]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
5517d86b | 66 | #include <linux/reciprocal_div.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
434d53b0 | 71 | #include <linux/bootmem.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
0a16b607 | 75 | #include <trace/sched.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 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
d0b27fa7 PZ |
116 | /* |
117 | * single value that denotes runtime == period, ie unlimited time. | |
118 | */ | |
119 | #define RUNTIME_INF ((u64)~0ULL) | |
120 | ||
7e066fb8 MD |
121 | DEFINE_TRACE(sched_wait_task); |
122 | DEFINE_TRACE(sched_wakeup); | |
123 | DEFINE_TRACE(sched_wakeup_new); | |
124 | DEFINE_TRACE(sched_switch); | |
125 | DEFINE_TRACE(sched_migrate_task); | |
126 | ||
5517d86b | 127 | #ifdef CONFIG_SMP |
fd2ab30b SN |
128 | |
129 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
130 | ||
5517d86b ED |
131 | /* |
132 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
133 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
134 | */ | |
135 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
136 | { | |
137 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Each time a sched group cpu_power is changed, | |
142 | * we must compute its reciprocal value | |
143 | */ | |
144 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
145 | { | |
146 | sg->__cpu_power += val; | |
147 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
148 | } | |
149 | #endif | |
150 | ||
e05606d3 IM |
151 | static inline int rt_policy(int policy) |
152 | { | |
3f33a7ce | 153 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
154 | return 1; |
155 | return 0; | |
156 | } | |
157 | ||
158 | static inline int task_has_rt_policy(struct task_struct *p) | |
159 | { | |
160 | return rt_policy(p->policy); | |
161 | } | |
162 | ||
1da177e4 | 163 | /* |
6aa645ea | 164 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 165 | */ |
6aa645ea IM |
166 | struct rt_prio_array { |
167 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
168 | struct list_head queue[MAX_RT_PRIO]; | |
169 | }; | |
170 | ||
d0b27fa7 | 171 | struct rt_bandwidth { |
ea736ed5 IM |
172 | /* nests inside the rq lock: */ |
173 | spinlock_t rt_runtime_lock; | |
174 | ktime_t rt_period; | |
175 | u64 rt_runtime; | |
176 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
177 | }; |
178 | ||
179 | static struct rt_bandwidth def_rt_bandwidth; | |
180 | ||
181 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
182 | ||
183 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
184 | { | |
185 | struct rt_bandwidth *rt_b = | |
186 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
187 | ktime_t now; | |
188 | int overrun; | |
189 | int idle = 0; | |
190 | ||
191 | for (;;) { | |
192 | now = hrtimer_cb_get_time(timer); | |
193 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
194 | ||
195 | if (!overrun) | |
196 | break; | |
197 | ||
198 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
199 | } | |
200 | ||
201 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
202 | } | |
203 | ||
204 | static | |
205 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
206 | { | |
207 | rt_b->rt_period = ns_to_ktime(period); | |
208 | rt_b->rt_runtime = runtime; | |
209 | ||
ac086bc2 PZ |
210 | spin_lock_init(&rt_b->rt_runtime_lock); |
211 | ||
d0b27fa7 PZ |
212 | hrtimer_init(&rt_b->rt_period_timer, |
213 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
214 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
215 | } |
216 | ||
c8bfff6d KH |
217 | static inline int rt_bandwidth_enabled(void) |
218 | { | |
219 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
220 | } |
221 | ||
222 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
223 | { | |
224 | ktime_t now; | |
225 | ||
cac64d00 | 226 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
227 | return; |
228 | ||
229 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
230 | return; | |
231 | ||
232 | spin_lock(&rt_b->rt_runtime_lock); | |
233 | for (;;) { | |
234 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
235 | break; | |
236 | ||
237 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
238 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
239 | hrtimer_start_expires(&rt_b->rt_period_timer, |
240 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
241 | } |
242 | spin_unlock(&rt_b->rt_runtime_lock); | |
243 | } | |
244 | ||
245 | #ifdef CONFIG_RT_GROUP_SCHED | |
246 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
247 | { | |
248 | hrtimer_cancel(&rt_b->rt_period_timer); | |
249 | } | |
250 | #endif | |
251 | ||
712555ee HC |
252 | /* |
253 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
254 | * detach_destroy_domains and partition_sched_domains. | |
255 | */ | |
256 | static DEFINE_MUTEX(sched_domains_mutex); | |
257 | ||
052f1dc7 | 258 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 259 | |
68318b8e SV |
260 | #include <linux/cgroup.h> |
261 | ||
29f59db3 SV |
262 | struct cfs_rq; |
263 | ||
6f505b16 PZ |
264 | static LIST_HEAD(task_groups); |
265 | ||
29f59db3 | 266 | /* task group related information */ |
4cf86d77 | 267 | struct task_group { |
052f1dc7 | 268 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
269 | struct cgroup_subsys_state css; |
270 | #endif | |
052f1dc7 | 271 | |
6c415b92 AB |
272 | #ifdef CONFIG_USER_SCHED |
273 | uid_t uid; | |
274 | #endif | |
275 | ||
052f1dc7 | 276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
277 | /* schedulable entities of this group on each cpu */ |
278 | struct sched_entity **se; | |
279 | /* runqueue "owned" by this group on each cpu */ | |
280 | struct cfs_rq **cfs_rq; | |
281 | unsigned long shares; | |
052f1dc7 PZ |
282 | #endif |
283 | ||
284 | #ifdef CONFIG_RT_GROUP_SCHED | |
285 | struct sched_rt_entity **rt_se; | |
286 | struct rt_rq **rt_rq; | |
287 | ||
d0b27fa7 | 288 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 289 | #endif |
6b2d7700 | 290 | |
ae8393e5 | 291 | struct rcu_head rcu; |
6f505b16 | 292 | struct list_head list; |
f473aa5e PZ |
293 | |
294 | struct task_group *parent; | |
295 | struct list_head siblings; | |
296 | struct list_head children; | |
29f59db3 SV |
297 | }; |
298 | ||
354d60c2 | 299 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 300 | |
6c415b92 AB |
301 | /* Helper function to pass uid information to create_sched_user() */ |
302 | void set_tg_uid(struct user_struct *user) | |
303 | { | |
304 | user->tg->uid = user->uid; | |
305 | } | |
306 | ||
eff766a6 PZ |
307 | /* |
308 | * Root task group. | |
309 | * Every UID task group (including init_task_group aka UID-0) will | |
310 | * be a child to this group. | |
311 | */ | |
312 | struct task_group root_task_group; | |
313 | ||
052f1dc7 | 314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
315 | /* Default task group's sched entity on each cpu */ |
316 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
317 | /* Default task group's cfs_rq on each cpu */ | |
318 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 319 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
320 | |
321 | #ifdef CONFIG_RT_GROUP_SCHED | |
322 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
323 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 325 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 326 | #define root_task_group init_task_group |
9a7e0b18 | 327 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 328 | |
8ed36996 | 329 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
330 | * a task group's cpu shares. |
331 | */ | |
8ed36996 | 332 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 333 | |
57310a98 PZ |
334 | #ifdef CONFIG_SMP |
335 | static int root_task_group_empty(void) | |
336 | { | |
337 | return list_empty(&root_task_group.children); | |
338 | } | |
339 | #endif | |
340 | ||
052f1dc7 | 341 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
342 | #ifdef CONFIG_USER_SCHED |
343 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 344 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 345 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 346 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 347 | |
cb4ad1ff | 348 | /* |
2e084786 LJ |
349 | * A weight of 0 or 1 can cause arithmetics problems. |
350 | * A weight of a cfs_rq is the sum of weights of which entities | |
351 | * are queued on this cfs_rq, so a weight of a entity should not be | |
352 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
353 | * (The default weight is 1024 - so there's no practical |
354 | * limitation from this.) | |
355 | */ | |
18d95a28 | 356 | #define MIN_SHARES 2 |
2e084786 | 357 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 358 | |
052f1dc7 PZ |
359 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
360 | #endif | |
361 | ||
29f59db3 | 362 | /* Default task group. |
3a252015 | 363 | * Every task in system belong to this group at bootup. |
29f59db3 | 364 | */ |
434d53b0 | 365 | struct task_group init_task_group; |
29f59db3 SV |
366 | |
367 | /* return group to which a task belongs */ | |
4cf86d77 | 368 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 369 | { |
4cf86d77 | 370 | struct task_group *tg; |
9b5b7751 | 371 | |
052f1dc7 | 372 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
373 | rcu_read_lock(); |
374 | tg = __task_cred(p)->user->tg; | |
375 | rcu_read_unlock(); | |
052f1dc7 | 376 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
377 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
378 | struct task_group, css); | |
24e377a8 | 379 | #else |
41a2d6cf | 380 | tg = &init_task_group; |
24e377a8 | 381 | #endif |
9b5b7751 | 382 | return tg; |
29f59db3 SV |
383 | } |
384 | ||
385 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 386 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 387 | { |
052f1dc7 | 388 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
389 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
390 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 391 | #endif |
6f505b16 | 392 | |
052f1dc7 | 393 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
394 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
395 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 396 | #endif |
29f59db3 SV |
397 | } |
398 | ||
399 | #else | |
400 | ||
57310a98 PZ |
401 | #ifdef CONFIG_SMP |
402 | static int root_task_group_empty(void) | |
403 | { | |
404 | return 1; | |
405 | } | |
406 | #endif | |
407 | ||
6f505b16 | 408 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
409 | static inline struct task_group *task_group(struct task_struct *p) |
410 | { | |
411 | return NULL; | |
412 | } | |
29f59db3 | 413 | |
052f1dc7 | 414 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 415 | |
6aa645ea IM |
416 | /* CFS-related fields in a runqueue */ |
417 | struct cfs_rq { | |
418 | struct load_weight load; | |
419 | unsigned long nr_running; | |
420 | ||
6aa645ea | 421 | u64 exec_clock; |
e9acbff6 | 422 | u64 min_vruntime; |
6aa645ea IM |
423 | |
424 | struct rb_root tasks_timeline; | |
425 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
426 | |
427 | struct list_head tasks; | |
428 | struct list_head *balance_iterator; | |
429 | ||
430 | /* | |
431 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
432 | * It is set to NULL otherwise (i.e when none are currently running). |
433 | */ | |
4793241b | 434 | struct sched_entity *curr, *next, *last; |
ddc97297 | 435 | |
5ac5c4d6 | 436 | unsigned int nr_spread_over; |
ddc97297 | 437 | |
62160e3f | 438 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
439 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
440 | ||
41a2d6cf IM |
441 | /* |
442 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
443 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
444 | * (like users, containers etc.) | |
445 | * | |
446 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
447 | * list is used during load balance. | |
448 | */ | |
41a2d6cf IM |
449 | struct list_head leaf_cfs_rq_list; |
450 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
451 | |
452 | #ifdef CONFIG_SMP | |
c09595f6 | 453 | /* |
c8cba857 | 454 | * the part of load.weight contributed by tasks |
c09595f6 | 455 | */ |
c8cba857 | 456 | unsigned long task_weight; |
c09595f6 | 457 | |
c8cba857 PZ |
458 | /* |
459 | * h_load = weight * f(tg) | |
460 | * | |
461 | * Where f(tg) is the recursive weight fraction assigned to | |
462 | * this group. | |
463 | */ | |
464 | unsigned long h_load; | |
c09595f6 | 465 | |
c8cba857 PZ |
466 | /* |
467 | * this cpu's part of tg->shares | |
468 | */ | |
469 | unsigned long shares; | |
f1d239f7 PZ |
470 | |
471 | /* | |
472 | * load.weight at the time we set shares | |
473 | */ | |
474 | unsigned long rq_weight; | |
c09595f6 | 475 | #endif |
6aa645ea IM |
476 | #endif |
477 | }; | |
1da177e4 | 478 | |
6aa645ea IM |
479 | /* Real-Time classes' related field in a runqueue: */ |
480 | struct rt_rq { | |
481 | struct rt_prio_array active; | |
63489e45 | 482 | unsigned long rt_nr_running; |
052f1dc7 | 483 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
484 | struct { |
485 | int curr; /* highest queued rt task prio */ | |
398a153b | 486 | #ifdef CONFIG_SMP |
e864c499 | 487 | int next; /* next highest */ |
398a153b | 488 | #endif |
e864c499 | 489 | } highest_prio; |
6f505b16 | 490 | #endif |
fa85ae24 | 491 | #ifdef CONFIG_SMP |
73fe6aae | 492 | unsigned long rt_nr_migratory; |
a22d7fc1 | 493 | int overloaded; |
917b627d | 494 | struct plist_head pushable_tasks; |
fa85ae24 | 495 | #endif |
6f505b16 | 496 | int rt_throttled; |
fa85ae24 | 497 | u64 rt_time; |
ac086bc2 | 498 | u64 rt_runtime; |
ea736ed5 | 499 | /* Nests inside the rq lock: */ |
ac086bc2 | 500 | spinlock_t rt_runtime_lock; |
6f505b16 | 501 | |
052f1dc7 | 502 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
503 | unsigned long rt_nr_boosted; |
504 | ||
6f505b16 PZ |
505 | struct rq *rq; |
506 | struct list_head leaf_rt_rq_list; | |
507 | struct task_group *tg; | |
508 | struct sched_rt_entity *rt_se; | |
509 | #endif | |
6aa645ea IM |
510 | }; |
511 | ||
57d885fe GH |
512 | #ifdef CONFIG_SMP |
513 | ||
514 | /* | |
515 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
516 | * variables. Each exclusive cpuset essentially defines an island domain by |
517 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
518 | * exclusive cpuset is created, we also create and attach a new root-domain |
519 | * object. | |
520 | * | |
57d885fe GH |
521 | */ |
522 | struct root_domain { | |
523 | atomic_t refcount; | |
c6c4927b RR |
524 | cpumask_var_t span; |
525 | cpumask_var_t online; | |
637f5085 | 526 | |
0eab9146 | 527 | /* |
637f5085 GH |
528 | * The "RT overload" flag: it gets set if a CPU has more than |
529 | * one runnable RT task. | |
530 | */ | |
c6c4927b | 531 | cpumask_var_t rto_mask; |
0eab9146 | 532 | atomic_t rto_count; |
6e0534f2 GH |
533 | #ifdef CONFIG_SMP |
534 | struct cpupri cpupri; | |
535 | #endif | |
7a09b1a2 VS |
536 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
537 | /* | |
538 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
539 | * used when most cpus are idle in the system indicating overall very | |
540 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
541 | */ | |
542 | unsigned int sched_mc_preferred_wakeup_cpu; | |
543 | #endif | |
57d885fe GH |
544 | }; |
545 | ||
dc938520 GH |
546 | /* |
547 | * By default the system creates a single root-domain with all cpus as | |
548 | * members (mimicking the global state we have today). | |
549 | */ | |
57d885fe GH |
550 | static struct root_domain def_root_domain; |
551 | ||
552 | #endif | |
553 | ||
1da177e4 LT |
554 | /* |
555 | * This is the main, per-CPU runqueue data structure. | |
556 | * | |
557 | * Locking rule: those places that want to lock multiple runqueues | |
558 | * (such as the load balancing or the thread migration code), lock | |
559 | * acquire operations must be ordered by ascending &runqueue. | |
560 | */ | |
70b97a7f | 561 | struct rq { |
d8016491 IM |
562 | /* runqueue lock: */ |
563 | spinlock_t lock; | |
1da177e4 LT |
564 | |
565 | /* | |
566 | * nr_running and cpu_load should be in the same cacheline because | |
567 | * remote CPUs use both these fields when doing load calculation. | |
568 | */ | |
569 | unsigned long nr_running; | |
6aa645ea IM |
570 | #define CPU_LOAD_IDX_MAX 5 |
571 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 572 | #ifdef CONFIG_NO_HZ |
15934a37 | 573 | unsigned long last_tick_seen; |
46cb4b7c SS |
574 | unsigned char in_nohz_recently; |
575 | #endif | |
d8016491 IM |
576 | /* capture load from *all* tasks on this cpu: */ |
577 | struct load_weight load; | |
6aa645ea IM |
578 | unsigned long nr_load_updates; |
579 | u64 nr_switches; | |
580 | ||
581 | struct cfs_rq cfs; | |
6f505b16 | 582 | struct rt_rq rt; |
6f505b16 | 583 | |
6aa645ea | 584 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
585 | /* list of leaf cfs_rq on this cpu: */ |
586 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
587 | #endif |
588 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 589 | struct list_head leaf_rt_rq_list; |
1da177e4 | 590 | #endif |
1da177e4 LT |
591 | |
592 | /* | |
593 | * This is part of a global counter where only the total sum | |
594 | * over all CPUs matters. A task can increase this counter on | |
595 | * one CPU and if it got migrated afterwards it may decrease | |
596 | * it on another CPU. Always updated under the runqueue lock: | |
597 | */ | |
598 | unsigned long nr_uninterruptible; | |
599 | ||
36c8b586 | 600 | struct task_struct *curr, *idle; |
c9819f45 | 601 | unsigned long next_balance; |
1da177e4 | 602 | struct mm_struct *prev_mm; |
6aa645ea | 603 | |
3e51f33f | 604 | u64 clock; |
6aa645ea | 605 | |
1da177e4 LT |
606 | atomic_t nr_iowait; |
607 | ||
608 | #ifdef CONFIG_SMP | |
0eab9146 | 609 | struct root_domain *rd; |
1da177e4 LT |
610 | struct sched_domain *sd; |
611 | ||
a0a522ce | 612 | unsigned char idle_at_tick; |
1da177e4 LT |
613 | /* For active balancing */ |
614 | int active_balance; | |
615 | int push_cpu; | |
d8016491 IM |
616 | /* cpu of this runqueue: */ |
617 | int cpu; | |
1f11eb6a | 618 | int online; |
1da177e4 | 619 | |
a8a51d5e | 620 | unsigned long avg_load_per_task; |
1da177e4 | 621 | |
36c8b586 | 622 | struct task_struct *migration_thread; |
1da177e4 LT |
623 | struct list_head migration_queue; |
624 | #endif | |
625 | ||
8f4d37ec | 626 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
627 | #ifdef CONFIG_SMP |
628 | int hrtick_csd_pending; | |
629 | struct call_single_data hrtick_csd; | |
630 | #endif | |
8f4d37ec PZ |
631 | struct hrtimer hrtick_timer; |
632 | #endif | |
633 | ||
1da177e4 LT |
634 | #ifdef CONFIG_SCHEDSTATS |
635 | /* latency stats */ | |
636 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
637 | unsigned long long rq_cpu_time; |
638 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
639 | |
640 | /* sys_sched_yield() stats */ | |
480b9434 | 641 | unsigned int yld_count; |
1da177e4 LT |
642 | |
643 | /* schedule() stats */ | |
480b9434 KC |
644 | unsigned int sched_switch; |
645 | unsigned int sched_count; | |
646 | unsigned int sched_goidle; | |
1da177e4 LT |
647 | |
648 | /* try_to_wake_up() stats */ | |
480b9434 KC |
649 | unsigned int ttwu_count; |
650 | unsigned int ttwu_local; | |
b8efb561 IM |
651 | |
652 | /* BKL stats */ | |
480b9434 | 653 | unsigned int bkl_count; |
1da177e4 LT |
654 | #endif |
655 | }; | |
656 | ||
f34e3b61 | 657 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 658 | |
15afe09b | 659 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 660 | { |
15afe09b | 661 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
662 | } |
663 | ||
0a2966b4 CL |
664 | static inline int cpu_of(struct rq *rq) |
665 | { | |
666 | #ifdef CONFIG_SMP | |
667 | return rq->cpu; | |
668 | #else | |
669 | return 0; | |
670 | #endif | |
671 | } | |
672 | ||
674311d5 NP |
673 | /* |
674 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 675 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
676 | * |
677 | * The domain tree of any CPU may only be accessed from within | |
678 | * preempt-disabled sections. | |
679 | */ | |
48f24c4d IM |
680 | #define for_each_domain(cpu, __sd) \ |
681 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
682 | |
683 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
684 | #define this_rq() (&__get_cpu_var(runqueues)) | |
685 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
686 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
687 | ||
3e51f33f PZ |
688 | static inline void update_rq_clock(struct rq *rq) |
689 | { | |
690 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
691 | } | |
692 | ||
bf5c91ba IM |
693 | /* |
694 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
695 | */ | |
696 | #ifdef CONFIG_SCHED_DEBUG | |
697 | # define const_debug __read_mostly | |
698 | #else | |
699 | # define const_debug static const | |
700 | #endif | |
701 | ||
017730c1 IM |
702 | /** |
703 | * runqueue_is_locked | |
704 | * | |
705 | * Returns true if the current cpu runqueue is locked. | |
706 | * This interface allows printk to be called with the runqueue lock | |
707 | * held and know whether or not it is OK to wake up the klogd. | |
708 | */ | |
709 | int runqueue_is_locked(void) | |
710 | { | |
711 | int cpu = get_cpu(); | |
712 | struct rq *rq = cpu_rq(cpu); | |
713 | int ret; | |
714 | ||
715 | ret = spin_is_locked(&rq->lock); | |
716 | put_cpu(); | |
717 | return ret; | |
718 | } | |
719 | ||
bf5c91ba IM |
720 | /* |
721 | * Debugging: various feature bits | |
722 | */ | |
f00b45c1 PZ |
723 | |
724 | #define SCHED_FEAT(name, enabled) \ | |
725 | __SCHED_FEAT_##name , | |
726 | ||
bf5c91ba | 727 | enum { |
f00b45c1 | 728 | #include "sched_features.h" |
bf5c91ba IM |
729 | }; |
730 | ||
f00b45c1 PZ |
731 | #undef SCHED_FEAT |
732 | ||
733 | #define SCHED_FEAT(name, enabled) \ | |
734 | (1UL << __SCHED_FEAT_##name) * enabled | | |
735 | ||
bf5c91ba | 736 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
737 | #include "sched_features.h" |
738 | 0; | |
739 | ||
740 | #undef SCHED_FEAT | |
741 | ||
742 | #ifdef CONFIG_SCHED_DEBUG | |
743 | #define SCHED_FEAT(name, enabled) \ | |
744 | #name , | |
745 | ||
983ed7a6 | 746 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
747 | #include "sched_features.h" |
748 | NULL | |
749 | }; | |
750 | ||
751 | #undef SCHED_FEAT | |
752 | ||
34f3a814 | 753 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 754 | { |
f00b45c1 PZ |
755 | int i; |
756 | ||
757 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
758 | if (!(sysctl_sched_features & (1UL << i))) |
759 | seq_puts(m, "NO_"); | |
760 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 761 | } |
34f3a814 | 762 | seq_puts(m, "\n"); |
f00b45c1 | 763 | |
34f3a814 | 764 | return 0; |
f00b45c1 PZ |
765 | } |
766 | ||
767 | static ssize_t | |
768 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
769 | size_t cnt, loff_t *ppos) | |
770 | { | |
771 | char buf[64]; | |
772 | char *cmp = buf; | |
773 | int neg = 0; | |
774 | int i; | |
775 | ||
776 | if (cnt > 63) | |
777 | cnt = 63; | |
778 | ||
779 | if (copy_from_user(&buf, ubuf, cnt)) | |
780 | return -EFAULT; | |
781 | ||
782 | buf[cnt] = 0; | |
783 | ||
c24b7c52 | 784 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
785 | neg = 1; |
786 | cmp += 3; | |
787 | } | |
788 | ||
789 | for (i = 0; sched_feat_names[i]; i++) { | |
790 | int len = strlen(sched_feat_names[i]); | |
791 | ||
792 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
793 | if (neg) | |
794 | sysctl_sched_features &= ~(1UL << i); | |
795 | else | |
796 | sysctl_sched_features |= (1UL << i); | |
797 | break; | |
798 | } | |
799 | } | |
800 | ||
801 | if (!sched_feat_names[i]) | |
802 | return -EINVAL; | |
803 | ||
804 | filp->f_pos += cnt; | |
805 | ||
806 | return cnt; | |
807 | } | |
808 | ||
34f3a814 LZ |
809 | static int sched_feat_open(struct inode *inode, struct file *filp) |
810 | { | |
811 | return single_open(filp, sched_feat_show, NULL); | |
812 | } | |
813 | ||
f00b45c1 | 814 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
815 | .open = sched_feat_open, |
816 | .write = sched_feat_write, | |
817 | .read = seq_read, | |
818 | .llseek = seq_lseek, | |
819 | .release = single_release, | |
f00b45c1 PZ |
820 | }; |
821 | ||
822 | static __init int sched_init_debug(void) | |
823 | { | |
f00b45c1 PZ |
824 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
825 | &sched_feat_fops); | |
826 | ||
827 | return 0; | |
828 | } | |
829 | late_initcall(sched_init_debug); | |
830 | ||
831 | #endif | |
832 | ||
833 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 834 | |
b82d9fdd PZ |
835 | /* |
836 | * Number of tasks to iterate in a single balance run. | |
837 | * Limited because this is done with IRQs disabled. | |
838 | */ | |
839 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
840 | ||
2398f2c6 PZ |
841 | /* |
842 | * ratelimit for updating the group shares. | |
55cd5340 | 843 | * default: 0.25ms |
2398f2c6 | 844 | */ |
55cd5340 | 845 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 846 | |
ffda12a1 PZ |
847 | /* |
848 | * Inject some fuzzyness into changing the per-cpu group shares | |
849 | * this avoids remote rq-locks at the expense of fairness. | |
850 | * default: 4 | |
851 | */ | |
852 | unsigned int sysctl_sched_shares_thresh = 4; | |
853 | ||
fa85ae24 | 854 | /* |
9f0c1e56 | 855 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
856 | * default: 1s |
857 | */ | |
9f0c1e56 | 858 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 859 | |
6892b75e IM |
860 | static __read_mostly int scheduler_running; |
861 | ||
9f0c1e56 PZ |
862 | /* |
863 | * part of the period that we allow rt tasks to run in us. | |
864 | * default: 0.95s | |
865 | */ | |
866 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 867 | |
d0b27fa7 PZ |
868 | static inline u64 global_rt_period(void) |
869 | { | |
870 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
871 | } | |
872 | ||
873 | static inline u64 global_rt_runtime(void) | |
874 | { | |
e26873bb | 875 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
876 | return RUNTIME_INF; |
877 | ||
878 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
879 | } | |
fa85ae24 | 880 | |
1da177e4 | 881 | #ifndef prepare_arch_switch |
4866cde0 NP |
882 | # define prepare_arch_switch(next) do { } while (0) |
883 | #endif | |
884 | #ifndef finish_arch_switch | |
885 | # define finish_arch_switch(prev) do { } while (0) | |
886 | #endif | |
887 | ||
051a1d1a DA |
888 | static inline int task_current(struct rq *rq, struct task_struct *p) |
889 | { | |
890 | return rq->curr == p; | |
891 | } | |
892 | ||
4866cde0 | 893 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 894 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 895 | { |
051a1d1a | 896 | return task_current(rq, p); |
4866cde0 NP |
897 | } |
898 | ||
70b97a7f | 899 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
900 | { |
901 | } | |
902 | ||
70b97a7f | 903 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 904 | { |
da04c035 IM |
905 | #ifdef CONFIG_DEBUG_SPINLOCK |
906 | /* this is a valid case when another task releases the spinlock */ | |
907 | rq->lock.owner = current; | |
908 | #endif | |
8a25d5de IM |
909 | /* |
910 | * If we are tracking spinlock dependencies then we have to | |
911 | * fix up the runqueue lock - which gets 'carried over' from | |
912 | * prev into current: | |
913 | */ | |
914 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
915 | ||
4866cde0 NP |
916 | spin_unlock_irq(&rq->lock); |
917 | } | |
918 | ||
919 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 920 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
921 | { |
922 | #ifdef CONFIG_SMP | |
923 | return p->oncpu; | |
924 | #else | |
051a1d1a | 925 | return task_current(rq, p); |
4866cde0 NP |
926 | #endif |
927 | } | |
928 | ||
70b97a7f | 929 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
930 | { |
931 | #ifdef CONFIG_SMP | |
932 | /* | |
933 | * We can optimise this out completely for !SMP, because the | |
934 | * SMP rebalancing from interrupt is the only thing that cares | |
935 | * here. | |
936 | */ | |
937 | next->oncpu = 1; | |
938 | #endif | |
939 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
940 | spin_unlock_irq(&rq->lock); | |
941 | #else | |
942 | spin_unlock(&rq->lock); | |
943 | #endif | |
944 | } | |
945 | ||
70b97a7f | 946 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
947 | { |
948 | #ifdef CONFIG_SMP | |
949 | /* | |
950 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
951 | * We must ensure this doesn't happen until the switch is completely | |
952 | * finished. | |
953 | */ | |
954 | smp_wmb(); | |
955 | prev->oncpu = 0; | |
956 | #endif | |
957 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
958 | local_irq_enable(); | |
1da177e4 | 959 | #endif |
4866cde0 NP |
960 | } |
961 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 962 | |
b29739f9 IM |
963 | /* |
964 | * __task_rq_lock - lock the runqueue a given task resides on. | |
965 | * Must be called interrupts disabled. | |
966 | */ | |
70b97a7f | 967 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
968 | __acquires(rq->lock) |
969 | { | |
3a5c359a AK |
970 | for (;;) { |
971 | struct rq *rq = task_rq(p); | |
972 | spin_lock(&rq->lock); | |
973 | if (likely(rq == task_rq(p))) | |
974 | return rq; | |
b29739f9 | 975 | spin_unlock(&rq->lock); |
b29739f9 | 976 | } |
b29739f9 IM |
977 | } |
978 | ||
1da177e4 LT |
979 | /* |
980 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 981 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
982 | * explicitly disabling preemption. |
983 | */ | |
70b97a7f | 984 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
985 | __acquires(rq->lock) |
986 | { | |
70b97a7f | 987 | struct rq *rq; |
1da177e4 | 988 | |
3a5c359a AK |
989 | for (;;) { |
990 | local_irq_save(*flags); | |
991 | rq = task_rq(p); | |
992 | spin_lock(&rq->lock); | |
993 | if (likely(rq == task_rq(p))) | |
994 | return rq; | |
1da177e4 | 995 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 996 | } |
1da177e4 LT |
997 | } |
998 | ||
ad474cac ON |
999 | void task_rq_unlock_wait(struct task_struct *p) |
1000 | { | |
1001 | struct rq *rq = task_rq(p); | |
1002 | ||
1003 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1004 | spin_unlock_wait(&rq->lock); | |
1005 | } | |
1006 | ||
a9957449 | 1007 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1008 | __releases(rq->lock) |
1009 | { | |
1010 | spin_unlock(&rq->lock); | |
1011 | } | |
1012 | ||
70b97a7f | 1013 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1014 | __releases(rq->lock) |
1015 | { | |
1016 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1017 | } | |
1018 | ||
1da177e4 | 1019 | /* |
cc2a73b5 | 1020 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1021 | */ |
a9957449 | 1022 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1023 | __acquires(rq->lock) |
1024 | { | |
70b97a7f | 1025 | struct rq *rq; |
1da177e4 LT |
1026 | |
1027 | local_irq_disable(); | |
1028 | rq = this_rq(); | |
1029 | spin_lock(&rq->lock); | |
1030 | ||
1031 | return rq; | |
1032 | } | |
1033 | ||
8f4d37ec PZ |
1034 | #ifdef CONFIG_SCHED_HRTICK |
1035 | /* | |
1036 | * Use HR-timers to deliver accurate preemption points. | |
1037 | * | |
1038 | * Its all a bit involved since we cannot program an hrt while holding the | |
1039 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1040 | * reschedule event. | |
1041 | * | |
1042 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1043 | * rq->lock. | |
1044 | */ | |
8f4d37ec PZ |
1045 | |
1046 | /* | |
1047 | * Use hrtick when: | |
1048 | * - enabled by features | |
1049 | * - hrtimer is actually high res | |
1050 | */ | |
1051 | static inline int hrtick_enabled(struct rq *rq) | |
1052 | { | |
1053 | if (!sched_feat(HRTICK)) | |
1054 | return 0; | |
ba42059f | 1055 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1056 | return 0; |
8f4d37ec PZ |
1057 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1058 | } | |
1059 | ||
8f4d37ec PZ |
1060 | static void hrtick_clear(struct rq *rq) |
1061 | { | |
1062 | if (hrtimer_active(&rq->hrtick_timer)) | |
1063 | hrtimer_cancel(&rq->hrtick_timer); | |
1064 | } | |
1065 | ||
8f4d37ec PZ |
1066 | /* |
1067 | * High-resolution timer tick. | |
1068 | * Runs from hardirq context with interrupts disabled. | |
1069 | */ | |
1070 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1071 | { | |
1072 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1073 | ||
1074 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1075 | ||
1076 | spin_lock(&rq->lock); | |
3e51f33f | 1077 | update_rq_clock(rq); |
8f4d37ec PZ |
1078 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1079 | spin_unlock(&rq->lock); | |
1080 | ||
1081 | return HRTIMER_NORESTART; | |
1082 | } | |
1083 | ||
95e904c7 | 1084 | #ifdef CONFIG_SMP |
31656519 PZ |
1085 | /* |
1086 | * called from hardirq (IPI) context | |
1087 | */ | |
1088 | static void __hrtick_start(void *arg) | |
b328ca18 | 1089 | { |
31656519 | 1090 | struct rq *rq = arg; |
b328ca18 | 1091 | |
31656519 PZ |
1092 | spin_lock(&rq->lock); |
1093 | hrtimer_restart(&rq->hrtick_timer); | |
1094 | rq->hrtick_csd_pending = 0; | |
1095 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1096 | } |
1097 | ||
31656519 PZ |
1098 | /* |
1099 | * Called to set the hrtick timer state. | |
1100 | * | |
1101 | * called with rq->lock held and irqs disabled | |
1102 | */ | |
1103 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1104 | { |
31656519 PZ |
1105 | struct hrtimer *timer = &rq->hrtick_timer; |
1106 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1107 | |
cc584b21 | 1108 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1109 | |
1110 | if (rq == this_rq()) { | |
1111 | hrtimer_restart(timer); | |
1112 | } else if (!rq->hrtick_csd_pending) { | |
1113 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1114 | rq->hrtick_csd_pending = 1; | |
1115 | } | |
b328ca18 PZ |
1116 | } |
1117 | ||
1118 | static int | |
1119 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1120 | { | |
1121 | int cpu = (int)(long)hcpu; | |
1122 | ||
1123 | switch (action) { | |
1124 | case CPU_UP_CANCELED: | |
1125 | case CPU_UP_CANCELED_FROZEN: | |
1126 | case CPU_DOWN_PREPARE: | |
1127 | case CPU_DOWN_PREPARE_FROZEN: | |
1128 | case CPU_DEAD: | |
1129 | case CPU_DEAD_FROZEN: | |
31656519 | 1130 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1131 | return NOTIFY_OK; |
1132 | } | |
1133 | ||
1134 | return NOTIFY_DONE; | |
1135 | } | |
1136 | ||
fa748203 | 1137 | static __init void init_hrtick(void) |
b328ca18 PZ |
1138 | { |
1139 | hotcpu_notifier(hotplug_hrtick, 0); | |
1140 | } | |
31656519 PZ |
1141 | #else |
1142 | /* | |
1143 | * Called to set the hrtick timer state. | |
1144 | * | |
1145 | * called with rq->lock held and irqs disabled | |
1146 | */ | |
1147 | static void hrtick_start(struct rq *rq, u64 delay) | |
1148 | { | |
1149 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1150 | } | |
b328ca18 | 1151 | |
006c75f1 | 1152 | static inline void init_hrtick(void) |
8f4d37ec | 1153 | { |
8f4d37ec | 1154 | } |
31656519 | 1155 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1156 | |
31656519 | 1157 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1158 | { |
31656519 PZ |
1159 | #ifdef CONFIG_SMP |
1160 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1161 | |
31656519 PZ |
1162 | rq->hrtick_csd.flags = 0; |
1163 | rq->hrtick_csd.func = __hrtick_start; | |
1164 | rq->hrtick_csd.info = rq; | |
1165 | #endif | |
8f4d37ec | 1166 | |
31656519 PZ |
1167 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1168 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1169 | } |
006c75f1 | 1170 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1171 | static inline void hrtick_clear(struct rq *rq) |
1172 | { | |
1173 | } | |
1174 | ||
8f4d37ec PZ |
1175 | static inline void init_rq_hrtick(struct rq *rq) |
1176 | { | |
1177 | } | |
1178 | ||
b328ca18 PZ |
1179 | static inline void init_hrtick(void) |
1180 | { | |
1181 | } | |
006c75f1 | 1182 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1183 | |
c24d20db IM |
1184 | /* |
1185 | * resched_task - mark a task 'to be rescheduled now'. | |
1186 | * | |
1187 | * On UP this means the setting of the need_resched flag, on SMP it | |
1188 | * might also involve a cross-CPU call to trigger the scheduler on | |
1189 | * the target CPU. | |
1190 | */ | |
1191 | #ifdef CONFIG_SMP | |
1192 | ||
1193 | #ifndef tsk_is_polling | |
1194 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1195 | #endif | |
1196 | ||
31656519 | 1197 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1198 | { |
1199 | int cpu; | |
1200 | ||
1201 | assert_spin_locked(&task_rq(p)->lock); | |
1202 | ||
5ed0cec0 | 1203 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1204 | return; |
1205 | ||
5ed0cec0 | 1206 | set_tsk_need_resched(p); |
c24d20db IM |
1207 | |
1208 | cpu = task_cpu(p); | |
1209 | if (cpu == smp_processor_id()) | |
1210 | return; | |
1211 | ||
1212 | /* NEED_RESCHED must be visible before we test polling */ | |
1213 | smp_mb(); | |
1214 | if (!tsk_is_polling(p)) | |
1215 | smp_send_reschedule(cpu); | |
1216 | } | |
1217 | ||
1218 | static void resched_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | unsigned long flags; | |
1222 | ||
1223 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1224 | return; | |
1225 | resched_task(cpu_curr(cpu)); | |
1226 | spin_unlock_irqrestore(&rq->lock, flags); | |
1227 | } | |
06d8308c TG |
1228 | |
1229 | #ifdef CONFIG_NO_HZ | |
1230 | /* | |
1231 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1232 | * idle CPU then this timer might expire before the next timer event | |
1233 | * which is scheduled to wake up that CPU. In case of a completely | |
1234 | * idle system the next event might even be infinite time into the | |
1235 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1236 | * leaves the inner idle loop so the newly added timer is taken into | |
1237 | * account when the CPU goes back to idle and evaluates the timer | |
1238 | * wheel for the next timer event. | |
1239 | */ | |
1240 | void wake_up_idle_cpu(int cpu) | |
1241 | { | |
1242 | struct rq *rq = cpu_rq(cpu); | |
1243 | ||
1244 | if (cpu == smp_processor_id()) | |
1245 | return; | |
1246 | ||
1247 | /* | |
1248 | * This is safe, as this function is called with the timer | |
1249 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1250 | * to idle and has not yet set rq->curr to idle then it will | |
1251 | * be serialized on the timer wheel base lock and take the new | |
1252 | * timer into account automatically. | |
1253 | */ | |
1254 | if (rq->curr != rq->idle) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1259 | * lockless. The worst case is that the other CPU runs the | |
1260 | * idle task through an additional NOOP schedule() | |
1261 | */ | |
5ed0cec0 | 1262 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1263 | |
1264 | /* NEED_RESCHED must be visible before we test polling */ | |
1265 | smp_mb(); | |
1266 | if (!tsk_is_polling(rq->idle)) | |
1267 | smp_send_reschedule(cpu); | |
1268 | } | |
6d6bc0ad | 1269 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1270 | |
6d6bc0ad | 1271 | #else /* !CONFIG_SMP */ |
31656519 | 1272 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1273 | { |
1274 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1275 | set_tsk_need_resched(p); |
c24d20db | 1276 | } |
6d6bc0ad | 1277 | #endif /* CONFIG_SMP */ |
c24d20db | 1278 | |
45bf76df IM |
1279 | #if BITS_PER_LONG == 32 |
1280 | # define WMULT_CONST (~0UL) | |
1281 | #else | |
1282 | # define WMULT_CONST (1UL << 32) | |
1283 | #endif | |
1284 | ||
1285 | #define WMULT_SHIFT 32 | |
1286 | ||
194081eb IM |
1287 | /* |
1288 | * Shift right and round: | |
1289 | */ | |
cf2ab469 | 1290 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1291 | |
a7be37ac PZ |
1292 | /* |
1293 | * delta *= weight / lw | |
1294 | */ | |
cb1c4fc9 | 1295 | static unsigned long |
45bf76df IM |
1296 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1297 | struct load_weight *lw) | |
1298 | { | |
1299 | u64 tmp; | |
1300 | ||
7a232e03 LJ |
1301 | if (!lw->inv_weight) { |
1302 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1303 | lw->inv_weight = 1; | |
1304 | else | |
1305 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1306 | / (lw->weight+1); | |
1307 | } | |
45bf76df IM |
1308 | |
1309 | tmp = (u64)delta_exec * weight; | |
1310 | /* | |
1311 | * Check whether we'd overflow the 64-bit multiplication: | |
1312 | */ | |
194081eb | 1313 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1314 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1315 | WMULT_SHIFT/2); |
1316 | else | |
cf2ab469 | 1317 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1318 | |
ecf691da | 1319 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1320 | } |
1321 | ||
1091985b | 1322 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1323 | { |
1324 | lw->weight += inc; | |
e89996ae | 1325 | lw->inv_weight = 0; |
45bf76df IM |
1326 | } |
1327 | ||
1091985b | 1328 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1329 | { |
1330 | lw->weight -= dec; | |
e89996ae | 1331 | lw->inv_weight = 0; |
45bf76df IM |
1332 | } |
1333 | ||
2dd73a4f PW |
1334 | /* |
1335 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1336 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1337 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1338 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1339 | * scaled version of the new time slice allocation that they receive on time |
1340 | * slice expiry etc. | |
1341 | */ | |
1342 | ||
cce7ade8 PZ |
1343 | #define WEIGHT_IDLEPRIO 3 |
1344 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1345 | |
1346 | /* | |
1347 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1348 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1349 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1350 | * that remained on nice 0. | |
1351 | * | |
1352 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1353 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1354 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1355 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1356 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1357 | */ |
1358 | static const int prio_to_weight[40] = { | |
254753dc IM |
1359 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1360 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1361 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1362 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1363 | /* 0 */ 1024, 820, 655, 526, 423, | |
1364 | /* 5 */ 335, 272, 215, 172, 137, | |
1365 | /* 10 */ 110, 87, 70, 56, 45, | |
1366 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1367 | }; |
1368 | ||
5714d2de IM |
1369 | /* |
1370 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1371 | * | |
1372 | * In cases where the weight does not change often, we can use the | |
1373 | * precalculated inverse to speed up arithmetics by turning divisions | |
1374 | * into multiplications: | |
1375 | */ | |
dd41f596 | 1376 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1377 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1378 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1379 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1380 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1381 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1382 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1383 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1384 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1385 | }; |
2dd73a4f | 1386 | |
dd41f596 IM |
1387 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1388 | ||
1389 | /* | |
1390 | * runqueue iterator, to support SMP load-balancing between different | |
1391 | * scheduling classes, without having to expose their internal data | |
1392 | * structures to the load-balancing proper: | |
1393 | */ | |
1394 | struct rq_iterator { | |
1395 | void *arg; | |
1396 | struct task_struct *(*start)(void *); | |
1397 | struct task_struct *(*next)(void *); | |
1398 | }; | |
1399 | ||
e1d1484f PW |
1400 | #ifdef CONFIG_SMP |
1401 | static unsigned long | |
1402 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1403 | unsigned long max_load_move, struct sched_domain *sd, | |
1404 | enum cpu_idle_type idle, int *all_pinned, | |
1405 | int *this_best_prio, struct rq_iterator *iterator); | |
1406 | ||
1407 | static int | |
1408 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1409 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1410 | struct rq_iterator *iterator); | |
e1d1484f | 1411 | #endif |
dd41f596 | 1412 | |
d842de87 SV |
1413 | #ifdef CONFIG_CGROUP_CPUACCT |
1414 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1415 | #else | |
1416 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1417 | #endif | |
1418 | ||
18d95a28 PZ |
1419 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1420 | { | |
1421 | update_load_add(&rq->load, load); | |
1422 | } | |
1423 | ||
1424 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1425 | { | |
1426 | update_load_sub(&rq->load, load); | |
1427 | } | |
1428 | ||
7940ca36 | 1429 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1430 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1431 | |
1432 | /* | |
1433 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1434 | * leaving it for the final time. | |
1435 | */ | |
eb755805 | 1436 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1437 | { |
1438 | struct task_group *parent, *child; | |
eb755805 | 1439 | int ret; |
c09595f6 PZ |
1440 | |
1441 | rcu_read_lock(); | |
1442 | parent = &root_task_group; | |
1443 | down: | |
eb755805 PZ |
1444 | ret = (*down)(parent, data); |
1445 | if (ret) | |
1446 | goto out_unlock; | |
c09595f6 PZ |
1447 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1448 | parent = child; | |
1449 | goto down; | |
1450 | ||
1451 | up: | |
1452 | continue; | |
1453 | } | |
eb755805 PZ |
1454 | ret = (*up)(parent, data); |
1455 | if (ret) | |
1456 | goto out_unlock; | |
c09595f6 PZ |
1457 | |
1458 | child = parent; | |
1459 | parent = parent->parent; | |
1460 | if (parent) | |
1461 | goto up; | |
eb755805 | 1462 | out_unlock: |
c09595f6 | 1463 | rcu_read_unlock(); |
eb755805 PZ |
1464 | |
1465 | return ret; | |
c09595f6 PZ |
1466 | } |
1467 | ||
eb755805 PZ |
1468 | static int tg_nop(struct task_group *tg, void *data) |
1469 | { | |
1470 | return 0; | |
c09595f6 | 1471 | } |
eb755805 PZ |
1472 | #endif |
1473 | ||
1474 | #ifdef CONFIG_SMP | |
1475 | static unsigned long source_load(int cpu, int type); | |
1476 | static unsigned long target_load(int cpu, int type); | |
1477 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1478 | ||
1479 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1480 | { | |
1481 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1482 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1483 | |
4cd42620 SR |
1484 | if (nr_running) |
1485 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1486 | else |
1487 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1488 | |
1489 | return rq->avg_load_per_task; | |
1490 | } | |
1491 | ||
1492 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1493 | |
c09595f6 PZ |
1494 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1495 | ||
1496 | /* | |
1497 | * Calculate and set the cpu's group shares. | |
1498 | */ | |
1499 | static void | |
ffda12a1 PZ |
1500 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1501 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1502 | { |
c09595f6 PZ |
1503 | unsigned long shares; |
1504 | unsigned long rq_weight; | |
1505 | ||
c8cba857 | 1506 | if (!tg->se[cpu]) |
c09595f6 PZ |
1507 | return; |
1508 | ||
ec4e0e2f | 1509 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1510 | |
c09595f6 PZ |
1511 | /* |
1512 | * \Sum shares * rq_weight | |
1513 | * shares = ----------------------- | |
1514 | * \Sum rq_weight | |
1515 | * | |
1516 | */ | |
ec4e0e2f | 1517 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1518 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1519 | |
ffda12a1 PZ |
1520 | if (abs(shares - tg->se[cpu]->load.weight) > |
1521 | sysctl_sched_shares_thresh) { | |
1522 | struct rq *rq = cpu_rq(cpu); | |
1523 | unsigned long flags; | |
c09595f6 | 1524 | |
ffda12a1 | 1525 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1526 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1527 | |
ffda12a1 PZ |
1528 | __set_se_shares(tg->se[cpu], shares); |
1529 | spin_unlock_irqrestore(&rq->lock, flags); | |
1530 | } | |
18d95a28 | 1531 | } |
c09595f6 PZ |
1532 | |
1533 | /* | |
c8cba857 PZ |
1534 | * Re-compute the task group their per cpu shares over the given domain. |
1535 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1536 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1537 | */ |
eb755805 | 1538 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1539 | { |
ec4e0e2f | 1540 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1541 | unsigned long shares = 0; |
eb755805 | 1542 | struct sched_domain *sd = data; |
c8cba857 | 1543 | int i; |
c09595f6 | 1544 | |
758b2cdc | 1545 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1546 | /* |
1547 | * If there are currently no tasks on the cpu pretend there | |
1548 | * is one of average load so that when a new task gets to | |
1549 | * run here it will not get delayed by group starvation. | |
1550 | */ | |
1551 | weight = tg->cfs_rq[i]->load.weight; | |
1552 | if (!weight) | |
1553 | weight = NICE_0_LOAD; | |
1554 | ||
1555 | tg->cfs_rq[i]->rq_weight = weight; | |
1556 | rq_weight += weight; | |
c8cba857 | 1557 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1558 | } |
c09595f6 | 1559 | |
c8cba857 PZ |
1560 | if ((!shares && rq_weight) || shares > tg->shares) |
1561 | shares = tg->shares; | |
1562 | ||
1563 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1564 | shares = tg->shares; | |
c09595f6 | 1565 | |
758b2cdc | 1566 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1567 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1568 | |
1569 | return 0; | |
c09595f6 PZ |
1570 | } |
1571 | ||
1572 | /* | |
c8cba857 PZ |
1573 | * Compute the cpu's hierarchical load factor for each task group. |
1574 | * This needs to be done in a top-down fashion because the load of a child | |
1575 | * group is a fraction of its parents load. | |
c09595f6 | 1576 | */ |
eb755805 | 1577 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1578 | { |
c8cba857 | 1579 | unsigned long load; |
eb755805 | 1580 | long cpu = (long)data; |
c09595f6 | 1581 | |
c8cba857 PZ |
1582 | if (!tg->parent) { |
1583 | load = cpu_rq(cpu)->load.weight; | |
1584 | } else { | |
1585 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1586 | load *= tg->cfs_rq[cpu]->shares; | |
1587 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1588 | } | |
c09595f6 | 1589 | |
c8cba857 | 1590 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1591 | |
eb755805 | 1592 | return 0; |
c09595f6 PZ |
1593 | } |
1594 | ||
c8cba857 | 1595 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1596 | { |
2398f2c6 PZ |
1597 | u64 now = cpu_clock(raw_smp_processor_id()); |
1598 | s64 elapsed = now - sd->last_update; | |
1599 | ||
1600 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1601 | sd->last_update = now; | |
eb755805 | 1602 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1603 | } |
4d8d595d PZ |
1604 | } |
1605 | ||
3e5459b4 PZ |
1606 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1607 | { | |
1608 | spin_unlock(&rq->lock); | |
1609 | update_shares(sd); | |
1610 | spin_lock(&rq->lock); | |
1611 | } | |
1612 | ||
eb755805 | 1613 | static void update_h_load(long cpu) |
c09595f6 | 1614 | { |
eb755805 | 1615 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1616 | } |
1617 | ||
c09595f6 PZ |
1618 | #else |
1619 | ||
c8cba857 | 1620 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1621 | { |
1622 | } | |
1623 | ||
3e5459b4 PZ |
1624 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1625 | { | |
1626 | } | |
1627 | ||
18d95a28 PZ |
1628 | #endif |
1629 | ||
8f45e2b5 GH |
1630 | #ifdef CONFIG_PREEMPT |
1631 | ||
70574a99 | 1632 | /* |
8f45e2b5 GH |
1633 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1634 | * way at the expense of forcing extra atomic operations in all | |
1635 | * invocations. This assures that the double_lock is acquired using the | |
1636 | * same underlying policy as the spinlock_t on this architecture, which | |
1637 | * reduces latency compared to the unfair variant below. However, it | |
1638 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1639 | */ |
8f45e2b5 GH |
1640 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1641 | __releases(this_rq->lock) | |
1642 | __acquires(busiest->lock) | |
1643 | __acquires(this_rq->lock) | |
1644 | { | |
1645 | spin_unlock(&this_rq->lock); | |
1646 | double_rq_lock(this_rq, busiest); | |
1647 | ||
1648 | return 1; | |
1649 | } | |
1650 | ||
1651 | #else | |
1652 | /* | |
1653 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1654 | * latency by eliminating extra atomic operations when the locks are | |
1655 | * already in proper order on entry. This favors lower cpu-ids and will | |
1656 | * grant the double lock to lower cpus over higher ids under contention, | |
1657 | * regardless of entry order into the function. | |
1658 | */ | |
1659 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1660 | __releases(this_rq->lock) |
1661 | __acquires(busiest->lock) | |
1662 | __acquires(this_rq->lock) | |
1663 | { | |
1664 | int ret = 0; | |
1665 | ||
70574a99 AD |
1666 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1667 | if (busiest < this_rq) { | |
1668 | spin_unlock(&this_rq->lock); | |
1669 | spin_lock(&busiest->lock); | |
1670 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1671 | ret = 1; | |
1672 | } else | |
1673 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1674 | } | |
1675 | return ret; | |
1676 | } | |
1677 | ||
8f45e2b5 GH |
1678 | #endif /* CONFIG_PREEMPT */ |
1679 | ||
1680 | /* | |
1681 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1682 | */ | |
1683 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1684 | { | |
1685 | if (unlikely(!irqs_disabled())) { | |
1686 | /* printk() doesn't work good under rq->lock */ | |
1687 | spin_unlock(&this_rq->lock); | |
1688 | BUG_ON(1); | |
1689 | } | |
1690 | ||
1691 | return _double_lock_balance(this_rq, busiest); | |
1692 | } | |
1693 | ||
70574a99 AD |
1694 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1695 | __releases(busiest->lock) | |
1696 | { | |
1697 | spin_unlock(&busiest->lock); | |
1698 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1699 | } | |
18d95a28 PZ |
1700 | #endif |
1701 | ||
30432094 | 1702 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1703 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1704 | { | |
30432094 | 1705 | #ifdef CONFIG_SMP |
34e83e85 IM |
1706 | cfs_rq->shares = shares; |
1707 | #endif | |
1708 | } | |
30432094 | 1709 | #endif |
e7693a36 | 1710 | |
dd41f596 | 1711 | #include "sched_stats.h" |
dd41f596 | 1712 | #include "sched_idletask.c" |
5522d5d5 IM |
1713 | #include "sched_fair.c" |
1714 | #include "sched_rt.c" | |
dd41f596 IM |
1715 | #ifdef CONFIG_SCHED_DEBUG |
1716 | # include "sched_debug.c" | |
1717 | #endif | |
1718 | ||
1719 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1720 | #define for_each_class(class) \ |
1721 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1722 | |
c09595f6 | 1723 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1724 | { |
1725 | rq->nr_running++; | |
9c217245 IM |
1726 | } |
1727 | ||
c09595f6 | 1728 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1729 | { |
1730 | rq->nr_running--; | |
9c217245 IM |
1731 | } |
1732 | ||
45bf76df IM |
1733 | static void set_load_weight(struct task_struct *p) |
1734 | { | |
1735 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1736 | p->se.load.weight = prio_to_weight[0] * 2; |
1737 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1738 | return; | |
1739 | } | |
45bf76df | 1740 | |
dd41f596 IM |
1741 | /* |
1742 | * SCHED_IDLE tasks get minimal weight: | |
1743 | */ | |
1744 | if (p->policy == SCHED_IDLE) { | |
1745 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1746 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1747 | return; | |
1748 | } | |
71f8bd46 | 1749 | |
dd41f596 IM |
1750 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1751 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1752 | } |
1753 | ||
2087a1ad GH |
1754 | static void update_avg(u64 *avg, u64 sample) |
1755 | { | |
1756 | s64 diff = sample - *avg; | |
1757 | *avg += diff >> 3; | |
1758 | } | |
1759 | ||
8159f87e | 1760 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1761 | { |
831451ac PZ |
1762 | if (wakeup) |
1763 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1764 | ||
dd41f596 | 1765 | sched_info_queued(p); |
fd390f6a | 1766 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1767 | p->se.on_rq = 1; |
71f8bd46 IM |
1768 | } |
1769 | ||
69be72c1 | 1770 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1771 | { |
831451ac PZ |
1772 | if (sleep) { |
1773 | if (p->se.last_wakeup) { | |
1774 | update_avg(&p->se.avg_overlap, | |
1775 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1776 | p->se.last_wakeup = 0; | |
1777 | } else { | |
1778 | update_avg(&p->se.avg_wakeup, | |
1779 | sysctl_sched_wakeup_granularity); | |
1780 | } | |
2087a1ad GH |
1781 | } |
1782 | ||
46ac22ba | 1783 | sched_info_dequeued(p); |
f02231e5 | 1784 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1785 | p->se.on_rq = 0; |
71f8bd46 IM |
1786 | } |
1787 | ||
14531189 | 1788 | /* |
dd41f596 | 1789 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1790 | */ |
14531189 IM |
1791 | static inline int __normal_prio(struct task_struct *p) |
1792 | { | |
dd41f596 | 1793 | return p->static_prio; |
14531189 IM |
1794 | } |
1795 | ||
b29739f9 IM |
1796 | /* |
1797 | * Calculate the expected normal priority: i.e. priority | |
1798 | * without taking RT-inheritance into account. Might be | |
1799 | * boosted by interactivity modifiers. Changes upon fork, | |
1800 | * setprio syscalls, and whenever the interactivity | |
1801 | * estimator recalculates. | |
1802 | */ | |
36c8b586 | 1803 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1804 | { |
1805 | int prio; | |
1806 | ||
e05606d3 | 1807 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1808 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1809 | else | |
1810 | prio = __normal_prio(p); | |
1811 | return prio; | |
1812 | } | |
1813 | ||
1814 | /* | |
1815 | * Calculate the current priority, i.e. the priority | |
1816 | * taken into account by the scheduler. This value might | |
1817 | * be boosted by RT tasks, or might be boosted by | |
1818 | * interactivity modifiers. Will be RT if the task got | |
1819 | * RT-boosted. If not then it returns p->normal_prio. | |
1820 | */ | |
36c8b586 | 1821 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1822 | { |
1823 | p->normal_prio = normal_prio(p); | |
1824 | /* | |
1825 | * If we are RT tasks or we were boosted to RT priority, | |
1826 | * keep the priority unchanged. Otherwise, update priority | |
1827 | * to the normal priority: | |
1828 | */ | |
1829 | if (!rt_prio(p->prio)) | |
1830 | return p->normal_prio; | |
1831 | return p->prio; | |
1832 | } | |
1833 | ||
1da177e4 | 1834 | /* |
dd41f596 | 1835 | * activate_task - move a task to the runqueue. |
1da177e4 | 1836 | */ |
dd41f596 | 1837 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1838 | { |
d9514f6c | 1839 | if (task_contributes_to_load(p)) |
dd41f596 | 1840 | rq->nr_uninterruptible--; |
1da177e4 | 1841 | |
8159f87e | 1842 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1843 | inc_nr_running(rq); |
1da177e4 LT |
1844 | } |
1845 | ||
1da177e4 LT |
1846 | /* |
1847 | * deactivate_task - remove a task from the runqueue. | |
1848 | */ | |
2e1cb74a | 1849 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1850 | { |
d9514f6c | 1851 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1852 | rq->nr_uninterruptible++; |
1853 | ||
69be72c1 | 1854 | dequeue_task(rq, p, sleep); |
c09595f6 | 1855 | dec_nr_running(rq); |
1da177e4 LT |
1856 | } |
1857 | ||
1da177e4 LT |
1858 | /** |
1859 | * task_curr - is this task currently executing on a CPU? | |
1860 | * @p: the task in question. | |
1861 | */ | |
36c8b586 | 1862 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1863 | { |
1864 | return cpu_curr(task_cpu(p)) == p; | |
1865 | } | |
1866 | ||
dd41f596 IM |
1867 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1868 | { | |
6f505b16 | 1869 | set_task_rq(p, cpu); |
dd41f596 | 1870 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1871 | /* |
1872 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1873 | * successfuly executed on another CPU. We must ensure that updates of | |
1874 | * per-task data have been completed by this moment. | |
1875 | */ | |
1876 | smp_wmb(); | |
dd41f596 | 1877 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1878 | #endif |
2dd73a4f PW |
1879 | } |
1880 | ||
cb469845 SR |
1881 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1882 | const struct sched_class *prev_class, | |
1883 | int oldprio, int running) | |
1884 | { | |
1885 | if (prev_class != p->sched_class) { | |
1886 | if (prev_class->switched_from) | |
1887 | prev_class->switched_from(rq, p, running); | |
1888 | p->sched_class->switched_to(rq, p, running); | |
1889 | } else | |
1890 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1891 | } | |
1892 | ||
1da177e4 | 1893 | #ifdef CONFIG_SMP |
c65cc870 | 1894 | |
e958b360 TG |
1895 | /* Used instead of source_load when we know the type == 0 */ |
1896 | static unsigned long weighted_cpuload(const int cpu) | |
1897 | { | |
1898 | return cpu_rq(cpu)->load.weight; | |
1899 | } | |
1900 | ||
cc367732 IM |
1901 | /* |
1902 | * Is this task likely cache-hot: | |
1903 | */ | |
e7693a36 | 1904 | static int |
cc367732 IM |
1905 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1906 | { | |
1907 | s64 delta; | |
1908 | ||
f540a608 IM |
1909 | /* |
1910 | * Buddy candidates are cache hot: | |
1911 | */ | |
4793241b PZ |
1912 | if (sched_feat(CACHE_HOT_BUDDY) && |
1913 | (&p->se == cfs_rq_of(&p->se)->next || | |
1914 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1915 | return 1; |
1916 | ||
cc367732 IM |
1917 | if (p->sched_class != &fair_sched_class) |
1918 | return 0; | |
1919 | ||
6bc1665b IM |
1920 | if (sysctl_sched_migration_cost == -1) |
1921 | return 1; | |
1922 | if (sysctl_sched_migration_cost == 0) | |
1923 | return 0; | |
1924 | ||
cc367732 IM |
1925 | delta = now - p->se.exec_start; |
1926 | ||
1927 | return delta < (s64)sysctl_sched_migration_cost; | |
1928 | } | |
1929 | ||
1930 | ||
dd41f596 | 1931 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1932 | { |
dd41f596 IM |
1933 | int old_cpu = task_cpu(p); |
1934 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1935 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1936 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1937 | u64 clock_offset; |
dd41f596 IM |
1938 | |
1939 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1940 | |
cbc34ed1 PZ |
1941 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1942 | ||
6cfb0d5d IM |
1943 | #ifdef CONFIG_SCHEDSTATS |
1944 | if (p->se.wait_start) | |
1945 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1946 | if (p->se.sleep_start) |
1947 | p->se.sleep_start -= clock_offset; | |
1948 | if (p->se.block_start) | |
1949 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1950 | if (old_cpu != new_cpu) { |
1951 | schedstat_inc(p, se.nr_migrations); | |
1952 | if (task_hot(p, old_rq->clock, NULL)) | |
1953 | schedstat_inc(p, se.nr_forced2_migrations); | |
1954 | } | |
6cfb0d5d | 1955 | #endif |
2830cf8c SV |
1956 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1957 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1958 | |
1959 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1960 | } |
1961 | ||
70b97a7f | 1962 | struct migration_req { |
1da177e4 | 1963 | struct list_head list; |
1da177e4 | 1964 | |
36c8b586 | 1965 | struct task_struct *task; |
1da177e4 LT |
1966 | int dest_cpu; |
1967 | ||
1da177e4 | 1968 | struct completion done; |
70b97a7f | 1969 | }; |
1da177e4 LT |
1970 | |
1971 | /* | |
1972 | * The task's runqueue lock must be held. | |
1973 | * Returns true if you have to wait for migration thread. | |
1974 | */ | |
36c8b586 | 1975 | static int |
70b97a7f | 1976 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1977 | { |
70b97a7f | 1978 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1979 | |
1980 | /* | |
1981 | * If the task is not on a runqueue (and not running), then | |
1982 | * it is sufficient to simply update the task's cpu field. | |
1983 | */ | |
dd41f596 | 1984 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1985 | set_task_cpu(p, dest_cpu); |
1986 | return 0; | |
1987 | } | |
1988 | ||
1989 | init_completion(&req->done); | |
1da177e4 LT |
1990 | req->task = p; |
1991 | req->dest_cpu = dest_cpu; | |
1992 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1993 | |
1da177e4 LT |
1994 | return 1; |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * wait_task_inactive - wait for a thread to unschedule. | |
1999 | * | |
85ba2d86 RM |
2000 | * If @match_state is nonzero, it's the @p->state value just checked and |
2001 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2002 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2003 | * we return a positive number (its total switch count). If a second call | |
2004 | * a short while later returns the same number, the caller can be sure that | |
2005 | * @p has remained unscheduled the whole time. | |
2006 | * | |
1da177e4 LT |
2007 | * The caller must ensure that the task *will* unschedule sometime soon, |
2008 | * else this function might spin for a *long* time. This function can't | |
2009 | * be called with interrupts off, or it may introduce deadlock with | |
2010 | * smp_call_function() if an IPI is sent by the same process we are | |
2011 | * waiting to become inactive. | |
2012 | */ | |
85ba2d86 | 2013 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2014 | { |
2015 | unsigned long flags; | |
dd41f596 | 2016 | int running, on_rq; |
85ba2d86 | 2017 | unsigned long ncsw; |
70b97a7f | 2018 | struct rq *rq; |
1da177e4 | 2019 | |
3a5c359a AK |
2020 | for (;;) { |
2021 | /* | |
2022 | * We do the initial early heuristics without holding | |
2023 | * any task-queue locks at all. We'll only try to get | |
2024 | * the runqueue lock when things look like they will | |
2025 | * work out! | |
2026 | */ | |
2027 | rq = task_rq(p); | |
fa490cfd | 2028 | |
3a5c359a AK |
2029 | /* |
2030 | * If the task is actively running on another CPU | |
2031 | * still, just relax and busy-wait without holding | |
2032 | * any locks. | |
2033 | * | |
2034 | * NOTE! Since we don't hold any locks, it's not | |
2035 | * even sure that "rq" stays as the right runqueue! | |
2036 | * But we don't care, since "task_running()" will | |
2037 | * return false if the runqueue has changed and p | |
2038 | * is actually now running somewhere else! | |
2039 | */ | |
85ba2d86 RM |
2040 | while (task_running(rq, p)) { |
2041 | if (match_state && unlikely(p->state != match_state)) | |
2042 | return 0; | |
3a5c359a | 2043 | cpu_relax(); |
85ba2d86 | 2044 | } |
fa490cfd | 2045 | |
3a5c359a AK |
2046 | /* |
2047 | * Ok, time to look more closely! We need the rq | |
2048 | * lock now, to be *sure*. If we're wrong, we'll | |
2049 | * just go back and repeat. | |
2050 | */ | |
2051 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2052 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2053 | running = task_running(rq, p); |
2054 | on_rq = p->se.on_rq; | |
85ba2d86 | 2055 | ncsw = 0; |
f31e11d8 | 2056 | if (!match_state || p->state == match_state) |
93dcf55f | 2057 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2058 | task_rq_unlock(rq, &flags); |
fa490cfd | 2059 | |
85ba2d86 RM |
2060 | /* |
2061 | * If it changed from the expected state, bail out now. | |
2062 | */ | |
2063 | if (unlikely(!ncsw)) | |
2064 | break; | |
2065 | ||
3a5c359a AK |
2066 | /* |
2067 | * Was it really running after all now that we | |
2068 | * checked with the proper locks actually held? | |
2069 | * | |
2070 | * Oops. Go back and try again.. | |
2071 | */ | |
2072 | if (unlikely(running)) { | |
2073 | cpu_relax(); | |
2074 | continue; | |
2075 | } | |
fa490cfd | 2076 | |
3a5c359a AK |
2077 | /* |
2078 | * It's not enough that it's not actively running, | |
2079 | * it must be off the runqueue _entirely_, and not | |
2080 | * preempted! | |
2081 | * | |
80dd99b3 | 2082 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2083 | * running right now), it's preempted, and we should |
2084 | * yield - it could be a while. | |
2085 | */ | |
2086 | if (unlikely(on_rq)) { | |
2087 | schedule_timeout_uninterruptible(1); | |
2088 | continue; | |
2089 | } | |
fa490cfd | 2090 | |
3a5c359a AK |
2091 | /* |
2092 | * Ahh, all good. It wasn't running, and it wasn't | |
2093 | * runnable, which means that it will never become | |
2094 | * running in the future either. We're all done! | |
2095 | */ | |
2096 | break; | |
2097 | } | |
85ba2d86 RM |
2098 | |
2099 | return ncsw; | |
1da177e4 LT |
2100 | } |
2101 | ||
2102 | /*** | |
2103 | * kick_process - kick a running thread to enter/exit the kernel | |
2104 | * @p: the to-be-kicked thread | |
2105 | * | |
2106 | * Cause a process which is running on another CPU to enter | |
2107 | * kernel-mode, without any delay. (to get signals handled.) | |
2108 | * | |
2109 | * NOTE: this function doesnt have to take the runqueue lock, | |
2110 | * because all it wants to ensure is that the remote task enters | |
2111 | * the kernel. If the IPI races and the task has been migrated | |
2112 | * to another CPU then no harm is done and the purpose has been | |
2113 | * achieved as well. | |
2114 | */ | |
36c8b586 | 2115 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2116 | { |
2117 | int cpu; | |
2118 | ||
2119 | preempt_disable(); | |
2120 | cpu = task_cpu(p); | |
2121 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2122 | smp_send_reschedule(cpu); | |
2123 | preempt_enable(); | |
2124 | } | |
2125 | ||
2126 | /* | |
2dd73a4f PW |
2127 | * Return a low guess at the load of a migration-source cpu weighted |
2128 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2129 | * |
2130 | * We want to under-estimate the load of migration sources, to | |
2131 | * balance conservatively. | |
2132 | */ | |
a9957449 | 2133 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2134 | { |
70b97a7f | 2135 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2136 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2137 | |
93b75217 | 2138 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2139 | return total; |
b910472d | 2140 | |
dd41f596 | 2141 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2142 | } |
2143 | ||
2144 | /* | |
2dd73a4f PW |
2145 | * Return a high guess at the load of a migration-target cpu weighted |
2146 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2147 | */ |
a9957449 | 2148 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2149 | { |
70b97a7f | 2150 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2151 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2152 | |
93b75217 | 2153 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2154 | return total; |
3b0bd9bc | 2155 | |
dd41f596 | 2156 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2157 | } |
2158 | ||
147cbb4b NP |
2159 | /* |
2160 | * find_idlest_group finds and returns the least busy CPU group within the | |
2161 | * domain. | |
2162 | */ | |
2163 | static struct sched_group * | |
2164 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2165 | { | |
2166 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2167 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2168 | int load_idx = sd->forkexec_idx; | |
2169 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2170 | ||
2171 | do { | |
2172 | unsigned long load, avg_load; | |
2173 | int local_group; | |
2174 | int i; | |
2175 | ||
da5a5522 | 2176 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2177 | if (!cpumask_intersects(sched_group_cpus(group), |
2178 | &p->cpus_allowed)) | |
3a5c359a | 2179 | continue; |
da5a5522 | 2180 | |
758b2cdc RR |
2181 | local_group = cpumask_test_cpu(this_cpu, |
2182 | sched_group_cpus(group)); | |
147cbb4b NP |
2183 | |
2184 | /* Tally up the load of all CPUs in the group */ | |
2185 | avg_load = 0; | |
2186 | ||
758b2cdc | 2187 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2188 | /* Bias balancing toward cpus of our domain */ |
2189 | if (local_group) | |
2190 | load = source_load(i, load_idx); | |
2191 | else | |
2192 | load = target_load(i, load_idx); | |
2193 | ||
2194 | avg_load += load; | |
2195 | } | |
2196 | ||
2197 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2198 | avg_load = sg_div_cpu_power(group, |
2199 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2200 | |
2201 | if (local_group) { | |
2202 | this_load = avg_load; | |
2203 | this = group; | |
2204 | } else if (avg_load < min_load) { | |
2205 | min_load = avg_load; | |
2206 | idlest = group; | |
2207 | } | |
3a5c359a | 2208 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2209 | |
2210 | if (!idlest || 100*this_load < imbalance*min_load) | |
2211 | return NULL; | |
2212 | return idlest; | |
2213 | } | |
2214 | ||
2215 | /* | |
0feaece9 | 2216 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2217 | */ |
95cdf3b7 | 2218 | static int |
758b2cdc | 2219 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2220 | { |
2221 | unsigned long load, min_load = ULONG_MAX; | |
2222 | int idlest = -1; | |
2223 | int i; | |
2224 | ||
da5a5522 | 2225 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2226 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2227 | load = weighted_cpuload(i); |
147cbb4b NP |
2228 | |
2229 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2230 | min_load = load; | |
2231 | idlest = i; | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | return idlest; | |
2236 | } | |
2237 | ||
476d139c NP |
2238 | /* |
2239 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2240 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2241 | * SD_BALANCE_EXEC. | |
2242 | * | |
2243 | * Balance, ie. select the least loaded group. | |
2244 | * | |
2245 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2246 | * | |
2247 | * preempt must be disabled. | |
2248 | */ | |
2249 | static int sched_balance_self(int cpu, int flag) | |
2250 | { | |
2251 | struct task_struct *t = current; | |
2252 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2253 | |
c96d145e | 2254 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2255 | /* |
2256 | * If power savings logic is enabled for a domain, stop there. | |
2257 | */ | |
5c45bf27 SS |
2258 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2259 | break; | |
476d139c NP |
2260 | if (tmp->flags & flag) |
2261 | sd = tmp; | |
c96d145e | 2262 | } |
476d139c | 2263 | |
039a1c41 PZ |
2264 | if (sd) |
2265 | update_shares(sd); | |
2266 | ||
476d139c | 2267 | while (sd) { |
476d139c | 2268 | struct sched_group *group; |
1a848870 SS |
2269 | int new_cpu, weight; |
2270 | ||
2271 | if (!(sd->flags & flag)) { | |
2272 | sd = sd->child; | |
2273 | continue; | |
2274 | } | |
476d139c | 2275 | |
476d139c | 2276 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2277 | if (!group) { |
2278 | sd = sd->child; | |
2279 | continue; | |
2280 | } | |
476d139c | 2281 | |
758b2cdc | 2282 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2283 | if (new_cpu == -1 || new_cpu == cpu) { |
2284 | /* Now try balancing at a lower domain level of cpu */ | |
2285 | sd = sd->child; | |
2286 | continue; | |
2287 | } | |
476d139c | 2288 | |
1a848870 | 2289 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2290 | cpu = new_cpu; |
758b2cdc | 2291 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2292 | sd = NULL; |
476d139c | 2293 | for_each_domain(cpu, tmp) { |
758b2cdc | 2294 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2295 | break; |
2296 | if (tmp->flags & flag) | |
2297 | sd = tmp; | |
2298 | } | |
2299 | /* while loop will break here if sd == NULL */ | |
2300 | } | |
2301 | ||
2302 | return cpu; | |
2303 | } | |
2304 | ||
2305 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2306 | |
1da177e4 LT |
2307 | /*** |
2308 | * try_to_wake_up - wake up a thread | |
2309 | * @p: the to-be-woken-up thread | |
2310 | * @state: the mask of task states that can be woken | |
2311 | * @sync: do a synchronous wakeup? | |
2312 | * | |
2313 | * Put it on the run-queue if it's not already there. The "current" | |
2314 | * thread is always on the run-queue (except when the actual | |
2315 | * re-schedule is in progress), and as such you're allowed to do | |
2316 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2317 | * runnable without the overhead of this. | |
2318 | * | |
2319 | * returns failure only if the task is already active. | |
2320 | */ | |
36c8b586 | 2321 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2322 | { |
cc367732 | 2323 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2324 | unsigned long flags; |
2325 | long old_state; | |
70b97a7f | 2326 | struct rq *rq; |
1da177e4 | 2327 | |
b85d0667 IM |
2328 | if (!sched_feat(SYNC_WAKEUPS)) |
2329 | sync = 0; | |
2330 | ||
2398f2c6 | 2331 | #ifdef CONFIG_SMP |
57310a98 | 2332 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2333 | struct sched_domain *sd; |
2334 | ||
2335 | this_cpu = raw_smp_processor_id(); | |
2336 | cpu = task_cpu(p); | |
2337 | ||
2338 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2339 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2340 | update_shares(sd); |
2341 | break; | |
2342 | } | |
2343 | } | |
2344 | } | |
2345 | #endif | |
2346 | ||
04e2f174 | 2347 | smp_wmb(); |
1da177e4 | 2348 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2349 | update_rq_clock(rq); |
1da177e4 LT |
2350 | old_state = p->state; |
2351 | if (!(old_state & state)) | |
2352 | goto out; | |
2353 | ||
dd41f596 | 2354 | if (p->se.on_rq) |
1da177e4 LT |
2355 | goto out_running; |
2356 | ||
2357 | cpu = task_cpu(p); | |
cc367732 | 2358 | orig_cpu = cpu; |
1da177e4 LT |
2359 | this_cpu = smp_processor_id(); |
2360 | ||
2361 | #ifdef CONFIG_SMP | |
2362 | if (unlikely(task_running(rq, p))) | |
2363 | goto out_activate; | |
2364 | ||
5d2f5a61 DA |
2365 | cpu = p->sched_class->select_task_rq(p, sync); |
2366 | if (cpu != orig_cpu) { | |
2367 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2368 | task_rq_unlock(rq, &flags); |
2369 | /* might preempt at this point */ | |
2370 | rq = task_rq_lock(p, &flags); | |
2371 | old_state = p->state; | |
2372 | if (!(old_state & state)) | |
2373 | goto out; | |
dd41f596 | 2374 | if (p->se.on_rq) |
1da177e4 LT |
2375 | goto out_running; |
2376 | ||
2377 | this_cpu = smp_processor_id(); | |
2378 | cpu = task_cpu(p); | |
2379 | } | |
2380 | ||
e7693a36 GH |
2381 | #ifdef CONFIG_SCHEDSTATS |
2382 | schedstat_inc(rq, ttwu_count); | |
2383 | if (cpu == this_cpu) | |
2384 | schedstat_inc(rq, ttwu_local); | |
2385 | else { | |
2386 | struct sched_domain *sd; | |
2387 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2388 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2389 | schedstat_inc(sd, ttwu_wake_remote); |
2390 | break; | |
2391 | } | |
2392 | } | |
2393 | } | |
6d6bc0ad | 2394 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2395 | |
1da177e4 LT |
2396 | out_activate: |
2397 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2398 | schedstat_inc(p, se.nr_wakeups); |
2399 | if (sync) | |
2400 | schedstat_inc(p, se.nr_wakeups_sync); | |
2401 | if (orig_cpu != cpu) | |
2402 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2403 | if (cpu == this_cpu) | |
2404 | schedstat_inc(p, se.nr_wakeups_local); | |
2405 | else | |
2406 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2407 | activate_task(rq, p, 1); |
1da177e4 LT |
2408 | success = 1; |
2409 | ||
831451ac PZ |
2410 | /* |
2411 | * Only attribute actual wakeups done by this task. | |
2412 | */ | |
2413 | if (!in_interrupt()) { | |
2414 | struct sched_entity *se = ¤t->se; | |
2415 | u64 sample = se->sum_exec_runtime; | |
2416 | ||
2417 | if (se->last_wakeup) | |
2418 | sample -= se->last_wakeup; | |
2419 | else | |
2420 | sample -= se->start_runtime; | |
2421 | update_avg(&se->avg_wakeup, sample); | |
2422 | ||
2423 | se->last_wakeup = se->sum_exec_runtime; | |
2424 | } | |
2425 | ||
1da177e4 | 2426 | out_running: |
468a15bb | 2427 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2428 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2429 | |
1da177e4 | 2430 | p->state = TASK_RUNNING; |
9a897c5a SR |
2431 | #ifdef CONFIG_SMP |
2432 | if (p->sched_class->task_wake_up) | |
2433 | p->sched_class->task_wake_up(rq, p); | |
2434 | #endif | |
1da177e4 LT |
2435 | out: |
2436 | task_rq_unlock(rq, &flags); | |
2437 | ||
2438 | return success; | |
2439 | } | |
2440 | ||
7ad5b3a5 | 2441 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2442 | { |
d9514f6c | 2443 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2444 | } |
1da177e4 LT |
2445 | EXPORT_SYMBOL(wake_up_process); |
2446 | ||
7ad5b3a5 | 2447 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2448 | { |
2449 | return try_to_wake_up(p, state, 0); | |
2450 | } | |
2451 | ||
1da177e4 LT |
2452 | /* |
2453 | * Perform scheduler related setup for a newly forked process p. | |
2454 | * p is forked by current. | |
dd41f596 IM |
2455 | * |
2456 | * __sched_fork() is basic setup used by init_idle() too: | |
2457 | */ | |
2458 | static void __sched_fork(struct task_struct *p) | |
2459 | { | |
dd41f596 IM |
2460 | p->se.exec_start = 0; |
2461 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2462 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2463 | p->se.last_wakeup = 0; |
2464 | p->se.avg_overlap = 0; | |
831451ac PZ |
2465 | p->se.start_runtime = 0; |
2466 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2467 | |
2468 | #ifdef CONFIG_SCHEDSTATS | |
2469 | p->se.wait_start = 0; | |
dd41f596 IM |
2470 | p->se.sum_sleep_runtime = 0; |
2471 | p->se.sleep_start = 0; | |
dd41f596 IM |
2472 | p->se.block_start = 0; |
2473 | p->se.sleep_max = 0; | |
2474 | p->se.block_max = 0; | |
2475 | p->se.exec_max = 0; | |
eba1ed4b | 2476 | p->se.slice_max = 0; |
dd41f596 | 2477 | p->se.wait_max = 0; |
6cfb0d5d | 2478 | #endif |
476d139c | 2479 | |
fa717060 | 2480 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2481 | p->se.on_rq = 0; |
4a55bd5e | 2482 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2483 | |
e107be36 AK |
2484 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2485 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2486 | #endif | |
2487 | ||
1da177e4 LT |
2488 | /* |
2489 | * We mark the process as running here, but have not actually | |
2490 | * inserted it onto the runqueue yet. This guarantees that | |
2491 | * nobody will actually run it, and a signal or other external | |
2492 | * event cannot wake it up and insert it on the runqueue either. | |
2493 | */ | |
2494 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2495 | } |
2496 | ||
2497 | /* | |
2498 | * fork()/clone()-time setup: | |
2499 | */ | |
2500 | void sched_fork(struct task_struct *p, int clone_flags) | |
2501 | { | |
2502 | int cpu = get_cpu(); | |
2503 | ||
2504 | __sched_fork(p); | |
2505 | ||
2506 | #ifdef CONFIG_SMP | |
2507 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2508 | #endif | |
02e4bac2 | 2509 | set_task_cpu(p, cpu); |
b29739f9 IM |
2510 | |
2511 | /* | |
2512 | * Make sure we do not leak PI boosting priority to the child: | |
2513 | */ | |
2514 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2515 | if (!rt_prio(p->prio)) |
2516 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2517 | |
52f17b6c | 2518 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2519 | if (likely(sched_info_on())) |
52f17b6c | 2520 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2521 | #endif |
d6077cb8 | 2522 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2523 | p->oncpu = 0; |
2524 | #endif | |
1da177e4 | 2525 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2526 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2527 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2528 | #endif |
917b627d GH |
2529 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2530 | ||
476d139c | 2531 | put_cpu(); |
1da177e4 LT |
2532 | } |
2533 | ||
2534 | /* | |
2535 | * wake_up_new_task - wake up a newly created task for the first time. | |
2536 | * | |
2537 | * This function will do some initial scheduler statistics housekeeping | |
2538 | * that must be done for every newly created context, then puts the task | |
2539 | * on the runqueue and wakes it. | |
2540 | */ | |
7ad5b3a5 | 2541 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2542 | { |
2543 | unsigned long flags; | |
dd41f596 | 2544 | struct rq *rq; |
1da177e4 LT |
2545 | |
2546 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2547 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2548 | update_rq_clock(rq); |
1da177e4 LT |
2549 | |
2550 | p->prio = effective_prio(p); | |
2551 | ||
b9dca1e0 | 2552 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2553 | activate_task(rq, p, 0); |
1da177e4 | 2554 | } else { |
1da177e4 | 2555 | /* |
dd41f596 IM |
2556 | * Let the scheduling class do new task startup |
2557 | * management (if any): | |
1da177e4 | 2558 | */ |
ee0827d8 | 2559 | p->sched_class->task_new(rq, p); |
c09595f6 | 2560 | inc_nr_running(rq); |
1da177e4 | 2561 | } |
c71dd42d | 2562 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2563 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2564 | #ifdef CONFIG_SMP |
2565 | if (p->sched_class->task_wake_up) | |
2566 | p->sched_class->task_wake_up(rq, p); | |
2567 | #endif | |
dd41f596 | 2568 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2569 | } |
2570 | ||
e107be36 AK |
2571 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2572 | ||
2573 | /** | |
80dd99b3 | 2574 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2575 | * @notifier: notifier struct to register |
e107be36 AK |
2576 | */ |
2577 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2578 | { | |
2579 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2580 | } | |
2581 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2582 | ||
2583 | /** | |
2584 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2585 | * @notifier: notifier struct to unregister |
e107be36 AK |
2586 | * |
2587 | * This is safe to call from within a preemption notifier. | |
2588 | */ | |
2589 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2590 | { | |
2591 | hlist_del(¬ifier->link); | |
2592 | } | |
2593 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2594 | ||
2595 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2596 | { | |
2597 | struct preempt_notifier *notifier; | |
2598 | struct hlist_node *node; | |
2599 | ||
2600 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2601 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2602 | } | |
2603 | ||
2604 | static void | |
2605 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2606 | struct task_struct *next) | |
2607 | { | |
2608 | struct preempt_notifier *notifier; | |
2609 | struct hlist_node *node; | |
2610 | ||
2611 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2612 | notifier->ops->sched_out(notifier, next); | |
2613 | } | |
2614 | ||
6d6bc0ad | 2615 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2616 | |
2617 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2618 | { | |
2619 | } | |
2620 | ||
2621 | static void | |
2622 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2623 | struct task_struct *next) | |
2624 | { | |
2625 | } | |
2626 | ||
6d6bc0ad | 2627 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2628 | |
4866cde0 NP |
2629 | /** |
2630 | * prepare_task_switch - prepare to switch tasks | |
2631 | * @rq: the runqueue preparing to switch | |
421cee29 | 2632 | * @prev: the current task that is being switched out |
4866cde0 NP |
2633 | * @next: the task we are going to switch to. |
2634 | * | |
2635 | * This is called with the rq lock held and interrupts off. It must | |
2636 | * be paired with a subsequent finish_task_switch after the context | |
2637 | * switch. | |
2638 | * | |
2639 | * prepare_task_switch sets up locking and calls architecture specific | |
2640 | * hooks. | |
2641 | */ | |
e107be36 AK |
2642 | static inline void |
2643 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2644 | struct task_struct *next) | |
4866cde0 | 2645 | { |
e107be36 | 2646 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2647 | prepare_lock_switch(rq, next); |
2648 | prepare_arch_switch(next); | |
2649 | } | |
2650 | ||
1da177e4 LT |
2651 | /** |
2652 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2653 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2654 | * @prev: the thread we just switched away from. |
2655 | * | |
4866cde0 NP |
2656 | * finish_task_switch must be called after the context switch, paired |
2657 | * with a prepare_task_switch call before the context switch. | |
2658 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2659 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2660 | * |
2661 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2662 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2663 | * with the lock held can cause deadlocks; see schedule() for |
2664 | * details.) | |
2665 | */ | |
a9957449 | 2666 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2667 | __releases(rq->lock) |
2668 | { | |
1da177e4 | 2669 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2670 | long prev_state; |
967fc046 GH |
2671 | #ifdef CONFIG_SMP |
2672 | int post_schedule = 0; | |
2673 | ||
2674 | if (current->sched_class->needs_post_schedule) | |
2675 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2676 | #endif | |
1da177e4 LT |
2677 | |
2678 | rq->prev_mm = NULL; | |
2679 | ||
2680 | /* | |
2681 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2682 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2683 | * schedule one last time. The schedule call will never return, and |
2684 | * the scheduled task must drop that reference. | |
c394cc9f | 2685 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2686 | * still held, otherwise prev could be scheduled on another cpu, die |
2687 | * there before we look at prev->state, and then the reference would | |
2688 | * be dropped twice. | |
2689 | * Manfred Spraul <manfred@colorfullife.com> | |
2690 | */ | |
55a101f8 | 2691 | prev_state = prev->state; |
4866cde0 NP |
2692 | finish_arch_switch(prev); |
2693 | finish_lock_switch(rq, prev); | |
9a897c5a | 2694 | #ifdef CONFIG_SMP |
967fc046 | 2695 | if (post_schedule) |
9a897c5a SR |
2696 | current->sched_class->post_schedule(rq); |
2697 | #endif | |
e8fa1362 | 2698 | |
e107be36 | 2699 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2700 | if (mm) |
2701 | mmdrop(mm); | |
c394cc9f | 2702 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2703 | /* |
2704 | * Remove function-return probe instances associated with this | |
2705 | * task and put them back on the free list. | |
9761eea8 | 2706 | */ |
c6fd91f0 | 2707 | kprobe_flush_task(prev); |
1da177e4 | 2708 | put_task_struct(prev); |
c6fd91f0 | 2709 | } |
1da177e4 LT |
2710 | } |
2711 | ||
2712 | /** | |
2713 | * schedule_tail - first thing a freshly forked thread must call. | |
2714 | * @prev: the thread we just switched away from. | |
2715 | */ | |
36c8b586 | 2716 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2717 | __releases(rq->lock) |
2718 | { | |
70b97a7f IM |
2719 | struct rq *rq = this_rq(); |
2720 | ||
4866cde0 NP |
2721 | finish_task_switch(rq, prev); |
2722 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2723 | /* In this case, finish_task_switch does not reenable preemption */ | |
2724 | preempt_enable(); | |
2725 | #endif | |
1da177e4 | 2726 | if (current->set_child_tid) |
b488893a | 2727 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2728 | } |
2729 | ||
2730 | /* | |
2731 | * context_switch - switch to the new MM and the new | |
2732 | * thread's register state. | |
2733 | */ | |
dd41f596 | 2734 | static inline void |
70b97a7f | 2735 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2736 | struct task_struct *next) |
1da177e4 | 2737 | { |
dd41f596 | 2738 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2739 | |
e107be36 | 2740 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2741 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2742 | mm = next->mm; |
2743 | oldmm = prev->active_mm; | |
9226d125 ZA |
2744 | /* |
2745 | * For paravirt, this is coupled with an exit in switch_to to | |
2746 | * combine the page table reload and the switch backend into | |
2747 | * one hypercall. | |
2748 | */ | |
7fd7d83d | 2749 | arch_start_context_switch(); |
9226d125 | 2750 | |
dd41f596 | 2751 | if (unlikely(!mm)) { |
1da177e4 LT |
2752 | next->active_mm = oldmm; |
2753 | atomic_inc(&oldmm->mm_count); | |
2754 | enter_lazy_tlb(oldmm, next); | |
2755 | } else | |
2756 | switch_mm(oldmm, mm, next); | |
2757 | ||
dd41f596 | 2758 | if (unlikely(!prev->mm)) { |
1da177e4 | 2759 | prev->active_mm = NULL; |
1da177e4 LT |
2760 | rq->prev_mm = oldmm; |
2761 | } | |
3a5f5e48 IM |
2762 | /* |
2763 | * Since the runqueue lock will be released by the next | |
2764 | * task (which is an invalid locking op but in the case | |
2765 | * of the scheduler it's an obvious special-case), so we | |
2766 | * do an early lockdep release here: | |
2767 | */ | |
2768 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2769 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2770 | #endif |
1da177e4 LT |
2771 | |
2772 | /* Here we just switch the register state and the stack. */ | |
2773 | switch_to(prev, next, prev); | |
2774 | ||
dd41f596 IM |
2775 | barrier(); |
2776 | /* | |
2777 | * this_rq must be evaluated again because prev may have moved | |
2778 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2779 | * frame will be invalid. | |
2780 | */ | |
2781 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2782 | } |
2783 | ||
2784 | /* | |
2785 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2786 | * | |
2787 | * externally visible scheduler statistics: current number of runnable | |
2788 | * threads, current number of uninterruptible-sleeping threads, total | |
2789 | * number of context switches performed since bootup. | |
2790 | */ | |
2791 | unsigned long nr_running(void) | |
2792 | { | |
2793 | unsigned long i, sum = 0; | |
2794 | ||
2795 | for_each_online_cpu(i) | |
2796 | sum += cpu_rq(i)->nr_running; | |
2797 | ||
2798 | return sum; | |
2799 | } | |
2800 | ||
2801 | unsigned long nr_uninterruptible(void) | |
2802 | { | |
2803 | unsigned long i, sum = 0; | |
2804 | ||
0a945022 | 2805 | for_each_possible_cpu(i) |
1da177e4 LT |
2806 | sum += cpu_rq(i)->nr_uninterruptible; |
2807 | ||
2808 | /* | |
2809 | * Since we read the counters lockless, it might be slightly | |
2810 | * inaccurate. Do not allow it to go below zero though: | |
2811 | */ | |
2812 | if (unlikely((long)sum < 0)) | |
2813 | sum = 0; | |
2814 | ||
2815 | return sum; | |
2816 | } | |
2817 | ||
2818 | unsigned long long nr_context_switches(void) | |
2819 | { | |
cc94abfc SR |
2820 | int i; |
2821 | unsigned long long sum = 0; | |
1da177e4 | 2822 | |
0a945022 | 2823 | for_each_possible_cpu(i) |
1da177e4 LT |
2824 | sum += cpu_rq(i)->nr_switches; |
2825 | ||
2826 | return sum; | |
2827 | } | |
2828 | ||
2829 | unsigned long nr_iowait(void) | |
2830 | { | |
2831 | unsigned long i, sum = 0; | |
2832 | ||
0a945022 | 2833 | for_each_possible_cpu(i) |
1da177e4 LT |
2834 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2835 | ||
2836 | return sum; | |
2837 | } | |
2838 | ||
db1b1fef JS |
2839 | unsigned long nr_active(void) |
2840 | { | |
2841 | unsigned long i, running = 0, uninterruptible = 0; | |
2842 | ||
2843 | for_each_online_cpu(i) { | |
2844 | running += cpu_rq(i)->nr_running; | |
2845 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2846 | } | |
2847 | ||
2848 | if (unlikely((long)uninterruptible < 0)) | |
2849 | uninterruptible = 0; | |
2850 | ||
2851 | return running + uninterruptible; | |
2852 | } | |
2853 | ||
48f24c4d | 2854 | /* |
dd41f596 IM |
2855 | * Update rq->cpu_load[] statistics. This function is usually called every |
2856 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2857 | */ |
dd41f596 | 2858 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2859 | { |
495eca49 | 2860 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2861 | int i, scale; |
2862 | ||
2863 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2864 | |
2865 | /* Update our load: */ | |
2866 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2867 | unsigned long old_load, new_load; | |
2868 | ||
2869 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2870 | ||
2871 | old_load = this_rq->cpu_load[i]; | |
2872 | new_load = this_load; | |
a25707f3 IM |
2873 | /* |
2874 | * Round up the averaging division if load is increasing. This | |
2875 | * prevents us from getting stuck on 9 if the load is 10, for | |
2876 | * example. | |
2877 | */ | |
2878 | if (new_load > old_load) | |
2879 | new_load += scale-1; | |
dd41f596 IM |
2880 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2881 | } | |
48f24c4d IM |
2882 | } |
2883 | ||
dd41f596 IM |
2884 | #ifdef CONFIG_SMP |
2885 | ||
1da177e4 LT |
2886 | /* |
2887 | * double_rq_lock - safely lock two runqueues | |
2888 | * | |
2889 | * Note this does not disable interrupts like task_rq_lock, | |
2890 | * you need to do so manually before calling. | |
2891 | */ | |
70b97a7f | 2892 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2893 | __acquires(rq1->lock) |
2894 | __acquires(rq2->lock) | |
2895 | { | |
054b9108 | 2896 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2897 | if (rq1 == rq2) { |
2898 | spin_lock(&rq1->lock); | |
2899 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2900 | } else { | |
c96d145e | 2901 | if (rq1 < rq2) { |
1da177e4 | 2902 | spin_lock(&rq1->lock); |
5e710e37 | 2903 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2904 | } else { |
2905 | spin_lock(&rq2->lock); | |
5e710e37 | 2906 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2907 | } |
2908 | } | |
6e82a3be IM |
2909 | update_rq_clock(rq1); |
2910 | update_rq_clock(rq2); | |
1da177e4 LT |
2911 | } |
2912 | ||
2913 | /* | |
2914 | * double_rq_unlock - safely unlock two runqueues | |
2915 | * | |
2916 | * Note this does not restore interrupts like task_rq_unlock, | |
2917 | * you need to do so manually after calling. | |
2918 | */ | |
70b97a7f | 2919 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2920 | __releases(rq1->lock) |
2921 | __releases(rq2->lock) | |
2922 | { | |
2923 | spin_unlock(&rq1->lock); | |
2924 | if (rq1 != rq2) | |
2925 | spin_unlock(&rq2->lock); | |
2926 | else | |
2927 | __release(rq2->lock); | |
2928 | } | |
2929 | ||
1da177e4 LT |
2930 | /* |
2931 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2932 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2933 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2934 | * the cpu_allowed mask is restored. |
2935 | */ | |
36c8b586 | 2936 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2937 | { |
70b97a7f | 2938 | struct migration_req req; |
1da177e4 | 2939 | unsigned long flags; |
70b97a7f | 2940 | struct rq *rq; |
1da177e4 LT |
2941 | |
2942 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2943 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2944 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2945 | goto out; |
2946 | ||
2947 | /* force the process onto the specified CPU */ | |
2948 | if (migrate_task(p, dest_cpu, &req)) { | |
2949 | /* Need to wait for migration thread (might exit: take ref). */ | |
2950 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2951 | |
1da177e4 LT |
2952 | get_task_struct(mt); |
2953 | task_rq_unlock(rq, &flags); | |
2954 | wake_up_process(mt); | |
2955 | put_task_struct(mt); | |
2956 | wait_for_completion(&req.done); | |
36c8b586 | 2957 | |
1da177e4 LT |
2958 | return; |
2959 | } | |
2960 | out: | |
2961 | task_rq_unlock(rq, &flags); | |
2962 | } | |
2963 | ||
2964 | /* | |
476d139c NP |
2965 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2966 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2967 | */ |
2968 | void sched_exec(void) | |
2969 | { | |
1da177e4 | 2970 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2971 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2972 | put_cpu(); |
476d139c NP |
2973 | if (new_cpu != this_cpu) |
2974 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2975 | } |
2976 | ||
2977 | /* | |
2978 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2979 | * Both runqueues must be locked. | |
2980 | */ | |
dd41f596 IM |
2981 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2982 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2983 | { |
2e1cb74a | 2984 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2985 | set_task_cpu(p, this_cpu); |
dd41f596 | 2986 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2987 | /* |
2988 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2989 | * to be always true for them. | |
2990 | */ | |
15afe09b | 2991 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2992 | } |
2993 | ||
2994 | /* | |
2995 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2996 | */ | |
858119e1 | 2997 | static |
70b97a7f | 2998 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2999 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3000 | int *all_pinned) |
1da177e4 | 3001 | { |
708dc512 | 3002 | int tsk_cache_hot = 0; |
1da177e4 LT |
3003 | /* |
3004 | * We do not migrate tasks that are: | |
3005 | * 1) running (obviously), or | |
3006 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3007 | * 3) are cache-hot on their current CPU. | |
3008 | */ | |
96f874e2 | 3009 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3010 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3011 | return 0; |
cc367732 | 3012 | } |
81026794 NP |
3013 | *all_pinned = 0; |
3014 | ||
cc367732 IM |
3015 | if (task_running(rq, p)) { |
3016 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3017 | return 0; |
cc367732 | 3018 | } |
1da177e4 | 3019 | |
da84d961 IM |
3020 | /* |
3021 | * Aggressive migration if: | |
3022 | * 1) task is cache cold, or | |
3023 | * 2) too many balance attempts have failed. | |
3024 | */ | |
3025 | ||
708dc512 LH |
3026 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3027 | if (!tsk_cache_hot || | |
3028 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3029 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3030 | if (tsk_cache_hot) { |
da84d961 | 3031 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3032 | schedstat_inc(p, se.nr_forced_migrations); |
3033 | } | |
da84d961 IM |
3034 | #endif |
3035 | return 1; | |
3036 | } | |
3037 | ||
708dc512 | 3038 | if (tsk_cache_hot) { |
cc367732 | 3039 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3040 | return 0; |
cc367732 | 3041 | } |
1da177e4 LT |
3042 | return 1; |
3043 | } | |
3044 | ||
e1d1484f PW |
3045 | static unsigned long |
3046 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3047 | unsigned long max_load_move, struct sched_domain *sd, | |
3048 | enum cpu_idle_type idle, int *all_pinned, | |
3049 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3050 | { |
051c6764 | 3051 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3052 | struct task_struct *p; |
3053 | long rem_load_move = max_load_move; | |
1da177e4 | 3054 | |
e1d1484f | 3055 | if (max_load_move == 0) |
1da177e4 LT |
3056 | goto out; |
3057 | ||
81026794 NP |
3058 | pinned = 1; |
3059 | ||
1da177e4 | 3060 | /* |
dd41f596 | 3061 | * Start the load-balancing iterator: |
1da177e4 | 3062 | */ |
dd41f596 IM |
3063 | p = iterator->start(iterator->arg); |
3064 | next: | |
b82d9fdd | 3065 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3066 | goto out; |
051c6764 PZ |
3067 | |
3068 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3069 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3070 | p = iterator->next(iterator->arg); |
3071 | goto next; | |
1da177e4 LT |
3072 | } |
3073 | ||
dd41f596 | 3074 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3075 | pulled++; |
dd41f596 | 3076 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3077 | |
7e96fa58 GH |
3078 | #ifdef CONFIG_PREEMPT |
3079 | /* | |
3080 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3081 | * will stop after the first task is pulled to minimize the critical | |
3082 | * section. | |
3083 | */ | |
3084 | if (idle == CPU_NEWLY_IDLE) | |
3085 | goto out; | |
3086 | #endif | |
3087 | ||
2dd73a4f | 3088 | /* |
b82d9fdd | 3089 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3090 | */ |
e1d1484f | 3091 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3092 | if (p->prio < *this_best_prio) |
3093 | *this_best_prio = p->prio; | |
dd41f596 IM |
3094 | p = iterator->next(iterator->arg); |
3095 | goto next; | |
1da177e4 LT |
3096 | } |
3097 | out: | |
3098 | /* | |
e1d1484f | 3099 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3100 | * so we can safely collect pull_task() stats here rather than |
3101 | * inside pull_task(). | |
3102 | */ | |
3103 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3104 | |
3105 | if (all_pinned) | |
3106 | *all_pinned = pinned; | |
e1d1484f PW |
3107 | |
3108 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3109 | } |
3110 | ||
dd41f596 | 3111 | /* |
43010659 PW |
3112 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3113 | * this_rq, as part of a balancing operation within domain "sd". | |
3114 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3115 | * |
3116 | * Called with both runqueues locked. | |
3117 | */ | |
3118 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3119 | unsigned long max_load_move, |
dd41f596 IM |
3120 | struct sched_domain *sd, enum cpu_idle_type idle, |
3121 | int *all_pinned) | |
3122 | { | |
5522d5d5 | 3123 | const struct sched_class *class = sched_class_highest; |
43010659 | 3124 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3125 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3126 | |
3127 | do { | |
43010659 PW |
3128 | total_load_moved += |
3129 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3130 | max_load_move - total_load_moved, |
a4ac01c3 | 3131 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3132 | class = class->next; |
c4acb2c0 | 3133 | |
7e96fa58 GH |
3134 | #ifdef CONFIG_PREEMPT |
3135 | /* | |
3136 | * NEWIDLE balancing is a source of latency, so preemptible | |
3137 | * kernels will stop after the first task is pulled to minimize | |
3138 | * the critical section. | |
3139 | */ | |
c4acb2c0 GH |
3140 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3141 | break; | |
7e96fa58 | 3142 | #endif |
43010659 | 3143 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3144 | |
43010659 PW |
3145 | return total_load_moved > 0; |
3146 | } | |
3147 | ||
e1d1484f PW |
3148 | static int |
3149 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3150 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3151 | struct rq_iterator *iterator) | |
3152 | { | |
3153 | struct task_struct *p = iterator->start(iterator->arg); | |
3154 | int pinned = 0; | |
3155 | ||
3156 | while (p) { | |
3157 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3158 | pull_task(busiest, p, this_rq, this_cpu); | |
3159 | /* | |
3160 | * Right now, this is only the second place pull_task() | |
3161 | * is called, so we can safely collect pull_task() | |
3162 | * stats here rather than inside pull_task(). | |
3163 | */ | |
3164 | schedstat_inc(sd, lb_gained[idle]); | |
3165 | ||
3166 | return 1; | |
3167 | } | |
3168 | p = iterator->next(iterator->arg); | |
3169 | } | |
3170 | ||
3171 | return 0; | |
3172 | } | |
3173 | ||
43010659 PW |
3174 | /* |
3175 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3176 | * part of active balancing operations within "domain". | |
3177 | * Returns 1 if successful and 0 otherwise. | |
3178 | * | |
3179 | * Called with both runqueues locked. | |
3180 | */ | |
3181 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3182 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3183 | { | |
5522d5d5 | 3184 | const struct sched_class *class; |
43010659 PW |
3185 | |
3186 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3187 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3188 | return 1; |
3189 | ||
3190 | return 0; | |
dd41f596 | 3191 | } |
67bb6c03 | 3192 | /********** Helpers for find_busiest_group ************************/ |
d5ac537e | 3193 | /* |
222d656d GS |
3194 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3195 | * during load balancing. | |
3196 | */ | |
3197 | struct sd_lb_stats { | |
3198 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3199 | struct sched_group *this; /* Local group in this sd */ | |
3200 | unsigned long total_load; /* Total load of all groups in sd */ | |
3201 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3202 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3203 | ||
3204 | /** Statistics of this group */ | |
3205 | unsigned long this_load; | |
3206 | unsigned long this_load_per_task; | |
3207 | unsigned long this_nr_running; | |
3208 | ||
3209 | /* Statistics of the busiest group */ | |
3210 | unsigned long max_load; | |
3211 | unsigned long busiest_load_per_task; | |
3212 | unsigned long busiest_nr_running; | |
3213 | ||
3214 | int group_imb; /* Is there imbalance in this sd */ | |
3215 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3216 | int power_savings_balance; /* Is powersave balance needed for this sd */ | |
3217 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3218 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3219 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3220 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3221 | unsigned long min_nr_running; /* Nr running of group_min */ | |
3222 | #endif | |
3223 | }; | |
67bb6c03 | 3224 | |
d5ac537e | 3225 | /* |
381be78f GS |
3226 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3227 | */ | |
3228 | struct sg_lb_stats { | |
3229 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3230 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3231 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3232 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3233 | unsigned long group_capacity; | |
3234 | int group_imb; /* Is there an imbalance in the group ? */ | |
3235 | }; | |
3236 | ||
67bb6c03 GS |
3237 | /** |
3238 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3239 | * @group: The group whose first cpu is to be returned. | |
3240 | */ | |
3241 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3242 | { | |
3243 | return cpumask_first(sched_group_cpus(group)); | |
3244 | } | |
3245 | ||
3246 | /** | |
3247 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3248 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3249 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3250 | */ | |
3251 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3252 | enum cpu_idle_type idle) | |
3253 | { | |
3254 | int load_idx; | |
3255 | ||
3256 | switch (idle) { | |
3257 | case CPU_NOT_IDLE: | |
3258 | load_idx = sd->busy_idx; | |
3259 | break; | |
3260 | ||
3261 | case CPU_NEWLY_IDLE: | |
3262 | load_idx = sd->newidle_idx; | |
3263 | break; | |
3264 | default: | |
3265 | load_idx = sd->idle_idx; | |
3266 | break; | |
3267 | } | |
3268 | ||
3269 | return load_idx; | |
3270 | } | |
1f8c553d GS |
3271 | |
3272 | ||
c071df18 GS |
3273 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3274 | /** | |
3275 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3276 | * the given sched_domain, during load balancing. | |
3277 | * | |
3278 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3279 | * @sds: Variable containing the statistics for sd. | |
3280 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3281 | */ | |
3282 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3283 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3284 | { | |
3285 | /* | |
3286 | * Busy processors will not participate in power savings | |
3287 | * balance. | |
3288 | */ | |
3289 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3290 | sds->power_savings_balance = 0; | |
3291 | else { | |
3292 | sds->power_savings_balance = 1; | |
3293 | sds->min_nr_running = ULONG_MAX; | |
3294 | sds->leader_nr_running = 0; | |
3295 | } | |
3296 | } | |
3297 | ||
3298 | /** | |
3299 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3300 | * sched_domain while performing load balancing. | |
3301 | * | |
3302 | * @group: sched_group belonging to the sched_domain under consideration. | |
3303 | * @sds: Variable containing the statistics of the sched_domain | |
3304 | * @local_group: Does group contain the CPU for which we're performing | |
3305 | * load balancing ? | |
3306 | * @sgs: Variable containing the statistics of the group. | |
3307 | */ | |
3308 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3309 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3310 | { | |
3311 | ||
3312 | if (!sds->power_savings_balance) | |
3313 | return; | |
3314 | ||
3315 | /* | |
3316 | * If the local group is idle or completely loaded | |
3317 | * no need to do power savings balance at this domain | |
3318 | */ | |
3319 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3320 | !sds->this_nr_running)) | |
3321 | sds->power_savings_balance = 0; | |
3322 | ||
3323 | /* | |
3324 | * If a group is already running at full capacity or idle, | |
3325 | * don't include that group in power savings calculations | |
3326 | */ | |
3327 | if (!sds->power_savings_balance || | |
3328 | sgs->sum_nr_running >= sgs->group_capacity || | |
3329 | !sgs->sum_nr_running) | |
3330 | return; | |
3331 | ||
3332 | /* | |
3333 | * Calculate the group which has the least non-idle load. | |
3334 | * This is the group from where we need to pick up the load | |
3335 | * for saving power | |
3336 | */ | |
3337 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3338 | (sgs->sum_nr_running == sds->min_nr_running && | |
3339 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3340 | sds->group_min = group; | |
3341 | sds->min_nr_running = sgs->sum_nr_running; | |
3342 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3343 | sgs->sum_nr_running; | |
3344 | } | |
3345 | ||
3346 | /* | |
3347 | * Calculate the group which is almost near its | |
3348 | * capacity but still has some space to pick up some load | |
3349 | * from other group and save more power | |
3350 | */ | |
3351 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3352 | return; | |
3353 | ||
3354 | if (sgs->sum_nr_running > sds->leader_nr_running || | |
3355 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3356 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3357 | sds->group_leader = group; | |
3358 | sds->leader_nr_running = sgs->sum_nr_running; | |
3359 | } | |
3360 | } | |
3361 | ||
3362 | /** | |
d5ac537e | 3363 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3364 | * @sds: Variable containing the statistics of the sched_domain |
3365 | * under consideration. | |
3366 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3367 | * @imbalance: Variable to store the imbalance. | |
3368 | * | |
d5ac537e RD |
3369 | * Description: |
3370 | * Check if we have potential to perform some power-savings balance. | |
3371 | * If yes, set the busiest group to be the least loaded group in the | |
3372 | * sched_domain, so that it's CPUs can be put to idle. | |
3373 | * | |
c071df18 GS |
3374 | * Returns 1 if there is potential to perform power-savings balance. |
3375 | * Else returns 0. | |
3376 | */ | |
3377 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3378 | int this_cpu, unsigned long *imbalance) | |
3379 | { | |
3380 | if (!sds->power_savings_balance) | |
3381 | return 0; | |
3382 | ||
3383 | if (sds->this != sds->group_leader || | |
3384 | sds->group_leader == sds->group_min) | |
3385 | return 0; | |
3386 | ||
3387 | *imbalance = sds->min_load_per_task; | |
3388 | sds->busiest = sds->group_min; | |
3389 | ||
3390 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { | |
3391 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3392 | group_first_cpu(sds->group_leader); | |
3393 | } | |
3394 | ||
3395 | return 1; | |
3396 | ||
3397 | } | |
3398 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3399 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3400 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3401 | { | |
3402 | return; | |
3403 | } | |
3404 | ||
3405 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3406 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3407 | { | |
3408 | return; | |
3409 | } | |
3410 | ||
3411 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3412 | int this_cpu, unsigned long *imbalance) | |
3413 | { | |
3414 | return 0; | |
3415 | } | |
3416 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3417 | ||
3418 | ||
1f8c553d GS |
3419 | /** |
3420 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3421 | * @group: sched_group whose statistics are to be updated. | |
3422 | * @this_cpu: Cpu for which load balance is currently performed. | |
3423 | * @idle: Idle status of this_cpu | |
3424 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3425 | * @sd_idle: Idle status of the sched_domain containing group. | |
3426 | * @local_group: Does group contain this_cpu. | |
3427 | * @cpus: Set of cpus considered for load balancing. | |
3428 | * @balance: Should we balance. | |
3429 | * @sgs: variable to hold the statistics for this group. | |
3430 | */ | |
3431 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3432 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3433 | int local_group, const struct cpumask *cpus, | |
3434 | int *balance, struct sg_lb_stats *sgs) | |
3435 | { | |
3436 | unsigned long load, max_cpu_load, min_cpu_load; | |
3437 | int i; | |
3438 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3439 | unsigned long sum_avg_load_per_task; | |
3440 | unsigned long avg_load_per_task; | |
3441 | ||
3442 | if (local_group) | |
3443 | balance_cpu = group_first_cpu(group); | |
3444 | ||
3445 | /* Tally up the load of all CPUs in the group */ | |
3446 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3447 | max_cpu_load = 0; | |
3448 | min_cpu_load = ~0UL; | |
3449 | ||
3450 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { | |
3451 | struct rq *rq = cpu_rq(i); | |
3452 | ||
3453 | if (*sd_idle && rq->nr_running) | |
3454 | *sd_idle = 0; | |
3455 | ||
3456 | /* Bias balancing toward cpus of our domain */ | |
3457 | if (local_group) { | |
3458 | if (idle_cpu(i) && !first_idle_cpu) { | |
3459 | first_idle_cpu = 1; | |
3460 | balance_cpu = i; | |
3461 | } | |
3462 | ||
3463 | load = target_load(i, load_idx); | |
3464 | } else { | |
3465 | load = source_load(i, load_idx); | |
3466 | if (load > max_cpu_load) | |
3467 | max_cpu_load = load; | |
3468 | if (min_cpu_load > load) | |
3469 | min_cpu_load = load; | |
3470 | } | |
3471 | ||
3472 | sgs->group_load += load; | |
3473 | sgs->sum_nr_running += rq->nr_running; | |
3474 | sgs->sum_weighted_load += weighted_cpuload(i); | |
3475 | ||
3476 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | * First idle cpu or the first cpu(busiest) in this sched group | |
3481 | * is eligible for doing load balancing at this and above | |
3482 | * domains. In the newly idle case, we will allow all the cpu's | |
3483 | * to do the newly idle load balance. | |
3484 | */ | |
3485 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3486 | balance_cpu != this_cpu && balance) { | |
3487 | *balance = 0; | |
3488 | return; | |
3489 | } | |
3490 | ||
3491 | /* Adjust by relative CPU power of the group */ | |
3492 | sgs->avg_load = sg_div_cpu_power(group, | |
3493 | sgs->group_load * SCHED_LOAD_SCALE); | |
3494 | ||
3495 | ||
3496 | /* | |
3497 | * Consider the group unbalanced when the imbalance is larger | |
3498 | * than the average weight of two tasks. | |
3499 | * | |
3500 | * APZ: with cgroup the avg task weight can vary wildly and | |
3501 | * might not be a suitable number - should we keep a | |
3502 | * normalized nr_running number somewhere that negates | |
3503 | * the hierarchy? | |
3504 | */ | |
3505 | avg_load_per_task = sg_div_cpu_power(group, | |
3506 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3507 | ||
3508 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3509 | sgs->group_imb = 1; | |
3510 | ||
3511 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3512 | ||
3513 | } | |
dd41f596 | 3514 | |
37abe198 GS |
3515 | /** |
3516 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3517 | * @sd: sched_domain whose statistics are to be updated. | |
3518 | * @this_cpu: Cpu for which load balance is currently performed. | |
3519 | * @idle: Idle status of this_cpu | |
3520 | * @sd_idle: Idle status of the sched_domain containing group. | |
3521 | * @cpus: Set of cpus considered for load balancing. | |
3522 | * @balance: Should we balance. | |
3523 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3524 | */ |
37abe198 GS |
3525 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3526 | enum cpu_idle_type idle, int *sd_idle, | |
3527 | const struct cpumask *cpus, int *balance, | |
3528 | struct sd_lb_stats *sds) | |
1da177e4 | 3529 | { |
222d656d | 3530 | struct sched_group *group = sd->groups; |
37abe198 | 3531 | struct sg_lb_stats sgs; |
222d656d GS |
3532 | int load_idx; |
3533 | ||
c071df18 | 3534 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3535 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3536 | |
3537 | do { | |
1da177e4 | 3538 | int local_group; |
1da177e4 | 3539 | |
758b2cdc RR |
3540 | local_group = cpumask_test_cpu(this_cpu, |
3541 | sched_group_cpus(group)); | |
381be78f | 3542 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3543 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3544 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3545 | |
37abe198 GS |
3546 | if (local_group && balance && !(*balance)) |
3547 | return; | |
783609c6 | 3548 | |
37abe198 GS |
3549 | sds->total_load += sgs.group_load; |
3550 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3551 | |
1da177e4 | 3552 | if (local_group) { |
37abe198 GS |
3553 | sds->this_load = sgs.avg_load; |
3554 | sds->this = group; | |
3555 | sds->this_nr_running = sgs.sum_nr_running; | |
3556 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3557 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3558 | (sgs.sum_nr_running > sgs.group_capacity || |
3559 | sgs.group_imb)) { | |
37abe198 GS |
3560 | sds->max_load = sgs.avg_load; |
3561 | sds->busiest = group; | |
3562 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3563 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3564 | sds->group_imb = sgs.group_imb; | |
1da177e4 | 3565 | } |
5c45bf27 | 3566 | |
c071df18 | 3567 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3568 | group = group->next; |
3569 | } while (group != sd->groups); | |
3570 | ||
37abe198 | 3571 | } |
2e6f44ae GS |
3572 | |
3573 | /** | |
3574 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3575 | * amongst the groups of a sched_domain, during |
3576 | * load balancing. | |
2e6f44ae GS |
3577 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3578 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3579 | * @imbalance: Variable to store the imbalance. | |
3580 | */ | |
3581 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3582 | int this_cpu, unsigned long *imbalance) | |
3583 | { | |
3584 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3585 | unsigned int imbn = 2; | |
3586 | ||
3587 | if (sds->this_nr_running) { | |
3588 | sds->this_load_per_task /= sds->this_nr_running; | |
3589 | if (sds->busiest_load_per_task > | |
3590 | sds->this_load_per_task) | |
3591 | imbn = 1; | |
3592 | } else | |
3593 | sds->this_load_per_task = | |
3594 | cpu_avg_load_per_task(this_cpu); | |
3595 | ||
3596 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= | |
3597 | sds->busiest_load_per_task * imbn) { | |
3598 | *imbalance = sds->busiest_load_per_task; | |
3599 | return; | |
3600 | } | |
3601 | ||
3602 | /* | |
3603 | * OK, we don't have enough imbalance to justify moving tasks, | |
3604 | * however we may be able to increase total CPU power used by | |
3605 | * moving them. | |
3606 | */ | |
3607 | ||
3608 | pwr_now += sds->busiest->__cpu_power * | |
3609 | min(sds->busiest_load_per_task, sds->max_load); | |
3610 | pwr_now += sds->this->__cpu_power * | |
3611 | min(sds->this_load_per_task, sds->this_load); | |
3612 | pwr_now /= SCHED_LOAD_SCALE; | |
3613 | ||
3614 | /* Amount of load we'd subtract */ | |
3615 | tmp = sg_div_cpu_power(sds->busiest, | |
3616 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3617 | if (sds->max_load > tmp) | |
3618 | pwr_move += sds->busiest->__cpu_power * | |
3619 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3620 | ||
3621 | /* Amount of load we'd add */ | |
3622 | if (sds->max_load * sds->busiest->__cpu_power < | |
3623 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3624 | tmp = sg_div_cpu_power(sds->this, | |
3625 | sds->max_load * sds->busiest->__cpu_power); | |
3626 | else | |
3627 | tmp = sg_div_cpu_power(sds->this, | |
3628 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3629 | pwr_move += sds->this->__cpu_power * | |
3630 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3631 | pwr_move /= SCHED_LOAD_SCALE; | |
3632 | ||
3633 | /* Move if we gain throughput */ | |
3634 | if (pwr_move > pwr_now) | |
3635 | *imbalance = sds->busiest_load_per_task; | |
3636 | } | |
dbc523a3 GS |
3637 | |
3638 | /** | |
3639 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3640 | * groups of a given sched_domain during load balance. | |
3641 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3642 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3643 | * @imbalance: The variable to store the imbalance. | |
3644 | */ | |
3645 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3646 | unsigned long *imbalance) | |
3647 | { | |
3648 | unsigned long max_pull; | |
3649 | /* | |
3650 | * In the presence of smp nice balancing, certain scenarios can have | |
3651 | * max load less than avg load(as we skip the groups at or below | |
3652 | * its cpu_power, while calculating max_load..) | |
3653 | */ | |
3654 | if (sds->max_load < sds->avg_load) { | |
3655 | *imbalance = 0; | |
3656 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
3657 | } | |
3658 | ||
3659 | /* Don't want to pull so many tasks that a group would go idle */ | |
3660 | max_pull = min(sds->max_load - sds->avg_load, | |
3661 | sds->max_load - sds->busiest_load_per_task); | |
3662 | ||
3663 | /* How much load to actually move to equalise the imbalance */ | |
3664 | *imbalance = min(max_pull * sds->busiest->__cpu_power, | |
3665 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
3666 | / SCHED_LOAD_SCALE; | |
3667 | ||
3668 | /* | |
3669 | * if *imbalance is less than the average load per runnable task | |
3670 | * there is no gaurantee that any tasks will be moved so we'll have | |
3671 | * a think about bumping its value to force at least one task to be | |
3672 | * moved | |
3673 | */ | |
3674 | if (*imbalance < sds->busiest_load_per_task) | |
3675 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
3676 | ||
3677 | } | |
37abe198 GS |
3678 | /******* find_busiest_group() helpers end here *********************/ |
3679 | ||
b7bb4c9b GS |
3680 | /** |
3681 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3682 | * if there is an imbalance. If there isn't an imbalance, and | |
3683 | * the user has opted for power-savings, it returns a group whose | |
3684 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3685 | * such a group exists. | |
3686 | * | |
3687 | * Also calculates the amount of weighted load which should be moved | |
3688 | * to restore balance. | |
3689 | * | |
3690 | * @sd: The sched_domain whose busiest group is to be returned. | |
3691 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3692 | * @imbalance: Variable which stores amount of weighted load which should | |
3693 | * be moved to restore balance/put a group to idle. | |
3694 | * @idle: The idle status of this_cpu. | |
3695 | * @sd_idle: The idleness of sd | |
3696 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3697 | * @balance: Pointer to a variable indicating if this_cpu | |
3698 | * is the appropriate cpu to perform load balancing at this_level. | |
3699 | * | |
3700 | * Returns: - the busiest group if imbalance exists. | |
3701 | * - If no imbalance and user has opted for power-savings balance, | |
3702 | * return the least loaded group whose CPUs can be | |
3703 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3704 | */ |
3705 | static struct sched_group * | |
3706 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3707 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3708 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3709 | { | |
3710 | struct sd_lb_stats sds; | |
37abe198 GS |
3711 | |
3712 | memset(&sds, 0, sizeof(sds)); | |
3713 | ||
3714 | /* | |
3715 | * Compute the various statistics relavent for load balancing at | |
3716 | * this level. | |
3717 | */ | |
3718 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3719 | balance, &sds); | |
3720 | ||
b7bb4c9b GS |
3721 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3722 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3723 | * at this level. | |
3724 | * 2) There is no busy sibling group to pull from. | |
3725 | * 3) This group is the busiest group. | |
3726 | * 4) This group is more busy than the avg busieness at this | |
3727 | * sched_domain. | |
3728 | * 5) The imbalance is within the specified limit. | |
3729 | * 6) Any rebalance would lead to ping-pong | |
3730 | */ | |
37abe198 GS |
3731 | if (balance && !(*balance)) |
3732 | goto ret; | |
3733 | ||
b7bb4c9b GS |
3734 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3735 | goto out_balanced; | |
3736 | ||
3737 | if (sds.this_load >= sds.max_load) | |
1da177e4 LT |
3738 | goto out_balanced; |
3739 | ||
222d656d | 3740 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3741 | |
b7bb4c9b GS |
3742 | if (sds.this_load >= sds.avg_load) |
3743 | goto out_balanced; | |
3744 | ||
3745 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3746 | goto out_balanced; |
3747 | ||
222d656d GS |
3748 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3749 | if (sds.group_imb) | |
3750 | sds.busiest_load_per_task = | |
3751 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3752 | |
1da177e4 LT |
3753 | /* |
3754 | * We're trying to get all the cpus to the average_load, so we don't | |
3755 | * want to push ourselves above the average load, nor do we wish to | |
3756 | * reduce the max loaded cpu below the average load, as either of these | |
3757 | * actions would just result in more rebalancing later, and ping-pong | |
3758 | * tasks around. Thus we look for the minimum possible imbalance. | |
3759 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3760 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3761 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3762 | * appear as very large values with unsigned longs. |
3763 | */ | |
222d656d | 3764 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3765 | goto out_balanced; |
3766 | ||
dbc523a3 GS |
3767 | /* Looks like there is an imbalance. Compute it */ |
3768 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3769 | return sds.busiest; |
1da177e4 LT |
3770 | |
3771 | out_balanced: | |
c071df18 GS |
3772 | /* |
3773 | * There is no obvious imbalance. But check if we can do some balancing | |
3774 | * to save power. | |
3775 | */ | |
3776 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3777 | return sds.busiest; | |
783609c6 | 3778 | ret: |
1da177e4 LT |
3779 | *imbalance = 0; |
3780 | return NULL; | |
3781 | } | |
3782 | ||
3783 | /* | |
3784 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3785 | */ | |
70b97a7f | 3786 | static struct rq * |
d15bcfdb | 3787 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3788 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3789 | { |
70b97a7f | 3790 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3791 | unsigned long max_load = 0; |
1da177e4 LT |
3792 | int i; |
3793 | ||
758b2cdc | 3794 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3795 | unsigned long wl; |
0a2966b4 | 3796 | |
96f874e2 | 3797 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3798 | continue; |
3799 | ||
48f24c4d | 3800 | rq = cpu_rq(i); |
dd41f596 | 3801 | wl = weighted_cpuload(i); |
2dd73a4f | 3802 | |
dd41f596 | 3803 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3804 | continue; |
1da177e4 | 3805 | |
dd41f596 IM |
3806 | if (wl > max_load) { |
3807 | max_load = wl; | |
48f24c4d | 3808 | busiest = rq; |
1da177e4 LT |
3809 | } |
3810 | } | |
3811 | ||
3812 | return busiest; | |
3813 | } | |
3814 | ||
77391d71 NP |
3815 | /* |
3816 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3817 | * so long as it is large enough. | |
3818 | */ | |
3819 | #define MAX_PINNED_INTERVAL 512 | |
3820 | ||
1da177e4 LT |
3821 | /* |
3822 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3823 | * tasks if there is an imbalance. | |
1da177e4 | 3824 | */ |
70b97a7f | 3825 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3826 | struct sched_domain *sd, enum cpu_idle_type idle, |
96f874e2 | 3827 | int *balance, struct cpumask *cpus) |
1da177e4 | 3828 | { |
43010659 | 3829 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3830 | struct sched_group *group; |
1da177e4 | 3831 | unsigned long imbalance; |
70b97a7f | 3832 | struct rq *busiest; |
fe2eea3f | 3833 | unsigned long flags; |
5969fe06 | 3834 | |
96f874e2 | 3835 | cpumask_setall(cpus); |
7c16ec58 | 3836 | |
89c4710e SS |
3837 | /* |
3838 | * When power savings policy is enabled for the parent domain, idle | |
3839 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3840 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3841 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3842 | */ |
d15bcfdb | 3843 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3844 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3845 | sd_idle = 1; |
1da177e4 | 3846 | |
2d72376b | 3847 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3848 | |
0a2966b4 | 3849 | redo: |
c8cba857 | 3850 | update_shares(sd); |
0a2966b4 | 3851 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3852 | cpus, balance); |
783609c6 | 3853 | |
06066714 | 3854 | if (*balance == 0) |
783609c6 | 3855 | goto out_balanced; |
783609c6 | 3856 | |
1da177e4 LT |
3857 | if (!group) { |
3858 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3859 | goto out_balanced; | |
3860 | } | |
3861 | ||
7c16ec58 | 3862 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3863 | if (!busiest) { |
3864 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3865 | goto out_balanced; | |
3866 | } | |
3867 | ||
db935dbd | 3868 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3869 | |
3870 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3871 | ||
43010659 | 3872 | ld_moved = 0; |
1da177e4 LT |
3873 | if (busiest->nr_running > 1) { |
3874 | /* | |
3875 | * Attempt to move tasks. If find_busiest_group has found | |
3876 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3877 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3878 | * correctly treated as an imbalance. |
3879 | */ | |
fe2eea3f | 3880 | local_irq_save(flags); |
e17224bf | 3881 | double_rq_lock(this_rq, busiest); |
43010659 | 3882 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3883 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3884 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3885 | local_irq_restore(flags); |
81026794 | 3886 | |
46cb4b7c SS |
3887 | /* |
3888 | * some other cpu did the load balance for us. | |
3889 | */ | |
43010659 | 3890 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3891 | resched_cpu(this_cpu); |
3892 | ||
81026794 | 3893 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3894 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3895 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3896 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3897 | goto redo; |
81026794 | 3898 | goto out_balanced; |
0a2966b4 | 3899 | } |
1da177e4 | 3900 | } |
81026794 | 3901 | |
43010659 | 3902 | if (!ld_moved) { |
1da177e4 LT |
3903 | schedstat_inc(sd, lb_failed[idle]); |
3904 | sd->nr_balance_failed++; | |
3905 | ||
3906 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3907 | |
fe2eea3f | 3908 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3909 | |
3910 | /* don't kick the migration_thread, if the curr | |
3911 | * task on busiest cpu can't be moved to this_cpu | |
3912 | */ | |
96f874e2 RR |
3913 | if (!cpumask_test_cpu(this_cpu, |
3914 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3915 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3916 | all_pinned = 1; |
3917 | goto out_one_pinned; | |
3918 | } | |
3919 | ||
1da177e4 LT |
3920 | if (!busiest->active_balance) { |
3921 | busiest->active_balance = 1; | |
3922 | busiest->push_cpu = this_cpu; | |
81026794 | 3923 | active_balance = 1; |
1da177e4 | 3924 | } |
fe2eea3f | 3925 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3926 | if (active_balance) |
1da177e4 LT |
3927 | wake_up_process(busiest->migration_thread); |
3928 | ||
3929 | /* | |
3930 | * We've kicked active balancing, reset the failure | |
3931 | * counter. | |
3932 | */ | |
39507451 | 3933 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3934 | } |
81026794 | 3935 | } else |
1da177e4 LT |
3936 | sd->nr_balance_failed = 0; |
3937 | ||
81026794 | 3938 | if (likely(!active_balance)) { |
1da177e4 LT |
3939 | /* We were unbalanced, so reset the balancing interval */ |
3940 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3941 | } else { |
3942 | /* | |
3943 | * If we've begun active balancing, start to back off. This | |
3944 | * case may not be covered by the all_pinned logic if there | |
3945 | * is only 1 task on the busy runqueue (because we don't call | |
3946 | * move_tasks). | |
3947 | */ | |
3948 | if (sd->balance_interval < sd->max_interval) | |
3949 | sd->balance_interval *= 2; | |
1da177e4 LT |
3950 | } |
3951 | ||
43010659 | 3952 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3953 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3954 | ld_moved = -1; |
3955 | ||
3956 | goto out; | |
1da177e4 LT |
3957 | |
3958 | out_balanced: | |
1da177e4 LT |
3959 | schedstat_inc(sd, lb_balanced[idle]); |
3960 | ||
16cfb1c0 | 3961 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3962 | |
3963 | out_one_pinned: | |
1da177e4 | 3964 | /* tune up the balancing interval */ |
77391d71 NP |
3965 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3966 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3967 | sd->balance_interval *= 2; |
3968 | ||
48f24c4d | 3969 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3970 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3971 | ld_moved = -1; |
3972 | else | |
3973 | ld_moved = 0; | |
3974 | out: | |
c8cba857 PZ |
3975 | if (ld_moved) |
3976 | update_shares(sd); | |
c09595f6 | 3977 | return ld_moved; |
1da177e4 LT |
3978 | } |
3979 | ||
3980 | /* | |
3981 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3982 | * tasks if there is an imbalance. | |
3983 | * | |
d15bcfdb | 3984 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3985 | * this_rq is locked. |
3986 | */ | |
48f24c4d | 3987 | static int |
7c16ec58 | 3988 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
96f874e2 | 3989 | struct cpumask *cpus) |
1da177e4 LT |
3990 | { |
3991 | struct sched_group *group; | |
70b97a7f | 3992 | struct rq *busiest = NULL; |
1da177e4 | 3993 | unsigned long imbalance; |
43010659 | 3994 | int ld_moved = 0; |
5969fe06 | 3995 | int sd_idle = 0; |
969bb4e4 | 3996 | int all_pinned = 0; |
7c16ec58 | 3997 | |
96f874e2 | 3998 | cpumask_setall(cpus); |
5969fe06 | 3999 | |
89c4710e SS |
4000 | /* |
4001 | * When power savings policy is enabled for the parent domain, idle | |
4002 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4003 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4004 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4005 | */ |
4006 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4007 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4008 | sd_idle = 1; |
1da177e4 | 4009 | |
2d72376b | 4010 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4011 | redo: |
3e5459b4 | 4012 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4013 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4014 | &sd_idle, cpus, NULL); |
1da177e4 | 4015 | if (!group) { |
d15bcfdb | 4016 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4017 | goto out_balanced; |
1da177e4 LT |
4018 | } |
4019 | ||
7c16ec58 | 4020 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4021 | if (!busiest) { |
d15bcfdb | 4022 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4023 | goto out_balanced; |
1da177e4 LT |
4024 | } |
4025 | ||
db935dbd NP |
4026 | BUG_ON(busiest == this_rq); |
4027 | ||
d15bcfdb | 4028 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4029 | |
43010659 | 4030 | ld_moved = 0; |
d6d5cfaf NP |
4031 | if (busiest->nr_running > 1) { |
4032 | /* Attempt to move tasks */ | |
4033 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4034 | /* this_rq->clock is already updated */ |
4035 | update_rq_clock(busiest); | |
43010659 | 4036 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4037 | imbalance, sd, CPU_NEWLY_IDLE, |
4038 | &all_pinned); | |
1b12bbc7 | 4039 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4040 | |
969bb4e4 | 4041 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4042 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4043 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4044 | goto redo; |
4045 | } | |
d6d5cfaf NP |
4046 | } |
4047 | ||
43010659 | 4048 | if (!ld_moved) { |
36dffab6 | 4049 | int active_balance = 0; |
ad273b32 | 4050 | |
d15bcfdb | 4051 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4052 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4053 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4054 | return -1; |
ad273b32 VS |
4055 | |
4056 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4057 | return -1; | |
4058 | ||
4059 | if (sd->nr_balance_failed++ < 2) | |
4060 | return -1; | |
4061 | ||
4062 | /* | |
4063 | * The only task running in a non-idle cpu can be moved to this | |
4064 | * cpu in an attempt to completely freeup the other CPU | |
4065 | * package. The same method used to move task in load_balance() | |
4066 | * have been extended for load_balance_newidle() to speedup | |
4067 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4068 | * | |
4069 | * The package power saving logic comes from | |
4070 | * find_busiest_group(). If there are no imbalance, then | |
4071 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4072 | * f_b_g() will select a group from which a running task may be | |
4073 | * pulled to this cpu in order to make the other package idle. | |
4074 | * If there is no opportunity to make a package idle and if | |
4075 | * there are no imbalance, then f_b_g() will return NULL and no | |
4076 | * action will be taken in load_balance_newidle(). | |
4077 | * | |
4078 | * Under normal task pull operation due to imbalance, there | |
4079 | * will be more than one task in the source run queue and | |
4080 | * move_tasks() will succeed. ld_moved will be true and this | |
4081 | * active balance code will not be triggered. | |
4082 | */ | |
4083 | ||
4084 | /* Lock busiest in correct order while this_rq is held */ | |
4085 | double_lock_balance(this_rq, busiest); | |
4086 | ||
4087 | /* | |
4088 | * don't kick the migration_thread, if the curr | |
4089 | * task on busiest cpu can't be moved to this_cpu | |
4090 | */ | |
6ca09dfc | 4091 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4092 | double_unlock_balance(this_rq, busiest); |
4093 | all_pinned = 1; | |
4094 | return ld_moved; | |
4095 | } | |
4096 | ||
4097 | if (!busiest->active_balance) { | |
4098 | busiest->active_balance = 1; | |
4099 | busiest->push_cpu = this_cpu; | |
4100 | active_balance = 1; | |
4101 | } | |
4102 | ||
4103 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4104 | /* |
4105 | * Should not call ttwu while holding a rq->lock | |
4106 | */ | |
4107 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4108 | if (active_balance) |
4109 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4110 | spin_lock(&this_rq->lock); |
ad273b32 | 4111 | |
5969fe06 | 4112 | } else |
16cfb1c0 | 4113 | sd->nr_balance_failed = 0; |
1da177e4 | 4114 | |
3e5459b4 | 4115 | update_shares_locked(this_rq, sd); |
43010659 | 4116 | return ld_moved; |
16cfb1c0 NP |
4117 | |
4118 | out_balanced: | |
d15bcfdb | 4119 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4120 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4121 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4122 | return -1; |
16cfb1c0 | 4123 | sd->nr_balance_failed = 0; |
48f24c4d | 4124 | |
16cfb1c0 | 4125 | return 0; |
1da177e4 LT |
4126 | } |
4127 | ||
4128 | /* | |
4129 | * idle_balance is called by schedule() if this_cpu is about to become | |
4130 | * idle. Attempts to pull tasks from other CPUs. | |
4131 | */ | |
70b97a7f | 4132 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4133 | { |
4134 | struct sched_domain *sd; | |
efbe027e | 4135 | int pulled_task = 0; |
dd41f596 | 4136 | unsigned long next_balance = jiffies + HZ; |
4d2732c6 RR |
4137 | cpumask_var_t tmpmask; |
4138 | ||
4139 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | |
4140 | return; | |
1da177e4 LT |
4141 | |
4142 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4143 | unsigned long interval; |
4144 | ||
4145 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4146 | continue; | |
4147 | ||
4148 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4149 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4150 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4d2732c6 | 4151 | sd, tmpmask); |
92c4ca5c CL |
4152 | |
4153 | interval = msecs_to_jiffies(sd->balance_interval); | |
4154 | if (time_after(next_balance, sd->last_balance + interval)) | |
4155 | next_balance = sd->last_balance + interval; | |
4156 | if (pulled_task) | |
4157 | break; | |
1da177e4 | 4158 | } |
dd41f596 | 4159 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4160 | /* |
4161 | * We are going idle. next_balance may be set based on | |
4162 | * a busy processor. So reset next_balance. | |
4163 | */ | |
4164 | this_rq->next_balance = next_balance; | |
dd41f596 | 4165 | } |
4d2732c6 | 4166 | free_cpumask_var(tmpmask); |
1da177e4 LT |
4167 | } |
4168 | ||
4169 | /* | |
4170 | * active_load_balance is run by migration threads. It pushes running tasks | |
4171 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4172 | * running on each physical CPU where possible, and avoids physical / | |
4173 | * logical imbalances. | |
4174 | * | |
4175 | * Called with busiest_rq locked. | |
4176 | */ | |
70b97a7f | 4177 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4178 | { |
39507451 | 4179 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4180 | struct sched_domain *sd; |
4181 | struct rq *target_rq; | |
39507451 | 4182 | |
48f24c4d | 4183 | /* Is there any task to move? */ |
39507451 | 4184 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4185 | return; |
4186 | ||
4187 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4188 | |
4189 | /* | |
39507451 | 4190 | * This condition is "impossible", if it occurs |
41a2d6cf | 4191 | * we need to fix it. Originally reported by |
39507451 | 4192 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4193 | */ |
39507451 | 4194 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4195 | |
39507451 NP |
4196 | /* move a task from busiest_rq to target_rq */ |
4197 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4198 | update_rq_clock(busiest_rq); |
4199 | update_rq_clock(target_rq); | |
39507451 NP |
4200 | |
4201 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4202 | for_each_domain(target_cpu, sd) { |
39507451 | 4203 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4204 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4205 | break; |
c96d145e | 4206 | } |
39507451 | 4207 | |
48f24c4d | 4208 | if (likely(sd)) { |
2d72376b | 4209 | schedstat_inc(sd, alb_count); |
39507451 | 4210 | |
43010659 PW |
4211 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4212 | sd, CPU_IDLE)) | |
48f24c4d IM |
4213 | schedstat_inc(sd, alb_pushed); |
4214 | else | |
4215 | schedstat_inc(sd, alb_failed); | |
4216 | } | |
1b12bbc7 | 4217 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4218 | } |
4219 | ||
46cb4b7c SS |
4220 | #ifdef CONFIG_NO_HZ |
4221 | static struct { | |
4222 | atomic_t load_balancer; | |
7d1e6a9b | 4223 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
4224 | } nohz ____cacheline_aligned = { |
4225 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4226 | }; |
4227 | ||
7835b98b | 4228 | /* |
46cb4b7c SS |
4229 | * This routine will try to nominate the ilb (idle load balancing) |
4230 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4231 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4232 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4233 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4234 | * arrives... | |
4235 | * | |
4236 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4237 | * for idle load balancing. ilb owner will still be part of | |
4238 | * nohz.cpu_mask.. | |
7835b98b | 4239 | * |
46cb4b7c SS |
4240 | * While stopping the tick, this cpu will become the ilb owner if there |
4241 | * is no other owner. And will be the owner till that cpu becomes busy | |
4242 | * or if all cpus in the system stop their ticks at which point | |
4243 | * there is no need for ilb owner. | |
4244 | * | |
4245 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4246 | * next busy scheduler_tick() | |
4247 | */ | |
4248 | int select_nohz_load_balancer(int stop_tick) | |
4249 | { | |
4250 | int cpu = smp_processor_id(); | |
4251 | ||
4252 | if (stop_tick) { | |
46cb4b7c SS |
4253 | cpu_rq(cpu)->in_nohz_recently = 1; |
4254 | ||
483b4ee6 SS |
4255 | if (!cpu_active(cpu)) { |
4256 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4257 | return 0; | |
4258 | ||
4259 | /* | |
4260 | * If we are going offline and still the leader, | |
4261 | * give up! | |
4262 | */ | |
46cb4b7c SS |
4263 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4264 | BUG(); | |
483b4ee6 | 4265 | |
46cb4b7c SS |
4266 | return 0; |
4267 | } | |
4268 | ||
483b4ee6 SS |
4269 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4270 | ||
46cb4b7c | 4271 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4272 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4273 | if (atomic_read(&nohz.load_balancer) == cpu) |
4274 | atomic_set(&nohz.load_balancer, -1); | |
4275 | return 0; | |
4276 | } | |
4277 | ||
4278 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4279 | /* make me the ilb owner */ | |
4280 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4281 | return 1; | |
4282 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
4283 | return 1; | |
4284 | } else { | |
7d1e6a9b | 4285 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4286 | return 0; |
4287 | ||
7d1e6a9b | 4288 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4289 | |
4290 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4291 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4292 | BUG(); | |
4293 | } | |
4294 | return 0; | |
4295 | } | |
4296 | #endif | |
4297 | ||
4298 | static DEFINE_SPINLOCK(balancing); | |
4299 | ||
4300 | /* | |
7835b98b CL |
4301 | * It checks each scheduling domain to see if it is due to be balanced, |
4302 | * and initiates a balancing operation if so. | |
4303 | * | |
4304 | * Balancing parameters are set up in arch_init_sched_domains. | |
4305 | */ | |
a9957449 | 4306 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4307 | { |
46cb4b7c SS |
4308 | int balance = 1; |
4309 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4310 | unsigned long interval; |
4311 | struct sched_domain *sd; | |
46cb4b7c | 4312 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4313 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4314 | int update_next_balance = 0; |
d07355f5 | 4315 | int need_serialize; |
a0e90245 RR |
4316 | cpumask_var_t tmp; |
4317 | ||
4318 | /* Fails alloc? Rebalancing probably not a priority right now. */ | |
4319 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | |
4320 | return; | |
1da177e4 | 4321 | |
46cb4b7c | 4322 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4323 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4324 | continue; | |
4325 | ||
4326 | interval = sd->balance_interval; | |
d15bcfdb | 4327 | if (idle != CPU_IDLE) |
1da177e4 LT |
4328 | interval *= sd->busy_factor; |
4329 | ||
4330 | /* scale ms to jiffies */ | |
4331 | interval = msecs_to_jiffies(interval); | |
4332 | if (unlikely(!interval)) | |
4333 | interval = 1; | |
dd41f596 IM |
4334 | if (interval > HZ*NR_CPUS/10) |
4335 | interval = HZ*NR_CPUS/10; | |
4336 | ||
d07355f5 | 4337 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4338 | |
d07355f5 | 4339 | if (need_serialize) { |
08c183f3 CL |
4340 | if (!spin_trylock(&balancing)) |
4341 | goto out; | |
4342 | } | |
4343 | ||
c9819f45 | 4344 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
a0e90245 | 4345 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { |
fa3b6ddc SS |
4346 | /* |
4347 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4348 | * longer idle, or one of our SMT siblings is |
4349 | * not idle. | |
4350 | */ | |
d15bcfdb | 4351 | idle = CPU_NOT_IDLE; |
1da177e4 | 4352 | } |
1bd77f2d | 4353 | sd->last_balance = jiffies; |
1da177e4 | 4354 | } |
d07355f5 | 4355 | if (need_serialize) |
08c183f3 CL |
4356 | spin_unlock(&balancing); |
4357 | out: | |
f549da84 | 4358 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4359 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4360 | update_next_balance = 1; |
4361 | } | |
783609c6 SS |
4362 | |
4363 | /* | |
4364 | * Stop the load balance at this level. There is another | |
4365 | * CPU in our sched group which is doing load balancing more | |
4366 | * actively. | |
4367 | */ | |
4368 | if (!balance) | |
4369 | break; | |
1da177e4 | 4370 | } |
f549da84 SS |
4371 | |
4372 | /* | |
4373 | * next_balance will be updated only when there is a need. | |
4374 | * When the cpu is attached to null domain for ex, it will not be | |
4375 | * updated. | |
4376 | */ | |
4377 | if (likely(update_next_balance)) | |
4378 | rq->next_balance = next_balance; | |
a0e90245 RR |
4379 | |
4380 | free_cpumask_var(tmp); | |
46cb4b7c SS |
4381 | } |
4382 | ||
4383 | /* | |
4384 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4385 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4386 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4387 | */ | |
4388 | static void run_rebalance_domains(struct softirq_action *h) | |
4389 | { | |
dd41f596 IM |
4390 | int this_cpu = smp_processor_id(); |
4391 | struct rq *this_rq = cpu_rq(this_cpu); | |
4392 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4393 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4394 | |
dd41f596 | 4395 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4396 | |
4397 | #ifdef CONFIG_NO_HZ | |
4398 | /* | |
4399 | * If this cpu is the owner for idle load balancing, then do the | |
4400 | * balancing on behalf of the other idle cpus whose ticks are | |
4401 | * stopped. | |
4402 | */ | |
dd41f596 IM |
4403 | if (this_rq->idle_at_tick && |
4404 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4405 | struct rq *rq; |
4406 | int balance_cpu; | |
4407 | ||
7d1e6a9b RR |
4408 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4409 | if (balance_cpu == this_cpu) | |
4410 | continue; | |
4411 | ||
46cb4b7c SS |
4412 | /* |
4413 | * If this cpu gets work to do, stop the load balancing | |
4414 | * work being done for other cpus. Next load | |
4415 | * balancing owner will pick it up. | |
4416 | */ | |
4417 | if (need_resched()) | |
4418 | break; | |
4419 | ||
de0cf899 | 4420 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4421 | |
4422 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4423 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4424 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4425 | } |
4426 | } | |
4427 | #endif | |
4428 | } | |
4429 | ||
8a0be9ef FW |
4430 | static inline int on_null_domain(int cpu) |
4431 | { | |
4432 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4433 | } | |
4434 | ||
46cb4b7c SS |
4435 | /* |
4436 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4437 | * | |
4438 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4439 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4440 | * if the whole system is idle. | |
4441 | */ | |
dd41f596 | 4442 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4443 | { |
46cb4b7c SS |
4444 | #ifdef CONFIG_NO_HZ |
4445 | /* | |
4446 | * If we were in the nohz mode recently and busy at the current | |
4447 | * scheduler tick, then check if we need to nominate new idle | |
4448 | * load balancer. | |
4449 | */ | |
4450 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4451 | rq->in_nohz_recently = 0; | |
4452 | ||
4453 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4454 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4455 | atomic_set(&nohz.load_balancer, -1); |
4456 | } | |
4457 | ||
4458 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4459 | /* | |
4460 | * simple selection for now: Nominate the | |
4461 | * first cpu in the nohz list to be the next | |
4462 | * ilb owner. | |
4463 | * | |
4464 | * TBD: Traverse the sched domains and nominate | |
4465 | * the nearest cpu in the nohz.cpu_mask. | |
4466 | */ | |
7d1e6a9b | 4467 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4468 | |
434d53b0 | 4469 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4470 | resched_cpu(ilb); |
4471 | } | |
4472 | } | |
4473 | ||
4474 | /* | |
4475 | * If this cpu is idle and doing idle load balancing for all the | |
4476 | * cpus with ticks stopped, is it time for that to stop? | |
4477 | */ | |
4478 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4479 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4480 | resched_cpu(cpu); |
4481 | return; | |
4482 | } | |
4483 | ||
4484 | /* | |
4485 | * If this cpu is idle and the idle load balancing is done by | |
4486 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4487 | */ | |
4488 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4489 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4490 | return; |
4491 | #endif | |
8a0be9ef FW |
4492 | /* Don't need to rebalance while attached to NULL domain */ |
4493 | if (time_after_eq(jiffies, rq->next_balance) && | |
4494 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4495 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4496 | } |
dd41f596 IM |
4497 | |
4498 | #else /* CONFIG_SMP */ | |
4499 | ||
1da177e4 LT |
4500 | /* |
4501 | * on UP we do not need to balance between CPUs: | |
4502 | */ | |
70b97a7f | 4503 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4504 | { |
4505 | } | |
dd41f596 | 4506 | |
1da177e4 LT |
4507 | #endif |
4508 | ||
1da177e4 LT |
4509 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4510 | ||
4511 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4512 | ||
4513 | /* | |
f06febc9 FM |
4514 | * Return any ns on the sched_clock that have not yet been banked in |
4515 | * @p in case that task is currently running. | |
1da177e4 | 4516 | */ |
bb34d92f | 4517 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4518 | { |
1da177e4 | 4519 | unsigned long flags; |
41b86e9c | 4520 | struct rq *rq; |
bb34d92f | 4521 | u64 ns = 0; |
48f24c4d | 4522 | |
41b86e9c | 4523 | rq = task_rq_lock(p, &flags); |
1508487e | 4524 | |
051a1d1a | 4525 | if (task_current(rq, p)) { |
f06febc9 FM |
4526 | u64 delta_exec; |
4527 | ||
a8e504d2 IM |
4528 | update_rq_clock(rq); |
4529 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4530 | if ((s64)delta_exec > 0) |
bb34d92f | 4531 | ns = delta_exec; |
41b86e9c | 4532 | } |
48f24c4d | 4533 | |
41b86e9c | 4534 | task_rq_unlock(rq, &flags); |
48f24c4d | 4535 | |
1da177e4 LT |
4536 | return ns; |
4537 | } | |
4538 | ||
1da177e4 LT |
4539 | /* |
4540 | * Account user cpu time to a process. | |
4541 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4542 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4543 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4544 | */ |
457533a7 MS |
4545 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4546 | cputime_t cputime_scaled) | |
1da177e4 LT |
4547 | { |
4548 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4549 | cputime64_t tmp; | |
4550 | ||
457533a7 | 4551 | /* Add user time to process. */ |
1da177e4 | 4552 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4553 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4554 | account_group_user_time(p, cputime); |
1da177e4 LT |
4555 | |
4556 | /* Add user time to cpustat. */ | |
4557 | tmp = cputime_to_cputime64(cputime); | |
4558 | if (TASK_NICE(p) > 0) | |
4559 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4560 | else | |
4561 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4562 | /* Account for user time used */ |
4563 | acct_update_integrals(p); | |
1da177e4 LT |
4564 | } |
4565 | ||
94886b84 LV |
4566 | /* |
4567 | * Account guest cpu time to a process. | |
4568 | * @p: the process that the cpu time gets accounted to | |
4569 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4570 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4571 | */ |
457533a7 MS |
4572 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4573 | cputime_t cputime_scaled) | |
94886b84 LV |
4574 | { |
4575 | cputime64_t tmp; | |
4576 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4577 | ||
4578 | tmp = cputime_to_cputime64(cputime); | |
4579 | ||
457533a7 | 4580 | /* Add guest time to process. */ |
94886b84 | 4581 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4582 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4583 | account_group_user_time(p, cputime); |
94886b84 LV |
4584 | p->gtime = cputime_add(p->gtime, cputime); |
4585 | ||
457533a7 | 4586 | /* Add guest time to cpustat. */ |
94886b84 LV |
4587 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4588 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4589 | } | |
4590 | ||
1da177e4 LT |
4591 | /* |
4592 | * Account system cpu time to a process. | |
4593 | * @p: the process that the cpu time gets accounted to | |
4594 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4595 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4596 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4597 | */ |
4598 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4599 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4600 | { |
4601 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4602 | cputime64_t tmp; |
4603 | ||
983ed7a6 | 4604 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4605 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4606 | return; |
4607 | } | |
94886b84 | 4608 | |
457533a7 | 4609 | /* Add system time to process. */ |
1da177e4 | 4610 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4611 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4612 | account_group_system_time(p, cputime); |
1da177e4 LT |
4613 | |
4614 | /* Add system time to cpustat. */ | |
4615 | tmp = cputime_to_cputime64(cputime); | |
4616 | if (hardirq_count() - hardirq_offset) | |
4617 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4618 | else if (softirq_count()) | |
4619 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4620 | else |
79741dd3 MS |
4621 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4622 | ||
1da177e4 LT |
4623 | /* Account for system time used */ |
4624 | acct_update_integrals(p); | |
1da177e4 LT |
4625 | } |
4626 | ||
c66f08be | 4627 | /* |
1da177e4 | 4628 | * Account for involuntary wait time. |
1da177e4 | 4629 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4630 | */ |
79741dd3 | 4631 | void account_steal_time(cputime_t cputime) |
c66f08be | 4632 | { |
79741dd3 MS |
4633 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4634 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4635 | ||
4636 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4637 | } |
4638 | ||
1da177e4 | 4639 | /* |
79741dd3 MS |
4640 | * Account for idle time. |
4641 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4642 | */ |
79741dd3 | 4643 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4644 | { |
4645 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4646 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4647 | struct rq *rq = this_rq(); |
1da177e4 | 4648 | |
79741dd3 MS |
4649 | if (atomic_read(&rq->nr_iowait) > 0) |
4650 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4651 | else | |
4652 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4653 | } |
4654 | ||
79741dd3 MS |
4655 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4656 | ||
4657 | /* | |
4658 | * Account a single tick of cpu time. | |
4659 | * @p: the process that the cpu time gets accounted to | |
4660 | * @user_tick: indicates if the tick is a user or a system tick | |
4661 | */ | |
4662 | void account_process_tick(struct task_struct *p, int user_tick) | |
4663 | { | |
4664 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4665 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4666 | struct rq *rq = this_rq(); | |
4667 | ||
4668 | if (user_tick) | |
4669 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
4670 | else if (p != rq->idle) | |
4671 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, | |
4672 | one_jiffy_scaled); | |
4673 | else | |
4674 | account_idle_time(one_jiffy); | |
4675 | } | |
4676 | ||
4677 | /* | |
4678 | * Account multiple ticks of steal time. | |
4679 | * @p: the process from which the cpu time has been stolen | |
4680 | * @ticks: number of stolen ticks | |
4681 | */ | |
4682 | void account_steal_ticks(unsigned long ticks) | |
4683 | { | |
4684 | account_steal_time(jiffies_to_cputime(ticks)); | |
4685 | } | |
4686 | ||
4687 | /* | |
4688 | * Account multiple ticks of idle time. | |
4689 | * @ticks: number of stolen ticks | |
4690 | */ | |
4691 | void account_idle_ticks(unsigned long ticks) | |
4692 | { | |
4693 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4694 | } |
4695 | ||
79741dd3 MS |
4696 | #endif |
4697 | ||
49048622 BS |
4698 | /* |
4699 | * Use precise platform statistics if available: | |
4700 | */ | |
4701 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4702 | cputime_t task_utime(struct task_struct *p) | |
4703 | { | |
4704 | return p->utime; | |
4705 | } | |
4706 | ||
4707 | cputime_t task_stime(struct task_struct *p) | |
4708 | { | |
4709 | return p->stime; | |
4710 | } | |
4711 | #else | |
4712 | cputime_t task_utime(struct task_struct *p) | |
4713 | { | |
4714 | clock_t utime = cputime_to_clock_t(p->utime), | |
4715 | total = utime + cputime_to_clock_t(p->stime); | |
4716 | u64 temp; | |
4717 | ||
4718 | /* | |
4719 | * Use CFS's precise accounting: | |
4720 | */ | |
4721 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4722 | ||
4723 | if (total) { | |
4724 | temp *= utime; | |
4725 | do_div(temp, total); | |
4726 | } | |
4727 | utime = (clock_t)temp; | |
4728 | ||
4729 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4730 | return p->prev_utime; | |
4731 | } | |
4732 | ||
4733 | cputime_t task_stime(struct task_struct *p) | |
4734 | { | |
4735 | clock_t stime; | |
4736 | ||
4737 | /* | |
4738 | * Use CFS's precise accounting. (we subtract utime from | |
4739 | * the total, to make sure the total observed by userspace | |
4740 | * grows monotonically - apps rely on that): | |
4741 | */ | |
4742 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4743 | cputime_to_clock_t(task_utime(p)); | |
4744 | ||
4745 | if (stime >= 0) | |
4746 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4747 | ||
4748 | return p->prev_stime; | |
4749 | } | |
4750 | #endif | |
4751 | ||
4752 | inline cputime_t task_gtime(struct task_struct *p) | |
4753 | { | |
4754 | return p->gtime; | |
4755 | } | |
4756 | ||
7835b98b CL |
4757 | /* |
4758 | * This function gets called by the timer code, with HZ frequency. | |
4759 | * We call it with interrupts disabled. | |
4760 | * | |
4761 | * It also gets called by the fork code, when changing the parent's | |
4762 | * timeslices. | |
4763 | */ | |
4764 | void scheduler_tick(void) | |
4765 | { | |
7835b98b CL |
4766 | int cpu = smp_processor_id(); |
4767 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4768 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4769 | |
4770 | sched_clock_tick(); | |
dd41f596 IM |
4771 | |
4772 | spin_lock(&rq->lock); | |
3e51f33f | 4773 | update_rq_clock(rq); |
f1a438d8 | 4774 | update_cpu_load(rq); |
fa85ae24 | 4775 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4776 | spin_unlock(&rq->lock); |
7835b98b | 4777 | |
e418e1c2 | 4778 | #ifdef CONFIG_SMP |
dd41f596 IM |
4779 | rq->idle_at_tick = idle_cpu(cpu); |
4780 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4781 | #endif |
1da177e4 LT |
4782 | } |
4783 | ||
6cd8a4bb SR |
4784 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4785 | defined(CONFIG_PREEMPT_TRACER)) | |
4786 | ||
4787 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4788 | { | |
4789 | if (in_lock_functions(addr)) { | |
4790 | addr = CALLER_ADDR2; | |
4791 | if (in_lock_functions(addr)) | |
4792 | addr = CALLER_ADDR3; | |
4793 | } | |
4794 | return addr; | |
4795 | } | |
1da177e4 | 4796 | |
43627582 | 4797 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4798 | { |
6cd8a4bb | 4799 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4800 | /* |
4801 | * Underflow? | |
4802 | */ | |
9a11b49a IM |
4803 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4804 | return; | |
6cd8a4bb | 4805 | #endif |
1da177e4 | 4806 | preempt_count() += val; |
6cd8a4bb | 4807 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4808 | /* |
4809 | * Spinlock count overflowing soon? | |
4810 | */ | |
33859f7f MOS |
4811 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4812 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4813 | #endif |
4814 | if (preempt_count() == val) | |
4815 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4816 | } |
4817 | EXPORT_SYMBOL(add_preempt_count); | |
4818 | ||
43627582 | 4819 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4820 | { |
6cd8a4bb | 4821 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4822 | /* |
4823 | * Underflow? | |
4824 | */ | |
01e3eb82 | 4825 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4826 | return; |
1da177e4 LT |
4827 | /* |
4828 | * Is the spinlock portion underflowing? | |
4829 | */ | |
9a11b49a IM |
4830 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4831 | !(preempt_count() & PREEMPT_MASK))) | |
4832 | return; | |
6cd8a4bb | 4833 | #endif |
9a11b49a | 4834 | |
6cd8a4bb SR |
4835 | if (preempt_count() == val) |
4836 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4837 | preempt_count() -= val; |
4838 | } | |
4839 | EXPORT_SYMBOL(sub_preempt_count); | |
4840 | ||
4841 | #endif | |
4842 | ||
4843 | /* | |
dd41f596 | 4844 | * Print scheduling while atomic bug: |
1da177e4 | 4845 | */ |
dd41f596 | 4846 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4847 | { |
838225b4 SS |
4848 | struct pt_regs *regs = get_irq_regs(); |
4849 | ||
4850 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4851 | prev->comm, prev->pid, preempt_count()); | |
4852 | ||
dd41f596 | 4853 | debug_show_held_locks(prev); |
e21f5b15 | 4854 | print_modules(); |
dd41f596 IM |
4855 | if (irqs_disabled()) |
4856 | print_irqtrace_events(prev); | |
838225b4 SS |
4857 | |
4858 | if (regs) | |
4859 | show_regs(regs); | |
4860 | else | |
4861 | dump_stack(); | |
dd41f596 | 4862 | } |
1da177e4 | 4863 | |
dd41f596 IM |
4864 | /* |
4865 | * Various schedule()-time debugging checks and statistics: | |
4866 | */ | |
4867 | static inline void schedule_debug(struct task_struct *prev) | |
4868 | { | |
1da177e4 | 4869 | /* |
41a2d6cf | 4870 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4871 | * schedule() atomically, we ignore that path for now. |
4872 | * Otherwise, whine if we are scheduling when we should not be. | |
4873 | */ | |
3f33a7ce | 4874 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4875 | __schedule_bug(prev); |
4876 | ||
1da177e4 LT |
4877 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4878 | ||
2d72376b | 4879 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4880 | #ifdef CONFIG_SCHEDSTATS |
4881 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4882 | schedstat_inc(this_rq(), bkl_count); |
4883 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4884 | } |
4885 | #endif | |
dd41f596 IM |
4886 | } |
4887 | ||
df1c99d4 MG |
4888 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
4889 | { | |
4890 | if (prev->state == TASK_RUNNING) { | |
4891 | u64 runtime = prev->se.sum_exec_runtime; | |
4892 | ||
4893 | runtime -= prev->se.prev_sum_exec_runtime; | |
4894 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
4895 | ||
4896 | /* | |
4897 | * In order to avoid avg_overlap growing stale when we are | |
4898 | * indeed overlapping and hence not getting put to sleep, grow | |
4899 | * the avg_overlap on preemption. | |
4900 | * | |
4901 | * We use the average preemption runtime because that | |
4902 | * correlates to the amount of cache footprint a task can | |
4903 | * build up. | |
4904 | */ | |
4905 | update_avg(&prev->se.avg_overlap, runtime); | |
4906 | } | |
4907 | prev->sched_class->put_prev_task(rq, prev); | |
4908 | } | |
4909 | ||
dd41f596 IM |
4910 | /* |
4911 | * Pick up the highest-prio task: | |
4912 | */ | |
4913 | static inline struct task_struct * | |
b67802ea | 4914 | pick_next_task(struct rq *rq) |
dd41f596 | 4915 | { |
5522d5d5 | 4916 | const struct sched_class *class; |
dd41f596 | 4917 | struct task_struct *p; |
1da177e4 LT |
4918 | |
4919 | /* | |
dd41f596 IM |
4920 | * Optimization: we know that if all tasks are in |
4921 | * the fair class we can call that function directly: | |
1da177e4 | 4922 | */ |
dd41f596 | 4923 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4924 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4925 | if (likely(p)) |
4926 | return p; | |
1da177e4 LT |
4927 | } |
4928 | ||
dd41f596 IM |
4929 | class = sched_class_highest; |
4930 | for ( ; ; ) { | |
fb8d4724 | 4931 | p = class->pick_next_task(rq); |
dd41f596 IM |
4932 | if (p) |
4933 | return p; | |
4934 | /* | |
4935 | * Will never be NULL as the idle class always | |
4936 | * returns a non-NULL p: | |
4937 | */ | |
4938 | class = class->next; | |
4939 | } | |
4940 | } | |
1da177e4 | 4941 | |
dd41f596 IM |
4942 | /* |
4943 | * schedule() is the main scheduler function. | |
4944 | */ | |
4945 | asmlinkage void __sched schedule(void) | |
4946 | { | |
4947 | struct task_struct *prev, *next; | |
67ca7bde | 4948 | unsigned long *switch_count; |
dd41f596 | 4949 | struct rq *rq; |
31656519 | 4950 | int cpu; |
dd41f596 IM |
4951 | |
4952 | need_resched: | |
4953 | preempt_disable(); | |
4954 | cpu = smp_processor_id(); | |
4955 | rq = cpu_rq(cpu); | |
4956 | rcu_qsctr_inc(cpu); | |
4957 | prev = rq->curr; | |
4958 | switch_count = &prev->nivcsw; | |
4959 | ||
4960 | release_kernel_lock(prev); | |
4961 | need_resched_nonpreemptible: | |
4962 | ||
4963 | schedule_debug(prev); | |
1da177e4 | 4964 | |
31656519 | 4965 | if (sched_feat(HRTICK)) |
f333fdc9 | 4966 | hrtick_clear(rq); |
8f4d37ec | 4967 | |
8cd162ce | 4968 | spin_lock_irq(&rq->lock); |
3e51f33f | 4969 | update_rq_clock(rq); |
1e819950 | 4970 | clear_tsk_need_resched(prev); |
1da177e4 | 4971 | |
1da177e4 | 4972 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4973 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4974 | prev->state = TASK_RUNNING; |
16882c1e | 4975 | else |
2e1cb74a | 4976 | deactivate_task(rq, prev, 1); |
dd41f596 | 4977 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4978 | } |
4979 | ||
9a897c5a SR |
4980 | #ifdef CONFIG_SMP |
4981 | if (prev->sched_class->pre_schedule) | |
4982 | prev->sched_class->pre_schedule(rq, prev); | |
4983 | #endif | |
f65eda4f | 4984 | |
dd41f596 | 4985 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4986 | idle_balance(cpu, rq); |
1da177e4 | 4987 | |
df1c99d4 | 4988 | put_prev_task(rq, prev); |
b67802ea | 4989 | next = pick_next_task(rq); |
1da177e4 | 4990 | |
1da177e4 | 4991 | if (likely(prev != next)) { |
673a90a1 DS |
4992 | sched_info_switch(prev, next); |
4993 | ||
1da177e4 LT |
4994 | rq->nr_switches++; |
4995 | rq->curr = next; | |
4996 | ++*switch_count; | |
4997 | ||
dd41f596 | 4998 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4999 | /* |
5000 | * the context switch might have flipped the stack from under | |
5001 | * us, hence refresh the local variables. | |
5002 | */ | |
5003 | cpu = smp_processor_id(); | |
5004 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5005 | } else |
5006 | spin_unlock_irq(&rq->lock); | |
5007 | ||
8f4d37ec | 5008 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5009 | goto need_resched_nonpreemptible; |
8f4d37ec | 5010 | |
1da177e4 LT |
5011 | preempt_enable_no_resched(); |
5012 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
5013 | goto need_resched; | |
5014 | } | |
1da177e4 LT |
5015 | EXPORT_SYMBOL(schedule); |
5016 | ||
5017 | #ifdef CONFIG_PREEMPT | |
5018 | /* | |
2ed6e34f | 5019 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5020 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5021 | * occur there and call schedule directly. |
5022 | */ | |
5023 | asmlinkage void __sched preempt_schedule(void) | |
5024 | { | |
5025 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5026 | |
1da177e4 LT |
5027 | /* |
5028 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5029 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5030 | */ |
beed33a8 | 5031 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5032 | return; |
5033 | ||
3a5c359a AK |
5034 | do { |
5035 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5036 | schedule(); |
3a5c359a | 5037 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5038 | |
3a5c359a AK |
5039 | /* |
5040 | * Check again in case we missed a preemption opportunity | |
5041 | * between schedule and now. | |
5042 | */ | |
5043 | barrier(); | |
5ed0cec0 | 5044 | } while (need_resched()); |
1da177e4 | 5045 | } |
1da177e4 LT |
5046 | EXPORT_SYMBOL(preempt_schedule); |
5047 | ||
5048 | /* | |
2ed6e34f | 5049 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5050 | * off of irq context. |
5051 | * Note, that this is called and return with irqs disabled. This will | |
5052 | * protect us against recursive calling from irq. | |
5053 | */ | |
5054 | asmlinkage void __sched preempt_schedule_irq(void) | |
5055 | { | |
5056 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5057 | |
2ed6e34f | 5058 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5059 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5060 | ||
3a5c359a AK |
5061 | do { |
5062 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5063 | local_irq_enable(); |
5064 | schedule(); | |
5065 | local_irq_disable(); | |
3a5c359a | 5066 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5067 | |
3a5c359a AK |
5068 | /* |
5069 | * Check again in case we missed a preemption opportunity | |
5070 | * between schedule and now. | |
5071 | */ | |
5072 | barrier(); | |
5ed0cec0 | 5073 | } while (need_resched()); |
1da177e4 LT |
5074 | } |
5075 | ||
5076 | #endif /* CONFIG_PREEMPT */ | |
5077 | ||
95cdf3b7 IM |
5078 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5079 | void *key) | |
1da177e4 | 5080 | { |
48f24c4d | 5081 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5082 | } |
1da177e4 LT |
5083 | EXPORT_SYMBOL(default_wake_function); |
5084 | ||
5085 | /* | |
41a2d6cf IM |
5086 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5087 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5088 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5089 | * | |
5090 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5091 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5092 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5093 | */ | |
777c6c5f JW |
5094 | void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
5095 | int nr_exclusive, int sync, void *key) | |
1da177e4 | 5096 | { |
2e45874c | 5097 | wait_queue_t *curr, *next; |
1da177e4 | 5098 | |
2e45874c | 5099 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5100 | unsigned flags = curr->flags; |
5101 | ||
1da177e4 | 5102 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5103 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5104 | break; |
5105 | } | |
5106 | } | |
5107 | ||
5108 | /** | |
5109 | * __wake_up - wake up threads blocked on a waitqueue. | |
5110 | * @q: the waitqueue | |
5111 | * @mode: which threads | |
5112 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5113 | * @key: is directly passed to the wakeup function |
1da177e4 | 5114 | */ |
7ad5b3a5 | 5115 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5116 | int nr_exclusive, void *key) |
1da177e4 LT |
5117 | { |
5118 | unsigned long flags; | |
5119 | ||
5120 | spin_lock_irqsave(&q->lock, flags); | |
5121 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5122 | spin_unlock_irqrestore(&q->lock, flags); | |
5123 | } | |
1da177e4 LT |
5124 | EXPORT_SYMBOL(__wake_up); |
5125 | ||
5126 | /* | |
5127 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5128 | */ | |
7ad5b3a5 | 5129 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5130 | { |
5131 | __wake_up_common(q, mode, 1, 0, NULL); | |
5132 | } | |
5133 | ||
5134 | /** | |
67be2dd1 | 5135 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5136 | * @q: the waitqueue |
5137 | * @mode: which threads | |
5138 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
5139 | * | |
5140 | * The sync wakeup differs that the waker knows that it will schedule | |
5141 | * away soon, so while the target thread will be woken up, it will not | |
5142 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5143 | * with each other. This can prevent needless bouncing between CPUs. | |
5144 | * | |
5145 | * On UP it can prevent extra preemption. | |
5146 | */ | |
7ad5b3a5 | 5147 | void |
95cdf3b7 | 5148 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
5149 | { |
5150 | unsigned long flags; | |
5151 | int sync = 1; | |
5152 | ||
5153 | if (unlikely(!q)) | |
5154 | return; | |
5155 | ||
5156 | if (unlikely(!nr_exclusive)) | |
5157 | sync = 0; | |
5158 | ||
5159 | spin_lock_irqsave(&q->lock, flags); | |
5160 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
5161 | spin_unlock_irqrestore(&q->lock, flags); | |
5162 | } | |
5163 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
5164 | ||
65eb3dc6 KD |
5165 | /** |
5166 | * complete: - signals a single thread waiting on this completion | |
5167 | * @x: holds the state of this particular completion | |
5168 | * | |
5169 | * This will wake up a single thread waiting on this completion. Threads will be | |
5170 | * awakened in the same order in which they were queued. | |
5171 | * | |
5172 | * See also complete_all(), wait_for_completion() and related routines. | |
5173 | */ | |
b15136e9 | 5174 | void complete(struct completion *x) |
1da177e4 LT |
5175 | { |
5176 | unsigned long flags; | |
5177 | ||
5178 | spin_lock_irqsave(&x->wait.lock, flags); | |
5179 | x->done++; | |
d9514f6c | 5180 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5181 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5182 | } | |
5183 | EXPORT_SYMBOL(complete); | |
5184 | ||
65eb3dc6 KD |
5185 | /** |
5186 | * complete_all: - signals all threads waiting on this completion | |
5187 | * @x: holds the state of this particular completion | |
5188 | * | |
5189 | * This will wake up all threads waiting on this particular completion event. | |
5190 | */ | |
b15136e9 | 5191 | void complete_all(struct completion *x) |
1da177e4 LT |
5192 | { |
5193 | unsigned long flags; | |
5194 | ||
5195 | spin_lock_irqsave(&x->wait.lock, flags); | |
5196 | x->done += UINT_MAX/2; | |
d9514f6c | 5197 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5198 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5199 | } | |
5200 | EXPORT_SYMBOL(complete_all); | |
5201 | ||
8cbbe86d AK |
5202 | static inline long __sched |
5203 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5204 | { |
1da177e4 LT |
5205 | if (!x->done) { |
5206 | DECLARE_WAITQUEUE(wait, current); | |
5207 | ||
5208 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5209 | __add_wait_queue_tail(&x->wait, &wait); | |
5210 | do { | |
94d3d824 | 5211 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5212 | timeout = -ERESTARTSYS; |
5213 | break; | |
8cbbe86d AK |
5214 | } |
5215 | __set_current_state(state); | |
1da177e4 LT |
5216 | spin_unlock_irq(&x->wait.lock); |
5217 | timeout = schedule_timeout(timeout); | |
5218 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5219 | } while (!x->done && timeout); |
1da177e4 | 5220 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5221 | if (!x->done) |
5222 | return timeout; | |
1da177e4 LT |
5223 | } |
5224 | x->done--; | |
ea71a546 | 5225 | return timeout ?: 1; |
1da177e4 | 5226 | } |
1da177e4 | 5227 | |
8cbbe86d AK |
5228 | static long __sched |
5229 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5230 | { |
1da177e4 LT |
5231 | might_sleep(); |
5232 | ||
5233 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5234 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5235 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5236 | return timeout; |
5237 | } | |
1da177e4 | 5238 | |
65eb3dc6 KD |
5239 | /** |
5240 | * wait_for_completion: - waits for completion of a task | |
5241 | * @x: holds the state of this particular completion | |
5242 | * | |
5243 | * This waits to be signaled for completion of a specific task. It is NOT | |
5244 | * interruptible and there is no timeout. | |
5245 | * | |
5246 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5247 | * and interrupt capability. Also see complete(). | |
5248 | */ | |
b15136e9 | 5249 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5250 | { |
5251 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5252 | } |
8cbbe86d | 5253 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5254 | |
65eb3dc6 KD |
5255 | /** |
5256 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5257 | * @x: holds the state of this particular completion | |
5258 | * @timeout: timeout value in jiffies | |
5259 | * | |
5260 | * This waits for either a completion of a specific task to be signaled or for a | |
5261 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5262 | * interruptible. | |
5263 | */ | |
b15136e9 | 5264 | unsigned long __sched |
8cbbe86d | 5265 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5266 | { |
8cbbe86d | 5267 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5268 | } |
8cbbe86d | 5269 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5270 | |
65eb3dc6 KD |
5271 | /** |
5272 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5273 | * @x: holds the state of this particular completion | |
5274 | * | |
5275 | * This waits for completion of a specific task to be signaled. It is | |
5276 | * interruptible. | |
5277 | */ | |
8cbbe86d | 5278 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5279 | { |
51e97990 AK |
5280 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5281 | if (t == -ERESTARTSYS) | |
5282 | return t; | |
5283 | return 0; | |
0fec171c | 5284 | } |
8cbbe86d | 5285 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5286 | |
65eb3dc6 KD |
5287 | /** |
5288 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5289 | * @x: holds the state of this particular completion | |
5290 | * @timeout: timeout value in jiffies | |
5291 | * | |
5292 | * This waits for either a completion of a specific task to be signaled or for a | |
5293 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5294 | */ | |
b15136e9 | 5295 | unsigned long __sched |
8cbbe86d AK |
5296 | wait_for_completion_interruptible_timeout(struct completion *x, |
5297 | unsigned long timeout) | |
0fec171c | 5298 | { |
8cbbe86d | 5299 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5300 | } |
8cbbe86d | 5301 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5302 | |
65eb3dc6 KD |
5303 | /** |
5304 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5305 | * @x: holds the state of this particular completion | |
5306 | * | |
5307 | * This waits to be signaled for completion of a specific task. It can be | |
5308 | * interrupted by a kill signal. | |
5309 | */ | |
009e577e MW |
5310 | int __sched wait_for_completion_killable(struct completion *x) |
5311 | { | |
5312 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5313 | if (t == -ERESTARTSYS) | |
5314 | return t; | |
5315 | return 0; | |
5316 | } | |
5317 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5318 | ||
be4de352 DC |
5319 | /** |
5320 | * try_wait_for_completion - try to decrement a completion without blocking | |
5321 | * @x: completion structure | |
5322 | * | |
5323 | * Returns: 0 if a decrement cannot be done without blocking | |
5324 | * 1 if a decrement succeeded. | |
5325 | * | |
5326 | * If a completion is being used as a counting completion, | |
5327 | * attempt to decrement the counter without blocking. This | |
5328 | * enables us to avoid waiting if the resource the completion | |
5329 | * is protecting is not available. | |
5330 | */ | |
5331 | bool try_wait_for_completion(struct completion *x) | |
5332 | { | |
5333 | int ret = 1; | |
5334 | ||
5335 | spin_lock_irq(&x->wait.lock); | |
5336 | if (!x->done) | |
5337 | ret = 0; | |
5338 | else | |
5339 | x->done--; | |
5340 | spin_unlock_irq(&x->wait.lock); | |
5341 | return ret; | |
5342 | } | |
5343 | EXPORT_SYMBOL(try_wait_for_completion); | |
5344 | ||
5345 | /** | |
5346 | * completion_done - Test to see if a completion has any waiters | |
5347 | * @x: completion structure | |
5348 | * | |
5349 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5350 | * 1 if there are no waiters. | |
5351 | * | |
5352 | */ | |
5353 | bool completion_done(struct completion *x) | |
5354 | { | |
5355 | int ret = 1; | |
5356 | ||
5357 | spin_lock_irq(&x->wait.lock); | |
5358 | if (!x->done) | |
5359 | ret = 0; | |
5360 | spin_unlock_irq(&x->wait.lock); | |
5361 | return ret; | |
5362 | } | |
5363 | EXPORT_SYMBOL(completion_done); | |
5364 | ||
8cbbe86d AK |
5365 | static long __sched |
5366 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5367 | { |
0fec171c IM |
5368 | unsigned long flags; |
5369 | wait_queue_t wait; | |
5370 | ||
5371 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5372 | |
8cbbe86d | 5373 | __set_current_state(state); |
1da177e4 | 5374 | |
8cbbe86d AK |
5375 | spin_lock_irqsave(&q->lock, flags); |
5376 | __add_wait_queue(q, &wait); | |
5377 | spin_unlock(&q->lock); | |
5378 | timeout = schedule_timeout(timeout); | |
5379 | spin_lock_irq(&q->lock); | |
5380 | __remove_wait_queue(q, &wait); | |
5381 | spin_unlock_irqrestore(&q->lock, flags); | |
5382 | ||
5383 | return timeout; | |
5384 | } | |
5385 | ||
5386 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5387 | { | |
5388 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5389 | } |
1da177e4 LT |
5390 | EXPORT_SYMBOL(interruptible_sleep_on); |
5391 | ||
0fec171c | 5392 | long __sched |
95cdf3b7 | 5393 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5394 | { |
8cbbe86d | 5395 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5396 | } |
1da177e4 LT |
5397 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5398 | ||
0fec171c | 5399 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5400 | { |
8cbbe86d | 5401 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5402 | } |
1da177e4 LT |
5403 | EXPORT_SYMBOL(sleep_on); |
5404 | ||
0fec171c | 5405 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5406 | { |
8cbbe86d | 5407 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5408 | } |
1da177e4 LT |
5409 | EXPORT_SYMBOL(sleep_on_timeout); |
5410 | ||
b29739f9 IM |
5411 | #ifdef CONFIG_RT_MUTEXES |
5412 | ||
5413 | /* | |
5414 | * rt_mutex_setprio - set the current priority of a task | |
5415 | * @p: task | |
5416 | * @prio: prio value (kernel-internal form) | |
5417 | * | |
5418 | * This function changes the 'effective' priority of a task. It does | |
5419 | * not touch ->normal_prio like __setscheduler(). | |
5420 | * | |
5421 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5422 | */ | |
36c8b586 | 5423 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5424 | { |
5425 | unsigned long flags; | |
83b699ed | 5426 | int oldprio, on_rq, running; |
70b97a7f | 5427 | struct rq *rq; |
cb469845 | 5428 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5429 | |
5430 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5431 | ||
5432 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5433 | update_rq_clock(rq); |
b29739f9 | 5434 | |
d5f9f942 | 5435 | oldprio = p->prio; |
dd41f596 | 5436 | on_rq = p->se.on_rq; |
051a1d1a | 5437 | running = task_current(rq, p); |
0e1f3483 | 5438 | if (on_rq) |
69be72c1 | 5439 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5440 | if (running) |
5441 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5442 | |
5443 | if (rt_prio(prio)) | |
5444 | p->sched_class = &rt_sched_class; | |
5445 | else | |
5446 | p->sched_class = &fair_sched_class; | |
5447 | ||
b29739f9 IM |
5448 | p->prio = prio; |
5449 | ||
0e1f3483 HS |
5450 | if (running) |
5451 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5452 | if (on_rq) { |
8159f87e | 5453 | enqueue_task(rq, p, 0); |
cb469845 SR |
5454 | |
5455 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5456 | } |
5457 | task_rq_unlock(rq, &flags); | |
5458 | } | |
5459 | ||
5460 | #endif | |
5461 | ||
36c8b586 | 5462 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5463 | { |
dd41f596 | 5464 | int old_prio, delta, on_rq; |
1da177e4 | 5465 | unsigned long flags; |
70b97a7f | 5466 | struct rq *rq; |
1da177e4 LT |
5467 | |
5468 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5469 | return; | |
5470 | /* | |
5471 | * We have to be careful, if called from sys_setpriority(), | |
5472 | * the task might be in the middle of scheduling on another CPU. | |
5473 | */ | |
5474 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5475 | update_rq_clock(rq); |
1da177e4 LT |
5476 | /* |
5477 | * The RT priorities are set via sched_setscheduler(), but we still | |
5478 | * allow the 'normal' nice value to be set - but as expected | |
5479 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5480 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5481 | */ |
e05606d3 | 5482 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5483 | p->static_prio = NICE_TO_PRIO(nice); |
5484 | goto out_unlock; | |
5485 | } | |
dd41f596 | 5486 | on_rq = p->se.on_rq; |
c09595f6 | 5487 | if (on_rq) |
69be72c1 | 5488 | dequeue_task(rq, p, 0); |
1da177e4 | 5489 | |
1da177e4 | 5490 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5491 | set_load_weight(p); |
b29739f9 IM |
5492 | old_prio = p->prio; |
5493 | p->prio = effective_prio(p); | |
5494 | delta = p->prio - old_prio; | |
1da177e4 | 5495 | |
dd41f596 | 5496 | if (on_rq) { |
8159f87e | 5497 | enqueue_task(rq, p, 0); |
1da177e4 | 5498 | /* |
d5f9f942 AM |
5499 | * If the task increased its priority or is running and |
5500 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5501 | */ |
d5f9f942 | 5502 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5503 | resched_task(rq->curr); |
5504 | } | |
5505 | out_unlock: | |
5506 | task_rq_unlock(rq, &flags); | |
5507 | } | |
1da177e4 LT |
5508 | EXPORT_SYMBOL(set_user_nice); |
5509 | ||
e43379f1 MM |
5510 | /* |
5511 | * can_nice - check if a task can reduce its nice value | |
5512 | * @p: task | |
5513 | * @nice: nice value | |
5514 | */ | |
36c8b586 | 5515 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5516 | { |
024f4747 MM |
5517 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5518 | int nice_rlim = 20 - nice; | |
48f24c4d | 5519 | |
e43379f1 MM |
5520 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5521 | capable(CAP_SYS_NICE)); | |
5522 | } | |
5523 | ||
1da177e4 LT |
5524 | #ifdef __ARCH_WANT_SYS_NICE |
5525 | ||
5526 | /* | |
5527 | * sys_nice - change the priority of the current process. | |
5528 | * @increment: priority increment | |
5529 | * | |
5530 | * sys_setpriority is a more generic, but much slower function that | |
5531 | * does similar things. | |
5532 | */ | |
5add95d4 | 5533 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5534 | { |
48f24c4d | 5535 | long nice, retval; |
1da177e4 LT |
5536 | |
5537 | /* | |
5538 | * Setpriority might change our priority at the same moment. | |
5539 | * We don't have to worry. Conceptually one call occurs first | |
5540 | * and we have a single winner. | |
5541 | */ | |
e43379f1 MM |
5542 | if (increment < -40) |
5543 | increment = -40; | |
1da177e4 LT |
5544 | if (increment > 40) |
5545 | increment = 40; | |
5546 | ||
2b8f836f | 5547 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5548 | if (nice < -20) |
5549 | nice = -20; | |
5550 | if (nice > 19) | |
5551 | nice = 19; | |
5552 | ||
e43379f1 MM |
5553 | if (increment < 0 && !can_nice(current, nice)) |
5554 | return -EPERM; | |
5555 | ||
1da177e4 LT |
5556 | retval = security_task_setnice(current, nice); |
5557 | if (retval) | |
5558 | return retval; | |
5559 | ||
5560 | set_user_nice(current, nice); | |
5561 | return 0; | |
5562 | } | |
5563 | ||
5564 | #endif | |
5565 | ||
5566 | /** | |
5567 | * task_prio - return the priority value of a given task. | |
5568 | * @p: the task in question. | |
5569 | * | |
5570 | * This is the priority value as seen by users in /proc. | |
5571 | * RT tasks are offset by -200. Normal tasks are centered | |
5572 | * around 0, value goes from -16 to +15. | |
5573 | */ | |
36c8b586 | 5574 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5575 | { |
5576 | return p->prio - MAX_RT_PRIO; | |
5577 | } | |
5578 | ||
5579 | /** | |
5580 | * task_nice - return the nice value of a given task. | |
5581 | * @p: the task in question. | |
5582 | */ | |
36c8b586 | 5583 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5584 | { |
5585 | return TASK_NICE(p); | |
5586 | } | |
150d8bed | 5587 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5588 | |
5589 | /** | |
5590 | * idle_cpu - is a given cpu idle currently? | |
5591 | * @cpu: the processor in question. | |
5592 | */ | |
5593 | int idle_cpu(int cpu) | |
5594 | { | |
5595 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5596 | } | |
5597 | ||
1da177e4 LT |
5598 | /** |
5599 | * idle_task - return the idle task for a given cpu. | |
5600 | * @cpu: the processor in question. | |
5601 | */ | |
36c8b586 | 5602 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5603 | { |
5604 | return cpu_rq(cpu)->idle; | |
5605 | } | |
5606 | ||
5607 | /** | |
5608 | * find_process_by_pid - find a process with a matching PID value. | |
5609 | * @pid: the pid in question. | |
5610 | */ | |
a9957449 | 5611 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5612 | { |
228ebcbe | 5613 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5614 | } |
5615 | ||
5616 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5617 | static void |
5618 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5619 | { |
dd41f596 | 5620 | BUG_ON(p->se.on_rq); |
48f24c4d | 5621 | |
1da177e4 | 5622 | p->policy = policy; |
dd41f596 IM |
5623 | switch (p->policy) { |
5624 | case SCHED_NORMAL: | |
5625 | case SCHED_BATCH: | |
5626 | case SCHED_IDLE: | |
5627 | p->sched_class = &fair_sched_class; | |
5628 | break; | |
5629 | case SCHED_FIFO: | |
5630 | case SCHED_RR: | |
5631 | p->sched_class = &rt_sched_class; | |
5632 | break; | |
5633 | } | |
5634 | ||
1da177e4 | 5635 | p->rt_priority = prio; |
b29739f9 IM |
5636 | p->normal_prio = normal_prio(p); |
5637 | /* we are holding p->pi_lock already */ | |
5638 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5639 | set_load_weight(p); |
1da177e4 LT |
5640 | } |
5641 | ||
c69e8d9c DH |
5642 | /* |
5643 | * check the target process has a UID that matches the current process's | |
5644 | */ | |
5645 | static bool check_same_owner(struct task_struct *p) | |
5646 | { | |
5647 | const struct cred *cred = current_cred(), *pcred; | |
5648 | bool match; | |
5649 | ||
5650 | rcu_read_lock(); | |
5651 | pcred = __task_cred(p); | |
5652 | match = (cred->euid == pcred->euid || | |
5653 | cred->euid == pcred->uid); | |
5654 | rcu_read_unlock(); | |
5655 | return match; | |
5656 | } | |
5657 | ||
961ccddd RR |
5658 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5659 | struct sched_param *param, bool user) | |
1da177e4 | 5660 | { |
83b699ed | 5661 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5662 | unsigned long flags; |
cb469845 | 5663 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5664 | struct rq *rq; |
1da177e4 | 5665 | |
66e5393a SR |
5666 | /* may grab non-irq protected spin_locks */ |
5667 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5668 | recheck: |
5669 | /* double check policy once rq lock held */ | |
5670 | if (policy < 0) | |
5671 | policy = oldpolicy = p->policy; | |
5672 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5673 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5674 | policy != SCHED_IDLE) | |
b0a9499c | 5675 | return -EINVAL; |
1da177e4 LT |
5676 | /* |
5677 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5678 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5679 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5680 | */ |
5681 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5682 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5683 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5684 | return -EINVAL; |
e05606d3 | 5685 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5686 | return -EINVAL; |
5687 | ||
37e4ab3f OC |
5688 | /* |
5689 | * Allow unprivileged RT tasks to decrease priority: | |
5690 | */ | |
961ccddd | 5691 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5692 | if (rt_policy(policy)) { |
8dc3e909 | 5693 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5694 | |
5695 | if (!lock_task_sighand(p, &flags)) | |
5696 | return -ESRCH; | |
5697 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5698 | unlock_task_sighand(p, &flags); | |
5699 | ||
5700 | /* can't set/change the rt policy */ | |
5701 | if (policy != p->policy && !rlim_rtprio) | |
5702 | return -EPERM; | |
5703 | ||
5704 | /* can't increase priority */ | |
5705 | if (param->sched_priority > p->rt_priority && | |
5706 | param->sched_priority > rlim_rtprio) | |
5707 | return -EPERM; | |
5708 | } | |
dd41f596 IM |
5709 | /* |
5710 | * Like positive nice levels, dont allow tasks to | |
5711 | * move out of SCHED_IDLE either: | |
5712 | */ | |
5713 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5714 | return -EPERM; | |
5fe1d75f | 5715 | |
37e4ab3f | 5716 | /* can't change other user's priorities */ |
c69e8d9c | 5717 | if (!check_same_owner(p)) |
37e4ab3f OC |
5718 | return -EPERM; |
5719 | } | |
1da177e4 | 5720 | |
725aad24 | 5721 | if (user) { |
b68aa230 | 5722 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5723 | /* |
5724 | * Do not allow realtime tasks into groups that have no runtime | |
5725 | * assigned. | |
5726 | */ | |
9a7e0b18 PZ |
5727 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5728 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5729 | return -EPERM; |
b68aa230 PZ |
5730 | #endif |
5731 | ||
725aad24 JF |
5732 | retval = security_task_setscheduler(p, policy, param); |
5733 | if (retval) | |
5734 | return retval; | |
5735 | } | |
5736 | ||
b29739f9 IM |
5737 | /* |
5738 | * make sure no PI-waiters arrive (or leave) while we are | |
5739 | * changing the priority of the task: | |
5740 | */ | |
5741 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5742 | /* |
5743 | * To be able to change p->policy safely, the apropriate | |
5744 | * runqueue lock must be held. | |
5745 | */ | |
b29739f9 | 5746 | rq = __task_rq_lock(p); |
1da177e4 LT |
5747 | /* recheck policy now with rq lock held */ |
5748 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5749 | policy = oldpolicy = -1; | |
b29739f9 IM |
5750 | __task_rq_unlock(rq); |
5751 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5752 | goto recheck; |
5753 | } | |
2daa3577 | 5754 | update_rq_clock(rq); |
dd41f596 | 5755 | on_rq = p->se.on_rq; |
051a1d1a | 5756 | running = task_current(rq, p); |
0e1f3483 | 5757 | if (on_rq) |
2e1cb74a | 5758 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5759 | if (running) |
5760 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5761 | |
1da177e4 | 5762 | oldprio = p->prio; |
dd41f596 | 5763 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5764 | |
0e1f3483 HS |
5765 | if (running) |
5766 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5767 | if (on_rq) { |
5768 | activate_task(rq, p, 0); | |
cb469845 SR |
5769 | |
5770 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5771 | } |
b29739f9 IM |
5772 | __task_rq_unlock(rq); |
5773 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5774 | ||
95e02ca9 TG |
5775 | rt_mutex_adjust_pi(p); |
5776 | ||
1da177e4 LT |
5777 | return 0; |
5778 | } | |
961ccddd RR |
5779 | |
5780 | /** | |
5781 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5782 | * @p: the task in question. | |
5783 | * @policy: new policy. | |
5784 | * @param: structure containing the new RT priority. | |
5785 | * | |
5786 | * NOTE that the task may be already dead. | |
5787 | */ | |
5788 | int sched_setscheduler(struct task_struct *p, int policy, | |
5789 | struct sched_param *param) | |
5790 | { | |
5791 | return __sched_setscheduler(p, policy, param, true); | |
5792 | } | |
1da177e4 LT |
5793 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5794 | ||
961ccddd RR |
5795 | /** |
5796 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5797 | * @p: the task in question. | |
5798 | * @policy: new policy. | |
5799 | * @param: structure containing the new RT priority. | |
5800 | * | |
5801 | * Just like sched_setscheduler, only don't bother checking if the | |
5802 | * current context has permission. For example, this is needed in | |
5803 | * stop_machine(): we create temporary high priority worker threads, | |
5804 | * but our caller might not have that capability. | |
5805 | */ | |
5806 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5807 | struct sched_param *param) | |
5808 | { | |
5809 | return __sched_setscheduler(p, policy, param, false); | |
5810 | } | |
5811 | ||
95cdf3b7 IM |
5812 | static int |
5813 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5814 | { |
1da177e4 LT |
5815 | struct sched_param lparam; |
5816 | struct task_struct *p; | |
36c8b586 | 5817 | int retval; |
1da177e4 LT |
5818 | |
5819 | if (!param || pid < 0) | |
5820 | return -EINVAL; | |
5821 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5822 | return -EFAULT; | |
5fe1d75f ON |
5823 | |
5824 | rcu_read_lock(); | |
5825 | retval = -ESRCH; | |
1da177e4 | 5826 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5827 | if (p != NULL) |
5828 | retval = sched_setscheduler(p, policy, &lparam); | |
5829 | rcu_read_unlock(); | |
36c8b586 | 5830 | |
1da177e4 LT |
5831 | return retval; |
5832 | } | |
5833 | ||
5834 | /** | |
5835 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5836 | * @pid: the pid in question. | |
5837 | * @policy: new policy. | |
5838 | * @param: structure containing the new RT priority. | |
5839 | */ | |
5add95d4 HC |
5840 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5841 | struct sched_param __user *, param) | |
1da177e4 | 5842 | { |
c21761f1 JB |
5843 | /* negative values for policy are not valid */ |
5844 | if (policy < 0) | |
5845 | return -EINVAL; | |
5846 | ||
1da177e4 LT |
5847 | return do_sched_setscheduler(pid, policy, param); |
5848 | } | |
5849 | ||
5850 | /** | |
5851 | * sys_sched_setparam - set/change the RT priority of a thread | |
5852 | * @pid: the pid in question. | |
5853 | * @param: structure containing the new RT priority. | |
5854 | */ | |
5add95d4 | 5855 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5856 | { |
5857 | return do_sched_setscheduler(pid, -1, param); | |
5858 | } | |
5859 | ||
5860 | /** | |
5861 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5862 | * @pid: the pid in question. | |
5863 | */ | |
5add95d4 | 5864 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5865 | { |
36c8b586 | 5866 | struct task_struct *p; |
3a5c359a | 5867 | int retval; |
1da177e4 LT |
5868 | |
5869 | if (pid < 0) | |
3a5c359a | 5870 | return -EINVAL; |
1da177e4 LT |
5871 | |
5872 | retval = -ESRCH; | |
5873 | read_lock(&tasklist_lock); | |
5874 | p = find_process_by_pid(pid); | |
5875 | if (p) { | |
5876 | retval = security_task_getscheduler(p); | |
5877 | if (!retval) | |
5878 | retval = p->policy; | |
5879 | } | |
5880 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5881 | return retval; |
5882 | } | |
5883 | ||
5884 | /** | |
5885 | * sys_sched_getscheduler - get the RT priority of a thread | |
5886 | * @pid: the pid in question. | |
5887 | * @param: structure containing the RT priority. | |
5888 | */ | |
5add95d4 | 5889 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5890 | { |
5891 | struct sched_param lp; | |
36c8b586 | 5892 | struct task_struct *p; |
3a5c359a | 5893 | int retval; |
1da177e4 LT |
5894 | |
5895 | if (!param || pid < 0) | |
3a5c359a | 5896 | return -EINVAL; |
1da177e4 LT |
5897 | |
5898 | read_lock(&tasklist_lock); | |
5899 | p = find_process_by_pid(pid); | |
5900 | retval = -ESRCH; | |
5901 | if (!p) | |
5902 | goto out_unlock; | |
5903 | ||
5904 | retval = security_task_getscheduler(p); | |
5905 | if (retval) | |
5906 | goto out_unlock; | |
5907 | ||
5908 | lp.sched_priority = p->rt_priority; | |
5909 | read_unlock(&tasklist_lock); | |
5910 | ||
5911 | /* | |
5912 | * This one might sleep, we cannot do it with a spinlock held ... | |
5913 | */ | |
5914 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5915 | ||
1da177e4 LT |
5916 | return retval; |
5917 | ||
5918 | out_unlock: | |
5919 | read_unlock(&tasklist_lock); | |
5920 | return retval; | |
5921 | } | |
5922 | ||
96f874e2 | 5923 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5924 | { |
5a16f3d3 | 5925 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5926 | struct task_struct *p; |
5927 | int retval; | |
1da177e4 | 5928 | |
95402b38 | 5929 | get_online_cpus(); |
1da177e4 LT |
5930 | read_lock(&tasklist_lock); |
5931 | ||
5932 | p = find_process_by_pid(pid); | |
5933 | if (!p) { | |
5934 | read_unlock(&tasklist_lock); | |
95402b38 | 5935 | put_online_cpus(); |
1da177e4 LT |
5936 | return -ESRCH; |
5937 | } | |
5938 | ||
5939 | /* | |
5940 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5941 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5942 | * usage count and then drop tasklist_lock. |
5943 | */ | |
5944 | get_task_struct(p); | |
5945 | read_unlock(&tasklist_lock); | |
5946 | ||
5a16f3d3 RR |
5947 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5948 | retval = -ENOMEM; | |
5949 | goto out_put_task; | |
5950 | } | |
5951 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5952 | retval = -ENOMEM; | |
5953 | goto out_free_cpus_allowed; | |
5954 | } | |
1da177e4 | 5955 | retval = -EPERM; |
c69e8d9c | 5956 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5957 | goto out_unlock; |
5958 | ||
e7834f8f DQ |
5959 | retval = security_task_setscheduler(p, 0, NULL); |
5960 | if (retval) | |
5961 | goto out_unlock; | |
5962 | ||
5a16f3d3 RR |
5963 | cpuset_cpus_allowed(p, cpus_allowed); |
5964 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 5965 | again: |
5a16f3d3 | 5966 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5967 | |
8707d8b8 | 5968 | if (!retval) { |
5a16f3d3 RR |
5969 | cpuset_cpus_allowed(p, cpus_allowed); |
5970 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5971 | /* |
5972 | * We must have raced with a concurrent cpuset | |
5973 | * update. Just reset the cpus_allowed to the | |
5974 | * cpuset's cpus_allowed | |
5975 | */ | |
5a16f3d3 | 5976 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5977 | goto again; |
5978 | } | |
5979 | } | |
1da177e4 | 5980 | out_unlock: |
5a16f3d3 RR |
5981 | free_cpumask_var(new_mask); |
5982 | out_free_cpus_allowed: | |
5983 | free_cpumask_var(cpus_allowed); | |
5984 | out_put_task: | |
1da177e4 | 5985 | put_task_struct(p); |
95402b38 | 5986 | put_online_cpus(); |
1da177e4 LT |
5987 | return retval; |
5988 | } | |
5989 | ||
5990 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5991 | struct cpumask *new_mask) |
1da177e4 | 5992 | { |
96f874e2 RR |
5993 | if (len < cpumask_size()) |
5994 | cpumask_clear(new_mask); | |
5995 | else if (len > cpumask_size()) | |
5996 | len = cpumask_size(); | |
5997 | ||
1da177e4 LT |
5998 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5999 | } | |
6000 | ||
6001 | /** | |
6002 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6003 | * @pid: pid of the process | |
6004 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6005 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6006 | */ | |
5add95d4 HC |
6007 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6008 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6009 | { |
5a16f3d3 | 6010 | cpumask_var_t new_mask; |
1da177e4 LT |
6011 | int retval; |
6012 | ||
5a16f3d3 RR |
6013 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6014 | return -ENOMEM; | |
1da177e4 | 6015 | |
5a16f3d3 RR |
6016 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6017 | if (retval == 0) | |
6018 | retval = sched_setaffinity(pid, new_mask); | |
6019 | free_cpumask_var(new_mask); | |
6020 | return retval; | |
1da177e4 LT |
6021 | } |
6022 | ||
96f874e2 | 6023 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6024 | { |
36c8b586 | 6025 | struct task_struct *p; |
1da177e4 | 6026 | int retval; |
1da177e4 | 6027 | |
95402b38 | 6028 | get_online_cpus(); |
1da177e4 LT |
6029 | read_lock(&tasklist_lock); |
6030 | ||
6031 | retval = -ESRCH; | |
6032 | p = find_process_by_pid(pid); | |
6033 | if (!p) | |
6034 | goto out_unlock; | |
6035 | ||
e7834f8f DQ |
6036 | retval = security_task_getscheduler(p); |
6037 | if (retval) | |
6038 | goto out_unlock; | |
6039 | ||
96f874e2 | 6040 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6041 | |
6042 | out_unlock: | |
6043 | read_unlock(&tasklist_lock); | |
95402b38 | 6044 | put_online_cpus(); |
1da177e4 | 6045 | |
9531b62f | 6046 | return retval; |
1da177e4 LT |
6047 | } |
6048 | ||
6049 | /** | |
6050 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6051 | * @pid: pid of the process | |
6052 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6053 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6054 | */ | |
5add95d4 HC |
6055 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6056 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6057 | { |
6058 | int ret; | |
f17c8607 | 6059 | cpumask_var_t mask; |
1da177e4 | 6060 | |
f17c8607 | 6061 | if (len < cpumask_size()) |
1da177e4 LT |
6062 | return -EINVAL; |
6063 | ||
f17c8607 RR |
6064 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6065 | return -ENOMEM; | |
1da177e4 | 6066 | |
f17c8607 RR |
6067 | ret = sched_getaffinity(pid, mask); |
6068 | if (ret == 0) { | |
6069 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6070 | ret = -EFAULT; | |
6071 | else | |
6072 | ret = cpumask_size(); | |
6073 | } | |
6074 | free_cpumask_var(mask); | |
1da177e4 | 6075 | |
f17c8607 | 6076 | return ret; |
1da177e4 LT |
6077 | } |
6078 | ||
6079 | /** | |
6080 | * sys_sched_yield - yield the current processor to other threads. | |
6081 | * | |
dd41f596 IM |
6082 | * This function yields the current CPU to other tasks. If there are no |
6083 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6084 | */ |
5add95d4 | 6085 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6086 | { |
70b97a7f | 6087 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6088 | |
2d72376b | 6089 | schedstat_inc(rq, yld_count); |
4530d7ab | 6090 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6091 | |
6092 | /* | |
6093 | * Since we are going to call schedule() anyway, there's | |
6094 | * no need to preempt or enable interrupts: | |
6095 | */ | |
6096 | __release(rq->lock); | |
8a25d5de | 6097 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6098 | _raw_spin_unlock(&rq->lock); |
6099 | preempt_enable_no_resched(); | |
6100 | ||
6101 | schedule(); | |
6102 | ||
6103 | return 0; | |
6104 | } | |
6105 | ||
e7b38404 | 6106 | static void __cond_resched(void) |
1da177e4 | 6107 | { |
8e0a43d8 IM |
6108 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6109 | __might_sleep(__FILE__, __LINE__); | |
6110 | #endif | |
5bbcfd90 IM |
6111 | /* |
6112 | * The BKS might be reacquired before we have dropped | |
6113 | * PREEMPT_ACTIVE, which could trigger a second | |
6114 | * cond_resched() call. | |
6115 | */ | |
1da177e4 LT |
6116 | do { |
6117 | add_preempt_count(PREEMPT_ACTIVE); | |
6118 | schedule(); | |
6119 | sub_preempt_count(PREEMPT_ACTIVE); | |
6120 | } while (need_resched()); | |
6121 | } | |
6122 | ||
02b67cc3 | 6123 | int __sched _cond_resched(void) |
1da177e4 | 6124 | { |
9414232f IM |
6125 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6126 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6127 | __cond_resched(); |
6128 | return 1; | |
6129 | } | |
6130 | return 0; | |
6131 | } | |
02b67cc3 | 6132 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6133 | |
6134 | /* | |
6135 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6136 | * call schedule, and on return reacquire the lock. | |
6137 | * | |
41a2d6cf | 6138 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6139 | * operations here to prevent schedule() from being called twice (once via |
6140 | * spin_unlock(), once by hand). | |
6141 | */ | |
95cdf3b7 | 6142 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6143 | { |
95c354fe | 6144 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6145 | int ret = 0; |
6146 | ||
95c354fe | 6147 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6148 | spin_unlock(lock); |
95c354fe NP |
6149 | if (resched && need_resched()) |
6150 | __cond_resched(); | |
6151 | else | |
6152 | cpu_relax(); | |
6df3cecb | 6153 | ret = 1; |
1da177e4 | 6154 | spin_lock(lock); |
1da177e4 | 6155 | } |
6df3cecb | 6156 | return ret; |
1da177e4 | 6157 | } |
1da177e4 LT |
6158 | EXPORT_SYMBOL(cond_resched_lock); |
6159 | ||
6160 | int __sched cond_resched_softirq(void) | |
6161 | { | |
6162 | BUG_ON(!in_softirq()); | |
6163 | ||
9414232f | 6164 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6165 | local_bh_enable(); |
1da177e4 LT |
6166 | __cond_resched(); |
6167 | local_bh_disable(); | |
6168 | return 1; | |
6169 | } | |
6170 | return 0; | |
6171 | } | |
1da177e4 LT |
6172 | EXPORT_SYMBOL(cond_resched_softirq); |
6173 | ||
1da177e4 LT |
6174 | /** |
6175 | * yield - yield the current processor to other threads. | |
6176 | * | |
72fd4a35 | 6177 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6178 | * thread runnable and calls sys_sched_yield(). |
6179 | */ | |
6180 | void __sched yield(void) | |
6181 | { | |
6182 | set_current_state(TASK_RUNNING); | |
6183 | sys_sched_yield(); | |
6184 | } | |
1da177e4 LT |
6185 | EXPORT_SYMBOL(yield); |
6186 | ||
6187 | /* | |
41a2d6cf | 6188 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6189 | * that process accounting knows that this is a task in IO wait state. |
6190 | * | |
6191 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6192 | * has set its backing_dev_info: the queue against which it should throttle) | |
6193 | */ | |
6194 | void __sched io_schedule(void) | |
6195 | { | |
70b97a7f | 6196 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6197 | |
0ff92245 | 6198 | delayacct_blkio_start(); |
1da177e4 LT |
6199 | atomic_inc(&rq->nr_iowait); |
6200 | schedule(); | |
6201 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6202 | delayacct_blkio_end(); |
1da177e4 | 6203 | } |
1da177e4 LT |
6204 | EXPORT_SYMBOL(io_schedule); |
6205 | ||
6206 | long __sched io_schedule_timeout(long timeout) | |
6207 | { | |
70b97a7f | 6208 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6209 | long ret; |
6210 | ||
0ff92245 | 6211 | delayacct_blkio_start(); |
1da177e4 LT |
6212 | atomic_inc(&rq->nr_iowait); |
6213 | ret = schedule_timeout(timeout); | |
6214 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6215 | delayacct_blkio_end(); |
1da177e4 LT |
6216 | return ret; |
6217 | } | |
6218 | ||
6219 | /** | |
6220 | * sys_sched_get_priority_max - return maximum RT priority. | |
6221 | * @policy: scheduling class. | |
6222 | * | |
6223 | * this syscall returns the maximum rt_priority that can be used | |
6224 | * by a given scheduling class. | |
6225 | */ | |
5add95d4 | 6226 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6227 | { |
6228 | int ret = -EINVAL; | |
6229 | ||
6230 | switch (policy) { | |
6231 | case SCHED_FIFO: | |
6232 | case SCHED_RR: | |
6233 | ret = MAX_USER_RT_PRIO-1; | |
6234 | break; | |
6235 | case SCHED_NORMAL: | |
b0a9499c | 6236 | case SCHED_BATCH: |
dd41f596 | 6237 | case SCHED_IDLE: |
1da177e4 LT |
6238 | ret = 0; |
6239 | break; | |
6240 | } | |
6241 | return ret; | |
6242 | } | |
6243 | ||
6244 | /** | |
6245 | * sys_sched_get_priority_min - return minimum RT priority. | |
6246 | * @policy: scheduling class. | |
6247 | * | |
6248 | * this syscall returns the minimum rt_priority that can be used | |
6249 | * by a given scheduling class. | |
6250 | */ | |
5add95d4 | 6251 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6252 | { |
6253 | int ret = -EINVAL; | |
6254 | ||
6255 | switch (policy) { | |
6256 | case SCHED_FIFO: | |
6257 | case SCHED_RR: | |
6258 | ret = 1; | |
6259 | break; | |
6260 | case SCHED_NORMAL: | |
b0a9499c | 6261 | case SCHED_BATCH: |
dd41f596 | 6262 | case SCHED_IDLE: |
1da177e4 LT |
6263 | ret = 0; |
6264 | } | |
6265 | return ret; | |
6266 | } | |
6267 | ||
6268 | /** | |
6269 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6270 | * @pid: pid of the process. | |
6271 | * @interval: userspace pointer to the timeslice value. | |
6272 | * | |
6273 | * this syscall writes the default timeslice value of a given process | |
6274 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6275 | */ | |
17da2bd9 | 6276 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6277 | struct timespec __user *, interval) |
1da177e4 | 6278 | { |
36c8b586 | 6279 | struct task_struct *p; |
a4ec24b4 | 6280 | unsigned int time_slice; |
3a5c359a | 6281 | int retval; |
1da177e4 | 6282 | struct timespec t; |
1da177e4 LT |
6283 | |
6284 | if (pid < 0) | |
3a5c359a | 6285 | return -EINVAL; |
1da177e4 LT |
6286 | |
6287 | retval = -ESRCH; | |
6288 | read_lock(&tasklist_lock); | |
6289 | p = find_process_by_pid(pid); | |
6290 | if (!p) | |
6291 | goto out_unlock; | |
6292 | ||
6293 | retval = security_task_getscheduler(p); | |
6294 | if (retval) | |
6295 | goto out_unlock; | |
6296 | ||
77034937 IM |
6297 | /* |
6298 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6299 | * tasks that are on an otherwise idle runqueue: | |
6300 | */ | |
6301 | time_slice = 0; | |
6302 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6303 | time_slice = DEF_TIMESLICE; |
1868f958 | 6304 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6305 | struct sched_entity *se = &p->se; |
6306 | unsigned long flags; | |
6307 | struct rq *rq; | |
6308 | ||
6309 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6310 | if (rq->cfs.load.weight) |
6311 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6312 | task_rq_unlock(rq, &flags); |
6313 | } | |
1da177e4 | 6314 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6315 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6316 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6317 | return retval; |
3a5c359a | 6318 | |
1da177e4 LT |
6319 | out_unlock: |
6320 | read_unlock(&tasklist_lock); | |
6321 | return retval; | |
6322 | } | |
6323 | ||
7c731e0a | 6324 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6325 | |
82a1fcb9 | 6326 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6327 | { |
1da177e4 | 6328 | unsigned long free = 0; |
36c8b586 | 6329 | unsigned state; |
1da177e4 | 6330 | |
1da177e4 | 6331 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6332 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6333 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6334 | #if BITS_PER_LONG == 32 |
1da177e4 | 6335 | if (state == TASK_RUNNING) |
cc4ea795 | 6336 | printk(KERN_CONT " running "); |
1da177e4 | 6337 | else |
cc4ea795 | 6338 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6339 | #else |
6340 | if (state == TASK_RUNNING) | |
cc4ea795 | 6341 | printk(KERN_CONT " running task "); |
1da177e4 | 6342 | else |
cc4ea795 | 6343 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6344 | #endif |
6345 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6346 | free = stack_not_used(p); |
1da177e4 | 6347 | #endif |
ba25f9dc | 6348 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 6349 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 6350 | |
5fb5e6de | 6351 | show_stack(p, NULL); |
1da177e4 LT |
6352 | } |
6353 | ||
e59e2ae2 | 6354 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6355 | { |
36c8b586 | 6356 | struct task_struct *g, *p; |
1da177e4 | 6357 | |
4bd77321 IM |
6358 | #if BITS_PER_LONG == 32 |
6359 | printk(KERN_INFO | |
6360 | " task PC stack pid father\n"); | |
1da177e4 | 6361 | #else |
4bd77321 IM |
6362 | printk(KERN_INFO |
6363 | " task PC stack pid father\n"); | |
1da177e4 LT |
6364 | #endif |
6365 | read_lock(&tasklist_lock); | |
6366 | do_each_thread(g, p) { | |
6367 | /* | |
6368 | * reset the NMI-timeout, listing all files on a slow | |
6369 | * console might take alot of time: | |
6370 | */ | |
6371 | touch_nmi_watchdog(); | |
39bc89fd | 6372 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6373 | sched_show_task(p); |
1da177e4 LT |
6374 | } while_each_thread(g, p); |
6375 | ||
04c9167f JF |
6376 | touch_all_softlockup_watchdogs(); |
6377 | ||
dd41f596 IM |
6378 | #ifdef CONFIG_SCHED_DEBUG |
6379 | sysrq_sched_debug_show(); | |
6380 | #endif | |
1da177e4 | 6381 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6382 | /* |
6383 | * Only show locks if all tasks are dumped: | |
6384 | */ | |
6385 | if (state_filter == -1) | |
6386 | debug_show_all_locks(); | |
1da177e4 LT |
6387 | } |
6388 | ||
1df21055 IM |
6389 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6390 | { | |
dd41f596 | 6391 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6392 | } |
6393 | ||
f340c0d1 IM |
6394 | /** |
6395 | * init_idle - set up an idle thread for a given CPU | |
6396 | * @idle: task in question | |
6397 | * @cpu: cpu the idle task belongs to | |
6398 | * | |
6399 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6400 | * flag, to make booting more robust. | |
6401 | */ | |
5c1e1767 | 6402 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6403 | { |
70b97a7f | 6404 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6405 | unsigned long flags; |
6406 | ||
5cbd54ef IM |
6407 | spin_lock_irqsave(&rq->lock, flags); |
6408 | ||
dd41f596 IM |
6409 | __sched_fork(idle); |
6410 | idle->se.exec_start = sched_clock(); | |
6411 | ||
b29739f9 | 6412 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6413 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6414 | __set_task_cpu(idle, cpu); |
1da177e4 | 6415 | |
1da177e4 | 6416 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6417 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6418 | idle->oncpu = 1; | |
6419 | #endif | |
1da177e4 LT |
6420 | spin_unlock_irqrestore(&rq->lock, flags); |
6421 | ||
6422 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6423 | #if defined(CONFIG_PREEMPT) |
6424 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6425 | #else | |
a1261f54 | 6426 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6427 | #endif |
dd41f596 IM |
6428 | /* |
6429 | * The idle tasks have their own, simple scheduling class: | |
6430 | */ | |
6431 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6432 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6433 | } |
6434 | ||
6435 | /* | |
6436 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6437 | * indicates which cpus entered this state. This is used | |
6438 | * in the rcu update to wait only for active cpus. For system | |
6439 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6440 | * always be CPU_BITS_NONE. |
1da177e4 | 6441 | */ |
6a7b3dc3 | 6442 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6443 | |
19978ca6 IM |
6444 | /* |
6445 | * Increase the granularity value when there are more CPUs, | |
6446 | * because with more CPUs the 'effective latency' as visible | |
6447 | * to users decreases. But the relationship is not linear, | |
6448 | * so pick a second-best guess by going with the log2 of the | |
6449 | * number of CPUs. | |
6450 | * | |
6451 | * This idea comes from the SD scheduler of Con Kolivas: | |
6452 | */ | |
6453 | static inline void sched_init_granularity(void) | |
6454 | { | |
6455 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6456 | const unsigned long limit = 200000000; | |
6457 | ||
6458 | sysctl_sched_min_granularity *= factor; | |
6459 | if (sysctl_sched_min_granularity > limit) | |
6460 | sysctl_sched_min_granularity = limit; | |
6461 | ||
6462 | sysctl_sched_latency *= factor; | |
6463 | if (sysctl_sched_latency > limit) | |
6464 | sysctl_sched_latency = limit; | |
6465 | ||
6466 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6467 | |
6468 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6469 | } |
6470 | ||
1da177e4 LT |
6471 | #ifdef CONFIG_SMP |
6472 | /* | |
6473 | * This is how migration works: | |
6474 | * | |
70b97a7f | 6475 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6476 | * runqueue and wake up that CPU's migration thread. |
6477 | * 2) we down() the locked semaphore => thread blocks. | |
6478 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6479 | * thread off the CPU) | |
6480 | * 4) it gets the migration request and checks whether the migrated | |
6481 | * task is still in the wrong runqueue. | |
6482 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6483 | * it and puts it into the right queue. | |
6484 | * 6) migration thread up()s the semaphore. | |
6485 | * 7) we wake up and the migration is done. | |
6486 | */ | |
6487 | ||
6488 | /* | |
6489 | * Change a given task's CPU affinity. Migrate the thread to a | |
6490 | * proper CPU and schedule it away if the CPU it's executing on | |
6491 | * is removed from the allowed bitmask. | |
6492 | * | |
6493 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6494 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6495 | * call is not atomic; no spinlocks may be held. |
6496 | */ | |
96f874e2 | 6497 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6498 | { |
70b97a7f | 6499 | struct migration_req req; |
1da177e4 | 6500 | unsigned long flags; |
70b97a7f | 6501 | struct rq *rq; |
48f24c4d | 6502 | int ret = 0; |
1da177e4 LT |
6503 | |
6504 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6505 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6506 | ret = -EINVAL; |
6507 | goto out; | |
6508 | } | |
6509 | ||
9985b0ba | 6510 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6511 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6512 | ret = -EINVAL; |
6513 | goto out; | |
6514 | } | |
6515 | ||
73fe6aae | 6516 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6517 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6518 | else { |
96f874e2 RR |
6519 | cpumask_copy(&p->cpus_allowed, new_mask); |
6520 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6521 | } |
6522 | ||
1da177e4 | 6523 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6524 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6525 | goto out; |
6526 | ||
1e5ce4f4 | 6527 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6528 | /* Need help from migration thread: drop lock and wait. */ |
6529 | task_rq_unlock(rq, &flags); | |
6530 | wake_up_process(rq->migration_thread); | |
6531 | wait_for_completion(&req.done); | |
6532 | tlb_migrate_finish(p->mm); | |
6533 | return 0; | |
6534 | } | |
6535 | out: | |
6536 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6537 | |
1da177e4 LT |
6538 | return ret; |
6539 | } | |
cd8ba7cd | 6540 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6541 | |
6542 | /* | |
41a2d6cf | 6543 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6544 | * this because either it can't run here any more (set_cpus_allowed() |
6545 | * away from this CPU, or CPU going down), or because we're | |
6546 | * attempting to rebalance this task on exec (sched_exec). | |
6547 | * | |
6548 | * So we race with normal scheduler movements, but that's OK, as long | |
6549 | * as the task is no longer on this CPU. | |
efc30814 KK |
6550 | * |
6551 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6552 | */ |
efc30814 | 6553 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6554 | { |
70b97a7f | 6555 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6556 | int ret = 0, on_rq; |
1da177e4 | 6557 | |
e761b772 | 6558 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6559 | return ret; |
1da177e4 LT |
6560 | |
6561 | rq_src = cpu_rq(src_cpu); | |
6562 | rq_dest = cpu_rq(dest_cpu); | |
6563 | ||
6564 | double_rq_lock(rq_src, rq_dest); | |
6565 | /* Already moved. */ | |
6566 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6567 | goto done; |
1da177e4 | 6568 | /* Affinity changed (again). */ |
96f874e2 | 6569 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6570 | goto fail; |
1da177e4 | 6571 | |
dd41f596 | 6572 | on_rq = p->se.on_rq; |
6e82a3be | 6573 | if (on_rq) |
2e1cb74a | 6574 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6575 | |
1da177e4 | 6576 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6577 | if (on_rq) { |
6578 | activate_task(rq_dest, p, 0); | |
15afe09b | 6579 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6580 | } |
b1e38734 | 6581 | done: |
efc30814 | 6582 | ret = 1; |
b1e38734 | 6583 | fail: |
1da177e4 | 6584 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6585 | return ret; |
1da177e4 LT |
6586 | } |
6587 | ||
6588 | /* | |
6589 | * migration_thread - this is a highprio system thread that performs | |
6590 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6591 | * another runqueue. | |
6592 | */ | |
95cdf3b7 | 6593 | static int migration_thread(void *data) |
1da177e4 | 6594 | { |
1da177e4 | 6595 | int cpu = (long)data; |
70b97a7f | 6596 | struct rq *rq; |
1da177e4 LT |
6597 | |
6598 | rq = cpu_rq(cpu); | |
6599 | BUG_ON(rq->migration_thread != current); | |
6600 | ||
6601 | set_current_state(TASK_INTERRUPTIBLE); | |
6602 | while (!kthread_should_stop()) { | |
70b97a7f | 6603 | struct migration_req *req; |
1da177e4 | 6604 | struct list_head *head; |
1da177e4 | 6605 | |
1da177e4 LT |
6606 | spin_lock_irq(&rq->lock); |
6607 | ||
6608 | if (cpu_is_offline(cpu)) { | |
6609 | spin_unlock_irq(&rq->lock); | |
6610 | goto wait_to_die; | |
6611 | } | |
6612 | ||
6613 | if (rq->active_balance) { | |
6614 | active_load_balance(rq, cpu); | |
6615 | rq->active_balance = 0; | |
6616 | } | |
6617 | ||
6618 | head = &rq->migration_queue; | |
6619 | ||
6620 | if (list_empty(head)) { | |
6621 | spin_unlock_irq(&rq->lock); | |
6622 | schedule(); | |
6623 | set_current_state(TASK_INTERRUPTIBLE); | |
6624 | continue; | |
6625 | } | |
70b97a7f | 6626 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6627 | list_del_init(head->next); |
6628 | ||
674311d5 NP |
6629 | spin_unlock(&rq->lock); |
6630 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6631 | local_irq_enable(); | |
1da177e4 LT |
6632 | |
6633 | complete(&req->done); | |
6634 | } | |
6635 | __set_current_state(TASK_RUNNING); | |
6636 | return 0; | |
6637 | ||
6638 | wait_to_die: | |
6639 | /* Wait for kthread_stop */ | |
6640 | set_current_state(TASK_INTERRUPTIBLE); | |
6641 | while (!kthread_should_stop()) { | |
6642 | schedule(); | |
6643 | set_current_state(TASK_INTERRUPTIBLE); | |
6644 | } | |
6645 | __set_current_state(TASK_RUNNING); | |
6646 | return 0; | |
6647 | } | |
6648 | ||
6649 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6650 | |
6651 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6652 | { | |
6653 | int ret; | |
6654 | ||
6655 | local_irq_disable(); | |
6656 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6657 | local_irq_enable(); | |
6658 | return ret; | |
6659 | } | |
6660 | ||
054b9108 | 6661 | /* |
3a4fa0a2 | 6662 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6663 | */ |
48f24c4d | 6664 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6665 | { |
70b97a7f | 6666 | int dest_cpu; |
6ca09dfc | 6667 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
6668 | |
6669 | again: | |
6670 | /* Look for allowed, online CPU in same node. */ | |
6671 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6672 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6673 | goto move; | |
6674 | ||
6675 | /* Any allowed, online CPU? */ | |
6676 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6677 | if (dest_cpu < nr_cpu_ids) | |
6678 | goto move; | |
6679 | ||
6680 | /* No more Mr. Nice Guy. */ | |
6681 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6682 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6683 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6684 | |
e76bd8d9 RR |
6685 | /* |
6686 | * Don't tell them about moving exiting tasks or | |
6687 | * kernel threads (both mm NULL), since they never | |
6688 | * leave kernel. | |
6689 | */ | |
6690 | if (p->mm && printk_ratelimit()) { | |
6691 | printk(KERN_INFO "process %d (%s) no " | |
6692 | "longer affine to cpu%d\n", | |
6693 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6694 | } |
e76bd8d9 RR |
6695 | } |
6696 | ||
6697 | move: | |
6698 | /* It can have affinity changed while we were choosing. */ | |
6699 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6700 | goto again; | |
1da177e4 LT |
6701 | } |
6702 | ||
6703 | /* | |
6704 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6705 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6706 | * for performance reasons the counter is not stricly tracking tasks to | |
6707 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6708 | * to keep the global sum constant after CPU-down: | |
6709 | */ | |
70b97a7f | 6710 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6711 | { |
1e5ce4f4 | 6712 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6713 | unsigned long flags; |
6714 | ||
6715 | local_irq_save(flags); | |
6716 | double_rq_lock(rq_src, rq_dest); | |
6717 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6718 | rq_src->nr_uninterruptible = 0; | |
6719 | double_rq_unlock(rq_src, rq_dest); | |
6720 | local_irq_restore(flags); | |
6721 | } | |
6722 | ||
6723 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6724 | static void migrate_live_tasks(int src_cpu) | |
6725 | { | |
48f24c4d | 6726 | struct task_struct *p, *t; |
1da177e4 | 6727 | |
f7b4cddc | 6728 | read_lock(&tasklist_lock); |
1da177e4 | 6729 | |
48f24c4d IM |
6730 | do_each_thread(t, p) { |
6731 | if (p == current) | |
1da177e4 LT |
6732 | continue; |
6733 | ||
48f24c4d IM |
6734 | if (task_cpu(p) == src_cpu) |
6735 | move_task_off_dead_cpu(src_cpu, p); | |
6736 | } while_each_thread(t, p); | |
1da177e4 | 6737 | |
f7b4cddc | 6738 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6739 | } |
6740 | ||
dd41f596 IM |
6741 | /* |
6742 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6743 | * It does so by boosting its priority to highest possible. |
6744 | * Used by CPU offline code. | |
1da177e4 LT |
6745 | */ |
6746 | void sched_idle_next(void) | |
6747 | { | |
48f24c4d | 6748 | int this_cpu = smp_processor_id(); |
70b97a7f | 6749 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6750 | struct task_struct *p = rq->idle; |
6751 | unsigned long flags; | |
6752 | ||
6753 | /* cpu has to be offline */ | |
48f24c4d | 6754 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6755 | |
48f24c4d IM |
6756 | /* |
6757 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6758 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6759 | */ |
6760 | spin_lock_irqsave(&rq->lock, flags); | |
6761 | ||
dd41f596 | 6762 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6763 | |
94bc9a7b DA |
6764 | update_rq_clock(rq); |
6765 | activate_task(rq, p, 0); | |
1da177e4 LT |
6766 | |
6767 | spin_unlock_irqrestore(&rq->lock, flags); | |
6768 | } | |
6769 | ||
48f24c4d IM |
6770 | /* |
6771 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6772 | * offline. |
6773 | */ | |
6774 | void idle_task_exit(void) | |
6775 | { | |
6776 | struct mm_struct *mm = current->active_mm; | |
6777 | ||
6778 | BUG_ON(cpu_online(smp_processor_id())); | |
6779 | ||
6780 | if (mm != &init_mm) | |
6781 | switch_mm(mm, &init_mm, current); | |
6782 | mmdrop(mm); | |
6783 | } | |
6784 | ||
054b9108 | 6785 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6786 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6787 | { |
70b97a7f | 6788 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6789 | |
6790 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6791 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6792 | |
6793 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6794 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6795 | |
48f24c4d | 6796 | get_task_struct(p); |
1da177e4 LT |
6797 | |
6798 | /* | |
6799 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6800 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6801 | * fine. |
6802 | */ | |
f7b4cddc | 6803 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6804 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6805 | spin_lock_irq(&rq->lock); |
1da177e4 | 6806 | |
48f24c4d | 6807 | put_task_struct(p); |
1da177e4 LT |
6808 | } |
6809 | ||
6810 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6811 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6812 | { | |
70b97a7f | 6813 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6814 | struct task_struct *next; |
48f24c4d | 6815 | |
dd41f596 IM |
6816 | for ( ; ; ) { |
6817 | if (!rq->nr_running) | |
6818 | break; | |
a8e504d2 | 6819 | update_rq_clock(rq); |
b67802ea | 6820 | next = pick_next_task(rq); |
dd41f596 IM |
6821 | if (!next) |
6822 | break; | |
79c53799 | 6823 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6824 | migrate_dead(dead_cpu, next); |
e692ab53 | 6825 | |
1da177e4 LT |
6826 | } |
6827 | } | |
6828 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6829 | ||
e692ab53 NP |
6830 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6831 | ||
6832 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6833 | { |
6834 | .procname = "sched_domain", | |
c57baf1e | 6835 | .mode = 0555, |
e0361851 | 6836 | }, |
38605cae | 6837 | {0, }, |
e692ab53 NP |
6838 | }; |
6839 | ||
6840 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6841 | { |
c57baf1e | 6842 | .ctl_name = CTL_KERN, |
e0361851 | 6843 | .procname = "kernel", |
c57baf1e | 6844 | .mode = 0555, |
e0361851 AD |
6845 | .child = sd_ctl_dir, |
6846 | }, | |
38605cae | 6847 | {0, }, |
e692ab53 NP |
6848 | }; |
6849 | ||
6850 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6851 | { | |
6852 | struct ctl_table *entry = | |
5cf9f062 | 6853 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6854 | |
e692ab53 NP |
6855 | return entry; |
6856 | } | |
6857 | ||
6382bc90 MM |
6858 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6859 | { | |
cd790076 | 6860 | struct ctl_table *entry; |
6382bc90 | 6861 | |
cd790076 MM |
6862 | /* |
6863 | * In the intermediate directories, both the child directory and | |
6864 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6865 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6866 | * static strings and all have proc handlers. |
6867 | */ | |
6868 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6869 | if (entry->child) |
6870 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6871 | if (entry->proc_handler == NULL) |
6872 | kfree(entry->procname); | |
6873 | } | |
6382bc90 MM |
6874 | |
6875 | kfree(*tablep); | |
6876 | *tablep = NULL; | |
6877 | } | |
6878 | ||
e692ab53 | 6879 | static void |
e0361851 | 6880 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6881 | const char *procname, void *data, int maxlen, |
6882 | mode_t mode, proc_handler *proc_handler) | |
6883 | { | |
e692ab53 NP |
6884 | entry->procname = procname; |
6885 | entry->data = data; | |
6886 | entry->maxlen = maxlen; | |
6887 | entry->mode = mode; | |
6888 | entry->proc_handler = proc_handler; | |
6889 | } | |
6890 | ||
6891 | static struct ctl_table * | |
6892 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6893 | { | |
a5d8c348 | 6894 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6895 | |
ad1cdc1d MM |
6896 | if (table == NULL) |
6897 | return NULL; | |
6898 | ||
e0361851 | 6899 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6900 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6901 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6902 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6903 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6904 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6905 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6906 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6907 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6908 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6909 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6910 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6911 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6912 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6913 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6914 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6915 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6916 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6917 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6918 | &sd->cache_nice_tries, |
6919 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6920 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6921 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6922 | set_table_entry(&table[11], "name", sd->name, |
6923 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6924 | /* &table[12] is terminator */ | |
e692ab53 NP |
6925 | |
6926 | return table; | |
6927 | } | |
6928 | ||
9a4e7159 | 6929 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6930 | { |
6931 | struct ctl_table *entry, *table; | |
6932 | struct sched_domain *sd; | |
6933 | int domain_num = 0, i; | |
6934 | char buf[32]; | |
6935 | ||
6936 | for_each_domain(cpu, sd) | |
6937 | domain_num++; | |
6938 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6939 | if (table == NULL) |
6940 | return NULL; | |
e692ab53 NP |
6941 | |
6942 | i = 0; | |
6943 | for_each_domain(cpu, sd) { | |
6944 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6945 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6946 | entry->mode = 0555; |
e692ab53 NP |
6947 | entry->child = sd_alloc_ctl_domain_table(sd); |
6948 | entry++; | |
6949 | i++; | |
6950 | } | |
6951 | return table; | |
6952 | } | |
6953 | ||
6954 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6955 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6956 | { |
6957 | int i, cpu_num = num_online_cpus(); | |
6958 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6959 | char buf[32]; | |
6960 | ||
7378547f MM |
6961 | WARN_ON(sd_ctl_dir[0].child); |
6962 | sd_ctl_dir[0].child = entry; | |
6963 | ||
ad1cdc1d MM |
6964 | if (entry == NULL) |
6965 | return; | |
6966 | ||
97b6ea7b | 6967 | for_each_online_cpu(i) { |
e692ab53 | 6968 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6969 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6970 | entry->mode = 0555; |
e692ab53 | 6971 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6972 | entry++; |
e692ab53 | 6973 | } |
7378547f MM |
6974 | |
6975 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6976 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6977 | } | |
6382bc90 | 6978 | |
7378547f | 6979 | /* may be called multiple times per register */ |
6382bc90 MM |
6980 | static void unregister_sched_domain_sysctl(void) |
6981 | { | |
7378547f MM |
6982 | if (sd_sysctl_header) |
6983 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6984 | sd_sysctl_header = NULL; |
7378547f MM |
6985 | if (sd_ctl_dir[0].child) |
6986 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6987 | } |
e692ab53 | 6988 | #else |
6382bc90 MM |
6989 | static void register_sched_domain_sysctl(void) |
6990 | { | |
6991 | } | |
6992 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6993 | { |
6994 | } | |
6995 | #endif | |
6996 | ||
1f11eb6a GH |
6997 | static void set_rq_online(struct rq *rq) |
6998 | { | |
6999 | if (!rq->online) { | |
7000 | const struct sched_class *class; | |
7001 | ||
c6c4927b | 7002 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7003 | rq->online = 1; |
7004 | ||
7005 | for_each_class(class) { | |
7006 | if (class->rq_online) | |
7007 | class->rq_online(rq); | |
7008 | } | |
7009 | } | |
7010 | } | |
7011 | ||
7012 | static void set_rq_offline(struct rq *rq) | |
7013 | { | |
7014 | if (rq->online) { | |
7015 | const struct sched_class *class; | |
7016 | ||
7017 | for_each_class(class) { | |
7018 | if (class->rq_offline) | |
7019 | class->rq_offline(rq); | |
7020 | } | |
7021 | ||
c6c4927b | 7022 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7023 | rq->online = 0; |
7024 | } | |
7025 | } | |
7026 | ||
1da177e4 LT |
7027 | /* |
7028 | * migration_call - callback that gets triggered when a CPU is added. | |
7029 | * Here we can start up the necessary migration thread for the new CPU. | |
7030 | */ | |
48f24c4d IM |
7031 | static int __cpuinit |
7032 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7033 | { |
1da177e4 | 7034 | struct task_struct *p; |
48f24c4d | 7035 | int cpu = (long)hcpu; |
1da177e4 | 7036 | unsigned long flags; |
70b97a7f | 7037 | struct rq *rq; |
1da177e4 LT |
7038 | |
7039 | switch (action) { | |
5be9361c | 7040 | |
1da177e4 | 7041 | case CPU_UP_PREPARE: |
8bb78442 | 7042 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7043 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7044 | if (IS_ERR(p)) |
7045 | return NOTIFY_BAD; | |
1da177e4 LT |
7046 | kthread_bind(p, cpu); |
7047 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7048 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7049 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7050 | task_rq_unlock(rq, &flags); |
7051 | cpu_rq(cpu)->migration_thread = p; | |
7052 | break; | |
48f24c4d | 7053 | |
1da177e4 | 7054 | case CPU_ONLINE: |
8bb78442 | 7055 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7056 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7057 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7058 | |
7059 | /* Update our root-domain */ | |
7060 | rq = cpu_rq(cpu); | |
7061 | spin_lock_irqsave(&rq->lock, flags); | |
7062 | if (rq->rd) { | |
c6c4927b | 7063 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7064 | |
7065 | set_rq_online(rq); | |
1f94ef59 GH |
7066 | } |
7067 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7068 | break; |
48f24c4d | 7069 | |
1da177e4 LT |
7070 | #ifdef CONFIG_HOTPLUG_CPU |
7071 | case CPU_UP_CANCELED: | |
8bb78442 | 7072 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7073 | if (!cpu_rq(cpu)->migration_thread) |
7074 | break; | |
41a2d6cf | 7075 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7076 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7077 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7078 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7079 | cpu_rq(cpu)->migration_thread = NULL; | |
7080 | break; | |
48f24c4d | 7081 | |
1da177e4 | 7082 | case CPU_DEAD: |
8bb78442 | 7083 | case CPU_DEAD_FROZEN: |
470fd646 | 7084 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7085 | migrate_live_tasks(cpu); |
7086 | rq = cpu_rq(cpu); | |
7087 | kthread_stop(rq->migration_thread); | |
7088 | rq->migration_thread = NULL; | |
7089 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7090 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7091 | update_rq_clock(rq); |
2e1cb74a | 7092 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7093 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7094 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7095 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7096 | migrate_dead_tasks(cpu); |
d2da272a | 7097 | spin_unlock_irq(&rq->lock); |
470fd646 | 7098 | cpuset_unlock(); |
1da177e4 LT |
7099 | migrate_nr_uninterruptible(rq); |
7100 | BUG_ON(rq->nr_running != 0); | |
7101 | ||
41a2d6cf IM |
7102 | /* |
7103 | * No need to migrate the tasks: it was best-effort if | |
7104 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7105 | * the requestors. | |
7106 | */ | |
1da177e4 LT |
7107 | spin_lock_irq(&rq->lock); |
7108 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7109 | struct migration_req *req; |
7110 | ||
1da177e4 | 7111 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7112 | struct migration_req, list); |
1da177e4 | 7113 | list_del_init(&req->list); |
9a2bd244 | 7114 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7115 | complete(&req->done); |
9a2bd244 | 7116 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7117 | } |
7118 | spin_unlock_irq(&rq->lock); | |
7119 | break; | |
57d885fe | 7120 | |
08f503b0 GH |
7121 | case CPU_DYING: |
7122 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7123 | /* Update our root-domain */ |
7124 | rq = cpu_rq(cpu); | |
7125 | spin_lock_irqsave(&rq->lock, flags); | |
7126 | if (rq->rd) { | |
c6c4927b | 7127 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7128 | set_rq_offline(rq); |
57d885fe GH |
7129 | } |
7130 | spin_unlock_irqrestore(&rq->lock, flags); | |
7131 | break; | |
1da177e4 LT |
7132 | #endif |
7133 | } | |
7134 | return NOTIFY_OK; | |
7135 | } | |
7136 | ||
7137 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
7138 | * happens before everything else. | |
7139 | */ | |
26c2143b | 7140 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7141 | .notifier_call = migration_call, |
7142 | .priority = 10 | |
7143 | }; | |
7144 | ||
7babe8db | 7145 | static int __init migration_init(void) |
1da177e4 LT |
7146 | { |
7147 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7148 | int err; |
48f24c4d IM |
7149 | |
7150 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7151 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7152 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7153 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7154 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7155 | |
7156 | return err; | |
1da177e4 | 7157 | } |
7babe8db | 7158 | early_initcall(migration_init); |
1da177e4 LT |
7159 | #endif |
7160 | ||
7161 | #ifdef CONFIG_SMP | |
476f3534 | 7162 | |
3e9830dc | 7163 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7164 | |
7c16ec58 | 7165 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7166 | struct cpumask *groupmask) |
1da177e4 | 7167 | { |
4dcf6aff | 7168 | struct sched_group *group = sd->groups; |
434d53b0 | 7169 | char str[256]; |
1da177e4 | 7170 | |
968ea6d8 | 7171 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7172 | cpumask_clear(groupmask); |
4dcf6aff IM |
7173 | |
7174 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7175 | ||
7176 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7177 | printk("does not load-balance\n"); | |
7178 | if (sd->parent) | |
7179 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7180 | " has parent"); | |
7181 | return -1; | |
41c7ce9a NP |
7182 | } |
7183 | ||
eefd796a | 7184 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7185 | |
758b2cdc | 7186 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7187 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7188 | "CPU%d\n", cpu); | |
7189 | } | |
758b2cdc | 7190 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7191 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7192 | " CPU%d\n", cpu); | |
7193 | } | |
1da177e4 | 7194 | |
4dcf6aff | 7195 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7196 | do { |
4dcf6aff IM |
7197 | if (!group) { |
7198 | printk("\n"); | |
7199 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7200 | break; |
7201 | } | |
7202 | ||
4dcf6aff IM |
7203 | if (!group->__cpu_power) { |
7204 | printk(KERN_CONT "\n"); | |
7205 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7206 | "set\n"); | |
7207 | break; | |
7208 | } | |
1da177e4 | 7209 | |
758b2cdc | 7210 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7211 | printk(KERN_CONT "\n"); |
7212 | printk(KERN_ERR "ERROR: empty group\n"); | |
7213 | break; | |
7214 | } | |
1da177e4 | 7215 | |
758b2cdc | 7216 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7217 | printk(KERN_CONT "\n"); |
7218 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7219 | break; | |
7220 | } | |
1da177e4 | 7221 | |
758b2cdc | 7222 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7223 | |
968ea6d8 | 7224 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
4dcf6aff | 7225 | printk(KERN_CONT " %s", str); |
1da177e4 | 7226 | |
4dcf6aff IM |
7227 | group = group->next; |
7228 | } while (group != sd->groups); | |
7229 | printk(KERN_CONT "\n"); | |
1da177e4 | 7230 | |
758b2cdc | 7231 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7232 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7233 | |
758b2cdc RR |
7234 | if (sd->parent && |
7235 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7236 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7237 | "of domain->span\n"); | |
7238 | return 0; | |
7239 | } | |
1da177e4 | 7240 | |
4dcf6aff IM |
7241 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7242 | { | |
d5dd3db1 | 7243 | cpumask_var_t groupmask; |
4dcf6aff | 7244 | int level = 0; |
1da177e4 | 7245 | |
4dcf6aff IM |
7246 | if (!sd) { |
7247 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7248 | return; | |
7249 | } | |
1da177e4 | 7250 | |
4dcf6aff IM |
7251 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7252 | ||
d5dd3db1 | 7253 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7254 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7255 | return; | |
7256 | } | |
7257 | ||
4dcf6aff | 7258 | for (;;) { |
7c16ec58 | 7259 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7260 | break; |
1da177e4 LT |
7261 | level++; |
7262 | sd = sd->parent; | |
33859f7f | 7263 | if (!sd) |
4dcf6aff IM |
7264 | break; |
7265 | } | |
d5dd3db1 | 7266 | free_cpumask_var(groupmask); |
1da177e4 | 7267 | } |
6d6bc0ad | 7268 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7269 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7270 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7271 | |
1a20ff27 | 7272 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7273 | { |
758b2cdc | 7274 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7275 | return 1; |
7276 | ||
7277 | /* Following flags need at least 2 groups */ | |
7278 | if (sd->flags & (SD_LOAD_BALANCE | | |
7279 | SD_BALANCE_NEWIDLE | | |
7280 | SD_BALANCE_FORK | | |
89c4710e SS |
7281 | SD_BALANCE_EXEC | |
7282 | SD_SHARE_CPUPOWER | | |
7283 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7284 | if (sd->groups != sd->groups->next) |
7285 | return 0; | |
7286 | } | |
7287 | ||
7288 | /* Following flags don't use groups */ | |
7289 | if (sd->flags & (SD_WAKE_IDLE | | |
7290 | SD_WAKE_AFFINE | | |
7291 | SD_WAKE_BALANCE)) | |
7292 | return 0; | |
7293 | ||
7294 | return 1; | |
7295 | } | |
7296 | ||
48f24c4d IM |
7297 | static int |
7298 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7299 | { |
7300 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7301 | ||
7302 | if (sd_degenerate(parent)) | |
7303 | return 1; | |
7304 | ||
758b2cdc | 7305 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7306 | return 0; |
7307 | ||
7308 | /* Does parent contain flags not in child? */ | |
7309 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7310 | if (cflags & SD_WAKE_AFFINE) | |
7311 | pflags &= ~SD_WAKE_BALANCE; | |
7312 | /* Flags needing groups don't count if only 1 group in parent */ | |
7313 | if (parent->groups == parent->groups->next) { | |
7314 | pflags &= ~(SD_LOAD_BALANCE | | |
7315 | SD_BALANCE_NEWIDLE | | |
7316 | SD_BALANCE_FORK | | |
89c4710e SS |
7317 | SD_BALANCE_EXEC | |
7318 | SD_SHARE_CPUPOWER | | |
7319 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7320 | if (nr_node_ids == 1) |
7321 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7322 | } |
7323 | if (~cflags & pflags) | |
7324 | return 0; | |
7325 | ||
7326 | return 1; | |
7327 | } | |
7328 | ||
c6c4927b RR |
7329 | static void free_rootdomain(struct root_domain *rd) |
7330 | { | |
68e74568 RR |
7331 | cpupri_cleanup(&rd->cpupri); |
7332 | ||
c6c4927b RR |
7333 | free_cpumask_var(rd->rto_mask); |
7334 | free_cpumask_var(rd->online); | |
7335 | free_cpumask_var(rd->span); | |
7336 | kfree(rd); | |
7337 | } | |
7338 | ||
57d885fe GH |
7339 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7340 | { | |
a0490fa3 | 7341 | struct root_domain *old_rd = NULL; |
57d885fe | 7342 | unsigned long flags; |
57d885fe GH |
7343 | |
7344 | spin_lock_irqsave(&rq->lock, flags); | |
7345 | ||
7346 | if (rq->rd) { | |
a0490fa3 | 7347 | old_rd = rq->rd; |
57d885fe | 7348 | |
c6c4927b | 7349 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7350 | set_rq_offline(rq); |
57d885fe | 7351 | |
c6c4927b | 7352 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7353 | |
a0490fa3 IM |
7354 | /* |
7355 | * If we dont want to free the old_rt yet then | |
7356 | * set old_rd to NULL to skip the freeing later | |
7357 | * in this function: | |
7358 | */ | |
7359 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7360 | old_rd = NULL; | |
57d885fe GH |
7361 | } |
7362 | ||
7363 | atomic_inc(&rd->refcount); | |
7364 | rq->rd = rd; | |
7365 | ||
c6c4927b RR |
7366 | cpumask_set_cpu(rq->cpu, rd->span); |
7367 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7368 | set_rq_online(rq); |
57d885fe GH |
7369 | |
7370 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7371 | |
7372 | if (old_rd) | |
7373 | free_rootdomain(old_rd); | |
57d885fe GH |
7374 | } |
7375 | ||
db2f59c8 | 7376 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7377 | { |
7378 | memset(rd, 0, sizeof(*rd)); | |
7379 | ||
c6c4927b RR |
7380 | if (bootmem) { |
7381 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7382 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7383 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7384 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7385 | return 0; |
7386 | } | |
7387 | ||
7388 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7389 | goto out; |
c6c4927b RR |
7390 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7391 | goto free_span; | |
7392 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7393 | goto free_online; | |
6e0534f2 | 7394 | |
68e74568 RR |
7395 | if (cpupri_init(&rd->cpupri, false) != 0) |
7396 | goto free_rto_mask; | |
c6c4927b | 7397 | return 0; |
6e0534f2 | 7398 | |
68e74568 RR |
7399 | free_rto_mask: |
7400 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7401 | free_online: |
7402 | free_cpumask_var(rd->online); | |
7403 | free_span: | |
7404 | free_cpumask_var(rd->span); | |
0c910d28 | 7405 | out: |
c6c4927b | 7406 | return -ENOMEM; |
57d885fe GH |
7407 | } |
7408 | ||
7409 | static void init_defrootdomain(void) | |
7410 | { | |
c6c4927b RR |
7411 | init_rootdomain(&def_root_domain, true); |
7412 | ||
57d885fe GH |
7413 | atomic_set(&def_root_domain.refcount, 1); |
7414 | } | |
7415 | ||
dc938520 | 7416 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7417 | { |
7418 | struct root_domain *rd; | |
7419 | ||
7420 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7421 | if (!rd) | |
7422 | return NULL; | |
7423 | ||
c6c4927b RR |
7424 | if (init_rootdomain(rd, false) != 0) { |
7425 | kfree(rd); | |
7426 | return NULL; | |
7427 | } | |
57d885fe GH |
7428 | |
7429 | return rd; | |
7430 | } | |
7431 | ||
1da177e4 | 7432 | /* |
0eab9146 | 7433 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7434 | * hold the hotplug lock. |
7435 | */ | |
0eab9146 IM |
7436 | static void |
7437 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7438 | { |
70b97a7f | 7439 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7440 | struct sched_domain *tmp; |
7441 | ||
7442 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7443 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7444 | struct sched_domain *parent = tmp->parent; |
7445 | if (!parent) | |
7446 | break; | |
f29c9b1c | 7447 | |
1a848870 | 7448 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7449 | tmp->parent = parent->parent; |
1a848870 SS |
7450 | if (parent->parent) |
7451 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7452 | } else |
7453 | tmp = tmp->parent; | |
245af2c7 SS |
7454 | } |
7455 | ||
1a848870 | 7456 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7457 | sd = sd->parent; |
1a848870 SS |
7458 | if (sd) |
7459 | sd->child = NULL; | |
7460 | } | |
1da177e4 LT |
7461 | |
7462 | sched_domain_debug(sd, cpu); | |
7463 | ||
57d885fe | 7464 | rq_attach_root(rq, rd); |
674311d5 | 7465 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7466 | } |
7467 | ||
7468 | /* cpus with isolated domains */ | |
dcc30a35 | 7469 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7470 | |
7471 | /* Setup the mask of cpus configured for isolated domains */ | |
7472 | static int __init isolated_cpu_setup(char *str) | |
7473 | { | |
968ea6d8 | 7474 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7475 | return 1; |
7476 | } | |
7477 | ||
8927f494 | 7478 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7479 | |
7480 | /* | |
6711cab4 SS |
7481 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7482 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7483 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7484 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7485 | * |
7486 | * init_sched_build_groups will build a circular linked list of the groups | |
7487 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7488 | * and ->cpu_power to 0. | |
7489 | */ | |
a616058b | 7490 | static void |
96f874e2 RR |
7491 | init_sched_build_groups(const struct cpumask *span, |
7492 | const struct cpumask *cpu_map, | |
7493 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7494 | struct sched_group **sg, |
96f874e2 RR |
7495 | struct cpumask *tmpmask), |
7496 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7497 | { |
7498 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7499 | int i; |
7500 | ||
96f874e2 | 7501 | cpumask_clear(covered); |
7c16ec58 | 7502 | |
abcd083a | 7503 | for_each_cpu(i, span) { |
6711cab4 | 7504 | struct sched_group *sg; |
7c16ec58 | 7505 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7506 | int j; |
7507 | ||
758b2cdc | 7508 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7509 | continue; |
7510 | ||
758b2cdc | 7511 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7512 | sg->__cpu_power = 0; |
1da177e4 | 7513 | |
abcd083a | 7514 | for_each_cpu(j, span) { |
7c16ec58 | 7515 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7516 | continue; |
7517 | ||
96f874e2 | 7518 | cpumask_set_cpu(j, covered); |
758b2cdc | 7519 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7520 | } |
7521 | if (!first) | |
7522 | first = sg; | |
7523 | if (last) | |
7524 | last->next = sg; | |
7525 | last = sg; | |
7526 | } | |
7527 | last->next = first; | |
7528 | } | |
7529 | ||
9c1cfda2 | 7530 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7531 | |
9c1cfda2 | 7532 | #ifdef CONFIG_NUMA |
198e2f18 | 7533 | |
9c1cfda2 JH |
7534 | /** |
7535 | * find_next_best_node - find the next node to include in a sched_domain | |
7536 | * @node: node whose sched_domain we're building | |
7537 | * @used_nodes: nodes already in the sched_domain | |
7538 | * | |
41a2d6cf | 7539 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7540 | * finds the closest node not already in the @used_nodes map. |
7541 | * | |
7542 | * Should use nodemask_t. | |
7543 | */ | |
c5f59f08 | 7544 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7545 | { |
7546 | int i, n, val, min_val, best_node = 0; | |
7547 | ||
7548 | min_val = INT_MAX; | |
7549 | ||
076ac2af | 7550 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7551 | /* Start at @node */ |
076ac2af | 7552 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7553 | |
7554 | if (!nr_cpus_node(n)) | |
7555 | continue; | |
7556 | ||
7557 | /* Skip already used nodes */ | |
c5f59f08 | 7558 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7559 | continue; |
7560 | ||
7561 | /* Simple min distance search */ | |
7562 | val = node_distance(node, n); | |
7563 | ||
7564 | if (val < min_val) { | |
7565 | min_val = val; | |
7566 | best_node = n; | |
7567 | } | |
7568 | } | |
7569 | ||
c5f59f08 | 7570 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7571 | return best_node; |
7572 | } | |
7573 | ||
7574 | /** | |
7575 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7576 | * @node: node whose cpumask we're constructing | |
73486722 | 7577 | * @span: resulting cpumask |
9c1cfda2 | 7578 | * |
41a2d6cf | 7579 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7580 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7581 | * out optimally. | |
7582 | */ | |
96f874e2 | 7583 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7584 | { |
c5f59f08 | 7585 | nodemask_t used_nodes; |
48f24c4d | 7586 | int i; |
9c1cfda2 | 7587 | |
6ca09dfc | 7588 | cpumask_clear(span); |
c5f59f08 | 7589 | nodes_clear(used_nodes); |
9c1cfda2 | 7590 | |
6ca09dfc | 7591 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7592 | node_set(node, used_nodes); |
9c1cfda2 JH |
7593 | |
7594 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7595 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7596 | |
6ca09dfc | 7597 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7598 | } |
9c1cfda2 | 7599 | } |
6d6bc0ad | 7600 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7601 | |
5c45bf27 | 7602 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7603 | |
6c99e9ad RR |
7604 | /* |
7605 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7606 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7607 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7608 | */ | |
7609 | struct static_sched_group { | |
7610 | struct sched_group sg; | |
7611 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7612 | }; | |
7613 | ||
7614 | struct static_sched_domain { | |
7615 | struct sched_domain sd; | |
7616 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7617 | }; | |
7618 | ||
9c1cfda2 | 7619 | /* |
48f24c4d | 7620 | * SMT sched-domains: |
9c1cfda2 | 7621 | */ |
1da177e4 | 7622 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7623 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7624 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7625 | |
41a2d6cf | 7626 | static int |
96f874e2 RR |
7627 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7628 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7629 | { |
6711cab4 | 7630 | if (sg) |
6c99e9ad | 7631 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7632 | return cpu; |
7633 | } | |
6d6bc0ad | 7634 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7635 | |
48f24c4d IM |
7636 | /* |
7637 | * multi-core sched-domains: | |
7638 | */ | |
1e9f28fa | 7639 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7640 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7641 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7642 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7643 | |
7644 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7645 | static int |
96f874e2 RR |
7646 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7647 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7648 | { |
6711cab4 | 7649 | int group; |
7c16ec58 | 7650 | |
96f874e2 RR |
7651 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7652 | group = cpumask_first(mask); | |
6711cab4 | 7653 | if (sg) |
6c99e9ad | 7654 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7655 | return group; |
1e9f28fa SS |
7656 | } |
7657 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7658 | static int |
96f874e2 RR |
7659 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7660 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7661 | { |
6711cab4 | 7662 | if (sg) |
6c99e9ad | 7663 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7664 | return cpu; |
7665 | } | |
7666 | #endif | |
7667 | ||
6c99e9ad RR |
7668 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7669 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7670 | |
41a2d6cf | 7671 | static int |
96f874e2 RR |
7672 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7673 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7674 | { |
6711cab4 | 7675 | int group; |
48f24c4d | 7676 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7677 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7678 | group = cpumask_first(mask); |
1e9f28fa | 7679 | #elif defined(CONFIG_SCHED_SMT) |
96f874e2 RR |
7680 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7681 | group = cpumask_first(mask); | |
1da177e4 | 7682 | #else |
6711cab4 | 7683 | group = cpu; |
1da177e4 | 7684 | #endif |
6711cab4 | 7685 | if (sg) |
6c99e9ad | 7686 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7687 | return group; |
1da177e4 LT |
7688 | } |
7689 | ||
7690 | #ifdef CONFIG_NUMA | |
1da177e4 | 7691 | /* |
9c1cfda2 JH |
7692 | * The init_sched_build_groups can't handle what we want to do with node |
7693 | * groups, so roll our own. Now each node has its own list of groups which | |
7694 | * gets dynamically allocated. | |
1da177e4 | 7695 | */ |
62ea9ceb | 7696 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7697 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7698 | |
62ea9ceb | 7699 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7700 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7701 | |
96f874e2 RR |
7702 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7703 | struct sched_group **sg, | |
7704 | struct cpumask *nodemask) | |
9c1cfda2 | 7705 | { |
6711cab4 SS |
7706 | int group; |
7707 | ||
6ca09dfc | 7708 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7709 | group = cpumask_first(nodemask); |
6711cab4 SS |
7710 | |
7711 | if (sg) | |
6c99e9ad | 7712 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7713 | return group; |
1da177e4 | 7714 | } |
6711cab4 | 7715 | |
08069033 SS |
7716 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7717 | { | |
7718 | struct sched_group *sg = group_head; | |
7719 | int j; | |
7720 | ||
7721 | if (!sg) | |
7722 | return; | |
3a5c359a | 7723 | do { |
758b2cdc | 7724 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7725 | struct sched_domain *sd; |
08069033 | 7726 | |
6c99e9ad | 7727 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7728 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7729 | /* |
7730 | * Only add "power" once for each | |
7731 | * physical package. | |
7732 | */ | |
7733 | continue; | |
7734 | } | |
08069033 | 7735 | |
3a5c359a AK |
7736 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7737 | } | |
7738 | sg = sg->next; | |
7739 | } while (sg != group_head); | |
08069033 | 7740 | } |
6d6bc0ad | 7741 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7742 | |
a616058b | 7743 | #ifdef CONFIG_NUMA |
51888ca2 | 7744 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7745 | static void free_sched_groups(const struct cpumask *cpu_map, |
7746 | struct cpumask *nodemask) | |
51888ca2 | 7747 | { |
a616058b | 7748 | int cpu, i; |
51888ca2 | 7749 | |
abcd083a | 7750 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7751 | struct sched_group **sched_group_nodes |
7752 | = sched_group_nodes_bycpu[cpu]; | |
7753 | ||
51888ca2 SV |
7754 | if (!sched_group_nodes) |
7755 | continue; | |
7756 | ||
076ac2af | 7757 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7758 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7759 | ||
6ca09dfc | 7760 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7761 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7762 | continue; |
7763 | ||
7764 | if (sg == NULL) | |
7765 | continue; | |
7766 | sg = sg->next; | |
7767 | next_sg: | |
7768 | oldsg = sg; | |
7769 | sg = sg->next; | |
7770 | kfree(oldsg); | |
7771 | if (oldsg != sched_group_nodes[i]) | |
7772 | goto next_sg; | |
7773 | } | |
7774 | kfree(sched_group_nodes); | |
7775 | sched_group_nodes_bycpu[cpu] = NULL; | |
7776 | } | |
51888ca2 | 7777 | } |
6d6bc0ad | 7778 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7779 | static void free_sched_groups(const struct cpumask *cpu_map, |
7780 | struct cpumask *nodemask) | |
a616058b SS |
7781 | { |
7782 | } | |
6d6bc0ad | 7783 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7784 | |
89c4710e SS |
7785 | /* |
7786 | * Initialize sched groups cpu_power. | |
7787 | * | |
7788 | * cpu_power indicates the capacity of sched group, which is used while | |
7789 | * distributing the load between different sched groups in a sched domain. | |
7790 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7791 | * there are asymmetries in the topology. If there are asymmetries, group | |
7792 | * having more cpu_power will pickup more load compared to the group having | |
7793 | * less cpu_power. | |
7794 | * | |
7795 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7796 | * the maximum number of tasks a group can handle in the presence of other idle | |
7797 | * or lightly loaded groups in the same sched domain. | |
7798 | */ | |
7799 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7800 | { | |
7801 | struct sched_domain *child; | |
7802 | struct sched_group *group; | |
7803 | ||
7804 | WARN_ON(!sd || !sd->groups); | |
7805 | ||
758b2cdc | 7806 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
7807 | return; |
7808 | ||
7809 | child = sd->child; | |
7810 | ||
5517d86b ED |
7811 | sd->groups->__cpu_power = 0; |
7812 | ||
89c4710e SS |
7813 | /* |
7814 | * For perf policy, if the groups in child domain share resources | |
7815 | * (for example cores sharing some portions of the cache hierarchy | |
7816 | * or SMT), then set this domain groups cpu_power such that each group | |
7817 | * can handle only one task, when there are other idle groups in the | |
7818 | * same sched domain. | |
7819 | */ | |
7820 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7821 | (child->flags & | |
7822 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7823 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7824 | return; |
7825 | } | |
7826 | ||
89c4710e SS |
7827 | /* |
7828 | * add cpu_power of each child group to this groups cpu_power | |
7829 | */ | |
7830 | group = child->groups; | |
7831 | do { | |
5517d86b | 7832 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7833 | group = group->next; |
7834 | } while (group != child->groups); | |
7835 | } | |
7836 | ||
7c16ec58 MT |
7837 | /* |
7838 | * Initializers for schedule domains | |
7839 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7840 | */ | |
7841 | ||
a5d8c348 IM |
7842 | #ifdef CONFIG_SCHED_DEBUG |
7843 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7844 | #else | |
7845 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7846 | #endif | |
7847 | ||
7c16ec58 | 7848 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7849 | |
7c16ec58 MT |
7850 | #define SD_INIT_FUNC(type) \ |
7851 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7852 | { \ | |
7853 | memset(sd, 0, sizeof(*sd)); \ | |
7854 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7855 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7856 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7857 | } |
7858 | ||
7859 | SD_INIT_FUNC(CPU) | |
7860 | #ifdef CONFIG_NUMA | |
7861 | SD_INIT_FUNC(ALLNODES) | |
7862 | SD_INIT_FUNC(NODE) | |
7863 | #endif | |
7864 | #ifdef CONFIG_SCHED_SMT | |
7865 | SD_INIT_FUNC(SIBLING) | |
7866 | #endif | |
7867 | #ifdef CONFIG_SCHED_MC | |
7868 | SD_INIT_FUNC(MC) | |
7869 | #endif | |
7870 | ||
1d3504fc HS |
7871 | static int default_relax_domain_level = -1; |
7872 | ||
7873 | static int __init setup_relax_domain_level(char *str) | |
7874 | { | |
30e0e178 LZ |
7875 | unsigned long val; |
7876 | ||
7877 | val = simple_strtoul(str, NULL, 0); | |
7878 | if (val < SD_LV_MAX) | |
7879 | default_relax_domain_level = val; | |
7880 | ||
1d3504fc HS |
7881 | return 1; |
7882 | } | |
7883 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7884 | ||
7885 | static void set_domain_attribute(struct sched_domain *sd, | |
7886 | struct sched_domain_attr *attr) | |
7887 | { | |
7888 | int request; | |
7889 | ||
7890 | if (!attr || attr->relax_domain_level < 0) { | |
7891 | if (default_relax_domain_level < 0) | |
7892 | return; | |
7893 | else | |
7894 | request = default_relax_domain_level; | |
7895 | } else | |
7896 | request = attr->relax_domain_level; | |
7897 | if (request < sd->level) { | |
7898 | /* turn off idle balance on this domain */ | |
7899 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7900 | } else { | |
7901 | /* turn on idle balance on this domain */ | |
7902 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7903 | } | |
7904 | } | |
7905 | ||
1da177e4 | 7906 | /* |
1a20ff27 DG |
7907 | * Build sched domains for a given set of cpus and attach the sched domains |
7908 | * to the individual cpus | |
1da177e4 | 7909 | */ |
96f874e2 | 7910 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 7911 | struct sched_domain_attr *attr) |
1da177e4 | 7912 | { |
3404c8d9 | 7913 | int i, err = -ENOMEM; |
57d885fe | 7914 | struct root_domain *rd; |
3404c8d9 RR |
7915 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
7916 | tmpmask; | |
d1b55138 | 7917 | #ifdef CONFIG_NUMA |
3404c8d9 | 7918 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 7919 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 7920 | int sd_allnodes = 0; |
d1b55138 | 7921 | |
3404c8d9 RR |
7922 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
7923 | goto out; | |
7924 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
7925 | goto free_domainspan; | |
7926 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
7927 | goto free_covered; | |
7928 | #endif | |
7929 | ||
7930 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
7931 | goto free_notcovered; | |
7932 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
7933 | goto free_nodemask; | |
7934 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
7935 | goto free_this_sibling_map; | |
7936 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
7937 | goto free_this_core_map; | |
7938 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
7939 | goto free_send_covered; | |
7940 | ||
7941 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
7942 | /* |
7943 | * Allocate the per-node list of sched groups | |
7944 | */ | |
076ac2af | 7945 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7946 | GFP_KERNEL); |
d1b55138 JH |
7947 | if (!sched_group_nodes) { |
7948 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 7949 | goto free_tmpmask; |
d1b55138 | 7950 | } |
d1b55138 | 7951 | #endif |
1da177e4 | 7952 | |
dc938520 | 7953 | rd = alloc_rootdomain(); |
57d885fe GH |
7954 | if (!rd) { |
7955 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 7956 | goto free_sched_groups; |
57d885fe GH |
7957 | } |
7958 | ||
7c16ec58 | 7959 | #ifdef CONFIG_NUMA |
96f874e2 | 7960 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
7961 | #endif |
7962 | ||
1da177e4 | 7963 | /* |
1a20ff27 | 7964 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7965 | */ |
abcd083a | 7966 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7967 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 7968 | |
6ca09dfc | 7969 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
7970 | |
7971 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
7972 | if (cpumask_weight(cpu_map) > |
7973 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 7974 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 7975 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7976 | set_domain_attribute(sd, attr); |
758b2cdc | 7977 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 7978 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7979 | p = sd; |
6711cab4 | 7980 | sd_allnodes = 1; |
9c1cfda2 JH |
7981 | } else |
7982 | p = NULL; | |
7983 | ||
62ea9ceb | 7984 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 7985 | SD_INIT(sd, NODE); |
1d3504fc | 7986 | set_domain_attribute(sd, attr); |
758b2cdc | 7987 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 7988 | sd->parent = p; |
1a848870 SS |
7989 | if (p) |
7990 | p->child = sd; | |
758b2cdc RR |
7991 | cpumask_and(sched_domain_span(sd), |
7992 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
7993 | #endif |
7994 | ||
7995 | p = sd; | |
6c99e9ad | 7996 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 7997 | SD_INIT(sd, CPU); |
1d3504fc | 7998 | set_domain_attribute(sd, attr); |
758b2cdc | 7999 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8000 | sd->parent = p; |
1a848870 SS |
8001 | if (p) |
8002 | p->child = sd; | |
7c16ec58 | 8003 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8004 | |
1e9f28fa SS |
8005 | #ifdef CONFIG_SCHED_MC |
8006 | p = sd; | |
6c99e9ad | 8007 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8008 | SD_INIT(sd, MC); |
1d3504fc | 8009 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8010 | cpumask_and(sched_domain_span(sd), cpu_map, |
8011 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8012 | sd->parent = p; |
1a848870 | 8013 | p->child = sd; |
7c16ec58 | 8014 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8015 | #endif |
8016 | ||
1da177e4 LT |
8017 | #ifdef CONFIG_SCHED_SMT |
8018 | p = sd; | |
6c99e9ad | 8019 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8020 | SD_INIT(sd, SIBLING); |
1d3504fc | 8021 | set_domain_attribute(sd, attr); |
758b2cdc RR |
8022 | cpumask_and(sched_domain_span(sd), |
8023 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
1da177e4 | 8024 | sd->parent = p; |
1a848870 | 8025 | p->child = sd; |
7c16ec58 | 8026 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8027 | #endif |
8028 | } | |
8029 | ||
8030 | #ifdef CONFIG_SCHED_SMT | |
8031 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8032 | for_each_cpu(i, cpu_map) { |
96f874e2 RR |
8033 | cpumask_and(this_sibling_map, |
8034 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
8035 | if (i != cpumask_first(this_sibling_map)) | |
1da177e4 LT |
8036 | continue; |
8037 | ||
dd41f596 | 8038 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8039 | &cpu_to_cpu_group, |
8040 | send_covered, tmpmask); | |
1da177e4 LT |
8041 | } |
8042 | #endif | |
8043 | ||
1e9f28fa SS |
8044 | #ifdef CONFIG_SCHED_MC |
8045 | /* Set up multi-core groups */ | |
abcd083a | 8046 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8047 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8048 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8049 | continue; |
7c16ec58 | 8050 | |
dd41f596 | 8051 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8052 | &cpu_to_core_group, |
8053 | send_covered, tmpmask); | |
1e9f28fa SS |
8054 | } |
8055 | #endif | |
8056 | ||
1da177e4 | 8057 | /* Set up physical groups */ |
076ac2af | 8058 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8059 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8060 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8061 | continue; |
8062 | ||
7c16ec58 MT |
8063 | init_sched_build_groups(nodemask, cpu_map, |
8064 | &cpu_to_phys_group, | |
8065 | send_covered, tmpmask); | |
1da177e4 LT |
8066 | } |
8067 | ||
8068 | #ifdef CONFIG_NUMA | |
8069 | /* Set up node groups */ | |
7c16ec58 | 8070 | if (sd_allnodes) { |
7c16ec58 MT |
8071 | init_sched_build_groups(cpu_map, cpu_map, |
8072 | &cpu_to_allnodes_group, | |
8073 | send_covered, tmpmask); | |
8074 | } | |
9c1cfda2 | 8075 | |
076ac2af | 8076 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8077 | /* Set up node groups */ |
8078 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8079 | int j; |
8080 | ||
96f874e2 | 8081 | cpumask_clear(covered); |
6ca09dfc | 8082 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8083 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8084 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8085 | continue; |
d1b55138 | 8086 | } |
9c1cfda2 | 8087 | |
4bdbaad3 | 8088 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8089 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8090 | |
6c99e9ad RR |
8091 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8092 | GFP_KERNEL, i); | |
51888ca2 SV |
8093 | if (!sg) { |
8094 | printk(KERN_WARNING "Can not alloc domain group for " | |
8095 | "node %d\n", i); | |
8096 | goto error; | |
8097 | } | |
9c1cfda2 | 8098 | sched_group_nodes[i] = sg; |
abcd083a | 8099 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8100 | struct sched_domain *sd; |
9761eea8 | 8101 | |
62ea9ceb | 8102 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8103 | sd->groups = sg; |
9c1cfda2 | 8104 | } |
5517d86b | 8105 | sg->__cpu_power = 0; |
758b2cdc | 8106 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8107 | sg->next = sg; |
96f874e2 | 8108 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8109 | prev = sg; |
8110 | ||
076ac2af | 8111 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8112 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8113 | |
96f874e2 RR |
8114 | cpumask_complement(notcovered, covered); |
8115 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8116 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8117 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8118 | break; |
8119 | ||
6ca09dfc | 8120 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8121 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8122 | continue; |
8123 | ||
6c99e9ad RR |
8124 | sg = kmalloc_node(sizeof(struct sched_group) + |
8125 | cpumask_size(), | |
15f0b676 | 8126 | GFP_KERNEL, i); |
9c1cfda2 JH |
8127 | if (!sg) { |
8128 | printk(KERN_WARNING | |
8129 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8130 | goto error; |
9c1cfda2 | 8131 | } |
5517d86b | 8132 | sg->__cpu_power = 0; |
758b2cdc | 8133 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8134 | sg->next = prev->next; |
96f874e2 | 8135 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8136 | prev->next = sg; |
8137 | prev = sg; | |
8138 | } | |
9c1cfda2 | 8139 | } |
1da177e4 LT |
8140 | #endif |
8141 | ||
8142 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8143 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8144 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8145 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8146 | |
89c4710e | 8147 | init_sched_groups_power(i, sd); |
5c45bf27 | 8148 | } |
1da177e4 | 8149 | #endif |
1e9f28fa | 8150 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8151 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8152 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8153 | |
89c4710e | 8154 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8155 | } |
8156 | #endif | |
1e9f28fa | 8157 | |
abcd083a | 8158 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8159 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8160 | |
89c4710e | 8161 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8162 | } |
8163 | ||
9c1cfda2 | 8164 | #ifdef CONFIG_NUMA |
076ac2af | 8165 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8166 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8167 | |
6711cab4 SS |
8168 | if (sd_allnodes) { |
8169 | struct sched_group *sg; | |
f712c0c7 | 8170 | |
96f874e2 | 8171 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8172 | tmpmask); |
f712c0c7 SS |
8173 | init_numa_sched_groups_power(sg); |
8174 | } | |
9c1cfda2 JH |
8175 | #endif |
8176 | ||
1da177e4 | 8177 | /* Attach the domains */ |
abcd083a | 8178 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8179 | struct sched_domain *sd; |
8180 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8181 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8182 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8183 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8184 | #else |
6c99e9ad | 8185 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8186 | #endif |
57d885fe | 8187 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8188 | } |
51888ca2 | 8189 | |
3404c8d9 RR |
8190 | err = 0; |
8191 | ||
8192 | free_tmpmask: | |
8193 | free_cpumask_var(tmpmask); | |
8194 | free_send_covered: | |
8195 | free_cpumask_var(send_covered); | |
8196 | free_this_core_map: | |
8197 | free_cpumask_var(this_core_map); | |
8198 | free_this_sibling_map: | |
8199 | free_cpumask_var(this_sibling_map); | |
8200 | free_nodemask: | |
8201 | free_cpumask_var(nodemask); | |
8202 | free_notcovered: | |
8203 | #ifdef CONFIG_NUMA | |
8204 | free_cpumask_var(notcovered); | |
8205 | free_covered: | |
8206 | free_cpumask_var(covered); | |
8207 | free_domainspan: | |
8208 | free_cpumask_var(domainspan); | |
8209 | out: | |
8210 | #endif | |
8211 | return err; | |
8212 | ||
8213 | free_sched_groups: | |
8214 | #ifdef CONFIG_NUMA | |
8215 | kfree(sched_group_nodes); | |
8216 | #endif | |
8217 | goto free_tmpmask; | |
51888ca2 | 8218 | |
a616058b | 8219 | #ifdef CONFIG_NUMA |
51888ca2 | 8220 | error: |
7c16ec58 | 8221 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8222 | free_rootdomain(rd); |
3404c8d9 | 8223 | goto free_tmpmask; |
a616058b | 8224 | #endif |
1da177e4 | 8225 | } |
029190c5 | 8226 | |
96f874e2 | 8227 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8228 | { |
8229 | return __build_sched_domains(cpu_map, NULL); | |
8230 | } | |
8231 | ||
96f874e2 | 8232 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8233 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8234 | static struct sched_domain_attr *dattr_cur; |
8235 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8236 | |
8237 | /* | |
8238 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8239 | * cpumask) fails, then fallback to a single sched domain, |
8240 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8241 | */ |
4212823f | 8242 | static cpumask_var_t fallback_doms; |
029190c5 | 8243 | |
ee79d1bd HC |
8244 | /* |
8245 | * arch_update_cpu_topology lets virtualized architectures update the | |
8246 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8247 | * or 0 if it stayed the same. | |
8248 | */ | |
8249 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8250 | { |
ee79d1bd | 8251 | return 0; |
22e52b07 HC |
8252 | } |
8253 | ||
1a20ff27 | 8254 | /* |
41a2d6cf | 8255 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8256 | * For now this just excludes isolated cpus, but could be used to |
8257 | * exclude other special cases in the future. | |
1a20ff27 | 8258 | */ |
96f874e2 | 8259 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8260 | { |
7378547f MM |
8261 | int err; |
8262 | ||
22e52b07 | 8263 | arch_update_cpu_topology(); |
029190c5 | 8264 | ndoms_cur = 1; |
96f874e2 | 8265 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8266 | if (!doms_cur) |
4212823f | 8267 | doms_cur = fallback_doms; |
dcc30a35 | 8268 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8269 | dattr_cur = NULL; |
7378547f | 8270 | err = build_sched_domains(doms_cur); |
6382bc90 | 8271 | register_sched_domain_sysctl(); |
7378547f MM |
8272 | |
8273 | return err; | |
1a20ff27 DG |
8274 | } |
8275 | ||
96f874e2 RR |
8276 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8277 | struct cpumask *tmpmask) | |
1da177e4 | 8278 | { |
7c16ec58 | 8279 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8280 | } |
1da177e4 | 8281 | |
1a20ff27 DG |
8282 | /* |
8283 | * Detach sched domains from a group of cpus specified in cpu_map | |
8284 | * These cpus will now be attached to the NULL domain | |
8285 | */ | |
96f874e2 | 8286 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8287 | { |
96f874e2 RR |
8288 | /* Save because hotplug lock held. */ |
8289 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8290 | int i; |
8291 | ||
abcd083a | 8292 | for_each_cpu(i, cpu_map) |
57d885fe | 8293 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8294 | synchronize_sched(); |
96f874e2 | 8295 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8296 | } |
8297 | ||
1d3504fc HS |
8298 | /* handle null as "default" */ |
8299 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8300 | struct sched_domain_attr *new, int idx_new) | |
8301 | { | |
8302 | struct sched_domain_attr tmp; | |
8303 | ||
8304 | /* fast path */ | |
8305 | if (!new && !cur) | |
8306 | return 1; | |
8307 | ||
8308 | tmp = SD_ATTR_INIT; | |
8309 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8310 | new ? (new + idx_new) : &tmp, | |
8311 | sizeof(struct sched_domain_attr)); | |
8312 | } | |
8313 | ||
029190c5 PJ |
8314 | /* |
8315 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8316 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8317 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8318 | * It destroys each deleted domain and builds each new domain. | |
8319 | * | |
96f874e2 | 8320 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8321 | * The masks don't intersect (don't overlap.) We should setup one |
8322 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8323 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8324 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8325 | * it as it is. | |
8326 | * | |
41a2d6cf IM |
8327 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8328 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8329 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8330 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8331 | * the single partition 'fallback_doms', it also forces the domains | |
8332 | * to be rebuilt. | |
029190c5 | 8333 | * |
96f874e2 | 8334 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8335 | * ndoms_new == 0 is a special case for destroying existing domains, |
8336 | * and it will not create the default domain. | |
dfb512ec | 8337 | * |
029190c5 PJ |
8338 | * Call with hotplug lock held |
8339 | */ | |
96f874e2 RR |
8340 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8341 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8342 | struct sched_domain_attr *dattr_new) |
029190c5 | 8343 | { |
dfb512ec | 8344 | int i, j, n; |
d65bd5ec | 8345 | int new_topology; |
029190c5 | 8346 | |
712555ee | 8347 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8348 | |
7378547f MM |
8349 | /* always unregister in case we don't destroy any domains */ |
8350 | unregister_sched_domain_sysctl(); | |
8351 | ||
d65bd5ec HC |
8352 | /* Let architecture update cpu core mappings. */ |
8353 | new_topology = arch_update_cpu_topology(); | |
8354 | ||
dfb512ec | 8355 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8356 | |
8357 | /* Destroy deleted domains */ | |
8358 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8359 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8360 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8361 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8362 | goto match1; |
8363 | } | |
8364 | /* no match - a current sched domain not in new doms_new[] */ | |
8365 | detach_destroy_domains(doms_cur + i); | |
8366 | match1: | |
8367 | ; | |
8368 | } | |
8369 | ||
e761b772 MK |
8370 | if (doms_new == NULL) { |
8371 | ndoms_cur = 0; | |
4212823f | 8372 | doms_new = fallback_doms; |
dcc30a35 | 8373 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8374 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8375 | } |
8376 | ||
029190c5 PJ |
8377 | /* Build new domains */ |
8378 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8379 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8380 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8381 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8382 | goto match2; |
8383 | } | |
8384 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8385 | __build_sched_domains(doms_new + i, |
8386 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8387 | match2: |
8388 | ; | |
8389 | } | |
8390 | ||
8391 | /* Remember the new sched domains */ | |
4212823f | 8392 | if (doms_cur != fallback_doms) |
029190c5 | 8393 | kfree(doms_cur); |
1d3504fc | 8394 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8395 | doms_cur = doms_new; |
1d3504fc | 8396 | dattr_cur = dattr_new; |
029190c5 | 8397 | ndoms_cur = ndoms_new; |
7378547f MM |
8398 | |
8399 | register_sched_domain_sysctl(); | |
a1835615 | 8400 | |
712555ee | 8401 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8402 | } |
8403 | ||
5c45bf27 | 8404 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8405 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8406 | { |
95402b38 | 8407 | get_online_cpus(); |
dfb512ec MK |
8408 | |
8409 | /* Destroy domains first to force the rebuild */ | |
8410 | partition_sched_domains(0, NULL, NULL); | |
8411 | ||
e761b772 | 8412 | rebuild_sched_domains(); |
95402b38 | 8413 | put_online_cpus(); |
5c45bf27 SS |
8414 | } |
8415 | ||
8416 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8417 | { | |
afb8a9b7 | 8418 | unsigned int level = 0; |
5c45bf27 | 8419 | |
afb8a9b7 GS |
8420 | if (sscanf(buf, "%u", &level) != 1) |
8421 | return -EINVAL; | |
8422 | ||
8423 | /* | |
8424 | * level is always be positive so don't check for | |
8425 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8426 | * What happens on 0 or 1 byte write, | |
8427 | * need to check for count as well? | |
8428 | */ | |
8429 | ||
8430 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8431 | return -EINVAL; |
8432 | ||
8433 | if (smt) | |
afb8a9b7 | 8434 | sched_smt_power_savings = level; |
5c45bf27 | 8435 | else |
afb8a9b7 | 8436 | sched_mc_power_savings = level; |
5c45bf27 | 8437 | |
c70f22d2 | 8438 | arch_reinit_sched_domains(); |
5c45bf27 | 8439 | |
c70f22d2 | 8440 | return count; |
5c45bf27 SS |
8441 | } |
8442 | ||
5c45bf27 | 8443 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8444 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8445 | char *page) | |
5c45bf27 SS |
8446 | { |
8447 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8448 | } | |
f718cd4a | 8449 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8450 | const char *buf, size_t count) |
5c45bf27 SS |
8451 | { |
8452 | return sched_power_savings_store(buf, count, 0); | |
8453 | } | |
f718cd4a AK |
8454 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8455 | sched_mc_power_savings_show, | |
8456 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8457 | #endif |
8458 | ||
8459 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8460 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8461 | char *page) | |
5c45bf27 SS |
8462 | { |
8463 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8464 | } | |
f718cd4a | 8465 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8466 | const char *buf, size_t count) |
5c45bf27 SS |
8467 | { |
8468 | return sched_power_savings_store(buf, count, 1); | |
8469 | } | |
f718cd4a AK |
8470 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8471 | sched_smt_power_savings_show, | |
6707de00 AB |
8472 | sched_smt_power_savings_store); |
8473 | #endif | |
8474 | ||
39aac648 | 8475 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8476 | { |
8477 | int err = 0; | |
8478 | ||
8479 | #ifdef CONFIG_SCHED_SMT | |
8480 | if (smt_capable()) | |
8481 | err = sysfs_create_file(&cls->kset.kobj, | |
8482 | &attr_sched_smt_power_savings.attr); | |
8483 | #endif | |
8484 | #ifdef CONFIG_SCHED_MC | |
8485 | if (!err && mc_capable()) | |
8486 | err = sysfs_create_file(&cls->kset.kobj, | |
8487 | &attr_sched_mc_power_savings.attr); | |
8488 | #endif | |
8489 | return err; | |
8490 | } | |
6d6bc0ad | 8491 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8492 | |
e761b772 | 8493 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8494 | /* |
e761b772 MK |
8495 | * Add online and remove offline CPUs from the scheduler domains. |
8496 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8497 | */ |
8498 | static int update_sched_domains(struct notifier_block *nfb, | |
8499 | unsigned long action, void *hcpu) | |
e761b772 MK |
8500 | { |
8501 | switch (action) { | |
8502 | case CPU_ONLINE: | |
8503 | case CPU_ONLINE_FROZEN: | |
8504 | case CPU_DEAD: | |
8505 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8506 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8507 | return NOTIFY_OK; |
8508 | ||
8509 | default: | |
8510 | return NOTIFY_DONE; | |
8511 | } | |
8512 | } | |
8513 | #endif | |
8514 | ||
8515 | static int update_runtime(struct notifier_block *nfb, | |
8516 | unsigned long action, void *hcpu) | |
1da177e4 | 8517 | { |
7def2be1 PZ |
8518 | int cpu = (int)(long)hcpu; |
8519 | ||
1da177e4 | 8520 | switch (action) { |
1da177e4 | 8521 | case CPU_DOWN_PREPARE: |
8bb78442 | 8522 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8523 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8524 | return NOTIFY_OK; |
8525 | ||
1da177e4 | 8526 | case CPU_DOWN_FAILED: |
8bb78442 | 8527 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8528 | case CPU_ONLINE: |
8bb78442 | 8529 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8530 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8531 | return NOTIFY_OK; |
8532 | ||
1da177e4 LT |
8533 | default: |
8534 | return NOTIFY_DONE; | |
8535 | } | |
1da177e4 | 8536 | } |
1da177e4 LT |
8537 | |
8538 | void __init sched_init_smp(void) | |
8539 | { | |
dcc30a35 RR |
8540 | cpumask_var_t non_isolated_cpus; |
8541 | ||
8542 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8543 | |
434d53b0 MT |
8544 | #if defined(CONFIG_NUMA) |
8545 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8546 | GFP_KERNEL); | |
8547 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8548 | #endif | |
95402b38 | 8549 | get_online_cpus(); |
712555ee | 8550 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8551 | arch_init_sched_domains(cpu_online_mask); |
8552 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8553 | if (cpumask_empty(non_isolated_cpus)) | |
8554 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8555 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8556 | put_online_cpus(); |
e761b772 MK |
8557 | |
8558 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8559 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8560 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8561 | #endif |
8562 | ||
8563 | /* RT runtime code needs to handle some hotplug events */ | |
8564 | hotcpu_notifier(update_runtime, 0); | |
8565 | ||
b328ca18 | 8566 | init_hrtick(); |
5c1e1767 NP |
8567 | |
8568 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8569 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8570 | BUG(); |
19978ca6 | 8571 | sched_init_granularity(); |
dcc30a35 | 8572 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8573 | |
8574 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8575 | init_sched_rt_class(); |
1da177e4 LT |
8576 | } |
8577 | #else | |
8578 | void __init sched_init_smp(void) | |
8579 | { | |
19978ca6 | 8580 | sched_init_granularity(); |
1da177e4 LT |
8581 | } |
8582 | #endif /* CONFIG_SMP */ | |
8583 | ||
8584 | int in_sched_functions(unsigned long addr) | |
8585 | { | |
1da177e4 LT |
8586 | return in_lock_functions(addr) || |
8587 | (addr >= (unsigned long)__sched_text_start | |
8588 | && addr < (unsigned long)__sched_text_end); | |
8589 | } | |
8590 | ||
a9957449 | 8591 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8592 | { |
8593 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8594 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8595 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8596 | cfs_rq->rq = rq; | |
8597 | #endif | |
67e9fb2a | 8598 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8599 | } |
8600 | ||
fa85ae24 PZ |
8601 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8602 | { | |
8603 | struct rt_prio_array *array; | |
8604 | int i; | |
8605 | ||
8606 | array = &rt_rq->active; | |
8607 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8608 | INIT_LIST_HEAD(array->queue + i); | |
8609 | __clear_bit(i, array->bitmap); | |
8610 | } | |
8611 | /* delimiter for bitsearch: */ | |
8612 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8613 | ||
052f1dc7 | 8614 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8615 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8616 | #ifdef CONFIG_SMP |
e864c499 | 8617 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8618 | #endif |
48d5e258 | 8619 | #endif |
fa85ae24 PZ |
8620 | #ifdef CONFIG_SMP |
8621 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8622 | rt_rq->overloaded = 0; |
917b627d | 8623 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8624 | #endif |
8625 | ||
8626 | rt_rq->rt_time = 0; | |
8627 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8628 | rt_rq->rt_runtime = 0; |
8629 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8630 | |
052f1dc7 | 8631 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8632 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8633 | rt_rq->rq = rq; |
8634 | #endif | |
fa85ae24 PZ |
8635 | } |
8636 | ||
6f505b16 | 8637 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8638 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8639 | struct sched_entity *se, int cpu, int add, | |
8640 | struct sched_entity *parent) | |
6f505b16 | 8641 | { |
ec7dc8ac | 8642 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8643 | tg->cfs_rq[cpu] = cfs_rq; |
8644 | init_cfs_rq(cfs_rq, rq); | |
8645 | cfs_rq->tg = tg; | |
8646 | if (add) | |
8647 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8648 | ||
8649 | tg->se[cpu] = se; | |
354d60c2 DG |
8650 | /* se could be NULL for init_task_group */ |
8651 | if (!se) | |
8652 | return; | |
8653 | ||
ec7dc8ac DG |
8654 | if (!parent) |
8655 | se->cfs_rq = &rq->cfs; | |
8656 | else | |
8657 | se->cfs_rq = parent->my_q; | |
8658 | ||
6f505b16 PZ |
8659 | se->my_q = cfs_rq; |
8660 | se->load.weight = tg->shares; | |
e05510d0 | 8661 | se->load.inv_weight = 0; |
ec7dc8ac | 8662 | se->parent = parent; |
6f505b16 | 8663 | } |
052f1dc7 | 8664 | #endif |
6f505b16 | 8665 | |
052f1dc7 | 8666 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8667 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8668 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8669 | struct sched_rt_entity *parent) | |
6f505b16 | 8670 | { |
ec7dc8ac DG |
8671 | struct rq *rq = cpu_rq(cpu); |
8672 | ||
6f505b16 PZ |
8673 | tg->rt_rq[cpu] = rt_rq; |
8674 | init_rt_rq(rt_rq, rq); | |
8675 | rt_rq->tg = tg; | |
8676 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8677 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8678 | if (add) |
8679 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8680 | ||
8681 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8682 | if (!rt_se) |
8683 | return; | |
8684 | ||
ec7dc8ac DG |
8685 | if (!parent) |
8686 | rt_se->rt_rq = &rq->rt; | |
8687 | else | |
8688 | rt_se->rt_rq = parent->my_q; | |
8689 | ||
6f505b16 | 8690 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8691 | rt_se->parent = parent; |
6f505b16 PZ |
8692 | INIT_LIST_HEAD(&rt_se->run_list); |
8693 | } | |
8694 | #endif | |
8695 | ||
1da177e4 LT |
8696 | void __init sched_init(void) |
8697 | { | |
dd41f596 | 8698 | int i, j; |
434d53b0 MT |
8699 | unsigned long alloc_size = 0, ptr; |
8700 | ||
8701 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8702 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8703 | #endif | |
8704 | #ifdef CONFIG_RT_GROUP_SCHED | |
8705 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8706 | #endif |
8707 | #ifdef CONFIG_USER_SCHED | |
8708 | alloc_size *= 2; | |
434d53b0 MT |
8709 | #endif |
8710 | /* | |
8711 | * As sched_init() is called before page_alloc is setup, | |
8712 | * we use alloc_bootmem(). | |
8713 | */ | |
8714 | if (alloc_size) { | |
5a9d3225 | 8715 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8716 | |
8717 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8718 | init_task_group.se = (struct sched_entity **)ptr; | |
8719 | ptr += nr_cpu_ids * sizeof(void **); | |
8720 | ||
8721 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8722 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8723 | |
8724 | #ifdef CONFIG_USER_SCHED | |
8725 | root_task_group.se = (struct sched_entity **)ptr; | |
8726 | ptr += nr_cpu_ids * sizeof(void **); | |
8727 | ||
8728 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8729 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8730 | #endif /* CONFIG_USER_SCHED */ |
8731 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8732 | #ifdef CONFIG_RT_GROUP_SCHED |
8733 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8734 | ptr += nr_cpu_ids * sizeof(void **); | |
8735 | ||
8736 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8737 | ptr += nr_cpu_ids * sizeof(void **); |
8738 | ||
8739 | #ifdef CONFIG_USER_SCHED | |
8740 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8741 | ptr += nr_cpu_ids * sizeof(void **); | |
8742 | ||
8743 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8744 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8745 | #endif /* CONFIG_USER_SCHED */ |
8746 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8747 | } |
dd41f596 | 8748 | |
57d885fe GH |
8749 | #ifdef CONFIG_SMP |
8750 | init_defrootdomain(); | |
8751 | #endif | |
8752 | ||
d0b27fa7 PZ |
8753 | init_rt_bandwidth(&def_rt_bandwidth, |
8754 | global_rt_period(), global_rt_runtime()); | |
8755 | ||
8756 | #ifdef CONFIG_RT_GROUP_SCHED | |
8757 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8758 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8759 | #ifdef CONFIG_USER_SCHED |
8760 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8761 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8762 | #endif /* CONFIG_USER_SCHED */ |
8763 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8764 | |
052f1dc7 | 8765 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8766 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8767 | INIT_LIST_HEAD(&init_task_group.children); |
8768 | ||
8769 | #ifdef CONFIG_USER_SCHED | |
8770 | INIT_LIST_HEAD(&root_task_group.children); | |
8771 | init_task_group.parent = &root_task_group; | |
8772 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8773 | #endif /* CONFIG_USER_SCHED */ |
8774 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8775 | |
0a945022 | 8776 | for_each_possible_cpu(i) { |
70b97a7f | 8777 | struct rq *rq; |
1da177e4 LT |
8778 | |
8779 | rq = cpu_rq(i); | |
8780 | spin_lock_init(&rq->lock); | |
7897986b | 8781 | rq->nr_running = 0; |
dd41f596 | 8782 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8783 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8784 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8785 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8786 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8787 | #ifdef CONFIG_CGROUP_SCHED |
8788 | /* | |
8789 | * How much cpu bandwidth does init_task_group get? | |
8790 | * | |
8791 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8792 | * gets 100% of the cpu resources in the system. This overall | |
8793 | * system cpu resource is divided among the tasks of | |
8794 | * init_task_group and its child task-groups in a fair manner, | |
8795 | * based on each entity's (task or task-group's) weight | |
8796 | * (se->load.weight). | |
8797 | * | |
8798 | * In other words, if init_task_group has 10 tasks of weight | |
8799 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8800 | * then A0's share of the cpu resource is: | |
8801 | * | |
8802 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8803 | * | |
8804 | * We achieve this by letting init_task_group's tasks sit | |
8805 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8806 | */ | |
ec7dc8ac | 8807 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8808 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8809 | root_task_group.shares = NICE_0_LOAD; |
8810 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8811 | /* |
8812 | * In case of task-groups formed thr' the user id of tasks, | |
8813 | * init_task_group represents tasks belonging to root user. | |
8814 | * Hence it forms a sibling of all subsequent groups formed. | |
8815 | * In this case, init_task_group gets only a fraction of overall | |
8816 | * system cpu resource, based on the weight assigned to root | |
8817 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8818 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8819 | * (init_cfs_rq) and having one entity represent this group of | |
8820 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8821 | */ | |
ec7dc8ac | 8822 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8823 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8824 | &per_cpu(init_sched_entity, i), i, 1, |
8825 | root_task_group.se[i]); | |
6f505b16 | 8826 | |
052f1dc7 | 8827 | #endif |
354d60c2 DG |
8828 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8829 | ||
8830 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8831 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8832 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8833 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8834 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8835 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8836 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8837 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8838 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8839 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8840 | root_task_group.rt_se[i]); | |
354d60c2 | 8841 | #endif |
dd41f596 | 8842 | #endif |
1da177e4 | 8843 | |
dd41f596 IM |
8844 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8845 | rq->cpu_load[j] = 0; | |
1da177e4 | 8846 | #ifdef CONFIG_SMP |
41c7ce9a | 8847 | rq->sd = NULL; |
57d885fe | 8848 | rq->rd = NULL; |
1da177e4 | 8849 | rq->active_balance = 0; |
dd41f596 | 8850 | rq->next_balance = jiffies; |
1da177e4 | 8851 | rq->push_cpu = 0; |
0a2966b4 | 8852 | rq->cpu = i; |
1f11eb6a | 8853 | rq->online = 0; |
1da177e4 LT |
8854 | rq->migration_thread = NULL; |
8855 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8856 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8857 | #endif |
8f4d37ec | 8858 | init_rq_hrtick(rq); |
1da177e4 | 8859 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8860 | } |
8861 | ||
2dd73a4f | 8862 | set_load_weight(&init_task); |
b50f60ce | 8863 | |
e107be36 AK |
8864 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8865 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8866 | #endif | |
8867 | ||
c9819f45 | 8868 | #ifdef CONFIG_SMP |
962cf36c | 8869 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8870 | #endif |
8871 | ||
b50f60ce HC |
8872 | #ifdef CONFIG_RT_MUTEXES |
8873 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8874 | #endif | |
8875 | ||
1da177e4 LT |
8876 | /* |
8877 | * The boot idle thread does lazy MMU switching as well: | |
8878 | */ | |
8879 | atomic_inc(&init_mm.mm_count); | |
8880 | enter_lazy_tlb(&init_mm, current); | |
8881 | ||
8882 | /* | |
8883 | * Make us the idle thread. Technically, schedule() should not be | |
8884 | * called from this thread, however somewhere below it might be, | |
8885 | * but because we are the idle thread, we just pick up running again | |
8886 | * when this runqueue becomes "idle". | |
8887 | */ | |
8888 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8889 | /* |
8890 | * During early bootup we pretend to be a normal task: | |
8891 | */ | |
8892 | current->sched_class = &fair_sched_class; | |
6892b75e | 8893 | |
6a7b3dc3 RR |
8894 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
8895 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 8896 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
8897 | #ifdef CONFIG_NO_HZ |
8898 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
8899 | #endif | |
dcc30a35 | 8900 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 8901 | #endif /* SMP */ |
6a7b3dc3 | 8902 | |
6892b75e | 8903 | scheduler_running = 1; |
1da177e4 LT |
8904 | } |
8905 | ||
8906 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8907 | void __might_sleep(char *file, int line) | |
8908 | { | |
48f24c4d | 8909 | #ifdef in_atomic |
1da177e4 LT |
8910 | static unsigned long prev_jiffy; /* ratelimiting */ |
8911 | ||
aef745fc IM |
8912 | if ((!in_atomic() && !irqs_disabled()) || |
8913 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8914 | return; | |
8915 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8916 | return; | |
8917 | prev_jiffy = jiffies; | |
8918 | ||
8919 | printk(KERN_ERR | |
8920 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8921 | file, line); | |
8922 | printk(KERN_ERR | |
8923 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8924 | in_atomic(), irqs_disabled(), | |
8925 | current->pid, current->comm); | |
8926 | ||
8927 | debug_show_held_locks(current); | |
8928 | if (irqs_disabled()) | |
8929 | print_irqtrace_events(current); | |
8930 | dump_stack(); | |
1da177e4 LT |
8931 | #endif |
8932 | } | |
8933 | EXPORT_SYMBOL(__might_sleep); | |
8934 | #endif | |
8935 | ||
8936 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8937 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8938 | { | |
8939 | int on_rq; | |
3e51f33f | 8940 | |
3a5e4dc1 AK |
8941 | update_rq_clock(rq); |
8942 | on_rq = p->se.on_rq; | |
8943 | if (on_rq) | |
8944 | deactivate_task(rq, p, 0); | |
8945 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8946 | if (on_rq) { | |
8947 | activate_task(rq, p, 0); | |
8948 | resched_task(rq->curr); | |
8949 | } | |
8950 | } | |
8951 | ||
1da177e4 LT |
8952 | void normalize_rt_tasks(void) |
8953 | { | |
a0f98a1c | 8954 | struct task_struct *g, *p; |
1da177e4 | 8955 | unsigned long flags; |
70b97a7f | 8956 | struct rq *rq; |
1da177e4 | 8957 | |
4cf5d77a | 8958 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8959 | do_each_thread(g, p) { |
178be793 IM |
8960 | /* |
8961 | * Only normalize user tasks: | |
8962 | */ | |
8963 | if (!p->mm) | |
8964 | continue; | |
8965 | ||
6cfb0d5d | 8966 | p->se.exec_start = 0; |
6cfb0d5d | 8967 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8968 | p->se.wait_start = 0; |
dd41f596 | 8969 | p->se.sleep_start = 0; |
dd41f596 | 8970 | p->se.block_start = 0; |
6cfb0d5d | 8971 | #endif |
dd41f596 IM |
8972 | |
8973 | if (!rt_task(p)) { | |
8974 | /* | |
8975 | * Renice negative nice level userspace | |
8976 | * tasks back to 0: | |
8977 | */ | |
8978 | if (TASK_NICE(p) < 0 && p->mm) | |
8979 | set_user_nice(p, 0); | |
1da177e4 | 8980 | continue; |
dd41f596 | 8981 | } |
1da177e4 | 8982 | |
4cf5d77a | 8983 | spin_lock(&p->pi_lock); |
b29739f9 | 8984 | rq = __task_rq_lock(p); |
1da177e4 | 8985 | |
178be793 | 8986 | normalize_task(rq, p); |
3a5e4dc1 | 8987 | |
b29739f9 | 8988 | __task_rq_unlock(rq); |
4cf5d77a | 8989 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8990 | } while_each_thread(g, p); |
8991 | ||
4cf5d77a | 8992 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8993 | } |
8994 | ||
8995 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8996 | |
8997 | #ifdef CONFIG_IA64 | |
8998 | /* | |
8999 | * These functions are only useful for the IA64 MCA handling. | |
9000 | * | |
9001 | * They can only be called when the whole system has been | |
9002 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9003 | * activity can take place. Using them for anything else would | |
9004 | * be a serious bug, and as a result, they aren't even visible | |
9005 | * under any other configuration. | |
9006 | */ | |
9007 | ||
9008 | /** | |
9009 | * curr_task - return the current task for a given cpu. | |
9010 | * @cpu: the processor in question. | |
9011 | * | |
9012 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9013 | */ | |
36c8b586 | 9014 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9015 | { |
9016 | return cpu_curr(cpu); | |
9017 | } | |
9018 | ||
9019 | /** | |
9020 | * set_curr_task - set the current task for a given cpu. | |
9021 | * @cpu: the processor in question. | |
9022 | * @p: the task pointer to set. | |
9023 | * | |
9024 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9025 | * are serviced on a separate stack. It allows the architecture to switch the |
9026 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9027 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9028 | * and caller must save the original value of the current task (see | |
9029 | * curr_task() above) and restore that value before reenabling interrupts and | |
9030 | * re-starting the system. | |
9031 | * | |
9032 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9033 | */ | |
36c8b586 | 9034 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9035 | { |
9036 | cpu_curr(cpu) = p; | |
9037 | } | |
9038 | ||
9039 | #endif | |
29f59db3 | 9040 | |
bccbe08a PZ |
9041 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9042 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9043 | { |
9044 | int i; | |
9045 | ||
9046 | for_each_possible_cpu(i) { | |
9047 | if (tg->cfs_rq) | |
9048 | kfree(tg->cfs_rq[i]); | |
9049 | if (tg->se) | |
9050 | kfree(tg->se[i]); | |
6f505b16 PZ |
9051 | } |
9052 | ||
9053 | kfree(tg->cfs_rq); | |
9054 | kfree(tg->se); | |
6f505b16 PZ |
9055 | } |
9056 | ||
ec7dc8ac DG |
9057 | static |
9058 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9059 | { |
29f59db3 | 9060 | struct cfs_rq *cfs_rq; |
eab17229 | 9061 | struct sched_entity *se; |
9b5b7751 | 9062 | struct rq *rq; |
29f59db3 SV |
9063 | int i; |
9064 | ||
434d53b0 | 9065 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9066 | if (!tg->cfs_rq) |
9067 | goto err; | |
434d53b0 | 9068 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9069 | if (!tg->se) |
9070 | goto err; | |
052f1dc7 PZ |
9071 | |
9072 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9073 | |
9074 | for_each_possible_cpu(i) { | |
9b5b7751 | 9075 | rq = cpu_rq(i); |
29f59db3 | 9076 | |
eab17229 LZ |
9077 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9078 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9079 | if (!cfs_rq) |
9080 | goto err; | |
9081 | ||
eab17229 LZ |
9082 | se = kzalloc_node(sizeof(struct sched_entity), |
9083 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9084 | if (!se) |
9085 | goto err; | |
9086 | ||
eab17229 | 9087 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9088 | } |
9089 | ||
9090 | return 1; | |
9091 | ||
9092 | err: | |
9093 | return 0; | |
9094 | } | |
9095 | ||
9096 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9097 | { | |
9098 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9099 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9100 | } | |
9101 | ||
9102 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9103 | { | |
9104 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9105 | } | |
6d6bc0ad | 9106 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9107 | static inline void free_fair_sched_group(struct task_group *tg) |
9108 | { | |
9109 | } | |
9110 | ||
ec7dc8ac DG |
9111 | static inline |
9112 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9113 | { |
9114 | return 1; | |
9115 | } | |
9116 | ||
9117 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9118 | { | |
9119 | } | |
9120 | ||
9121 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9122 | { | |
9123 | } | |
6d6bc0ad | 9124 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9125 | |
9126 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9127 | static void free_rt_sched_group(struct task_group *tg) |
9128 | { | |
9129 | int i; | |
9130 | ||
d0b27fa7 PZ |
9131 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9132 | ||
bccbe08a PZ |
9133 | for_each_possible_cpu(i) { |
9134 | if (tg->rt_rq) | |
9135 | kfree(tg->rt_rq[i]); | |
9136 | if (tg->rt_se) | |
9137 | kfree(tg->rt_se[i]); | |
9138 | } | |
9139 | ||
9140 | kfree(tg->rt_rq); | |
9141 | kfree(tg->rt_se); | |
9142 | } | |
9143 | ||
ec7dc8ac DG |
9144 | static |
9145 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9146 | { |
9147 | struct rt_rq *rt_rq; | |
eab17229 | 9148 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9149 | struct rq *rq; |
9150 | int i; | |
9151 | ||
434d53b0 | 9152 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9153 | if (!tg->rt_rq) |
9154 | goto err; | |
434d53b0 | 9155 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9156 | if (!tg->rt_se) |
9157 | goto err; | |
9158 | ||
d0b27fa7 PZ |
9159 | init_rt_bandwidth(&tg->rt_bandwidth, |
9160 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9161 | |
9162 | for_each_possible_cpu(i) { | |
9163 | rq = cpu_rq(i); | |
9164 | ||
eab17229 LZ |
9165 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9166 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9167 | if (!rt_rq) |
9168 | goto err; | |
29f59db3 | 9169 | |
eab17229 LZ |
9170 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9171 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9172 | if (!rt_se) |
9173 | goto err; | |
29f59db3 | 9174 | |
eab17229 | 9175 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9176 | } |
9177 | ||
bccbe08a PZ |
9178 | return 1; |
9179 | ||
9180 | err: | |
9181 | return 0; | |
9182 | } | |
9183 | ||
9184 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9185 | { | |
9186 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9187 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9188 | } | |
9189 | ||
9190 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9191 | { | |
9192 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9193 | } | |
6d6bc0ad | 9194 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9195 | static inline void free_rt_sched_group(struct task_group *tg) |
9196 | { | |
9197 | } | |
9198 | ||
ec7dc8ac DG |
9199 | static inline |
9200 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9201 | { |
9202 | return 1; | |
9203 | } | |
9204 | ||
9205 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9206 | { | |
9207 | } | |
9208 | ||
9209 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9210 | { | |
9211 | } | |
6d6bc0ad | 9212 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9213 | |
d0b27fa7 | 9214 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9215 | static void free_sched_group(struct task_group *tg) |
9216 | { | |
9217 | free_fair_sched_group(tg); | |
9218 | free_rt_sched_group(tg); | |
9219 | kfree(tg); | |
9220 | } | |
9221 | ||
9222 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9223 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9224 | { |
9225 | struct task_group *tg; | |
9226 | unsigned long flags; | |
9227 | int i; | |
9228 | ||
9229 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9230 | if (!tg) | |
9231 | return ERR_PTR(-ENOMEM); | |
9232 | ||
ec7dc8ac | 9233 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9234 | goto err; |
9235 | ||
ec7dc8ac | 9236 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9237 | goto err; |
9238 | ||
8ed36996 | 9239 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9240 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9241 | register_fair_sched_group(tg, i); |
9242 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9243 | } |
6f505b16 | 9244 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9245 | |
9246 | WARN_ON(!parent); /* root should already exist */ | |
9247 | ||
9248 | tg->parent = parent; | |
f473aa5e | 9249 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9250 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9251 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9252 | |
9b5b7751 | 9253 | return tg; |
29f59db3 SV |
9254 | |
9255 | err: | |
6f505b16 | 9256 | free_sched_group(tg); |
29f59db3 SV |
9257 | return ERR_PTR(-ENOMEM); |
9258 | } | |
9259 | ||
9b5b7751 | 9260 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9261 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9262 | { |
29f59db3 | 9263 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9264 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9265 | } |
9266 | ||
9b5b7751 | 9267 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9268 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9269 | { |
8ed36996 | 9270 | unsigned long flags; |
9b5b7751 | 9271 | int i; |
29f59db3 | 9272 | |
8ed36996 | 9273 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9274 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9275 | unregister_fair_sched_group(tg, i); |
9276 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9277 | } |
6f505b16 | 9278 | list_del_rcu(&tg->list); |
f473aa5e | 9279 | list_del_rcu(&tg->siblings); |
8ed36996 | 9280 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9281 | |
9b5b7751 | 9282 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9283 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9284 | } |
9285 | ||
9b5b7751 | 9286 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9287 | * The caller of this function should have put the task in its new group |
9288 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9289 | * reflect its new group. | |
9b5b7751 SV |
9290 | */ |
9291 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9292 | { |
9293 | int on_rq, running; | |
9294 | unsigned long flags; | |
9295 | struct rq *rq; | |
9296 | ||
9297 | rq = task_rq_lock(tsk, &flags); | |
9298 | ||
29f59db3 SV |
9299 | update_rq_clock(rq); |
9300 | ||
051a1d1a | 9301 | running = task_current(rq, tsk); |
29f59db3 SV |
9302 | on_rq = tsk->se.on_rq; |
9303 | ||
0e1f3483 | 9304 | if (on_rq) |
29f59db3 | 9305 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9306 | if (unlikely(running)) |
9307 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9308 | |
6f505b16 | 9309 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9310 | |
810b3817 PZ |
9311 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9312 | if (tsk->sched_class->moved_group) | |
9313 | tsk->sched_class->moved_group(tsk); | |
9314 | #endif | |
9315 | ||
0e1f3483 HS |
9316 | if (unlikely(running)) |
9317 | tsk->sched_class->set_curr_task(rq); | |
9318 | if (on_rq) | |
7074badb | 9319 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9320 | |
29f59db3 SV |
9321 | task_rq_unlock(rq, &flags); |
9322 | } | |
6d6bc0ad | 9323 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9324 | |
052f1dc7 | 9325 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9326 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9327 | { |
9328 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9329 | int on_rq; |
9330 | ||
29f59db3 | 9331 | on_rq = se->on_rq; |
62fb1851 | 9332 | if (on_rq) |
29f59db3 SV |
9333 | dequeue_entity(cfs_rq, se, 0); |
9334 | ||
9335 | se->load.weight = shares; | |
e05510d0 | 9336 | se->load.inv_weight = 0; |
29f59db3 | 9337 | |
62fb1851 | 9338 | if (on_rq) |
29f59db3 | 9339 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9340 | } |
62fb1851 | 9341 | |
c09595f6 PZ |
9342 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9343 | { | |
9344 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9345 | struct rq *rq = cfs_rq->rq; | |
9346 | unsigned long flags; | |
9347 | ||
9348 | spin_lock_irqsave(&rq->lock, flags); | |
9349 | __set_se_shares(se, shares); | |
9350 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9351 | } |
9352 | ||
8ed36996 PZ |
9353 | static DEFINE_MUTEX(shares_mutex); |
9354 | ||
4cf86d77 | 9355 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9356 | { |
9357 | int i; | |
8ed36996 | 9358 | unsigned long flags; |
c61935fd | 9359 | |
ec7dc8ac DG |
9360 | /* |
9361 | * We can't change the weight of the root cgroup. | |
9362 | */ | |
9363 | if (!tg->se[0]) | |
9364 | return -EINVAL; | |
9365 | ||
18d95a28 PZ |
9366 | if (shares < MIN_SHARES) |
9367 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9368 | else if (shares > MAX_SHARES) |
9369 | shares = MAX_SHARES; | |
62fb1851 | 9370 | |
8ed36996 | 9371 | mutex_lock(&shares_mutex); |
9b5b7751 | 9372 | if (tg->shares == shares) |
5cb350ba | 9373 | goto done; |
29f59db3 | 9374 | |
8ed36996 | 9375 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9376 | for_each_possible_cpu(i) |
9377 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9378 | list_del_rcu(&tg->siblings); |
8ed36996 | 9379 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9380 | |
9381 | /* wait for any ongoing reference to this group to finish */ | |
9382 | synchronize_sched(); | |
9383 | ||
9384 | /* | |
9385 | * Now we are free to modify the group's share on each cpu | |
9386 | * w/o tripping rebalance_share or load_balance_fair. | |
9387 | */ | |
9b5b7751 | 9388 | tg->shares = shares; |
c09595f6 PZ |
9389 | for_each_possible_cpu(i) { |
9390 | /* | |
9391 | * force a rebalance | |
9392 | */ | |
9393 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9394 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9395 | } |
29f59db3 | 9396 | |
6b2d7700 SV |
9397 | /* |
9398 | * Enable load balance activity on this group, by inserting it back on | |
9399 | * each cpu's rq->leaf_cfs_rq_list. | |
9400 | */ | |
8ed36996 | 9401 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9402 | for_each_possible_cpu(i) |
9403 | register_fair_sched_group(tg, i); | |
f473aa5e | 9404 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9405 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9406 | done: |
8ed36996 | 9407 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9408 | return 0; |
29f59db3 SV |
9409 | } |
9410 | ||
5cb350ba DG |
9411 | unsigned long sched_group_shares(struct task_group *tg) |
9412 | { | |
9413 | return tg->shares; | |
9414 | } | |
052f1dc7 | 9415 | #endif |
5cb350ba | 9416 | |
052f1dc7 | 9417 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9418 | /* |
9f0c1e56 | 9419 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9420 | */ |
9f0c1e56 PZ |
9421 | static DEFINE_MUTEX(rt_constraints_mutex); |
9422 | ||
9423 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9424 | { | |
9425 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9426 | return 1ULL << 20; |
9f0c1e56 | 9427 | |
9a7e0b18 | 9428 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9429 | } |
9430 | ||
9a7e0b18 PZ |
9431 | /* Must be called with tasklist_lock held */ |
9432 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9433 | { |
9a7e0b18 | 9434 | struct task_struct *g, *p; |
b40b2e8e | 9435 | |
9a7e0b18 PZ |
9436 | do_each_thread(g, p) { |
9437 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9438 | return 1; | |
9439 | } while_each_thread(g, p); | |
b40b2e8e | 9440 | |
9a7e0b18 PZ |
9441 | return 0; |
9442 | } | |
b40b2e8e | 9443 | |
9a7e0b18 PZ |
9444 | struct rt_schedulable_data { |
9445 | struct task_group *tg; | |
9446 | u64 rt_period; | |
9447 | u64 rt_runtime; | |
9448 | }; | |
b40b2e8e | 9449 | |
9a7e0b18 PZ |
9450 | static int tg_schedulable(struct task_group *tg, void *data) |
9451 | { | |
9452 | struct rt_schedulable_data *d = data; | |
9453 | struct task_group *child; | |
9454 | unsigned long total, sum = 0; | |
9455 | u64 period, runtime; | |
b40b2e8e | 9456 | |
9a7e0b18 PZ |
9457 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9458 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9459 | |
9a7e0b18 PZ |
9460 | if (tg == d->tg) { |
9461 | period = d->rt_period; | |
9462 | runtime = d->rt_runtime; | |
b40b2e8e | 9463 | } |
b40b2e8e | 9464 | |
98a4826b PZ |
9465 | #ifdef CONFIG_USER_SCHED |
9466 | if (tg == &root_task_group) { | |
9467 | period = global_rt_period(); | |
9468 | runtime = global_rt_runtime(); | |
9469 | } | |
9470 | #endif | |
9471 | ||
4653f803 PZ |
9472 | /* |
9473 | * Cannot have more runtime than the period. | |
9474 | */ | |
9475 | if (runtime > period && runtime != RUNTIME_INF) | |
9476 | return -EINVAL; | |
6f505b16 | 9477 | |
4653f803 PZ |
9478 | /* |
9479 | * Ensure we don't starve existing RT tasks. | |
9480 | */ | |
9a7e0b18 PZ |
9481 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9482 | return -EBUSY; | |
6f505b16 | 9483 | |
9a7e0b18 | 9484 | total = to_ratio(period, runtime); |
6f505b16 | 9485 | |
4653f803 PZ |
9486 | /* |
9487 | * Nobody can have more than the global setting allows. | |
9488 | */ | |
9489 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9490 | return -EINVAL; | |
6f505b16 | 9491 | |
4653f803 PZ |
9492 | /* |
9493 | * The sum of our children's runtime should not exceed our own. | |
9494 | */ | |
9a7e0b18 PZ |
9495 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9496 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9497 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9498 | |
9a7e0b18 PZ |
9499 | if (child == d->tg) { |
9500 | period = d->rt_period; | |
9501 | runtime = d->rt_runtime; | |
9502 | } | |
6f505b16 | 9503 | |
9a7e0b18 | 9504 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9505 | } |
6f505b16 | 9506 | |
9a7e0b18 PZ |
9507 | if (sum > total) |
9508 | return -EINVAL; | |
9509 | ||
9510 | return 0; | |
6f505b16 PZ |
9511 | } |
9512 | ||
9a7e0b18 | 9513 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9514 | { |
9a7e0b18 PZ |
9515 | struct rt_schedulable_data data = { |
9516 | .tg = tg, | |
9517 | .rt_period = period, | |
9518 | .rt_runtime = runtime, | |
9519 | }; | |
9520 | ||
9521 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9522 | } |
9523 | ||
d0b27fa7 PZ |
9524 | static int tg_set_bandwidth(struct task_group *tg, |
9525 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9526 | { |
ac086bc2 | 9527 | int i, err = 0; |
9f0c1e56 | 9528 | |
9f0c1e56 | 9529 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9530 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9531 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9532 | if (err) | |
9f0c1e56 | 9533 | goto unlock; |
ac086bc2 PZ |
9534 | |
9535 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9536 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9537 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9538 | |
9539 | for_each_possible_cpu(i) { | |
9540 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9541 | ||
9542 | spin_lock(&rt_rq->rt_runtime_lock); | |
9543 | rt_rq->rt_runtime = rt_runtime; | |
9544 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9545 | } | |
9546 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9547 | unlock: |
521f1a24 | 9548 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9549 | mutex_unlock(&rt_constraints_mutex); |
9550 | ||
9551 | return err; | |
6f505b16 PZ |
9552 | } |
9553 | ||
d0b27fa7 PZ |
9554 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9555 | { | |
9556 | u64 rt_runtime, rt_period; | |
9557 | ||
9558 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9559 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9560 | if (rt_runtime_us < 0) | |
9561 | rt_runtime = RUNTIME_INF; | |
9562 | ||
9563 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9564 | } | |
9565 | ||
9f0c1e56 PZ |
9566 | long sched_group_rt_runtime(struct task_group *tg) |
9567 | { | |
9568 | u64 rt_runtime_us; | |
9569 | ||
d0b27fa7 | 9570 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9571 | return -1; |
9572 | ||
d0b27fa7 | 9573 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9574 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9575 | return rt_runtime_us; | |
9576 | } | |
d0b27fa7 PZ |
9577 | |
9578 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9579 | { | |
9580 | u64 rt_runtime, rt_period; | |
9581 | ||
9582 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9583 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9584 | ||
619b0488 R |
9585 | if (rt_period == 0) |
9586 | return -EINVAL; | |
9587 | ||
d0b27fa7 PZ |
9588 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9589 | } | |
9590 | ||
9591 | long sched_group_rt_period(struct task_group *tg) | |
9592 | { | |
9593 | u64 rt_period_us; | |
9594 | ||
9595 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9596 | do_div(rt_period_us, NSEC_PER_USEC); | |
9597 | return rt_period_us; | |
9598 | } | |
9599 | ||
9600 | static int sched_rt_global_constraints(void) | |
9601 | { | |
4653f803 | 9602 | u64 runtime, period; |
d0b27fa7 PZ |
9603 | int ret = 0; |
9604 | ||
ec5d4989 HS |
9605 | if (sysctl_sched_rt_period <= 0) |
9606 | return -EINVAL; | |
9607 | ||
4653f803 PZ |
9608 | runtime = global_rt_runtime(); |
9609 | period = global_rt_period(); | |
9610 | ||
9611 | /* | |
9612 | * Sanity check on the sysctl variables. | |
9613 | */ | |
9614 | if (runtime > period && runtime != RUNTIME_INF) | |
9615 | return -EINVAL; | |
10b612f4 | 9616 | |
d0b27fa7 | 9617 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9618 | read_lock(&tasklist_lock); |
4653f803 | 9619 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9620 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9621 | mutex_unlock(&rt_constraints_mutex); |
9622 | ||
9623 | return ret; | |
9624 | } | |
54e99124 DG |
9625 | |
9626 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9627 | { | |
9628 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9629 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9630 | return 0; | |
9631 | ||
9632 | return 1; | |
9633 | } | |
9634 | ||
6d6bc0ad | 9635 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9636 | static int sched_rt_global_constraints(void) |
9637 | { | |
ac086bc2 PZ |
9638 | unsigned long flags; |
9639 | int i; | |
9640 | ||
ec5d4989 HS |
9641 | if (sysctl_sched_rt_period <= 0) |
9642 | return -EINVAL; | |
9643 | ||
ac086bc2 PZ |
9644 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9645 | for_each_possible_cpu(i) { | |
9646 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9647 | ||
9648 | spin_lock(&rt_rq->rt_runtime_lock); | |
9649 | rt_rq->rt_runtime = global_rt_runtime(); | |
9650 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9651 | } | |
9652 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9653 | ||
d0b27fa7 PZ |
9654 | return 0; |
9655 | } | |
6d6bc0ad | 9656 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9657 | |
9658 | int sched_rt_handler(struct ctl_table *table, int write, | |
9659 | struct file *filp, void __user *buffer, size_t *lenp, | |
9660 | loff_t *ppos) | |
9661 | { | |
9662 | int ret; | |
9663 | int old_period, old_runtime; | |
9664 | static DEFINE_MUTEX(mutex); | |
9665 | ||
9666 | mutex_lock(&mutex); | |
9667 | old_period = sysctl_sched_rt_period; | |
9668 | old_runtime = sysctl_sched_rt_runtime; | |
9669 | ||
9670 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9671 | ||
9672 | if (!ret && write) { | |
9673 | ret = sched_rt_global_constraints(); | |
9674 | if (ret) { | |
9675 | sysctl_sched_rt_period = old_period; | |
9676 | sysctl_sched_rt_runtime = old_runtime; | |
9677 | } else { | |
9678 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9679 | def_rt_bandwidth.rt_period = | |
9680 | ns_to_ktime(global_rt_period()); | |
9681 | } | |
9682 | } | |
9683 | mutex_unlock(&mutex); | |
9684 | ||
9685 | return ret; | |
9686 | } | |
68318b8e | 9687 | |
052f1dc7 | 9688 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9689 | |
9690 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9691 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9692 | { |
2b01dfe3 PM |
9693 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9694 | struct task_group, css); | |
68318b8e SV |
9695 | } |
9696 | ||
9697 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9698 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9699 | { |
ec7dc8ac | 9700 | struct task_group *tg, *parent; |
68318b8e | 9701 | |
2b01dfe3 | 9702 | if (!cgrp->parent) { |
68318b8e | 9703 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9704 | return &init_task_group.css; |
9705 | } | |
9706 | ||
ec7dc8ac DG |
9707 | parent = cgroup_tg(cgrp->parent); |
9708 | tg = sched_create_group(parent); | |
68318b8e SV |
9709 | if (IS_ERR(tg)) |
9710 | return ERR_PTR(-ENOMEM); | |
9711 | ||
68318b8e SV |
9712 | return &tg->css; |
9713 | } | |
9714 | ||
41a2d6cf IM |
9715 | static void |
9716 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9717 | { |
2b01dfe3 | 9718 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9719 | |
9720 | sched_destroy_group(tg); | |
9721 | } | |
9722 | ||
41a2d6cf IM |
9723 | static int |
9724 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9725 | struct task_struct *tsk) | |
68318b8e | 9726 | { |
b68aa230 | 9727 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9728 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9729 | return -EINVAL; |
9730 | #else | |
68318b8e SV |
9731 | /* We don't support RT-tasks being in separate groups */ |
9732 | if (tsk->sched_class != &fair_sched_class) | |
9733 | return -EINVAL; | |
b68aa230 | 9734 | #endif |
68318b8e SV |
9735 | |
9736 | return 0; | |
9737 | } | |
9738 | ||
9739 | static void | |
2b01dfe3 | 9740 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9741 | struct cgroup *old_cont, struct task_struct *tsk) |
9742 | { | |
9743 | sched_move_task(tsk); | |
9744 | } | |
9745 | ||
052f1dc7 | 9746 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9747 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9748 | u64 shareval) |
68318b8e | 9749 | { |
2b01dfe3 | 9750 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9751 | } |
9752 | ||
f4c753b7 | 9753 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9754 | { |
2b01dfe3 | 9755 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9756 | |
9757 | return (u64) tg->shares; | |
9758 | } | |
6d6bc0ad | 9759 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9760 | |
052f1dc7 | 9761 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9762 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9763 | s64 val) |
6f505b16 | 9764 | { |
06ecb27c | 9765 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9766 | } |
9767 | ||
06ecb27c | 9768 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9769 | { |
06ecb27c | 9770 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9771 | } |
d0b27fa7 PZ |
9772 | |
9773 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9774 | u64 rt_period_us) | |
9775 | { | |
9776 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9777 | } | |
9778 | ||
9779 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9780 | { | |
9781 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9782 | } | |
6d6bc0ad | 9783 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9784 | |
fe5c7cc2 | 9785 | static struct cftype cpu_files[] = { |
052f1dc7 | 9786 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9787 | { |
9788 | .name = "shares", | |
f4c753b7 PM |
9789 | .read_u64 = cpu_shares_read_u64, |
9790 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9791 | }, |
052f1dc7 PZ |
9792 | #endif |
9793 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9794 | { |
9f0c1e56 | 9795 | .name = "rt_runtime_us", |
06ecb27c PM |
9796 | .read_s64 = cpu_rt_runtime_read, |
9797 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9798 | }, |
d0b27fa7 PZ |
9799 | { |
9800 | .name = "rt_period_us", | |
f4c753b7 PM |
9801 | .read_u64 = cpu_rt_period_read_uint, |
9802 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9803 | }, |
052f1dc7 | 9804 | #endif |
68318b8e SV |
9805 | }; |
9806 | ||
9807 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9808 | { | |
fe5c7cc2 | 9809 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9810 | } |
9811 | ||
9812 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9813 | .name = "cpu", |
9814 | .create = cpu_cgroup_create, | |
9815 | .destroy = cpu_cgroup_destroy, | |
9816 | .can_attach = cpu_cgroup_can_attach, | |
9817 | .attach = cpu_cgroup_attach, | |
9818 | .populate = cpu_cgroup_populate, | |
9819 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9820 | .early_init = 1, |
9821 | }; | |
9822 | ||
052f1dc7 | 9823 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9824 | |
9825 | #ifdef CONFIG_CGROUP_CPUACCT | |
9826 | ||
9827 | /* | |
9828 | * CPU accounting code for task groups. | |
9829 | * | |
9830 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9831 | * (balbir@in.ibm.com). | |
9832 | */ | |
9833 | ||
934352f2 | 9834 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9835 | struct cpuacct { |
9836 | struct cgroup_subsys_state css; | |
9837 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9838 | u64 *cpuusage; | |
934352f2 | 9839 | struct cpuacct *parent; |
d842de87 SV |
9840 | }; |
9841 | ||
9842 | struct cgroup_subsys cpuacct_subsys; | |
9843 | ||
9844 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9845 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9846 | { |
32cd756a | 9847 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9848 | struct cpuacct, css); |
9849 | } | |
9850 | ||
9851 | /* return cpu accounting group to which this task belongs */ | |
9852 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9853 | { | |
9854 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9855 | struct cpuacct, css); | |
9856 | } | |
9857 | ||
9858 | /* create a new cpu accounting group */ | |
9859 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9860 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9861 | { |
9862 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9863 | ||
9864 | if (!ca) | |
9865 | return ERR_PTR(-ENOMEM); | |
9866 | ||
9867 | ca->cpuusage = alloc_percpu(u64); | |
9868 | if (!ca->cpuusage) { | |
9869 | kfree(ca); | |
9870 | return ERR_PTR(-ENOMEM); | |
9871 | } | |
9872 | ||
934352f2 BR |
9873 | if (cgrp->parent) |
9874 | ca->parent = cgroup_ca(cgrp->parent); | |
9875 | ||
d842de87 SV |
9876 | return &ca->css; |
9877 | } | |
9878 | ||
9879 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9880 | static void |
32cd756a | 9881 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9882 | { |
32cd756a | 9883 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9884 | |
9885 | free_percpu(ca->cpuusage); | |
9886 | kfree(ca); | |
9887 | } | |
9888 | ||
720f5498 KC |
9889 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9890 | { | |
b36128c8 | 9891 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9892 | u64 data; |
9893 | ||
9894 | #ifndef CONFIG_64BIT | |
9895 | /* | |
9896 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9897 | */ | |
9898 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9899 | data = *cpuusage; | |
9900 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9901 | #else | |
9902 | data = *cpuusage; | |
9903 | #endif | |
9904 | ||
9905 | return data; | |
9906 | } | |
9907 | ||
9908 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9909 | { | |
b36128c8 | 9910 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9911 | |
9912 | #ifndef CONFIG_64BIT | |
9913 | /* | |
9914 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9915 | */ | |
9916 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9917 | *cpuusage = val; | |
9918 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9919 | #else | |
9920 | *cpuusage = val; | |
9921 | #endif | |
9922 | } | |
9923 | ||
d842de87 | 9924 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9925 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9926 | { |
32cd756a | 9927 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9928 | u64 totalcpuusage = 0; |
9929 | int i; | |
9930 | ||
720f5498 KC |
9931 | for_each_present_cpu(i) |
9932 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9933 | |
9934 | return totalcpuusage; | |
9935 | } | |
9936 | ||
0297b803 DG |
9937 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9938 | u64 reset) | |
9939 | { | |
9940 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9941 | int err = 0; | |
9942 | int i; | |
9943 | ||
9944 | if (reset) { | |
9945 | err = -EINVAL; | |
9946 | goto out; | |
9947 | } | |
9948 | ||
720f5498 KC |
9949 | for_each_present_cpu(i) |
9950 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9951 | |
0297b803 DG |
9952 | out: |
9953 | return err; | |
9954 | } | |
9955 | ||
e9515c3c KC |
9956 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9957 | struct seq_file *m) | |
9958 | { | |
9959 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9960 | u64 percpu; | |
9961 | int i; | |
9962 | ||
9963 | for_each_present_cpu(i) { | |
9964 | percpu = cpuacct_cpuusage_read(ca, i); | |
9965 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9966 | } | |
9967 | seq_printf(m, "\n"); | |
9968 | return 0; | |
9969 | } | |
9970 | ||
d842de87 SV |
9971 | static struct cftype files[] = { |
9972 | { | |
9973 | .name = "usage", | |
f4c753b7 PM |
9974 | .read_u64 = cpuusage_read, |
9975 | .write_u64 = cpuusage_write, | |
d842de87 | 9976 | }, |
e9515c3c KC |
9977 | { |
9978 | .name = "usage_percpu", | |
9979 | .read_seq_string = cpuacct_percpu_seq_read, | |
9980 | }, | |
9981 | ||
d842de87 SV |
9982 | }; |
9983 | ||
32cd756a | 9984 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9985 | { |
32cd756a | 9986 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9987 | } |
9988 | ||
9989 | /* | |
9990 | * charge this task's execution time to its accounting group. | |
9991 | * | |
9992 | * called with rq->lock held. | |
9993 | */ | |
9994 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9995 | { | |
9996 | struct cpuacct *ca; | |
934352f2 | 9997 | int cpu; |
d842de87 | 9998 | |
c40c6f85 | 9999 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10000 | return; |
10001 | ||
934352f2 | 10002 | cpu = task_cpu(tsk); |
d842de87 | 10003 | ca = task_ca(tsk); |
d842de87 | 10004 | |
934352f2 | 10005 | for (; ca; ca = ca->parent) { |
b36128c8 | 10006 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10007 | *cpuusage += cputime; |
10008 | } | |
10009 | } | |
10010 | ||
10011 | struct cgroup_subsys cpuacct_subsys = { | |
10012 | .name = "cpuacct", | |
10013 | .create = cpuacct_create, | |
10014 | .destroy = cpuacct_destroy, | |
10015 | .populate = cpuacct_populate, | |
10016 | .subsys_id = cpuacct_subsys_id, | |
10017 | }; | |
10018 | #endif /* CONFIG_CGROUP_CPUACCT */ |