /** | |
5662 | * __wake_up - wake up threads blocked on a waitqueue. | |
5663 | * @q: the waitqueue | |
5664 | * @mode: which threads | |
5665 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5666 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5667 | * |
5668 | * It may be assumed that this function implies a write memory barrier before | |
5669 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5670 | */ |
7ad5b3a5 | 5671 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5672 | int nr_exclusive, void *key) |
1da177e4 LT |
5673 | { |
5674 | unsigned long flags; | |
5675 | ||
5676 | spin_lock_irqsave(&q->lock, flags); | |
5677 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5678 | spin_unlock_irqrestore(&q->lock, flags); | |
5679 | } | |
1da177e4 LT |
5680 | EXPORT_SYMBOL(__wake_up); |
5681 | ||
5682 | /* | |
5683 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5684 | */ | |
7ad5b3a5 | 5685 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5686 | { |
5687 | __wake_up_common(q, mode, 1, 0, NULL); | |
5688 | } | |
5689 | ||
4ede816a DL |
5690 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5691 | { | |
5692 | __wake_up_common(q, mode, 1, 0, key); | |
5693 | } | |
5694 | ||
1da177e4 | 5695 | /** |
4ede816a | 5696 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5697 | * @q: the waitqueue |
5698 | * @mode: which threads | |
5699 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5700 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5701 | * |
5702 | * The sync wakeup differs that the waker knows that it will schedule | |
5703 | * away soon, so while the target thread will be woken up, it will not | |
5704 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5705 | * with each other. This can prevent needless bouncing between CPUs. | |
5706 | * | |
5707 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5708 | * |
5709 | * It may be assumed that this function implies a write memory barrier before | |
5710 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5711 | */ |
4ede816a DL |
5712 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5713 | int nr_exclusive, void *key) | |
1da177e4 LT |
5714 | { |
5715 | unsigned long flags; | |
7d478721 | 5716 | int wake_flags = WF_SYNC; |
1da177e4 LT |
5717 | |
5718 | if (unlikely(!q)) | |
5719 | return; | |
5720 | ||
5721 | if (unlikely(!nr_exclusive)) | |
7d478721 | 5722 | wake_flags = 0; |
1da177e4 LT |
5723 | |
5724 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 5725 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
5726 | spin_unlock_irqrestore(&q->lock, flags); |
5727 | } | |
4ede816a DL |
5728 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5729 | ||
5730 | /* | |
5731 | * __wake_up_sync - see __wake_up_sync_key() | |
5732 | */ | |
5733 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5734 | { | |
5735 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5736 | } | |
1da177e4 LT |
5737 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5738 | ||
65eb3dc6 KD |
5739 | /** |
5740 | * complete: - signals a single thread waiting on this completion | |
5741 | * @x: holds the state of this particular completion | |
5742 | * | |
5743 | * This will wake up a single thread waiting on this completion. Threads will be | |
5744 | * awakened in the same order in which they were queued. | |
5745 | * | |
5746 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5747 | * |
5748 | * It may be assumed that this function implies a write memory barrier before | |
5749 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5750 | */ |
b15136e9 | 5751 | void complete(struct completion *x) |
1da177e4 LT |
5752 | { |
5753 | unsigned long flags; | |
5754 | ||
5755 | spin_lock_irqsave(&x->wait.lock, flags); | |
5756 | x->done++; | |
d9514f6c | 5757 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5758 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5759 | } | |
5760 | EXPORT_SYMBOL(complete); | |
5761 | ||
65eb3dc6 KD |
5762 | /** |
5763 | * complete_all: - signals all threads waiting on this completion | |
5764 | * @x: holds the state of this particular completion | |
5765 | * | |
5766 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5767 | * |
5768 | * It may be assumed that this function implies a write memory barrier before | |
5769 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5770 | */ |
b15136e9 | 5771 | void complete_all(struct completion *x) |
1da177e4 LT |
5772 | { |
5773 | unsigned long flags; | |
5774 | ||
5775 | spin_lock_irqsave(&x->wait.lock, flags); | |
5776 | x->done += UINT_MAX/2; | |
d9514f6c | 5777 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5778 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5779 | } | |
5780 | EXPORT_SYMBOL(complete_all); | |
5781 | ||
8cbbe86d AK |
5782 | static inline long __sched |
5783 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5784 | { |
1da177e4 LT |
5785 | if (!x->done) { |
5786 | DECLARE_WAITQUEUE(wait, current); | |
5787 | ||
5788 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5789 | __add_wait_queue_tail(&x->wait, &wait); | |
5790 | do { | |
94d3d824 | 5791 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5792 | timeout = -ERESTARTSYS; |
5793 | break; | |
8cbbe86d AK |
5794 | } |
5795 | __set_current_state(state); | |
1da177e4 LT |
5796 | spin_unlock_irq(&x->wait.lock); |
5797 | timeout = schedule_timeout(timeout); | |
5798 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5799 | } while (!x->done && timeout); |
1da177e4 | 5800 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5801 | if (!x->done) |
5802 | return timeout; | |
1da177e4 LT |
5803 | } |
5804 | x->done--; | |
ea71a546 | 5805 | return timeout ?: 1; |
1da177e4 | 5806 | } |
1da177e4 | 5807 | |
8cbbe86d AK |
5808 | static long __sched |
5809 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5810 | { |
1da177e4 LT |
5811 | might_sleep(); |
5812 | ||
5813 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5814 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5815 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5816 | return timeout; |
5817 | } | |
1da177e4 | 5818 | |
65eb3dc6 KD |
5819 | /** |
5820 | * wait_for_completion: - waits for completion of a task | |
5821 | * @x: holds the state of this particular completion | |
5822 | * | |
5823 | * This waits to be signaled for completion of a specific task. It is NOT | |
5824 | * interruptible and there is no timeout. | |
5825 | * | |
5826 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5827 | * and interrupt capability. Also see complete(). | |
5828 | */ | |
b15136e9 | 5829 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5830 | { |
5831 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5832 | } |
8cbbe86d | 5833 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5834 | |
65eb3dc6 KD |
5835 | /** |
5836 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5837 | * @x: holds the state of this particular completion | |
5838 | * @timeout: timeout value in jiffies | |
5839 | * | |
5840 | * This waits for either a completion of a specific task to be signaled or for a | |
5841 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5842 | * interruptible. | |
5843 | */ | |
b15136e9 | 5844 | unsigned long __sched |
8cbbe86d | 5845 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5846 | { |
8cbbe86d | 5847 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5848 | } |
8cbbe86d | 5849 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5850 | |
65eb3dc6 KD |
5851 | /** |
5852 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5853 | * @x: holds the state of this particular completion | |
5854 | * | |
5855 | * This waits for completion of a specific task to be signaled. It is | |
5856 | * interruptible. | |
5857 | */ | |
8cbbe86d | 5858 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5859 | { |
51e97990 AK |
5860 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5861 | if (t == -ERESTARTSYS) | |
5862 | return t; | |
5863 | return 0; | |
0fec171c | 5864 | } |
8cbbe86d | 5865 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5866 | |
65eb3dc6 KD |
5867 | /** |
5868 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5869 | * @x: holds the state of this particular completion | |
5870 | * @timeout: timeout value in jiffies | |
5871 | * | |
5872 | * This waits for either a completion of a specific task to be signaled or for a | |
5873 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5874 | */ | |
b15136e9 | 5875 | unsigned long __sched |
8cbbe86d AK |
5876 | wait_for_completion_interruptible_timeout(struct completion *x, |
5877 | unsigned long timeout) | |
0fec171c | 5878 | { |
8cbbe86d | 5879 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5880 | } |
8cbbe86d | 5881 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5882 | |
65eb3dc6 KD |
5883 | /** |
5884 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5885 | * @x: holds the state of this particular completion | |
5886 | * | |
5887 | * This waits to be signaled for completion of a specific task. It can be | |
5888 | * interrupted by a kill signal. | |
5889 | */ | |
009e577e MW |
5890 | int __sched wait_for_completion_killable(struct completion *x) |
5891 | { | |
5892 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5893 | if (t == -ERESTARTSYS) | |
5894 | return t; | |
5895 | return 0; | |
5896 | } | |
5897 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5898 | ||
be4de352 DC |
5899 | /** |
5900 | * try_wait_for_completion - try to decrement a completion without blocking | |
5901 | * @x: completion structure | |
5902 | * | |
5903 | * Returns: 0 if a decrement cannot be done without blocking | |
5904 | * 1 if a decrement succeeded. | |
5905 | * | |
5906 | * If a completion is being used as a counting completion, | |
5907 | * attempt to decrement the counter without blocking. This | |
5908 | * enables us to avoid waiting if the resource the completion | |
5909 | * is protecting is not available. | |
5910 | */ | |
5911 | bool try_wait_for_completion(struct completion *x) | |
5912 | { | |
7539a3b3 | 5913 | unsigned long flags; |
be4de352 DC |
5914 | int ret = 1; |
5915 | ||
7539a3b3 | 5916 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
5917 | if (!x->done) |
5918 | ret = 0; | |
5919 | else | |
5920 | x->done--; | |
7539a3b3 | 5921 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
5922 | return ret; |
5923 | } | |
5924 | EXPORT_SYMBOL(try_wait_for_completion); | |
5925 | ||
5926 | /** | |
5927 | * completion_done - Test to see if a completion has any waiters | |
5928 | * @x: completion structure | |
5929 | * | |
5930 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5931 | * 1 if there are no waiters. | |
5932 | * | |
5933 | */ | |
5934 | bool completion_done(struct completion *x) | |
5935 | { | |
7539a3b3 | 5936 | unsigned long flags; |
be4de352 DC |
5937 | int ret = 1; |
5938 | ||
7539a3b3 | 5939 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
5940 | if (!x->done) |
5941 | ret = 0; | |
7539a3b3 | 5942 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
5943 | return ret; |
5944 | } | |
5945 | EXPORT_SYMBOL(completion_done); | |
5946 | ||
8cbbe86d AK |
5947 | static long __sched |
5948 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5949 | { |
0fec171c IM |
5950 | unsigned long flags; |
5951 | wait_queue_t wait; | |
5952 | ||
5953 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5954 | |
8cbbe86d | 5955 | __set_current_state(state); |
1da177e4 | 5956 | |
8cbbe86d AK |
5957 | spin_lock_irqsave(&q->lock, flags); |
5958 | __add_wait_queue(q, &wait); | |
5959 | spin_unlock(&q->lock); | |
5960 | timeout = schedule_timeout(timeout); | |
5961 | spin_lock_irq(&q->lock); | |
5962 | __remove_wait_queue(q, &wait); | |
5963 | spin_unlock_irqrestore(&q->lock, flags); | |
5964 | ||
5965 | return timeout; | |
5966 | } | |
5967 | ||
5968 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5969 | { | |
5970 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5971 | } |
1da177e4 LT |
5972 | EXPORT_SYMBOL(interruptible_sleep_on); |
5973 | ||
0fec171c | 5974 | long __sched |
95cdf3b7 | 5975 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5976 | { |
8cbbe86d | 5977 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5978 | } |
1da177e4 LT |
5979 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5980 | ||
0fec171c | 5981 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5982 | { |
8cbbe86d | 5983 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5984 | } |
1da177e4 LT |
5985 | EXPORT_SYMBOL(sleep_on); |
5986 | ||
0fec171c | 5987 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5988 | { |
8cbbe86d | 5989 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5990 | } |
1da177e4 LT |
5991 | EXPORT_SYMBOL(sleep_on_timeout); |
5992 | ||
b29739f9 IM |
5993 | #ifdef CONFIG_RT_MUTEXES |
5994 | ||
5995 | /* | |
5996 | * rt_mutex_setprio - set the current priority of a task | |
5997 | * @p: task | |
5998 | * @prio: prio value (kernel-internal form) | |
5999 | * | |
6000 | * This function changes the 'effective' priority of a task. It does | |
6001 | * not touch ->normal_prio like __setscheduler(). | |
6002 | * | |
6003 | * Used by the rt_mutex code to implement priority inheritance logic. | |
6004 | */ | |
36c8b586 | 6005 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
6006 | { |
6007 | unsigned long flags; | |
83b699ed | 6008 | int oldprio, on_rq, running; |
70b97a7f | 6009 | struct rq *rq; |
cb469845 | 6010 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
6011 | |
6012 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
6013 | ||
6014 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6015 | update_rq_clock(rq); |
b29739f9 | 6016 | |
d5f9f942 | 6017 | oldprio = p->prio; |
dd41f596 | 6018 | on_rq = p->se.on_rq; |
051a1d1a | 6019 | running = task_current(rq, p); |
0e1f3483 | 6020 | if (on_rq) |
69be72c1 | 6021 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
6022 | if (running) |
6023 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
6024 | |
6025 | if (rt_prio(prio)) | |
6026 | p->sched_class = &rt_sched_class; | |
6027 | else | |
6028 | p->sched_class = &fair_sched_class; | |
6029 | ||
b29739f9 IM |
6030 | p->prio = prio; |
6031 | ||
0e1f3483 HS |
6032 | if (running) |
6033 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 6034 | if (on_rq) { |
8159f87e | 6035 | enqueue_task(rq, p, 0); |
cb469845 SR |
6036 | |
6037 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
6038 | } |
6039 | task_rq_unlock(rq, &flags); | |
6040 | } | |
6041 | ||
6042 | #endif | |
6043 | ||
36c8b586 | 6044 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6045 | { |
dd41f596 | 6046 | int old_prio, delta, on_rq; |
1da177e4 | 6047 | unsigned long flags; |
70b97a7f | 6048 | struct rq *rq; |
1da177e4 LT |
6049 | |
6050 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6051 | return; | |
6052 | /* | |
6053 | * We have to be careful, if called from sys_setpriority(), | |
6054 | * the task might be in the middle of scheduling on another CPU. | |
6055 | */ | |
6056 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6057 | update_rq_clock(rq); |
1da177e4 LT |
6058 | /* |
6059 | * The RT priorities are set via sched_setscheduler(), but we still | |
6060 | * allow the 'normal' nice value to be set - but as expected | |
6061 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6062 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6063 | */ |
e05606d3 | 6064 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6065 | p->static_prio = NICE_TO_PRIO(nice); |
6066 | goto out_unlock; | |
6067 | } | |
dd41f596 | 6068 | on_rq = p->se.on_rq; |
c09595f6 | 6069 | if (on_rq) |
69be72c1 | 6070 | dequeue_task(rq, p, 0); |
1da177e4 | 6071 | |
1da177e4 | 6072 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6073 | set_load_weight(p); |
b29739f9 IM |
6074 | old_prio = p->prio; |
6075 | p->prio = effective_prio(p); | |
6076 | delta = p->prio - old_prio; | |
1da177e4 | 6077 | |
dd41f596 | 6078 | if (on_rq) { |
8159f87e | 6079 | enqueue_task(rq, p, 0); |
1da177e4 | 6080 | /* |
d5f9f942 AM |
6081 | * If the task increased its priority or is running and |
6082 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6083 | */ |
d5f9f942 | 6084 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6085 | resched_task(rq->curr); |
6086 | } | |
6087 | out_unlock: | |
6088 | task_rq_unlock(rq, &flags); | |
6089 | } | |
1da177e4 LT |
6090 | EXPORT_SYMBOL(set_user_nice); |
6091 | ||
e43379f1 MM |
6092 | /* |
6093 | * can_nice - check if a task can reduce its nice value | |
6094 | * @p: task | |
6095 | * @nice: nice value | |
6096 | */ | |
36c8b586 | 6097 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6098 | { |
024f4747 MM |
6099 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6100 | int nice_rlim = 20 - nice; | |
48f24c4d | 6101 | |
e43379f1 MM |
6102 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6103 | capable(CAP_SYS_NICE)); | |
6104 | } | |
6105 | ||
1da177e4 LT |
6106 | #ifdef __ARCH_WANT_SYS_NICE |
6107 | ||
6108 | /* | |
6109 | * sys_nice - change the priority of the current process. | |
6110 | * @increment: priority increment | |
6111 | * | |
6112 | * sys_setpriority is a more generic, but much slower function that | |
6113 | * does similar things. | |
6114 | */ | |
5add95d4 | 6115 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6116 | { |
48f24c4d | 6117 | long nice, retval; |
1da177e4 LT |
6118 | |
6119 | /* | |
6120 | * Setpriority might change our priority at the same moment. | |
6121 | * We don't have to worry. Conceptually one call occurs first | |
6122 | * and we have a single winner. | |
6123 | */ | |
e43379f1 MM |
6124 | if (increment < -40) |
6125 | increment = -40; | |
1da177e4 LT |
6126 | if (increment > 40) |
6127 | increment = 40; | |
6128 | ||
2b8f836f | 6129 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6130 | if (nice < -20) |
6131 | nice = -20; | |
6132 | if (nice > 19) | |
6133 | nice = 19; | |
6134 | ||
e43379f1 MM |
6135 | if (increment < 0 && !can_nice(current, nice)) |
6136 | return -EPERM; | |
6137 | ||
1da177e4 LT |
6138 | retval = security_task_setnice(current, nice); |
6139 | if (retval) | |
6140 | return retval; | |
6141 | ||
6142 | set_user_nice(current, nice); | |
6143 | return 0; | |
6144 | } | |
6145 | ||
6146 | #endif | |
6147 | ||
6148 | /** | |
6149 | * task_prio - return the priority value of a given task. | |
6150 | * @p: the task in question. | |
6151 | * | |
6152 | * This is the priority value as seen by users in /proc. | |
6153 | * RT tasks are offset by -200. Normal tasks are centered | |
6154 | * around 0, value goes from -16 to +15. | |
6155 | */ | |
36c8b586 | 6156 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6157 | { |
6158 | return p->prio - MAX_RT_PRIO; | |
6159 | } | |
6160 | ||
6161 | /** | |
6162 | * task_nice - return the nice value of a given task. | |
6163 | * @p: the task in question. | |
6164 | */ | |
36c8b586 | 6165 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6166 | { |
6167 | return TASK_NICE(p); | |
6168 | } | |
150d8bed | 6169 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6170 | |
6171 | /** | |
6172 | * idle_cpu - is a given cpu idle currently? | |
6173 | * @cpu: the processor in question. | |
6174 | */ | |
6175 | int idle_cpu(int cpu) | |
6176 | { | |
6177 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6178 | } | |
6179 | ||
1da177e4 LT |
6180 | /** |
6181 | * idle_task - return the idle task for a given cpu. | |
6182 | * @cpu: the processor in question. | |
6183 | */ | |
36c8b586 | 6184 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6185 | { |
6186 | return cpu_rq(cpu)->idle; | |
6187 | } | |
6188 | ||
6189 | /** | |
6190 | * find_process_by_pid - find a process with a matching PID value. | |
6191 | * @pid: the pid in question. | |
6192 | */ | |
a9957449 | 6193 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6194 | { |
228ebcbe | 6195 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6196 | } |
6197 | ||
6198 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6199 | static void |
6200 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6201 | { |
dd41f596 | 6202 | BUG_ON(p->se.on_rq); |
48f24c4d | 6203 | |
1da177e4 LT |
6204 | p->policy = policy; |
6205 | p->rt_priority = prio; | |
b29739f9 IM |
6206 | p->normal_prio = normal_prio(p); |
6207 | /* we are holding p->pi_lock already */ | |
6208 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
6209 | if (rt_prio(p->prio)) |
6210 | p->sched_class = &rt_sched_class; | |
6211 | else | |
6212 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 6213 | set_load_weight(p); |
1da177e4 LT |
6214 | } |
6215 | ||
c69e8d9c DH |
6216 | /* |
6217 | * check the target process has a UID that matches the current process's | |
6218 | */ | |
6219 | static bool check_same_owner(struct task_struct *p) | |
6220 | { | |
6221 | const struct cred *cred = current_cred(), *pcred; | |
6222 | bool match; | |
6223 | ||
6224 | rcu_read_lock(); | |
6225 | pcred = __task_cred(p); | |
6226 | match = (cred->euid == pcred->euid || | |
6227 | cred->euid == pcred->uid); | |
6228 | rcu_read_unlock(); | |
6229 | return match; | |
6230 | } | |
6231 | ||
961ccddd RR |
6232 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6233 | struct sched_param *param, bool user) | |
1da177e4 | 6234 | { |
83b699ed | 6235 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6236 | unsigned long flags; |
cb469845 | 6237 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6238 | struct rq *rq; |
ca94c442 | 6239 | int reset_on_fork; |
1da177e4 | 6240 | |
66e5393a SR |
6241 | /* may grab non-irq protected spin_locks */ |
6242 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6243 | recheck: |
6244 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6245 | if (policy < 0) { |
6246 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6247 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6248 | } else { |
6249 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6250 | policy &= ~SCHED_RESET_ON_FORK; | |
6251 | ||
6252 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6253 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6254 | policy != SCHED_IDLE) | |
6255 | return -EINVAL; | |
6256 | } | |
6257 | ||
1da177e4 LT |
6258 | /* |
6259 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6260 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6261 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6262 | */ |
6263 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6264 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6265 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6266 | return -EINVAL; |
e05606d3 | 6267 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6268 | return -EINVAL; |
6269 | ||
37e4ab3f OC |
6270 | /* |
6271 | * Allow unprivileged RT tasks to decrease priority: | |
6272 | */ | |
961ccddd | 6273 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6274 | if (rt_policy(policy)) { |
8dc3e909 | 6275 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6276 | |
6277 | if (!lock_task_sighand(p, &flags)) | |
6278 | return -ESRCH; | |
6279 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6280 | unlock_task_sighand(p, &flags); | |
6281 | ||
6282 | /* can't set/change the rt policy */ | |
6283 | if (policy != p->policy && !rlim_rtprio) | |
6284 | return -EPERM; | |
6285 | ||
6286 | /* can't increase priority */ | |
6287 | if (param->sched_priority > p->rt_priority && | |
6288 | param->sched_priority > rlim_rtprio) | |
6289 | return -EPERM; | |
6290 | } | |
dd41f596 IM |
6291 | /* |
6292 | * Like positive nice levels, dont allow tasks to | |
6293 | * move out of SCHED_IDLE either: | |
6294 | */ | |
6295 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6296 | return -EPERM; | |
5fe1d75f | 6297 | |
37e4ab3f | 6298 | /* can't change other user's priorities */ |
c69e8d9c | 6299 | if (!check_same_owner(p)) |
37e4ab3f | 6300 | return -EPERM; |
ca94c442 LP |
6301 | |
6302 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6303 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6304 | return -EPERM; | |
37e4ab3f | 6305 | } |
1da177e4 | 6306 | |
725aad24 | 6307 | if (user) { |
b68aa230 | 6308 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6309 | /* |
6310 | * Do not allow realtime tasks into groups that have no runtime | |
6311 | * assigned. | |
6312 | */ | |
9a7e0b18 PZ |
6313 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6314 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6315 | return -EPERM; |
b68aa230 PZ |
6316 | #endif |
6317 | ||
725aad24 JF |
6318 | retval = security_task_setscheduler(p, policy, param); |
6319 | if (retval) | |
6320 | return retval; | |
6321 | } | |
6322 | ||
b29739f9 IM |
6323 | /* |
6324 | * make sure no PI-waiters arrive (or leave) while we are | |
6325 | * changing the priority of the task: | |
6326 | */ | |
6327 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6328 | /* |
6329 | * To be able to change p->policy safely, the apropriate | |
6330 | * runqueue lock must be held. | |
6331 | */ | |
b29739f9 | 6332 | rq = __task_rq_lock(p); |
1da177e4 LT |
6333 | /* recheck policy now with rq lock held */ |
6334 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6335 | policy = oldpolicy = -1; | |
b29739f9 IM |
6336 | __task_rq_unlock(rq); |
6337 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6338 | goto recheck; |
6339 | } | |
2daa3577 | 6340 | update_rq_clock(rq); |
dd41f596 | 6341 | on_rq = p->se.on_rq; |
051a1d1a | 6342 | running = task_current(rq, p); |
0e1f3483 | 6343 | if (on_rq) |
2e1cb74a | 6344 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6345 | if (running) |
6346 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6347 | |
ca94c442 LP |
6348 | p->sched_reset_on_fork = reset_on_fork; |
6349 | ||
1da177e4 | 6350 | oldprio = p->prio; |
dd41f596 | 6351 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6352 | |
0e1f3483 HS |
6353 | if (running) |
6354 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6355 | if (on_rq) { |
6356 | activate_task(rq, p, 0); | |
cb469845 SR |
6357 | |
6358 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6359 | } |
b29739f9 IM |
6360 | __task_rq_unlock(rq); |
6361 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6362 | ||
95e02ca9 TG |
6363 | rt_mutex_adjust_pi(p); |
6364 | ||
1da177e4 LT |
6365 | return 0; |
6366 | } | |
961ccddd RR |
6367 | |
6368 | /** | |
6369 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6370 | * @p: the task in question. | |
6371 | * @policy: new policy. | |
6372 | * @param: structure containing the new RT priority. | |
6373 | * | |
6374 | * NOTE that the task may be already dead. | |
6375 | */ | |
6376 | int sched_setscheduler(struct task_struct *p, int policy, | |
6377 | struct sched_param *param) | |
6378 | { | |
6379 | return __sched_setscheduler(p, policy, param, true); | |
6380 | } | |
1da177e4 LT |
6381 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6382 | ||
961ccddd RR |
6383 | /** |
6384 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6385 | * @p: the task in question. | |
6386 | * @policy: new policy. | |
6387 | * @param: structure containing the new RT priority. | |
6388 | * | |
6389 | * Just like sched_setscheduler, only don't bother checking if the | |
6390 | * current context has permission. For example, this is needed in | |
6391 | * stop_machine(): we create temporary high priority worker threads, | |
6392 | * but our caller might not have that capability. | |
6393 | */ | |
6394 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6395 | struct sched_param *param) | |
6396 | { | |
6397 | return __sched_setscheduler(p, policy, param, false); | |
6398 | } | |
6399 | ||
95cdf3b7 IM |
6400 | static int |
6401 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6402 | { |
1da177e4 LT |
6403 | struct sched_param lparam; |
6404 | struct task_struct *p; | |
36c8b586 | 6405 | int retval; |
1da177e4 LT |
6406 | |
6407 | if (!param || pid < 0) | |
6408 | return -EINVAL; | |
6409 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6410 | return -EFAULT; | |
5fe1d75f ON |
6411 | |
6412 | rcu_read_lock(); | |
6413 | retval = -ESRCH; | |
1da177e4 | 6414 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6415 | if (p != NULL) |
6416 | retval = sched_setscheduler(p, policy, &lparam); | |
6417 | rcu_read_unlock(); | |
36c8b586 | 6418 | |
1da177e4 LT |
6419 | return retval; |
6420 | } | |
6421 | ||
6422 | /** | |
6423 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6424 | * @pid: the pid in question. | |
6425 | * @policy: new policy. | |
6426 | * @param: structure containing the new RT priority. | |
6427 | */ | |
5add95d4 HC |
6428 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6429 | struct sched_param __user *, param) | |
1da177e4 | 6430 | { |
c21761f1 JB |
6431 | /* negative values for policy are not valid */ |
6432 | if (policy < 0) | |
6433 | return -EINVAL; | |
6434 | ||
1da177e4 LT |
6435 | return do_sched_setscheduler(pid, policy, param); |
6436 | } | |
6437 | ||
6438 | /** | |
6439 | * sys_sched_setparam - set/change the RT priority of a thread | |
6440 | * @pid: the pid in question. | |
6441 | * @param: structure containing the new RT priority. | |
6442 | */ | |
5add95d4 | 6443 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6444 | { |
6445 | return do_sched_setscheduler(pid, -1, param); | |
6446 | } | |
6447 | ||
6448 | /** | |
6449 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6450 | * @pid: the pid in question. | |
6451 | */ | |
5add95d4 | 6452 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6453 | { |
36c8b586 | 6454 | struct task_struct *p; |
3a5c359a | 6455 | int retval; |
1da177e4 LT |
6456 | |
6457 | if (pid < 0) | |
3a5c359a | 6458 | return -EINVAL; |
1da177e4 LT |
6459 | |
6460 | retval = -ESRCH; | |
6461 | read_lock(&tasklist_lock); | |
6462 | p = find_process_by_pid(pid); | |
6463 | if (p) { | |
6464 | retval = security_task_getscheduler(p); | |
6465 | if (!retval) | |
ca94c442 LP |
6466 | retval = p->policy |
6467 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6468 | } |
6469 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6470 | return retval; |
6471 | } | |
6472 | ||
6473 | /** | |
ca94c442 | 6474 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6475 | * @pid: the pid in question. |
6476 | * @param: structure containing the RT priority. | |
6477 | */ | |
5add95d4 | 6478 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6479 | { |
6480 | struct sched_param lp; | |
36c8b586 | 6481 | struct task_struct *p; |
3a5c359a | 6482 | int retval; |
1da177e4 LT |
6483 | |
6484 | if (!param || pid < 0) | |
3a5c359a | 6485 | return -EINVAL; |
1da177e4 LT |
6486 | |
6487 | read_lock(&tasklist_lock); | |
6488 | p = find_process_by_pid(pid); | |
6489 | retval = -ESRCH; | |
6490 | if (!p) | |
6491 | goto out_unlock; | |
6492 | ||
6493 | retval = security_task_getscheduler(p); | |
6494 | if (retval) | |
6495 | goto out_unlock; | |
6496 | ||
6497 | lp.sched_priority = p->rt_priority; | |
6498 | read_unlock(&tasklist_lock); | |
6499 | ||
6500 | /* | |
6501 | * This one might sleep, we cannot do it with a spinlock held ... | |
6502 | */ | |
6503 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6504 | ||
1da177e4 LT |
6505 | return retval; |
6506 | ||
6507 | out_unlock: | |
6508 | read_unlock(&tasklist_lock); | |
6509 | return retval; | |
6510 | } | |
6511 | ||
96f874e2 | 6512 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6513 | { |
5a16f3d3 | 6514 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6515 | struct task_struct *p; |
6516 | int retval; | |
1da177e4 | 6517 | |
95402b38 | 6518 | get_online_cpus(); |
1da177e4 LT |
6519 | read_lock(&tasklist_lock); |
6520 | ||
6521 | p = find_process_by_pid(pid); | |
6522 | if (!p) { | |
6523 | read_unlock(&tasklist_lock); | |
95402b38 | 6524 | put_online_cpus(); |
1da177e4 LT |
6525 | return -ESRCH; |
6526 | } | |
6527 | ||
6528 | /* | |
6529 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6530 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6531 | * usage count and then drop tasklist_lock. |
6532 | */ | |
6533 | get_task_struct(p); | |
6534 | read_unlock(&tasklist_lock); | |
6535 | ||
5a16f3d3 RR |
6536 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6537 | retval = -ENOMEM; | |
6538 | goto out_put_task; | |
6539 | } | |
6540 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6541 | retval = -ENOMEM; | |
6542 | goto out_free_cpus_allowed; | |
6543 | } | |
1da177e4 | 6544 | retval = -EPERM; |
c69e8d9c | 6545 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6546 | goto out_unlock; |
6547 | ||
e7834f8f DQ |
6548 | retval = security_task_setscheduler(p, 0, NULL); |
6549 | if (retval) | |
6550 | goto out_unlock; | |
6551 | ||
5a16f3d3 RR |
6552 | cpuset_cpus_allowed(p, cpus_allowed); |
6553 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6554 | again: |
5a16f3d3 | 6555 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6556 | |
8707d8b8 | 6557 | if (!retval) { |
5a16f3d3 RR |
6558 | cpuset_cpus_allowed(p, cpus_allowed); |
6559 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6560 | /* |
6561 | * We must have raced with a concurrent cpuset | |
6562 | * update. Just reset the cpus_allowed to the | |
6563 | * cpuset's cpus_allowed | |
6564 | */ | |
5a16f3d3 | 6565 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6566 | goto again; |
6567 | } | |
6568 | } | |
1da177e4 | 6569 | out_unlock: |
5a16f3d3 RR |
6570 | free_cpumask_var(new_mask); |
6571 | out_free_cpus_allowed: | |
6572 | free_cpumask_var(cpus_allowed); | |
6573 | out_put_task: | |
1da177e4 | 6574 | put_task_struct(p); |
95402b38 | 6575 | put_online_cpus(); |
1da177e4 LT |
6576 | return retval; |
6577 | } | |
6578 | ||
6579 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6580 | struct cpumask *new_mask) |
1da177e4 | 6581 | { |
96f874e2 RR |
6582 | if (len < cpumask_size()) |
6583 | cpumask_clear(new_mask); | |
6584 | else if (len > cpumask_size()) | |
6585 | len = cpumask_size(); | |
6586 | ||
1da177e4 LT |
6587 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6588 | } | |
6589 | ||
6590 | /** | |
6591 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6592 | * @pid: pid of the process | |
6593 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6594 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6595 | */ | |
5add95d4 HC |
6596 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6597 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6598 | { |
5a16f3d3 | 6599 | cpumask_var_t new_mask; |
1da177e4 LT |
6600 | int retval; |
6601 | ||
5a16f3d3 RR |
6602 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6603 | return -ENOMEM; | |
1da177e4 | 6604 | |
5a16f3d3 RR |
6605 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6606 | if (retval == 0) | |
6607 | retval = sched_setaffinity(pid, new_mask); | |
6608 | free_cpumask_var(new_mask); | |
6609 | return retval; | |
1da177e4 LT |
6610 | } |
6611 | ||
96f874e2 | 6612 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6613 | { |
36c8b586 | 6614 | struct task_struct *p; |
31605683 TG |
6615 | unsigned long flags; |
6616 | struct rq *rq; | |
1da177e4 | 6617 | int retval; |
1da177e4 | 6618 | |
95402b38 | 6619 | get_online_cpus(); |
1da177e4 LT |
6620 | read_lock(&tasklist_lock); |
6621 | ||
6622 | retval = -ESRCH; | |
6623 | p = find_process_by_pid(pid); | |
6624 | if (!p) | |
6625 | goto out_unlock; | |
6626 | ||
e7834f8f DQ |
6627 | retval = security_task_getscheduler(p); |
6628 | if (retval) | |
6629 | goto out_unlock; | |
6630 | ||
31605683 | 6631 | rq = task_rq_lock(p, &flags); |
96f874e2 | 6632 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 6633 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
6634 | |
6635 | out_unlock: | |
6636 | read_unlock(&tasklist_lock); | |
95402b38 | 6637 | put_online_cpus(); |
1da177e4 | 6638 | |
9531b62f | 6639 | return retval; |
1da177e4 LT |
6640 | } |
6641 | ||
6642 | /** | |
6643 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6644 | * @pid: pid of the process | |
6645 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6646 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6647 | */ | |
5add95d4 HC |
6648 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6649 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6650 | { |
6651 | int ret; | |
f17c8607 | 6652 | cpumask_var_t mask; |
1da177e4 | 6653 | |
f17c8607 | 6654 | if (len < cpumask_size()) |
1da177e4 LT |
6655 | return -EINVAL; |
6656 | ||
f17c8607 RR |
6657 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6658 | return -ENOMEM; | |
1da177e4 | 6659 | |
f17c8607 RR |
6660 | ret = sched_getaffinity(pid, mask); |
6661 | if (ret == 0) { | |
6662 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6663 | ret = -EFAULT; | |
6664 | else | |
6665 | ret = cpumask_size(); | |
6666 | } | |
6667 | free_cpumask_var(mask); | |
1da177e4 | 6668 | |
f17c8607 | 6669 | return ret; |
1da177e4 LT |
6670 | } |
6671 | ||
6672 | /** | |
6673 | * sys_sched_yield - yield the current processor to other threads. | |
6674 | * | |
dd41f596 IM |
6675 | * This function yields the current CPU to other tasks. If there are no |
6676 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6677 | */ |
5add95d4 | 6678 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6679 | { |
70b97a7f | 6680 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6681 | |
2d72376b | 6682 | schedstat_inc(rq, yld_count); |
4530d7ab | 6683 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6684 | |
6685 | /* | |
6686 | * Since we are going to call schedule() anyway, there's | |
6687 | * no need to preempt or enable interrupts: | |
6688 | */ | |
6689 | __release(rq->lock); | |
8a25d5de | 6690 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6691 | _raw_spin_unlock(&rq->lock); |
6692 | preempt_enable_no_resched(); | |
6693 | ||
6694 | schedule(); | |
6695 | ||
6696 | return 0; | |
6697 | } | |
6698 | ||
d86ee480 PZ |
6699 | static inline int should_resched(void) |
6700 | { | |
6701 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6702 | } | |
6703 | ||
e7b38404 | 6704 | static void __cond_resched(void) |
1da177e4 | 6705 | { |
e7aaaa69 FW |
6706 | add_preempt_count(PREEMPT_ACTIVE); |
6707 | schedule(); | |
6708 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6709 | } |
6710 | ||
02b67cc3 | 6711 | int __sched _cond_resched(void) |
1da177e4 | 6712 | { |
d86ee480 | 6713 | if (should_resched()) { |
1da177e4 LT |
6714 | __cond_resched(); |
6715 | return 1; | |
6716 | } | |
6717 | return 0; | |
6718 | } | |
02b67cc3 | 6719 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6720 | |
6721 | /* | |
613afbf8 | 6722 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6723 | * call schedule, and on return reacquire the lock. |
6724 | * | |
41a2d6cf | 6725 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6726 | * operations here to prevent schedule() from being called twice (once via |
6727 | * spin_unlock(), once by hand). | |
6728 | */ | |
613afbf8 | 6729 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6730 | { |
d86ee480 | 6731 | int resched = should_resched(); |
6df3cecb JK |
6732 | int ret = 0; |
6733 | ||
f607c668 PZ |
6734 | lockdep_assert_held(lock); |
6735 | ||
95c354fe | 6736 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6737 | spin_unlock(lock); |
d86ee480 | 6738 | if (resched) |
95c354fe NP |
6739 | __cond_resched(); |
6740 | else | |
6741 | cpu_relax(); | |
6df3cecb | 6742 | ret = 1; |
1da177e4 | 6743 | spin_lock(lock); |
1da177e4 | 6744 | } |
6df3cecb | 6745 | return ret; |
1da177e4 | 6746 | } |
613afbf8 | 6747 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6748 | |
613afbf8 | 6749 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6750 | { |
6751 | BUG_ON(!in_softirq()); | |
6752 | ||
d86ee480 | 6753 | if (should_resched()) { |
98d82567 | 6754 | local_bh_enable(); |
1da177e4 LT |
6755 | __cond_resched(); |
6756 | local_bh_disable(); | |
6757 | return 1; | |
6758 | } | |
6759 | return 0; | |
6760 | } | |
613afbf8 | 6761 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6762 | |
1da177e4 LT |
6763 | /** |
6764 | * yield - yield the current processor to other threads. | |
6765 | * | |
72fd4a35 | 6766 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6767 | * thread runnable and calls sys_sched_yield(). |
6768 | */ | |
6769 | void __sched yield(void) | |
6770 | { | |
6771 | set_current_state(TASK_RUNNING); | |
6772 | sys_sched_yield(); | |
6773 | } | |
1da177e4 LT |
6774 | EXPORT_SYMBOL(yield); |
6775 | ||
6776 | /* | |
41a2d6cf | 6777 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6778 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
6779 | */ |
6780 | void __sched io_schedule(void) | |
6781 | { | |
54d35f29 | 6782 | struct rq *rq = raw_rq(); |
1da177e4 | 6783 | |
0ff92245 | 6784 | delayacct_blkio_start(); |
1da177e4 | 6785 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6786 | current->in_iowait = 1; |
1da177e4 | 6787 | schedule(); |
8f0dfc34 | 6788 | current->in_iowait = 0; |
1da177e4 | 6789 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6790 | delayacct_blkio_end(); |
1da177e4 | 6791 | } |
1da177e4 LT |
6792 | EXPORT_SYMBOL(io_schedule); |
6793 | ||
6794 | long __sched io_schedule_timeout(long timeout) | |
6795 | { | |
54d35f29 | 6796 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6797 | long ret; |
6798 | ||
0ff92245 | 6799 | delayacct_blkio_start(); |
1da177e4 | 6800 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6801 | current->in_iowait = 1; |
1da177e4 | 6802 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6803 | current->in_iowait = 0; |
1da177e4 | 6804 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6805 | delayacct_blkio_end(); |
1da177e4 LT |
6806 | return ret; |
6807 | } | |
6808 | ||
6809 | /** | |
6810 | * sys_sched_get_priority_max - return maximum RT priority. | |
6811 | * @policy: scheduling class. | |
6812 | * | |
6813 | * this syscall returns the maximum rt_priority that can be used | |
6814 | * by a given scheduling class. | |
6815 | */ | |
5add95d4 | 6816 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6817 | { |
6818 | int ret = -EINVAL; | |
6819 | ||
6820 | switch (policy) { | |
6821 | case SCHED_FIFO: | |
6822 | case SCHED_RR: | |
6823 | ret = MAX_USER_RT_PRIO-1; | |
6824 | break; | |
6825 | case SCHED_NORMAL: | |
b0a9499c | 6826 | case SCHED_BATCH: |
dd41f596 | 6827 | case SCHED_IDLE: |
1da177e4 LT |
6828 | ret = 0; |
6829 | break; | |
6830 | } | |
6831 | return ret; | |
6832 | } | |
6833 | ||
6834 | /** | |
6835 | * sys_sched_get_priority_min - return minimum RT priority. | |
6836 | * @policy: scheduling class. | |
6837 | * | |
6838 | * this syscall returns the minimum rt_priority that can be used | |
6839 | * by a given scheduling class. | |
6840 | */ | |
5add95d4 | 6841 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6842 | { |
6843 | int ret = -EINVAL; | |
6844 | ||
6845 | switch (policy) { | |
6846 | case SCHED_FIFO: | |
6847 | case SCHED_RR: | |
6848 | ret = 1; | |
6849 | break; | |
6850 | case SCHED_NORMAL: | |
b0a9499c | 6851 | case SCHED_BATCH: |
dd41f596 | 6852 | case SCHED_IDLE: |
1da177e4 LT |
6853 | ret = 0; |
6854 | } | |
6855 | return ret; | |
6856 | } | |
6857 | ||
6858 | /** | |
6859 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6860 | * @pid: pid of the process. | |
6861 | * @interval: userspace pointer to the timeslice value. | |
6862 | * | |
6863 | * this syscall writes the default timeslice value of a given process | |
6864 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6865 | */ | |
17da2bd9 | 6866 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6867 | struct timespec __user *, interval) |
1da177e4 | 6868 | { |
36c8b586 | 6869 | struct task_struct *p; |
a4ec24b4 | 6870 | unsigned int time_slice; |
dba091b9 TG |
6871 | unsigned long flags; |
6872 | struct rq *rq; | |
3a5c359a | 6873 | int retval; |
1da177e4 | 6874 | struct timespec t; |
1da177e4 LT |
6875 | |
6876 | if (pid < 0) | |
3a5c359a | 6877 | return -EINVAL; |
1da177e4 LT |
6878 | |
6879 | retval = -ESRCH; | |
6880 | read_lock(&tasklist_lock); | |
6881 | p = find_process_by_pid(pid); | |
6882 | if (!p) | |
6883 | goto out_unlock; | |
6884 | ||
6885 | retval = security_task_getscheduler(p); | |
6886 | if (retval) | |
6887 | goto out_unlock; | |
6888 | ||
dba091b9 TG |
6889 | rq = task_rq_lock(p, &flags); |
6890 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
6891 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 6892 | |
1da177e4 | 6893 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6894 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6895 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6896 | return retval; |
3a5c359a | 6897 | |
1da177e4 LT |
6898 | out_unlock: |
6899 | read_unlock(&tasklist_lock); | |
6900 | return retval; | |
6901 | } | |
6902 | ||
7c731e0a | 6903 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6904 | |
82a1fcb9 | 6905 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6906 | { |
1da177e4 | 6907 | unsigned long free = 0; |
36c8b586 | 6908 | unsigned state; |
1da177e4 | 6909 | |
1da177e4 | 6910 | state = p->state ? __ffs(p->state) + 1 : 0; |
663997d4 | 6911 | pr_info("%-13.13s %c", p->comm, |
2ed6e34f | 6912 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6913 | #if BITS_PER_LONG == 32 |
1da177e4 | 6914 | if (state == TASK_RUNNING) |
663997d4 | 6915 | pr_cont(" running "); |
1da177e4 | 6916 | else |
663997d4 | 6917 | pr_cont(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6918 | #else |
6919 | if (state == TASK_RUNNING) | |
663997d4 | 6920 | pr_cont(" running task "); |
1da177e4 | 6921 | else |
663997d4 | 6922 | pr_cont(" %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6923 | #endif |
6924 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6925 | free = stack_not_used(p); |
1da177e4 | 6926 | #endif |
663997d4 | 6927 | pr_cont("%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
6928 | task_pid_nr(p), task_pid_nr(p->real_parent), |
6929 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6930 | |
5fb5e6de | 6931 | show_stack(p, NULL); |
1da177e4 LT |
6932 | } |
6933 | ||
e59e2ae2 | 6934 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6935 | { |
36c8b586 | 6936 | struct task_struct *g, *p; |
1da177e4 | 6937 | |
4bd77321 | 6938 | #if BITS_PER_LONG == 32 |
663997d4 | 6939 | pr_info(" task PC stack pid father\n"); |
1da177e4 | 6940 | #else |
663997d4 | 6941 | pr_info(" task PC stack pid father\n"); |
1da177e4 LT |
6942 | #endif |
6943 | read_lock(&tasklist_lock); | |
6944 | do_each_thread(g, p) { | |
6945 | /* | |
6946 | * reset the NMI-timeout, listing all files on a slow | |
6947 | * console might take alot of time: | |
6948 | */ | |
6949 | touch_nmi_watchdog(); | |
39bc89fd | 6950 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6951 | sched_show_task(p); |
1da177e4 LT |
6952 | } while_each_thread(g, p); |
6953 | ||
04c9167f JF |
6954 | touch_all_softlockup_watchdogs(); |
6955 | ||
dd41f596 IM |
6956 | #ifdef CONFIG_SCHED_DEBUG |
6957 | sysrq_sched_debug_show(); | |
6958 | #endif | |
1da177e4 | 6959 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6960 | /* |
6961 | * Only show locks if all tasks are dumped: | |
6962 | */ | |
93335a21 | 6963 | if (!state_filter) |
e59e2ae2 | 6964 | debug_show_all_locks(); |
1da177e4 LT |
6965 | } |
6966 | ||
1df21055 IM |
6967 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6968 | { | |
dd41f596 | 6969 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6970 | } |
6971 | ||
f340c0d1 IM |
6972 | /** |
6973 | * init_idle - set up an idle thread for a given CPU | |
6974 | * @idle: task in question | |
6975 | * @cpu: cpu the idle task belongs to | |
6976 | * | |
6977 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6978 | * flag, to make booting more robust. | |
6979 | */ | |
5c1e1767 | 6980 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6981 | { |
70b97a7f | 6982 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6983 | unsigned long flags; |
6984 | ||
5cbd54ef IM |
6985 | spin_lock_irqsave(&rq->lock, flags); |
6986 | ||
dd41f596 IM |
6987 | __sched_fork(idle); |
6988 | idle->se.exec_start = sched_clock(); | |
6989 | ||
96f874e2 | 6990 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6991 | __set_task_cpu(idle, cpu); |
1da177e4 | 6992 | |
1da177e4 | 6993 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6994 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6995 | idle->oncpu = 1; | |
6996 | #endif | |
1da177e4 LT |
6997 | spin_unlock_irqrestore(&rq->lock, flags); |
6998 | ||
6999 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
7000 | #if defined(CONFIG_PREEMPT) |
7001 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
7002 | #else | |
a1261f54 | 7003 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 7004 | #endif |
dd41f596 IM |
7005 | /* |
7006 | * The idle tasks have their own, simple scheduling class: | |
7007 | */ | |
7008 | idle->sched_class = &idle_sched_class; | |
fb52607a | 7009 | ftrace_graph_init_task(idle); |
1da177e4 LT |
7010 | } |
7011 | ||
7012 | /* | |
7013 | * In a system that switches off the HZ timer nohz_cpu_mask | |
7014 | * indicates which cpus entered this state. This is used | |
7015 | * in the rcu update to wait only for active cpus. For system | |
7016 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 7017 | * always be CPU_BITS_NONE. |
1da177e4 | 7018 | */ |
6a7b3dc3 | 7019 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 7020 | |
19978ca6 IM |
7021 | /* |
7022 | * Increase the granularity value when there are more CPUs, | |
7023 | * because with more CPUs the 'effective latency' as visible | |
7024 | * to users decreases. But the relationship is not linear, | |
7025 | * so pick a second-best guess by going with the log2 of the | |
7026 | * number of CPUs. | |
7027 | * | |
7028 | * This idea comes from the SD scheduler of Con Kolivas: | |
7029 | */ | |
acb4a848 | 7030 | static int get_update_sysctl_factor(void) |
19978ca6 | 7031 | { |
4ca3ef71 | 7032 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
7033 | unsigned int factor; |
7034 | ||
7035 | switch (sysctl_sched_tunable_scaling) { | |
7036 | case SCHED_TUNABLESCALING_NONE: | |
7037 | factor = 1; | |
7038 | break; | |
7039 | case SCHED_TUNABLESCALING_LINEAR: | |
7040 | factor = cpus; | |
7041 | break; | |
7042 | case SCHED_TUNABLESCALING_LOG: | |
7043 | default: | |
7044 | factor = 1 + ilog2(cpus); | |
7045 | break; | |
7046 | } | |
19978ca6 | 7047 | |
acb4a848 CE |
7048 | return factor; |
7049 | } | |
19978ca6 | 7050 | |
acb4a848 CE |
7051 | static void update_sysctl(void) |
7052 | { | |
7053 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 7054 | |
0bcdcf28 CE |
7055 | #define SET_SYSCTL(name) \ |
7056 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
7057 | SET_SYSCTL(sched_min_granularity); | |
7058 | SET_SYSCTL(sched_latency); | |
7059 | SET_SYSCTL(sched_wakeup_granularity); | |
7060 | SET_SYSCTL(sched_shares_ratelimit); | |
7061 | #undef SET_SYSCTL | |
7062 | } | |
55cd5340 | 7063 | |
0bcdcf28 CE |
7064 | static inline void sched_init_granularity(void) |
7065 | { | |
7066 | update_sysctl(); | |
19978ca6 IM |
7067 | } |
7068 | ||
1da177e4 LT |
7069 | #ifdef CONFIG_SMP |
7070 | /* | |
7071 | * This is how migration works: | |
7072 | * | |
70b97a7f | 7073 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7074 | * runqueue and wake up that CPU's migration thread. |
7075 | * 2) we down() the locked semaphore => thread blocks. | |
7076 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7077 | * thread off the CPU) | |
7078 | * 4) it gets the migration request and checks whether the migrated | |
7079 | * task is still in the wrong runqueue. | |
7080 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7081 | * it and puts it into the right queue. | |
7082 | * 6) migration thread up()s the semaphore. | |
7083 | * 7) we wake up and the migration is done. | |
7084 | */ | |
7085 | ||
7086 | /* | |
7087 | * Change a given task's CPU affinity. Migrate the thread to a | |
7088 | * proper CPU and schedule it away if the CPU it's executing on | |
7089 | * is removed from the allowed bitmask. | |
7090 | * | |
7091 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7092 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7093 | * call is not atomic; no spinlocks may be held. |
7094 | */ | |
96f874e2 | 7095 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7096 | { |
70b97a7f | 7097 | struct migration_req req; |
1da177e4 | 7098 | unsigned long flags; |
70b97a7f | 7099 | struct rq *rq; |
48f24c4d | 7100 | int ret = 0; |
1da177e4 LT |
7101 | |
7102 | rq = task_rq_lock(p, &flags); | |
6ad4c188 | 7103 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
7104 | ret = -EINVAL; |
7105 | goto out; | |
7106 | } | |
7107 | ||
9985b0ba | 7108 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7109 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7110 | ret = -EINVAL; |
7111 | goto out; | |
7112 | } | |
7113 | ||
73fe6aae | 7114 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7115 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7116 | else { |
96f874e2 RR |
7117 | cpumask_copy(&p->cpus_allowed, new_mask); |
7118 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7119 | } |
7120 | ||
1da177e4 | 7121 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7122 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7123 | goto out; |
7124 | ||
6ad4c188 | 7125 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 7126 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7127 | struct task_struct *mt = rq->migration_thread; |
7128 | ||
7129 | get_task_struct(mt); | |
1da177e4 LT |
7130 | task_rq_unlock(rq, &flags); |
7131 | wake_up_process(rq->migration_thread); | |
693525e3 | 7132 | put_task_struct(mt); |
1da177e4 LT |
7133 | wait_for_completion(&req.done); |
7134 | tlb_migrate_finish(p->mm); | |
7135 | return 0; | |
7136 | } | |
7137 | out: | |
7138 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7139 | |
1da177e4 LT |
7140 | return ret; |
7141 | } | |
cd8ba7cd | 7142 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7143 | |
7144 | /* | |
41a2d6cf | 7145 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7146 | * this because either it can't run here any more (set_cpus_allowed() |
7147 | * away from this CPU, or CPU going down), or because we're | |
7148 | * attempting to rebalance this task on exec (sched_exec). | |
7149 | * | |
7150 | * So we race with normal scheduler movements, but that's OK, as long | |
7151 | * as the task is no longer on this CPU. | |
efc30814 KK |
7152 | * |
7153 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7154 | */ |
efc30814 | 7155 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7156 | { |
70b97a7f | 7157 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7158 | int ret = 0, on_rq; |
1da177e4 | 7159 | |
e761b772 | 7160 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7161 | return ret; |
1da177e4 LT |
7162 | |
7163 | rq_src = cpu_rq(src_cpu); | |
7164 | rq_dest = cpu_rq(dest_cpu); | |
7165 | ||
7166 | double_rq_lock(rq_src, rq_dest); | |
7167 | /* Already moved. */ | |
7168 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7169 | goto done; |
1da177e4 | 7170 | /* Affinity changed (again). */ |
96f874e2 | 7171 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7172 | goto fail; |
1da177e4 | 7173 | |
dd41f596 | 7174 | on_rq = p->se.on_rq; |
6e82a3be | 7175 | if (on_rq) |
2e1cb74a | 7176 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7177 | |
1da177e4 | 7178 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7179 | if (on_rq) { |
7180 | activate_task(rq_dest, p, 0); | |
15afe09b | 7181 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7182 | } |
b1e38734 | 7183 | done: |
efc30814 | 7184 | ret = 1; |
b1e38734 | 7185 | fail: |
1da177e4 | 7186 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7187 | return ret; |
1da177e4 LT |
7188 | } |
7189 | ||
03b042bf PM |
7190 | #define RCU_MIGRATION_IDLE 0 |
7191 | #define RCU_MIGRATION_NEED_QS 1 | |
7192 | #define RCU_MIGRATION_GOT_QS 2 | |
7193 | #define RCU_MIGRATION_MUST_SYNC 3 | |
7194 | ||
1da177e4 LT |
7195 | /* |
7196 | * migration_thread - this is a highprio system thread that performs | |
7197 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7198 | * another runqueue. | |
7199 | */ | |
95cdf3b7 | 7200 | static int migration_thread(void *data) |
1da177e4 | 7201 | { |
03b042bf | 7202 | int badcpu; |
1da177e4 | 7203 | int cpu = (long)data; |
70b97a7f | 7204 | struct rq *rq; |
1da177e4 LT |
7205 | |
7206 | rq = cpu_rq(cpu); | |
7207 | BUG_ON(rq->migration_thread != current); | |
7208 | ||
7209 | set_current_state(TASK_INTERRUPTIBLE); | |
7210 | while (!kthread_should_stop()) { | |
70b97a7f | 7211 | struct migration_req *req; |
1da177e4 | 7212 | struct list_head *head; |
1da177e4 | 7213 | |
1da177e4 LT |
7214 | spin_lock_irq(&rq->lock); |
7215 | ||
7216 | if (cpu_is_offline(cpu)) { | |
7217 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7218 | break; |
1da177e4 LT |
7219 | } |
7220 | ||
7221 | if (rq->active_balance) { | |
7222 | active_load_balance(rq, cpu); | |
7223 | rq->active_balance = 0; | |
7224 | } | |
7225 | ||
7226 | head = &rq->migration_queue; | |
7227 | ||
7228 | if (list_empty(head)) { | |
7229 | spin_unlock_irq(&rq->lock); | |
7230 | schedule(); | |
7231 | set_current_state(TASK_INTERRUPTIBLE); | |
7232 | continue; | |
7233 | } | |
70b97a7f | 7234 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7235 | list_del_init(head->next); |
7236 | ||
03b042bf PM |
7237 | if (req->task != NULL) { |
7238 | spin_unlock(&rq->lock); | |
7239 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7240 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
7241 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
7242 | spin_unlock(&rq->lock); | |
7243 | } else { | |
7244 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
7245 | spin_unlock(&rq->lock); | |
7246 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); | |
7247 | } | |
674311d5 | 7248 | local_irq_enable(); |
1da177e4 LT |
7249 | |
7250 | complete(&req->done); | |
7251 | } | |
7252 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7253 | |
1da177e4 LT |
7254 | return 0; |
7255 | } | |
7256 | ||
7257 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7258 | |
7259 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7260 | { | |
7261 | int ret; | |
7262 | ||
7263 | local_irq_disable(); | |
7264 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7265 | local_irq_enable(); | |
7266 | return ret; | |
7267 | } | |
7268 | ||
054b9108 | 7269 | /* |
3a4fa0a2 | 7270 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7271 | */ |
48f24c4d | 7272 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7273 | { |
70b97a7f | 7274 | int dest_cpu; |
6ca09dfc | 7275 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7276 | |
7277 | again: | |
7278 | /* Look for allowed, online CPU in same node. */ | |
6ad4c188 | 7279 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) |
e76bd8d9 RR |
7280 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
7281 | goto move; | |
7282 | ||
7283 | /* Any allowed, online CPU? */ | |
6ad4c188 | 7284 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); |
e76bd8d9 RR |
7285 | if (dest_cpu < nr_cpu_ids) |
7286 | goto move; | |
7287 | ||
7288 | /* No more Mr. Nice Guy. */ | |
7289 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 | 7290 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6ad4c188 | 7291 | dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed); |
1da177e4 | 7292 | |
e76bd8d9 RR |
7293 | /* |
7294 | * Don't tell them about moving exiting tasks or | |
7295 | * kernel threads (both mm NULL), since they never | |
7296 | * leave kernel. | |
7297 | */ | |
7298 | if (p->mm && printk_ratelimit()) { | |
663997d4 JP |
7299 | pr_info("process %d (%s) no longer affine to cpu%d\n", |
7300 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7301 | } |
e76bd8d9 RR |
7302 | } |
7303 | ||
7304 | move: | |
7305 | /* It can have affinity changed while we were choosing. */ | |
7306 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7307 | goto again; | |
1da177e4 LT |
7308 | } |
7309 | ||
7310 | /* | |
7311 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7312 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7313 | * for performance reasons the counter is not stricly tracking tasks to | |
7314 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7315 | * to keep the global sum constant after CPU-down: | |
7316 | */ | |
70b97a7f | 7317 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7318 | { |
6ad4c188 | 7319 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
7320 | unsigned long flags; |
7321 | ||
7322 | local_irq_save(flags); | |
7323 | double_rq_lock(rq_src, rq_dest); | |
7324 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7325 | rq_src->nr_uninterruptible = 0; | |
7326 | double_rq_unlock(rq_src, rq_dest); | |
7327 | local_irq_restore(flags); | |
7328 | } | |
7329 | ||
7330 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7331 | static void migrate_live_tasks(int src_cpu) | |
7332 | { | |
48f24c4d | 7333 | struct task_struct *p, *t; |
1da177e4 | 7334 | |
f7b4cddc | 7335 | read_lock(&tasklist_lock); |
1da177e4 | 7336 | |
48f24c4d IM |
7337 | do_each_thread(t, p) { |
7338 | if (p == current) | |
1da177e4 LT |
7339 | continue; |
7340 | ||
48f24c4d IM |
7341 | if (task_cpu(p) == src_cpu) |
7342 | move_task_off_dead_cpu(src_cpu, p); | |
7343 | } while_each_thread(t, p); | |
1da177e4 | 7344 | |
f7b4cddc | 7345 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7346 | } |
7347 | ||
dd41f596 IM |
7348 | /* |
7349 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7350 | * It does so by boosting its priority to highest possible. |
7351 | * Used by CPU offline code. | |
1da177e4 LT |
7352 | */ |
7353 | void sched_idle_next(void) | |
7354 | { | |
48f24c4d | 7355 | int this_cpu = smp_processor_id(); |
70b97a7f | 7356 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7357 | struct task_struct *p = rq->idle; |
7358 | unsigned long flags; | |
7359 | ||
7360 | /* cpu has to be offline */ | |
48f24c4d | 7361 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7362 | |
48f24c4d IM |
7363 | /* |
7364 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7365 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7366 | */ |
7367 | spin_lock_irqsave(&rq->lock, flags); | |
7368 | ||
dd41f596 | 7369 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7370 | |
94bc9a7b DA |
7371 | update_rq_clock(rq); |
7372 | activate_task(rq, p, 0); | |
1da177e4 LT |
7373 | |
7374 | spin_unlock_irqrestore(&rq->lock, flags); | |
7375 | } | |
7376 | ||
48f24c4d IM |
7377 | /* |
7378 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7379 | * offline. |
7380 | */ | |
7381 | void idle_task_exit(void) | |
7382 | { | |
7383 | struct mm_struct *mm = current->active_mm; | |
7384 | ||
7385 | BUG_ON(cpu_online(smp_processor_id())); | |
7386 | ||
7387 | if (mm != &init_mm) | |
7388 | switch_mm(mm, &init_mm, current); | |
7389 | mmdrop(mm); | |
7390 | } | |
7391 | ||
054b9108 | 7392 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7393 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7394 | { |
70b97a7f | 7395 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7396 | |
7397 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7398 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7399 | |
7400 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7401 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7402 | |
48f24c4d | 7403 | get_task_struct(p); |
1da177e4 LT |
7404 | |
7405 | /* | |
7406 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7407 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7408 | * fine. |
7409 | */ | |
f7b4cddc | 7410 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7411 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7412 | spin_lock_irq(&rq->lock); |
1da177e4 | 7413 | |
48f24c4d | 7414 | put_task_struct(p); |
1da177e4 LT |
7415 | } |
7416 | ||
7417 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7418 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7419 | { | |
70b97a7f | 7420 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7421 | struct task_struct *next; |
48f24c4d | 7422 | |
dd41f596 IM |
7423 | for ( ; ; ) { |
7424 | if (!rq->nr_running) | |
7425 | break; | |
a8e504d2 | 7426 | update_rq_clock(rq); |
b67802ea | 7427 | next = pick_next_task(rq); |
dd41f596 IM |
7428 | if (!next) |
7429 | break; | |
79c53799 | 7430 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7431 | migrate_dead(dead_cpu, next); |
e692ab53 | 7432 | |
1da177e4 LT |
7433 | } |
7434 | } | |
dce48a84 TG |
7435 | |
7436 | /* | |
7437 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7438 | */ | |
7439 | static void calc_global_load_remove(struct rq *rq) | |
7440 | { | |
7441 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7442 | rq->calc_load_active = 0; |
dce48a84 | 7443 | } |
1da177e4 LT |
7444 | #endif /* CONFIG_HOTPLUG_CPU */ |
7445 | ||
e692ab53 NP |
7446 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7447 | ||
7448 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7449 | { |
7450 | .procname = "sched_domain", | |
c57baf1e | 7451 | .mode = 0555, |
e0361851 | 7452 | }, |
56992309 | 7453 | {} |
e692ab53 NP |
7454 | }; |
7455 | ||
7456 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
7457 | { |
7458 | .procname = "kernel", | |
c57baf1e | 7459 | .mode = 0555, |
e0361851 AD |
7460 | .child = sd_ctl_dir, |
7461 | }, | |
56992309 | 7462 | {} |
e692ab53 NP |
7463 | }; |
7464 | ||
7465 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7466 | { | |
7467 | struct ctl_table *entry = | |
5cf9f062 | 7468 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7469 | |
e692ab53 NP |
7470 | return entry; |
7471 | } | |
7472 | ||
6382bc90 MM |
7473 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7474 | { | |
cd790076 | 7475 | struct ctl_table *entry; |
6382bc90 | 7476 | |
cd790076 MM |
7477 | /* |
7478 | * In the intermediate directories, both the child directory and | |
7479 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7480 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7481 | * static strings and all have proc handlers. |
7482 | */ | |
7483 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7484 | if (entry->child) |
7485 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7486 | if (entry->proc_handler == NULL) |
7487 | kfree(entry->procname); | |
7488 | } | |
6382bc90 MM |
7489 | |
7490 | kfree(*tablep); | |
7491 | *tablep = NULL; | |
7492 | } | |
7493 | ||
e692ab53 | 7494 | static void |
e0361851 | 7495 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7496 | const char *procname, void *data, int maxlen, |
7497 | mode_t mode, proc_handler *proc_handler) | |
7498 | { | |
e692ab53 NP |
7499 | entry->procname = procname; |
7500 | entry->data = data; | |
7501 | entry->maxlen = maxlen; | |
7502 | entry->mode = mode; | |
7503 | entry->proc_handler = proc_handler; | |
7504 | } | |
7505 | ||
7506 | static struct ctl_table * | |
7507 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7508 | { | |
a5d8c348 | 7509 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7510 | |
ad1cdc1d MM |
7511 | if (table == NULL) |
7512 | return NULL; | |
7513 | ||
e0361851 | 7514 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7515 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7516 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7517 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7518 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7519 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7520 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7521 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7522 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7523 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7524 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7525 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7526 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7527 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7528 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7529 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7530 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7531 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7532 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7533 | &sd->cache_nice_tries, |
7534 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7535 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7536 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7537 | set_table_entry(&table[11], "name", sd->name, |
7538 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7539 | /* &table[12] is terminator */ | |
e692ab53 NP |
7540 | |
7541 | return table; | |
7542 | } | |
7543 | ||
9a4e7159 | 7544 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7545 | { |
7546 | struct ctl_table *entry, *table; | |
7547 | struct sched_domain *sd; | |
7548 | int domain_num = 0, i; | |
7549 | char buf[32]; | |
7550 | ||
7551 | for_each_domain(cpu, sd) | |
7552 | domain_num++; | |
7553 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7554 | if (table == NULL) |
7555 | return NULL; | |
e692ab53 NP |
7556 | |
7557 | i = 0; | |
7558 | for_each_domain(cpu, sd) { | |
7559 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7560 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7561 | entry->mode = 0555; |
e692ab53 NP |
7562 | entry->child = sd_alloc_ctl_domain_table(sd); |
7563 | entry++; | |
7564 | i++; | |
7565 | } | |
7566 | return table; | |
7567 | } | |
7568 | ||
7569 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7570 | static void register_sched_domain_sysctl(void) |
e692ab53 | 7571 | { |
6ad4c188 | 7572 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
7573 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
7574 | char buf[32]; | |
7575 | ||
7378547f MM |
7576 | WARN_ON(sd_ctl_dir[0].child); |
7577 | sd_ctl_dir[0].child = entry; | |
7578 | ||
ad1cdc1d MM |
7579 | if (entry == NULL) |
7580 | return; | |
7581 | ||
6ad4c188 | 7582 | for_each_possible_cpu(i) { |
e692ab53 | 7583 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7584 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7585 | entry->mode = 0555; |
e692ab53 | 7586 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7587 | entry++; |
e692ab53 | 7588 | } |
7378547f MM |
7589 | |
7590 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7591 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7592 | } | |
6382bc90 | 7593 | |
7378547f | 7594 | /* may be called multiple times per register */ |
6382bc90 MM |
7595 | static void unregister_sched_domain_sysctl(void) |
7596 | { | |
7378547f MM |
7597 | if (sd_sysctl_header) |
7598 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7599 | sd_sysctl_header = NULL; |
7378547f MM |
7600 | if (sd_ctl_dir[0].child) |
7601 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7602 | } |
e692ab53 | 7603 | #else |
6382bc90 MM |
7604 | static void register_sched_domain_sysctl(void) |
7605 | { | |
7606 | } | |
7607 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7608 | { |
7609 | } | |
7610 | #endif | |
7611 | ||
1f11eb6a GH |
7612 | static void set_rq_online(struct rq *rq) |
7613 | { | |
7614 | if (!rq->online) { | |
7615 | const struct sched_class *class; | |
7616 | ||
c6c4927b | 7617 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7618 | rq->online = 1; |
7619 | ||
7620 | for_each_class(class) { | |
7621 | if (class->rq_online) | |
7622 | class->rq_online(rq); | |
7623 | } | |
7624 | } | |
7625 | } | |
7626 | ||
7627 | static void set_rq_offline(struct rq *rq) | |
7628 | { | |
7629 | if (rq->online) { | |
7630 | const struct sched_class *class; | |
7631 | ||
7632 | for_each_class(class) { | |
7633 | if (class->rq_offline) | |
7634 | class->rq_offline(rq); | |
7635 | } | |
7636 | ||
c6c4927b | 7637 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7638 | rq->online = 0; |
7639 | } | |
7640 | } | |
7641 | ||
1da177e4 LT |
7642 | /* |
7643 | * migration_call - callback that gets triggered when a CPU is added. | |
7644 | * Here we can start up the necessary migration thread for the new CPU. | |
7645 | */ | |
48f24c4d IM |
7646 | static int __cpuinit |
7647 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7648 | { |
1da177e4 | 7649 | struct task_struct *p; |
48f24c4d | 7650 | int cpu = (long)hcpu; |
1da177e4 | 7651 | unsigned long flags; |
70b97a7f | 7652 | struct rq *rq; |
1da177e4 LT |
7653 | |
7654 | switch (action) { | |
5be9361c | 7655 | |
1da177e4 | 7656 | case CPU_UP_PREPARE: |
8bb78442 | 7657 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7658 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7659 | if (IS_ERR(p)) |
7660 | return NOTIFY_BAD; | |
1da177e4 LT |
7661 | kthread_bind(p, cpu); |
7662 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7663 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7664 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7665 | task_rq_unlock(rq, &flags); |
371cbb38 | 7666 | get_task_struct(p); |
1da177e4 | 7667 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7668 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7669 | break; |
48f24c4d | 7670 | |
1da177e4 | 7671 | case CPU_ONLINE: |
8bb78442 | 7672 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7673 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7674 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7675 | |
7676 | /* Update our root-domain */ | |
7677 | rq = cpu_rq(cpu); | |
7678 | spin_lock_irqsave(&rq->lock, flags); | |
7679 | if (rq->rd) { | |
c6c4927b | 7680 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7681 | |
7682 | set_rq_online(rq); | |
1f94ef59 GH |
7683 | } |
7684 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7685 | break; |
48f24c4d | 7686 | |
1da177e4 LT |
7687 | #ifdef CONFIG_HOTPLUG_CPU |
7688 | case CPU_UP_CANCELED: | |
8bb78442 | 7689 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7690 | if (!cpu_rq(cpu)->migration_thread) |
7691 | break; | |
41a2d6cf | 7692 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7693 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7694 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7695 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7696 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7697 | cpu_rq(cpu)->migration_thread = NULL; |
7698 | break; | |
48f24c4d | 7699 | |
1da177e4 | 7700 | case CPU_DEAD: |
8bb78442 | 7701 | case CPU_DEAD_FROZEN: |
470fd646 | 7702 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7703 | migrate_live_tasks(cpu); |
7704 | rq = cpu_rq(cpu); | |
7705 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7706 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7707 | rq->migration_thread = NULL; |
7708 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7709 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7710 | update_rq_clock(rq); |
2e1cb74a | 7711 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
7712 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7713 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7714 | migrate_dead_tasks(cpu); |
d2da272a | 7715 | spin_unlock_irq(&rq->lock); |
470fd646 | 7716 | cpuset_unlock(); |
1da177e4 LT |
7717 | migrate_nr_uninterruptible(rq); |
7718 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7719 | calc_global_load_remove(rq); |
41a2d6cf IM |
7720 | /* |
7721 | * No need to migrate the tasks: it was best-effort if | |
7722 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7723 | * the requestors. | |
7724 | */ | |
1da177e4 LT |
7725 | spin_lock_irq(&rq->lock); |
7726 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7727 | struct migration_req *req; |
7728 | ||
1da177e4 | 7729 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7730 | struct migration_req, list); |
1da177e4 | 7731 | list_del_init(&req->list); |
9a2bd244 | 7732 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7733 | complete(&req->done); |
9a2bd244 | 7734 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7735 | } |
7736 | spin_unlock_irq(&rq->lock); | |
7737 | break; | |
57d885fe | 7738 | |
08f503b0 GH |
7739 | case CPU_DYING: |
7740 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7741 | /* Update our root-domain */ |
7742 | rq = cpu_rq(cpu); | |
7743 | spin_lock_irqsave(&rq->lock, flags); | |
7744 | if (rq->rd) { | |
c6c4927b | 7745 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7746 | set_rq_offline(rq); |
57d885fe GH |
7747 | } |
7748 | spin_unlock_irqrestore(&rq->lock, flags); | |
7749 | break; | |
1da177e4 LT |
7750 | #endif |
7751 | } | |
7752 | return NOTIFY_OK; | |
7753 | } | |
7754 | ||
f38b0820 PM |
7755 | /* |
7756 | * Register at high priority so that task migration (migrate_all_tasks) | |
7757 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 7758 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 7759 | */ |
26c2143b | 7760 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7761 | .notifier_call = migration_call, |
7762 | .priority = 10 | |
7763 | }; | |
7764 | ||
7babe8db | 7765 | static int __init migration_init(void) |
1da177e4 LT |
7766 | { |
7767 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7768 | int err; |
48f24c4d IM |
7769 | |
7770 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7771 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7772 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7773 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7774 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7775 | |
a004cd42 | 7776 | return 0; |
1da177e4 | 7777 | } |
7babe8db | 7778 | early_initcall(migration_init); |
1da177e4 LT |
7779 | #endif |
7780 | ||
7781 | #ifdef CONFIG_SMP | |
476f3534 | 7782 | |
3e9830dc | 7783 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7784 | |
f6630114 MT |
7785 | static __read_mostly int sched_domain_debug_enabled; |
7786 | ||
7787 | static int __init sched_domain_debug_setup(char *str) | |
7788 | { | |
7789 | sched_domain_debug_enabled = 1; | |
7790 | ||
7791 | return 0; | |
7792 | } | |
7793 | early_param("sched_debug", sched_domain_debug_setup); | |
7794 | ||
7c16ec58 | 7795 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7796 | struct cpumask *groupmask) |
1da177e4 | 7797 | { |
4dcf6aff | 7798 | struct sched_group *group = sd->groups; |
434d53b0 | 7799 | char str[256]; |
1da177e4 | 7800 | |
968ea6d8 | 7801 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7802 | cpumask_clear(groupmask); |
4dcf6aff IM |
7803 | |
7804 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7805 | ||
7806 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
663997d4 | 7807 | pr_cont("does not load-balance\n"); |
4dcf6aff | 7808 | if (sd->parent) |
663997d4 | 7809 | pr_err("ERROR: !SD_LOAD_BALANCE domain has parent\n"); |
4dcf6aff | 7810 | return -1; |
41c7ce9a NP |
7811 | } |
7812 | ||
663997d4 | 7813 | pr_cont("span %s level %s\n", str, sd->name); |
4dcf6aff | 7814 | |
758b2cdc | 7815 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
663997d4 | 7816 | pr_err("ERROR: domain->span does not contain CPU%d\n", cpu); |
4dcf6aff | 7817 | } |
758b2cdc | 7818 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
663997d4 | 7819 | pr_err("ERROR: domain->groups does not contain CPU%d\n", cpu); |
4dcf6aff | 7820 | } |
1da177e4 | 7821 | |
4dcf6aff | 7822 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7823 | do { |
4dcf6aff | 7824 | if (!group) { |
663997d4 JP |
7825 | pr_cont("\n"); |
7826 | pr_err("ERROR: group is NULL\n"); | |
1da177e4 LT |
7827 | break; |
7828 | } | |
7829 | ||
18a3885f | 7830 | if (!group->cpu_power) { |
663997d4 JP |
7831 | pr_cont("\n"); |
7832 | pr_err("ERROR: domain->cpu_power not set\n"); | |
4dcf6aff IM |
7833 | break; |
7834 | } | |
1da177e4 | 7835 | |
758b2cdc | 7836 | if (!cpumask_weight(sched_group_cpus(group))) { |
663997d4 JP |
7837 | pr_cont("\n"); |
7838 | pr_err("ERROR: empty group\n"); | |
4dcf6aff IM |
7839 | break; |
7840 | } | |
1da177e4 | 7841 | |
758b2cdc | 7842 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
663997d4 JP |
7843 | pr_cont("\n"); |
7844 | pr_err("ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
7845 | break; |
7846 | } | |
1da177e4 | 7847 | |
758b2cdc | 7848 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7849 | |
968ea6d8 | 7850 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 7851 | |
663997d4 | 7852 | pr_cont(" %s", str); |
18a3885f | 7853 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
663997d4 | 7854 | pr_cont(" (cpu_power = %d)", group->cpu_power); |
381512cf | 7855 | } |
1da177e4 | 7856 | |
4dcf6aff IM |
7857 | group = group->next; |
7858 | } while (group != sd->groups); | |
663997d4 | 7859 | pr_cont("\n"); |
1da177e4 | 7860 | |
758b2cdc | 7861 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
663997d4 | 7862 | pr_err("ERROR: groups don't span domain->span\n"); |
1da177e4 | 7863 | |
758b2cdc RR |
7864 | if (sd->parent && |
7865 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
663997d4 | 7866 | pr_err("ERROR: parent span is not a superset of domain->span\n"); |
4dcf6aff IM |
7867 | return 0; |
7868 | } | |
1da177e4 | 7869 | |
4dcf6aff IM |
7870 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7871 | { | |
d5dd3db1 | 7872 | cpumask_var_t groupmask; |
4dcf6aff | 7873 | int level = 0; |
1da177e4 | 7874 | |
f6630114 MT |
7875 | if (!sched_domain_debug_enabled) |
7876 | return; | |
7877 | ||
4dcf6aff IM |
7878 | if (!sd) { |
7879 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7880 | return; | |
7881 | } | |
1da177e4 | 7882 | |
4dcf6aff IM |
7883 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7884 | ||
d5dd3db1 | 7885 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7886 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7887 | return; | |
7888 | } | |
7889 | ||
4dcf6aff | 7890 | for (;;) { |
7c16ec58 | 7891 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7892 | break; |
1da177e4 LT |
7893 | level++; |
7894 | sd = sd->parent; | |
33859f7f | 7895 | if (!sd) |
4dcf6aff IM |
7896 | break; |
7897 | } | |
d5dd3db1 | 7898 | free_cpumask_var(groupmask); |
1da177e4 | 7899 | } |
6d6bc0ad | 7900 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7901 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7902 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7903 | |
1a20ff27 | 7904 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7905 | { |
758b2cdc | 7906 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7907 | return 1; |
7908 | ||
7909 | /* Following flags need at least 2 groups */ | |
7910 | if (sd->flags & (SD_LOAD_BALANCE | | |
7911 | SD_BALANCE_NEWIDLE | | |
7912 | SD_BALANCE_FORK | | |
89c4710e SS |
7913 | SD_BALANCE_EXEC | |
7914 | SD_SHARE_CPUPOWER | | |
7915 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7916 | if (sd->groups != sd->groups->next) |
7917 | return 0; | |
7918 | } | |
7919 | ||
7920 | /* Following flags don't use groups */ | |
c88d5910 | 7921 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
7922 | return 0; |
7923 | ||
7924 | return 1; | |
7925 | } | |
7926 | ||
48f24c4d IM |
7927 | static int |
7928 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7929 | { |
7930 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7931 | ||
7932 | if (sd_degenerate(parent)) | |
7933 | return 1; | |
7934 | ||
758b2cdc | 7935 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7936 | return 0; |
7937 | ||
245af2c7 SS |
7938 | /* Flags needing groups don't count if only 1 group in parent */ |
7939 | if (parent->groups == parent->groups->next) { | |
7940 | pflags &= ~(SD_LOAD_BALANCE | | |
7941 | SD_BALANCE_NEWIDLE | | |
7942 | SD_BALANCE_FORK | | |
89c4710e SS |
7943 | SD_BALANCE_EXEC | |
7944 | SD_SHARE_CPUPOWER | | |
7945 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7946 | if (nr_node_ids == 1) |
7947 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7948 | } |
7949 | if (~cflags & pflags) | |
7950 | return 0; | |
7951 | ||
7952 | return 1; | |
7953 | } | |
7954 | ||
c6c4927b RR |
7955 | static void free_rootdomain(struct root_domain *rd) |
7956 | { | |
047106ad PZ |
7957 | synchronize_sched(); |
7958 | ||
68e74568 RR |
7959 | cpupri_cleanup(&rd->cpupri); |
7960 | ||
c6c4927b RR |
7961 | free_cpumask_var(rd->rto_mask); |
7962 | free_cpumask_var(rd->online); | |
7963 | free_cpumask_var(rd->span); | |
7964 | kfree(rd); | |
7965 | } | |
7966 | ||
57d885fe GH |
7967 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7968 | { | |
a0490fa3 | 7969 | struct root_domain *old_rd = NULL; |
57d885fe | 7970 | unsigned long flags; |
57d885fe GH |
7971 | |
7972 | spin_lock_irqsave(&rq->lock, flags); | |
7973 | ||
7974 | if (rq->rd) { | |
a0490fa3 | 7975 | old_rd = rq->rd; |
57d885fe | 7976 | |
c6c4927b | 7977 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7978 | set_rq_offline(rq); |
57d885fe | 7979 | |
c6c4927b | 7980 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7981 | |
a0490fa3 IM |
7982 | /* |
7983 | * If we dont want to free the old_rt yet then | |
7984 | * set old_rd to NULL to skip the freeing later | |
7985 | * in this function: | |
7986 | */ | |
7987 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7988 | old_rd = NULL; | |
57d885fe GH |
7989 | } |
7990 | ||
7991 | atomic_inc(&rd->refcount); | |
7992 | rq->rd = rd; | |
7993 | ||
c6c4927b | 7994 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 7995 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 7996 | set_rq_online(rq); |
57d885fe GH |
7997 | |
7998 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7999 | |
8000 | if (old_rd) | |
8001 | free_rootdomain(old_rd); | |
57d885fe GH |
8002 | } |
8003 | ||
fd5e1b5d | 8004 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 8005 | { |
36b7b6d4 PE |
8006 | gfp_t gfp = GFP_KERNEL; |
8007 | ||
57d885fe GH |
8008 | memset(rd, 0, sizeof(*rd)); |
8009 | ||
36b7b6d4 PE |
8010 | if (bootmem) |
8011 | gfp = GFP_NOWAIT; | |
c6c4927b | 8012 | |
36b7b6d4 | 8013 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 8014 | goto out; |
36b7b6d4 | 8015 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 8016 | goto free_span; |
36b7b6d4 | 8017 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 8018 | goto free_online; |
6e0534f2 | 8019 | |
0fb53029 | 8020 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 8021 | goto free_rto_mask; |
c6c4927b | 8022 | return 0; |
6e0534f2 | 8023 | |
68e74568 RR |
8024 | free_rto_mask: |
8025 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
8026 | free_online: |
8027 | free_cpumask_var(rd->online); | |
8028 | free_span: | |
8029 | free_cpumask_var(rd->span); | |
0c910d28 | 8030 | out: |
c6c4927b | 8031 | return -ENOMEM; |
57d885fe GH |
8032 | } |
8033 | ||
8034 | static void init_defrootdomain(void) | |
8035 | { | |
c6c4927b RR |
8036 | init_rootdomain(&def_root_domain, true); |
8037 | ||
57d885fe GH |
8038 | atomic_set(&def_root_domain.refcount, 1); |
8039 | } | |
8040 | ||
dc938520 | 8041 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
8042 | { |
8043 | struct root_domain *rd; | |
8044 | ||
8045 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
8046 | if (!rd) | |
8047 | return NULL; | |
8048 | ||
c6c4927b RR |
8049 | if (init_rootdomain(rd, false) != 0) { |
8050 | kfree(rd); | |
8051 | return NULL; | |
8052 | } | |
57d885fe GH |
8053 | |
8054 | return rd; | |
8055 | } | |
8056 | ||
1da177e4 | 8057 | /* |
0eab9146 | 8058 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
8059 | * hold the hotplug lock. |
8060 | */ | |
0eab9146 IM |
8061 | static void |
8062 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 8063 | { |
70b97a7f | 8064 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8065 | struct sched_domain *tmp; |
8066 | ||
8067 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8068 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8069 | struct sched_domain *parent = tmp->parent; |
8070 | if (!parent) | |
8071 | break; | |
f29c9b1c | 8072 | |
1a848870 | 8073 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8074 | tmp->parent = parent->parent; |
1a848870 SS |
8075 | if (parent->parent) |
8076 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8077 | } else |
8078 | tmp = tmp->parent; | |
245af2c7 SS |
8079 | } |
8080 | ||
1a848870 | 8081 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8082 | sd = sd->parent; |
1a848870 SS |
8083 | if (sd) |
8084 | sd->child = NULL; | |
8085 | } | |
1da177e4 LT |
8086 | |
8087 | sched_domain_debug(sd, cpu); | |
8088 | ||
57d885fe | 8089 | rq_attach_root(rq, rd); |
674311d5 | 8090 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8091 | } |
8092 | ||
8093 | /* cpus with isolated domains */ | |
dcc30a35 | 8094 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8095 | |
8096 | /* Setup the mask of cpus configured for isolated domains */ | |
8097 | static int __init isolated_cpu_setup(char *str) | |
8098 | { | |
bdddd296 | 8099 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 8100 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8101 | return 1; |
8102 | } | |
8103 | ||
8927f494 | 8104 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8105 | |
8106 | /* | |
6711cab4 SS |
8107 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8108 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8109 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8110 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8111 | * |
8112 | * init_sched_build_groups will build a circular linked list of the groups | |
8113 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8114 | * and ->cpu_power to 0. | |
8115 | */ | |
a616058b | 8116 | static void |
96f874e2 RR |
8117 | init_sched_build_groups(const struct cpumask *span, |
8118 | const struct cpumask *cpu_map, | |
8119 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8120 | struct sched_group **sg, |
96f874e2 RR |
8121 | struct cpumask *tmpmask), |
8122 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8123 | { |
8124 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8125 | int i; |
8126 | ||
96f874e2 | 8127 | cpumask_clear(covered); |
7c16ec58 | 8128 | |
abcd083a | 8129 | for_each_cpu(i, span) { |
6711cab4 | 8130 | struct sched_group *sg; |
7c16ec58 | 8131 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8132 | int j; |
8133 | ||
758b2cdc | 8134 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8135 | continue; |
8136 | ||
758b2cdc | 8137 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 8138 | sg->cpu_power = 0; |
1da177e4 | 8139 | |
abcd083a | 8140 | for_each_cpu(j, span) { |
7c16ec58 | 8141 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8142 | continue; |
8143 | ||
96f874e2 | 8144 | cpumask_set_cpu(j, covered); |
758b2cdc | 8145 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8146 | } |
8147 | if (!first) | |
8148 | first = sg; | |
8149 | if (last) | |
8150 | last->next = sg; | |
8151 | last = sg; | |
8152 | } | |
8153 | last->next = first; | |
8154 | } | |
8155 | ||
9c1cfda2 | 8156 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8157 | |
9c1cfda2 | 8158 | #ifdef CONFIG_NUMA |
198e2f18 | 8159 | |
9c1cfda2 JH |
8160 | /** |
8161 | * find_next_best_node - find the next node to include in a sched_domain | |
8162 | * @node: node whose sched_domain we're building | |
8163 | * @used_nodes: nodes already in the sched_domain | |
8164 | * | |
41a2d6cf | 8165 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8166 | * finds the closest node not already in the @used_nodes map. |
8167 | * | |
8168 | * Should use nodemask_t. | |
8169 | */ | |
c5f59f08 | 8170 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8171 | { |
8172 | int i, n, val, min_val, best_node = 0; | |
8173 | ||
8174 | min_val = INT_MAX; | |
8175 | ||
076ac2af | 8176 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8177 | /* Start at @node */ |
076ac2af | 8178 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8179 | |
8180 | if (!nr_cpus_node(n)) | |
8181 | continue; | |
8182 | ||
8183 | /* Skip already used nodes */ | |
c5f59f08 | 8184 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8185 | continue; |
8186 | ||
8187 | /* Simple min distance search */ | |
8188 | val = node_distance(node, n); | |
8189 | ||
8190 | if (val < min_val) { | |
8191 | min_val = val; | |
8192 | best_node = n; | |
8193 | } | |
8194 | } | |
8195 | ||
c5f59f08 | 8196 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8197 | return best_node; |
8198 | } | |
8199 | ||
8200 | /** | |
8201 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8202 | * @node: node whose cpumask we're constructing | |
73486722 | 8203 | * @span: resulting cpumask |
9c1cfda2 | 8204 | * |
41a2d6cf | 8205 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8206 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8207 | * out optimally. | |
8208 | */ | |
96f874e2 | 8209 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8210 | { |
c5f59f08 | 8211 | nodemask_t used_nodes; |
48f24c4d | 8212 | int i; |
9c1cfda2 | 8213 | |
6ca09dfc | 8214 | cpumask_clear(span); |
c5f59f08 | 8215 | nodes_clear(used_nodes); |
9c1cfda2 | 8216 | |
6ca09dfc | 8217 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8218 | node_set(node, used_nodes); |
9c1cfda2 JH |
8219 | |
8220 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8221 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8222 | |
6ca09dfc | 8223 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8224 | } |
9c1cfda2 | 8225 | } |
6d6bc0ad | 8226 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8227 | |
5c45bf27 | 8228 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8229 | |
6c99e9ad RR |
8230 | /* |
8231 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8232 | * |
8233 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8234 | * and struct sched_domain. ) | |
6c99e9ad RR |
8235 | */ |
8236 | struct static_sched_group { | |
8237 | struct sched_group sg; | |
8238 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8239 | }; | |
8240 | ||
8241 | struct static_sched_domain { | |
8242 | struct sched_domain sd; | |
8243 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8244 | }; | |
8245 | ||
49a02c51 AH |
8246 | struct s_data { |
8247 | #ifdef CONFIG_NUMA | |
8248 | int sd_allnodes; | |
8249 | cpumask_var_t domainspan; | |
8250 | cpumask_var_t covered; | |
8251 | cpumask_var_t notcovered; | |
8252 | #endif | |
8253 | cpumask_var_t nodemask; | |
8254 | cpumask_var_t this_sibling_map; | |
8255 | cpumask_var_t this_core_map; | |
8256 | cpumask_var_t send_covered; | |
8257 | cpumask_var_t tmpmask; | |
8258 | struct sched_group **sched_group_nodes; | |
8259 | struct root_domain *rd; | |
8260 | }; | |
8261 | ||
2109b99e AH |
8262 | enum s_alloc { |
8263 | sa_sched_groups = 0, | |
8264 | sa_rootdomain, | |
8265 | sa_tmpmask, | |
8266 | sa_send_covered, | |
8267 | sa_this_core_map, | |
8268 | sa_this_sibling_map, | |
8269 | sa_nodemask, | |
8270 | sa_sched_group_nodes, | |
8271 | #ifdef CONFIG_NUMA | |
8272 | sa_notcovered, | |
8273 | sa_covered, | |
8274 | sa_domainspan, | |
8275 | #endif | |
8276 | sa_none, | |
8277 | }; | |
8278 | ||
9c1cfda2 | 8279 | /* |
48f24c4d | 8280 | * SMT sched-domains: |
9c1cfda2 | 8281 | */ |
1da177e4 | 8282 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8283 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8284 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8285 | |
41a2d6cf | 8286 | static int |
96f874e2 RR |
8287 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8288 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8289 | { |
6711cab4 | 8290 | if (sg) |
6c99e9ad | 8291 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8292 | return cpu; |
8293 | } | |
6d6bc0ad | 8294 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8295 | |
48f24c4d IM |
8296 | /* |
8297 | * multi-core sched-domains: | |
8298 | */ | |
1e9f28fa | 8299 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8300 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8301 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8302 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8303 | |
8304 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8305 | static int |
96f874e2 RR |
8306 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8307 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8308 | { |
6711cab4 | 8309 | int group; |
7c16ec58 | 8310 | |
c69fc56d | 8311 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8312 | group = cpumask_first(mask); |
6711cab4 | 8313 | if (sg) |
6c99e9ad | 8314 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8315 | return group; |
1e9f28fa SS |
8316 | } |
8317 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8318 | static int |
96f874e2 RR |
8319 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8320 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8321 | { |
6711cab4 | 8322 | if (sg) |
6c99e9ad | 8323 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8324 | return cpu; |
8325 | } | |
8326 | #endif | |
8327 | ||
6c99e9ad RR |
8328 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8329 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8330 | |
41a2d6cf | 8331 | static int |
96f874e2 RR |
8332 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8333 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8334 | { |
6711cab4 | 8335 | int group; |
48f24c4d | 8336 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8337 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8338 | group = cpumask_first(mask); |
1e9f28fa | 8339 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8340 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8341 | group = cpumask_first(mask); |
1da177e4 | 8342 | #else |
6711cab4 | 8343 | group = cpu; |
1da177e4 | 8344 | #endif |
6711cab4 | 8345 | if (sg) |
6c99e9ad | 8346 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8347 | return group; |
1da177e4 LT |
8348 | } |
8349 | ||
8350 | #ifdef CONFIG_NUMA | |
1da177e4 | 8351 | /* |
9c1cfda2 JH |
8352 | * The init_sched_build_groups can't handle what we want to do with node |
8353 | * groups, so roll our own. Now each node has its own list of groups which | |
8354 | * gets dynamically allocated. | |
1da177e4 | 8355 | */ |
62ea9ceb | 8356 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8357 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8358 | |
62ea9ceb | 8359 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8360 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8361 | |
96f874e2 RR |
8362 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8363 | struct sched_group **sg, | |
8364 | struct cpumask *nodemask) | |
9c1cfda2 | 8365 | { |
6711cab4 SS |
8366 | int group; |
8367 | ||
6ca09dfc | 8368 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8369 | group = cpumask_first(nodemask); |
6711cab4 SS |
8370 | |
8371 | if (sg) | |
6c99e9ad | 8372 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8373 | return group; |
1da177e4 | 8374 | } |
6711cab4 | 8375 | |
08069033 SS |
8376 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8377 | { | |
8378 | struct sched_group *sg = group_head; | |
8379 | int j; | |
8380 | ||
8381 | if (!sg) | |
8382 | return; | |
3a5c359a | 8383 | do { |
758b2cdc | 8384 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8385 | struct sched_domain *sd; |
08069033 | 8386 | |
6c99e9ad | 8387 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8388 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8389 | /* |
8390 | * Only add "power" once for each | |
8391 | * physical package. | |
8392 | */ | |
8393 | continue; | |
8394 | } | |
08069033 | 8395 | |
18a3885f | 8396 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
8397 | } |
8398 | sg = sg->next; | |
8399 | } while (sg != group_head); | |
08069033 | 8400 | } |
0601a88d AH |
8401 | |
8402 | static int build_numa_sched_groups(struct s_data *d, | |
8403 | const struct cpumask *cpu_map, int num) | |
8404 | { | |
8405 | struct sched_domain *sd; | |
8406 | struct sched_group *sg, *prev; | |
8407 | int n, j; | |
8408 | ||
8409 | cpumask_clear(d->covered); | |
8410 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8411 | if (cpumask_empty(d->nodemask)) { | |
8412 | d->sched_group_nodes[num] = NULL; | |
8413 | goto out; | |
8414 | } | |
8415 | ||
8416 | sched_domain_node_span(num, d->domainspan); | |
8417 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8418 | ||
8419 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8420 | GFP_KERNEL, num); | |
8421 | if (!sg) { | |
663997d4 | 8422 | pr_warning("Can not alloc domain group for node %d\n", num); |
0601a88d AH |
8423 | return -ENOMEM; |
8424 | } | |
8425 | d->sched_group_nodes[num] = sg; | |
8426 | ||
8427 | for_each_cpu(j, d->nodemask) { | |
8428 | sd = &per_cpu(node_domains, j).sd; | |
8429 | sd->groups = sg; | |
8430 | } | |
8431 | ||
18a3885f | 8432 | sg->cpu_power = 0; |
0601a88d AH |
8433 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
8434 | sg->next = sg; | |
8435 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8436 | ||
8437 | prev = sg; | |
8438 | for (j = 0; j < nr_node_ids; j++) { | |
8439 | n = (num + j) % nr_node_ids; | |
8440 | cpumask_complement(d->notcovered, d->covered); | |
8441 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8442 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8443 | if (cpumask_empty(d->tmpmask)) | |
8444 | break; | |
8445 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8446 | if (cpumask_empty(d->tmpmask)) | |
8447 | continue; | |
8448 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8449 | GFP_KERNEL, num); | |
8450 | if (!sg) { | |
663997d4 JP |
8451 | pr_warning("Can not alloc domain group for node %d\n", |
8452 | j); | |
0601a88d AH |
8453 | return -ENOMEM; |
8454 | } | |
18a3885f | 8455 | sg->cpu_power = 0; |
0601a88d AH |
8456 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
8457 | sg->next = prev->next; | |
8458 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8459 | prev->next = sg; | |
8460 | prev = sg; | |
8461 | } | |
8462 | out: | |
8463 | return 0; | |
8464 | } | |
6d6bc0ad | 8465 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8466 | |
a616058b | 8467 | #ifdef CONFIG_NUMA |
51888ca2 | 8468 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8469 | static void free_sched_groups(const struct cpumask *cpu_map, |
8470 | struct cpumask *nodemask) | |
51888ca2 | 8471 | { |
a616058b | 8472 | int cpu, i; |
51888ca2 | 8473 | |
abcd083a | 8474 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8475 | struct sched_group **sched_group_nodes |
8476 | = sched_group_nodes_bycpu[cpu]; | |
8477 | ||
51888ca2 SV |
8478 | if (!sched_group_nodes) |
8479 | continue; | |
8480 | ||
076ac2af | 8481 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8482 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8483 | ||
6ca09dfc | 8484 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8485 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8486 | continue; |
8487 | ||
8488 | if (sg == NULL) | |
8489 | continue; | |
8490 | sg = sg->next; | |
8491 | next_sg: | |
8492 | oldsg = sg; | |
8493 | sg = sg->next; | |
8494 | kfree(oldsg); | |
8495 | if (oldsg != sched_group_nodes[i]) | |
8496 | goto next_sg; | |
8497 | } | |
8498 | kfree(sched_group_nodes); | |
8499 | sched_group_nodes_bycpu[cpu] = NULL; | |
8500 | } | |
51888ca2 | 8501 | } |
6d6bc0ad | 8502 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8503 | static void free_sched_groups(const struct cpumask *cpu_map, |
8504 | struct cpumask *nodemask) | |
a616058b SS |
8505 | { |
8506 | } | |
6d6bc0ad | 8507 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8508 | |
89c4710e SS |
8509 | /* |
8510 | * Initialize sched groups cpu_power. | |
8511 | * | |
8512 | * cpu_power indicates the capacity of sched group, which is used while | |
8513 | * distributing the load between different sched groups in a sched domain. | |
8514 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8515 | * there are asymmetries in the topology. If there are asymmetries, group | |
8516 | * having more cpu_power will pickup more load compared to the group having | |
8517 | * less cpu_power. | |
89c4710e SS |
8518 | */ |
8519 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8520 | { | |
8521 | struct sched_domain *child; | |
8522 | struct sched_group *group; | |
f93e65c1 PZ |
8523 | long power; |
8524 | int weight; | |
89c4710e SS |
8525 | |
8526 | WARN_ON(!sd || !sd->groups); | |
8527 | ||
13318a71 | 8528 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8529 | return; |
8530 | ||
8531 | child = sd->child; | |
8532 | ||
18a3885f | 8533 | sd->groups->cpu_power = 0; |
5517d86b | 8534 | |
f93e65c1 PZ |
8535 | if (!child) { |
8536 | power = SCHED_LOAD_SCALE; | |
8537 | weight = cpumask_weight(sched_domain_span(sd)); | |
8538 | /* | |
8539 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8540 | * Usually multiple threads get a better yield out of |
8541 | * that one core than a single thread would have, | |
8542 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8543 | */ |
a52bfd73 PZ |
8544 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8545 | power *= sd->smt_gain; | |
f93e65c1 | 8546 | power /= weight; |
a52bfd73 PZ |
8547 | power >>= SCHED_LOAD_SHIFT; |
8548 | } | |
18a3885f | 8549 | sd->groups->cpu_power += power; |
89c4710e SS |
8550 | return; |
8551 | } | |
8552 | ||
89c4710e | 8553 | /* |
f93e65c1 | 8554 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8555 | */ |
8556 | group = child->groups; | |
8557 | do { | |
18a3885f | 8558 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
8559 | group = group->next; |
8560 | } while (group != child->groups); | |
8561 | } | |
8562 | ||
7c16ec58 MT |
8563 | /* |
8564 | * Initializers for schedule domains | |
8565 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8566 | */ | |
8567 | ||
a5d8c348 IM |
8568 | #ifdef CONFIG_SCHED_DEBUG |
8569 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8570 | #else | |
8571 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8572 | #endif | |
8573 | ||
7c16ec58 | 8574 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8575 | |
7c16ec58 MT |
8576 | #define SD_INIT_FUNC(type) \ |
8577 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8578 | { \ | |
8579 | memset(sd, 0, sizeof(*sd)); \ | |
8580 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8581 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8582 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8583 | } |
8584 | ||
8585 | SD_INIT_FUNC(CPU) | |
8586 | #ifdef CONFIG_NUMA | |
8587 | SD_INIT_FUNC(ALLNODES) | |
8588 | SD_INIT_FUNC(NODE) | |
8589 | #endif | |
8590 | #ifdef CONFIG_SCHED_SMT | |
8591 | SD_INIT_FUNC(SIBLING) | |
8592 | #endif | |
8593 | #ifdef CONFIG_SCHED_MC | |
8594 | SD_INIT_FUNC(MC) | |
8595 | #endif | |
8596 | ||
1d3504fc HS |
8597 | static int default_relax_domain_level = -1; |
8598 | ||
8599 | static int __init setup_relax_domain_level(char *str) | |
8600 | { | |
30e0e178 LZ |
8601 | unsigned long val; |
8602 | ||
8603 | val = simple_strtoul(str, NULL, 0); | |
8604 | if (val < SD_LV_MAX) | |
8605 | default_relax_domain_level = val; | |
8606 | ||
1d3504fc HS |
8607 | return 1; |
8608 | } | |
8609 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8610 | ||
8611 | static void set_domain_attribute(struct sched_domain *sd, | |
8612 | struct sched_domain_attr *attr) | |
8613 | { | |
8614 | int request; | |
8615 | ||
8616 | if (!attr || attr->relax_domain_level < 0) { | |
8617 | if (default_relax_domain_level < 0) | |
8618 | return; | |
8619 | else | |
8620 | request = default_relax_domain_level; | |
8621 | } else | |
8622 | request = attr->relax_domain_level; | |
8623 | if (request < sd->level) { | |
8624 | /* turn off idle balance on this domain */ | |
c88d5910 | 8625 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8626 | } else { |
8627 | /* turn on idle balance on this domain */ | |
c88d5910 | 8628 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8629 | } |
8630 | } | |
8631 | ||
2109b99e AH |
8632 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8633 | const struct cpumask *cpu_map) | |
8634 | { | |
8635 | switch (what) { | |
8636 | case sa_sched_groups: | |
8637 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8638 | d->sched_group_nodes = NULL; | |
8639 | case sa_rootdomain: | |
8640 | free_rootdomain(d->rd); /* fall through */ | |
8641 | case sa_tmpmask: | |
8642 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8643 | case sa_send_covered: | |
8644 | free_cpumask_var(d->send_covered); /* fall through */ | |
8645 | case sa_this_core_map: | |
8646 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8647 | case sa_this_sibling_map: | |
8648 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8649 | case sa_nodemask: | |
8650 | free_cpumask_var(d->nodemask); /* fall through */ | |
8651 | case sa_sched_group_nodes: | |
d1b55138 | 8652 | #ifdef CONFIG_NUMA |
2109b99e AH |
8653 | kfree(d->sched_group_nodes); /* fall through */ |
8654 | case sa_notcovered: | |
8655 | free_cpumask_var(d->notcovered); /* fall through */ | |
8656 | case sa_covered: | |
8657 | free_cpumask_var(d->covered); /* fall through */ | |
8658 | case sa_domainspan: | |
8659 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8660 | #endif |
2109b99e AH |
8661 | case sa_none: |
8662 | break; | |
8663 | } | |
8664 | } | |
3404c8d9 | 8665 | |
2109b99e AH |
8666 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8667 | const struct cpumask *cpu_map) | |
8668 | { | |
3404c8d9 | 8669 | #ifdef CONFIG_NUMA |
2109b99e AH |
8670 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8671 | return sa_none; | |
8672 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8673 | return sa_domainspan; | |
8674 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8675 | return sa_covered; | |
8676 | /* Allocate the per-node list of sched groups */ | |
8677 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8678 | sizeof(struct sched_group *), GFP_KERNEL); | |
8679 | if (!d->sched_group_nodes) { | |
663997d4 | 8680 | pr_warning("Can not alloc sched group node list\n"); |
2109b99e | 8681 | return sa_notcovered; |
d1b55138 | 8682 | } |
2109b99e | 8683 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8684 | #endif |
2109b99e AH |
8685 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8686 | return sa_sched_group_nodes; | |
8687 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8688 | return sa_nodemask; | |
8689 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8690 | return sa_this_sibling_map; | |
8691 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8692 | return sa_this_core_map; | |
8693 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8694 | return sa_send_covered; | |
8695 | d->rd = alloc_rootdomain(); | |
8696 | if (!d->rd) { | |
663997d4 | 8697 | pr_warning("Cannot alloc root domain\n"); |
2109b99e | 8698 | return sa_tmpmask; |
57d885fe | 8699 | } |
2109b99e AH |
8700 | return sa_rootdomain; |
8701 | } | |
57d885fe | 8702 | |
7f4588f3 AH |
8703 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8704 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8705 | { | |
8706 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8707 | #ifdef CONFIG_NUMA |
7f4588f3 | 8708 | struct sched_domain *parent; |
1da177e4 | 8709 | |
7f4588f3 AH |
8710 | d->sd_allnodes = 0; |
8711 | if (cpumask_weight(cpu_map) > | |
8712 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8713 | sd = &per_cpu(allnodes_domains, i).sd; | |
8714 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8715 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8716 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8717 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8718 | d->sd_allnodes = 1; | |
8719 | } | |
8720 | parent = sd; | |
8721 | ||
8722 | sd = &per_cpu(node_domains, i).sd; | |
8723 | SD_INIT(sd, NODE); | |
8724 | set_domain_attribute(sd, attr); | |
8725 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8726 | sd->parent = parent; | |
8727 | if (parent) | |
8728 | parent->child = sd; | |
8729 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8730 | #endif |
7f4588f3 AH |
8731 | return sd; |
8732 | } | |
1da177e4 | 8733 | |
87cce662 AH |
8734 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8735 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8736 | struct sched_domain *parent, int i) | |
8737 | { | |
8738 | struct sched_domain *sd; | |
8739 | sd = &per_cpu(phys_domains, i).sd; | |
8740 | SD_INIT(sd, CPU); | |
8741 | set_domain_attribute(sd, attr); | |
8742 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8743 | sd->parent = parent; | |
8744 | if (parent) | |
8745 | parent->child = sd; | |
8746 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8747 | return sd; | |
8748 | } | |
1da177e4 | 8749 | |
410c4081 AH |
8750 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8751 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8752 | struct sched_domain *parent, int i) | |
8753 | { | |
8754 | struct sched_domain *sd = parent; | |
1e9f28fa | 8755 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8756 | sd = &per_cpu(core_domains, i).sd; |
8757 | SD_INIT(sd, MC); | |
8758 | set_domain_attribute(sd, attr); | |
8759 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8760 | sd->parent = parent; | |
8761 | parent->child = sd; | |
8762 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8763 | #endif |
410c4081 AH |
8764 | return sd; |
8765 | } | |
1e9f28fa | 8766 | |
d8173535 AH |
8767 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8768 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8769 | struct sched_domain *parent, int i) | |
8770 | { | |
8771 | struct sched_domain *sd = parent; | |
1da177e4 | 8772 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8773 | sd = &per_cpu(cpu_domains, i).sd; |
8774 | SD_INIT(sd, SIBLING); | |
8775 | set_domain_attribute(sd, attr); | |
8776 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8777 | sd->parent = parent; | |
8778 | parent->child = sd; | |
8779 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8780 | #endif |
d8173535 AH |
8781 | return sd; |
8782 | } | |
1da177e4 | 8783 | |
0e8e85c9 AH |
8784 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8785 | const struct cpumask *cpu_map, int cpu) | |
8786 | { | |
8787 | switch (l) { | |
1da177e4 | 8788 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8789 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8790 | cpumask_and(d->this_sibling_map, cpu_map, | |
8791 | topology_thread_cpumask(cpu)); | |
8792 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8793 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8794 | &cpu_to_cpu_group, | |
8795 | d->send_covered, d->tmpmask); | |
8796 | break; | |
1da177e4 | 8797 | #endif |
1e9f28fa | 8798 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8799 | case SD_LV_MC: /* set up multi-core groups */ |
8800 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8801 | if (cpu == cpumask_first(d->this_core_map)) | |
8802 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8803 | &cpu_to_core_group, | |
8804 | d->send_covered, d->tmpmask); | |
8805 | break; | |
1e9f28fa | 8806 | #endif |
86548096 AH |
8807 | case SD_LV_CPU: /* set up physical groups */ |
8808 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8809 | if (!cpumask_empty(d->nodemask)) | |
8810 | init_sched_build_groups(d->nodemask, cpu_map, | |
8811 | &cpu_to_phys_group, | |
8812 | d->send_covered, d->tmpmask); | |
8813 | break; | |
1da177e4 | 8814 | #ifdef CONFIG_NUMA |
de616e36 AH |
8815 | case SD_LV_ALLNODES: |
8816 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8817 | d->send_covered, d->tmpmask); | |
8818 | break; | |
8819 | #endif | |
0e8e85c9 AH |
8820 | default: |
8821 | break; | |
7c16ec58 | 8822 | } |
0e8e85c9 | 8823 | } |
9c1cfda2 | 8824 | |
2109b99e AH |
8825 | /* |
8826 | * Build sched domains for a given set of cpus and attach the sched domains | |
8827 | * to the individual cpus | |
8828 | */ | |
8829 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8830 | struct sched_domain_attr *attr) | |
8831 | { | |
8832 | enum s_alloc alloc_state = sa_none; | |
8833 | struct s_data d; | |
294b0c96 | 8834 | struct sched_domain *sd; |
2109b99e | 8835 | int i; |
7c16ec58 | 8836 | #ifdef CONFIG_NUMA |
2109b99e | 8837 | d.sd_allnodes = 0; |
7c16ec58 | 8838 | #endif |
9c1cfda2 | 8839 | |
2109b99e AH |
8840 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8841 | if (alloc_state != sa_rootdomain) | |
8842 | goto error; | |
8843 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8844 | |
1da177e4 | 8845 | /* |
1a20ff27 | 8846 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8847 | */ |
abcd083a | 8848 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8849 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8850 | cpu_map); | |
9761eea8 | 8851 | |
7f4588f3 | 8852 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8853 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8854 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8855 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8856 | } |
9c1cfda2 | 8857 | |
abcd083a | 8858 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8859 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8860 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8861 | } |
9c1cfda2 | 8862 | |
1da177e4 | 8863 | /* Set up physical groups */ |
86548096 AH |
8864 | for (i = 0; i < nr_node_ids; i++) |
8865 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8866 | |
1da177e4 LT |
8867 | #ifdef CONFIG_NUMA |
8868 | /* Set up node groups */ | |
de616e36 AH |
8869 | if (d.sd_allnodes) |
8870 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8871 | |
0601a88d AH |
8872 | for (i = 0; i < nr_node_ids; i++) |
8873 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8874 | goto error; |
1da177e4 LT |
8875 | #endif |
8876 | ||
8877 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8878 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8879 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8880 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8881 | init_sched_groups_power(i, sd); |
5c45bf27 | 8882 | } |
1da177e4 | 8883 | #endif |
1e9f28fa | 8884 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8885 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8886 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8887 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8888 | } |
8889 | #endif | |
1e9f28fa | 8890 | |
abcd083a | 8891 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8892 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8893 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8894 | } |
8895 | ||
9c1cfda2 | 8896 | #ifdef CONFIG_NUMA |
076ac2af | 8897 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8898 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8899 | |
49a02c51 | 8900 | if (d.sd_allnodes) { |
6711cab4 | 8901 | struct sched_group *sg; |
f712c0c7 | 8902 | |
96f874e2 | 8903 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 8904 | d.tmpmask); |
f712c0c7 SS |
8905 | init_numa_sched_groups_power(sg); |
8906 | } | |
9c1cfda2 JH |
8907 | #endif |
8908 | ||
1da177e4 | 8909 | /* Attach the domains */ |
abcd083a | 8910 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8911 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8912 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8913 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8914 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8915 | #else |
6c99e9ad | 8916 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8917 | #endif |
49a02c51 | 8918 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 8919 | } |
51888ca2 | 8920 | |
2109b99e AH |
8921 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
8922 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
8923 | return 0; | |
51888ca2 | 8924 | |
51888ca2 | 8925 | error: |
2109b99e AH |
8926 | __free_domain_allocs(&d, alloc_state, cpu_map); |
8927 | return -ENOMEM; | |
1da177e4 | 8928 | } |
029190c5 | 8929 | |
96f874e2 | 8930 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8931 | { |
8932 | return __build_sched_domains(cpu_map, NULL); | |
8933 | } | |
8934 | ||
acc3f5d7 | 8935 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 8936 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8937 | static struct sched_domain_attr *dattr_cur; |
8938 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8939 | |
8940 | /* | |
8941 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8942 | * cpumask) fails, then fallback to a single sched domain, |
8943 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8944 | */ |
4212823f | 8945 | static cpumask_var_t fallback_doms; |
029190c5 | 8946 | |
ee79d1bd HC |
8947 | /* |
8948 | * arch_update_cpu_topology lets virtualized architectures update the | |
8949 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8950 | * or 0 if it stayed the same. | |
8951 | */ | |
8952 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8953 | { |
ee79d1bd | 8954 | return 0; |
22e52b07 HC |
8955 | } |
8956 | ||
acc3f5d7 RR |
8957 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
8958 | { | |
8959 | int i; | |
8960 | cpumask_var_t *doms; | |
8961 | ||
8962 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
8963 | if (!doms) | |
8964 | return NULL; | |
8965 | for (i = 0; i < ndoms; i++) { | |
8966 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
8967 | free_sched_domains(doms, i); | |
8968 | return NULL; | |
8969 | } | |
8970 | } | |
8971 | return doms; | |
8972 | } | |
8973 | ||
8974 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
8975 | { | |
8976 | unsigned int i; | |
8977 | for (i = 0; i < ndoms; i++) | |
8978 | free_cpumask_var(doms[i]); | |
8979 | kfree(doms); | |
8980 | } | |
8981 | ||
1a20ff27 | 8982 | /* |
41a2d6cf | 8983 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8984 | * For now this just excludes isolated cpus, but could be used to |
8985 | * exclude other special cases in the future. | |
1a20ff27 | 8986 | */ |
96f874e2 | 8987 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8988 | { |
7378547f MM |
8989 | int err; |
8990 | ||
22e52b07 | 8991 | arch_update_cpu_topology(); |
029190c5 | 8992 | ndoms_cur = 1; |
acc3f5d7 | 8993 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 8994 | if (!doms_cur) |
acc3f5d7 RR |
8995 | doms_cur = &fallback_doms; |
8996 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 8997 | dattr_cur = NULL; |
acc3f5d7 | 8998 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 8999 | register_sched_domain_sysctl(); |
7378547f MM |
9000 | |
9001 | return err; | |
1a20ff27 DG |
9002 | } |
9003 | ||
96f874e2 RR |
9004 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
9005 | struct cpumask *tmpmask) | |
1da177e4 | 9006 | { |
7c16ec58 | 9007 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 9008 | } |
1da177e4 | 9009 | |
1a20ff27 DG |
9010 | /* |
9011 | * Detach sched domains from a group of cpus specified in cpu_map | |
9012 | * These cpus will now be attached to the NULL domain | |
9013 | */ | |
96f874e2 | 9014 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9015 | { |
96f874e2 RR |
9016 | /* Save because hotplug lock held. */ |
9017 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
9018 | int i; |
9019 | ||
abcd083a | 9020 | for_each_cpu(i, cpu_map) |
57d885fe | 9021 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 9022 | synchronize_sched(); |
96f874e2 | 9023 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
9024 | } |
9025 | ||
1d3504fc HS |
9026 | /* handle null as "default" */ |
9027 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
9028 | struct sched_domain_attr *new, int idx_new) | |
9029 | { | |
9030 | struct sched_domain_attr tmp; | |
9031 | ||
9032 | /* fast path */ | |
9033 | if (!new && !cur) | |
9034 | return 1; | |
9035 | ||
9036 | tmp = SD_ATTR_INIT; | |
9037 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
9038 | new ? (new + idx_new) : &tmp, | |
9039 | sizeof(struct sched_domain_attr)); | |
9040 | } | |
9041 | ||
029190c5 PJ |
9042 | /* |
9043 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 9044 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
9045 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
9046 | * It destroys each deleted domain and builds each new domain. | |
9047 | * | |
acc3f5d7 | 9048 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
9049 | * The masks don't intersect (don't overlap.) We should setup one |
9050 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
9051 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
9052 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
9053 | * it as it is. | |
9054 | * | |
acc3f5d7 RR |
9055 | * The passed in 'doms_new' should be allocated using |
9056 | * alloc_sched_domains. This routine takes ownership of it and will | |
9057 | * free_sched_domains it when done with it. If the caller failed the | |
9058 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
9059 | * and partition_sched_domains() will fallback to the single partition | |
9060 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 9061 | * |
96f874e2 | 9062 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
9063 | * ndoms_new == 0 is a special case for destroying existing domains, |
9064 | * and it will not create the default domain. | |
dfb512ec | 9065 | * |
029190c5 PJ |
9066 | * Call with hotplug lock held |
9067 | */ | |
acc3f5d7 | 9068 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 9069 | struct sched_domain_attr *dattr_new) |
029190c5 | 9070 | { |
dfb512ec | 9071 | int i, j, n; |
d65bd5ec | 9072 | int new_topology; |
029190c5 | 9073 | |
712555ee | 9074 | mutex_lock(&sched_domains_mutex); |
a1835615 | 9075 | |
7378547f MM |
9076 | /* always unregister in case we don't destroy any domains */ |
9077 | unregister_sched_domain_sysctl(); | |
9078 | ||
d65bd5ec HC |
9079 | /* Let architecture update cpu core mappings. */ |
9080 | new_topology = arch_update_cpu_topology(); | |
9081 | ||
dfb512ec | 9082 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
9083 | |
9084 | /* Destroy deleted domains */ | |
9085 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 9086 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 9087 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 9088 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9089 | goto match1; |
9090 | } | |
9091 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 9092 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
9093 | match1: |
9094 | ; | |
9095 | } | |
9096 | ||
e761b772 MK |
9097 | if (doms_new == NULL) { |
9098 | ndoms_cur = 0; | |
acc3f5d7 | 9099 | doms_new = &fallback_doms; |
6ad4c188 | 9100 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 9101 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9102 | } |
9103 | ||
029190c5 PJ |
9104 | /* Build new domains */ |
9105 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9106 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 9107 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 9108 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9109 | goto match2; |
9110 | } | |
9111 | /* no match - add a new doms_new */ | |
acc3f5d7 | 9112 | __build_sched_domains(doms_new[i], |
1d3504fc | 9113 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
9114 | match2: |
9115 | ; | |
9116 | } | |
9117 | ||
9118 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
9119 | if (doms_cur != &fallback_doms) |
9120 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 9121 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9122 | doms_cur = doms_new; |
1d3504fc | 9123 | dattr_cur = dattr_new; |
029190c5 | 9124 | ndoms_cur = ndoms_new; |
7378547f MM |
9125 | |
9126 | register_sched_domain_sysctl(); | |
a1835615 | 9127 | |
712555ee | 9128 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9129 | } |
9130 | ||
5c45bf27 | 9131 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9132 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9133 | { |
95402b38 | 9134 | get_online_cpus(); |
dfb512ec MK |
9135 | |
9136 | /* Destroy domains first to force the rebuild */ | |
9137 | partition_sched_domains(0, NULL, NULL); | |
9138 | ||
e761b772 | 9139 | rebuild_sched_domains(); |
95402b38 | 9140 | put_online_cpus(); |
5c45bf27 SS |
9141 | } |
9142 | ||
9143 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9144 | { | |
afb8a9b7 | 9145 | unsigned int level = 0; |
5c45bf27 | 9146 | |
afb8a9b7 GS |
9147 | if (sscanf(buf, "%u", &level) != 1) |
9148 | return -EINVAL; | |
9149 | ||
9150 | /* | |
9151 | * level is always be positive so don't check for | |
9152 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9153 | * What happens on 0 or 1 byte write, | |
9154 | * need to check for count as well? | |
9155 | */ | |
9156 | ||
9157 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9158 | return -EINVAL; |
9159 | ||
9160 | if (smt) | |
afb8a9b7 | 9161 | sched_smt_power_savings = level; |
5c45bf27 | 9162 | else |
afb8a9b7 | 9163 | sched_mc_power_savings = level; |
5c45bf27 | 9164 | |
c70f22d2 | 9165 | arch_reinit_sched_domains(); |
5c45bf27 | 9166 | |
c70f22d2 | 9167 | return count; |
5c45bf27 SS |
9168 | } |
9169 | ||
5c45bf27 | 9170 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9171 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9172 | char *page) | |
5c45bf27 SS |
9173 | { |
9174 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9175 | } | |
f718cd4a | 9176 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9177 | const char *buf, size_t count) |
5c45bf27 SS |
9178 | { |
9179 | return sched_power_savings_store(buf, count, 0); | |
9180 | } | |
f718cd4a AK |
9181 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9182 | sched_mc_power_savings_show, | |
9183 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9184 | #endif |
9185 | ||
9186 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9187 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9188 | char *page) | |
5c45bf27 SS |
9189 | { |
9190 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9191 | } | |
f718cd4a | 9192 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9193 | const char *buf, size_t count) |
5c45bf27 SS |
9194 | { |
9195 | return sched_power_savings_store(buf, count, 1); | |
9196 | } | |
f718cd4a AK |
9197 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9198 | sched_smt_power_savings_show, | |
6707de00 AB |
9199 | sched_smt_power_savings_store); |
9200 | #endif | |
9201 | ||
39aac648 | 9202 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9203 | { |
9204 | int err = 0; | |
9205 | ||
9206 | #ifdef CONFIG_SCHED_SMT | |
9207 | if (smt_capable()) | |
9208 | err = sysfs_create_file(&cls->kset.kobj, | |
9209 | &attr_sched_smt_power_savings.attr); | |
9210 | #endif | |
9211 | #ifdef CONFIG_SCHED_MC | |
9212 | if (!err && mc_capable()) | |
9213 | err = sysfs_create_file(&cls->kset.kobj, | |
9214 | &attr_sched_mc_power_savings.attr); | |
9215 | #endif | |
9216 | return err; | |
9217 | } | |
6d6bc0ad | 9218 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9219 | |
e761b772 | 9220 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9221 | /* |
e761b772 MK |
9222 | * Add online and remove offline CPUs from the scheduler domains. |
9223 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9224 | */ |
9225 | static int update_sched_domains(struct notifier_block *nfb, | |
9226 | unsigned long action, void *hcpu) | |
e761b772 MK |
9227 | { |
9228 | switch (action) { | |
9229 | case CPU_ONLINE: | |
9230 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
9231 | case CPU_DOWN_PREPARE: |
9232 | case CPU_DOWN_PREPARE_FROZEN: | |
9233 | case CPU_DOWN_FAILED: | |
9234 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 9235 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9236 | return NOTIFY_OK; |
9237 | ||
9238 | default: | |
9239 | return NOTIFY_DONE; | |
9240 | } | |
9241 | } | |
9242 | #endif | |
9243 | ||
9244 | static int update_runtime(struct notifier_block *nfb, | |
9245 | unsigned long action, void *hcpu) | |
1da177e4 | 9246 | { |
7def2be1 PZ |
9247 | int cpu = (int)(long)hcpu; |
9248 | ||
1da177e4 | 9249 | switch (action) { |
1da177e4 | 9250 | case CPU_DOWN_PREPARE: |
8bb78442 | 9251 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9252 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9253 | return NOTIFY_OK; |
9254 | ||
1da177e4 | 9255 | case CPU_DOWN_FAILED: |
8bb78442 | 9256 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9257 | case CPU_ONLINE: |
8bb78442 | 9258 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9259 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9260 | return NOTIFY_OK; |
9261 | ||
1da177e4 LT |
9262 | default: |
9263 | return NOTIFY_DONE; | |
9264 | } | |
1da177e4 | 9265 | } |
1da177e4 LT |
9266 | |
9267 | void __init sched_init_smp(void) | |
9268 | { | |
dcc30a35 RR |
9269 | cpumask_var_t non_isolated_cpus; |
9270 | ||
9271 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 9272 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 9273 | |
434d53b0 MT |
9274 | #if defined(CONFIG_NUMA) |
9275 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9276 | GFP_KERNEL); | |
9277 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9278 | #endif | |
95402b38 | 9279 | get_online_cpus(); |
712555ee | 9280 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 9281 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
9282 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
9283 | if (cpumask_empty(non_isolated_cpus)) | |
9284 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9285 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9286 | put_online_cpus(); |
e761b772 MK |
9287 | |
9288 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9289 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9290 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9291 | #endif |
9292 | ||
9293 | /* RT runtime code needs to handle some hotplug events */ | |
9294 | hotcpu_notifier(update_runtime, 0); | |
9295 | ||
b328ca18 | 9296 | init_hrtick(); |
5c1e1767 NP |
9297 | |
9298 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9299 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9300 | BUG(); |
19978ca6 | 9301 | sched_init_granularity(); |
dcc30a35 | 9302 | free_cpumask_var(non_isolated_cpus); |
4212823f | 9303 | |
0e3900e6 | 9304 | init_sched_rt_class(); |
1da177e4 LT |
9305 | } |
9306 | #else | |
9307 | void __init sched_init_smp(void) | |
9308 | { | |
19978ca6 | 9309 | sched_init_granularity(); |
1da177e4 LT |
9310 | } |
9311 | #endif /* CONFIG_SMP */ | |
9312 | ||
cd1bb94b AB |
9313 | const_debug unsigned int sysctl_timer_migration = 1; |
9314 | ||
1da177e4 LT |
9315 | int in_sched_functions(unsigned long addr) |
9316 | { | |
1da177e4 LT |
9317 | return in_lock_functions(addr) || |
9318 | (addr >= (unsigned long)__sched_text_start | |
9319 | && addr < (unsigned long)__sched_text_end); | |
9320 | } | |
9321 | ||
a9957449 | 9322 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9323 | { |
9324 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9325 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9326 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9327 | cfs_rq->rq = rq; | |
9328 | #endif | |
67e9fb2a | 9329 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9330 | } |
9331 | ||
fa85ae24 PZ |
9332 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9333 | { | |
9334 | struct rt_prio_array *array; | |
9335 | int i; | |
9336 | ||
9337 | array = &rt_rq->active; | |
9338 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9339 | INIT_LIST_HEAD(array->queue + i); | |
9340 | __clear_bit(i, array->bitmap); | |
9341 | } | |
9342 | /* delimiter for bitsearch: */ | |
9343 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9344 | ||
052f1dc7 | 9345 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9346 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9347 | #ifdef CONFIG_SMP |
e864c499 | 9348 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9349 | #endif |
48d5e258 | 9350 | #endif |
fa85ae24 PZ |
9351 | #ifdef CONFIG_SMP |
9352 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9353 | rt_rq->overloaded = 0; |
c20b08e3 | 9354 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9355 | #endif |
9356 | ||
9357 | rt_rq->rt_time = 0; | |
9358 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9359 | rt_rq->rt_runtime = 0; |
9360 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9361 | |
052f1dc7 | 9362 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9363 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9364 | rt_rq->rq = rq; |
9365 | #endif | |
fa85ae24 PZ |
9366 | } |
9367 | ||
6f505b16 | 9368 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9369 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9370 | struct sched_entity *se, int cpu, int add, | |
9371 | struct sched_entity *parent) | |
6f505b16 | 9372 | { |
ec7dc8ac | 9373 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9374 | tg->cfs_rq[cpu] = cfs_rq; |
9375 | init_cfs_rq(cfs_rq, rq); | |
9376 | cfs_rq->tg = tg; | |
9377 | if (add) | |
9378 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9379 | ||
9380 | tg->se[cpu] = se; | |
354d60c2 DG |
9381 | /* se could be NULL for init_task_group */ |
9382 | if (!se) | |
9383 | return; | |
9384 | ||
ec7dc8ac DG |
9385 | if (!parent) |
9386 | se->cfs_rq = &rq->cfs; | |
9387 | else | |
9388 | se->cfs_rq = parent->my_q; | |
9389 | ||
6f505b16 PZ |
9390 | se->my_q = cfs_rq; |
9391 | se->load.weight = tg->shares; | |
e05510d0 | 9392 | se->load.inv_weight = 0; |
ec7dc8ac | 9393 | se->parent = parent; |
6f505b16 | 9394 | } |
052f1dc7 | 9395 | #endif |
6f505b16 | 9396 | |
052f1dc7 | 9397 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9398 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9399 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9400 | struct sched_rt_entity *parent) | |
6f505b16 | 9401 | { |
ec7dc8ac DG |
9402 | struct rq *rq = cpu_rq(cpu); |
9403 | ||
6f505b16 PZ |
9404 | tg->rt_rq[cpu] = rt_rq; |
9405 | init_rt_rq(rt_rq, rq); | |
9406 | rt_rq->tg = tg; | |
9407 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9408 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9409 | if (add) |
9410 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9411 | ||
9412 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9413 | if (!rt_se) |
9414 | return; | |
9415 | ||
ec7dc8ac DG |
9416 | if (!parent) |
9417 | rt_se->rt_rq = &rq->rt; | |
9418 | else | |
9419 | rt_se->rt_rq = parent->my_q; | |
9420 | ||
6f505b16 | 9421 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9422 | rt_se->parent = parent; |
6f505b16 PZ |
9423 | INIT_LIST_HEAD(&rt_se->run_list); |
9424 | } | |
9425 | #endif | |
9426 | ||
1da177e4 LT |
9427 | void __init sched_init(void) |
9428 | { | |
dd41f596 | 9429 | int i, j; |
434d53b0 MT |
9430 | unsigned long alloc_size = 0, ptr; |
9431 | ||
9432 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9433 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9434 | #endif | |
9435 | #ifdef CONFIG_RT_GROUP_SCHED | |
9436 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9437 | #endif |
9438 | #ifdef CONFIG_USER_SCHED | |
9439 | alloc_size *= 2; | |
df7c8e84 RR |
9440 | #endif |
9441 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9442 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 9443 | #endif |
434d53b0 | 9444 | if (alloc_size) { |
36b7b6d4 | 9445 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9446 | |
9447 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9448 | init_task_group.se = (struct sched_entity **)ptr; | |
9449 | ptr += nr_cpu_ids * sizeof(void **); | |
9450 | ||
9451 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9452 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9453 | |
9454 | #ifdef CONFIG_USER_SCHED | |
9455 | root_task_group.se = (struct sched_entity **)ptr; | |
9456 | ptr += nr_cpu_ids * sizeof(void **); | |
9457 | ||
9458 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9459 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9460 | #endif /* CONFIG_USER_SCHED */ |
9461 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9462 | #ifdef CONFIG_RT_GROUP_SCHED |
9463 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9464 | ptr += nr_cpu_ids * sizeof(void **); | |
9465 | ||
9466 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9467 | ptr += nr_cpu_ids * sizeof(void **); |
9468 | ||
9469 | #ifdef CONFIG_USER_SCHED | |
9470 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9471 | ptr += nr_cpu_ids * sizeof(void **); | |
9472 | ||
9473 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9474 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9475 | #endif /* CONFIG_USER_SCHED */ |
9476 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9477 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9478 | for_each_possible_cpu(i) { | |
9479 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9480 | ptr += cpumask_size(); | |
9481 | } | |
9482 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9483 | } |
dd41f596 | 9484 | |
57d885fe GH |
9485 | #ifdef CONFIG_SMP |
9486 | init_defrootdomain(); | |
9487 | #endif | |
9488 | ||
d0b27fa7 PZ |
9489 | init_rt_bandwidth(&def_rt_bandwidth, |
9490 | global_rt_period(), global_rt_runtime()); | |
9491 | ||
9492 | #ifdef CONFIG_RT_GROUP_SCHED | |
9493 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9494 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9495 | #ifdef CONFIG_USER_SCHED |
9496 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9497 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9498 | #endif /* CONFIG_USER_SCHED */ |
9499 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9500 | |
052f1dc7 | 9501 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9502 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9503 | INIT_LIST_HEAD(&init_task_group.children); |
9504 | ||
9505 | #ifdef CONFIG_USER_SCHED | |
9506 | INIT_LIST_HEAD(&root_task_group.children); | |
9507 | init_task_group.parent = &root_task_group; | |
9508 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9509 | #endif /* CONFIG_USER_SCHED */ |
9510 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9511 | |
4a6cc4bd JK |
9512 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
9513 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
9514 | __alignof__(unsigned long)); | |
9515 | #endif | |
0a945022 | 9516 | for_each_possible_cpu(i) { |
70b97a7f | 9517 | struct rq *rq; |
1da177e4 LT |
9518 | |
9519 | rq = cpu_rq(i); | |
9520 | spin_lock_init(&rq->lock); | |
7897986b | 9521 | rq->nr_running = 0; |
dce48a84 TG |
9522 | rq->calc_load_active = 0; |
9523 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9524 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9525 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9526 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9527 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9528 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9529 | #ifdef CONFIG_CGROUP_SCHED |
9530 | /* | |
9531 | * How much cpu bandwidth does init_task_group get? | |
9532 | * | |
9533 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9534 | * gets 100% of the cpu resources in the system. This overall | |
9535 | * system cpu resource is divided among the tasks of | |
9536 | * init_task_group and its child task-groups in a fair manner, | |
9537 | * based on each entity's (task or task-group's) weight | |
9538 | * (se->load.weight). | |
9539 | * | |
9540 | * In other words, if init_task_group has 10 tasks of weight | |
9541 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9542 | * then A0's share of the cpu resource is: | |
9543 | * | |
0d905bca | 9544 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9545 | * |
9546 | * We achieve this by letting init_task_group's tasks sit | |
9547 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9548 | */ | |
ec7dc8ac | 9549 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9550 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9551 | root_task_group.shares = NICE_0_LOAD; |
9552 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9553 | /* |
9554 | * In case of task-groups formed thr' the user id of tasks, | |
9555 | * init_task_group represents tasks belonging to root user. | |
9556 | * Hence it forms a sibling of all subsequent groups formed. | |
9557 | * In this case, init_task_group gets only a fraction of overall | |
9558 | * system cpu resource, based on the weight assigned to root | |
9559 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9560 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9561 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9562 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9563 | */ | |
ec7dc8ac | 9564 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9565 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9566 | &per_cpu(init_sched_entity, i), i, 1, |
9567 | root_task_group.se[i]); | |
6f505b16 | 9568 | |
052f1dc7 | 9569 | #endif |
354d60c2 DG |
9570 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9571 | ||
9572 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9573 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9574 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9575 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9576 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9577 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9578 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9579 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9580 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9581 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9582 | root_task_group.rt_se[i]); | |
354d60c2 | 9583 | #endif |
dd41f596 | 9584 | #endif |
1da177e4 | 9585 | |
dd41f596 IM |
9586 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9587 | rq->cpu_load[j] = 0; | |
1da177e4 | 9588 | #ifdef CONFIG_SMP |
41c7ce9a | 9589 | rq->sd = NULL; |
57d885fe | 9590 | rq->rd = NULL; |
3f029d3c | 9591 | rq->post_schedule = 0; |
1da177e4 | 9592 | rq->active_balance = 0; |
dd41f596 | 9593 | rq->next_balance = jiffies; |
1da177e4 | 9594 | rq->push_cpu = 0; |
0a2966b4 | 9595 | rq->cpu = i; |
1f11eb6a | 9596 | rq->online = 0; |
1da177e4 | 9597 | rq->migration_thread = NULL; |
eae0c9df MG |
9598 | rq->idle_stamp = 0; |
9599 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 9600 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 9601 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9602 | #endif |
8f4d37ec | 9603 | init_rq_hrtick(rq); |
1da177e4 | 9604 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9605 | } |
9606 | ||
2dd73a4f | 9607 | set_load_weight(&init_task); |
b50f60ce | 9608 | |
e107be36 AK |
9609 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9610 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9611 | #endif | |
9612 | ||
c9819f45 | 9613 | #ifdef CONFIG_SMP |
962cf36c | 9614 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9615 | #endif |
9616 | ||
b50f60ce HC |
9617 | #ifdef CONFIG_RT_MUTEXES |
9618 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9619 | #endif | |
9620 | ||
1da177e4 LT |
9621 | /* |
9622 | * The boot idle thread does lazy MMU switching as well: | |
9623 | */ | |
9624 | atomic_inc(&init_mm.mm_count); | |
9625 | enter_lazy_tlb(&init_mm, current); | |
9626 | ||
9627 | /* | |
9628 | * Make us the idle thread. Technically, schedule() should not be | |
9629 | * called from this thread, however somewhere below it might be, | |
9630 | * but because we are the idle thread, we just pick up running again | |
9631 | * when this runqueue becomes "idle". | |
9632 | */ | |
9633 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9634 | |
9635 | calc_load_update = jiffies + LOAD_FREQ; | |
9636 | ||
dd41f596 IM |
9637 | /* |
9638 | * During early bootup we pretend to be a normal task: | |
9639 | */ | |
9640 | current->sched_class = &fair_sched_class; | |
6892b75e | 9641 | |
6a7b3dc3 | 9642 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 9643 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9644 | #ifdef CONFIG_SMP |
7d1e6a9b | 9645 | #ifdef CONFIG_NO_HZ |
49557e62 | 9646 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 9647 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 9648 | #endif |
bdddd296 RR |
9649 | /* May be allocated at isolcpus cmdline parse time */ |
9650 | if (cpu_isolated_map == NULL) | |
9651 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 9652 | #endif /* SMP */ |
6a7b3dc3 | 9653 | |
cdd6c482 | 9654 | perf_event_init(); |
0d905bca | 9655 | |
6892b75e | 9656 | scheduler_running = 1; |
1da177e4 LT |
9657 | } |
9658 | ||
9659 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9660 | static inline int preempt_count_equals(int preempt_offset) |
9661 | { | |
9662 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9663 | ||
9664 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9665 | } | |
9666 | ||
9667 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9668 | { |
48f24c4d | 9669 | #ifdef in_atomic |
1da177e4 LT |
9670 | static unsigned long prev_jiffy; /* ratelimiting */ |
9671 | ||
e4aafea2 FW |
9672 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9673 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9674 | return; |
9675 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9676 | return; | |
9677 | prev_jiffy = jiffies; | |
9678 | ||
663997d4 JP |
9679 | pr_err("BUG: sleeping function called from invalid context at %s:%d\n", |
9680 | file, line); | |
9681 | pr_err("in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9682 | in_atomic(), irqs_disabled(), | |
9683 | current->pid, current->comm); | |
aef745fc IM |
9684 | |
9685 | debug_show_held_locks(current); | |
9686 | if (irqs_disabled()) | |
9687 | print_irqtrace_events(current); | |
9688 | dump_stack(); | |
1da177e4 LT |
9689 | #endif |
9690 | } | |
9691 | EXPORT_SYMBOL(__might_sleep); | |
9692 | #endif | |
9693 | ||
9694 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9695 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9696 | { | |
9697 | int on_rq; | |
3e51f33f | 9698 | |
3a5e4dc1 AK |
9699 | update_rq_clock(rq); |
9700 | on_rq = p->se.on_rq; | |
9701 | if (on_rq) | |
9702 | deactivate_task(rq, p, 0); | |
9703 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9704 | if (on_rq) { | |
9705 | activate_task(rq, p, 0); | |
9706 | resched_task(rq->curr); | |
9707 | } | |
9708 | } | |
9709 | ||
1da177e4 LT |
9710 | void normalize_rt_tasks(void) |
9711 | { | |
a0f98a1c | 9712 | struct task_struct *g, *p; |
1da177e4 | 9713 | unsigned long flags; |
70b97a7f | 9714 | struct rq *rq; |
1da177e4 | 9715 | |
4cf5d77a | 9716 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9717 | do_each_thread(g, p) { |
178be793 IM |
9718 | /* |
9719 | * Only normalize user tasks: | |
9720 | */ | |
9721 | if (!p->mm) | |
9722 | continue; | |
9723 | ||
6cfb0d5d | 9724 | p->se.exec_start = 0; |
6cfb0d5d | 9725 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9726 | p->se.wait_start = 0; |
dd41f596 | 9727 | p->se.sleep_start = 0; |
dd41f596 | 9728 | p->se.block_start = 0; |
6cfb0d5d | 9729 | #endif |
dd41f596 IM |
9730 | |
9731 | if (!rt_task(p)) { | |
9732 | /* | |
9733 | * Renice negative nice level userspace | |
9734 | * tasks back to 0: | |
9735 | */ | |
9736 | if (TASK_NICE(p) < 0 && p->mm) | |
9737 | set_user_nice(p, 0); | |
1da177e4 | 9738 | continue; |
dd41f596 | 9739 | } |
1da177e4 | 9740 | |
4cf5d77a | 9741 | spin_lock(&p->pi_lock); |
b29739f9 | 9742 | rq = __task_rq_lock(p); |
1da177e4 | 9743 | |
178be793 | 9744 | normalize_task(rq, p); |
3a5e4dc1 | 9745 | |
b29739f9 | 9746 | __task_rq_unlock(rq); |
4cf5d77a | 9747 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9748 | } while_each_thread(g, p); |
9749 | ||
4cf5d77a | 9750 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9751 | } |
9752 | ||
9753 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9754 | |
9755 | #ifdef CONFIG_IA64 | |
9756 | /* | |
9757 | * These functions are only useful for the IA64 MCA handling. | |
9758 | * | |
9759 | * They can only be called when the whole system has been | |
9760 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9761 | * activity can take place. Using them for anything else would | |
9762 | * be a serious bug, and as a result, they aren't even visible | |
9763 | * under any other configuration. | |
9764 | */ | |
9765 | ||
9766 | /** | |
9767 | * curr_task - return the current task for a given cpu. | |
9768 | * @cpu: the processor in question. | |
9769 | * | |
9770 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9771 | */ | |
36c8b586 | 9772 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9773 | { |
9774 | return cpu_curr(cpu); | |
9775 | } | |
9776 | ||
9777 | /** | |
9778 | * set_curr_task - set the current task for a given cpu. | |
9779 | * @cpu: the processor in question. | |
9780 | * @p: the task pointer to set. | |
9781 | * | |
9782 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9783 | * are serviced on a separate stack. It allows the architecture to switch the |
9784 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9785 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9786 | * and caller must save the original value of the current task (see | |
9787 | * curr_task() above) and restore that value before reenabling interrupts and | |
9788 | * re-starting the system. | |
9789 | * | |
9790 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9791 | */ | |
36c8b586 | 9792 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9793 | { |
9794 | cpu_curr(cpu) = p; | |
9795 | } | |
9796 | ||
9797 | #endif | |
29f59db3 | 9798 | |
bccbe08a PZ |
9799 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9800 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9801 | { |
9802 | int i; | |
9803 | ||
9804 | for_each_possible_cpu(i) { | |
9805 | if (tg->cfs_rq) | |
9806 | kfree(tg->cfs_rq[i]); | |
9807 | if (tg->se) | |
9808 | kfree(tg->se[i]); | |
6f505b16 PZ |
9809 | } |
9810 | ||
9811 | kfree(tg->cfs_rq); | |
9812 | kfree(tg->se); | |
6f505b16 PZ |
9813 | } |
9814 | ||
ec7dc8ac DG |
9815 | static |
9816 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9817 | { |
29f59db3 | 9818 | struct cfs_rq *cfs_rq; |
eab17229 | 9819 | struct sched_entity *se; |
9b5b7751 | 9820 | struct rq *rq; |
29f59db3 SV |
9821 | int i; |
9822 | ||
434d53b0 | 9823 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9824 | if (!tg->cfs_rq) |
9825 | goto err; | |
434d53b0 | 9826 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9827 | if (!tg->se) |
9828 | goto err; | |
052f1dc7 PZ |
9829 | |
9830 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9831 | |
9832 | for_each_possible_cpu(i) { | |
9b5b7751 | 9833 | rq = cpu_rq(i); |
29f59db3 | 9834 | |
eab17229 LZ |
9835 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9836 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9837 | if (!cfs_rq) |
9838 | goto err; | |
9839 | ||
eab17229 LZ |
9840 | se = kzalloc_node(sizeof(struct sched_entity), |
9841 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 9842 | if (!se) |
dfc12eb2 | 9843 | goto err_free_rq; |
29f59db3 | 9844 | |
eab17229 | 9845 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9846 | } |
9847 | ||
9848 | return 1; | |
9849 | ||
dfc12eb2 PC |
9850 | err_free_rq: |
9851 | kfree(cfs_rq); | |
bccbe08a PZ |
9852 | err: |
9853 | return 0; | |
9854 | } | |
9855 | ||
9856 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9857 | { | |
9858 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9859 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9860 | } | |
9861 | ||
9862 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9863 | { | |
9864 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9865 | } | |
6d6bc0ad | 9866 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9867 | static inline void free_fair_sched_group(struct task_group *tg) |
9868 | { | |
9869 | } | |
9870 | ||
ec7dc8ac DG |
9871 | static inline |
9872 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9873 | { |
9874 | return 1; | |
9875 | } | |
9876 | ||
9877 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9878 | { | |
9879 | } | |
9880 | ||
9881 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9882 | { | |
9883 | } | |
6d6bc0ad | 9884 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9885 | |
9886 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9887 | static void free_rt_sched_group(struct task_group *tg) |
9888 | { | |
9889 | int i; | |
9890 | ||
d0b27fa7 PZ |
9891 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9892 | ||
bccbe08a PZ |
9893 | for_each_possible_cpu(i) { |
9894 | if (tg->rt_rq) | |
9895 | kfree(tg->rt_rq[i]); | |
9896 | if (tg->rt_se) | |
9897 | kfree(tg->rt_se[i]); | |
9898 | } | |
9899 | ||
9900 | kfree(tg->rt_rq); | |
9901 | kfree(tg->rt_se); | |
9902 | } | |
9903 | ||
ec7dc8ac DG |
9904 | static |
9905 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9906 | { |
9907 | struct rt_rq *rt_rq; | |
eab17229 | 9908 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9909 | struct rq *rq; |
9910 | int i; | |
9911 | ||
434d53b0 | 9912 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9913 | if (!tg->rt_rq) |
9914 | goto err; | |
434d53b0 | 9915 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9916 | if (!tg->rt_se) |
9917 | goto err; | |
9918 | ||
d0b27fa7 PZ |
9919 | init_rt_bandwidth(&tg->rt_bandwidth, |
9920 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9921 | |
9922 | for_each_possible_cpu(i) { | |
9923 | rq = cpu_rq(i); | |
9924 | ||
eab17229 LZ |
9925 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9926 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9927 | if (!rt_rq) |
9928 | goto err; | |
29f59db3 | 9929 | |
eab17229 LZ |
9930 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9931 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 9932 | if (!rt_se) |
dfc12eb2 | 9933 | goto err_free_rq; |
29f59db3 | 9934 | |
eab17229 | 9935 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9936 | } |
9937 | ||
bccbe08a PZ |
9938 | return 1; |
9939 | ||
dfc12eb2 PC |
9940 | err_free_rq: |
9941 | kfree(rt_rq); | |
bccbe08a PZ |
9942 | err: |
9943 | return 0; | |
9944 | } | |
9945 | ||
9946 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9947 | { | |
9948 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9949 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9950 | } | |
9951 | ||
9952 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9953 | { | |
9954 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9955 | } | |
6d6bc0ad | 9956 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9957 | static inline void free_rt_sched_group(struct task_group *tg) |
9958 | { | |
9959 | } | |
9960 | ||
ec7dc8ac DG |
9961 | static inline |
9962 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9963 | { |
9964 | return 1; | |
9965 | } | |
9966 | ||
9967 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9968 | { | |
9969 | } | |
9970 | ||
9971 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9972 | { | |
9973 | } | |
6d6bc0ad | 9974 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9975 | |
d0b27fa7 | 9976 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9977 | static void free_sched_group(struct task_group *tg) |
9978 | { | |
9979 | free_fair_sched_group(tg); | |
9980 | free_rt_sched_group(tg); | |
9981 | kfree(tg); | |
9982 | } | |
9983 | ||
9984 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9985 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9986 | { |
9987 | struct task_group *tg; | |
9988 | unsigned long flags; | |
9989 | int i; | |
9990 | ||
9991 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9992 | if (!tg) | |
9993 | return ERR_PTR(-ENOMEM); | |
9994 | ||
ec7dc8ac | 9995 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9996 | goto err; |
9997 | ||
ec7dc8ac | 9998 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9999 | goto err; |
10000 | ||
8ed36996 | 10001 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10002 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10003 | register_fair_sched_group(tg, i); |
10004 | register_rt_sched_group(tg, i); | |
9b5b7751 | 10005 | } |
6f505b16 | 10006 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
10007 | |
10008 | WARN_ON(!parent); /* root should already exist */ | |
10009 | ||
10010 | tg->parent = parent; | |
f473aa5e | 10011 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10012 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10013 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 10014 | |
9b5b7751 | 10015 | return tg; |
29f59db3 SV |
10016 | |
10017 | err: | |
6f505b16 | 10018 | free_sched_group(tg); |
29f59db3 SV |
10019 | return ERR_PTR(-ENOMEM); |
10020 | } | |
10021 | ||
9b5b7751 | 10022 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 10023 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10024 | { |
29f59db3 | 10025 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 10026 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10027 | } |
10028 | ||
9b5b7751 | 10029 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 10030 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 10031 | { |
8ed36996 | 10032 | unsigned long flags; |
9b5b7751 | 10033 | int i; |
29f59db3 | 10034 | |
8ed36996 | 10035 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10036 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10037 | unregister_fair_sched_group(tg, i); |
10038 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 10039 | } |
6f505b16 | 10040 | list_del_rcu(&tg->list); |
f473aa5e | 10041 | list_del_rcu(&tg->siblings); |
8ed36996 | 10042 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 10043 | |
9b5b7751 | 10044 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 10045 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
10046 | } |
10047 | ||
9b5b7751 | 10048 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
10049 | * The caller of this function should have put the task in its new group |
10050 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
10051 | * reflect its new group. | |
9b5b7751 SV |
10052 | */ |
10053 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
10054 | { |
10055 | int on_rq, running; | |
10056 | unsigned long flags; | |
10057 | struct rq *rq; | |
10058 | ||
10059 | rq = task_rq_lock(tsk, &flags); | |
10060 | ||
29f59db3 SV |
10061 | update_rq_clock(rq); |
10062 | ||
051a1d1a | 10063 | running = task_current(rq, tsk); |
29f59db3 SV |
10064 | on_rq = tsk->se.on_rq; |
10065 | ||
0e1f3483 | 10066 | if (on_rq) |
29f59db3 | 10067 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
10068 | if (unlikely(running)) |
10069 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 10070 | |
6f505b16 | 10071 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 10072 | |
810b3817 PZ |
10073 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10074 | if (tsk->sched_class->moved_group) | |
10075 | tsk->sched_class->moved_group(tsk); | |
10076 | #endif | |
10077 | ||
0e1f3483 HS |
10078 | if (unlikely(running)) |
10079 | tsk->sched_class->set_curr_task(rq); | |
10080 | if (on_rq) | |
7074badb | 10081 | enqueue_task(rq, tsk, 0); |
29f59db3 | 10082 | |
29f59db3 SV |
10083 | task_rq_unlock(rq, &flags); |
10084 | } | |
6d6bc0ad | 10085 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 10086 | |
052f1dc7 | 10087 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 10088 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
10089 | { |
10090 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
10091 | int on_rq; |
10092 | ||
29f59db3 | 10093 | on_rq = se->on_rq; |
62fb1851 | 10094 | if (on_rq) |
29f59db3 SV |
10095 | dequeue_entity(cfs_rq, se, 0); |
10096 | ||
10097 | se->load.weight = shares; | |
e05510d0 | 10098 | se->load.inv_weight = 0; |
29f59db3 | 10099 | |
62fb1851 | 10100 | if (on_rq) |
29f59db3 | 10101 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10102 | } |
62fb1851 | 10103 | |
c09595f6 PZ |
10104 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10105 | { | |
10106 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10107 | struct rq *rq = cfs_rq->rq; | |
10108 | unsigned long flags; | |
10109 | ||
10110 | spin_lock_irqsave(&rq->lock, flags); | |
10111 | __set_se_shares(se, shares); | |
10112 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
10113 | } |
10114 | ||
8ed36996 PZ |
10115 | static DEFINE_MUTEX(shares_mutex); |
10116 | ||
4cf86d77 | 10117 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10118 | { |
10119 | int i; | |
8ed36996 | 10120 | unsigned long flags; |
c61935fd | 10121 | |
ec7dc8ac DG |
10122 | /* |
10123 | * We can't change the weight of the root cgroup. | |
10124 | */ | |
10125 | if (!tg->se[0]) | |
10126 | return -EINVAL; | |
10127 | ||
18d95a28 PZ |
10128 | if (shares < MIN_SHARES) |
10129 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10130 | else if (shares > MAX_SHARES) |
10131 | shares = MAX_SHARES; | |
62fb1851 | 10132 | |
8ed36996 | 10133 | mutex_lock(&shares_mutex); |
9b5b7751 | 10134 | if (tg->shares == shares) |
5cb350ba | 10135 | goto done; |
29f59db3 | 10136 | |
8ed36996 | 10137 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10138 | for_each_possible_cpu(i) |
10139 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10140 | list_del_rcu(&tg->siblings); |
8ed36996 | 10141 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10142 | |
10143 | /* wait for any ongoing reference to this group to finish */ | |
10144 | synchronize_sched(); | |
10145 | ||
10146 | /* | |
10147 | * Now we are free to modify the group's share on each cpu | |
10148 | * w/o tripping rebalance_share or load_balance_fair. | |
10149 | */ | |
9b5b7751 | 10150 | tg->shares = shares; |
c09595f6 PZ |
10151 | for_each_possible_cpu(i) { |
10152 | /* | |
10153 | * force a rebalance | |
10154 | */ | |
10155 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10156 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10157 | } |
29f59db3 | 10158 | |
6b2d7700 SV |
10159 | /* |
10160 | * Enable load balance activity on this group, by inserting it back on | |
10161 | * each cpu's rq->leaf_cfs_rq_list. | |
10162 | */ | |
8ed36996 | 10163 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10164 | for_each_possible_cpu(i) |
10165 | register_fair_sched_group(tg, i); | |
f473aa5e | 10166 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10167 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10168 | done: |
8ed36996 | 10169 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10170 | return 0; |
29f59db3 SV |
10171 | } |
10172 | ||
5cb350ba DG |
10173 | unsigned long sched_group_shares(struct task_group *tg) |
10174 | { | |
10175 | return tg->shares; | |
10176 | } | |
052f1dc7 | 10177 | #endif |
5cb350ba | 10178 | |
052f1dc7 | 10179 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10180 | /* |
9f0c1e56 | 10181 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10182 | */ |
9f0c1e56 PZ |
10183 | static DEFINE_MUTEX(rt_constraints_mutex); |
10184 | ||
10185 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10186 | { | |
10187 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10188 | return 1ULL << 20; |
9f0c1e56 | 10189 | |
9a7e0b18 | 10190 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10191 | } |
10192 | ||
9a7e0b18 PZ |
10193 | /* Must be called with tasklist_lock held */ |
10194 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10195 | { |
9a7e0b18 | 10196 | struct task_struct *g, *p; |
b40b2e8e | 10197 | |
9a7e0b18 PZ |
10198 | do_each_thread(g, p) { |
10199 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10200 | return 1; | |
10201 | } while_each_thread(g, p); | |
b40b2e8e | 10202 | |
9a7e0b18 PZ |
10203 | return 0; |
10204 | } | |
b40b2e8e | 10205 | |
9a7e0b18 PZ |
10206 | struct rt_schedulable_data { |
10207 | struct task_group *tg; | |
10208 | u64 rt_period; | |
10209 | u64 rt_runtime; | |
10210 | }; | |
b40b2e8e | 10211 | |
9a7e0b18 PZ |
10212 | static int tg_schedulable(struct task_group *tg, void *data) |
10213 | { | |
10214 | struct rt_schedulable_data *d = data; | |
10215 | struct task_group *child; | |
10216 | unsigned long total, sum = 0; | |
10217 | u64 period, runtime; | |
b40b2e8e | 10218 | |
9a7e0b18 PZ |
10219 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10220 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10221 | |
9a7e0b18 PZ |
10222 | if (tg == d->tg) { |
10223 | period = d->rt_period; | |
10224 | runtime = d->rt_runtime; | |
b40b2e8e | 10225 | } |
b40b2e8e | 10226 | |
98a4826b PZ |
10227 | #ifdef CONFIG_USER_SCHED |
10228 | if (tg == &root_task_group) { | |
10229 | period = global_rt_period(); | |
10230 | runtime = global_rt_runtime(); | |
10231 | } | |
10232 | #endif | |
10233 | ||
4653f803 PZ |
10234 | /* |
10235 | * Cannot have more runtime than the period. | |
10236 | */ | |
10237 | if (runtime > period && runtime != RUNTIME_INF) | |
10238 | return -EINVAL; | |
6f505b16 | 10239 | |
4653f803 PZ |
10240 | /* |
10241 | * Ensure we don't starve existing RT tasks. | |
10242 | */ | |
9a7e0b18 PZ |
10243 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10244 | return -EBUSY; | |
6f505b16 | 10245 | |
9a7e0b18 | 10246 | total = to_ratio(period, runtime); |
6f505b16 | 10247 | |
4653f803 PZ |
10248 | /* |
10249 | * Nobody can have more than the global setting allows. | |
10250 | */ | |
10251 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10252 | return -EINVAL; | |
6f505b16 | 10253 | |
4653f803 PZ |
10254 | /* |
10255 | * The sum of our children's runtime should not exceed our own. | |
10256 | */ | |
9a7e0b18 PZ |
10257 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10258 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10259 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10260 | |
9a7e0b18 PZ |
10261 | if (child == d->tg) { |
10262 | period = d->rt_period; | |
10263 | runtime = d->rt_runtime; | |
10264 | } | |
6f505b16 | 10265 | |
9a7e0b18 | 10266 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10267 | } |
6f505b16 | 10268 | |
9a7e0b18 PZ |
10269 | if (sum > total) |
10270 | return -EINVAL; | |
10271 | ||
10272 | return 0; | |
6f505b16 PZ |
10273 | } |
10274 | ||
9a7e0b18 | 10275 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10276 | { |
9a7e0b18 PZ |
10277 | struct rt_schedulable_data data = { |
10278 | .tg = tg, | |
10279 | .rt_period = period, | |
10280 | .rt_runtime = runtime, | |
10281 | }; | |
10282 | ||
10283 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10284 | } |
10285 | ||
d0b27fa7 PZ |
10286 | static int tg_set_bandwidth(struct task_group *tg, |
10287 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10288 | { |
ac086bc2 | 10289 | int i, err = 0; |
9f0c1e56 | 10290 | |
9f0c1e56 | 10291 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10292 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10293 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10294 | if (err) | |
9f0c1e56 | 10295 | goto unlock; |
ac086bc2 PZ |
10296 | |
10297 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10298 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10299 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10300 | |
10301 | for_each_possible_cpu(i) { | |
10302 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10303 | ||
10304 | spin_lock(&rt_rq->rt_runtime_lock); | |
10305 | rt_rq->rt_runtime = rt_runtime; | |
10306 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10307 | } | |
10308 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10309 | unlock: |
521f1a24 | 10310 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10311 | mutex_unlock(&rt_constraints_mutex); |
10312 | ||
10313 | return err; | |
6f505b16 PZ |
10314 | } |
10315 | ||
d0b27fa7 PZ |
10316 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10317 | { | |
10318 | u64 rt_runtime, rt_period; | |
10319 | ||
10320 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10321 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10322 | if (rt_runtime_us < 0) | |
10323 | rt_runtime = RUNTIME_INF; | |
10324 | ||
10325 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10326 | } | |
10327 | ||
9f0c1e56 PZ |
10328 | long sched_group_rt_runtime(struct task_group *tg) |
10329 | { | |
10330 | u64 rt_runtime_us; | |
10331 | ||
d0b27fa7 | 10332 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10333 | return -1; |
10334 | ||
d0b27fa7 | 10335 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10336 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10337 | return rt_runtime_us; | |
10338 | } | |
d0b27fa7 PZ |
10339 | |
10340 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10341 | { | |
10342 | u64 rt_runtime, rt_period; | |
10343 | ||
10344 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10345 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10346 | ||
619b0488 R |
10347 | if (rt_period == 0) |
10348 | return -EINVAL; | |
10349 | ||
d0b27fa7 PZ |
10350 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10351 | } | |
10352 | ||
10353 | long sched_group_rt_period(struct task_group *tg) | |
10354 | { | |
10355 | u64 rt_period_us; | |
10356 | ||
10357 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10358 | do_div(rt_period_us, NSEC_PER_USEC); | |
10359 | return rt_period_us; | |
10360 | } | |
10361 | ||
10362 | static int sched_rt_global_constraints(void) | |
10363 | { | |
4653f803 | 10364 | u64 runtime, period; |
d0b27fa7 PZ |
10365 | int ret = 0; |
10366 | ||
ec5d4989 HS |
10367 | if (sysctl_sched_rt_period <= 0) |
10368 | return -EINVAL; | |
10369 | ||
4653f803 PZ |
10370 | runtime = global_rt_runtime(); |
10371 | period = global_rt_period(); | |
10372 | ||
10373 | /* | |
10374 | * Sanity check on the sysctl variables. | |
10375 | */ | |
10376 | if (runtime > period && runtime != RUNTIME_INF) | |
10377 | return -EINVAL; | |
10b612f4 | 10378 | |
d0b27fa7 | 10379 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10380 | read_lock(&tasklist_lock); |
4653f803 | 10381 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10382 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10383 | mutex_unlock(&rt_constraints_mutex); |
10384 | ||
10385 | return ret; | |
10386 | } | |
54e99124 DG |
10387 | |
10388 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10389 | { | |
10390 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10391 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10392 | return 0; | |
10393 | ||
10394 | return 1; | |
10395 | } | |
10396 | ||
6d6bc0ad | 10397 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10398 | static int sched_rt_global_constraints(void) |
10399 | { | |
ac086bc2 PZ |
10400 | unsigned long flags; |
10401 | int i; | |
10402 | ||
ec5d4989 HS |
10403 | if (sysctl_sched_rt_period <= 0) |
10404 | return -EINVAL; | |
10405 | ||
60aa605d PZ |
10406 | /* |
10407 | * There's always some RT tasks in the root group | |
10408 | * -- migration, kstopmachine etc.. | |
10409 | */ | |
10410 | if (sysctl_sched_rt_runtime == 0) | |
10411 | return -EBUSY; | |
10412 | ||
ac086bc2 PZ |
10413 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10414 | for_each_possible_cpu(i) { | |
10415 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10416 | ||
10417 | spin_lock(&rt_rq->rt_runtime_lock); | |
10418 | rt_rq->rt_runtime = global_rt_runtime(); | |
10419 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10420 | } | |
10421 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10422 | ||
d0b27fa7 PZ |
10423 | return 0; |
10424 | } | |
6d6bc0ad | 10425 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10426 | |
10427 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 10428 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
10429 | loff_t *ppos) |
10430 | { | |
10431 | int ret; | |
10432 | int old_period, old_runtime; | |
10433 | static DEFINE_MUTEX(mutex); | |
10434 | ||
10435 | mutex_lock(&mutex); | |
10436 | old_period = sysctl_sched_rt_period; | |
10437 | old_runtime = sysctl_sched_rt_runtime; | |
10438 | ||
8d65af78 | 10439 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
10440 | |
10441 | if (!ret && write) { | |
10442 | ret = sched_rt_global_constraints(); | |
10443 | if (ret) { | |
10444 | sysctl_sched_rt_period = old_period; | |
10445 | sysctl_sched_rt_runtime = old_runtime; | |
10446 | } else { | |
10447 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10448 | def_rt_bandwidth.rt_period = | |
10449 | ns_to_ktime(global_rt_period()); | |
10450 | } | |
10451 | } | |
10452 | mutex_unlock(&mutex); | |
10453 | ||
10454 | return ret; | |
10455 | } | |
68318b8e | 10456 | |
052f1dc7 | 10457 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10458 | |
10459 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10460 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10461 | { |
2b01dfe3 PM |
10462 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10463 | struct task_group, css); | |
68318b8e SV |
10464 | } |
10465 | ||
10466 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10467 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10468 | { |
ec7dc8ac | 10469 | struct task_group *tg, *parent; |
68318b8e | 10470 | |
2b01dfe3 | 10471 | if (!cgrp->parent) { |
68318b8e | 10472 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10473 | return &init_task_group.css; |
10474 | } | |
10475 | ||
ec7dc8ac DG |
10476 | parent = cgroup_tg(cgrp->parent); |
10477 | tg = sched_create_group(parent); | |
68318b8e SV |
10478 | if (IS_ERR(tg)) |
10479 | return ERR_PTR(-ENOMEM); | |
10480 | ||
68318b8e SV |
10481 | return &tg->css; |
10482 | } | |
10483 | ||
41a2d6cf IM |
10484 | static void |
10485 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10486 | { |
2b01dfe3 | 10487 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10488 | |
10489 | sched_destroy_group(tg); | |
10490 | } | |
10491 | ||
41a2d6cf | 10492 | static int |
be367d09 | 10493 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 10494 | { |
b68aa230 | 10495 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10496 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10497 | return -EINVAL; |
10498 | #else | |
68318b8e SV |
10499 | /* We don't support RT-tasks being in separate groups */ |
10500 | if (tsk->sched_class != &fair_sched_class) | |
10501 | return -EINVAL; | |
b68aa230 | 10502 | #endif |
be367d09 BB |
10503 | return 0; |
10504 | } | |
68318b8e | 10505 | |
be367d09 BB |
10506 | static int |
10507 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10508 | struct task_struct *tsk, bool threadgroup) | |
10509 | { | |
10510 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
10511 | if (retval) | |
10512 | return retval; | |
10513 | if (threadgroup) { | |
10514 | struct task_struct *c; | |
10515 | rcu_read_lock(); | |
10516 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10517 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
10518 | if (retval) { | |
10519 | rcu_read_unlock(); | |
10520 | return retval; | |
10521 | } | |
10522 | } | |
10523 | rcu_read_unlock(); | |
10524 | } | |
68318b8e SV |
10525 | return 0; |
10526 | } | |
10527 | ||
10528 | static void | |
2b01dfe3 | 10529 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
10530 | struct cgroup *old_cont, struct task_struct *tsk, |
10531 | bool threadgroup) | |
68318b8e SV |
10532 | { |
10533 | sched_move_task(tsk); | |
be367d09 BB |
10534 | if (threadgroup) { |
10535 | struct task_struct *c; | |
10536 | rcu_read_lock(); | |
10537 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10538 | sched_move_task(c); | |
10539 | } | |
10540 | rcu_read_unlock(); | |
10541 | } | |
68318b8e SV |
10542 | } |
10543 | ||
052f1dc7 | 10544 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10545 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10546 | u64 shareval) |
68318b8e | 10547 | { |
2b01dfe3 | 10548 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10549 | } |
10550 | ||
f4c753b7 | 10551 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10552 | { |
2b01dfe3 | 10553 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10554 | |
10555 | return (u64) tg->shares; | |
10556 | } | |
6d6bc0ad | 10557 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10558 | |
052f1dc7 | 10559 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10560 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10561 | s64 val) |
6f505b16 | 10562 | { |
06ecb27c | 10563 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10564 | } |
10565 | ||
06ecb27c | 10566 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10567 | { |
06ecb27c | 10568 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10569 | } |
d0b27fa7 PZ |
10570 | |
10571 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10572 | u64 rt_period_us) | |
10573 | { | |
10574 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10575 | } | |
10576 | ||
10577 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10578 | { | |
10579 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10580 | } | |
6d6bc0ad | 10581 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10582 | |
fe5c7cc2 | 10583 | static struct cftype cpu_files[] = { |
052f1dc7 | 10584 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10585 | { |
10586 | .name = "shares", | |
f4c753b7 PM |
10587 | .read_u64 = cpu_shares_read_u64, |
10588 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10589 | }, |
052f1dc7 PZ |
10590 | #endif |
10591 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10592 | { |
9f0c1e56 | 10593 | .name = "rt_runtime_us", |
06ecb27c PM |
10594 | .read_s64 = cpu_rt_runtime_read, |
10595 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10596 | }, |
d0b27fa7 PZ |
10597 | { |
10598 | .name = "rt_period_us", | |
f4c753b7 PM |
10599 | .read_u64 = cpu_rt_period_read_uint, |
10600 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10601 | }, |
052f1dc7 | 10602 | #endif |
68318b8e SV |
10603 | }; |
10604 | ||
10605 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10606 | { | |
fe5c7cc2 | 10607 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10608 | } |
10609 | ||
10610 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10611 | .name = "cpu", |
10612 | .create = cpu_cgroup_create, | |
10613 | .destroy = cpu_cgroup_destroy, | |
10614 | .can_attach = cpu_cgroup_can_attach, | |
10615 | .attach = cpu_cgroup_attach, | |
10616 | .populate = cpu_cgroup_populate, | |
10617 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10618 | .early_init = 1, |
10619 | }; | |
10620 | ||
052f1dc7 | 10621 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10622 | |
10623 | #ifdef CONFIG_CGROUP_CPUACCT | |
10624 | ||
10625 | /* | |
10626 | * CPU accounting code for task groups. | |
10627 | * | |
10628 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10629 | * (balbir@in.ibm.com). | |
10630 | */ | |
10631 | ||
934352f2 | 10632 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10633 | struct cpuacct { |
10634 | struct cgroup_subsys_state css; | |
10635 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10636 | u64 *cpuusage; | |
ef12fefa | 10637 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10638 | struct cpuacct *parent; |
d842de87 SV |
10639 | }; |
10640 | ||
10641 | struct cgroup_subsys cpuacct_subsys; | |
10642 | ||
10643 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10644 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10645 | { |
32cd756a | 10646 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10647 | struct cpuacct, css); |
10648 | } | |
10649 | ||
10650 | /* return cpu accounting group to which this task belongs */ | |
10651 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10652 | { | |
10653 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10654 | struct cpuacct, css); | |
10655 | } | |
10656 | ||
10657 | /* create a new cpu accounting group */ | |
10658 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10659 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10660 | { |
10661 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10662 | int i; |
d842de87 SV |
10663 | |
10664 | if (!ca) | |
ef12fefa | 10665 | goto out; |
d842de87 SV |
10666 | |
10667 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10668 | if (!ca->cpuusage) |
10669 | goto out_free_ca; | |
10670 | ||
10671 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10672 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10673 | goto out_free_counters; | |
d842de87 | 10674 | |
934352f2 BR |
10675 | if (cgrp->parent) |
10676 | ca->parent = cgroup_ca(cgrp->parent); | |
10677 | ||
d842de87 | 10678 | return &ca->css; |
ef12fefa BR |
10679 | |
10680 | out_free_counters: | |
10681 | while (--i >= 0) | |
10682 | percpu_counter_destroy(&ca->cpustat[i]); | |
10683 | free_percpu(ca->cpuusage); | |
10684 | out_free_ca: | |
10685 | kfree(ca); | |
10686 | out: | |
10687 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10688 | } |
10689 | ||
10690 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10691 | static void |
32cd756a | 10692 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10693 | { |
32cd756a | 10694 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10695 | int i; |
d842de87 | 10696 | |
ef12fefa BR |
10697 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10698 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10699 | free_percpu(ca->cpuusage); |
10700 | kfree(ca); | |
10701 | } | |
10702 | ||
720f5498 KC |
10703 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10704 | { | |
b36128c8 | 10705 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10706 | u64 data; |
10707 | ||
10708 | #ifndef CONFIG_64BIT | |
10709 | /* | |
10710 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10711 | */ | |
10712 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10713 | data = *cpuusage; | |
10714 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10715 | #else | |
10716 | data = *cpuusage; | |
10717 | #endif | |
10718 | ||
10719 | return data; | |
10720 | } | |
10721 | ||
10722 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10723 | { | |
b36128c8 | 10724 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10725 | |
10726 | #ifndef CONFIG_64BIT | |
10727 | /* | |
10728 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10729 | */ | |
10730 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10731 | *cpuusage = val; | |
10732 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10733 | #else | |
10734 | *cpuusage = val; | |
10735 | #endif | |
10736 | } | |
10737 | ||
d842de87 | 10738 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10739 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10740 | { |
32cd756a | 10741 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10742 | u64 totalcpuusage = 0; |
10743 | int i; | |
10744 | ||
720f5498 KC |
10745 | for_each_present_cpu(i) |
10746 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10747 | |
10748 | return totalcpuusage; | |
10749 | } | |
10750 | ||
0297b803 DG |
10751 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10752 | u64 reset) | |
10753 | { | |
10754 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10755 | int err = 0; | |
10756 | int i; | |
10757 | ||
10758 | if (reset) { | |
10759 | err = -EINVAL; | |
10760 | goto out; | |
10761 | } | |
10762 | ||
720f5498 KC |
10763 | for_each_present_cpu(i) |
10764 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10765 | |
0297b803 DG |
10766 | out: |
10767 | return err; | |
10768 | } | |
10769 | ||
e9515c3c KC |
10770 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10771 | struct seq_file *m) | |
10772 | { | |
10773 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10774 | u64 percpu; | |
10775 | int i; | |
10776 | ||
10777 | for_each_present_cpu(i) { | |
10778 | percpu = cpuacct_cpuusage_read(ca, i); | |
10779 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10780 | } | |
10781 | seq_printf(m, "\n"); | |
10782 | return 0; | |
10783 | } | |
10784 | ||
ef12fefa BR |
10785 | static const char *cpuacct_stat_desc[] = { |
10786 | [CPUACCT_STAT_USER] = "user", | |
10787 | [CPUACCT_STAT_SYSTEM] = "system", | |
10788 | }; | |
10789 | ||
10790 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10791 | struct cgroup_map_cb *cb) | |
10792 | { | |
10793 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10794 | int i; | |
10795 | ||
10796 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10797 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10798 | val = cputime64_to_clock_t(val); | |
10799 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10800 | } | |
10801 | return 0; | |
10802 | } | |
10803 | ||
d842de87 SV |
10804 | static struct cftype files[] = { |
10805 | { | |
10806 | .name = "usage", | |
f4c753b7 PM |
10807 | .read_u64 = cpuusage_read, |
10808 | .write_u64 = cpuusage_write, | |
d842de87 | 10809 | }, |
e9515c3c KC |
10810 | { |
10811 | .name = "usage_percpu", | |
10812 | .read_seq_string = cpuacct_percpu_seq_read, | |
10813 | }, | |
ef12fefa BR |
10814 | { |
10815 | .name = "stat", | |
10816 | .read_map = cpuacct_stats_show, | |
10817 | }, | |
d842de87 SV |
10818 | }; |
10819 | ||
32cd756a | 10820 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10821 | { |
32cd756a | 10822 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10823 | } |
10824 | ||
10825 | /* | |
10826 | * charge this task's execution time to its accounting group. | |
10827 | * | |
10828 | * called with rq->lock held. | |
10829 | */ | |
10830 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10831 | { | |
10832 | struct cpuacct *ca; | |
934352f2 | 10833 | int cpu; |
d842de87 | 10834 | |
c40c6f85 | 10835 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10836 | return; |
10837 | ||
934352f2 | 10838 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10839 | |
10840 | rcu_read_lock(); | |
10841 | ||
d842de87 | 10842 | ca = task_ca(tsk); |
d842de87 | 10843 | |
934352f2 | 10844 | for (; ca; ca = ca->parent) { |
b36128c8 | 10845 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10846 | *cpuusage += cputime; |
10847 | } | |
a18b83b7 BR |
10848 | |
10849 | rcu_read_unlock(); | |
d842de87 SV |
10850 | } |
10851 | ||
ef12fefa BR |
10852 | /* |
10853 | * Charge the system/user time to the task's accounting group. | |
10854 | */ | |
10855 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10856 | enum cpuacct_stat_index idx, cputime_t val) | |
10857 | { | |
10858 | struct cpuacct *ca; | |
10859 | ||
10860 | if (unlikely(!cpuacct_subsys.active)) | |
10861 | return; | |
10862 | ||
10863 | rcu_read_lock(); | |
10864 | ca = task_ca(tsk); | |
10865 | ||
10866 | do { | |
10867 | percpu_counter_add(&ca->cpustat[idx], val); | |
10868 | ca = ca->parent; | |
10869 | } while (ca); | |
10870 | rcu_read_unlock(); | |
10871 | } | |
10872 | ||
d842de87 SV |
10873 | struct cgroup_subsys cpuacct_subsys = { |
10874 | .name = "cpuacct", | |
10875 | .create = cpuacct_create, | |
10876 | .destroy = cpuacct_destroy, | |
10877 | .populate = cpuacct_populate, | |
10878 | .subsys_id = cpuacct_subsys_id, | |
10879 | }; | |
10880 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
10881 | |
10882 | #ifndef CONFIG_SMP | |
10883 | ||
10884 | int rcu_expedited_torture_stats(char *page) | |
10885 | { | |
10886 | return 0; | |
10887 | } | |
10888 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10889 | ||
10890 | void synchronize_sched_expedited(void) | |
10891 | { | |
10892 | } | |
10893 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10894 | ||
10895 | #else /* #ifndef CONFIG_SMP */ | |
10896 | ||
10897 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
10898 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
10899 | ||
10900 | #define RCU_EXPEDITED_STATE_POST -2 | |
10901 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
10902 | ||
10903 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10904 | ||
10905 | int rcu_expedited_torture_stats(char *page) | |
10906 | { | |
10907 | int cnt = 0; | |
10908 | int cpu; | |
10909 | ||
10910 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
10911 | for_each_online_cpu(cpu) { | |
10912 | cnt += sprintf(&page[cnt], " %d:%d", | |
10913 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
10914 | } | |
10915 | cnt += sprintf(&page[cnt], "\n"); | |
10916 | return cnt; | |
10917 | } | |
10918 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10919 | ||
10920 | static long synchronize_sched_expedited_count; | |
10921 | ||
10922 | /* | |
10923 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
10924 | * approach to force grace period to end quickly. This consumes | |
10925 | * significant time on all CPUs, and is thus not recommended for | |
10926 | * any sort of common-case code. | |
10927 | * | |
10928 | * Note that it is illegal to call this function while holding any | |
10929 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
10930 | * observe this restriction will result in deadlock. | |
10931 | */ | |
10932 | void synchronize_sched_expedited(void) | |
10933 | { | |
10934 | int cpu; | |
10935 | unsigned long flags; | |
10936 | bool need_full_sync = 0; | |
10937 | struct rq *rq; | |
10938 | struct migration_req *req; | |
10939 | long snap; | |
10940 | int trycount = 0; | |
10941 | ||
10942 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
10943 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
10944 | get_online_cpus(); | |
10945 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
10946 | put_online_cpus(); | |
10947 | if (trycount++ < 10) | |
10948 | udelay(trycount * num_online_cpus()); | |
10949 | else { | |
10950 | synchronize_sched(); | |
10951 | return; | |
10952 | } | |
10953 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
10954 | smp_mb(); /* ensure test happens before caller kfree */ | |
10955 | return; | |
10956 | } | |
10957 | get_online_cpus(); | |
10958 | } | |
10959 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
10960 | for_each_online_cpu(cpu) { | |
10961 | rq = cpu_rq(cpu); | |
10962 | req = &per_cpu(rcu_migration_req, cpu); | |
10963 | init_completion(&req->done); | |
10964 | req->task = NULL; | |
10965 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
10966 | spin_lock_irqsave(&rq->lock, flags); | |
10967 | list_add(&req->list, &rq->migration_queue); | |
10968 | spin_unlock_irqrestore(&rq->lock, flags); | |
10969 | wake_up_process(rq->migration_thread); | |
10970 | } | |
10971 | for_each_online_cpu(cpu) { | |
10972 | rcu_expedited_state = cpu; | |
10973 | req = &per_cpu(rcu_migration_req, cpu); | |
10974 | rq = cpu_rq(cpu); | |
10975 | wait_for_completion(&req->done); | |
10976 | spin_lock_irqsave(&rq->lock, flags); | |
10977 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) | |
10978 | need_full_sync = 1; | |
10979 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
10980 | spin_unlock_irqrestore(&rq->lock, flags); | |
10981 | } | |
10982 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 10983 | synchronize_sched_expedited_count++; |
03b042bf PM |
10984 | mutex_unlock(&rcu_sched_expedited_mutex); |
10985 | put_online_cpus(); | |
10986 | if (need_full_sync) | |
10987 | synchronize_sched(); | |
10988 | } | |
10989 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10990 | ||
10991 | #endif /* #else #ifndef CONFIG_SMP */ |