]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
0d905bca | 42 | #include <linux/perf_counter.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
a8d154b0 | 80 | #define CREATE_TRACE_POINTS |
ad8d75ff | 81 | #include <trace/events/sched.h> |
a8d154b0 | 82 | |
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
d0b27fa7 PZ |
117 | /* |
118 | * single value that denotes runtime == period, ie unlimited time. | |
119 | */ | |
120 | #define RUNTIME_INF ((u64)~0ULL) | |
121 | ||
fd2ab30b SN |
122 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
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 */ | |
84e9dabf | 298 | static DEFINE_PER_CPU(struct cfs_rq, init_tg_cfs_rq) ____cacheline_aligned_in_smp; |
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); | |
303 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
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 | |
57310a98 PZ |
314 | #ifdef CONFIG_SMP |
315 | static int root_task_group_empty(void) | |
316 | { | |
317 | return list_empty(&root_task_group.children); | |
318 | } | |
319 | #endif | |
320 | ||
052f1dc7 | 321 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
322 | #ifdef CONFIG_USER_SCHED |
323 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 324 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 325 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 326 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 327 | |
cb4ad1ff | 328 | /* |
2e084786 LJ |
329 | * A weight of 0 or 1 can cause arithmetics problems. |
330 | * A weight of a cfs_rq is the sum of weights of which entities | |
331 | * are queued on this cfs_rq, so a weight of a entity should not be | |
332 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
333 | * (The default weight is 1024 - so there's no practical |
334 | * limitation from this.) | |
335 | */ | |
18d95a28 | 336 | #define MIN_SHARES 2 |
2e084786 | 337 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 338 | |
052f1dc7 PZ |
339 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
340 | #endif | |
341 | ||
29f59db3 | 342 | /* Default task group. |
3a252015 | 343 | * Every task in system belong to this group at bootup. |
29f59db3 | 344 | */ |
434d53b0 | 345 | struct task_group init_task_group; |
29f59db3 SV |
346 | |
347 | /* return group to which a task belongs */ | |
4cf86d77 | 348 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 349 | { |
4cf86d77 | 350 | struct task_group *tg; |
9b5b7751 | 351 | |
052f1dc7 | 352 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
353 | rcu_read_lock(); |
354 | tg = __task_cred(p)->user->tg; | |
355 | rcu_read_unlock(); | |
052f1dc7 | 356 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
357 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
358 | struct task_group, css); | |
24e377a8 | 359 | #else |
41a2d6cf | 360 | tg = &init_task_group; |
24e377a8 | 361 | #endif |
9b5b7751 | 362 | return tg; |
29f59db3 SV |
363 | } |
364 | ||
365 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 366 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 367 | { |
052f1dc7 | 368 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
369 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
370 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 371 | #endif |
6f505b16 | 372 | |
052f1dc7 | 373 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
374 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
375 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 376 | #endif |
29f59db3 SV |
377 | } |
378 | ||
379 | #else | |
380 | ||
57310a98 PZ |
381 | #ifdef CONFIG_SMP |
382 | static int root_task_group_empty(void) | |
383 | { | |
384 | return 1; | |
385 | } | |
386 | #endif | |
387 | ||
6f505b16 | 388 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
389 | static inline struct task_group *task_group(struct task_struct *p) |
390 | { | |
391 | return NULL; | |
392 | } | |
29f59db3 | 393 | |
052f1dc7 | 394 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 395 | |
6aa645ea IM |
396 | /* CFS-related fields in a runqueue */ |
397 | struct cfs_rq { | |
398 | struct load_weight load; | |
399 | unsigned long nr_running; | |
400 | ||
6aa645ea | 401 | u64 exec_clock; |
e9acbff6 | 402 | u64 min_vruntime; |
6aa645ea IM |
403 | |
404 | struct rb_root tasks_timeline; | |
405 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
406 | |
407 | struct list_head tasks; | |
408 | struct list_head *balance_iterator; | |
409 | ||
410 | /* | |
411 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
412 | * It is set to NULL otherwise (i.e when none are currently running). |
413 | */ | |
4793241b | 414 | struct sched_entity *curr, *next, *last; |
ddc97297 | 415 | |
5ac5c4d6 | 416 | unsigned int nr_spread_over; |
ddc97297 | 417 | |
62160e3f | 418 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
419 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
420 | ||
41a2d6cf IM |
421 | /* |
422 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
423 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
424 | * (like users, containers etc.) | |
425 | * | |
426 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
427 | * list is used during load balance. | |
428 | */ | |
41a2d6cf IM |
429 | struct list_head leaf_cfs_rq_list; |
430 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
431 | |
432 | #ifdef CONFIG_SMP | |
c09595f6 | 433 | /* |
c8cba857 | 434 | * the part of load.weight contributed by tasks |
c09595f6 | 435 | */ |
c8cba857 | 436 | unsigned long task_weight; |
c09595f6 | 437 | |
c8cba857 PZ |
438 | /* |
439 | * h_load = weight * f(tg) | |
440 | * | |
441 | * Where f(tg) is the recursive weight fraction assigned to | |
442 | * this group. | |
443 | */ | |
444 | unsigned long h_load; | |
c09595f6 | 445 | |
c8cba857 PZ |
446 | /* |
447 | * this cpu's part of tg->shares | |
448 | */ | |
449 | unsigned long shares; | |
f1d239f7 PZ |
450 | |
451 | /* | |
452 | * load.weight at the time we set shares | |
453 | */ | |
454 | unsigned long rq_weight; | |
c09595f6 | 455 | #endif |
6aa645ea IM |
456 | #endif |
457 | }; | |
1da177e4 | 458 | |
6aa645ea IM |
459 | /* Real-Time classes' related field in a runqueue: */ |
460 | struct rt_rq { | |
461 | struct rt_prio_array active; | |
63489e45 | 462 | unsigned long rt_nr_running; |
052f1dc7 | 463 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
464 | struct { |
465 | int curr; /* highest queued rt task prio */ | |
398a153b | 466 | #ifdef CONFIG_SMP |
e864c499 | 467 | int next; /* next highest */ |
398a153b | 468 | #endif |
e864c499 | 469 | } highest_prio; |
6f505b16 | 470 | #endif |
fa85ae24 | 471 | #ifdef CONFIG_SMP |
73fe6aae | 472 | unsigned long rt_nr_migratory; |
a1ba4d8b | 473 | unsigned long rt_nr_total; |
a22d7fc1 | 474 | int overloaded; |
917b627d | 475 | struct plist_head pushable_tasks; |
fa85ae24 | 476 | #endif |
6f505b16 | 477 | int rt_throttled; |
fa85ae24 | 478 | u64 rt_time; |
ac086bc2 | 479 | u64 rt_runtime; |
ea736ed5 | 480 | /* Nests inside the rq lock: */ |
ac086bc2 | 481 | spinlock_t rt_runtime_lock; |
6f505b16 | 482 | |
052f1dc7 | 483 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
484 | unsigned long rt_nr_boosted; |
485 | ||
6f505b16 PZ |
486 | struct rq *rq; |
487 | struct list_head leaf_rt_rq_list; | |
488 | struct task_group *tg; | |
489 | struct sched_rt_entity *rt_se; | |
490 | #endif | |
6aa645ea IM |
491 | }; |
492 | ||
57d885fe GH |
493 | #ifdef CONFIG_SMP |
494 | ||
495 | /* | |
496 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
497 | * variables. Each exclusive cpuset essentially defines an island domain by |
498 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
499 | * exclusive cpuset is created, we also create and attach a new root-domain |
500 | * object. | |
501 | * | |
57d885fe GH |
502 | */ |
503 | struct root_domain { | |
504 | atomic_t refcount; | |
c6c4927b RR |
505 | cpumask_var_t span; |
506 | cpumask_var_t online; | |
637f5085 | 507 | |
0eab9146 | 508 | /* |
637f5085 GH |
509 | * The "RT overload" flag: it gets set if a CPU has more than |
510 | * one runnable RT task. | |
511 | */ | |
c6c4927b | 512 | cpumask_var_t rto_mask; |
0eab9146 | 513 | atomic_t rto_count; |
6e0534f2 GH |
514 | #ifdef CONFIG_SMP |
515 | struct cpupri cpupri; | |
516 | #endif | |
7a09b1a2 VS |
517 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
518 | /* | |
519 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
520 | * used when most cpus are idle in the system indicating overall very | |
521 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
522 | */ | |
523 | unsigned int sched_mc_preferred_wakeup_cpu; | |
524 | #endif | |
57d885fe GH |
525 | }; |
526 | ||
dc938520 GH |
527 | /* |
528 | * By default the system creates a single root-domain with all cpus as | |
529 | * members (mimicking the global state we have today). | |
530 | */ | |
57d885fe GH |
531 | static struct root_domain def_root_domain; |
532 | ||
533 | #endif | |
534 | ||
1da177e4 LT |
535 | /* |
536 | * This is the main, per-CPU runqueue data structure. | |
537 | * | |
538 | * Locking rule: those places that want to lock multiple runqueues | |
539 | * (such as the load balancing or the thread migration code), lock | |
540 | * acquire operations must be ordered by ascending &runqueue. | |
541 | */ | |
70b97a7f | 542 | struct rq { |
d8016491 IM |
543 | /* runqueue lock: */ |
544 | spinlock_t lock; | |
1da177e4 LT |
545 | |
546 | /* | |
547 | * nr_running and cpu_load should be in the same cacheline because | |
548 | * remote CPUs use both these fields when doing load calculation. | |
549 | */ | |
550 | unsigned long nr_running; | |
6aa645ea IM |
551 | #define CPU_LOAD_IDX_MAX 5 |
552 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 553 | #ifdef CONFIG_NO_HZ |
15934a37 | 554 | unsigned long last_tick_seen; |
46cb4b7c SS |
555 | unsigned char in_nohz_recently; |
556 | #endif | |
d8016491 IM |
557 | /* capture load from *all* tasks on this cpu: */ |
558 | struct load_weight load; | |
6aa645ea IM |
559 | unsigned long nr_load_updates; |
560 | u64 nr_switches; | |
23a185ca | 561 | u64 nr_migrations_in; |
6aa645ea IM |
562 | |
563 | struct cfs_rq cfs; | |
6f505b16 | 564 | struct rt_rq rt; |
6f505b16 | 565 | |
6aa645ea | 566 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
567 | /* list of leaf cfs_rq on this cpu: */ |
568 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
569 | #endif |
570 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 571 | struct list_head leaf_rt_rq_list; |
1da177e4 | 572 | #endif |
1da177e4 LT |
573 | |
574 | /* | |
575 | * This is part of a global counter where only the total sum | |
576 | * over all CPUs matters. A task can increase this counter on | |
577 | * one CPU and if it got migrated afterwards it may decrease | |
578 | * it on another CPU. Always updated under the runqueue lock: | |
579 | */ | |
580 | unsigned long nr_uninterruptible; | |
581 | ||
36c8b586 | 582 | struct task_struct *curr, *idle; |
c9819f45 | 583 | unsigned long next_balance; |
1da177e4 | 584 | struct mm_struct *prev_mm; |
6aa645ea | 585 | |
3e51f33f | 586 | u64 clock; |
6aa645ea | 587 | |
1da177e4 LT |
588 | atomic_t nr_iowait; |
589 | ||
590 | #ifdef CONFIG_SMP | |
0eab9146 | 591 | struct root_domain *rd; |
1da177e4 LT |
592 | struct sched_domain *sd; |
593 | ||
a0a522ce | 594 | unsigned char idle_at_tick; |
1da177e4 | 595 | /* For active balancing */ |
3f029d3c | 596 | int post_schedule; |
1da177e4 LT |
597 | int active_balance; |
598 | int push_cpu; | |
d8016491 IM |
599 | /* cpu of this runqueue: */ |
600 | int cpu; | |
1f11eb6a | 601 | int online; |
1da177e4 | 602 | |
a8a51d5e | 603 | unsigned long avg_load_per_task; |
1da177e4 | 604 | |
36c8b586 | 605 | struct task_struct *migration_thread; |
1da177e4 | 606 | struct list_head migration_queue; |
e9e9250b PZ |
607 | |
608 | u64 rt_avg; | |
609 | u64 age_stamp; | |
1da177e4 LT |
610 | #endif |
611 | ||
dce48a84 TG |
612 | /* calc_load related fields */ |
613 | unsigned long calc_load_update; | |
614 | long calc_load_active; | |
615 | ||
8f4d37ec | 616 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
617 | #ifdef CONFIG_SMP |
618 | int hrtick_csd_pending; | |
619 | struct call_single_data hrtick_csd; | |
620 | #endif | |
8f4d37ec PZ |
621 | struct hrtimer hrtick_timer; |
622 | #endif | |
623 | ||
1da177e4 LT |
624 | #ifdef CONFIG_SCHEDSTATS |
625 | /* latency stats */ | |
626 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
627 | unsigned long long rq_cpu_time; |
628 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
629 | |
630 | /* sys_sched_yield() stats */ | |
480b9434 | 631 | unsigned int yld_count; |
1da177e4 LT |
632 | |
633 | /* schedule() stats */ | |
480b9434 KC |
634 | unsigned int sched_switch; |
635 | unsigned int sched_count; | |
636 | unsigned int sched_goidle; | |
1da177e4 LT |
637 | |
638 | /* try_to_wake_up() stats */ | |
480b9434 KC |
639 | unsigned int ttwu_count; |
640 | unsigned int ttwu_local; | |
b8efb561 IM |
641 | |
642 | /* BKL stats */ | |
480b9434 | 643 | unsigned int bkl_count; |
1da177e4 LT |
644 | #endif |
645 | }; | |
646 | ||
f34e3b61 | 647 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 648 | |
15afe09b | 649 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 650 | { |
15afe09b | 651 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
652 | } |
653 | ||
0a2966b4 CL |
654 | static inline int cpu_of(struct rq *rq) |
655 | { | |
656 | #ifdef CONFIG_SMP | |
657 | return rq->cpu; | |
658 | #else | |
659 | return 0; | |
660 | #endif | |
661 | } | |
662 | ||
674311d5 NP |
663 | /* |
664 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 665 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
666 | * |
667 | * The domain tree of any CPU may only be accessed from within | |
668 | * preempt-disabled sections. | |
669 | */ | |
48f24c4d IM |
670 | #define for_each_domain(cpu, __sd) \ |
671 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
672 | |
673 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
674 | #define this_rq() (&__get_cpu_var(runqueues)) | |
675 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
676 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 677 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 678 | |
aa9c4c0f | 679 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
680 | { |
681 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
682 | } | |
683 | ||
bf5c91ba IM |
684 | /* |
685 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
686 | */ | |
687 | #ifdef CONFIG_SCHED_DEBUG | |
688 | # define const_debug __read_mostly | |
689 | #else | |
690 | # define const_debug static const | |
691 | #endif | |
692 | ||
017730c1 IM |
693 | /** |
694 | * runqueue_is_locked | |
695 | * | |
696 | * Returns true if the current cpu runqueue is locked. | |
697 | * This interface allows printk to be called with the runqueue lock | |
698 | * held and know whether or not it is OK to wake up the klogd. | |
699 | */ | |
700 | int runqueue_is_locked(void) | |
701 | { | |
702 | int cpu = get_cpu(); | |
703 | struct rq *rq = cpu_rq(cpu); | |
704 | int ret; | |
705 | ||
706 | ret = spin_is_locked(&rq->lock); | |
707 | put_cpu(); | |
708 | return ret; | |
709 | } | |
710 | ||
bf5c91ba IM |
711 | /* |
712 | * Debugging: various feature bits | |
713 | */ | |
f00b45c1 PZ |
714 | |
715 | #define SCHED_FEAT(name, enabled) \ | |
716 | __SCHED_FEAT_##name , | |
717 | ||
bf5c91ba | 718 | enum { |
f00b45c1 | 719 | #include "sched_features.h" |
bf5c91ba IM |
720 | }; |
721 | ||
f00b45c1 PZ |
722 | #undef SCHED_FEAT |
723 | ||
724 | #define SCHED_FEAT(name, enabled) \ | |
725 | (1UL << __SCHED_FEAT_##name) * enabled | | |
726 | ||
bf5c91ba | 727 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
728 | #include "sched_features.h" |
729 | 0; | |
730 | ||
731 | #undef SCHED_FEAT | |
732 | ||
733 | #ifdef CONFIG_SCHED_DEBUG | |
734 | #define SCHED_FEAT(name, enabled) \ | |
735 | #name , | |
736 | ||
983ed7a6 | 737 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
738 | #include "sched_features.h" |
739 | NULL | |
740 | }; | |
741 | ||
742 | #undef SCHED_FEAT | |
743 | ||
34f3a814 | 744 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 745 | { |
f00b45c1 PZ |
746 | int i; |
747 | ||
748 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
749 | if (!(sysctl_sched_features & (1UL << i))) |
750 | seq_puts(m, "NO_"); | |
751 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 752 | } |
34f3a814 | 753 | seq_puts(m, "\n"); |
f00b45c1 | 754 | |
34f3a814 | 755 | return 0; |
f00b45c1 PZ |
756 | } |
757 | ||
758 | static ssize_t | |
759 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
760 | size_t cnt, loff_t *ppos) | |
761 | { | |
762 | char buf[64]; | |
763 | char *cmp = buf; | |
764 | int neg = 0; | |
765 | int i; | |
766 | ||
767 | if (cnt > 63) | |
768 | cnt = 63; | |
769 | ||
770 | if (copy_from_user(&buf, ubuf, cnt)) | |
771 | return -EFAULT; | |
772 | ||
773 | buf[cnt] = 0; | |
774 | ||
c24b7c52 | 775 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
776 | neg = 1; |
777 | cmp += 3; | |
778 | } | |
779 | ||
780 | for (i = 0; sched_feat_names[i]; i++) { | |
781 | int len = strlen(sched_feat_names[i]); | |
782 | ||
783 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
784 | if (neg) | |
785 | sysctl_sched_features &= ~(1UL << i); | |
786 | else | |
787 | sysctl_sched_features |= (1UL << i); | |
788 | break; | |
789 | } | |
790 | } | |
791 | ||
792 | if (!sched_feat_names[i]) | |
793 | return -EINVAL; | |
794 | ||
795 | filp->f_pos += cnt; | |
796 | ||
797 | return cnt; | |
798 | } | |
799 | ||
34f3a814 LZ |
800 | static int sched_feat_open(struct inode *inode, struct file *filp) |
801 | { | |
802 | return single_open(filp, sched_feat_show, NULL); | |
803 | } | |
804 | ||
f00b45c1 | 805 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
806 | .open = sched_feat_open, |
807 | .write = sched_feat_write, | |
808 | .read = seq_read, | |
809 | .llseek = seq_lseek, | |
810 | .release = single_release, | |
f00b45c1 PZ |
811 | }; |
812 | ||
813 | static __init int sched_init_debug(void) | |
814 | { | |
f00b45c1 PZ |
815 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
816 | &sched_feat_fops); | |
817 | ||
818 | return 0; | |
819 | } | |
820 | late_initcall(sched_init_debug); | |
821 | ||
822 | #endif | |
823 | ||
824 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 825 | |
b82d9fdd PZ |
826 | /* |
827 | * Number of tasks to iterate in a single balance run. | |
828 | * Limited because this is done with IRQs disabled. | |
829 | */ | |
830 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
831 | ||
2398f2c6 PZ |
832 | /* |
833 | * ratelimit for updating the group shares. | |
55cd5340 | 834 | * default: 0.25ms |
2398f2c6 | 835 | */ |
55cd5340 | 836 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 837 | |
ffda12a1 PZ |
838 | /* |
839 | * Inject some fuzzyness into changing the per-cpu group shares | |
840 | * this avoids remote rq-locks at the expense of fairness. | |
841 | * default: 4 | |
842 | */ | |
843 | unsigned int sysctl_sched_shares_thresh = 4; | |
844 | ||
e9e9250b PZ |
845 | /* |
846 | * period over which we average the RT time consumption, measured | |
847 | * in ms. | |
848 | * | |
849 | * default: 1s | |
850 | */ | |
851 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
852 | ||
fa85ae24 | 853 | /* |
9f0c1e56 | 854 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
855 | * default: 1s |
856 | */ | |
9f0c1e56 | 857 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 858 | |
6892b75e IM |
859 | static __read_mostly int scheduler_running; |
860 | ||
9f0c1e56 PZ |
861 | /* |
862 | * part of the period that we allow rt tasks to run in us. | |
863 | * default: 0.95s | |
864 | */ | |
865 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 866 | |
d0b27fa7 PZ |
867 | static inline u64 global_rt_period(void) |
868 | { | |
869 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
870 | } | |
871 | ||
872 | static inline u64 global_rt_runtime(void) | |
873 | { | |
e26873bb | 874 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
875 | return RUNTIME_INF; |
876 | ||
877 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
878 | } | |
fa85ae24 | 879 | |
1da177e4 | 880 | #ifndef prepare_arch_switch |
4866cde0 NP |
881 | # define prepare_arch_switch(next) do { } while (0) |
882 | #endif | |
883 | #ifndef finish_arch_switch | |
884 | # define finish_arch_switch(prev) do { } while (0) | |
885 | #endif | |
886 | ||
051a1d1a DA |
887 | static inline int task_current(struct rq *rq, struct task_struct *p) |
888 | { | |
889 | return rq->curr == p; | |
890 | } | |
891 | ||
4866cde0 | 892 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 893 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 894 | { |
051a1d1a | 895 | return task_current(rq, p); |
4866cde0 NP |
896 | } |
897 | ||
70b97a7f | 898 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
899 | { |
900 | } | |
901 | ||
70b97a7f | 902 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 903 | { |
da04c035 IM |
904 | #ifdef CONFIG_DEBUG_SPINLOCK |
905 | /* this is a valid case when another task releases the spinlock */ | |
906 | rq->lock.owner = current; | |
907 | #endif | |
8a25d5de IM |
908 | /* |
909 | * If we are tracking spinlock dependencies then we have to | |
910 | * fix up the runqueue lock - which gets 'carried over' from | |
911 | * prev into current: | |
912 | */ | |
913 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
914 | ||
4866cde0 NP |
915 | spin_unlock_irq(&rq->lock); |
916 | } | |
917 | ||
918 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 919 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
920 | { |
921 | #ifdef CONFIG_SMP | |
922 | return p->oncpu; | |
923 | #else | |
051a1d1a | 924 | return task_current(rq, p); |
4866cde0 NP |
925 | #endif |
926 | } | |
927 | ||
70b97a7f | 928 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
929 | { |
930 | #ifdef CONFIG_SMP | |
931 | /* | |
932 | * We can optimise this out completely for !SMP, because the | |
933 | * SMP rebalancing from interrupt is the only thing that cares | |
934 | * here. | |
935 | */ | |
936 | next->oncpu = 1; | |
937 | #endif | |
938 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
939 | spin_unlock_irq(&rq->lock); | |
940 | #else | |
941 | spin_unlock(&rq->lock); | |
942 | #endif | |
943 | } | |
944 | ||
70b97a7f | 945 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
946 | { |
947 | #ifdef CONFIG_SMP | |
948 | /* | |
949 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
950 | * We must ensure this doesn't happen until the switch is completely | |
951 | * finished. | |
952 | */ | |
953 | smp_wmb(); | |
954 | prev->oncpu = 0; | |
955 | #endif | |
956 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
957 | local_irq_enable(); | |
1da177e4 | 958 | #endif |
4866cde0 NP |
959 | } |
960 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 961 | |
b29739f9 IM |
962 | /* |
963 | * __task_rq_lock - lock the runqueue a given task resides on. | |
964 | * Must be called interrupts disabled. | |
965 | */ | |
70b97a7f | 966 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
967 | __acquires(rq->lock) |
968 | { | |
3a5c359a AK |
969 | for (;;) { |
970 | struct rq *rq = task_rq(p); | |
971 | spin_lock(&rq->lock); | |
972 | if (likely(rq == task_rq(p))) | |
973 | return rq; | |
b29739f9 | 974 | spin_unlock(&rq->lock); |
b29739f9 | 975 | } |
b29739f9 IM |
976 | } |
977 | ||
1da177e4 LT |
978 | /* |
979 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 980 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
981 | * explicitly disabling preemption. |
982 | */ | |
70b97a7f | 983 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
984 | __acquires(rq->lock) |
985 | { | |
70b97a7f | 986 | struct rq *rq; |
1da177e4 | 987 | |
3a5c359a AK |
988 | for (;;) { |
989 | local_irq_save(*flags); | |
990 | rq = task_rq(p); | |
991 | spin_lock(&rq->lock); | |
992 | if (likely(rq == task_rq(p))) | |
993 | return rq; | |
1da177e4 | 994 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 995 | } |
1da177e4 LT |
996 | } |
997 | ||
ad474cac ON |
998 | void task_rq_unlock_wait(struct task_struct *p) |
999 | { | |
1000 | struct rq *rq = task_rq(p); | |
1001 | ||
1002 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1003 | spin_unlock_wait(&rq->lock); | |
1004 | } | |
1005 | ||
a9957449 | 1006 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1007 | __releases(rq->lock) |
1008 | { | |
1009 | spin_unlock(&rq->lock); | |
1010 | } | |
1011 | ||
70b97a7f | 1012 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1013 | __releases(rq->lock) |
1014 | { | |
1015 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1016 | } | |
1017 | ||
1da177e4 | 1018 | /* |
cc2a73b5 | 1019 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1020 | */ |
a9957449 | 1021 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1022 | __acquires(rq->lock) |
1023 | { | |
70b97a7f | 1024 | struct rq *rq; |
1da177e4 LT |
1025 | |
1026 | local_irq_disable(); | |
1027 | rq = this_rq(); | |
1028 | spin_lock(&rq->lock); | |
1029 | ||
1030 | return rq; | |
1031 | } | |
1032 | ||
8f4d37ec PZ |
1033 | #ifdef CONFIG_SCHED_HRTICK |
1034 | /* | |
1035 | * Use HR-timers to deliver accurate preemption points. | |
1036 | * | |
1037 | * Its all a bit involved since we cannot program an hrt while holding the | |
1038 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1039 | * reschedule event. | |
1040 | * | |
1041 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1042 | * rq->lock. | |
1043 | */ | |
8f4d37ec PZ |
1044 | |
1045 | /* | |
1046 | * Use hrtick when: | |
1047 | * - enabled by features | |
1048 | * - hrtimer is actually high res | |
1049 | */ | |
1050 | static inline int hrtick_enabled(struct rq *rq) | |
1051 | { | |
1052 | if (!sched_feat(HRTICK)) | |
1053 | return 0; | |
ba42059f | 1054 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1055 | return 0; |
8f4d37ec PZ |
1056 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1057 | } | |
1058 | ||
8f4d37ec PZ |
1059 | static void hrtick_clear(struct rq *rq) |
1060 | { | |
1061 | if (hrtimer_active(&rq->hrtick_timer)) | |
1062 | hrtimer_cancel(&rq->hrtick_timer); | |
1063 | } | |
1064 | ||
8f4d37ec PZ |
1065 | /* |
1066 | * High-resolution timer tick. | |
1067 | * Runs from hardirq context with interrupts disabled. | |
1068 | */ | |
1069 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1070 | { | |
1071 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1072 | ||
1073 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1074 | ||
1075 | spin_lock(&rq->lock); | |
3e51f33f | 1076 | update_rq_clock(rq); |
8f4d37ec PZ |
1077 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1078 | spin_unlock(&rq->lock); | |
1079 | ||
1080 | return HRTIMER_NORESTART; | |
1081 | } | |
1082 | ||
95e904c7 | 1083 | #ifdef CONFIG_SMP |
31656519 PZ |
1084 | /* |
1085 | * called from hardirq (IPI) context | |
1086 | */ | |
1087 | static void __hrtick_start(void *arg) | |
b328ca18 | 1088 | { |
31656519 | 1089 | struct rq *rq = arg; |
b328ca18 | 1090 | |
31656519 PZ |
1091 | spin_lock(&rq->lock); |
1092 | hrtimer_restart(&rq->hrtick_timer); | |
1093 | rq->hrtick_csd_pending = 0; | |
1094 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1095 | } |
1096 | ||
31656519 PZ |
1097 | /* |
1098 | * Called to set the hrtick timer state. | |
1099 | * | |
1100 | * called with rq->lock held and irqs disabled | |
1101 | */ | |
1102 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1103 | { |
31656519 PZ |
1104 | struct hrtimer *timer = &rq->hrtick_timer; |
1105 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1106 | |
cc584b21 | 1107 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1108 | |
1109 | if (rq == this_rq()) { | |
1110 | hrtimer_restart(timer); | |
1111 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1112 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1113 | rq->hrtick_csd_pending = 1; |
1114 | } | |
b328ca18 PZ |
1115 | } |
1116 | ||
1117 | static int | |
1118 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1119 | { | |
1120 | int cpu = (int)(long)hcpu; | |
1121 | ||
1122 | switch (action) { | |
1123 | case CPU_UP_CANCELED: | |
1124 | case CPU_UP_CANCELED_FROZEN: | |
1125 | case CPU_DOWN_PREPARE: | |
1126 | case CPU_DOWN_PREPARE_FROZEN: | |
1127 | case CPU_DEAD: | |
1128 | case CPU_DEAD_FROZEN: | |
31656519 | 1129 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1130 | return NOTIFY_OK; |
1131 | } | |
1132 | ||
1133 | return NOTIFY_DONE; | |
1134 | } | |
1135 | ||
fa748203 | 1136 | static __init void init_hrtick(void) |
b328ca18 PZ |
1137 | { |
1138 | hotcpu_notifier(hotplug_hrtick, 0); | |
1139 | } | |
31656519 PZ |
1140 | #else |
1141 | /* | |
1142 | * Called to set the hrtick timer state. | |
1143 | * | |
1144 | * called with rq->lock held and irqs disabled | |
1145 | */ | |
1146 | static void hrtick_start(struct rq *rq, u64 delay) | |
1147 | { | |
7f1e2ca9 | 1148 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1149 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1150 | } |
b328ca18 | 1151 | |
006c75f1 | 1152 | static inline void init_hrtick(void) |
8f4d37ec | 1153 | { |
8f4d37ec | 1154 | } |
31656519 | 1155 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1156 | |
31656519 | 1157 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1158 | { |
31656519 PZ |
1159 | #ifdef CONFIG_SMP |
1160 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1161 | |
31656519 PZ |
1162 | rq->hrtick_csd.flags = 0; |
1163 | rq->hrtick_csd.func = __hrtick_start; | |
1164 | rq->hrtick_csd.info = rq; | |
1165 | #endif | |
8f4d37ec | 1166 | |
31656519 PZ |
1167 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1168 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1169 | } |
006c75f1 | 1170 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1171 | static inline void hrtick_clear(struct rq *rq) |
1172 | { | |
1173 | } | |
1174 | ||
8f4d37ec PZ |
1175 | static inline void init_rq_hrtick(struct rq *rq) |
1176 | { | |
1177 | } | |
1178 | ||
b328ca18 PZ |
1179 | static inline void init_hrtick(void) |
1180 | { | |
1181 | } | |
006c75f1 | 1182 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1183 | |
c24d20db IM |
1184 | /* |
1185 | * resched_task - mark a task 'to be rescheduled now'. | |
1186 | * | |
1187 | * On UP this means the setting of the need_resched flag, on SMP it | |
1188 | * might also involve a cross-CPU call to trigger the scheduler on | |
1189 | * the target CPU. | |
1190 | */ | |
1191 | #ifdef CONFIG_SMP | |
1192 | ||
1193 | #ifndef tsk_is_polling | |
1194 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1195 | #endif | |
1196 | ||
31656519 | 1197 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1198 | { |
1199 | int cpu; | |
1200 | ||
1201 | assert_spin_locked(&task_rq(p)->lock); | |
1202 | ||
5ed0cec0 | 1203 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1204 | return; |
1205 | ||
5ed0cec0 | 1206 | set_tsk_need_resched(p); |
c24d20db IM |
1207 | |
1208 | cpu = task_cpu(p); | |
1209 | if (cpu == smp_processor_id()) | |
1210 | return; | |
1211 | ||
1212 | /* NEED_RESCHED must be visible before we test polling */ | |
1213 | smp_mb(); | |
1214 | if (!tsk_is_polling(p)) | |
1215 | smp_send_reschedule(cpu); | |
1216 | } | |
1217 | ||
1218 | static void resched_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | unsigned long flags; | |
1222 | ||
1223 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1224 | return; | |
1225 | resched_task(cpu_curr(cpu)); | |
1226 | spin_unlock_irqrestore(&rq->lock, flags); | |
1227 | } | |
06d8308c TG |
1228 | |
1229 | #ifdef CONFIG_NO_HZ | |
1230 | /* | |
1231 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1232 | * idle CPU then this timer might expire before the next timer event | |
1233 | * which is scheduled to wake up that CPU. In case of a completely | |
1234 | * idle system the next event might even be infinite time into the | |
1235 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1236 | * leaves the inner idle loop so the newly added timer is taken into | |
1237 | * account when the CPU goes back to idle and evaluates the timer | |
1238 | * wheel for the next timer event. | |
1239 | */ | |
1240 | void wake_up_idle_cpu(int cpu) | |
1241 | { | |
1242 | struct rq *rq = cpu_rq(cpu); | |
1243 | ||
1244 | if (cpu == smp_processor_id()) | |
1245 | return; | |
1246 | ||
1247 | /* | |
1248 | * This is safe, as this function is called with the timer | |
1249 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1250 | * to idle and has not yet set rq->curr to idle then it will | |
1251 | * be serialized on the timer wheel base lock and take the new | |
1252 | * timer into account automatically. | |
1253 | */ | |
1254 | if (rq->curr != rq->idle) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1259 | * lockless. The worst case is that the other CPU runs the | |
1260 | * idle task through an additional NOOP schedule() | |
1261 | */ | |
5ed0cec0 | 1262 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1263 | |
1264 | /* NEED_RESCHED must be visible before we test polling */ | |
1265 | smp_mb(); | |
1266 | if (!tsk_is_polling(rq->idle)) | |
1267 | smp_send_reschedule(cpu); | |
1268 | } | |
6d6bc0ad | 1269 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1270 | |
e9e9250b PZ |
1271 | static u64 sched_avg_period(void) |
1272 | { | |
1273 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1274 | } | |
1275 | ||
1276 | static void sched_avg_update(struct rq *rq) | |
1277 | { | |
1278 | s64 period = sched_avg_period(); | |
1279 | ||
1280 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1281 | rq->age_stamp += period; | |
1282 | rq->rt_avg /= 2; | |
1283 | } | |
1284 | } | |
1285 | ||
1286 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1287 | { | |
1288 | rq->rt_avg += rt_delta; | |
1289 | sched_avg_update(rq); | |
1290 | } | |
1291 | ||
6d6bc0ad | 1292 | #else /* !CONFIG_SMP */ |
31656519 | 1293 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1294 | { |
1295 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1296 | set_tsk_need_resched(p); |
c24d20db | 1297 | } |
e9e9250b PZ |
1298 | |
1299 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1300 | { | |
1301 | } | |
6d6bc0ad | 1302 | #endif /* CONFIG_SMP */ |
c24d20db | 1303 | |
45bf76df IM |
1304 | #if BITS_PER_LONG == 32 |
1305 | # define WMULT_CONST (~0UL) | |
1306 | #else | |
1307 | # define WMULT_CONST (1UL << 32) | |
1308 | #endif | |
1309 | ||
1310 | #define WMULT_SHIFT 32 | |
1311 | ||
194081eb IM |
1312 | /* |
1313 | * Shift right and round: | |
1314 | */ | |
cf2ab469 | 1315 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1316 | |
a7be37ac PZ |
1317 | /* |
1318 | * delta *= weight / lw | |
1319 | */ | |
cb1c4fc9 | 1320 | static unsigned long |
45bf76df IM |
1321 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1322 | struct load_weight *lw) | |
1323 | { | |
1324 | u64 tmp; | |
1325 | ||
7a232e03 LJ |
1326 | if (!lw->inv_weight) { |
1327 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1328 | lw->inv_weight = 1; | |
1329 | else | |
1330 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1331 | / (lw->weight+1); | |
1332 | } | |
45bf76df IM |
1333 | |
1334 | tmp = (u64)delta_exec * weight; | |
1335 | /* | |
1336 | * Check whether we'd overflow the 64-bit multiplication: | |
1337 | */ | |
194081eb | 1338 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1339 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1340 | WMULT_SHIFT/2); |
1341 | else | |
cf2ab469 | 1342 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1343 | |
ecf691da | 1344 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1345 | } |
1346 | ||
1091985b | 1347 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1348 | { |
1349 | lw->weight += inc; | |
e89996ae | 1350 | lw->inv_weight = 0; |
45bf76df IM |
1351 | } |
1352 | ||
1091985b | 1353 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1354 | { |
1355 | lw->weight -= dec; | |
e89996ae | 1356 | lw->inv_weight = 0; |
45bf76df IM |
1357 | } |
1358 | ||
2dd73a4f PW |
1359 | /* |
1360 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1361 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1362 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1363 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1364 | * scaled version of the new time slice allocation that they receive on time |
1365 | * slice expiry etc. | |
1366 | */ | |
1367 | ||
cce7ade8 PZ |
1368 | #define WEIGHT_IDLEPRIO 3 |
1369 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1370 | |
1371 | /* | |
1372 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1373 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1374 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1375 | * that remained on nice 0. | |
1376 | * | |
1377 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1378 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1379 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1380 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1381 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1382 | */ |
1383 | static const int prio_to_weight[40] = { | |
254753dc IM |
1384 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1385 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1386 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1387 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1388 | /* 0 */ 1024, 820, 655, 526, 423, | |
1389 | /* 5 */ 335, 272, 215, 172, 137, | |
1390 | /* 10 */ 110, 87, 70, 56, 45, | |
1391 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1392 | }; |
1393 | ||
5714d2de IM |
1394 | /* |
1395 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1396 | * | |
1397 | * In cases where the weight does not change often, we can use the | |
1398 | * precalculated inverse to speed up arithmetics by turning divisions | |
1399 | * into multiplications: | |
1400 | */ | |
dd41f596 | 1401 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1402 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1403 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1404 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1405 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1406 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1407 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1408 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1409 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1410 | }; |
2dd73a4f | 1411 | |
dd41f596 IM |
1412 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1413 | ||
1414 | /* | |
1415 | * runqueue iterator, to support SMP load-balancing between different | |
1416 | * scheduling classes, without having to expose their internal data | |
1417 | * structures to the load-balancing proper: | |
1418 | */ | |
1419 | struct rq_iterator { | |
1420 | void *arg; | |
1421 | struct task_struct *(*start)(void *); | |
1422 | struct task_struct *(*next)(void *); | |
1423 | }; | |
1424 | ||
e1d1484f PW |
1425 | #ifdef CONFIG_SMP |
1426 | static unsigned long | |
1427 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1428 | unsigned long max_load_move, struct sched_domain *sd, | |
1429 | enum cpu_idle_type idle, int *all_pinned, | |
1430 | int *this_best_prio, struct rq_iterator *iterator); | |
1431 | ||
1432 | static int | |
1433 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1434 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1435 | struct rq_iterator *iterator); | |
e1d1484f | 1436 | #endif |
dd41f596 | 1437 | |
ef12fefa BR |
1438 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1439 | enum cpuacct_stat_index { | |
1440 | CPUACCT_STAT_USER, /* ... user mode */ | |
1441 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1442 | ||
1443 | CPUACCT_STAT_NSTATS, | |
1444 | }; | |
1445 | ||
d842de87 SV |
1446 | #ifdef CONFIG_CGROUP_CPUACCT |
1447 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1448 | static void cpuacct_update_stats(struct task_struct *tsk, |
1449 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1450 | #else |
1451 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1452 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1453 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1454 | #endif |
1455 | ||
18d95a28 PZ |
1456 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1457 | { | |
1458 | update_load_add(&rq->load, load); | |
1459 | } | |
1460 | ||
1461 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1462 | { | |
1463 | update_load_sub(&rq->load, load); | |
1464 | } | |
1465 | ||
7940ca36 | 1466 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1467 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1468 | |
1469 | /* | |
1470 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1471 | * leaving it for the final time. | |
1472 | */ | |
eb755805 | 1473 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1474 | { |
1475 | struct task_group *parent, *child; | |
eb755805 | 1476 | int ret; |
c09595f6 PZ |
1477 | |
1478 | rcu_read_lock(); | |
1479 | parent = &root_task_group; | |
1480 | down: | |
eb755805 PZ |
1481 | ret = (*down)(parent, data); |
1482 | if (ret) | |
1483 | goto out_unlock; | |
c09595f6 PZ |
1484 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1485 | parent = child; | |
1486 | goto down; | |
1487 | ||
1488 | up: | |
1489 | continue; | |
1490 | } | |
eb755805 PZ |
1491 | ret = (*up)(parent, data); |
1492 | if (ret) | |
1493 | goto out_unlock; | |
c09595f6 PZ |
1494 | |
1495 | child = parent; | |
1496 | parent = parent->parent; | |
1497 | if (parent) | |
1498 | goto up; | |
eb755805 | 1499 | out_unlock: |
c09595f6 | 1500 | rcu_read_unlock(); |
eb755805 PZ |
1501 | |
1502 | return ret; | |
c09595f6 PZ |
1503 | } |
1504 | ||
eb755805 PZ |
1505 | static int tg_nop(struct task_group *tg, void *data) |
1506 | { | |
1507 | return 0; | |
c09595f6 | 1508 | } |
eb755805 PZ |
1509 | #endif |
1510 | ||
1511 | #ifdef CONFIG_SMP | |
1512 | static unsigned long source_load(int cpu, int type); | |
1513 | static unsigned long target_load(int cpu, int type); | |
1514 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1515 | ||
1516 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1517 | { | |
1518 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1519 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1520 | |
4cd42620 SR |
1521 | if (nr_running) |
1522 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1523 | else |
1524 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1525 | |
1526 | return rq->avg_load_per_task; | |
1527 | } | |
1528 | ||
1529 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1530 | |
34d76c41 PZ |
1531 | struct update_shares_data { |
1532 | unsigned long rq_weight[NR_CPUS]; | |
1533 | }; | |
1534 | ||
1535 | static DEFINE_PER_CPU(struct update_shares_data, update_shares_data); | |
1536 | ||
c09595f6 PZ |
1537 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1538 | ||
1539 | /* | |
1540 | * Calculate and set the cpu's group shares. | |
1541 | */ | |
34d76c41 PZ |
1542 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1543 | unsigned long sd_shares, | |
1544 | unsigned long sd_rq_weight, | |
1545 | struct update_shares_data *usd) | |
18d95a28 | 1546 | { |
34d76c41 | 1547 | unsigned long shares, rq_weight; |
a5004278 | 1548 | int boost = 0; |
c09595f6 | 1549 | |
34d76c41 | 1550 | rq_weight = usd->rq_weight[cpu]; |
a5004278 PZ |
1551 | if (!rq_weight) { |
1552 | boost = 1; | |
1553 | rq_weight = NICE_0_LOAD; | |
1554 | } | |
c8cba857 | 1555 | |
c09595f6 | 1556 | /* |
a8af7246 PZ |
1557 | * \Sum_j shares_j * rq_weight_i |
1558 | * shares_i = ----------------------------- | |
1559 | * \Sum_j rq_weight_j | |
c09595f6 | 1560 | */ |
ec4e0e2f | 1561 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1562 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1563 | |
ffda12a1 PZ |
1564 | if (abs(shares - tg->se[cpu]->load.weight) > |
1565 | sysctl_sched_shares_thresh) { | |
1566 | struct rq *rq = cpu_rq(cpu); | |
1567 | unsigned long flags; | |
c09595f6 | 1568 | |
ffda12a1 | 1569 | spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1570 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1571 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1572 | __set_se_shares(tg->se[cpu], shares); |
1573 | spin_unlock_irqrestore(&rq->lock, flags); | |
1574 | } | |
18d95a28 | 1575 | } |
c09595f6 PZ |
1576 | |
1577 | /* | |
c8cba857 PZ |
1578 | * Re-compute the task group their per cpu shares over the given domain. |
1579 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1580 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1581 | */ |
eb755805 | 1582 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1583 | { |
34d76c41 PZ |
1584 | unsigned long weight, rq_weight = 0, shares = 0; |
1585 | struct update_shares_data *usd; | |
eb755805 | 1586 | struct sched_domain *sd = data; |
34d76c41 | 1587 | unsigned long flags; |
c8cba857 | 1588 | int i; |
c09595f6 | 1589 | |
34d76c41 PZ |
1590 | if (!tg->se[0]) |
1591 | return 0; | |
1592 | ||
1593 | local_irq_save(flags); | |
1594 | usd = &__get_cpu_var(update_shares_data); | |
1595 | ||
758b2cdc | 1596 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 PZ |
1597 | weight = tg->cfs_rq[i]->load.weight; |
1598 | usd->rq_weight[i] = weight; | |
1599 | ||
ec4e0e2f KC |
1600 | /* |
1601 | * If there are currently no tasks on the cpu pretend there | |
1602 | * is one of average load so that when a new task gets to | |
1603 | * run here it will not get delayed by group starvation. | |
1604 | */ | |
ec4e0e2f KC |
1605 | if (!weight) |
1606 | weight = NICE_0_LOAD; | |
1607 | ||
ec4e0e2f | 1608 | rq_weight += weight; |
c8cba857 | 1609 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1610 | } |
c09595f6 | 1611 | |
c8cba857 PZ |
1612 | if ((!shares && rq_weight) || shares > tg->shares) |
1613 | shares = tg->shares; | |
1614 | ||
1615 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1616 | shares = tg->shares; | |
c09595f6 | 1617 | |
758b2cdc | 1618 | for_each_cpu(i, sched_domain_span(sd)) |
34d76c41 PZ |
1619 | update_group_shares_cpu(tg, i, shares, rq_weight, usd); |
1620 | ||
1621 | local_irq_restore(flags); | |
eb755805 PZ |
1622 | |
1623 | return 0; | |
c09595f6 PZ |
1624 | } |
1625 | ||
1626 | /* | |
c8cba857 PZ |
1627 | * Compute the cpu's hierarchical load factor for each task group. |
1628 | * This needs to be done in a top-down fashion because the load of a child | |
1629 | * group is a fraction of its parents load. | |
c09595f6 | 1630 | */ |
eb755805 | 1631 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1632 | { |
c8cba857 | 1633 | unsigned long load; |
eb755805 | 1634 | long cpu = (long)data; |
c09595f6 | 1635 | |
c8cba857 PZ |
1636 | if (!tg->parent) { |
1637 | load = cpu_rq(cpu)->load.weight; | |
1638 | } else { | |
1639 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1640 | load *= tg->cfs_rq[cpu]->shares; | |
1641 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1642 | } | |
c09595f6 | 1643 | |
c8cba857 | 1644 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1645 | |
eb755805 | 1646 | return 0; |
c09595f6 PZ |
1647 | } |
1648 | ||
c8cba857 | 1649 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1650 | { |
e7097159 PZ |
1651 | s64 elapsed; |
1652 | u64 now; | |
1653 | ||
1654 | if (root_task_group_empty()) | |
1655 | return; | |
1656 | ||
1657 | now = cpu_clock(raw_smp_processor_id()); | |
1658 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1659 | |
1660 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1661 | sd->last_update = now; | |
eb755805 | 1662 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1663 | } |
4d8d595d PZ |
1664 | } |
1665 | ||
3e5459b4 PZ |
1666 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1667 | { | |
e7097159 PZ |
1668 | if (root_task_group_empty()) |
1669 | return; | |
1670 | ||
3e5459b4 PZ |
1671 | spin_unlock(&rq->lock); |
1672 | update_shares(sd); | |
1673 | spin_lock(&rq->lock); | |
1674 | } | |
1675 | ||
eb755805 | 1676 | static void update_h_load(long cpu) |
c09595f6 | 1677 | { |
e7097159 PZ |
1678 | if (root_task_group_empty()) |
1679 | return; | |
1680 | ||
eb755805 | 1681 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1682 | } |
1683 | ||
c09595f6 PZ |
1684 | #else |
1685 | ||
c8cba857 | 1686 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1687 | { |
1688 | } | |
1689 | ||
3e5459b4 PZ |
1690 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1691 | { | |
1692 | } | |
1693 | ||
18d95a28 PZ |
1694 | #endif |
1695 | ||
8f45e2b5 GH |
1696 | #ifdef CONFIG_PREEMPT |
1697 | ||
70574a99 | 1698 | /* |
8f45e2b5 GH |
1699 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1700 | * way at the expense of forcing extra atomic operations in all | |
1701 | * invocations. This assures that the double_lock is acquired using the | |
1702 | * same underlying policy as the spinlock_t on this architecture, which | |
1703 | * reduces latency compared to the unfair variant below. However, it | |
1704 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1705 | */ |
8f45e2b5 GH |
1706 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1707 | __releases(this_rq->lock) | |
1708 | __acquires(busiest->lock) | |
1709 | __acquires(this_rq->lock) | |
1710 | { | |
1711 | spin_unlock(&this_rq->lock); | |
1712 | double_rq_lock(this_rq, busiest); | |
1713 | ||
1714 | return 1; | |
1715 | } | |
1716 | ||
1717 | #else | |
1718 | /* | |
1719 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1720 | * latency by eliminating extra atomic operations when the locks are | |
1721 | * already in proper order on entry. This favors lower cpu-ids and will | |
1722 | * grant the double lock to lower cpus over higher ids under contention, | |
1723 | * regardless of entry order into the function. | |
1724 | */ | |
1725 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1726 | __releases(this_rq->lock) |
1727 | __acquires(busiest->lock) | |
1728 | __acquires(this_rq->lock) | |
1729 | { | |
1730 | int ret = 0; | |
1731 | ||
70574a99 AD |
1732 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1733 | if (busiest < this_rq) { | |
1734 | spin_unlock(&this_rq->lock); | |
1735 | spin_lock(&busiest->lock); | |
1736 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1737 | ret = 1; | |
1738 | } else | |
1739 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1740 | } | |
1741 | return ret; | |
1742 | } | |
1743 | ||
8f45e2b5 GH |
1744 | #endif /* CONFIG_PREEMPT */ |
1745 | ||
1746 | /* | |
1747 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1748 | */ | |
1749 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1750 | { | |
1751 | if (unlikely(!irqs_disabled())) { | |
1752 | /* printk() doesn't work good under rq->lock */ | |
1753 | spin_unlock(&this_rq->lock); | |
1754 | BUG_ON(1); | |
1755 | } | |
1756 | ||
1757 | return _double_lock_balance(this_rq, busiest); | |
1758 | } | |
1759 | ||
70574a99 AD |
1760 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1761 | __releases(busiest->lock) | |
1762 | { | |
1763 | spin_unlock(&busiest->lock); | |
1764 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1765 | } | |
18d95a28 PZ |
1766 | #endif |
1767 | ||
30432094 | 1768 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1769 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1770 | { | |
30432094 | 1771 | #ifdef CONFIG_SMP |
34e83e85 IM |
1772 | cfs_rq->shares = shares; |
1773 | #endif | |
1774 | } | |
30432094 | 1775 | #endif |
e7693a36 | 1776 | |
dce48a84 TG |
1777 | static void calc_load_account_active(struct rq *this_rq); |
1778 | ||
dd41f596 | 1779 | #include "sched_stats.h" |
dd41f596 | 1780 | #include "sched_idletask.c" |
5522d5d5 IM |
1781 | #include "sched_fair.c" |
1782 | #include "sched_rt.c" | |
dd41f596 IM |
1783 | #ifdef CONFIG_SCHED_DEBUG |
1784 | # include "sched_debug.c" | |
1785 | #endif | |
1786 | ||
1787 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1788 | #define for_each_class(class) \ |
1789 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1790 | |
c09595f6 | 1791 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1792 | { |
1793 | rq->nr_running++; | |
9c217245 IM |
1794 | } |
1795 | ||
c09595f6 | 1796 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1797 | { |
1798 | rq->nr_running--; | |
9c217245 IM |
1799 | } |
1800 | ||
45bf76df IM |
1801 | static void set_load_weight(struct task_struct *p) |
1802 | { | |
1803 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1804 | p->se.load.weight = prio_to_weight[0] * 2; |
1805 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1806 | return; | |
1807 | } | |
45bf76df | 1808 | |
dd41f596 IM |
1809 | /* |
1810 | * SCHED_IDLE tasks get minimal weight: | |
1811 | */ | |
1812 | if (p->policy == SCHED_IDLE) { | |
1813 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1814 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1815 | return; | |
1816 | } | |
71f8bd46 | 1817 | |
dd41f596 IM |
1818 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1819 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1820 | } |
1821 | ||
2087a1ad GH |
1822 | static void update_avg(u64 *avg, u64 sample) |
1823 | { | |
1824 | s64 diff = sample - *avg; | |
1825 | *avg += diff >> 3; | |
1826 | } | |
1827 | ||
8159f87e | 1828 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1829 | { |
831451ac PZ |
1830 | if (wakeup) |
1831 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1832 | ||
dd41f596 | 1833 | sched_info_queued(p); |
fd390f6a | 1834 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1835 | p->se.on_rq = 1; |
71f8bd46 IM |
1836 | } |
1837 | ||
69be72c1 | 1838 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1839 | { |
831451ac PZ |
1840 | if (sleep) { |
1841 | if (p->se.last_wakeup) { | |
1842 | update_avg(&p->se.avg_overlap, | |
1843 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1844 | p->se.last_wakeup = 0; | |
1845 | } else { | |
1846 | update_avg(&p->se.avg_wakeup, | |
1847 | sysctl_sched_wakeup_granularity); | |
1848 | } | |
2087a1ad GH |
1849 | } |
1850 | ||
46ac22ba | 1851 | sched_info_dequeued(p); |
f02231e5 | 1852 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1853 | p->se.on_rq = 0; |
71f8bd46 IM |
1854 | } |
1855 | ||
14531189 | 1856 | /* |
dd41f596 | 1857 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1858 | */ |
14531189 IM |
1859 | static inline int __normal_prio(struct task_struct *p) |
1860 | { | |
dd41f596 | 1861 | return p->static_prio; |
14531189 IM |
1862 | } |
1863 | ||
b29739f9 IM |
1864 | /* |
1865 | * Calculate the expected normal priority: i.e. priority | |
1866 | * without taking RT-inheritance into account. Might be | |
1867 | * boosted by interactivity modifiers. Changes upon fork, | |
1868 | * setprio syscalls, and whenever the interactivity | |
1869 | * estimator recalculates. | |
1870 | */ | |
36c8b586 | 1871 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1872 | { |
1873 | int prio; | |
1874 | ||
e05606d3 | 1875 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1876 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1877 | else | |
1878 | prio = __normal_prio(p); | |
1879 | return prio; | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * Calculate the current priority, i.e. the priority | |
1884 | * taken into account by the scheduler. This value might | |
1885 | * be boosted by RT tasks, or might be boosted by | |
1886 | * interactivity modifiers. Will be RT if the task got | |
1887 | * RT-boosted. If not then it returns p->normal_prio. | |
1888 | */ | |
36c8b586 | 1889 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1890 | { |
1891 | p->normal_prio = normal_prio(p); | |
1892 | /* | |
1893 | * If we are RT tasks or we were boosted to RT priority, | |
1894 | * keep the priority unchanged. Otherwise, update priority | |
1895 | * to the normal priority: | |
1896 | */ | |
1897 | if (!rt_prio(p->prio)) | |
1898 | return p->normal_prio; | |
1899 | return p->prio; | |
1900 | } | |
1901 | ||
1da177e4 | 1902 | /* |
dd41f596 | 1903 | * activate_task - move a task to the runqueue. |
1da177e4 | 1904 | */ |
dd41f596 | 1905 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1906 | { |
d9514f6c | 1907 | if (task_contributes_to_load(p)) |
dd41f596 | 1908 | rq->nr_uninterruptible--; |
1da177e4 | 1909 | |
8159f87e | 1910 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1911 | inc_nr_running(rq); |
1da177e4 LT |
1912 | } |
1913 | ||
1da177e4 LT |
1914 | /* |
1915 | * deactivate_task - remove a task from the runqueue. | |
1916 | */ | |
2e1cb74a | 1917 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1918 | { |
d9514f6c | 1919 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1920 | rq->nr_uninterruptible++; |
1921 | ||
69be72c1 | 1922 | dequeue_task(rq, p, sleep); |
c09595f6 | 1923 | dec_nr_running(rq); |
1da177e4 LT |
1924 | } |
1925 | ||
1da177e4 LT |
1926 | /** |
1927 | * task_curr - is this task currently executing on a CPU? | |
1928 | * @p: the task in question. | |
1929 | */ | |
36c8b586 | 1930 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1931 | { |
1932 | return cpu_curr(task_cpu(p)) == p; | |
1933 | } | |
1934 | ||
dd41f596 IM |
1935 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1936 | { | |
6f505b16 | 1937 | set_task_rq(p, cpu); |
dd41f596 | 1938 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1939 | /* |
1940 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1941 | * successfuly executed on another CPU. We must ensure that updates of | |
1942 | * per-task data have been completed by this moment. | |
1943 | */ | |
1944 | smp_wmb(); | |
dd41f596 | 1945 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1946 | #endif |
2dd73a4f PW |
1947 | } |
1948 | ||
cb469845 SR |
1949 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1950 | const struct sched_class *prev_class, | |
1951 | int oldprio, int running) | |
1952 | { | |
1953 | if (prev_class != p->sched_class) { | |
1954 | if (prev_class->switched_from) | |
1955 | prev_class->switched_from(rq, p, running); | |
1956 | p->sched_class->switched_to(rq, p, running); | |
1957 | } else | |
1958 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1959 | } | |
1960 | ||
1da177e4 | 1961 | #ifdef CONFIG_SMP |
c65cc870 | 1962 | |
e958b360 TG |
1963 | /* Used instead of source_load when we know the type == 0 */ |
1964 | static unsigned long weighted_cpuload(const int cpu) | |
1965 | { | |
1966 | return cpu_rq(cpu)->load.weight; | |
1967 | } | |
1968 | ||
cc367732 IM |
1969 | /* |
1970 | * Is this task likely cache-hot: | |
1971 | */ | |
e7693a36 | 1972 | static int |
cc367732 IM |
1973 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1974 | { | |
1975 | s64 delta; | |
1976 | ||
f540a608 IM |
1977 | /* |
1978 | * Buddy candidates are cache hot: | |
1979 | */ | |
4793241b PZ |
1980 | if (sched_feat(CACHE_HOT_BUDDY) && |
1981 | (&p->se == cfs_rq_of(&p->se)->next || | |
1982 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1983 | return 1; |
1984 | ||
cc367732 IM |
1985 | if (p->sched_class != &fair_sched_class) |
1986 | return 0; | |
1987 | ||
6bc1665b IM |
1988 | if (sysctl_sched_migration_cost == -1) |
1989 | return 1; | |
1990 | if (sysctl_sched_migration_cost == 0) | |
1991 | return 0; | |
1992 | ||
cc367732 IM |
1993 | delta = now - p->se.exec_start; |
1994 | ||
1995 | return delta < (s64)sysctl_sched_migration_cost; | |
1996 | } | |
1997 | ||
1998 | ||
dd41f596 | 1999 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2000 | { |
dd41f596 IM |
2001 | int old_cpu = task_cpu(p); |
2002 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
2003 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
2004 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 2005 | u64 clock_offset; |
dd41f596 IM |
2006 | |
2007 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 2008 | |
de1d7286 | 2009 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2010 | |
6cfb0d5d IM |
2011 | #ifdef CONFIG_SCHEDSTATS |
2012 | if (p->se.wait_start) | |
2013 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
2014 | if (p->se.sleep_start) |
2015 | p->se.sleep_start -= clock_offset; | |
2016 | if (p->se.block_start) | |
2017 | p->se.block_start -= clock_offset; | |
6c594c21 | 2018 | #endif |
cc367732 | 2019 | if (old_cpu != new_cpu) { |
6c594c21 | 2020 | p->se.nr_migrations++; |
23a185ca | 2021 | new_rq->nr_migrations_in++; |
6c594c21 | 2022 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
2023 | if (task_hot(p, old_rq->clock, NULL)) |
2024 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 2025 | #endif |
e5289d4a PZ |
2026 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
2027 | 1, 1, NULL, 0); | |
6c594c21 | 2028 | } |
2830cf8c SV |
2029 | p->se.vruntime -= old_cfsrq->min_vruntime - |
2030 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
2031 | |
2032 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2033 | } |
2034 | ||
70b97a7f | 2035 | struct migration_req { |
1da177e4 | 2036 | struct list_head list; |
1da177e4 | 2037 | |
36c8b586 | 2038 | struct task_struct *task; |
1da177e4 LT |
2039 | int dest_cpu; |
2040 | ||
1da177e4 | 2041 | struct completion done; |
70b97a7f | 2042 | }; |
1da177e4 LT |
2043 | |
2044 | /* | |
2045 | * The task's runqueue lock must be held. | |
2046 | * Returns true if you have to wait for migration thread. | |
2047 | */ | |
36c8b586 | 2048 | static int |
70b97a7f | 2049 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2050 | { |
70b97a7f | 2051 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2052 | |
2053 | /* | |
2054 | * If the task is not on a runqueue (and not running), then | |
2055 | * it is sufficient to simply update the task's cpu field. | |
2056 | */ | |
dd41f596 | 2057 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2058 | set_task_cpu(p, dest_cpu); |
2059 | return 0; | |
2060 | } | |
2061 | ||
2062 | init_completion(&req->done); | |
1da177e4 LT |
2063 | req->task = p; |
2064 | req->dest_cpu = dest_cpu; | |
2065 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2066 | |
1da177e4 LT |
2067 | return 1; |
2068 | } | |
2069 | ||
a26b89f0 MM |
2070 | /* |
2071 | * wait_task_context_switch - wait for a thread to complete at least one | |
2072 | * context switch. | |
2073 | * | |
2074 | * @p must not be current. | |
2075 | */ | |
2076 | void wait_task_context_switch(struct task_struct *p) | |
2077 | { | |
2078 | unsigned long nvcsw, nivcsw, flags; | |
2079 | int running; | |
2080 | struct rq *rq; | |
2081 | ||
2082 | nvcsw = p->nvcsw; | |
2083 | nivcsw = p->nivcsw; | |
2084 | for (;;) { | |
2085 | /* | |
2086 | * The runqueue is assigned before the actual context | |
2087 | * switch. We need to take the runqueue lock. | |
2088 | * | |
2089 | * We could check initially without the lock but it is | |
2090 | * very likely that we need to take the lock in every | |
2091 | * iteration. | |
2092 | */ | |
2093 | rq = task_rq_lock(p, &flags); | |
2094 | running = task_running(rq, p); | |
2095 | task_rq_unlock(rq, &flags); | |
2096 | ||
2097 | if (likely(!running)) | |
2098 | break; | |
2099 | /* | |
2100 | * The switch count is incremented before the actual | |
2101 | * context switch. We thus wait for two switches to be | |
2102 | * sure at least one completed. | |
2103 | */ | |
2104 | if ((p->nvcsw - nvcsw) > 1) | |
2105 | break; | |
2106 | if ((p->nivcsw - nivcsw) > 1) | |
2107 | break; | |
2108 | ||
2109 | cpu_relax(); | |
2110 | } | |
2111 | } | |
2112 | ||
1da177e4 LT |
2113 | /* |
2114 | * wait_task_inactive - wait for a thread to unschedule. | |
2115 | * | |
85ba2d86 RM |
2116 | * If @match_state is nonzero, it's the @p->state value just checked and |
2117 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2118 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2119 | * we return a positive number (its total switch count). If a second call | |
2120 | * a short while later returns the same number, the caller can be sure that | |
2121 | * @p has remained unscheduled the whole time. | |
2122 | * | |
1da177e4 LT |
2123 | * The caller must ensure that the task *will* unschedule sometime soon, |
2124 | * else this function might spin for a *long* time. This function can't | |
2125 | * be called with interrupts off, or it may introduce deadlock with | |
2126 | * smp_call_function() if an IPI is sent by the same process we are | |
2127 | * waiting to become inactive. | |
2128 | */ | |
85ba2d86 | 2129 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2130 | { |
2131 | unsigned long flags; | |
dd41f596 | 2132 | int running, on_rq; |
85ba2d86 | 2133 | unsigned long ncsw; |
70b97a7f | 2134 | struct rq *rq; |
1da177e4 | 2135 | |
3a5c359a AK |
2136 | for (;;) { |
2137 | /* | |
2138 | * We do the initial early heuristics without holding | |
2139 | * any task-queue locks at all. We'll only try to get | |
2140 | * the runqueue lock when things look like they will | |
2141 | * work out! | |
2142 | */ | |
2143 | rq = task_rq(p); | |
fa490cfd | 2144 | |
3a5c359a AK |
2145 | /* |
2146 | * If the task is actively running on another CPU | |
2147 | * still, just relax and busy-wait without holding | |
2148 | * any locks. | |
2149 | * | |
2150 | * NOTE! Since we don't hold any locks, it's not | |
2151 | * even sure that "rq" stays as the right runqueue! | |
2152 | * But we don't care, since "task_running()" will | |
2153 | * return false if the runqueue has changed and p | |
2154 | * is actually now running somewhere else! | |
2155 | */ | |
85ba2d86 RM |
2156 | while (task_running(rq, p)) { |
2157 | if (match_state && unlikely(p->state != match_state)) | |
2158 | return 0; | |
3a5c359a | 2159 | cpu_relax(); |
85ba2d86 | 2160 | } |
fa490cfd | 2161 | |
3a5c359a AK |
2162 | /* |
2163 | * Ok, time to look more closely! We need the rq | |
2164 | * lock now, to be *sure*. If we're wrong, we'll | |
2165 | * just go back and repeat. | |
2166 | */ | |
2167 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2168 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2169 | running = task_running(rq, p); |
2170 | on_rq = p->se.on_rq; | |
85ba2d86 | 2171 | ncsw = 0; |
f31e11d8 | 2172 | if (!match_state || p->state == match_state) |
93dcf55f | 2173 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2174 | task_rq_unlock(rq, &flags); |
fa490cfd | 2175 | |
85ba2d86 RM |
2176 | /* |
2177 | * If it changed from the expected state, bail out now. | |
2178 | */ | |
2179 | if (unlikely(!ncsw)) | |
2180 | break; | |
2181 | ||
3a5c359a AK |
2182 | /* |
2183 | * Was it really running after all now that we | |
2184 | * checked with the proper locks actually held? | |
2185 | * | |
2186 | * Oops. Go back and try again.. | |
2187 | */ | |
2188 | if (unlikely(running)) { | |
2189 | cpu_relax(); | |
2190 | continue; | |
2191 | } | |
fa490cfd | 2192 | |
3a5c359a AK |
2193 | /* |
2194 | * It's not enough that it's not actively running, | |
2195 | * it must be off the runqueue _entirely_, and not | |
2196 | * preempted! | |
2197 | * | |
80dd99b3 | 2198 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2199 | * running right now), it's preempted, and we should |
2200 | * yield - it could be a while. | |
2201 | */ | |
2202 | if (unlikely(on_rq)) { | |
2203 | schedule_timeout_uninterruptible(1); | |
2204 | continue; | |
2205 | } | |
fa490cfd | 2206 | |
3a5c359a AK |
2207 | /* |
2208 | * Ahh, all good. It wasn't running, and it wasn't | |
2209 | * runnable, which means that it will never become | |
2210 | * running in the future either. We're all done! | |
2211 | */ | |
2212 | break; | |
2213 | } | |
85ba2d86 RM |
2214 | |
2215 | return ncsw; | |
1da177e4 LT |
2216 | } |
2217 | ||
2218 | /*** | |
2219 | * kick_process - kick a running thread to enter/exit the kernel | |
2220 | * @p: the to-be-kicked thread | |
2221 | * | |
2222 | * Cause a process which is running on another CPU to enter | |
2223 | * kernel-mode, without any delay. (to get signals handled.) | |
2224 | * | |
2225 | * NOTE: this function doesnt have to take the runqueue lock, | |
2226 | * because all it wants to ensure is that the remote task enters | |
2227 | * the kernel. If the IPI races and the task has been migrated | |
2228 | * to another CPU then no harm is done and the purpose has been | |
2229 | * achieved as well. | |
2230 | */ | |
36c8b586 | 2231 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2232 | { |
2233 | int cpu; | |
2234 | ||
2235 | preempt_disable(); | |
2236 | cpu = task_cpu(p); | |
2237 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2238 | smp_send_reschedule(cpu); | |
2239 | preempt_enable(); | |
2240 | } | |
b43e3521 | 2241 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2242 | |
2243 | /* | |
2dd73a4f PW |
2244 | * Return a low guess at the load of a migration-source cpu weighted |
2245 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2246 | * |
2247 | * We want to under-estimate the load of migration sources, to | |
2248 | * balance conservatively. | |
2249 | */ | |
a9957449 | 2250 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2251 | { |
70b97a7f | 2252 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2253 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2254 | |
93b75217 | 2255 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2256 | return total; |
b910472d | 2257 | |
dd41f596 | 2258 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2259 | } |
2260 | ||
2261 | /* | |
2dd73a4f PW |
2262 | * Return a high guess at the load of a migration-target cpu weighted |
2263 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2264 | */ |
a9957449 | 2265 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2266 | { |
70b97a7f | 2267 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2268 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2269 | |
93b75217 | 2270 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2271 | return total; |
3b0bd9bc | 2272 | |
dd41f596 | 2273 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2274 | } |
2275 | ||
147cbb4b NP |
2276 | /* |
2277 | * find_idlest_group finds and returns the least busy CPU group within the | |
2278 | * domain. | |
2279 | */ | |
2280 | static struct sched_group * | |
2281 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2282 | { | |
2283 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2284 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2285 | int load_idx = sd->forkexec_idx; | |
2286 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2287 | ||
2288 | do { | |
2289 | unsigned long load, avg_load; | |
2290 | int local_group; | |
2291 | int i; | |
2292 | ||
da5a5522 | 2293 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2294 | if (!cpumask_intersects(sched_group_cpus(group), |
2295 | &p->cpus_allowed)) | |
3a5c359a | 2296 | continue; |
da5a5522 | 2297 | |
758b2cdc RR |
2298 | local_group = cpumask_test_cpu(this_cpu, |
2299 | sched_group_cpus(group)); | |
147cbb4b NP |
2300 | |
2301 | /* Tally up the load of all CPUs in the group */ | |
2302 | avg_load = 0; | |
2303 | ||
758b2cdc | 2304 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2305 | /* Bias balancing toward cpus of our domain */ |
2306 | if (local_group) | |
2307 | load = source_load(i, load_idx); | |
2308 | else | |
2309 | load = target_load(i, load_idx); | |
2310 | ||
2311 | avg_load += load; | |
2312 | } | |
2313 | ||
2314 | /* Adjust by relative CPU power of the group */ | |
18a3885f | 2315 | avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power; |
147cbb4b NP |
2316 | |
2317 | if (local_group) { | |
2318 | this_load = avg_load; | |
2319 | this = group; | |
2320 | } else if (avg_load < min_load) { | |
2321 | min_load = avg_load; | |
2322 | idlest = group; | |
2323 | } | |
3a5c359a | 2324 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2325 | |
2326 | if (!idlest || 100*this_load < imbalance*min_load) | |
2327 | return NULL; | |
2328 | return idlest; | |
2329 | } | |
2330 | ||
2331 | /* | |
0feaece9 | 2332 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2333 | */ |
95cdf3b7 | 2334 | static int |
758b2cdc | 2335 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2336 | { |
2337 | unsigned long load, min_load = ULONG_MAX; | |
2338 | int idlest = -1; | |
2339 | int i; | |
2340 | ||
da5a5522 | 2341 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2342 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2343 | load = weighted_cpuload(i); |
147cbb4b NP |
2344 | |
2345 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2346 | min_load = load; | |
2347 | idlest = i; | |
2348 | } | |
2349 | } | |
2350 | ||
2351 | return idlest; | |
2352 | } | |
2353 | ||
476d139c NP |
2354 | /* |
2355 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2356 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2357 | * SD_BALANCE_EXEC. | |
2358 | * | |
2359 | * Balance, ie. select the least loaded group. | |
2360 | * | |
2361 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2362 | * | |
2363 | * preempt must be disabled. | |
2364 | */ | |
2365 | static int sched_balance_self(int cpu, int flag) | |
2366 | { | |
2367 | struct task_struct *t = current; | |
2368 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2369 | |
c96d145e | 2370 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2371 | /* |
2372 | * If power savings logic is enabled for a domain, stop there. | |
2373 | */ | |
5c45bf27 SS |
2374 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2375 | break; | |
476d139c NP |
2376 | if (tmp->flags & flag) |
2377 | sd = tmp; | |
c96d145e | 2378 | } |
476d139c | 2379 | |
039a1c41 PZ |
2380 | if (sd) |
2381 | update_shares(sd); | |
2382 | ||
476d139c | 2383 | while (sd) { |
476d139c | 2384 | struct sched_group *group; |
1a848870 SS |
2385 | int new_cpu, weight; |
2386 | ||
2387 | if (!(sd->flags & flag)) { | |
2388 | sd = sd->child; | |
2389 | continue; | |
2390 | } | |
476d139c | 2391 | |
476d139c | 2392 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2393 | if (!group) { |
2394 | sd = sd->child; | |
2395 | continue; | |
2396 | } | |
476d139c | 2397 | |
758b2cdc | 2398 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2399 | if (new_cpu == -1 || new_cpu == cpu) { |
2400 | /* Now try balancing at a lower domain level of cpu */ | |
2401 | sd = sd->child; | |
2402 | continue; | |
2403 | } | |
476d139c | 2404 | |
1a848870 | 2405 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2406 | cpu = new_cpu; |
758b2cdc | 2407 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2408 | sd = NULL; |
476d139c | 2409 | for_each_domain(cpu, tmp) { |
758b2cdc | 2410 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2411 | break; |
2412 | if (tmp->flags & flag) | |
2413 | sd = tmp; | |
2414 | } | |
2415 | /* while loop will break here if sd == NULL */ | |
2416 | } | |
2417 | ||
2418 | return cpu; | |
2419 | } | |
2420 | ||
2421 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2422 | |
0793a61d TG |
2423 | /** |
2424 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2425 | * @p: the task to evaluate | |
2426 | * @func: the function to be called | |
2427 | * @info: the function call argument | |
2428 | * | |
2429 | * Calls the function @func when the task is currently running. This might | |
2430 | * be on the current CPU, which just calls the function directly | |
2431 | */ | |
2432 | void task_oncpu_function_call(struct task_struct *p, | |
2433 | void (*func) (void *info), void *info) | |
2434 | { | |
2435 | int cpu; | |
2436 | ||
2437 | preempt_disable(); | |
2438 | cpu = task_cpu(p); | |
2439 | if (task_curr(p)) | |
2440 | smp_call_function_single(cpu, func, info, 1); | |
2441 | preempt_enable(); | |
2442 | } | |
2443 | ||
1da177e4 LT |
2444 | /*** |
2445 | * try_to_wake_up - wake up a thread | |
2446 | * @p: the to-be-woken-up thread | |
2447 | * @state: the mask of task states that can be woken | |
2448 | * @sync: do a synchronous wakeup? | |
2449 | * | |
2450 | * Put it on the run-queue if it's not already there. The "current" | |
2451 | * thread is always on the run-queue (except when the actual | |
2452 | * re-schedule is in progress), and as such you're allowed to do | |
2453 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2454 | * runnable without the overhead of this. | |
2455 | * | |
2456 | * returns failure only if the task is already active. | |
2457 | */ | |
36c8b586 | 2458 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2459 | { |
cc367732 | 2460 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2461 | unsigned long flags; |
2462 | long old_state; | |
70b97a7f | 2463 | struct rq *rq; |
1da177e4 | 2464 | |
b85d0667 IM |
2465 | if (!sched_feat(SYNC_WAKEUPS)) |
2466 | sync = 0; | |
2467 | ||
2398f2c6 | 2468 | #ifdef CONFIG_SMP |
57310a98 | 2469 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2470 | struct sched_domain *sd; |
2471 | ||
2472 | this_cpu = raw_smp_processor_id(); | |
2473 | cpu = task_cpu(p); | |
2474 | ||
2475 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2476 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2477 | update_shares(sd); |
2478 | break; | |
2479 | } | |
2480 | } | |
2481 | } | |
2482 | #endif | |
2483 | ||
04e2f174 | 2484 | smp_wmb(); |
1da177e4 | 2485 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2486 | update_rq_clock(rq); |
1da177e4 LT |
2487 | old_state = p->state; |
2488 | if (!(old_state & state)) | |
2489 | goto out; | |
2490 | ||
dd41f596 | 2491 | if (p->se.on_rq) |
1da177e4 LT |
2492 | goto out_running; |
2493 | ||
2494 | cpu = task_cpu(p); | |
cc367732 | 2495 | orig_cpu = cpu; |
1da177e4 LT |
2496 | this_cpu = smp_processor_id(); |
2497 | ||
2498 | #ifdef CONFIG_SMP | |
2499 | if (unlikely(task_running(rq, p))) | |
2500 | goto out_activate; | |
2501 | ||
5d2f5a61 DA |
2502 | cpu = p->sched_class->select_task_rq(p, sync); |
2503 | if (cpu != orig_cpu) { | |
2504 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2505 | task_rq_unlock(rq, &flags); |
2506 | /* might preempt at this point */ | |
2507 | rq = task_rq_lock(p, &flags); | |
2508 | old_state = p->state; | |
2509 | if (!(old_state & state)) | |
2510 | goto out; | |
dd41f596 | 2511 | if (p->se.on_rq) |
1da177e4 LT |
2512 | goto out_running; |
2513 | ||
2514 | this_cpu = smp_processor_id(); | |
2515 | cpu = task_cpu(p); | |
2516 | } | |
2517 | ||
e7693a36 GH |
2518 | #ifdef CONFIG_SCHEDSTATS |
2519 | schedstat_inc(rq, ttwu_count); | |
2520 | if (cpu == this_cpu) | |
2521 | schedstat_inc(rq, ttwu_local); | |
2522 | else { | |
2523 | struct sched_domain *sd; | |
2524 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2525 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2526 | schedstat_inc(sd, ttwu_wake_remote); |
2527 | break; | |
2528 | } | |
2529 | } | |
2530 | } | |
6d6bc0ad | 2531 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2532 | |
1da177e4 LT |
2533 | out_activate: |
2534 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2535 | schedstat_inc(p, se.nr_wakeups); |
2536 | if (sync) | |
2537 | schedstat_inc(p, se.nr_wakeups_sync); | |
2538 | if (orig_cpu != cpu) | |
2539 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2540 | if (cpu == this_cpu) | |
2541 | schedstat_inc(p, se.nr_wakeups_local); | |
2542 | else | |
2543 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2544 | activate_task(rq, p, 1); |
1da177e4 LT |
2545 | success = 1; |
2546 | ||
831451ac PZ |
2547 | /* |
2548 | * Only attribute actual wakeups done by this task. | |
2549 | */ | |
2550 | if (!in_interrupt()) { | |
2551 | struct sched_entity *se = ¤t->se; | |
2552 | u64 sample = se->sum_exec_runtime; | |
2553 | ||
2554 | if (se->last_wakeup) | |
2555 | sample -= se->last_wakeup; | |
2556 | else | |
2557 | sample -= se->start_runtime; | |
2558 | update_avg(&se->avg_wakeup, sample); | |
2559 | ||
2560 | se->last_wakeup = se->sum_exec_runtime; | |
2561 | } | |
2562 | ||
1da177e4 | 2563 | out_running: |
468a15bb | 2564 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2565 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2566 | |
1da177e4 | 2567 | p->state = TASK_RUNNING; |
9a897c5a SR |
2568 | #ifdef CONFIG_SMP |
2569 | if (p->sched_class->task_wake_up) | |
2570 | p->sched_class->task_wake_up(rq, p); | |
2571 | #endif | |
1da177e4 LT |
2572 | out: |
2573 | task_rq_unlock(rq, &flags); | |
2574 | ||
2575 | return success; | |
2576 | } | |
2577 | ||
50fa610a DH |
2578 | /** |
2579 | * wake_up_process - Wake up a specific process | |
2580 | * @p: The process to be woken up. | |
2581 | * | |
2582 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2583 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2584 | * running. | |
2585 | * | |
2586 | * It may be assumed that this function implies a write memory barrier before | |
2587 | * changing the task state if and only if any tasks are woken up. | |
2588 | */ | |
7ad5b3a5 | 2589 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2590 | { |
d9514f6c | 2591 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2592 | } |
1da177e4 LT |
2593 | EXPORT_SYMBOL(wake_up_process); |
2594 | ||
7ad5b3a5 | 2595 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2596 | { |
2597 | return try_to_wake_up(p, state, 0); | |
2598 | } | |
2599 | ||
1da177e4 LT |
2600 | /* |
2601 | * Perform scheduler related setup for a newly forked process p. | |
2602 | * p is forked by current. | |
dd41f596 IM |
2603 | * |
2604 | * __sched_fork() is basic setup used by init_idle() too: | |
2605 | */ | |
2606 | static void __sched_fork(struct task_struct *p) | |
2607 | { | |
dd41f596 IM |
2608 | p->se.exec_start = 0; |
2609 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2610 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2611 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2612 | p->se.last_wakeup = 0; |
2613 | p->se.avg_overlap = 0; | |
831451ac PZ |
2614 | p->se.start_runtime = 0; |
2615 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2616 | |
2617 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2618 | p->se.wait_start = 0; |
2619 | p->se.wait_max = 0; | |
2620 | p->se.wait_count = 0; | |
2621 | p->se.wait_sum = 0; | |
2622 | ||
2623 | p->se.sleep_start = 0; | |
2624 | p->se.sleep_max = 0; | |
2625 | p->se.sum_sleep_runtime = 0; | |
2626 | ||
2627 | p->se.block_start = 0; | |
2628 | p->se.block_max = 0; | |
2629 | p->se.exec_max = 0; | |
2630 | p->se.slice_max = 0; | |
2631 | ||
2632 | p->se.nr_migrations_cold = 0; | |
2633 | p->se.nr_failed_migrations_affine = 0; | |
2634 | p->se.nr_failed_migrations_running = 0; | |
2635 | p->se.nr_failed_migrations_hot = 0; | |
2636 | p->se.nr_forced_migrations = 0; | |
2637 | p->se.nr_forced2_migrations = 0; | |
2638 | ||
2639 | p->se.nr_wakeups = 0; | |
2640 | p->se.nr_wakeups_sync = 0; | |
2641 | p->se.nr_wakeups_migrate = 0; | |
2642 | p->se.nr_wakeups_local = 0; | |
2643 | p->se.nr_wakeups_remote = 0; | |
2644 | p->se.nr_wakeups_affine = 0; | |
2645 | p->se.nr_wakeups_affine_attempts = 0; | |
2646 | p->se.nr_wakeups_passive = 0; | |
2647 | p->se.nr_wakeups_idle = 0; | |
2648 | ||
6cfb0d5d | 2649 | #endif |
476d139c | 2650 | |
fa717060 | 2651 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2652 | p->se.on_rq = 0; |
4a55bd5e | 2653 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2654 | |
e107be36 AK |
2655 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2656 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2657 | #endif | |
2658 | ||
1da177e4 LT |
2659 | /* |
2660 | * We mark the process as running here, but have not actually | |
2661 | * inserted it onto the runqueue yet. This guarantees that | |
2662 | * nobody will actually run it, and a signal or other external | |
2663 | * event cannot wake it up and insert it on the runqueue either. | |
2664 | */ | |
2665 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2666 | } |
2667 | ||
2668 | /* | |
2669 | * fork()/clone()-time setup: | |
2670 | */ | |
2671 | void sched_fork(struct task_struct *p, int clone_flags) | |
2672 | { | |
2673 | int cpu = get_cpu(); | |
2674 | ||
2675 | __sched_fork(p); | |
2676 | ||
2677 | #ifdef CONFIG_SMP | |
2678 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2679 | #endif | |
02e4bac2 | 2680 | set_task_cpu(p, cpu); |
b29739f9 IM |
2681 | |
2682 | /* | |
b9dc29e7 | 2683 | * Make sure we do not leak PI boosting priority to the child. |
b29739f9 IM |
2684 | */ |
2685 | p->prio = current->normal_prio; | |
ca94c442 | 2686 | |
b9dc29e7 MG |
2687 | /* |
2688 | * Revert to default priority/policy on fork if requested. | |
2689 | */ | |
2690 | if (unlikely(p->sched_reset_on_fork)) { | |
2691 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) | |
2692 | p->policy = SCHED_NORMAL; | |
2693 | ||
2694 | if (p->normal_prio < DEFAULT_PRIO) | |
2695 | p->prio = DEFAULT_PRIO; | |
2696 | ||
6c697bdf MG |
2697 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2698 | p->static_prio = NICE_TO_PRIO(0); | |
2699 | set_load_weight(p); | |
2700 | } | |
2701 | ||
b9dc29e7 MG |
2702 | /* |
2703 | * We don't need the reset flag anymore after the fork. It has | |
2704 | * fulfilled its duty: | |
2705 | */ | |
2706 | p->sched_reset_on_fork = 0; | |
2707 | } | |
ca94c442 | 2708 | |
2ddbf952 HS |
2709 | if (!rt_prio(p->prio)) |
2710 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2711 | |
52f17b6c | 2712 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2713 | if (likely(sched_info_on())) |
52f17b6c | 2714 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2715 | #endif |
d6077cb8 | 2716 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2717 | p->oncpu = 0; |
2718 | #endif | |
1da177e4 | 2719 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2720 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2721 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2722 | #endif |
917b627d GH |
2723 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2724 | ||
476d139c | 2725 | put_cpu(); |
1da177e4 LT |
2726 | } |
2727 | ||
2728 | /* | |
2729 | * wake_up_new_task - wake up a newly created task for the first time. | |
2730 | * | |
2731 | * This function will do some initial scheduler statistics housekeeping | |
2732 | * that must be done for every newly created context, then puts the task | |
2733 | * on the runqueue and wakes it. | |
2734 | */ | |
7ad5b3a5 | 2735 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2736 | { |
2737 | unsigned long flags; | |
dd41f596 | 2738 | struct rq *rq; |
1da177e4 LT |
2739 | |
2740 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2741 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2742 | update_rq_clock(rq); |
1da177e4 LT |
2743 | |
2744 | p->prio = effective_prio(p); | |
2745 | ||
b9dca1e0 | 2746 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2747 | activate_task(rq, p, 0); |
1da177e4 | 2748 | } else { |
1da177e4 | 2749 | /* |
dd41f596 IM |
2750 | * Let the scheduling class do new task startup |
2751 | * management (if any): | |
1da177e4 | 2752 | */ |
ee0827d8 | 2753 | p->sched_class->task_new(rq, p); |
c09595f6 | 2754 | inc_nr_running(rq); |
1da177e4 | 2755 | } |
c71dd42d | 2756 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2757 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2758 | #ifdef CONFIG_SMP |
2759 | if (p->sched_class->task_wake_up) | |
2760 | p->sched_class->task_wake_up(rq, p); | |
2761 | #endif | |
dd41f596 | 2762 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2763 | } |
2764 | ||
e107be36 AK |
2765 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2766 | ||
2767 | /** | |
80dd99b3 | 2768 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2769 | * @notifier: notifier struct to register |
e107be36 AK |
2770 | */ |
2771 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2772 | { | |
2773 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2774 | } | |
2775 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2776 | ||
2777 | /** | |
2778 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2779 | * @notifier: notifier struct to unregister |
e107be36 AK |
2780 | * |
2781 | * This is safe to call from within a preemption notifier. | |
2782 | */ | |
2783 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2784 | { | |
2785 | hlist_del(¬ifier->link); | |
2786 | } | |
2787 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2788 | ||
2789 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2790 | { | |
2791 | struct preempt_notifier *notifier; | |
2792 | struct hlist_node *node; | |
2793 | ||
2794 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2795 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2796 | } | |
2797 | ||
2798 | static void | |
2799 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2800 | struct task_struct *next) | |
2801 | { | |
2802 | struct preempt_notifier *notifier; | |
2803 | struct hlist_node *node; | |
2804 | ||
2805 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2806 | notifier->ops->sched_out(notifier, next); | |
2807 | } | |
2808 | ||
6d6bc0ad | 2809 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2810 | |
2811 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2812 | { | |
2813 | } | |
2814 | ||
2815 | static void | |
2816 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2817 | struct task_struct *next) | |
2818 | { | |
2819 | } | |
2820 | ||
6d6bc0ad | 2821 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2822 | |
4866cde0 NP |
2823 | /** |
2824 | * prepare_task_switch - prepare to switch tasks | |
2825 | * @rq: the runqueue preparing to switch | |
421cee29 | 2826 | * @prev: the current task that is being switched out |
4866cde0 NP |
2827 | * @next: the task we are going to switch to. |
2828 | * | |
2829 | * This is called with the rq lock held and interrupts off. It must | |
2830 | * be paired with a subsequent finish_task_switch after the context | |
2831 | * switch. | |
2832 | * | |
2833 | * prepare_task_switch sets up locking and calls architecture specific | |
2834 | * hooks. | |
2835 | */ | |
e107be36 AK |
2836 | static inline void |
2837 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2838 | struct task_struct *next) | |
4866cde0 | 2839 | { |
e107be36 | 2840 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2841 | prepare_lock_switch(rq, next); |
2842 | prepare_arch_switch(next); | |
2843 | } | |
2844 | ||
1da177e4 LT |
2845 | /** |
2846 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2847 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2848 | * @prev: the thread we just switched away from. |
2849 | * | |
4866cde0 NP |
2850 | * finish_task_switch must be called after the context switch, paired |
2851 | * with a prepare_task_switch call before the context switch. | |
2852 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2853 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2854 | * |
2855 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2856 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2857 | * with the lock held can cause deadlocks; see schedule() for |
2858 | * details.) | |
2859 | */ | |
a9957449 | 2860 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2861 | __releases(rq->lock) |
2862 | { | |
1da177e4 | 2863 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2864 | long prev_state; |
1da177e4 LT |
2865 | |
2866 | rq->prev_mm = NULL; | |
2867 | ||
2868 | /* | |
2869 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2870 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2871 | * schedule one last time. The schedule call will never return, and |
2872 | * the scheduled task must drop that reference. | |
c394cc9f | 2873 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2874 | * still held, otherwise prev could be scheduled on another cpu, die |
2875 | * there before we look at prev->state, and then the reference would | |
2876 | * be dropped twice. | |
2877 | * Manfred Spraul <manfred@colorfullife.com> | |
2878 | */ | |
55a101f8 | 2879 | prev_state = prev->state; |
4866cde0 | 2880 | finish_arch_switch(prev); |
0793a61d | 2881 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2882 | finish_lock_switch(rq, prev); |
e8fa1362 | 2883 | |
e107be36 | 2884 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2885 | if (mm) |
2886 | mmdrop(mm); | |
c394cc9f | 2887 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2888 | /* |
2889 | * Remove function-return probe instances associated with this | |
2890 | * task and put them back on the free list. | |
9761eea8 | 2891 | */ |
c6fd91f0 | 2892 | kprobe_flush_task(prev); |
1da177e4 | 2893 | put_task_struct(prev); |
c6fd91f0 | 2894 | } |
1da177e4 LT |
2895 | } |
2896 | ||
3f029d3c GH |
2897 | #ifdef CONFIG_SMP |
2898 | ||
2899 | /* assumes rq->lock is held */ | |
2900 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2901 | { | |
2902 | if (prev->sched_class->pre_schedule) | |
2903 | prev->sched_class->pre_schedule(rq, prev); | |
2904 | } | |
2905 | ||
2906 | /* rq->lock is NOT held, but preemption is disabled */ | |
2907 | static inline void post_schedule(struct rq *rq) | |
2908 | { | |
2909 | if (rq->post_schedule) { | |
2910 | unsigned long flags; | |
2911 | ||
2912 | spin_lock_irqsave(&rq->lock, flags); | |
2913 | if (rq->curr->sched_class->post_schedule) | |
2914 | rq->curr->sched_class->post_schedule(rq); | |
2915 | spin_unlock_irqrestore(&rq->lock, flags); | |
2916 | ||
2917 | rq->post_schedule = 0; | |
2918 | } | |
2919 | } | |
2920 | ||
2921 | #else | |
da19ab51 | 2922 | |
3f029d3c GH |
2923 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2924 | { | |
2925 | } | |
2926 | ||
2927 | static inline void post_schedule(struct rq *rq) | |
2928 | { | |
1da177e4 LT |
2929 | } |
2930 | ||
3f029d3c GH |
2931 | #endif |
2932 | ||
1da177e4 LT |
2933 | /** |
2934 | * schedule_tail - first thing a freshly forked thread must call. | |
2935 | * @prev: the thread we just switched away from. | |
2936 | */ | |
36c8b586 | 2937 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2938 | __releases(rq->lock) |
2939 | { | |
70b97a7f IM |
2940 | struct rq *rq = this_rq(); |
2941 | ||
4866cde0 | 2942 | finish_task_switch(rq, prev); |
da19ab51 | 2943 | |
3f029d3c GH |
2944 | /* |
2945 | * FIXME: do we need to worry about rq being invalidated by the | |
2946 | * task_switch? | |
2947 | */ | |
2948 | post_schedule(rq); | |
70b97a7f | 2949 | |
4866cde0 NP |
2950 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2951 | /* In this case, finish_task_switch does not reenable preemption */ | |
2952 | preempt_enable(); | |
2953 | #endif | |
1da177e4 | 2954 | if (current->set_child_tid) |
b488893a | 2955 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2956 | } |
2957 | ||
2958 | /* | |
2959 | * context_switch - switch to the new MM and the new | |
2960 | * thread's register state. | |
2961 | */ | |
dd41f596 | 2962 | static inline void |
70b97a7f | 2963 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2964 | struct task_struct *next) |
1da177e4 | 2965 | { |
dd41f596 | 2966 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2967 | |
e107be36 | 2968 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2969 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2970 | mm = next->mm; |
2971 | oldmm = prev->active_mm; | |
9226d125 ZA |
2972 | /* |
2973 | * For paravirt, this is coupled with an exit in switch_to to | |
2974 | * combine the page table reload and the switch backend into | |
2975 | * one hypercall. | |
2976 | */ | |
224101ed | 2977 | arch_start_context_switch(prev); |
9226d125 | 2978 | |
dd41f596 | 2979 | if (unlikely(!mm)) { |
1da177e4 LT |
2980 | next->active_mm = oldmm; |
2981 | atomic_inc(&oldmm->mm_count); | |
2982 | enter_lazy_tlb(oldmm, next); | |
2983 | } else | |
2984 | switch_mm(oldmm, mm, next); | |
2985 | ||
dd41f596 | 2986 | if (unlikely(!prev->mm)) { |
1da177e4 | 2987 | prev->active_mm = NULL; |
1da177e4 LT |
2988 | rq->prev_mm = oldmm; |
2989 | } | |
3a5f5e48 IM |
2990 | /* |
2991 | * Since the runqueue lock will be released by the next | |
2992 | * task (which is an invalid locking op but in the case | |
2993 | * of the scheduler it's an obvious special-case), so we | |
2994 | * do an early lockdep release here: | |
2995 | */ | |
2996 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2997 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2998 | #endif |
1da177e4 LT |
2999 | |
3000 | /* Here we just switch the register state and the stack. */ | |
3001 | switch_to(prev, next, prev); | |
3002 | ||
dd41f596 IM |
3003 | barrier(); |
3004 | /* | |
3005 | * this_rq must be evaluated again because prev may have moved | |
3006 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3007 | * frame will be invalid. | |
3008 | */ | |
3009 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3010 | } |
3011 | ||
3012 | /* | |
3013 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3014 | * | |
3015 | * externally visible scheduler statistics: current number of runnable | |
3016 | * threads, current number of uninterruptible-sleeping threads, total | |
3017 | * number of context switches performed since bootup. | |
3018 | */ | |
3019 | unsigned long nr_running(void) | |
3020 | { | |
3021 | unsigned long i, sum = 0; | |
3022 | ||
3023 | for_each_online_cpu(i) | |
3024 | sum += cpu_rq(i)->nr_running; | |
3025 | ||
3026 | return sum; | |
3027 | } | |
3028 | ||
3029 | unsigned long nr_uninterruptible(void) | |
3030 | { | |
3031 | unsigned long i, sum = 0; | |
3032 | ||
0a945022 | 3033 | for_each_possible_cpu(i) |
1da177e4 LT |
3034 | sum += cpu_rq(i)->nr_uninterruptible; |
3035 | ||
3036 | /* | |
3037 | * Since we read the counters lockless, it might be slightly | |
3038 | * inaccurate. Do not allow it to go below zero though: | |
3039 | */ | |
3040 | if (unlikely((long)sum < 0)) | |
3041 | sum = 0; | |
3042 | ||
3043 | return sum; | |
3044 | } | |
3045 | ||
3046 | unsigned long long nr_context_switches(void) | |
3047 | { | |
cc94abfc SR |
3048 | int i; |
3049 | unsigned long long sum = 0; | |
1da177e4 | 3050 | |
0a945022 | 3051 | for_each_possible_cpu(i) |
1da177e4 LT |
3052 | sum += cpu_rq(i)->nr_switches; |
3053 | ||
3054 | return sum; | |
3055 | } | |
3056 | ||
3057 | unsigned long nr_iowait(void) | |
3058 | { | |
3059 | unsigned long i, sum = 0; | |
3060 | ||
0a945022 | 3061 | for_each_possible_cpu(i) |
1da177e4 LT |
3062 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
3063 | ||
3064 | return sum; | |
3065 | } | |
3066 | ||
dce48a84 TG |
3067 | /* Variables and functions for calc_load */ |
3068 | static atomic_long_t calc_load_tasks; | |
3069 | static unsigned long calc_load_update; | |
3070 | unsigned long avenrun[3]; | |
3071 | EXPORT_SYMBOL(avenrun); | |
3072 | ||
2d02494f TG |
3073 | /** |
3074 | * get_avenrun - get the load average array | |
3075 | * @loads: pointer to dest load array | |
3076 | * @offset: offset to add | |
3077 | * @shift: shift count to shift the result left | |
3078 | * | |
3079 | * These values are estimates at best, so no need for locking. | |
3080 | */ | |
3081 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3082 | { | |
3083 | loads[0] = (avenrun[0] + offset) << shift; | |
3084 | loads[1] = (avenrun[1] + offset) << shift; | |
3085 | loads[2] = (avenrun[2] + offset) << shift; | |
3086 | } | |
3087 | ||
dce48a84 TG |
3088 | static unsigned long |
3089 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3090 | { |
dce48a84 TG |
3091 | load *= exp; |
3092 | load += active * (FIXED_1 - exp); | |
3093 | return load >> FSHIFT; | |
3094 | } | |
db1b1fef | 3095 | |
dce48a84 TG |
3096 | /* |
3097 | * calc_load - update the avenrun load estimates 10 ticks after the | |
3098 | * CPUs have updated calc_load_tasks. | |
3099 | */ | |
3100 | void calc_global_load(void) | |
3101 | { | |
3102 | unsigned long upd = calc_load_update + 10; | |
3103 | long active; | |
3104 | ||
3105 | if (time_before(jiffies, upd)) | |
3106 | return; | |
db1b1fef | 3107 | |
dce48a84 TG |
3108 | active = atomic_long_read(&calc_load_tasks); |
3109 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3110 | |
dce48a84 TG |
3111 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3112 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3113 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3114 | ||
3115 | calc_load_update += LOAD_FREQ; | |
3116 | } | |
3117 | ||
3118 | /* | |
3119 | * Either called from update_cpu_load() or from a cpu going idle | |
3120 | */ | |
3121 | static void calc_load_account_active(struct rq *this_rq) | |
3122 | { | |
3123 | long nr_active, delta; | |
3124 | ||
3125 | nr_active = this_rq->nr_running; | |
3126 | nr_active += (long) this_rq->nr_uninterruptible; | |
3127 | ||
3128 | if (nr_active != this_rq->calc_load_active) { | |
3129 | delta = nr_active - this_rq->calc_load_active; | |
3130 | this_rq->calc_load_active = nr_active; | |
3131 | atomic_long_add(delta, &calc_load_tasks); | |
3132 | } | |
db1b1fef JS |
3133 | } |
3134 | ||
23a185ca PM |
3135 | /* |
3136 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3137 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3138 | */ | |
23a185ca PM |
3139 | u64 cpu_nr_migrations(int cpu) |
3140 | { | |
3141 | return cpu_rq(cpu)->nr_migrations_in; | |
3142 | } | |
3143 | ||
48f24c4d | 3144 | /* |
dd41f596 IM |
3145 | * Update rq->cpu_load[] statistics. This function is usually called every |
3146 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3147 | */ |
dd41f596 | 3148 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3149 | { |
495eca49 | 3150 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3151 | int i, scale; |
3152 | ||
3153 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3154 | |
3155 | /* Update our load: */ | |
3156 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3157 | unsigned long old_load, new_load; | |
3158 | ||
3159 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3160 | ||
3161 | old_load = this_rq->cpu_load[i]; | |
3162 | new_load = this_load; | |
a25707f3 IM |
3163 | /* |
3164 | * Round up the averaging division if load is increasing. This | |
3165 | * prevents us from getting stuck on 9 if the load is 10, for | |
3166 | * example. | |
3167 | */ | |
3168 | if (new_load > old_load) | |
3169 | new_load += scale-1; | |
dd41f596 IM |
3170 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3171 | } | |
dce48a84 TG |
3172 | |
3173 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3174 | this_rq->calc_load_update += LOAD_FREQ; | |
3175 | calc_load_account_active(this_rq); | |
3176 | } | |
48f24c4d IM |
3177 | } |
3178 | ||
dd41f596 IM |
3179 | #ifdef CONFIG_SMP |
3180 | ||
1da177e4 LT |
3181 | /* |
3182 | * double_rq_lock - safely lock two runqueues | |
3183 | * | |
3184 | * Note this does not disable interrupts like task_rq_lock, | |
3185 | * you need to do so manually before calling. | |
3186 | */ | |
70b97a7f | 3187 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3188 | __acquires(rq1->lock) |
3189 | __acquires(rq2->lock) | |
3190 | { | |
054b9108 | 3191 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3192 | if (rq1 == rq2) { |
3193 | spin_lock(&rq1->lock); | |
3194 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3195 | } else { | |
c96d145e | 3196 | if (rq1 < rq2) { |
1da177e4 | 3197 | spin_lock(&rq1->lock); |
5e710e37 | 3198 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3199 | } else { |
3200 | spin_lock(&rq2->lock); | |
5e710e37 | 3201 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3202 | } |
3203 | } | |
6e82a3be IM |
3204 | update_rq_clock(rq1); |
3205 | update_rq_clock(rq2); | |
1da177e4 LT |
3206 | } |
3207 | ||
3208 | /* | |
3209 | * double_rq_unlock - safely unlock two runqueues | |
3210 | * | |
3211 | * Note this does not restore interrupts like task_rq_unlock, | |
3212 | * you need to do so manually after calling. | |
3213 | */ | |
70b97a7f | 3214 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3215 | __releases(rq1->lock) |
3216 | __releases(rq2->lock) | |
3217 | { | |
3218 | spin_unlock(&rq1->lock); | |
3219 | if (rq1 != rq2) | |
3220 | spin_unlock(&rq2->lock); | |
3221 | else | |
3222 | __release(rq2->lock); | |
3223 | } | |
3224 | ||
1da177e4 LT |
3225 | /* |
3226 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3227 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3228 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3229 | * the cpu_allowed mask is restored. |
3230 | */ | |
36c8b586 | 3231 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3232 | { |
70b97a7f | 3233 | struct migration_req req; |
1da177e4 | 3234 | unsigned long flags; |
70b97a7f | 3235 | struct rq *rq; |
1da177e4 LT |
3236 | |
3237 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3238 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3239 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3240 | goto out; |
3241 | ||
3242 | /* force the process onto the specified CPU */ | |
3243 | if (migrate_task(p, dest_cpu, &req)) { | |
3244 | /* Need to wait for migration thread (might exit: take ref). */ | |
3245 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3246 | |
1da177e4 LT |
3247 | get_task_struct(mt); |
3248 | task_rq_unlock(rq, &flags); | |
3249 | wake_up_process(mt); | |
3250 | put_task_struct(mt); | |
3251 | wait_for_completion(&req.done); | |
36c8b586 | 3252 | |
1da177e4 LT |
3253 | return; |
3254 | } | |
3255 | out: | |
3256 | task_rq_unlock(rq, &flags); | |
3257 | } | |
3258 | ||
3259 | /* | |
476d139c NP |
3260 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3261 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3262 | */ |
3263 | void sched_exec(void) | |
3264 | { | |
1da177e4 | 3265 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3266 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3267 | put_cpu(); |
476d139c NP |
3268 | if (new_cpu != this_cpu) |
3269 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3270 | } |
3271 | ||
3272 | /* | |
3273 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3274 | * Both runqueues must be locked. | |
3275 | */ | |
dd41f596 IM |
3276 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3277 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3278 | { |
2e1cb74a | 3279 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3280 | set_task_cpu(p, this_cpu); |
dd41f596 | 3281 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3282 | /* |
3283 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3284 | * to be always true for them. | |
3285 | */ | |
15afe09b | 3286 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3287 | } |
3288 | ||
3289 | /* | |
3290 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3291 | */ | |
858119e1 | 3292 | static |
70b97a7f | 3293 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3294 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3295 | int *all_pinned) |
1da177e4 | 3296 | { |
708dc512 | 3297 | int tsk_cache_hot = 0; |
1da177e4 LT |
3298 | /* |
3299 | * We do not migrate tasks that are: | |
3300 | * 1) running (obviously), or | |
3301 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3302 | * 3) are cache-hot on their current CPU. | |
3303 | */ | |
96f874e2 | 3304 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3305 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3306 | return 0; |
cc367732 | 3307 | } |
81026794 NP |
3308 | *all_pinned = 0; |
3309 | ||
cc367732 IM |
3310 | if (task_running(rq, p)) { |
3311 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3312 | return 0; |
cc367732 | 3313 | } |
1da177e4 | 3314 | |
da84d961 IM |
3315 | /* |
3316 | * Aggressive migration if: | |
3317 | * 1) task is cache cold, or | |
3318 | * 2) too many balance attempts have failed. | |
3319 | */ | |
3320 | ||
708dc512 LH |
3321 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3322 | if (!tsk_cache_hot || | |
3323 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3324 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3325 | if (tsk_cache_hot) { |
da84d961 | 3326 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3327 | schedstat_inc(p, se.nr_forced_migrations); |
3328 | } | |
da84d961 IM |
3329 | #endif |
3330 | return 1; | |
3331 | } | |
3332 | ||
708dc512 | 3333 | if (tsk_cache_hot) { |
cc367732 | 3334 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3335 | return 0; |
cc367732 | 3336 | } |
1da177e4 LT |
3337 | return 1; |
3338 | } | |
3339 | ||
e1d1484f PW |
3340 | static unsigned long |
3341 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3342 | unsigned long max_load_move, struct sched_domain *sd, | |
3343 | enum cpu_idle_type idle, int *all_pinned, | |
3344 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3345 | { |
051c6764 | 3346 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3347 | struct task_struct *p; |
3348 | long rem_load_move = max_load_move; | |
1da177e4 | 3349 | |
e1d1484f | 3350 | if (max_load_move == 0) |
1da177e4 LT |
3351 | goto out; |
3352 | ||
81026794 NP |
3353 | pinned = 1; |
3354 | ||
1da177e4 | 3355 | /* |
dd41f596 | 3356 | * Start the load-balancing iterator: |
1da177e4 | 3357 | */ |
dd41f596 IM |
3358 | p = iterator->start(iterator->arg); |
3359 | next: | |
b82d9fdd | 3360 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3361 | goto out; |
051c6764 PZ |
3362 | |
3363 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3364 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3365 | p = iterator->next(iterator->arg); |
3366 | goto next; | |
1da177e4 LT |
3367 | } |
3368 | ||
dd41f596 | 3369 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3370 | pulled++; |
dd41f596 | 3371 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3372 | |
7e96fa58 GH |
3373 | #ifdef CONFIG_PREEMPT |
3374 | /* | |
3375 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3376 | * will stop after the first task is pulled to minimize the critical | |
3377 | * section. | |
3378 | */ | |
3379 | if (idle == CPU_NEWLY_IDLE) | |
3380 | goto out; | |
3381 | #endif | |
3382 | ||
2dd73a4f | 3383 | /* |
b82d9fdd | 3384 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3385 | */ |
e1d1484f | 3386 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3387 | if (p->prio < *this_best_prio) |
3388 | *this_best_prio = p->prio; | |
dd41f596 IM |
3389 | p = iterator->next(iterator->arg); |
3390 | goto next; | |
1da177e4 LT |
3391 | } |
3392 | out: | |
3393 | /* | |
e1d1484f | 3394 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3395 | * so we can safely collect pull_task() stats here rather than |
3396 | * inside pull_task(). | |
3397 | */ | |
3398 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3399 | |
3400 | if (all_pinned) | |
3401 | *all_pinned = pinned; | |
e1d1484f PW |
3402 | |
3403 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3404 | } |
3405 | ||
dd41f596 | 3406 | /* |
43010659 PW |
3407 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3408 | * this_rq, as part of a balancing operation within domain "sd". | |
3409 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3410 | * |
3411 | * Called with both runqueues locked. | |
3412 | */ | |
3413 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3414 | unsigned long max_load_move, |
dd41f596 IM |
3415 | struct sched_domain *sd, enum cpu_idle_type idle, |
3416 | int *all_pinned) | |
3417 | { | |
5522d5d5 | 3418 | const struct sched_class *class = sched_class_highest; |
43010659 | 3419 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3420 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3421 | |
3422 | do { | |
43010659 PW |
3423 | total_load_moved += |
3424 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3425 | max_load_move - total_load_moved, |
a4ac01c3 | 3426 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3427 | class = class->next; |
c4acb2c0 | 3428 | |
7e96fa58 GH |
3429 | #ifdef CONFIG_PREEMPT |
3430 | /* | |
3431 | * NEWIDLE balancing is a source of latency, so preemptible | |
3432 | * kernels will stop after the first task is pulled to minimize | |
3433 | * the critical section. | |
3434 | */ | |
c4acb2c0 GH |
3435 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3436 | break; | |
7e96fa58 | 3437 | #endif |
43010659 | 3438 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3439 | |
43010659 PW |
3440 | return total_load_moved > 0; |
3441 | } | |
3442 | ||
e1d1484f PW |
3443 | static int |
3444 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3445 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3446 | struct rq_iterator *iterator) | |
3447 | { | |
3448 | struct task_struct *p = iterator->start(iterator->arg); | |
3449 | int pinned = 0; | |
3450 | ||
3451 | while (p) { | |
3452 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3453 | pull_task(busiest, p, this_rq, this_cpu); | |
3454 | /* | |
3455 | * Right now, this is only the second place pull_task() | |
3456 | * is called, so we can safely collect pull_task() | |
3457 | * stats here rather than inside pull_task(). | |
3458 | */ | |
3459 | schedstat_inc(sd, lb_gained[idle]); | |
3460 | ||
3461 | return 1; | |
3462 | } | |
3463 | p = iterator->next(iterator->arg); | |
3464 | } | |
3465 | ||
3466 | return 0; | |
3467 | } | |
3468 | ||
43010659 PW |
3469 | /* |
3470 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3471 | * part of active balancing operations within "domain". | |
3472 | * Returns 1 if successful and 0 otherwise. | |
3473 | * | |
3474 | * Called with both runqueues locked. | |
3475 | */ | |
3476 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3477 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3478 | { | |
5522d5d5 | 3479 | const struct sched_class *class; |
43010659 | 3480 | |
cde7e5ca | 3481 | for_each_class(class) { |
e1d1484f | 3482 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3483 | return 1; |
cde7e5ca | 3484 | } |
43010659 PW |
3485 | |
3486 | return 0; | |
dd41f596 | 3487 | } |
67bb6c03 | 3488 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3489 | /* |
222d656d GS |
3490 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3491 | * during load balancing. | |
1da177e4 | 3492 | */ |
222d656d GS |
3493 | struct sd_lb_stats { |
3494 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3495 | struct sched_group *this; /* Local group in this sd */ | |
3496 | unsigned long total_load; /* Total load of all groups in sd */ | |
3497 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3498 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3499 | ||
3500 | /** Statistics of this group */ | |
3501 | unsigned long this_load; | |
3502 | unsigned long this_load_per_task; | |
3503 | unsigned long this_nr_running; | |
3504 | ||
3505 | /* Statistics of the busiest group */ | |
3506 | unsigned long max_load; | |
3507 | unsigned long busiest_load_per_task; | |
3508 | unsigned long busiest_nr_running; | |
3509 | ||
3510 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3511 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3512 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3513 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3514 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3515 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3516 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3517 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3518 | #endif |
222d656d | 3519 | }; |
1da177e4 | 3520 | |
d5ac537e | 3521 | /* |
381be78f GS |
3522 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3523 | */ | |
3524 | struct sg_lb_stats { | |
3525 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3526 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3527 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3528 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3529 | unsigned long group_capacity; | |
3530 | int group_imb; /* Is there an imbalance in the group ? */ | |
3531 | }; | |
408ed066 | 3532 | |
67bb6c03 GS |
3533 | /** |
3534 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3535 | * @group: The group whose first cpu is to be returned. | |
3536 | */ | |
3537 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3538 | { | |
3539 | return cpumask_first(sched_group_cpus(group)); | |
3540 | } | |
3541 | ||
3542 | /** | |
3543 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3544 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3545 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3546 | */ | |
3547 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3548 | enum cpu_idle_type idle) | |
3549 | { | |
3550 | int load_idx; | |
3551 | ||
3552 | switch (idle) { | |
3553 | case CPU_NOT_IDLE: | |
7897986b | 3554 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3555 | break; |
3556 | ||
3557 | case CPU_NEWLY_IDLE: | |
7897986b | 3558 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3559 | break; |
3560 | default: | |
7897986b | 3561 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3562 | break; |
3563 | } | |
1da177e4 | 3564 | |
67bb6c03 GS |
3565 | return load_idx; |
3566 | } | |
1da177e4 | 3567 | |
1da177e4 | 3568 | |
c071df18 GS |
3569 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3570 | /** | |
3571 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3572 | * the given sched_domain, during load balancing. | |
3573 | * | |
3574 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3575 | * @sds: Variable containing the statistics for sd. | |
3576 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3577 | */ | |
3578 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3579 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3580 | { | |
3581 | /* | |
3582 | * Busy processors will not participate in power savings | |
3583 | * balance. | |
3584 | */ | |
3585 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3586 | sds->power_savings_balance = 0; | |
3587 | else { | |
3588 | sds->power_savings_balance = 1; | |
3589 | sds->min_nr_running = ULONG_MAX; | |
3590 | sds->leader_nr_running = 0; | |
3591 | } | |
3592 | } | |
783609c6 | 3593 | |
c071df18 GS |
3594 | /** |
3595 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3596 | * sched_domain while performing load balancing. | |
3597 | * | |
3598 | * @group: sched_group belonging to the sched_domain under consideration. | |
3599 | * @sds: Variable containing the statistics of the sched_domain | |
3600 | * @local_group: Does group contain the CPU for which we're performing | |
3601 | * load balancing ? | |
3602 | * @sgs: Variable containing the statistics of the group. | |
3603 | */ | |
3604 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3605 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3606 | { | |
408ed066 | 3607 | |
c071df18 GS |
3608 | if (!sds->power_savings_balance) |
3609 | return; | |
1da177e4 | 3610 | |
c071df18 GS |
3611 | /* |
3612 | * If the local group is idle or completely loaded | |
3613 | * no need to do power savings balance at this domain | |
3614 | */ | |
3615 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3616 | !sds->this_nr_running)) | |
3617 | sds->power_savings_balance = 0; | |
2dd73a4f | 3618 | |
c071df18 GS |
3619 | /* |
3620 | * If a group is already running at full capacity or idle, | |
3621 | * don't include that group in power savings calculations | |
3622 | */ | |
3623 | if (!sds->power_savings_balance || | |
3624 | sgs->sum_nr_running >= sgs->group_capacity || | |
3625 | !sgs->sum_nr_running) | |
3626 | return; | |
5969fe06 | 3627 | |
c071df18 GS |
3628 | /* |
3629 | * Calculate the group which has the least non-idle load. | |
3630 | * This is the group from where we need to pick up the load | |
3631 | * for saving power | |
3632 | */ | |
3633 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3634 | (sgs->sum_nr_running == sds->min_nr_running && | |
3635 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3636 | sds->group_min = group; | |
3637 | sds->min_nr_running = sgs->sum_nr_running; | |
3638 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3639 | sgs->sum_nr_running; | |
3640 | } | |
783609c6 | 3641 | |
c071df18 GS |
3642 | /* |
3643 | * Calculate the group which is almost near its | |
3644 | * capacity but still has some space to pick up some load | |
3645 | * from other group and save more power | |
3646 | */ | |
d899a789 | 3647 | if (sgs->sum_nr_running + 1 > sgs->group_capacity) |
c071df18 | 3648 | return; |
1da177e4 | 3649 | |
c071df18 GS |
3650 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3651 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3652 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3653 | sds->group_leader = group; | |
3654 | sds->leader_nr_running = sgs->sum_nr_running; | |
3655 | } | |
3656 | } | |
408ed066 | 3657 | |
c071df18 | 3658 | /** |
d5ac537e | 3659 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3660 | * @sds: Variable containing the statistics of the sched_domain |
3661 | * under consideration. | |
3662 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3663 | * @imbalance: Variable to store the imbalance. | |
3664 | * | |
d5ac537e RD |
3665 | * Description: |
3666 | * Check if we have potential to perform some power-savings balance. | |
3667 | * If yes, set the busiest group to be the least loaded group in the | |
3668 | * sched_domain, so that it's CPUs can be put to idle. | |
3669 | * | |
c071df18 GS |
3670 | * Returns 1 if there is potential to perform power-savings balance. |
3671 | * Else returns 0. | |
3672 | */ | |
3673 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3674 | int this_cpu, unsigned long *imbalance) | |
3675 | { | |
3676 | if (!sds->power_savings_balance) | |
3677 | return 0; | |
1da177e4 | 3678 | |
c071df18 GS |
3679 | if (sds->this != sds->group_leader || |
3680 | sds->group_leader == sds->group_min) | |
3681 | return 0; | |
783609c6 | 3682 | |
c071df18 GS |
3683 | *imbalance = sds->min_load_per_task; |
3684 | sds->busiest = sds->group_min; | |
1da177e4 | 3685 | |
c071df18 GS |
3686 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3687 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3688 | group_first_cpu(sds->group_leader); | |
3689 | } | |
3690 | ||
3691 | return 1; | |
1da177e4 | 3692 | |
c071df18 GS |
3693 | } |
3694 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3695 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3696 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3697 | { | |
3698 | return; | |
3699 | } | |
408ed066 | 3700 | |
c071df18 GS |
3701 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3702 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3703 | { | |
3704 | return; | |
3705 | } | |
3706 | ||
3707 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3708 | int this_cpu, unsigned long *imbalance) | |
3709 | { | |
3710 | return 0; | |
3711 | } | |
3712 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3713 | ||
e9e9250b | 3714 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
ab29230e PZ |
3715 | { |
3716 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3717 | unsigned long smt_gain = sd->smt_gain; | |
3718 | ||
3719 | smt_gain /= weight; | |
3720 | ||
3721 | return smt_gain; | |
3722 | } | |
3723 | ||
e9e9250b PZ |
3724 | unsigned long scale_rt_power(int cpu) |
3725 | { | |
3726 | struct rq *rq = cpu_rq(cpu); | |
3727 | u64 total, available; | |
3728 | ||
3729 | sched_avg_update(rq); | |
3730 | ||
3731 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3732 | available = total - rq->rt_avg; | |
3733 | ||
3734 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3735 | total = SCHED_LOAD_SCALE; | |
3736 | ||
3737 | total >>= SCHED_LOAD_SHIFT; | |
3738 | ||
3739 | return div_u64(available, total); | |
3740 | } | |
3741 | ||
ab29230e PZ |
3742 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3743 | { | |
3744 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3745 | unsigned long power = SCHED_LOAD_SCALE; | |
3746 | struct sched_group *sdg = sd->groups; | |
ab29230e PZ |
3747 | |
3748 | /* here we could scale based on cpufreq */ | |
3749 | ||
3750 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
e9e9250b | 3751 | power *= arch_scale_smt_power(sd, cpu); |
ab29230e PZ |
3752 | power >>= SCHED_LOAD_SHIFT; |
3753 | } | |
3754 | ||
e9e9250b PZ |
3755 | power *= scale_rt_power(cpu); |
3756 | power >>= SCHED_LOAD_SHIFT; | |
3757 | ||
3758 | if (!power) | |
3759 | power = 1; | |
ab29230e | 3760 | |
18a3885f | 3761 | sdg->cpu_power = power; |
ab29230e PZ |
3762 | } |
3763 | ||
3764 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3765 | { |
3766 | struct sched_domain *child = sd->child; | |
3767 | struct sched_group *group, *sdg = sd->groups; | |
d7ea17a7 | 3768 | unsigned long power; |
cc9fba7d PZ |
3769 | |
3770 | if (!child) { | |
ab29230e | 3771 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3772 | return; |
3773 | } | |
3774 | ||
d7ea17a7 | 3775 | power = 0; |
cc9fba7d PZ |
3776 | |
3777 | group = child->groups; | |
3778 | do { | |
d7ea17a7 | 3779 | power += group->cpu_power; |
cc9fba7d PZ |
3780 | group = group->next; |
3781 | } while (group != child->groups); | |
d7ea17a7 IM |
3782 | |
3783 | sdg->cpu_power = power; | |
cc9fba7d | 3784 | } |
c071df18 | 3785 | |
1f8c553d GS |
3786 | /** |
3787 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3788 | * @group: sched_group whose statistics are to be updated. | |
3789 | * @this_cpu: Cpu for which load balance is currently performed. | |
3790 | * @idle: Idle status of this_cpu | |
3791 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3792 | * @sd_idle: Idle status of the sched_domain containing group. | |
3793 | * @local_group: Does group contain this_cpu. | |
3794 | * @cpus: Set of cpus considered for load balancing. | |
3795 | * @balance: Should we balance. | |
3796 | * @sgs: variable to hold the statistics for this group. | |
3797 | */ | |
cc9fba7d PZ |
3798 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3799 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3800 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3801 | int local_group, const struct cpumask *cpus, | |
3802 | int *balance, struct sg_lb_stats *sgs) | |
3803 | { | |
3804 | unsigned long load, max_cpu_load, min_cpu_load; | |
3805 | int i; | |
3806 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3807 | unsigned long sum_avg_load_per_task; | |
3808 | unsigned long avg_load_per_task; | |
3809 | ||
cc9fba7d | 3810 | if (local_group) { |
1f8c553d | 3811 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3812 | if (balance_cpu == this_cpu) |
ab29230e | 3813 | update_group_power(sd, this_cpu); |
cc9fba7d | 3814 | } |
1f8c553d GS |
3815 | |
3816 | /* Tally up the load of all CPUs in the group */ | |
3817 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3818 | max_cpu_load = 0; | |
3819 | min_cpu_load = ~0UL; | |
408ed066 | 3820 | |
1f8c553d GS |
3821 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3822 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3823 | |
1f8c553d GS |
3824 | if (*sd_idle && rq->nr_running) |
3825 | *sd_idle = 0; | |
5c45bf27 | 3826 | |
1f8c553d | 3827 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3828 | if (local_group) { |
1f8c553d GS |
3829 | if (idle_cpu(i) && !first_idle_cpu) { |
3830 | first_idle_cpu = 1; | |
3831 | balance_cpu = i; | |
3832 | } | |
3833 | ||
3834 | load = target_load(i, load_idx); | |
3835 | } else { | |
3836 | load = source_load(i, load_idx); | |
3837 | if (load > max_cpu_load) | |
3838 | max_cpu_load = load; | |
3839 | if (min_cpu_load > load) | |
3840 | min_cpu_load = load; | |
1da177e4 | 3841 | } |
5c45bf27 | 3842 | |
1f8c553d GS |
3843 | sgs->group_load += load; |
3844 | sgs->sum_nr_running += rq->nr_running; | |
3845 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3846 | |
1f8c553d GS |
3847 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3848 | } | |
5c45bf27 | 3849 | |
1f8c553d GS |
3850 | /* |
3851 | * First idle cpu or the first cpu(busiest) in this sched group | |
3852 | * is eligible for doing load balancing at this and above | |
3853 | * domains. In the newly idle case, we will allow all the cpu's | |
3854 | * to do the newly idle load balance. | |
3855 | */ | |
3856 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3857 | balance_cpu != this_cpu && balance) { | |
3858 | *balance = 0; | |
3859 | return; | |
3860 | } | |
5c45bf27 | 3861 | |
1f8c553d | 3862 | /* Adjust by relative CPU power of the group */ |
18a3885f | 3863 | sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power; |
5c45bf27 | 3864 | |
1f8c553d GS |
3865 | |
3866 | /* | |
3867 | * Consider the group unbalanced when the imbalance is larger | |
3868 | * than the average weight of two tasks. | |
3869 | * | |
3870 | * APZ: with cgroup the avg task weight can vary wildly and | |
3871 | * might not be a suitable number - should we keep a | |
3872 | * normalized nr_running number somewhere that negates | |
3873 | * the hierarchy? | |
3874 | */ | |
18a3885f PZ |
3875 | avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) / |
3876 | group->cpu_power; | |
1f8c553d GS |
3877 | |
3878 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3879 | sgs->group_imb = 1; | |
3880 | ||
bdb94aa5 | 3881 | sgs->group_capacity = |
18a3885f | 3882 | DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE); |
1f8c553d | 3883 | } |
dd41f596 | 3884 | |
37abe198 GS |
3885 | /** |
3886 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3887 | * @sd: sched_domain whose statistics are to be updated. | |
3888 | * @this_cpu: Cpu for which load balance is currently performed. | |
3889 | * @idle: Idle status of this_cpu | |
3890 | * @sd_idle: Idle status of the sched_domain containing group. | |
3891 | * @cpus: Set of cpus considered for load balancing. | |
3892 | * @balance: Should we balance. | |
3893 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3894 | */ |
37abe198 GS |
3895 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3896 | enum cpu_idle_type idle, int *sd_idle, | |
3897 | const struct cpumask *cpus, int *balance, | |
3898 | struct sd_lb_stats *sds) | |
1da177e4 | 3899 | { |
b5d978e0 | 3900 | struct sched_domain *child = sd->child; |
222d656d | 3901 | struct sched_group *group = sd->groups; |
37abe198 | 3902 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3903 | int load_idx, prefer_sibling = 0; |
3904 | ||
3905 | if (child && child->flags & SD_PREFER_SIBLING) | |
3906 | prefer_sibling = 1; | |
222d656d | 3907 | |
c071df18 | 3908 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3909 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3910 | |
3911 | do { | |
1da177e4 | 3912 | int local_group; |
1da177e4 | 3913 | |
758b2cdc RR |
3914 | local_group = cpumask_test_cpu(this_cpu, |
3915 | sched_group_cpus(group)); | |
381be78f | 3916 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3917 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3918 | local_group, cpus, balance, &sgs); |
1da177e4 | 3919 | |
37abe198 GS |
3920 | if (local_group && balance && !(*balance)) |
3921 | return; | |
783609c6 | 3922 | |
37abe198 | 3923 | sds->total_load += sgs.group_load; |
18a3885f | 3924 | sds->total_pwr += group->cpu_power; |
1da177e4 | 3925 | |
b5d978e0 PZ |
3926 | /* |
3927 | * In case the child domain prefers tasks go to siblings | |
3928 | * first, lower the group capacity to one so that we'll try | |
3929 | * and move all the excess tasks away. | |
3930 | */ | |
3931 | if (prefer_sibling) | |
bdb94aa5 | 3932 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
1da177e4 | 3933 | |
1da177e4 | 3934 | if (local_group) { |
37abe198 GS |
3935 | sds->this_load = sgs.avg_load; |
3936 | sds->this = group; | |
3937 | sds->this_nr_running = sgs.sum_nr_running; | |
3938 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3939 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3940 | (sgs.sum_nr_running > sgs.group_capacity || |
3941 | sgs.group_imb)) { | |
37abe198 GS |
3942 | sds->max_load = sgs.avg_load; |
3943 | sds->busiest = group; | |
3944 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3945 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3946 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3947 | } |
5c45bf27 | 3948 | |
c071df18 | 3949 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3950 | group = group->next; |
3951 | } while (group != sd->groups); | |
37abe198 | 3952 | } |
1da177e4 | 3953 | |
2e6f44ae GS |
3954 | /** |
3955 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3956 | * amongst the groups of a sched_domain, during |
3957 | * load balancing. | |
2e6f44ae GS |
3958 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3959 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3960 | * @imbalance: Variable to store the imbalance. | |
3961 | */ | |
3962 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3963 | int this_cpu, unsigned long *imbalance) | |
3964 | { | |
3965 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3966 | unsigned int imbn = 2; | |
3967 | ||
3968 | if (sds->this_nr_running) { | |
3969 | sds->this_load_per_task /= sds->this_nr_running; | |
3970 | if (sds->busiest_load_per_task > | |
3971 | sds->this_load_per_task) | |
3972 | imbn = 1; | |
3973 | } else | |
3974 | sds->this_load_per_task = | |
3975 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3976 | |
2e6f44ae GS |
3977 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3978 | sds->busiest_load_per_task * imbn) { | |
3979 | *imbalance = sds->busiest_load_per_task; | |
3980 | return; | |
3981 | } | |
908a7c1b | 3982 | |
1da177e4 | 3983 | /* |
2e6f44ae GS |
3984 | * OK, we don't have enough imbalance to justify moving tasks, |
3985 | * however we may be able to increase total CPU power used by | |
3986 | * moving them. | |
1da177e4 | 3987 | */ |
2dd73a4f | 3988 | |
18a3885f | 3989 | pwr_now += sds->busiest->cpu_power * |
2e6f44ae | 3990 | min(sds->busiest_load_per_task, sds->max_load); |
18a3885f | 3991 | pwr_now += sds->this->cpu_power * |
2e6f44ae GS |
3992 | min(sds->this_load_per_task, sds->this_load); |
3993 | pwr_now /= SCHED_LOAD_SCALE; | |
3994 | ||
3995 | /* Amount of load we'd subtract */ | |
18a3885f PZ |
3996 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3997 | sds->busiest->cpu_power; | |
2e6f44ae | 3998 | if (sds->max_load > tmp) |
18a3885f | 3999 | pwr_move += sds->busiest->cpu_power * |
2e6f44ae GS |
4000 | min(sds->busiest_load_per_task, sds->max_load - tmp); |
4001 | ||
4002 | /* Amount of load we'd add */ | |
18a3885f | 4003 | if (sds->max_load * sds->busiest->cpu_power < |
2e6f44ae | 4004 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) |
18a3885f PZ |
4005 | tmp = (sds->max_load * sds->busiest->cpu_power) / |
4006 | sds->this->cpu_power; | |
2e6f44ae | 4007 | else |
18a3885f PZ |
4008 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
4009 | sds->this->cpu_power; | |
4010 | pwr_move += sds->this->cpu_power * | |
2e6f44ae GS |
4011 | min(sds->this_load_per_task, sds->this_load + tmp); |
4012 | pwr_move /= SCHED_LOAD_SCALE; | |
4013 | ||
4014 | /* Move if we gain throughput */ | |
4015 | if (pwr_move > pwr_now) | |
4016 | *imbalance = sds->busiest_load_per_task; | |
4017 | } | |
dbc523a3 GS |
4018 | |
4019 | /** | |
4020 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
4021 | * groups of a given sched_domain during load balance. | |
4022 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
4023 | * @this_cpu: Cpu for which currently load balance is being performed. | |
4024 | * @imbalance: The variable to store the imbalance. | |
4025 | */ | |
4026 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
4027 | unsigned long *imbalance) | |
4028 | { | |
4029 | unsigned long max_pull; | |
2dd73a4f PW |
4030 | /* |
4031 | * In the presence of smp nice balancing, certain scenarios can have | |
4032 | * max load less than avg load(as we skip the groups at or below | |
4033 | * its cpu_power, while calculating max_load..) | |
4034 | */ | |
dbc523a3 | 4035 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 4036 | *imbalance = 0; |
dbc523a3 | 4037 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 4038 | } |
0c117f1b SS |
4039 | |
4040 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
4041 | max_pull = min(sds->max_load - sds->avg_load, |
4042 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 4043 | |
1da177e4 | 4044 | /* How much load to actually move to equalise the imbalance */ |
18a3885f PZ |
4045 | *imbalance = min(max_pull * sds->busiest->cpu_power, |
4046 | (sds->avg_load - sds->this_load) * sds->this->cpu_power) | |
1da177e4 LT |
4047 | / SCHED_LOAD_SCALE; |
4048 | ||
2dd73a4f PW |
4049 | /* |
4050 | * if *imbalance is less than the average load per runnable task | |
4051 | * there is no gaurantee that any tasks will be moved so we'll have | |
4052 | * a think about bumping its value to force at least one task to be | |
4053 | * moved | |
4054 | */ | |
dbc523a3 GS |
4055 | if (*imbalance < sds->busiest_load_per_task) |
4056 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 4057 | |
dbc523a3 | 4058 | } |
37abe198 | 4059 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 4060 | |
b7bb4c9b GS |
4061 | /** |
4062 | * find_busiest_group - Returns the busiest group within the sched_domain | |
4063 | * if there is an imbalance. If there isn't an imbalance, and | |
4064 | * the user has opted for power-savings, it returns a group whose | |
4065 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
4066 | * such a group exists. | |
4067 | * | |
4068 | * Also calculates the amount of weighted load which should be moved | |
4069 | * to restore balance. | |
4070 | * | |
4071 | * @sd: The sched_domain whose busiest group is to be returned. | |
4072 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
4073 | * @imbalance: Variable which stores amount of weighted load which should | |
4074 | * be moved to restore balance/put a group to idle. | |
4075 | * @idle: The idle status of this_cpu. | |
4076 | * @sd_idle: The idleness of sd | |
4077 | * @cpus: The set of CPUs under consideration for load-balancing. | |
4078 | * @balance: Pointer to a variable indicating if this_cpu | |
4079 | * is the appropriate cpu to perform load balancing at this_level. | |
4080 | * | |
4081 | * Returns: - the busiest group if imbalance exists. | |
4082 | * - If no imbalance and user has opted for power-savings balance, | |
4083 | * return the least loaded group whose CPUs can be | |
4084 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
4085 | */ |
4086 | static struct sched_group * | |
4087 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
4088 | unsigned long *imbalance, enum cpu_idle_type idle, | |
4089 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
4090 | { | |
4091 | struct sd_lb_stats sds; | |
1da177e4 | 4092 | |
37abe198 | 4093 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 4094 | |
37abe198 GS |
4095 | /* |
4096 | * Compute the various statistics relavent for load balancing at | |
4097 | * this level. | |
4098 | */ | |
4099 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
4100 | balance, &sds); | |
4101 | ||
b7bb4c9b GS |
4102 | /* Cases where imbalance does not exist from POV of this_cpu */ |
4103 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
4104 | * at this level. | |
4105 | * 2) There is no busy sibling group to pull from. | |
4106 | * 3) This group is the busiest group. | |
4107 | * 4) This group is more busy than the avg busieness at this | |
4108 | * sched_domain. | |
4109 | * 5) The imbalance is within the specified limit. | |
4110 | * 6) Any rebalance would lead to ping-pong | |
4111 | */ | |
37abe198 GS |
4112 | if (balance && !(*balance)) |
4113 | goto ret; | |
1da177e4 | 4114 | |
b7bb4c9b GS |
4115 | if (!sds.busiest || sds.busiest_nr_running == 0) |
4116 | goto out_balanced; | |
1da177e4 | 4117 | |
b7bb4c9b | 4118 | if (sds.this_load >= sds.max_load) |
1da177e4 | 4119 | goto out_balanced; |
1da177e4 | 4120 | |
222d656d | 4121 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4122 | |
b7bb4c9b GS |
4123 | if (sds.this_load >= sds.avg_load) |
4124 | goto out_balanced; | |
4125 | ||
4126 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4127 | goto out_balanced; |
4128 | ||
222d656d GS |
4129 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4130 | if (sds.group_imb) | |
4131 | sds.busiest_load_per_task = | |
4132 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4133 | |
1da177e4 LT |
4134 | /* |
4135 | * We're trying to get all the cpus to the average_load, so we don't | |
4136 | * want to push ourselves above the average load, nor do we wish to | |
4137 | * reduce the max loaded cpu below the average load, as either of these | |
4138 | * actions would just result in more rebalancing later, and ping-pong | |
4139 | * tasks around. Thus we look for the minimum possible imbalance. | |
4140 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4141 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4142 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4143 | * appear as very large values with unsigned longs. |
4144 | */ | |
222d656d | 4145 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4146 | goto out_balanced; |
4147 | ||
dbc523a3 GS |
4148 | /* Looks like there is an imbalance. Compute it */ |
4149 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4150 | return sds.busiest; |
1da177e4 LT |
4151 | |
4152 | out_balanced: | |
c071df18 GS |
4153 | /* |
4154 | * There is no obvious imbalance. But check if we can do some balancing | |
4155 | * to save power. | |
4156 | */ | |
4157 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4158 | return sds.busiest; | |
783609c6 | 4159 | ret: |
1da177e4 LT |
4160 | *imbalance = 0; |
4161 | return NULL; | |
4162 | } | |
4163 | ||
bdb94aa5 PZ |
4164 | static struct sched_group *group_of(int cpu) |
4165 | { | |
4166 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
4167 | ||
4168 | if (!sd) | |
4169 | return NULL; | |
4170 | ||
4171 | return sd->groups; | |
4172 | } | |
4173 | ||
4174 | static unsigned long power_of(int cpu) | |
4175 | { | |
4176 | struct sched_group *group = group_of(cpu); | |
4177 | ||
4178 | if (!group) | |
4179 | return SCHED_LOAD_SCALE; | |
4180 | ||
18a3885f | 4181 | return group->cpu_power; |
bdb94aa5 PZ |
4182 | } |
4183 | ||
1da177e4 LT |
4184 | /* |
4185 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4186 | */ | |
70b97a7f | 4187 | static struct rq * |
d15bcfdb | 4188 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4189 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4190 | { |
70b97a7f | 4191 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4192 | unsigned long max_load = 0; |
1da177e4 LT |
4193 | int i; |
4194 | ||
758b2cdc | 4195 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4196 | unsigned long power = power_of(i); |
4197 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4198 | unsigned long wl; |
0a2966b4 | 4199 | |
96f874e2 | 4200 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4201 | continue; |
4202 | ||
48f24c4d | 4203 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4204 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4205 | wl /= power; | |
2dd73a4f | 4206 | |
bdb94aa5 | 4207 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4208 | continue; |
1da177e4 | 4209 | |
dd41f596 IM |
4210 | if (wl > max_load) { |
4211 | max_load = wl; | |
48f24c4d | 4212 | busiest = rq; |
1da177e4 LT |
4213 | } |
4214 | } | |
4215 | ||
4216 | return busiest; | |
4217 | } | |
4218 | ||
77391d71 NP |
4219 | /* |
4220 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4221 | * so long as it is large enough. | |
4222 | */ | |
4223 | #define MAX_PINNED_INTERVAL 512 | |
4224 | ||
df7c8e84 RR |
4225 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4226 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4227 | ||
1da177e4 LT |
4228 | /* |
4229 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4230 | * tasks if there is an imbalance. | |
1da177e4 | 4231 | */ |
70b97a7f | 4232 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4233 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4234 | int *balance) |
1da177e4 | 4235 | { |
43010659 | 4236 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4237 | struct sched_group *group; |
1da177e4 | 4238 | unsigned long imbalance; |
70b97a7f | 4239 | struct rq *busiest; |
fe2eea3f | 4240 | unsigned long flags; |
df7c8e84 | 4241 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4242 | |
96f874e2 | 4243 | cpumask_setall(cpus); |
7c16ec58 | 4244 | |
89c4710e SS |
4245 | /* |
4246 | * When power savings policy is enabled for the parent domain, idle | |
4247 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4248 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4249 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4250 | */ |
d15bcfdb | 4251 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4252 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4253 | sd_idle = 1; |
1da177e4 | 4254 | |
2d72376b | 4255 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4256 | |
0a2966b4 | 4257 | redo: |
c8cba857 | 4258 | update_shares(sd); |
0a2966b4 | 4259 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4260 | cpus, balance); |
783609c6 | 4261 | |
06066714 | 4262 | if (*balance == 0) |
783609c6 | 4263 | goto out_balanced; |
783609c6 | 4264 | |
1da177e4 LT |
4265 | if (!group) { |
4266 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4267 | goto out_balanced; | |
4268 | } | |
4269 | ||
7c16ec58 | 4270 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4271 | if (!busiest) { |
4272 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4273 | goto out_balanced; | |
4274 | } | |
4275 | ||
db935dbd | 4276 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4277 | |
4278 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4279 | ||
43010659 | 4280 | ld_moved = 0; |
1da177e4 LT |
4281 | if (busiest->nr_running > 1) { |
4282 | /* | |
4283 | * Attempt to move tasks. If find_busiest_group has found | |
4284 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4285 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4286 | * correctly treated as an imbalance. |
4287 | */ | |
fe2eea3f | 4288 | local_irq_save(flags); |
e17224bf | 4289 | double_rq_lock(this_rq, busiest); |
43010659 | 4290 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4291 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4292 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4293 | local_irq_restore(flags); |
81026794 | 4294 | |
46cb4b7c SS |
4295 | /* |
4296 | * some other cpu did the load balance for us. | |
4297 | */ | |
43010659 | 4298 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4299 | resched_cpu(this_cpu); |
4300 | ||
81026794 | 4301 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4302 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4303 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4304 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4305 | goto redo; |
81026794 | 4306 | goto out_balanced; |
0a2966b4 | 4307 | } |
1da177e4 | 4308 | } |
81026794 | 4309 | |
43010659 | 4310 | if (!ld_moved) { |
1da177e4 LT |
4311 | schedstat_inc(sd, lb_failed[idle]); |
4312 | sd->nr_balance_failed++; | |
4313 | ||
4314 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4315 | |
fe2eea3f | 4316 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4317 | |
4318 | /* don't kick the migration_thread, if the curr | |
4319 | * task on busiest cpu can't be moved to this_cpu | |
4320 | */ | |
96f874e2 RR |
4321 | if (!cpumask_test_cpu(this_cpu, |
4322 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4323 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4324 | all_pinned = 1; |
4325 | goto out_one_pinned; | |
4326 | } | |
4327 | ||
1da177e4 LT |
4328 | if (!busiest->active_balance) { |
4329 | busiest->active_balance = 1; | |
4330 | busiest->push_cpu = this_cpu; | |
81026794 | 4331 | active_balance = 1; |
1da177e4 | 4332 | } |
fe2eea3f | 4333 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4334 | if (active_balance) |
1da177e4 LT |
4335 | wake_up_process(busiest->migration_thread); |
4336 | ||
4337 | /* | |
4338 | * We've kicked active balancing, reset the failure | |
4339 | * counter. | |
4340 | */ | |
39507451 | 4341 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4342 | } |
81026794 | 4343 | } else |
1da177e4 LT |
4344 | sd->nr_balance_failed = 0; |
4345 | ||
81026794 | 4346 | if (likely(!active_balance)) { |
1da177e4 LT |
4347 | /* We were unbalanced, so reset the balancing interval */ |
4348 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4349 | } else { |
4350 | /* | |
4351 | * If we've begun active balancing, start to back off. This | |
4352 | * case may not be covered by the all_pinned logic if there | |
4353 | * is only 1 task on the busy runqueue (because we don't call | |
4354 | * move_tasks). | |
4355 | */ | |
4356 | if (sd->balance_interval < sd->max_interval) | |
4357 | sd->balance_interval *= 2; | |
1da177e4 LT |
4358 | } |
4359 | ||
43010659 | 4360 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4361 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4362 | ld_moved = -1; |
4363 | ||
4364 | goto out; | |
1da177e4 LT |
4365 | |
4366 | out_balanced: | |
1da177e4 LT |
4367 | schedstat_inc(sd, lb_balanced[idle]); |
4368 | ||
16cfb1c0 | 4369 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4370 | |
4371 | out_one_pinned: | |
1da177e4 | 4372 | /* tune up the balancing interval */ |
77391d71 NP |
4373 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4374 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4375 | sd->balance_interval *= 2; |
4376 | ||
48f24c4d | 4377 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4378 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4379 | ld_moved = -1; |
4380 | else | |
4381 | ld_moved = 0; | |
4382 | out: | |
c8cba857 PZ |
4383 | if (ld_moved) |
4384 | update_shares(sd); | |
c09595f6 | 4385 | return ld_moved; |
1da177e4 LT |
4386 | } |
4387 | ||
4388 | /* | |
4389 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4390 | * tasks if there is an imbalance. | |
4391 | * | |
d15bcfdb | 4392 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4393 | * this_rq is locked. |
4394 | */ | |
48f24c4d | 4395 | static int |
df7c8e84 | 4396 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4397 | { |
4398 | struct sched_group *group; | |
70b97a7f | 4399 | struct rq *busiest = NULL; |
1da177e4 | 4400 | unsigned long imbalance; |
43010659 | 4401 | int ld_moved = 0; |
5969fe06 | 4402 | int sd_idle = 0; |
969bb4e4 | 4403 | int all_pinned = 0; |
df7c8e84 | 4404 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4405 | |
96f874e2 | 4406 | cpumask_setall(cpus); |
5969fe06 | 4407 | |
89c4710e SS |
4408 | /* |
4409 | * When power savings policy is enabled for the parent domain, idle | |
4410 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4411 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4412 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4413 | */ |
4414 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4415 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4416 | sd_idle = 1; |
1da177e4 | 4417 | |
2d72376b | 4418 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4419 | redo: |
3e5459b4 | 4420 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4421 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4422 | &sd_idle, cpus, NULL); |
1da177e4 | 4423 | if (!group) { |
d15bcfdb | 4424 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4425 | goto out_balanced; |
1da177e4 LT |
4426 | } |
4427 | ||
7c16ec58 | 4428 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4429 | if (!busiest) { |
d15bcfdb | 4430 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4431 | goto out_balanced; |
1da177e4 LT |
4432 | } |
4433 | ||
db935dbd NP |
4434 | BUG_ON(busiest == this_rq); |
4435 | ||
d15bcfdb | 4436 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4437 | |
43010659 | 4438 | ld_moved = 0; |
d6d5cfaf NP |
4439 | if (busiest->nr_running > 1) { |
4440 | /* Attempt to move tasks */ | |
4441 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4442 | /* this_rq->clock is already updated */ |
4443 | update_rq_clock(busiest); | |
43010659 | 4444 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4445 | imbalance, sd, CPU_NEWLY_IDLE, |
4446 | &all_pinned); | |
1b12bbc7 | 4447 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4448 | |
969bb4e4 | 4449 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4450 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4451 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4452 | goto redo; |
4453 | } | |
d6d5cfaf NP |
4454 | } |
4455 | ||
43010659 | 4456 | if (!ld_moved) { |
36dffab6 | 4457 | int active_balance = 0; |
ad273b32 | 4458 | |
d15bcfdb | 4459 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4460 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4461 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4462 | return -1; |
ad273b32 VS |
4463 | |
4464 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4465 | return -1; | |
4466 | ||
4467 | if (sd->nr_balance_failed++ < 2) | |
4468 | return -1; | |
4469 | ||
4470 | /* | |
4471 | * The only task running in a non-idle cpu can be moved to this | |
4472 | * cpu in an attempt to completely freeup the other CPU | |
4473 | * package. The same method used to move task in load_balance() | |
4474 | * have been extended for load_balance_newidle() to speedup | |
4475 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4476 | * | |
4477 | * The package power saving logic comes from | |
4478 | * find_busiest_group(). If there are no imbalance, then | |
4479 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4480 | * f_b_g() will select a group from which a running task may be | |
4481 | * pulled to this cpu in order to make the other package idle. | |
4482 | * If there is no opportunity to make a package idle and if | |
4483 | * there are no imbalance, then f_b_g() will return NULL and no | |
4484 | * action will be taken in load_balance_newidle(). | |
4485 | * | |
4486 | * Under normal task pull operation due to imbalance, there | |
4487 | * will be more than one task in the source run queue and | |
4488 | * move_tasks() will succeed. ld_moved will be true and this | |
4489 | * active balance code will not be triggered. | |
4490 | */ | |
4491 | ||
4492 | /* Lock busiest in correct order while this_rq is held */ | |
4493 | double_lock_balance(this_rq, busiest); | |
4494 | ||
4495 | /* | |
4496 | * don't kick the migration_thread, if the curr | |
4497 | * task on busiest cpu can't be moved to this_cpu | |
4498 | */ | |
6ca09dfc | 4499 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4500 | double_unlock_balance(this_rq, busiest); |
4501 | all_pinned = 1; | |
4502 | return ld_moved; | |
4503 | } | |
4504 | ||
4505 | if (!busiest->active_balance) { | |
4506 | busiest->active_balance = 1; | |
4507 | busiest->push_cpu = this_cpu; | |
4508 | active_balance = 1; | |
4509 | } | |
4510 | ||
4511 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4512 | /* |
4513 | * Should not call ttwu while holding a rq->lock | |
4514 | */ | |
4515 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4516 | if (active_balance) |
4517 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4518 | spin_lock(&this_rq->lock); |
ad273b32 | 4519 | |
5969fe06 | 4520 | } else |
16cfb1c0 | 4521 | sd->nr_balance_failed = 0; |
1da177e4 | 4522 | |
3e5459b4 | 4523 | update_shares_locked(this_rq, sd); |
43010659 | 4524 | return ld_moved; |
16cfb1c0 NP |
4525 | |
4526 | out_balanced: | |
d15bcfdb | 4527 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4528 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4529 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4530 | return -1; |
16cfb1c0 | 4531 | sd->nr_balance_failed = 0; |
48f24c4d | 4532 | |
16cfb1c0 | 4533 | return 0; |
1da177e4 LT |
4534 | } |
4535 | ||
4536 | /* | |
4537 | * idle_balance is called by schedule() if this_cpu is about to become | |
4538 | * idle. Attempts to pull tasks from other CPUs. | |
4539 | */ | |
70b97a7f | 4540 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4541 | { |
4542 | struct sched_domain *sd; | |
efbe027e | 4543 | int pulled_task = 0; |
dd41f596 | 4544 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4545 | |
4546 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4547 | unsigned long interval; |
4548 | ||
4549 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4550 | continue; | |
4551 | ||
4552 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4553 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4554 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4555 | sd); |
92c4ca5c CL |
4556 | |
4557 | interval = msecs_to_jiffies(sd->balance_interval); | |
4558 | if (time_after(next_balance, sd->last_balance + interval)) | |
4559 | next_balance = sd->last_balance + interval; | |
4560 | if (pulled_task) | |
4561 | break; | |
1da177e4 | 4562 | } |
dd41f596 | 4563 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4564 | /* |
4565 | * We are going idle. next_balance may be set based on | |
4566 | * a busy processor. So reset next_balance. | |
4567 | */ | |
4568 | this_rq->next_balance = next_balance; | |
dd41f596 | 4569 | } |
1da177e4 LT |
4570 | } |
4571 | ||
4572 | /* | |
4573 | * active_load_balance is run by migration threads. It pushes running tasks | |
4574 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4575 | * running on each physical CPU where possible, and avoids physical / | |
4576 | * logical imbalances. | |
4577 | * | |
4578 | * Called with busiest_rq locked. | |
4579 | */ | |
70b97a7f | 4580 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4581 | { |
39507451 | 4582 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4583 | struct sched_domain *sd; |
4584 | struct rq *target_rq; | |
39507451 | 4585 | |
48f24c4d | 4586 | /* Is there any task to move? */ |
39507451 | 4587 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4588 | return; |
4589 | ||
4590 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4591 | |
4592 | /* | |
39507451 | 4593 | * This condition is "impossible", if it occurs |
41a2d6cf | 4594 | * we need to fix it. Originally reported by |
39507451 | 4595 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4596 | */ |
39507451 | 4597 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4598 | |
39507451 NP |
4599 | /* move a task from busiest_rq to target_rq */ |
4600 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4601 | update_rq_clock(busiest_rq); |
4602 | update_rq_clock(target_rq); | |
39507451 NP |
4603 | |
4604 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4605 | for_each_domain(target_cpu, sd) { |
39507451 | 4606 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4607 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4608 | break; |
c96d145e | 4609 | } |
39507451 | 4610 | |
48f24c4d | 4611 | if (likely(sd)) { |
2d72376b | 4612 | schedstat_inc(sd, alb_count); |
39507451 | 4613 | |
43010659 PW |
4614 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4615 | sd, CPU_IDLE)) | |
48f24c4d IM |
4616 | schedstat_inc(sd, alb_pushed); |
4617 | else | |
4618 | schedstat_inc(sd, alb_failed); | |
4619 | } | |
1b12bbc7 | 4620 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4621 | } |
4622 | ||
46cb4b7c SS |
4623 | #ifdef CONFIG_NO_HZ |
4624 | static struct { | |
4625 | atomic_t load_balancer; | |
7d1e6a9b | 4626 | cpumask_var_t cpu_mask; |
f711f609 | 4627 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4628 | } nohz ____cacheline_aligned = { |
4629 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4630 | }; |
4631 | ||
eea08f32 AB |
4632 | int get_nohz_load_balancer(void) |
4633 | { | |
4634 | return atomic_read(&nohz.load_balancer); | |
4635 | } | |
4636 | ||
f711f609 GS |
4637 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4638 | /** | |
4639 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4640 | * @cpu: The cpu whose lowest level of sched domain is to | |
4641 | * be returned. | |
4642 | * @flag: The flag to check for the lowest sched_domain | |
4643 | * for the given cpu. | |
4644 | * | |
4645 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4646 | */ | |
4647 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4648 | { | |
4649 | struct sched_domain *sd; | |
4650 | ||
4651 | for_each_domain(cpu, sd) | |
4652 | if (sd && (sd->flags & flag)) | |
4653 | break; | |
4654 | ||
4655 | return sd; | |
4656 | } | |
4657 | ||
4658 | /** | |
4659 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4660 | * @cpu: The cpu whose domains we're iterating over. | |
4661 | * @sd: variable holding the value of the power_savings_sd | |
4662 | * for cpu. | |
4663 | * @flag: The flag to filter the sched_domains to be iterated. | |
4664 | * | |
4665 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4666 | * set, starting from the lowest sched_domain to the highest. | |
4667 | */ | |
4668 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4669 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4670 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4671 | ||
4672 | /** | |
4673 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4674 | * @ilb_group: group to be checked for semi-idleness | |
4675 | * | |
4676 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4677 | * | |
4678 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4679 | * and atleast one non-idle CPU. This helper function checks if the given | |
4680 | * sched_group is semi-idle or not. | |
4681 | */ | |
4682 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4683 | { | |
4684 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4685 | sched_group_cpus(ilb_group)); | |
4686 | ||
4687 | /* | |
4688 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4689 | * and atleast one idle cpu. | |
4690 | */ | |
4691 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4692 | return 0; | |
4693 | ||
4694 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4695 | return 0; | |
4696 | ||
4697 | return 1; | |
4698 | } | |
4699 | /** | |
4700 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4701 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4702 | * | |
4703 | * Returns: Returns the id of the idle load balancer if it exists, | |
4704 | * Else, returns >= nr_cpu_ids. | |
4705 | * | |
4706 | * This algorithm picks the idle load balancer such that it belongs to a | |
4707 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4708 | * completely idle packages/cores just for the purpose of idle load balancing | |
4709 | * when there are other idle cpu's which are better suited for that job. | |
4710 | */ | |
4711 | static int find_new_ilb(int cpu) | |
4712 | { | |
4713 | struct sched_domain *sd; | |
4714 | struct sched_group *ilb_group; | |
4715 | ||
4716 | /* | |
4717 | * Have idle load balancer selection from semi-idle packages only | |
4718 | * when power-aware load balancing is enabled | |
4719 | */ | |
4720 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4721 | goto out_done; | |
4722 | ||
4723 | /* | |
4724 | * Optimize for the case when we have no idle CPUs or only one | |
4725 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4726 | */ | |
4727 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4728 | goto out_done; | |
4729 | ||
4730 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4731 | ilb_group = sd->groups; | |
4732 | ||
4733 | do { | |
4734 | if (is_semi_idle_group(ilb_group)) | |
4735 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4736 | ||
4737 | ilb_group = ilb_group->next; | |
4738 | ||
4739 | } while (ilb_group != sd->groups); | |
4740 | } | |
4741 | ||
4742 | out_done: | |
4743 | return cpumask_first(nohz.cpu_mask); | |
4744 | } | |
4745 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4746 | static inline int find_new_ilb(int call_cpu) | |
4747 | { | |
6e29ec57 | 4748 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4749 | } |
4750 | #endif | |
4751 | ||
7835b98b | 4752 | /* |
46cb4b7c SS |
4753 | * This routine will try to nominate the ilb (idle load balancing) |
4754 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4755 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4756 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4757 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4758 | * arrives... | |
4759 | * | |
4760 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4761 | * for idle load balancing. ilb owner will still be part of | |
4762 | * nohz.cpu_mask.. | |
7835b98b | 4763 | * |
46cb4b7c SS |
4764 | * While stopping the tick, this cpu will become the ilb owner if there |
4765 | * is no other owner. And will be the owner till that cpu becomes busy | |
4766 | * or if all cpus in the system stop their ticks at which point | |
4767 | * there is no need for ilb owner. | |
4768 | * | |
4769 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4770 | * next busy scheduler_tick() | |
4771 | */ | |
4772 | int select_nohz_load_balancer(int stop_tick) | |
4773 | { | |
4774 | int cpu = smp_processor_id(); | |
4775 | ||
4776 | if (stop_tick) { | |
46cb4b7c SS |
4777 | cpu_rq(cpu)->in_nohz_recently = 1; |
4778 | ||
483b4ee6 SS |
4779 | if (!cpu_active(cpu)) { |
4780 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4781 | return 0; | |
4782 | ||
4783 | /* | |
4784 | * If we are going offline and still the leader, | |
4785 | * give up! | |
4786 | */ | |
46cb4b7c SS |
4787 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4788 | BUG(); | |
483b4ee6 | 4789 | |
46cb4b7c SS |
4790 | return 0; |
4791 | } | |
4792 | ||
483b4ee6 SS |
4793 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4794 | ||
46cb4b7c | 4795 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4796 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4797 | if (atomic_read(&nohz.load_balancer) == cpu) |
4798 | atomic_set(&nohz.load_balancer, -1); | |
4799 | return 0; | |
4800 | } | |
4801 | ||
4802 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4803 | /* make me the ilb owner */ | |
4804 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4805 | return 1; | |
e790fb0b GS |
4806 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4807 | int new_ilb; | |
4808 | ||
4809 | if (!(sched_smt_power_savings || | |
4810 | sched_mc_power_savings)) | |
4811 | return 1; | |
4812 | /* | |
4813 | * Check to see if there is a more power-efficient | |
4814 | * ilb. | |
4815 | */ | |
4816 | new_ilb = find_new_ilb(cpu); | |
4817 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4818 | atomic_set(&nohz.load_balancer, -1); | |
4819 | resched_cpu(new_ilb); | |
4820 | return 0; | |
4821 | } | |
46cb4b7c | 4822 | return 1; |
e790fb0b | 4823 | } |
46cb4b7c | 4824 | } else { |
7d1e6a9b | 4825 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4826 | return 0; |
4827 | ||
7d1e6a9b | 4828 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4829 | |
4830 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4831 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4832 | BUG(); | |
4833 | } | |
4834 | return 0; | |
4835 | } | |
4836 | #endif | |
4837 | ||
4838 | static DEFINE_SPINLOCK(balancing); | |
4839 | ||
4840 | /* | |
7835b98b CL |
4841 | * It checks each scheduling domain to see if it is due to be balanced, |
4842 | * and initiates a balancing operation if so. | |
4843 | * | |
4844 | * Balancing parameters are set up in arch_init_sched_domains. | |
4845 | */ | |
a9957449 | 4846 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4847 | { |
46cb4b7c SS |
4848 | int balance = 1; |
4849 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4850 | unsigned long interval; |
4851 | struct sched_domain *sd; | |
46cb4b7c | 4852 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4853 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4854 | int update_next_balance = 0; |
d07355f5 | 4855 | int need_serialize; |
1da177e4 | 4856 | |
46cb4b7c | 4857 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4858 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4859 | continue; | |
4860 | ||
4861 | interval = sd->balance_interval; | |
d15bcfdb | 4862 | if (idle != CPU_IDLE) |
1da177e4 LT |
4863 | interval *= sd->busy_factor; |
4864 | ||
4865 | /* scale ms to jiffies */ | |
4866 | interval = msecs_to_jiffies(interval); | |
4867 | if (unlikely(!interval)) | |
4868 | interval = 1; | |
dd41f596 IM |
4869 | if (interval > HZ*NR_CPUS/10) |
4870 | interval = HZ*NR_CPUS/10; | |
4871 | ||
d07355f5 | 4872 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4873 | |
d07355f5 | 4874 | if (need_serialize) { |
08c183f3 CL |
4875 | if (!spin_trylock(&balancing)) |
4876 | goto out; | |
4877 | } | |
4878 | ||
c9819f45 | 4879 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4880 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4881 | /* |
4882 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4883 | * longer idle, or one of our SMT siblings is |
4884 | * not idle. | |
4885 | */ | |
d15bcfdb | 4886 | idle = CPU_NOT_IDLE; |
1da177e4 | 4887 | } |
1bd77f2d | 4888 | sd->last_balance = jiffies; |
1da177e4 | 4889 | } |
d07355f5 | 4890 | if (need_serialize) |
08c183f3 CL |
4891 | spin_unlock(&balancing); |
4892 | out: | |
f549da84 | 4893 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4894 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4895 | update_next_balance = 1; |
4896 | } | |
783609c6 SS |
4897 | |
4898 | /* | |
4899 | * Stop the load balance at this level. There is another | |
4900 | * CPU in our sched group which is doing load balancing more | |
4901 | * actively. | |
4902 | */ | |
4903 | if (!balance) | |
4904 | break; | |
1da177e4 | 4905 | } |
f549da84 SS |
4906 | |
4907 | /* | |
4908 | * next_balance will be updated only when there is a need. | |
4909 | * When the cpu is attached to null domain for ex, it will not be | |
4910 | * updated. | |
4911 | */ | |
4912 | if (likely(update_next_balance)) | |
4913 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4914 | } |
4915 | ||
4916 | /* | |
4917 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4918 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4919 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4920 | */ | |
4921 | static void run_rebalance_domains(struct softirq_action *h) | |
4922 | { | |
dd41f596 IM |
4923 | int this_cpu = smp_processor_id(); |
4924 | struct rq *this_rq = cpu_rq(this_cpu); | |
4925 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4926 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4927 | |
dd41f596 | 4928 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4929 | |
4930 | #ifdef CONFIG_NO_HZ | |
4931 | /* | |
4932 | * If this cpu is the owner for idle load balancing, then do the | |
4933 | * balancing on behalf of the other idle cpus whose ticks are | |
4934 | * stopped. | |
4935 | */ | |
dd41f596 IM |
4936 | if (this_rq->idle_at_tick && |
4937 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4938 | struct rq *rq; |
4939 | int balance_cpu; | |
4940 | ||
7d1e6a9b RR |
4941 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4942 | if (balance_cpu == this_cpu) | |
4943 | continue; | |
4944 | ||
46cb4b7c SS |
4945 | /* |
4946 | * If this cpu gets work to do, stop the load balancing | |
4947 | * work being done for other cpus. Next load | |
4948 | * balancing owner will pick it up. | |
4949 | */ | |
4950 | if (need_resched()) | |
4951 | break; | |
4952 | ||
de0cf899 | 4953 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4954 | |
4955 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4956 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4957 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4958 | } |
4959 | } | |
4960 | #endif | |
4961 | } | |
4962 | ||
8a0be9ef FW |
4963 | static inline int on_null_domain(int cpu) |
4964 | { | |
4965 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4966 | } | |
4967 | ||
46cb4b7c SS |
4968 | /* |
4969 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4970 | * | |
4971 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4972 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4973 | * if the whole system is idle. | |
4974 | */ | |
dd41f596 | 4975 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4976 | { |
46cb4b7c SS |
4977 | #ifdef CONFIG_NO_HZ |
4978 | /* | |
4979 | * If we were in the nohz mode recently and busy at the current | |
4980 | * scheduler tick, then check if we need to nominate new idle | |
4981 | * load balancer. | |
4982 | */ | |
4983 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4984 | rq->in_nohz_recently = 0; | |
4985 | ||
4986 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4987 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4988 | atomic_set(&nohz.load_balancer, -1); |
4989 | } | |
4990 | ||
4991 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4992 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4993 | |
434d53b0 | 4994 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4995 | resched_cpu(ilb); |
4996 | } | |
4997 | } | |
4998 | ||
4999 | /* | |
5000 | * If this cpu is idle and doing idle load balancing for all the | |
5001 | * cpus with ticks stopped, is it time for that to stop? | |
5002 | */ | |
5003 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 5004 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
5005 | resched_cpu(cpu); |
5006 | return; | |
5007 | } | |
5008 | ||
5009 | /* | |
5010 | * If this cpu is idle and the idle load balancing is done by | |
5011 | * someone else, then no need raise the SCHED_SOFTIRQ | |
5012 | */ | |
5013 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 5014 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
5015 | return; |
5016 | #endif | |
8a0be9ef FW |
5017 | /* Don't need to rebalance while attached to NULL domain */ |
5018 | if (time_after_eq(jiffies, rq->next_balance) && | |
5019 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 5020 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 5021 | } |
dd41f596 IM |
5022 | |
5023 | #else /* CONFIG_SMP */ | |
5024 | ||
1da177e4 LT |
5025 | /* |
5026 | * on UP we do not need to balance between CPUs: | |
5027 | */ | |
70b97a7f | 5028 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
5029 | { |
5030 | } | |
dd41f596 | 5031 | |
1da177e4 LT |
5032 | #endif |
5033 | ||
1da177e4 LT |
5034 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
5035 | ||
5036 | EXPORT_PER_CPU_SYMBOL(kstat); | |
5037 | ||
5038 | /* | |
c5f8d995 | 5039 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 5040 | * @p in case that task is currently running. |
c5f8d995 HS |
5041 | * |
5042 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 5043 | */ |
c5f8d995 HS |
5044 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
5045 | { | |
5046 | u64 ns = 0; | |
5047 | ||
5048 | if (task_current(rq, p)) { | |
5049 | update_rq_clock(rq); | |
5050 | ns = rq->clock - p->se.exec_start; | |
5051 | if ((s64)ns < 0) | |
5052 | ns = 0; | |
5053 | } | |
5054 | ||
5055 | return ns; | |
5056 | } | |
5057 | ||
bb34d92f | 5058 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 5059 | { |
1da177e4 | 5060 | unsigned long flags; |
41b86e9c | 5061 | struct rq *rq; |
bb34d92f | 5062 | u64 ns = 0; |
48f24c4d | 5063 | |
41b86e9c | 5064 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
5065 | ns = do_task_delta_exec(p, rq); |
5066 | task_rq_unlock(rq, &flags); | |
1508487e | 5067 | |
c5f8d995 HS |
5068 | return ns; |
5069 | } | |
f06febc9 | 5070 | |
c5f8d995 HS |
5071 | /* |
5072 | * Return accounted runtime for the task. | |
5073 | * In case the task is currently running, return the runtime plus current's | |
5074 | * pending runtime that have not been accounted yet. | |
5075 | */ | |
5076 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5077 | { | |
5078 | unsigned long flags; | |
5079 | struct rq *rq; | |
5080 | u64 ns = 0; | |
5081 | ||
5082 | rq = task_rq_lock(p, &flags); | |
5083 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
5084 | task_rq_unlock(rq, &flags); | |
5085 | ||
5086 | return ns; | |
5087 | } | |
48f24c4d | 5088 | |
c5f8d995 HS |
5089 | /* |
5090 | * Return sum_exec_runtime for the thread group. | |
5091 | * In case the task is currently running, return the sum plus current's | |
5092 | * pending runtime that have not been accounted yet. | |
5093 | * | |
5094 | * Note that the thread group might have other running tasks as well, | |
5095 | * so the return value not includes other pending runtime that other | |
5096 | * running tasks might have. | |
5097 | */ | |
5098 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
5099 | { | |
5100 | struct task_cputime totals; | |
5101 | unsigned long flags; | |
5102 | struct rq *rq; | |
5103 | u64 ns; | |
5104 | ||
5105 | rq = task_rq_lock(p, &flags); | |
5106 | thread_group_cputime(p, &totals); | |
5107 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 5108 | task_rq_unlock(rq, &flags); |
48f24c4d | 5109 | |
1da177e4 LT |
5110 | return ns; |
5111 | } | |
5112 | ||
1da177e4 LT |
5113 | /* |
5114 | * Account user cpu time to a process. | |
5115 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 5116 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 5117 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 5118 | */ |
457533a7 MS |
5119 | void account_user_time(struct task_struct *p, cputime_t cputime, |
5120 | cputime_t cputime_scaled) | |
1da177e4 LT |
5121 | { |
5122 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5123 | cputime64_t tmp; | |
5124 | ||
457533a7 | 5125 | /* Add user time to process. */ |
1da177e4 | 5126 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5127 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5128 | account_group_user_time(p, cputime); |
1da177e4 LT |
5129 | |
5130 | /* Add user time to cpustat. */ | |
5131 | tmp = cputime_to_cputime64(cputime); | |
5132 | if (TASK_NICE(p) > 0) | |
5133 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5134 | else | |
5135 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5136 | |
5137 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5138 | /* Account for user time used */ |
5139 | acct_update_integrals(p); | |
1da177e4 LT |
5140 | } |
5141 | ||
94886b84 LV |
5142 | /* |
5143 | * Account guest cpu time to a process. | |
5144 | * @p: the process that the cpu time gets accounted to | |
5145 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5146 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5147 | */ |
457533a7 MS |
5148 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5149 | cputime_t cputime_scaled) | |
94886b84 LV |
5150 | { |
5151 | cputime64_t tmp; | |
5152 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5153 | ||
5154 | tmp = cputime_to_cputime64(cputime); | |
5155 | ||
457533a7 | 5156 | /* Add guest time to process. */ |
94886b84 | 5157 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5158 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5159 | account_group_user_time(p, cputime); |
94886b84 LV |
5160 | p->gtime = cputime_add(p->gtime, cputime); |
5161 | ||
457533a7 | 5162 | /* Add guest time to cpustat. */ |
94886b84 LV |
5163 | cpustat->user = cputime64_add(cpustat->user, tmp); |
5164 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5165 | } | |
5166 | ||
1da177e4 LT |
5167 | /* |
5168 | * Account system cpu time to a process. | |
5169 | * @p: the process that the cpu time gets accounted to | |
5170 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5171 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5172 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5173 | */ |
5174 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5175 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5176 | { |
5177 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5178 | cputime64_t tmp; |
5179 | ||
983ed7a6 | 5180 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5181 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5182 | return; |
5183 | } | |
94886b84 | 5184 | |
457533a7 | 5185 | /* Add system time to process. */ |
1da177e4 | 5186 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5187 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5188 | account_group_system_time(p, cputime); |
1da177e4 LT |
5189 | |
5190 | /* Add system time to cpustat. */ | |
5191 | tmp = cputime_to_cputime64(cputime); | |
5192 | if (hardirq_count() - hardirq_offset) | |
5193 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5194 | else if (softirq_count()) | |
5195 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5196 | else |
79741dd3 MS |
5197 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5198 | ||
ef12fefa BR |
5199 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5200 | ||
1da177e4 LT |
5201 | /* Account for system time used */ |
5202 | acct_update_integrals(p); | |
1da177e4 LT |
5203 | } |
5204 | ||
c66f08be | 5205 | /* |
1da177e4 | 5206 | * Account for involuntary wait time. |
1da177e4 | 5207 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5208 | */ |
79741dd3 | 5209 | void account_steal_time(cputime_t cputime) |
c66f08be | 5210 | { |
79741dd3 MS |
5211 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5212 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5213 | ||
5214 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5215 | } |
5216 | ||
1da177e4 | 5217 | /* |
79741dd3 MS |
5218 | * Account for idle time. |
5219 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5220 | */ |
79741dd3 | 5221 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5222 | { |
5223 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5224 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5225 | struct rq *rq = this_rq(); |
1da177e4 | 5226 | |
79741dd3 MS |
5227 | if (atomic_read(&rq->nr_iowait) > 0) |
5228 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5229 | else | |
5230 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5231 | } |
5232 | ||
79741dd3 MS |
5233 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5234 | ||
5235 | /* | |
5236 | * Account a single tick of cpu time. | |
5237 | * @p: the process that the cpu time gets accounted to | |
5238 | * @user_tick: indicates if the tick is a user or a system tick | |
5239 | */ | |
5240 | void account_process_tick(struct task_struct *p, int user_tick) | |
5241 | { | |
5242 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5243 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5244 | struct rq *rq = this_rq(); | |
5245 | ||
5246 | if (user_tick) | |
5247 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5248 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5249 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5250 | one_jiffy_scaled); | |
5251 | else | |
5252 | account_idle_time(one_jiffy); | |
5253 | } | |
5254 | ||
5255 | /* | |
5256 | * Account multiple ticks of steal time. | |
5257 | * @p: the process from which the cpu time has been stolen | |
5258 | * @ticks: number of stolen ticks | |
5259 | */ | |
5260 | void account_steal_ticks(unsigned long ticks) | |
5261 | { | |
5262 | account_steal_time(jiffies_to_cputime(ticks)); | |
5263 | } | |
5264 | ||
5265 | /* | |
5266 | * Account multiple ticks of idle time. | |
5267 | * @ticks: number of stolen ticks | |
5268 | */ | |
5269 | void account_idle_ticks(unsigned long ticks) | |
5270 | { | |
5271 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5272 | } |
5273 | ||
79741dd3 MS |
5274 | #endif |
5275 | ||
49048622 BS |
5276 | /* |
5277 | * Use precise platform statistics if available: | |
5278 | */ | |
5279 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5280 | cputime_t task_utime(struct task_struct *p) | |
5281 | { | |
5282 | return p->utime; | |
5283 | } | |
5284 | ||
5285 | cputime_t task_stime(struct task_struct *p) | |
5286 | { | |
5287 | return p->stime; | |
5288 | } | |
5289 | #else | |
5290 | cputime_t task_utime(struct task_struct *p) | |
5291 | { | |
5292 | clock_t utime = cputime_to_clock_t(p->utime), | |
5293 | total = utime + cputime_to_clock_t(p->stime); | |
5294 | u64 temp; | |
5295 | ||
5296 | /* | |
5297 | * Use CFS's precise accounting: | |
5298 | */ | |
5299 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5300 | ||
5301 | if (total) { | |
5302 | temp *= utime; | |
5303 | do_div(temp, total); | |
5304 | } | |
5305 | utime = (clock_t)temp; | |
5306 | ||
5307 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5308 | return p->prev_utime; | |
5309 | } | |
5310 | ||
5311 | cputime_t task_stime(struct task_struct *p) | |
5312 | { | |
5313 | clock_t stime; | |
5314 | ||
5315 | /* | |
5316 | * Use CFS's precise accounting. (we subtract utime from | |
5317 | * the total, to make sure the total observed by userspace | |
5318 | * grows monotonically - apps rely on that): | |
5319 | */ | |
5320 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5321 | cputime_to_clock_t(task_utime(p)); | |
5322 | ||
5323 | if (stime >= 0) | |
5324 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5325 | ||
5326 | return p->prev_stime; | |
5327 | } | |
5328 | #endif | |
5329 | ||
5330 | inline cputime_t task_gtime(struct task_struct *p) | |
5331 | { | |
5332 | return p->gtime; | |
5333 | } | |
5334 | ||
7835b98b CL |
5335 | /* |
5336 | * This function gets called by the timer code, with HZ frequency. | |
5337 | * We call it with interrupts disabled. | |
5338 | * | |
5339 | * It also gets called by the fork code, when changing the parent's | |
5340 | * timeslices. | |
5341 | */ | |
5342 | void scheduler_tick(void) | |
5343 | { | |
7835b98b CL |
5344 | int cpu = smp_processor_id(); |
5345 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5346 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5347 | |
5348 | sched_clock_tick(); | |
dd41f596 IM |
5349 | |
5350 | spin_lock(&rq->lock); | |
3e51f33f | 5351 | update_rq_clock(rq); |
f1a438d8 | 5352 | update_cpu_load(rq); |
fa85ae24 | 5353 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5354 | spin_unlock(&rq->lock); |
7835b98b | 5355 | |
e220d2dc PZ |
5356 | perf_counter_task_tick(curr, cpu); |
5357 | ||
e418e1c2 | 5358 | #ifdef CONFIG_SMP |
dd41f596 IM |
5359 | rq->idle_at_tick = idle_cpu(cpu); |
5360 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5361 | #endif |
1da177e4 LT |
5362 | } |
5363 | ||
132380a0 | 5364 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5365 | { |
5366 | if (in_lock_functions(addr)) { | |
5367 | addr = CALLER_ADDR2; | |
5368 | if (in_lock_functions(addr)) | |
5369 | addr = CALLER_ADDR3; | |
5370 | } | |
5371 | return addr; | |
5372 | } | |
1da177e4 | 5373 | |
7e49fcce SR |
5374 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5375 | defined(CONFIG_PREEMPT_TRACER)) | |
5376 | ||
43627582 | 5377 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5378 | { |
6cd8a4bb | 5379 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5380 | /* |
5381 | * Underflow? | |
5382 | */ | |
9a11b49a IM |
5383 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5384 | return; | |
6cd8a4bb | 5385 | #endif |
1da177e4 | 5386 | preempt_count() += val; |
6cd8a4bb | 5387 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5388 | /* |
5389 | * Spinlock count overflowing soon? | |
5390 | */ | |
33859f7f MOS |
5391 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5392 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5393 | #endif |
5394 | if (preempt_count() == val) | |
5395 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5396 | } |
5397 | EXPORT_SYMBOL(add_preempt_count); | |
5398 | ||
43627582 | 5399 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5400 | { |
6cd8a4bb | 5401 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5402 | /* |
5403 | * Underflow? | |
5404 | */ | |
01e3eb82 | 5405 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5406 | return; |
1da177e4 LT |
5407 | /* |
5408 | * Is the spinlock portion underflowing? | |
5409 | */ | |
9a11b49a IM |
5410 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5411 | !(preempt_count() & PREEMPT_MASK))) | |
5412 | return; | |
6cd8a4bb | 5413 | #endif |
9a11b49a | 5414 | |
6cd8a4bb SR |
5415 | if (preempt_count() == val) |
5416 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5417 | preempt_count() -= val; |
5418 | } | |
5419 | EXPORT_SYMBOL(sub_preempt_count); | |
5420 | ||
5421 | #endif | |
5422 | ||
5423 | /* | |
dd41f596 | 5424 | * Print scheduling while atomic bug: |
1da177e4 | 5425 | */ |
dd41f596 | 5426 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5427 | { |
838225b4 SS |
5428 | struct pt_regs *regs = get_irq_regs(); |
5429 | ||
5430 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5431 | prev->comm, prev->pid, preempt_count()); | |
5432 | ||
dd41f596 | 5433 | debug_show_held_locks(prev); |
e21f5b15 | 5434 | print_modules(); |
dd41f596 IM |
5435 | if (irqs_disabled()) |
5436 | print_irqtrace_events(prev); | |
838225b4 SS |
5437 | |
5438 | if (regs) | |
5439 | show_regs(regs); | |
5440 | else | |
5441 | dump_stack(); | |
dd41f596 | 5442 | } |
1da177e4 | 5443 | |
dd41f596 IM |
5444 | /* |
5445 | * Various schedule()-time debugging checks and statistics: | |
5446 | */ | |
5447 | static inline void schedule_debug(struct task_struct *prev) | |
5448 | { | |
1da177e4 | 5449 | /* |
41a2d6cf | 5450 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5451 | * schedule() atomically, we ignore that path for now. |
5452 | * Otherwise, whine if we are scheduling when we should not be. | |
5453 | */ | |
3f33a7ce | 5454 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5455 | __schedule_bug(prev); |
5456 | ||
1da177e4 LT |
5457 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5458 | ||
2d72376b | 5459 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5460 | #ifdef CONFIG_SCHEDSTATS |
5461 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5462 | schedstat_inc(this_rq(), bkl_count); |
5463 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5464 | } |
5465 | #endif | |
dd41f596 IM |
5466 | } |
5467 | ||
df1c99d4 MG |
5468 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5469 | { | |
5470 | if (prev->state == TASK_RUNNING) { | |
5471 | u64 runtime = prev->se.sum_exec_runtime; | |
5472 | ||
5473 | runtime -= prev->se.prev_sum_exec_runtime; | |
5474 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5475 | ||
5476 | /* | |
5477 | * In order to avoid avg_overlap growing stale when we are | |
5478 | * indeed overlapping and hence not getting put to sleep, grow | |
5479 | * the avg_overlap on preemption. | |
5480 | * | |
5481 | * We use the average preemption runtime because that | |
5482 | * correlates to the amount of cache footprint a task can | |
5483 | * build up. | |
5484 | */ | |
5485 | update_avg(&prev->se.avg_overlap, runtime); | |
5486 | } | |
5487 | prev->sched_class->put_prev_task(rq, prev); | |
5488 | } | |
5489 | ||
dd41f596 IM |
5490 | /* |
5491 | * Pick up the highest-prio task: | |
5492 | */ | |
5493 | static inline struct task_struct * | |
b67802ea | 5494 | pick_next_task(struct rq *rq) |
dd41f596 | 5495 | { |
5522d5d5 | 5496 | const struct sched_class *class; |
dd41f596 | 5497 | struct task_struct *p; |
1da177e4 LT |
5498 | |
5499 | /* | |
dd41f596 IM |
5500 | * Optimization: we know that if all tasks are in |
5501 | * the fair class we can call that function directly: | |
1da177e4 | 5502 | */ |
dd41f596 | 5503 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5504 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5505 | if (likely(p)) |
5506 | return p; | |
1da177e4 LT |
5507 | } |
5508 | ||
dd41f596 IM |
5509 | class = sched_class_highest; |
5510 | for ( ; ; ) { | |
fb8d4724 | 5511 | p = class->pick_next_task(rq); |
dd41f596 IM |
5512 | if (p) |
5513 | return p; | |
5514 | /* | |
5515 | * Will never be NULL as the idle class always | |
5516 | * returns a non-NULL p: | |
5517 | */ | |
5518 | class = class->next; | |
5519 | } | |
5520 | } | |
1da177e4 | 5521 | |
dd41f596 IM |
5522 | /* |
5523 | * schedule() is the main scheduler function. | |
5524 | */ | |
ff743345 | 5525 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5526 | { |
5527 | struct task_struct *prev, *next; | |
67ca7bde | 5528 | unsigned long *switch_count; |
dd41f596 | 5529 | struct rq *rq; |
31656519 | 5530 | int cpu; |
dd41f596 | 5531 | |
ff743345 PZ |
5532 | need_resched: |
5533 | preempt_disable(); | |
dd41f596 IM |
5534 | cpu = smp_processor_id(); |
5535 | rq = cpu_rq(cpu); | |
d6714c22 | 5536 | rcu_sched_qs(cpu); |
dd41f596 IM |
5537 | prev = rq->curr; |
5538 | switch_count = &prev->nivcsw; | |
5539 | ||
5540 | release_kernel_lock(prev); | |
5541 | need_resched_nonpreemptible: | |
5542 | ||
5543 | schedule_debug(prev); | |
1da177e4 | 5544 | |
31656519 | 5545 | if (sched_feat(HRTICK)) |
f333fdc9 | 5546 | hrtick_clear(rq); |
8f4d37ec | 5547 | |
8cd162ce | 5548 | spin_lock_irq(&rq->lock); |
3e51f33f | 5549 | update_rq_clock(rq); |
1e819950 | 5550 | clear_tsk_need_resched(prev); |
1da177e4 | 5551 | |
1da177e4 | 5552 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5553 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5554 | prev->state = TASK_RUNNING; |
16882c1e | 5555 | else |
2e1cb74a | 5556 | deactivate_task(rq, prev, 1); |
dd41f596 | 5557 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5558 | } |
5559 | ||
3f029d3c | 5560 | pre_schedule(rq, prev); |
f65eda4f | 5561 | |
dd41f596 | 5562 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5563 | idle_balance(cpu, rq); |
1da177e4 | 5564 | |
df1c99d4 | 5565 | put_prev_task(rq, prev); |
b67802ea | 5566 | next = pick_next_task(rq); |
1da177e4 | 5567 | |
1da177e4 | 5568 | if (likely(prev != next)) { |
673a90a1 | 5569 | sched_info_switch(prev, next); |
564c2b21 | 5570 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5571 | |
1da177e4 LT |
5572 | rq->nr_switches++; |
5573 | rq->curr = next; | |
5574 | ++*switch_count; | |
5575 | ||
dd41f596 | 5576 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5577 | /* |
5578 | * the context switch might have flipped the stack from under | |
5579 | * us, hence refresh the local variables. | |
5580 | */ | |
5581 | cpu = smp_processor_id(); | |
5582 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5583 | } else |
5584 | spin_unlock_irq(&rq->lock); | |
5585 | ||
3f029d3c | 5586 | post_schedule(rq); |
1da177e4 | 5587 | |
8f4d37ec | 5588 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5589 | goto need_resched_nonpreemptible; |
8f4d37ec | 5590 | |
1da177e4 | 5591 | preempt_enable_no_resched(); |
ff743345 | 5592 | if (need_resched()) |
1da177e4 LT |
5593 | goto need_resched; |
5594 | } | |
1da177e4 LT |
5595 | EXPORT_SYMBOL(schedule); |
5596 | ||
0d66bf6d PZ |
5597 | #ifdef CONFIG_SMP |
5598 | /* | |
5599 | * Look out! "owner" is an entirely speculative pointer | |
5600 | * access and not reliable. | |
5601 | */ | |
5602 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5603 | { | |
5604 | unsigned int cpu; | |
5605 | struct rq *rq; | |
5606 | ||
5607 | if (!sched_feat(OWNER_SPIN)) | |
5608 | return 0; | |
5609 | ||
5610 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5611 | /* | |
5612 | * Need to access the cpu field knowing that | |
5613 | * DEBUG_PAGEALLOC could have unmapped it if | |
5614 | * the mutex owner just released it and exited. | |
5615 | */ | |
5616 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5617 | goto out; | |
5618 | #else | |
5619 | cpu = owner->cpu; | |
5620 | #endif | |
5621 | ||
5622 | /* | |
5623 | * Even if the access succeeded (likely case), | |
5624 | * the cpu field may no longer be valid. | |
5625 | */ | |
5626 | if (cpu >= nr_cpumask_bits) | |
5627 | goto out; | |
5628 | ||
5629 | /* | |
5630 | * We need to validate that we can do a | |
5631 | * get_cpu() and that we have the percpu area. | |
5632 | */ | |
5633 | if (!cpu_online(cpu)) | |
5634 | goto out; | |
5635 | ||
5636 | rq = cpu_rq(cpu); | |
5637 | ||
5638 | for (;;) { | |
5639 | /* | |
5640 | * Owner changed, break to re-assess state. | |
5641 | */ | |
5642 | if (lock->owner != owner) | |
5643 | break; | |
5644 | ||
5645 | /* | |
5646 | * Is that owner really running on that cpu? | |
5647 | */ | |
5648 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5649 | return 0; | |
5650 | ||
5651 | cpu_relax(); | |
5652 | } | |
5653 | out: | |
5654 | return 1; | |
5655 | } | |
5656 | #endif | |
5657 | ||
1da177e4 LT |
5658 | #ifdef CONFIG_PREEMPT |
5659 | /* | |
2ed6e34f | 5660 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5661 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5662 | * occur there and call schedule directly. |
5663 | */ | |
5664 | asmlinkage void __sched preempt_schedule(void) | |
5665 | { | |
5666 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5667 | |
1da177e4 LT |
5668 | /* |
5669 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5670 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5671 | */ |
beed33a8 | 5672 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5673 | return; |
5674 | ||
3a5c359a AK |
5675 | do { |
5676 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5677 | schedule(); |
3a5c359a | 5678 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5679 | |
3a5c359a AK |
5680 | /* |
5681 | * Check again in case we missed a preemption opportunity | |
5682 | * between schedule and now. | |
5683 | */ | |
5684 | barrier(); | |
5ed0cec0 | 5685 | } while (need_resched()); |
1da177e4 | 5686 | } |
1da177e4 LT |
5687 | EXPORT_SYMBOL(preempt_schedule); |
5688 | ||
5689 | /* | |
2ed6e34f | 5690 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5691 | * off of irq context. |
5692 | * Note, that this is called and return with irqs disabled. This will | |
5693 | * protect us against recursive calling from irq. | |
5694 | */ | |
5695 | asmlinkage void __sched preempt_schedule_irq(void) | |
5696 | { | |
5697 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5698 | |
2ed6e34f | 5699 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5700 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5701 | ||
3a5c359a AK |
5702 | do { |
5703 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5704 | local_irq_enable(); |
5705 | schedule(); | |
5706 | local_irq_disable(); | |
3a5c359a | 5707 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5708 | |
3a5c359a AK |
5709 | /* |
5710 | * Check again in case we missed a preemption opportunity | |
5711 | * between schedule and now. | |
5712 | */ | |
5713 | barrier(); | |
5ed0cec0 | 5714 | } while (need_resched()); |
1da177e4 LT |
5715 | } |
5716 | ||
5717 | #endif /* CONFIG_PREEMPT */ | |
5718 | ||
95cdf3b7 IM |
5719 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5720 | void *key) | |
1da177e4 | 5721 | { |
48f24c4d | 5722 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5723 | } |
1da177e4 LT |
5724 | EXPORT_SYMBOL(default_wake_function); |
5725 | ||
5726 | /* | |
41a2d6cf IM |
5727 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5728 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5729 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5730 | * | |
5731 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5732 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5733 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5734 | */ | |
78ddb08f | 5735 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5736 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5737 | { |
2e45874c | 5738 | wait_queue_t *curr, *next; |
1da177e4 | 5739 | |
2e45874c | 5740 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5741 | unsigned flags = curr->flags; |
5742 | ||
1da177e4 | 5743 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5744 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5745 | break; |
5746 | } | |
5747 | } | |
5748 | ||
5749 | /** | |
5750 | * __wake_up - wake up threads blocked on a waitqueue. | |
5751 | * @q: the waitqueue | |
5752 | * @mode: which threads | |
5753 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5754 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5755 | * |
5756 | * It may be assumed that this function implies a write memory barrier before | |
5757 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5758 | */ |
7ad5b3a5 | 5759 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5760 | int nr_exclusive, void *key) |
1da177e4 LT |
5761 | { |
5762 | unsigned long flags; | |
5763 | ||
5764 | spin_lock_irqsave(&q->lock, flags); | |
5765 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5766 | spin_unlock_irqrestore(&q->lock, flags); | |
5767 | } | |
1da177e4 LT |
5768 | EXPORT_SYMBOL(__wake_up); |
5769 | ||
5770 | /* | |
5771 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5772 | */ | |
7ad5b3a5 | 5773 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5774 | { |
5775 | __wake_up_common(q, mode, 1, 0, NULL); | |
5776 | } | |
5777 | ||
4ede816a DL |
5778 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5779 | { | |
5780 | __wake_up_common(q, mode, 1, 0, key); | |
5781 | } | |
5782 | ||
1da177e4 | 5783 | /** |
4ede816a | 5784 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5785 | * @q: the waitqueue |
5786 | * @mode: which threads | |
5787 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5788 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5789 | * |
5790 | * The sync wakeup differs that the waker knows that it will schedule | |
5791 | * away soon, so while the target thread will be woken up, it will not | |
5792 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5793 | * with each other. This can prevent needless bouncing between CPUs. | |
5794 | * | |
5795 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5796 | * |
5797 | * It may be assumed that this function implies a write memory barrier before | |
5798 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5799 | */ |
4ede816a DL |
5800 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5801 | int nr_exclusive, void *key) | |
1da177e4 LT |
5802 | { |
5803 | unsigned long flags; | |
5804 | int sync = 1; | |
5805 | ||
5806 | if (unlikely(!q)) | |
5807 | return; | |
5808 | ||
5809 | if (unlikely(!nr_exclusive)) | |
5810 | sync = 0; | |
5811 | ||
5812 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5813 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5814 | spin_unlock_irqrestore(&q->lock, flags); |
5815 | } | |
4ede816a DL |
5816 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5817 | ||
5818 | /* | |
5819 | * __wake_up_sync - see __wake_up_sync_key() | |
5820 | */ | |
5821 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5822 | { | |
5823 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5824 | } | |
1da177e4 LT |
5825 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5826 | ||
65eb3dc6 KD |
5827 | /** |
5828 | * complete: - signals a single thread waiting on this completion | |
5829 | * @x: holds the state of this particular completion | |
5830 | * | |
5831 | * This will wake up a single thread waiting on this completion. Threads will be | |
5832 | * awakened in the same order in which they were queued. | |
5833 | * | |
5834 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5835 | * |
5836 | * It may be assumed that this function implies a write memory barrier before | |
5837 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5838 | */ |
b15136e9 | 5839 | void complete(struct completion *x) |
1da177e4 LT |
5840 | { |
5841 | unsigned long flags; | |
5842 | ||
5843 | spin_lock_irqsave(&x->wait.lock, flags); | |
5844 | x->done++; | |
d9514f6c | 5845 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5846 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5847 | } | |
5848 | EXPORT_SYMBOL(complete); | |
5849 | ||
65eb3dc6 KD |
5850 | /** |
5851 | * complete_all: - signals all threads waiting on this completion | |
5852 | * @x: holds the state of this particular completion | |
5853 | * | |
5854 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5855 | * |
5856 | * It may be assumed that this function implies a write memory barrier before | |
5857 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5858 | */ |
b15136e9 | 5859 | void complete_all(struct completion *x) |
1da177e4 LT |
5860 | { |
5861 | unsigned long flags; | |
5862 | ||
5863 | spin_lock_irqsave(&x->wait.lock, flags); | |
5864 | x->done += UINT_MAX/2; | |
d9514f6c | 5865 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5866 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5867 | } | |
5868 | EXPORT_SYMBOL(complete_all); | |
5869 | ||
8cbbe86d AK |
5870 | static inline long __sched |
5871 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5872 | { |
1da177e4 LT |
5873 | if (!x->done) { |
5874 | DECLARE_WAITQUEUE(wait, current); | |
5875 | ||
5876 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5877 | __add_wait_queue_tail(&x->wait, &wait); | |
5878 | do { | |
94d3d824 | 5879 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5880 | timeout = -ERESTARTSYS; |
5881 | break; | |
8cbbe86d AK |
5882 | } |
5883 | __set_current_state(state); | |
1da177e4 LT |
5884 | spin_unlock_irq(&x->wait.lock); |
5885 | timeout = schedule_timeout(timeout); | |
5886 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5887 | } while (!x->done && timeout); |
1da177e4 | 5888 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5889 | if (!x->done) |
5890 | return timeout; | |
1da177e4 LT |
5891 | } |
5892 | x->done--; | |
ea71a546 | 5893 | return timeout ?: 1; |
1da177e4 | 5894 | } |
1da177e4 | 5895 | |
8cbbe86d AK |
5896 | static long __sched |
5897 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5898 | { |
1da177e4 LT |
5899 | might_sleep(); |
5900 | ||
5901 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5902 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5903 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5904 | return timeout; |
5905 | } | |
1da177e4 | 5906 | |
65eb3dc6 KD |
5907 | /** |
5908 | * wait_for_completion: - waits for completion of a task | |
5909 | * @x: holds the state of this particular completion | |
5910 | * | |
5911 | * This waits to be signaled for completion of a specific task. It is NOT | |
5912 | * interruptible and there is no timeout. | |
5913 | * | |
5914 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5915 | * and interrupt capability. Also see complete(). | |
5916 | */ | |
b15136e9 | 5917 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5918 | { |
5919 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5920 | } |
8cbbe86d | 5921 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5922 | |
65eb3dc6 KD |
5923 | /** |
5924 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5925 | * @x: holds the state of this particular completion | |
5926 | * @timeout: timeout value in jiffies | |
5927 | * | |
5928 | * This waits for either a completion of a specific task to be signaled or for a | |
5929 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5930 | * interruptible. | |
5931 | */ | |
b15136e9 | 5932 | unsigned long __sched |
8cbbe86d | 5933 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5934 | { |
8cbbe86d | 5935 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5936 | } |
8cbbe86d | 5937 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5938 | |
65eb3dc6 KD |
5939 | /** |
5940 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5941 | * @x: holds the state of this particular completion | |
5942 | * | |
5943 | * This waits for completion of a specific task to be signaled. It is | |
5944 | * interruptible. | |
5945 | */ | |
8cbbe86d | 5946 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5947 | { |
51e97990 AK |
5948 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5949 | if (t == -ERESTARTSYS) | |
5950 | return t; | |
5951 | return 0; | |
0fec171c | 5952 | } |
8cbbe86d | 5953 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5954 | |
65eb3dc6 KD |
5955 | /** |
5956 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5957 | * @x: holds the state of this particular completion | |
5958 | * @timeout: timeout value in jiffies | |
5959 | * | |
5960 | * This waits for either a completion of a specific task to be signaled or for a | |
5961 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5962 | */ | |
b15136e9 | 5963 | unsigned long __sched |
8cbbe86d AK |
5964 | wait_for_completion_interruptible_timeout(struct completion *x, |
5965 | unsigned long timeout) | |
0fec171c | 5966 | { |
8cbbe86d | 5967 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5968 | } |
8cbbe86d | 5969 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5970 | |
65eb3dc6 KD |
5971 | /** |
5972 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5973 | * @x: holds the state of this particular completion | |
5974 | * | |
5975 | * This waits to be signaled for completion of a specific task. It can be | |
5976 | * interrupted by a kill signal. | |
5977 | */ | |
009e577e MW |
5978 | int __sched wait_for_completion_killable(struct completion *x) |
5979 | { | |
5980 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5981 | if (t == -ERESTARTSYS) | |
5982 | return t; | |
5983 | return 0; | |
5984 | } | |
5985 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5986 | ||
be4de352 DC |
5987 | /** |
5988 | * try_wait_for_completion - try to decrement a completion without blocking | |
5989 | * @x: completion structure | |
5990 | * | |
5991 | * Returns: 0 if a decrement cannot be done without blocking | |
5992 | * 1 if a decrement succeeded. | |
5993 | * | |
5994 | * If a completion is being used as a counting completion, | |
5995 | * attempt to decrement the counter without blocking. This | |
5996 | * enables us to avoid waiting if the resource the completion | |
5997 | * is protecting is not available. | |
5998 | */ | |
5999 | bool try_wait_for_completion(struct completion *x) | |
6000 | { | |
6001 | int ret = 1; | |
6002 | ||
6003 | spin_lock_irq(&x->wait.lock); | |
6004 | if (!x->done) | |
6005 | ret = 0; | |
6006 | else | |
6007 | x->done--; | |
6008 | spin_unlock_irq(&x->wait.lock); | |
6009 | return ret; | |
6010 | } | |
6011 | EXPORT_SYMBOL(try_wait_for_completion); | |
6012 | ||
6013 | /** | |
6014 | * completion_done - Test to see if a completion has any waiters | |
6015 | * @x: completion structure | |
6016 | * | |
6017 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
6018 | * 1 if there are no waiters. | |
6019 | * | |
6020 | */ | |
6021 | bool completion_done(struct completion *x) | |
6022 | { | |
6023 | int ret = 1; | |
6024 | ||
6025 | spin_lock_irq(&x->wait.lock); | |
6026 | if (!x->done) | |
6027 | ret = 0; | |
6028 | spin_unlock_irq(&x->wait.lock); | |
6029 | return ret; | |
6030 | } | |
6031 | EXPORT_SYMBOL(completion_done); | |
6032 | ||
8cbbe86d AK |
6033 | static long __sched |
6034 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 6035 | { |
0fec171c IM |
6036 | unsigned long flags; |
6037 | wait_queue_t wait; | |
6038 | ||
6039 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 6040 | |
8cbbe86d | 6041 | __set_current_state(state); |
1da177e4 | 6042 | |
8cbbe86d AK |
6043 | spin_lock_irqsave(&q->lock, flags); |
6044 | __add_wait_queue(q, &wait); | |
6045 | spin_unlock(&q->lock); | |
6046 | timeout = schedule_timeout(timeout); | |
6047 | spin_lock_irq(&q->lock); | |
6048 | __remove_wait_queue(q, &wait); | |
6049 | spin_unlock_irqrestore(&q->lock, flags); | |
6050 | ||
6051 | return timeout; | |
6052 | } | |
6053 | ||
6054 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
6055 | { | |
6056 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 6057 | } |
1da177e4 LT |
6058 | EXPORT_SYMBOL(interruptible_sleep_on); |
6059 | ||
0fec171c | 6060 | long __sched |
95cdf3b7 | 6061 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6062 | { |
8cbbe86d | 6063 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 6064 | } |
1da177e4 LT |
6065 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
6066 | ||
0fec171c | 6067 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 6068 | { |
8cbbe86d | 6069 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 6070 | } |
1da177e4 LT |
6071 | EXPORT_SYMBOL(sleep_on); |
6072 | ||
0fec171c | 6073 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6074 | { |
8cbbe86d | 6075 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 6076 | } |
1da177e4 LT |
6077 | EXPORT_SYMBOL(sleep_on_timeout); |
6078 | ||
b29739f9 IM |
6079 | #ifdef CONFIG_RT_MUTEXES |
6080 | ||
6081 | /* | |
6082 | * rt_mutex_setprio - set the current priority of a task | |
6083 | * @p: task | |
6084 | * @prio: prio value (kernel-internal form) | |
6085 | * | |
6086 | * This function changes the 'effective' priority of a task. It does | |
6087 | * not touch ->normal_prio like __setscheduler(). | |
6088 | * | |
6089 | * Used by the rt_mutex code to implement priority inheritance logic. | |
6090 | */ | |
36c8b586 | 6091 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
6092 | { |
6093 | unsigned long flags; | |
83b699ed | 6094 | int oldprio, on_rq, running; |
70b97a7f | 6095 | struct rq *rq; |
cb469845 | 6096 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
6097 | |
6098 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
6099 | ||
6100 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6101 | update_rq_clock(rq); |
b29739f9 | 6102 | |
d5f9f942 | 6103 | oldprio = p->prio; |
dd41f596 | 6104 | on_rq = p->se.on_rq; |
051a1d1a | 6105 | running = task_current(rq, p); |
0e1f3483 | 6106 | if (on_rq) |
69be72c1 | 6107 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
6108 | if (running) |
6109 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
6110 | |
6111 | if (rt_prio(prio)) | |
6112 | p->sched_class = &rt_sched_class; | |
6113 | else | |
6114 | p->sched_class = &fair_sched_class; | |
6115 | ||
b29739f9 IM |
6116 | p->prio = prio; |
6117 | ||
0e1f3483 HS |
6118 | if (running) |
6119 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 6120 | if (on_rq) { |
8159f87e | 6121 | enqueue_task(rq, p, 0); |
cb469845 SR |
6122 | |
6123 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
6124 | } |
6125 | task_rq_unlock(rq, &flags); | |
6126 | } | |
6127 | ||
6128 | #endif | |
6129 | ||
36c8b586 | 6130 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6131 | { |
dd41f596 | 6132 | int old_prio, delta, on_rq; |
1da177e4 | 6133 | unsigned long flags; |
70b97a7f | 6134 | struct rq *rq; |
1da177e4 LT |
6135 | |
6136 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6137 | return; | |
6138 | /* | |
6139 | * We have to be careful, if called from sys_setpriority(), | |
6140 | * the task might be in the middle of scheduling on another CPU. | |
6141 | */ | |
6142 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6143 | update_rq_clock(rq); |
1da177e4 LT |
6144 | /* |
6145 | * The RT priorities are set via sched_setscheduler(), but we still | |
6146 | * allow the 'normal' nice value to be set - but as expected | |
6147 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6148 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6149 | */ |
e05606d3 | 6150 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6151 | p->static_prio = NICE_TO_PRIO(nice); |
6152 | goto out_unlock; | |
6153 | } | |
dd41f596 | 6154 | on_rq = p->se.on_rq; |
c09595f6 | 6155 | if (on_rq) |
69be72c1 | 6156 | dequeue_task(rq, p, 0); |
1da177e4 | 6157 | |
1da177e4 | 6158 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6159 | set_load_weight(p); |
b29739f9 IM |
6160 | old_prio = p->prio; |
6161 | p->prio = effective_prio(p); | |
6162 | delta = p->prio - old_prio; | |
1da177e4 | 6163 | |
dd41f596 | 6164 | if (on_rq) { |
8159f87e | 6165 | enqueue_task(rq, p, 0); |
1da177e4 | 6166 | /* |
d5f9f942 AM |
6167 | * If the task increased its priority or is running and |
6168 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6169 | */ |
d5f9f942 | 6170 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6171 | resched_task(rq->curr); |
6172 | } | |
6173 | out_unlock: | |
6174 | task_rq_unlock(rq, &flags); | |
6175 | } | |
1da177e4 LT |
6176 | EXPORT_SYMBOL(set_user_nice); |
6177 | ||
e43379f1 MM |
6178 | /* |
6179 | * can_nice - check if a task can reduce its nice value | |
6180 | * @p: task | |
6181 | * @nice: nice value | |
6182 | */ | |
36c8b586 | 6183 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6184 | { |
024f4747 MM |
6185 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6186 | int nice_rlim = 20 - nice; | |
48f24c4d | 6187 | |
e43379f1 MM |
6188 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6189 | capable(CAP_SYS_NICE)); | |
6190 | } | |
6191 | ||
1da177e4 LT |
6192 | #ifdef __ARCH_WANT_SYS_NICE |
6193 | ||
6194 | /* | |
6195 | * sys_nice - change the priority of the current process. | |
6196 | * @increment: priority increment | |
6197 | * | |
6198 | * sys_setpriority is a more generic, but much slower function that | |
6199 | * does similar things. | |
6200 | */ | |
5add95d4 | 6201 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6202 | { |
48f24c4d | 6203 | long nice, retval; |
1da177e4 LT |
6204 | |
6205 | /* | |
6206 | * Setpriority might change our priority at the same moment. | |
6207 | * We don't have to worry. Conceptually one call occurs first | |
6208 | * and we have a single winner. | |
6209 | */ | |
e43379f1 MM |
6210 | if (increment < -40) |
6211 | increment = -40; | |
1da177e4 LT |
6212 | if (increment > 40) |
6213 | increment = 40; | |
6214 | ||
2b8f836f | 6215 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6216 | if (nice < -20) |
6217 | nice = -20; | |
6218 | if (nice > 19) | |
6219 | nice = 19; | |
6220 | ||
e43379f1 MM |
6221 | if (increment < 0 && !can_nice(current, nice)) |
6222 | return -EPERM; | |
6223 | ||
1da177e4 LT |
6224 | retval = security_task_setnice(current, nice); |
6225 | if (retval) | |
6226 | return retval; | |
6227 | ||
6228 | set_user_nice(current, nice); | |
6229 | return 0; | |
6230 | } | |
6231 | ||
6232 | #endif | |
6233 | ||
6234 | /** | |
6235 | * task_prio - return the priority value of a given task. | |
6236 | * @p: the task in question. | |
6237 | * | |
6238 | * This is the priority value as seen by users in /proc. | |
6239 | * RT tasks are offset by -200. Normal tasks are centered | |
6240 | * around 0, value goes from -16 to +15. | |
6241 | */ | |
36c8b586 | 6242 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6243 | { |
6244 | return p->prio - MAX_RT_PRIO; | |
6245 | } | |
6246 | ||
6247 | /** | |
6248 | * task_nice - return the nice value of a given task. | |
6249 | * @p: the task in question. | |
6250 | */ | |
36c8b586 | 6251 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6252 | { |
6253 | return TASK_NICE(p); | |
6254 | } | |
150d8bed | 6255 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6256 | |
6257 | /** | |
6258 | * idle_cpu - is a given cpu idle currently? | |
6259 | * @cpu: the processor in question. | |
6260 | */ | |
6261 | int idle_cpu(int cpu) | |
6262 | { | |
6263 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6264 | } | |
6265 | ||
1da177e4 LT |
6266 | /** |
6267 | * idle_task - return the idle task for a given cpu. | |
6268 | * @cpu: the processor in question. | |
6269 | */ | |
36c8b586 | 6270 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6271 | { |
6272 | return cpu_rq(cpu)->idle; | |
6273 | } | |
6274 | ||
6275 | /** | |
6276 | * find_process_by_pid - find a process with a matching PID value. | |
6277 | * @pid: the pid in question. | |
6278 | */ | |
a9957449 | 6279 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6280 | { |
228ebcbe | 6281 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6282 | } |
6283 | ||
6284 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6285 | static void |
6286 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6287 | { |
dd41f596 | 6288 | BUG_ON(p->se.on_rq); |
48f24c4d | 6289 | |
1da177e4 | 6290 | p->policy = policy; |
dd41f596 IM |
6291 | switch (p->policy) { |
6292 | case SCHED_NORMAL: | |
6293 | case SCHED_BATCH: | |
6294 | case SCHED_IDLE: | |
6295 | p->sched_class = &fair_sched_class; | |
6296 | break; | |
6297 | case SCHED_FIFO: | |
6298 | case SCHED_RR: | |
6299 | p->sched_class = &rt_sched_class; | |
6300 | break; | |
6301 | } | |
6302 | ||
1da177e4 | 6303 | p->rt_priority = prio; |
b29739f9 IM |
6304 | p->normal_prio = normal_prio(p); |
6305 | /* we are holding p->pi_lock already */ | |
6306 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6307 | set_load_weight(p); |
1da177e4 LT |
6308 | } |
6309 | ||
c69e8d9c DH |
6310 | /* |
6311 | * check the target process has a UID that matches the current process's | |
6312 | */ | |
6313 | static bool check_same_owner(struct task_struct *p) | |
6314 | { | |
6315 | const struct cred *cred = current_cred(), *pcred; | |
6316 | bool match; | |
6317 | ||
6318 | rcu_read_lock(); | |
6319 | pcred = __task_cred(p); | |
6320 | match = (cred->euid == pcred->euid || | |
6321 | cred->euid == pcred->uid); | |
6322 | rcu_read_unlock(); | |
6323 | return match; | |
6324 | } | |
6325 | ||
961ccddd RR |
6326 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6327 | struct sched_param *param, bool user) | |
1da177e4 | 6328 | { |
83b699ed | 6329 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6330 | unsigned long flags; |
cb469845 | 6331 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6332 | struct rq *rq; |
ca94c442 | 6333 | int reset_on_fork; |
1da177e4 | 6334 | |
66e5393a SR |
6335 | /* may grab non-irq protected spin_locks */ |
6336 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6337 | recheck: |
6338 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6339 | if (policy < 0) { |
6340 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6341 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6342 | } else { |
6343 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6344 | policy &= ~SCHED_RESET_ON_FORK; | |
6345 | ||
6346 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6347 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6348 | policy != SCHED_IDLE) | |
6349 | return -EINVAL; | |
6350 | } | |
6351 | ||
1da177e4 LT |
6352 | /* |
6353 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6354 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6355 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6356 | */ |
6357 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6358 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6359 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6360 | return -EINVAL; |
e05606d3 | 6361 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6362 | return -EINVAL; |
6363 | ||
37e4ab3f OC |
6364 | /* |
6365 | * Allow unprivileged RT tasks to decrease priority: | |
6366 | */ | |
961ccddd | 6367 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6368 | if (rt_policy(policy)) { |
8dc3e909 | 6369 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6370 | |
6371 | if (!lock_task_sighand(p, &flags)) | |
6372 | return -ESRCH; | |
6373 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6374 | unlock_task_sighand(p, &flags); | |
6375 | ||
6376 | /* can't set/change the rt policy */ | |
6377 | if (policy != p->policy && !rlim_rtprio) | |
6378 | return -EPERM; | |
6379 | ||
6380 | /* can't increase priority */ | |
6381 | if (param->sched_priority > p->rt_priority && | |
6382 | param->sched_priority > rlim_rtprio) | |
6383 | return -EPERM; | |
6384 | } | |
dd41f596 IM |
6385 | /* |
6386 | * Like positive nice levels, dont allow tasks to | |
6387 | * move out of SCHED_IDLE either: | |
6388 | */ | |
6389 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6390 | return -EPERM; | |
5fe1d75f | 6391 | |
37e4ab3f | 6392 | /* can't change other user's priorities */ |
c69e8d9c | 6393 | if (!check_same_owner(p)) |
37e4ab3f | 6394 | return -EPERM; |
ca94c442 LP |
6395 | |
6396 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6397 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6398 | return -EPERM; | |
37e4ab3f | 6399 | } |
1da177e4 | 6400 | |
725aad24 | 6401 | if (user) { |
b68aa230 | 6402 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6403 | /* |
6404 | * Do not allow realtime tasks into groups that have no runtime | |
6405 | * assigned. | |
6406 | */ | |
9a7e0b18 PZ |
6407 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6408 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6409 | return -EPERM; |
b68aa230 PZ |
6410 | #endif |
6411 | ||
725aad24 JF |
6412 | retval = security_task_setscheduler(p, policy, param); |
6413 | if (retval) | |
6414 | return retval; | |
6415 | } | |
6416 | ||
b29739f9 IM |
6417 | /* |
6418 | * make sure no PI-waiters arrive (or leave) while we are | |
6419 | * changing the priority of the task: | |
6420 | */ | |
6421 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6422 | /* |
6423 | * To be able to change p->policy safely, the apropriate | |
6424 | * runqueue lock must be held. | |
6425 | */ | |
b29739f9 | 6426 | rq = __task_rq_lock(p); |
1da177e4 LT |
6427 | /* recheck policy now with rq lock held */ |
6428 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6429 | policy = oldpolicy = -1; | |
b29739f9 IM |
6430 | __task_rq_unlock(rq); |
6431 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6432 | goto recheck; |
6433 | } | |
2daa3577 | 6434 | update_rq_clock(rq); |
dd41f596 | 6435 | on_rq = p->se.on_rq; |
051a1d1a | 6436 | running = task_current(rq, p); |
0e1f3483 | 6437 | if (on_rq) |
2e1cb74a | 6438 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6439 | if (running) |
6440 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6441 | |
ca94c442 LP |
6442 | p->sched_reset_on_fork = reset_on_fork; |
6443 | ||
1da177e4 | 6444 | oldprio = p->prio; |
dd41f596 | 6445 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6446 | |
0e1f3483 HS |
6447 | if (running) |
6448 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6449 | if (on_rq) { |
6450 | activate_task(rq, p, 0); | |
cb469845 SR |
6451 | |
6452 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6453 | } |
b29739f9 IM |
6454 | __task_rq_unlock(rq); |
6455 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6456 | ||
95e02ca9 TG |
6457 | rt_mutex_adjust_pi(p); |
6458 | ||
1da177e4 LT |
6459 | return 0; |
6460 | } | |
961ccddd RR |
6461 | |
6462 | /** | |
6463 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6464 | * @p: the task in question. | |
6465 | * @policy: new policy. | |
6466 | * @param: structure containing the new RT priority. | |
6467 | * | |
6468 | * NOTE that the task may be already dead. | |
6469 | */ | |
6470 | int sched_setscheduler(struct task_struct *p, int policy, | |
6471 | struct sched_param *param) | |
6472 | { | |
6473 | return __sched_setscheduler(p, policy, param, true); | |
6474 | } | |
1da177e4 LT |
6475 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6476 | ||
961ccddd RR |
6477 | /** |
6478 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6479 | * @p: the task in question. | |
6480 | * @policy: new policy. | |
6481 | * @param: structure containing the new RT priority. | |
6482 | * | |
6483 | * Just like sched_setscheduler, only don't bother checking if the | |
6484 | * current context has permission. For example, this is needed in | |
6485 | * stop_machine(): we create temporary high priority worker threads, | |
6486 | * but our caller might not have that capability. | |
6487 | */ | |
6488 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6489 | struct sched_param *param) | |
6490 | { | |
6491 | return __sched_setscheduler(p, policy, param, false); | |
6492 | } | |
6493 | ||
95cdf3b7 IM |
6494 | static int |
6495 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6496 | { |
1da177e4 LT |
6497 | struct sched_param lparam; |
6498 | struct task_struct *p; | |
36c8b586 | 6499 | int retval; |
1da177e4 LT |
6500 | |
6501 | if (!param || pid < 0) | |
6502 | return -EINVAL; | |
6503 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6504 | return -EFAULT; | |
5fe1d75f ON |
6505 | |
6506 | rcu_read_lock(); | |
6507 | retval = -ESRCH; | |
1da177e4 | 6508 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6509 | if (p != NULL) |
6510 | retval = sched_setscheduler(p, policy, &lparam); | |
6511 | rcu_read_unlock(); | |
36c8b586 | 6512 | |
1da177e4 LT |
6513 | return retval; |
6514 | } | |
6515 | ||
6516 | /** | |
6517 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6518 | * @pid: the pid in question. | |
6519 | * @policy: new policy. | |
6520 | * @param: structure containing the new RT priority. | |
6521 | */ | |
5add95d4 HC |
6522 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6523 | struct sched_param __user *, param) | |
1da177e4 | 6524 | { |
c21761f1 JB |
6525 | /* negative values for policy are not valid */ |
6526 | if (policy < 0) | |
6527 | return -EINVAL; | |
6528 | ||
1da177e4 LT |
6529 | return do_sched_setscheduler(pid, policy, param); |
6530 | } | |
6531 | ||
6532 | /** | |
6533 | * sys_sched_setparam - set/change the RT priority of a thread | |
6534 | * @pid: the pid in question. | |
6535 | * @param: structure containing the new RT priority. | |
6536 | */ | |
5add95d4 | 6537 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6538 | { |
6539 | return do_sched_setscheduler(pid, -1, param); | |
6540 | } | |
6541 | ||
6542 | /** | |
6543 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6544 | * @pid: the pid in question. | |
6545 | */ | |
5add95d4 | 6546 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6547 | { |
36c8b586 | 6548 | struct task_struct *p; |
3a5c359a | 6549 | int retval; |
1da177e4 LT |
6550 | |
6551 | if (pid < 0) | |
3a5c359a | 6552 | return -EINVAL; |
1da177e4 LT |
6553 | |
6554 | retval = -ESRCH; | |
6555 | read_lock(&tasklist_lock); | |
6556 | p = find_process_by_pid(pid); | |
6557 | if (p) { | |
6558 | retval = security_task_getscheduler(p); | |
6559 | if (!retval) | |
ca94c442 LP |
6560 | retval = p->policy |
6561 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6562 | } |
6563 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6564 | return retval; |
6565 | } | |
6566 | ||
6567 | /** | |
ca94c442 | 6568 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6569 | * @pid: the pid in question. |
6570 | * @param: structure containing the RT priority. | |
6571 | */ | |
5add95d4 | 6572 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6573 | { |
6574 | struct sched_param lp; | |
36c8b586 | 6575 | struct task_struct *p; |
3a5c359a | 6576 | int retval; |
1da177e4 LT |
6577 | |
6578 | if (!param || pid < 0) | |
3a5c359a | 6579 | return -EINVAL; |
1da177e4 LT |
6580 | |
6581 | read_lock(&tasklist_lock); | |
6582 | p = find_process_by_pid(pid); | |
6583 | retval = -ESRCH; | |
6584 | if (!p) | |
6585 | goto out_unlock; | |
6586 | ||
6587 | retval = security_task_getscheduler(p); | |
6588 | if (retval) | |
6589 | goto out_unlock; | |
6590 | ||
6591 | lp.sched_priority = p->rt_priority; | |
6592 | read_unlock(&tasklist_lock); | |
6593 | ||
6594 | /* | |
6595 | * This one might sleep, we cannot do it with a spinlock held ... | |
6596 | */ | |
6597 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6598 | ||
1da177e4 LT |
6599 | return retval; |
6600 | ||
6601 | out_unlock: | |
6602 | read_unlock(&tasklist_lock); | |
6603 | return retval; | |
6604 | } | |
6605 | ||
96f874e2 | 6606 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6607 | { |
5a16f3d3 | 6608 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6609 | struct task_struct *p; |
6610 | int retval; | |
1da177e4 | 6611 | |
95402b38 | 6612 | get_online_cpus(); |
1da177e4 LT |
6613 | read_lock(&tasklist_lock); |
6614 | ||
6615 | p = find_process_by_pid(pid); | |
6616 | if (!p) { | |
6617 | read_unlock(&tasklist_lock); | |
95402b38 | 6618 | put_online_cpus(); |
1da177e4 LT |
6619 | return -ESRCH; |
6620 | } | |
6621 | ||
6622 | /* | |
6623 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6624 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6625 | * usage count and then drop tasklist_lock. |
6626 | */ | |
6627 | get_task_struct(p); | |
6628 | read_unlock(&tasklist_lock); | |
6629 | ||
5a16f3d3 RR |
6630 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6631 | retval = -ENOMEM; | |
6632 | goto out_put_task; | |
6633 | } | |
6634 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6635 | retval = -ENOMEM; | |
6636 | goto out_free_cpus_allowed; | |
6637 | } | |
1da177e4 | 6638 | retval = -EPERM; |
c69e8d9c | 6639 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6640 | goto out_unlock; |
6641 | ||
e7834f8f DQ |
6642 | retval = security_task_setscheduler(p, 0, NULL); |
6643 | if (retval) | |
6644 | goto out_unlock; | |
6645 | ||
5a16f3d3 RR |
6646 | cpuset_cpus_allowed(p, cpus_allowed); |
6647 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6648 | again: |
5a16f3d3 | 6649 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6650 | |
8707d8b8 | 6651 | if (!retval) { |
5a16f3d3 RR |
6652 | cpuset_cpus_allowed(p, cpus_allowed); |
6653 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6654 | /* |
6655 | * We must have raced with a concurrent cpuset | |
6656 | * update. Just reset the cpus_allowed to the | |
6657 | * cpuset's cpus_allowed | |
6658 | */ | |
5a16f3d3 | 6659 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6660 | goto again; |
6661 | } | |
6662 | } | |
1da177e4 | 6663 | out_unlock: |
5a16f3d3 RR |
6664 | free_cpumask_var(new_mask); |
6665 | out_free_cpus_allowed: | |
6666 | free_cpumask_var(cpus_allowed); | |
6667 | out_put_task: | |
1da177e4 | 6668 | put_task_struct(p); |
95402b38 | 6669 | put_online_cpus(); |
1da177e4 LT |
6670 | return retval; |
6671 | } | |
6672 | ||
6673 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6674 | struct cpumask *new_mask) |
1da177e4 | 6675 | { |
96f874e2 RR |
6676 | if (len < cpumask_size()) |
6677 | cpumask_clear(new_mask); | |
6678 | else if (len > cpumask_size()) | |
6679 | len = cpumask_size(); | |
6680 | ||
1da177e4 LT |
6681 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6682 | } | |
6683 | ||
6684 | /** | |
6685 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6686 | * @pid: pid of the process | |
6687 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6688 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6689 | */ | |
5add95d4 HC |
6690 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6691 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6692 | { |
5a16f3d3 | 6693 | cpumask_var_t new_mask; |
1da177e4 LT |
6694 | int retval; |
6695 | ||
5a16f3d3 RR |
6696 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6697 | return -ENOMEM; | |
1da177e4 | 6698 | |
5a16f3d3 RR |
6699 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6700 | if (retval == 0) | |
6701 | retval = sched_setaffinity(pid, new_mask); | |
6702 | free_cpumask_var(new_mask); | |
6703 | return retval; | |
1da177e4 LT |
6704 | } |
6705 | ||
96f874e2 | 6706 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6707 | { |
36c8b586 | 6708 | struct task_struct *p; |
1da177e4 | 6709 | int retval; |
1da177e4 | 6710 | |
95402b38 | 6711 | get_online_cpus(); |
1da177e4 LT |
6712 | read_lock(&tasklist_lock); |
6713 | ||
6714 | retval = -ESRCH; | |
6715 | p = find_process_by_pid(pid); | |
6716 | if (!p) | |
6717 | goto out_unlock; | |
6718 | ||
e7834f8f DQ |
6719 | retval = security_task_getscheduler(p); |
6720 | if (retval) | |
6721 | goto out_unlock; | |
6722 | ||
96f874e2 | 6723 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6724 | |
6725 | out_unlock: | |
6726 | read_unlock(&tasklist_lock); | |
95402b38 | 6727 | put_online_cpus(); |
1da177e4 | 6728 | |
9531b62f | 6729 | return retval; |
1da177e4 LT |
6730 | } |
6731 | ||
6732 | /** | |
6733 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6734 | * @pid: pid of the process | |
6735 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6736 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6737 | */ | |
5add95d4 HC |
6738 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6739 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6740 | { |
6741 | int ret; | |
f17c8607 | 6742 | cpumask_var_t mask; |
1da177e4 | 6743 | |
f17c8607 | 6744 | if (len < cpumask_size()) |
1da177e4 LT |
6745 | return -EINVAL; |
6746 | ||
f17c8607 RR |
6747 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6748 | return -ENOMEM; | |
1da177e4 | 6749 | |
f17c8607 RR |
6750 | ret = sched_getaffinity(pid, mask); |
6751 | if (ret == 0) { | |
6752 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6753 | ret = -EFAULT; | |
6754 | else | |
6755 | ret = cpumask_size(); | |
6756 | } | |
6757 | free_cpumask_var(mask); | |
1da177e4 | 6758 | |
f17c8607 | 6759 | return ret; |
1da177e4 LT |
6760 | } |
6761 | ||
6762 | /** | |
6763 | * sys_sched_yield - yield the current processor to other threads. | |
6764 | * | |
dd41f596 IM |
6765 | * This function yields the current CPU to other tasks. If there are no |
6766 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6767 | */ |
5add95d4 | 6768 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6769 | { |
70b97a7f | 6770 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6771 | |
2d72376b | 6772 | schedstat_inc(rq, yld_count); |
4530d7ab | 6773 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6774 | |
6775 | /* | |
6776 | * Since we are going to call schedule() anyway, there's | |
6777 | * no need to preempt or enable interrupts: | |
6778 | */ | |
6779 | __release(rq->lock); | |
8a25d5de | 6780 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6781 | _raw_spin_unlock(&rq->lock); |
6782 | preempt_enable_no_resched(); | |
6783 | ||
6784 | schedule(); | |
6785 | ||
6786 | return 0; | |
6787 | } | |
6788 | ||
d86ee480 PZ |
6789 | static inline int should_resched(void) |
6790 | { | |
6791 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6792 | } | |
6793 | ||
e7b38404 | 6794 | static void __cond_resched(void) |
1da177e4 | 6795 | { |
e7aaaa69 FW |
6796 | add_preempt_count(PREEMPT_ACTIVE); |
6797 | schedule(); | |
6798 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6799 | } |
6800 | ||
02b67cc3 | 6801 | int __sched _cond_resched(void) |
1da177e4 | 6802 | { |
d86ee480 | 6803 | if (should_resched()) { |
1da177e4 LT |
6804 | __cond_resched(); |
6805 | return 1; | |
6806 | } | |
6807 | return 0; | |
6808 | } | |
02b67cc3 | 6809 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6810 | |
6811 | /* | |
613afbf8 | 6812 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6813 | * call schedule, and on return reacquire the lock. |
6814 | * | |
41a2d6cf | 6815 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6816 | * operations here to prevent schedule() from being called twice (once via |
6817 | * spin_unlock(), once by hand). | |
6818 | */ | |
613afbf8 | 6819 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6820 | { |
d86ee480 | 6821 | int resched = should_resched(); |
6df3cecb JK |
6822 | int ret = 0; |
6823 | ||
f607c668 PZ |
6824 | lockdep_assert_held(lock); |
6825 | ||
95c354fe | 6826 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6827 | spin_unlock(lock); |
d86ee480 | 6828 | if (resched) |
95c354fe NP |
6829 | __cond_resched(); |
6830 | else | |
6831 | cpu_relax(); | |
6df3cecb | 6832 | ret = 1; |
1da177e4 | 6833 | spin_lock(lock); |
1da177e4 | 6834 | } |
6df3cecb | 6835 | return ret; |
1da177e4 | 6836 | } |
613afbf8 | 6837 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6838 | |
613afbf8 | 6839 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6840 | { |
6841 | BUG_ON(!in_softirq()); | |
6842 | ||
d86ee480 | 6843 | if (should_resched()) { |
98d82567 | 6844 | local_bh_enable(); |
1da177e4 LT |
6845 | __cond_resched(); |
6846 | local_bh_disable(); | |
6847 | return 1; | |
6848 | } | |
6849 | return 0; | |
6850 | } | |
613afbf8 | 6851 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6852 | |
1da177e4 LT |
6853 | /** |
6854 | * yield - yield the current processor to other threads. | |
6855 | * | |
72fd4a35 | 6856 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6857 | * thread runnable and calls sys_sched_yield(). |
6858 | */ | |
6859 | void __sched yield(void) | |
6860 | { | |
6861 | set_current_state(TASK_RUNNING); | |
6862 | sys_sched_yield(); | |
6863 | } | |
1da177e4 LT |
6864 | EXPORT_SYMBOL(yield); |
6865 | ||
6866 | /* | |
41a2d6cf | 6867 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6868 | * that process accounting knows that this is a task in IO wait state. |
6869 | * | |
6870 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6871 | * has set its backing_dev_info: the queue against which it should throttle) | |
6872 | */ | |
6873 | void __sched io_schedule(void) | |
6874 | { | |
54d35f29 | 6875 | struct rq *rq = raw_rq(); |
1da177e4 | 6876 | |
0ff92245 | 6877 | delayacct_blkio_start(); |
1da177e4 | 6878 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6879 | current->in_iowait = 1; |
1da177e4 | 6880 | schedule(); |
8f0dfc34 | 6881 | current->in_iowait = 0; |
1da177e4 | 6882 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6883 | delayacct_blkio_end(); |
1da177e4 | 6884 | } |
1da177e4 LT |
6885 | EXPORT_SYMBOL(io_schedule); |
6886 | ||
6887 | long __sched io_schedule_timeout(long timeout) | |
6888 | { | |
54d35f29 | 6889 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6890 | long ret; |
6891 | ||
0ff92245 | 6892 | delayacct_blkio_start(); |
1da177e4 | 6893 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6894 | current->in_iowait = 1; |
1da177e4 | 6895 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6896 | current->in_iowait = 0; |
1da177e4 | 6897 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6898 | delayacct_blkio_end(); |
1da177e4 LT |
6899 | return ret; |
6900 | } | |
6901 | ||
6902 | /** | |
6903 | * sys_sched_get_priority_max - return maximum RT priority. | |
6904 | * @policy: scheduling class. | |
6905 | * | |
6906 | * this syscall returns the maximum rt_priority that can be used | |
6907 | * by a given scheduling class. | |
6908 | */ | |
5add95d4 | 6909 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6910 | { |
6911 | int ret = -EINVAL; | |
6912 | ||
6913 | switch (policy) { | |
6914 | case SCHED_FIFO: | |
6915 | case SCHED_RR: | |
6916 | ret = MAX_USER_RT_PRIO-1; | |
6917 | break; | |
6918 | case SCHED_NORMAL: | |
b0a9499c | 6919 | case SCHED_BATCH: |
dd41f596 | 6920 | case SCHED_IDLE: |
1da177e4 LT |
6921 | ret = 0; |
6922 | break; | |
6923 | } | |
6924 | return ret; | |
6925 | } | |
6926 | ||
6927 | /** | |
6928 | * sys_sched_get_priority_min - return minimum RT priority. | |
6929 | * @policy: scheduling class. | |
6930 | * | |
6931 | * this syscall returns the minimum rt_priority that can be used | |
6932 | * by a given scheduling class. | |
6933 | */ | |
5add95d4 | 6934 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6935 | { |
6936 | int ret = -EINVAL; | |
6937 | ||
6938 | switch (policy) { | |
6939 | case SCHED_FIFO: | |
6940 | case SCHED_RR: | |
6941 | ret = 1; | |
6942 | break; | |
6943 | case SCHED_NORMAL: | |
b0a9499c | 6944 | case SCHED_BATCH: |
dd41f596 | 6945 | case SCHED_IDLE: |
1da177e4 LT |
6946 | ret = 0; |
6947 | } | |
6948 | return ret; | |
6949 | } | |
6950 | ||
6951 | /** | |
6952 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6953 | * @pid: pid of the process. | |
6954 | * @interval: userspace pointer to the timeslice value. | |
6955 | * | |
6956 | * this syscall writes the default timeslice value of a given process | |
6957 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6958 | */ | |
17da2bd9 | 6959 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6960 | struct timespec __user *, interval) |
1da177e4 | 6961 | { |
36c8b586 | 6962 | struct task_struct *p; |
a4ec24b4 | 6963 | unsigned int time_slice; |
3a5c359a | 6964 | int retval; |
1da177e4 | 6965 | struct timespec t; |
1da177e4 LT |
6966 | |
6967 | if (pid < 0) | |
3a5c359a | 6968 | return -EINVAL; |
1da177e4 LT |
6969 | |
6970 | retval = -ESRCH; | |
6971 | read_lock(&tasklist_lock); | |
6972 | p = find_process_by_pid(pid); | |
6973 | if (!p) | |
6974 | goto out_unlock; | |
6975 | ||
6976 | retval = security_task_getscheduler(p); | |
6977 | if (retval) | |
6978 | goto out_unlock; | |
6979 | ||
77034937 IM |
6980 | /* |
6981 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6982 | * tasks that are on an otherwise idle runqueue: | |
6983 | */ | |
6984 | time_slice = 0; | |
6985 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6986 | time_slice = DEF_TIMESLICE; |
1868f958 | 6987 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6988 | struct sched_entity *se = &p->se; |
6989 | unsigned long flags; | |
6990 | struct rq *rq; | |
6991 | ||
6992 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6993 | if (rq->cfs.load.weight) |
6994 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6995 | task_rq_unlock(rq, &flags); |
6996 | } | |
1da177e4 | 6997 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6998 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6999 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 7000 | return retval; |
3a5c359a | 7001 | |
1da177e4 LT |
7002 | out_unlock: |
7003 | read_unlock(&tasklist_lock); | |
7004 | return retval; | |
7005 | } | |
7006 | ||
7c731e0a | 7007 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 7008 | |
82a1fcb9 | 7009 | void sched_show_task(struct task_struct *p) |
1da177e4 | 7010 | { |
1da177e4 | 7011 | unsigned long free = 0; |
36c8b586 | 7012 | unsigned state; |
1da177e4 | 7013 | |
1da177e4 | 7014 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 7015 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 7016 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 7017 | #if BITS_PER_LONG == 32 |
1da177e4 | 7018 | if (state == TASK_RUNNING) |
cc4ea795 | 7019 | printk(KERN_CONT " running "); |
1da177e4 | 7020 | else |
cc4ea795 | 7021 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
7022 | #else |
7023 | if (state == TASK_RUNNING) | |
cc4ea795 | 7024 | printk(KERN_CONT " running task "); |
1da177e4 | 7025 | else |
cc4ea795 | 7026 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
7027 | #endif |
7028 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 7029 | free = stack_not_used(p); |
1da177e4 | 7030 | #endif |
aa47b7e0 DR |
7031 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
7032 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
7033 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 7034 | |
5fb5e6de | 7035 | show_stack(p, NULL); |
1da177e4 LT |
7036 | } |
7037 | ||
e59e2ae2 | 7038 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7039 | { |
36c8b586 | 7040 | struct task_struct *g, *p; |
1da177e4 | 7041 | |
4bd77321 IM |
7042 | #if BITS_PER_LONG == 32 |
7043 | printk(KERN_INFO | |
7044 | " task PC stack pid father\n"); | |
1da177e4 | 7045 | #else |
4bd77321 IM |
7046 | printk(KERN_INFO |
7047 | " task PC stack pid father\n"); | |
1da177e4 LT |
7048 | #endif |
7049 | read_lock(&tasklist_lock); | |
7050 | do_each_thread(g, p) { | |
7051 | /* | |
7052 | * reset the NMI-timeout, listing all files on a slow | |
7053 | * console might take alot of time: | |
7054 | */ | |
7055 | touch_nmi_watchdog(); | |
39bc89fd | 7056 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 7057 | sched_show_task(p); |
1da177e4 LT |
7058 | } while_each_thread(g, p); |
7059 | ||
04c9167f JF |
7060 | touch_all_softlockup_watchdogs(); |
7061 | ||
dd41f596 IM |
7062 | #ifdef CONFIG_SCHED_DEBUG |
7063 | sysrq_sched_debug_show(); | |
7064 | #endif | |
1da177e4 | 7065 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
7066 | /* |
7067 | * Only show locks if all tasks are dumped: | |
7068 | */ | |
7069 | if (state_filter == -1) | |
7070 | debug_show_all_locks(); | |
1da177e4 LT |
7071 | } |
7072 | ||
1df21055 IM |
7073 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
7074 | { | |
dd41f596 | 7075 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
7076 | } |
7077 | ||
f340c0d1 IM |
7078 | /** |
7079 | * init_idle - set up an idle thread for a given CPU | |
7080 | * @idle: task in question | |
7081 | * @cpu: cpu the idle task belongs to | |
7082 | * | |
7083 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7084 | * flag, to make booting more robust. | |
7085 | */ | |
5c1e1767 | 7086 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7087 | { |
70b97a7f | 7088 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7089 | unsigned long flags; |
7090 | ||
5cbd54ef IM |
7091 | spin_lock_irqsave(&rq->lock, flags); |
7092 | ||
dd41f596 IM |
7093 | __sched_fork(idle); |
7094 | idle->se.exec_start = sched_clock(); | |
7095 | ||
b29739f9 | 7096 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 7097 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 7098 | __set_task_cpu(idle, cpu); |
1da177e4 | 7099 | |
1da177e4 | 7100 | rq->curr = rq->idle = idle; |
4866cde0 NP |
7101 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
7102 | idle->oncpu = 1; | |
7103 | #endif | |
1da177e4 LT |
7104 | spin_unlock_irqrestore(&rq->lock, flags); |
7105 | ||
7106 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
7107 | #if defined(CONFIG_PREEMPT) |
7108 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
7109 | #else | |
a1261f54 | 7110 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 7111 | #endif |
dd41f596 IM |
7112 | /* |
7113 | * The idle tasks have their own, simple scheduling class: | |
7114 | */ | |
7115 | idle->sched_class = &idle_sched_class; | |
fb52607a | 7116 | ftrace_graph_init_task(idle); |
1da177e4 LT |
7117 | } |
7118 | ||
7119 | /* | |
7120 | * In a system that switches off the HZ timer nohz_cpu_mask | |
7121 | * indicates which cpus entered this state. This is used | |
7122 | * in the rcu update to wait only for active cpus. For system | |
7123 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 7124 | * always be CPU_BITS_NONE. |
1da177e4 | 7125 | */ |
6a7b3dc3 | 7126 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 7127 | |
19978ca6 IM |
7128 | /* |
7129 | * Increase the granularity value when there are more CPUs, | |
7130 | * because with more CPUs the 'effective latency' as visible | |
7131 | * to users decreases. But the relationship is not linear, | |
7132 | * so pick a second-best guess by going with the log2 of the | |
7133 | * number of CPUs. | |
7134 | * | |
7135 | * This idea comes from the SD scheduler of Con Kolivas: | |
7136 | */ | |
7137 | static inline void sched_init_granularity(void) | |
7138 | { | |
7139 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
7140 | const unsigned long limit = 200000000; | |
7141 | ||
7142 | sysctl_sched_min_granularity *= factor; | |
7143 | if (sysctl_sched_min_granularity > limit) | |
7144 | sysctl_sched_min_granularity = limit; | |
7145 | ||
7146 | sysctl_sched_latency *= factor; | |
7147 | if (sysctl_sched_latency > limit) | |
7148 | sysctl_sched_latency = limit; | |
7149 | ||
7150 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
7151 | |
7152 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
7153 | } |
7154 | ||
1da177e4 LT |
7155 | #ifdef CONFIG_SMP |
7156 | /* | |
7157 | * This is how migration works: | |
7158 | * | |
70b97a7f | 7159 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7160 | * runqueue and wake up that CPU's migration thread. |
7161 | * 2) we down() the locked semaphore => thread blocks. | |
7162 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7163 | * thread off the CPU) | |
7164 | * 4) it gets the migration request and checks whether the migrated | |
7165 | * task is still in the wrong runqueue. | |
7166 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7167 | * it and puts it into the right queue. | |
7168 | * 6) migration thread up()s the semaphore. | |
7169 | * 7) we wake up and the migration is done. | |
7170 | */ | |
7171 | ||
7172 | /* | |
7173 | * Change a given task's CPU affinity. Migrate the thread to a | |
7174 | * proper CPU and schedule it away if the CPU it's executing on | |
7175 | * is removed from the allowed bitmask. | |
7176 | * | |
7177 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7178 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7179 | * call is not atomic; no spinlocks may be held. |
7180 | */ | |
96f874e2 | 7181 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7182 | { |
70b97a7f | 7183 | struct migration_req req; |
1da177e4 | 7184 | unsigned long flags; |
70b97a7f | 7185 | struct rq *rq; |
48f24c4d | 7186 | int ret = 0; |
1da177e4 LT |
7187 | |
7188 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7189 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7190 | ret = -EINVAL; |
7191 | goto out; | |
7192 | } | |
7193 | ||
9985b0ba | 7194 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7195 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7196 | ret = -EINVAL; |
7197 | goto out; | |
7198 | } | |
7199 | ||
73fe6aae | 7200 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7201 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7202 | else { |
96f874e2 RR |
7203 | cpumask_copy(&p->cpus_allowed, new_mask); |
7204 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7205 | } |
7206 | ||
1da177e4 | 7207 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7208 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7209 | goto out; |
7210 | ||
1e5ce4f4 | 7211 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 | 7212 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7213 | struct task_struct *mt = rq->migration_thread; |
7214 | ||
7215 | get_task_struct(mt); | |
1da177e4 LT |
7216 | task_rq_unlock(rq, &flags); |
7217 | wake_up_process(rq->migration_thread); | |
693525e3 | 7218 | put_task_struct(mt); |
1da177e4 LT |
7219 | wait_for_completion(&req.done); |
7220 | tlb_migrate_finish(p->mm); | |
7221 | return 0; | |
7222 | } | |
7223 | out: | |
7224 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7225 | |
1da177e4 LT |
7226 | return ret; |
7227 | } | |
cd8ba7cd | 7228 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7229 | |
7230 | /* | |
41a2d6cf | 7231 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7232 | * this because either it can't run here any more (set_cpus_allowed() |
7233 | * away from this CPU, or CPU going down), or because we're | |
7234 | * attempting to rebalance this task on exec (sched_exec). | |
7235 | * | |
7236 | * So we race with normal scheduler movements, but that's OK, as long | |
7237 | * as the task is no longer on this CPU. | |
efc30814 KK |
7238 | * |
7239 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7240 | */ |
efc30814 | 7241 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7242 | { |
70b97a7f | 7243 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7244 | int ret = 0, on_rq; |
1da177e4 | 7245 | |
e761b772 | 7246 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7247 | return ret; |
1da177e4 LT |
7248 | |
7249 | rq_src = cpu_rq(src_cpu); | |
7250 | rq_dest = cpu_rq(dest_cpu); | |
7251 | ||
7252 | double_rq_lock(rq_src, rq_dest); | |
7253 | /* Already moved. */ | |
7254 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7255 | goto done; |
1da177e4 | 7256 | /* Affinity changed (again). */ |
96f874e2 | 7257 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7258 | goto fail; |
1da177e4 | 7259 | |
dd41f596 | 7260 | on_rq = p->se.on_rq; |
6e82a3be | 7261 | if (on_rq) |
2e1cb74a | 7262 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7263 | |
1da177e4 | 7264 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7265 | if (on_rq) { |
7266 | activate_task(rq_dest, p, 0); | |
15afe09b | 7267 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7268 | } |
b1e38734 | 7269 | done: |
efc30814 | 7270 | ret = 1; |
b1e38734 | 7271 | fail: |
1da177e4 | 7272 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7273 | return ret; |
1da177e4 LT |
7274 | } |
7275 | ||
03b042bf PM |
7276 | #define RCU_MIGRATION_IDLE 0 |
7277 | #define RCU_MIGRATION_NEED_QS 1 | |
7278 | #define RCU_MIGRATION_GOT_QS 2 | |
7279 | #define RCU_MIGRATION_MUST_SYNC 3 | |
7280 | ||
1da177e4 LT |
7281 | /* |
7282 | * migration_thread - this is a highprio system thread that performs | |
7283 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7284 | * another runqueue. | |
7285 | */ | |
95cdf3b7 | 7286 | static int migration_thread(void *data) |
1da177e4 | 7287 | { |
03b042bf | 7288 | int badcpu; |
1da177e4 | 7289 | int cpu = (long)data; |
70b97a7f | 7290 | struct rq *rq; |
1da177e4 LT |
7291 | |
7292 | rq = cpu_rq(cpu); | |
7293 | BUG_ON(rq->migration_thread != current); | |
7294 | ||
7295 | set_current_state(TASK_INTERRUPTIBLE); | |
7296 | while (!kthread_should_stop()) { | |
70b97a7f | 7297 | struct migration_req *req; |
1da177e4 | 7298 | struct list_head *head; |
1da177e4 | 7299 | |
1da177e4 LT |
7300 | spin_lock_irq(&rq->lock); |
7301 | ||
7302 | if (cpu_is_offline(cpu)) { | |
7303 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7304 | break; |
1da177e4 LT |
7305 | } |
7306 | ||
7307 | if (rq->active_balance) { | |
7308 | active_load_balance(rq, cpu); | |
7309 | rq->active_balance = 0; | |
7310 | } | |
7311 | ||
7312 | head = &rq->migration_queue; | |
7313 | ||
7314 | if (list_empty(head)) { | |
7315 | spin_unlock_irq(&rq->lock); | |
7316 | schedule(); | |
7317 | set_current_state(TASK_INTERRUPTIBLE); | |
7318 | continue; | |
7319 | } | |
70b97a7f | 7320 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7321 | list_del_init(head->next); |
7322 | ||
03b042bf PM |
7323 | if (req->task != NULL) { |
7324 | spin_unlock(&rq->lock); | |
7325 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7326 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
7327 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
7328 | spin_unlock(&rq->lock); | |
7329 | } else { | |
7330 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
7331 | spin_unlock(&rq->lock); | |
7332 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); | |
7333 | } | |
674311d5 | 7334 | local_irq_enable(); |
1da177e4 LT |
7335 | |
7336 | complete(&req->done); | |
7337 | } | |
7338 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7339 | |
1da177e4 LT |
7340 | return 0; |
7341 | } | |
7342 | ||
7343 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7344 | |
7345 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7346 | { | |
7347 | int ret; | |
7348 | ||
7349 | local_irq_disable(); | |
7350 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7351 | local_irq_enable(); | |
7352 | return ret; | |
7353 | } | |
7354 | ||
054b9108 | 7355 | /* |
3a4fa0a2 | 7356 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7357 | */ |
48f24c4d | 7358 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7359 | { |
70b97a7f | 7360 | int dest_cpu; |
6ca09dfc | 7361 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7362 | |
7363 | again: | |
7364 | /* Look for allowed, online CPU in same node. */ | |
7365 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7366 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7367 | goto move; | |
7368 | ||
7369 | /* Any allowed, online CPU? */ | |
7370 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7371 | if (dest_cpu < nr_cpu_ids) | |
7372 | goto move; | |
7373 | ||
7374 | /* No more Mr. Nice Guy. */ | |
7375 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7376 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7377 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7378 | |
e76bd8d9 RR |
7379 | /* |
7380 | * Don't tell them about moving exiting tasks or | |
7381 | * kernel threads (both mm NULL), since they never | |
7382 | * leave kernel. | |
7383 | */ | |
7384 | if (p->mm && printk_ratelimit()) { | |
7385 | printk(KERN_INFO "process %d (%s) no " | |
7386 | "longer affine to cpu%d\n", | |
7387 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7388 | } |
e76bd8d9 RR |
7389 | } |
7390 | ||
7391 | move: | |
7392 | /* It can have affinity changed while we were choosing. */ | |
7393 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7394 | goto again; | |
1da177e4 LT |
7395 | } |
7396 | ||
7397 | /* | |
7398 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7399 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7400 | * for performance reasons the counter is not stricly tracking tasks to | |
7401 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7402 | * to keep the global sum constant after CPU-down: | |
7403 | */ | |
70b97a7f | 7404 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7405 | { |
1e5ce4f4 | 7406 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7407 | unsigned long flags; |
7408 | ||
7409 | local_irq_save(flags); | |
7410 | double_rq_lock(rq_src, rq_dest); | |
7411 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7412 | rq_src->nr_uninterruptible = 0; | |
7413 | double_rq_unlock(rq_src, rq_dest); | |
7414 | local_irq_restore(flags); | |
7415 | } | |
7416 | ||
7417 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7418 | static void migrate_live_tasks(int src_cpu) | |
7419 | { | |
48f24c4d | 7420 | struct task_struct *p, *t; |
1da177e4 | 7421 | |
f7b4cddc | 7422 | read_lock(&tasklist_lock); |
1da177e4 | 7423 | |
48f24c4d IM |
7424 | do_each_thread(t, p) { |
7425 | if (p == current) | |
1da177e4 LT |
7426 | continue; |
7427 | ||
48f24c4d IM |
7428 | if (task_cpu(p) == src_cpu) |
7429 | move_task_off_dead_cpu(src_cpu, p); | |
7430 | } while_each_thread(t, p); | |
1da177e4 | 7431 | |
f7b4cddc | 7432 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7433 | } |
7434 | ||
dd41f596 IM |
7435 | /* |
7436 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7437 | * It does so by boosting its priority to highest possible. |
7438 | * Used by CPU offline code. | |
1da177e4 LT |
7439 | */ |
7440 | void sched_idle_next(void) | |
7441 | { | |
48f24c4d | 7442 | int this_cpu = smp_processor_id(); |
70b97a7f | 7443 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7444 | struct task_struct *p = rq->idle; |
7445 | unsigned long flags; | |
7446 | ||
7447 | /* cpu has to be offline */ | |
48f24c4d | 7448 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7449 | |
48f24c4d IM |
7450 | /* |
7451 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7452 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7453 | */ |
7454 | spin_lock_irqsave(&rq->lock, flags); | |
7455 | ||
dd41f596 | 7456 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7457 | |
94bc9a7b DA |
7458 | update_rq_clock(rq); |
7459 | activate_task(rq, p, 0); | |
1da177e4 LT |
7460 | |
7461 | spin_unlock_irqrestore(&rq->lock, flags); | |
7462 | } | |
7463 | ||
48f24c4d IM |
7464 | /* |
7465 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7466 | * offline. |
7467 | */ | |
7468 | void idle_task_exit(void) | |
7469 | { | |
7470 | struct mm_struct *mm = current->active_mm; | |
7471 | ||
7472 | BUG_ON(cpu_online(smp_processor_id())); | |
7473 | ||
7474 | if (mm != &init_mm) | |
7475 | switch_mm(mm, &init_mm, current); | |
7476 | mmdrop(mm); | |
7477 | } | |
7478 | ||
054b9108 | 7479 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7480 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7481 | { |
70b97a7f | 7482 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7483 | |
7484 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7485 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7486 | |
7487 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7488 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7489 | |
48f24c4d | 7490 | get_task_struct(p); |
1da177e4 LT |
7491 | |
7492 | /* | |
7493 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7494 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7495 | * fine. |
7496 | */ | |
f7b4cddc | 7497 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7498 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7499 | spin_lock_irq(&rq->lock); |
1da177e4 | 7500 | |
48f24c4d | 7501 | put_task_struct(p); |
1da177e4 LT |
7502 | } |
7503 | ||
7504 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7505 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7506 | { | |
70b97a7f | 7507 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7508 | struct task_struct *next; |
48f24c4d | 7509 | |
dd41f596 IM |
7510 | for ( ; ; ) { |
7511 | if (!rq->nr_running) | |
7512 | break; | |
a8e504d2 | 7513 | update_rq_clock(rq); |
b67802ea | 7514 | next = pick_next_task(rq); |
dd41f596 IM |
7515 | if (!next) |
7516 | break; | |
79c53799 | 7517 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7518 | migrate_dead(dead_cpu, next); |
e692ab53 | 7519 | |
1da177e4 LT |
7520 | } |
7521 | } | |
dce48a84 TG |
7522 | |
7523 | /* | |
7524 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7525 | */ | |
7526 | static void calc_global_load_remove(struct rq *rq) | |
7527 | { | |
7528 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7529 | rq->calc_load_active = 0; |
dce48a84 | 7530 | } |
1da177e4 LT |
7531 | #endif /* CONFIG_HOTPLUG_CPU */ |
7532 | ||
e692ab53 NP |
7533 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7534 | ||
7535 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7536 | { |
7537 | .procname = "sched_domain", | |
c57baf1e | 7538 | .mode = 0555, |
e0361851 | 7539 | }, |
38605cae | 7540 | {0, }, |
e692ab53 NP |
7541 | }; |
7542 | ||
7543 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7544 | { |
c57baf1e | 7545 | .ctl_name = CTL_KERN, |
e0361851 | 7546 | .procname = "kernel", |
c57baf1e | 7547 | .mode = 0555, |
e0361851 AD |
7548 | .child = sd_ctl_dir, |
7549 | }, | |
38605cae | 7550 | {0, }, |
e692ab53 NP |
7551 | }; |
7552 | ||
7553 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7554 | { | |
7555 | struct ctl_table *entry = | |
5cf9f062 | 7556 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7557 | |
e692ab53 NP |
7558 | return entry; |
7559 | } | |
7560 | ||
6382bc90 MM |
7561 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7562 | { | |
cd790076 | 7563 | struct ctl_table *entry; |
6382bc90 | 7564 | |
cd790076 MM |
7565 | /* |
7566 | * In the intermediate directories, both the child directory and | |
7567 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7568 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7569 | * static strings and all have proc handlers. |
7570 | */ | |
7571 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7572 | if (entry->child) |
7573 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7574 | if (entry->proc_handler == NULL) |
7575 | kfree(entry->procname); | |
7576 | } | |
6382bc90 MM |
7577 | |
7578 | kfree(*tablep); | |
7579 | *tablep = NULL; | |
7580 | } | |
7581 | ||
e692ab53 | 7582 | static void |
e0361851 | 7583 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7584 | const char *procname, void *data, int maxlen, |
7585 | mode_t mode, proc_handler *proc_handler) | |
7586 | { | |
e692ab53 NP |
7587 | entry->procname = procname; |
7588 | entry->data = data; | |
7589 | entry->maxlen = maxlen; | |
7590 | entry->mode = mode; | |
7591 | entry->proc_handler = proc_handler; | |
7592 | } | |
7593 | ||
7594 | static struct ctl_table * | |
7595 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7596 | { | |
a5d8c348 | 7597 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7598 | |
ad1cdc1d MM |
7599 | if (table == NULL) |
7600 | return NULL; | |
7601 | ||
e0361851 | 7602 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7603 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7604 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7605 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7606 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7607 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7608 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7609 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7610 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7611 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7612 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7613 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7614 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7615 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7616 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7617 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7618 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7619 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7620 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7621 | &sd->cache_nice_tries, |
7622 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7623 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7624 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7625 | set_table_entry(&table[11], "name", sd->name, |
7626 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7627 | /* &table[12] is terminator */ | |
e692ab53 NP |
7628 | |
7629 | return table; | |
7630 | } | |
7631 | ||
9a4e7159 | 7632 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7633 | { |
7634 | struct ctl_table *entry, *table; | |
7635 | struct sched_domain *sd; | |
7636 | int domain_num = 0, i; | |
7637 | char buf[32]; | |
7638 | ||
7639 | for_each_domain(cpu, sd) | |
7640 | domain_num++; | |
7641 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7642 | if (table == NULL) |
7643 | return NULL; | |
e692ab53 NP |
7644 | |
7645 | i = 0; | |
7646 | for_each_domain(cpu, sd) { | |
7647 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7648 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7649 | entry->mode = 0555; |
e692ab53 NP |
7650 | entry->child = sd_alloc_ctl_domain_table(sd); |
7651 | entry++; | |
7652 | i++; | |
7653 | } | |
7654 | return table; | |
7655 | } | |
7656 | ||
7657 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7658 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7659 | { |
7660 | int i, cpu_num = num_online_cpus(); | |
7661 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7662 | char buf[32]; | |
7663 | ||
7378547f MM |
7664 | WARN_ON(sd_ctl_dir[0].child); |
7665 | sd_ctl_dir[0].child = entry; | |
7666 | ||
ad1cdc1d MM |
7667 | if (entry == NULL) |
7668 | return; | |
7669 | ||
97b6ea7b | 7670 | for_each_online_cpu(i) { |
e692ab53 | 7671 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7672 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7673 | entry->mode = 0555; |
e692ab53 | 7674 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7675 | entry++; |
e692ab53 | 7676 | } |
7378547f MM |
7677 | |
7678 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7679 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7680 | } | |
6382bc90 | 7681 | |
7378547f | 7682 | /* may be called multiple times per register */ |
6382bc90 MM |
7683 | static void unregister_sched_domain_sysctl(void) |
7684 | { | |
7378547f MM |
7685 | if (sd_sysctl_header) |
7686 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7687 | sd_sysctl_header = NULL; |
7378547f MM |
7688 | if (sd_ctl_dir[0].child) |
7689 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7690 | } |
e692ab53 | 7691 | #else |
6382bc90 MM |
7692 | static void register_sched_domain_sysctl(void) |
7693 | { | |
7694 | } | |
7695 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7696 | { |
7697 | } | |
7698 | #endif | |
7699 | ||
1f11eb6a GH |
7700 | static void set_rq_online(struct rq *rq) |
7701 | { | |
7702 | if (!rq->online) { | |
7703 | const struct sched_class *class; | |
7704 | ||
c6c4927b | 7705 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7706 | rq->online = 1; |
7707 | ||
7708 | for_each_class(class) { | |
7709 | if (class->rq_online) | |
7710 | class->rq_online(rq); | |
7711 | } | |
7712 | } | |
7713 | } | |
7714 | ||
7715 | static void set_rq_offline(struct rq *rq) | |
7716 | { | |
7717 | if (rq->online) { | |
7718 | const struct sched_class *class; | |
7719 | ||
7720 | for_each_class(class) { | |
7721 | if (class->rq_offline) | |
7722 | class->rq_offline(rq); | |
7723 | } | |
7724 | ||
c6c4927b | 7725 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7726 | rq->online = 0; |
7727 | } | |
7728 | } | |
7729 | ||
1da177e4 LT |
7730 | /* |
7731 | * migration_call - callback that gets triggered when a CPU is added. | |
7732 | * Here we can start up the necessary migration thread for the new CPU. | |
7733 | */ | |
48f24c4d IM |
7734 | static int __cpuinit |
7735 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7736 | { |
1da177e4 | 7737 | struct task_struct *p; |
48f24c4d | 7738 | int cpu = (long)hcpu; |
1da177e4 | 7739 | unsigned long flags; |
70b97a7f | 7740 | struct rq *rq; |
1da177e4 LT |
7741 | |
7742 | switch (action) { | |
5be9361c | 7743 | |
1da177e4 | 7744 | case CPU_UP_PREPARE: |
8bb78442 | 7745 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7746 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7747 | if (IS_ERR(p)) |
7748 | return NOTIFY_BAD; | |
1da177e4 LT |
7749 | kthread_bind(p, cpu); |
7750 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7751 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7752 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7753 | task_rq_unlock(rq, &flags); |
371cbb38 | 7754 | get_task_struct(p); |
1da177e4 | 7755 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7756 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7757 | break; |
48f24c4d | 7758 | |
1da177e4 | 7759 | case CPU_ONLINE: |
8bb78442 | 7760 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7761 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7762 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7763 | |
7764 | /* Update our root-domain */ | |
7765 | rq = cpu_rq(cpu); | |
7766 | spin_lock_irqsave(&rq->lock, flags); | |
7767 | if (rq->rd) { | |
c6c4927b | 7768 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7769 | |
7770 | set_rq_online(rq); | |
1f94ef59 GH |
7771 | } |
7772 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7773 | break; |
48f24c4d | 7774 | |
1da177e4 LT |
7775 | #ifdef CONFIG_HOTPLUG_CPU |
7776 | case CPU_UP_CANCELED: | |
8bb78442 | 7777 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7778 | if (!cpu_rq(cpu)->migration_thread) |
7779 | break; | |
41a2d6cf | 7780 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7781 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7782 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7783 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7784 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7785 | cpu_rq(cpu)->migration_thread = NULL; |
7786 | break; | |
48f24c4d | 7787 | |
1da177e4 | 7788 | case CPU_DEAD: |
8bb78442 | 7789 | case CPU_DEAD_FROZEN: |
470fd646 | 7790 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7791 | migrate_live_tasks(cpu); |
7792 | rq = cpu_rq(cpu); | |
7793 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7794 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7795 | rq->migration_thread = NULL; |
7796 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7797 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7798 | update_rq_clock(rq); |
2e1cb74a | 7799 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7800 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7801 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7802 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7803 | migrate_dead_tasks(cpu); |
d2da272a | 7804 | spin_unlock_irq(&rq->lock); |
470fd646 | 7805 | cpuset_unlock(); |
1da177e4 LT |
7806 | migrate_nr_uninterruptible(rq); |
7807 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7808 | calc_global_load_remove(rq); |
41a2d6cf IM |
7809 | /* |
7810 | * No need to migrate the tasks: it was best-effort if | |
7811 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7812 | * the requestors. | |
7813 | */ | |
1da177e4 LT |
7814 | spin_lock_irq(&rq->lock); |
7815 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7816 | struct migration_req *req; |
7817 | ||
1da177e4 | 7818 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7819 | struct migration_req, list); |
1da177e4 | 7820 | list_del_init(&req->list); |
9a2bd244 | 7821 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7822 | complete(&req->done); |
9a2bd244 | 7823 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7824 | } |
7825 | spin_unlock_irq(&rq->lock); | |
7826 | break; | |
57d885fe | 7827 | |
08f503b0 GH |
7828 | case CPU_DYING: |
7829 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7830 | /* Update our root-domain */ |
7831 | rq = cpu_rq(cpu); | |
7832 | spin_lock_irqsave(&rq->lock, flags); | |
7833 | if (rq->rd) { | |
c6c4927b | 7834 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7835 | set_rq_offline(rq); |
57d885fe GH |
7836 | } |
7837 | spin_unlock_irqrestore(&rq->lock, flags); | |
7838 | break; | |
1da177e4 LT |
7839 | #endif |
7840 | } | |
7841 | return NOTIFY_OK; | |
7842 | } | |
7843 | ||
f38b0820 PM |
7844 | /* |
7845 | * Register at high priority so that task migration (migrate_all_tasks) | |
7846 | * happens before everything else. This has to be lower priority than | |
7847 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7848 | */ |
26c2143b | 7849 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7850 | .notifier_call = migration_call, |
7851 | .priority = 10 | |
7852 | }; | |
7853 | ||
7babe8db | 7854 | static int __init migration_init(void) |
1da177e4 LT |
7855 | { |
7856 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7857 | int err; |
48f24c4d IM |
7858 | |
7859 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7860 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7861 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7862 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7863 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7864 | |
a004cd42 | 7865 | return 0; |
1da177e4 | 7866 | } |
7babe8db | 7867 | early_initcall(migration_init); |
1da177e4 LT |
7868 | #endif |
7869 | ||
7870 | #ifdef CONFIG_SMP | |
476f3534 | 7871 | |
3e9830dc | 7872 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7873 | |
7c16ec58 | 7874 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7875 | struct cpumask *groupmask) |
1da177e4 | 7876 | { |
4dcf6aff | 7877 | struct sched_group *group = sd->groups; |
434d53b0 | 7878 | char str[256]; |
1da177e4 | 7879 | |
968ea6d8 | 7880 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7881 | cpumask_clear(groupmask); |
4dcf6aff IM |
7882 | |
7883 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7884 | ||
7885 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7886 | printk("does not load-balance\n"); | |
7887 | if (sd->parent) | |
7888 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7889 | " has parent"); | |
7890 | return -1; | |
41c7ce9a NP |
7891 | } |
7892 | ||
eefd796a | 7893 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7894 | |
758b2cdc | 7895 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7896 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7897 | "CPU%d\n", cpu); | |
7898 | } | |
758b2cdc | 7899 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7900 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7901 | " CPU%d\n", cpu); | |
7902 | } | |
1da177e4 | 7903 | |
4dcf6aff | 7904 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7905 | do { |
4dcf6aff IM |
7906 | if (!group) { |
7907 | printk("\n"); | |
7908 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7909 | break; |
7910 | } | |
7911 | ||
18a3885f | 7912 | if (!group->cpu_power) { |
4dcf6aff IM |
7913 | printk(KERN_CONT "\n"); |
7914 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7915 | "set\n"); | |
7916 | break; | |
7917 | } | |
1da177e4 | 7918 | |
758b2cdc | 7919 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7920 | printk(KERN_CONT "\n"); |
7921 | printk(KERN_ERR "ERROR: empty group\n"); | |
7922 | break; | |
7923 | } | |
1da177e4 | 7924 | |
758b2cdc | 7925 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7926 | printk(KERN_CONT "\n"); |
7927 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7928 | break; | |
7929 | } | |
1da177e4 | 7930 | |
758b2cdc | 7931 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7932 | |
968ea6d8 | 7933 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7934 | |
7935 | printk(KERN_CONT " %s", str); | |
18a3885f PZ |
7936 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
7937 | printk(KERN_CONT " (cpu_power = %d)", | |
7938 | group->cpu_power); | |
381512cf | 7939 | } |
1da177e4 | 7940 | |
4dcf6aff IM |
7941 | group = group->next; |
7942 | } while (group != sd->groups); | |
7943 | printk(KERN_CONT "\n"); | |
1da177e4 | 7944 | |
758b2cdc | 7945 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7946 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7947 | |
758b2cdc RR |
7948 | if (sd->parent && |
7949 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7950 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7951 | "of domain->span\n"); | |
7952 | return 0; | |
7953 | } | |
1da177e4 | 7954 | |
4dcf6aff IM |
7955 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7956 | { | |
d5dd3db1 | 7957 | cpumask_var_t groupmask; |
4dcf6aff | 7958 | int level = 0; |
1da177e4 | 7959 | |
4dcf6aff IM |
7960 | if (!sd) { |
7961 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7962 | return; | |
7963 | } | |
1da177e4 | 7964 | |
4dcf6aff IM |
7965 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7966 | ||
d5dd3db1 | 7967 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7968 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7969 | return; | |
7970 | } | |
7971 | ||
4dcf6aff | 7972 | for (;;) { |
7c16ec58 | 7973 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7974 | break; |
1da177e4 LT |
7975 | level++; |
7976 | sd = sd->parent; | |
33859f7f | 7977 | if (!sd) |
4dcf6aff IM |
7978 | break; |
7979 | } | |
d5dd3db1 | 7980 | free_cpumask_var(groupmask); |
1da177e4 | 7981 | } |
6d6bc0ad | 7982 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7983 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7984 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7985 | |
1a20ff27 | 7986 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7987 | { |
758b2cdc | 7988 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7989 | return 1; |
7990 | ||
7991 | /* Following flags need at least 2 groups */ | |
7992 | if (sd->flags & (SD_LOAD_BALANCE | | |
7993 | SD_BALANCE_NEWIDLE | | |
7994 | SD_BALANCE_FORK | | |
89c4710e SS |
7995 | SD_BALANCE_EXEC | |
7996 | SD_SHARE_CPUPOWER | | |
7997 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7998 | if (sd->groups != sd->groups->next) |
7999 | return 0; | |
8000 | } | |
8001 | ||
8002 | /* Following flags don't use groups */ | |
8003 | if (sd->flags & (SD_WAKE_IDLE | | |
8004 | SD_WAKE_AFFINE | | |
8005 | SD_WAKE_BALANCE)) | |
8006 | return 0; | |
8007 | ||
8008 | return 1; | |
8009 | } | |
8010 | ||
48f24c4d IM |
8011 | static int |
8012 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
8013 | { |
8014 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
8015 | ||
8016 | if (sd_degenerate(parent)) | |
8017 | return 1; | |
8018 | ||
758b2cdc | 8019 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
8020 | return 0; |
8021 | ||
8022 | /* Does parent contain flags not in child? */ | |
8023 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
8024 | if (cflags & SD_WAKE_AFFINE) | |
8025 | pflags &= ~SD_WAKE_BALANCE; | |
8026 | /* Flags needing groups don't count if only 1 group in parent */ | |
8027 | if (parent->groups == parent->groups->next) { | |
8028 | pflags &= ~(SD_LOAD_BALANCE | | |
8029 | SD_BALANCE_NEWIDLE | | |
8030 | SD_BALANCE_FORK | | |
89c4710e SS |
8031 | SD_BALANCE_EXEC | |
8032 | SD_SHARE_CPUPOWER | | |
8033 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
8034 | if (nr_node_ids == 1) |
8035 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
8036 | } |
8037 | if (~cflags & pflags) | |
8038 | return 0; | |
8039 | ||
8040 | return 1; | |
8041 | } | |
8042 | ||
c6c4927b RR |
8043 | static void free_rootdomain(struct root_domain *rd) |
8044 | { | |
68e74568 RR |
8045 | cpupri_cleanup(&rd->cpupri); |
8046 | ||
c6c4927b RR |
8047 | free_cpumask_var(rd->rto_mask); |
8048 | free_cpumask_var(rd->online); | |
8049 | free_cpumask_var(rd->span); | |
8050 | kfree(rd); | |
8051 | } | |
8052 | ||
57d885fe GH |
8053 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
8054 | { | |
a0490fa3 | 8055 | struct root_domain *old_rd = NULL; |
57d885fe | 8056 | unsigned long flags; |
57d885fe GH |
8057 | |
8058 | spin_lock_irqsave(&rq->lock, flags); | |
8059 | ||
8060 | if (rq->rd) { | |
a0490fa3 | 8061 | old_rd = rq->rd; |
57d885fe | 8062 | |
c6c4927b | 8063 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 8064 | set_rq_offline(rq); |
57d885fe | 8065 | |
c6c4927b | 8066 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 8067 | |
a0490fa3 IM |
8068 | /* |
8069 | * If we dont want to free the old_rt yet then | |
8070 | * set old_rd to NULL to skip the freeing later | |
8071 | * in this function: | |
8072 | */ | |
8073 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
8074 | old_rd = NULL; | |
57d885fe GH |
8075 | } |
8076 | ||
8077 | atomic_inc(&rd->refcount); | |
8078 | rq->rd = rd; | |
8079 | ||
c6c4927b | 8080 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 8081 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 8082 | set_rq_online(rq); |
57d885fe GH |
8083 | |
8084 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
8085 | |
8086 | if (old_rd) | |
8087 | free_rootdomain(old_rd); | |
57d885fe GH |
8088 | } |
8089 | ||
fd5e1b5d | 8090 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 8091 | { |
36b7b6d4 PE |
8092 | gfp_t gfp = GFP_KERNEL; |
8093 | ||
57d885fe GH |
8094 | memset(rd, 0, sizeof(*rd)); |
8095 | ||
36b7b6d4 PE |
8096 | if (bootmem) |
8097 | gfp = GFP_NOWAIT; | |
c6c4927b | 8098 | |
36b7b6d4 | 8099 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 8100 | goto out; |
36b7b6d4 | 8101 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 8102 | goto free_span; |
36b7b6d4 | 8103 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 8104 | goto free_online; |
6e0534f2 | 8105 | |
0fb53029 | 8106 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 8107 | goto free_rto_mask; |
c6c4927b | 8108 | return 0; |
6e0534f2 | 8109 | |
68e74568 RR |
8110 | free_rto_mask: |
8111 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
8112 | free_online: |
8113 | free_cpumask_var(rd->online); | |
8114 | free_span: | |
8115 | free_cpumask_var(rd->span); | |
0c910d28 | 8116 | out: |
c6c4927b | 8117 | return -ENOMEM; |
57d885fe GH |
8118 | } |
8119 | ||
8120 | static void init_defrootdomain(void) | |
8121 | { | |
c6c4927b RR |
8122 | init_rootdomain(&def_root_domain, true); |
8123 | ||
57d885fe GH |
8124 | atomic_set(&def_root_domain.refcount, 1); |
8125 | } | |
8126 | ||
dc938520 | 8127 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
8128 | { |
8129 | struct root_domain *rd; | |
8130 | ||
8131 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
8132 | if (!rd) | |
8133 | return NULL; | |
8134 | ||
c6c4927b RR |
8135 | if (init_rootdomain(rd, false) != 0) { |
8136 | kfree(rd); | |
8137 | return NULL; | |
8138 | } | |
57d885fe GH |
8139 | |
8140 | return rd; | |
8141 | } | |
8142 | ||
1da177e4 | 8143 | /* |
0eab9146 | 8144 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
8145 | * hold the hotplug lock. |
8146 | */ | |
0eab9146 IM |
8147 | static void |
8148 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 8149 | { |
70b97a7f | 8150 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8151 | struct sched_domain *tmp; |
8152 | ||
8153 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8154 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8155 | struct sched_domain *parent = tmp->parent; |
8156 | if (!parent) | |
8157 | break; | |
f29c9b1c | 8158 | |
1a848870 | 8159 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8160 | tmp->parent = parent->parent; |
1a848870 SS |
8161 | if (parent->parent) |
8162 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8163 | } else |
8164 | tmp = tmp->parent; | |
245af2c7 SS |
8165 | } |
8166 | ||
1a848870 | 8167 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8168 | sd = sd->parent; |
1a848870 SS |
8169 | if (sd) |
8170 | sd->child = NULL; | |
8171 | } | |
1da177e4 LT |
8172 | |
8173 | sched_domain_debug(sd, cpu); | |
8174 | ||
57d885fe | 8175 | rq_attach_root(rq, rd); |
674311d5 | 8176 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8177 | } |
8178 | ||
8179 | /* cpus with isolated domains */ | |
dcc30a35 | 8180 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8181 | |
8182 | /* Setup the mask of cpus configured for isolated domains */ | |
8183 | static int __init isolated_cpu_setup(char *str) | |
8184 | { | |
968ea6d8 | 8185 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8186 | return 1; |
8187 | } | |
8188 | ||
8927f494 | 8189 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8190 | |
8191 | /* | |
6711cab4 SS |
8192 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8193 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8194 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8195 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8196 | * |
8197 | * init_sched_build_groups will build a circular linked list of the groups | |
8198 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8199 | * and ->cpu_power to 0. | |
8200 | */ | |
a616058b | 8201 | static void |
96f874e2 RR |
8202 | init_sched_build_groups(const struct cpumask *span, |
8203 | const struct cpumask *cpu_map, | |
8204 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8205 | struct sched_group **sg, |
96f874e2 RR |
8206 | struct cpumask *tmpmask), |
8207 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8208 | { |
8209 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8210 | int i; |
8211 | ||
96f874e2 | 8212 | cpumask_clear(covered); |
7c16ec58 | 8213 | |
abcd083a | 8214 | for_each_cpu(i, span) { |
6711cab4 | 8215 | struct sched_group *sg; |
7c16ec58 | 8216 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8217 | int j; |
8218 | ||
758b2cdc | 8219 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8220 | continue; |
8221 | ||
758b2cdc | 8222 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 8223 | sg->cpu_power = 0; |
1da177e4 | 8224 | |
abcd083a | 8225 | for_each_cpu(j, span) { |
7c16ec58 | 8226 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8227 | continue; |
8228 | ||
96f874e2 | 8229 | cpumask_set_cpu(j, covered); |
758b2cdc | 8230 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8231 | } |
8232 | if (!first) | |
8233 | first = sg; | |
8234 | if (last) | |
8235 | last->next = sg; | |
8236 | last = sg; | |
8237 | } | |
8238 | last->next = first; | |
8239 | } | |
8240 | ||
9c1cfda2 | 8241 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8242 | |
9c1cfda2 | 8243 | #ifdef CONFIG_NUMA |
198e2f18 | 8244 | |
9c1cfda2 JH |
8245 | /** |
8246 | * find_next_best_node - find the next node to include in a sched_domain | |
8247 | * @node: node whose sched_domain we're building | |
8248 | * @used_nodes: nodes already in the sched_domain | |
8249 | * | |
41a2d6cf | 8250 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8251 | * finds the closest node not already in the @used_nodes map. |
8252 | * | |
8253 | * Should use nodemask_t. | |
8254 | */ | |
c5f59f08 | 8255 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8256 | { |
8257 | int i, n, val, min_val, best_node = 0; | |
8258 | ||
8259 | min_val = INT_MAX; | |
8260 | ||
076ac2af | 8261 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8262 | /* Start at @node */ |
076ac2af | 8263 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8264 | |
8265 | if (!nr_cpus_node(n)) | |
8266 | continue; | |
8267 | ||
8268 | /* Skip already used nodes */ | |
c5f59f08 | 8269 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8270 | continue; |
8271 | ||
8272 | /* Simple min distance search */ | |
8273 | val = node_distance(node, n); | |
8274 | ||
8275 | if (val < min_val) { | |
8276 | min_val = val; | |
8277 | best_node = n; | |
8278 | } | |
8279 | } | |
8280 | ||
c5f59f08 | 8281 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8282 | return best_node; |
8283 | } | |
8284 | ||
8285 | /** | |
8286 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8287 | * @node: node whose cpumask we're constructing | |
73486722 | 8288 | * @span: resulting cpumask |
9c1cfda2 | 8289 | * |
41a2d6cf | 8290 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8291 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8292 | * out optimally. | |
8293 | */ | |
96f874e2 | 8294 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8295 | { |
c5f59f08 | 8296 | nodemask_t used_nodes; |
48f24c4d | 8297 | int i; |
9c1cfda2 | 8298 | |
6ca09dfc | 8299 | cpumask_clear(span); |
c5f59f08 | 8300 | nodes_clear(used_nodes); |
9c1cfda2 | 8301 | |
6ca09dfc | 8302 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8303 | node_set(node, used_nodes); |
9c1cfda2 JH |
8304 | |
8305 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8306 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8307 | |
6ca09dfc | 8308 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8309 | } |
9c1cfda2 | 8310 | } |
6d6bc0ad | 8311 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8312 | |
5c45bf27 | 8313 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8314 | |
6c99e9ad RR |
8315 | /* |
8316 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8317 | * |
8318 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8319 | * and struct sched_domain. ) | |
6c99e9ad RR |
8320 | */ |
8321 | struct static_sched_group { | |
8322 | struct sched_group sg; | |
8323 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8324 | }; | |
8325 | ||
8326 | struct static_sched_domain { | |
8327 | struct sched_domain sd; | |
8328 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8329 | }; | |
8330 | ||
49a02c51 AH |
8331 | struct s_data { |
8332 | #ifdef CONFIG_NUMA | |
8333 | int sd_allnodes; | |
8334 | cpumask_var_t domainspan; | |
8335 | cpumask_var_t covered; | |
8336 | cpumask_var_t notcovered; | |
8337 | #endif | |
8338 | cpumask_var_t nodemask; | |
8339 | cpumask_var_t this_sibling_map; | |
8340 | cpumask_var_t this_core_map; | |
8341 | cpumask_var_t send_covered; | |
8342 | cpumask_var_t tmpmask; | |
8343 | struct sched_group **sched_group_nodes; | |
8344 | struct root_domain *rd; | |
8345 | }; | |
8346 | ||
2109b99e AH |
8347 | enum s_alloc { |
8348 | sa_sched_groups = 0, | |
8349 | sa_rootdomain, | |
8350 | sa_tmpmask, | |
8351 | sa_send_covered, | |
8352 | sa_this_core_map, | |
8353 | sa_this_sibling_map, | |
8354 | sa_nodemask, | |
8355 | sa_sched_group_nodes, | |
8356 | #ifdef CONFIG_NUMA | |
8357 | sa_notcovered, | |
8358 | sa_covered, | |
8359 | sa_domainspan, | |
8360 | #endif | |
8361 | sa_none, | |
8362 | }; | |
8363 | ||
9c1cfda2 | 8364 | /* |
48f24c4d | 8365 | * SMT sched-domains: |
9c1cfda2 | 8366 | */ |
1da177e4 | 8367 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8368 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8369 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8370 | |
41a2d6cf | 8371 | static int |
96f874e2 RR |
8372 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8373 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8374 | { |
6711cab4 | 8375 | if (sg) |
6c99e9ad | 8376 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8377 | return cpu; |
8378 | } | |
6d6bc0ad | 8379 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8380 | |
48f24c4d IM |
8381 | /* |
8382 | * multi-core sched-domains: | |
8383 | */ | |
1e9f28fa | 8384 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8385 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8386 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8387 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8388 | |
8389 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8390 | static int |
96f874e2 RR |
8391 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8392 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8393 | { |
6711cab4 | 8394 | int group; |
7c16ec58 | 8395 | |
c69fc56d | 8396 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8397 | group = cpumask_first(mask); |
6711cab4 | 8398 | if (sg) |
6c99e9ad | 8399 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8400 | return group; |
1e9f28fa SS |
8401 | } |
8402 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8403 | static int |
96f874e2 RR |
8404 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8405 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8406 | { |
6711cab4 | 8407 | if (sg) |
6c99e9ad | 8408 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8409 | return cpu; |
8410 | } | |
8411 | #endif | |
8412 | ||
6c99e9ad RR |
8413 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8414 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8415 | |
41a2d6cf | 8416 | static int |
96f874e2 RR |
8417 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8418 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8419 | { |
6711cab4 | 8420 | int group; |
48f24c4d | 8421 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8422 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8423 | group = cpumask_first(mask); |
1e9f28fa | 8424 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8425 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8426 | group = cpumask_first(mask); |
1da177e4 | 8427 | #else |
6711cab4 | 8428 | group = cpu; |
1da177e4 | 8429 | #endif |
6711cab4 | 8430 | if (sg) |
6c99e9ad | 8431 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8432 | return group; |
1da177e4 LT |
8433 | } |
8434 | ||
8435 | #ifdef CONFIG_NUMA | |
1da177e4 | 8436 | /* |
9c1cfda2 JH |
8437 | * The init_sched_build_groups can't handle what we want to do with node |
8438 | * groups, so roll our own. Now each node has its own list of groups which | |
8439 | * gets dynamically allocated. | |
1da177e4 | 8440 | */ |
62ea9ceb | 8441 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8442 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8443 | |
62ea9ceb | 8444 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8445 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8446 | |
96f874e2 RR |
8447 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8448 | struct sched_group **sg, | |
8449 | struct cpumask *nodemask) | |
9c1cfda2 | 8450 | { |
6711cab4 SS |
8451 | int group; |
8452 | ||
6ca09dfc | 8453 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8454 | group = cpumask_first(nodemask); |
6711cab4 SS |
8455 | |
8456 | if (sg) | |
6c99e9ad | 8457 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8458 | return group; |
1da177e4 | 8459 | } |
6711cab4 | 8460 | |
08069033 SS |
8461 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8462 | { | |
8463 | struct sched_group *sg = group_head; | |
8464 | int j; | |
8465 | ||
8466 | if (!sg) | |
8467 | return; | |
3a5c359a | 8468 | do { |
758b2cdc | 8469 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8470 | struct sched_domain *sd; |
08069033 | 8471 | |
6c99e9ad | 8472 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8473 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8474 | /* |
8475 | * Only add "power" once for each | |
8476 | * physical package. | |
8477 | */ | |
8478 | continue; | |
8479 | } | |
08069033 | 8480 | |
18a3885f | 8481 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
8482 | } |
8483 | sg = sg->next; | |
8484 | } while (sg != group_head); | |
08069033 | 8485 | } |
0601a88d AH |
8486 | |
8487 | static int build_numa_sched_groups(struct s_data *d, | |
8488 | const struct cpumask *cpu_map, int num) | |
8489 | { | |
8490 | struct sched_domain *sd; | |
8491 | struct sched_group *sg, *prev; | |
8492 | int n, j; | |
8493 | ||
8494 | cpumask_clear(d->covered); | |
8495 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8496 | if (cpumask_empty(d->nodemask)) { | |
8497 | d->sched_group_nodes[num] = NULL; | |
8498 | goto out; | |
8499 | } | |
8500 | ||
8501 | sched_domain_node_span(num, d->domainspan); | |
8502 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8503 | ||
8504 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8505 | GFP_KERNEL, num); | |
8506 | if (!sg) { | |
8507 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", | |
8508 | num); | |
8509 | return -ENOMEM; | |
8510 | } | |
8511 | d->sched_group_nodes[num] = sg; | |
8512 | ||
8513 | for_each_cpu(j, d->nodemask) { | |
8514 | sd = &per_cpu(node_domains, j).sd; | |
8515 | sd->groups = sg; | |
8516 | } | |
8517 | ||
18a3885f | 8518 | sg->cpu_power = 0; |
0601a88d AH |
8519 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
8520 | sg->next = sg; | |
8521 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8522 | ||
8523 | prev = sg; | |
8524 | for (j = 0; j < nr_node_ids; j++) { | |
8525 | n = (num + j) % nr_node_ids; | |
8526 | cpumask_complement(d->notcovered, d->covered); | |
8527 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8528 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8529 | if (cpumask_empty(d->tmpmask)) | |
8530 | break; | |
8531 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8532 | if (cpumask_empty(d->tmpmask)) | |
8533 | continue; | |
8534 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8535 | GFP_KERNEL, num); | |
8536 | if (!sg) { | |
8537 | printk(KERN_WARNING | |
8538 | "Can not alloc domain group for node %d\n", j); | |
8539 | return -ENOMEM; | |
8540 | } | |
18a3885f | 8541 | sg->cpu_power = 0; |
0601a88d AH |
8542 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
8543 | sg->next = prev->next; | |
8544 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8545 | prev->next = sg; | |
8546 | prev = sg; | |
8547 | } | |
8548 | out: | |
8549 | return 0; | |
8550 | } | |
6d6bc0ad | 8551 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8552 | |
a616058b | 8553 | #ifdef CONFIG_NUMA |
51888ca2 | 8554 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8555 | static void free_sched_groups(const struct cpumask *cpu_map, |
8556 | struct cpumask *nodemask) | |
51888ca2 | 8557 | { |
a616058b | 8558 | int cpu, i; |
51888ca2 | 8559 | |
abcd083a | 8560 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8561 | struct sched_group **sched_group_nodes |
8562 | = sched_group_nodes_bycpu[cpu]; | |
8563 | ||
51888ca2 SV |
8564 | if (!sched_group_nodes) |
8565 | continue; | |
8566 | ||
076ac2af | 8567 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8568 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8569 | ||
6ca09dfc | 8570 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8571 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8572 | continue; |
8573 | ||
8574 | if (sg == NULL) | |
8575 | continue; | |
8576 | sg = sg->next; | |
8577 | next_sg: | |
8578 | oldsg = sg; | |
8579 | sg = sg->next; | |
8580 | kfree(oldsg); | |
8581 | if (oldsg != sched_group_nodes[i]) | |
8582 | goto next_sg; | |
8583 | } | |
8584 | kfree(sched_group_nodes); | |
8585 | sched_group_nodes_bycpu[cpu] = NULL; | |
8586 | } | |
51888ca2 | 8587 | } |
6d6bc0ad | 8588 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8589 | static void free_sched_groups(const struct cpumask *cpu_map, |
8590 | struct cpumask *nodemask) | |
a616058b SS |
8591 | { |
8592 | } | |
6d6bc0ad | 8593 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8594 | |
89c4710e SS |
8595 | /* |
8596 | * Initialize sched groups cpu_power. | |
8597 | * | |
8598 | * cpu_power indicates the capacity of sched group, which is used while | |
8599 | * distributing the load between different sched groups in a sched domain. | |
8600 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8601 | * there are asymmetries in the topology. If there are asymmetries, group | |
8602 | * having more cpu_power will pickup more load compared to the group having | |
8603 | * less cpu_power. | |
89c4710e SS |
8604 | */ |
8605 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8606 | { | |
8607 | struct sched_domain *child; | |
8608 | struct sched_group *group; | |
f93e65c1 PZ |
8609 | long power; |
8610 | int weight; | |
89c4710e SS |
8611 | |
8612 | WARN_ON(!sd || !sd->groups); | |
8613 | ||
13318a71 | 8614 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8615 | return; |
8616 | ||
8617 | child = sd->child; | |
8618 | ||
18a3885f | 8619 | sd->groups->cpu_power = 0; |
5517d86b | 8620 | |
f93e65c1 PZ |
8621 | if (!child) { |
8622 | power = SCHED_LOAD_SCALE; | |
8623 | weight = cpumask_weight(sched_domain_span(sd)); | |
8624 | /* | |
8625 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8626 | * Usually multiple threads get a better yield out of |
8627 | * that one core than a single thread would have, | |
8628 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8629 | */ |
a52bfd73 PZ |
8630 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8631 | power *= sd->smt_gain; | |
f93e65c1 | 8632 | power /= weight; |
a52bfd73 PZ |
8633 | power >>= SCHED_LOAD_SHIFT; |
8634 | } | |
18a3885f | 8635 | sd->groups->cpu_power += power; |
89c4710e SS |
8636 | return; |
8637 | } | |
8638 | ||
89c4710e | 8639 | /* |
f93e65c1 | 8640 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8641 | */ |
8642 | group = child->groups; | |
8643 | do { | |
18a3885f | 8644 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
8645 | group = group->next; |
8646 | } while (group != child->groups); | |
8647 | } | |
8648 | ||
7c16ec58 MT |
8649 | /* |
8650 | * Initializers for schedule domains | |
8651 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8652 | */ | |
8653 | ||
a5d8c348 IM |
8654 | #ifdef CONFIG_SCHED_DEBUG |
8655 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8656 | #else | |
8657 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8658 | #endif | |
8659 | ||
7c16ec58 | 8660 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8661 | |
7c16ec58 MT |
8662 | #define SD_INIT_FUNC(type) \ |
8663 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8664 | { \ | |
8665 | memset(sd, 0, sizeof(*sd)); \ | |
8666 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8667 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8668 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8669 | } |
8670 | ||
8671 | SD_INIT_FUNC(CPU) | |
8672 | #ifdef CONFIG_NUMA | |
8673 | SD_INIT_FUNC(ALLNODES) | |
8674 | SD_INIT_FUNC(NODE) | |
8675 | #endif | |
8676 | #ifdef CONFIG_SCHED_SMT | |
8677 | SD_INIT_FUNC(SIBLING) | |
8678 | #endif | |
8679 | #ifdef CONFIG_SCHED_MC | |
8680 | SD_INIT_FUNC(MC) | |
8681 | #endif | |
8682 | ||
1d3504fc HS |
8683 | static int default_relax_domain_level = -1; |
8684 | ||
8685 | static int __init setup_relax_domain_level(char *str) | |
8686 | { | |
30e0e178 LZ |
8687 | unsigned long val; |
8688 | ||
8689 | val = simple_strtoul(str, NULL, 0); | |
8690 | if (val < SD_LV_MAX) | |
8691 | default_relax_domain_level = val; | |
8692 | ||
1d3504fc HS |
8693 | return 1; |
8694 | } | |
8695 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8696 | ||
8697 | static void set_domain_attribute(struct sched_domain *sd, | |
8698 | struct sched_domain_attr *attr) | |
8699 | { | |
8700 | int request; | |
8701 | ||
8702 | if (!attr || attr->relax_domain_level < 0) { | |
8703 | if (default_relax_domain_level < 0) | |
8704 | return; | |
8705 | else | |
8706 | request = default_relax_domain_level; | |
8707 | } else | |
8708 | request = attr->relax_domain_level; | |
8709 | if (request < sd->level) { | |
8710 | /* turn off idle balance on this domain */ | |
8711 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8712 | } else { | |
8713 | /* turn on idle balance on this domain */ | |
8714 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8715 | } | |
8716 | } | |
8717 | ||
2109b99e AH |
8718 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8719 | const struct cpumask *cpu_map) | |
8720 | { | |
8721 | switch (what) { | |
8722 | case sa_sched_groups: | |
8723 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8724 | d->sched_group_nodes = NULL; | |
8725 | case sa_rootdomain: | |
8726 | free_rootdomain(d->rd); /* fall through */ | |
8727 | case sa_tmpmask: | |
8728 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8729 | case sa_send_covered: | |
8730 | free_cpumask_var(d->send_covered); /* fall through */ | |
8731 | case sa_this_core_map: | |
8732 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8733 | case sa_this_sibling_map: | |
8734 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8735 | case sa_nodemask: | |
8736 | free_cpumask_var(d->nodemask); /* fall through */ | |
8737 | case sa_sched_group_nodes: | |
d1b55138 | 8738 | #ifdef CONFIG_NUMA |
2109b99e AH |
8739 | kfree(d->sched_group_nodes); /* fall through */ |
8740 | case sa_notcovered: | |
8741 | free_cpumask_var(d->notcovered); /* fall through */ | |
8742 | case sa_covered: | |
8743 | free_cpumask_var(d->covered); /* fall through */ | |
8744 | case sa_domainspan: | |
8745 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8746 | #endif |
2109b99e AH |
8747 | case sa_none: |
8748 | break; | |
8749 | } | |
8750 | } | |
3404c8d9 | 8751 | |
2109b99e AH |
8752 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8753 | const struct cpumask *cpu_map) | |
8754 | { | |
3404c8d9 | 8755 | #ifdef CONFIG_NUMA |
2109b99e AH |
8756 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8757 | return sa_none; | |
8758 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8759 | return sa_domainspan; | |
8760 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8761 | return sa_covered; | |
8762 | /* Allocate the per-node list of sched groups */ | |
8763 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8764 | sizeof(struct sched_group *), GFP_KERNEL); | |
8765 | if (!d->sched_group_nodes) { | |
d1b55138 | 8766 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 8767 | return sa_notcovered; |
d1b55138 | 8768 | } |
2109b99e | 8769 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8770 | #endif |
2109b99e AH |
8771 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8772 | return sa_sched_group_nodes; | |
8773 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8774 | return sa_nodemask; | |
8775 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8776 | return sa_this_sibling_map; | |
8777 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8778 | return sa_this_core_map; | |
8779 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8780 | return sa_send_covered; | |
8781 | d->rd = alloc_rootdomain(); | |
8782 | if (!d->rd) { | |
57d885fe | 8783 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 8784 | return sa_tmpmask; |
57d885fe | 8785 | } |
2109b99e AH |
8786 | return sa_rootdomain; |
8787 | } | |
57d885fe | 8788 | |
7f4588f3 AH |
8789 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8790 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8791 | { | |
8792 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8793 | #ifdef CONFIG_NUMA |
7f4588f3 | 8794 | struct sched_domain *parent; |
1da177e4 | 8795 | |
7f4588f3 AH |
8796 | d->sd_allnodes = 0; |
8797 | if (cpumask_weight(cpu_map) > | |
8798 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8799 | sd = &per_cpu(allnodes_domains, i).sd; | |
8800 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8801 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8802 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8803 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8804 | d->sd_allnodes = 1; | |
8805 | } | |
8806 | parent = sd; | |
8807 | ||
8808 | sd = &per_cpu(node_domains, i).sd; | |
8809 | SD_INIT(sd, NODE); | |
8810 | set_domain_attribute(sd, attr); | |
8811 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8812 | sd->parent = parent; | |
8813 | if (parent) | |
8814 | parent->child = sd; | |
8815 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8816 | #endif |
7f4588f3 AH |
8817 | return sd; |
8818 | } | |
1da177e4 | 8819 | |
87cce662 AH |
8820 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8821 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8822 | struct sched_domain *parent, int i) | |
8823 | { | |
8824 | struct sched_domain *sd; | |
8825 | sd = &per_cpu(phys_domains, i).sd; | |
8826 | SD_INIT(sd, CPU); | |
8827 | set_domain_attribute(sd, attr); | |
8828 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8829 | sd->parent = parent; | |
8830 | if (parent) | |
8831 | parent->child = sd; | |
8832 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8833 | return sd; | |
8834 | } | |
1da177e4 | 8835 | |
410c4081 AH |
8836 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8837 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8838 | struct sched_domain *parent, int i) | |
8839 | { | |
8840 | struct sched_domain *sd = parent; | |
1e9f28fa | 8841 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8842 | sd = &per_cpu(core_domains, i).sd; |
8843 | SD_INIT(sd, MC); | |
8844 | set_domain_attribute(sd, attr); | |
8845 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8846 | sd->parent = parent; | |
8847 | parent->child = sd; | |
8848 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8849 | #endif |
410c4081 AH |
8850 | return sd; |
8851 | } | |
1e9f28fa | 8852 | |
d8173535 AH |
8853 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8854 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8855 | struct sched_domain *parent, int i) | |
8856 | { | |
8857 | struct sched_domain *sd = parent; | |
1da177e4 | 8858 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8859 | sd = &per_cpu(cpu_domains, i).sd; |
8860 | SD_INIT(sd, SIBLING); | |
8861 | set_domain_attribute(sd, attr); | |
8862 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8863 | sd->parent = parent; | |
8864 | parent->child = sd; | |
8865 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8866 | #endif |
d8173535 AH |
8867 | return sd; |
8868 | } | |
1da177e4 | 8869 | |
0e8e85c9 AH |
8870 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8871 | const struct cpumask *cpu_map, int cpu) | |
8872 | { | |
8873 | switch (l) { | |
1da177e4 | 8874 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8875 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8876 | cpumask_and(d->this_sibling_map, cpu_map, | |
8877 | topology_thread_cpumask(cpu)); | |
8878 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8879 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8880 | &cpu_to_cpu_group, | |
8881 | d->send_covered, d->tmpmask); | |
8882 | break; | |
1da177e4 | 8883 | #endif |
1e9f28fa | 8884 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8885 | case SD_LV_MC: /* set up multi-core groups */ |
8886 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8887 | if (cpu == cpumask_first(d->this_core_map)) | |
8888 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8889 | &cpu_to_core_group, | |
8890 | d->send_covered, d->tmpmask); | |
8891 | break; | |
1e9f28fa | 8892 | #endif |
86548096 AH |
8893 | case SD_LV_CPU: /* set up physical groups */ |
8894 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8895 | if (!cpumask_empty(d->nodemask)) | |
8896 | init_sched_build_groups(d->nodemask, cpu_map, | |
8897 | &cpu_to_phys_group, | |
8898 | d->send_covered, d->tmpmask); | |
8899 | break; | |
1da177e4 | 8900 | #ifdef CONFIG_NUMA |
de616e36 AH |
8901 | case SD_LV_ALLNODES: |
8902 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8903 | d->send_covered, d->tmpmask); | |
8904 | break; | |
8905 | #endif | |
0e8e85c9 AH |
8906 | default: |
8907 | break; | |
7c16ec58 | 8908 | } |
0e8e85c9 | 8909 | } |
9c1cfda2 | 8910 | |
2109b99e AH |
8911 | /* |
8912 | * Build sched domains for a given set of cpus and attach the sched domains | |
8913 | * to the individual cpus | |
8914 | */ | |
8915 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8916 | struct sched_domain_attr *attr) | |
8917 | { | |
8918 | enum s_alloc alloc_state = sa_none; | |
8919 | struct s_data d; | |
294b0c96 | 8920 | struct sched_domain *sd; |
2109b99e | 8921 | int i; |
7c16ec58 | 8922 | #ifdef CONFIG_NUMA |
2109b99e | 8923 | d.sd_allnodes = 0; |
7c16ec58 | 8924 | #endif |
9c1cfda2 | 8925 | |
2109b99e AH |
8926 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8927 | if (alloc_state != sa_rootdomain) | |
8928 | goto error; | |
8929 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8930 | |
1da177e4 | 8931 | /* |
1a20ff27 | 8932 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8933 | */ |
abcd083a | 8934 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8935 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8936 | cpu_map); | |
9761eea8 | 8937 | |
7f4588f3 | 8938 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8939 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8940 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8941 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8942 | } |
9c1cfda2 | 8943 | |
abcd083a | 8944 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8945 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8946 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8947 | } |
9c1cfda2 | 8948 | |
1da177e4 | 8949 | /* Set up physical groups */ |
86548096 AH |
8950 | for (i = 0; i < nr_node_ids; i++) |
8951 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8952 | |
1da177e4 LT |
8953 | #ifdef CONFIG_NUMA |
8954 | /* Set up node groups */ | |
de616e36 AH |
8955 | if (d.sd_allnodes) |
8956 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8957 | |
0601a88d AH |
8958 | for (i = 0; i < nr_node_ids; i++) |
8959 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8960 | goto error; |
1da177e4 LT |
8961 | #endif |
8962 | ||
8963 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8964 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8965 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8966 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8967 | init_sched_groups_power(i, sd); |
5c45bf27 | 8968 | } |
1da177e4 | 8969 | #endif |
1e9f28fa | 8970 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8971 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8972 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8973 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8974 | } |
8975 | #endif | |
1e9f28fa | 8976 | |
abcd083a | 8977 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8978 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8979 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8980 | } |
8981 | ||
9c1cfda2 | 8982 | #ifdef CONFIG_NUMA |
076ac2af | 8983 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8984 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8985 | |
49a02c51 | 8986 | if (d.sd_allnodes) { |
6711cab4 | 8987 | struct sched_group *sg; |
f712c0c7 | 8988 | |
96f874e2 | 8989 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 8990 | d.tmpmask); |
f712c0c7 SS |
8991 | init_numa_sched_groups_power(sg); |
8992 | } | |
9c1cfda2 JH |
8993 | #endif |
8994 | ||
1da177e4 | 8995 | /* Attach the domains */ |
abcd083a | 8996 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8997 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8998 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8999 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 9000 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 9001 | #else |
6c99e9ad | 9002 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 9003 | #endif |
49a02c51 | 9004 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 9005 | } |
51888ca2 | 9006 | |
2109b99e AH |
9007 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
9008 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
9009 | return 0; | |
51888ca2 | 9010 | |
51888ca2 | 9011 | error: |
2109b99e AH |
9012 | __free_domain_allocs(&d, alloc_state, cpu_map); |
9013 | return -ENOMEM; | |
1da177e4 | 9014 | } |
029190c5 | 9015 | |
96f874e2 | 9016 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
9017 | { |
9018 | return __build_sched_domains(cpu_map, NULL); | |
9019 | } | |
9020 | ||
96f874e2 | 9021 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 9022 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
9023 | static struct sched_domain_attr *dattr_cur; |
9024 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
9025 | |
9026 | /* | |
9027 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
9028 | * cpumask) fails, then fallback to a single sched domain, |
9029 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 9030 | */ |
4212823f | 9031 | static cpumask_var_t fallback_doms; |
029190c5 | 9032 | |
ee79d1bd HC |
9033 | /* |
9034 | * arch_update_cpu_topology lets virtualized architectures update the | |
9035 | * cpu core maps. It is supposed to return 1 if the topology changed | |
9036 | * or 0 if it stayed the same. | |
9037 | */ | |
9038 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 9039 | { |
ee79d1bd | 9040 | return 0; |
22e52b07 HC |
9041 | } |
9042 | ||
1a20ff27 | 9043 | /* |
41a2d6cf | 9044 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
9045 | * For now this just excludes isolated cpus, but could be used to |
9046 | * exclude other special cases in the future. | |
1a20ff27 | 9047 | */ |
96f874e2 | 9048 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9049 | { |
7378547f MM |
9050 | int err; |
9051 | ||
22e52b07 | 9052 | arch_update_cpu_topology(); |
029190c5 | 9053 | ndoms_cur = 1; |
96f874e2 | 9054 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 9055 | if (!doms_cur) |
4212823f | 9056 | doms_cur = fallback_doms; |
dcc30a35 | 9057 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 9058 | dattr_cur = NULL; |
7378547f | 9059 | err = build_sched_domains(doms_cur); |
6382bc90 | 9060 | register_sched_domain_sysctl(); |
7378547f MM |
9061 | |
9062 | return err; | |
1a20ff27 DG |
9063 | } |
9064 | ||
96f874e2 RR |
9065 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
9066 | struct cpumask *tmpmask) | |
1da177e4 | 9067 | { |
7c16ec58 | 9068 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 9069 | } |
1da177e4 | 9070 | |
1a20ff27 DG |
9071 | /* |
9072 | * Detach sched domains from a group of cpus specified in cpu_map | |
9073 | * These cpus will now be attached to the NULL domain | |
9074 | */ | |
96f874e2 | 9075 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9076 | { |
96f874e2 RR |
9077 | /* Save because hotplug lock held. */ |
9078 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
9079 | int i; |
9080 | ||
abcd083a | 9081 | for_each_cpu(i, cpu_map) |
57d885fe | 9082 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 9083 | synchronize_sched(); |
96f874e2 | 9084 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
9085 | } |
9086 | ||
1d3504fc HS |
9087 | /* handle null as "default" */ |
9088 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
9089 | struct sched_domain_attr *new, int idx_new) | |
9090 | { | |
9091 | struct sched_domain_attr tmp; | |
9092 | ||
9093 | /* fast path */ | |
9094 | if (!new && !cur) | |
9095 | return 1; | |
9096 | ||
9097 | tmp = SD_ATTR_INIT; | |
9098 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
9099 | new ? (new + idx_new) : &tmp, | |
9100 | sizeof(struct sched_domain_attr)); | |
9101 | } | |
9102 | ||
029190c5 PJ |
9103 | /* |
9104 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 9105 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
9106 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
9107 | * It destroys each deleted domain and builds each new domain. | |
9108 | * | |
96f874e2 | 9109 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
9110 | * The masks don't intersect (don't overlap.) We should setup one |
9111 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
9112 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
9113 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
9114 | * it as it is. | |
9115 | * | |
41a2d6cf IM |
9116 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
9117 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
9118 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
9119 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
9120 | * the single partition 'fallback_doms', it also forces the domains | |
9121 | * to be rebuilt. | |
029190c5 | 9122 | * |
96f874e2 | 9123 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
9124 | * ndoms_new == 0 is a special case for destroying existing domains, |
9125 | * and it will not create the default domain. | |
dfb512ec | 9126 | * |
029190c5 PJ |
9127 | * Call with hotplug lock held |
9128 | */ | |
96f874e2 RR |
9129 | /* FIXME: Change to struct cpumask *doms_new[] */ |
9130 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 9131 | struct sched_domain_attr *dattr_new) |
029190c5 | 9132 | { |
dfb512ec | 9133 | int i, j, n; |
d65bd5ec | 9134 | int new_topology; |
029190c5 | 9135 | |
712555ee | 9136 | mutex_lock(&sched_domains_mutex); |
a1835615 | 9137 | |
7378547f MM |
9138 | /* always unregister in case we don't destroy any domains */ |
9139 | unregister_sched_domain_sysctl(); | |
9140 | ||
d65bd5ec HC |
9141 | /* Let architecture update cpu core mappings. */ |
9142 | new_topology = arch_update_cpu_topology(); | |
9143 | ||
dfb512ec | 9144 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
9145 | |
9146 | /* Destroy deleted domains */ | |
9147 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 9148 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 9149 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 9150 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9151 | goto match1; |
9152 | } | |
9153 | /* no match - a current sched domain not in new doms_new[] */ | |
9154 | detach_destroy_domains(doms_cur + i); | |
9155 | match1: | |
9156 | ; | |
9157 | } | |
9158 | ||
e761b772 MK |
9159 | if (doms_new == NULL) { |
9160 | ndoms_cur = 0; | |
4212823f | 9161 | doms_new = fallback_doms; |
dcc30a35 | 9162 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 9163 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9164 | } |
9165 | ||
029190c5 PJ |
9166 | /* Build new domains */ |
9167 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9168 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 9169 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 9170 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9171 | goto match2; |
9172 | } | |
9173 | /* no match - add a new doms_new */ | |
1d3504fc HS |
9174 | __build_sched_domains(doms_new + i, |
9175 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
9176 | match2: |
9177 | ; | |
9178 | } | |
9179 | ||
9180 | /* Remember the new sched domains */ | |
4212823f | 9181 | if (doms_cur != fallback_doms) |
029190c5 | 9182 | kfree(doms_cur); |
1d3504fc | 9183 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9184 | doms_cur = doms_new; |
1d3504fc | 9185 | dattr_cur = dattr_new; |
029190c5 | 9186 | ndoms_cur = ndoms_new; |
7378547f MM |
9187 | |
9188 | register_sched_domain_sysctl(); | |
a1835615 | 9189 | |
712555ee | 9190 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9191 | } |
9192 | ||
5c45bf27 | 9193 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9194 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9195 | { |
95402b38 | 9196 | get_online_cpus(); |
dfb512ec MK |
9197 | |
9198 | /* Destroy domains first to force the rebuild */ | |
9199 | partition_sched_domains(0, NULL, NULL); | |
9200 | ||
e761b772 | 9201 | rebuild_sched_domains(); |
95402b38 | 9202 | put_online_cpus(); |
5c45bf27 SS |
9203 | } |
9204 | ||
9205 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9206 | { | |
afb8a9b7 | 9207 | unsigned int level = 0; |
5c45bf27 | 9208 | |
afb8a9b7 GS |
9209 | if (sscanf(buf, "%u", &level) != 1) |
9210 | return -EINVAL; | |
9211 | ||
9212 | /* | |
9213 | * level is always be positive so don't check for | |
9214 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9215 | * What happens on 0 or 1 byte write, | |
9216 | * need to check for count as well? | |
9217 | */ | |
9218 | ||
9219 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9220 | return -EINVAL; |
9221 | ||
9222 | if (smt) | |
afb8a9b7 | 9223 | sched_smt_power_savings = level; |
5c45bf27 | 9224 | else |
afb8a9b7 | 9225 | sched_mc_power_savings = level; |
5c45bf27 | 9226 | |
c70f22d2 | 9227 | arch_reinit_sched_domains(); |
5c45bf27 | 9228 | |
c70f22d2 | 9229 | return count; |
5c45bf27 SS |
9230 | } |
9231 | ||
5c45bf27 | 9232 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9233 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9234 | char *page) | |
5c45bf27 SS |
9235 | { |
9236 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9237 | } | |
f718cd4a | 9238 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9239 | const char *buf, size_t count) |
5c45bf27 SS |
9240 | { |
9241 | return sched_power_savings_store(buf, count, 0); | |
9242 | } | |
f718cd4a AK |
9243 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9244 | sched_mc_power_savings_show, | |
9245 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9246 | #endif |
9247 | ||
9248 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9249 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9250 | char *page) | |
5c45bf27 SS |
9251 | { |
9252 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9253 | } | |
f718cd4a | 9254 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9255 | const char *buf, size_t count) |
5c45bf27 SS |
9256 | { |
9257 | return sched_power_savings_store(buf, count, 1); | |
9258 | } | |
f718cd4a AK |
9259 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9260 | sched_smt_power_savings_show, | |
6707de00 AB |
9261 | sched_smt_power_savings_store); |
9262 | #endif | |
9263 | ||
39aac648 | 9264 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9265 | { |
9266 | int err = 0; | |
9267 | ||
9268 | #ifdef CONFIG_SCHED_SMT | |
9269 | if (smt_capable()) | |
9270 | err = sysfs_create_file(&cls->kset.kobj, | |
9271 | &attr_sched_smt_power_savings.attr); | |
9272 | #endif | |
9273 | #ifdef CONFIG_SCHED_MC | |
9274 | if (!err && mc_capable()) | |
9275 | err = sysfs_create_file(&cls->kset.kobj, | |
9276 | &attr_sched_mc_power_savings.attr); | |
9277 | #endif | |
9278 | return err; | |
9279 | } | |
6d6bc0ad | 9280 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9281 | |
e761b772 | 9282 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9283 | /* |
e761b772 MK |
9284 | * Add online and remove offline CPUs from the scheduler domains. |
9285 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9286 | */ |
9287 | static int update_sched_domains(struct notifier_block *nfb, | |
9288 | unsigned long action, void *hcpu) | |
e761b772 MK |
9289 | { |
9290 | switch (action) { | |
9291 | case CPU_ONLINE: | |
9292 | case CPU_ONLINE_FROZEN: | |
9293 | case CPU_DEAD: | |
9294 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9295 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9296 | return NOTIFY_OK; |
9297 | ||
9298 | default: | |
9299 | return NOTIFY_DONE; | |
9300 | } | |
9301 | } | |
9302 | #endif | |
9303 | ||
9304 | static int update_runtime(struct notifier_block *nfb, | |
9305 | unsigned long action, void *hcpu) | |
1da177e4 | 9306 | { |
7def2be1 PZ |
9307 | int cpu = (int)(long)hcpu; |
9308 | ||
1da177e4 | 9309 | switch (action) { |
1da177e4 | 9310 | case CPU_DOWN_PREPARE: |
8bb78442 | 9311 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9312 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9313 | return NOTIFY_OK; |
9314 | ||
1da177e4 | 9315 | case CPU_DOWN_FAILED: |
8bb78442 | 9316 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9317 | case CPU_ONLINE: |
8bb78442 | 9318 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9319 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9320 | return NOTIFY_OK; |
9321 | ||
1da177e4 LT |
9322 | default: |
9323 | return NOTIFY_DONE; | |
9324 | } | |
1da177e4 | 9325 | } |
1da177e4 LT |
9326 | |
9327 | void __init sched_init_smp(void) | |
9328 | { | |
dcc30a35 RR |
9329 | cpumask_var_t non_isolated_cpus; |
9330 | ||
9331 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 9332 | |
434d53b0 MT |
9333 | #if defined(CONFIG_NUMA) |
9334 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9335 | GFP_KERNEL); | |
9336 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9337 | #endif | |
95402b38 | 9338 | get_online_cpus(); |
712555ee | 9339 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9340 | arch_init_sched_domains(cpu_online_mask); |
9341 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9342 | if (cpumask_empty(non_isolated_cpus)) | |
9343 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9344 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9345 | put_online_cpus(); |
e761b772 MK |
9346 | |
9347 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9348 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9349 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9350 | #endif |
9351 | ||
9352 | /* RT runtime code needs to handle some hotplug events */ | |
9353 | hotcpu_notifier(update_runtime, 0); | |
9354 | ||
b328ca18 | 9355 | init_hrtick(); |
5c1e1767 NP |
9356 | |
9357 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9358 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9359 | BUG(); |
19978ca6 | 9360 | sched_init_granularity(); |
dcc30a35 | 9361 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9362 | |
9363 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9364 | init_sched_rt_class(); |
1da177e4 LT |
9365 | } |
9366 | #else | |
9367 | void __init sched_init_smp(void) | |
9368 | { | |
19978ca6 | 9369 | sched_init_granularity(); |
1da177e4 LT |
9370 | } |
9371 | #endif /* CONFIG_SMP */ | |
9372 | ||
cd1bb94b AB |
9373 | const_debug unsigned int sysctl_timer_migration = 1; |
9374 | ||
1da177e4 LT |
9375 | int in_sched_functions(unsigned long addr) |
9376 | { | |
1da177e4 LT |
9377 | return in_lock_functions(addr) || |
9378 | (addr >= (unsigned long)__sched_text_start | |
9379 | && addr < (unsigned long)__sched_text_end); | |
9380 | } | |
9381 | ||
a9957449 | 9382 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9383 | { |
9384 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9385 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9386 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9387 | cfs_rq->rq = rq; | |
9388 | #endif | |
67e9fb2a | 9389 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9390 | } |
9391 | ||
fa85ae24 PZ |
9392 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9393 | { | |
9394 | struct rt_prio_array *array; | |
9395 | int i; | |
9396 | ||
9397 | array = &rt_rq->active; | |
9398 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9399 | INIT_LIST_HEAD(array->queue + i); | |
9400 | __clear_bit(i, array->bitmap); | |
9401 | } | |
9402 | /* delimiter for bitsearch: */ | |
9403 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9404 | ||
052f1dc7 | 9405 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9406 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9407 | #ifdef CONFIG_SMP |
e864c499 | 9408 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9409 | #endif |
48d5e258 | 9410 | #endif |
fa85ae24 PZ |
9411 | #ifdef CONFIG_SMP |
9412 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9413 | rt_rq->overloaded = 0; |
c20b08e3 | 9414 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9415 | #endif |
9416 | ||
9417 | rt_rq->rt_time = 0; | |
9418 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9419 | rt_rq->rt_runtime = 0; |
9420 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9421 | |
052f1dc7 | 9422 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9423 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9424 | rt_rq->rq = rq; |
9425 | #endif | |
fa85ae24 PZ |
9426 | } |
9427 | ||
6f505b16 | 9428 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9429 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9430 | struct sched_entity *se, int cpu, int add, | |
9431 | struct sched_entity *parent) | |
6f505b16 | 9432 | { |
ec7dc8ac | 9433 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9434 | tg->cfs_rq[cpu] = cfs_rq; |
9435 | init_cfs_rq(cfs_rq, rq); | |
9436 | cfs_rq->tg = tg; | |
9437 | if (add) | |
9438 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9439 | ||
9440 | tg->se[cpu] = se; | |
354d60c2 DG |
9441 | /* se could be NULL for init_task_group */ |
9442 | if (!se) | |
9443 | return; | |
9444 | ||
ec7dc8ac DG |
9445 | if (!parent) |
9446 | se->cfs_rq = &rq->cfs; | |
9447 | else | |
9448 | se->cfs_rq = parent->my_q; | |
9449 | ||
6f505b16 PZ |
9450 | se->my_q = cfs_rq; |
9451 | se->load.weight = tg->shares; | |
e05510d0 | 9452 | se->load.inv_weight = 0; |
ec7dc8ac | 9453 | se->parent = parent; |
6f505b16 | 9454 | } |
052f1dc7 | 9455 | #endif |
6f505b16 | 9456 | |
052f1dc7 | 9457 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9458 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9459 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9460 | struct sched_rt_entity *parent) | |
6f505b16 | 9461 | { |
ec7dc8ac DG |
9462 | struct rq *rq = cpu_rq(cpu); |
9463 | ||
6f505b16 PZ |
9464 | tg->rt_rq[cpu] = rt_rq; |
9465 | init_rt_rq(rt_rq, rq); | |
9466 | rt_rq->tg = tg; | |
9467 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9468 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9469 | if (add) |
9470 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9471 | ||
9472 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9473 | if (!rt_se) |
9474 | return; | |
9475 | ||
ec7dc8ac DG |
9476 | if (!parent) |
9477 | rt_se->rt_rq = &rq->rt; | |
9478 | else | |
9479 | rt_se->rt_rq = parent->my_q; | |
9480 | ||
6f505b16 | 9481 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9482 | rt_se->parent = parent; |
6f505b16 PZ |
9483 | INIT_LIST_HEAD(&rt_se->run_list); |
9484 | } | |
9485 | #endif | |
9486 | ||
1da177e4 LT |
9487 | void __init sched_init(void) |
9488 | { | |
dd41f596 | 9489 | int i, j; |
434d53b0 MT |
9490 | unsigned long alloc_size = 0, ptr; |
9491 | ||
9492 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9493 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9494 | #endif | |
9495 | #ifdef CONFIG_RT_GROUP_SCHED | |
9496 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9497 | #endif |
9498 | #ifdef CONFIG_USER_SCHED | |
9499 | alloc_size *= 2; | |
df7c8e84 RR |
9500 | #endif |
9501 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9502 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9503 | #endif |
9504 | /* | |
9505 | * As sched_init() is called before page_alloc is setup, | |
9506 | * we use alloc_bootmem(). | |
9507 | */ | |
9508 | if (alloc_size) { | |
36b7b6d4 | 9509 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9510 | |
9511 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9512 | init_task_group.se = (struct sched_entity **)ptr; | |
9513 | ptr += nr_cpu_ids * sizeof(void **); | |
9514 | ||
9515 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9516 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9517 | |
9518 | #ifdef CONFIG_USER_SCHED | |
9519 | root_task_group.se = (struct sched_entity **)ptr; | |
9520 | ptr += nr_cpu_ids * sizeof(void **); | |
9521 | ||
9522 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9523 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9524 | #endif /* CONFIG_USER_SCHED */ |
9525 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9526 | #ifdef CONFIG_RT_GROUP_SCHED |
9527 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9528 | ptr += nr_cpu_ids * sizeof(void **); | |
9529 | ||
9530 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9531 | ptr += nr_cpu_ids * sizeof(void **); |
9532 | ||
9533 | #ifdef CONFIG_USER_SCHED | |
9534 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9535 | ptr += nr_cpu_ids * sizeof(void **); | |
9536 | ||
9537 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9538 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9539 | #endif /* CONFIG_USER_SCHED */ |
9540 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9541 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9542 | for_each_possible_cpu(i) { | |
9543 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9544 | ptr += cpumask_size(); | |
9545 | } | |
9546 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9547 | } |
dd41f596 | 9548 | |
57d885fe GH |
9549 | #ifdef CONFIG_SMP |
9550 | init_defrootdomain(); | |
9551 | #endif | |
9552 | ||
d0b27fa7 PZ |
9553 | init_rt_bandwidth(&def_rt_bandwidth, |
9554 | global_rt_period(), global_rt_runtime()); | |
9555 | ||
9556 | #ifdef CONFIG_RT_GROUP_SCHED | |
9557 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9558 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9559 | #ifdef CONFIG_USER_SCHED |
9560 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9561 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9562 | #endif /* CONFIG_USER_SCHED */ |
9563 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9564 | |
052f1dc7 | 9565 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9566 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9567 | INIT_LIST_HEAD(&init_task_group.children); |
9568 | ||
9569 | #ifdef CONFIG_USER_SCHED | |
9570 | INIT_LIST_HEAD(&root_task_group.children); | |
9571 | init_task_group.parent = &root_task_group; | |
9572 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9573 | #endif /* CONFIG_USER_SCHED */ |
9574 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9575 | |
0a945022 | 9576 | for_each_possible_cpu(i) { |
70b97a7f | 9577 | struct rq *rq; |
1da177e4 LT |
9578 | |
9579 | rq = cpu_rq(i); | |
9580 | spin_lock_init(&rq->lock); | |
7897986b | 9581 | rq->nr_running = 0; |
dce48a84 TG |
9582 | rq->calc_load_active = 0; |
9583 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9584 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9585 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9586 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9587 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9588 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9589 | #ifdef CONFIG_CGROUP_SCHED |
9590 | /* | |
9591 | * How much cpu bandwidth does init_task_group get? | |
9592 | * | |
9593 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9594 | * gets 100% of the cpu resources in the system. This overall | |
9595 | * system cpu resource is divided among the tasks of | |
9596 | * init_task_group and its child task-groups in a fair manner, | |
9597 | * based on each entity's (task or task-group's) weight | |
9598 | * (se->load.weight). | |
9599 | * | |
9600 | * In other words, if init_task_group has 10 tasks of weight | |
9601 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9602 | * then A0's share of the cpu resource is: | |
9603 | * | |
0d905bca | 9604 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9605 | * |
9606 | * We achieve this by letting init_task_group's tasks sit | |
9607 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9608 | */ | |
ec7dc8ac | 9609 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9610 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9611 | root_task_group.shares = NICE_0_LOAD; |
9612 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9613 | /* |
9614 | * In case of task-groups formed thr' the user id of tasks, | |
9615 | * init_task_group represents tasks belonging to root user. | |
9616 | * Hence it forms a sibling of all subsequent groups formed. | |
9617 | * In this case, init_task_group gets only a fraction of overall | |
9618 | * system cpu resource, based on the weight assigned to root | |
9619 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9620 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9621 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9622 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9623 | */ | |
ec7dc8ac | 9624 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9625 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9626 | &per_cpu(init_sched_entity, i), i, 1, |
9627 | root_task_group.se[i]); | |
6f505b16 | 9628 | |
052f1dc7 | 9629 | #endif |
354d60c2 DG |
9630 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9631 | ||
9632 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9633 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9634 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9635 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9636 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9637 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9638 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9639 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9640 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9641 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9642 | root_task_group.rt_se[i]); | |
354d60c2 | 9643 | #endif |
dd41f596 | 9644 | #endif |
1da177e4 | 9645 | |
dd41f596 IM |
9646 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9647 | rq->cpu_load[j] = 0; | |
1da177e4 | 9648 | #ifdef CONFIG_SMP |
41c7ce9a | 9649 | rq->sd = NULL; |
57d885fe | 9650 | rq->rd = NULL; |
3f029d3c | 9651 | rq->post_schedule = 0; |
1da177e4 | 9652 | rq->active_balance = 0; |
dd41f596 | 9653 | rq->next_balance = jiffies; |
1da177e4 | 9654 | rq->push_cpu = 0; |
0a2966b4 | 9655 | rq->cpu = i; |
1f11eb6a | 9656 | rq->online = 0; |
1da177e4 LT |
9657 | rq->migration_thread = NULL; |
9658 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9659 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9660 | #endif |
8f4d37ec | 9661 | init_rq_hrtick(rq); |
1da177e4 | 9662 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9663 | } |
9664 | ||
2dd73a4f | 9665 | set_load_weight(&init_task); |
b50f60ce | 9666 | |
e107be36 AK |
9667 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9668 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9669 | #endif | |
9670 | ||
c9819f45 | 9671 | #ifdef CONFIG_SMP |
962cf36c | 9672 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9673 | #endif |
9674 | ||
b50f60ce HC |
9675 | #ifdef CONFIG_RT_MUTEXES |
9676 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9677 | #endif | |
9678 | ||
1da177e4 LT |
9679 | /* |
9680 | * The boot idle thread does lazy MMU switching as well: | |
9681 | */ | |
9682 | atomic_inc(&init_mm.mm_count); | |
9683 | enter_lazy_tlb(&init_mm, current); | |
9684 | ||
9685 | /* | |
9686 | * Make us the idle thread. Technically, schedule() should not be | |
9687 | * called from this thread, however somewhere below it might be, | |
9688 | * but because we are the idle thread, we just pick up running again | |
9689 | * when this runqueue becomes "idle". | |
9690 | */ | |
9691 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9692 | |
9693 | calc_load_update = jiffies + LOAD_FREQ; | |
9694 | ||
dd41f596 IM |
9695 | /* |
9696 | * During early bootup we pretend to be a normal task: | |
9697 | */ | |
9698 | current->sched_class = &fair_sched_class; | |
6892b75e | 9699 | |
6a7b3dc3 | 9700 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9701 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9702 | #ifdef CONFIG_SMP |
7d1e6a9b | 9703 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9704 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9705 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9706 | #endif |
4bdddf8f | 9707 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9708 | #endif /* SMP */ |
6a7b3dc3 | 9709 | |
0d905bca IM |
9710 | perf_counter_init(); |
9711 | ||
6892b75e | 9712 | scheduler_running = 1; |
1da177e4 LT |
9713 | } |
9714 | ||
9715 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9716 | static inline int preempt_count_equals(int preempt_offset) |
9717 | { | |
9718 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9719 | ||
9720 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9721 | } | |
9722 | ||
9723 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9724 | { |
48f24c4d | 9725 | #ifdef in_atomic |
1da177e4 LT |
9726 | static unsigned long prev_jiffy; /* ratelimiting */ |
9727 | ||
e4aafea2 FW |
9728 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9729 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9730 | return; |
9731 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9732 | return; | |
9733 | prev_jiffy = jiffies; | |
9734 | ||
9735 | printk(KERN_ERR | |
9736 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9737 | file, line); | |
9738 | printk(KERN_ERR | |
9739 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9740 | in_atomic(), irqs_disabled(), | |
9741 | current->pid, current->comm); | |
9742 | ||
9743 | debug_show_held_locks(current); | |
9744 | if (irqs_disabled()) | |
9745 | print_irqtrace_events(current); | |
9746 | dump_stack(); | |
1da177e4 LT |
9747 | #endif |
9748 | } | |
9749 | EXPORT_SYMBOL(__might_sleep); | |
9750 | #endif | |
9751 | ||
9752 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9753 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9754 | { | |
9755 | int on_rq; | |
3e51f33f | 9756 | |
3a5e4dc1 AK |
9757 | update_rq_clock(rq); |
9758 | on_rq = p->se.on_rq; | |
9759 | if (on_rq) | |
9760 | deactivate_task(rq, p, 0); | |
9761 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9762 | if (on_rq) { | |
9763 | activate_task(rq, p, 0); | |
9764 | resched_task(rq->curr); | |
9765 | } | |
9766 | } | |
9767 | ||
1da177e4 LT |
9768 | void normalize_rt_tasks(void) |
9769 | { | |
a0f98a1c | 9770 | struct task_struct *g, *p; |
1da177e4 | 9771 | unsigned long flags; |
70b97a7f | 9772 | struct rq *rq; |
1da177e4 | 9773 | |
4cf5d77a | 9774 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9775 | do_each_thread(g, p) { |
178be793 IM |
9776 | /* |
9777 | * Only normalize user tasks: | |
9778 | */ | |
9779 | if (!p->mm) | |
9780 | continue; | |
9781 | ||
6cfb0d5d | 9782 | p->se.exec_start = 0; |
6cfb0d5d | 9783 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9784 | p->se.wait_start = 0; |
dd41f596 | 9785 | p->se.sleep_start = 0; |
dd41f596 | 9786 | p->se.block_start = 0; |
6cfb0d5d | 9787 | #endif |
dd41f596 IM |
9788 | |
9789 | if (!rt_task(p)) { | |
9790 | /* | |
9791 | * Renice negative nice level userspace | |
9792 | * tasks back to 0: | |
9793 | */ | |
9794 | if (TASK_NICE(p) < 0 && p->mm) | |
9795 | set_user_nice(p, 0); | |
1da177e4 | 9796 | continue; |
dd41f596 | 9797 | } |
1da177e4 | 9798 | |
4cf5d77a | 9799 | spin_lock(&p->pi_lock); |
b29739f9 | 9800 | rq = __task_rq_lock(p); |
1da177e4 | 9801 | |
178be793 | 9802 | normalize_task(rq, p); |
3a5e4dc1 | 9803 | |
b29739f9 | 9804 | __task_rq_unlock(rq); |
4cf5d77a | 9805 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9806 | } while_each_thread(g, p); |
9807 | ||
4cf5d77a | 9808 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9809 | } |
9810 | ||
9811 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9812 | |
9813 | #ifdef CONFIG_IA64 | |
9814 | /* | |
9815 | * These functions are only useful for the IA64 MCA handling. | |
9816 | * | |
9817 | * They can only be called when the whole system has been | |
9818 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9819 | * activity can take place. Using them for anything else would | |
9820 | * be a serious bug, and as a result, they aren't even visible | |
9821 | * under any other configuration. | |
9822 | */ | |
9823 | ||
9824 | /** | |
9825 | * curr_task - return the current task for a given cpu. | |
9826 | * @cpu: the processor in question. | |
9827 | * | |
9828 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9829 | */ | |
36c8b586 | 9830 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9831 | { |
9832 | return cpu_curr(cpu); | |
9833 | } | |
9834 | ||
9835 | /** | |
9836 | * set_curr_task - set the current task for a given cpu. | |
9837 | * @cpu: the processor in question. | |
9838 | * @p: the task pointer to set. | |
9839 | * | |
9840 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9841 | * are serviced on a separate stack. It allows the architecture to switch the |
9842 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9843 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9844 | * and caller must save the original value of the current task (see | |
9845 | * curr_task() above) and restore that value before reenabling interrupts and | |
9846 | * re-starting the system. | |
9847 | * | |
9848 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9849 | */ | |
36c8b586 | 9850 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9851 | { |
9852 | cpu_curr(cpu) = p; | |
9853 | } | |
9854 | ||
9855 | #endif | |
29f59db3 | 9856 | |
bccbe08a PZ |
9857 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9858 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9859 | { |
9860 | int i; | |
9861 | ||
9862 | for_each_possible_cpu(i) { | |
9863 | if (tg->cfs_rq) | |
9864 | kfree(tg->cfs_rq[i]); | |
9865 | if (tg->se) | |
9866 | kfree(tg->se[i]); | |
6f505b16 PZ |
9867 | } |
9868 | ||
9869 | kfree(tg->cfs_rq); | |
9870 | kfree(tg->se); | |
6f505b16 PZ |
9871 | } |
9872 | ||
ec7dc8ac DG |
9873 | static |
9874 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9875 | { |
29f59db3 | 9876 | struct cfs_rq *cfs_rq; |
eab17229 | 9877 | struct sched_entity *se; |
9b5b7751 | 9878 | struct rq *rq; |
29f59db3 SV |
9879 | int i; |
9880 | ||
434d53b0 | 9881 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9882 | if (!tg->cfs_rq) |
9883 | goto err; | |
434d53b0 | 9884 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9885 | if (!tg->se) |
9886 | goto err; | |
052f1dc7 PZ |
9887 | |
9888 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9889 | |
9890 | for_each_possible_cpu(i) { | |
9b5b7751 | 9891 | rq = cpu_rq(i); |
29f59db3 | 9892 | |
eab17229 LZ |
9893 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9894 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9895 | if (!cfs_rq) |
9896 | goto err; | |
9897 | ||
eab17229 LZ |
9898 | se = kzalloc_node(sizeof(struct sched_entity), |
9899 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9900 | if (!se) |
9901 | goto err; | |
9902 | ||
eab17229 | 9903 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9904 | } |
9905 | ||
9906 | return 1; | |
9907 | ||
9908 | err: | |
9909 | return 0; | |
9910 | } | |
9911 | ||
9912 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9913 | { | |
9914 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9915 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9916 | } | |
9917 | ||
9918 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9919 | { | |
9920 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9921 | } | |
6d6bc0ad | 9922 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9923 | static inline void free_fair_sched_group(struct task_group *tg) |
9924 | { | |
9925 | } | |
9926 | ||
ec7dc8ac DG |
9927 | static inline |
9928 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9929 | { |
9930 | return 1; | |
9931 | } | |
9932 | ||
9933 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9934 | { | |
9935 | } | |
9936 | ||
9937 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9938 | { | |
9939 | } | |
6d6bc0ad | 9940 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9941 | |
9942 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9943 | static void free_rt_sched_group(struct task_group *tg) |
9944 | { | |
9945 | int i; | |
9946 | ||
d0b27fa7 PZ |
9947 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9948 | ||
bccbe08a PZ |
9949 | for_each_possible_cpu(i) { |
9950 | if (tg->rt_rq) | |
9951 | kfree(tg->rt_rq[i]); | |
9952 | if (tg->rt_se) | |
9953 | kfree(tg->rt_se[i]); | |
9954 | } | |
9955 | ||
9956 | kfree(tg->rt_rq); | |
9957 | kfree(tg->rt_se); | |
9958 | } | |
9959 | ||
ec7dc8ac DG |
9960 | static |
9961 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9962 | { |
9963 | struct rt_rq *rt_rq; | |
eab17229 | 9964 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9965 | struct rq *rq; |
9966 | int i; | |
9967 | ||
434d53b0 | 9968 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9969 | if (!tg->rt_rq) |
9970 | goto err; | |
434d53b0 | 9971 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9972 | if (!tg->rt_se) |
9973 | goto err; | |
9974 | ||
d0b27fa7 PZ |
9975 | init_rt_bandwidth(&tg->rt_bandwidth, |
9976 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9977 | |
9978 | for_each_possible_cpu(i) { | |
9979 | rq = cpu_rq(i); | |
9980 | ||
eab17229 LZ |
9981 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9982 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9983 | if (!rt_rq) |
9984 | goto err; | |
29f59db3 | 9985 | |
eab17229 LZ |
9986 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9987 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9988 | if (!rt_se) |
9989 | goto err; | |
29f59db3 | 9990 | |
eab17229 | 9991 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9992 | } |
9993 | ||
bccbe08a PZ |
9994 | return 1; |
9995 | ||
9996 | err: | |
9997 | return 0; | |
9998 | } | |
9999 | ||
10000 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
10001 | { | |
10002 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
10003 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
10004 | } | |
10005 | ||
10006 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
10007 | { | |
10008 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
10009 | } | |
6d6bc0ad | 10010 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
10011 | static inline void free_rt_sched_group(struct task_group *tg) |
10012 | { | |
10013 | } | |
10014 | ||
ec7dc8ac DG |
10015 | static inline |
10016 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
10017 | { |
10018 | return 1; | |
10019 | } | |
10020 | ||
10021 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
10022 | { | |
10023 | } | |
10024 | ||
10025 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
10026 | { | |
10027 | } | |
6d6bc0ad | 10028 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 10029 | |
d0b27fa7 | 10030 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
10031 | static void free_sched_group(struct task_group *tg) |
10032 | { | |
10033 | free_fair_sched_group(tg); | |
10034 | free_rt_sched_group(tg); | |
10035 | kfree(tg); | |
10036 | } | |
10037 | ||
10038 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 10039 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
10040 | { |
10041 | struct task_group *tg; | |
10042 | unsigned long flags; | |
10043 | int i; | |
10044 | ||
10045 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
10046 | if (!tg) | |
10047 | return ERR_PTR(-ENOMEM); | |
10048 | ||
ec7dc8ac | 10049 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
10050 | goto err; |
10051 | ||
ec7dc8ac | 10052 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
10053 | goto err; |
10054 | ||
8ed36996 | 10055 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10056 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10057 | register_fair_sched_group(tg, i); |
10058 | register_rt_sched_group(tg, i); | |
9b5b7751 | 10059 | } |
6f505b16 | 10060 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
10061 | |
10062 | WARN_ON(!parent); /* root should already exist */ | |
10063 | ||
10064 | tg->parent = parent; | |
f473aa5e | 10065 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10066 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10067 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 10068 | |
9b5b7751 | 10069 | return tg; |
29f59db3 SV |
10070 | |
10071 | err: | |
6f505b16 | 10072 | free_sched_group(tg); |
29f59db3 SV |
10073 | return ERR_PTR(-ENOMEM); |
10074 | } | |
10075 | ||
9b5b7751 | 10076 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 10077 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10078 | { |
29f59db3 | 10079 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 10080 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10081 | } |
10082 | ||
9b5b7751 | 10083 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 10084 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 10085 | { |
8ed36996 | 10086 | unsigned long flags; |
9b5b7751 | 10087 | int i; |
29f59db3 | 10088 | |
8ed36996 | 10089 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10090 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10091 | unregister_fair_sched_group(tg, i); |
10092 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 10093 | } |
6f505b16 | 10094 | list_del_rcu(&tg->list); |
f473aa5e | 10095 | list_del_rcu(&tg->siblings); |
8ed36996 | 10096 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 10097 | |
9b5b7751 | 10098 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 10099 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
10100 | } |
10101 | ||
9b5b7751 | 10102 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
10103 | * The caller of this function should have put the task in its new group |
10104 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
10105 | * reflect its new group. | |
9b5b7751 SV |
10106 | */ |
10107 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
10108 | { |
10109 | int on_rq, running; | |
10110 | unsigned long flags; | |
10111 | struct rq *rq; | |
10112 | ||
10113 | rq = task_rq_lock(tsk, &flags); | |
10114 | ||
29f59db3 SV |
10115 | update_rq_clock(rq); |
10116 | ||
051a1d1a | 10117 | running = task_current(rq, tsk); |
29f59db3 SV |
10118 | on_rq = tsk->se.on_rq; |
10119 | ||
0e1f3483 | 10120 | if (on_rq) |
29f59db3 | 10121 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
10122 | if (unlikely(running)) |
10123 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 10124 | |
6f505b16 | 10125 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 10126 | |
810b3817 PZ |
10127 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10128 | if (tsk->sched_class->moved_group) | |
10129 | tsk->sched_class->moved_group(tsk); | |
10130 | #endif | |
10131 | ||
0e1f3483 HS |
10132 | if (unlikely(running)) |
10133 | tsk->sched_class->set_curr_task(rq); | |
10134 | if (on_rq) | |
7074badb | 10135 | enqueue_task(rq, tsk, 0); |
29f59db3 | 10136 | |
29f59db3 SV |
10137 | task_rq_unlock(rq, &flags); |
10138 | } | |
6d6bc0ad | 10139 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 10140 | |
052f1dc7 | 10141 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 10142 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
10143 | { |
10144 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
10145 | int on_rq; |
10146 | ||
29f59db3 | 10147 | on_rq = se->on_rq; |
62fb1851 | 10148 | if (on_rq) |
29f59db3 SV |
10149 | dequeue_entity(cfs_rq, se, 0); |
10150 | ||
10151 | se->load.weight = shares; | |
e05510d0 | 10152 | se->load.inv_weight = 0; |
29f59db3 | 10153 | |
62fb1851 | 10154 | if (on_rq) |
29f59db3 | 10155 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10156 | } |
62fb1851 | 10157 | |
c09595f6 PZ |
10158 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10159 | { | |
10160 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10161 | struct rq *rq = cfs_rq->rq; | |
10162 | unsigned long flags; | |
10163 | ||
10164 | spin_lock_irqsave(&rq->lock, flags); | |
10165 | __set_se_shares(se, shares); | |
10166 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
10167 | } |
10168 | ||
8ed36996 PZ |
10169 | static DEFINE_MUTEX(shares_mutex); |
10170 | ||
4cf86d77 | 10171 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10172 | { |
10173 | int i; | |
8ed36996 | 10174 | unsigned long flags; |
c61935fd | 10175 | |
ec7dc8ac DG |
10176 | /* |
10177 | * We can't change the weight of the root cgroup. | |
10178 | */ | |
10179 | if (!tg->se[0]) | |
10180 | return -EINVAL; | |
10181 | ||
18d95a28 PZ |
10182 | if (shares < MIN_SHARES) |
10183 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10184 | else if (shares > MAX_SHARES) |
10185 | shares = MAX_SHARES; | |
62fb1851 | 10186 | |
8ed36996 | 10187 | mutex_lock(&shares_mutex); |
9b5b7751 | 10188 | if (tg->shares == shares) |
5cb350ba | 10189 | goto done; |
29f59db3 | 10190 | |
8ed36996 | 10191 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10192 | for_each_possible_cpu(i) |
10193 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10194 | list_del_rcu(&tg->siblings); |
8ed36996 | 10195 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10196 | |
10197 | /* wait for any ongoing reference to this group to finish */ | |
10198 | synchronize_sched(); | |
10199 | ||
10200 | /* | |
10201 | * Now we are free to modify the group's share on each cpu | |
10202 | * w/o tripping rebalance_share or load_balance_fair. | |
10203 | */ | |
9b5b7751 | 10204 | tg->shares = shares; |
c09595f6 PZ |
10205 | for_each_possible_cpu(i) { |
10206 | /* | |
10207 | * force a rebalance | |
10208 | */ | |
10209 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10210 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10211 | } |
29f59db3 | 10212 | |
6b2d7700 SV |
10213 | /* |
10214 | * Enable load balance activity on this group, by inserting it back on | |
10215 | * each cpu's rq->leaf_cfs_rq_list. | |
10216 | */ | |
8ed36996 | 10217 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10218 | for_each_possible_cpu(i) |
10219 | register_fair_sched_group(tg, i); | |
f473aa5e | 10220 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10221 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10222 | done: |
8ed36996 | 10223 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10224 | return 0; |
29f59db3 SV |
10225 | } |
10226 | ||
5cb350ba DG |
10227 | unsigned long sched_group_shares(struct task_group *tg) |
10228 | { | |
10229 | return tg->shares; | |
10230 | } | |
052f1dc7 | 10231 | #endif |
5cb350ba | 10232 | |
052f1dc7 | 10233 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10234 | /* |
9f0c1e56 | 10235 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10236 | */ |
9f0c1e56 PZ |
10237 | static DEFINE_MUTEX(rt_constraints_mutex); |
10238 | ||
10239 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10240 | { | |
10241 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10242 | return 1ULL << 20; |
9f0c1e56 | 10243 | |
9a7e0b18 | 10244 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10245 | } |
10246 | ||
9a7e0b18 PZ |
10247 | /* Must be called with tasklist_lock held */ |
10248 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10249 | { |
9a7e0b18 | 10250 | struct task_struct *g, *p; |
b40b2e8e | 10251 | |
9a7e0b18 PZ |
10252 | do_each_thread(g, p) { |
10253 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10254 | return 1; | |
10255 | } while_each_thread(g, p); | |
b40b2e8e | 10256 | |
9a7e0b18 PZ |
10257 | return 0; |
10258 | } | |
b40b2e8e | 10259 | |
9a7e0b18 PZ |
10260 | struct rt_schedulable_data { |
10261 | struct task_group *tg; | |
10262 | u64 rt_period; | |
10263 | u64 rt_runtime; | |
10264 | }; | |
b40b2e8e | 10265 | |
9a7e0b18 PZ |
10266 | static int tg_schedulable(struct task_group *tg, void *data) |
10267 | { | |
10268 | struct rt_schedulable_data *d = data; | |
10269 | struct task_group *child; | |
10270 | unsigned long total, sum = 0; | |
10271 | u64 period, runtime; | |
b40b2e8e | 10272 | |
9a7e0b18 PZ |
10273 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10274 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10275 | |
9a7e0b18 PZ |
10276 | if (tg == d->tg) { |
10277 | period = d->rt_period; | |
10278 | runtime = d->rt_runtime; | |
b40b2e8e | 10279 | } |
b40b2e8e | 10280 | |
98a4826b PZ |
10281 | #ifdef CONFIG_USER_SCHED |
10282 | if (tg == &root_task_group) { | |
10283 | period = global_rt_period(); | |
10284 | runtime = global_rt_runtime(); | |
10285 | } | |
10286 | #endif | |
10287 | ||
4653f803 PZ |
10288 | /* |
10289 | * Cannot have more runtime than the period. | |
10290 | */ | |
10291 | if (runtime > period && runtime != RUNTIME_INF) | |
10292 | return -EINVAL; | |
6f505b16 | 10293 | |
4653f803 PZ |
10294 | /* |
10295 | * Ensure we don't starve existing RT tasks. | |
10296 | */ | |
9a7e0b18 PZ |
10297 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10298 | return -EBUSY; | |
6f505b16 | 10299 | |
9a7e0b18 | 10300 | total = to_ratio(period, runtime); |
6f505b16 | 10301 | |
4653f803 PZ |
10302 | /* |
10303 | * Nobody can have more than the global setting allows. | |
10304 | */ | |
10305 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10306 | return -EINVAL; | |
6f505b16 | 10307 | |
4653f803 PZ |
10308 | /* |
10309 | * The sum of our children's runtime should not exceed our own. | |
10310 | */ | |
9a7e0b18 PZ |
10311 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10312 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10313 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10314 | |
9a7e0b18 PZ |
10315 | if (child == d->tg) { |
10316 | period = d->rt_period; | |
10317 | runtime = d->rt_runtime; | |
10318 | } | |
6f505b16 | 10319 | |
9a7e0b18 | 10320 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10321 | } |
6f505b16 | 10322 | |
9a7e0b18 PZ |
10323 | if (sum > total) |
10324 | return -EINVAL; | |
10325 | ||
10326 | return 0; | |
6f505b16 PZ |
10327 | } |
10328 | ||
9a7e0b18 | 10329 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10330 | { |
9a7e0b18 PZ |
10331 | struct rt_schedulable_data data = { |
10332 | .tg = tg, | |
10333 | .rt_period = period, | |
10334 | .rt_runtime = runtime, | |
10335 | }; | |
10336 | ||
10337 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10338 | } |
10339 | ||
d0b27fa7 PZ |
10340 | static int tg_set_bandwidth(struct task_group *tg, |
10341 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10342 | { |
ac086bc2 | 10343 | int i, err = 0; |
9f0c1e56 | 10344 | |
9f0c1e56 | 10345 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10346 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10347 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10348 | if (err) | |
9f0c1e56 | 10349 | goto unlock; |
ac086bc2 PZ |
10350 | |
10351 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10352 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10353 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10354 | |
10355 | for_each_possible_cpu(i) { | |
10356 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10357 | ||
10358 | spin_lock(&rt_rq->rt_runtime_lock); | |
10359 | rt_rq->rt_runtime = rt_runtime; | |
10360 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10361 | } | |
10362 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10363 | unlock: |
521f1a24 | 10364 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10365 | mutex_unlock(&rt_constraints_mutex); |
10366 | ||
10367 | return err; | |
6f505b16 PZ |
10368 | } |
10369 | ||
d0b27fa7 PZ |
10370 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10371 | { | |
10372 | u64 rt_runtime, rt_period; | |
10373 | ||
10374 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10375 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10376 | if (rt_runtime_us < 0) | |
10377 | rt_runtime = RUNTIME_INF; | |
10378 | ||
10379 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10380 | } | |
10381 | ||
9f0c1e56 PZ |
10382 | long sched_group_rt_runtime(struct task_group *tg) |
10383 | { | |
10384 | u64 rt_runtime_us; | |
10385 | ||
d0b27fa7 | 10386 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10387 | return -1; |
10388 | ||
d0b27fa7 | 10389 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10390 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10391 | return rt_runtime_us; | |
10392 | } | |
d0b27fa7 PZ |
10393 | |
10394 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10395 | { | |
10396 | u64 rt_runtime, rt_period; | |
10397 | ||
10398 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10399 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10400 | ||
619b0488 R |
10401 | if (rt_period == 0) |
10402 | return -EINVAL; | |
10403 | ||
d0b27fa7 PZ |
10404 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10405 | } | |
10406 | ||
10407 | long sched_group_rt_period(struct task_group *tg) | |
10408 | { | |
10409 | u64 rt_period_us; | |
10410 | ||
10411 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10412 | do_div(rt_period_us, NSEC_PER_USEC); | |
10413 | return rt_period_us; | |
10414 | } | |
10415 | ||
10416 | static int sched_rt_global_constraints(void) | |
10417 | { | |
4653f803 | 10418 | u64 runtime, period; |
d0b27fa7 PZ |
10419 | int ret = 0; |
10420 | ||
ec5d4989 HS |
10421 | if (sysctl_sched_rt_period <= 0) |
10422 | return -EINVAL; | |
10423 | ||
4653f803 PZ |
10424 | runtime = global_rt_runtime(); |
10425 | period = global_rt_period(); | |
10426 | ||
10427 | /* | |
10428 | * Sanity check on the sysctl variables. | |
10429 | */ | |
10430 | if (runtime > period && runtime != RUNTIME_INF) | |
10431 | return -EINVAL; | |
10b612f4 | 10432 | |
d0b27fa7 | 10433 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10434 | read_lock(&tasklist_lock); |
4653f803 | 10435 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10436 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10437 | mutex_unlock(&rt_constraints_mutex); |
10438 | ||
10439 | return ret; | |
10440 | } | |
54e99124 DG |
10441 | |
10442 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10443 | { | |
10444 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10445 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10446 | return 0; | |
10447 | ||
10448 | return 1; | |
10449 | } | |
10450 | ||
6d6bc0ad | 10451 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10452 | static int sched_rt_global_constraints(void) |
10453 | { | |
ac086bc2 PZ |
10454 | unsigned long flags; |
10455 | int i; | |
10456 | ||
ec5d4989 HS |
10457 | if (sysctl_sched_rt_period <= 0) |
10458 | return -EINVAL; | |
10459 | ||
60aa605d PZ |
10460 | /* |
10461 | * There's always some RT tasks in the root group | |
10462 | * -- migration, kstopmachine etc.. | |
10463 | */ | |
10464 | if (sysctl_sched_rt_runtime == 0) | |
10465 | return -EBUSY; | |
10466 | ||
ac086bc2 PZ |
10467 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10468 | for_each_possible_cpu(i) { | |
10469 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10470 | ||
10471 | spin_lock(&rt_rq->rt_runtime_lock); | |
10472 | rt_rq->rt_runtime = global_rt_runtime(); | |
10473 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10474 | } | |
10475 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10476 | ||
d0b27fa7 PZ |
10477 | return 0; |
10478 | } | |
6d6bc0ad | 10479 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10480 | |
10481 | int sched_rt_handler(struct ctl_table *table, int write, | |
10482 | struct file *filp, void __user *buffer, size_t *lenp, | |
10483 | loff_t *ppos) | |
10484 | { | |
10485 | int ret; | |
10486 | int old_period, old_runtime; | |
10487 | static DEFINE_MUTEX(mutex); | |
10488 | ||
10489 | mutex_lock(&mutex); | |
10490 | old_period = sysctl_sched_rt_period; | |
10491 | old_runtime = sysctl_sched_rt_runtime; | |
10492 | ||
10493 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10494 | ||
10495 | if (!ret && write) { | |
10496 | ret = sched_rt_global_constraints(); | |
10497 | if (ret) { | |
10498 | sysctl_sched_rt_period = old_period; | |
10499 | sysctl_sched_rt_runtime = old_runtime; | |
10500 | } else { | |
10501 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10502 | def_rt_bandwidth.rt_period = | |
10503 | ns_to_ktime(global_rt_period()); | |
10504 | } | |
10505 | } | |
10506 | mutex_unlock(&mutex); | |
10507 | ||
10508 | return ret; | |
10509 | } | |
68318b8e | 10510 | |
052f1dc7 | 10511 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10512 | |
10513 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10514 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10515 | { |
2b01dfe3 PM |
10516 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10517 | struct task_group, css); | |
68318b8e SV |
10518 | } |
10519 | ||
10520 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10521 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10522 | { |
ec7dc8ac | 10523 | struct task_group *tg, *parent; |
68318b8e | 10524 | |
2b01dfe3 | 10525 | if (!cgrp->parent) { |
68318b8e | 10526 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10527 | return &init_task_group.css; |
10528 | } | |
10529 | ||
ec7dc8ac DG |
10530 | parent = cgroup_tg(cgrp->parent); |
10531 | tg = sched_create_group(parent); | |
68318b8e SV |
10532 | if (IS_ERR(tg)) |
10533 | return ERR_PTR(-ENOMEM); | |
10534 | ||
68318b8e SV |
10535 | return &tg->css; |
10536 | } | |
10537 | ||
41a2d6cf IM |
10538 | static void |
10539 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10540 | { |
2b01dfe3 | 10541 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10542 | |
10543 | sched_destroy_group(tg); | |
10544 | } | |
10545 | ||
41a2d6cf IM |
10546 | static int |
10547 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10548 | struct task_struct *tsk) | |
68318b8e | 10549 | { |
b68aa230 | 10550 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10551 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10552 | return -EINVAL; |
10553 | #else | |
68318b8e SV |
10554 | /* We don't support RT-tasks being in separate groups */ |
10555 | if (tsk->sched_class != &fair_sched_class) | |
10556 | return -EINVAL; | |
b68aa230 | 10557 | #endif |
68318b8e SV |
10558 | |
10559 | return 0; | |
10560 | } | |
10561 | ||
10562 | static void | |
2b01dfe3 | 10563 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10564 | struct cgroup *old_cont, struct task_struct *tsk) |
10565 | { | |
10566 | sched_move_task(tsk); | |
10567 | } | |
10568 | ||
052f1dc7 | 10569 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10570 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10571 | u64 shareval) |
68318b8e | 10572 | { |
2b01dfe3 | 10573 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10574 | } |
10575 | ||
f4c753b7 | 10576 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10577 | { |
2b01dfe3 | 10578 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10579 | |
10580 | return (u64) tg->shares; | |
10581 | } | |
6d6bc0ad | 10582 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10583 | |
052f1dc7 | 10584 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10585 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10586 | s64 val) |
6f505b16 | 10587 | { |
06ecb27c | 10588 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10589 | } |
10590 | ||
06ecb27c | 10591 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10592 | { |
06ecb27c | 10593 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10594 | } |
d0b27fa7 PZ |
10595 | |
10596 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10597 | u64 rt_period_us) | |
10598 | { | |
10599 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10600 | } | |
10601 | ||
10602 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10603 | { | |
10604 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10605 | } | |
6d6bc0ad | 10606 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10607 | |
fe5c7cc2 | 10608 | static struct cftype cpu_files[] = { |
052f1dc7 | 10609 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10610 | { |
10611 | .name = "shares", | |
f4c753b7 PM |
10612 | .read_u64 = cpu_shares_read_u64, |
10613 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10614 | }, |
052f1dc7 PZ |
10615 | #endif |
10616 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10617 | { |
9f0c1e56 | 10618 | .name = "rt_runtime_us", |
06ecb27c PM |
10619 | .read_s64 = cpu_rt_runtime_read, |
10620 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10621 | }, |
d0b27fa7 PZ |
10622 | { |
10623 | .name = "rt_period_us", | |
f4c753b7 PM |
10624 | .read_u64 = cpu_rt_period_read_uint, |
10625 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10626 | }, |
052f1dc7 | 10627 | #endif |
68318b8e SV |
10628 | }; |
10629 | ||
10630 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10631 | { | |
fe5c7cc2 | 10632 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10633 | } |
10634 | ||
10635 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10636 | .name = "cpu", |
10637 | .create = cpu_cgroup_create, | |
10638 | .destroy = cpu_cgroup_destroy, | |
10639 | .can_attach = cpu_cgroup_can_attach, | |
10640 | .attach = cpu_cgroup_attach, | |
10641 | .populate = cpu_cgroup_populate, | |
10642 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10643 | .early_init = 1, |
10644 | }; | |
10645 | ||
052f1dc7 | 10646 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10647 | |
10648 | #ifdef CONFIG_CGROUP_CPUACCT | |
10649 | ||
10650 | /* | |
10651 | * CPU accounting code for task groups. | |
10652 | * | |
10653 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10654 | * (balbir@in.ibm.com). | |
10655 | */ | |
10656 | ||
934352f2 | 10657 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10658 | struct cpuacct { |
10659 | struct cgroup_subsys_state css; | |
10660 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10661 | u64 *cpuusage; | |
ef12fefa | 10662 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10663 | struct cpuacct *parent; |
d842de87 SV |
10664 | }; |
10665 | ||
10666 | struct cgroup_subsys cpuacct_subsys; | |
10667 | ||
10668 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10669 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10670 | { |
32cd756a | 10671 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10672 | struct cpuacct, css); |
10673 | } | |
10674 | ||
10675 | /* return cpu accounting group to which this task belongs */ | |
10676 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10677 | { | |
10678 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10679 | struct cpuacct, css); | |
10680 | } | |
10681 | ||
10682 | /* create a new cpu accounting group */ | |
10683 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10684 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10685 | { |
10686 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10687 | int i; |
d842de87 SV |
10688 | |
10689 | if (!ca) | |
ef12fefa | 10690 | goto out; |
d842de87 SV |
10691 | |
10692 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10693 | if (!ca->cpuusage) |
10694 | goto out_free_ca; | |
10695 | ||
10696 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10697 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10698 | goto out_free_counters; | |
d842de87 | 10699 | |
934352f2 BR |
10700 | if (cgrp->parent) |
10701 | ca->parent = cgroup_ca(cgrp->parent); | |
10702 | ||
d842de87 | 10703 | return &ca->css; |
ef12fefa BR |
10704 | |
10705 | out_free_counters: | |
10706 | while (--i >= 0) | |
10707 | percpu_counter_destroy(&ca->cpustat[i]); | |
10708 | free_percpu(ca->cpuusage); | |
10709 | out_free_ca: | |
10710 | kfree(ca); | |
10711 | out: | |
10712 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10713 | } |
10714 | ||
10715 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10716 | static void |
32cd756a | 10717 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10718 | { |
32cd756a | 10719 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10720 | int i; |
d842de87 | 10721 | |
ef12fefa BR |
10722 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10723 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10724 | free_percpu(ca->cpuusage); |
10725 | kfree(ca); | |
10726 | } | |
10727 | ||
720f5498 KC |
10728 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10729 | { | |
b36128c8 | 10730 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10731 | u64 data; |
10732 | ||
10733 | #ifndef CONFIG_64BIT | |
10734 | /* | |
10735 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10736 | */ | |
10737 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10738 | data = *cpuusage; | |
10739 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10740 | #else | |
10741 | data = *cpuusage; | |
10742 | #endif | |
10743 | ||
10744 | return data; | |
10745 | } | |
10746 | ||
10747 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10748 | { | |
b36128c8 | 10749 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10750 | |
10751 | #ifndef CONFIG_64BIT | |
10752 | /* | |
10753 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10754 | */ | |
10755 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10756 | *cpuusage = val; | |
10757 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10758 | #else | |
10759 | *cpuusage = val; | |
10760 | #endif | |
10761 | } | |
10762 | ||
d842de87 | 10763 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10764 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10765 | { |
32cd756a | 10766 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10767 | u64 totalcpuusage = 0; |
10768 | int i; | |
10769 | ||
720f5498 KC |
10770 | for_each_present_cpu(i) |
10771 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10772 | |
10773 | return totalcpuusage; | |
10774 | } | |
10775 | ||
0297b803 DG |
10776 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10777 | u64 reset) | |
10778 | { | |
10779 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10780 | int err = 0; | |
10781 | int i; | |
10782 | ||
10783 | if (reset) { | |
10784 | err = -EINVAL; | |
10785 | goto out; | |
10786 | } | |
10787 | ||
720f5498 KC |
10788 | for_each_present_cpu(i) |
10789 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10790 | |
0297b803 DG |
10791 | out: |
10792 | return err; | |
10793 | } | |
10794 | ||
e9515c3c KC |
10795 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10796 | struct seq_file *m) | |
10797 | { | |
10798 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10799 | u64 percpu; | |
10800 | int i; | |
10801 | ||
10802 | for_each_present_cpu(i) { | |
10803 | percpu = cpuacct_cpuusage_read(ca, i); | |
10804 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10805 | } | |
10806 | seq_printf(m, "\n"); | |
10807 | return 0; | |
10808 | } | |
10809 | ||
ef12fefa BR |
10810 | static const char *cpuacct_stat_desc[] = { |
10811 | [CPUACCT_STAT_USER] = "user", | |
10812 | [CPUACCT_STAT_SYSTEM] = "system", | |
10813 | }; | |
10814 | ||
10815 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10816 | struct cgroup_map_cb *cb) | |
10817 | { | |
10818 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10819 | int i; | |
10820 | ||
10821 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10822 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10823 | val = cputime64_to_clock_t(val); | |
10824 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10825 | } | |
10826 | return 0; | |
10827 | } | |
10828 | ||
d842de87 SV |
10829 | static struct cftype files[] = { |
10830 | { | |
10831 | .name = "usage", | |
f4c753b7 PM |
10832 | .read_u64 = cpuusage_read, |
10833 | .write_u64 = cpuusage_write, | |
d842de87 | 10834 | }, |
e9515c3c KC |
10835 | { |
10836 | .name = "usage_percpu", | |
10837 | .read_seq_string = cpuacct_percpu_seq_read, | |
10838 | }, | |
ef12fefa BR |
10839 | { |
10840 | .name = "stat", | |
10841 | .read_map = cpuacct_stats_show, | |
10842 | }, | |
d842de87 SV |
10843 | }; |
10844 | ||
32cd756a | 10845 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10846 | { |
32cd756a | 10847 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10848 | } |
10849 | ||
10850 | /* | |
10851 | * charge this task's execution time to its accounting group. | |
10852 | * | |
10853 | * called with rq->lock held. | |
10854 | */ | |
10855 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10856 | { | |
10857 | struct cpuacct *ca; | |
934352f2 | 10858 | int cpu; |
d842de87 | 10859 | |
c40c6f85 | 10860 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10861 | return; |
10862 | ||
934352f2 | 10863 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10864 | |
10865 | rcu_read_lock(); | |
10866 | ||
d842de87 | 10867 | ca = task_ca(tsk); |
d842de87 | 10868 | |
934352f2 | 10869 | for (; ca; ca = ca->parent) { |
b36128c8 | 10870 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10871 | *cpuusage += cputime; |
10872 | } | |
a18b83b7 BR |
10873 | |
10874 | rcu_read_unlock(); | |
d842de87 SV |
10875 | } |
10876 | ||
ef12fefa BR |
10877 | /* |
10878 | * Charge the system/user time to the task's accounting group. | |
10879 | */ | |
10880 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10881 | enum cpuacct_stat_index idx, cputime_t val) | |
10882 | { | |
10883 | struct cpuacct *ca; | |
10884 | ||
10885 | if (unlikely(!cpuacct_subsys.active)) | |
10886 | return; | |
10887 | ||
10888 | rcu_read_lock(); | |
10889 | ca = task_ca(tsk); | |
10890 | ||
10891 | do { | |
10892 | percpu_counter_add(&ca->cpustat[idx], val); | |
10893 | ca = ca->parent; | |
10894 | } while (ca); | |
10895 | rcu_read_unlock(); | |
10896 | } | |
10897 | ||
d842de87 SV |
10898 | struct cgroup_subsys cpuacct_subsys = { |
10899 | .name = "cpuacct", | |
10900 | .create = cpuacct_create, | |
10901 | .destroy = cpuacct_destroy, | |
10902 | .populate = cpuacct_populate, | |
10903 | .subsys_id = cpuacct_subsys_id, | |
10904 | }; | |
10905 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
10906 | |
10907 | #ifndef CONFIG_SMP | |
10908 | ||
10909 | int rcu_expedited_torture_stats(char *page) | |
10910 | { | |
10911 | return 0; | |
10912 | } | |
10913 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10914 | ||
10915 | void synchronize_sched_expedited(void) | |
10916 | { | |
10917 | } | |
10918 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10919 | ||
10920 | #else /* #ifndef CONFIG_SMP */ | |
10921 | ||
10922 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
10923 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
10924 | ||
10925 | #define RCU_EXPEDITED_STATE_POST -2 | |
10926 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
10927 | ||
10928 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10929 | ||
10930 | int rcu_expedited_torture_stats(char *page) | |
10931 | { | |
10932 | int cnt = 0; | |
10933 | int cpu; | |
10934 | ||
10935 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
10936 | for_each_online_cpu(cpu) { | |
10937 | cnt += sprintf(&page[cnt], " %d:%d", | |
10938 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
10939 | } | |
10940 | cnt += sprintf(&page[cnt], "\n"); | |
10941 | return cnt; | |
10942 | } | |
10943 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10944 | ||
10945 | static long synchronize_sched_expedited_count; | |
10946 | ||
10947 | /* | |
10948 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
10949 | * approach to force grace period to end quickly. This consumes | |
10950 | * significant time on all CPUs, and is thus not recommended for | |
10951 | * any sort of common-case code. | |
10952 | * | |
10953 | * Note that it is illegal to call this function while holding any | |
10954 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
10955 | * observe this restriction will result in deadlock. | |
10956 | */ | |
10957 | void synchronize_sched_expedited(void) | |
10958 | { | |
10959 | int cpu; | |
10960 | unsigned long flags; | |
10961 | bool need_full_sync = 0; | |
10962 | struct rq *rq; | |
10963 | struct migration_req *req; | |
10964 | long snap; | |
10965 | int trycount = 0; | |
10966 | ||
10967 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
10968 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
10969 | get_online_cpus(); | |
10970 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
10971 | put_online_cpus(); | |
10972 | if (trycount++ < 10) | |
10973 | udelay(trycount * num_online_cpus()); | |
10974 | else { | |
10975 | synchronize_sched(); | |
10976 | return; | |
10977 | } | |
10978 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
10979 | smp_mb(); /* ensure test happens before caller kfree */ | |
10980 | return; | |
10981 | } | |
10982 | get_online_cpus(); | |
10983 | } | |
10984 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
10985 | for_each_online_cpu(cpu) { | |
10986 | rq = cpu_rq(cpu); | |
10987 | req = &per_cpu(rcu_migration_req, cpu); | |
10988 | init_completion(&req->done); | |
10989 | req->task = NULL; | |
10990 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
10991 | spin_lock_irqsave(&rq->lock, flags); | |
10992 | list_add(&req->list, &rq->migration_queue); | |
10993 | spin_unlock_irqrestore(&rq->lock, flags); | |
10994 | wake_up_process(rq->migration_thread); | |
10995 | } | |
10996 | for_each_online_cpu(cpu) { | |
10997 | rcu_expedited_state = cpu; | |
10998 | req = &per_cpu(rcu_migration_req, cpu); | |
10999 | rq = cpu_rq(cpu); | |
11000 | wait_for_completion(&req->done); | |
11001 | spin_lock_irqsave(&rq->lock, flags); | |
11002 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) | |
11003 | need_full_sync = 1; | |
11004 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
11005 | spin_unlock_irqrestore(&rq->lock, flags); | |
11006 | } | |
11007 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
11008 | mutex_unlock(&rcu_sched_expedited_mutex); | |
11009 | put_online_cpus(); | |
11010 | if (need_full_sync) | |
11011 | synchronize_sched(); | |
11012 | } | |
11013 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
11014 | ||
11015 | #endif /* #else #ifndef CONFIG_SMP */ |