]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
663997d4 JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
1da177e4 LT |
31 | #include <linux/mm.h> |
32 | #include <linux/module.h> | |
33 | #include <linux/nmi.h> | |
34 | #include <linux/init.h> | |
dff06c15 | 35 | #include <linux/uaccess.h> |
1da177e4 LT |
36 | #include <linux/highmem.h> |
37 | #include <linux/smp_lock.h> | |
38 | #include <asm/mmu_context.h> | |
39 | #include <linux/interrupt.h> | |
c59ede7b | 40 | #include <linux/capability.h> |
1da177e4 LT |
41 | #include <linux/completion.h> |
42 | #include <linux/kernel_stat.h> | |
9a11b49a | 43 | #include <linux/debug_locks.h> |
cdd6c482 | 44 | #include <linux/perf_event.h> |
1da177e4 LT |
45 | #include <linux/security.h> |
46 | #include <linux/notifier.h> | |
47 | #include <linux/profile.h> | |
7dfb7103 | 48 | #include <linux/freezer.h> |
198e2f18 | 49 | #include <linux/vmalloc.h> |
1da177e4 LT |
50 | #include <linux/blkdev.h> |
51 | #include <linux/delay.h> | |
b488893a | 52 | #include <linux/pid_namespace.h> |
1da177e4 LT |
53 | #include <linux/smp.h> |
54 | #include <linux/threads.h> | |
55 | #include <linux/timer.h> | |
56 | #include <linux/rcupdate.h> | |
57 | #include <linux/cpu.h> | |
58 | #include <linux/cpuset.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/kthread.h> | |
b5aadf7f | 61 | #include <linux/proc_fs.h> |
1da177e4 | 62 | #include <linux/seq_file.h> |
e692ab53 | 63 | #include <linux/sysctl.h> |
1da177e4 LT |
64 | #include <linux/syscalls.h> |
65 | #include <linux/times.h> | |
8f0ab514 | 66 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 67 | #include <linux/kprobes.h> |
0ff92245 | 68 | #include <linux/delayacct.h> |
dff06c15 | 69 | #include <linux/unistd.h> |
f5ff8422 | 70 | #include <linux/pagemap.h> |
8f4d37ec | 71 | #include <linux/hrtimer.h> |
30914a58 | 72 | #include <linux/tick.h> |
f00b45c1 PZ |
73 | #include <linux/debugfs.h> |
74 | #include <linux/ctype.h> | |
6cd8a4bb | 75 | #include <linux/ftrace.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 | 145 | /* nests inside the rq lock: */ |
0986b11b | 146 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
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 | ||
0986b11b | 183 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 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 | ||
0986b11b | 205 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 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 | 221 | } |
0986b11b | 222 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
223 | } |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
052f1dc7 | 238 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
052f1dc7 | 248 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
249 | struct cgroup_subsys_state css; |
250 | #endif | |
052f1dc7 | 251 | |
6c415b92 AB |
252 | #ifdef CONFIG_USER_SCHED |
253 | uid_t uid; | |
254 | #endif | |
255 | ||
052f1dc7 | 256 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
257 | /* schedulable entities of this group on each cpu */ |
258 | struct sched_entity **se; | |
259 | /* runqueue "owned" by this group on each cpu */ | |
260 | struct cfs_rq **cfs_rq; | |
261 | unsigned long shares; | |
052f1dc7 PZ |
262 | #endif |
263 | ||
264 | #ifdef CONFIG_RT_GROUP_SCHED | |
265 | struct sched_rt_entity **rt_se; | |
266 | struct rt_rq **rt_rq; | |
267 | ||
d0b27fa7 | 268 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 269 | #endif |
6b2d7700 | 270 | |
ae8393e5 | 271 | struct rcu_head rcu; |
6f505b16 | 272 | struct list_head list; |
f473aa5e PZ |
273 | |
274 | struct task_group *parent; | |
275 | struct list_head siblings; | |
276 | struct list_head children; | |
29f59db3 SV |
277 | }; |
278 | ||
354d60c2 | 279 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 280 | |
6c415b92 AB |
281 | /* Helper function to pass uid information to create_sched_user() */ |
282 | void set_tg_uid(struct user_struct *user) | |
283 | { | |
284 | user->tg->uid = user->uid; | |
285 | } | |
286 | ||
eff766a6 PZ |
287 | /* |
288 | * Root task group. | |
84e9dabf AS |
289 | * Every UID task group (including init_task_group aka UID-0) will |
290 | * be a child to this group. | |
eff766a6 PZ |
291 | */ |
292 | struct task_group root_task_group; | |
293 | ||
052f1dc7 | 294 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
295 | /* Default task group's sched entity on each cpu */ |
296 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
297 | /* Default task group's cfs_rq on each cpu */ | |
ada3fa15 | 298 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct cfs_rq, init_tg_cfs_rq); |
6d6bc0ad | 299 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
300 | |
301 | #ifdef CONFIG_RT_GROUP_SCHED | |
302 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
1871e52c | 303 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rt_rq, init_rt_rq_var); |
6d6bc0ad | 304 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 305 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 306 | #define root_task_group init_task_group |
9a7e0b18 | 307 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 308 | |
8ed36996 | 309 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
310 | * a task group's cpu shares. |
311 | */ | |
8ed36996 | 312 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 313 | |
e9036b36 CG |
314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
315 | ||
57310a98 PZ |
316 | #ifdef CONFIG_SMP |
317 | static int root_task_group_empty(void) | |
318 | { | |
319 | return list_empty(&root_task_group.children); | |
320 | } | |
321 | #endif | |
322 | ||
052f1dc7 PZ |
323 | #ifdef CONFIG_USER_SCHED |
324 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 325 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 326 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 327 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 328 | |
cb4ad1ff | 329 | /* |
2e084786 LJ |
330 | * A weight of 0 or 1 can cause arithmetics problems. |
331 | * A weight of a cfs_rq is the sum of weights of which entities | |
332 | * are queued on this cfs_rq, so a weight of a entity should not be | |
333 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
334 | * (The default weight is 1024 - so there's no practical |
335 | * limitation from this.) | |
336 | */ | |
18d95a28 | 337 | #define MIN_SHARES 2 |
2e084786 | 338 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 339 | |
052f1dc7 PZ |
340 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
341 | #endif | |
342 | ||
29f59db3 | 343 | /* Default task group. |
3a252015 | 344 | * Every task in system belong to this group at bootup. |
29f59db3 | 345 | */ |
434d53b0 | 346 | struct task_group init_task_group; |
29f59db3 SV |
347 | |
348 | /* return group to which a task belongs */ | |
4cf86d77 | 349 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 350 | { |
4cf86d77 | 351 | struct task_group *tg; |
9b5b7751 | 352 | |
052f1dc7 | 353 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
354 | rcu_read_lock(); |
355 | tg = __task_cred(p)->user->tg; | |
356 | rcu_read_unlock(); | |
052f1dc7 | 357 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
358 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
359 | struct task_group, css); | |
24e377a8 | 360 | #else |
41a2d6cf | 361 | tg = &init_task_group; |
24e377a8 | 362 | #endif |
9b5b7751 | 363 | return tg; |
29f59db3 SV |
364 | } |
365 | ||
366 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 367 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 368 | { |
052f1dc7 | 369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
370 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
371 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 372 | #endif |
6f505b16 | 373 | |
052f1dc7 | 374 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
375 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
376 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 377 | #endif |
29f59db3 SV |
378 | } |
379 | ||
380 | #else | |
381 | ||
6f505b16 | 382 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
383 | static inline struct task_group *task_group(struct task_struct *p) |
384 | { | |
385 | return NULL; | |
386 | } | |
29f59db3 | 387 | |
052f1dc7 | 388 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 389 | |
6aa645ea IM |
390 | /* CFS-related fields in a runqueue */ |
391 | struct cfs_rq { | |
392 | struct load_weight load; | |
393 | unsigned long nr_running; | |
394 | ||
6aa645ea | 395 | u64 exec_clock; |
e9acbff6 | 396 | u64 min_vruntime; |
6aa645ea IM |
397 | |
398 | struct rb_root tasks_timeline; | |
399 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
400 | |
401 | struct list_head tasks; | |
402 | struct list_head *balance_iterator; | |
403 | ||
404 | /* | |
405 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
406 | * It is set to NULL otherwise (i.e when none are currently running). |
407 | */ | |
4793241b | 408 | struct sched_entity *curr, *next, *last; |
ddc97297 | 409 | |
5ac5c4d6 | 410 | unsigned int nr_spread_over; |
ddc97297 | 411 | |
62160e3f | 412 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
413 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
414 | ||
41a2d6cf IM |
415 | /* |
416 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
417 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
418 | * (like users, containers etc.) | |
419 | * | |
420 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
421 | * list is used during load balance. | |
422 | */ | |
41a2d6cf IM |
423 | struct list_head leaf_cfs_rq_list; |
424 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
425 | |
426 | #ifdef CONFIG_SMP | |
c09595f6 | 427 | /* |
c8cba857 | 428 | * the part of load.weight contributed by tasks |
c09595f6 | 429 | */ |
c8cba857 | 430 | unsigned long task_weight; |
c09595f6 | 431 | |
c8cba857 PZ |
432 | /* |
433 | * h_load = weight * f(tg) | |
434 | * | |
435 | * Where f(tg) is the recursive weight fraction assigned to | |
436 | * this group. | |
437 | */ | |
438 | unsigned long h_load; | |
c09595f6 | 439 | |
c8cba857 PZ |
440 | /* |
441 | * this cpu's part of tg->shares | |
442 | */ | |
443 | unsigned long shares; | |
f1d239f7 PZ |
444 | |
445 | /* | |
446 | * load.weight at the time we set shares | |
447 | */ | |
448 | unsigned long rq_weight; | |
c09595f6 | 449 | #endif |
6aa645ea IM |
450 | #endif |
451 | }; | |
1da177e4 | 452 | |
6aa645ea IM |
453 | /* Real-Time classes' related field in a runqueue: */ |
454 | struct rt_rq { | |
455 | struct rt_prio_array active; | |
63489e45 | 456 | unsigned long rt_nr_running; |
052f1dc7 | 457 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
458 | struct { |
459 | int curr; /* highest queued rt task prio */ | |
398a153b | 460 | #ifdef CONFIG_SMP |
e864c499 | 461 | int next; /* next highest */ |
398a153b | 462 | #endif |
e864c499 | 463 | } highest_prio; |
6f505b16 | 464 | #endif |
fa85ae24 | 465 | #ifdef CONFIG_SMP |
73fe6aae | 466 | unsigned long rt_nr_migratory; |
a1ba4d8b | 467 | unsigned long rt_nr_total; |
a22d7fc1 | 468 | int overloaded; |
917b627d | 469 | struct plist_head pushable_tasks; |
fa85ae24 | 470 | #endif |
6f505b16 | 471 | int rt_throttled; |
fa85ae24 | 472 | u64 rt_time; |
ac086bc2 | 473 | u64 rt_runtime; |
ea736ed5 | 474 | /* Nests inside the rq lock: */ |
0986b11b | 475 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 476 | |
052f1dc7 | 477 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
478 | unsigned long rt_nr_boosted; |
479 | ||
6f505b16 PZ |
480 | struct rq *rq; |
481 | struct list_head leaf_rt_rq_list; | |
482 | struct task_group *tg; | |
483 | struct sched_rt_entity *rt_se; | |
484 | #endif | |
6aa645ea IM |
485 | }; |
486 | ||
57d885fe GH |
487 | #ifdef CONFIG_SMP |
488 | ||
489 | /* | |
490 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
491 | * variables. Each exclusive cpuset essentially defines an island domain by |
492 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
493 | * exclusive cpuset is created, we also create and attach a new root-domain |
494 | * object. | |
495 | * | |
57d885fe GH |
496 | */ |
497 | struct root_domain { | |
498 | atomic_t refcount; | |
c6c4927b RR |
499 | cpumask_var_t span; |
500 | cpumask_var_t online; | |
637f5085 | 501 | |
0eab9146 | 502 | /* |
637f5085 GH |
503 | * The "RT overload" flag: it gets set if a CPU has more than |
504 | * one runnable RT task. | |
505 | */ | |
c6c4927b | 506 | cpumask_var_t rto_mask; |
0eab9146 | 507 | atomic_t rto_count; |
6e0534f2 GH |
508 | #ifdef CONFIG_SMP |
509 | struct cpupri cpupri; | |
510 | #endif | |
57d885fe GH |
511 | }; |
512 | ||
dc938520 GH |
513 | /* |
514 | * By default the system creates a single root-domain with all cpus as | |
515 | * members (mimicking the global state we have today). | |
516 | */ | |
57d885fe GH |
517 | static struct root_domain def_root_domain; |
518 | ||
519 | #endif | |
520 | ||
1da177e4 LT |
521 | /* |
522 | * This is the main, per-CPU runqueue data structure. | |
523 | * | |
524 | * Locking rule: those places that want to lock multiple runqueues | |
525 | * (such as the load balancing or the thread migration code), lock | |
526 | * acquire operations must be ordered by ascending &runqueue. | |
527 | */ | |
70b97a7f | 528 | struct rq { |
d8016491 | 529 | /* runqueue lock: */ |
05fa785c | 530 | raw_spinlock_t lock; |
1da177e4 LT |
531 | |
532 | /* | |
533 | * nr_running and cpu_load should be in the same cacheline because | |
534 | * remote CPUs use both these fields when doing load calculation. | |
535 | */ | |
536 | unsigned long nr_running; | |
6aa645ea IM |
537 | #define CPU_LOAD_IDX_MAX 5 |
538 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c SS |
539 | #ifdef CONFIG_NO_HZ |
540 | unsigned char in_nohz_recently; | |
541 | #endif | |
d8016491 IM |
542 | /* capture load from *all* tasks on this cpu: */ |
543 | struct load_weight load; | |
6aa645ea IM |
544 | unsigned long nr_load_updates; |
545 | u64 nr_switches; | |
546 | ||
547 | struct cfs_rq cfs; | |
6f505b16 | 548 | struct rt_rq rt; |
6f505b16 | 549 | |
6aa645ea | 550 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
551 | /* list of leaf cfs_rq on this cpu: */ |
552 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
553 | #endif |
554 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 555 | struct list_head leaf_rt_rq_list; |
1da177e4 | 556 | #endif |
1da177e4 LT |
557 | |
558 | /* | |
559 | * This is part of a global counter where only the total sum | |
560 | * over all CPUs matters. A task can increase this counter on | |
561 | * one CPU and if it got migrated afterwards it may decrease | |
562 | * it on another CPU. Always updated under the runqueue lock: | |
563 | */ | |
564 | unsigned long nr_uninterruptible; | |
565 | ||
36c8b586 | 566 | struct task_struct *curr, *idle; |
c9819f45 | 567 | unsigned long next_balance; |
1da177e4 | 568 | struct mm_struct *prev_mm; |
6aa645ea | 569 | |
3e51f33f | 570 | u64 clock; |
6aa645ea | 571 | |
1da177e4 LT |
572 | atomic_t nr_iowait; |
573 | ||
574 | #ifdef CONFIG_SMP | |
0eab9146 | 575 | struct root_domain *rd; |
1da177e4 LT |
576 | struct sched_domain *sd; |
577 | ||
a0a522ce | 578 | unsigned char idle_at_tick; |
1da177e4 | 579 | /* For active balancing */ |
3f029d3c | 580 | int post_schedule; |
1da177e4 LT |
581 | int active_balance; |
582 | int push_cpu; | |
d8016491 IM |
583 | /* cpu of this runqueue: */ |
584 | int cpu; | |
1f11eb6a | 585 | int online; |
1da177e4 | 586 | |
a8a51d5e | 587 | unsigned long avg_load_per_task; |
1da177e4 | 588 | |
36c8b586 | 589 | struct task_struct *migration_thread; |
1da177e4 | 590 | struct list_head migration_queue; |
e9e9250b PZ |
591 | |
592 | u64 rt_avg; | |
593 | u64 age_stamp; | |
1b9508f6 MG |
594 | u64 idle_stamp; |
595 | u64 avg_idle; | |
1da177e4 LT |
596 | #endif |
597 | ||
dce48a84 TG |
598 | /* calc_load related fields */ |
599 | unsigned long calc_load_update; | |
600 | long calc_load_active; | |
601 | ||
8f4d37ec | 602 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
603 | #ifdef CONFIG_SMP |
604 | int hrtick_csd_pending; | |
605 | struct call_single_data hrtick_csd; | |
606 | #endif | |
8f4d37ec PZ |
607 | struct hrtimer hrtick_timer; |
608 | #endif | |
609 | ||
1da177e4 LT |
610 | #ifdef CONFIG_SCHEDSTATS |
611 | /* latency stats */ | |
612 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
613 | unsigned long long rq_cpu_time; |
614 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
615 | |
616 | /* sys_sched_yield() stats */ | |
480b9434 | 617 | unsigned int yld_count; |
1da177e4 LT |
618 | |
619 | /* schedule() stats */ | |
480b9434 KC |
620 | unsigned int sched_switch; |
621 | unsigned int sched_count; | |
622 | unsigned int sched_goidle; | |
1da177e4 LT |
623 | |
624 | /* try_to_wake_up() stats */ | |
480b9434 KC |
625 | unsigned int ttwu_count; |
626 | unsigned int ttwu_local; | |
b8efb561 IM |
627 | |
628 | /* BKL stats */ | |
480b9434 | 629 | unsigned int bkl_count; |
1da177e4 LT |
630 | #endif |
631 | }; | |
632 | ||
f34e3b61 | 633 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 634 | |
7d478721 PZ |
635 | static inline |
636 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 637 | { |
7d478721 | 638 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
dd41f596 IM |
639 | } |
640 | ||
0a2966b4 CL |
641 | static inline int cpu_of(struct rq *rq) |
642 | { | |
643 | #ifdef CONFIG_SMP | |
644 | return rq->cpu; | |
645 | #else | |
646 | return 0; | |
647 | #endif | |
648 | } | |
649 | ||
674311d5 NP |
650 | /* |
651 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 652 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
653 | * |
654 | * The domain tree of any CPU may only be accessed from within | |
655 | * preempt-disabled sections. | |
656 | */ | |
48f24c4d IM |
657 | #define for_each_domain(cpu, __sd) \ |
658 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
659 | |
660 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
661 | #define this_rq() (&__get_cpu_var(runqueues)) | |
662 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
663 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 664 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 665 | |
aa9c4c0f | 666 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
667 | { |
668 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
669 | } | |
670 | ||
bf5c91ba IM |
671 | /* |
672 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
673 | */ | |
674 | #ifdef CONFIG_SCHED_DEBUG | |
675 | # define const_debug __read_mostly | |
676 | #else | |
677 | # define const_debug static const | |
678 | #endif | |
679 | ||
017730c1 IM |
680 | /** |
681 | * runqueue_is_locked | |
e17b38bf | 682 | * @cpu: the processor in question. |
017730c1 IM |
683 | * |
684 | * Returns true if the current cpu runqueue is locked. | |
685 | * This interface allows printk to be called with the runqueue lock | |
686 | * held and know whether or not it is OK to wake up the klogd. | |
687 | */ | |
89f19f04 | 688 | int runqueue_is_locked(int cpu) |
017730c1 | 689 | { |
05fa785c | 690 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
691 | } |
692 | ||
bf5c91ba IM |
693 | /* |
694 | * Debugging: various feature bits | |
695 | */ | |
f00b45c1 PZ |
696 | |
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | __SCHED_FEAT_##name , | |
699 | ||
bf5c91ba | 700 | enum { |
f00b45c1 | 701 | #include "sched_features.h" |
bf5c91ba IM |
702 | }; |
703 | ||
f00b45c1 PZ |
704 | #undef SCHED_FEAT |
705 | ||
706 | #define SCHED_FEAT(name, enabled) \ | |
707 | (1UL << __SCHED_FEAT_##name) * enabled | | |
708 | ||
bf5c91ba | 709 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
710 | #include "sched_features.h" |
711 | 0; | |
712 | ||
713 | #undef SCHED_FEAT | |
714 | ||
715 | #ifdef CONFIG_SCHED_DEBUG | |
716 | #define SCHED_FEAT(name, enabled) \ | |
717 | #name , | |
718 | ||
983ed7a6 | 719 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
720 | #include "sched_features.h" |
721 | NULL | |
722 | }; | |
723 | ||
724 | #undef SCHED_FEAT | |
725 | ||
34f3a814 | 726 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 727 | { |
f00b45c1 PZ |
728 | int i; |
729 | ||
730 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
731 | if (!(sysctl_sched_features & (1UL << i))) |
732 | seq_puts(m, "NO_"); | |
733 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 734 | } |
34f3a814 | 735 | seq_puts(m, "\n"); |
f00b45c1 | 736 | |
34f3a814 | 737 | return 0; |
f00b45c1 PZ |
738 | } |
739 | ||
740 | static ssize_t | |
741 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
742 | size_t cnt, loff_t *ppos) | |
743 | { | |
744 | char buf[64]; | |
745 | char *cmp = buf; | |
746 | int neg = 0; | |
747 | int i; | |
748 | ||
749 | if (cnt > 63) | |
750 | cnt = 63; | |
751 | ||
752 | if (copy_from_user(&buf, ubuf, cnt)) | |
753 | return -EFAULT; | |
754 | ||
755 | buf[cnt] = 0; | |
756 | ||
c24b7c52 | 757 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
758 | neg = 1; |
759 | cmp += 3; | |
760 | } | |
761 | ||
762 | for (i = 0; sched_feat_names[i]; i++) { | |
763 | int len = strlen(sched_feat_names[i]); | |
764 | ||
765 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
766 | if (neg) | |
767 | sysctl_sched_features &= ~(1UL << i); | |
768 | else | |
769 | sysctl_sched_features |= (1UL << i); | |
770 | break; | |
771 | } | |
772 | } | |
773 | ||
774 | if (!sched_feat_names[i]) | |
775 | return -EINVAL; | |
776 | ||
42994724 | 777 | *ppos += cnt; |
f00b45c1 PZ |
778 | |
779 | return cnt; | |
780 | } | |
781 | ||
34f3a814 LZ |
782 | static int sched_feat_open(struct inode *inode, struct file *filp) |
783 | { | |
784 | return single_open(filp, sched_feat_show, NULL); | |
785 | } | |
786 | ||
828c0950 | 787 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
788 | .open = sched_feat_open, |
789 | .write = sched_feat_write, | |
790 | .read = seq_read, | |
791 | .llseek = seq_lseek, | |
792 | .release = single_release, | |
f00b45c1 PZ |
793 | }; |
794 | ||
795 | static __init int sched_init_debug(void) | |
796 | { | |
f00b45c1 PZ |
797 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
798 | &sched_feat_fops); | |
799 | ||
800 | return 0; | |
801 | } | |
802 | late_initcall(sched_init_debug); | |
803 | ||
804 | #endif | |
805 | ||
806 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 807 | |
b82d9fdd PZ |
808 | /* |
809 | * Number of tasks to iterate in a single balance run. | |
810 | * Limited because this is done with IRQs disabled. | |
811 | */ | |
812 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
813 | ||
2398f2c6 PZ |
814 | /* |
815 | * ratelimit for updating the group shares. | |
55cd5340 | 816 | * default: 0.25ms |
2398f2c6 | 817 | */ |
55cd5340 | 818 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 819 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 820 | |
ffda12a1 PZ |
821 | /* |
822 | * Inject some fuzzyness into changing the per-cpu group shares | |
823 | * this avoids remote rq-locks at the expense of fairness. | |
824 | * default: 4 | |
825 | */ | |
826 | unsigned int sysctl_sched_shares_thresh = 4; | |
827 | ||
e9e9250b PZ |
828 | /* |
829 | * period over which we average the RT time consumption, measured | |
830 | * in ms. | |
831 | * | |
832 | * default: 1s | |
833 | */ | |
834 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
835 | ||
fa85ae24 | 836 | /* |
9f0c1e56 | 837 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
838 | * default: 1s |
839 | */ | |
9f0c1e56 | 840 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 841 | |
6892b75e IM |
842 | static __read_mostly int scheduler_running; |
843 | ||
9f0c1e56 PZ |
844 | /* |
845 | * part of the period that we allow rt tasks to run in us. | |
846 | * default: 0.95s | |
847 | */ | |
848 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 849 | |
d0b27fa7 PZ |
850 | static inline u64 global_rt_period(void) |
851 | { | |
852 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
853 | } | |
854 | ||
855 | static inline u64 global_rt_runtime(void) | |
856 | { | |
e26873bb | 857 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
858 | return RUNTIME_INF; |
859 | ||
860 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
861 | } | |
fa85ae24 | 862 | |
1da177e4 | 863 | #ifndef prepare_arch_switch |
4866cde0 NP |
864 | # define prepare_arch_switch(next) do { } while (0) |
865 | #endif | |
866 | #ifndef finish_arch_switch | |
867 | # define finish_arch_switch(prev) do { } while (0) | |
868 | #endif | |
869 | ||
051a1d1a DA |
870 | static inline int task_current(struct rq *rq, struct task_struct *p) |
871 | { | |
872 | return rq->curr == p; | |
873 | } | |
874 | ||
4866cde0 | 875 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 876 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 877 | { |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | } |
880 | ||
70b97a7f | 881 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
882 | { |
883 | } | |
884 | ||
70b97a7f | 885 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 886 | { |
da04c035 IM |
887 | #ifdef CONFIG_DEBUG_SPINLOCK |
888 | /* this is a valid case when another task releases the spinlock */ | |
889 | rq->lock.owner = current; | |
890 | #endif | |
8a25d5de IM |
891 | /* |
892 | * If we are tracking spinlock dependencies then we have to | |
893 | * fix up the runqueue lock - which gets 'carried over' from | |
894 | * prev into current: | |
895 | */ | |
896 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
897 | ||
05fa785c | 898 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
899 | } |
900 | ||
901 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 902 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
903 | { |
904 | #ifdef CONFIG_SMP | |
905 | return p->oncpu; | |
906 | #else | |
051a1d1a | 907 | return task_current(rq, p); |
4866cde0 NP |
908 | #endif |
909 | } | |
910 | ||
70b97a7f | 911 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
912 | { |
913 | #ifdef CONFIG_SMP | |
914 | /* | |
915 | * We can optimise this out completely for !SMP, because the | |
916 | * SMP rebalancing from interrupt is the only thing that cares | |
917 | * here. | |
918 | */ | |
919 | next->oncpu = 1; | |
920 | #endif | |
921 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 922 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 923 | #else |
05fa785c | 924 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
925 | #endif |
926 | } | |
927 | ||
70b97a7f | 928 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
929 | { |
930 | #ifdef CONFIG_SMP | |
931 | /* | |
932 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
933 | * We must ensure this doesn't happen until the switch is completely | |
934 | * finished. | |
935 | */ | |
936 | smp_wmb(); | |
937 | prev->oncpu = 0; | |
938 | #endif | |
939 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
940 | local_irq_enable(); | |
1da177e4 | 941 | #endif |
4866cde0 NP |
942 | } |
943 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 944 | |
b29739f9 IM |
945 | /* |
946 | * __task_rq_lock - lock the runqueue a given task resides on. | |
947 | * Must be called interrupts disabled. | |
948 | */ | |
70b97a7f | 949 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
950 | __acquires(rq->lock) |
951 | { | |
3a5c359a AK |
952 | for (;;) { |
953 | struct rq *rq = task_rq(p); | |
05fa785c | 954 | raw_spin_lock(&rq->lock); |
3a5c359a AK |
955 | if (likely(rq == task_rq(p))) |
956 | return rq; | |
05fa785c | 957 | raw_spin_unlock(&rq->lock); |
b29739f9 | 958 | } |
b29739f9 IM |
959 | } |
960 | ||
1da177e4 LT |
961 | /* |
962 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 963 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
964 | * explicitly disabling preemption. |
965 | */ | |
70b97a7f | 966 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
967 | __acquires(rq->lock) |
968 | { | |
70b97a7f | 969 | struct rq *rq; |
1da177e4 | 970 | |
3a5c359a AK |
971 | for (;;) { |
972 | local_irq_save(*flags); | |
973 | rq = task_rq(p); | |
05fa785c | 974 | raw_spin_lock(&rq->lock); |
3a5c359a AK |
975 | if (likely(rq == task_rq(p))) |
976 | return rq; | |
05fa785c | 977 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 978 | } |
1da177e4 LT |
979 | } |
980 | ||
ad474cac ON |
981 | void task_rq_unlock_wait(struct task_struct *p) |
982 | { | |
983 | struct rq *rq = task_rq(p); | |
984 | ||
985 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 986 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
987 | } |
988 | ||
a9957449 | 989 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
990 | __releases(rq->lock) |
991 | { | |
05fa785c | 992 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
993 | } |
994 | ||
70b97a7f | 995 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
996 | __releases(rq->lock) |
997 | { | |
05fa785c | 998 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
999 | } |
1000 | ||
1da177e4 | 1001 | /* |
cc2a73b5 | 1002 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1003 | */ |
a9957449 | 1004 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1005 | __acquires(rq->lock) |
1006 | { | |
70b97a7f | 1007 | struct rq *rq; |
1da177e4 LT |
1008 | |
1009 | local_irq_disable(); | |
1010 | rq = this_rq(); | |
05fa785c | 1011 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1012 | |
1013 | return rq; | |
1014 | } | |
1015 | ||
8f4d37ec PZ |
1016 | #ifdef CONFIG_SCHED_HRTICK |
1017 | /* | |
1018 | * Use HR-timers to deliver accurate preemption points. | |
1019 | * | |
1020 | * Its all a bit involved since we cannot program an hrt while holding the | |
1021 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1022 | * reschedule event. | |
1023 | * | |
1024 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1025 | * rq->lock. | |
1026 | */ | |
8f4d37ec PZ |
1027 | |
1028 | /* | |
1029 | * Use hrtick when: | |
1030 | * - enabled by features | |
1031 | * - hrtimer is actually high res | |
1032 | */ | |
1033 | static inline int hrtick_enabled(struct rq *rq) | |
1034 | { | |
1035 | if (!sched_feat(HRTICK)) | |
1036 | return 0; | |
ba42059f | 1037 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1038 | return 0; |
8f4d37ec PZ |
1039 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1040 | } | |
1041 | ||
8f4d37ec PZ |
1042 | static void hrtick_clear(struct rq *rq) |
1043 | { | |
1044 | if (hrtimer_active(&rq->hrtick_timer)) | |
1045 | hrtimer_cancel(&rq->hrtick_timer); | |
1046 | } | |
1047 | ||
8f4d37ec PZ |
1048 | /* |
1049 | * High-resolution timer tick. | |
1050 | * Runs from hardirq context with interrupts disabled. | |
1051 | */ | |
1052 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1053 | { | |
1054 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1055 | ||
1056 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1057 | ||
05fa785c | 1058 | raw_spin_lock(&rq->lock); |
3e51f33f | 1059 | update_rq_clock(rq); |
8f4d37ec | 1060 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1061 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1062 | |
1063 | return HRTIMER_NORESTART; | |
1064 | } | |
1065 | ||
95e904c7 | 1066 | #ifdef CONFIG_SMP |
31656519 PZ |
1067 | /* |
1068 | * called from hardirq (IPI) context | |
1069 | */ | |
1070 | static void __hrtick_start(void *arg) | |
b328ca18 | 1071 | { |
31656519 | 1072 | struct rq *rq = arg; |
b328ca18 | 1073 | |
05fa785c | 1074 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1075 | hrtimer_restart(&rq->hrtick_timer); |
1076 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1077 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1078 | } |
1079 | ||
31656519 PZ |
1080 | /* |
1081 | * Called to set the hrtick timer state. | |
1082 | * | |
1083 | * called with rq->lock held and irqs disabled | |
1084 | */ | |
1085 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1086 | { |
31656519 PZ |
1087 | struct hrtimer *timer = &rq->hrtick_timer; |
1088 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1089 | |
cc584b21 | 1090 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1091 | |
1092 | if (rq == this_rq()) { | |
1093 | hrtimer_restart(timer); | |
1094 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1095 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1096 | rq->hrtick_csd_pending = 1; |
1097 | } | |
b328ca18 PZ |
1098 | } |
1099 | ||
1100 | static int | |
1101 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1102 | { | |
1103 | int cpu = (int)(long)hcpu; | |
1104 | ||
1105 | switch (action) { | |
1106 | case CPU_UP_CANCELED: | |
1107 | case CPU_UP_CANCELED_FROZEN: | |
1108 | case CPU_DOWN_PREPARE: | |
1109 | case CPU_DOWN_PREPARE_FROZEN: | |
1110 | case CPU_DEAD: | |
1111 | case CPU_DEAD_FROZEN: | |
31656519 | 1112 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1113 | return NOTIFY_OK; |
1114 | } | |
1115 | ||
1116 | return NOTIFY_DONE; | |
1117 | } | |
1118 | ||
fa748203 | 1119 | static __init void init_hrtick(void) |
b328ca18 PZ |
1120 | { |
1121 | hotcpu_notifier(hotplug_hrtick, 0); | |
1122 | } | |
31656519 PZ |
1123 | #else |
1124 | /* | |
1125 | * Called to set the hrtick timer state. | |
1126 | * | |
1127 | * called with rq->lock held and irqs disabled | |
1128 | */ | |
1129 | static void hrtick_start(struct rq *rq, u64 delay) | |
1130 | { | |
7f1e2ca9 | 1131 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1132 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1133 | } |
b328ca18 | 1134 | |
006c75f1 | 1135 | static inline void init_hrtick(void) |
8f4d37ec | 1136 | { |
8f4d37ec | 1137 | } |
31656519 | 1138 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1139 | |
31656519 | 1140 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1141 | { |
31656519 PZ |
1142 | #ifdef CONFIG_SMP |
1143 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1144 | |
31656519 PZ |
1145 | rq->hrtick_csd.flags = 0; |
1146 | rq->hrtick_csd.func = __hrtick_start; | |
1147 | rq->hrtick_csd.info = rq; | |
1148 | #endif | |
8f4d37ec | 1149 | |
31656519 PZ |
1150 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1151 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1152 | } |
006c75f1 | 1153 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1154 | static inline void hrtick_clear(struct rq *rq) |
1155 | { | |
1156 | } | |
1157 | ||
8f4d37ec PZ |
1158 | static inline void init_rq_hrtick(struct rq *rq) |
1159 | { | |
1160 | } | |
1161 | ||
b328ca18 PZ |
1162 | static inline void init_hrtick(void) |
1163 | { | |
1164 | } | |
006c75f1 | 1165 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1166 | |
c24d20db IM |
1167 | /* |
1168 | * resched_task - mark a task 'to be rescheduled now'. | |
1169 | * | |
1170 | * On UP this means the setting of the need_resched flag, on SMP it | |
1171 | * might also involve a cross-CPU call to trigger the scheduler on | |
1172 | * the target CPU. | |
1173 | */ | |
1174 | #ifdef CONFIG_SMP | |
1175 | ||
1176 | #ifndef tsk_is_polling | |
1177 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1178 | #endif | |
1179 | ||
31656519 | 1180 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1181 | { |
1182 | int cpu; | |
1183 | ||
05fa785c | 1184 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1185 | |
5ed0cec0 | 1186 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1187 | return; |
1188 | ||
5ed0cec0 | 1189 | set_tsk_need_resched(p); |
c24d20db IM |
1190 | |
1191 | cpu = task_cpu(p); | |
1192 | if (cpu == smp_processor_id()) | |
1193 | return; | |
1194 | ||
1195 | /* NEED_RESCHED must be visible before we test polling */ | |
1196 | smp_mb(); | |
1197 | if (!tsk_is_polling(p)) | |
1198 | smp_send_reschedule(cpu); | |
1199 | } | |
1200 | ||
1201 | static void resched_cpu(int cpu) | |
1202 | { | |
1203 | struct rq *rq = cpu_rq(cpu); | |
1204 | unsigned long flags; | |
1205 | ||
05fa785c | 1206 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1207 | return; |
1208 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1209 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1210 | } |
06d8308c TG |
1211 | |
1212 | #ifdef CONFIG_NO_HZ | |
1213 | /* | |
1214 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1215 | * idle CPU then this timer might expire before the next timer event | |
1216 | * which is scheduled to wake up that CPU. In case of a completely | |
1217 | * idle system the next event might even be infinite time into the | |
1218 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1219 | * leaves the inner idle loop so the newly added timer is taken into | |
1220 | * account when the CPU goes back to idle and evaluates the timer | |
1221 | * wheel for the next timer event. | |
1222 | */ | |
1223 | void wake_up_idle_cpu(int cpu) | |
1224 | { | |
1225 | struct rq *rq = cpu_rq(cpu); | |
1226 | ||
1227 | if (cpu == smp_processor_id()) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * This is safe, as this function is called with the timer | |
1232 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1233 | * to idle and has not yet set rq->curr to idle then it will | |
1234 | * be serialized on the timer wheel base lock and take the new | |
1235 | * timer into account automatically. | |
1236 | */ | |
1237 | if (rq->curr != rq->idle) | |
1238 | return; | |
1239 | ||
1240 | /* | |
1241 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1242 | * lockless. The worst case is that the other CPU runs the | |
1243 | * idle task through an additional NOOP schedule() | |
1244 | */ | |
5ed0cec0 | 1245 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1246 | |
1247 | /* NEED_RESCHED must be visible before we test polling */ | |
1248 | smp_mb(); | |
1249 | if (!tsk_is_polling(rq->idle)) | |
1250 | smp_send_reschedule(cpu); | |
1251 | } | |
6d6bc0ad | 1252 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1253 | |
e9e9250b PZ |
1254 | static u64 sched_avg_period(void) |
1255 | { | |
1256 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1257 | } | |
1258 | ||
1259 | static void sched_avg_update(struct rq *rq) | |
1260 | { | |
1261 | s64 period = sched_avg_period(); | |
1262 | ||
1263 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1264 | rq->age_stamp += period; | |
1265 | rq->rt_avg /= 2; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1270 | { | |
1271 | rq->rt_avg += rt_delta; | |
1272 | sched_avg_update(rq); | |
1273 | } | |
1274 | ||
6d6bc0ad | 1275 | #else /* !CONFIG_SMP */ |
31656519 | 1276 | static void resched_task(struct task_struct *p) |
c24d20db | 1277 | { |
05fa785c | 1278 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1279 | set_tsk_need_resched(p); |
c24d20db | 1280 | } |
e9e9250b PZ |
1281 | |
1282 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1283 | { | |
1284 | } | |
6d6bc0ad | 1285 | #endif /* CONFIG_SMP */ |
c24d20db | 1286 | |
45bf76df IM |
1287 | #if BITS_PER_LONG == 32 |
1288 | # define WMULT_CONST (~0UL) | |
1289 | #else | |
1290 | # define WMULT_CONST (1UL << 32) | |
1291 | #endif | |
1292 | ||
1293 | #define WMULT_SHIFT 32 | |
1294 | ||
194081eb IM |
1295 | /* |
1296 | * Shift right and round: | |
1297 | */ | |
cf2ab469 | 1298 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1299 | |
a7be37ac PZ |
1300 | /* |
1301 | * delta *= weight / lw | |
1302 | */ | |
cb1c4fc9 | 1303 | static unsigned long |
45bf76df IM |
1304 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1305 | struct load_weight *lw) | |
1306 | { | |
1307 | u64 tmp; | |
1308 | ||
7a232e03 LJ |
1309 | if (!lw->inv_weight) { |
1310 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1311 | lw->inv_weight = 1; | |
1312 | else | |
1313 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1314 | / (lw->weight+1); | |
1315 | } | |
45bf76df IM |
1316 | |
1317 | tmp = (u64)delta_exec * weight; | |
1318 | /* | |
1319 | * Check whether we'd overflow the 64-bit multiplication: | |
1320 | */ | |
194081eb | 1321 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1322 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1323 | WMULT_SHIFT/2); |
1324 | else | |
cf2ab469 | 1325 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1326 | |
ecf691da | 1327 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1331 | { |
1332 | lw->weight += inc; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1337 | { |
1338 | lw->weight -= dec; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
2dd73a4f PW |
1342 | /* |
1343 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1344 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1345 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1346 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1347 | * scaled version of the new time slice allocation that they receive on time |
1348 | * slice expiry etc. | |
1349 | */ | |
1350 | ||
cce7ade8 PZ |
1351 | #define WEIGHT_IDLEPRIO 3 |
1352 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1353 | |
1354 | /* | |
1355 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1356 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1357 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1358 | * that remained on nice 0. | |
1359 | * | |
1360 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1361 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1362 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1363 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1364 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1365 | */ |
1366 | static const int prio_to_weight[40] = { | |
254753dc IM |
1367 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1368 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1369 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1370 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1371 | /* 0 */ 1024, 820, 655, 526, 423, | |
1372 | /* 5 */ 335, 272, 215, 172, 137, | |
1373 | /* 10 */ 110, 87, 70, 56, 45, | |
1374 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1375 | }; |
1376 | ||
5714d2de IM |
1377 | /* |
1378 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1379 | * | |
1380 | * In cases where the weight does not change often, we can use the | |
1381 | * precalculated inverse to speed up arithmetics by turning divisions | |
1382 | * into multiplications: | |
1383 | */ | |
dd41f596 | 1384 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1385 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1386 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1387 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1388 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1389 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1390 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1391 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1392 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1393 | }; |
2dd73a4f | 1394 | |
dd41f596 IM |
1395 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1396 | ||
1397 | /* | |
1398 | * runqueue iterator, to support SMP load-balancing between different | |
1399 | * scheduling classes, without having to expose their internal data | |
1400 | * structures to the load-balancing proper: | |
1401 | */ | |
1402 | struct rq_iterator { | |
1403 | void *arg; | |
1404 | struct task_struct *(*start)(void *); | |
1405 | struct task_struct *(*next)(void *); | |
1406 | }; | |
1407 | ||
e1d1484f PW |
1408 | #ifdef CONFIG_SMP |
1409 | static unsigned long | |
1410 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1411 | unsigned long max_load_move, struct sched_domain *sd, | |
1412 | enum cpu_idle_type idle, int *all_pinned, | |
1413 | int *this_best_prio, struct rq_iterator *iterator); | |
1414 | ||
1415 | static int | |
1416 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1417 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1418 | struct rq_iterator *iterator); | |
e1d1484f | 1419 | #endif |
dd41f596 | 1420 | |
ef12fefa BR |
1421 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1422 | enum cpuacct_stat_index { | |
1423 | CPUACCT_STAT_USER, /* ... user mode */ | |
1424 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1425 | ||
1426 | CPUACCT_STAT_NSTATS, | |
1427 | }; | |
1428 | ||
d842de87 SV |
1429 | #ifdef CONFIG_CGROUP_CPUACCT |
1430 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1431 | static void cpuacct_update_stats(struct task_struct *tsk, |
1432 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1433 | #else |
1434 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1435 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1437 | #endif |
1438 | ||
18d95a28 PZ |
1439 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1440 | { | |
1441 | update_load_add(&rq->load, load); | |
1442 | } | |
1443 | ||
1444 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1445 | { | |
1446 | update_load_sub(&rq->load, load); | |
1447 | } | |
1448 | ||
7940ca36 | 1449 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1450 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1451 | |
1452 | /* | |
1453 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1454 | * leaving it for the final time. | |
1455 | */ | |
eb755805 | 1456 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1457 | { |
1458 | struct task_group *parent, *child; | |
eb755805 | 1459 | int ret; |
c09595f6 PZ |
1460 | |
1461 | rcu_read_lock(); | |
1462 | parent = &root_task_group; | |
1463 | down: | |
eb755805 PZ |
1464 | ret = (*down)(parent, data); |
1465 | if (ret) | |
1466 | goto out_unlock; | |
c09595f6 PZ |
1467 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1468 | parent = child; | |
1469 | goto down; | |
1470 | ||
1471 | up: | |
1472 | continue; | |
1473 | } | |
eb755805 PZ |
1474 | ret = (*up)(parent, data); |
1475 | if (ret) | |
1476 | goto out_unlock; | |
c09595f6 PZ |
1477 | |
1478 | child = parent; | |
1479 | parent = parent->parent; | |
1480 | if (parent) | |
1481 | goto up; | |
eb755805 | 1482 | out_unlock: |
c09595f6 | 1483 | rcu_read_unlock(); |
eb755805 PZ |
1484 | |
1485 | return ret; | |
c09595f6 PZ |
1486 | } |
1487 | ||
eb755805 PZ |
1488 | static int tg_nop(struct task_group *tg, void *data) |
1489 | { | |
1490 | return 0; | |
c09595f6 | 1491 | } |
eb755805 PZ |
1492 | #endif |
1493 | ||
1494 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1495 | /* Used instead of source_load when we know the type == 0 */ |
1496 | static unsigned long weighted_cpuload(const int cpu) | |
1497 | { | |
1498 | return cpu_rq(cpu)->load.weight; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * Return a low guess at the load of a migration-source cpu weighted | |
1503 | * according to the scheduling class and "nice" value. | |
1504 | * | |
1505 | * We want to under-estimate the load of migration sources, to | |
1506 | * balance conservatively. | |
1507 | */ | |
1508 | static unsigned long source_load(int cpu, int type) | |
1509 | { | |
1510 | struct rq *rq = cpu_rq(cpu); | |
1511 | unsigned long total = weighted_cpuload(cpu); | |
1512 | ||
1513 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1514 | return total; | |
1515 | ||
1516 | return min(rq->cpu_load[type-1], total); | |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Return a high guess at the load of a migration-target cpu weighted | |
1521 | * according to the scheduling class and "nice" value. | |
1522 | */ | |
1523 | static unsigned long target_load(int cpu, int type) | |
1524 | { | |
1525 | struct rq *rq = cpu_rq(cpu); | |
1526 | unsigned long total = weighted_cpuload(cpu); | |
1527 | ||
1528 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1529 | return total; | |
1530 | ||
1531 | return max(rq->cpu_load[type-1], total); | |
1532 | } | |
1533 | ||
ae154be1 PZ |
1534 | static struct sched_group *group_of(int cpu) |
1535 | { | |
1536 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
1537 | ||
1538 | if (!sd) | |
1539 | return NULL; | |
1540 | ||
1541 | return sd->groups; | |
1542 | } | |
1543 | ||
1544 | static unsigned long power_of(int cpu) | |
1545 | { | |
1546 | struct sched_group *group = group_of(cpu); | |
1547 | ||
1548 | if (!group) | |
1549 | return SCHED_LOAD_SCALE; | |
1550 | ||
1551 | return group->cpu_power; | |
1552 | } | |
1553 | ||
eb755805 PZ |
1554 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1555 | ||
1556 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1557 | { | |
1558 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1559 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1560 | |
4cd42620 SR |
1561 | if (nr_running) |
1562 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1563 | else |
1564 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1565 | |
1566 | return rq->avg_load_per_task; | |
1567 | } | |
1568 | ||
1569 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1570 | |
4a6cc4bd | 1571 | static __read_mostly unsigned long *update_shares_data; |
34d76c41 | 1572 | |
c09595f6 PZ |
1573 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1574 | ||
1575 | /* | |
1576 | * Calculate and set the cpu's group shares. | |
1577 | */ | |
34d76c41 PZ |
1578 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1579 | unsigned long sd_shares, | |
1580 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1581 | unsigned long *usd_rq_weight) |
18d95a28 | 1582 | { |
34d76c41 | 1583 | unsigned long shares, rq_weight; |
a5004278 | 1584 | int boost = 0; |
c09595f6 | 1585 | |
4a6cc4bd | 1586 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1587 | if (!rq_weight) { |
1588 | boost = 1; | |
1589 | rq_weight = NICE_0_LOAD; | |
1590 | } | |
c8cba857 | 1591 | |
c09595f6 | 1592 | /* |
a8af7246 PZ |
1593 | * \Sum_j shares_j * rq_weight_i |
1594 | * shares_i = ----------------------------- | |
1595 | * \Sum_j rq_weight_j | |
c09595f6 | 1596 | */ |
ec4e0e2f | 1597 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1598 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1599 | |
ffda12a1 PZ |
1600 | if (abs(shares - tg->se[cpu]->load.weight) > |
1601 | sysctl_sched_shares_thresh) { | |
1602 | struct rq *rq = cpu_rq(cpu); | |
1603 | unsigned long flags; | |
c09595f6 | 1604 | |
05fa785c | 1605 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1606 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1607 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1608 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1609 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1610 | } |
18d95a28 | 1611 | } |
c09595f6 PZ |
1612 | |
1613 | /* | |
c8cba857 PZ |
1614 | * Re-compute the task group their per cpu shares over the given domain. |
1615 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1616 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1617 | */ |
eb755805 | 1618 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1619 | { |
cd8ad40d | 1620 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1621 | unsigned long *usd_rq_weight; |
eb755805 | 1622 | struct sched_domain *sd = data; |
34d76c41 | 1623 | unsigned long flags; |
c8cba857 | 1624 | int i; |
c09595f6 | 1625 | |
34d76c41 PZ |
1626 | if (!tg->se[0]) |
1627 | return 0; | |
1628 | ||
1629 | local_irq_save(flags); | |
4a6cc4bd | 1630 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1631 | |
758b2cdc | 1632 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1633 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1634 | usd_rq_weight[i] = weight; |
34d76c41 | 1635 | |
cd8ad40d | 1636 | rq_weight += weight; |
ec4e0e2f KC |
1637 | /* |
1638 | * If there are currently no tasks on the cpu pretend there | |
1639 | * is one of average load so that when a new task gets to | |
1640 | * run here it will not get delayed by group starvation. | |
1641 | */ | |
ec4e0e2f KC |
1642 | if (!weight) |
1643 | weight = NICE_0_LOAD; | |
1644 | ||
cd8ad40d | 1645 | sum_weight += weight; |
c8cba857 | 1646 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1647 | } |
c09595f6 | 1648 | |
cd8ad40d PZ |
1649 | if (!rq_weight) |
1650 | rq_weight = sum_weight; | |
1651 | ||
c8cba857 PZ |
1652 | if ((!shares && rq_weight) || shares > tg->shares) |
1653 | shares = tg->shares; | |
1654 | ||
1655 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1656 | shares = tg->shares; | |
c09595f6 | 1657 | |
758b2cdc | 1658 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1659 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1660 | |
1661 | local_irq_restore(flags); | |
eb755805 PZ |
1662 | |
1663 | return 0; | |
c09595f6 PZ |
1664 | } |
1665 | ||
1666 | /* | |
c8cba857 PZ |
1667 | * Compute the cpu's hierarchical load factor for each task group. |
1668 | * This needs to be done in a top-down fashion because the load of a child | |
1669 | * group is a fraction of its parents load. | |
c09595f6 | 1670 | */ |
eb755805 | 1671 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1672 | { |
c8cba857 | 1673 | unsigned long load; |
eb755805 | 1674 | long cpu = (long)data; |
c09595f6 | 1675 | |
c8cba857 PZ |
1676 | if (!tg->parent) { |
1677 | load = cpu_rq(cpu)->load.weight; | |
1678 | } else { | |
1679 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1680 | load *= tg->cfs_rq[cpu]->shares; | |
1681 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1682 | } | |
c09595f6 | 1683 | |
c8cba857 | 1684 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1685 | |
eb755805 | 1686 | return 0; |
c09595f6 PZ |
1687 | } |
1688 | ||
c8cba857 | 1689 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1690 | { |
e7097159 PZ |
1691 | s64 elapsed; |
1692 | u64 now; | |
1693 | ||
1694 | if (root_task_group_empty()) | |
1695 | return; | |
1696 | ||
1697 | now = cpu_clock(raw_smp_processor_id()); | |
1698 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1699 | |
1700 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1701 | sd->last_update = now; | |
eb755805 | 1702 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1703 | } |
4d8d595d PZ |
1704 | } |
1705 | ||
3e5459b4 PZ |
1706 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1707 | { | |
e7097159 PZ |
1708 | if (root_task_group_empty()) |
1709 | return; | |
1710 | ||
05fa785c | 1711 | raw_spin_unlock(&rq->lock); |
3e5459b4 | 1712 | update_shares(sd); |
05fa785c | 1713 | raw_spin_lock(&rq->lock); |
3e5459b4 PZ |
1714 | } |
1715 | ||
eb755805 | 1716 | static void update_h_load(long cpu) |
c09595f6 | 1717 | { |
e7097159 PZ |
1718 | if (root_task_group_empty()) |
1719 | return; | |
1720 | ||
eb755805 | 1721 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1722 | } |
1723 | ||
c09595f6 PZ |
1724 | #else |
1725 | ||
c8cba857 | 1726 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1727 | { |
1728 | } | |
1729 | ||
3e5459b4 PZ |
1730 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1731 | { | |
1732 | } | |
1733 | ||
18d95a28 PZ |
1734 | #endif |
1735 | ||
8f45e2b5 GH |
1736 | #ifdef CONFIG_PREEMPT |
1737 | ||
b78bb868 PZ |
1738 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1739 | ||
70574a99 | 1740 | /* |
8f45e2b5 GH |
1741 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1742 | * way at the expense of forcing extra atomic operations in all | |
1743 | * invocations. This assures that the double_lock is acquired using the | |
1744 | * same underlying policy as the spinlock_t on this architecture, which | |
1745 | * reduces latency compared to the unfair variant below. However, it | |
1746 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1747 | */ |
8f45e2b5 GH |
1748 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1749 | __releases(this_rq->lock) | |
1750 | __acquires(busiest->lock) | |
1751 | __acquires(this_rq->lock) | |
1752 | { | |
05fa785c | 1753 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1754 | double_rq_lock(this_rq, busiest); |
1755 | ||
1756 | return 1; | |
1757 | } | |
1758 | ||
1759 | #else | |
1760 | /* | |
1761 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1762 | * latency by eliminating extra atomic operations when the locks are | |
1763 | * already in proper order on entry. This favors lower cpu-ids and will | |
1764 | * grant the double lock to lower cpus over higher ids under contention, | |
1765 | * regardless of entry order into the function. | |
1766 | */ | |
1767 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1768 | __releases(this_rq->lock) |
1769 | __acquires(busiest->lock) | |
1770 | __acquires(this_rq->lock) | |
1771 | { | |
1772 | int ret = 0; | |
1773 | ||
05fa785c | 1774 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1775 | if (busiest < this_rq) { |
05fa785c TG |
1776 | raw_spin_unlock(&this_rq->lock); |
1777 | raw_spin_lock(&busiest->lock); | |
1778 | raw_spin_lock_nested(&this_rq->lock, | |
1779 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1780 | ret = 1; |
1781 | } else | |
05fa785c TG |
1782 | raw_spin_lock_nested(&busiest->lock, |
1783 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1784 | } |
1785 | return ret; | |
1786 | } | |
1787 | ||
8f45e2b5 GH |
1788 | #endif /* CONFIG_PREEMPT */ |
1789 | ||
1790 | /* | |
1791 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1792 | */ | |
1793 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1794 | { | |
1795 | if (unlikely(!irqs_disabled())) { | |
1796 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1797 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1798 | BUG_ON(1); |
1799 | } | |
1800 | ||
1801 | return _double_lock_balance(this_rq, busiest); | |
1802 | } | |
1803 | ||
70574a99 AD |
1804 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1805 | __releases(busiest->lock) | |
1806 | { | |
05fa785c | 1807 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1808 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1809 | } | |
18d95a28 PZ |
1810 | #endif |
1811 | ||
30432094 | 1812 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1813 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1814 | { | |
30432094 | 1815 | #ifdef CONFIG_SMP |
34e83e85 IM |
1816 | cfs_rq->shares = shares; |
1817 | #endif | |
1818 | } | |
30432094 | 1819 | #endif |
e7693a36 | 1820 | |
dce48a84 | 1821 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1822 | static void update_sysctl(void); |
acb4a848 | 1823 | static int get_update_sysctl_factor(void); |
dce48a84 | 1824 | |
cd29fe6f PZ |
1825 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1826 | { | |
1827 | set_task_rq(p, cpu); | |
1828 | #ifdef CONFIG_SMP | |
1829 | /* | |
1830 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1831 | * successfuly executed on another CPU. We must ensure that updates of | |
1832 | * per-task data have been completed by this moment. | |
1833 | */ | |
1834 | smp_wmb(); | |
1835 | task_thread_info(p)->cpu = cpu; | |
1836 | #endif | |
1837 | } | |
dce48a84 | 1838 | |
dd41f596 | 1839 | #include "sched_stats.h" |
dd41f596 | 1840 | #include "sched_idletask.c" |
5522d5d5 IM |
1841 | #include "sched_fair.c" |
1842 | #include "sched_rt.c" | |
dd41f596 IM |
1843 | #ifdef CONFIG_SCHED_DEBUG |
1844 | # include "sched_debug.c" | |
1845 | #endif | |
1846 | ||
1847 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1848 | #define for_each_class(class) \ |
1849 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1850 | |
c09595f6 | 1851 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1852 | { |
1853 | rq->nr_running++; | |
9c217245 IM |
1854 | } |
1855 | ||
c09595f6 | 1856 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1857 | { |
1858 | rq->nr_running--; | |
9c217245 IM |
1859 | } |
1860 | ||
45bf76df IM |
1861 | static void set_load_weight(struct task_struct *p) |
1862 | { | |
1863 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1864 | p->se.load.weight = prio_to_weight[0] * 2; |
1865 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1866 | return; | |
1867 | } | |
45bf76df | 1868 | |
dd41f596 IM |
1869 | /* |
1870 | * SCHED_IDLE tasks get minimal weight: | |
1871 | */ | |
1872 | if (p->policy == SCHED_IDLE) { | |
1873 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1874 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1875 | return; | |
1876 | } | |
71f8bd46 | 1877 | |
dd41f596 IM |
1878 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1879 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1880 | } |
1881 | ||
2087a1ad GH |
1882 | static void update_avg(u64 *avg, u64 sample) |
1883 | { | |
1884 | s64 diff = sample - *avg; | |
1885 | *avg += diff >> 3; | |
1886 | } | |
1887 | ||
8159f87e | 1888 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1889 | { |
831451ac PZ |
1890 | if (wakeup) |
1891 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1892 | ||
dd41f596 | 1893 | sched_info_queued(p); |
fd390f6a | 1894 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1895 | p->se.on_rq = 1; |
71f8bd46 IM |
1896 | } |
1897 | ||
69be72c1 | 1898 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1899 | { |
831451ac PZ |
1900 | if (sleep) { |
1901 | if (p->se.last_wakeup) { | |
1902 | update_avg(&p->se.avg_overlap, | |
1903 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1904 | p->se.last_wakeup = 0; | |
1905 | } else { | |
1906 | update_avg(&p->se.avg_wakeup, | |
1907 | sysctl_sched_wakeup_granularity); | |
1908 | } | |
2087a1ad GH |
1909 | } |
1910 | ||
46ac22ba | 1911 | sched_info_dequeued(p); |
f02231e5 | 1912 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1913 | p->se.on_rq = 0; |
71f8bd46 IM |
1914 | } |
1915 | ||
14531189 | 1916 | /* |
dd41f596 | 1917 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1918 | */ |
14531189 IM |
1919 | static inline int __normal_prio(struct task_struct *p) |
1920 | { | |
dd41f596 | 1921 | return p->static_prio; |
14531189 IM |
1922 | } |
1923 | ||
b29739f9 IM |
1924 | /* |
1925 | * Calculate the expected normal priority: i.e. priority | |
1926 | * without taking RT-inheritance into account. Might be | |
1927 | * boosted by interactivity modifiers. Changes upon fork, | |
1928 | * setprio syscalls, and whenever the interactivity | |
1929 | * estimator recalculates. | |
1930 | */ | |
36c8b586 | 1931 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1932 | { |
1933 | int prio; | |
1934 | ||
e05606d3 | 1935 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1936 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1937 | else | |
1938 | prio = __normal_prio(p); | |
1939 | return prio; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * Calculate the current priority, i.e. the priority | |
1944 | * taken into account by the scheduler. This value might | |
1945 | * be boosted by RT tasks, or might be boosted by | |
1946 | * interactivity modifiers. Will be RT if the task got | |
1947 | * RT-boosted. If not then it returns p->normal_prio. | |
1948 | */ | |
36c8b586 | 1949 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1950 | { |
1951 | p->normal_prio = normal_prio(p); | |
1952 | /* | |
1953 | * If we are RT tasks or we were boosted to RT priority, | |
1954 | * keep the priority unchanged. Otherwise, update priority | |
1955 | * to the normal priority: | |
1956 | */ | |
1957 | if (!rt_prio(p->prio)) | |
1958 | return p->normal_prio; | |
1959 | return p->prio; | |
1960 | } | |
1961 | ||
1da177e4 | 1962 | /* |
dd41f596 | 1963 | * activate_task - move a task to the runqueue. |
1da177e4 | 1964 | */ |
dd41f596 | 1965 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1966 | { |
d9514f6c | 1967 | if (task_contributes_to_load(p)) |
dd41f596 | 1968 | rq->nr_uninterruptible--; |
1da177e4 | 1969 | |
8159f87e | 1970 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1971 | inc_nr_running(rq); |
1da177e4 LT |
1972 | } |
1973 | ||
1da177e4 LT |
1974 | /* |
1975 | * deactivate_task - remove a task from the runqueue. | |
1976 | */ | |
2e1cb74a | 1977 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1978 | { |
d9514f6c | 1979 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1980 | rq->nr_uninterruptible++; |
1981 | ||
69be72c1 | 1982 | dequeue_task(rq, p, sleep); |
c09595f6 | 1983 | dec_nr_running(rq); |
1da177e4 LT |
1984 | } |
1985 | ||
1da177e4 LT |
1986 | /** |
1987 | * task_curr - is this task currently executing on a CPU? | |
1988 | * @p: the task in question. | |
1989 | */ | |
36c8b586 | 1990 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1991 | { |
1992 | return cpu_curr(task_cpu(p)) == p; | |
1993 | } | |
1994 | ||
cb469845 SR |
1995 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1996 | const struct sched_class *prev_class, | |
1997 | int oldprio, int running) | |
1998 | { | |
1999 | if (prev_class != p->sched_class) { | |
2000 | if (prev_class->switched_from) | |
2001 | prev_class->switched_from(rq, p, running); | |
2002 | p->sched_class->switched_to(rq, p, running); | |
2003 | } else | |
2004 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2005 | } | |
2006 | ||
1da177e4 | 2007 | #ifdef CONFIG_SMP |
cc367732 IM |
2008 | /* |
2009 | * Is this task likely cache-hot: | |
2010 | */ | |
e7693a36 | 2011 | static int |
cc367732 IM |
2012 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2013 | { | |
2014 | s64 delta; | |
2015 | ||
e6c8fba7 PZ |
2016 | if (p->sched_class != &fair_sched_class) |
2017 | return 0; | |
2018 | ||
f540a608 IM |
2019 | /* |
2020 | * Buddy candidates are cache hot: | |
2021 | */ | |
f685ceac | 2022 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2023 | (&p->se == cfs_rq_of(&p->se)->next || |
2024 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2025 | return 1; |
2026 | ||
6bc1665b IM |
2027 | if (sysctl_sched_migration_cost == -1) |
2028 | return 1; | |
2029 | if (sysctl_sched_migration_cost == 0) | |
2030 | return 0; | |
2031 | ||
cc367732 IM |
2032 | delta = now - p->se.exec_start; |
2033 | ||
2034 | return delta < (s64)sysctl_sched_migration_cost; | |
2035 | } | |
2036 | ||
dd41f596 | 2037 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2038 | { |
e2912009 PZ |
2039 | #ifdef CONFIG_SCHED_DEBUG |
2040 | /* | |
2041 | * We should never call set_task_cpu() on a blocked task, | |
2042 | * ttwu() will sort out the placement. | |
2043 | */ | |
416eb395 | 2044 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING); |
e2912009 PZ |
2045 | #endif |
2046 | ||
de1d7286 | 2047 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2048 | |
738d2be4 PZ |
2049 | if (task_cpu(p) == new_cpu) |
2050 | return; | |
2051 | ||
2052 | p->se.nr_migrations++; | |
2053 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
dd41f596 IM |
2054 | |
2055 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2056 | } |
2057 | ||
70b97a7f | 2058 | struct migration_req { |
1da177e4 | 2059 | struct list_head list; |
1da177e4 | 2060 | |
36c8b586 | 2061 | struct task_struct *task; |
1da177e4 LT |
2062 | int dest_cpu; |
2063 | ||
1da177e4 | 2064 | struct completion done; |
70b97a7f | 2065 | }; |
1da177e4 LT |
2066 | |
2067 | /* | |
2068 | * The task's runqueue lock must be held. | |
2069 | * Returns true if you have to wait for migration thread. | |
2070 | */ | |
36c8b586 | 2071 | static int |
70b97a7f | 2072 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2073 | { |
70b97a7f | 2074 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2075 | |
2076 | /* | |
2077 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2078 | * the next wake-up will properly place the task. |
1da177e4 | 2079 | */ |
e2912009 | 2080 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2081 | return 0; |
1da177e4 LT |
2082 | |
2083 | init_completion(&req->done); | |
1da177e4 LT |
2084 | req->task = p; |
2085 | req->dest_cpu = dest_cpu; | |
2086 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2087 | |
1da177e4 LT |
2088 | return 1; |
2089 | } | |
2090 | ||
a26b89f0 MM |
2091 | /* |
2092 | * wait_task_context_switch - wait for a thread to complete at least one | |
2093 | * context switch. | |
2094 | * | |
2095 | * @p must not be current. | |
2096 | */ | |
2097 | void wait_task_context_switch(struct task_struct *p) | |
2098 | { | |
2099 | unsigned long nvcsw, nivcsw, flags; | |
2100 | int running; | |
2101 | struct rq *rq; | |
2102 | ||
2103 | nvcsw = p->nvcsw; | |
2104 | nivcsw = p->nivcsw; | |
2105 | for (;;) { | |
2106 | /* | |
2107 | * The runqueue is assigned before the actual context | |
2108 | * switch. We need to take the runqueue lock. | |
2109 | * | |
2110 | * We could check initially without the lock but it is | |
2111 | * very likely that we need to take the lock in every | |
2112 | * iteration. | |
2113 | */ | |
2114 | rq = task_rq_lock(p, &flags); | |
2115 | running = task_running(rq, p); | |
2116 | task_rq_unlock(rq, &flags); | |
2117 | ||
2118 | if (likely(!running)) | |
2119 | break; | |
2120 | /* | |
2121 | * The switch count is incremented before the actual | |
2122 | * context switch. We thus wait for two switches to be | |
2123 | * sure at least one completed. | |
2124 | */ | |
2125 | if ((p->nvcsw - nvcsw) > 1) | |
2126 | break; | |
2127 | if ((p->nivcsw - nivcsw) > 1) | |
2128 | break; | |
2129 | ||
2130 | cpu_relax(); | |
2131 | } | |
2132 | } | |
2133 | ||
1da177e4 LT |
2134 | /* |
2135 | * wait_task_inactive - wait for a thread to unschedule. | |
2136 | * | |
85ba2d86 RM |
2137 | * If @match_state is nonzero, it's the @p->state value just checked and |
2138 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2139 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2140 | * we return a positive number (its total switch count). If a second call | |
2141 | * a short while later returns the same number, the caller can be sure that | |
2142 | * @p has remained unscheduled the whole time. | |
2143 | * | |
1da177e4 LT |
2144 | * The caller must ensure that the task *will* unschedule sometime soon, |
2145 | * else this function might spin for a *long* time. This function can't | |
2146 | * be called with interrupts off, or it may introduce deadlock with | |
2147 | * smp_call_function() if an IPI is sent by the same process we are | |
2148 | * waiting to become inactive. | |
2149 | */ | |
85ba2d86 | 2150 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2151 | { |
2152 | unsigned long flags; | |
dd41f596 | 2153 | int running, on_rq; |
85ba2d86 | 2154 | unsigned long ncsw; |
70b97a7f | 2155 | struct rq *rq; |
1da177e4 | 2156 | |
3a5c359a AK |
2157 | for (;;) { |
2158 | /* | |
2159 | * We do the initial early heuristics without holding | |
2160 | * any task-queue locks at all. We'll only try to get | |
2161 | * the runqueue lock when things look like they will | |
2162 | * work out! | |
2163 | */ | |
2164 | rq = task_rq(p); | |
fa490cfd | 2165 | |
3a5c359a AK |
2166 | /* |
2167 | * If the task is actively running on another CPU | |
2168 | * still, just relax and busy-wait without holding | |
2169 | * any locks. | |
2170 | * | |
2171 | * NOTE! Since we don't hold any locks, it's not | |
2172 | * even sure that "rq" stays as the right runqueue! | |
2173 | * But we don't care, since "task_running()" will | |
2174 | * return false if the runqueue has changed and p | |
2175 | * is actually now running somewhere else! | |
2176 | */ | |
85ba2d86 RM |
2177 | while (task_running(rq, p)) { |
2178 | if (match_state && unlikely(p->state != match_state)) | |
2179 | return 0; | |
3a5c359a | 2180 | cpu_relax(); |
85ba2d86 | 2181 | } |
fa490cfd | 2182 | |
3a5c359a AK |
2183 | /* |
2184 | * Ok, time to look more closely! We need the rq | |
2185 | * lock now, to be *sure*. If we're wrong, we'll | |
2186 | * just go back and repeat. | |
2187 | */ | |
2188 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2189 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2190 | running = task_running(rq, p); |
2191 | on_rq = p->se.on_rq; | |
85ba2d86 | 2192 | ncsw = 0; |
f31e11d8 | 2193 | if (!match_state || p->state == match_state) |
93dcf55f | 2194 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2195 | task_rq_unlock(rq, &flags); |
fa490cfd | 2196 | |
85ba2d86 RM |
2197 | /* |
2198 | * If it changed from the expected state, bail out now. | |
2199 | */ | |
2200 | if (unlikely(!ncsw)) | |
2201 | break; | |
2202 | ||
3a5c359a AK |
2203 | /* |
2204 | * Was it really running after all now that we | |
2205 | * checked with the proper locks actually held? | |
2206 | * | |
2207 | * Oops. Go back and try again.. | |
2208 | */ | |
2209 | if (unlikely(running)) { | |
2210 | cpu_relax(); | |
2211 | continue; | |
2212 | } | |
fa490cfd | 2213 | |
3a5c359a AK |
2214 | /* |
2215 | * It's not enough that it's not actively running, | |
2216 | * it must be off the runqueue _entirely_, and not | |
2217 | * preempted! | |
2218 | * | |
80dd99b3 | 2219 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2220 | * running right now), it's preempted, and we should |
2221 | * yield - it could be a while. | |
2222 | */ | |
2223 | if (unlikely(on_rq)) { | |
2224 | schedule_timeout_uninterruptible(1); | |
2225 | continue; | |
2226 | } | |
fa490cfd | 2227 | |
3a5c359a AK |
2228 | /* |
2229 | * Ahh, all good. It wasn't running, and it wasn't | |
2230 | * runnable, which means that it will never become | |
2231 | * running in the future either. We're all done! | |
2232 | */ | |
2233 | break; | |
2234 | } | |
85ba2d86 RM |
2235 | |
2236 | return ncsw; | |
1da177e4 LT |
2237 | } |
2238 | ||
2239 | /*** | |
2240 | * kick_process - kick a running thread to enter/exit the kernel | |
2241 | * @p: the to-be-kicked thread | |
2242 | * | |
2243 | * Cause a process which is running on another CPU to enter | |
2244 | * kernel-mode, without any delay. (to get signals handled.) | |
2245 | * | |
2246 | * NOTE: this function doesnt have to take the runqueue lock, | |
2247 | * because all it wants to ensure is that the remote task enters | |
2248 | * the kernel. If the IPI races and the task has been migrated | |
2249 | * to another CPU then no harm is done and the purpose has been | |
2250 | * achieved as well. | |
2251 | */ | |
36c8b586 | 2252 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2253 | { |
2254 | int cpu; | |
2255 | ||
2256 | preempt_disable(); | |
2257 | cpu = task_cpu(p); | |
2258 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2259 | smp_send_reschedule(cpu); | |
2260 | preempt_enable(); | |
2261 | } | |
b43e3521 | 2262 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2263 | #endif /* CONFIG_SMP */ |
1da177e4 | 2264 | |
0793a61d TG |
2265 | /** |
2266 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2267 | * @p: the task to evaluate | |
2268 | * @func: the function to be called | |
2269 | * @info: the function call argument | |
2270 | * | |
2271 | * Calls the function @func when the task is currently running. This might | |
2272 | * be on the current CPU, which just calls the function directly | |
2273 | */ | |
2274 | void task_oncpu_function_call(struct task_struct *p, | |
2275 | void (*func) (void *info), void *info) | |
2276 | { | |
2277 | int cpu; | |
2278 | ||
2279 | preempt_disable(); | |
2280 | cpu = task_cpu(p); | |
2281 | if (task_curr(p)) | |
2282 | smp_call_function_single(cpu, func, info, 1); | |
2283 | preempt_enable(); | |
2284 | } | |
2285 | ||
970b13ba | 2286 | #ifdef CONFIG_SMP |
5da9a0fb PZ |
2287 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2288 | { | |
2289 | int dest_cpu; | |
2290 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2291 | ||
2292 | /* Look for allowed, online CPU in same node. */ | |
2293 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2294 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2295 | return dest_cpu; | |
2296 | ||
2297 | /* Any allowed, online CPU? */ | |
2298 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2299 | if (dest_cpu < nr_cpu_ids) | |
2300 | return dest_cpu; | |
2301 | ||
2302 | /* No more Mr. Nice Guy. */ | |
2303 | if (dest_cpu >= nr_cpu_ids) { | |
2304 | rcu_read_lock(); | |
2305 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); | |
2306 | rcu_read_unlock(); | |
2307 | dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed); | |
2308 | ||
2309 | /* | |
2310 | * Don't tell them about moving exiting tasks or | |
2311 | * kernel threads (both mm NULL), since they never | |
2312 | * leave kernel. | |
2313 | */ | |
2314 | if (p->mm && printk_ratelimit()) { | |
2315 | printk(KERN_INFO "process %d (%s) no " | |
2316 | "longer affine to cpu%d\n", | |
2317 | task_pid_nr(p), p->comm, cpu); | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | return dest_cpu; | |
2322 | } | |
2323 | ||
e2912009 PZ |
2324 | /* |
2325 | * Called from: | |
2326 | * | |
2327 | * - fork, @p is stable because it isn't on the tasklist yet | |
2328 | * | |
38022906 | 2329 | * - exec, @p is unstable, retry loop |
e2912009 PZ |
2330 | * |
2331 | * - wake-up, we serialize ->cpus_allowed against TASK_WAKING so | |
2332 | * we should be good. | |
2333 | */ | |
970b13ba PZ |
2334 | static inline |
2335 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2336 | { | |
e2912009 PZ |
2337 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
2338 | ||
2339 | /* | |
2340 | * In order not to call set_task_cpu() on a blocking task we need | |
2341 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2342 | * cpu. | |
2343 | * | |
2344 | * Since this is common to all placement strategies, this lives here. | |
2345 | * | |
2346 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2347 | * not worry about this generic constraint ] | |
2348 | */ | |
2349 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
5da9a0fb PZ |
2350 | !cpu_active(cpu))) |
2351 | cpu = select_fallback_rq(task_cpu(p), p); | |
e2912009 PZ |
2352 | |
2353 | return cpu; | |
970b13ba PZ |
2354 | } |
2355 | #endif | |
2356 | ||
1da177e4 LT |
2357 | /*** |
2358 | * try_to_wake_up - wake up a thread | |
2359 | * @p: the to-be-woken-up thread | |
2360 | * @state: the mask of task states that can be woken | |
2361 | * @sync: do a synchronous wakeup? | |
2362 | * | |
2363 | * Put it on the run-queue if it's not already there. The "current" | |
2364 | * thread is always on the run-queue (except when the actual | |
2365 | * re-schedule is in progress), and as such you're allowed to do | |
2366 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2367 | * runnable without the overhead of this. | |
2368 | * | |
2369 | * returns failure only if the task is already active. | |
2370 | */ | |
7d478721 PZ |
2371 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2372 | int wake_flags) | |
1da177e4 | 2373 | { |
cc367732 | 2374 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2375 | unsigned long flags; |
f5dc3753 | 2376 | struct rq *rq, *orig_rq; |
1da177e4 | 2377 | |
b85d0667 | 2378 | if (!sched_feat(SYNC_WAKEUPS)) |
7d478721 | 2379 | wake_flags &= ~WF_SYNC; |
2398f2c6 | 2380 | |
e9c84311 | 2381 | this_cpu = get_cpu(); |
2398f2c6 | 2382 | |
04e2f174 | 2383 | smp_wmb(); |
f5dc3753 | 2384 | rq = orig_rq = task_rq_lock(p, &flags); |
03e89e45 | 2385 | update_rq_clock(rq); |
e9c84311 | 2386 | if (!(p->state & state)) |
1da177e4 LT |
2387 | goto out; |
2388 | ||
dd41f596 | 2389 | if (p->se.on_rq) |
1da177e4 LT |
2390 | goto out_running; |
2391 | ||
2392 | cpu = task_cpu(p); | |
cc367732 | 2393 | orig_cpu = cpu; |
1da177e4 LT |
2394 | |
2395 | #ifdef CONFIG_SMP | |
2396 | if (unlikely(task_running(rq, p))) | |
2397 | goto out_activate; | |
2398 | ||
e9c84311 PZ |
2399 | /* |
2400 | * In order to handle concurrent wakeups and release the rq->lock | |
2401 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2402 | * |
2403 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2404 | */ |
eb24073b IM |
2405 | if (task_contributes_to_load(p)) |
2406 | rq->nr_uninterruptible--; | |
e9c84311 | 2407 | p->state = TASK_WAKING; |
efbbd05a PZ |
2408 | |
2409 | if (p->sched_class->task_waking) | |
2410 | p->sched_class->task_waking(rq, p); | |
2411 | ||
ab19cb23 | 2412 | __task_rq_unlock(rq); |
e9c84311 | 2413 | |
970b13ba | 2414 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
ab19cb23 | 2415 | if (cpu != orig_cpu) |
5d2f5a61 | 2416 | set_task_cpu(p, cpu); |
ab19cb23 PZ |
2417 | |
2418 | rq = __task_rq_lock(p); | |
2419 | update_rq_clock(rq); | |
f5dc3753 | 2420 | |
e9c84311 PZ |
2421 | WARN_ON(p->state != TASK_WAKING); |
2422 | cpu = task_cpu(p); | |
1da177e4 | 2423 | |
e7693a36 GH |
2424 | #ifdef CONFIG_SCHEDSTATS |
2425 | schedstat_inc(rq, ttwu_count); | |
2426 | if (cpu == this_cpu) | |
2427 | schedstat_inc(rq, ttwu_local); | |
2428 | else { | |
2429 | struct sched_domain *sd; | |
2430 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2431 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2432 | schedstat_inc(sd, ttwu_wake_remote); |
2433 | break; | |
2434 | } | |
2435 | } | |
2436 | } | |
6d6bc0ad | 2437 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2438 | |
1da177e4 LT |
2439 | out_activate: |
2440 | #endif /* CONFIG_SMP */ | |
cc367732 | 2441 | schedstat_inc(p, se.nr_wakeups); |
7d478721 | 2442 | if (wake_flags & WF_SYNC) |
cc367732 IM |
2443 | schedstat_inc(p, se.nr_wakeups_sync); |
2444 | if (orig_cpu != cpu) | |
2445 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2446 | if (cpu == this_cpu) | |
2447 | schedstat_inc(p, se.nr_wakeups_local); | |
2448 | else | |
2449 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2450 | activate_task(rq, p, 1); |
1da177e4 LT |
2451 | success = 1; |
2452 | ||
831451ac PZ |
2453 | /* |
2454 | * Only attribute actual wakeups done by this task. | |
2455 | */ | |
2456 | if (!in_interrupt()) { | |
2457 | struct sched_entity *se = ¤t->se; | |
2458 | u64 sample = se->sum_exec_runtime; | |
2459 | ||
2460 | if (se->last_wakeup) | |
2461 | sample -= se->last_wakeup; | |
2462 | else | |
2463 | sample -= se->start_runtime; | |
2464 | update_avg(&se->avg_wakeup, sample); | |
2465 | ||
2466 | se->last_wakeup = se->sum_exec_runtime; | |
2467 | } | |
2468 | ||
1da177e4 | 2469 | out_running: |
468a15bb | 2470 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2471 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2472 | |
1da177e4 | 2473 | p->state = TASK_RUNNING; |
9a897c5a | 2474 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2475 | if (p->sched_class->task_woken) |
2476 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2477 | |
2478 | if (unlikely(rq->idle_stamp)) { | |
2479 | u64 delta = rq->clock - rq->idle_stamp; | |
2480 | u64 max = 2*sysctl_sched_migration_cost; | |
2481 | ||
2482 | if (delta > max) | |
2483 | rq->avg_idle = max; | |
2484 | else | |
2485 | update_avg(&rq->avg_idle, delta); | |
2486 | rq->idle_stamp = 0; | |
2487 | } | |
9a897c5a | 2488 | #endif |
1da177e4 LT |
2489 | out: |
2490 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2491 | put_cpu(); |
1da177e4 LT |
2492 | |
2493 | return success; | |
2494 | } | |
2495 | ||
50fa610a DH |
2496 | /** |
2497 | * wake_up_process - Wake up a specific process | |
2498 | * @p: The process to be woken up. | |
2499 | * | |
2500 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2501 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2502 | * running. | |
2503 | * | |
2504 | * It may be assumed that this function implies a write memory barrier before | |
2505 | * changing the task state if and only if any tasks are woken up. | |
2506 | */ | |
7ad5b3a5 | 2507 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2508 | { |
d9514f6c | 2509 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2510 | } |
1da177e4 LT |
2511 | EXPORT_SYMBOL(wake_up_process); |
2512 | ||
7ad5b3a5 | 2513 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2514 | { |
2515 | return try_to_wake_up(p, state, 0); | |
2516 | } | |
2517 | ||
1da177e4 LT |
2518 | /* |
2519 | * Perform scheduler related setup for a newly forked process p. | |
2520 | * p is forked by current. | |
dd41f596 IM |
2521 | * |
2522 | * __sched_fork() is basic setup used by init_idle() too: | |
2523 | */ | |
2524 | static void __sched_fork(struct task_struct *p) | |
2525 | { | |
dd41f596 IM |
2526 | p->se.exec_start = 0; |
2527 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2528 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2529 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2530 | p->se.last_wakeup = 0; |
2531 | p->se.avg_overlap = 0; | |
831451ac PZ |
2532 | p->se.start_runtime = 0; |
2533 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2534 | |
2535 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2536 | p->se.wait_start = 0; |
2537 | p->se.wait_max = 0; | |
2538 | p->se.wait_count = 0; | |
2539 | p->se.wait_sum = 0; | |
2540 | ||
2541 | p->se.sleep_start = 0; | |
2542 | p->se.sleep_max = 0; | |
2543 | p->se.sum_sleep_runtime = 0; | |
2544 | ||
2545 | p->se.block_start = 0; | |
2546 | p->se.block_max = 0; | |
2547 | p->se.exec_max = 0; | |
2548 | p->se.slice_max = 0; | |
2549 | ||
2550 | p->se.nr_migrations_cold = 0; | |
2551 | p->se.nr_failed_migrations_affine = 0; | |
2552 | p->se.nr_failed_migrations_running = 0; | |
2553 | p->se.nr_failed_migrations_hot = 0; | |
2554 | p->se.nr_forced_migrations = 0; | |
7793527b LDM |
2555 | |
2556 | p->se.nr_wakeups = 0; | |
2557 | p->se.nr_wakeups_sync = 0; | |
2558 | p->se.nr_wakeups_migrate = 0; | |
2559 | p->se.nr_wakeups_local = 0; | |
2560 | p->se.nr_wakeups_remote = 0; | |
2561 | p->se.nr_wakeups_affine = 0; | |
2562 | p->se.nr_wakeups_affine_attempts = 0; | |
2563 | p->se.nr_wakeups_passive = 0; | |
2564 | p->se.nr_wakeups_idle = 0; | |
2565 | ||
6cfb0d5d | 2566 | #endif |
476d139c | 2567 | |
fa717060 | 2568 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2569 | p->se.on_rq = 0; |
4a55bd5e | 2570 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2571 | |
e107be36 AK |
2572 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2573 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2574 | #endif | |
dd41f596 IM |
2575 | } |
2576 | ||
2577 | /* | |
2578 | * fork()/clone()-time setup: | |
2579 | */ | |
2580 | void sched_fork(struct task_struct *p, int clone_flags) | |
2581 | { | |
2582 | int cpu = get_cpu(); | |
2583 | ||
2584 | __sched_fork(p); | |
06b83b5f PZ |
2585 | /* |
2586 | * We mark the process as waking here. This guarantees that | |
2587 | * nobody will actually run it, and a signal or other external | |
2588 | * event cannot wake it up and insert it on the runqueue either. | |
2589 | */ | |
2590 | p->state = TASK_WAKING; | |
dd41f596 | 2591 | |
b9dc29e7 MG |
2592 | /* |
2593 | * Revert to default priority/policy on fork if requested. | |
2594 | */ | |
2595 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2596 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2597 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2598 | p->normal_prio = p->static_prio; |
2599 | } | |
b9dc29e7 | 2600 | |
6c697bdf MG |
2601 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2602 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2603 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2604 | set_load_weight(p); |
2605 | } | |
2606 | ||
b9dc29e7 MG |
2607 | /* |
2608 | * We don't need the reset flag anymore after the fork. It has | |
2609 | * fulfilled its duty: | |
2610 | */ | |
2611 | p->sched_reset_on_fork = 0; | |
2612 | } | |
ca94c442 | 2613 | |
f83f9ac2 PW |
2614 | /* |
2615 | * Make sure we do not leak PI boosting priority to the child. | |
2616 | */ | |
2617 | p->prio = current->normal_prio; | |
2618 | ||
2ddbf952 HS |
2619 | if (!rt_prio(p->prio)) |
2620 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2621 | |
cd29fe6f PZ |
2622 | if (p->sched_class->task_fork) |
2623 | p->sched_class->task_fork(p); | |
2624 | ||
5f3edc1b | 2625 | #ifdef CONFIG_SMP |
970b13ba | 2626 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); |
5f3edc1b PZ |
2627 | #endif |
2628 | set_task_cpu(p, cpu); | |
2629 | ||
52f17b6c | 2630 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2631 | if (likely(sched_info_on())) |
52f17b6c | 2632 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2633 | #endif |
d6077cb8 | 2634 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2635 | p->oncpu = 0; |
2636 | #endif | |
1da177e4 | 2637 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2638 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2639 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2640 | #endif |
917b627d GH |
2641 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2642 | ||
476d139c | 2643 | put_cpu(); |
1da177e4 LT |
2644 | } |
2645 | ||
2646 | /* | |
2647 | * wake_up_new_task - wake up a newly created task for the first time. | |
2648 | * | |
2649 | * This function will do some initial scheduler statistics housekeeping | |
2650 | * that must be done for every newly created context, then puts the task | |
2651 | * on the runqueue and wakes it. | |
2652 | */ | |
7ad5b3a5 | 2653 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2654 | { |
2655 | unsigned long flags; | |
dd41f596 | 2656 | struct rq *rq; |
1da177e4 LT |
2657 | |
2658 | rq = task_rq_lock(p, &flags); | |
06b83b5f PZ |
2659 | BUG_ON(p->state != TASK_WAKING); |
2660 | p->state = TASK_RUNNING; | |
a8e504d2 | 2661 | update_rq_clock(rq); |
cd29fe6f | 2662 | activate_task(rq, p, 0); |
c71dd42d | 2663 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2664 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2665 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2666 | if (p->sched_class->task_woken) |
2667 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2668 | #endif |
dd41f596 | 2669 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2670 | } |
2671 | ||
e107be36 AK |
2672 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2673 | ||
2674 | /** | |
80dd99b3 | 2675 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2676 | * @notifier: notifier struct to register |
e107be36 AK |
2677 | */ |
2678 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2679 | { | |
2680 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2681 | } | |
2682 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2683 | ||
2684 | /** | |
2685 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2686 | * @notifier: notifier struct to unregister |
e107be36 AK |
2687 | * |
2688 | * This is safe to call from within a preemption notifier. | |
2689 | */ | |
2690 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2691 | { | |
2692 | hlist_del(¬ifier->link); | |
2693 | } | |
2694 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2695 | ||
2696 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2697 | { | |
2698 | struct preempt_notifier *notifier; | |
2699 | struct hlist_node *node; | |
2700 | ||
2701 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2702 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2703 | } | |
2704 | ||
2705 | static void | |
2706 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2707 | struct task_struct *next) | |
2708 | { | |
2709 | struct preempt_notifier *notifier; | |
2710 | struct hlist_node *node; | |
2711 | ||
2712 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2713 | notifier->ops->sched_out(notifier, next); | |
2714 | } | |
2715 | ||
6d6bc0ad | 2716 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2717 | |
2718 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2719 | { | |
2720 | } | |
2721 | ||
2722 | static void | |
2723 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2724 | struct task_struct *next) | |
2725 | { | |
2726 | } | |
2727 | ||
6d6bc0ad | 2728 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2729 | |
4866cde0 NP |
2730 | /** |
2731 | * prepare_task_switch - prepare to switch tasks | |
2732 | * @rq: the runqueue preparing to switch | |
421cee29 | 2733 | * @prev: the current task that is being switched out |
4866cde0 NP |
2734 | * @next: the task we are going to switch to. |
2735 | * | |
2736 | * This is called with the rq lock held and interrupts off. It must | |
2737 | * be paired with a subsequent finish_task_switch after the context | |
2738 | * switch. | |
2739 | * | |
2740 | * prepare_task_switch sets up locking and calls architecture specific | |
2741 | * hooks. | |
2742 | */ | |
e107be36 AK |
2743 | static inline void |
2744 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2745 | struct task_struct *next) | |
4866cde0 | 2746 | { |
e107be36 | 2747 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2748 | prepare_lock_switch(rq, next); |
2749 | prepare_arch_switch(next); | |
2750 | } | |
2751 | ||
1da177e4 LT |
2752 | /** |
2753 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2754 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2755 | * @prev: the thread we just switched away from. |
2756 | * | |
4866cde0 NP |
2757 | * finish_task_switch must be called after the context switch, paired |
2758 | * with a prepare_task_switch call before the context switch. | |
2759 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2760 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2761 | * |
2762 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2763 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2764 | * with the lock held can cause deadlocks; see schedule() for |
2765 | * details.) | |
2766 | */ | |
a9957449 | 2767 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2768 | __releases(rq->lock) |
2769 | { | |
1da177e4 | 2770 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2771 | long prev_state; |
1da177e4 LT |
2772 | |
2773 | rq->prev_mm = NULL; | |
2774 | ||
2775 | /* | |
2776 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2777 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2778 | * schedule one last time. The schedule call will never return, and |
2779 | * the scheduled task must drop that reference. | |
c394cc9f | 2780 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2781 | * still held, otherwise prev could be scheduled on another cpu, die |
2782 | * there before we look at prev->state, and then the reference would | |
2783 | * be dropped twice. | |
2784 | * Manfred Spraul <manfred@colorfullife.com> | |
2785 | */ | |
55a101f8 | 2786 | prev_state = prev->state; |
4866cde0 | 2787 | finish_arch_switch(prev); |
cdd6c482 | 2788 | perf_event_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2789 | finish_lock_switch(rq, prev); |
e8fa1362 | 2790 | |
e107be36 | 2791 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2792 | if (mm) |
2793 | mmdrop(mm); | |
c394cc9f | 2794 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2795 | /* |
2796 | * Remove function-return probe instances associated with this | |
2797 | * task and put them back on the free list. | |
9761eea8 | 2798 | */ |
c6fd91f0 | 2799 | kprobe_flush_task(prev); |
1da177e4 | 2800 | put_task_struct(prev); |
c6fd91f0 | 2801 | } |
1da177e4 LT |
2802 | } |
2803 | ||
3f029d3c GH |
2804 | #ifdef CONFIG_SMP |
2805 | ||
2806 | /* assumes rq->lock is held */ | |
2807 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2808 | { | |
2809 | if (prev->sched_class->pre_schedule) | |
2810 | prev->sched_class->pre_schedule(rq, prev); | |
2811 | } | |
2812 | ||
2813 | /* rq->lock is NOT held, but preemption is disabled */ | |
2814 | static inline void post_schedule(struct rq *rq) | |
2815 | { | |
2816 | if (rq->post_schedule) { | |
2817 | unsigned long flags; | |
2818 | ||
05fa785c | 2819 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2820 | if (rq->curr->sched_class->post_schedule) |
2821 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2822 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2823 | |
2824 | rq->post_schedule = 0; | |
2825 | } | |
2826 | } | |
2827 | ||
2828 | #else | |
da19ab51 | 2829 | |
3f029d3c GH |
2830 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2831 | { | |
2832 | } | |
2833 | ||
2834 | static inline void post_schedule(struct rq *rq) | |
2835 | { | |
1da177e4 LT |
2836 | } |
2837 | ||
3f029d3c GH |
2838 | #endif |
2839 | ||
1da177e4 LT |
2840 | /** |
2841 | * schedule_tail - first thing a freshly forked thread must call. | |
2842 | * @prev: the thread we just switched away from. | |
2843 | */ | |
36c8b586 | 2844 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2845 | __releases(rq->lock) |
2846 | { | |
70b97a7f IM |
2847 | struct rq *rq = this_rq(); |
2848 | ||
4866cde0 | 2849 | finish_task_switch(rq, prev); |
da19ab51 | 2850 | |
3f029d3c GH |
2851 | /* |
2852 | * FIXME: do we need to worry about rq being invalidated by the | |
2853 | * task_switch? | |
2854 | */ | |
2855 | post_schedule(rq); | |
70b97a7f | 2856 | |
4866cde0 NP |
2857 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2858 | /* In this case, finish_task_switch does not reenable preemption */ | |
2859 | preempt_enable(); | |
2860 | #endif | |
1da177e4 | 2861 | if (current->set_child_tid) |
b488893a | 2862 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /* | |
2866 | * context_switch - switch to the new MM and the new | |
2867 | * thread's register state. | |
2868 | */ | |
dd41f596 | 2869 | static inline void |
70b97a7f | 2870 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2871 | struct task_struct *next) |
1da177e4 | 2872 | { |
dd41f596 | 2873 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2874 | |
e107be36 | 2875 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2876 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2877 | mm = next->mm; |
2878 | oldmm = prev->active_mm; | |
9226d125 ZA |
2879 | /* |
2880 | * For paravirt, this is coupled with an exit in switch_to to | |
2881 | * combine the page table reload and the switch backend into | |
2882 | * one hypercall. | |
2883 | */ | |
224101ed | 2884 | arch_start_context_switch(prev); |
9226d125 | 2885 | |
710390d9 | 2886 | if (likely(!mm)) { |
1da177e4 LT |
2887 | next->active_mm = oldmm; |
2888 | atomic_inc(&oldmm->mm_count); | |
2889 | enter_lazy_tlb(oldmm, next); | |
2890 | } else | |
2891 | switch_mm(oldmm, mm, next); | |
2892 | ||
710390d9 | 2893 | if (likely(!prev->mm)) { |
1da177e4 | 2894 | prev->active_mm = NULL; |
1da177e4 LT |
2895 | rq->prev_mm = oldmm; |
2896 | } | |
3a5f5e48 IM |
2897 | /* |
2898 | * Since the runqueue lock will be released by the next | |
2899 | * task (which is an invalid locking op but in the case | |
2900 | * of the scheduler it's an obvious special-case), so we | |
2901 | * do an early lockdep release here: | |
2902 | */ | |
2903 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2904 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2905 | #endif |
1da177e4 LT |
2906 | |
2907 | /* Here we just switch the register state and the stack. */ | |
2908 | switch_to(prev, next, prev); | |
2909 | ||
dd41f596 IM |
2910 | barrier(); |
2911 | /* | |
2912 | * this_rq must be evaluated again because prev may have moved | |
2913 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2914 | * frame will be invalid. | |
2915 | */ | |
2916 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2917 | } |
2918 | ||
2919 | /* | |
2920 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2921 | * | |
2922 | * externally visible scheduler statistics: current number of runnable | |
2923 | * threads, current number of uninterruptible-sleeping threads, total | |
2924 | * number of context switches performed since bootup. | |
2925 | */ | |
2926 | unsigned long nr_running(void) | |
2927 | { | |
2928 | unsigned long i, sum = 0; | |
2929 | ||
2930 | for_each_online_cpu(i) | |
2931 | sum += cpu_rq(i)->nr_running; | |
2932 | ||
2933 | return sum; | |
2934 | } | |
2935 | ||
2936 | unsigned long nr_uninterruptible(void) | |
2937 | { | |
2938 | unsigned long i, sum = 0; | |
2939 | ||
0a945022 | 2940 | for_each_possible_cpu(i) |
1da177e4 LT |
2941 | sum += cpu_rq(i)->nr_uninterruptible; |
2942 | ||
2943 | /* | |
2944 | * Since we read the counters lockless, it might be slightly | |
2945 | * inaccurate. Do not allow it to go below zero though: | |
2946 | */ | |
2947 | if (unlikely((long)sum < 0)) | |
2948 | sum = 0; | |
2949 | ||
2950 | return sum; | |
2951 | } | |
2952 | ||
2953 | unsigned long long nr_context_switches(void) | |
2954 | { | |
cc94abfc SR |
2955 | int i; |
2956 | unsigned long long sum = 0; | |
1da177e4 | 2957 | |
0a945022 | 2958 | for_each_possible_cpu(i) |
1da177e4 LT |
2959 | sum += cpu_rq(i)->nr_switches; |
2960 | ||
2961 | return sum; | |
2962 | } | |
2963 | ||
2964 | unsigned long nr_iowait(void) | |
2965 | { | |
2966 | unsigned long i, sum = 0; | |
2967 | ||
0a945022 | 2968 | for_each_possible_cpu(i) |
1da177e4 LT |
2969 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2970 | ||
2971 | return sum; | |
2972 | } | |
2973 | ||
69d25870 AV |
2974 | unsigned long nr_iowait_cpu(void) |
2975 | { | |
2976 | struct rq *this = this_rq(); | |
2977 | return atomic_read(&this->nr_iowait); | |
2978 | } | |
2979 | ||
2980 | unsigned long this_cpu_load(void) | |
2981 | { | |
2982 | struct rq *this = this_rq(); | |
2983 | return this->cpu_load[0]; | |
2984 | } | |
2985 | ||
2986 | ||
dce48a84 TG |
2987 | /* Variables and functions for calc_load */ |
2988 | static atomic_long_t calc_load_tasks; | |
2989 | static unsigned long calc_load_update; | |
2990 | unsigned long avenrun[3]; | |
2991 | EXPORT_SYMBOL(avenrun); | |
2992 | ||
2d02494f TG |
2993 | /** |
2994 | * get_avenrun - get the load average array | |
2995 | * @loads: pointer to dest load array | |
2996 | * @offset: offset to add | |
2997 | * @shift: shift count to shift the result left | |
2998 | * | |
2999 | * These values are estimates at best, so no need for locking. | |
3000 | */ | |
3001 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3002 | { | |
3003 | loads[0] = (avenrun[0] + offset) << shift; | |
3004 | loads[1] = (avenrun[1] + offset) << shift; | |
3005 | loads[2] = (avenrun[2] + offset) << shift; | |
3006 | } | |
3007 | ||
dce48a84 TG |
3008 | static unsigned long |
3009 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3010 | { |
dce48a84 TG |
3011 | load *= exp; |
3012 | load += active * (FIXED_1 - exp); | |
3013 | return load >> FSHIFT; | |
3014 | } | |
db1b1fef | 3015 | |
dce48a84 TG |
3016 | /* |
3017 | * calc_load - update the avenrun load estimates 10 ticks after the | |
3018 | * CPUs have updated calc_load_tasks. | |
3019 | */ | |
3020 | void calc_global_load(void) | |
3021 | { | |
3022 | unsigned long upd = calc_load_update + 10; | |
3023 | long active; | |
3024 | ||
3025 | if (time_before(jiffies, upd)) | |
3026 | return; | |
db1b1fef | 3027 | |
dce48a84 TG |
3028 | active = atomic_long_read(&calc_load_tasks); |
3029 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3030 | |
dce48a84 TG |
3031 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3032 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3033 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3034 | ||
3035 | calc_load_update += LOAD_FREQ; | |
3036 | } | |
3037 | ||
3038 | /* | |
3039 | * Either called from update_cpu_load() or from a cpu going idle | |
3040 | */ | |
3041 | static void calc_load_account_active(struct rq *this_rq) | |
3042 | { | |
3043 | long nr_active, delta; | |
3044 | ||
3045 | nr_active = this_rq->nr_running; | |
3046 | nr_active += (long) this_rq->nr_uninterruptible; | |
3047 | ||
3048 | if (nr_active != this_rq->calc_load_active) { | |
3049 | delta = nr_active - this_rq->calc_load_active; | |
3050 | this_rq->calc_load_active = nr_active; | |
3051 | atomic_long_add(delta, &calc_load_tasks); | |
3052 | } | |
db1b1fef JS |
3053 | } |
3054 | ||
48f24c4d | 3055 | /* |
dd41f596 IM |
3056 | * Update rq->cpu_load[] statistics. This function is usually called every |
3057 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3058 | */ |
dd41f596 | 3059 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3060 | { |
495eca49 | 3061 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3062 | int i, scale; |
3063 | ||
3064 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3065 | |
3066 | /* Update our load: */ | |
3067 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3068 | unsigned long old_load, new_load; | |
3069 | ||
3070 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3071 | ||
3072 | old_load = this_rq->cpu_load[i]; | |
3073 | new_load = this_load; | |
a25707f3 IM |
3074 | /* |
3075 | * Round up the averaging division if load is increasing. This | |
3076 | * prevents us from getting stuck on 9 if the load is 10, for | |
3077 | * example. | |
3078 | */ | |
3079 | if (new_load > old_load) | |
3080 | new_load += scale-1; | |
dd41f596 IM |
3081 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3082 | } | |
dce48a84 TG |
3083 | |
3084 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3085 | this_rq->calc_load_update += LOAD_FREQ; | |
3086 | calc_load_account_active(this_rq); | |
3087 | } | |
48f24c4d IM |
3088 | } |
3089 | ||
dd41f596 IM |
3090 | #ifdef CONFIG_SMP |
3091 | ||
1da177e4 LT |
3092 | /* |
3093 | * double_rq_lock - safely lock two runqueues | |
3094 | * | |
3095 | * Note this does not disable interrupts like task_rq_lock, | |
3096 | * you need to do so manually before calling. | |
3097 | */ | |
70b97a7f | 3098 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3099 | __acquires(rq1->lock) |
3100 | __acquires(rq2->lock) | |
3101 | { | |
054b9108 | 3102 | BUG_ON(!irqs_disabled()); |
1da177e4 | 3103 | if (rq1 == rq2) { |
05fa785c | 3104 | raw_spin_lock(&rq1->lock); |
1da177e4 LT |
3105 | __acquire(rq2->lock); /* Fake it out ;) */ |
3106 | } else { | |
c96d145e | 3107 | if (rq1 < rq2) { |
05fa785c TG |
3108 | raw_spin_lock(&rq1->lock); |
3109 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1da177e4 | 3110 | } else { |
05fa785c TG |
3111 | raw_spin_lock(&rq2->lock); |
3112 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1da177e4 LT |
3113 | } |
3114 | } | |
6e82a3be IM |
3115 | update_rq_clock(rq1); |
3116 | update_rq_clock(rq2); | |
1da177e4 LT |
3117 | } |
3118 | ||
3119 | /* | |
3120 | * double_rq_unlock - safely unlock two runqueues | |
3121 | * | |
3122 | * Note this does not restore interrupts like task_rq_unlock, | |
3123 | * you need to do so manually after calling. | |
3124 | */ | |
70b97a7f | 3125 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3126 | __releases(rq1->lock) |
3127 | __releases(rq2->lock) | |
3128 | { | |
05fa785c | 3129 | raw_spin_unlock(&rq1->lock); |
1da177e4 | 3130 | if (rq1 != rq2) |
05fa785c | 3131 | raw_spin_unlock(&rq2->lock); |
1da177e4 LT |
3132 | else |
3133 | __release(rq2->lock); | |
3134 | } | |
3135 | ||
1da177e4 | 3136 | /* |
38022906 PZ |
3137 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3138 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 | 3139 | */ |
38022906 | 3140 | void sched_exec(void) |
1da177e4 | 3141 | { |
38022906 | 3142 | struct task_struct *p = current; |
70b97a7f | 3143 | struct migration_req req; |
38022906 | 3144 | int dest_cpu, this_cpu; |
1da177e4 | 3145 | unsigned long flags; |
70b97a7f | 3146 | struct rq *rq; |
1da177e4 | 3147 | |
38022906 PZ |
3148 | again: |
3149 | this_cpu = get_cpu(); | |
3150 | dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0); | |
3151 | if (dest_cpu == this_cpu) { | |
3152 | put_cpu(); | |
3153 | return; | |
3154 | } | |
3155 | ||
1da177e4 | 3156 | rq = task_rq_lock(p, &flags); |
38022906 PZ |
3157 | put_cpu(); |
3158 | ||
3159 | /* | |
3160 | * select_task_rq() can race against ->cpus_allowed | |
3161 | */ | |
96f874e2 | 3162 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
38022906 PZ |
3163 | || unlikely(!cpu_active(dest_cpu))) { |
3164 | task_rq_unlock(rq, &flags); | |
3165 | goto again; | |
3166 | } | |
1da177e4 LT |
3167 | |
3168 | /* force the process onto the specified CPU */ | |
3169 | if (migrate_task(p, dest_cpu, &req)) { | |
3170 | /* Need to wait for migration thread (might exit: take ref). */ | |
3171 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3172 | |
1da177e4 LT |
3173 | get_task_struct(mt); |
3174 | task_rq_unlock(rq, &flags); | |
3175 | wake_up_process(mt); | |
3176 | put_task_struct(mt); | |
3177 | wait_for_completion(&req.done); | |
36c8b586 | 3178 | |
1da177e4 LT |
3179 | return; |
3180 | } | |
1da177e4 LT |
3181 | task_rq_unlock(rq, &flags); |
3182 | } | |
3183 | ||
1da177e4 LT |
3184 | /* |
3185 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3186 | * Both runqueues must be locked. | |
3187 | */ | |
dd41f596 IM |
3188 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3189 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3190 | { |
2e1cb74a | 3191 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3192 | set_task_cpu(p, this_cpu); |
dd41f596 | 3193 | activate_task(this_rq, p, 0); |
15afe09b | 3194 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3195 | } |
3196 | ||
3197 | /* | |
3198 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3199 | */ | |
858119e1 | 3200 | static |
70b97a7f | 3201 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3202 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3203 | int *all_pinned) |
1da177e4 | 3204 | { |
708dc512 | 3205 | int tsk_cache_hot = 0; |
1da177e4 LT |
3206 | /* |
3207 | * We do not migrate tasks that are: | |
3208 | * 1) running (obviously), or | |
3209 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3210 | * 3) are cache-hot on their current CPU. | |
3211 | */ | |
96f874e2 | 3212 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3213 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3214 | return 0; |
cc367732 | 3215 | } |
81026794 NP |
3216 | *all_pinned = 0; |
3217 | ||
cc367732 IM |
3218 | if (task_running(rq, p)) { |
3219 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3220 | return 0; |
cc367732 | 3221 | } |
1da177e4 | 3222 | |
da84d961 IM |
3223 | /* |
3224 | * Aggressive migration if: | |
3225 | * 1) task is cache cold, or | |
3226 | * 2) too many balance attempts have failed. | |
3227 | */ | |
3228 | ||
708dc512 LH |
3229 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3230 | if (!tsk_cache_hot || | |
3231 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3232 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3233 | if (tsk_cache_hot) { |
da84d961 | 3234 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3235 | schedstat_inc(p, se.nr_forced_migrations); |
3236 | } | |
da84d961 IM |
3237 | #endif |
3238 | return 1; | |
3239 | } | |
3240 | ||
708dc512 | 3241 | if (tsk_cache_hot) { |
cc367732 | 3242 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3243 | return 0; |
cc367732 | 3244 | } |
1da177e4 LT |
3245 | return 1; |
3246 | } | |
3247 | ||
e1d1484f PW |
3248 | static unsigned long |
3249 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3250 | unsigned long max_load_move, struct sched_domain *sd, | |
3251 | enum cpu_idle_type idle, int *all_pinned, | |
3252 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3253 | { |
051c6764 | 3254 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3255 | struct task_struct *p; |
3256 | long rem_load_move = max_load_move; | |
1da177e4 | 3257 | |
e1d1484f | 3258 | if (max_load_move == 0) |
1da177e4 LT |
3259 | goto out; |
3260 | ||
81026794 NP |
3261 | pinned = 1; |
3262 | ||
1da177e4 | 3263 | /* |
dd41f596 | 3264 | * Start the load-balancing iterator: |
1da177e4 | 3265 | */ |
dd41f596 IM |
3266 | p = iterator->start(iterator->arg); |
3267 | next: | |
b82d9fdd | 3268 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3269 | goto out; |
051c6764 PZ |
3270 | |
3271 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3272 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3273 | p = iterator->next(iterator->arg); |
3274 | goto next; | |
1da177e4 LT |
3275 | } |
3276 | ||
dd41f596 | 3277 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3278 | pulled++; |
dd41f596 | 3279 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3280 | |
7e96fa58 GH |
3281 | #ifdef CONFIG_PREEMPT |
3282 | /* | |
3283 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3284 | * will stop after the first task is pulled to minimize the critical | |
3285 | * section. | |
3286 | */ | |
3287 | if (idle == CPU_NEWLY_IDLE) | |
3288 | goto out; | |
3289 | #endif | |
3290 | ||
2dd73a4f | 3291 | /* |
b82d9fdd | 3292 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3293 | */ |
e1d1484f | 3294 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3295 | if (p->prio < *this_best_prio) |
3296 | *this_best_prio = p->prio; | |
dd41f596 IM |
3297 | p = iterator->next(iterator->arg); |
3298 | goto next; | |
1da177e4 LT |
3299 | } |
3300 | out: | |
3301 | /* | |
e1d1484f | 3302 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3303 | * so we can safely collect pull_task() stats here rather than |
3304 | * inside pull_task(). | |
3305 | */ | |
3306 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3307 | |
3308 | if (all_pinned) | |
3309 | *all_pinned = pinned; | |
e1d1484f PW |
3310 | |
3311 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3312 | } |
3313 | ||
dd41f596 | 3314 | /* |
43010659 PW |
3315 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3316 | * this_rq, as part of a balancing operation within domain "sd". | |
3317 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3318 | * |
3319 | * Called with both runqueues locked. | |
3320 | */ | |
3321 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3322 | unsigned long max_load_move, |
dd41f596 IM |
3323 | struct sched_domain *sd, enum cpu_idle_type idle, |
3324 | int *all_pinned) | |
3325 | { | |
5522d5d5 | 3326 | const struct sched_class *class = sched_class_highest; |
43010659 | 3327 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3328 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3329 | |
3330 | do { | |
43010659 PW |
3331 | total_load_moved += |
3332 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3333 | max_load_move - total_load_moved, |
a4ac01c3 | 3334 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3335 | class = class->next; |
c4acb2c0 | 3336 | |
7e96fa58 GH |
3337 | #ifdef CONFIG_PREEMPT |
3338 | /* | |
3339 | * NEWIDLE balancing is a source of latency, so preemptible | |
3340 | * kernels will stop after the first task is pulled to minimize | |
3341 | * the critical section. | |
3342 | */ | |
c4acb2c0 GH |
3343 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3344 | break; | |
7e96fa58 | 3345 | #endif |
43010659 | 3346 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3347 | |
43010659 PW |
3348 | return total_load_moved > 0; |
3349 | } | |
3350 | ||
e1d1484f PW |
3351 | static int |
3352 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3353 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3354 | struct rq_iterator *iterator) | |
3355 | { | |
3356 | struct task_struct *p = iterator->start(iterator->arg); | |
3357 | int pinned = 0; | |
3358 | ||
3359 | while (p) { | |
3360 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3361 | pull_task(busiest, p, this_rq, this_cpu); | |
3362 | /* | |
3363 | * Right now, this is only the second place pull_task() | |
3364 | * is called, so we can safely collect pull_task() | |
3365 | * stats here rather than inside pull_task(). | |
3366 | */ | |
3367 | schedstat_inc(sd, lb_gained[idle]); | |
3368 | ||
3369 | return 1; | |
3370 | } | |
3371 | p = iterator->next(iterator->arg); | |
3372 | } | |
3373 | ||
3374 | return 0; | |
3375 | } | |
3376 | ||
43010659 PW |
3377 | /* |
3378 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3379 | * part of active balancing operations within "domain". | |
3380 | * Returns 1 if successful and 0 otherwise. | |
3381 | * | |
3382 | * Called with both runqueues locked. | |
3383 | */ | |
3384 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3385 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3386 | { | |
5522d5d5 | 3387 | const struct sched_class *class; |
43010659 | 3388 | |
cde7e5ca | 3389 | for_each_class(class) { |
e1d1484f | 3390 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3391 | return 1; |
cde7e5ca | 3392 | } |
43010659 PW |
3393 | |
3394 | return 0; | |
dd41f596 | 3395 | } |
67bb6c03 | 3396 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3397 | /* |
222d656d GS |
3398 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3399 | * during load balancing. | |
1da177e4 | 3400 | */ |
222d656d GS |
3401 | struct sd_lb_stats { |
3402 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3403 | struct sched_group *this; /* Local group in this sd */ | |
3404 | unsigned long total_load; /* Total load of all groups in sd */ | |
3405 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3406 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3407 | ||
3408 | /** Statistics of this group */ | |
3409 | unsigned long this_load; | |
3410 | unsigned long this_load_per_task; | |
3411 | unsigned long this_nr_running; | |
3412 | ||
3413 | /* Statistics of the busiest group */ | |
3414 | unsigned long max_load; | |
3415 | unsigned long busiest_load_per_task; | |
3416 | unsigned long busiest_nr_running; | |
3417 | ||
3418 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3419 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3420 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3421 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3422 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3423 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3424 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3425 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3426 | #endif |
222d656d | 3427 | }; |
1da177e4 | 3428 | |
d5ac537e | 3429 | /* |
381be78f GS |
3430 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3431 | */ | |
3432 | struct sg_lb_stats { | |
3433 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3434 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3435 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3436 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3437 | unsigned long group_capacity; | |
3438 | int group_imb; /* Is there an imbalance in the group ? */ | |
3439 | }; | |
408ed066 | 3440 | |
67bb6c03 GS |
3441 | /** |
3442 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3443 | * @group: The group whose first cpu is to be returned. | |
3444 | */ | |
3445 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3446 | { | |
3447 | return cpumask_first(sched_group_cpus(group)); | |
3448 | } | |
3449 | ||
3450 | /** | |
3451 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3452 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3453 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3454 | */ | |
3455 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3456 | enum cpu_idle_type idle) | |
3457 | { | |
3458 | int load_idx; | |
3459 | ||
3460 | switch (idle) { | |
3461 | case CPU_NOT_IDLE: | |
7897986b | 3462 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3463 | break; |
3464 | ||
3465 | case CPU_NEWLY_IDLE: | |
7897986b | 3466 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3467 | break; |
3468 | default: | |
7897986b | 3469 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3470 | break; |
3471 | } | |
1da177e4 | 3472 | |
67bb6c03 GS |
3473 | return load_idx; |
3474 | } | |
1da177e4 | 3475 | |
1da177e4 | 3476 | |
c071df18 GS |
3477 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3478 | /** | |
3479 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3480 | * the given sched_domain, during load balancing. | |
3481 | * | |
3482 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3483 | * @sds: Variable containing the statistics for sd. | |
3484 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3485 | */ | |
3486 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3487 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3488 | { | |
3489 | /* | |
3490 | * Busy processors will not participate in power savings | |
3491 | * balance. | |
3492 | */ | |
3493 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3494 | sds->power_savings_balance = 0; | |
3495 | else { | |
3496 | sds->power_savings_balance = 1; | |
3497 | sds->min_nr_running = ULONG_MAX; | |
3498 | sds->leader_nr_running = 0; | |
3499 | } | |
3500 | } | |
783609c6 | 3501 | |
c071df18 GS |
3502 | /** |
3503 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3504 | * sched_domain while performing load balancing. | |
3505 | * | |
3506 | * @group: sched_group belonging to the sched_domain under consideration. | |
3507 | * @sds: Variable containing the statistics of the sched_domain | |
3508 | * @local_group: Does group contain the CPU for which we're performing | |
3509 | * load balancing ? | |
3510 | * @sgs: Variable containing the statistics of the group. | |
3511 | */ | |
3512 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3513 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3514 | { | |
408ed066 | 3515 | |
c071df18 GS |
3516 | if (!sds->power_savings_balance) |
3517 | return; | |
1da177e4 | 3518 | |
c071df18 GS |
3519 | /* |
3520 | * If the local group is idle or completely loaded | |
3521 | * no need to do power savings balance at this domain | |
3522 | */ | |
3523 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3524 | !sds->this_nr_running)) | |
3525 | sds->power_savings_balance = 0; | |
2dd73a4f | 3526 | |
c071df18 GS |
3527 | /* |
3528 | * If a group is already running at full capacity or idle, | |
3529 | * don't include that group in power savings calculations | |
3530 | */ | |
3531 | if (!sds->power_savings_balance || | |
3532 | sgs->sum_nr_running >= sgs->group_capacity || | |
3533 | !sgs->sum_nr_running) | |
3534 | return; | |
5969fe06 | 3535 | |
c071df18 GS |
3536 | /* |
3537 | * Calculate the group which has the least non-idle load. | |
3538 | * This is the group from where we need to pick up the load | |
3539 | * for saving power | |
3540 | */ | |
3541 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3542 | (sgs->sum_nr_running == sds->min_nr_running && | |
3543 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3544 | sds->group_min = group; | |
3545 | sds->min_nr_running = sgs->sum_nr_running; | |
3546 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3547 | sgs->sum_nr_running; | |
3548 | } | |
783609c6 | 3549 | |
c071df18 GS |
3550 | /* |
3551 | * Calculate the group which is almost near its | |
3552 | * capacity but still has some space to pick up some load | |
3553 | * from other group and save more power | |
3554 | */ | |
d899a789 | 3555 | if (sgs->sum_nr_running + 1 > sgs->group_capacity) |
c071df18 | 3556 | return; |
1da177e4 | 3557 | |
c071df18 GS |
3558 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3559 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3560 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3561 | sds->group_leader = group; | |
3562 | sds->leader_nr_running = sgs->sum_nr_running; | |
3563 | } | |
3564 | } | |
408ed066 | 3565 | |
c071df18 | 3566 | /** |
d5ac537e | 3567 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3568 | * @sds: Variable containing the statistics of the sched_domain |
3569 | * under consideration. | |
3570 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3571 | * @imbalance: Variable to store the imbalance. | |
3572 | * | |
d5ac537e RD |
3573 | * Description: |
3574 | * Check if we have potential to perform some power-savings balance. | |
3575 | * If yes, set the busiest group to be the least loaded group in the | |
3576 | * sched_domain, so that it's CPUs can be put to idle. | |
3577 | * | |
c071df18 GS |
3578 | * Returns 1 if there is potential to perform power-savings balance. |
3579 | * Else returns 0. | |
3580 | */ | |
3581 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3582 | int this_cpu, unsigned long *imbalance) | |
3583 | { | |
3584 | if (!sds->power_savings_balance) | |
3585 | return 0; | |
1da177e4 | 3586 | |
c071df18 GS |
3587 | if (sds->this != sds->group_leader || |
3588 | sds->group_leader == sds->group_min) | |
3589 | return 0; | |
783609c6 | 3590 | |
c071df18 GS |
3591 | *imbalance = sds->min_load_per_task; |
3592 | sds->busiest = sds->group_min; | |
1da177e4 | 3593 | |
c071df18 | 3594 | return 1; |
1da177e4 | 3595 | |
c071df18 GS |
3596 | } |
3597 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3598 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3599 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3600 | { | |
3601 | return; | |
3602 | } | |
408ed066 | 3603 | |
c071df18 GS |
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 | { | |
3607 | return; | |
3608 | } | |
3609 | ||
3610 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3611 | int this_cpu, unsigned long *imbalance) | |
3612 | { | |
3613 | return 0; | |
3614 | } | |
3615 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3616 | ||
d6a59aa3 PZ |
3617 | |
3618 | unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu) | |
3619 | { | |
3620 | return SCHED_LOAD_SCALE; | |
3621 | } | |
3622 | ||
3623 | unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu) | |
3624 | { | |
3625 | return default_scale_freq_power(sd, cpu); | |
3626 | } | |
3627 | ||
3628 | unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu) | |
ab29230e PZ |
3629 | { |
3630 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3631 | unsigned long smt_gain = sd->smt_gain; | |
3632 | ||
3633 | smt_gain /= weight; | |
3634 | ||
3635 | return smt_gain; | |
3636 | } | |
3637 | ||
d6a59aa3 PZ |
3638 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
3639 | { | |
3640 | return default_scale_smt_power(sd, cpu); | |
3641 | } | |
3642 | ||
e9e9250b PZ |
3643 | unsigned long scale_rt_power(int cpu) |
3644 | { | |
3645 | struct rq *rq = cpu_rq(cpu); | |
3646 | u64 total, available; | |
3647 | ||
3648 | sched_avg_update(rq); | |
3649 | ||
3650 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3651 | available = total - rq->rt_avg; | |
3652 | ||
3653 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3654 | total = SCHED_LOAD_SCALE; | |
3655 | ||
3656 | total >>= SCHED_LOAD_SHIFT; | |
3657 | ||
3658 | return div_u64(available, total); | |
3659 | } | |
3660 | ||
ab29230e PZ |
3661 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3662 | { | |
3663 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3664 | unsigned long power = SCHED_LOAD_SCALE; | |
3665 | struct sched_group *sdg = sd->groups; | |
ab29230e | 3666 | |
8e6598af PZ |
3667 | if (sched_feat(ARCH_POWER)) |
3668 | power *= arch_scale_freq_power(sd, cpu); | |
3669 | else | |
3670 | power *= default_scale_freq_power(sd, cpu); | |
3671 | ||
d6a59aa3 | 3672 | power >>= SCHED_LOAD_SHIFT; |
ab29230e PZ |
3673 | |
3674 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
8e6598af PZ |
3675 | if (sched_feat(ARCH_POWER)) |
3676 | power *= arch_scale_smt_power(sd, cpu); | |
3677 | else | |
3678 | power *= default_scale_smt_power(sd, cpu); | |
3679 | ||
ab29230e PZ |
3680 | power >>= SCHED_LOAD_SHIFT; |
3681 | } | |
3682 | ||
e9e9250b PZ |
3683 | power *= scale_rt_power(cpu); |
3684 | power >>= SCHED_LOAD_SHIFT; | |
3685 | ||
3686 | if (!power) | |
3687 | power = 1; | |
ab29230e | 3688 | |
18a3885f | 3689 | sdg->cpu_power = power; |
ab29230e PZ |
3690 | } |
3691 | ||
3692 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3693 | { |
3694 | struct sched_domain *child = sd->child; | |
3695 | struct sched_group *group, *sdg = sd->groups; | |
d7ea17a7 | 3696 | unsigned long power; |
cc9fba7d PZ |
3697 | |
3698 | if (!child) { | |
ab29230e | 3699 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3700 | return; |
3701 | } | |
3702 | ||
d7ea17a7 | 3703 | power = 0; |
cc9fba7d PZ |
3704 | |
3705 | group = child->groups; | |
3706 | do { | |
d7ea17a7 | 3707 | power += group->cpu_power; |
cc9fba7d PZ |
3708 | group = group->next; |
3709 | } while (group != child->groups); | |
d7ea17a7 IM |
3710 | |
3711 | sdg->cpu_power = power; | |
cc9fba7d | 3712 | } |
c071df18 | 3713 | |
1f8c553d GS |
3714 | /** |
3715 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
e17b38bf | 3716 | * @sd: The sched_domain whose statistics are to be updated. |
1f8c553d GS |
3717 | * @group: sched_group whose statistics are to be updated. |
3718 | * @this_cpu: Cpu for which load balance is currently performed. | |
3719 | * @idle: Idle status of this_cpu | |
3720 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3721 | * @sd_idle: Idle status of the sched_domain containing group. | |
3722 | * @local_group: Does group contain this_cpu. | |
3723 | * @cpus: Set of cpus considered for load balancing. | |
3724 | * @balance: Should we balance. | |
3725 | * @sgs: variable to hold the statistics for this group. | |
3726 | */ | |
cc9fba7d PZ |
3727 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3728 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3729 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3730 | int local_group, const struct cpumask *cpus, | |
3731 | int *balance, struct sg_lb_stats *sgs) | |
3732 | { | |
3733 | unsigned long load, max_cpu_load, min_cpu_load; | |
3734 | int i; | |
3735 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3736 | unsigned long sum_avg_load_per_task; | |
3737 | unsigned long avg_load_per_task; | |
3738 | ||
cc9fba7d | 3739 | if (local_group) { |
1f8c553d | 3740 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3741 | if (balance_cpu == this_cpu) |
ab29230e | 3742 | update_group_power(sd, this_cpu); |
cc9fba7d | 3743 | } |
1f8c553d GS |
3744 | |
3745 | /* Tally up the load of all CPUs in the group */ | |
3746 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3747 | max_cpu_load = 0; | |
3748 | min_cpu_load = ~0UL; | |
408ed066 | 3749 | |
1f8c553d GS |
3750 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3751 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3752 | |
1f8c553d GS |
3753 | if (*sd_idle && rq->nr_running) |
3754 | *sd_idle = 0; | |
5c45bf27 | 3755 | |
1f8c553d | 3756 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3757 | if (local_group) { |
1f8c553d GS |
3758 | if (idle_cpu(i) && !first_idle_cpu) { |
3759 | first_idle_cpu = 1; | |
3760 | balance_cpu = i; | |
3761 | } | |
3762 | ||
3763 | load = target_load(i, load_idx); | |
3764 | } else { | |
3765 | load = source_load(i, load_idx); | |
3766 | if (load > max_cpu_load) | |
3767 | max_cpu_load = load; | |
3768 | if (min_cpu_load > load) | |
3769 | min_cpu_load = load; | |
1da177e4 | 3770 | } |
5c45bf27 | 3771 | |
1f8c553d GS |
3772 | sgs->group_load += load; |
3773 | sgs->sum_nr_running += rq->nr_running; | |
3774 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3775 | |
1f8c553d GS |
3776 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3777 | } | |
5c45bf27 | 3778 | |
1f8c553d GS |
3779 | /* |
3780 | * First idle cpu or the first cpu(busiest) in this sched group | |
3781 | * is eligible for doing load balancing at this and above | |
3782 | * domains. In the newly idle case, we will allow all the cpu's | |
3783 | * to do the newly idle load balance. | |
3784 | */ | |
3785 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3786 | balance_cpu != this_cpu && balance) { | |
3787 | *balance = 0; | |
3788 | return; | |
3789 | } | |
5c45bf27 | 3790 | |
1f8c553d | 3791 | /* Adjust by relative CPU power of the group */ |
18a3885f | 3792 | sgs->avg_load = (sgs->group_load * SCHED_LOAD_SCALE) / group->cpu_power; |
5c45bf27 | 3793 | |
1f8c553d GS |
3794 | |
3795 | /* | |
3796 | * Consider the group unbalanced when the imbalance is larger | |
3797 | * than the average weight of two tasks. | |
3798 | * | |
3799 | * APZ: with cgroup the avg task weight can vary wildly and | |
3800 | * might not be a suitable number - should we keep a | |
3801 | * normalized nr_running number somewhere that negates | |
3802 | * the hierarchy? | |
3803 | */ | |
18a3885f PZ |
3804 | avg_load_per_task = (sum_avg_load_per_task * SCHED_LOAD_SCALE) / |
3805 | group->cpu_power; | |
1f8c553d GS |
3806 | |
3807 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3808 | sgs->group_imb = 1; | |
3809 | ||
bdb94aa5 | 3810 | sgs->group_capacity = |
18a3885f | 3811 | DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE); |
1f8c553d | 3812 | } |
dd41f596 | 3813 | |
37abe198 GS |
3814 | /** |
3815 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3816 | * @sd: sched_domain whose statistics are to be updated. | |
3817 | * @this_cpu: Cpu for which load balance is currently performed. | |
3818 | * @idle: Idle status of this_cpu | |
3819 | * @sd_idle: Idle status of the sched_domain containing group. | |
3820 | * @cpus: Set of cpus considered for load balancing. | |
3821 | * @balance: Should we balance. | |
3822 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3823 | */ |
37abe198 GS |
3824 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3825 | enum cpu_idle_type idle, int *sd_idle, | |
3826 | const struct cpumask *cpus, int *balance, | |
3827 | struct sd_lb_stats *sds) | |
1da177e4 | 3828 | { |
b5d978e0 | 3829 | struct sched_domain *child = sd->child; |
222d656d | 3830 | struct sched_group *group = sd->groups; |
37abe198 | 3831 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3832 | int load_idx, prefer_sibling = 0; |
3833 | ||
3834 | if (child && child->flags & SD_PREFER_SIBLING) | |
3835 | prefer_sibling = 1; | |
222d656d | 3836 | |
c071df18 | 3837 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3838 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3839 | |
3840 | do { | |
1da177e4 | 3841 | int local_group; |
1da177e4 | 3842 | |
758b2cdc RR |
3843 | local_group = cpumask_test_cpu(this_cpu, |
3844 | sched_group_cpus(group)); | |
381be78f | 3845 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3846 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3847 | local_group, cpus, balance, &sgs); |
1da177e4 | 3848 | |
37abe198 GS |
3849 | if (local_group && balance && !(*balance)) |
3850 | return; | |
783609c6 | 3851 | |
37abe198 | 3852 | sds->total_load += sgs.group_load; |
18a3885f | 3853 | sds->total_pwr += group->cpu_power; |
1da177e4 | 3854 | |
b5d978e0 PZ |
3855 | /* |
3856 | * In case the child domain prefers tasks go to siblings | |
3857 | * first, lower the group capacity to one so that we'll try | |
3858 | * and move all the excess tasks away. | |
3859 | */ | |
3860 | if (prefer_sibling) | |
bdb94aa5 | 3861 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
1da177e4 | 3862 | |
1da177e4 | 3863 | if (local_group) { |
37abe198 GS |
3864 | sds->this_load = sgs.avg_load; |
3865 | sds->this = group; | |
3866 | sds->this_nr_running = sgs.sum_nr_running; | |
3867 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3868 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3869 | (sgs.sum_nr_running > sgs.group_capacity || |
3870 | sgs.group_imb)) { | |
37abe198 GS |
3871 | sds->max_load = sgs.avg_load; |
3872 | sds->busiest = group; | |
3873 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3874 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3875 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3876 | } |
5c45bf27 | 3877 | |
c071df18 | 3878 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3879 | group = group->next; |
3880 | } while (group != sd->groups); | |
37abe198 | 3881 | } |
1da177e4 | 3882 | |
2e6f44ae GS |
3883 | /** |
3884 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3885 | * amongst the groups of a sched_domain, during |
3886 | * load balancing. | |
2e6f44ae GS |
3887 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3888 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3889 | * @imbalance: Variable to store the imbalance. | |
3890 | */ | |
3891 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3892 | int this_cpu, unsigned long *imbalance) | |
3893 | { | |
3894 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3895 | unsigned int imbn = 2; | |
3896 | ||
3897 | if (sds->this_nr_running) { | |
3898 | sds->this_load_per_task /= sds->this_nr_running; | |
3899 | if (sds->busiest_load_per_task > | |
3900 | sds->this_load_per_task) | |
3901 | imbn = 1; | |
3902 | } else | |
3903 | sds->this_load_per_task = | |
3904 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3905 | |
2e6f44ae GS |
3906 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3907 | sds->busiest_load_per_task * imbn) { | |
3908 | *imbalance = sds->busiest_load_per_task; | |
3909 | return; | |
3910 | } | |
908a7c1b | 3911 | |
1da177e4 | 3912 | /* |
2e6f44ae GS |
3913 | * OK, we don't have enough imbalance to justify moving tasks, |
3914 | * however we may be able to increase total CPU power used by | |
3915 | * moving them. | |
1da177e4 | 3916 | */ |
2dd73a4f | 3917 | |
18a3885f | 3918 | pwr_now += sds->busiest->cpu_power * |
2e6f44ae | 3919 | min(sds->busiest_load_per_task, sds->max_load); |
18a3885f | 3920 | pwr_now += sds->this->cpu_power * |
2e6f44ae GS |
3921 | min(sds->this_load_per_task, sds->this_load); |
3922 | pwr_now /= SCHED_LOAD_SCALE; | |
3923 | ||
3924 | /* Amount of load we'd subtract */ | |
18a3885f PZ |
3925 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3926 | sds->busiest->cpu_power; | |
2e6f44ae | 3927 | if (sds->max_load > tmp) |
18a3885f | 3928 | pwr_move += sds->busiest->cpu_power * |
2e6f44ae GS |
3929 | min(sds->busiest_load_per_task, sds->max_load - tmp); |
3930 | ||
3931 | /* Amount of load we'd add */ | |
18a3885f | 3932 | if (sds->max_load * sds->busiest->cpu_power < |
2e6f44ae | 3933 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) |
18a3885f PZ |
3934 | tmp = (sds->max_load * sds->busiest->cpu_power) / |
3935 | sds->this->cpu_power; | |
2e6f44ae | 3936 | else |
18a3885f PZ |
3937 | tmp = (sds->busiest_load_per_task * SCHED_LOAD_SCALE) / |
3938 | sds->this->cpu_power; | |
3939 | pwr_move += sds->this->cpu_power * | |
2e6f44ae GS |
3940 | min(sds->this_load_per_task, sds->this_load + tmp); |
3941 | pwr_move /= SCHED_LOAD_SCALE; | |
3942 | ||
3943 | /* Move if we gain throughput */ | |
3944 | if (pwr_move > pwr_now) | |
3945 | *imbalance = sds->busiest_load_per_task; | |
3946 | } | |
dbc523a3 GS |
3947 | |
3948 | /** | |
3949 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3950 | * groups of a given sched_domain during load balance. | |
3951 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3952 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3953 | * @imbalance: The variable to store the imbalance. | |
3954 | */ | |
3955 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3956 | unsigned long *imbalance) | |
3957 | { | |
3958 | unsigned long max_pull; | |
2dd73a4f PW |
3959 | /* |
3960 | * In the presence of smp nice balancing, certain scenarios can have | |
3961 | * max load less than avg load(as we skip the groups at or below | |
3962 | * its cpu_power, while calculating max_load..) | |
3963 | */ | |
dbc523a3 | 3964 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3965 | *imbalance = 0; |
dbc523a3 | 3966 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3967 | } |
0c117f1b SS |
3968 | |
3969 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3970 | max_pull = min(sds->max_load - sds->avg_load, |
3971 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3972 | |
1da177e4 | 3973 | /* How much load to actually move to equalise the imbalance */ |
18a3885f PZ |
3974 | *imbalance = min(max_pull * sds->busiest->cpu_power, |
3975 | (sds->avg_load - sds->this_load) * sds->this->cpu_power) | |
1da177e4 LT |
3976 | / SCHED_LOAD_SCALE; |
3977 | ||
2dd73a4f PW |
3978 | /* |
3979 | * if *imbalance is less than the average load per runnable task | |
3980 | * there is no gaurantee that any tasks will be moved so we'll have | |
3981 | * a think about bumping its value to force at least one task to be | |
3982 | * moved | |
3983 | */ | |
dbc523a3 GS |
3984 | if (*imbalance < sds->busiest_load_per_task) |
3985 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3986 | |
dbc523a3 | 3987 | } |
37abe198 | 3988 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3989 | |
b7bb4c9b GS |
3990 | /** |
3991 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3992 | * if there is an imbalance. If there isn't an imbalance, and | |
3993 | * the user has opted for power-savings, it returns a group whose | |
3994 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3995 | * such a group exists. | |
3996 | * | |
3997 | * Also calculates the amount of weighted load which should be moved | |
3998 | * to restore balance. | |
3999 | * | |
4000 | * @sd: The sched_domain whose busiest group is to be returned. | |
4001 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
4002 | * @imbalance: Variable which stores amount of weighted load which should | |
4003 | * be moved to restore balance/put a group to idle. | |
4004 | * @idle: The idle status of this_cpu. | |
4005 | * @sd_idle: The idleness of sd | |
4006 | * @cpus: The set of CPUs under consideration for load-balancing. | |
4007 | * @balance: Pointer to a variable indicating if this_cpu | |
4008 | * is the appropriate cpu to perform load balancing at this_level. | |
4009 | * | |
4010 | * Returns: - the busiest group if imbalance exists. | |
4011 | * - If no imbalance and user has opted for power-savings balance, | |
4012 | * return the least loaded group whose CPUs can be | |
4013 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
4014 | */ |
4015 | static struct sched_group * | |
4016 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
4017 | unsigned long *imbalance, enum cpu_idle_type idle, | |
4018 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
4019 | { | |
4020 | struct sd_lb_stats sds; | |
1da177e4 | 4021 | |
37abe198 | 4022 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 4023 | |
37abe198 GS |
4024 | /* |
4025 | * Compute the various statistics relavent for load balancing at | |
4026 | * this level. | |
4027 | */ | |
4028 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
4029 | balance, &sds); | |
4030 | ||
b7bb4c9b GS |
4031 | /* Cases where imbalance does not exist from POV of this_cpu */ |
4032 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
4033 | * at this level. | |
4034 | * 2) There is no busy sibling group to pull from. | |
4035 | * 3) This group is the busiest group. | |
4036 | * 4) This group is more busy than the avg busieness at this | |
4037 | * sched_domain. | |
4038 | * 5) The imbalance is within the specified limit. | |
4039 | * 6) Any rebalance would lead to ping-pong | |
4040 | */ | |
37abe198 GS |
4041 | if (balance && !(*balance)) |
4042 | goto ret; | |
1da177e4 | 4043 | |
b7bb4c9b GS |
4044 | if (!sds.busiest || sds.busiest_nr_running == 0) |
4045 | goto out_balanced; | |
1da177e4 | 4046 | |
b7bb4c9b | 4047 | if (sds.this_load >= sds.max_load) |
1da177e4 | 4048 | goto out_balanced; |
1da177e4 | 4049 | |
222d656d | 4050 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4051 | |
b7bb4c9b GS |
4052 | if (sds.this_load >= sds.avg_load) |
4053 | goto out_balanced; | |
4054 | ||
4055 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4056 | goto out_balanced; |
4057 | ||
222d656d GS |
4058 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4059 | if (sds.group_imb) | |
4060 | sds.busiest_load_per_task = | |
4061 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4062 | |
1da177e4 LT |
4063 | /* |
4064 | * We're trying to get all the cpus to the average_load, so we don't | |
4065 | * want to push ourselves above the average load, nor do we wish to | |
4066 | * reduce the max loaded cpu below the average load, as either of these | |
4067 | * actions would just result in more rebalancing later, and ping-pong | |
4068 | * tasks around. Thus we look for the minimum possible imbalance. | |
4069 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4070 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4071 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4072 | * appear as very large values with unsigned longs. |
4073 | */ | |
222d656d | 4074 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4075 | goto out_balanced; |
4076 | ||
dbc523a3 GS |
4077 | /* Looks like there is an imbalance. Compute it */ |
4078 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4079 | return sds.busiest; |
1da177e4 LT |
4080 | |
4081 | out_balanced: | |
c071df18 GS |
4082 | /* |
4083 | * There is no obvious imbalance. But check if we can do some balancing | |
4084 | * to save power. | |
4085 | */ | |
4086 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4087 | return sds.busiest; | |
783609c6 | 4088 | ret: |
1da177e4 LT |
4089 | *imbalance = 0; |
4090 | return NULL; | |
4091 | } | |
4092 | ||
4093 | /* | |
4094 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4095 | */ | |
70b97a7f | 4096 | static struct rq * |
d15bcfdb | 4097 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4098 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4099 | { |
70b97a7f | 4100 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4101 | unsigned long max_load = 0; |
1da177e4 LT |
4102 | int i; |
4103 | ||
758b2cdc | 4104 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4105 | unsigned long power = power_of(i); |
4106 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4107 | unsigned long wl; |
0a2966b4 | 4108 | |
96f874e2 | 4109 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4110 | continue; |
4111 | ||
48f24c4d | 4112 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4113 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4114 | wl /= power; | |
2dd73a4f | 4115 | |
bdb94aa5 | 4116 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4117 | continue; |
1da177e4 | 4118 | |
dd41f596 IM |
4119 | if (wl > max_load) { |
4120 | max_load = wl; | |
48f24c4d | 4121 | busiest = rq; |
1da177e4 LT |
4122 | } |
4123 | } | |
4124 | ||
4125 | return busiest; | |
4126 | } | |
4127 | ||
77391d71 NP |
4128 | /* |
4129 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4130 | * so long as it is large enough. | |
4131 | */ | |
4132 | #define MAX_PINNED_INTERVAL 512 | |
4133 | ||
df7c8e84 RR |
4134 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4135 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4136 | ||
1da177e4 LT |
4137 | /* |
4138 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4139 | * tasks if there is an imbalance. | |
1da177e4 | 4140 | */ |
70b97a7f | 4141 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4142 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4143 | int *balance) |
1da177e4 | 4144 | { |
43010659 | 4145 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4146 | struct sched_group *group; |
1da177e4 | 4147 | unsigned long imbalance; |
70b97a7f | 4148 | struct rq *busiest; |
fe2eea3f | 4149 | unsigned long flags; |
df7c8e84 | 4150 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4151 | |
6ad4c188 | 4152 | cpumask_copy(cpus, cpu_active_mask); |
7c16ec58 | 4153 | |
89c4710e SS |
4154 | /* |
4155 | * When power savings policy is enabled for the parent domain, idle | |
4156 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4157 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4158 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4159 | */ |
d15bcfdb | 4160 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4161 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4162 | sd_idle = 1; |
1da177e4 | 4163 | |
2d72376b | 4164 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4165 | |
0a2966b4 | 4166 | redo: |
c8cba857 | 4167 | update_shares(sd); |
0a2966b4 | 4168 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4169 | cpus, balance); |
783609c6 | 4170 | |
06066714 | 4171 | if (*balance == 0) |
783609c6 | 4172 | goto out_balanced; |
783609c6 | 4173 | |
1da177e4 LT |
4174 | if (!group) { |
4175 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4176 | goto out_balanced; | |
4177 | } | |
4178 | ||
7c16ec58 | 4179 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4180 | if (!busiest) { |
4181 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4182 | goto out_balanced; | |
4183 | } | |
4184 | ||
db935dbd | 4185 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4186 | |
4187 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4188 | ||
43010659 | 4189 | ld_moved = 0; |
1da177e4 LT |
4190 | if (busiest->nr_running > 1) { |
4191 | /* | |
4192 | * Attempt to move tasks. If find_busiest_group has found | |
4193 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4194 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4195 | * correctly treated as an imbalance. |
4196 | */ | |
fe2eea3f | 4197 | local_irq_save(flags); |
e17224bf | 4198 | double_rq_lock(this_rq, busiest); |
43010659 | 4199 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4200 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4201 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4202 | local_irq_restore(flags); |
81026794 | 4203 | |
46cb4b7c SS |
4204 | /* |
4205 | * some other cpu did the load balance for us. | |
4206 | */ | |
43010659 | 4207 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4208 | resched_cpu(this_cpu); |
4209 | ||
81026794 | 4210 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4211 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4212 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4213 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4214 | goto redo; |
81026794 | 4215 | goto out_balanced; |
0a2966b4 | 4216 | } |
1da177e4 | 4217 | } |
81026794 | 4218 | |
43010659 | 4219 | if (!ld_moved) { |
1da177e4 LT |
4220 | schedstat_inc(sd, lb_failed[idle]); |
4221 | sd->nr_balance_failed++; | |
4222 | ||
4223 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4224 | |
05fa785c | 4225 | raw_spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4226 | |
4227 | /* don't kick the migration_thread, if the curr | |
4228 | * task on busiest cpu can't be moved to this_cpu | |
4229 | */ | |
96f874e2 RR |
4230 | if (!cpumask_test_cpu(this_cpu, |
4231 | &busiest->curr->cpus_allowed)) { | |
05fa785c TG |
4232 | raw_spin_unlock_irqrestore(&busiest->lock, |
4233 | flags); | |
fa3b6ddc SS |
4234 | all_pinned = 1; |
4235 | goto out_one_pinned; | |
4236 | } | |
4237 | ||
1da177e4 LT |
4238 | if (!busiest->active_balance) { |
4239 | busiest->active_balance = 1; | |
4240 | busiest->push_cpu = this_cpu; | |
81026794 | 4241 | active_balance = 1; |
1da177e4 | 4242 | } |
05fa785c | 4243 | raw_spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4244 | if (active_balance) |
1da177e4 LT |
4245 | wake_up_process(busiest->migration_thread); |
4246 | ||
4247 | /* | |
4248 | * We've kicked active balancing, reset the failure | |
4249 | * counter. | |
4250 | */ | |
39507451 | 4251 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4252 | } |
81026794 | 4253 | } else |
1da177e4 LT |
4254 | sd->nr_balance_failed = 0; |
4255 | ||
81026794 | 4256 | if (likely(!active_balance)) { |
1da177e4 LT |
4257 | /* We were unbalanced, so reset the balancing interval */ |
4258 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4259 | } else { |
4260 | /* | |
4261 | * If we've begun active balancing, start to back off. This | |
4262 | * case may not be covered by the all_pinned logic if there | |
4263 | * is only 1 task on the busy runqueue (because we don't call | |
4264 | * move_tasks). | |
4265 | */ | |
4266 | if (sd->balance_interval < sd->max_interval) | |
4267 | sd->balance_interval *= 2; | |
1da177e4 LT |
4268 | } |
4269 | ||
43010659 | 4270 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4271 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4272 | ld_moved = -1; |
4273 | ||
4274 | goto out; | |
1da177e4 LT |
4275 | |
4276 | out_balanced: | |
1da177e4 LT |
4277 | schedstat_inc(sd, lb_balanced[idle]); |
4278 | ||
16cfb1c0 | 4279 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4280 | |
4281 | out_one_pinned: | |
1da177e4 | 4282 | /* tune up the balancing interval */ |
77391d71 NP |
4283 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4284 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4285 | sd->balance_interval *= 2; |
4286 | ||
48f24c4d | 4287 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4288 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4289 | ld_moved = -1; |
4290 | else | |
4291 | ld_moved = 0; | |
4292 | out: | |
c8cba857 PZ |
4293 | if (ld_moved) |
4294 | update_shares(sd); | |
c09595f6 | 4295 | return ld_moved; |
1da177e4 LT |
4296 | } |
4297 | ||
4298 | /* | |
4299 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4300 | * tasks if there is an imbalance. | |
4301 | * | |
d15bcfdb | 4302 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4303 | * this_rq is locked. |
4304 | */ | |
48f24c4d | 4305 | static int |
df7c8e84 | 4306 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4307 | { |
4308 | struct sched_group *group; | |
70b97a7f | 4309 | struct rq *busiest = NULL; |
1da177e4 | 4310 | unsigned long imbalance; |
43010659 | 4311 | int ld_moved = 0; |
5969fe06 | 4312 | int sd_idle = 0; |
969bb4e4 | 4313 | int all_pinned = 0; |
df7c8e84 | 4314 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4315 | |
6ad4c188 | 4316 | cpumask_copy(cpus, cpu_active_mask); |
5969fe06 | 4317 | |
89c4710e SS |
4318 | /* |
4319 | * When power savings policy is enabled for the parent domain, idle | |
4320 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4321 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4322 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4323 | */ |
4324 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4325 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4326 | sd_idle = 1; |
1da177e4 | 4327 | |
2d72376b | 4328 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4329 | redo: |
3e5459b4 | 4330 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4331 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4332 | &sd_idle, cpus, NULL); |
1da177e4 | 4333 | if (!group) { |
d15bcfdb | 4334 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4335 | goto out_balanced; |
1da177e4 LT |
4336 | } |
4337 | ||
7c16ec58 | 4338 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4339 | if (!busiest) { |
d15bcfdb | 4340 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4341 | goto out_balanced; |
1da177e4 LT |
4342 | } |
4343 | ||
db935dbd NP |
4344 | BUG_ON(busiest == this_rq); |
4345 | ||
d15bcfdb | 4346 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4347 | |
43010659 | 4348 | ld_moved = 0; |
d6d5cfaf NP |
4349 | if (busiest->nr_running > 1) { |
4350 | /* Attempt to move tasks */ | |
4351 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4352 | /* this_rq->clock is already updated */ |
4353 | update_rq_clock(busiest); | |
43010659 | 4354 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4355 | imbalance, sd, CPU_NEWLY_IDLE, |
4356 | &all_pinned); | |
1b12bbc7 | 4357 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4358 | |
969bb4e4 | 4359 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4360 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4361 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4362 | goto redo; |
4363 | } | |
d6d5cfaf NP |
4364 | } |
4365 | ||
43010659 | 4366 | if (!ld_moved) { |
36dffab6 | 4367 | int active_balance = 0; |
ad273b32 | 4368 | |
d15bcfdb | 4369 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4370 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4371 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4372 | return -1; |
ad273b32 VS |
4373 | |
4374 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4375 | return -1; | |
4376 | ||
4377 | if (sd->nr_balance_failed++ < 2) | |
4378 | return -1; | |
4379 | ||
4380 | /* | |
4381 | * The only task running in a non-idle cpu can be moved to this | |
4382 | * cpu in an attempt to completely freeup the other CPU | |
4383 | * package. The same method used to move task in load_balance() | |
4384 | * have been extended for load_balance_newidle() to speedup | |
4385 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4386 | * | |
4387 | * The package power saving logic comes from | |
4388 | * find_busiest_group(). If there are no imbalance, then | |
4389 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4390 | * f_b_g() will select a group from which a running task may be | |
4391 | * pulled to this cpu in order to make the other package idle. | |
4392 | * If there is no opportunity to make a package idle and if | |
4393 | * there are no imbalance, then f_b_g() will return NULL and no | |
4394 | * action will be taken in load_balance_newidle(). | |
4395 | * | |
4396 | * Under normal task pull operation due to imbalance, there | |
4397 | * will be more than one task in the source run queue and | |
4398 | * move_tasks() will succeed. ld_moved will be true and this | |
4399 | * active balance code will not be triggered. | |
4400 | */ | |
4401 | ||
4402 | /* Lock busiest in correct order while this_rq is held */ | |
4403 | double_lock_balance(this_rq, busiest); | |
4404 | ||
4405 | /* | |
4406 | * don't kick the migration_thread, if the curr | |
4407 | * task on busiest cpu can't be moved to this_cpu | |
4408 | */ | |
6ca09dfc | 4409 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4410 | double_unlock_balance(this_rq, busiest); |
4411 | all_pinned = 1; | |
4412 | return ld_moved; | |
4413 | } | |
4414 | ||
4415 | if (!busiest->active_balance) { | |
4416 | busiest->active_balance = 1; | |
4417 | busiest->push_cpu = this_cpu; | |
4418 | active_balance = 1; | |
4419 | } | |
4420 | ||
4421 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4422 | /* |
4423 | * Should not call ttwu while holding a rq->lock | |
4424 | */ | |
05fa785c | 4425 | raw_spin_unlock(&this_rq->lock); |
ad273b32 VS |
4426 | if (active_balance) |
4427 | wake_up_process(busiest->migration_thread); | |
05fa785c | 4428 | raw_spin_lock(&this_rq->lock); |
ad273b32 | 4429 | |
5969fe06 | 4430 | } else |
16cfb1c0 | 4431 | sd->nr_balance_failed = 0; |
1da177e4 | 4432 | |
3e5459b4 | 4433 | update_shares_locked(this_rq, sd); |
43010659 | 4434 | return ld_moved; |
16cfb1c0 NP |
4435 | |
4436 | out_balanced: | |
d15bcfdb | 4437 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4438 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4439 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4440 | return -1; |
16cfb1c0 | 4441 | sd->nr_balance_failed = 0; |
48f24c4d | 4442 | |
16cfb1c0 | 4443 | return 0; |
1da177e4 LT |
4444 | } |
4445 | ||
4446 | /* | |
4447 | * idle_balance is called by schedule() if this_cpu is about to become | |
4448 | * idle. Attempts to pull tasks from other CPUs. | |
4449 | */ | |
70b97a7f | 4450 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4451 | { |
4452 | struct sched_domain *sd; | |
efbe027e | 4453 | int pulled_task = 0; |
dd41f596 | 4454 | unsigned long next_balance = jiffies + HZ; |
1da177e4 | 4455 | |
1b9508f6 MG |
4456 | this_rq->idle_stamp = this_rq->clock; |
4457 | ||
4458 | if (this_rq->avg_idle < sysctl_sched_migration_cost) | |
4459 | return; | |
4460 | ||
1da177e4 | 4461 | for_each_domain(this_cpu, sd) { |
92c4ca5c CL |
4462 | unsigned long interval; |
4463 | ||
4464 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4465 | continue; | |
4466 | ||
4467 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4468 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4469 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4470 | sd); |
92c4ca5c CL |
4471 | |
4472 | interval = msecs_to_jiffies(sd->balance_interval); | |
4473 | if (time_after(next_balance, sd->last_balance + interval)) | |
4474 | next_balance = sd->last_balance + interval; | |
1b9508f6 MG |
4475 | if (pulled_task) { |
4476 | this_rq->idle_stamp = 0; | |
92c4ca5c | 4477 | break; |
1b9508f6 | 4478 | } |
1da177e4 | 4479 | } |
dd41f596 | 4480 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4481 | /* |
4482 | * We are going idle. next_balance may be set based on | |
4483 | * a busy processor. So reset next_balance. | |
4484 | */ | |
4485 | this_rq->next_balance = next_balance; | |
dd41f596 | 4486 | } |
1da177e4 LT |
4487 | } |
4488 | ||
4489 | /* | |
4490 | * active_load_balance is run by migration threads. It pushes running tasks | |
4491 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4492 | * running on each physical CPU where possible, and avoids physical / | |
4493 | * logical imbalances. | |
4494 | * | |
4495 | * Called with busiest_rq locked. | |
4496 | */ | |
70b97a7f | 4497 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4498 | { |
39507451 | 4499 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4500 | struct sched_domain *sd; |
4501 | struct rq *target_rq; | |
39507451 | 4502 | |
48f24c4d | 4503 | /* Is there any task to move? */ |
39507451 | 4504 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4505 | return; |
4506 | ||
4507 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4508 | |
4509 | /* | |
39507451 | 4510 | * This condition is "impossible", if it occurs |
41a2d6cf | 4511 | * we need to fix it. Originally reported by |
39507451 | 4512 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4513 | */ |
39507451 | 4514 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4515 | |
39507451 NP |
4516 | /* move a task from busiest_rq to target_rq */ |
4517 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4518 | update_rq_clock(busiest_rq); |
4519 | update_rq_clock(target_rq); | |
39507451 NP |
4520 | |
4521 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4522 | for_each_domain(target_cpu, sd) { |
39507451 | 4523 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4524 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4525 | break; |
c96d145e | 4526 | } |
39507451 | 4527 | |
48f24c4d | 4528 | if (likely(sd)) { |
2d72376b | 4529 | schedstat_inc(sd, alb_count); |
39507451 | 4530 | |
43010659 PW |
4531 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4532 | sd, CPU_IDLE)) | |
48f24c4d IM |
4533 | schedstat_inc(sd, alb_pushed); |
4534 | else | |
4535 | schedstat_inc(sd, alb_failed); | |
4536 | } | |
1b12bbc7 | 4537 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4538 | } |
4539 | ||
46cb4b7c SS |
4540 | #ifdef CONFIG_NO_HZ |
4541 | static struct { | |
4542 | atomic_t load_balancer; | |
7d1e6a9b | 4543 | cpumask_var_t cpu_mask; |
f711f609 | 4544 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4545 | } nohz ____cacheline_aligned = { |
4546 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4547 | }; |
4548 | ||
eea08f32 AB |
4549 | int get_nohz_load_balancer(void) |
4550 | { | |
4551 | return atomic_read(&nohz.load_balancer); | |
4552 | } | |
4553 | ||
f711f609 GS |
4554 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4555 | /** | |
4556 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4557 | * @cpu: The cpu whose lowest level of sched domain is to | |
4558 | * be returned. | |
4559 | * @flag: The flag to check for the lowest sched_domain | |
4560 | * for the given cpu. | |
4561 | * | |
4562 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4563 | */ | |
4564 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4565 | { | |
4566 | struct sched_domain *sd; | |
4567 | ||
4568 | for_each_domain(cpu, sd) | |
4569 | if (sd && (sd->flags & flag)) | |
4570 | break; | |
4571 | ||
4572 | return sd; | |
4573 | } | |
4574 | ||
4575 | /** | |
4576 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4577 | * @cpu: The cpu whose domains we're iterating over. | |
4578 | * @sd: variable holding the value of the power_savings_sd | |
4579 | * for cpu. | |
4580 | * @flag: The flag to filter the sched_domains to be iterated. | |
4581 | * | |
4582 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4583 | * set, starting from the lowest sched_domain to the highest. | |
4584 | */ | |
4585 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4586 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4587 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4588 | ||
4589 | /** | |
4590 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4591 | * @ilb_group: group to be checked for semi-idleness | |
4592 | * | |
4593 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4594 | * | |
4595 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4596 | * and atleast one non-idle CPU. This helper function checks if the given | |
4597 | * sched_group is semi-idle or not. | |
4598 | */ | |
4599 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4600 | { | |
4601 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4602 | sched_group_cpus(ilb_group)); | |
4603 | ||
4604 | /* | |
4605 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4606 | * and atleast one idle cpu. | |
4607 | */ | |
4608 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4609 | return 0; | |
4610 | ||
4611 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4612 | return 0; | |
4613 | ||
4614 | return 1; | |
4615 | } | |
4616 | /** | |
4617 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4618 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4619 | * | |
4620 | * Returns: Returns the id of the idle load balancer if it exists, | |
4621 | * Else, returns >= nr_cpu_ids. | |
4622 | * | |
4623 | * This algorithm picks the idle load balancer such that it belongs to a | |
4624 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4625 | * completely idle packages/cores just for the purpose of idle load balancing | |
4626 | * when there are other idle cpu's which are better suited for that job. | |
4627 | */ | |
4628 | static int find_new_ilb(int cpu) | |
4629 | { | |
4630 | struct sched_domain *sd; | |
4631 | struct sched_group *ilb_group; | |
4632 | ||
4633 | /* | |
4634 | * Have idle load balancer selection from semi-idle packages only | |
4635 | * when power-aware load balancing is enabled | |
4636 | */ | |
4637 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4638 | goto out_done; | |
4639 | ||
4640 | /* | |
4641 | * Optimize for the case when we have no idle CPUs or only one | |
4642 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4643 | */ | |
4644 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4645 | goto out_done; | |
4646 | ||
4647 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4648 | ilb_group = sd->groups; | |
4649 | ||
4650 | do { | |
4651 | if (is_semi_idle_group(ilb_group)) | |
4652 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4653 | ||
4654 | ilb_group = ilb_group->next; | |
4655 | ||
4656 | } while (ilb_group != sd->groups); | |
4657 | } | |
4658 | ||
4659 | out_done: | |
4660 | return cpumask_first(nohz.cpu_mask); | |
4661 | } | |
4662 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4663 | static inline int find_new_ilb(int call_cpu) | |
4664 | { | |
6e29ec57 | 4665 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4666 | } |
4667 | #endif | |
4668 | ||
7835b98b | 4669 | /* |
46cb4b7c SS |
4670 | * This routine will try to nominate the ilb (idle load balancing) |
4671 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4672 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4673 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4674 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4675 | * arrives... | |
4676 | * | |
4677 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4678 | * for idle load balancing. ilb owner will still be part of | |
4679 | * nohz.cpu_mask.. | |
7835b98b | 4680 | * |
46cb4b7c SS |
4681 | * While stopping the tick, this cpu will become the ilb owner if there |
4682 | * is no other owner. And will be the owner till that cpu becomes busy | |
4683 | * or if all cpus in the system stop their ticks at which point | |
4684 | * there is no need for ilb owner. | |
4685 | * | |
4686 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4687 | * next busy scheduler_tick() | |
4688 | */ | |
4689 | int select_nohz_load_balancer(int stop_tick) | |
4690 | { | |
4691 | int cpu = smp_processor_id(); | |
4692 | ||
4693 | if (stop_tick) { | |
46cb4b7c SS |
4694 | cpu_rq(cpu)->in_nohz_recently = 1; |
4695 | ||
483b4ee6 SS |
4696 | if (!cpu_active(cpu)) { |
4697 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4698 | return 0; | |
4699 | ||
4700 | /* | |
4701 | * If we are going offline and still the leader, | |
4702 | * give up! | |
4703 | */ | |
46cb4b7c SS |
4704 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4705 | BUG(); | |
483b4ee6 | 4706 | |
46cb4b7c SS |
4707 | return 0; |
4708 | } | |
4709 | ||
483b4ee6 SS |
4710 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4711 | ||
46cb4b7c | 4712 | /* time for ilb owner also to sleep */ |
6ad4c188 | 4713 | if (cpumask_weight(nohz.cpu_mask) == num_active_cpus()) { |
46cb4b7c SS |
4714 | if (atomic_read(&nohz.load_balancer) == cpu) |
4715 | atomic_set(&nohz.load_balancer, -1); | |
4716 | return 0; | |
4717 | } | |
4718 | ||
4719 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4720 | /* make me the ilb owner */ | |
4721 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4722 | return 1; | |
e790fb0b GS |
4723 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4724 | int new_ilb; | |
4725 | ||
4726 | if (!(sched_smt_power_savings || | |
4727 | sched_mc_power_savings)) | |
4728 | return 1; | |
4729 | /* | |
4730 | * Check to see if there is a more power-efficient | |
4731 | * ilb. | |
4732 | */ | |
4733 | new_ilb = find_new_ilb(cpu); | |
4734 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4735 | atomic_set(&nohz.load_balancer, -1); | |
4736 | resched_cpu(new_ilb); | |
4737 | return 0; | |
4738 | } | |
46cb4b7c | 4739 | return 1; |
e790fb0b | 4740 | } |
46cb4b7c | 4741 | } else { |
7d1e6a9b | 4742 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4743 | return 0; |
4744 | ||
7d1e6a9b | 4745 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4746 | |
4747 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4748 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4749 | BUG(); | |
4750 | } | |
4751 | return 0; | |
4752 | } | |
4753 | #endif | |
4754 | ||
4755 | static DEFINE_SPINLOCK(balancing); | |
4756 | ||
4757 | /* | |
7835b98b CL |
4758 | * It checks each scheduling domain to see if it is due to be balanced, |
4759 | * and initiates a balancing operation if so. | |
4760 | * | |
4761 | * Balancing parameters are set up in arch_init_sched_domains. | |
4762 | */ | |
a9957449 | 4763 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4764 | { |
46cb4b7c SS |
4765 | int balance = 1; |
4766 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4767 | unsigned long interval; |
4768 | struct sched_domain *sd; | |
46cb4b7c | 4769 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4770 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4771 | int update_next_balance = 0; |
d07355f5 | 4772 | int need_serialize; |
1da177e4 | 4773 | |
46cb4b7c | 4774 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4775 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4776 | continue; | |
4777 | ||
4778 | interval = sd->balance_interval; | |
d15bcfdb | 4779 | if (idle != CPU_IDLE) |
1da177e4 LT |
4780 | interval *= sd->busy_factor; |
4781 | ||
4782 | /* scale ms to jiffies */ | |
4783 | interval = msecs_to_jiffies(interval); | |
4784 | if (unlikely(!interval)) | |
4785 | interval = 1; | |
dd41f596 IM |
4786 | if (interval > HZ*NR_CPUS/10) |
4787 | interval = HZ*NR_CPUS/10; | |
4788 | ||
d07355f5 | 4789 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4790 | |
d07355f5 | 4791 | if (need_serialize) { |
08c183f3 CL |
4792 | if (!spin_trylock(&balancing)) |
4793 | goto out; | |
4794 | } | |
4795 | ||
c9819f45 | 4796 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4797 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4798 | /* |
4799 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4800 | * longer idle, or one of our SMT siblings is |
4801 | * not idle. | |
4802 | */ | |
d15bcfdb | 4803 | idle = CPU_NOT_IDLE; |
1da177e4 | 4804 | } |
1bd77f2d | 4805 | sd->last_balance = jiffies; |
1da177e4 | 4806 | } |
d07355f5 | 4807 | if (need_serialize) |
08c183f3 CL |
4808 | spin_unlock(&balancing); |
4809 | out: | |
f549da84 | 4810 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4811 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4812 | update_next_balance = 1; |
4813 | } | |
783609c6 SS |
4814 | |
4815 | /* | |
4816 | * Stop the load balance at this level. There is another | |
4817 | * CPU in our sched group which is doing load balancing more | |
4818 | * actively. | |
4819 | */ | |
4820 | if (!balance) | |
4821 | break; | |
1da177e4 | 4822 | } |
f549da84 SS |
4823 | |
4824 | /* | |
4825 | * next_balance will be updated only when there is a need. | |
4826 | * When the cpu is attached to null domain for ex, it will not be | |
4827 | * updated. | |
4828 | */ | |
4829 | if (likely(update_next_balance)) | |
4830 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4831 | } |
4832 | ||
4833 | /* | |
4834 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4835 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4836 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4837 | */ | |
4838 | static void run_rebalance_domains(struct softirq_action *h) | |
4839 | { | |
dd41f596 IM |
4840 | int this_cpu = smp_processor_id(); |
4841 | struct rq *this_rq = cpu_rq(this_cpu); | |
4842 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4843 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4844 | |
dd41f596 | 4845 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4846 | |
4847 | #ifdef CONFIG_NO_HZ | |
4848 | /* | |
4849 | * If this cpu is the owner for idle load balancing, then do the | |
4850 | * balancing on behalf of the other idle cpus whose ticks are | |
4851 | * stopped. | |
4852 | */ | |
dd41f596 IM |
4853 | if (this_rq->idle_at_tick && |
4854 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4855 | struct rq *rq; |
4856 | int balance_cpu; | |
4857 | ||
7d1e6a9b RR |
4858 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4859 | if (balance_cpu == this_cpu) | |
4860 | continue; | |
4861 | ||
46cb4b7c SS |
4862 | /* |
4863 | * If this cpu gets work to do, stop the load balancing | |
4864 | * work being done for other cpus. Next load | |
4865 | * balancing owner will pick it up. | |
4866 | */ | |
4867 | if (need_resched()) | |
4868 | break; | |
4869 | ||
de0cf899 | 4870 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4871 | |
4872 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4873 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4874 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4875 | } |
4876 | } | |
4877 | #endif | |
4878 | } | |
4879 | ||
8a0be9ef FW |
4880 | static inline int on_null_domain(int cpu) |
4881 | { | |
4882 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4883 | } | |
4884 | ||
46cb4b7c SS |
4885 | /* |
4886 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4887 | * | |
4888 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4889 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4890 | * if the whole system is idle. | |
4891 | */ | |
dd41f596 | 4892 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4893 | { |
46cb4b7c SS |
4894 | #ifdef CONFIG_NO_HZ |
4895 | /* | |
4896 | * If we were in the nohz mode recently and busy at the current | |
4897 | * scheduler tick, then check if we need to nominate new idle | |
4898 | * load balancer. | |
4899 | */ | |
4900 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4901 | rq->in_nohz_recently = 0; | |
4902 | ||
4903 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4904 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4905 | atomic_set(&nohz.load_balancer, -1); |
4906 | } | |
4907 | ||
4908 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4909 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4910 | |
434d53b0 | 4911 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4912 | resched_cpu(ilb); |
4913 | } | |
4914 | } | |
4915 | ||
4916 | /* | |
4917 | * If this cpu is idle and doing idle load balancing for all the | |
4918 | * cpus with ticks stopped, is it time for that to stop? | |
4919 | */ | |
4920 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4921 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4922 | resched_cpu(cpu); |
4923 | return; | |
4924 | } | |
4925 | ||
4926 | /* | |
4927 | * If this cpu is idle and the idle load balancing is done by | |
4928 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4929 | */ | |
4930 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4931 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4932 | return; |
4933 | #endif | |
8a0be9ef FW |
4934 | /* Don't need to rebalance while attached to NULL domain */ |
4935 | if (time_after_eq(jiffies, rq->next_balance) && | |
4936 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4937 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4938 | } |
dd41f596 IM |
4939 | |
4940 | #else /* CONFIG_SMP */ | |
4941 | ||
1da177e4 LT |
4942 | /* |
4943 | * on UP we do not need to balance between CPUs: | |
4944 | */ | |
70b97a7f | 4945 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4946 | { |
4947 | } | |
dd41f596 | 4948 | |
1da177e4 LT |
4949 | #endif |
4950 | ||
1da177e4 LT |
4951 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4952 | ||
4953 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4954 | ||
4955 | /* | |
c5f8d995 | 4956 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4957 | * @p in case that task is currently running. |
c5f8d995 HS |
4958 | * |
4959 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4960 | */ |
c5f8d995 HS |
4961 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4962 | { | |
4963 | u64 ns = 0; | |
4964 | ||
4965 | if (task_current(rq, p)) { | |
4966 | update_rq_clock(rq); | |
4967 | ns = rq->clock - p->se.exec_start; | |
4968 | if ((s64)ns < 0) | |
4969 | ns = 0; | |
4970 | } | |
4971 | ||
4972 | return ns; | |
4973 | } | |
4974 | ||
bb34d92f | 4975 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4976 | { |
1da177e4 | 4977 | unsigned long flags; |
41b86e9c | 4978 | struct rq *rq; |
bb34d92f | 4979 | u64 ns = 0; |
48f24c4d | 4980 | |
41b86e9c | 4981 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4982 | ns = do_task_delta_exec(p, rq); |
4983 | task_rq_unlock(rq, &flags); | |
1508487e | 4984 | |
c5f8d995 HS |
4985 | return ns; |
4986 | } | |
f06febc9 | 4987 | |
c5f8d995 HS |
4988 | /* |
4989 | * Return accounted runtime for the task. | |
4990 | * In case the task is currently running, return the runtime plus current's | |
4991 | * pending runtime that have not been accounted yet. | |
4992 | */ | |
4993 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4994 | { | |
4995 | unsigned long flags; | |
4996 | struct rq *rq; | |
4997 | u64 ns = 0; | |
4998 | ||
4999 | rq = task_rq_lock(p, &flags); | |
5000 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
5001 | task_rq_unlock(rq, &flags); | |
5002 | ||
5003 | return ns; | |
5004 | } | |
48f24c4d | 5005 | |
c5f8d995 HS |
5006 | /* |
5007 | * Return sum_exec_runtime for the thread group. | |
5008 | * In case the task is currently running, return the sum plus current's | |
5009 | * pending runtime that have not been accounted yet. | |
5010 | * | |
5011 | * Note that the thread group might have other running tasks as well, | |
5012 | * so the return value not includes other pending runtime that other | |
5013 | * running tasks might have. | |
5014 | */ | |
5015 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
5016 | { | |
5017 | struct task_cputime totals; | |
5018 | unsigned long flags; | |
5019 | struct rq *rq; | |
5020 | u64 ns; | |
5021 | ||
5022 | rq = task_rq_lock(p, &flags); | |
5023 | thread_group_cputime(p, &totals); | |
5024 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 5025 | task_rq_unlock(rq, &flags); |
48f24c4d | 5026 | |
1da177e4 LT |
5027 | return ns; |
5028 | } | |
5029 | ||
1da177e4 LT |
5030 | /* |
5031 | * Account user cpu time to a process. | |
5032 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 5033 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 5034 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 5035 | */ |
457533a7 MS |
5036 | void account_user_time(struct task_struct *p, cputime_t cputime, |
5037 | cputime_t cputime_scaled) | |
1da177e4 LT |
5038 | { |
5039 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5040 | cputime64_t tmp; | |
5041 | ||
457533a7 | 5042 | /* Add user time to process. */ |
1da177e4 | 5043 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5044 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5045 | account_group_user_time(p, cputime); |
1da177e4 LT |
5046 | |
5047 | /* Add user time to cpustat. */ | |
5048 | tmp = cputime_to_cputime64(cputime); | |
5049 | if (TASK_NICE(p) > 0) | |
5050 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5051 | else | |
5052 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5053 | |
5054 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5055 | /* Account for user time used */ |
5056 | acct_update_integrals(p); | |
1da177e4 LT |
5057 | } |
5058 | ||
94886b84 LV |
5059 | /* |
5060 | * Account guest cpu time to a process. | |
5061 | * @p: the process that the cpu time gets accounted to | |
5062 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5063 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5064 | */ |
457533a7 MS |
5065 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5066 | cputime_t cputime_scaled) | |
94886b84 LV |
5067 | { |
5068 | cputime64_t tmp; | |
5069 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5070 | ||
5071 | tmp = cputime_to_cputime64(cputime); | |
5072 | ||
457533a7 | 5073 | /* Add guest time to process. */ |
94886b84 | 5074 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5075 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5076 | account_group_user_time(p, cputime); |
94886b84 LV |
5077 | p->gtime = cputime_add(p->gtime, cputime); |
5078 | ||
457533a7 | 5079 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
5080 | if (TASK_NICE(p) > 0) { |
5081 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5082 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
5083 | } else { | |
5084 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
5085 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5086 | } | |
94886b84 LV |
5087 | } |
5088 | ||
1da177e4 LT |
5089 | /* |
5090 | * Account system cpu time to a process. | |
5091 | * @p: the process that the cpu time gets accounted to | |
5092 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5093 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5094 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5095 | */ |
5096 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5097 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5098 | { |
5099 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5100 | cputime64_t tmp; |
5101 | ||
983ed7a6 | 5102 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5103 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5104 | return; |
5105 | } | |
94886b84 | 5106 | |
457533a7 | 5107 | /* Add system time to process. */ |
1da177e4 | 5108 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5109 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5110 | account_group_system_time(p, cputime); |
1da177e4 LT |
5111 | |
5112 | /* Add system time to cpustat. */ | |
5113 | tmp = cputime_to_cputime64(cputime); | |
5114 | if (hardirq_count() - hardirq_offset) | |
5115 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5116 | else if (softirq_count()) | |
5117 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5118 | else |
79741dd3 MS |
5119 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5120 | ||
ef12fefa BR |
5121 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5122 | ||
1da177e4 LT |
5123 | /* Account for system time used */ |
5124 | acct_update_integrals(p); | |
1da177e4 LT |
5125 | } |
5126 | ||
c66f08be | 5127 | /* |
1da177e4 | 5128 | * Account for involuntary wait time. |
1da177e4 | 5129 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5130 | */ |
79741dd3 | 5131 | void account_steal_time(cputime_t cputime) |
c66f08be | 5132 | { |
79741dd3 MS |
5133 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5134 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5135 | ||
5136 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5137 | } |
5138 | ||
1da177e4 | 5139 | /* |
79741dd3 MS |
5140 | * Account for idle time. |
5141 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5142 | */ |
79741dd3 | 5143 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5144 | { |
5145 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5146 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5147 | struct rq *rq = this_rq(); |
1da177e4 | 5148 | |
79741dd3 MS |
5149 | if (atomic_read(&rq->nr_iowait) > 0) |
5150 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5151 | else | |
5152 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5153 | } |
5154 | ||
79741dd3 MS |
5155 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5156 | ||
5157 | /* | |
5158 | * Account a single tick of cpu time. | |
5159 | * @p: the process that the cpu time gets accounted to | |
5160 | * @user_tick: indicates if the tick is a user or a system tick | |
5161 | */ | |
5162 | void account_process_tick(struct task_struct *p, int user_tick) | |
5163 | { | |
a42548a1 | 5164 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
5165 | struct rq *rq = this_rq(); |
5166 | ||
5167 | if (user_tick) | |
a42548a1 | 5168 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 5169 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 5170 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
5171 | one_jiffy_scaled); |
5172 | else | |
a42548a1 | 5173 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
5174 | } |
5175 | ||
5176 | /* | |
5177 | * Account multiple ticks of steal time. | |
5178 | * @p: the process from which the cpu time has been stolen | |
5179 | * @ticks: number of stolen ticks | |
5180 | */ | |
5181 | void account_steal_ticks(unsigned long ticks) | |
5182 | { | |
5183 | account_steal_time(jiffies_to_cputime(ticks)); | |
5184 | } | |
5185 | ||
5186 | /* | |
5187 | * Account multiple ticks of idle time. | |
5188 | * @ticks: number of stolen ticks | |
5189 | */ | |
5190 | void account_idle_ticks(unsigned long ticks) | |
5191 | { | |
5192 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5193 | } |
5194 | ||
79741dd3 MS |
5195 | #endif |
5196 | ||
49048622 BS |
5197 | /* |
5198 | * Use precise platform statistics if available: | |
5199 | */ | |
5200 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 5201 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5202 | { |
d99ca3b9 HS |
5203 | *ut = p->utime; |
5204 | *st = p->stime; | |
49048622 BS |
5205 | } |
5206 | ||
0cf55e1e | 5207 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5208 | { |
0cf55e1e HS |
5209 | struct task_cputime cputime; |
5210 | ||
5211 | thread_group_cputime(p, &cputime); | |
5212 | ||
5213 | *ut = cputime.utime; | |
5214 | *st = cputime.stime; | |
49048622 BS |
5215 | } |
5216 | #else | |
761b1d26 HS |
5217 | |
5218 | #ifndef nsecs_to_cputime | |
b7b20df9 | 5219 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
5220 | #endif |
5221 | ||
d180c5bc | 5222 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 5223 | { |
d99ca3b9 | 5224 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
5225 | |
5226 | /* | |
5227 | * Use CFS's precise accounting: | |
5228 | */ | |
d180c5bc | 5229 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
5230 | |
5231 | if (total) { | |
d180c5bc HS |
5232 | u64 temp; |
5233 | ||
5234 | temp = (u64)(rtime * utime); | |
49048622 | 5235 | do_div(temp, total); |
d180c5bc HS |
5236 | utime = (cputime_t)temp; |
5237 | } else | |
5238 | utime = rtime; | |
49048622 | 5239 | |
d180c5bc HS |
5240 | /* |
5241 | * Compare with previous values, to keep monotonicity: | |
5242 | */ | |
761b1d26 | 5243 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 5244 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 5245 | |
d99ca3b9 HS |
5246 | *ut = p->prev_utime; |
5247 | *st = p->prev_stime; | |
49048622 BS |
5248 | } |
5249 | ||
0cf55e1e HS |
5250 | /* |
5251 | * Must be called with siglock held. | |
5252 | */ | |
5253 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 5254 | { |
0cf55e1e HS |
5255 | struct signal_struct *sig = p->signal; |
5256 | struct task_cputime cputime; | |
5257 | cputime_t rtime, utime, total; | |
49048622 | 5258 | |
0cf55e1e | 5259 | thread_group_cputime(p, &cputime); |
49048622 | 5260 | |
0cf55e1e HS |
5261 | total = cputime_add(cputime.utime, cputime.stime); |
5262 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 5263 | |
0cf55e1e HS |
5264 | if (total) { |
5265 | u64 temp; | |
49048622 | 5266 | |
0cf55e1e HS |
5267 | temp = (u64)(rtime * cputime.utime); |
5268 | do_div(temp, total); | |
5269 | utime = (cputime_t)temp; | |
5270 | } else | |
5271 | utime = rtime; | |
5272 | ||
5273 | sig->prev_utime = max(sig->prev_utime, utime); | |
5274 | sig->prev_stime = max(sig->prev_stime, | |
5275 | cputime_sub(rtime, sig->prev_utime)); | |
5276 | ||
5277 | *ut = sig->prev_utime; | |
5278 | *st = sig->prev_stime; | |
49048622 | 5279 | } |
49048622 | 5280 | #endif |
49048622 | 5281 | |
7835b98b CL |
5282 | /* |
5283 | * This function gets called by the timer code, with HZ frequency. | |
5284 | * We call it with interrupts disabled. | |
5285 | * | |
5286 | * It also gets called by the fork code, when changing the parent's | |
5287 | * timeslices. | |
5288 | */ | |
5289 | void scheduler_tick(void) | |
5290 | { | |
7835b98b CL |
5291 | int cpu = smp_processor_id(); |
5292 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5293 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5294 | |
5295 | sched_clock_tick(); | |
dd41f596 | 5296 | |
05fa785c | 5297 | raw_spin_lock(&rq->lock); |
3e51f33f | 5298 | update_rq_clock(rq); |
f1a438d8 | 5299 | update_cpu_load(rq); |
fa85ae24 | 5300 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 5301 | raw_spin_unlock(&rq->lock); |
7835b98b | 5302 | |
cdd6c482 | 5303 | perf_event_task_tick(curr, cpu); |
e220d2dc | 5304 | |
e418e1c2 | 5305 | #ifdef CONFIG_SMP |
dd41f596 IM |
5306 | rq->idle_at_tick = idle_cpu(cpu); |
5307 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5308 | #endif |
1da177e4 LT |
5309 | } |
5310 | ||
132380a0 | 5311 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5312 | { |
5313 | if (in_lock_functions(addr)) { | |
5314 | addr = CALLER_ADDR2; | |
5315 | if (in_lock_functions(addr)) | |
5316 | addr = CALLER_ADDR3; | |
5317 | } | |
5318 | return addr; | |
5319 | } | |
1da177e4 | 5320 | |
7e49fcce SR |
5321 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5322 | defined(CONFIG_PREEMPT_TRACER)) | |
5323 | ||
43627582 | 5324 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5325 | { |
6cd8a4bb | 5326 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5327 | /* |
5328 | * Underflow? | |
5329 | */ | |
9a11b49a IM |
5330 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5331 | return; | |
6cd8a4bb | 5332 | #endif |
1da177e4 | 5333 | preempt_count() += val; |
6cd8a4bb | 5334 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5335 | /* |
5336 | * Spinlock count overflowing soon? | |
5337 | */ | |
33859f7f MOS |
5338 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5339 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5340 | #endif |
5341 | if (preempt_count() == val) | |
5342 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5343 | } |
5344 | EXPORT_SYMBOL(add_preempt_count); | |
5345 | ||
43627582 | 5346 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5347 | { |
6cd8a4bb | 5348 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5349 | /* |
5350 | * Underflow? | |
5351 | */ | |
01e3eb82 | 5352 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5353 | return; |
1da177e4 LT |
5354 | /* |
5355 | * Is the spinlock portion underflowing? | |
5356 | */ | |
9a11b49a IM |
5357 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5358 | !(preempt_count() & PREEMPT_MASK))) | |
5359 | return; | |
6cd8a4bb | 5360 | #endif |
9a11b49a | 5361 | |
6cd8a4bb SR |
5362 | if (preempt_count() == val) |
5363 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5364 | preempt_count() -= val; |
5365 | } | |
5366 | EXPORT_SYMBOL(sub_preempt_count); | |
5367 | ||
5368 | #endif | |
5369 | ||
5370 | /* | |
dd41f596 | 5371 | * Print scheduling while atomic bug: |
1da177e4 | 5372 | */ |
dd41f596 | 5373 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5374 | { |
838225b4 SS |
5375 | struct pt_regs *regs = get_irq_regs(); |
5376 | ||
663997d4 JP |
5377 | pr_err("BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5378 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5379 | |
dd41f596 | 5380 | debug_show_held_locks(prev); |
e21f5b15 | 5381 | print_modules(); |
dd41f596 IM |
5382 | if (irqs_disabled()) |
5383 | print_irqtrace_events(prev); | |
838225b4 SS |
5384 | |
5385 | if (regs) | |
5386 | show_regs(regs); | |
5387 | else | |
5388 | dump_stack(); | |
dd41f596 | 5389 | } |
1da177e4 | 5390 | |
dd41f596 IM |
5391 | /* |
5392 | * Various schedule()-time debugging checks and statistics: | |
5393 | */ | |
5394 | static inline void schedule_debug(struct task_struct *prev) | |
5395 | { | |
1da177e4 | 5396 | /* |
41a2d6cf | 5397 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5398 | * schedule() atomically, we ignore that path for now. |
5399 | * Otherwise, whine if we are scheduling when we should not be. | |
5400 | */ | |
3f33a7ce | 5401 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5402 | __schedule_bug(prev); |
5403 | ||
1da177e4 LT |
5404 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5405 | ||
2d72376b | 5406 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5407 | #ifdef CONFIG_SCHEDSTATS |
5408 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5409 | schedstat_inc(this_rq(), bkl_count); |
5410 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5411 | } |
5412 | #endif | |
dd41f596 IM |
5413 | } |
5414 | ||
6cecd084 | 5415 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 5416 | { |
6cecd084 PZ |
5417 | if (prev->state == TASK_RUNNING) { |
5418 | u64 runtime = prev->se.sum_exec_runtime; | |
df1c99d4 | 5419 | |
6cecd084 PZ |
5420 | runtime -= prev->se.prev_sum_exec_runtime; |
5421 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
df1c99d4 MG |
5422 | |
5423 | /* | |
5424 | * In order to avoid avg_overlap growing stale when we are | |
5425 | * indeed overlapping and hence not getting put to sleep, grow | |
5426 | * the avg_overlap on preemption. | |
5427 | * | |
5428 | * We use the average preemption runtime because that | |
5429 | * correlates to the amount of cache footprint a task can | |
5430 | * build up. | |
5431 | */ | |
6cecd084 | 5432 | update_avg(&prev->se.avg_overlap, runtime); |
df1c99d4 | 5433 | } |
6cecd084 | 5434 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
5435 | } |
5436 | ||
dd41f596 IM |
5437 | /* |
5438 | * Pick up the highest-prio task: | |
5439 | */ | |
5440 | static inline struct task_struct * | |
b67802ea | 5441 | pick_next_task(struct rq *rq) |
dd41f596 | 5442 | { |
5522d5d5 | 5443 | const struct sched_class *class; |
dd41f596 | 5444 | struct task_struct *p; |
1da177e4 LT |
5445 | |
5446 | /* | |
dd41f596 IM |
5447 | * Optimization: we know that if all tasks are in |
5448 | * the fair class we can call that function directly: | |
1da177e4 | 5449 | */ |
dd41f596 | 5450 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5451 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5452 | if (likely(p)) |
5453 | return p; | |
1da177e4 LT |
5454 | } |
5455 | ||
dd41f596 IM |
5456 | class = sched_class_highest; |
5457 | for ( ; ; ) { | |
fb8d4724 | 5458 | p = class->pick_next_task(rq); |
dd41f596 IM |
5459 | if (p) |
5460 | return p; | |
5461 | /* | |
5462 | * Will never be NULL as the idle class always | |
5463 | * returns a non-NULL p: | |
5464 | */ | |
5465 | class = class->next; | |
5466 | } | |
5467 | } | |
1da177e4 | 5468 | |
dd41f596 IM |
5469 | /* |
5470 | * schedule() is the main scheduler function. | |
5471 | */ | |
ff743345 | 5472 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5473 | { |
5474 | struct task_struct *prev, *next; | |
67ca7bde | 5475 | unsigned long *switch_count; |
dd41f596 | 5476 | struct rq *rq; |
31656519 | 5477 | int cpu; |
dd41f596 | 5478 | |
ff743345 PZ |
5479 | need_resched: |
5480 | preempt_disable(); | |
dd41f596 IM |
5481 | cpu = smp_processor_id(); |
5482 | rq = cpu_rq(cpu); | |
d6714c22 | 5483 | rcu_sched_qs(cpu); |
dd41f596 IM |
5484 | prev = rq->curr; |
5485 | switch_count = &prev->nivcsw; | |
5486 | ||
5487 | release_kernel_lock(prev); | |
5488 | need_resched_nonpreemptible: | |
5489 | ||
5490 | schedule_debug(prev); | |
1da177e4 | 5491 | |
31656519 | 5492 | if (sched_feat(HRTICK)) |
f333fdc9 | 5493 | hrtick_clear(rq); |
8f4d37ec | 5494 | |
05fa785c | 5495 | raw_spin_lock_irq(&rq->lock); |
3e51f33f | 5496 | update_rq_clock(rq); |
1e819950 | 5497 | clear_tsk_need_resched(prev); |
1da177e4 | 5498 | |
1da177e4 | 5499 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5500 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5501 | prev->state = TASK_RUNNING; |
16882c1e | 5502 | else |
2e1cb74a | 5503 | deactivate_task(rq, prev, 1); |
dd41f596 | 5504 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5505 | } |
5506 | ||
3f029d3c | 5507 | pre_schedule(rq, prev); |
f65eda4f | 5508 | |
dd41f596 | 5509 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5510 | idle_balance(cpu, rq); |
1da177e4 | 5511 | |
df1c99d4 | 5512 | put_prev_task(rq, prev); |
b67802ea | 5513 | next = pick_next_task(rq); |
1da177e4 | 5514 | |
1da177e4 | 5515 | if (likely(prev != next)) { |
673a90a1 | 5516 | sched_info_switch(prev, next); |
cdd6c482 | 5517 | perf_event_task_sched_out(prev, next, cpu); |
673a90a1 | 5518 | |
1da177e4 LT |
5519 | rq->nr_switches++; |
5520 | rq->curr = next; | |
5521 | ++*switch_count; | |
5522 | ||
dd41f596 | 5523 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5524 | /* |
5525 | * the context switch might have flipped the stack from under | |
5526 | * us, hence refresh the local variables. | |
5527 | */ | |
5528 | cpu = smp_processor_id(); | |
5529 | rq = cpu_rq(cpu); | |
1da177e4 | 5530 | } else |
05fa785c | 5531 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5532 | |
3f029d3c | 5533 | post_schedule(rq); |
1da177e4 | 5534 | |
8f4d37ec | 5535 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5536 | goto need_resched_nonpreemptible; |
8f4d37ec | 5537 | |
1da177e4 | 5538 | preempt_enable_no_resched(); |
ff743345 | 5539 | if (need_resched()) |
1da177e4 LT |
5540 | goto need_resched; |
5541 | } | |
1da177e4 LT |
5542 | EXPORT_SYMBOL(schedule); |
5543 | ||
c08f7829 | 5544 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
5545 | /* |
5546 | * Look out! "owner" is an entirely speculative pointer | |
5547 | * access and not reliable. | |
5548 | */ | |
5549 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5550 | { | |
5551 | unsigned int cpu; | |
5552 | struct rq *rq; | |
5553 | ||
5554 | if (!sched_feat(OWNER_SPIN)) | |
5555 | return 0; | |
5556 | ||
5557 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5558 | /* | |
5559 | * Need to access the cpu field knowing that | |
5560 | * DEBUG_PAGEALLOC could have unmapped it if | |
5561 | * the mutex owner just released it and exited. | |
5562 | */ | |
5563 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5564 | goto out; | |
5565 | #else | |
5566 | cpu = owner->cpu; | |
5567 | #endif | |
5568 | ||
5569 | /* | |
5570 | * Even if the access succeeded (likely case), | |
5571 | * the cpu field may no longer be valid. | |
5572 | */ | |
5573 | if (cpu >= nr_cpumask_bits) | |
5574 | goto out; | |
5575 | ||
5576 | /* | |
5577 | * We need to validate that we can do a | |
5578 | * get_cpu() and that we have the percpu area. | |
5579 | */ | |
5580 | if (!cpu_online(cpu)) | |
5581 | goto out; | |
5582 | ||
5583 | rq = cpu_rq(cpu); | |
5584 | ||
5585 | for (;;) { | |
5586 | /* | |
5587 | * Owner changed, break to re-assess state. | |
5588 | */ | |
5589 | if (lock->owner != owner) | |
5590 | break; | |
5591 | ||
5592 | /* | |
5593 | * Is that owner really running on that cpu? | |
5594 | */ | |
5595 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5596 | return 0; | |
5597 | ||
5598 | cpu_relax(); | |
5599 | } | |
5600 | out: | |
5601 | return 1; | |
5602 | } | |
5603 | #endif | |
5604 | ||
1da177e4 LT |
5605 | #ifdef CONFIG_PREEMPT |
5606 | /* | |
2ed6e34f | 5607 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5608 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5609 | * occur there and call schedule directly. |
5610 | */ | |
5611 | asmlinkage void __sched preempt_schedule(void) | |
5612 | { | |
5613 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5614 | |
1da177e4 LT |
5615 | /* |
5616 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5617 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5618 | */ |
beed33a8 | 5619 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5620 | return; |
5621 | ||
3a5c359a AK |
5622 | do { |
5623 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5624 | schedule(); |
3a5c359a | 5625 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5626 | |
3a5c359a AK |
5627 | /* |
5628 | * Check again in case we missed a preemption opportunity | |
5629 | * between schedule and now. | |
5630 | */ | |
5631 | barrier(); | |
5ed0cec0 | 5632 | } while (need_resched()); |
1da177e4 | 5633 | } |
1da177e4 LT |
5634 | EXPORT_SYMBOL(preempt_schedule); |
5635 | ||
5636 | /* | |
2ed6e34f | 5637 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5638 | * off of irq context. |
5639 | * Note, that this is called and return with irqs disabled. This will | |
5640 | * protect us against recursive calling from irq. | |
5641 | */ | |
5642 | asmlinkage void __sched preempt_schedule_irq(void) | |
5643 | { | |
5644 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5645 | |
2ed6e34f | 5646 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5647 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5648 | ||
3a5c359a AK |
5649 | do { |
5650 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5651 | local_irq_enable(); |
5652 | schedule(); | |
5653 | local_irq_disable(); | |
3a5c359a | 5654 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5655 | |
3a5c359a AK |
5656 | /* |
5657 | * Check again in case we missed a preemption opportunity | |
5658 | * between schedule and now. | |
5659 | */ | |
5660 | barrier(); | |
5ed0cec0 | 5661 | } while (need_resched()); |
1da177e4 LT |
5662 | } |
5663 | ||
5664 | #endif /* CONFIG_PREEMPT */ | |
5665 | ||
63859d4f | 5666 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5667 | void *key) |
1da177e4 | 5668 | { |
63859d4f | 5669 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5670 | } |
1da177e4 LT |
5671 | EXPORT_SYMBOL(default_wake_function); |
5672 | ||
5673 | /* | |
41a2d6cf IM |
5674 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5675 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5676 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5677 | * | |
5678 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5679 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5680 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5681 | */ | |
78ddb08f | 5682 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 5683 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 5684 | { |
2e45874c | 5685 | wait_queue_t *curr, *next; |
1da177e4 | 5686 | |
2e45874c | 5687 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5688 | unsigned flags = curr->flags; |
5689 | ||
63859d4f | 5690 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 5691 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5692 | break; |
5693 | } | |
5694 | } | |
5695 | ||
5696 | /** | |
5697 | * __wake_up - wake up threads blocked on a waitqueue. | |
5698 | * @q: the waitqueue | |
5699 | * @mode: which threads | |
5700 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5701 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5702 | * |
5703 | * It may be assumed that this function implies a write memory barrier before | |
5704 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5705 | */ |
7ad5b3a5 | 5706 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5707 | int nr_exclusive, void *key) |
1da177e4 LT |
5708 | { |
5709 | unsigned long flags; | |
5710 | ||
5711 | spin_lock_irqsave(&q->lock, flags); | |
5712 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5713 | spin_unlock_irqrestore(&q->lock, flags); | |
5714 | } | |
1da177e4 LT |
5715 | EXPORT_SYMBOL(__wake_up); |
5716 | ||
5717 | /* | |
5718 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5719 | */ | |
7ad5b3a5 | 5720 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5721 | { |
5722 | __wake_up_common(q, mode, 1, 0, NULL); | |
5723 | } | |
5724 | ||
4ede816a DL |
5725 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5726 | { | |
5727 | __wake_up_common(q, mode, 1, 0, key); | |
5728 | } | |
5729 | ||
1da177e4 | 5730 | /** |
4ede816a | 5731 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5732 | * @q: the waitqueue |
5733 | * @mode: which threads | |
5734 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5735 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5736 | * |
5737 | * The sync wakeup differs that the waker knows that it will schedule | |
5738 | * away soon, so while the target thread will be woken up, it will not | |
5739 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5740 | * with each other. This can prevent needless bouncing between CPUs. | |
5741 | * | |
5742 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5743 | * |
5744 | * It may be assumed that this function implies a write memory barrier before | |
5745 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5746 | */ |
4ede816a DL |
5747 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5748 | int nr_exclusive, void *key) | |
1da177e4 LT |
5749 | { |
5750 | unsigned long flags; | |
7d478721 | 5751 | int wake_flags = WF_SYNC; |
1da177e4 LT |
5752 | |
5753 | if (unlikely(!q)) | |
5754 | return; | |
5755 | ||
5756 | if (unlikely(!nr_exclusive)) | |
7d478721 | 5757 | wake_flags = 0; |
1da177e4 LT |
5758 | |
5759 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 5760 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
5761 | spin_unlock_irqrestore(&q->lock, flags); |
5762 | } | |
4ede816a DL |
5763 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5764 | ||
5765 | /* | |
5766 | * __wake_up_sync - see __wake_up_sync_key() | |
5767 | */ | |
5768 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5769 | { | |
5770 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5771 | } | |
1da177e4 LT |
5772 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5773 | ||
65eb3dc6 KD |
5774 | /** |
5775 | * complete: - signals a single thread waiting on this completion | |
5776 | * @x: holds the state of this particular completion | |
5777 | * | |
5778 | * This will wake up a single thread waiting on this completion. Threads will be | |
5779 | * awakened in the same order in which they were queued. | |
5780 | * | |
5781 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5782 | * |
5783 | * It may be assumed that this function implies a write memory barrier before | |
5784 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5785 | */ |
b15136e9 | 5786 | void complete(struct completion *x) |
1da177e4 LT |
5787 | { |
5788 | unsigned long flags; | |
5789 | ||
5790 | spin_lock_irqsave(&x->wait.lock, flags); | |
5791 | x->done++; | |
d9514f6c | 5792 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5793 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5794 | } | |
5795 | EXPORT_SYMBOL(complete); | |
5796 | ||
65eb3dc6 KD |
5797 | /** |
5798 | * complete_all: - signals all threads waiting on this completion | |
5799 | * @x: holds the state of this particular completion | |
5800 | * | |
5801 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5802 | * |
5803 | * It may be assumed that this function implies a write memory barrier before | |
5804 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5805 | */ |
b15136e9 | 5806 | void complete_all(struct completion *x) |
1da177e4 LT |
5807 | { |
5808 | unsigned long flags; | |
5809 | ||
5810 | spin_lock_irqsave(&x->wait.lock, flags); | |
5811 | x->done += UINT_MAX/2; | |
d9514f6c | 5812 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5813 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5814 | } | |
5815 | EXPORT_SYMBOL(complete_all); | |
5816 | ||
8cbbe86d AK |
5817 | static inline long __sched |
5818 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5819 | { |
1da177e4 LT |
5820 | if (!x->done) { |
5821 | DECLARE_WAITQUEUE(wait, current); | |
5822 | ||
5823 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5824 | __add_wait_queue_tail(&x->wait, &wait); | |
5825 | do { | |
94d3d824 | 5826 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5827 | timeout = -ERESTARTSYS; |
5828 | break; | |
8cbbe86d AK |
5829 | } |
5830 | __set_current_state(state); | |
1da177e4 LT |
5831 | spin_unlock_irq(&x->wait.lock); |
5832 | timeout = schedule_timeout(timeout); | |
5833 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5834 | } while (!x->done && timeout); |
1da177e4 | 5835 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5836 | if (!x->done) |
5837 | return timeout; | |
1da177e4 LT |
5838 | } |
5839 | x->done--; | |
ea71a546 | 5840 | return timeout ?: 1; |
1da177e4 | 5841 | } |
1da177e4 | 5842 | |
8cbbe86d AK |
5843 | static long __sched |
5844 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5845 | { |
1da177e4 LT |
5846 | might_sleep(); |
5847 | ||
5848 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5849 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5850 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5851 | return timeout; |
5852 | } | |
1da177e4 | 5853 | |
65eb3dc6 KD |
5854 | /** |
5855 | * wait_for_completion: - waits for completion of a task | |
5856 | * @x: holds the state of this particular completion | |
5857 | * | |
5858 | * This waits to be signaled for completion of a specific task. It is NOT | |
5859 | * interruptible and there is no timeout. | |
5860 | * | |
5861 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5862 | * and interrupt capability. Also see complete(). | |
5863 | */ | |
b15136e9 | 5864 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5865 | { |
5866 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5867 | } |
8cbbe86d | 5868 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5869 | |
65eb3dc6 KD |
5870 | /** |
5871 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5872 | * @x: holds the state of this particular completion | |
5873 | * @timeout: timeout value in jiffies | |
5874 | * | |
5875 | * This waits for either a completion of a specific task to be signaled or for a | |
5876 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5877 | * interruptible. | |
5878 | */ | |
b15136e9 | 5879 | unsigned long __sched |
8cbbe86d | 5880 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5881 | { |
8cbbe86d | 5882 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5883 | } |
8cbbe86d | 5884 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5885 | |
65eb3dc6 KD |
5886 | /** |
5887 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5888 | * @x: holds the state of this particular completion | |
5889 | * | |
5890 | * This waits for completion of a specific task to be signaled. It is | |
5891 | * interruptible. | |
5892 | */ | |
8cbbe86d | 5893 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5894 | { |
51e97990 AK |
5895 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5896 | if (t == -ERESTARTSYS) | |
5897 | return t; | |
5898 | return 0; | |
0fec171c | 5899 | } |
8cbbe86d | 5900 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5901 | |
65eb3dc6 KD |
5902 | /** |
5903 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5904 | * @x: holds the state of this particular completion | |
5905 | * @timeout: timeout value in jiffies | |
5906 | * | |
5907 | * This waits for either a completion of a specific task to be signaled or for a | |
5908 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5909 | */ | |
b15136e9 | 5910 | unsigned long __sched |
8cbbe86d AK |
5911 | wait_for_completion_interruptible_timeout(struct completion *x, |
5912 | unsigned long timeout) | |
0fec171c | 5913 | { |
8cbbe86d | 5914 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5915 | } |
8cbbe86d | 5916 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5917 | |
65eb3dc6 KD |
5918 | /** |
5919 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5920 | * @x: holds the state of this particular completion | |
5921 | * | |
5922 | * This waits to be signaled for completion of a specific task. It can be | |
5923 | * interrupted by a kill signal. | |
5924 | */ | |
009e577e MW |
5925 | int __sched wait_for_completion_killable(struct completion *x) |
5926 | { | |
5927 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5928 | if (t == -ERESTARTSYS) | |
5929 | return t; | |
5930 | return 0; | |
5931 | } | |
5932 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5933 | ||
be4de352 DC |
5934 | /** |
5935 | * try_wait_for_completion - try to decrement a completion without blocking | |
5936 | * @x: completion structure | |
5937 | * | |
5938 | * Returns: 0 if a decrement cannot be done without blocking | |
5939 | * 1 if a decrement succeeded. | |
5940 | * | |
5941 | * If a completion is being used as a counting completion, | |
5942 | * attempt to decrement the counter without blocking. This | |
5943 | * enables us to avoid waiting if the resource the completion | |
5944 | * is protecting is not available. | |
5945 | */ | |
5946 | bool try_wait_for_completion(struct completion *x) | |
5947 | { | |
7539a3b3 | 5948 | unsigned long flags; |
be4de352 DC |
5949 | int ret = 1; |
5950 | ||
7539a3b3 | 5951 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
5952 | if (!x->done) |
5953 | ret = 0; | |
5954 | else | |
5955 | x->done--; | |
7539a3b3 | 5956 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
5957 | return ret; |
5958 | } | |
5959 | EXPORT_SYMBOL(try_wait_for_completion); | |
5960 | ||
5961 | /** | |
5962 | * completion_done - Test to see if a completion has any waiters | |
5963 | * @x: completion structure | |
5964 | * | |
5965 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5966 | * 1 if there are no waiters. | |
5967 | * | |
5968 | */ | |
5969 | bool completion_done(struct completion *x) | |
5970 | { | |
7539a3b3 | 5971 | unsigned long flags; |
be4de352 DC |
5972 | int ret = 1; |
5973 | ||
7539a3b3 | 5974 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
5975 | if (!x->done) |
5976 | ret = 0; | |
7539a3b3 | 5977 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
5978 | return ret; |
5979 | } | |
5980 | EXPORT_SYMBOL(completion_done); | |
5981 | ||
8cbbe86d AK |
5982 | static long __sched |
5983 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5984 | { |
0fec171c IM |
5985 | unsigned long flags; |
5986 | wait_queue_t wait; | |
5987 | ||
5988 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5989 | |
8cbbe86d | 5990 | __set_current_state(state); |
1da177e4 | 5991 | |
8cbbe86d AK |
5992 | spin_lock_irqsave(&q->lock, flags); |
5993 | __add_wait_queue(q, &wait); | |
5994 | spin_unlock(&q->lock); | |
5995 | timeout = schedule_timeout(timeout); | |
5996 | spin_lock_irq(&q->lock); | |
5997 | __remove_wait_queue(q, &wait); | |
5998 | spin_unlock_irqrestore(&q->lock, flags); | |
5999 | ||
6000 | return timeout; | |
6001 | } | |
6002 | ||
6003 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
6004 | { | |
6005 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 6006 | } |
1da177e4 LT |
6007 | EXPORT_SYMBOL(interruptible_sleep_on); |
6008 | ||
0fec171c | 6009 | long __sched |
95cdf3b7 | 6010 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6011 | { |
8cbbe86d | 6012 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 6013 | } |
1da177e4 LT |
6014 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
6015 | ||
0fec171c | 6016 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 6017 | { |
8cbbe86d | 6018 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 6019 | } |
1da177e4 LT |
6020 | EXPORT_SYMBOL(sleep_on); |
6021 | ||
0fec171c | 6022 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6023 | { |
8cbbe86d | 6024 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 6025 | } |
1da177e4 LT |
6026 | EXPORT_SYMBOL(sleep_on_timeout); |
6027 | ||
b29739f9 IM |
6028 | #ifdef CONFIG_RT_MUTEXES |
6029 | ||
6030 | /* | |
6031 | * rt_mutex_setprio - set the current priority of a task | |
6032 | * @p: task | |
6033 | * @prio: prio value (kernel-internal form) | |
6034 | * | |
6035 | * This function changes the 'effective' priority of a task. It does | |
6036 | * not touch ->normal_prio like __setscheduler(). | |
6037 | * | |
6038 | * Used by the rt_mutex code to implement priority inheritance logic. | |
6039 | */ | |
36c8b586 | 6040 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
6041 | { |
6042 | unsigned long flags; | |
83b699ed | 6043 | int oldprio, on_rq, running; |
70b97a7f | 6044 | struct rq *rq; |
cb469845 | 6045 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
6046 | |
6047 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
6048 | ||
6049 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6050 | update_rq_clock(rq); |
b29739f9 | 6051 | |
d5f9f942 | 6052 | oldprio = p->prio; |
dd41f596 | 6053 | on_rq = p->se.on_rq; |
051a1d1a | 6054 | running = task_current(rq, p); |
0e1f3483 | 6055 | if (on_rq) |
69be72c1 | 6056 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
6057 | if (running) |
6058 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
6059 | |
6060 | if (rt_prio(prio)) | |
6061 | p->sched_class = &rt_sched_class; | |
6062 | else | |
6063 | p->sched_class = &fair_sched_class; | |
6064 | ||
b29739f9 IM |
6065 | p->prio = prio; |
6066 | ||
0e1f3483 HS |
6067 | if (running) |
6068 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 6069 | if (on_rq) { |
8159f87e | 6070 | enqueue_task(rq, p, 0); |
cb469845 SR |
6071 | |
6072 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
6073 | } |
6074 | task_rq_unlock(rq, &flags); | |
6075 | } | |
6076 | ||
6077 | #endif | |
6078 | ||
36c8b586 | 6079 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6080 | { |
dd41f596 | 6081 | int old_prio, delta, on_rq; |
1da177e4 | 6082 | unsigned long flags; |
70b97a7f | 6083 | struct rq *rq; |
1da177e4 LT |
6084 | |
6085 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6086 | return; | |
6087 | /* | |
6088 | * We have to be careful, if called from sys_setpriority(), | |
6089 | * the task might be in the middle of scheduling on another CPU. | |
6090 | */ | |
6091 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6092 | update_rq_clock(rq); |
1da177e4 LT |
6093 | /* |
6094 | * The RT priorities are set via sched_setscheduler(), but we still | |
6095 | * allow the 'normal' nice value to be set - but as expected | |
6096 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6097 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6098 | */ |
e05606d3 | 6099 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6100 | p->static_prio = NICE_TO_PRIO(nice); |
6101 | goto out_unlock; | |
6102 | } | |
dd41f596 | 6103 | on_rq = p->se.on_rq; |
c09595f6 | 6104 | if (on_rq) |
69be72c1 | 6105 | dequeue_task(rq, p, 0); |
1da177e4 | 6106 | |
1da177e4 | 6107 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6108 | set_load_weight(p); |
b29739f9 IM |
6109 | old_prio = p->prio; |
6110 | p->prio = effective_prio(p); | |
6111 | delta = p->prio - old_prio; | |
1da177e4 | 6112 | |
dd41f596 | 6113 | if (on_rq) { |
8159f87e | 6114 | enqueue_task(rq, p, 0); |
1da177e4 | 6115 | /* |
d5f9f942 AM |
6116 | * If the task increased its priority or is running and |
6117 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6118 | */ |
d5f9f942 | 6119 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6120 | resched_task(rq->curr); |
6121 | } | |
6122 | out_unlock: | |
6123 | task_rq_unlock(rq, &flags); | |
6124 | } | |
1da177e4 LT |
6125 | EXPORT_SYMBOL(set_user_nice); |
6126 | ||
e43379f1 MM |
6127 | /* |
6128 | * can_nice - check if a task can reduce its nice value | |
6129 | * @p: task | |
6130 | * @nice: nice value | |
6131 | */ | |
36c8b586 | 6132 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6133 | { |
024f4747 MM |
6134 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6135 | int nice_rlim = 20 - nice; | |
48f24c4d | 6136 | |
e43379f1 MM |
6137 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6138 | capable(CAP_SYS_NICE)); | |
6139 | } | |
6140 | ||
1da177e4 LT |
6141 | #ifdef __ARCH_WANT_SYS_NICE |
6142 | ||
6143 | /* | |
6144 | * sys_nice - change the priority of the current process. | |
6145 | * @increment: priority increment | |
6146 | * | |
6147 | * sys_setpriority is a more generic, but much slower function that | |
6148 | * does similar things. | |
6149 | */ | |
5add95d4 | 6150 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6151 | { |
48f24c4d | 6152 | long nice, retval; |
1da177e4 LT |
6153 | |
6154 | /* | |
6155 | * Setpriority might change our priority at the same moment. | |
6156 | * We don't have to worry. Conceptually one call occurs first | |
6157 | * and we have a single winner. | |
6158 | */ | |
e43379f1 MM |
6159 | if (increment < -40) |
6160 | increment = -40; | |
1da177e4 LT |
6161 | if (increment > 40) |
6162 | increment = 40; | |
6163 | ||
2b8f836f | 6164 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6165 | if (nice < -20) |
6166 | nice = -20; | |
6167 | if (nice > 19) | |
6168 | nice = 19; | |
6169 | ||
e43379f1 MM |
6170 | if (increment < 0 && !can_nice(current, nice)) |
6171 | return -EPERM; | |
6172 | ||
1da177e4 LT |
6173 | retval = security_task_setnice(current, nice); |
6174 | if (retval) | |
6175 | return retval; | |
6176 | ||
6177 | set_user_nice(current, nice); | |
6178 | return 0; | |
6179 | } | |
6180 | ||
6181 | #endif | |
6182 | ||
6183 | /** | |
6184 | * task_prio - return the priority value of a given task. | |
6185 | * @p: the task in question. | |
6186 | * | |
6187 | * This is the priority value as seen by users in /proc. | |
6188 | * RT tasks are offset by -200. Normal tasks are centered | |
6189 | * around 0, value goes from -16 to +15. | |
6190 | */ | |
36c8b586 | 6191 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6192 | { |
6193 | return p->prio - MAX_RT_PRIO; | |
6194 | } | |
6195 | ||
6196 | /** | |
6197 | * task_nice - return the nice value of a given task. | |
6198 | * @p: the task in question. | |
6199 | */ | |
36c8b586 | 6200 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6201 | { |
6202 | return TASK_NICE(p); | |
6203 | } | |
150d8bed | 6204 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6205 | |
6206 | /** | |
6207 | * idle_cpu - is a given cpu idle currently? | |
6208 | * @cpu: the processor in question. | |
6209 | */ | |
6210 | int idle_cpu(int cpu) | |
6211 | { | |
6212 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6213 | } | |
6214 | ||
1da177e4 LT |
6215 | /** |
6216 | * idle_task - return the idle task for a given cpu. | |
6217 | * @cpu: the processor in question. | |
6218 | */ | |
36c8b586 | 6219 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6220 | { |
6221 | return cpu_rq(cpu)->idle; | |
6222 | } | |
6223 | ||
6224 | /** | |
6225 | * find_process_by_pid - find a process with a matching PID value. | |
6226 | * @pid: the pid in question. | |
6227 | */ | |
a9957449 | 6228 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6229 | { |
228ebcbe | 6230 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6231 | } |
6232 | ||
6233 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6234 | static void |
6235 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6236 | { |
dd41f596 | 6237 | BUG_ON(p->se.on_rq); |
48f24c4d | 6238 | |
1da177e4 LT |
6239 | p->policy = policy; |
6240 | p->rt_priority = prio; | |
b29739f9 IM |
6241 | p->normal_prio = normal_prio(p); |
6242 | /* we are holding p->pi_lock already */ | |
6243 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
6244 | if (rt_prio(p->prio)) |
6245 | p->sched_class = &rt_sched_class; | |
6246 | else | |
6247 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 6248 | set_load_weight(p); |
1da177e4 LT |
6249 | } |
6250 | ||
c69e8d9c DH |
6251 | /* |
6252 | * check the target process has a UID that matches the current process's | |
6253 | */ | |
6254 | static bool check_same_owner(struct task_struct *p) | |
6255 | { | |
6256 | const struct cred *cred = current_cred(), *pcred; | |
6257 | bool match; | |
6258 | ||
6259 | rcu_read_lock(); | |
6260 | pcred = __task_cred(p); | |
6261 | match = (cred->euid == pcred->euid || | |
6262 | cred->euid == pcred->uid); | |
6263 | rcu_read_unlock(); | |
6264 | return match; | |
6265 | } | |
6266 | ||
961ccddd RR |
6267 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6268 | struct sched_param *param, bool user) | |
1da177e4 | 6269 | { |
83b699ed | 6270 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6271 | unsigned long flags; |
cb469845 | 6272 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6273 | struct rq *rq; |
ca94c442 | 6274 | int reset_on_fork; |
1da177e4 | 6275 | |
66e5393a SR |
6276 | /* may grab non-irq protected spin_locks */ |
6277 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6278 | recheck: |
6279 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6280 | if (policy < 0) { |
6281 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6282 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6283 | } else { |
6284 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6285 | policy &= ~SCHED_RESET_ON_FORK; | |
6286 | ||
6287 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6288 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6289 | policy != SCHED_IDLE) | |
6290 | return -EINVAL; | |
6291 | } | |
6292 | ||
1da177e4 LT |
6293 | /* |
6294 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6295 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6296 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6297 | */ |
6298 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6299 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6300 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6301 | return -EINVAL; |
e05606d3 | 6302 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6303 | return -EINVAL; |
6304 | ||
37e4ab3f OC |
6305 | /* |
6306 | * Allow unprivileged RT tasks to decrease priority: | |
6307 | */ | |
961ccddd | 6308 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6309 | if (rt_policy(policy)) { |
8dc3e909 | 6310 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6311 | |
6312 | if (!lock_task_sighand(p, &flags)) | |
6313 | return -ESRCH; | |
6314 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6315 | unlock_task_sighand(p, &flags); | |
6316 | ||
6317 | /* can't set/change the rt policy */ | |
6318 | if (policy != p->policy && !rlim_rtprio) | |
6319 | return -EPERM; | |
6320 | ||
6321 | /* can't increase priority */ | |
6322 | if (param->sched_priority > p->rt_priority && | |
6323 | param->sched_priority > rlim_rtprio) | |
6324 | return -EPERM; | |
6325 | } | |
dd41f596 IM |
6326 | /* |
6327 | * Like positive nice levels, dont allow tasks to | |
6328 | * move out of SCHED_IDLE either: | |
6329 | */ | |
6330 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6331 | return -EPERM; | |
5fe1d75f | 6332 | |
37e4ab3f | 6333 | /* can't change other user's priorities */ |
c69e8d9c | 6334 | if (!check_same_owner(p)) |
37e4ab3f | 6335 | return -EPERM; |
ca94c442 LP |
6336 | |
6337 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6338 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6339 | return -EPERM; | |
37e4ab3f | 6340 | } |
1da177e4 | 6341 | |
725aad24 | 6342 | if (user) { |
b68aa230 | 6343 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6344 | /* |
6345 | * Do not allow realtime tasks into groups that have no runtime | |
6346 | * assigned. | |
6347 | */ | |
9a7e0b18 PZ |
6348 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6349 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6350 | return -EPERM; |
b68aa230 PZ |
6351 | #endif |
6352 | ||
725aad24 JF |
6353 | retval = security_task_setscheduler(p, policy, param); |
6354 | if (retval) | |
6355 | return retval; | |
6356 | } | |
6357 | ||
b29739f9 IM |
6358 | /* |
6359 | * make sure no PI-waiters arrive (or leave) while we are | |
6360 | * changing the priority of the task: | |
6361 | */ | |
1d615482 | 6362 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
6363 | /* |
6364 | * To be able to change p->policy safely, the apropriate | |
6365 | * runqueue lock must be held. | |
6366 | */ | |
b29739f9 | 6367 | rq = __task_rq_lock(p); |
1da177e4 LT |
6368 | /* recheck policy now with rq lock held */ |
6369 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6370 | policy = oldpolicy = -1; | |
b29739f9 | 6371 | __task_rq_unlock(rq); |
1d615482 | 6372 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6373 | goto recheck; |
6374 | } | |
2daa3577 | 6375 | update_rq_clock(rq); |
dd41f596 | 6376 | on_rq = p->se.on_rq; |
051a1d1a | 6377 | running = task_current(rq, p); |
0e1f3483 | 6378 | if (on_rq) |
2e1cb74a | 6379 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6380 | if (running) |
6381 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6382 | |
ca94c442 LP |
6383 | p->sched_reset_on_fork = reset_on_fork; |
6384 | ||
1da177e4 | 6385 | oldprio = p->prio; |
dd41f596 | 6386 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6387 | |
0e1f3483 HS |
6388 | if (running) |
6389 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6390 | if (on_rq) { |
6391 | activate_task(rq, p, 0); | |
cb469845 SR |
6392 | |
6393 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6394 | } |
b29739f9 | 6395 | __task_rq_unlock(rq); |
1d615482 | 6396 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 6397 | |
95e02ca9 TG |
6398 | rt_mutex_adjust_pi(p); |
6399 | ||
1da177e4 LT |
6400 | return 0; |
6401 | } | |
961ccddd RR |
6402 | |
6403 | /** | |
6404 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6405 | * @p: the task in question. | |
6406 | * @policy: new policy. | |
6407 | * @param: structure containing the new RT priority. | |
6408 | * | |
6409 | * NOTE that the task may be already dead. | |
6410 | */ | |
6411 | int sched_setscheduler(struct task_struct *p, int policy, | |
6412 | struct sched_param *param) | |
6413 | { | |
6414 | return __sched_setscheduler(p, policy, param, true); | |
6415 | } | |
1da177e4 LT |
6416 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6417 | ||
961ccddd RR |
6418 | /** |
6419 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6420 | * @p: the task in question. | |
6421 | * @policy: new policy. | |
6422 | * @param: structure containing the new RT priority. | |
6423 | * | |
6424 | * Just like sched_setscheduler, only don't bother checking if the | |
6425 | * current context has permission. For example, this is needed in | |
6426 | * stop_machine(): we create temporary high priority worker threads, | |
6427 | * but our caller might not have that capability. | |
6428 | */ | |
6429 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6430 | struct sched_param *param) | |
6431 | { | |
6432 | return __sched_setscheduler(p, policy, param, false); | |
6433 | } | |
6434 | ||
95cdf3b7 IM |
6435 | static int |
6436 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6437 | { |
1da177e4 LT |
6438 | struct sched_param lparam; |
6439 | struct task_struct *p; | |
36c8b586 | 6440 | int retval; |
1da177e4 LT |
6441 | |
6442 | if (!param || pid < 0) | |
6443 | return -EINVAL; | |
6444 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6445 | return -EFAULT; | |
5fe1d75f ON |
6446 | |
6447 | rcu_read_lock(); | |
6448 | retval = -ESRCH; | |
1da177e4 | 6449 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6450 | if (p != NULL) |
6451 | retval = sched_setscheduler(p, policy, &lparam); | |
6452 | rcu_read_unlock(); | |
36c8b586 | 6453 | |
1da177e4 LT |
6454 | return retval; |
6455 | } | |
6456 | ||
6457 | /** | |
6458 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6459 | * @pid: the pid in question. | |
6460 | * @policy: new policy. | |
6461 | * @param: structure containing the new RT priority. | |
6462 | */ | |
5add95d4 HC |
6463 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6464 | struct sched_param __user *, param) | |
1da177e4 | 6465 | { |
c21761f1 JB |
6466 | /* negative values for policy are not valid */ |
6467 | if (policy < 0) | |
6468 | return -EINVAL; | |
6469 | ||
1da177e4 LT |
6470 | return do_sched_setscheduler(pid, policy, param); |
6471 | } | |
6472 | ||
6473 | /** | |
6474 | * sys_sched_setparam - set/change the RT priority of a thread | |
6475 | * @pid: the pid in question. | |
6476 | * @param: structure containing the new RT priority. | |
6477 | */ | |
5add95d4 | 6478 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6479 | { |
6480 | return do_sched_setscheduler(pid, -1, param); | |
6481 | } | |
6482 | ||
6483 | /** | |
6484 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6485 | * @pid: the pid in question. | |
6486 | */ | |
5add95d4 | 6487 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6488 | { |
36c8b586 | 6489 | struct task_struct *p; |
3a5c359a | 6490 | int retval; |
1da177e4 LT |
6491 | |
6492 | if (pid < 0) | |
3a5c359a | 6493 | return -EINVAL; |
1da177e4 LT |
6494 | |
6495 | retval = -ESRCH; | |
5fe85be0 | 6496 | rcu_read_lock(); |
1da177e4 LT |
6497 | p = find_process_by_pid(pid); |
6498 | if (p) { | |
6499 | retval = security_task_getscheduler(p); | |
6500 | if (!retval) | |
ca94c442 LP |
6501 | retval = p->policy |
6502 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 6503 | } |
5fe85be0 | 6504 | rcu_read_unlock(); |
1da177e4 LT |
6505 | return retval; |
6506 | } | |
6507 | ||
6508 | /** | |
ca94c442 | 6509 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6510 | * @pid: the pid in question. |
6511 | * @param: structure containing the RT priority. | |
6512 | */ | |
5add95d4 | 6513 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6514 | { |
6515 | struct sched_param lp; | |
36c8b586 | 6516 | struct task_struct *p; |
3a5c359a | 6517 | int retval; |
1da177e4 LT |
6518 | |
6519 | if (!param || pid < 0) | |
3a5c359a | 6520 | return -EINVAL; |
1da177e4 | 6521 | |
5fe85be0 | 6522 | rcu_read_lock(); |
1da177e4 LT |
6523 | p = find_process_by_pid(pid); |
6524 | retval = -ESRCH; | |
6525 | if (!p) | |
6526 | goto out_unlock; | |
6527 | ||
6528 | retval = security_task_getscheduler(p); | |
6529 | if (retval) | |
6530 | goto out_unlock; | |
6531 | ||
6532 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 6533 | rcu_read_unlock(); |
1da177e4 LT |
6534 | |
6535 | /* | |
6536 | * This one might sleep, we cannot do it with a spinlock held ... | |
6537 | */ | |
6538 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6539 | ||
1da177e4 LT |
6540 | return retval; |
6541 | ||
6542 | out_unlock: | |
5fe85be0 | 6543 | rcu_read_unlock(); |
1da177e4 LT |
6544 | return retval; |
6545 | } | |
6546 | ||
96f874e2 | 6547 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6548 | { |
5a16f3d3 | 6549 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6550 | struct task_struct *p; |
6551 | int retval; | |
1da177e4 | 6552 | |
95402b38 | 6553 | get_online_cpus(); |
23f5d142 | 6554 | rcu_read_lock(); |
1da177e4 LT |
6555 | |
6556 | p = find_process_by_pid(pid); | |
6557 | if (!p) { | |
23f5d142 | 6558 | rcu_read_unlock(); |
95402b38 | 6559 | put_online_cpus(); |
1da177e4 LT |
6560 | return -ESRCH; |
6561 | } | |
6562 | ||
23f5d142 | 6563 | /* Prevent p going away */ |
1da177e4 | 6564 | get_task_struct(p); |
23f5d142 | 6565 | rcu_read_unlock(); |
1da177e4 | 6566 | |
5a16f3d3 RR |
6567 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6568 | retval = -ENOMEM; | |
6569 | goto out_put_task; | |
6570 | } | |
6571 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6572 | retval = -ENOMEM; | |
6573 | goto out_free_cpus_allowed; | |
6574 | } | |
1da177e4 | 6575 | retval = -EPERM; |
c69e8d9c | 6576 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6577 | goto out_unlock; |
6578 | ||
e7834f8f DQ |
6579 | retval = security_task_setscheduler(p, 0, NULL); |
6580 | if (retval) | |
6581 | goto out_unlock; | |
6582 | ||
5a16f3d3 RR |
6583 | cpuset_cpus_allowed(p, cpus_allowed); |
6584 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6585 | again: |
5a16f3d3 | 6586 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6587 | |
8707d8b8 | 6588 | if (!retval) { |
5a16f3d3 RR |
6589 | cpuset_cpus_allowed(p, cpus_allowed); |
6590 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6591 | /* |
6592 | * We must have raced with a concurrent cpuset | |
6593 | * update. Just reset the cpus_allowed to the | |
6594 | * cpuset's cpus_allowed | |
6595 | */ | |
5a16f3d3 | 6596 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6597 | goto again; |
6598 | } | |
6599 | } | |
1da177e4 | 6600 | out_unlock: |
5a16f3d3 RR |
6601 | free_cpumask_var(new_mask); |
6602 | out_free_cpus_allowed: | |
6603 | free_cpumask_var(cpus_allowed); | |
6604 | out_put_task: | |
1da177e4 | 6605 | put_task_struct(p); |
95402b38 | 6606 | put_online_cpus(); |
1da177e4 LT |
6607 | return retval; |
6608 | } | |
6609 | ||
6610 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6611 | struct cpumask *new_mask) |
1da177e4 | 6612 | { |
96f874e2 RR |
6613 | if (len < cpumask_size()) |
6614 | cpumask_clear(new_mask); | |
6615 | else if (len > cpumask_size()) | |
6616 | len = cpumask_size(); | |
6617 | ||
1da177e4 LT |
6618 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6619 | } | |
6620 | ||
6621 | /** | |
6622 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6623 | * @pid: pid of the process | |
6624 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6625 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6626 | */ | |
5add95d4 HC |
6627 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6628 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6629 | { |
5a16f3d3 | 6630 | cpumask_var_t new_mask; |
1da177e4 LT |
6631 | int retval; |
6632 | ||
5a16f3d3 RR |
6633 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6634 | return -ENOMEM; | |
1da177e4 | 6635 | |
5a16f3d3 RR |
6636 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6637 | if (retval == 0) | |
6638 | retval = sched_setaffinity(pid, new_mask); | |
6639 | free_cpumask_var(new_mask); | |
6640 | return retval; | |
1da177e4 LT |
6641 | } |
6642 | ||
96f874e2 | 6643 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6644 | { |
36c8b586 | 6645 | struct task_struct *p; |
31605683 TG |
6646 | unsigned long flags; |
6647 | struct rq *rq; | |
1da177e4 | 6648 | int retval; |
1da177e4 | 6649 | |
95402b38 | 6650 | get_online_cpus(); |
23f5d142 | 6651 | rcu_read_lock(); |
1da177e4 LT |
6652 | |
6653 | retval = -ESRCH; | |
6654 | p = find_process_by_pid(pid); | |
6655 | if (!p) | |
6656 | goto out_unlock; | |
6657 | ||
e7834f8f DQ |
6658 | retval = security_task_getscheduler(p); |
6659 | if (retval) | |
6660 | goto out_unlock; | |
6661 | ||
31605683 | 6662 | rq = task_rq_lock(p, &flags); |
96f874e2 | 6663 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 6664 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
6665 | |
6666 | out_unlock: | |
23f5d142 | 6667 | rcu_read_unlock(); |
95402b38 | 6668 | put_online_cpus(); |
1da177e4 | 6669 | |
9531b62f | 6670 | return retval; |
1da177e4 LT |
6671 | } |
6672 | ||
6673 | /** | |
6674 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6675 | * @pid: pid of the process | |
6676 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6677 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6678 | */ | |
5add95d4 HC |
6679 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6680 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6681 | { |
6682 | int ret; | |
f17c8607 | 6683 | cpumask_var_t mask; |
1da177e4 | 6684 | |
f17c8607 | 6685 | if (len < cpumask_size()) |
1da177e4 LT |
6686 | return -EINVAL; |
6687 | ||
f17c8607 RR |
6688 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6689 | return -ENOMEM; | |
1da177e4 | 6690 | |
f17c8607 RR |
6691 | ret = sched_getaffinity(pid, mask); |
6692 | if (ret == 0) { | |
6693 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6694 | ret = -EFAULT; | |
6695 | else | |
6696 | ret = cpumask_size(); | |
6697 | } | |
6698 | free_cpumask_var(mask); | |
1da177e4 | 6699 | |
f17c8607 | 6700 | return ret; |
1da177e4 LT |
6701 | } |
6702 | ||
6703 | /** | |
6704 | * sys_sched_yield - yield the current processor to other threads. | |
6705 | * | |
dd41f596 IM |
6706 | * This function yields the current CPU to other tasks. If there are no |
6707 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6708 | */ |
5add95d4 | 6709 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6710 | { |
70b97a7f | 6711 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6712 | |
2d72376b | 6713 | schedstat_inc(rq, yld_count); |
4530d7ab | 6714 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6715 | |
6716 | /* | |
6717 | * Since we are going to call schedule() anyway, there's | |
6718 | * no need to preempt or enable interrupts: | |
6719 | */ | |
6720 | __release(rq->lock); | |
8a25d5de | 6721 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 6722 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
6723 | preempt_enable_no_resched(); |
6724 | ||
6725 | schedule(); | |
6726 | ||
6727 | return 0; | |
6728 | } | |
6729 | ||
d86ee480 PZ |
6730 | static inline int should_resched(void) |
6731 | { | |
6732 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6733 | } | |
6734 | ||
e7b38404 | 6735 | static void __cond_resched(void) |
1da177e4 | 6736 | { |
e7aaaa69 FW |
6737 | add_preempt_count(PREEMPT_ACTIVE); |
6738 | schedule(); | |
6739 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6740 | } |
6741 | ||
02b67cc3 | 6742 | int __sched _cond_resched(void) |
1da177e4 | 6743 | { |
d86ee480 | 6744 | if (should_resched()) { |
1da177e4 LT |
6745 | __cond_resched(); |
6746 | return 1; | |
6747 | } | |
6748 | return 0; | |
6749 | } | |
02b67cc3 | 6750 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6751 | |
6752 | /* | |
613afbf8 | 6753 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6754 | * call schedule, and on return reacquire the lock. |
6755 | * | |
41a2d6cf | 6756 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6757 | * operations here to prevent schedule() from being called twice (once via |
6758 | * spin_unlock(), once by hand). | |
6759 | */ | |
613afbf8 | 6760 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6761 | { |
d86ee480 | 6762 | int resched = should_resched(); |
6df3cecb JK |
6763 | int ret = 0; |
6764 | ||
f607c668 PZ |
6765 | lockdep_assert_held(lock); |
6766 | ||
95c354fe | 6767 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6768 | spin_unlock(lock); |
d86ee480 | 6769 | if (resched) |
95c354fe NP |
6770 | __cond_resched(); |
6771 | else | |
6772 | cpu_relax(); | |
6df3cecb | 6773 | ret = 1; |
1da177e4 | 6774 | spin_lock(lock); |
1da177e4 | 6775 | } |
6df3cecb | 6776 | return ret; |
1da177e4 | 6777 | } |
613afbf8 | 6778 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6779 | |
613afbf8 | 6780 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6781 | { |
6782 | BUG_ON(!in_softirq()); | |
6783 | ||
d86ee480 | 6784 | if (should_resched()) { |
98d82567 | 6785 | local_bh_enable(); |
1da177e4 LT |
6786 | __cond_resched(); |
6787 | local_bh_disable(); | |
6788 | return 1; | |
6789 | } | |
6790 | return 0; | |
6791 | } | |
613afbf8 | 6792 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6793 | |
1da177e4 LT |
6794 | /** |
6795 | * yield - yield the current processor to other threads. | |
6796 | * | |
72fd4a35 | 6797 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6798 | * thread runnable and calls sys_sched_yield(). |
6799 | */ | |
6800 | void __sched yield(void) | |
6801 | { | |
6802 | set_current_state(TASK_RUNNING); | |
6803 | sys_sched_yield(); | |
6804 | } | |
1da177e4 LT |
6805 | EXPORT_SYMBOL(yield); |
6806 | ||
6807 | /* | |
41a2d6cf | 6808 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6809 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
6810 | */ |
6811 | void __sched io_schedule(void) | |
6812 | { | |
54d35f29 | 6813 | struct rq *rq = raw_rq(); |
1da177e4 | 6814 | |
0ff92245 | 6815 | delayacct_blkio_start(); |
1da177e4 | 6816 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6817 | current->in_iowait = 1; |
1da177e4 | 6818 | schedule(); |
8f0dfc34 | 6819 | current->in_iowait = 0; |
1da177e4 | 6820 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6821 | delayacct_blkio_end(); |
1da177e4 | 6822 | } |
1da177e4 LT |
6823 | EXPORT_SYMBOL(io_schedule); |
6824 | ||
6825 | long __sched io_schedule_timeout(long timeout) | |
6826 | { | |
54d35f29 | 6827 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6828 | long ret; |
6829 | ||
0ff92245 | 6830 | delayacct_blkio_start(); |
1da177e4 | 6831 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6832 | current->in_iowait = 1; |
1da177e4 | 6833 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6834 | current->in_iowait = 0; |
1da177e4 | 6835 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6836 | delayacct_blkio_end(); |
1da177e4 LT |
6837 | return ret; |
6838 | } | |
6839 | ||
6840 | /** | |
6841 | * sys_sched_get_priority_max - return maximum RT priority. | |
6842 | * @policy: scheduling class. | |
6843 | * | |
6844 | * this syscall returns the maximum rt_priority that can be used | |
6845 | * by a given scheduling class. | |
6846 | */ | |
5add95d4 | 6847 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6848 | { |
6849 | int ret = -EINVAL; | |
6850 | ||
6851 | switch (policy) { | |
6852 | case SCHED_FIFO: | |
6853 | case SCHED_RR: | |
6854 | ret = MAX_USER_RT_PRIO-1; | |
6855 | break; | |
6856 | case SCHED_NORMAL: | |
b0a9499c | 6857 | case SCHED_BATCH: |
dd41f596 | 6858 | case SCHED_IDLE: |
1da177e4 LT |
6859 | ret = 0; |
6860 | break; | |
6861 | } | |
6862 | return ret; | |
6863 | } | |
6864 | ||
6865 | /** | |
6866 | * sys_sched_get_priority_min - return minimum RT priority. | |
6867 | * @policy: scheduling class. | |
6868 | * | |
6869 | * this syscall returns the minimum rt_priority that can be used | |
6870 | * by a given scheduling class. | |
6871 | */ | |
5add95d4 | 6872 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6873 | { |
6874 | int ret = -EINVAL; | |
6875 | ||
6876 | switch (policy) { | |
6877 | case SCHED_FIFO: | |
6878 | case SCHED_RR: | |
6879 | ret = 1; | |
6880 | break; | |
6881 | case SCHED_NORMAL: | |
b0a9499c | 6882 | case SCHED_BATCH: |
dd41f596 | 6883 | case SCHED_IDLE: |
1da177e4 LT |
6884 | ret = 0; |
6885 | } | |
6886 | return ret; | |
6887 | } | |
6888 | ||
6889 | /** | |
6890 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6891 | * @pid: pid of the process. | |
6892 | * @interval: userspace pointer to the timeslice value. | |
6893 | * | |
6894 | * this syscall writes the default timeslice value of a given process | |
6895 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6896 | */ | |
17da2bd9 | 6897 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6898 | struct timespec __user *, interval) |
1da177e4 | 6899 | { |
36c8b586 | 6900 | struct task_struct *p; |
a4ec24b4 | 6901 | unsigned int time_slice; |
dba091b9 TG |
6902 | unsigned long flags; |
6903 | struct rq *rq; | |
3a5c359a | 6904 | int retval; |
1da177e4 | 6905 | struct timespec t; |
1da177e4 LT |
6906 | |
6907 | if (pid < 0) | |
3a5c359a | 6908 | return -EINVAL; |
1da177e4 LT |
6909 | |
6910 | retval = -ESRCH; | |
1a551ae7 | 6911 | rcu_read_lock(); |
1da177e4 LT |
6912 | p = find_process_by_pid(pid); |
6913 | if (!p) | |
6914 | goto out_unlock; | |
6915 | ||
6916 | retval = security_task_getscheduler(p); | |
6917 | if (retval) | |
6918 | goto out_unlock; | |
6919 | ||
dba091b9 TG |
6920 | rq = task_rq_lock(p, &flags); |
6921 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
6922 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 6923 | |
1a551ae7 | 6924 | rcu_read_unlock(); |
a4ec24b4 | 6925 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6926 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6927 | return retval; |
3a5c359a | 6928 | |
1da177e4 | 6929 | out_unlock: |
1a551ae7 | 6930 | rcu_read_unlock(); |
1da177e4 LT |
6931 | return retval; |
6932 | } | |
6933 | ||
7c731e0a | 6934 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6935 | |
82a1fcb9 | 6936 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6937 | { |
1da177e4 | 6938 | unsigned long free = 0; |
36c8b586 | 6939 | unsigned state; |
1da177e4 | 6940 | |
1da177e4 | 6941 | state = p->state ? __ffs(p->state) + 1 : 0; |
663997d4 | 6942 | pr_info("%-13.13s %c", p->comm, |
2ed6e34f | 6943 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6944 | #if BITS_PER_LONG == 32 |
1da177e4 | 6945 | if (state == TASK_RUNNING) |
663997d4 | 6946 | pr_cont(" running "); |
1da177e4 | 6947 | else |
663997d4 | 6948 | pr_cont(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6949 | #else |
6950 | if (state == TASK_RUNNING) | |
663997d4 | 6951 | pr_cont(" running task "); |
1da177e4 | 6952 | else |
663997d4 | 6953 | pr_cont(" %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6954 | #endif |
6955 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6956 | free = stack_not_used(p); |
1da177e4 | 6957 | #endif |
663997d4 | 6958 | pr_cont("%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
6959 | task_pid_nr(p), task_pid_nr(p->real_parent), |
6960 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6961 | |
5fb5e6de | 6962 | show_stack(p, NULL); |
1da177e4 LT |
6963 | } |
6964 | ||
e59e2ae2 | 6965 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6966 | { |
36c8b586 | 6967 | struct task_struct *g, *p; |
1da177e4 | 6968 | |
4bd77321 | 6969 | #if BITS_PER_LONG == 32 |
663997d4 | 6970 | pr_info(" task PC stack pid father\n"); |
1da177e4 | 6971 | #else |
663997d4 | 6972 | pr_info(" task PC stack pid father\n"); |
1da177e4 LT |
6973 | #endif |
6974 | read_lock(&tasklist_lock); | |
6975 | do_each_thread(g, p) { | |
6976 | /* | |
6977 | * reset the NMI-timeout, listing all files on a slow | |
6978 | * console might take alot of time: | |
6979 | */ | |
6980 | touch_nmi_watchdog(); | |
39bc89fd | 6981 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6982 | sched_show_task(p); |
1da177e4 LT |
6983 | } while_each_thread(g, p); |
6984 | ||
04c9167f JF |
6985 | touch_all_softlockup_watchdogs(); |
6986 | ||
dd41f596 IM |
6987 | #ifdef CONFIG_SCHED_DEBUG |
6988 | sysrq_sched_debug_show(); | |
6989 | #endif | |
1da177e4 | 6990 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6991 | /* |
6992 | * Only show locks if all tasks are dumped: | |
6993 | */ | |
93335a21 | 6994 | if (!state_filter) |
e59e2ae2 | 6995 | debug_show_all_locks(); |
1da177e4 LT |
6996 | } |
6997 | ||
1df21055 IM |
6998 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6999 | { | |
dd41f596 | 7000 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
7001 | } |
7002 | ||
f340c0d1 IM |
7003 | /** |
7004 | * init_idle - set up an idle thread for a given CPU | |
7005 | * @idle: task in question | |
7006 | * @cpu: cpu the idle task belongs to | |
7007 | * | |
7008 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7009 | * flag, to make booting more robust. | |
7010 | */ | |
5c1e1767 | 7011 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7012 | { |
70b97a7f | 7013 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7014 | unsigned long flags; |
7015 | ||
05fa785c | 7016 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 7017 | |
dd41f596 | 7018 | __sched_fork(idle); |
06b83b5f | 7019 | idle->state = TASK_RUNNING; |
dd41f596 IM |
7020 | idle->se.exec_start = sched_clock(); |
7021 | ||
96f874e2 | 7022 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 7023 | __set_task_cpu(idle, cpu); |
1da177e4 | 7024 | |
1da177e4 | 7025 | rq->curr = rq->idle = idle; |
4866cde0 NP |
7026 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
7027 | idle->oncpu = 1; | |
7028 | #endif | |
05fa785c | 7029 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
7030 | |
7031 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
7032 | #if defined(CONFIG_PREEMPT) |
7033 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
7034 | #else | |
a1261f54 | 7035 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 7036 | #endif |
dd41f596 IM |
7037 | /* |
7038 | * The idle tasks have their own, simple scheduling class: | |
7039 | */ | |
7040 | idle->sched_class = &idle_sched_class; | |
fb52607a | 7041 | ftrace_graph_init_task(idle); |
1da177e4 LT |
7042 | } |
7043 | ||
7044 | /* | |
7045 | * In a system that switches off the HZ timer nohz_cpu_mask | |
7046 | * indicates which cpus entered this state. This is used | |
7047 | * in the rcu update to wait only for active cpus. For system | |
7048 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 7049 | * always be CPU_BITS_NONE. |
1da177e4 | 7050 | */ |
6a7b3dc3 | 7051 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 7052 | |
19978ca6 IM |
7053 | /* |
7054 | * Increase the granularity value when there are more CPUs, | |
7055 | * because with more CPUs the 'effective latency' as visible | |
7056 | * to users decreases. But the relationship is not linear, | |
7057 | * so pick a second-best guess by going with the log2 of the | |
7058 | * number of CPUs. | |
7059 | * | |
7060 | * This idea comes from the SD scheduler of Con Kolivas: | |
7061 | */ | |
acb4a848 | 7062 | static int get_update_sysctl_factor(void) |
19978ca6 | 7063 | { |
4ca3ef71 | 7064 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
7065 | unsigned int factor; |
7066 | ||
7067 | switch (sysctl_sched_tunable_scaling) { | |
7068 | case SCHED_TUNABLESCALING_NONE: | |
7069 | factor = 1; | |
7070 | break; | |
7071 | case SCHED_TUNABLESCALING_LINEAR: | |
7072 | factor = cpus; | |
7073 | break; | |
7074 | case SCHED_TUNABLESCALING_LOG: | |
7075 | default: | |
7076 | factor = 1 + ilog2(cpus); | |
7077 | break; | |
7078 | } | |
19978ca6 | 7079 | |
acb4a848 CE |
7080 | return factor; |
7081 | } | |
19978ca6 | 7082 | |
acb4a848 CE |
7083 | static void update_sysctl(void) |
7084 | { | |
7085 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 7086 | |
0bcdcf28 CE |
7087 | #define SET_SYSCTL(name) \ |
7088 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
7089 | SET_SYSCTL(sched_min_granularity); | |
7090 | SET_SYSCTL(sched_latency); | |
7091 | SET_SYSCTL(sched_wakeup_granularity); | |
7092 | SET_SYSCTL(sched_shares_ratelimit); | |
7093 | #undef SET_SYSCTL | |
7094 | } | |
55cd5340 | 7095 | |
0bcdcf28 CE |
7096 | static inline void sched_init_granularity(void) |
7097 | { | |
7098 | update_sysctl(); | |
19978ca6 IM |
7099 | } |
7100 | ||
1da177e4 LT |
7101 | #ifdef CONFIG_SMP |
7102 | /* | |
7103 | * This is how migration works: | |
7104 | * | |
70b97a7f | 7105 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7106 | * runqueue and wake up that CPU's migration thread. |
7107 | * 2) we down() the locked semaphore => thread blocks. | |
7108 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7109 | * thread off the CPU) | |
7110 | * 4) it gets the migration request and checks whether the migrated | |
7111 | * task is still in the wrong runqueue. | |
7112 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7113 | * it and puts it into the right queue. | |
7114 | * 6) migration thread up()s the semaphore. | |
7115 | * 7) we wake up and the migration is done. | |
7116 | */ | |
7117 | ||
7118 | /* | |
7119 | * Change a given task's CPU affinity. Migrate the thread to a | |
7120 | * proper CPU and schedule it away if the CPU it's executing on | |
7121 | * is removed from the allowed bitmask. | |
7122 | * | |
7123 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7124 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7125 | * call is not atomic; no spinlocks may be held. |
7126 | */ | |
96f874e2 | 7127 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7128 | { |
70b97a7f | 7129 | struct migration_req req; |
1da177e4 | 7130 | unsigned long flags; |
70b97a7f | 7131 | struct rq *rq; |
48f24c4d | 7132 | int ret = 0; |
1da177e4 | 7133 | |
e2912009 PZ |
7134 | /* |
7135 | * Since we rely on wake-ups to migrate sleeping tasks, don't change | |
7136 | * the ->cpus_allowed mask from under waking tasks, which would be | |
7137 | * possible when we change rq->lock in ttwu(), so synchronize against | |
7138 | * TASK_WAKING to avoid that. | |
7139 | */ | |
7140 | again: | |
7141 | while (p->state == TASK_WAKING) | |
7142 | cpu_relax(); | |
7143 | ||
1da177e4 | 7144 | rq = task_rq_lock(p, &flags); |
e2912009 PZ |
7145 | |
7146 | if (p->state == TASK_WAKING) { | |
7147 | task_rq_unlock(rq, &flags); | |
7148 | goto again; | |
7149 | } | |
7150 | ||
6ad4c188 | 7151 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
7152 | ret = -EINVAL; |
7153 | goto out; | |
7154 | } | |
7155 | ||
9985b0ba | 7156 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7157 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7158 | ret = -EINVAL; |
7159 | goto out; | |
7160 | } | |
7161 | ||
73fe6aae | 7162 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7163 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7164 | else { |
96f874e2 RR |
7165 | cpumask_copy(&p->cpus_allowed, new_mask); |
7166 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7167 | } |
7168 | ||
1da177e4 | 7169 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7170 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7171 | goto out; |
7172 | ||
6ad4c188 | 7173 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 7174 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7175 | struct task_struct *mt = rq->migration_thread; |
7176 | ||
7177 | get_task_struct(mt); | |
1da177e4 LT |
7178 | task_rq_unlock(rq, &flags); |
7179 | wake_up_process(rq->migration_thread); | |
693525e3 | 7180 | put_task_struct(mt); |
1da177e4 LT |
7181 | wait_for_completion(&req.done); |
7182 | tlb_migrate_finish(p->mm); | |
7183 | return 0; | |
7184 | } | |
7185 | out: | |
7186 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7187 | |
1da177e4 LT |
7188 | return ret; |
7189 | } | |
cd8ba7cd | 7190 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7191 | |
7192 | /* | |
41a2d6cf | 7193 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7194 | * this because either it can't run here any more (set_cpus_allowed() |
7195 | * away from this CPU, or CPU going down), or because we're | |
7196 | * attempting to rebalance this task on exec (sched_exec). | |
7197 | * | |
7198 | * So we race with normal scheduler movements, but that's OK, as long | |
7199 | * as the task is no longer on this CPU. | |
efc30814 KK |
7200 | * |
7201 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7202 | */ |
efc30814 | 7203 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7204 | { |
70b97a7f | 7205 | struct rq *rq_dest, *rq_src; |
e2912009 | 7206 | int ret = 0; |
1da177e4 | 7207 | |
e761b772 | 7208 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7209 | return ret; |
1da177e4 LT |
7210 | |
7211 | rq_src = cpu_rq(src_cpu); | |
7212 | rq_dest = cpu_rq(dest_cpu); | |
7213 | ||
7214 | double_rq_lock(rq_src, rq_dest); | |
7215 | /* Already moved. */ | |
7216 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7217 | goto done; |
1da177e4 | 7218 | /* Affinity changed (again). */ |
96f874e2 | 7219 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7220 | goto fail; |
1da177e4 | 7221 | |
e2912009 PZ |
7222 | /* |
7223 | * If we're not on a rq, the next wake-up will ensure we're | |
7224 | * placed properly. | |
7225 | */ | |
7226 | if (p->se.on_rq) { | |
2e1cb74a | 7227 | deactivate_task(rq_src, p, 0); |
e2912009 | 7228 | set_task_cpu(p, dest_cpu); |
dd41f596 | 7229 | activate_task(rq_dest, p, 0); |
15afe09b | 7230 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7231 | } |
b1e38734 | 7232 | done: |
efc30814 | 7233 | ret = 1; |
b1e38734 | 7234 | fail: |
1da177e4 | 7235 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7236 | return ret; |
1da177e4 LT |
7237 | } |
7238 | ||
03b042bf PM |
7239 | #define RCU_MIGRATION_IDLE 0 |
7240 | #define RCU_MIGRATION_NEED_QS 1 | |
7241 | #define RCU_MIGRATION_GOT_QS 2 | |
7242 | #define RCU_MIGRATION_MUST_SYNC 3 | |
7243 | ||
1da177e4 LT |
7244 | /* |
7245 | * migration_thread - this is a highprio system thread that performs | |
7246 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7247 | * another runqueue. | |
7248 | */ | |
95cdf3b7 | 7249 | static int migration_thread(void *data) |
1da177e4 | 7250 | { |
03b042bf | 7251 | int badcpu; |
1da177e4 | 7252 | int cpu = (long)data; |
70b97a7f | 7253 | struct rq *rq; |
1da177e4 LT |
7254 | |
7255 | rq = cpu_rq(cpu); | |
7256 | BUG_ON(rq->migration_thread != current); | |
7257 | ||
7258 | set_current_state(TASK_INTERRUPTIBLE); | |
7259 | while (!kthread_should_stop()) { | |
70b97a7f | 7260 | struct migration_req *req; |
1da177e4 | 7261 | struct list_head *head; |
1da177e4 | 7262 | |
05fa785c | 7263 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
7264 | |
7265 | if (cpu_is_offline(cpu)) { | |
05fa785c | 7266 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 7267 | break; |
1da177e4 LT |
7268 | } |
7269 | ||
7270 | if (rq->active_balance) { | |
7271 | active_load_balance(rq, cpu); | |
7272 | rq->active_balance = 0; | |
7273 | } | |
7274 | ||
7275 | head = &rq->migration_queue; | |
7276 | ||
7277 | if (list_empty(head)) { | |
05fa785c | 7278 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
7279 | schedule(); |
7280 | set_current_state(TASK_INTERRUPTIBLE); | |
7281 | continue; | |
7282 | } | |
70b97a7f | 7283 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7284 | list_del_init(head->next); |
7285 | ||
03b042bf | 7286 | if (req->task != NULL) { |
05fa785c | 7287 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
7288 | __migrate_task(req->task, cpu, req->dest_cpu); |
7289 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
7290 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 7291 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
7292 | } else { |
7293 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 7294 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
7295 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
7296 | } | |
674311d5 | 7297 | local_irq_enable(); |
1da177e4 LT |
7298 | |
7299 | complete(&req->done); | |
7300 | } | |
7301 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7302 | |
1da177e4 LT |
7303 | return 0; |
7304 | } | |
7305 | ||
7306 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7307 | |
7308 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7309 | { | |
7310 | int ret; | |
7311 | ||
7312 | local_irq_disable(); | |
7313 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7314 | local_irq_enable(); | |
7315 | return ret; | |
7316 | } | |
7317 | ||
054b9108 | 7318 | /* |
3a4fa0a2 | 7319 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7320 | */ |
48f24c4d | 7321 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7322 | { |
70b97a7f | 7323 | int dest_cpu; |
e76bd8d9 RR |
7324 | |
7325 | again: | |
5da9a0fb | 7326 | dest_cpu = select_fallback_rq(dead_cpu, p); |
e76bd8d9 | 7327 | |
e76bd8d9 RR |
7328 | /* It can have affinity changed while we were choosing. */ |
7329 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7330 | goto again; | |
1da177e4 LT |
7331 | } |
7332 | ||
7333 | /* | |
7334 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7335 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7336 | * for performance reasons the counter is not stricly tracking tasks to | |
7337 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7338 | * to keep the global sum constant after CPU-down: | |
7339 | */ | |
70b97a7f | 7340 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7341 | { |
6ad4c188 | 7342 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
7343 | unsigned long flags; |
7344 | ||
7345 | local_irq_save(flags); | |
7346 | double_rq_lock(rq_src, rq_dest); | |
7347 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7348 | rq_src->nr_uninterruptible = 0; | |
7349 | double_rq_unlock(rq_src, rq_dest); | |
7350 | local_irq_restore(flags); | |
7351 | } | |
7352 | ||
7353 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7354 | static void migrate_live_tasks(int src_cpu) | |
7355 | { | |
48f24c4d | 7356 | struct task_struct *p, *t; |
1da177e4 | 7357 | |
f7b4cddc | 7358 | read_lock(&tasklist_lock); |
1da177e4 | 7359 | |
48f24c4d IM |
7360 | do_each_thread(t, p) { |
7361 | if (p == current) | |
1da177e4 LT |
7362 | continue; |
7363 | ||
48f24c4d IM |
7364 | if (task_cpu(p) == src_cpu) |
7365 | move_task_off_dead_cpu(src_cpu, p); | |
7366 | } while_each_thread(t, p); | |
1da177e4 | 7367 | |
f7b4cddc | 7368 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7369 | } |
7370 | ||
dd41f596 IM |
7371 | /* |
7372 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7373 | * It does so by boosting its priority to highest possible. |
7374 | * Used by CPU offline code. | |
1da177e4 LT |
7375 | */ |
7376 | void sched_idle_next(void) | |
7377 | { | |
48f24c4d | 7378 | int this_cpu = smp_processor_id(); |
70b97a7f | 7379 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7380 | struct task_struct *p = rq->idle; |
7381 | unsigned long flags; | |
7382 | ||
7383 | /* cpu has to be offline */ | |
48f24c4d | 7384 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7385 | |
48f24c4d IM |
7386 | /* |
7387 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7388 | * and interrupts disabled on the current cpu. | |
1da177e4 | 7389 | */ |
05fa785c | 7390 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 7391 | |
dd41f596 | 7392 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7393 | |
94bc9a7b DA |
7394 | update_rq_clock(rq); |
7395 | activate_task(rq, p, 0); | |
1da177e4 | 7396 | |
05fa785c | 7397 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
7398 | } |
7399 | ||
48f24c4d IM |
7400 | /* |
7401 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7402 | * offline. |
7403 | */ | |
7404 | void idle_task_exit(void) | |
7405 | { | |
7406 | struct mm_struct *mm = current->active_mm; | |
7407 | ||
7408 | BUG_ON(cpu_online(smp_processor_id())); | |
7409 | ||
7410 | if (mm != &init_mm) | |
7411 | switch_mm(mm, &init_mm, current); | |
7412 | mmdrop(mm); | |
7413 | } | |
7414 | ||
054b9108 | 7415 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7416 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7417 | { |
70b97a7f | 7418 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7419 | |
7420 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7421 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7422 | |
7423 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7424 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7425 | |
48f24c4d | 7426 | get_task_struct(p); |
1da177e4 LT |
7427 | |
7428 | /* | |
7429 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7430 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7431 | * fine. |
7432 | */ | |
05fa785c | 7433 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 7434 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 7435 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 7436 | |
48f24c4d | 7437 | put_task_struct(p); |
1da177e4 LT |
7438 | } |
7439 | ||
7440 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7441 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7442 | { | |
70b97a7f | 7443 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7444 | struct task_struct *next; |
48f24c4d | 7445 | |
dd41f596 IM |
7446 | for ( ; ; ) { |
7447 | if (!rq->nr_running) | |
7448 | break; | |
a8e504d2 | 7449 | update_rq_clock(rq); |
b67802ea | 7450 | next = pick_next_task(rq); |
dd41f596 IM |
7451 | if (!next) |
7452 | break; | |
79c53799 | 7453 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7454 | migrate_dead(dead_cpu, next); |
e692ab53 | 7455 | |
1da177e4 LT |
7456 | } |
7457 | } | |
dce48a84 TG |
7458 | |
7459 | /* | |
7460 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7461 | */ | |
7462 | static void calc_global_load_remove(struct rq *rq) | |
7463 | { | |
7464 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7465 | rq->calc_load_active = 0; |
dce48a84 | 7466 | } |
1da177e4 LT |
7467 | #endif /* CONFIG_HOTPLUG_CPU */ |
7468 | ||
e692ab53 NP |
7469 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7470 | ||
7471 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7472 | { |
7473 | .procname = "sched_domain", | |
c57baf1e | 7474 | .mode = 0555, |
e0361851 | 7475 | }, |
56992309 | 7476 | {} |
e692ab53 NP |
7477 | }; |
7478 | ||
7479 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
7480 | { |
7481 | .procname = "kernel", | |
c57baf1e | 7482 | .mode = 0555, |
e0361851 AD |
7483 | .child = sd_ctl_dir, |
7484 | }, | |
56992309 | 7485 | {} |
e692ab53 NP |
7486 | }; |
7487 | ||
7488 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7489 | { | |
7490 | struct ctl_table *entry = | |
5cf9f062 | 7491 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7492 | |
e692ab53 NP |
7493 | return entry; |
7494 | } | |
7495 | ||
6382bc90 MM |
7496 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7497 | { | |
cd790076 | 7498 | struct ctl_table *entry; |
6382bc90 | 7499 | |
cd790076 MM |
7500 | /* |
7501 | * In the intermediate directories, both the child directory and | |
7502 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7503 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7504 | * static strings and all have proc handlers. |
7505 | */ | |
7506 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7507 | if (entry->child) |
7508 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7509 | if (entry->proc_handler == NULL) |
7510 | kfree(entry->procname); | |
7511 | } | |
6382bc90 MM |
7512 | |
7513 | kfree(*tablep); | |
7514 | *tablep = NULL; | |
7515 | } | |
7516 | ||
e692ab53 | 7517 | static void |
e0361851 | 7518 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7519 | const char *procname, void *data, int maxlen, |
7520 | mode_t mode, proc_handler *proc_handler) | |
7521 | { | |
e692ab53 NP |
7522 | entry->procname = procname; |
7523 | entry->data = data; | |
7524 | entry->maxlen = maxlen; | |
7525 | entry->mode = mode; | |
7526 | entry->proc_handler = proc_handler; | |
7527 | } | |
7528 | ||
7529 | static struct ctl_table * | |
7530 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7531 | { | |
a5d8c348 | 7532 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7533 | |
ad1cdc1d MM |
7534 | if (table == NULL) |
7535 | return NULL; | |
7536 | ||
e0361851 | 7537 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7538 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7539 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7540 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7541 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7542 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7543 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7544 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7545 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7546 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7547 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7548 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7549 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7550 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7551 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7552 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7553 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7554 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7555 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7556 | &sd->cache_nice_tries, |
7557 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7558 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7559 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7560 | set_table_entry(&table[11], "name", sd->name, |
7561 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7562 | /* &table[12] is terminator */ | |
e692ab53 NP |
7563 | |
7564 | return table; | |
7565 | } | |
7566 | ||
9a4e7159 | 7567 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7568 | { |
7569 | struct ctl_table *entry, *table; | |
7570 | struct sched_domain *sd; | |
7571 | int domain_num = 0, i; | |
7572 | char buf[32]; | |
7573 | ||
7574 | for_each_domain(cpu, sd) | |
7575 | domain_num++; | |
7576 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7577 | if (table == NULL) |
7578 | return NULL; | |
e692ab53 NP |
7579 | |
7580 | i = 0; | |
7581 | for_each_domain(cpu, sd) { | |
7582 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7583 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7584 | entry->mode = 0555; |
e692ab53 NP |
7585 | entry->child = sd_alloc_ctl_domain_table(sd); |
7586 | entry++; | |
7587 | i++; | |
7588 | } | |
7589 | return table; | |
7590 | } | |
7591 | ||
7592 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7593 | static void register_sched_domain_sysctl(void) |
e692ab53 | 7594 | { |
6ad4c188 | 7595 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
7596 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
7597 | char buf[32]; | |
7598 | ||
7378547f MM |
7599 | WARN_ON(sd_ctl_dir[0].child); |
7600 | sd_ctl_dir[0].child = entry; | |
7601 | ||
ad1cdc1d MM |
7602 | if (entry == NULL) |
7603 | return; | |
7604 | ||
6ad4c188 | 7605 | for_each_possible_cpu(i) { |
e692ab53 | 7606 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7607 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7608 | entry->mode = 0555; |
e692ab53 | 7609 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7610 | entry++; |
e692ab53 | 7611 | } |
7378547f MM |
7612 | |
7613 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7614 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7615 | } | |
6382bc90 | 7616 | |
7378547f | 7617 | /* may be called multiple times per register */ |
6382bc90 MM |
7618 | static void unregister_sched_domain_sysctl(void) |
7619 | { | |
7378547f MM |
7620 | if (sd_sysctl_header) |
7621 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7622 | sd_sysctl_header = NULL; |
7378547f MM |
7623 | if (sd_ctl_dir[0].child) |
7624 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7625 | } |
e692ab53 | 7626 | #else |
6382bc90 MM |
7627 | static void register_sched_domain_sysctl(void) |
7628 | { | |
7629 | } | |
7630 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7631 | { |
7632 | } | |
7633 | #endif | |
7634 | ||
1f11eb6a GH |
7635 | static void set_rq_online(struct rq *rq) |
7636 | { | |
7637 | if (!rq->online) { | |
7638 | const struct sched_class *class; | |
7639 | ||
c6c4927b | 7640 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7641 | rq->online = 1; |
7642 | ||
7643 | for_each_class(class) { | |
7644 | if (class->rq_online) | |
7645 | class->rq_online(rq); | |
7646 | } | |
7647 | } | |
7648 | } | |
7649 | ||
7650 | static void set_rq_offline(struct rq *rq) | |
7651 | { | |
7652 | if (rq->online) { | |
7653 | const struct sched_class *class; | |
7654 | ||
7655 | for_each_class(class) { | |
7656 | if (class->rq_offline) | |
7657 | class->rq_offline(rq); | |
7658 | } | |
7659 | ||
c6c4927b | 7660 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7661 | rq->online = 0; |
7662 | } | |
7663 | } | |
7664 | ||
1da177e4 LT |
7665 | /* |
7666 | * migration_call - callback that gets triggered when a CPU is added. | |
7667 | * Here we can start up the necessary migration thread for the new CPU. | |
7668 | */ | |
48f24c4d IM |
7669 | static int __cpuinit |
7670 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7671 | { |
1da177e4 | 7672 | struct task_struct *p; |
48f24c4d | 7673 | int cpu = (long)hcpu; |
1da177e4 | 7674 | unsigned long flags; |
70b97a7f | 7675 | struct rq *rq; |
1da177e4 LT |
7676 | |
7677 | switch (action) { | |
5be9361c | 7678 | |
1da177e4 | 7679 | case CPU_UP_PREPARE: |
8bb78442 | 7680 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7681 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7682 | if (IS_ERR(p)) |
7683 | return NOTIFY_BAD; | |
1da177e4 LT |
7684 | kthread_bind(p, cpu); |
7685 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7686 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7687 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7688 | task_rq_unlock(rq, &flags); |
371cbb38 | 7689 | get_task_struct(p); |
1da177e4 | 7690 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7691 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7692 | break; |
48f24c4d | 7693 | |
1da177e4 | 7694 | case CPU_ONLINE: |
8bb78442 | 7695 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7696 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7697 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7698 | |
7699 | /* Update our root-domain */ | |
7700 | rq = cpu_rq(cpu); | |
05fa785c | 7701 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 7702 | if (rq->rd) { |
c6c4927b | 7703 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7704 | |
7705 | set_rq_online(rq); | |
1f94ef59 | 7706 | } |
05fa785c | 7707 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 7708 | break; |
48f24c4d | 7709 | |
1da177e4 LT |
7710 | #ifdef CONFIG_HOTPLUG_CPU |
7711 | case CPU_UP_CANCELED: | |
8bb78442 | 7712 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7713 | if (!cpu_rq(cpu)->migration_thread) |
7714 | break; | |
41a2d6cf | 7715 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7716 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7717 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7718 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7719 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7720 | cpu_rq(cpu)->migration_thread = NULL; |
7721 | break; | |
48f24c4d | 7722 | |
1da177e4 | 7723 | case CPU_DEAD: |
8bb78442 | 7724 | case CPU_DEAD_FROZEN: |
470fd646 | 7725 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7726 | migrate_live_tasks(cpu); |
7727 | rq = cpu_rq(cpu); | |
7728 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7729 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7730 | rq->migration_thread = NULL; |
7731 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 7732 | raw_spin_lock_irq(&rq->lock); |
a8e504d2 | 7733 | update_rq_clock(rq); |
2e1cb74a | 7734 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
7735 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7736 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7737 | migrate_dead_tasks(cpu); |
05fa785c | 7738 | raw_spin_unlock_irq(&rq->lock); |
470fd646 | 7739 | cpuset_unlock(); |
1da177e4 LT |
7740 | migrate_nr_uninterruptible(rq); |
7741 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7742 | calc_global_load_remove(rq); |
41a2d6cf IM |
7743 | /* |
7744 | * No need to migrate the tasks: it was best-effort if | |
7745 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7746 | * the requestors. | |
7747 | */ | |
05fa785c | 7748 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 7749 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
7750 | struct migration_req *req; |
7751 | ||
1da177e4 | 7752 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7753 | struct migration_req, list); |
1da177e4 | 7754 | list_del_init(&req->list); |
05fa785c | 7755 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 7756 | complete(&req->done); |
05fa785c | 7757 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 7758 | } |
05fa785c | 7759 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 7760 | break; |
57d885fe | 7761 | |
08f503b0 GH |
7762 | case CPU_DYING: |
7763 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7764 | /* Update our root-domain */ |
7765 | rq = cpu_rq(cpu); | |
05fa785c | 7766 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 7767 | if (rq->rd) { |
c6c4927b | 7768 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7769 | set_rq_offline(rq); |
57d885fe | 7770 | } |
05fa785c | 7771 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 7772 | break; |
1da177e4 LT |
7773 | #endif |
7774 | } | |
7775 | return NOTIFY_OK; | |
7776 | } | |
7777 | ||
f38b0820 PM |
7778 | /* |
7779 | * Register at high priority so that task migration (migrate_all_tasks) | |
7780 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 7781 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 7782 | */ |
26c2143b | 7783 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7784 | .notifier_call = migration_call, |
7785 | .priority = 10 | |
7786 | }; | |
7787 | ||
7babe8db | 7788 | static int __init migration_init(void) |
1da177e4 LT |
7789 | { |
7790 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7791 | int err; |
48f24c4d IM |
7792 | |
7793 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7794 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7795 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7796 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7797 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7798 | |
a004cd42 | 7799 | return 0; |
1da177e4 | 7800 | } |
7babe8db | 7801 | early_initcall(migration_init); |
1da177e4 LT |
7802 | #endif |
7803 | ||
7804 | #ifdef CONFIG_SMP | |
476f3534 | 7805 | |
3e9830dc | 7806 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7807 | |
f6630114 MT |
7808 | static __read_mostly int sched_domain_debug_enabled; |
7809 | ||
7810 | static int __init sched_domain_debug_setup(char *str) | |
7811 | { | |
7812 | sched_domain_debug_enabled = 1; | |
7813 | ||
7814 | return 0; | |
7815 | } | |
7816 | early_param("sched_debug", sched_domain_debug_setup); | |
7817 | ||
7c16ec58 | 7818 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7819 | struct cpumask *groupmask) |
1da177e4 | 7820 | { |
4dcf6aff | 7821 | struct sched_group *group = sd->groups; |
434d53b0 | 7822 | char str[256]; |
1da177e4 | 7823 | |
968ea6d8 | 7824 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7825 | cpumask_clear(groupmask); |
4dcf6aff IM |
7826 | |
7827 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7828 | ||
7829 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
663997d4 | 7830 | pr_cont("does not load-balance\n"); |
4dcf6aff | 7831 | if (sd->parent) |
663997d4 | 7832 | pr_err("ERROR: !SD_LOAD_BALANCE domain has parent\n"); |
4dcf6aff | 7833 | return -1; |
41c7ce9a NP |
7834 | } |
7835 | ||
663997d4 | 7836 | pr_cont("span %s level %s\n", str, sd->name); |
4dcf6aff | 7837 | |
758b2cdc | 7838 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
663997d4 | 7839 | pr_err("ERROR: domain->span does not contain CPU%d\n", cpu); |
4dcf6aff | 7840 | } |
758b2cdc | 7841 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
663997d4 | 7842 | pr_err("ERROR: domain->groups does not contain CPU%d\n", cpu); |
4dcf6aff | 7843 | } |
1da177e4 | 7844 | |
4dcf6aff | 7845 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7846 | do { |
4dcf6aff | 7847 | if (!group) { |
663997d4 JP |
7848 | pr_cont("\n"); |
7849 | pr_err("ERROR: group is NULL\n"); | |
1da177e4 LT |
7850 | break; |
7851 | } | |
7852 | ||
18a3885f | 7853 | if (!group->cpu_power) { |
663997d4 JP |
7854 | pr_cont("\n"); |
7855 | pr_err("ERROR: domain->cpu_power not set\n"); | |
4dcf6aff IM |
7856 | break; |
7857 | } | |
1da177e4 | 7858 | |
758b2cdc | 7859 | if (!cpumask_weight(sched_group_cpus(group))) { |
663997d4 JP |
7860 | pr_cont("\n"); |
7861 | pr_err("ERROR: empty group\n"); | |
4dcf6aff IM |
7862 | break; |
7863 | } | |
1da177e4 | 7864 | |
758b2cdc | 7865 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
663997d4 JP |
7866 | pr_cont("\n"); |
7867 | pr_err("ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
7868 | break; |
7869 | } | |
1da177e4 | 7870 | |
758b2cdc | 7871 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7872 | |
968ea6d8 | 7873 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 7874 | |
663997d4 | 7875 | pr_cont(" %s", str); |
18a3885f | 7876 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
663997d4 | 7877 | pr_cont(" (cpu_power = %d)", group->cpu_power); |
381512cf | 7878 | } |
1da177e4 | 7879 | |
4dcf6aff IM |
7880 | group = group->next; |
7881 | } while (group != sd->groups); | |
663997d4 | 7882 | pr_cont("\n"); |
1da177e4 | 7883 | |
758b2cdc | 7884 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
663997d4 | 7885 | pr_err("ERROR: groups don't span domain->span\n"); |
1da177e4 | 7886 | |
758b2cdc RR |
7887 | if (sd->parent && |
7888 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
663997d4 | 7889 | pr_err("ERROR: parent span is not a superset of domain->span\n"); |
4dcf6aff IM |
7890 | return 0; |
7891 | } | |
1da177e4 | 7892 | |
4dcf6aff IM |
7893 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7894 | { | |
d5dd3db1 | 7895 | cpumask_var_t groupmask; |
4dcf6aff | 7896 | int level = 0; |
1da177e4 | 7897 | |
f6630114 MT |
7898 | if (!sched_domain_debug_enabled) |
7899 | return; | |
7900 | ||
4dcf6aff IM |
7901 | if (!sd) { |
7902 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7903 | return; | |
7904 | } | |
1da177e4 | 7905 | |
4dcf6aff IM |
7906 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7907 | ||
d5dd3db1 | 7908 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7909 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7910 | return; | |
7911 | } | |
7912 | ||
4dcf6aff | 7913 | for (;;) { |
7c16ec58 | 7914 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7915 | break; |
1da177e4 LT |
7916 | level++; |
7917 | sd = sd->parent; | |
33859f7f | 7918 | if (!sd) |
4dcf6aff IM |
7919 | break; |
7920 | } | |
d5dd3db1 | 7921 | free_cpumask_var(groupmask); |
1da177e4 | 7922 | } |
6d6bc0ad | 7923 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7924 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7925 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7926 | |
1a20ff27 | 7927 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7928 | { |
758b2cdc | 7929 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7930 | return 1; |
7931 | ||
7932 | /* Following flags need at least 2 groups */ | |
7933 | if (sd->flags & (SD_LOAD_BALANCE | | |
7934 | SD_BALANCE_NEWIDLE | | |
7935 | SD_BALANCE_FORK | | |
89c4710e SS |
7936 | SD_BALANCE_EXEC | |
7937 | SD_SHARE_CPUPOWER | | |
7938 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7939 | if (sd->groups != sd->groups->next) |
7940 | return 0; | |
7941 | } | |
7942 | ||
7943 | /* Following flags don't use groups */ | |
c88d5910 | 7944 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
7945 | return 0; |
7946 | ||
7947 | return 1; | |
7948 | } | |
7949 | ||
48f24c4d IM |
7950 | static int |
7951 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7952 | { |
7953 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7954 | ||
7955 | if (sd_degenerate(parent)) | |
7956 | return 1; | |
7957 | ||
758b2cdc | 7958 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7959 | return 0; |
7960 | ||
245af2c7 SS |
7961 | /* Flags needing groups don't count if only 1 group in parent */ |
7962 | if (parent->groups == parent->groups->next) { | |
7963 | pflags &= ~(SD_LOAD_BALANCE | | |
7964 | SD_BALANCE_NEWIDLE | | |
7965 | SD_BALANCE_FORK | | |
89c4710e SS |
7966 | SD_BALANCE_EXEC | |
7967 | SD_SHARE_CPUPOWER | | |
7968 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7969 | if (nr_node_ids == 1) |
7970 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7971 | } |
7972 | if (~cflags & pflags) | |
7973 | return 0; | |
7974 | ||
7975 | return 1; | |
7976 | } | |
7977 | ||
c6c4927b RR |
7978 | static void free_rootdomain(struct root_domain *rd) |
7979 | { | |
047106ad PZ |
7980 | synchronize_sched(); |
7981 | ||
68e74568 RR |
7982 | cpupri_cleanup(&rd->cpupri); |
7983 | ||
c6c4927b RR |
7984 | free_cpumask_var(rd->rto_mask); |
7985 | free_cpumask_var(rd->online); | |
7986 | free_cpumask_var(rd->span); | |
7987 | kfree(rd); | |
7988 | } | |
7989 | ||
57d885fe GH |
7990 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7991 | { | |
a0490fa3 | 7992 | struct root_domain *old_rd = NULL; |
57d885fe | 7993 | unsigned long flags; |
57d885fe | 7994 | |
05fa785c | 7995 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
7996 | |
7997 | if (rq->rd) { | |
a0490fa3 | 7998 | old_rd = rq->rd; |
57d885fe | 7999 | |
c6c4927b | 8000 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 8001 | set_rq_offline(rq); |
57d885fe | 8002 | |
c6c4927b | 8003 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 8004 | |
a0490fa3 IM |
8005 | /* |
8006 | * If we dont want to free the old_rt yet then | |
8007 | * set old_rd to NULL to skip the freeing later | |
8008 | * in this function: | |
8009 | */ | |
8010 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
8011 | old_rd = NULL; | |
57d885fe GH |
8012 | } |
8013 | ||
8014 | atomic_inc(&rd->refcount); | |
8015 | rq->rd = rd; | |
8016 | ||
c6c4927b | 8017 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 8018 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 8019 | set_rq_online(rq); |
57d885fe | 8020 | |
05fa785c | 8021 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
8022 | |
8023 | if (old_rd) | |
8024 | free_rootdomain(old_rd); | |
57d885fe GH |
8025 | } |
8026 | ||
fd5e1b5d | 8027 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 8028 | { |
36b7b6d4 PE |
8029 | gfp_t gfp = GFP_KERNEL; |
8030 | ||
57d885fe GH |
8031 | memset(rd, 0, sizeof(*rd)); |
8032 | ||
36b7b6d4 PE |
8033 | if (bootmem) |
8034 | gfp = GFP_NOWAIT; | |
c6c4927b | 8035 | |
36b7b6d4 | 8036 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 8037 | goto out; |
36b7b6d4 | 8038 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 8039 | goto free_span; |
36b7b6d4 | 8040 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 8041 | goto free_online; |
6e0534f2 | 8042 | |
0fb53029 | 8043 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 8044 | goto free_rto_mask; |
c6c4927b | 8045 | return 0; |
6e0534f2 | 8046 | |
68e74568 RR |
8047 | free_rto_mask: |
8048 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
8049 | free_online: |
8050 | free_cpumask_var(rd->online); | |
8051 | free_span: | |
8052 | free_cpumask_var(rd->span); | |
0c910d28 | 8053 | out: |
c6c4927b | 8054 | return -ENOMEM; |
57d885fe GH |
8055 | } |
8056 | ||
8057 | static void init_defrootdomain(void) | |
8058 | { | |
c6c4927b RR |
8059 | init_rootdomain(&def_root_domain, true); |
8060 | ||
57d885fe GH |
8061 | atomic_set(&def_root_domain.refcount, 1); |
8062 | } | |
8063 | ||
dc938520 | 8064 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
8065 | { |
8066 | struct root_domain *rd; | |
8067 | ||
8068 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
8069 | if (!rd) | |
8070 | return NULL; | |
8071 | ||
c6c4927b RR |
8072 | if (init_rootdomain(rd, false) != 0) { |
8073 | kfree(rd); | |
8074 | return NULL; | |
8075 | } | |
57d885fe GH |
8076 | |
8077 | return rd; | |
8078 | } | |
8079 | ||
1da177e4 | 8080 | /* |
0eab9146 | 8081 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
8082 | * hold the hotplug lock. |
8083 | */ | |
0eab9146 IM |
8084 | static void |
8085 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 8086 | { |
70b97a7f | 8087 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8088 | struct sched_domain *tmp; |
8089 | ||
8090 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8091 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8092 | struct sched_domain *parent = tmp->parent; |
8093 | if (!parent) | |
8094 | break; | |
f29c9b1c | 8095 | |
1a848870 | 8096 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8097 | tmp->parent = parent->parent; |
1a848870 SS |
8098 | if (parent->parent) |
8099 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8100 | } else |
8101 | tmp = tmp->parent; | |
245af2c7 SS |
8102 | } |
8103 | ||
1a848870 | 8104 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8105 | sd = sd->parent; |
1a848870 SS |
8106 | if (sd) |
8107 | sd->child = NULL; | |
8108 | } | |
1da177e4 LT |
8109 | |
8110 | sched_domain_debug(sd, cpu); | |
8111 | ||
57d885fe | 8112 | rq_attach_root(rq, rd); |
674311d5 | 8113 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8114 | } |
8115 | ||
8116 | /* cpus with isolated domains */ | |
dcc30a35 | 8117 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8118 | |
8119 | /* Setup the mask of cpus configured for isolated domains */ | |
8120 | static int __init isolated_cpu_setup(char *str) | |
8121 | { | |
bdddd296 | 8122 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 8123 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8124 | return 1; |
8125 | } | |
8126 | ||
8927f494 | 8127 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8128 | |
8129 | /* | |
6711cab4 SS |
8130 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8131 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8132 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8133 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8134 | * |
8135 | * init_sched_build_groups will build a circular linked list of the groups | |
8136 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8137 | * and ->cpu_power to 0. | |
8138 | */ | |
a616058b | 8139 | static void |
96f874e2 RR |
8140 | init_sched_build_groups(const struct cpumask *span, |
8141 | const struct cpumask *cpu_map, | |
8142 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8143 | struct sched_group **sg, |
96f874e2 RR |
8144 | struct cpumask *tmpmask), |
8145 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8146 | { |
8147 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8148 | int i; |
8149 | ||
96f874e2 | 8150 | cpumask_clear(covered); |
7c16ec58 | 8151 | |
abcd083a | 8152 | for_each_cpu(i, span) { |
6711cab4 | 8153 | struct sched_group *sg; |
7c16ec58 | 8154 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8155 | int j; |
8156 | ||
758b2cdc | 8157 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8158 | continue; |
8159 | ||
758b2cdc | 8160 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 8161 | sg->cpu_power = 0; |
1da177e4 | 8162 | |
abcd083a | 8163 | for_each_cpu(j, span) { |
7c16ec58 | 8164 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8165 | continue; |
8166 | ||
96f874e2 | 8167 | cpumask_set_cpu(j, covered); |
758b2cdc | 8168 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8169 | } |
8170 | if (!first) | |
8171 | first = sg; | |
8172 | if (last) | |
8173 | last->next = sg; | |
8174 | last = sg; | |
8175 | } | |
8176 | last->next = first; | |
8177 | } | |
8178 | ||
9c1cfda2 | 8179 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8180 | |
9c1cfda2 | 8181 | #ifdef CONFIG_NUMA |
198e2f18 | 8182 | |
9c1cfda2 JH |
8183 | /** |
8184 | * find_next_best_node - find the next node to include in a sched_domain | |
8185 | * @node: node whose sched_domain we're building | |
8186 | * @used_nodes: nodes already in the sched_domain | |
8187 | * | |
41a2d6cf | 8188 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8189 | * finds the closest node not already in the @used_nodes map. |
8190 | * | |
8191 | * Should use nodemask_t. | |
8192 | */ | |
c5f59f08 | 8193 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8194 | { |
8195 | int i, n, val, min_val, best_node = 0; | |
8196 | ||
8197 | min_val = INT_MAX; | |
8198 | ||
076ac2af | 8199 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8200 | /* Start at @node */ |
076ac2af | 8201 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8202 | |
8203 | if (!nr_cpus_node(n)) | |
8204 | continue; | |
8205 | ||
8206 | /* Skip already used nodes */ | |
c5f59f08 | 8207 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8208 | continue; |
8209 | ||
8210 | /* Simple min distance search */ | |
8211 | val = node_distance(node, n); | |
8212 | ||
8213 | if (val < min_val) { | |
8214 | min_val = val; | |
8215 | best_node = n; | |
8216 | } | |
8217 | } | |
8218 | ||
c5f59f08 | 8219 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8220 | return best_node; |
8221 | } | |
8222 | ||
8223 | /** | |
8224 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8225 | * @node: node whose cpumask we're constructing | |
73486722 | 8226 | * @span: resulting cpumask |
9c1cfda2 | 8227 | * |
41a2d6cf | 8228 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8229 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8230 | * out optimally. | |
8231 | */ | |
96f874e2 | 8232 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8233 | { |
c5f59f08 | 8234 | nodemask_t used_nodes; |
48f24c4d | 8235 | int i; |
9c1cfda2 | 8236 | |
6ca09dfc | 8237 | cpumask_clear(span); |
c5f59f08 | 8238 | nodes_clear(used_nodes); |
9c1cfda2 | 8239 | |
6ca09dfc | 8240 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8241 | node_set(node, used_nodes); |
9c1cfda2 JH |
8242 | |
8243 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8244 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8245 | |
6ca09dfc | 8246 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8247 | } |
9c1cfda2 | 8248 | } |
6d6bc0ad | 8249 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8250 | |
5c45bf27 | 8251 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8252 | |
6c99e9ad RR |
8253 | /* |
8254 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8255 | * |
8256 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8257 | * and struct sched_domain. ) | |
6c99e9ad RR |
8258 | */ |
8259 | struct static_sched_group { | |
8260 | struct sched_group sg; | |
8261 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8262 | }; | |
8263 | ||
8264 | struct static_sched_domain { | |
8265 | struct sched_domain sd; | |
8266 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8267 | }; | |
8268 | ||
49a02c51 AH |
8269 | struct s_data { |
8270 | #ifdef CONFIG_NUMA | |
8271 | int sd_allnodes; | |
8272 | cpumask_var_t domainspan; | |
8273 | cpumask_var_t covered; | |
8274 | cpumask_var_t notcovered; | |
8275 | #endif | |
8276 | cpumask_var_t nodemask; | |
8277 | cpumask_var_t this_sibling_map; | |
8278 | cpumask_var_t this_core_map; | |
8279 | cpumask_var_t send_covered; | |
8280 | cpumask_var_t tmpmask; | |
8281 | struct sched_group **sched_group_nodes; | |
8282 | struct root_domain *rd; | |
8283 | }; | |
8284 | ||
2109b99e AH |
8285 | enum s_alloc { |
8286 | sa_sched_groups = 0, | |
8287 | sa_rootdomain, | |
8288 | sa_tmpmask, | |
8289 | sa_send_covered, | |
8290 | sa_this_core_map, | |
8291 | sa_this_sibling_map, | |
8292 | sa_nodemask, | |
8293 | sa_sched_group_nodes, | |
8294 | #ifdef CONFIG_NUMA | |
8295 | sa_notcovered, | |
8296 | sa_covered, | |
8297 | sa_domainspan, | |
8298 | #endif | |
8299 | sa_none, | |
8300 | }; | |
8301 | ||
9c1cfda2 | 8302 | /* |
48f24c4d | 8303 | * SMT sched-domains: |
9c1cfda2 | 8304 | */ |
1da177e4 | 8305 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8306 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 8307 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 8308 | |
41a2d6cf | 8309 | static int |
96f874e2 RR |
8310 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8311 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8312 | { |
6711cab4 | 8313 | if (sg) |
1871e52c | 8314 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
8315 | return cpu; |
8316 | } | |
6d6bc0ad | 8317 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8318 | |
48f24c4d IM |
8319 | /* |
8320 | * multi-core sched-domains: | |
8321 | */ | |
1e9f28fa | 8322 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8323 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8324 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8325 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8326 | |
8327 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8328 | static int |
96f874e2 RR |
8329 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8330 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8331 | { |
6711cab4 | 8332 | int group; |
7c16ec58 | 8333 | |
c69fc56d | 8334 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8335 | group = cpumask_first(mask); |
6711cab4 | 8336 | if (sg) |
6c99e9ad | 8337 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8338 | return group; |
1e9f28fa SS |
8339 | } |
8340 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8341 | static int |
96f874e2 RR |
8342 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8343 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8344 | { |
6711cab4 | 8345 | if (sg) |
6c99e9ad | 8346 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8347 | return cpu; |
8348 | } | |
8349 | #endif | |
8350 | ||
6c99e9ad RR |
8351 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8352 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8353 | |
41a2d6cf | 8354 | static int |
96f874e2 RR |
8355 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8356 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8357 | { |
6711cab4 | 8358 | int group; |
48f24c4d | 8359 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8360 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8361 | group = cpumask_first(mask); |
1e9f28fa | 8362 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8363 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8364 | group = cpumask_first(mask); |
1da177e4 | 8365 | #else |
6711cab4 | 8366 | group = cpu; |
1da177e4 | 8367 | #endif |
6711cab4 | 8368 | if (sg) |
6c99e9ad | 8369 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8370 | return group; |
1da177e4 LT |
8371 | } |
8372 | ||
8373 | #ifdef CONFIG_NUMA | |
1da177e4 | 8374 | /* |
9c1cfda2 JH |
8375 | * The init_sched_build_groups can't handle what we want to do with node |
8376 | * groups, so roll our own. Now each node has its own list of groups which | |
8377 | * gets dynamically allocated. | |
1da177e4 | 8378 | */ |
62ea9ceb | 8379 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8380 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8381 | |
62ea9ceb | 8382 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8383 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8384 | |
96f874e2 RR |
8385 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8386 | struct sched_group **sg, | |
8387 | struct cpumask *nodemask) | |
9c1cfda2 | 8388 | { |
6711cab4 SS |
8389 | int group; |
8390 | ||
6ca09dfc | 8391 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8392 | group = cpumask_first(nodemask); |
6711cab4 SS |
8393 | |
8394 | if (sg) | |
6c99e9ad | 8395 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8396 | return group; |
1da177e4 | 8397 | } |
6711cab4 | 8398 | |
08069033 SS |
8399 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8400 | { | |
8401 | struct sched_group *sg = group_head; | |
8402 | int j; | |
8403 | ||
8404 | if (!sg) | |
8405 | return; | |
3a5c359a | 8406 | do { |
758b2cdc | 8407 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8408 | struct sched_domain *sd; |
08069033 | 8409 | |
6c99e9ad | 8410 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8411 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8412 | /* |
8413 | * Only add "power" once for each | |
8414 | * physical package. | |
8415 | */ | |
8416 | continue; | |
8417 | } | |
08069033 | 8418 | |
18a3885f | 8419 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
8420 | } |
8421 | sg = sg->next; | |
8422 | } while (sg != group_head); | |
08069033 | 8423 | } |
0601a88d AH |
8424 | |
8425 | static int build_numa_sched_groups(struct s_data *d, | |
8426 | const struct cpumask *cpu_map, int num) | |
8427 | { | |
8428 | struct sched_domain *sd; | |
8429 | struct sched_group *sg, *prev; | |
8430 | int n, j; | |
8431 | ||
8432 | cpumask_clear(d->covered); | |
8433 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8434 | if (cpumask_empty(d->nodemask)) { | |
8435 | d->sched_group_nodes[num] = NULL; | |
8436 | goto out; | |
8437 | } | |
8438 | ||
8439 | sched_domain_node_span(num, d->domainspan); | |
8440 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8441 | ||
8442 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8443 | GFP_KERNEL, num); | |
8444 | if (!sg) { | |
663997d4 | 8445 | pr_warning("Can not alloc domain group for node %d\n", num); |
0601a88d AH |
8446 | return -ENOMEM; |
8447 | } | |
8448 | d->sched_group_nodes[num] = sg; | |
8449 | ||
8450 | for_each_cpu(j, d->nodemask) { | |
8451 | sd = &per_cpu(node_domains, j).sd; | |
8452 | sd->groups = sg; | |
8453 | } | |
8454 | ||
18a3885f | 8455 | sg->cpu_power = 0; |
0601a88d AH |
8456 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
8457 | sg->next = sg; | |
8458 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8459 | ||
8460 | prev = sg; | |
8461 | for (j = 0; j < nr_node_ids; j++) { | |
8462 | n = (num + j) % nr_node_ids; | |
8463 | cpumask_complement(d->notcovered, d->covered); | |
8464 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8465 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8466 | if (cpumask_empty(d->tmpmask)) | |
8467 | break; | |
8468 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8469 | if (cpumask_empty(d->tmpmask)) | |
8470 | continue; | |
8471 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8472 | GFP_KERNEL, num); | |
8473 | if (!sg) { | |
663997d4 JP |
8474 | pr_warning("Can not alloc domain group for node %d\n", |
8475 | j); | |
0601a88d AH |
8476 | return -ENOMEM; |
8477 | } | |
18a3885f | 8478 | sg->cpu_power = 0; |
0601a88d AH |
8479 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
8480 | sg->next = prev->next; | |
8481 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8482 | prev->next = sg; | |
8483 | prev = sg; | |
8484 | } | |
8485 | out: | |
8486 | return 0; | |
8487 | } | |
6d6bc0ad | 8488 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8489 | |
a616058b | 8490 | #ifdef CONFIG_NUMA |
51888ca2 | 8491 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8492 | static void free_sched_groups(const struct cpumask *cpu_map, |
8493 | struct cpumask *nodemask) | |
51888ca2 | 8494 | { |
a616058b | 8495 | int cpu, i; |
51888ca2 | 8496 | |
abcd083a | 8497 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8498 | struct sched_group **sched_group_nodes |
8499 | = sched_group_nodes_bycpu[cpu]; | |
8500 | ||
51888ca2 SV |
8501 | if (!sched_group_nodes) |
8502 | continue; | |
8503 | ||
076ac2af | 8504 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8505 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8506 | ||
6ca09dfc | 8507 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8508 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8509 | continue; |
8510 | ||
8511 | if (sg == NULL) | |
8512 | continue; | |
8513 | sg = sg->next; | |
8514 | next_sg: | |
8515 | oldsg = sg; | |
8516 | sg = sg->next; | |
8517 | kfree(oldsg); | |
8518 | if (oldsg != sched_group_nodes[i]) | |
8519 | goto next_sg; | |
8520 | } | |
8521 | kfree(sched_group_nodes); | |
8522 | sched_group_nodes_bycpu[cpu] = NULL; | |
8523 | } | |
51888ca2 | 8524 | } |
6d6bc0ad | 8525 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8526 | static void free_sched_groups(const struct cpumask *cpu_map, |
8527 | struct cpumask *nodemask) | |
a616058b SS |
8528 | { |
8529 | } | |
6d6bc0ad | 8530 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8531 | |
89c4710e SS |
8532 | /* |
8533 | * Initialize sched groups cpu_power. | |
8534 | * | |
8535 | * cpu_power indicates the capacity of sched group, which is used while | |
8536 | * distributing the load between different sched groups in a sched domain. | |
8537 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8538 | * there are asymmetries in the topology. If there are asymmetries, group | |
8539 | * having more cpu_power will pickup more load compared to the group having | |
8540 | * less cpu_power. | |
89c4710e SS |
8541 | */ |
8542 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8543 | { | |
8544 | struct sched_domain *child; | |
8545 | struct sched_group *group; | |
f93e65c1 PZ |
8546 | long power; |
8547 | int weight; | |
89c4710e SS |
8548 | |
8549 | WARN_ON(!sd || !sd->groups); | |
8550 | ||
13318a71 | 8551 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8552 | return; |
8553 | ||
8554 | child = sd->child; | |
8555 | ||
18a3885f | 8556 | sd->groups->cpu_power = 0; |
5517d86b | 8557 | |
f93e65c1 PZ |
8558 | if (!child) { |
8559 | power = SCHED_LOAD_SCALE; | |
8560 | weight = cpumask_weight(sched_domain_span(sd)); | |
8561 | /* | |
8562 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8563 | * Usually multiple threads get a better yield out of |
8564 | * that one core than a single thread would have, | |
8565 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8566 | */ |
a52bfd73 PZ |
8567 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8568 | power *= sd->smt_gain; | |
f93e65c1 | 8569 | power /= weight; |
a52bfd73 PZ |
8570 | power >>= SCHED_LOAD_SHIFT; |
8571 | } | |
18a3885f | 8572 | sd->groups->cpu_power += power; |
89c4710e SS |
8573 | return; |
8574 | } | |
8575 | ||
89c4710e | 8576 | /* |
f93e65c1 | 8577 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8578 | */ |
8579 | group = child->groups; | |
8580 | do { | |
18a3885f | 8581 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
8582 | group = group->next; |
8583 | } while (group != child->groups); | |
8584 | } | |
8585 | ||
7c16ec58 MT |
8586 | /* |
8587 | * Initializers for schedule domains | |
8588 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8589 | */ | |
8590 | ||
a5d8c348 IM |
8591 | #ifdef CONFIG_SCHED_DEBUG |
8592 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8593 | #else | |
8594 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8595 | #endif | |
8596 | ||
7c16ec58 | 8597 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8598 | |
7c16ec58 MT |
8599 | #define SD_INIT_FUNC(type) \ |
8600 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8601 | { \ | |
8602 | memset(sd, 0, sizeof(*sd)); \ | |
8603 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8604 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8605 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8606 | } |
8607 | ||
8608 | SD_INIT_FUNC(CPU) | |
8609 | #ifdef CONFIG_NUMA | |
8610 | SD_INIT_FUNC(ALLNODES) | |
8611 | SD_INIT_FUNC(NODE) | |
8612 | #endif | |
8613 | #ifdef CONFIG_SCHED_SMT | |
8614 | SD_INIT_FUNC(SIBLING) | |
8615 | #endif | |
8616 | #ifdef CONFIG_SCHED_MC | |
8617 | SD_INIT_FUNC(MC) | |
8618 | #endif | |
8619 | ||
1d3504fc HS |
8620 | static int default_relax_domain_level = -1; |
8621 | ||
8622 | static int __init setup_relax_domain_level(char *str) | |
8623 | { | |
30e0e178 LZ |
8624 | unsigned long val; |
8625 | ||
8626 | val = simple_strtoul(str, NULL, 0); | |
8627 | if (val < SD_LV_MAX) | |
8628 | default_relax_domain_level = val; | |
8629 | ||
1d3504fc HS |
8630 | return 1; |
8631 | } | |
8632 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8633 | ||
8634 | static void set_domain_attribute(struct sched_domain *sd, | |
8635 | struct sched_domain_attr *attr) | |
8636 | { | |
8637 | int request; | |
8638 | ||
8639 | if (!attr || attr->relax_domain_level < 0) { | |
8640 | if (default_relax_domain_level < 0) | |
8641 | return; | |
8642 | else | |
8643 | request = default_relax_domain_level; | |
8644 | } else | |
8645 | request = attr->relax_domain_level; | |
8646 | if (request < sd->level) { | |
8647 | /* turn off idle balance on this domain */ | |
c88d5910 | 8648 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8649 | } else { |
8650 | /* turn on idle balance on this domain */ | |
c88d5910 | 8651 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
8652 | } |
8653 | } | |
8654 | ||
2109b99e AH |
8655 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8656 | const struct cpumask *cpu_map) | |
8657 | { | |
8658 | switch (what) { | |
8659 | case sa_sched_groups: | |
8660 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8661 | d->sched_group_nodes = NULL; | |
8662 | case sa_rootdomain: | |
8663 | free_rootdomain(d->rd); /* fall through */ | |
8664 | case sa_tmpmask: | |
8665 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8666 | case sa_send_covered: | |
8667 | free_cpumask_var(d->send_covered); /* fall through */ | |
8668 | case sa_this_core_map: | |
8669 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8670 | case sa_this_sibling_map: | |
8671 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8672 | case sa_nodemask: | |
8673 | free_cpumask_var(d->nodemask); /* fall through */ | |
8674 | case sa_sched_group_nodes: | |
d1b55138 | 8675 | #ifdef CONFIG_NUMA |
2109b99e AH |
8676 | kfree(d->sched_group_nodes); /* fall through */ |
8677 | case sa_notcovered: | |
8678 | free_cpumask_var(d->notcovered); /* fall through */ | |
8679 | case sa_covered: | |
8680 | free_cpumask_var(d->covered); /* fall through */ | |
8681 | case sa_domainspan: | |
8682 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8683 | #endif |
2109b99e AH |
8684 | case sa_none: |
8685 | break; | |
8686 | } | |
8687 | } | |
3404c8d9 | 8688 | |
2109b99e AH |
8689 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8690 | const struct cpumask *cpu_map) | |
8691 | { | |
3404c8d9 | 8692 | #ifdef CONFIG_NUMA |
2109b99e AH |
8693 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8694 | return sa_none; | |
8695 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8696 | return sa_domainspan; | |
8697 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8698 | return sa_covered; | |
8699 | /* Allocate the per-node list of sched groups */ | |
8700 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8701 | sizeof(struct sched_group *), GFP_KERNEL); | |
8702 | if (!d->sched_group_nodes) { | |
663997d4 | 8703 | pr_warning("Can not alloc sched group node list\n"); |
2109b99e | 8704 | return sa_notcovered; |
d1b55138 | 8705 | } |
2109b99e | 8706 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8707 | #endif |
2109b99e AH |
8708 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8709 | return sa_sched_group_nodes; | |
8710 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8711 | return sa_nodemask; | |
8712 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8713 | return sa_this_sibling_map; | |
8714 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8715 | return sa_this_core_map; | |
8716 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8717 | return sa_send_covered; | |
8718 | d->rd = alloc_rootdomain(); | |
8719 | if (!d->rd) { | |
663997d4 | 8720 | pr_warning("Cannot alloc root domain\n"); |
2109b99e | 8721 | return sa_tmpmask; |
57d885fe | 8722 | } |
2109b99e AH |
8723 | return sa_rootdomain; |
8724 | } | |
57d885fe | 8725 | |
7f4588f3 AH |
8726 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8727 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8728 | { | |
8729 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8730 | #ifdef CONFIG_NUMA |
7f4588f3 | 8731 | struct sched_domain *parent; |
1da177e4 | 8732 | |
7f4588f3 AH |
8733 | d->sd_allnodes = 0; |
8734 | if (cpumask_weight(cpu_map) > | |
8735 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8736 | sd = &per_cpu(allnodes_domains, i).sd; | |
8737 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8738 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8739 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8740 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8741 | d->sd_allnodes = 1; | |
8742 | } | |
8743 | parent = sd; | |
8744 | ||
8745 | sd = &per_cpu(node_domains, i).sd; | |
8746 | SD_INIT(sd, NODE); | |
8747 | set_domain_attribute(sd, attr); | |
8748 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8749 | sd->parent = parent; | |
8750 | if (parent) | |
8751 | parent->child = sd; | |
8752 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8753 | #endif |
7f4588f3 AH |
8754 | return sd; |
8755 | } | |
1da177e4 | 8756 | |
87cce662 AH |
8757 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8758 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8759 | struct sched_domain *parent, int i) | |
8760 | { | |
8761 | struct sched_domain *sd; | |
8762 | sd = &per_cpu(phys_domains, i).sd; | |
8763 | SD_INIT(sd, CPU); | |
8764 | set_domain_attribute(sd, attr); | |
8765 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8766 | sd->parent = parent; | |
8767 | if (parent) | |
8768 | parent->child = sd; | |
8769 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8770 | return sd; | |
8771 | } | |
1da177e4 | 8772 | |
410c4081 AH |
8773 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8774 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8775 | struct sched_domain *parent, int i) | |
8776 | { | |
8777 | struct sched_domain *sd = parent; | |
1e9f28fa | 8778 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8779 | sd = &per_cpu(core_domains, i).sd; |
8780 | SD_INIT(sd, MC); | |
8781 | set_domain_attribute(sd, attr); | |
8782 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8783 | sd->parent = parent; | |
8784 | parent->child = sd; | |
8785 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8786 | #endif |
410c4081 AH |
8787 | return sd; |
8788 | } | |
1e9f28fa | 8789 | |
d8173535 AH |
8790 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8791 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8792 | struct sched_domain *parent, int i) | |
8793 | { | |
8794 | struct sched_domain *sd = parent; | |
1da177e4 | 8795 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8796 | sd = &per_cpu(cpu_domains, i).sd; |
8797 | SD_INIT(sd, SIBLING); | |
8798 | set_domain_attribute(sd, attr); | |
8799 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8800 | sd->parent = parent; | |
8801 | parent->child = sd; | |
8802 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8803 | #endif |
d8173535 AH |
8804 | return sd; |
8805 | } | |
1da177e4 | 8806 | |
0e8e85c9 AH |
8807 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8808 | const struct cpumask *cpu_map, int cpu) | |
8809 | { | |
8810 | switch (l) { | |
1da177e4 | 8811 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8812 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8813 | cpumask_and(d->this_sibling_map, cpu_map, | |
8814 | topology_thread_cpumask(cpu)); | |
8815 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8816 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8817 | &cpu_to_cpu_group, | |
8818 | d->send_covered, d->tmpmask); | |
8819 | break; | |
1da177e4 | 8820 | #endif |
1e9f28fa | 8821 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8822 | case SD_LV_MC: /* set up multi-core groups */ |
8823 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8824 | if (cpu == cpumask_first(d->this_core_map)) | |
8825 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8826 | &cpu_to_core_group, | |
8827 | d->send_covered, d->tmpmask); | |
8828 | break; | |
1e9f28fa | 8829 | #endif |
86548096 AH |
8830 | case SD_LV_CPU: /* set up physical groups */ |
8831 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8832 | if (!cpumask_empty(d->nodemask)) | |
8833 | init_sched_build_groups(d->nodemask, cpu_map, | |
8834 | &cpu_to_phys_group, | |
8835 | d->send_covered, d->tmpmask); | |
8836 | break; | |
1da177e4 | 8837 | #ifdef CONFIG_NUMA |
de616e36 AH |
8838 | case SD_LV_ALLNODES: |
8839 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8840 | d->send_covered, d->tmpmask); | |
8841 | break; | |
8842 | #endif | |
0e8e85c9 AH |
8843 | default: |
8844 | break; | |
7c16ec58 | 8845 | } |
0e8e85c9 | 8846 | } |
9c1cfda2 | 8847 | |
2109b99e AH |
8848 | /* |
8849 | * Build sched domains for a given set of cpus and attach the sched domains | |
8850 | * to the individual cpus | |
8851 | */ | |
8852 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8853 | struct sched_domain_attr *attr) | |
8854 | { | |
8855 | enum s_alloc alloc_state = sa_none; | |
8856 | struct s_data d; | |
294b0c96 | 8857 | struct sched_domain *sd; |
2109b99e | 8858 | int i; |
7c16ec58 | 8859 | #ifdef CONFIG_NUMA |
2109b99e | 8860 | d.sd_allnodes = 0; |
7c16ec58 | 8861 | #endif |
9c1cfda2 | 8862 | |
2109b99e AH |
8863 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8864 | if (alloc_state != sa_rootdomain) | |
8865 | goto error; | |
8866 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8867 | |
1da177e4 | 8868 | /* |
1a20ff27 | 8869 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8870 | */ |
abcd083a | 8871 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8872 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8873 | cpu_map); | |
9761eea8 | 8874 | |
7f4588f3 | 8875 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8876 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8877 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8878 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8879 | } |
9c1cfda2 | 8880 | |
abcd083a | 8881 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8882 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8883 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8884 | } |
9c1cfda2 | 8885 | |
1da177e4 | 8886 | /* Set up physical groups */ |
86548096 AH |
8887 | for (i = 0; i < nr_node_ids; i++) |
8888 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8889 | |
1da177e4 LT |
8890 | #ifdef CONFIG_NUMA |
8891 | /* Set up node groups */ | |
de616e36 AH |
8892 | if (d.sd_allnodes) |
8893 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8894 | |
0601a88d AH |
8895 | for (i = 0; i < nr_node_ids; i++) |
8896 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8897 | goto error; |
1da177e4 LT |
8898 | #endif |
8899 | ||
8900 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8901 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8902 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8903 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8904 | init_sched_groups_power(i, sd); |
5c45bf27 | 8905 | } |
1da177e4 | 8906 | #endif |
1e9f28fa | 8907 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8908 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8909 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8910 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8911 | } |
8912 | #endif | |
1e9f28fa | 8913 | |
abcd083a | 8914 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8915 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8916 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8917 | } |
8918 | ||
9c1cfda2 | 8919 | #ifdef CONFIG_NUMA |
076ac2af | 8920 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8921 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8922 | |
49a02c51 | 8923 | if (d.sd_allnodes) { |
6711cab4 | 8924 | struct sched_group *sg; |
f712c0c7 | 8925 | |
96f874e2 | 8926 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 8927 | d.tmpmask); |
f712c0c7 SS |
8928 | init_numa_sched_groups_power(sg); |
8929 | } | |
9c1cfda2 JH |
8930 | #endif |
8931 | ||
1da177e4 | 8932 | /* Attach the domains */ |
abcd083a | 8933 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8934 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 8935 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8936 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8937 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8938 | #else |
6c99e9ad | 8939 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8940 | #endif |
49a02c51 | 8941 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 8942 | } |
51888ca2 | 8943 | |
2109b99e AH |
8944 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
8945 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
8946 | return 0; | |
51888ca2 | 8947 | |
51888ca2 | 8948 | error: |
2109b99e AH |
8949 | __free_domain_allocs(&d, alloc_state, cpu_map); |
8950 | return -ENOMEM; | |
1da177e4 | 8951 | } |
029190c5 | 8952 | |
96f874e2 | 8953 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8954 | { |
8955 | return __build_sched_domains(cpu_map, NULL); | |
8956 | } | |
8957 | ||
acc3f5d7 | 8958 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 8959 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8960 | static struct sched_domain_attr *dattr_cur; |
8961 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8962 | |
8963 | /* | |
8964 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8965 | * cpumask) fails, then fallback to a single sched domain, |
8966 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8967 | */ |
4212823f | 8968 | static cpumask_var_t fallback_doms; |
029190c5 | 8969 | |
ee79d1bd HC |
8970 | /* |
8971 | * arch_update_cpu_topology lets virtualized architectures update the | |
8972 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8973 | * or 0 if it stayed the same. | |
8974 | */ | |
8975 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8976 | { |
ee79d1bd | 8977 | return 0; |
22e52b07 HC |
8978 | } |
8979 | ||
acc3f5d7 RR |
8980 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
8981 | { | |
8982 | int i; | |
8983 | cpumask_var_t *doms; | |
8984 | ||
8985 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
8986 | if (!doms) | |
8987 | return NULL; | |
8988 | for (i = 0; i < ndoms; i++) { | |
8989 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
8990 | free_sched_domains(doms, i); | |
8991 | return NULL; | |
8992 | } | |
8993 | } | |
8994 | return doms; | |
8995 | } | |
8996 | ||
8997 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
8998 | { | |
8999 | unsigned int i; | |
9000 | for (i = 0; i < ndoms; i++) | |
9001 | free_cpumask_var(doms[i]); | |
9002 | kfree(doms); | |
9003 | } | |
9004 | ||
1a20ff27 | 9005 | /* |
41a2d6cf | 9006 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
9007 | * For now this just excludes isolated cpus, but could be used to |
9008 | * exclude other special cases in the future. | |
1a20ff27 | 9009 | */ |
96f874e2 | 9010 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9011 | { |
7378547f MM |
9012 | int err; |
9013 | ||
22e52b07 | 9014 | arch_update_cpu_topology(); |
029190c5 | 9015 | ndoms_cur = 1; |
acc3f5d7 | 9016 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 9017 | if (!doms_cur) |
acc3f5d7 RR |
9018 | doms_cur = &fallback_doms; |
9019 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 9020 | dattr_cur = NULL; |
acc3f5d7 | 9021 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 9022 | register_sched_domain_sysctl(); |
7378547f MM |
9023 | |
9024 | return err; | |
1a20ff27 DG |
9025 | } |
9026 | ||
96f874e2 RR |
9027 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
9028 | struct cpumask *tmpmask) | |
1da177e4 | 9029 | { |
7c16ec58 | 9030 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 9031 | } |
1da177e4 | 9032 | |
1a20ff27 DG |
9033 | /* |
9034 | * Detach sched domains from a group of cpus specified in cpu_map | |
9035 | * These cpus will now be attached to the NULL domain | |
9036 | */ | |
96f874e2 | 9037 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9038 | { |
96f874e2 RR |
9039 | /* Save because hotplug lock held. */ |
9040 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
9041 | int i; |
9042 | ||
abcd083a | 9043 | for_each_cpu(i, cpu_map) |
57d885fe | 9044 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 9045 | synchronize_sched(); |
96f874e2 | 9046 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
9047 | } |
9048 | ||
1d3504fc HS |
9049 | /* handle null as "default" */ |
9050 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
9051 | struct sched_domain_attr *new, int idx_new) | |
9052 | { | |
9053 | struct sched_domain_attr tmp; | |
9054 | ||
9055 | /* fast path */ | |
9056 | if (!new && !cur) | |
9057 | return 1; | |
9058 | ||
9059 | tmp = SD_ATTR_INIT; | |
9060 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
9061 | new ? (new + idx_new) : &tmp, | |
9062 | sizeof(struct sched_domain_attr)); | |
9063 | } | |
9064 | ||
029190c5 PJ |
9065 | /* |
9066 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 9067 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
9068 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
9069 | * It destroys each deleted domain and builds each new domain. | |
9070 | * | |
acc3f5d7 | 9071 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
9072 | * The masks don't intersect (don't overlap.) We should setup one |
9073 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
9074 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
9075 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
9076 | * it as it is. | |
9077 | * | |
acc3f5d7 RR |
9078 | * The passed in 'doms_new' should be allocated using |
9079 | * alloc_sched_domains. This routine takes ownership of it and will | |
9080 | * free_sched_domains it when done with it. If the caller failed the | |
9081 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
9082 | * and partition_sched_domains() will fallback to the single partition | |
9083 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 9084 | * |
96f874e2 | 9085 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
9086 | * ndoms_new == 0 is a special case for destroying existing domains, |
9087 | * and it will not create the default domain. | |
dfb512ec | 9088 | * |
029190c5 PJ |
9089 | * Call with hotplug lock held |
9090 | */ | |
acc3f5d7 | 9091 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 9092 | struct sched_domain_attr *dattr_new) |
029190c5 | 9093 | { |
dfb512ec | 9094 | int i, j, n; |
d65bd5ec | 9095 | int new_topology; |
029190c5 | 9096 | |
712555ee | 9097 | mutex_lock(&sched_domains_mutex); |
a1835615 | 9098 | |
7378547f MM |
9099 | /* always unregister in case we don't destroy any domains */ |
9100 | unregister_sched_domain_sysctl(); | |
9101 | ||
d65bd5ec HC |
9102 | /* Let architecture update cpu core mappings. */ |
9103 | new_topology = arch_update_cpu_topology(); | |
9104 | ||
dfb512ec | 9105 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
9106 | |
9107 | /* Destroy deleted domains */ | |
9108 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 9109 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 9110 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 9111 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9112 | goto match1; |
9113 | } | |
9114 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 9115 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
9116 | match1: |
9117 | ; | |
9118 | } | |
9119 | ||
e761b772 MK |
9120 | if (doms_new == NULL) { |
9121 | ndoms_cur = 0; | |
acc3f5d7 | 9122 | doms_new = &fallback_doms; |
6ad4c188 | 9123 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 9124 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9125 | } |
9126 | ||
029190c5 PJ |
9127 | /* Build new domains */ |
9128 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9129 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 9130 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 9131 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9132 | goto match2; |
9133 | } | |
9134 | /* no match - add a new doms_new */ | |
acc3f5d7 | 9135 | __build_sched_domains(doms_new[i], |
1d3504fc | 9136 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
9137 | match2: |
9138 | ; | |
9139 | } | |
9140 | ||
9141 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
9142 | if (doms_cur != &fallback_doms) |
9143 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 9144 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9145 | doms_cur = doms_new; |
1d3504fc | 9146 | dattr_cur = dattr_new; |
029190c5 | 9147 | ndoms_cur = ndoms_new; |
7378547f MM |
9148 | |
9149 | register_sched_domain_sysctl(); | |
a1835615 | 9150 | |
712555ee | 9151 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9152 | } |
9153 | ||
5c45bf27 | 9154 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9155 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9156 | { |
95402b38 | 9157 | get_online_cpus(); |
dfb512ec MK |
9158 | |
9159 | /* Destroy domains first to force the rebuild */ | |
9160 | partition_sched_domains(0, NULL, NULL); | |
9161 | ||
e761b772 | 9162 | rebuild_sched_domains(); |
95402b38 | 9163 | put_online_cpus(); |
5c45bf27 SS |
9164 | } |
9165 | ||
9166 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9167 | { | |
afb8a9b7 | 9168 | unsigned int level = 0; |
5c45bf27 | 9169 | |
afb8a9b7 GS |
9170 | if (sscanf(buf, "%u", &level) != 1) |
9171 | return -EINVAL; | |
9172 | ||
9173 | /* | |
9174 | * level is always be positive so don't check for | |
9175 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9176 | * What happens on 0 or 1 byte write, | |
9177 | * need to check for count as well? | |
9178 | */ | |
9179 | ||
9180 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9181 | return -EINVAL; |
9182 | ||
9183 | if (smt) | |
afb8a9b7 | 9184 | sched_smt_power_savings = level; |
5c45bf27 | 9185 | else |
afb8a9b7 | 9186 | sched_mc_power_savings = level; |
5c45bf27 | 9187 | |
c70f22d2 | 9188 | arch_reinit_sched_domains(); |
5c45bf27 | 9189 | |
c70f22d2 | 9190 | return count; |
5c45bf27 SS |
9191 | } |
9192 | ||
5c45bf27 | 9193 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9194 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9195 | char *page) | |
5c45bf27 SS |
9196 | { |
9197 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9198 | } | |
f718cd4a | 9199 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9200 | const char *buf, size_t count) |
5c45bf27 SS |
9201 | { |
9202 | return sched_power_savings_store(buf, count, 0); | |
9203 | } | |
f718cd4a AK |
9204 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9205 | sched_mc_power_savings_show, | |
9206 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9207 | #endif |
9208 | ||
9209 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9210 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9211 | char *page) | |
5c45bf27 SS |
9212 | { |
9213 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9214 | } | |
f718cd4a | 9215 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9216 | const char *buf, size_t count) |
5c45bf27 SS |
9217 | { |
9218 | return sched_power_savings_store(buf, count, 1); | |
9219 | } | |
f718cd4a AK |
9220 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9221 | sched_smt_power_savings_show, | |
6707de00 AB |
9222 | sched_smt_power_savings_store); |
9223 | #endif | |
9224 | ||
39aac648 | 9225 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9226 | { |
9227 | int err = 0; | |
9228 | ||
9229 | #ifdef CONFIG_SCHED_SMT | |
9230 | if (smt_capable()) | |
9231 | err = sysfs_create_file(&cls->kset.kobj, | |
9232 | &attr_sched_smt_power_savings.attr); | |
9233 | #endif | |
9234 | #ifdef CONFIG_SCHED_MC | |
9235 | if (!err && mc_capable()) | |
9236 | err = sysfs_create_file(&cls->kset.kobj, | |
9237 | &attr_sched_mc_power_savings.attr); | |
9238 | #endif | |
9239 | return err; | |
9240 | } | |
6d6bc0ad | 9241 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9242 | |
e761b772 | 9243 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9244 | /* |
e761b772 MK |
9245 | * Add online and remove offline CPUs from the scheduler domains. |
9246 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9247 | */ |
9248 | static int update_sched_domains(struct notifier_block *nfb, | |
9249 | unsigned long action, void *hcpu) | |
e761b772 MK |
9250 | { |
9251 | switch (action) { | |
9252 | case CPU_ONLINE: | |
9253 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
9254 | case CPU_DOWN_PREPARE: |
9255 | case CPU_DOWN_PREPARE_FROZEN: | |
9256 | case CPU_DOWN_FAILED: | |
9257 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 9258 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9259 | return NOTIFY_OK; |
9260 | ||
9261 | default: | |
9262 | return NOTIFY_DONE; | |
9263 | } | |
9264 | } | |
9265 | #endif | |
9266 | ||
9267 | static int update_runtime(struct notifier_block *nfb, | |
9268 | unsigned long action, void *hcpu) | |
1da177e4 | 9269 | { |
7def2be1 PZ |
9270 | int cpu = (int)(long)hcpu; |
9271 | ||
1da177e4 | 9272 | switch (action) { |
1da177e4 | 9273 | case CPU_DOWN_PREPARE: |
8bb78442 | 9274 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9275 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9276 | return NOTIFY_OK; |
9277 | ||
1da177e4 | 9278 | case CPU_DOWN_FAILED: |
8bb78442 | 9279 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9280 | case CPU_ONLINE: |
8bb78442 | 9281 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9282 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9283 | return NOTIFY_OK; |
9284 | ||
1da177e4 LT |
9285 | default: |
9286 | return NOTIFY_DONE; | |
9287 | } | |
1da177e4 | 9288 | } |
1da177e4 LT |
9289 | |
9290 | void __init sched_init_smp(void) | |
9291 | { | |
dcc30a35 RR |
9292 | cpumask_var_t non_isolated_cpus; |
9293 | ||
9294 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 9295 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 9296 | |
434d53b0 MT |
9297 | #if defined(CONFIG_NUMA) |
9298 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9299 | GFP_KERNEL); | |
9300 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9301 | #endif | |
95402b38 | 9302 | get_online_cpus(); |
712555ee | 9303 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 9304 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
9305 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
9306 | if (cpumask_empty(non_isolated_cpus)) | |
9307 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9308 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9309 | put_online_cpus(); |
e761b772 MK |
9310 | |
9311 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9312 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9313 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9314 | #endif |
9315 | ||
9316 | /* RT runtime code needs to handle some hotplug events */ | |
9317 | hotcpu_notifier(update_runtime, 0); | |
9318 | ||
b328ca18 | 9319 | init_hrtick(); |
5c1e1767 NP |
9320 | |
9321 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9322 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9323 | BUG(); |
19978ca6 | 9324 | sched_init_granularity(); |
dcc30a35 | 9325 | free_cpumask_var(non_isolated_cpus); |
4212823f | 9326 | |
0e3900e6 | 9327 | init_sched_rt_class(); |
1da177e4 LT |
9328 | } |
9329 | #else | |
9330 | void __init sched_init_smp(void) | |
9331 | { | |
19978ca6 | 9332 | sched_init_granularity(); |
1da177e4 LT |
9333 | } |
9334 | #endif /* CONFIG_SMP */ | |
9335 | ||
cd1bb94b AB |
9336 | const_debug unsigned int sysctl_timer_migration = 1; |
9337 | ||
1da177e4 LT |
9338 | int in_sched_functions(unsigned long addr) |
9339 | { | |
1da177e4 LT |
9340 | return in_lock_functions(addr) || |
9341 | (addr >= (unsigned long)__sched_text_start | |
9342 | && addr < (unsigned long)__sched_text_end); | |
9343 | } | |
9344 | ||
a9957449 | 9345 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9346 | { |
9347 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9348 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9349 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9350 | cfs_rq->rq = rq; | |
9351 | #endif | |
67e9fb2a | 9352 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9353 | } |
9354 | ||
fa85ae24 PZ |
9355 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9356 | { | |
9357 | struct rt_prio_array *array; | |
9358 | int i; | |
9359 | ||
9360 | array = &rt_rq->active; | |
9361 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9362 | INIT_LIST_HEAD(array->queue + i); | |
9363 | __clear_bit(i, array->bitmap); | |
9364 | } | |
9365 | /* delimiter for bitsearch: */ | |
9366 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9367 | ||
052f1dc7 | 9368 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9369 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9370 | #ifdef CONFIG_SMP |
e864c499 | 9371 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9372 | #endif |
48d5e258 | 9373 | #endif |
fa85ae24 PZ |
9374 | #ifdef CONFIG_SMP |
9375 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9376 | rt_rq->overloaded = 0; |
05fa785c | 9377 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9378 | #endif |
9379 | ||
9380 | rt_rq->rt_time = 0; | |
9381 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 9382 | rt_rq->rt_runtime = 0; |
0986b11b | 9383 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 9384 | |
052f1dc7 | 9385 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9386 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9387 | rt_rq->rq = rq; |
9388 | #endif | |
fa85ae24 PZ |
9389 | } |
9390 | ||
6f505b16 | 9391 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9392 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9393 | struct sched_entity *se, int cpu, int add, | |
9394 | struct sched_entity *parent) | |
6f505b16 | 9395 | { |
ec7dc8ac | 9396 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9397 | tg->cfs_rq[cpu] = cfs_rq; |
9398 | init_cfs_rq(cfs_rq, rq); | |
9399 | cfs_rq->tg = tg; | |
9400 | if (add) | |
9401 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9402 | ||
9403 | tg->se[cpu] = se; | |
354d60c2 DG |
9404 | /* se could be NULL for init_task_group */ |
9405 | if (!se) | |
9406 | return; | |
9407 | ||
ec7dc8ac DG |
9408 | if (!parent) |
9409 | se->cfs_rq = &rq->cfs; | |
9410 | else | |
9411 | se->cfs_rq = parent->my_q; | |
9412 | ||
6f505b16 PZ |
9413 | se->my_q = cfs_rq; |
9414 | se->load.weight = tg->shares; | |
e05510d0 | 9415 | se->load.inv_weight = 0; |
ec7dc8ac | 9416 | se->parent = parent; |
6f505b16 | 9417 | } |
052f1dc7 | 9418 | #endif |
6f505b16 | 9419 | |
052f1dc7 | 9420 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9421 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9422 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9423 | struct sched_rt_entity *parent) | |
6f505b16 | 9424 | { |
ec7dc8ac DG |
9425 | struct rq *rq = cpu_rq(cpu); |
9426 | ||
6f505b16 PZ |
9427 | tg->rt_rq[cpu] = rt_rq; |
9428 | init_rt_rq(rt_rq, rq); | |
9429 | rt_rq->tg = tg; | |
9430 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9431 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9432 | if (add) |
9433 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9434 | ||
9435 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9436 | if (!rt_se) |
9437 | return; | |
9438 | ||
ec7dc8ac DG |
9439 | if (!parent) |
9440 | rt_se->rt_rq = &rq->rt; | |
9441 | else | |
9442 | rt_se->rt_rq = parent->my_q; | |
9443 | ||
6f505b16 | 9444 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9445 | rt_se->parent = parent; |
6f505b16 PZ |
9446 | INIT_LIST_HEAD(&rt_se->run_list); |
9447 | } | |
9448 | #endif | |
9449 | ||
1da177e4 LT |
9450 | void __init sched_init(void) |
9451 | { | |
dd41f596 | 9452 | int i, j; |
434d53b0 MT |
9453 | unsigned long alloc_size = 0, ptr; |
9454 | ||
9455 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9456 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9457 | #endif | |
9458 | #ifdef CONFIG_RT_GROUP_SCHED | |
9459 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9460 | #endif |
9461 | #ifdef CONFIG_USER_SCHED | |
9462 | alloc_size *= 2; | |
df7c8e84 RR |
9463 | #endif |
9464 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9465 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 9466 | #endif |
434d53b0 | 9467 | if (alloc_size) { |
36b7b6d4 | 9468 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9469 | |
9470 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9471 | init_task_group.se = (struct sched_entity **)ptr; | |
9472 | ptr += nr_cpu_ids * sizeof(void **); | |
9473 | ||
9474 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9475 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9476 | |
9477 | #ifdef CONFIG_USER_SCHED | |
9478 | root_task_group.se = (struct sched_entity **)ptr; | |
9479 | ptr += nr_cpu_ids * sizeof(void **); | |
9480 | ||
9481 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9482 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9483 | #endif /* CONFIG_USER_SCHED */ |
9484 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9485 | #ifdef CONFIG_RT_GROUP_SCHED |
9486 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9487 | ptr += nr_cpu_ids * sizeof(void **); | |
9488 | ||
9489 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9490 | ptr += nr_cpu_ids * sizeof(void **); |
9491 | ||
9492 | #ifdef CONFIG_USER_SCHED | |
9493 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9494 | ptr += nr_cpu_ids * sizeof(void **); | |
9495 | ||
9496 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9497 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9498 | #endif /* CONFIG_USER_SCHED */ |
9499 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9500 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9501 | for_each_possible_cpu(i) { | |
9502 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9503 | ptr += cpumask_size(); | |
9504 | } | |
9505 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9506 | } |
dd41f596 | 9507 | |
57d885fe GH |
9508 | #ifdef CONFIG_SMP |
9509 | init_defrootdomain(); | |
9510 | #endif | |
9511 | ||
d0b27fa7 PZ |
9512 | init_rt_bandwidth(&def_rt_bandwidth, |
9513 | global_rt_period(), global_rt_runtime()); | |
9514 | ||
9515 | #ifdef CONFIG_RT_GROUP_SCHED | |
9516 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9517 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9518 | #ifdef CONFIG_USER_SCHED |
9519 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9520 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9521 | #endif /* CONFIG_USER_SCHED */ |
9522 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9523 | |
052f1dc7 | 9524 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9525 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9526 | INIT_LIST_HEAD(&init_task_group.children); |
9527 | ||
9528 | #ifdef CONFIG_USER_SCHED | |
9529 | INIT_LIST_HEAD(&root_task_group.children); | |
9530 | init_task_group.parent = &root_task_group; | |
9531 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9532 | #endif /* CONFIG_USER_SCHED */ |
9533 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9534 | |
4a6cc4bd JK |
9535 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
9536 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
9537 | __alignof__(unsigned long)); | |
9538 | #endif | |
0a945022 | 9539 | for_each_possible_cpu(i) { |
70b97a7f | 9540 | struct rq *rq; |
1da177e4 LT |
9541 | |
9542 | rq = cpu_rq(i); | |
05fa785c | 9543 | raw_spin_lock_init(&rq->lock); |
7897986b | 9544 | rq->nr_running = 0; |
dce48a84 TG |
9545 | rq->calc_load_active = 0; |
9546 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9547 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9548 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9549 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9550 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9551 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9552 | #ifdef CONFIG_CGROUP_SCHED |
9553 | /* | |
9554 | * How much cpu bandwidth does init_task_group get? | |
9555 | * | |
9556 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9557 | * gets 100% of the cpu resources in the system. This overall | |
9558 | * system cpu resource is divided among the tasks of | |
9559 | * init_task_group and its child task-groups in a fair manner, | |
9560 | * based on each entity's (task or task-group's) weight | |
9561 | * (se->load.weight). | |
9562 | * | |
9563 | * In other words, if init_task_group has 10 tasks of weight | |
9564 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9565 | * then A0's share of the cpu resource is: | |
9566 | * | |
0d905bca | 9567 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9568 | * |
9569 | * We achieve this by letting init_task_group's tasks sit | |
9570 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9571 | */ | |
ec7dc8ac | 9572 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9573 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9574 | root_task_group.shares = NICE_0_LOAD; |
9575 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9576 | /* |
9577 | * In case of task-groups formed thr' the user id of tasks, | |
9578 | * init_task_group represents tasks belonging to root user. | |
9579 | * Hence it forms a sibling of all subsequent groups formed. | |
9580 | * In this case, init_task_group gets only a fraction of overall | |
9581 | * system cpu resource, based on the weight assigned to root | |
9582 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9583 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9584 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9585 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9586 | */ | |
ec7dc8ac | 9587 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9588 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9589 | &per_cpu(init_sched_entity, i), i, 1, |
9590 | root_task_group.se[i]); | |
6f505b16 | 9591 | |
052f1dc7 | 9592 | #endif |
354d60c2 DG |
9593 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9594 | ||
9595 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9596 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9597 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9598 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9599 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9600 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9601 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9602 | init_tg_rt_entry(&init_task_group, |
1871e52c | 9603 | &per_cpu(init_rt_rq_var, i), |
eff766a6 PZ |
9604 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9605 | root_task_group.rt_se[i]); | |
354d60c2 | 9606 | #endif |
dd41f596 | 9607 | #endif |
1da177e4 | 9608 | |
dd41f596 IM |
9609 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9610 | rq->cpu_load[j] = 0; | |
1da177e4 | 9611 | #ifdef CONFIG_SMP |
41c7ce9a | 9612 | rq->sd = NULL; |
57d885fe | 9613 | rq->rd = NULL; |
3f029d3c | 9614 | rq->post_schedule = 0; |
1da177e4 | 9615 | rq->active_balance = 0; |
dd41f596 | 9616 | rq->next_balance = jiffies; |
1da177e4 | 9617 | rq->push_cpu = 0; |
0a2966b4 | 9618 | rq->cpu = i; |
1f11eb6a | 9619 | rq->online = 0; |
1da177e4 | 9620 | rq->migration_thread = NULL; |
eae0c9df MG |
9621 | rq->idle_stamp = 0; |
9622 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 9623 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 9624 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9625 | #endif |
8f4d37ec | 9626 | init_rq_hrtick(rq); |
1da177e4 | 9627 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9628 | } |
9629 | ||
2dd73a4f | 9630 | set_load_weight(&init_task); |
b50f60ce | 9631 | |
e107be36 AK |
9632 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9633 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9634 | #endif | |
9635 | ||
c9819f45 | 9636 | #ifdef CONFIG_SMP |
962cf36c | 9637 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9638 | #endif |
9639 | ||
b50f60ce | 9640 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 9641 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
9642 | #endif |
9643 | ||
1da177e4 LT |
9644 | /* |
9645 | * The boot idle thread does lazy MMU switching as well: | |
9646 | */ | |
9647 | atomic_inc(&init_mm.mm_count); | |
9648 | enter_lazy_tlb(&init_mm, current); | |
9649 | ||
9650 | /* | |
9651 | * Make us the idle thread. Technically, schedule() should not be | |
9652 | * called from this thread, however somewhere below it might be, | |
9653 | * but because we are the idle thread, we just pick up running again | |
9654 | * when this runqueue becomes "idle". | |
9655 | */ | |
9656 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9657 | |
9658 | calc_load_update = jiffies + LOAD_FREQ; | |
9659 | ||
dd41f596 IM |
9660 | /* |
9661 | * During early bootup we pretend to be a normal task: | |
9662 | */ | |
9663 | current->sched_class = &fair_sched_class; | |
6892b75e | 9664 | |
6a7b3dc3 | 9665 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 9666 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9667 | #ifdef CONFIG_SMP |
7d1e6a9b | 9668 | #ifdef CONFIG_NO_HZ |
49557e62 | 9669 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 9670 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 9671 | #endif |
bdddd296 RR |
9672 | /* May be allocated at isolcpus cmdline parse time */ |
9673 | if (cpu_isolated_map == NULL) | |
9674 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 9675 | #endif /* SMP */ |
6a7b3dc3 | 9676 | |
cdd6c482 | 9677 | perf_event_init(); |
0d905bca | 9678 | |
6892b75e | 9679 | scheduler_running = 1; |
1da177e4 LT |
9680 | } |
9681 | ||
9682 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9683 | static inline int preempt_count_equals(int preempt_offset) |
9684 | { | |
234da7bc | 9685 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
9686 | |
9687 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9688 | } | |
9689 | ||
9690 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9691 | { |
48f24c4d | 9692 | #ifdef in_atomic |
1da177e4 LT |
9693 | static unsigned long prev_jiffy; /* ratelimiting */ |
9694 | ||
e4aafea2 FW |
9695 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9696 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9697 | return; |
9698 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9699 | return; | |
9700 | prev_jiffy = jiffies; | |
9701 | ||
663997d4 JP |
9702 | pr_err("BUG: sleeping function called from invalid context at %s:%d\n", |
9703 | file, line); | |
9704 | pr_err("in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9705 | in_atomic(), irqs_disabled(), | |
9706 | current->pid, current->comm); | |
aef745fc IM |
9707 | |
9708 | debug_show_held_locks(current); | |
9709 | if (irqs_disabled()) | |
9710 | print_irqtrace_events(current); | |
9711 | dump_stack(); | |
1da177e4 LT |
9712 | #endif |
9713 | } | |
9714 | EXPORT_SYMBOL(__might_sleep); | |
9715 | #endif | |
9716 | ||
9717 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9718 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9719 | { | |
9720 | int on_rq; | |
3e51f33f | 9721 | |
3a5e4dc1 AK |
9722 | update_rq_clock(rq); |
9723 | on_rq = p->se.on_rq; | |
9724 | if (on_rq) | |
9725 | deactivate_task(rq, p, 0); | |
9726 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9727 | if (on_rq) { | |
9728 | activate_task(rq, p, 0); | |
9729 | resched_task(rq->curr); | |
9730 | } | |
9731 | } | |
9732 | ||
1da177e4 LT |
9733 | void normalize_rt_tasks(void) |
9734 | { | |
a0f98a1c | 9735 | struct task_struct *g, *p; |
1da177e4 | 9736 | unsigned long flags; |
70b97a7f | 9737 | struct rq *rq; |
1da177e4 | 9738 | |
4cf5d77a | 9739 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9740 | do_each_thread(g, p) { |
178be793 IM |
9741 | /* |
9742 | * Only normalize user tasks: | |
9743 | */ | |
9744 | if (!p->mm) | |
9745 | continue; | |
9746 | ||
6cfb0d5d | 9747 | p->se.exec_start = 0; |
6cfb0d5d | 9748 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9749 | p->se.wait_start = 0; |
dd41f596 | 9750 | p->se.sleep_start = 0; |
dd41f596 | 9751 | p->se.block_start = 0; |
6cfb0d5d | 9752 | #endif |
dd41f596 IM |
9753 | |
9754 | if (!rt_task(p)) { | |
9755 | /* | |
9756 | * Renice negative nice level userspace | |
9757 | * tasks back to 0: | |
9758 | */ | |
9759 | if (TASK_NICE(p) < 0 && p->mm) | |
9760 | set_user_nice(p, 0); | |
1da177e4 | 9761 | continue; |
dd41f596 | 9762 | } |
1da177e4 | 9763 | |
1d615482 | 9764 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 9765 | rq = __task_rq_lock(p); |
1da177e4 | 9766 | |
178be793 | 9767 | normalize_task(rq, p); |
3a5e4dc1 | 9768 | |
b29739f9 | 9769 | __task_rq_unlock(rq); |
1d615482 | 9770 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9771 | } while_each_thread(g, p); |
9772 | ||
4cf5d77a | 9773 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9774 | } |
9775 | ||
9776 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9777 | |
9778 | #ifdef CONFIG_IA64 | |
9779 | /* | |
9780 | * These functions are only useful for the IA64 MCA handling. | |
9781 | * | |
9782 | * They can only be called when the whole system has been | |
9783 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9784 | * activity can take place. Using them for anything else would | |
9785 | * be a serious bug, and as a result, they aren't even visible | |
9786 | * under any other configuration. | |
9787 | */ | |
9788 | ||
9789 | /** | |
9790 | * curr_task - return the current task for a given cpu. | |
9791 | * @cpu: the processor in question. | |
9792 | * | |
9793 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9794 | */ | |
36c8b586 | 9795 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9796 | { |
9797 | return cpu_curr(cpu); | |
9798 | } | |
9799 | ||
9800 | /** | |
9801 | * set_curr_task - set the current task for a given cpu. | |
9802 | * @cpu: the processor in question. | |
9803 | * @p: the task pointer to set. | |
9804 | * | |
9805 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9806 | * are serviced on a separate stack. It allows the architecture to switch the |
9807 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9808 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9809 | * and caller must save the original value of the current task (see | |
9810 | * curr_task() above) and restore that value before reenabling interrupts and | |
9811 | * re-starting the system. | |
9812 | * | |
9813 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9814 | */ | |
36c8b586 | 9815 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9816 | { |
9817 | cpu_curr(cpu) = p; | |
9818 | } | |
9819 | ||
9820 | #endif | |
29f59db3 | 9821 | |
bccbe08a PZ |
9822 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9823 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9824 | { |
9825 | int i; | |
9826 | ||
9827 | for_each_possible_cpu(i) { | |
9828 | if (tg->cfs_rq) | |
9829 | kfree(tg->cfs_rq[i]); | |
9830 | if (tg->se) | |
9831 | kfree(tg->se[i]); | |
6f505b16 PZ |
9832 | } |
9833 | ||
9834 | kfree(tg->cfs_rq); | |
9835 | kfree(tg->se); | |
6f505b16 PZ |
9836 | } |
9837 | ||
ec7dc8ac DG |
9838 | static |
9839 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9840 | { |
29f59db3 | 9841 | struct cfs_rq *cfs_rq; |
eab17229 | 9842 | struct sched_entity *se; |
9b5b7751 | 9843 | struct rq *rq; |
29f59db3 SV |
9844 | int i; |
9845 | ||
434d53b0 | 9846 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9847 | if (!tg->cfs_rq) |
9848 | goto err; | |
434d53b0 | 9849 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9850 | if (!tg->se) |
9851 | goto err; | |
052f1dc7 PZ |
9852 | |
9853 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9854 | |
9855 | for_each_possible_cpu(i) { | |
9b5b7751 | 9856 | rq = cpu_rq(i); |
29f59db3 | 9857 | |
eab17229 LZ |
9858 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9859 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9860 | if (!cfs_rq) |
9861 | goto err; | |
9862 | ||
eab17229 LZ |
9863 | se = kzalloc_node(sizeof(struct sched_entity), |
9864 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 9865 | if (!se) |
dfc12eb2 | 9866 | goto err_free_rq; |
29f59db3 | 9867 | |
eab17229 | 9868 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9869 | } |
9870 | ||
9871 | return 1; | |
9872 | ||
dfc12eb2 PC |
9873 | err_free_rq: |
9874 | kfree(cfs_rq); | |
bccbe08a PZ |
9875 | err: |
9876 | return 0; | |
9877 | } | |
9878 | ||
9879 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9880 | { | |
9881 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9882 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9883 | } | |
9884 | ||
9885 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9886 | { | |
9887 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9888 | } | |
6d6bc0ad | 9889 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9890 | static inline void free_fair_sched_group(struct task_group *tg) |
9891 | { | |
9892 | } | |
9893 | ||
ec7dc8ac DG |
9894 | static inline |
9895 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9896 | { |
9897 | return 1; | |
9898 | } | |
9899 | ||
9900 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9901 | { | |
9902 | } | |
9903 | ||
9904 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9905 | { | |
9906 | } | |
6d6bc0ad | 9907 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9908 | |
9909 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9910 | static void free_rt_sched_group(struct task_group *tg) |
9911 | { | |
9912 | int i; | |
9913 | ||
d0b27fa7 PZ |
9914 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9915 | ||
bccbe08a PZ |
9916 | for_each_possible_cpu(i) { |
9917 | if (tg->rt_rq) | |
9918 | kfree(tg->rt_rq[i]); | |
9919 | if (tg->rt_se) | |
9920 | kfree(tg->rt_se[i]); | |
9921 | } | |
9922 | ||
9923 | kfree(tg->rt_rq); | |
9924 | kfree(tg->rt_se); | |
9925 | } | |
9926 | ||
ec7dc8ac DG |
9927 | static |
9928 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9929 | { |
9930 | struct rt_rq *rt_rq; | |
eab17229 | 9931 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9932 | struct rq *rq; |
9933 | int i; | |
9934 | ||
434d53b0 | 9935 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9936 | if (!tg->rt_rq) |
9937 | goto err; | |
434d53b0 | 9938 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9939 | if (!tg->rt_se) |
9940 | goto err; | |
9941 | ||
d0b27fa7 PZ |
9942 | init_rt_bandwidth(&tg->rt_bandwidth, |
9943 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9944 | |
9945 | for_each_possible_cpu(i) { | |
9946 | rq = cpu_rq(i); | |
9947 | ||
eab17229 LZ |
9948 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9949 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9950 | if (!rt_rq) |
9951 | goto err; | |
29f59db3 | 9952 | |
eab17229 LZ |
9953 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9954 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 9955 | if (!rt_se) |
dfc12eb2 | 9956 | goto err_free_rq; |
29f59db3 | 9957 | |
eab17229 | 9958 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9959 | } |
9960 | ||
bccbe08a PZ |
9961 | return 1; |
9962 | ||
dfc12eb2 PC |
9963 | err_free_rq: |
9964 | kfree(rt_rq); | |
bccbe08a PZ |
9965 | err: |
9966 | return 0; | |
9967 | } | |
9968 | ||
9969 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9970 | { | |
9971 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9972 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9973 | } | |
9974 | ||
9975 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9976 | { | |
9977 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9978 | } | |
6d6bc0ad | 9979 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9980 | static inline void free_rt_sched_group(struct task_group *tg) |
9981 | { | |
9982 | } | |
9983 | ||
ec7dc8ac DG |
9984 | static inline |
9985 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9986 | { |
9987 | return 1; | |
9988 | } | |
9989 | ||
9990 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9991 | { | |
9992 | } | |
9993 | ||
9994 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9995 | { | |
9996 | } | |
6d6bc0ad | 9997 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9998 | |
d0b27fa7 | 9999 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
10000 | static void free_sched_group(struct task_group *tg) |
10001 | { | |
10002 | free_fair_sched_group(tg); | |
10003 | free_rt_sched_group(tg); | |
10004 | kfree(tg); | |
10005 | } | |
10006 | ||
10007 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 10008 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
10009 | { |
10010 | struct task_group *tg; | |
10011 | unsigned long flags; | |
10012 | int i; | |
10013 | ||
10014 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
10015 | if (!tg) | |
10016 | return ERR_PTR(-ENOMEM); | |
10017 | ||
ec7dc8ac | 10018 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
10019 | goto err; |
10020 | ||
ec7dc8ac | 10021 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
10022 | goto err; |
10023 | ||
8ed36996 | 10024 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10025 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10026 | register_fair_sched_group(tg, i); |
10027 | register_rt_sched_group(tg, i); | |
9b5b7751 | 10028 | } |
6f505b16 | 10029 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
10030 | |
10031 | WARN_ON(!parent); /* root should already exist */ | |
10032 | ||
10033 | tg->parent = parent; | |
f473aa5e | 10034 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10035 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10036 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 10037 | |
9b5b7751 | 10038 | return tg; |
29f59db3 SV |
10039 | |
10040 | err: | |
6f505b16 | 10041 | free_sched_group(tg); |
29f59db3 SV |
10042 | return ERR_PTR(-ENOMEM); |
10043 | } | |
10044 | ||
9b5b7751 | 10045 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 10046 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10047 | { |
29f59db3 | 10048 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 10049 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10050 | } |
10051 | ||
9b5b7751 | 10052 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 10053 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 10054 | { |
8ed36996 | 10055 | unsigned long flags; |
9b5b7751 | 10056 | int i; |
29f59db3 | 10057 | |
8ed36996 | 10058 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10059 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10060 | unregister_fair_sched_group(tg, i); |
10061 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 10062 | } |
6f505b16 | 10063 | list_del_rcu(&tg->list); |
f473aa5e | 10064 | list_del_rcu(&tg->siblings); |
8ed36996 | 10065 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 10066 | |
9b5b7751 | 10067 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 10068 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
10069 | } |
10070 | ||
9b5b7751 | 10071 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
10072 | * The caller of this function should have put the task in its new group |
10073 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
10074 | * reflect its new group. | |
9b5b7751 SV |
10075 | */ |
10076 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
10077 | { |
10078 | int on_rq, running; | |
10079 | unsigned long flags; | |
10080 | struct rq *rq; | |
10081 | ||
10082 | rq = task_rq_lock(tsk, &flags); | |
10083 | ||
29f59db3 SV |
10084 | update_rq_clock(rq); |
10085 | ||
051a1d1a | 10086 | running = task_current(rq, tsk); |
29f59db3 SV |
10087 | on_rq = tsk->se.on_rq; |
10088 | ||
0e1f3483 | 10089 | if (on_rq) |
29f59db3 | 10090 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
10091 | if (unlikely(running)) |
10092 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 10093 | |
6f505b16 | 10094 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 10095 | |
810b3817 PZ |
10096 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10097 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 10098 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
10099 | #endif |
10100 | ||
0e1f3483 HS |
10101 | if (unlikely(running)) |
10102 | tsk->sched_class->set_curr_task(rq); | |
10103 | if (on_rq) | |
7074badb | 10104 | enqueue_task(rq, tsk, 0); |
29f59db3 | 10105 | |
29f59db3 SV |
10106 | task_rq_unlock(rq, &flags); |
10107 | } | |
6d6bc0ad | 10108 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 10109 | |
052f1dc7 | 10110 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 10111 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
10112 | { |
10113 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
10114 | int on_rq; |
10115 | ||
29f59db3 | 10116 | on_rq = se->on_rq; |
62fb1851 | 10117 | if (on_rq) |
29f59db3 SV |
10118 | dequeue_entity(cfs_rq, se, 0); |
10119 | ||
10120 | se->load.weight = shares; | |
e05510d0 | 10121 | se->load.inv_weight = 0; |
29f59db3 | 10122 | |
62fb1851 | 10123 | if (on_rq) |
29f59db3 | 10124 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10125 | } |
62fb1851 | 10126 | |
c09595f6 PZ |
10127 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10128 | { | |
10129 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10130 | struct rq *rq = cfs_rq->rq; | |
10131 | unsigned long flags; | |
10132 | ||
05fa785c | 10133 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 10134 | __set_se_shares(se, shares); |
05fa785c | 10135 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
10136 | } |
10137 | ||
8ed36996 PZ |
10138 | static DEFINE_MUTEX(shares_mutex); |
10139 | ||
4cf86d77 | 10140 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10141 | { |
10142 | int i; | |
8ed36996 | 10143 | unsigned long flags; |
c61935fd | 10144 | |
ec7dc8ac DG |
10145 | /* |
10146 | * We can't change the weight of the root cgroup. | |
10147 | */ | |
10148 | if (!tg->se[0]) | |
10149 | return -EINVAL; | |
10150 | ||
18d95a28 PZ |
10151 | if (shares < MIN_SHARES) |
10152 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10153 | else if (shares > MAX_SHARES) |
10154 | shares = MAX_SHARES; | |
62fb1851 | 10155 | |
8ed36996 | 10156 | mutex_lock(&shares_mutex); |
9b5b7751 | 10157 | if (tg->shares == shares) |
5cb350ba | 10158 | goto done; |
29f59db3 | 10159 | |
8ed36996 | 10160 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10161 | for_each_possible_cpu(i) |
10162 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10163 | list_del_rcu(&tg->siblings); |
8ed36996 | 10164 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10165 | |
10166 | /* wait for any ongoing reference to this group to finish */ | |
10167 | synchronize_sched(); | |
10168 | ||
10169 | /* | |
10170 | * Now we are free to modify the group's share on each cpu | |
10171 | * w/o tripping rebalance_share or load_balance_fair. | |
10172 | */ | |
9b5b7751 | 10173 | tg->shares = shares; |
c09595f6 PZ |
10174 | for_each_possible_cpu(i) { |
10175 | /* | |
10176 | * force a rebalance | |
10177 | */ | |
10178 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10179 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10180 | } |
29f59db3 | 10181 | |
6b2d7700 SV |
10182 | /* |
10183 | * Enable load balance activity on this group, by inserting it back on | |
10184 | * each cpu's rq->leaf_cfs_rq_list. | |
10185 | */ | |
8ed36996 | 10186 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10187 | for_each_possible_cpu(i) |
10188 | register_fair_sched_group(tg, i); | |
f473aa5e | 10189 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10190 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10191 | done: |
8ed36996 | 10192 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10193 | return 0; |
29f59db3 SV |
10194 | } |
10195 | ||
5cb350ba DG |
10196 | unsigned long sched_group_shares(struct task_group *tg) |
10197 | { | |
10198 | return tg->shares; | |
10199 | } | |
052f1dc7 | 10200 | #endif |
5cb350ba | 10201 | |
052f1dc7 | 10202 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10203 | /* |
9f0c1e56 | 10204 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10205 | */ |
9f0c1e56 PZ |
10206 | static DEFINE_MUTEX(rt_constraints_mutex); |
10207 | ||
10208 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10209 | { | |
10210 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10211 | return 1ULL << 20; |
9f0c1e56 | 10212 | |
9a7e0b18 | 10213 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10214 | } |
10215 | ||
9a7e0b18 PZ |
10216 | /* Must be called with tasklist_lock held */ |
10217 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10218 | { |
9a7e0b18 | 10219 | struct task_struct *g, *p; |
b40b2e8e | 10220 | |
9a7e0b18 PZ |
10221 | do_each_thread(g, p) { |
10222 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10223 | return 1; | |
10224 | } while_each_thread(g, p); | |
b40b2e8e | 10225 | |
9a7e0b18 PZ |
10226 | return 0; |
10227 | } | |
b40b2e8e | 10228 | |
9a7e0b18 PZ |
10229 | struct rt_schedulable_data { |
10230 | struct task_group *tg; | |
10231 | u64 rt_period; | |
10232 | u64 rt_runtime; | |
10233 | }; | |
b40b2e8e | 10234 | |
9a7e0b18 PZ |
10235 | static int tg_schedulable(struct task_group *tg, void *data) |
10236 | { | |
10237 | struct rt_schedulable_data *d = data; | |
10238 | struct task_group *child; | |
10239 | unsigned long total, sum = 0; | |
10240 | u64 period, runtime; | |
b40b2e8e | 10241 | |
9a7e0b18 PZ |
10242 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10243 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10244 | |
9a7e0b18 PZ |
10245 | if (tg == d->tg) { |
10246 | period = d->rt_period; | |
10247 | runtime = d->rt_runtime; | |
b40b2e8e | 10248 | } |
b40b2e8e | 10249 | |
98a4826b PZ |
10250 | #ifdef CONFIG_USER_SCHED |
10251 | if (tg == &root_task_group) { | |
10252 | period = global_rt_period(); | |
10253 | runtime = global_rt_runtime(); | |
10254 | } | |
10255 | #endif | |
10256 | ||
4653f803 PZ |
10257 | /* |
10258 | * Cannot have more runtime than the period. | |
10259 | */ | |
10260 | if (runtime > period && runtime != RUNTIME_INF) | |
10261 | return -EINVAL; | |
6f505b16 | 10262 | |
4653f803 PZ |
10263 | /* |
10264 | * Ensure we don't starve existing RT tasks. | |
10265 | */ | |
9a7e0b18 PZ |
10266 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10267 | return -EBUSY; | |
6f505b16 | 10268 | |
9a7e0b18 | 10269 | total = to_ratio(period, runtime); |
6f505b16 | 10270 | |
4653f803 PZ |
10271 | /* |
10272 | * Nobody can have more than the global setting allows. | |
10273 | */ | |
10274 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10275 | return -EINVAL; | |
6f505b16 | 10276 | |
4653f803 PZ |
10277 | /* |
10278 | * The sum of our children's runtime should not exceed our own. | |
10279 | */ | |
9a7e0b18 PZ |
10280 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10281 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10282 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10283 | |
9a7e0b18 PZ |
10284 | if (child == d->tg) { |
10285 | period = d->rt_period; | |
10286 | runtime = d->rt_runtime; | |
10287 | } | |
6f505b16 | 10288 | |
9a7e0b18 | 10289 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10290 | } |
6f505b16 | 10291 | |
9a7e0b18 PZ |
10292 | if (sum > total) |
10293 | return -EINVAL; | |
10294 | ||
10295 | return 0; | |
6f505b16 PZ |
10296 | } |
10297 | ||
9a7e0b18 | 10298 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10299 | { |
9a7e0b18 PZ |
10300 | struct rt_schedulable_data data = { |
10301 | .tg = tg, | |
10302 | .rt_period = period, | |
10303 | .rt_runtime = runtime, | |
10304 | }; | |
10305 | ||
10306 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10307 | } |
10308 | ||
d0b27fa7 PZ |
10309 | static int tg_set_bandwidth(struct task_group *tg, |
10310 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10311 | { |
ac086bc2 | 10312 | int i, err = 0; |
9f0c1e56 | 10313 | |
9f0c1e56 | 10314 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10315 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10316 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10317 | if (err) | |
9f0c1e56 | 10318 | goto unlock; |
ac086bc2 | 10319 | |
0986b11b | 10320 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
10321 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10322 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10323 | |
10324 | for_each_possible_cpu(i) { | |
10325 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10326 | ||
0986b11b | 10327 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 10328 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 10329 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 10330 | } |
0986b11b | 10331 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 10332 | unlock: |
521f1a24 | 10333 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10334 | mutex_unlock(&rt_constraints_mutex); |
10335 | ||
10336 | return err; | |
6f505b16 PZ |
10337 | } |
10338 | ||
d0b27fa7 PZ |
10339 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10340 | { | |
10341 | u64 rt_runtime, rt_period; | |
10342 | ||
10343 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10344 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10345 | if (rt_runtime_us < 0) | |
10346 | rt_runtime = RUNTIME_INF; | |
10347 | ||
10348 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10349 | } | |
10350 | ||
9f0c1e56 PZ |
10351 | long sched_group_rt_runtime(struct task_group *tg) |
10352 | { | |
10353 | u64 rt_runtime_us; | |
10354 | ||
d0b27fa7 | 10355 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10356 | return -1; |
10357 | ||
d0b27fa7 | 10358 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10359 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10360 | return rt_runtime_us; | |
10361 | } | |
d0b27fa7 PZ |
10362 | |
10363 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10364 | { | |
10365 | u64 rt_runtime, rt_period; | |
10366 | ||
10367 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10368 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10369 | ||
619b0488 R |
10370 | if (rt_period == 0) |
10371 | return -EINVAL; | |
10372 | ||
d0b27fa7 PZ |
10373 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10374 | } | |
10375 | ||
10376 | long sched_group_rt_period(struct task_group *tg) | |
10377 | { | |
10378 | u64 rt_period_us; | |
10379 | ||
10380 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10381 | do_div(rt_period_us, NSEC_PER_USEC); | |
10382 | return rt_period_us; | |
10383 | } | |
10384 | ||
10385 | static int sched_rt_global_constraints(void) | |
10386 | { | |
4653f803 | 10387 | u64 runtime, period; |
d0b27fa7 PZ |
10388 | int ret = 0; |
10389 | ||
ec5d4989 HS |
10390 | if (sysctl_sched_rt_period <= 0) |
10391 | return -EINVAL; | |
10392 | ||
4653f803 PZ |
10393 | runtime = global_rt_runtime(); |
10394 | period = global_rt_period(); | |
10395 | ||
10396 | /* | |
10397 | * Sanity check on the sysctl variables. | |
10398 | */ | |
10399 | if (runtime > period && runtime != RUNTIME_INF) | |
10400 | return -EINVAL; | |
10b612f4 | 10401 | |
d0b27fa7 | 10402 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10403 | read_lock(&tasklist_lock); |
4653f803 | 10404 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10405 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10406 | mutex_unlock(&rt_constraints_mutex); |
10407 | ||
10408 | return ret; | |
10409 | } | |
54e99124 DG |
10410 | |
10411 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10412 | { | |
10413 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10414 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10415 | return 0; | |
10416 | ||
10417 | return 1; | |
10418 | } | |
10419 | ||
6d6bc0ad | 10420 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10421 | static int sched_rt_global_constraints(void) |
10422 | { | |
ac086bc2 PZ |
10423 | unsigned long flags; |
10424 | int i; | |
10425 | ||
ec5d4989 HS |
10426 | if (sysctl_sched_rt_period <= 0) |
10427 | return -EINVAL; | |
10428 | ||
60aa605d PZ |
10429 | /* |
10430 | * There's always some RT tasks in the root group | |
10431 | * -- migration, kstopmachine etc.. | |
10432 | */ | |
10433 | if (sysctl_sched_rt_runtime == 0) | |
10434 | return -EBUSY; | |
10435 | ||
0986b11b | 10436 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
10437 | for_each_possible_cpu(i) { |
10438 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10439 | ||
0986b11b | 10440 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 10441 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 10442 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 10443 | } |
0986b11b | 10444 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 10445 | |
d0b27fa7 PZ |
10446 | return 0; |
10447 | } | |
6d6bc0ad | 10448 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10449 | |
10450 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 10451 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
10452 | loff_t *ppos) |
10453 | { | |
10454 | int ret; | |
10455 | int old_period, old_runtime; | |
10456 | static DEFINE_MUTEX(mutex); | |
10457 | ||
10458 | mutex_lock(&mutex); | |
10459 | old_period = sysctl_sched_rt_period; | |
10460 | old_runtime = sysctl_sched_rt_runtime; | |
10461 | ||
8d65af78 | 10462 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
10463 | |
10464 | if (!ret && write) { | |
10465 | ret = sched_rt_global_constraints(); | |
10466 | if (ret) { | |
10467 | sysctl_sched_rt_period = old_period; | |
10468 | sysctl_sched_rt_runtime = old_runtime; | |
10469 | } else { | |
10470 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10471 | def_rt_bandwidth.rt_period = | |
10472 | ns_to_ktime(global_rt_period()); | |
10473 | } | |
10474 | } | |
10475 | mutex_unlock(&mutex); | |
10476 | ||
10477 | return ret; | |
10478 | } | |
68318b8e | 10479 | |
052f1dc7 | 10480 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10481 | |
10482 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10483 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10484 | { |
2b01dfe3 PM |
10485 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10486 | struct task_group, css); | |
68318b8e SV |
10487 | } |
10488 | ||
10489 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10490 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10491 | { |
ec7dc8ac | 10492 | struct task_group *tg, *parent; |
68318b8e | 10493 | |
2b01dfe3 | 10494 | if (!cgrp->parent) { |
68318b8e | 10495 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10496 | return &init_task_group.css; |
10497 | } | |
10498 | ||
ec7dc8ac DG |
10499 | parent = cgroup_tg(cgrp->parent); |
10500 | tg = sched_create_group(parent); | |
68318b8e SV |
10501 | if (IS_ERR(tg)) |
10502 | return ERR_PTR(-ENOMEM); | |
10503 | ||
68318b8e SV |
10504 | return &tg->css; |
10505 | } | |
10506 | ||
41a2d6cf IM |
10507 | static void |
10508 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10509 | { |
2b01dfe3 | 10510 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10511 | |
10512 | sched_destroy_group(tg); | |
10513 | } | |
10514 | ||
41a2d6cf | 10515 | static int |
be367d09 | 10516 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 10517 | { |
b68aa230 | 10518 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10519 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10520 | return -EINVAL; |
10521 | #else | |
68318b8e SV |
10522 | /* We don't support RT-tasks being in separate groups */ |
10523 | if (tsk->sched_class != &fair_sched_class) | |
10524 | return -EINVAL; | |
b68aa230 | 10525 | #endif |
be367d09 BB |
10526 | return 0; |
10527 | } | |
68318b8e | 10528 | |
be367d09 BB |
10529 | static int |
10530 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10531 | struct task_struct *tsk, bool threadgroup) | |
10532 | { | |
10533 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
10534 | if (retval) | |
10535 | return retval; | |
10536 | if (threadgroup) { | |
10537 | struct task_struct *c; | |
10538 | rcu_read_lock(); | |
10539 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10540 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
10541 | if (retval) { | |
10542 | rcu_read_unlock(); | |
10543 | return retval; | |
10544 | } | |
10545 | } | |
10546 | rcu_read_unlock(); | |
10547 | } | |
68318b8e SV |
10548 | return 0; |
10549 | } | |
10550 | ||
10551 | static void | |
2b01dfe3 | 10552 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
10553 | struct cgroup *old_cont, struct task_struct *tsk, |
10554 | bool threadgroup) | |
68318b8e SV |
10555 | { |
10556 | sched_move_task(tsk); | |
be367d09 BB |
10557 | if (threadgroup) { |
10558 | struct task_struct *c; | |
10559 | rcu_read_lock(); | |
10560 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
10561 | sched_move_task(c); | |
10562 | } | |
10563 | rcu_read_unlock(); | |
10564 | } | |
68318b8e SV |
10565 | } |
10566 | ||
052f1dc7 | 10567 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10568 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10569 | u64 shareval) |
68318b8e | 10570 | { |
2b01dfe3 | 10571 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10572 | } |
10573 | ||
f4c753b7 | 10574 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10575 | { |
2b01dfe3 | 10576 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10577 | |
10578 | return (u64) tg->shares; | |
10579 | } | |
6d6bc0ad | 10580 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10581 | |
052f1dc7 | 10582 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10583 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10584 | s64 val) |
6f505b16 | 10585 | { |
06ecb27c | 10586 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10587 | } |
10588 | ||
06ecb27c | 10589 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10590 | { |
06ecb27c | 10591 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10592 | } |
d0b27fa7 PZ |
10593 | |
10594 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10595 | u64 rt_period_us) | |
10596 | { | |
10597 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10598 | } | |
10599 | ||
10600 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10601 | { | |
10602 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10603 | } | |
6d6bc0ad | 10604 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10605 | |
fe5c7cc2 | 10606 | static struct cftype cpu_files[] = { |
052f1dc7 | 10607 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10608 | { |
10609 | .name = "shares", | |
f4c753b7 PM |
10610 | .read_u64 = cpu_shares_read_u64, |
10611 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10612 | }, |
052f1dc7 PZ |
10613 | #endif |
10614 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10615 | { |
9f0c1e56 | 10616 | .name = "rt_runtime_us", |
06ecb27c PM |
10617 | .read_s64 = cpu_rt_runtime_read, |
10618 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10619 | }, |
d0b27fa7 PZ |
10620 | { |
10621 | .name = "rt_period_us", | |
f4c753b7 PM |
10622 | .read_u64 = cpu_rt_period_read_uint, |
10623 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10624 | }, |
052f1dc7 | 10625 | #endif |
68318b8e SV |
10626 | }; |
10627 | ||
10628 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10629 | { | |
fe5c7cc2 | 10630 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10631 | } |
10632 | ||
10633 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10634 | .name = "cpu", |
10635 | .create = cpu_cgroup_create, | |
10636 | .destroy = cpu_cgroup_destroy, | |
10637 | .can_attach = cpu_cgroup_can_attach, | |
10638 | .attach = cpu_cgroup_attach, | |
10639 | .populate = cpu_cgroup_populate, | |
10640 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10641 | .early_init = 1, |
10642 | }; | |
10643 | ||
052f1dc7 | 10644 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10645 | |
10646 | #ifdef CONFIG_CGROUP_CPUACCT | |
10647 | ||
10648 | /* | |
10649 | * CPU accounting code for task groups. | |
10650 | * | |
10651 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10652 | * (balbir@in.ibm.com). | |
10653 | */ | |
10654 | ||
934352f2 | 10655 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10656 | struct cpuacct { |
10657 | struct cgroup_subsys_state css; | |
10658 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10659 | u64 *cpuusage; | |
ef12fefa | 10660 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10661 | struct cpuacct *parent; |
d842de87 SV |
10662 | }; |
10663 | ||
10664 | struct cgroup_subsys cpuacct_subsys; | |
10665 | ||
10666 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10667 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10668 | { |
32cd756a | 10669 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10670 | struct cpuacct, css); |
10671 | } | |
10672 | ||
10673 | /* return cpu accounting group to which this task belongs */ | |
10674 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10675 | { | |
10676 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10677 | struct cpuacct, css); | |
10678 | } | |
10679 | ||
10680 | /* create a new cpu accounting group */ | |
10681 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10682 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10683 | { |
10684 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10685 | int i; |
d842de87 SV |
10686 | |
10687 | if (!ca) | |
ef12fefa | 10688 | goto out; |
d842de87 SV |
10689 | |
10690 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10691 | if (!ca->cpuusage) |
10692 | goto out_free_ca; | |
10693 | ||
10694 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10695 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10696 | goto out_free_counters; | |
d842de87 | 10697 | |
934352f2 BR |
10698 | if (cgrp->parent) |
10699 | ca->parent = cgroup_ca(cgrp->parent); | |
10700 | ||
d842de87 | 10701 | return &ca->css; |
ef12fefa BR |
10702 | |
10703 | out_free_counters: | |
10704 | while (--i >= 0) | |
10705 | percpu_counter_destroy(&ca->cpustat[i]); | |
10706 | free_percpu(ca->cpuusage); | |
10707 | out_free_ca: | |
10708 | kfree(ca); | |
10709 | out: | |
10710 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10711 | } |
10712 | ||
10713 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10714 | static void |
32cd756a | 10715 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10716 | { |
32cd756a | 10717 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10718 | int i; |
d842de87 | 10719 | |
ef12fefa BR |
10720 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10721 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10722 | free_percpu(ca->cpuusage); |
10723 | kfree(ca); | |
10724 | } | |
10725 | ||
720f5498 KC |
10726 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10727 | { | |
b36128c8 | 10728 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10729 | u64 data; |
10730 | ||
10731 | #ifndef CONFIG_64BIT | |
10732 | /* | |
10733 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10734 | */ | |
05fa785c | 10735 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 10736 | data = *cpuusage; |
05fa785c | 10737 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
10738 | #else |
10739 | data = *cpuusage; | |
10740 | #endif | |
10741 | ||
10742 | return data; | |
10743 | } | |
10744 | ||
10745 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10746 | { | |
b36128c8 | 10747 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10748 | |
10749 | #ifndef CONFIG_64BIT | |
10750 | /* | |
10751 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10752 | */ | |
05fa785c | 10753 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 10754 | *cpuusage = val; |
05fa785c | 10755 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
10756 | #else |
10757 | *cpuusage = val; | |
10758 | #endif | |
10759 | } | |
10760 | ||
d842de87 | 10761 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10762 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10763 | { |
32cd756a | 10764 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10765 | u64 totalcpuusage = 0; |
10766 | int i; | |
10767 | ||
720f5498 KC |
10768 | for_each_present_cpu(i) |
10769 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10770 | |
10771 | return totalcpuusage; | |
10772 | } | |
10773 | ||
0297b803 DG |
10774 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10775 | u64 reset) | |
10776 | { | |
10777 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10778 | int err = 0; | |
10779 | int i; | |
10780 | ||
10781 | if (reset) { | |
10782 | err = -EINVAL; | |
10783 | goto out; | |
10784 | } | |
10785 | ||
720f5498 KC |
10786 | for_each_present_cpu(i) |
10787 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10788 | |
0297b803 DG |
10789 | out: |
10790 | return err; | |
10791 | } | |
10792 | ||
e9515c3c KC |
10793 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10794 | struct seq_file *m) | |
10795 | { | |
10796 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10797 | u64 percpu; | |
10798 | int i; | |
10799 | ||
10800 | for_each_present_cpu(i) { | |
10801 | percpu = cpuacct_cpuusage_read(ca, i); | |
10802 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10803 | } | |
10804 | seq_printf(m, "\n"); | |
10805 | return 0; | |
10806 | } | |
10807 | ||
ef12fefa BR |
10808 | static const char *cpuacct_stat_desc[] = { |
10809 | [CPUACCT_STAT_USER] = "user", | |
10810 | [CPUACCT_STAT_SYSTEM] = "system", | |
10811 | }; | |
10812 | ||
10813 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10814 | struct cgroup_map_cb *cb) | |
10815 | { | |
10816 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10817 | int i; | |
10818 | ||
10819 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10820 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10821 | val = cputime64_to_clock_t(val); | |
10822 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10823 | } | |
10824 | return 0; | |
10825 | } | |
10826 | ||
d842de87 SV |
10827 | static struct cftype files[] = { |
10828 | { | |
10829 | .name = "usage", | |
f4c753b7 PM |
10830 | .read_u64 = cpuusage_read, |
10831 | .write_u64 = cpuusage_write, | |
d842de87 | 10832 | }, |
e9515c3c KC |
10833 | { |
10834 | .name = "usage_percpu", | |
10835 | .read_seq_string = cpuacct_percpu_seq_read, | |
10836 | }, | |
ef12fefa BR |
10837 | { |
10838 | .name = "stat", | |
10839 | .read_map = cpuacct_stats_show, | |
10840 | }, | |
d842de87 SV |
10841 | }; |
10842 | ||
32cd756a | 10843 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10844 | { |
32cd756a | 10845 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10846 | } |
10847 | ||
10848 | /* | |
10849 | * charge this task's execution time to its accounting group. | |
10850 | * | |
10851 | * called with rq->lock held. | |
10852 | */ | |
10853 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10854 | { | |
10855 | struct cpuacct *ca; | |
934352f2 | 10856 | int cpu; |
d842de87 | 10857 | |
c40c6f85 | 10858 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10859 | return; |
10860 | ||
934352f2 | 10861 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10862 | |
10863 | rcu_read_lock(); | |
10864 | ||
d842de87 | 10865 | ca = task_ca(tsk); |
d842de87 | 10866 | |
934352f2 | 10867 | for (; ca; ca = ca->parent) { |
b36128c8 | 10868 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10869 | *cpuusage += cputime; |
10870 | } | |
a18b83b7 BR |
10871 | |
10872 | rcu_read_unlock(); | |
d842de87 SV |
10873 | } |
10874 | ||
ef12fefa BR |
10875 | /* |
10876 | * Charge the system/user time to the task's accounting group. | |
10877 | */ | |
10878 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10879 | enum cpuacct_stat_index idx, cputime_t val) | |
10880 | { | |
10881 | struct cpuacct *ca; | |
10882 | ||
10883 | if (unlikely(!cpuacct_subsys.active)) | |
10884 | return; | |
10885 | ||
10886 | rcu_read_lock(); | |
10887 | ca = task_ca(tsk); | |
10888 | ||
10889 | do { | |
10890 | percpu_counter_add(&ca->cpustat[idx], val); | |
10891 | ca = ca->parent; | |
10892 | } while (ca); | |
10893 | rcu_read_unlock(); | |
10894 | } | |
10895 | ||
d842de87 SV |
10896 | struct cgroup_subsys cpuacct_subsys = { |
10897 | .name = "cpuacct", | |
10898 | .create = cpuacct_create, | |
10899 | .destroy = cpuacct_destroy, | |
10900 | .populate = cpuacct_populate, | |
10901 | .subsys_id = cpuacct_subsys_id, | |
10902 | }; | |
10903 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
10904 | |
10905 | #ifndef CONFIG_SMP | |
10906 | ||
10907 | int rcu_expedited_torture_stats(char *page) | |
10908 | { | |
10909 | return 0; | |
10910 | } | |
10911 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10912 | ||
10913 | void synchronize_sched_expedited(void) | |
10914 | { | |
10915 | } | |
10916 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
10917 | ||
10918 | #else /* #ifndef CONFIG_SMP */ | |
10919 | ||
10920 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
10921 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
10922 | ||
10923 | #define RCU_EXPEDITED_STATE_POST -2 | |
10924 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
10925 | ||
10926 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
10927 | ||
10928 | int rcu_expedited_torture_stats(char *page) | |
10929 | { | |
10930 | int cnt = 0; | |
10931 | int cpu; | |
10932 | ||
10933 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
10934 | for_each_online_cpu(cpu) { | |
10935 | cnt += sprintf(&page[cnt], " %d:%d", | |
10936 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
10937 | } | |
10938 | cnt += sprintf(&page[cnt], "\n"); | |
10939 | return cnt; | |
10940 | } | |
10941 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
10942 | ||
10943 | static long synchronize_sched_expedited_count; | |
10944 | ||
10945 | /* | |
10946 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
10947 | * approach to force grace period to end quickly. This consumes | |
10948 | * significant time on all CPUs, and is thus not recommended for | |
10949 | * any sort of common-case code. | |
10950 | * | |
10951 | * Note that it is illegal to call this function while holding any | |
10952 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
10953 | * observe this restriction will result in deadlock. | |
10954 | */ | |
10955 | void synchronize_sched_expedited(void) | |
10956 | { | |
10957 | int cpu; | |
10958 | unsigned long flags; | |
10959 | bool need_full_sync = 0; | |
10960 | struct rq *rq; | |
10961 | struct migration_req *req; | |
10962 | long snap; | |
10963 | int trycount = 0; | |
10964 | ||
10965 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
10966 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
10967 | get_online_cpus(); | |
10968 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
10969 | put_online_cpus(); | |
10970 | if (trycount++ < 10) | |
10971 | udelay(trycount * num_online_cpus()); | |
10972 | else { | |
10973 | synchronize_sched(); | |
10974 | return; | |
10975 | } | |
10976 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
10977 | smp_mb(); /* ensure test happens before caller kfree */ | |
10978 | return; | |
10979 | } | |
10980 | get_online_cpus(); | |
10981 | } | |
10982 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
10983 | for_each_online_cpu(cpu) { | |
10984 | rq = cpu_rq(cpu); | |
10985 | req = &per_cpu(rcu_migration_req, cpu); | |
10986 | init_completion(&req->done); | |
10987 | req->task = NULL; | |
10988 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 10989 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 10990 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 10991 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
10992 | wake_up_process(rq->migration_thread); |
10993 | } | |
10994 | for_each_online_cpu(cpu) { | |
10995 | rcu_expedited_state = cpu; | |
10996 | req = &per_cpu(rcu_migration_req, cpu); | |
10997 | rq = cpu_rq(cpu); | |
10998 | wait_for_completion(&req->done); | |
05fa785c | 10999 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
11000 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
11001 | need_full_sync = 1; | |
11002 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 11003 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
11004 | } |
11005 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 11006 | synchronize_sched_expedited_count++; |
03b042bf PM |
11007 | mutex_unlock(&rcu_sched_expedited_mutex); |
11008 | put_online_cpus(); | |
11009 | if (need_full_sync) | |
11010 | synchronize_sched(); | |
11011 | } | |
11012 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
11013 | ||
11014 | #endif /* #else #ifndef CONFIG_SMP */ |