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 (;;) { | |
7f1e2ca9 PZ |
234 | unsigned long delta; |
235 | ktime_t soft, hard; | |
236 | ||
d0b27fa7 PZ |
237 | if (hrtimer_active(&rt_b->rt_period_timer)) |
238 | break; | |
239 | ||
240 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
241 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
242 | |
243 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
244 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
245 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
246 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
247 | HRTIMER_MODE_ABS, 0); | |
d0b27fa7 PZ |
248 | } |
249 | spin_unlock(&rt_b->rt_runtime_lock); | |
250 | } | |
251 | ||
252 | #ifdef CONFIG_RT_GROUP_SCHED | |
253 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
254 | { | |
255 | hrtimer_cancel(&rt_b->rt_period_timer); | |
256 | } | |
257 | #endif | |
258 | ||
712555ee HC |
259 | /* |
260 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
261 | * detach_destroy_domains and partition_sched_domains. | |
262 | */ | |
263 | static DEFINE_MUTEX(sched_domains_mutex); | |
264 | ||
052f1dc7 | 265 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 266 | |
68318b8e SV |
267 | #include <linux/cgroup.h> |
268 | ||
29f59db3 SV |
269 | struct cfs_rq; |
270 | ||
6f505b16 PZ |
271 | static LIST_HEAD(task_groups); |
272 | ||
29f59db3 | 273 | /* task group related information */ |
4cf86d77 | 274 | struct task_group { |
052f1dc7 | 275 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
276 | struct cgroup_subsys_state css; |
277 | #endif | |
052f1dc7 | 278 | |
6c415b92 AB |
279 | #ifdef CONFIG_USER_SCHED |
280 | uid_t uid; | |
281 | #endif | |
282 | ||
052f1dc7 | 283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
284 | /* schedulable entities of this group on each cpu */ |
285 | struct sched_entity **se; | |
286 | /* runqueue "owned" by this group on each cpu */ | |
287 | struct cfs_rq **cfs_rq; | |
288 | unsigned long shares; | |
052f1dc7 PZ |
289 | #endif |
290 | ||
291 | #ifdef CONFIG_RT_GROUP_SCHED | |
292 | struct sched_rt_entity **rt_se; | |
293 | struct rt_rq **rt_rq; | |
294 | ||
d0b27fa7 | 295 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 296 | #endif |
6b2d7700 | 297 | |
ae8393e5 | 298 | struct rcu_head rcu; |
6f505b16 | 299 | struct list_head list; |
f473aa5e PZ |
300 | |
301 | struct task_group *parent; | |
302 | struct list_head siblings; | |
303 | struct list_head children; | |
29f59db3 SV |
304 | }; |
305 | ||
354d60c2 | 306 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 307 | |
6c415b92 AB |
308 | /* Helper function to pass uid information to create_sched_user() */ |
309 | void set_tg_uid(struct user_struct *user) | |
310 | { | |
311 | user->tg->uid = user->uid; | |
312 | } | |
313 | ||
eff766a6 PZ |
314 | /* |
315 | * Root task group. | |
316 | * Every UID task group (including init_task_group aka UID-0) will | |
317 | * be a child to this group. | |
318 | */ | |
319 | struct task_group root_task_group; | |
320 | ||
052f1dc7 | 321 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
322 | /* Default task group's sched entity on each cpu */ |
323 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
324 | /* Default task group's cfs_rq on each cpu */ | |
325 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 326 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
327 | |
328 | #ifdef CONFIG_RT_GROUP_SCHED | |
329 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
330 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 331 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 332 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 333 | #define root_task_group init_task_group |
9a7e0b18 | 334 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 335 | |
8ed36996 | 336 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
337 | * a task group's cpu shares. |
338 | */ | |
8ed36996 | 339 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 340 | |
57310a98 PZ |
341 | #ifdef CONFIG_SMP |
342 | static int root_task_group_empty(void) | |
343 | { | |
344 | return list_empty(&root_task_group.children); | |
345 | } | |
346 | #endif | |
347 | ||
052f1dc7 | 348 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
349 | #ifdef CONFIG_USER_SCHED |
350 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 351 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 352 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 353 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 354 | |
cb4ad1ff | 355 | /* |
2e084786 LJ |
356 | * A weight of 0 or 1 can cause arithmetics problems. |
357 | * A weight of a cfs_rq is the sum of weights of which entities | |
358 | * are queued on this cfs_rq, so a weight of a entity should not be | |
359 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
360 | * (The default weight is 1024 - so there's no practical |
361 | * limitation from this.) | |
362 | */ | |
18d95a28 | 363 | #define MIN_SHARES 2 |
2e084786 | 364 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 365 | |
052f1dc7 PZ |
366 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
367 | #endif | |
368 | ||
29f59db3 | 369 | /* Default task group. |
3a252015 | 370 | * Every task in system belong to this group at bootup. |
29f59db3 | 371 | */ |
434d53b0 | 372 | struct task_group init_task_group; |
29f59db3 SV |
373 | |
374 | /* return group to which a task belongs */ | |
4cf86d77 | 375 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 376 | { |
4cf86d77 | 377 | struct task_group *tg; |
9b5b7751 | 378 | |
052f1dc7 | 379 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
380 | rcu_read_lock(); |
381 | tg = __task_cred(p)->user->tg; | |
382 | rcu_read_unlock(); | |
052f1dc7 | 383 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
384 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
385 | struct task_group, css); | |
24e377a8 | 386 | #else |
41a2d6cf | 387 | tg = &init_task_group; |
24e377a8 | 388 | #endif |
9b5b7751 | 389 | return tg; |
29f59db3 SV |
390 | } |
391 | ||
392 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 393 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 394 | { |
052f1dc7 | 395 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
396 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
397 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 398 | #endif |
6f505b16 | 399 | |
052f1dc7 | 400 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
401 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
402 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 403 | #endif |
29f59db3 SV |
404 | } |
405 | ||
406 | #else | |
407 | ||
57310a98 PZ |
408 | #ifdef CONFIG_SMP |
409 | static int root_task_group_empty(void) | |
410 | { | |
411 | return 1; | |
412 | } | |
413 | #endif | |
414 | ||
6f505b16 | 415 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
416 | static inline struct task_group *task_group(struct task_struct *p) |
417 | { | |
418 | return NULL; | |
419 | } | |
29f59db3 | 420 | |
052f1dc7 | 421 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 422 | |
6aa645ea IM |
423 | /* CFS-related fields in a runqueue */ |
424 | struct cfs_rq { | |
425 | struct load_weight load; | |
426 | unsigned long nr_running; | |
427 | ||
6aa645ea | 428 | u64 exec_clock; |
e9acbff6 | 429 | u64 min_vruntime; |
6aa645ea IM |
430 | |
431 | struct rb_root tasks_timeline; | |
432 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
433 | |
434 | struct list_head tasks; | |
435 | struct list_head *balance_iterator; | |
436 | ||
437 | /* | |
438 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
439 | * It is set to NULL otherwise (i.e when none are currently running). |
440 | */ | |
4793241b | 441 | struct sched_entity *curr, *next, *last; |
ddc97297 | 442 | |
5ac5c4d6 | 443 | unsigned int nr_spread_over; |
ddc97297 | 444 | |
62160e3f | 445 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
446 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
447 | ||
41a2d6cf IM |
448 | /* |
449 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
450 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
451 | * (like users, containers etc.) | |
452 | * | |
453 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
454 | * list is used during load balance. | |
455 | */ | |
41a2d6cf IM |
456 | struct list_head leaf_cfs_rq_list; |
457 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
458 | |
459 | #ifdef CONFIG_SMP | |
c09595f6 | 460 | /* |
c8cba857 | 461 | * the part of load.weight contributed by tasks |
c09595f6 | 462 | */ |
c8cba857 | 463 | unsigned long task_weight; |
c09595f6 | 464 | |
c8cba857 PZ |
465 | /* |
466 | * h_load = weight * f(tg) | |
467 | * | |
468 | * Where f(tg) is the recursive weight fraction assigned to | |
469 | * this group. | |
470 | */ | |
471 | unsigned long h_load; | |
c09595f6 | 472 | |
c8cba857 PZ |
473 | /* |
474 | * this cpu's part of tg->shares | |
475 | */ | |
476 | unsigned long shares; | |
f1d239f7 PZ |
477 | |
478 | /* | |
479 | * load.weight at the time we set shares | |
480 | */ | |
481 | unsigned long rq_weight; | |
c09595f6 | 482 | #endif |
6aa645ea IM |
483 | #endif |
484 | }; | |
1da177e4 | 485 | |
6aa645ea IM |
486 | /* Real-Time classes' related field in a runqueue: */ |
487 | struct rt_rq { | |
488 | struct rt_prio_array active; | |
63489e45 | 489 | unsigned long rt_nr_running; |
052f1dc7 | 490 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
491 | struct { |
492 | int curr; /* highest queued rt task prio */ | |
398a153b | 493 | #ifdef CONFIG_SMP |
e864c499 | 494 | int next; /* next highest */ |
398a153b | 495 | #endif |
e864c499 | 496 | } highest_prio; |
6f505b16 | 497 | #endif |
fa85ae24 | 498 | #ifdef CONFIG_SMP |
73fe6aae | 499 | unsigned long rt_nr_migratory; |
a22d7fc1 | 500 | int overloaded; |
917b627d | 501 | struct plist_head pushable_tasks; |
fa85ae24 | 502 | #endif |
6f505b16 | 503 | int rt_throttled; |
fa85ae24 | 504 | u64 rt_time; |
ac086bc2 | 505 | u64 rt_runtime; |
ea736ed5 | 506 | /* Nests inside the rq lock: */ |
ac086bc2 | 507 | spinlock_t rt_runtime_lock; |
6f505b16 | 508 | |
052f1dc7 | 509 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
510 | unsigned long rt_nr_boosted; |
511 | ||
6f505b16 PZ |
512 | struct rq *rq; |
513 | struct list_head leaf_rt_rq_list; | |
514 | struct task_group *tg; | |
515 | struct sched_rt_entity *rt_se; | |
516 | #endif | |
6aa645ea IM |
517 | }; |
518 | ||
57d885fe GH |
519 | #ifdef CONFIG_SMP |
520 | ||
521 | /* | |
522 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
523 | * variables. Each exclusive cpuset essentially defines an island domain by |
524 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
525 | * exclusive cpuset is created, we also create and attach a new root-domain |
526 | * object. | |
527 | * | |
57d885fe GH |
528 | */ |
529 | struct root_domain { | |
530 | atomic_t refcount; | |
c6c4927b RR |
531 | cpumask_var_t span; |
532 | cpumask_var_t online; | |
637f5085 | 533 | |
0eab9146 | 534 | /* |
637f5085 GH |
535 | * The "RT overload" flag: it gets set if a CPU has more than |
536 | * one runnable RT task. | |
537 | */ | |
c6c4927b | 538 | cpumask_var_t rto_mask; |
0eab9146 | 539 | atomic_t rto_count; |
6e0534f2 GH |
540 | #ifdef CONFIG_SMP |
541 | struct cpupri cpupri; | |
542 | #endif | |
7a09b1a2 VS |
543 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
544 | /* | |
545 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
546 | * used when most cpus are idle in the system indicating overall very | |
547 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
548 | */ | |
549 | unsigned int sched_mc_preferred_wakeup_cpu; | |
550 | #endif | |
57d885fe GH |
551 | }; |
552 | ||
dc938520 GH |
553 | /* |
554 | * By default the system creates a single root-domain with all cpus as | |
555 | * members (mimicking the global state we have today). | |
556 | */ | |
57d885fe GH |
557 | static struct root_domain def_root_domain; |
558 | ||
559 | #endif | |
560 | ||
1da177e4 LT |
561 | /* |
562 | * This is the main, per-CPU runqueue data structure. | |
563 | * | |
564 | * Locking rule: those places that want to lock multiple runqueues | |
565 | * (such as the load balancing or the thread migration code), lock | |
566 | * acquire operations must be ordered by ascending &runqueue. | |
567 | */ | |
70b97a7f | 568 | struct rq { |
d8016491 IM |
569 | /* runqueue lock: */ |
570 | spinlock_t lock; | |
1da177e4 LT |
571 | |
572 | /* | |
573 | * nr_running and cpu_load should be in the same cacheline because | |
574 | * remote CPUs use both these fields when doing load calculation. | |
575 | */ | |
576 | unsigned long nr_running; | |
6aa645ea IM |
577 | #define CPU_LOAD_IDX_MAX 5 |
578 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 579 | #ifdef CONFIG_NO_HZ |
15934a37 | 580 | unsigned long last_tick_seen; |
46cb4b7c SS |
581 | unsigned char in_nohz_recently; |
582 | #endif | |
d8016491 IM |
583 | /* capture load from *all* tasks on this cpu: */ |
584 | struct load_weight load; | |
6aa645ea IM |
585 | unsigned long nr_load_updates; |
586 | u64 nr_switches; | |
587 | ||
588 | struct cfs_rq cfs; | |
6f505b16 | 589 | struct rt_rq rt; |
6f505b16 | 590 | |
6aa645ea | 591 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
592 | /* list of leaf cfs_rq on this cpu: */ |
593 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
594 | #endif |
595 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 596 | struct list_head leaf_rt_rq_list; |
1da177e4 | 597 | #endif |
1da177e4 LT |
598 | |
599 | /* | |
600 | * This is part of a global counter where only the total sum | |
601 | * over all CPUs matters. A task can increase this counter on | |
602 | * one CPU and if it got migrated afterwards it may decrease | |
603 | * it on another CPU. Always updated under the runqueue lock: | |
604 | */ | |
605 | unsigned long nr_uninterruptible; | |
606 | ||
36c8b586 | 607 | struct task_struct *curr, *idle; |
c9819f45 | 608 | unsigned long next_balance; |
1da177e4 | 609 | struct mm_struct *prev_mm; |
6aa645ea | 610 | |
3e51f33f | 611 | u64 clock; |
6aa645ea | 612 | |
1da177e4 LT |
613 | atomic_t nr_iowait; |
614 | ||
615 | #ifdef CONFIG_SMP | |
0eab9146 | 616 | struct root_domain *rd; |
1da177e4 LT |
617 | struct sched_domain *sd; |
618 | ||
a0a522ce | 619 | unsigned char idle_at_tick; |
1da177e4 LT |
620 | /* For active balancing */ |
621 | int active_balance; | |
622 | int push_cpu; | |
d8016491 IM |
623 | /* cpu of this runqueue: */ |
624 | int cpu; | |
1f11eb6a | 625 | int online; |
1da177e4 | 626 | |
a8a51d5e | 627 | unsigned long avg_load_per_task; |
1da177e4 | 628 | |
36c8b586 | 629 | struct task_struct *migration_thread; |
1da177e4 LT |
630 | struct list_head migration_queue; |
631 | #endif | |
632 | ||
dce48a84 TG |
633 | /* calc_load related fields */ |
634 | unsigned long calc_load_update; | |
635 | long calc_load_active; | |
636 | ||
8f4d37ec | 637 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
638 | #ifdef CONFIG_SMP |
639 | int hrtick_csd_pending; | |
640 | struct call_single_data hrtick_csd; | |
641 | #endif | |
8f4d37ec PZ |
642 | struct hrtimer hrtick_timer; |
643 | #endif | |
644 | ||
1da177e4 LT |
645 | #ifdef CONFIG_SCHEDSTATS |
646 | /* latency stats */ | |
647 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
648 | unsigned long long rq_cpu_time; |
649 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
650 | |
651 | /* sys_sched_yield() stats */ | |
480b9434 | 652 | unsigned int yld_count; |
1da177e4 LT |
653 | |
654 | /* schedule() stats */ | |
480b9434 KC |
655 | unsigned int sched_switch; |
656 | unsigned int sched_count; | |
657 | unsigned int sched_goidle; | |
1da177e4 LT |
658 | |
659 | /* try_to_wake_up() stats */ | |
480b9434 KC |
660 | unsigned int ttwu_count; |
661 | unsigned int ttwu_local; | |
b8efb561 IM |
662 | |
663 | /* BKL stats */ | |
480b9434 | 664 | unsigned int bkl_count; |
1da177e4 LT |
665 | #endif |
666 | }; | |
667 | ||
f34e3b61 | 668 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 669 | |
15afe09b | 670 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 671 | { |
15afe09b | 672 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
673 | } |
674 | ||
0a2966b4 CL |
675 | static inline int cpu_of(struct rq *rq) |
676 | { | |
677 | #ifdef CONFIG_SMP | |
678 | return rq->cpu; | |
679 | #else | |
680 | return 0; | |
681 | #endif | |
682 | } | |
683 | ||
674311d5 NP |
684 | /* |
685 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 686 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
687 | * |
688 | * The domain tree of any CPU may only be accessed from within | |
689 | * preempt-disabled sections. | |
690 | */ | |
48f24c4d IM |
691 | #define for_each_domain(cpu, __sd) \ |
692 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
693 | |
694 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
695 | #define this_rq() (&__get_cpu_var(runqueues)) | |
696 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
697 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
698 | ||
3e51f33f PZ |
699 | static inline void update_rq_clock(struct rq *rq) |
700 | { | |
701 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
702 | } | |
703 | ||
bf5c91ba IM |
704 | /* |
705 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
706 | */ | |
707 | #ifdef CONFIG_SCHED_DEBUG | |
708 | # define const_debug __read_mostly | |
709 | #else | |
710 | # define const_debug static const | |
711 | #endif | |
712 | ||
017730c1 IM |
713 | /** |
714 | * runqueue_is_locked | |
715 | * | |
716 | * Returns true if the current cpu runqueue is locked. | |
717 | * This interface allows printk to be called with the runqueue lock | |
718 | * held and know whether or not it is OK to wake up the klogd. | |
719 | */ | |
720 | int runqueue_is_locked(void) | |
721 | { | |
722 | int cpu = get_cpu(); | |
723 | struct rq *rq = cpu_rq(cpu); | |
724 | int ret; | |
725 | ||
726 | ret = spin_is_locked(&rq->lock); | |
727 | put_cpu(); | |
728 | return ret; | |
729 | } | |
730 | ||
bf5c91ba IM |
731 | /* |
732 | * Debugging: various feature bits | |
733 | */ | |
f00b45c1 PZ |
734 | |
735 | #define SCHED_FEAT(name, enabled) \ | |
736 | __SCHED_FEAT_##name , | |
737 | ||
bf5c91ba | 738 | enum { |
f00b45c1 | 739 | #include "sched_features.h" |
bf5c91ba IM |
740 | }; |
741 | ||
f00b45c1 PZ |
742 | #undef SCHED_FEAT |
743 | ||
744 | #define SCHED_FEAT(name, enabled) \ | |
745 | (1UL << __SCHED_FEAT_##name) * enabled | | |
746 | ||
bf5c91ba | 747 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
748 | #include "sched_features.h" |
749 | 0; | |
750 | ||
751 | #undef SCHED_FEAT | |
752 | ||
753 | #ifdef CONFIG_SCHED_DEBUG | |
754 | #define SCHED_FEAT(name, enabled) \ | |
755 | #name , | |
756 | ||
983ed7a6 | 757 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
758 | #include "sched_features.h" |
759 | NULL | |
760 | }; | |
761 | ||
762 | #undef SCHED_FEAT | |
763 | ||
34f3a814 | 764 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 765 | { |
f00b45c1 PZ |
766 | int i; |
767 | ||
768 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
769 | if (!(sysctl_sched_features & (1UL << i))) |
770 | seq_puts(m, "NO_"); | |
771 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 772 | } |
34f3a814 | 773 | seq_puts(m, "\n"); |
f00b45c1 | 774 | |
34f3a814 | 775 | return 0; |
f00b45c1 PZ |
776 | } |
777 | ||
778 | static ssize_t | |
779 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
780 | size_t cnt, loff_t *ppos) | |
781 | { | |
782 | char buf[64]; | |
783 | char *cmp = buf; | |
784 | int neg = 0; | |
785 | int i; | |
786 | ||
787 | if (cnt > 63) | |
788 | cnt = 63; | |
789 | ||
790 | if (copy_from_user(&buf, ubuf, cnt)) | |
791 | return -EFAULT; | |
792 | ||
793 | buf[cnt] = 0; | |
794 | ||
c24b7c52 | 795 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
796 | neg = 1; |
797 | cmp += 3; | |
798 | } | |
799 | ||
800 | for (i = 0; sched_feat_names[i]; i++) { | |
801 | int len = strlen(sched_feat_names[i]); | |
802 | ||
803 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
804 | if (neg) | |
805 | sysctl_sched_features &= ~(1UL << i); | |
806 | else | |
807 | sysctl_sched_features |= (1UL << i); | |
808 | break; | |
809 | } | |
810 | } | |
811 | ||
812 | if (!sched_feat_names[i]) | |
813 | return -EINVAL; | |
814 | ||
815 | filp->f_pos += cnt; | |
816 | ||
817 | return cnt; | |
818 | } | |
819 | ||
34f3a814 LZ |
820 | static int sched_feat_open(struct inode *inode, struct file *filp) |
821 | { | |
822 | return single_open(filp, sched_feat_show, NULL); | |
823 | } | |
824 | ||
f00b45c1 | 825 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
826 | .open = sched_feat_open, |
827 | .write = sched_feat_write, | |
828 | .read = seq_read, | |
829 | .llseek = seq_lseek, | |
830 | .release = single_release, | |
f00b45c1 PZ |
831 | }; |
832 | ||
833 | static __init int sched_init_debug(void) | |
834 | { | |
f00b45c1 PZ |
835 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
836 | &sched_feat_fops); | |
837 | ||
838 | return 0; | |
839 | } | |
840 | late_initcall(sched_init_debug); | |
841 | ||
842 | #endif | |
843 | ||
844 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 845 | |
b82d9fdd PZ |
846 | /* |
847 | * Number of tasks to iterate in a single balance run. | |
848 | * Limited because this is done with IRQs disabled. | |
849 | */ | |
850 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
851 | ||
2398f2c6 PZ |
852 | /* |
853 | * ratelimit for updating the group shares. | |
55cd5340 | 854 | * default: 0.25ms |
2398f2c6 | 855 | */ |
55cd5340 | 856 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 857 | |
ffda12a1 PZ |
858 | /* |
859 | * Inject some fuzzyness into changing the per-cpu group shares | |
860 | * this avoids remote rq-locks at the expense of fairness. | |
861 | * default: 4 | |
862 | */ | |
863 | unsigned int sysctl_sched_shares_thresh = 4; | |
864 | ||
fa85ae24 | 865 | /* |
9f0c1e56 | 866 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
867 | * default: 1s |
868 | */ | |
9f0c1e56 | 869 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 870 | |
6892b75e IM |
871 | static __read_mostly int scheduler_running; |
872 | ||
9f0c1e56 PZ |
873 | /* |
874 | * part of the period that we allow rt tasks to run in us. | |
875 | * default: 0.95s | |
876 | */ | |
877 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 878 | |
d0b27fa7 PZ |
879 | static inline u64 global_rt_period(void) |
880 | { | |
881 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
882 | } | |
883 | ||
884 | static inline u64 global_rt_runtime(void) | |
885 | { | |
e26873bb | 886 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
887 | return RUNTIME_INF; |
888 | ||
889 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
890 | } | |
fa85ae24 | 891 | |
1da177e4 | 892 | #ifndef prepare_arch_switch |
4866cde0 NP |
893 | # define prepare_arch_switch(next) do { } while (0) |
894 | #endif | |
895 | #ifndef finish_arch_switch | |
896 | # define finish_arch_switch(prev) do { } while (0) | |
897 | #endif | |
898 | ||
051a1d1a DA |
899 | static inline int task_current(struct rq *rq, struct task_struct *p) |
900 | { | |
901 | return rq->curr == p; | |
902 | } | |
903 | ||
4866cde0 | 904 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 905 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 906 | { |
051a1d1a | 907 | return task_current(rq, p); |
4866cde0 NP |
908 | } |
909 | ||
70b97a7f | 910 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
911 | { |
912 | } | |
913 | ||
70b97a7f | 914 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 915 | { |
da04c035 IM |
916 | #ifdef CONFIG_DEBUG_SPINLOCK |
917 | /* this is a valid case when another task releases the spinlock */ | |
918 | rq->lock.owner = current; | |
919 | #endif | |
8a25d5de IM |
920 | /* |
921 | * If we are tracking spinlock dependencies then we have to | |
922 | * fix up the runqueue lock - which gets 'carried over' from | |
923 | * prev into current: | |
924 | */ | |
925 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
926 | ||
4866cde0 NP |
927 | spin_unlock_irq(&rq->lock); |
928 | } | |
929 | ||
930 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 931 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
932 | { |
933 | #ifdef CONFIG_SMP | |
934 | return p->oncpu; | |
935 | #else | |
051a1d1a | 936 | return task_current(rq, p); |
4866cde0 NP |
937 | #endif |
938 | } | |
939 | ||
70b97a7f | 940 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
941 | { |
942 | #ifdef CONFIG_SMP | |
943 | /* | |
944 | * We can optimise this out completely for !SMP, because the | |
945 | * SMP rebalancing from interrupt is the only thing that cares | |
946 | * here. | |
947 | */ | |
948 | next->oncpu = 1; | |
949 | #endif | |
950 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
951 | spin_unlock_irq(&rq->lock); | |
952 | #else | |
953 | spin_unlock(&rq->lock); | |
954 | #endif | |
955 | } | |
956 | ||
70b97a7f | 957 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
958 | { |
959 | #ifdef CONFIG_SMP | |
960 | /* | |
961 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
962 | * We must ensure this doesn't happen until the switch is completely | |
963 | * finished. | |
964 | */ | |
965 | smp_wmb(); | |
966 | prev->oncpu = 0; | |
967 | #endif | |
968 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
969 | local_irq_enable(); | |
1da177e4 | 970 | #endif |
4866cde0 NP |
971 | } |
972 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 973 | |
b29739f9 IM |
974 | /* |
975 | * __task_rq_lock - lock the runqueue a given task resides on. | |
976 | * Must be called interrupts disabled. | |
977 | */ | |
70b97a7f | 978 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
979 | __acquires(rq->lock) |
980 | { | |
3a5c359a AK |
981 | for (;;) { |
982 | struct rq *rq = task_rq(p); | |
983 | spin_lock(&rq->lock); | |
984 | if (likely(rq == task_rq(p))) | |
985 | return rq; | |
b29739f9 | 986 | spin_unlock(&rq->lock); |
b29739f9 | 987 | } |
b29739f9 IM |
988 | } |
989 | ||
1da177e4 LT |
990 | /* |
991 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 992 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
993 | * explicitly disabling preemption. |
994 | */ | |
70b97a7f | 995 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
996 | __acquires(rq->lock) |
997 | { | |
70b97a7f | 998 | struct rq *rq; |
1da177e4 | 999 | |
3a5c359a AK |
1000 | for (;;) { |
1001 | local_irq_save(*flags); | |
1002 | rq = task_rq(p); | |
1003 | spin_lock(&rq->lock); | |
1004 | if (likely(rq == task_rq(p))) | |
1005 | return rq; | |
1da177e4 | 1006 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1007 | } |
1da177e4 LT |
1008 | } |
1009 | ||
ad474cac ON |
1010 | void task_rq_unlock_wait(struct task_struct *p) |
1011 | { | |
1012 | struct rq *rq = task_rq(p); | |
1013 | ||
1014 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1015 | spin_unlock_wait(&rq->lock); | |
1016 | } | |
1017 | ||
a9957449 | 1018 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1019 | __releases(rq->lock) |
1020 | { | |
1021 | spin_unlock(&rq->lock); | |
1022 | } | |
1023 | ||
70b97a7f | 1024 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1025 | __releases(rq->lock) |
1026 | { | |
1027 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1028 | } | |
1029 | ||
1da177e4 | 1030 | /* |
cc2a73b5 | 1031 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1032 | */ |
a9957449 | 1033 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1034 | __acquires(rq->lock) |
1035 | { | |
70b97a7f | 1036 | struct rq *rq; |
1da177e4 LT |
1037 | |
1038 | local_irq_disable(); | |
1039 | rq = this_rq(); | |
1040 | spin_lock(&rq->lock); | |
1041 | ||
1042 | return rq; | |
1043 | } | |
1044 | ||
8f4d37ec PZ |
1045 | #ifdef CONFIG_SCHED_HRTICK |
1046 | /* | |
1047 | * Use HR-timers to deliver accurate preemption points. | |
1048 | * | |
1049 | * Its all a bit involved since we cannot program an hrt while holding the | |
1050 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1051 | * reschedule event. | |
1052 | * | |
1053 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1054 | * rq->lock. | |
1055 | */ | |
8f4d37ec PZ |
1056 | |
1057 | /* | |
1058 | * Use hrtick when: | |
1059 | * - enabled by features | |
1060 | * - hrtimer is actually high res | |
1061 | */ | |
1062 | static inline int hrtick_enabled(struct rq *rq) | |
1063 | { | |
1064 | if (!sched_feat(HRTICK)) | |
1065 | return 0; | |
ba42059f | 1066 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1067 | return 0; |
8f4d37ec PZ |
1068 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1069 | } | |
1070 | ||
8f4d37ec PZ |
1071 | static void hrtick_clear(struct rq *rq) |
1072 | { | |
1073 | if (hrtimer_active(&rq->hrtick_timer)) | |
1074 | hrtimer_cancel(&rq->hrtick_timer); | |
1075 | } | |
1076 | ||
8f4d37ec PZ |
1077 | /* |
1078 | * High-resolution timer tick. | |
1079 | * Runs from hardirq context with interrupts disabled. | |
1080 | */ | |
1081 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1082 | { | |
1083 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1084 | ||
1085 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1086 | ||
1087 | spin_lock(&rq->lock); | |
3e51f33f | 1088 | update_rq_clock(rq); |
8f4d37ec PZ |
1089 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1090 | spin_unlock(&rq->lock); | |
1091 | ||
1092 | return HRTIMER_NORESTART; | |
1093 | } | |
1094 | ||
95e904c7 | 1095 | #ifdef CONFIG_SMP |
31656519 PZ |
1096 | /* |
1097 | * called from hardirq (IPI) context | |
1098 | */ | |
1099 | static void __hrtick_start(void *arg) | |
b328ca18 | 1100 | { |
31656519 | 1101 | struct rq *rq = arg; |
b328ca18 | 1102 | |
31656519 PZ |
1103 | spin_lock(&rq->lock); |
1104 | hrtimer_restart(&rq->hrtick_timer); | |
1105 | rq->hrtick_csd_pending = 0; | |
1106 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1107 | } |
1108 | ||
31656519 PZ |
1109 | /* |
1110 | * Called to set the hrtick timer state. | |
1111 | * | |
1112 | * called with rq->lock held and irqs disabled | |
1113 | */ | |
1114 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1115 | { |
31656519 PZ |
1116 | struct hrtimer *timer = &rq->hrtick_timer; |
1117 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1118 | |
cc584b21 | 1119 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1120 | |
1121 | if (rq == this_rq()) { | |
1122 | hrtimer_restart(timer); | |
1123 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1124 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1125 | rq->hrtick_csd_pending = 1; |
1126 | } | |
b328ca18 PZ |
1127 | } |
1128 | ||
1129 | static int | |
1130 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1131 | { | |
1132 | int cpu = (int)(long)hcpu; | |
1133 | ||
1134 | switch (action) { | |
1135 | case CPU_UP_CANCELED: | |
1136 | case CPU_UP_CANCELED_FROZEN: | |
1137 | case CPU_DOWN_PREPARE: | |
1138 | case CPU_DOWN_PREPARE_FROZEN: | |
1139 | case CPU_DEAD: | |
1140 | case CPU_DEAD_FROZEN: | |
31656519 | 1141 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1142 | return NOTIFY_OK; |
1143 | } | |
1144 | ||
1145 | return NOTIFY_DONE; | |
1146 | } | |
1147 | ||
fa748203 | 1148 | static __init void init_hrtick(void) |
b328ca18 PZ |
1149 | { |
1150 | hotcpu_notifier(hotplug_hrtick, 0); | |
1151 | } | |
31656519 PZ |
1152 | #else |
1153 | /* | |
1154 | * Called to set the hrtick timer state. | |
1155 | * | |
1156 | * called with rq->lock held and irqs disabled | |
1157 | */ | |
1158 | static void hrtick_start(struct rq *rq, u64 delay) | |
1159 | { | |
7f1e2ca9 PZ |
1160 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
1161 | HRTIMER_MODE_REL, 0); | |
31656519 | 1162 | } |
b328ca18 | 1163 | |
006c75f1 | 1164 | static inline void init_hrtick(void) |
8f4d37ec | 1165 | { |
8f4d37ec | 1166 | } |
31656519 | 1167 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1168 | |
31656519 | 1169 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1170 | { |
31656519 PZ |
1171 | #ifdef CONFIG_SMP |
1172 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1173 | |
31656519 PZ |
1174 | rq->hrtick_csd.flags = 0; |
1175 | rq->hrtick_csd.func = __hrtick_start; | |
1176 | rq->hrtick_csd.info = rq; | |
1177 | #endif | |
8f4d37ec | 1178 | |
31656519 PZ |
1179 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1180 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1181 | } |
006c75f1 | 1182 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1183 | static inline void hrtick_clear(struct rq *rq) |
1184 | { | |
1185 | } | |
1186 | ||
8f4d37ec PZ |
1187 | static inline void init_rq_hrtick(struct rq *rq) |
1188 | { | |
1189 | } | |
1190 | ||
b328ca18 PZ |
1191 | static inline void init_hrtick(void) |
1192 | { | |
1193 | } | |
006c75f1 | 1194 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1195 | |
c24d20db IM |
1196 | /* |
1197 | * resched_task - mark a task 'to be rescheduled now'. | |
1198 | * | |
1199 | * On UP this means the setting of the need_resched flag, on SMP it | |
1200 | * might also involve a cross-CPU call to trigger the scheduler on | |
1201 | * the target CPU. | |
1202 | */ | |
1203 | #ifdef CONFIG_SMP | |
1204 | ||
1205 | #ifndef tsk_is_polling | |
1206 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1207 | #endif | |
1208 | ||
31656519 | 1209 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1210 | { |
1211 | int cpu; | |
1212 | ||
1213 | assert_spin_locked(&task_rq(p)->lock); | |
1214 | ||
5ed0cec0 | 1215 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1216 | return; |
1217 | ||
5ed0cec0 | 1218 | set_tsk_need_resched(p); |
c24d20db IM |
1219 | |
1220 | cpu = task_cpu(p); | |
1221 | if (cpu == smp_processor_id()) | |
1222 | return; | |
1223 | ||
1224 | /* NEED_RESCHED must be visible before we test polling */ | |
1225 | smp_mb(); | |
1226 | if (!tsk_is_polling(p)) | |
1227 | smp_send_reschedule(cpu); | |
1228 | } | |
1229 | ||
1230 | static void resched_cpu(int cpu) | |
1231 | { | |
1232 | struct rq *rq = cpu_rq(cpu); | |
1233 | unsigned long flags; | |
1234 | ||
1235 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1236 | return; | |
1237 | resched_task(cpu_curr(cpu)); | |
1238 | spin_unlock_irqrestore(&rq->lock, flags); | |
1239 | } | |
06d8308c TG |
1240 | |
1241 | #ifdef CONFIG_NO_HZ | |
1242 | /* | |
1243 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1244 | * idle CPU then this timer might expire before the next timer event | |
1245 | * which is scheduled to wake up that CPU. In case of a completely | |
1246 | * idle system the next event might even be infinite time into the | |
1247 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1248 | * leaves the inner idle loop so the newly added timer is taken into | |
1249 | * account when the CPU goes back to idle and evaluates the timer | |
1250 | * wheel for the next timer event. | |
1251 | */ | |
1252 | void wake_up_idle_cpu(int cpu) | |
1253 | { | |
1254 | struct rq *rq = cpu_rq(cpu); | |
1255 | ||
1256 | if (cpu == smp_processor_id()) | |
1257 | return; | |
1258 | ||
1259 | /* | |
1260 | * This is safe, as this function is called with the timer | |
1261 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1262 | * to idle and has not yet set rq->curr to idle then it will | |
1263 | * be serialized on the timer wheel base lock and take the new | |
1264 | * timer into account automatically. | |
1265 | */ | |
1266 | if (rq->curr != rq->idle) | |
1267 | return; | |
1268 | ||
1269 | /* | |
1270 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1271 | * lockless. The worst case is that the other CPU runs the | |
1272 | * idle task through an additional NOOP schedule() | |
1273 | */ | |
5ed0cec0 | 1274 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1275 | |
1276 | /* NEED_RESCHED must be visible before we test polling */ | |
1277 | smp_mb(); | |
1278 | if (!tsk_is_polling(rq->idle)) | |
1279 | smp_send_reschedule(cpu); | |
1280 | } | |
6d6bc0ad | 1281 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1282 | |
6d6bc0ad | 1283 | #else /* !CONFIG_SMP */ |
31656519 | 1284 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1285 | { |
1286 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1287 | set_tsk_need_resched(p); |
c24d20db | 1288 | } |
6d6bc0ad | 1289 | #endif /* CONFIG_SMP */ |
c24d20db | 1290 | |
45bf76df IM |
1291 | #if BITS_PER_LONG == 32 |
1292 | # define WMULT_CONST (~0UL) | |
1293 | #else | |
1294 | # define WMULT_CONST (1UL << 32) | |
1295 | #endif | |
1296 | ||
1297 | #define WMULT_SHIFT 32 | |
1298 | ||
194081eb IM |
1299 | /* |
1300 | * Shift right and round: | |
1301 | */ | |
cf2ab469 | 1302 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1303 | |
a7be37ac PZ |
1304 | /* |
1305 | * delta *= weight / lw | |
1306 | */ | |
cb1c4fc9 | 1307 | static unsigned long |
45bf76df IM |
1308 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1309 | struct load_weight *lw) | |
1310 | { | |
1311 | u64 tmp; | |
1312 | ||
7a232e03 LJ |
1313 | if (!lw->inv_weight) { |
1314 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1315 | lw->inv_weight = 1; | |
1316 | else | |
1317 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1318 | / (lw->weight+1); | |
1319 | } | |
45bf76df IM |
1320 | |
1321 | tmp = (u64)delta_exec * weight; | |
1322 | /* | |
1323 | * Check whether we'd overflow the 64-bit multiplication: | |
1324 | */ | |
194081eb | 1325 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1326 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1327 | WMULT_SHIFT/2); |
1328 | else | |
cf2ab469 | 1329 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1330 | |
ecf691da | 1331 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1332 | } |
1333 | ||
1091985b | 1334 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1335 | { |
1336 | lw->weight += inc; | |
e89996ae | 1337 | lw->inv_weight = 0; |
45bf76df IM |
1338 | } |
1339 | ||
1091985b | 1340 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1341 | { |
1342 | lw->weight -= dec; | |
e89996ae | 1343 | lw->inv_weight = 0; |
45bf76df IM |
1344 | } |
1345 | ||
2dd73a4f PW |
1346 | /* |
1347 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1348 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1349 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1350 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1351 | * scaled version of the new time slice allocation that they receive on time |
1352 | * slice expiry etc. | |
1353 | */ | |
1354 | ||
cce7ade8 PZ |
1355 | #define WEIGHT_IDLEPRIO 3 |
1356 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1357 | |
1358 | /* | |
1359 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1360 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1361 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1362 | * that remained on nice 0. | |
1363 | * | |
1364 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1365 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1366 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1367 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1368 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1369 | */ |
1370 | static const int prio_to_weight[40] = { | |
254753dc IM |
1371 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1372 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1373 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1374 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1375 | /* 0 */ 1024, 820, 655, 526, 423, | |
1376 | /* 5 */ 335, 272, 215, 172, 137, | |
1377 | /* 10 */ 110, 87, 70, 56, 45, | |
1378 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1379 | }; |
1380 | ||
5714d2de IM |
1381 | /* |
1382 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1383 | * | |
1384 | * In cases where the weight does not change often, we can use the | |
1385 | * precalculated inverse to speed up arithmetics by turning divisions | |
1386 | * into multiplications: | |
1387 | */ | |
dd41f596 | 1388 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1389 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1390 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1391 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1392 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1393 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1394 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1395 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1396 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1397 | }; |
2dd73a4f | 1398 | |
dd41f596 IM |
1399 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1400 | ||
1401 | /* | |
1402 | * runqueue iterator, to support SMP load-balancing between different | |
1403 | * scheduling classes, without having to expose their internal data | |
1404 | * structures to the load-balancing proper: | |
1405 | */ | |
1406 | struct rq_iterator { | |
1407 | void *arg; | |
1408 | struct task_struct *(*start)(void *); | |
1409 | struct task_struct *(*next)(void *); | |
1410 | }; | |
1411 | ||
e1d1484f PW |
1412 | #ifdef CONFIG_SMP |
1413 | static unsigned long | |
1414 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1415 | unsigned long max_load_move, struct sched_domain *sd, | |
1416 | enum cpu_idle_type idle, int *all_pinned, | |
1417 | int *this_best_prio, struct rq_iterator *iterator); | |
1418 | ||
1419 | static int | |
1420 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1421 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1422 | struct rq_iterator *iterator); | |
e1d1484f | 1423 | #endif |
dd41f596 | 1424 | |
ef12fefa BR |
1425 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1426 | enum cpuacct_stat_index { | |
1427 | CPUACCT_STAT_USER, /* ... user mode */ | |
1428 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1429 | ||
1430 | CPUACCT_STAT_NSTATS, | |
1431 | }; | |
1432 | ||
d842de87 SV |
1433 | #ifdef CONFIG_CGROUP_CPUACCT |
1434 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1435 | static void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1437 | #else |
1438 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1439 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1440 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1441 | #endif |
1442 | ||
18d95a28 PZ |
1443 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1444 | { | |
1445 | update_load_add(&rq->load, load); | |
1446 | } | |
1447 | ||
1448 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1449 | { | |
1450 | update_load_sub(&rq->load, load); | |
1451 | } | |
1452 | ||
7940ca36 | 1453 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1454 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1455 | |
1456 | /* | |
1457 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1458 | * leaving it for the final time. | |
1459 | */ | |
eb755805 | 1460 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1461 | { |
1462 | struct task_group *parent, *child; | |
eb755805 | 1463 | int ret; |
c09595f6 PZ |
1464 | |
1465 | rcu_read_lock(); | |
1466 | parent = &root_task_group; | |
1467 | down: | |
eb755805 PZ |
1468 | ret = (*down)(parent, data); |
1469 | if (ret) | |
1470 | goto out_unlock; | |
c09595f6 PZ |
1471 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1472 | parent = child; | |
1473 | goto down; | |
1474 | ||
1475 | up: | |
1476 | continue; | |
1477 | } | |
eb755805 PZ |
1478 | ret = (*up)(parent, data); |
1479 | if (ret) | |
1480 | goto out_unlock; | |
c09595f6 PZ |
1481 | |
1482 | child = parent; | |
1483 | parent = parent->parent; | |
1484 | if (parent) | |
1485 | goto up; | |
eb755805 | 1486 | out_unlock: |
c09595f6 | 1487 | rcu_read_unlock(); |
eb755805 PZ |
1488 | |
1489 | return ret; | |
c09595f6 PZ |
1490 | } |
1491 | ||
eb755805 PZ |
1492 | static int tg_nop(struct task_group *tg, void *data) |
1493 | { | |
1494 | return 0; | |
c09595f6 | 1495 | } |
eb755805 PZ |
1496 | #endif |
1497 | ||
1498 | #ifdef CONFIG_SMP | |
1499 | static unsigned long source_load(int cpu, int type); | |
1500 | static unsigned long target_load(int cpu, int type); | |
1501 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1502 | ||
1503 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1504 | { | |
1505 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1506 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1507 | |
4cd42620 SR |
1508 | if (nr_running) |
1509 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1510 | else |
1511 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1512 | |
1513 | return rq->avg_load_per_task; | |
1514 | } | |
1515 | ||
1516 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1517 | |
c09595f6 PZ |
1518 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1519 | ||
1520 | /* | |
1521 | * Calculate and set the cpu's group shares. | |
1522 | */ | |
1523 | static void | |
ffda12a1 PZ |
1524 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1525 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1526 | { |
c09595f6 PZ |
1527 | unsigned long shares; |
1528 | unsigned long rq_weight; | |
1529 | ||
c8cba857 | 1530 | if (!tg->se[cpu]) |
c09595f6 PZ |
1531 | return; |
1532 | ||
ec4e0e2f | 1533 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1534 | |
c09595f6 PZ |
1535 | /* |
1536 | * \Sum shares * rq_weight | |
1537 | * shares = ----------------------- | |
1538 | * \Sum rq_weight | |
1539 | * | |
1540 | */ | |
ec4e0e2f | 1541 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1542 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1543 | |
ffda12a1 PZ |
1544 | if (abs(shares - tg->se[cpu]->load.weight) > |
1545 | sysctl_sched_shares_thresh) { | |
1546 | struct rq *rq = cpu_rq(cpu); | |
1547 | unsigned long flags; | |
c09595f6 | 1548 | |
ffda12a1 | 1549 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1550 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1551 | |
ffda12a1 PZ |
1552 | __set_se_shares(tg->se[cpu], shares); |
1553 | spin_unlock_irqrestore(&rq->lock, flags); | |
1554 | } | |
18d95a28 | 1555 | } |
c09595f6 PZ |
1556 | |
1557 | /* | |
c8cba857 PZ |
1558 | * Re-compute the task group their per cpu shares over the given domain. |
1559 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1560 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1561 | */ |
eb755805 | 1562 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1563 | { |
ec4e0e2f | 1564 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1565 | unsigned long shares = 0; |
eb755805 | 1566 | struct sched_domain *sd = data; |
c8cba857 | 1567 | int i; |
c09595f6 | 1568 | |
758b2cdc | 1569 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1570 | /* |
1571 | * If there are currently no tasks on the cpu pretend there | |
1572 | * is one of average load so that when a new task gets to | |
1573 | * run here it will not get delayed by group starvation. | |
1574 | */ | |
1575 | weight = tg->cfs_rq[i]->load.weight; | |
1576 | if (!weight) | |
1577 | weight = NICE_0_LOAD; | |
1578 | ||
1579 | tg->cfs_rq[i]->rq_weight = weight; | |
1580 | rq_weight += weight; | |
c8cba857 | 1581 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1582 | } |
c09595f6 | 1583 | |
c8cba857 PZ |
1584 | if ((!shares && rq_weight) || shares > tg->shares) |
1585 | shares = tg->shares; | |
1586 | ||
1587 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1588 | shares = tg->shares; | |
c09595f6 | 1589 | |
758b2cdc | 1590 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1591 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1592 | |
1593 | return 0; | |
c09595f6 PZ |
1594 | } |
1595 | ||
1596 | /* | |
c8cba857 PZ |
1597 | * Compute the cpu's hierarchical load factor for each task group. |
1598 | * This needs to be done in a top-down fashion because the load of a child | |
1599 | * group is a fraction of its parents load. | |
c09595f6 | 1600 | */ |
eb755805 | 1601 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1602 | { |
c8cba857 | 1603 | unsigned long load; |
eb755805 | 1604 | long cpu = (long)data; |
c09595f6 | 1605 | |
c8cba857 PZ |
1606 | if (!tg->parent) { |
1607 | load = cpu_rq(cpu)->load.weight; | |
1608 | } else { | |
1609 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1610 | load *= tg->cfs_rq[cpu]->shares; | |
1611 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1612 | } | |
c09595f6 | 1613 | |
c8cba857 | 1614 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1615 | |
eb755805 | 1616 | return 0; |
c09595f6 PZ |
1617 | } |
1618 | ||
c8cba857 | 1619 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1620 | { |
2398f2c6 PZ |
1621 | u64 now = cpu_clock(raw_smp_processor_id()); |
1622 | s64 elapsed = now - sd->last_update; | |
1623 | ||
1624 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1625 | sd->last_update = now; | |
eb755805 | 1626 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1627 | } |
4d8d595d PZ |
1628 | } |
1629 | ||
3e5459b4 PZ |
1630 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1631 | { | |
1632 | spin_unlock(&rq->lock); | |
1633 | update_shares(sd); | |
1634 | spin_lock(&rq->lock); | |
1635 | } | |
1636 | ||
eb755805 | 1637 | static void update_h_load(long cpu) |
c09595f6 | 1638 | { |
eb755805 | 1639 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1640 | } |
1641 | ||
c09595f6 PZ |
1642 | #else |
1643 | ||
c8cba857 | 1644 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1645 | { |
1646 | } | |
1647 | ||
3e5459b4 PZ |
1648 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1649 | { | |
1650 | } | |
1651 | ||
18d95a28 PZ |
1652 | #endif |
1653 | ||
8f45e2b5 GH |
1654 | #ifdef CONFIG_PREEMPT |
1655 | ||
70574a99 | 1656 | /* |
8f45e2b5 GH |
1657 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1658 | * way at the expense of forcing extra atomic operations in all | |
1659 | * invocations. This assures that the double_lock is acquired using the | |
1660 | * same underlying policy as the spinlock_t on this architecture, which | |
1661 | * reduces latency compared to the unfair variant below. However, it | |
1662 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1663 | */ |
8f45e2b5 GH |
1664 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1665 | __releases(this_rq->lock) | |
1666 | __acquires(busiest->lock) | |
1667 | __acquires(this_rq->lock) | |
1668 | { | |
1669 | spin_unlock(&this_rq->lock); | |
1670 | double_rq_lock(this_rq, busiest); | |
1671 | ||
1672 | return 1; | |
1673 | } | |
1674 | ||
1675 | #else | |
1676 | /* | |
1677 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1678 | * latency by eliminating extra atomic operations when the locks are | |
1679 | * already in proper order on entry. This favors lower cpu-ids and will | |
1680 | * grant the double lock to lower cpus over higher ids under contention, | |
1681 | * regardless of entry order into the function. | |
1682 | */ | |
1683 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1684 | __releases(this_rq->lock) |
1685 | __acquires(busiest->lock) | |
1686 | __acquires(this_rq->lock) | |
1687 | { | |
1688 | int ret = 0; | |
1689 | ||
70574a99 AD |
1690 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1691 | if (busiest < this_rq) { | |
1692 | spin_unlock(&this_rq->lock); | |
1693 | spin_lock(&busiest->lock); | |
1694 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1695 | ret = 1; | |
1696 | } else | |
1697 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1698 | } | |
1699 | return ret; | |
1700 | } | |
1701 | ||
8f45e2b5 GH |
1702 | #endif /* CONFIG_PREEMPT */ |
1703 | ||
1704 | /* | |
1705 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1706 | */ | |
1707 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1708 | { | |
1709 | if (unlikely(!irqs_disabled())) { | |
1710 | /* printk() doesn't work good under rq->lock */ | |
1711 | spin_unlock(&this_rq->lock); | |
1712 | BUG_ON(1); | |
1713 | } | |
1714 | ||
1715 | return _double_lock_balance(this_rq, busiest); | |
1716 | } | |
1717 | ||
70574a99 AD |
1718 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1719 | __releases(busiest->lock) | |
1720 | { | |
1721 | spin_unlock(&busiest->lock); | |
1722 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1723 | } | |
18d95a28 PZ |
1724 | #endif |
1725 | ||
30432094 | 1726 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1727 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1728 | { | |
30432094 | 1729 | #ifdef CONFIG_SMP |
34e83e85 IM |
1730 | cfs_rq->shares = shares; |
1731 | #endif | |
1732 | } | |
30432094 | 1733 | #endif |
e7693a36 | 1734 | |
dce48a84 TG |
1735 | static void calc_load_account_active(struct rq *this_rq); |
1736 | ||
dd41f596 | 1737 | #include "sched_stats.h" |
dd41f596 | 1738 | #include "sched_idletask.c" |
5522d5d5 IM |
1739 | #include "sched_fair.c" |
1740 | #include "sched_rt.c" | |
dd41f596 IM |
1741 | #ifdef CONFIG_SCHED_DEBUG |
1742 | # include "sched_debug.c" | |
1743 | #endif | |
1744 | ||
1745 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1746 | #define for_each_class(class) \ |
1747 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1748 | |
c09595f6 | 1749 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1750 | { |
1751 | rq->nr_running++; | |
9c217245 IM |
1752 | } |
1753 | ||
c09595f6 | 1754 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1755 | { |
1756 | rq->nr_running--; | |
9c217245 IM |
1757 | } |
1758 | ||
45bf76df IM |
1759 | static void set_load_weight(struct task_struct *p) |
1760 | { | |
1761 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1762 | p->se.load.weight = prio_to_weight[0] * 2; |
1763 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1764 | return; | |
1765 | } | |
45bf76df | 1766 | |
dd41f596 IM |
1767 | /* |
1768 | * SCHED_IDLE tasks get minimal weight: | |
1769 | */ | |
1770 | if (p->policy == SCHED_IDLE) { | |
1771 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1772 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1773 | return; | |
1774 | } | |
71f8bd46 | 1775 | |
dd41f596 IM |
1776 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1777 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1778 | } |
1779 | ||
2087a1ad GH |
1780 | static void update_avg(u64 *avg, u64 sample) |
1781 | { | |
1782 | s64 diff = sample - *avg; | |
1783 | *avg += diff >> 3; | |
1784 | } | |
1785 | ||
8159f87e | 1786 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1787 | { |
831451ac PZ |
1788 | if (wakeup) |
1789 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1790 | ||
dd41f596 | 1791 | sched_info_queued(p); |
fd390f6a | 1792 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1793 | p->se.on_rq = 1; |
71f8bd46 IM |
1794 | } |
1795 | ||
69be72c1 | 1796 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1797 | { |
831451ac PZ |
1798 | if (sleep) { |
1799 | if (p->se.last_wakeup) { | |
1800 | update_avg(&p->se.avg_overlap, | |
1801 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1802 | p->se.last_wakeup = 0; | |
1803 | } else { | |
1804 | update_avg(&p->se.avg_wakeup, | |
1805 | sysctl_sched_wakeup_granularity); | |
1806 | } | |
2087a1ad GH |
1807 | } |
1808 | ||
46ac22ba | 1809 | sched_info_dequeued(p); |
f02231e5 | 1810 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1811 | p->se.on_rq = 0; |
71f8bd46 IM |
1812 | } |
1813 | ||
14531189 | 1814 | /* |
dd41f596 | 1815 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1816 | */ |
14531189 IM |
1817 | static inline int __normal_prio(struct task_struct *p) |
1818 | { | |
dd41f596 | 1819 | return p->static_prio; |
14531189 IM |
1820 | } |
1821 | ||
b29739f9 IM |
1822 | /* |
1823 | * Calculate the expected normal priority: i.e. priority | |
1824 | * without taking RT-inheritance into account. Might be | |
1825 | * boosted by interactivity modifiers. Changes upon fork, | |
1826 | * setprio syscalls, and whenever the interactivity | |
1827 | * estimator recalculates. | |
1828 | */ | |
36c8b586 | 1829 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1830 | { |
1831 | int prio; | |
1832 | ||
e05606d3 | 1833 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1834 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1835 | else | |
1836 | prio = __normal_prio(p); | |
1837 | return prio; | |
1838 | } | |
1839 | ||
1840 | /* | |
1841 | * Calculate the current priority, i.e. the priority | |
1842 | * taken into account by the scheduler. This value might | |
1843 | * be boosted by RT tasks, or might be boosted by | |
1844 | * interactivity modifiers. Will be RT if the task got | |
1845 | * RT-boosted. If not then it returns p->normal_prio. | |
1846 | */ | |
36c8b586 | 1847 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1848 | { |
1849 | p->normal_prio = normal_prio(p); | |
1850 | /* | |
1851 | * If we are RT tasks or we were boosted to RT priority, | |
1852 | * keep the priority unchanged. Otherwise, update priority | |
1853 | * to the normal priority: | |
1854 | */ | |
1855 | if (!rt_prio(p->prio)) | |
1856 | return p->normal_prio; | |
1857 | return p->prio; | |
1858 | } | |
1859 | ||
1da177e4 | 1860 | /* |
dd41f596 | 1861 | * activate_task - move a task to the runqueue. |
1da177e4 | 1862 | */ |
dd41f596 | 1863 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1864 | { |
d9514f6c | 1865 | if (task_contributes_to_load(p)) |
dd41f596 | 1866 | rq->nr_uninterruptible--; |
1da177e4 | 1867 | |
8159f87e | 1868 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1869 | inc_nr_running(rq); |
1da177e4 LT |
1870 | } |
1871 | ||
1da177e4 LT |
1872 | /* |
1873 | * deactivate_task - remove a task from the runqueue. | |
1874 | */ | |
2e1cb74a | 1875 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1876 | { |
d9514f6c | 1877 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1878 | rq->nr_uninterruptible++; |
1879 | ||
69be72c1 | 1880 | dequeue_task(rq, p, sleep); |
c09595f6 | 1881 | dec_nr_running(rq); |
1da177e4 LT |
1882 | } |
1883 | ||
1da177e4 LT |
1884 | /** |
1885 | * task_curr - is this task currently executing on a CPU? | |
1886 | * @p: the task in question. | |
1887 | */ | |
36c8b586 | 1888 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1889 | { |
1890 | return cpu_curr(task_cpu(p)) == p; | |
1891 | } | |
1892 | ||
dd41f596 IM |
1893 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1894 | { | |
6f505b16 | 1895 | set_task_rq(p, cpu); |
dd41f596 | 1896 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1897 | /* |
1898 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1899 | * successfuly executed on another CPU. We must ensure that updates of | |
1900 | * per-task data have been completed by this moment. | |
1901 | */ | |
1902 | smp_wmb(); | |
dd41f596 | 1903 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1904 | #endif |
2dd73a4f PW |
1905 | } |
1906 | ||
cb469845 SR |
1907 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1908 | const struct sched_class *prev_class, | |
1909 | int oldprio, int running) | |
1910 | { | |
1911 | if (prev_class != p->sched_class) { | |
1912 | if (prev_class->switched_from) | |
1913 | prev_class->switched_from(rq, p, running); | |
1914 | p->sched_class->switched_to(rq, p, running); | |
1915 | } else | |
1916 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1917 | } | |
1918 | ||
1da177e4 | 1919 | #ifdef CONFIG_SMP |
c65cc870 | 1920 | |
e958b360 TG |
1921 | /* Used instead of source_load when we know the type == 0 */ |
1922 | static unsigned long weighted_cpuload(const int cpu) | |
1923 | { | |
1924 | return cpu_rq(cpu)->load.weight; | |
1925 | } | |
1926 | ||
cc367732 IM |
1927 | /* |
1928 | * Is this task likely cache-hot: | |
1929 | */ | |
e7693a36 | 1930 | static int |
cc367732 IM |
1931 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1932 | { | |
1933 | s64 delta; | |
1934 | ||
f540a608 IM |
1935 | /* |
1936 | * Buddy candidates are cache hot: | |
1937 | */ | |
4793241b PZ |
1938 | if (sched_feat(CACHE_HOT_BUDDY) && |
1939 | (&p->se == cfs_rq_of(&p->se)->next || | |
1940 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1941 | return 1; |
1942 | ||
cc367732 IM |
1943 | if (p->sched_class != &fair_sched_class) |
1944 | return 0; | |
1945 | ||
6bc1665b IM |
1946 | if (sysctl_sched_migration_cost == -1) |
1947 | return 1; | |
1948 | if (sysctl_sched_migration_cost == 0) | |
1949 | return 0; | |
1950 | ||
cc367732 IM |
1951 | delta = now - p->se.exec_start; |
1952 | ||
1953 | return delta < (s64)sysctl_sched_migration_cost; | |
1954 | } | |
1955 | ||
1956 | ||
dd41f596 | 1957 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1958 | { |
dd41f596 IM |
1959 | int old_cpu = task_cpu(p); |
1960 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1961 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1962 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1963 | u64 clock_offset; |
dd41f596 IM |
1964 | |
1965 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1966 | |
cbc34ed1 PZ |
1967 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1968 | ||
6cfb0d5d IM |
1969 | #ifdef CONFIG_SCHEDSTATS |
1970 | if (p->se.wait_start) | |
1971 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1972 | if (p->se.sleep_start) |
1973 | p->se.sleep_start -= clock_offset; | |
1974 | if (p->se.block_start) | |
1975 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1976 | if (old_cpu != new_cpu) { |
1977 | schedstat_inc(p, se.nr_migrations); | |
1978 | if (task_hot(p, old_rq->clock, NULL)) | |
1979 | schedstat_inc(p, se.nr_forced2_migrations); | |
1980 | } | |
6cfb0d5d | 1981 | #endif |
2830cf8c SV |
1982 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1983 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1984 | |
1985 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1986 | } |
1987 | ||
70b97a7f | 1988 | struct migration_req { |
1da177e4 | 1989 | struct list_head list; |
1da177e4 | 1990 | |
36c8b586 | 1991 | struct task_struct *task; |
1da177e4 LT |
1992 | int dest_cpu; |
1993 | ||
1da177e4 | 1994 | struct completion done; |
70b97a7f | 1995 | }; |
1da177e4 LT |
1996 | |
1997 | /* | |
1998 | * The task's runqueue lock must be held. | |
1999 | * Returns true if you have to wait for migration thread. | |
2000 | */ | |
36c8b586 | 2001 | static int |
70b97a7f | 2002 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2003 | { |
70b97a7f | 2004 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2005 | |
2006 | /* | |
2007 | * If the task is not on a runqueue (and not running), then | |
2008 | * it is sufficient to simply update the task's cpu field. | |
2009 | */ | |
dd41f596 | 2010 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2011 | set_task_cpu(p, dest_cpu); |
2012 | return 0; | |
2013 | } | |
2014 | ||
2015 | init_completion(&req->done); | |
1da177e4 LT |
2016 | req->task = p; |
2017 | req->dest_cpu = dest_cpu; | |
2018 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2019 | |
1da177e4 LT |
2020 | return 1; |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * wait_task_inactive - wait for a thread to unschedule. | |
2025 | * | |
85ba2d86 RM |
2026 | * If @match_state is nonzero, it's the @p->state value just checked and |
2027 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2028 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2029 | * we return a positive number (its total switch count). If a second call | |
2030 | * a short while later returns the same number, the caller can be sure that | |
2031 | * @p has remained unscheduled the whole time. | |
2032 | * | |
1da177e4 LT |
2033 | * The caller must ensure that the task *will* unschedule sometime soon, |
2034 | * else this function might spin for a *long* time. This function can't | |
2035 | * be called with interrupts off, or it may introduce deadlock with | |
2036 | * smp_call_function() if an IPI is sent by the same process we are | |
2037 | * waiting to become inactive. | |
2038 | */ | |
85ba2d86 | 2039 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2040 | { |
2041 | unsigned long flags; | |
dd41f596 | 2042 | int running, on_rq; |
85ba2d86 | 2043 | unsigned long ncsw; |
70b97a7f | 2044 | struct rq *rq; |
1da177e4 | 2045 | |
3a5c359a AK |
2046 | for (;;) { |
2047 | /* | |
2048 | * We do the initial early heuristics without holding | |
2049 | * any task-queue locks at all. We'll only try to get | |
2050 | * the runqueue lock when things look like they will | |
2051 | * work out! | |
2052 | */ | |
2053 | rq = task_rq(p); | |
fa490cfd | 2054 | |
3a5c359a AK |
2055 | /* |
2056 | * If the task is actively running on another CPU | |
2057 | * still, just relax and busy-wait without holding | |
2058 | * any locks. | |
2059 | * | |
2060 | * NOTE! Since we don't hold any locks, it's not | |
2061 | * even sure that "rq" stays as the right runqueue! | |
2062 | * But we don't care, since "task_running()" will | |
2063 | * return false if the runqueue has changed and p | |
2064 | * is actually now running somewhere else! | |
2065 | */ | |
85ba2d86 RM |
2066 | while (task_running(rq, p)) { |
2067 | if (match_state && unlikely(p->state != match_state)) | |
2068 | return 0; | |
3a5c359a | 2069 | cpu_relax(); |
85ba2d86 | 2070 | } |
fa490cfd | 2071 | |
3a5c359a AK |
2072 | /* |
2073 | * Ok, time to look more closely! We need the rq | |
2074 | * lock now, to be *sure*. If we're wrong, we'll | |
2075 | * just go back and repeat. | |
2076 | */ | |
2077 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2078 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2079 | running = task_running(rq, p); |
2080 | on_rq = p->se.on_rq; | |
85ba2d86 | 2081 | ncsw = 0; |
f31e11d8 | 2082 | if (!match_state || p->state == match_state) |
93dcf55f | 2083 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2084 | task_rq_unlock(rq, &flags); |
fa490cfd | 2085 | |
85ba2d86 RM |
2086 | /* |
2087 | * If it changed from the expected state, bail out now. | |
2088 | */ | |
2089 | if (unlikely(!ncsw)) | |
2090 | break; | |
2091 | ||
3a5c359a AK |
2092 | /* |
2093 | * Was it really running after all now that we | |
2094 | * checked with the proper locks actually held? | |
2095 | * | |
2096 | * Oops. Go back and try again.. | |
2097 | */ | |
2098 | if (unlikely(running)) { | |
2099 | cpu_relax(); | |
2100 | continue; | |
2101 | } | |
fa490cfd | 2102 | |
3a5c359a AK |
2103 | /* |
2104 | * It's not enough that it's not actively running, | |
2105 | * it must be off the runqueue _entirely_, and not | |
2106 | * preempted! | |
2107 | * | |
80dd99b3 | 2108 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2109 | * running right now), it's preempted, and we should |
2110 | * yield - it could be a while. | |
2111 | */ | |
2112 | if (unlikely(on_rq)) { | |
2113 | schedule_timeout_uninterruptible(1); | |
2114 | continue; | |
2115 | } | |
fa490cfd | 2116 | |
3a5c359a AK |
2117 | /* |
2118 | * Ahh, all good. It wasn't running, and it wasn't | |
2119 | * runnable, which means that it will never become | |
2120 | * running in the future either. We're all done! | |
2121 | */ | |
2122 | break; | |
2123 | } | |
85ba2d86 RM |
2124 | |
2125 | return ncsw; | |
1da177e4 LT |
2126 | } |
2127 | ||
2128 | /*** | |
2129 | * kick_process - kick a running thread to enter/exit the kernel | |
2130 | * @p: the to-be-kicked thread | |
2131 | * | |
2132 | * Cause a process which is running on another CPU to enter | |
2133 | * kernel-mode, without any delay. (to get signals handled.) | |
2134 | * | |
2135 | * NOTE: this function doesnt have to take the runqueue lock, | |
2136 | * because all it wants to ensure is that the remote task enters | |
2137 | * the kernel. If the IPI races and the task has been migrated | |
2138 | * to another CPU then no harm is done and the purpose has been | |
2139 | * achieved as well. | |
2140 | */ | |
36c8b586 | 2141 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2142 | { |
2143 | int cpu; | |
2144 | ||
2145 | preempt_disable(); | |
2146 | cpu = task_cpu(p); | |
2147 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2148 | smp_send_reschedule(cpu); | |
2149 | preempt_enable(); | |
2150 | } | |
2151 | ||
2152 | /* | |
2dd73a4f PW |
2153 | * Return a low guess at the load of a migration-source cpu weighted |
2154 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2155 | * |
2156 | * We want to under-estimate the load of migration sources, to | |
2157 | * balance conservatively. | |
2158 | */ | |
a9957449 | 2159 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2160 | { |
70b97a7f | 2161 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2162 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2163 | |
93b75217 | 2164 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2165 | return total; |
b910472d | 2166 | |
dd41f596 | 2167 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | /* | |
2dd73a4f PW |
2171 | * Return a high guess at the load of a migration-target cpu weighted |
2172 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2173 | */ |
a9957449 | 2174 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2175 | { |
70b97a7f | 2176 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2177 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2178 | |
93b75217 | 2179 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2180 | return total; |
3b0bd9bc | 2181 | |
dd41f596 | 2182 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2183 | } |
2184 | ||
147cbb4b NP |
2185 | /* |
2186 | * find_idlest_group finds and returns the least busy CPU group within the | |
2187 | * domain. | |
2188 | */ | |
2189 | static struct sched_group * | |
2190 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2191 | { | |
2192 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2193 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2194 | int load_idx = sd->forkexec_idx; | |
2195 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2196 | ||
2197 | do { | |
2198 | unsigned long load, avg_load; | |
2199 | int local_group; | |
2200 | int i; | |
2201 | ||
da5a5522 | 2202 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2203 | if (!cpumask_intersects(sched_group_cpus(group), |
2204 | &p->cpus_allowed)) | |
3a5c359a | 2205 | continue; |
da5a5522 | 2206 | |
758b2cdc RR |
2207 | local_group = cpumask_test_cpu(this_cpu, |
2208 | sched_group_cpus(group)); | |
147cbb4b NP |
2209 | |
2210 | /* Tally up the load of all CPUs in the group */ | |
2211 | avg_load = 0; | |
2212 | ||
758b2cdc | 2213 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2214 | /* Bias balancing toward cpus of our domain */ |
2215 | if (local_group) | |
2216 | load = source_load(i, load_idx); | |
2217 | else | |
2218 | load = target_load(i, load_idx); | |
2219 | ||
2220 | avg_load += load; | |
2221 | } | |
2222 | ||
2223 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2224 | avg_load = sg_div_cpu_power(group, |
2225 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2226 | |
2227 | if (local_group) { | |
2228 | this_load = avg_load; | |
2229 | this = group; | |
2230 | } else if (avg_load < min_load) { | |
2231 | min_load = avg_load; | |
2232 | idlest = group; | |
2233 | } | |
3a5c359a | 2234 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2235 | |
2236 | if (!idlest || 100*this_load < imbalance*min_load) | |
2237 | return NULL; | |
2238 | return idlest; | |
2239 | } | |
2240 | ||
2241 | /* | |
0feaece9 | 2242 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2243 | */ |
95cdf3b7 | 2244 | static int |
758b2cdc | 2245 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2246 | { |
2247 | unsigned long load, min_load = ULONG_MAX; | |
2248 | int idlest = -1; | |
2249 | int i; | |
2250 | ||
da5a5522 | 2251 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2252 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2253 | load = weighted_cpuload(i); |
147cbb4b NP |
2254 | |
2255 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2256 | min_load = load; | |
2257 | idlest = i; | |
2258 | } | |
2259 | } | |
2260 | ||
2261 | return idlest; | |
2262 | } | |
2263 | ||
476d139c NP |
2264 | /* |
2265 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2266 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2267 | * SD_BALANCE_EXEC. | |
2268 | * | |
2269 | * Balance, ie. select the least loaded group. | |
2270 | * | |
2271 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2272 | * | |
2273 | * preempt must be disabled. | |
2274 | */ | |
2275 | static int sched_balance_self(int cpu, int flag) | |
2276 | { | |
2277 | struct task_struct *t = current; | |
2278 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2279 | |
c96d145e | 2280 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2281 | /* |
2282 | * If power savings logic is enabled for a domain, stop there. | |
2283 | */ | |
5c45bf27 SS |
2284 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2285 | break; | |
476d139c NP |
2286 | if (tmp->flags & flag) |
2287 | sd = tmp; | |
c96d145e | 2288 | } |
476d139c | 2289 | |
039a1c41 PZ |
2290 | if (sd) |
2291 | update_shares(sd); | |
2292 | ||
476d139c | 2293 | while (sd) { |
476d139c | 2294 | struct sched_group *group; |
1a848870 SS |
2295 | int new_cpu, weight; |
2296 | ||
2297 | if (!(sd->flags & flag)) { | |
2298 | sd = sd->child; | |
2299 | continue; | |
2300 | } | |
476d139c | 2301 | |
476d139c | 2302 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2303 | if (!group) { |
2304 | sd = sd->child; | |
2305 | continue; | |
2306 | } | |
476d139c | 2307 | |
758b2cdc | 2308 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2309 | if (new_cpu == -1 || new_cpu == cpu) { |
2310 | /* Now try balancing at a lower domain level of cpu */ | |
2311 | sd = sd->child; | |
2312 | continue; | |
2313 | } | |
476d139c | 2314 | |
1a848870 | 2315 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2316 | cpu = new_cpu; |
758b2cdc | 2317 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2318 | sd = NULL; |
476d139c | 2319 | for_each_domain(cpu, tmp) { |
758b2cdc | 2320 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2321 | break; |
2322 | if (tmp->flags & flag) | |
2323 | sd = tmp; | |
2324 | } | |
2325 | /* while loop will break here if sd == NULL */ | |
2326 | } | |
2327 | ||
2328 | return cpu; | |
2329 | } | |
2330 | ||
2331 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2332 | |
1da177e4 LT |
2333 | /*** |
2334 | * try_to_wake_up - wake up a thread | |
2335 | * @p: the to-be-woken-up thread | |
2336 | * @state: the mask of task states that can be woken | |
2337 | * @sync: do a synchronous wakeup? | |
2338 | * | |
2339 | * Put it on the run-queue if it's not already there. The "current" | |
2340 | * thread is always on the run-queue (except when the actual | |
2341 | * re-schedule is in progress), and as such you're allowed to do | |
2342 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2343 | * runnable without the overhead of this. | |
2344 | * | |
2345 | * returns failure only if the task is already active. | |
2346 | */ | |
36c8b586 | 2347 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2348 | { |
cc367732 | 2349 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2350 | unsigned long flags; |
2351 | long old_state; | |
70b97a7f | 2352 | struct rq *rq; |
1da177e4 | 2353 | |
b85d0667 IM |
2354 | if (!sched_feat(SYNC_WAKEUPS)) |
2355 | sync = 0; | |
2356 | ||
2398f2c6 | 2357 | #ifdef CONFIG_SMP |
57310a98 | 2358 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2359 | struct sched_domain *sd; |
2360 | ||
2361 | this_cpu = raw_smp_processor_id(); | |
2362 | cpu = task_cpu(p); | |
2363 | ||
2364 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2365 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2366 | update_shares(sd); |
2367 | break; | |
2368 | } | |
2369 | } | |
2370 | } | |
2371 | #endif | |
2372 | ||
04e2f174 | 2373 | smp_wmb(); |
1da177e4 | 2374 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2375 | update_rq_clock(rq); |
1da177e4 LT |
2376 | old_state = p->state; |
2377 | if (!(old_state & state)) | |
2378 | goto out; | |
2379 | ||
dd41f596 | 2380 | if (p->se.on_rq) |
1da177e4 LT |
2381 | goto out_running; |
2382 | ||
2383 | cpu = task_cpu(p); | |
cc367732 | 2384 | orig_cpu = cpu; |
1da177e4 LT |
2385 | this_cpu = smp_processor_id(); |
2386 | ||
2387 | #ifdef CONFIG_SMP | |
2388 | if (unlikely(task_running(rq, p))) | |
2389 | goto out_activate; | |
2390 | ||
5d2f5a61 DA |
2391 | cpu = p->sched_class->select_task_rq(p, sync); |
2392 | if (cpu != orig_cpu) { | |
2393 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2394 | task_rq_unlock(rq, &flags); |
2395 | /* might preempt at this point */ | |
2396 | rq = task_rq_lock(p, &flags); | |
2397 | old_state = p->state; | |
2398 | if (!(old_state & state)) | |
2399 | goto out; | |
dd41f596 | 2400 | if (p->se.on_rq) |
1da177e4 LT |
2401 | goto out_running; |
2402 | ||
2403 | this_cpu = smp_processor_id(); | |
2404 | cpu = task_cpu(p); | |
2405 | } | |
2406 | ||
e7693a36 GH |
2407 | #ifdef CONFIG_SCHEDSTATS |
2408 | schedstat_inc(rq, ttwu_count); | |
2409 | if (cpu == this_cpu) | |
2410 | schedstat_inc(rq, ttwu_local); | |
2411 | else { | |
2412 | struct sched_domain *sd; | |
2413 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2414 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2415 | schedstat_inc(sd, ttwu_wake_remote); |
2416 | break; | |
2417 | } | |
2418 | } | |
2419 | } | |
6d6bc0ad | 2420 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2421 | |
1da177e4 LT |
2422 | out_activate: |
2423 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2424 | schedstat_inc(p, se.nr_wakeups); |
2425 | if (sync) | |
2426 | schedstat_inc(p, se.nr_wakeups_sync); | |
2427 | if (orig_cpu != cpu) | |
2428 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2429 | if (cpu == this_cpu) | |
2430 | schedstat_inc(p, se.nr_wakeups_local); | |
2431 | else | |
2432 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2433 | activate_task(rq, p, 1); |
1da177e4 LT |
2434 | success = 1; |
2435 | ||
831451ac PZ |
2436 | /* |
2437 | * Only attribute actual wakeups done by this task. | |
2438 | */ | |
2439 | if (!in_interrupt()) { | |
2440 | struct sched_entity *se = ¤t->se; | |
2441 | u64 sample = se->sum_exec_runtime; | |
2442 | ||
2443 | if (se->last_wakeup) | |
2444 | sample -= se->last_wakeup; | |
2445 | else | |
2446 | sample -= se->start_runtime; | |
2447 | update_avg(&se->avg_wakeup, sample); | |
2448 | ||
2449 | se->last_wakeup = se->sum_exec_runtime; | |
2450 | } | |
2451 | ||
1da177e4 | 2452 | out_running: |
468a15bb | 2453 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2454 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2455 | |
1da177e4 | 2456 | p->state = TASK_RUNNING; |
9a897c5a SR |
2457 | #ifdef CONFIG_SMP |
2458 | if (p->sched_class->task_wake_up) | |
2459 | p->sched_class->task_wake_up(rq, p); | |
2460 | #endif | |
1da177e4 LT |
2461 | out: |
2462 | task_rq_unlock(rq, &flags); | |
2463 | ||
2464 | return success; | |
2465 | } | |
2466 | ||
50fa610a DH |
2467 | /** |
2468 | * wake_up_process - Wake up a specific process | |
2469 | * @p: The process to be woken up. | |
2470 | * | |
2471 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2472 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2473 | * running. | |
2474 | * | |
2475 | * It may be assumed that this function implies a write memory barrier before | |
2476 | * changing the task state if and only if any tasks are woken up. | |
2477 | */ | |
7ad5b3a5 | 2478 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2479 | { |
d9514f6c | 2480 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2481 | } |
1da177e4 LT |
2482 | EXPORT_SYMBOL(wake_up_process); |
2483 | ||
7ad5b3a5 | 2484 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2485 | { |
2486 | return try_to_wake_up(p, state, 0); | |
2487 | } | |
2488 | ||
1da177e4 LT |
2489 | /* |
2490 | * Perform scheduler related setup for a newly forked process p. | |
2491 | * p is forked by current. | |
dd41f596 IM |
2492 | * |
2493 | * __sched_fork() is basic setup used by init_idle() too: | |
2494 | */ | |
2495 | static void __sched_fork(struct task_struct *p) | |
2496 | { | |
dd41f596 IM |
2497 | p->se.exec_start = 0; |
2498 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2499 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2500 | p->se.last_wakeup = 0; |
2501 | p->se.avg_overlap = 0; | |
831451ac PZ |
2502 | p->se.start_runtime = 0; |
2503 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2504 | |
2505 | #ifdef CONFIG_SCHEDSTATS | |
2506 | p->se.wait_start = 0; | |
dd41f596 IM |
2507 | p->se.sum_sleep_runtime = 0; |
2508 | p->se.sleep_start = 0; | |
dd41f596 IM |
2509 | p->se.block_start = 0; |
2510 | p->se.sleep_max = 0; | |
2511 | p->se.block_max = 0; | |
2512 | p->se.exec_max = 0; | |
eba1ed4b | 2513 | p->se.slice_max = 0; |
dd41f596 | 2514 | p->se.wait_max = 0; |
6cfb0d5d | 2515 | #endif |
476d139c | 2516 | |
fa717060 | 2517 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2518 | p->se.on_rq = 0; |
4a55bd5e | 2519 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2520 | |
e107be36 AK |
2521 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2522 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2523 | #endif | |
2524 | ||
1da177e4 LT |
2525 | /* |
2526 | * We mark the process as running here, but have not actually | |
2527 | * inserted it onto the runqueue yet. This guarantees that | |
2528 | * nobody will actually run it, and a signal or other external | |
2529 | * event cannot wake it up and insert it on the runqueue either. | |
2530 | */ | |
2531 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2532 | } |
2533 | ||
2534 | /* | |
2535 | * fork()/clone()-time setup: | |
2536 | */ | |
2537 | void sched_fork(struct task_struct *p, int clone_flags) | |
2538 | { | |
2539 | int cpu = get_cpu(); | |
2540 | ||
2541 | __sched_fork(p); | |
2542 | ||
2543 | #ifdef CONFIG_SMP | |
2544 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2545 | #endif | |
02e4bac2 | 2546 | set_task_cpu(p, cpu); |
b29739f9 IM |
2547 | |
2548 | /* | |
2549 | * Make sure we do not leak PI boosting priority to the child: | |
2550 | */ | |
2551 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2552 | if (!rt_prio(p->prio)) |
2553 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2554 | |
52f17b6c | 2555 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2556 | if (likely(sched_info_on())) |
52f17b6c | 2557 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2558 | #endif |
d6077cb8 | 2559 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2560 | p->oncpu = 0; |
2561 | #endif | |
1da177e4 | 2562 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2563 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2564 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2565 | #endif |
917b627d GH |
2566 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2567 | ||
476d139c | 2568 | put_cpu(); |
1da177e4 LT |
2569 | } |
2570 | ||
2571 | /* | |
2572 | * wake_up_new_task - wake up a newly created task for the first time. | |
2573 | * | |
2574 | * This function will do some initial scheduler statistics housekeeping | |
2575 | * that must be done for every newly created context, then puts the task | |
2576 | * on the runqueue and wakes it. | |
2577 | */ | |
7ad5b3a5 | 2578 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2579 | { |
2580 | unsigned long flags; | |
dd41f596 | 2581 | struct rq *rq; |
1da177e4 LT |
2582 | |
2583 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2584 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2585 | update_rq_clock(rq); |
1da177e4 LT |
2586 | |
2587 | p->prio = effective_prio(p); | |
2588 | ||
b9dca1e0 | 2589 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2590 | activate_task(rq, p, 0); |
1da177e4 | 2591 | } else { |
1da177e4 | 2592 | /* |
dd41f596 IM |
2593 | * Let the scheduling class do new task startup |
2594 | * management (if any): | |
1da177e4 | 2595 | */ |
ee0827d8 | 2596 | p->sched_class->task_new(rq, p); |
c09595f6 | 2597 | inc_nr_running(rq); |
1da177e4 | 2598 | } |
c71dd42d | 2599 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2600 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2601 | #ifdef CONFIG_SMP |
2602 | if (p->sched_class->task_wake_up) | |
2603 | p->sched_class->task_wake_up(rq, p); | |
2604 | #endif | |
dd41f596 | 2605 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2606 | } |
2607 | ||
e107be36 AK |
2608 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2609 | ||
2610 | /** | |
80dd99b3 | 2611 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2612 | * @notifier: notifier struct to register |
e107be36 AK |
2613 | */ |
2614 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2615 | { | |
2616 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2617 | } | |
2618 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2619 | ||
2620 | /** | |
2621 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2622 | * @notifier: notifier struct to unregister |
e107be36 AK |
2623 | * |
2624 | * This is safe to call from within a preemption notifier. | |
2625 | */ | |
2626 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2627 | { | |
2628 | hlist_del(¬ifier->link); | |
2629 | } | |
2630 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2631 | ||
2632 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2633 | { | |
2634 | struct preempt_notifier *notifier; | |
2635 | struct hlist_node *node; | |
2636 | ||
2637 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2638 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2639 | } | |
2640 | ||
2641 | static void | |
2642 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2643 | struct task_struct *next) | |
2644 | { | |
2645 | struct preempt_notifier *notifier; | |
2646 | struct hlist_node *node; | |
2647 | ||
2648 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2649 | notifier->ops->sched_out(notifier, next); | |
2650 | } | |
2651 | ||
6d6bc0ad | 2652 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2653 | |
2654 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2655 | { | |
2656 | } | |
2657 | ||
2658 | static void | |
2659 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2660 | struct task_struct *next) | |
2661 | { | |
2662 | } | |
2663 | ||
6d6bc0ad | 2664 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2665 | |
4866cde0 NP |
2666 | /** |
2667 | * prepare_task_switch - prepare to switch tasks | |
2668 | * @rq: the runqueue preparing to switch | |
421cee29 | 2669 | * @prev: the current task that is being switched out |
4866cde0 NP |
2670 | * @next: the task we are going to switch to. |
2671 | * | |
2672 | * This is called with the rq lock held and interrupts off. It must | |
2673 | * be paired with a subsequent finish_task_switch after the context | |
2674 | * switch. | |
2675 | * | |
2676 | * prepare_task_switch sets up locking and calls architecture specific | |
2677 | * hooks. | |
2678 | */ | |
e107be36 AK |
2679 | static inline void |
2680 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2681 | struct task_struct *next) | |
4866cde0 | 2682 | { |
e107be36 | 2683 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2684 | prepare_lock_switch(rq, next); |
2685 | prepare_arch_switch(next); | |
2686 | } | |
2687 | ||
1da177e4 LT |
2688 | /** |
2689 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2690 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2691 | * @prev: the thread we just switched away from. |
2692 | * | |
4866cde0 NP |
2693 | * finish_task_switch must be called after the context switch, paired |
2694 | * with a prepare_task_switch call before the context switch. | |
2695 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2696 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2697 | * |
2698 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2699 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2700 | * with the lock held can cause deadlocks; see schedule() for |
2701 | * details.) | |
2702 | */ | |
a9957449 | 2703 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2704 | __releases(rq->lock) |
2705 | { | |
1da177e4 | 2706 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2707 | long prev_state; |
967fc046 GH |
2708 | #ifdef CONFIG_SMP |
2709 | int post_schedule = 0; | |
2710 | ||
2711 | if (current->sched_class->needs_post_schedule) | |
2712 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2713 | #endif | |
1da177e4 LT |
2714 | |
2715 | rq->prev_mm = NULL; | |
2716 | ||
2717 | /* | |
2718 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2719 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2720 | * schedule one last time. The schedule call will never return, and |
2721 | * the scheduled task must drop that reference. | |
c394cc9f | 2722 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2723 | * still held, otherwise prev could be scheduled on another cpu, die |
2724 | * there before we look at prev->state, and then the reference would | |
2725 | * be dropped twice. | |
2726 | * Manfred Spraul <manfred@colorfullife.com> | |
2727 | */ | |
55a101f8 | 2728 | prev_state = prev->state; |
4866cde0 NP |
2729 | finish_arch_switch(prev); |
2730 | finish_lock_switch(rq, prev); | |
9a897c5a | 2731 | #ifdef CONFIG_SMP |
967fc046 | 2732 | if (post_schedule) |
9a897c5a SR |
2733 | current->sched_class->post_schedule(rq); |
2734 | #endif | |
e8fa1362 | 2735 | |
e107be36 | 2736 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2737 | if (mm) |
2738 | mmdrop(mm); | |
c394cc9f | 2739 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2740 | /* |
2741 | * Remove function-return probe instances associated with this | |
2742 | * task and put them back on the free list. | |
9761eea8 | 2743 | */ |
c6fd91f0 | 2744 | kprobe_flush_task(prev); |
1da177e4 | 2745 | put_task_struct(prev); |
c6fd91f0 | 2746 | } |
1da177e4 LT |
2747 | } |
2748 | ||
2749 | /** | |
2750 | * schedule_tail - first thing a freshly forked thread must call. | |
2751 | * @prev: the thread we just switched away from. | |
2752 | */ | |
36c8b586 | 2753 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2754 | __releases(rq->lock) |
2755 | { | |
70b97a7f IM |
2756 | struct rq *rq = this_rq(); |
2757 | ||
4866cde0 NP |
2758 | finish_task_switch(rq, prev); |
2759 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2760 | /* In this case, finish_task_switch does not reenable preemption */ | |
2761 | preempt_enable(); | |
2762 | #endif | |
1da177e4 | 2763 | if (current->set_child_tid) |
b488893a | 2764 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2765 | } |
2766 | ||
2767 | /* | |
2768 | * context_switch - switch to the new MM and the new | |
2769 | * thread's register state. | |
2770 | */ | |
dd41f596 | 2771 | static inline void |
70b97a7f | 2772 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2773 | struct task_struct *next) |
1da177e4 | 2774 | { |
dd41f596 | 2775 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2776 | |
e107be36 | 2777 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2778 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2779 | mm = next->mm; |
2780 | oldmm = prev->active_mm; | |
9226d125 ZA |
2781 | /* |
2782 | * For paravirt, this is coupled with an exit in switch_to to | |
2783 | * combine the page table reload and the switch backend into | |
2784 | * one hypercall. | |
2785 | */ | |
224101ed | 2786 | arch_start_context_switch(prev); |
9226d125 | 2787 | |
dd41f596 | 2788 | if (unlikely(!mm)) { |
1da177e4 LT |
2789 | next->active_mm = oldmm; |
2790 | atomic_inc(&oldmm->mm_count); | |
2791 | enter_lazy_tlb(oldmm, next); | |
2792 | } else | |
2793 | switch_mm(oldmm, mm, next); | |
2794 | ||
dd41f596 | 2795 | if (unlikely(!prev->mm)) { |
1da177e4 | 2796 | prev->active_mm = NULL; |
1da177e4 LT |
2797 | rq->prev_mm = oldmm; |
2798 | } | |
3a5f5e48 IM |
2799 | /* |
2800 | * Since the runqueue lock will be released by the next | |
2801 | * task (which is an invalid locking op but in the case | |
2802 | * of the scheduler it's an obvious special-case), so we | |
2803 | * do an early lockdep release here: | |
2804 | */ | |
2805 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2806 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2807 | #endif |
1da177e4 LT |
2808 | |
2809 | /* Here we just switch the register state and the stack. */ | |
2810 | switch_to(prev, next, prev); | |
2811 | ||
dd41f596 IM |
2812 | barrier(); |
2813 | /* | |
2814 | * this_rq must be evaluated again because prev may have moved | |
2815 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2816 | * frame will be invalid. | |
2817 | */ | |
2818 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2819 | } |
2820 | ||
2821 | /* | |
2822 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2823 | * | |
2824 | * externally visible scheduler statistics: current number of runnable | |
2825 | * threads, current number of uninterruptible-sleeping threads, total | |
2826 | * number of context switches performed since bootup. | |
2827 | */ | |
2828 | unsigned long nr_running(void) | |
2829 | { | |
2830 | unsigned long i, sum = 0; | |
2831 | ||
2832 | for_each_online_cpu(i) | |
2833 | sum += cpu_rq(i)->nr_running; | |
2834 | ||
2835 | return sum; | |
2836 | } | |
2837 | ||
2838 | unsigned long nr_uninterruptible(void) | |
2839 | { | |
2840 | unsigned long i, sum = 0; | |
2841 | ||
0a945022 | 2842 | for_each_possible_cpu(i) |
1da177e4 LT |
2843 | sum += cpu_rq(i)->nr_uninterruptible; |
2844 | ||
2845 | /* | |
2846 | * Since we read the counters lockless, it might be slightly | |
2847 | * inaccurate. Do not allow it to go below zero though: | |
2848 | */ | |
2849 | if (unlikely((long)sum < 0)) | |
2850 | sum = 0; | |
2851 | ||
2852 | return sum; | |
2853 | } | |
2854 | ||
2855 | unsigned long long nr_context_switches(void) | |
2856 | { | |
cc94abfc SR |
2857 | int i; |
2858 | unsigned long long sum = 0; | |
1da177e4 | 2859 | |
0a945022 | 2860 | for_each_possible_cpu(i) |
1da177e4 LT |
2861 | sum += cpu_rq(i)->nr_switches; |
2862 | ||
2863 | return sum; | |
2864 | } | |
2865 | ||
2866 | unsigned long nr_iowait(void) | |
2867 | { | |
2868 | unsigned long i, sum = 0; | |
2869 | ||
0a945022 | 2870 | for_each_possible_cpu(i) |
1da177e4 LT |
2871 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2872 | ||
2873 | return sum; | |
2874 | } | |
2875 | ||
dce48a84 TG |
2876 | /* Variables and functions for calc_load */ |
2877 | static atomic_long_t calc_load_tasks; | |
2878 | static unsigned long calc_load_update; | |
2879 | unsigned long avenrun[3]; | |
2880 | EXPORT_SYMBOL(avenrun); | |
2881 | ||
2d02494f TG |
2882 | /** |
2883 | * get_avenrun - get the load average array | |
2884 | * @loads: pointer to dest load array | |
2885 | * @offset: offset to add | |
2886 | * @shift: shift count to shift the result left | |
2887 | * | |
2888 | * These values are estimates at best, so no need for locking. | |
2889 | */ | |
2890 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2891 | { | |
2892 | loads[0] = (avenrun[0] + offset) << shift; | |
2893 | loads[1] = (avenrun[1] + offset) << shift; | |
2894 | loads[2] = (avenrun[2] + offset) << shift; | |
2895 | } | |
2896 | ||
dce48a84 TG |
2897 | static unsigned long |
2898 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2899 | { |
dce48a84 TG |
2900 | load *= exp; |
2901 | load += active * (FIXED_1 - exp); | |
2902 | return load >> FSHIFT; | |
2903 | } | |
db1b1fef | 2904 | |
dce48a84 TG |
2905 | /* |
2906 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2907 | * CPUs have updated calc_load_tasks. | |
2908 | */ | |
2909 | void calc_global_load(void) | |
2910 | { | |
2911 | unsigned long upd = calc_load_update + 10; | |
2912 | long active; | |
2913 | ||
2914 | if (time_before(jiffies, upd)) | |
2915 | return; | |
db1b1fef | 2916 | |
dce48a84 TG |
2917 | active = atomic_long_read(&calc_load_tasks); |
2918 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2919 | |
dce48a84 TG |
2920 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2921 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2922 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2923 | ||
2924 | calc_load_update += LOAD_FREQ; | |
2925 | } | |
2926 | ||
2927 | /* | |
2928 | * Either called from update_cpu_load() or from a cpu going idle | |
2929 | */ | |
2930 | static void calc_load_account_active(struct rq *this_rq) | |
2931 | { | |
2932 | long nr_active, delta; | |
2933 | ||
2934 | nr_active = this_rq->nr_running; | |
2935 | nr_active += (long) this_rq->nr_uninterruptible; | |
2936 | ||
2937 | if (nr_active != this_rq->calc_load_active) { | |
2938 | delta = nr_active - this_rq->calc_load_active; | |
2939 | this_rq->calc_load_active = nr_active; | |
2940 | atomic_long_add(delta, &calc_load_tasks); | |
2941 | } | |
db1b1fef JS |
2942 | } |
2943 | ||
48f24c4d | 2944 | /* |
dd41f596 IM |
2945 | * Update rq->cpu_load[] statistics. This function is usually called every |
2946 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2947 | */ |
dd41f596 | 2948 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2949 | { |
495eca49 | 2950 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2951 | int i, scale; |
2952 | ||
2953 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2954 | |
2955 | /* Update our load: */ | |
2956 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2957 | unsigned long old_load, new_load; | |
2958 | ||
2959 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2960 | ||
2961 | old_load = this_rq->cpu_load[i]; | |
2962 | new_load = this_load; | |
a25707f3 IM |
2963 | /* |
2964 | * Round up the averaging division if load is increasing. This | |
2965 | * prevents us from getting stuck on 9 if the load is 10, for | |
2966 | * example. | |
2967 | */ | |
2968 | if (new_load > old_load) | |
2969 | new_load += scale-1; | |
dd41f596 IM |
2970 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2971 | } | |
dce48a84 TG |
2972 | |
2973 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
2974 | this_rq->calc_load_update += LOAD_FREQ; | |
2975 | calc_load_account_active(this_rq); | |
2976 | } | |
48f24c4d IM |
2977 | } |
2978 | ||
dd41f596 IM |
2979 | #ifdef CONFIG_SMP |
2980 | ||
1da177e4 LT |
2981 | /* |
2982 | * double_rq_lock - safely lock two runqueues | |
2983 | * | |
2984 | * Note this does not disable interrupts like task_rq_lock, | |
2985 | * you need to do so manually before calling. | |
2986 | */ | |
70b97a7f | 2987 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2988 | __acquires(rq1->lock) |
2989 | __acquires(rq2->lock) | |
2990 | { | |
054b9108 | 2991 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2992 | if (rq1 == rq2) { |
2993 | spin_lock(&rq1->lock); | |
2994 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2995 | } else { | |
c96d145e | 2996 | if (rq1 < rq2) { |
1da177e4 | 2997 | spin_lock(&rq1->lock); |
5e710e37 | 2998 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2999 | } else { |
3000 | spin_lock(&rq2->lock); | |
5e710e37 | 3001 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3002 | } |
3003 | } | |
6e82a3be IM |
3004 | update_rq_clock(rq1); |
3005 | update_rq_clock(rq2); | |
1da177e4 LT |
3006 | } |
3007 | ||
3008 | /* | |
3009 | * double_rq_unlock - safely unlock two runqueues | |
3010 | * | |
3011 | * Note this does not restore interrupts like task_rq_unlock, | |
3012 | * you need to do so manually after calling. | |
3013 | */ | |
70b97a7f | 3014 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3015 | __releases(rq1->lock) |
3016 | __releases(rq2->lock) | |
3017 | { | |
3018 | spin_unlock(&rq1->lock); | |
3019 | if (rq1 != rq2) | |
3020 | spin_unlock(&rq2->lock); | |
3021 | else | |
3022 | __release(rq2->lock); | |
3023 | } | |
3024 | ||
1da177e4 LT |
3025 | /* |
3026 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3027 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3028 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3029 | * the cpu_allowed mask is restored. |
3030 | */ | |
36c8b586 | 3031 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3032 | { |
70b97a7f | 3033 | struct migration_req req; |
1da177e4 | 3034 | unsigned long flags; |
70b97a7f | 3035 | struct rq *rq; |
1da177e4 LT |
3036 | |
3037 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3038 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3039 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3040 | goto out; |
3041 | ||
3042 | /* force the process onto the specified CPU */ | |
3043 | if (migrate_task(p, dest_cpu, &req)) { | |
3044 | /* Need to wait for migration thread (might exit: take ref). */ | |
3045 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3046 | |
1da177e4 LT |
3047 | get_task_struct(mt); |
3048 | task_rq_unlock(rq, &flags); | |
3049 | wake_up_process(mt); | |
3050 | put_task_struct(mt); | |
3051 | wait_for_completion(&req.done); | |
36c8b586 | 3052 | |
1da177e4 LT |
3053 | return; |
3054 | } | |
3055 | out: | |
3056 | task_rq_unlock(rq, &flags); | |
3057 | } | |
3058 | ||
3059 | /* | |
476d139c NP |
3060 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3061 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3062 | */ |
3063 | void sched_exec(void) | |
3064 | { | |
1da177e4 | 3065 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3066 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3067 | put_cpu(); |
476d139c NP |
3068 | if (new_cpu != this_cpu) |
3069 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3070 | } |
3071 | ||
3072 | /* | |
3073 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3074 | * Both runqueues must be locked. | |
3075 | */ | |
dd41f596 IM |
3076 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3077 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3078 | { |
2e1cb74a | 3079 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3080 | set_task_cpu(p, this_cpu); |
dd41f596 | 3081 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3082 | /* |
3083 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3084 | * to be always true for them. | |
3085 | */ | |
15afe09b | 3086 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3087 | } |
3088 | ||
3089 | /* | |
3090 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3091 | */ | |
858119e1 | 3092 | static |
70b97a7f | 3093 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3094 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3095 | int *all_pinned) |
1da177e4 | 3096 | { |
708dc512 | 3097 | int tsk_cache_hot = 0; |
1da177e4 LT |
3098 | /* |
3099 | * We do not migrate tasks that are: | |
3100 | * 1) running (obviously), or | |
3101 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3102 | * 3) are cache-hot on their current CPU. | |
3103 | */ | |
96f874e2 | 3104 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3105 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3106 | return 0; |
cc367732 | 3107 | } |
81026794 NP |
3108 | *all_pinned = 0; |
3109 | ||
cc367732 IM |
3110 | if (task_running(rq, p)) { |
3111 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3112 | return 0; |
cc367732 | 3113 | } |
1da177e4 | 3114 | |
da84d961 IM |
3115 | /* |
3116 | * Aggressive migration if: | |
3117 | * 1) task is cache cold, or | |
3118 | * 2) too many balance attempts have failed. | |
3119 | */ | |
3120 | ||
708dc512 LH |
3121 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3122 | if (!tsk_cache_hot || | |
3123 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3124 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3125 | if (tsk_cache_hot) { |
da84d961 | 3126 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3127 | schedstat_inc(p, se.nr_forced_migrations); |
3128 | } | |
da84d961 IM |
3129 | #endif |
3130 | return 1; | |
3131 | } | |
3132 | ||
708dc512 | 3133 | if (tsk_cache_hot) { |
cc367732 | 3134 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3135 | return 0; |
cc367732 | 3136 | } |
1da177e4 LT |
3137 | return 1; |
3138 | } | |
3139 | ||
e1d1484f PW |
3140 | static unsigned long |
3141 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3142 | unsigned long max_load_move, struct sched_domain *sd, | |
3143 | enum cpu_idle_type idle, int *all_pinned, | |
3144 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3145 | { |
051c6764 | 3146 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3147 | struct task_struct *p; |
3148 | long rem_load_move = max_load_move; | |
1da177e4 | 3149 | |
e1d1484f | 3150 | if (max_load_move == 0) |
1da177e4 LT |
3151 | goto out; |
3152 | ||
81026794 NP |
3153 | pinned = 1; |
3154 | ||
1da177e4 | 3155 | /* |
dd41f596 | 3156 | * Start the load-balancing iterator: |
1da177e4 | 3157 | */ |
dd41f596 IM |
3158 | p = iterator->start(iterator->arg); |
3159 | next: | |
b82d9fdd | 3160 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3161 | goto out; |
051c6764 PZ |
3162 | |
3163 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3164 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3165 | p = iterator->next(iterator->arg); |
3166 | goto next; | |
1da177e4 LT |
3167 | } |
3168 | ||
dd41f596 | 3169 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3170 | pulled++; |
dd41f596 | 3171 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3172 | |
7e96fa58 GH |
3173 | #ifdef CONFIG_PREEMPT |
3174 | /* | |
3175 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3176 | * will stop after the first task is pulled to minimize the critical | |
3177 | * section. | |
3178 | */ | |
3179 | if (idle == CPU_NEWLY_IDLE) | |
3180 | goto out; | |
3181 | #endif | |
3182 | ||
2dd73a4f | 3183 | /* |
b82d9fdd | 3184 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3185 | */ |
e1d1484f | 3186 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3187 | if (p->prio < *this_best_prio) |
3188 | *this_best_prio = p->prio; | |
dd41f596 IM |
3189 | p = iterator->next(iterator->arg); |
3190 | goto next; | |
1da177e4 LT |
3191 | } |
3192 | out: | |
3193 | /* | |
e1d1484f | 3194 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3195 | * so we can safely collect pull_task() stats here rather than |
3196 | * inside pull_task(). | |
3197 | */ | |
3198 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3199 | |
3200 | if (all_pinned) | |
3201 | *all_pinned = pinned; | |
e1d1484f PW |
3202 | |
3203 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3204 | } |
3205 | ||
dd41f596 | 3206 | /* |
43010659 PW |
3207 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3208 | * this_rq, as part of a balancing operation within domain "sd". | |
3209 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3210 | * |
3211 | * Called with both runqueues locked. | |
3212 | */ | |
3213 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3214 | unsigned long max_load_move, |
dd41f596 IM |
3215 | struct sched_domain *sd, enum cpu_idle_type idle, |
3216 | int *all_pinned) | |
3217 | { | |
5522d5d5 | 3218 | const struct sched_class *class = sched_class_highest; |
43010659 | 3219 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3220 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3221 | |
3222 | do { | |
43010659 PW |
3223 | total_load_moved += |
3224 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3225 | max_load_move - total_load_moved, |
a4ac01c3 | 3226 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3227 | class = class->next; |
c4acb2c0 | 3228 | |
7e96fa58 GH |
3229 | #ifdef CONFIG_PREEMPT |
3230 | /* | |
3231 | * NEWIDLE balancing is a source of latency, so preemptible | |
3232 | * kernels will stop after the first task is pulled to minimize | |
3233 | * the critical section. | |
3234 | */ | |
c4acb2c0 GH |
3235 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3236 | break; | |
7e96fa58 | 3237 | #endif |
43010659 | 3238 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3239 | |
43010659 PW |
3240 | return total_load_moved > 0; |
3241 | } | |
3242 | ||
e1d1484f PW |
3243 | static int |
3244 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3245 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3246 | struct rq_iterator *iterator) | |
3247 | { | |
3248 | struct task_struct *p = iterator->start(iterator->arg); | |
3249 | int pinned = 0; | |
3250 | ||
3251 | while (p) { | |
3252 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3253 | pull_task(busiest, p, this_rq, this_cpu); | |
3254 | /* | |
3255 | * Right now, this is only the second place pull_task() | |
3256 | * is called, so we can safely collect pull_task() | |
3257 | * stats here rather than inside pull_task(). | |
3258 | */ | |
3259 | schedstat_inc(sd, lb_gained[idle]); | |
3260 | ||
3261 | return 1; | |
3262 | } | |
3263 | p = iterator->next(iterator->arg); | |
3264 | } | |
3265 | ||
3266 | return 0; | |
3267 | } | |
3268 | ||
43010659 PW |
3269 | /* |
3270 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3271 | * part of active balancing operations within "domain". | |
3272 | * Returns 1 if successful and 0 otherwise. | |
3273 | * | |
3274 | * Called with both runqueues locked. | |
3275 | */ | |
3276 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3277 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3278 | { | |
5522d5d5 | 3279 | const struct sched_class *class; |
43010659 PW |
3280 | |
3281 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3282 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3283 | return 1; |
3284 | ||
3285 | return 0; | |
dd41f596 | 3286 | } |
67bb6c03 | 3287 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3288 | /* |
222d656d GS |
3289 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3290 | * during load balancing. | |
1da177e4 | 3291 | */ |
222d656d GS |
3292 | struct sd_lb_stats { |
3293 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3294 | struct sched_group *this; /* Local group in this sd */ | |
3295 | unsigned long total_load; /* Total load of all groups in sd */ | |
3296 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3297 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3298 | ||
3299 | /** Statistics of this group */ | |
3300 | unsigned long this_load; | |
3301 | unsigned long this_load_per_task; | |
3302 | unsigned long this_nr_running; | |
3303 | ||
3304 | /* Statistics of the busiest group */ | |
3305 | unsigned long max_load; | |
3306 | unsigned long busiest_load_per_task; | |
3307 | unsigned long busiest_nr_running; | |
3308 | ||
3309 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3310 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3311 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3312 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3313 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3314 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3315 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3316 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3317 | #endif |
222d656d | 3318 | }; |
1da177e4 | 3319 | |
d5ac537e | 3320 | /* |
381be78f GS |
3321 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3322 | */ | |
3323 | struct sg_lb_stats { | |
3324 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3325 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3326 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3327 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3328 | unsigned long group_capacity; | |
3329 | int group_imb; /* Is there an imbalance in the group ? */ | |
3330 | }; | |
408ed066 | 3331 | |
67bb6c03 GS |
3332 | /** |
3333 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3334 | * @group: The group whose first cpu is to be returned. | |
3335 | */ | |
3336 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3337 | { | |
3338 | return cpumask_first(sched_group_cpus(group)); | |
3339 | } | |
3340 | ||
3341 | /** | |
3342 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3343 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3344 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3345 | */ | |
3346 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3347 | enum cpu_idle_type idle) | |
3348 | { | |
3349 | int load_idx; | |
3350 | ||
3351 | switch (idle) { | |
3352 | case CPU_NOT_IDLE: | |
7897986b | 3353 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3354 | break; |
3355 | ||
3356 | case CPU_NEWLY_IDLE: | |
7897986b | 3357 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3358 | break; |
3359 | default: | |
7897986b | 3360 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3361 | break; |
3362 | } | |
1da177e4 | 3363 | |
67bb6c03 GS |
3364 | return load_idx; |
3365 | } | |
1da177e4 | 3366 | |
1da177e4 | 3367 | |
c071df18 GS |
3368 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3369 | /** | |
3370 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3371 | * the given sched_domain, during load balancing. | |
3372 | * | |
3373 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3374 | * @sds: Variable containing the statistics for sd. | |
3375 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3376 | */ | |
3377 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3378 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3379 | { | |
3380 | /* | |
3381 | * Busy processors will not participate in power savings | |
3382 | * balance. | |
3383 | */ | |
3384 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3385 | sds->power_savings_balance = 0; | |
3386 | else { | |
3387 | sds->power_savings_balance = 1; | |
3388 | sds->min_nr_running = ULONG_MAX; | |
3389 | sds->leader_nr_running = 0; | |
3390 | } | |
3391 | } | |
783609c6 | 3392 | |
c071df18 GS |
3393 | /** |
3394 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3395 | * sched_domain while performing load balancing. | |
3396 | * | |
3397 | * @group: sched_group belonging to the sched_domain under consideration. | |
3398 | * @sds: Variable containing the statistics of the sched_domain | |
3399 | * @local_group: Does group contain the CPU for which we're performing | |
3400 | * load balancing ? | |
3401 | * @sgs: Variable containing the statistics of the group. | |
3402 | */ | |
3403 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3404 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3405 | { | |
408ed066 | 3406 | |
c071df18 GS |
3407 | if (!sds->power_savings_balance) |
3408 | return; | |
1da177e4 | 3409 | |
c071df18 GS |
3410 | /* |
3411 | * If the local group is idle or completely loaded | |
3412 | * no need to do power savings balance at this domain | |
3413 | */ | |
3414 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3415 | !sds->this_nr_running)) | |
3416 | sds->power_savings_balance = 0; | |
2dd73a4f | 3417 | |
c071df18 GS |
3418 | /* |
3419 | * If a group is already running at full capacity or idle, | |
3420 | * don't include that group in power savings calculations | |
3421 | */ | |
3422 | if (!sds->power_savings_balance || | |
3423 | sgs->sum_nr_running >= sgs->group_capacity || | |
3424 | !sgs->sum_nr_running) | |
3425 | return; | |
5969fe06 | 3426 | |
c071df18 GS |
3427 | /* |
3428 | * Calculate the group which has the least non-idle load. | |
3429 | * This is the group from where we need to pick up the load | |
3430 | * for saving power | |
3431 | */ | |
3432 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3433 | (sgs->sum_nr_running == sds->min_nr_running && | |
3434 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3435 | sds->group_min = group; | |
3436 | sds->min_nr_running = sgs->sum_nr_running; | |
3437 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3438 | sgs->sum_nr_running; | |
3439 | } | |
783609c6 | 3440 | |
c071df18 GS |
3441 | /* |
3442 | * Calculate the group which is almost near its | |
3443 | * capacity but still has some space to pick up some load | |
3444 | * from other group and save more power | |
3445 | */ | |
3446 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3447 | return; | |
1da177e4 | 3448 | |
c071df18 GS |
3449 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3450 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3451 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3452 | sds->group_leader = group; | |
3453 | sds->leader_nr_running = sgs->sum_nr_running; | |
3454 | } | |
3455 | } | |
408ed066 | 3456 | |
c071df18 | 3457 | /** |
d5ac537e | 3458 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3459 | * @sds: Variable containing the statistics of the sched_domain |
3460 | * under consideration. | |
3461 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3462 | * @imbalance: Variable to store the imbalance. | |
3463 | * | |
d5ac537e RD |
3464 | * Description: |
3465 | * Check if we have potential to perform some power-savings balance. | |
3466 | * If yes, set the busiest group to be the least loaded group in the | |
3467 | * sched_domain, so that it's CPUs can be put to idle. | |
3468 | * | |
c071df18 GS |
3469 | * Returns 1 if there is potential to perform power-savings balance. |
3470 | * Else returns 0. | |
3471 | */ | |
3472 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3473 | int this_cpu, unsigned long *imbalance) | |
3474 | { | |
3475 | if (!sds->power_savings_balance) | |
3476 | return 0; | |
1da177e4 | 3477 | |
c071df18 GS |
3478 | if (sds->this != sds->group_leader || |
3479 | sds->group_leader == sds->group_min) | |
3480 | return 0; | |
783609c6 | 3481 | |
c071df18 GS |
3482 | *imbalance = sds->min_load_per_task; |
3483 | sds->busiest = sds->group_min; | |
1da177e4 | 3484 | |
c071df18 GS |
3485 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3486 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3487 | group_first_cpu(sds->group_leader); | |
3488 | } | |
3489 | ||
3490 | return 1; | |
1da177e4 | 3491 | |
c071df18 GS |
3492 | } |
3493 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3494 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3495 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3496 | { | |
3497 | return; | |
3498 | } | |
408ed066 | 3499 | |
c071df18 GS |
3500 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3501 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3502 | { | |
3503 | return; | |
3504 | } | |
3505 | ||
3506 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3507 | int this_cpu, unsigned long *imbalance) | |
3508 | { | |
3509 | return 0; | |
3510 | } | |
3511 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3512 | ||
3513 | ||
1f8c553d GS |
3514 | /** |
3515 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3516 | * @group: sched_group whose statistics are to be updated. | |
3517 | * @this_cpu: Cpu for which load balance is currently performed. | |
3518 | * @idle: Idle status of this_cpu | |
3519 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3520 | * @sd_idle: Idle status of the sched_domain containing group. | |
3521 | * @local_group: Does group contain this_cpu. | |
3522 | * @cpus: Set of cpus considered for load balancing. | |
3523 | * @balance: Should we balance. | |
3524 | * @sgs: variable to hold the statistics for this group. | |
3525 | */ | |
3526 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3527 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3528 | int local_group, const struct cpumask *cpus, | |
3529 | int *balance, struct sg_lb_stats *sgs) | |
3530 | { | |
3531 | unsigned long load, max_cpu_load, min_cpu_load; | |
3532 | int i; | |
3533 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3534 | unsigned long sum_avg_load_per_task; | |
3535 | unsigned long avg_load_per_task; | |
3536 | ||
3537 | if (local_group) | |
3538 | balance_cpu = group_first_cpu(group); | |
3539 | ||
3540 | /* Tally up the load of all CPUs in the group */ | |
3541 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3542 | max_cpu_load = 0; | |
3543 | min_cpu_load = ~0UL; | |
408ed066 | 3544 | |
1f8c553d GS |
3545 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3546 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3547 | |
1f8c553d GS |
3548 | if (*sd_idle && rq->nr_running) |
3549 | *sd_idle = 0; | |
5c45bf27 | 3550 | |
1f8c553d | 3551 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3552 | if (local_group) { |
1f8c553d GS |
3553 | if (idle_cpu(i) && !first_idle_cpu) { |
3554 | first_idle_cpu = 1; | |
3555 | balance_cpu = i; | |
3556 | } | |
3557 | ||
3558 | load = target_load(i, load_idx); | |
3559 | } else { | |
3560 | load = source_load(i, load_idx); | |
3561 | if (load > max_cpu_load) | |
3562 | max_cpu_load = load; | |
3563 | if (min_cpu_load > load) | |
3564 | min_cpu_load = load; | |
1da177e4 | 3565 | } |
5c45bf27 | 3566 | |
1f8c553d GS |
3567 | sgs->group_load += load; |
3568 | sgs->sum_nr_running += rq->nr_running; | |
3569 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3570 | |
1f8c553d GS |
3571 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3572 | } | |
5c45bf27 | 3573 | |
1f8c553d GS |
3574 | /* |
3575 | * First idle cpu or the first cpu(busiest) in this sched group | |
3576 | * is eligible for doing load balancing at this and above | |
3577 | * domains. In the newly idle case, we will allow all the cpu's | |
3578 | * to do the newly idle load balance. | |
3579 | */ | |
3580 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3581 | balance_cpu != this_cpu && balance) { | |
3582 | *balance = 0; | |
3583 | return; | |
3584 | } | |
5c45bf27 | 3585 | |
1f8c553d GS |
3586 | /* Adjust by relative CPU power of the group */ |
3587 | sgs->avg_load = sg_div_cpu_power(group, | |
3588 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3589 | |
1f8c553d GS |
3590 | |
3591 | /* | |
3592 | * Consider the group unbalanced when the imbalance is larger | |
3593 | * than the average weight of two tasks. | |
3594 | * | |
3595 | * APZ: with cgroup the avg task weight can vary wildly and | |
3596 | * might not be a suitable number - should we keep a | |
3597 | * normalized nr_running number somewhere that negates | |
3598 | * the hierarchy? | |
3599 | */ | |
3600 | avg_load_per_task = sg_div_cpu_power(group, | |
3601 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3602 | ||
3603 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3604 | sgs->group_imb = 1; | |
3605 | ||
3606 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3607 | ||
3608 | } | |
dd41f596 | 3609 | |
37abe198 GS |
3610 | /** |
3611 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3612 | * @sd: sched_domain whose statistics are to be updated. | |
3613 | * @this_cpu: Cpu for which load balance is currently performed. | |
3614 | * @idle: Idle status of this_cpu | |
3615 | * @sd_idle: Idle status of the sched_domain containing group. | |
3616 | * @cpus: Set of cpus considered for load balancing. | |
3617 | * @balance: Should we balance. | |
3618 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3619 | */ |
37abe198 GS |
3620 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3621 | enum cpu_idle_type idle, int *sd_idle, | |
3622 | const struct cpumask *cpus, int *balance, | |
3623 | struct sd_lb_stats *sds) | |
1da177e4 | 3624 | { |
222d656d | 3625 | struct sched_group *group = sd->groups; |
37abe198 | 3626 | struct sg_lb_stats sgs; |
222d656d GS |
3627 | int load_idx; |
3628 | ||
c071df18 | 3629 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3630 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3631 | |
3632 | do { | |
1da177e4 | 3633 | int local_group; |
1da177e4 | 3634 | |
758b2cdc RR |
3635 | local_group = cpumask_test_cpu(this_cpu, |
3636 | sched_group_cpus(group)); | |
381be78f | 3637 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3638 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3639 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3640 | |
37abe198 GS |
3641 | if (local_group && balance && !(*balance)) |
3642 | return; | |
783609c6 | 3643 | |
37abe198 GS |
3644 | sds->total_load += sgs.group_load; |
3645 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3646 | |
1da177e4 | 3647 | if (local_group) { |
37abe198 GS |
3648 | sds->this_load = sgs.avg_load; |
3649 | sds->this = group; | |
3650 | sds->this_nr_running = sgs.sum_nr_running; | |
3651 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3652 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3653 | (sgs.sum_nr_running > sgs.group_capacity || |
3654 | sgs.group_imb)) { | |
37abe198 GS |
3655 | sds->max_load = sgs.avg_load; |
3656 | sds->busiest = group; | |
3657 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3658 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3659 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3660 | } |
5c45bf27 | 3661 | |
c071df18 | 3662 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3663 | group = group->next; |
3664 | } while (group != sd->groups); | |
3665 | ||
37abe198 | 3666 | } |
1da177e4 | 3667 | |
2e6f44ae GS |
3668 | /** |
3669 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3670 | * amongst the groups of a sched_domain, during |
3671 | * load balancing. | |
2e6f44ae GS |
3672 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3673 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3674 | * @imbalance: Variable to store the imbalance. | |
3675 | */ | |
3676 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3677 | int this_cpu, unsigned long *imbalance) | |
3678 | { | |
3679 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3680 | unsigned int imbn = 2; | |
3681 | ||
3682 | if (sds->this_nr_running) { | |
3683 | sds->this_load_per_task /= sds->this_nr_running; | |
3684 | if (sds->busiest_load_per_task > | |
3685 | sds->this_load_per_task) | |
3686 | imbn = 1; | |
3687 | } else | |
3688 | sds->this_load_per_task = | |
3689 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3690 | |
2e6f44ae GS |
3691 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3692 | sds->busiest_load_per_task * imbn) { | |
3693 | *imbalance = sds->busiest_load_per_task; | |
3694 | return; | |
3695 | } | |
908a7c1b | 3696 | |
1da177e4 | 3697 | /* |
2e6f44ae GS |
3698 | * OK, we don't have enough imbalance to justify moving tasks, |
3699 | * however we may be able to increase total CPU power used by | |
3700 | * moving them. | |
1da177e4 | 3701 | */ |
2dd73a4f | 3702 | |
2e6f44ae GS |
3703 | pwr_now += sds->busiest->__cpu_power * |
3704 | min(sds->busiest_load_per_task, sds->max_load); | |
3705 | pwr_now += sds->this->__cpu_power * | |
3706 | min(sds->this_load_per_task, sds->this_load); | |
3707 | pwr_now /= SCHED_LOAD_SCALE; | |
3708 | ||
3709 | /* Amount of load we'd subtract */ | |
3710 | tmp = sg_div_cpu_power(sds->busiest, | |
3711 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3712 | if (sds->max_load > tmp) | |
3713 | pwr_move += sds->busiest->__cpu_power * | |
3714 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3715 | ||
3716 | /* Amount of load we'd add */ | |
3717 | if (sds->max_load * sds->busiest->__cpu_power < | |
3718 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3719 | tmp = sg_div_cpu_power(sds->this, | |
3720 | sds->max_load * sds->busiest->__cpu_power); | |
3721 | else | |
3722 | tmp = sg_div_cpu_power(sds->this, | |
3723 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3724 | pwr_move += sds->this->__cpu_power * | |
3725 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3726 | pwr_move /= SCHED_LOAD_SCALE; | |
3727 | ||
3728 | /* Move if we gain throughput */ | |
3729 | if (pwr_move > pwr_now) | |
3730 | *imbalance = sds->busiest_load_per_task; | |
3731 | } | |
dbc523a3 GS |
3732 | |
3733 | /** | |
3734 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3735 | * groups of a given sched_domain during load balance. | |
3736 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3737 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3738 | * @imbalance: The variable to store the imbalance. | |
3739 | */ | |
3740 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3741 | unsigned long *imbalance) | |
3742 | { | |
3743 | unsigned long max_pull; | |
2dd73a4f PW |
3744 | /* |
3745 | * In the presence of smp nice balancing, certain scenarios can have | |
3746 | * max load less than avg load(as we skip the groups at or below | |
3747 | * its cpu_power, while calculating max_load..) | |
3748 | */ | |
dbc523a3 | 3749 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3750 | *imbalance = 0; |
dbc523a3 | 3751 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3752 | } |
0c117f1b SS |
3753 | |
3754 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3755 | max_pull = min(sds->max_load - sds->avg_load, |
3756 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3757 | |
1da177e4 | 3758 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3759 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3760 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3761 | / SCHED_LOAD_SCALE; |
3762 | ||
2dd73a4f PW |
3763 | /* |
3764 | * if *imbalance is less than the average load per runnable task | |
3765 | * there is no gaurantee that any tasks will be moved so we'll have | |
3766 | * a think about bumping its value to force at least one task to be | |
3767 | * moved | |
3768 | */ | |
dbc523a3 GS |
3769 | if (*imbalance < sds->busiest_load_per_task) |
3770 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3771 | |
dbc523a3 | 3772 | } |
37abe198 | 3773 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3774 | |
b7bb4c9b GS |
3775 | /** |
3776 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3777 | * if there is an imbalance. If there isn't an imbalance, and | |
3778 | * the user has opted for power-savings, it returns a group whose | |
3779 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3780 | * such a group exists. | |
3781 | * | |
3782 | * Also calculates the amount of weighted load which should be moved | |
3783 | * to restore balance. | |
3784 | * | |
3785 | * @sd: The sched_domain whose busiest group is to be returned. | |
3786 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3787 | * @imbalance: Variable which stores amount of weighted load which should | |
3788 | * be moved to restore balance/put a group to idle. | |
3789 | * @idle: The idle status of this_cpu. | |
3790 | * @sd_idle: The idleness of sd | |
3791 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3792 | * @balance: Pointer to a variable indicating if this_cpu | |
3793 | * is the appropriate cpu to perform load balancing at this_level. | |
3794 | * | |
3795 | * Returns: - the busiest group if imbalance exists. | |
3796 | * - If no imbalance and user has opted for power-savings balance, | |
3797 | * return the least loaded group whose CPUs can be | |
3798 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3799 | */ |
3800 | static struct sched_group * | |
3801 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3802 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3803 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3804 | { | |
3805 | struct sd_lb_stats sds; | |
1da177e4 | 3806 | |
37abe198 | 3807 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3808 | |
37abe198 GS |
3809 | /* |
3810 | * Compute the various statistics relavent for load balancing at | |
3811 | * this level. | |
3812 | */ | |
3813 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3814 | balance, &sds); | |
3815 | ||
b7bb4c9b GS |
3816 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3817 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3818 | * at this level. | |
3819 | * 2) There is no busy sibling group to pull from. | |
3820 | * 3) This group is the busiest group. | |
3821 | * 4) This group is more busy than the avg busieness at this | |
3822 | * sched_domain. | |
3823 | * 5) The imbalance is within the specified limit. | |
3824 | * 6) Any rebalance would lead to ping-pong | |
3825 | */ | |
37abe198 GS |
3826 | if (balance && !(*balance)) |
3827 | goto ret; | |
1da177e4 | 3828 | |
b7bb4c9b GS |
3829 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3830 | goto out_balanced; | |
1da177e4 | 3831 | |
b7bb4c9b | 3832 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3833 | goto out_balanced; |
1da177e4 | 3834 | |
222d656d | 3835 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3836 | |
b7bb4c9b GS |
3837 | if (sds.this_load >= sds.avg_load) |
3838 | goto out_balanced; | |
3839 | ||
3840 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3841 | goto out_balanced; |
3842 | ||
222d656d GS |
3843 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3844 | if (sds.group_imb) | |
3845 | sds.busiest_load_per_task = | |
3846 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3847 | |
1da177e4 LT |
3848 | /* |
3849 | * We're trying to get all the cpus to the average_load, so we don't | |
3850 | * want to push ourselves above the average load, nor do we wish to | |
3851 | * reduce the max loaded cpu below the average load, as either of these | |
3852 | * actions would just result in more rebalancing later, and ping-pong | |
3853 | * tasks around. Thus we look for the minimum possible imbalance. | |
3854 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3855 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3856 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3857 | * appear as very large values with unsigned longs. |
3858 | */ | |
222d656d | 3859 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3860 | goto out_balanced; |
3861 | ||
dbc523a3 GS |
3862 | /* Looks like there is an imbalance. Compute it */ |
3863 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3864 | return sds.busiest; |
1da177e4 LT |
3865 | |
3866 | out_balanced: | |
c071df18 GS |
3867 | /* |
3868 | * There is no obvious imbalance. But check if we can do some balancing | |
3869 | * to save power. | |
3870 | */ | |
3871 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3872 | return sds.busiest; | |
783609c6 | 3873 | ret: |
1da177e4 LT |
3874 | *imbalance = 0; |
3875 | return NULL; | |
3876 | } | |
3877 | ||
3878 | /* | |
3879 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3880 | */ | |
70b97a7f | 3881 | static struct rq * |
d15bcfdb | 3882 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3883 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3884 | { |
70b97a7f | 3885 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3886 | unsigned long max_load = 0; |
1da177e4 LT |
3887 | int i; |
3888 | ||
758b2cdc | 3889 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3890 | unsigned long wl; |
0a2966b4 | 3891 | |
96f874e2 | 3892 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3893 | continue; |
3894 | ||
48f24c4d | 3895 | rq = cpu_rq(i); |
dd41f596 | 3896 | wl = weighted_cpuload(i); |
2dd73a4f | 3897 | |
dd41f596 | 3898 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3899 | continue; |
1da177e4 | 3900 | |
dd41f596 IM |
3901 | if (wl > max_load) { |
3902 | max_load = wl; | |
48f24c4d | 3903 | busiest = rq; |
1da177e4 LT |
3904 | } |
3905 | } | |
3906 | ||
3907 | return busiest; | |
3908 | } | |
3909 | ||
77391d71 NP |
3910 | /* |
3911 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3912 | * so long as it is large enough. | |
3913 | */ | |
3914 | #define MAX_PINNED_INTERVAL 512 | |
3915 | ||
df7c8e84 RR |
3916 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3917 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3918 | ||
1da177e4 LT |
3919 | /* |
3920 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3921 | * tasks if there is an imbalance. | |
1da177e4 | 3922 | */ |
70b97a7f | 3923 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3924 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 3925 | int *balance) |
1da177e4 | 3926 | { |
43010659 | 3927 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3928 | struct sched_group *group; |
1da177e4 | 3929 | unsigned long imbalance; |
70b97a7f | 3930 | struct rq *busiest; |
fe2eea3f | 3931 | unsigned long flags; |
df7c8e84 | 3932 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 3933 | |
96f874e2 | 3934 | cpumask_setall(cpus); |
7c16ec58 | 3935 | |
89c4710e SS |
3936 | /* |
3937 | * When power savings policy is enabled for the parent domain, idle | |
3938 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3939 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3940 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3941 | */ |
d15bcfdb | 3942 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3943 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3944 | sd_idle = 1; |
1da177e4 | 3945 | |
2d72376b | 3946 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3947 | |
0a2966b4 | 3948 | redo: |
c8cba857 | 3949 | update_shares(sd); |
0a2966b4 | 3950 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3951 | cpus, balance); |
783609c6 | 3952 | |
06066714 | 3953 | if (*balance == 0) |
783609c6 | 3954 | goto out_balanced; |
783609c6 | 3955 | |
1da177e4 LT |
3956 | if (!group) { |
3957 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3958 | goto out_balanced; | |
3959 | } | |
3960 | ||
7c16ec58 | 3961 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3962 | if (!busiest) { |
3963 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3964 | goto out_balanced; | |
3965 | } | |
3966 | ||
db935dbd | 3967 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3968 | |
3969 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3970 | ||
43010659 | 3971 | ld_moved = 0; |
1da177e4 LT |
3972 | if (busiest->nr_running > 1) { |
3973 | /* | |
3974 | * Attempt to move tasks. If find_busiest_group has found | |
3975 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3976 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3977 | * correctly treated as an imbalance. |
3978 | */ | |
fe2eea3f | 3979 | local_irq_save(flags); |
e17224bf | 3980 | double_rq_lock(this_rq, busiest); |
43010659 | 3981 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3982 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3983 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3984 | local_irq_restore(flags); |
81026794 | 3985 | |
46cb4b7c SS |
3986 | /* |
3987 | * some other cpu did the load balance for us. | |
3988 | */ | |
43010659 | 3989 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3990 | resched_cpu(this_cpu); |
3991 | ||
81026794 | 3992 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3993 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3994 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3995 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3996 | goto redo; |
81026794 | 3997 | goto out_balanced; |
0a2966b4 | 3998 | } |
1da177e4 | 3999 | } |
81026794 | 4000 | |
43010659 | 4001 | if (!ld_moved) { |
1da177e4 LT |
4002 | schedstat_inc(sd, lb_failed[idle]); |
4003 | sd->nr_balance_failed++; | |
4004 | ||
4005 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4006 | |
fe2eea3f | 4007 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4008 | |
4009 | /* don't kick the migration_thread, if the curr | |
4010 | * task on busiest cpu can't be moved to this_cpu | |
4011 | */ | |
96f874e2 RR |
4012 | if (!cpumask_test_cpu(this_cpu, |
4013 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4014 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4015 | all_pinned = 1; |
4016 | goto out_one_pinned; | |
4017 | } | |
4018 | ||
1da177e4 LT |
4019 | if (!busiest->active_balance) { |
4020 | busiest->active_balance = 1; | |
4021 | busiest->push_cpu = this_cpu; | |
81026794 | 4022 | active_balance = 1; |
1da177e4 | 4023 | } |
fe2eea3f | 4024 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4025 | if (active_balance) |
1da177e4 LT |
4026 | wake_up_process(busiest->migration_thread); |
4027 | ||
4028 | /* | |
4029 | * We've kicked active balancing, reset the failure | |
4030 | * counter. | |
4031 | */ | |
39507451 | 4032 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4033 | } |
81026794 | 4034 | } else |
1da177e4 LT |
4035 | sd->nr_balance_failed = 0; |
4036 | ||
81026794 | 4037 | if (likely(!active_balance)) { |
1da177e4 LT |
4038 | /* We were unbalanced, so reset the balancing interval */ |
4039 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4040 | } else { |
4041 | /* | |
4042 | * If we've begun active balancing, start to back off. This | |
4043 | * case may not be covered by the all_pinned logic if there | |
4044 | * is only 1 task on the busy runqueue (because we don't call | |
4045 | * move_tasks). | |
4046 | */ | |
4047 | if (sd->balance_interval < sd->max_interval) | |
4048 | sd->balance_interval *= 2; | |
1da177e4 LT |
4049 | } |
4050 | ||
43010659 | 4051 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4052 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4053 | ld_moved = -1; |
4054 | ||
4055 | goto out; | |
1da177e4 LT |
4056 | |
4057 | out_balanced: | |
1da177e4 LT |
4058 | schedstat_inc(sd, lb_balanced[idle]); |
4059 | ||
16cfb1c0 | 4060 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4061 | |
4062 | out_one_pinned: | |
1da177e4 | 4063 | /* tune up the balancing interval */ |
77391d71 NP |
4064 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4065 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4066 | sd->balance_interval *= 2; |
4067 | ||
48f24c4d | 4068 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4069 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4070 | ld_moved = -1; |
4071 | else | |
4072 | ld_moved = 0; | |
4073 | out: | |
c8cba857 PZ |
4074 | if (ld_moved) |
4075 | update_shares(sd); | |
c09595f6 | 4076 | return ld_moved; |
1da177e4 LT |
4077 | } |
4078 | ||
4079 | /* | |
4080 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4081 | * tasks if there is an imbalance. | |
4082 | * | |
d15bcfdb | 4083 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4084 | * this_rq is locked. |
4085 | */ | |
48f24c4d | 4086 | static int |
df7c8e84 | 4087 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4088 | { |
4089 | struct sched_group *group; | |
70b97a7f | 4090 | struct rq *busiest = NULL; |
1da177e4 | 4091 | unsigned long imbalance; |
43010659 | 4092 | int ld_moved = 0; |
5969fe06 | 4093 | int sd_idle = 0; |
969bb4e4 | 4094 | int all_pinned = 0; |
df7c8e84 | 4095 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4096 | |
96f874e2 | 4097 | cpumask_setall(cpus); |
5969fe06 | 4098 | |
89c4710e SS |
4099 | /* |
4100 | * When power savings policy is enabled for the parent domain, idle | |
4101 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4102 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4103 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4104 | */ |
4105 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4106 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4107 | sd_idle = 1; |
1da177e4 | 4108 | |
2d72376b | 4109 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4110 | redo: |
3e5459b4 | 4111 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4112 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4113 | &sd_idle, cpus, NULL); |
1da177e4 | 4114 | if (!group) { |
d15bcfdb | 4115 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4116 | goto out_balanced; |
1da177e4 LT |
4117 | } |
4118 | ||
7c16ec58 | 4119 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4120 | if (!busiest) { |
d15bcfdb | 4121 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4122 | goto out_balanced; |
1da177e4 LT |
4123 | } |
4124 | ||
db935dbd NP |
4125 | BUG_ON(busiest == this_rq); |
4126 | ||
d15bcfdb | 4127 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4128 | |
43010659 | 4129 | ld_moved = 0; |
d6d5cfaf NP |
4130 | if (busiest->nr_running > 1) { |
4131 | /* Attempt to move tasks */ | |
4132 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4133 | /* this_rq->clock is already updated */ |
4134 | update_rq_clock(busiest); | |
43010659 | 4135 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4136 | imbalance, sd, CPU_NEWLY_IDLE, |
4137 | &all_pinned); | |
1b12bbc7 | 4138 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4139 | |
969bb4e4 | 4140 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4141 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4142 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4143 | goto redo; |
4144 | } | |
d6d5cfaf NP |
4145 | } |
4146 | ||
43010659 | 4147 | if (!ld_moved) { |
36dffab6 | 4148 | int active_balance = 0; |
ad273b32 | 4149 | |
d15bcfdb | 4150 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4151 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4152 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4153 | return -1; |
ad273b32 VS |
4154 | |
4155 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4156 | return -1; | |
4157 | ||
4158 | if (sd->nr_balance_failed++ < 2) | |
4159 | return -1; | |
4160 | ||
4161 | /* | |
4162 | * The only task running in a non-idle cpu can be moved to this | |
4163 | * cpu in an attempt to completely freeup the other CPU | |
4164 | * package. The same method used to move task in load_balance() | |
4165 | * have been extended for load_balance_newidle() to speedup | |
4166 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4167 | * | |
4168 | * The package power saving logic comes from | |
4169 | * find_busiest_group(). If there are no imbalance, then | |
4170 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4171 | * f_b_g() will select a group from which a running task may be | |
4172 | * pulled to this cpu in order to make the other package idle. | |
4173 | * If there is no opportunity to make a package idle and if | |
4174 | * there are no imbalance, then f_b_g() will return NULL and no | |
4175 | * action will be taken in load_balance_newidle(). | |
4176 | * | |
4177 | * Under normal task pull operation due to imbalance, there | |
4178 | * will be more than one task in the source run queue and | |
4179 | * move_tasks() will succeed. ld_moved will be true and this | |
4180 | * active balance code will not be triggered. | |
4181 | */ | |
4182 | ||
4183 | /* Lock busiest in correct order while this_rq is held */ | |
4184 | double_lock_balance(this_rq, busiest); | |
4185 | ||
4186 | /* | |
4187 | * don't kick the migration_thread, if the curr | |
4188 | * task on busiest cpu can't be moved to this_cpu | |
4189 | */ | |
6ca09dfc | 4190 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4191 | double_unlock_balance(this_rq, busiest); |
4192 | all_pinned = 1; | |
4193 | return ld_moved; | |
4194 | } | |
4195 | ||
4196 | if (!busiest->active_balance) { | |
4197 | busiest->active_balance = 1; | |
4198 | busiest->push_cpu = this_cpu; | |
4199 | active_balance = 1; | |
4200 | } | |
4201 | ||
4202 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4203 | /* |
4204 | * Should not call ttwu while holding a rq->lock | |
4205 | */ | |
4206 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4207 | if (active_balance) |
4208 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4209 | spin_lock(&this_rq->lock); |
ad273b32 | 4210 | |
5969fe06 | 4211 | } else |
16cfb1c0 | 4212 | sd->nr_balance_failed = 0; |
1da177e4 | 4213 | |
3e5459b4 | 4214 | update_shares_locked(this_rq, sd); |
43010659 | 4215 | return ld_moved; |
16cfb1c0 NP |
4216 | |
4217 | out_balanced: | |
d15bcfdb | 4218 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4219 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4220 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4221 | return -1; |
16cfb1c0 | 4222 | sd->nr_balance_failed = 0; |
48f24c4d | 4223 | |
16cfb1c0 | 4224 | return 0; |
1da177e4 LT |
4225 | } |
4226 | ||
4227 | /* | |
4228 | * idle_balance is called by schedule() if this_cpu is about to become | |
4229 | * idle. Attempts to pull tasks from other CPUs. | |
4230 | */ | |
70b97a7f | 4231 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4232 | { |
4233 | struct sched_domain *sd; | |
efbe027e | 4234 | int pulled_task = 0; |
dd41f596 | 4235 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4236 | |
4237 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4238 | unsigned long interval; |
4239 | ||
4240 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4241 | continue; | |
4242 | ||
4243 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4244 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4245 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4246 | sd); |
92c4ca5c CL |
4247 | |
4248 | interval = msecs_to_jiffies(sd->balance_interval); | |
4249 | if (time_after(next_balance, sd->last_balance + interval)) | |
4250 | next_balance = sd->last_balance + interval; | |
4251 | if (pulled_task) | |
4252 | break; | |
1da177e4 | 4253 | } |
dd41f596 | 4254 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4255 | /* |
4256 | * We are going idle. next_balance may be set based on | |
4257 | * a busy processor. So reset next_balance. | |
4258 | */ | |
4259 | this_rq->next_balance = next_balance; | |
dd41f596 | 4260 | } |
1da177e4 LT |
4261 | } |
4262 | ||
4263 | /* | |
4264 | * active_load_balance is run by migration threads. It pushes running tasks | |
4265 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4266 | * running on each physical CPU where possible, and avoids physical / | |
4267 | * logical imbalances. | |
4268 | * | |
4269 | * Called with busiest_rq locked. | |
4270 | */ | |
70b97a7f | 4271 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4272 | { |
39507451 | 4273 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4274 | struct sched_domain *sd; |
4275 | struct rq *target_rq; | |
39507451 | 4276 | |
48f24c4d | 4277 | /* Is there any task to move? */ |
39507451 | 4278 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4279 | return; |
4280 | ||
4281 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4282 | |
4283 | /* | |
39507451 | 4284 | * This condition is "impossible", if it occurs |
41a2d6cf | 4285 | * we need to fix it. Originally reported by |
39507451 | 4286 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4287 | */ |
39507451 | 4288 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4289 | |
39507451 NP |
4290 | /* move a task from busiest_rq to target_rq */ |
4291 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4292 | update_rq_clock(busiest_rq); |
4293 | update_rq_clock(target_rq); | |
39507451 NP |
4294 | |
4295 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4296 | for_each_domain(target_cpu, sd) { |
39507451 | 4297 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4298 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4299 | break; |
c96d145e | 4300 | } |
39507451 | 4301 | |
48f24c4d | 4302 | if (likely(sd)) { |
2d72376b | 4303 | schedstat_inc(sd, alb_count); |
39507451 | 4304 | |
43010659 PW |
4305 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4306 | sd, CPU_IDLE)) | |
48f24c4d IM |
4307 | schedstat_inc(sd, alb_pushed); |
4308 | else | |
4309 | schedstat_inc(sd, alb_failed); | |
4310 | } | |
1b12bbc7 | 4311 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4312 | } |
4313 | ||
46cb4b7c SS |
4314 | #ifdef CONFIG_NO_HZ |
4315 | static struct { | |
4316 | atomic_t load_balancer; | |
7d1e6a9b | 4317 | cpumask_var_t cpu_mask; |
f711f609 | 4318 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4319 | } nohz ____cacheline_aligned = { |
4320 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4321 | }; |
4322 | ||
f711f609 GS |
4323 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4324 | /** | |
4325 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4326 | * @cpu: The cpu whose lowest level of sched domain is to | |
4327 | * be returned. | |
4328 | * @flag: The flag to check for the lowest sched_domain | |
4329 | * for the given cpu. | |
4330 | * | |
4331 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4332 | */ | |
4333 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4334 | { | |
4335 | struct sched_domain *sd; | |
4336 | ||
4337 | for_each_domain(cpu, sd) | |
4338 | if (sd && (sd->flags & flag)) | |
4339 | break; | |
4340 | ||
4341 | return sd; | |
4342 | } | |
4343 | ||
4344 | /** | |
4345 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4346 | * @cpu: The cpu whose domains we're iterating over. | |
4347 | * @sd: variable holding the value of the power_savings_sd | |
4348 | * for cpu. | |
4349 | * @flag: The flag to filter the sched_domains to be iterated. | |
4350 | * | |
4351 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4352 | * set, starting from the lowest sched_domain to the highest. | |
4353 | */ | |
4354 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4355 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4356 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4357 | ||
4358 | /** | |
4359 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4360 | * @ilb_group: group to be checked for semi-idleness | |
4361 | * | |
4362 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4363 | * | |
4364 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4365 | * and atleast one non-idle CPU. This helper function checks if the given | |
4366 | * sched_group is semi-idle or not. | |
4367 | */ | |
4368 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4369 | { | |
4370 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4371 | sched_group_cpus(ilb_group)); | |
4372 | ||
4373 | /* | |
4374 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4375 | * and atleast one idle cpu. | |
4376 | */ | |
4377 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4378 | return 0; | |
4379 | ||
4380 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4381 | return 0; | |
4382 | ||
4383 | return 1; | |
4384 | } | |
4385 | /** | |
4386 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4387 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4388 | * | |
4389 | * Returns: Returns the id of the idle load balancer if it exists, | |
4390 | * Else, returns >= nr_cpu_ids. | |
4391 | * | |
4392 | * This algorithm picks the idle load balancer such that it belongs to a | |
4393 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4394 | * completely idle packages/cores just for the purpose of idle load balancing | |
4395 | * when there are other idle cpu's which are better suited for that job. | |
4396 | */ | |
4397 | static int find_new_ilb(int cpu) | |
4398 | { | |
4399 | struct sched_domain *sd; | |
4400 | struct sched_group *ilb_group; | |
4401 | ||
4402 | /* | |
4403 | * Have idle load balancer selection from semi-idle packages only | |
4404 | * when power-aware load balancing is enabled | |
4405 | */ | |
4406 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4407 | goto out_done; | |
4408 | ||
4409 | /* | |
4410 | * Optimize for the case when we have no idle CPUs or only one | |
4411 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4412 | */ | |
4413 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4414 | goto out_done; | |
4415 | ||
4416 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4417 | ilb_group = sd->groups; | |
4418 | ||
4419 | do { | |
4420 | if (is_semi_idle_group(ilb_group)) | |
4421 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4422 | ||
4423 | ilb_group = ilb_group->next; | |
4424 | ||
4425 | } while (ilb_group != sd->groups); | |
4426 | } | |
4427 | ||
4428 | out_done: | |
4429 | return cpumask_first(nohz.cpu_mask); | |
4430 | } | |
4431 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4432 | static inline int find_new_ilb(int call_cpu) | |
4433 | { | |
6e29ec57 | 4434 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4435 | } |
4436 | #endif | |
4437 | ||
7835b98b | 4438 | /* |
46cb4b7c SS |
4439 | * This routine will try to nominate the ilb (idle load balancing) |
4440 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4441 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4442 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4443 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4444 | * arrives... | |
4445 | * | |
4446 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4447 | * for idle load balancing. ilb owner will still be part of | |
4448 | * nohz.cpu_mask.. | |
7835b98b | 4449 | * |
46cb4b7c SS |
4450 | * While stopping the tick, this cpu will become the ilb owner if there |
4451 | * is no other owner. And will be the owner till that cpu becomes busy | |
4452 | * or if all cpus in the system stop their ticks at which point | |
4453 | * there is no need for ilb owner. | |
4454 | * | |
4455 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4456 | * next busy scheduler_tick() | |
4457 | */ | |
4458 | int select_nohz_load_balancer(int stop_tick) | |
4459 | { | |
4460 | int cpu = smp_processor_id(); | |
4461 | ||
4462 | if (stop_tick) { | |
46cb4b7c SS |
4463 | cpu_rq(cpu)->in_nohz_recently = 1; |
4464 | ||
483b4ee6 SS |
4465 | if (!cpu_active(cpu)) { |
4466 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4467 | return 0; | |
4468 | ||
4469 | /* | |
4470 | * If we are going offline and still the leader, | |
4471 | * give up! | |
4472 | */ | |
46cb4b7c SS |
4473 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4474 | BUG(); | |
483b4ee6 | 4475 | |
46cb4b7c SS |
4476 | return 0; |
4477 | } | |
4478 | ||
483b4ee6 SS |
4479 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4480 | ||
46cb4b7c | 4481 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4482 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4483 | if (atomic_read(&nohz.load_balancer) == cpu) |
4484 | atomic_set(&nohz.load_balancer, -1); | |
4485 | return 0; | |
4486 | } | |
4487 | ||
4488 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4489 | /* make me the ilb owner */ | |
4490 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4491 | return 1; | |
e790fb0b GS |
4492 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4493 | int new_ilb; | |
4494 | ||
4495 | if (!(sched_smt_power_savings || | |
4496 | sched_mc_power_savings)) | |
4497 | return 1; | |
4498 | /* | |
4499 | * Check to see if there is a more power-efficient | |
4500 | * ilb. | |
4501 | */ | |
4502 | new_ilb = find_new_ilb(cpu); | |
4503 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4504 | atomic_set(&nohz.load_balancer, -1); | |
4505 | resched_cpu(new_ilb); | |
4506 | return 0; | |
4507 | } | |
46cb4b7c | 4508 | return 1; |
e790fb0b | 4509 | } |
46cb4b7c | 4510 | } else { |
7d1e6a9b | 4511 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4512 | return 0; |
4513 | ||
7d1e6a9b | 4514 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4515 | |
4516 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4517 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4518 | BUG(); | |
4519 | } | |
4520 | return 0; | |
4521 | } | |
4522 | #endif | |
4523 | ||
4524 | static DEFINE_SPINLOCK(balancing); | |
4525 | ||
4526 | /* | |
7835b98b CL |
4527 | * It checks each scheduling domain to see if it is due to be balanced, |
4528 | * and initiates a balancing operation if so. | |
4529 | * | |
4530 | * Balancing parameters are set up in arch_init_sched_domains. | |
4531 | */ | |
a9957449 | 4532 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4533 | { |
46cb4b7c SS |
4534 | int balance = 1; |
4535 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4536 | unsigned long interval; |
4537 | struct sched_domain *sd; | |
46cb4b7c | 4538 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4539 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4540 | int update_next_balance = 0; |
d07355f5 | 4541 | int need_serialize; |
1da177e4 | 4542 | |
46cb4b7c | 4543 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4544 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4545 | continue; | |
4546 | ||
4547 | interval = sd->balance_interval; | |
d15bcfdb | 4548 | if (idle != CPU_IDLE) |
1da177e4 LT |
4549 | interval *= sd->busy_factor; |
4550 | ||
4551 | /* scale ms to jiffies */ | |
4552 | interval = msecs_to_jiffies(interval); | |
4553 | if (unlikely(!interval)) | |
4554 | interval = 1; | |
dd41f596 IM |
4555 | if (interval > HZ*NR_CPUS/10) |
4556 | interval = HZ*NR_CPUS/10; | |
4557 | ||
d07355f5 | 4558 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4559 | |
d07355f5 | 4560 | if (need_serialize) { |
08c183f3 CL |
4561 | if (!spin_trylock(&balancing)) |
4562 | goto out; | |
4563 | } | |
4564 | ||
c9819f45 | 4565 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4566 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4567 | /* |
4568 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4569 | * longer idle, or one of our SMT siblings is |
4570 | * not idle. | |
4571 | */ | |
d15bcfdb | 4572 | idle = CPU_NOT_IDLE; |
1da177e4 | 4573 | } |
1bd77f2d | 4574 | sd->last_balance = jiffies; |
1da177e4 | 4575 | } |
d07355f5 | 4576 | if (need_serialize) |
08c183f3 CL |
4577 | spin_unlock(&balancing); |
4578 | out: | |
f549da84 | 4579 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4580 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4581 | update_next_balance = 1; |
4582 | } | |
783609c6 SS |
4583 | |
4584 | /* | |
4585 | * Stop the load balance at this level. There is another | |
4586 | * CPU in our sched group which is doing load balancing more | |
4587 | * actively. | |
4588 | */ | |
4589 | if (!balance) | |
4590 | break; | |
1da177e4 | 4591 | } |
f549da84 SS |
4592 | |
4593 | /* | |
4594 | * next_balance will be updated only when there is a need. | |
4595 | * When the cpu is attached to null domain for ex, it will not be | |
4596 | * updated. | |
4597 | */ | |
4598 | if (likely(update_next_balance)) | |
4599 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4600 | } |
4601 | ||
4602 | /* | |
4603 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4604 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4605 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4606 | */ | |
4607 | static void run_rebalance_domains(struct softirq_action *h) | |
4608 | { | |
dd41f596 IM |
4609 | int this_cpu = smp_processor_id(); |
4610 | struct rq *this_rq = cpu_rq(this_cpu); | |
4611 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4612 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4613 | |
dd41f596 | 4614 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4615 | |
4616 | #ifdef CONFIG_NO_HZ | |
4617 | /* | |
4618 | * If this cpu is the owner for idle load balancing, then do the | |
4619 | * balancing on behalf of the other idle cpus whose ticks are | |
4620 | * stopped. | |
4621 | */ | |
dd41f596 IM |
4622 | if (this_rq->idle_at_tick && |
4623 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4624 | struct rq *rq; |
4625 | int balance_cpu; | |
4626 | ||
7d1e6a9b RR |
4627 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4628 | if (balance_cpu == this_cpu) | |
4629 | continue; | |
4630 | ||
46cb4b7c SS |
4631 | /* |
4632 | * If this cpu gets work to do, stop the load balancing | |
4633 | * work being done for other cpus. Next load | |
4634 | * balancing owner will pick it up. | |
4635 | */ | |
4636 | if (need_resched()) | |
4637 | break; | |
4638 | ||
de0cf899 | 4639 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4640 | |
4641 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4642 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4643 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4644 | } |
4645 | } | |
4646 | #endif | |
4647 | } | |
4648 | ||
8a0be9ef FW |
4649 | static inline int on_null_domain(int cpu) |
4650 | { | |
4651 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4652 | } | |
4653 | ||
46cb4b7c SS |
4654 | /* |
4655 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4656 | * | |
4657 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4658 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4659 | * if the whole system is idle. | |
4660 | */ | |
dd41f596 | 4661 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4662 | { |
46cb4b7c SS |
4663 | #ifdef CONFIG_NO_HZ |
4664 | /* | |
4665 | * If we were in the nohz mode recently and busy at the current | |
4666 | * scheduler tick, then check if we need to nominate new idle | |
4667 | * load balancer. | |
4668 | */ | |
4669 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4670 | rq->in_nohz_recently = 0; | |
4671 | ||
4672 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4673 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4674 | atomic_set(&nohz.load_balancer, -1); |
4675 | } | |
4676 | ||
4677 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4678 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4679 | |
434d53b0 | 4680 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4681 | resched_cpu(ilb); |
4682 | } | |
4683 | } | |
4684 | ||
4685 | /* | |
4686 | * If this cpu is idle and doing idle load balancing for all the | |
4687 | * cpus with ticks stopped, is it time for that to stop? | |
4688 | */ | |
4689 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4690 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4691 | resched_cpu(cpu); |
4692 | return; | |
4693 | } | |
4694 | ||
4695 | /* | |
4696 | * If this cpu is idle and the idle load balancing is done by | |
4697 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4698 | */ | |
4699 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4700 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4701 | return; |
4702 | #endif | |
8a0be9ef FW |
4703 | /* Don't need to rebalance while attached to NULL domain */ |
4704 | if (time_after_eq(jiffies, rq->next_balance) && | |
4705 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4706 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4707 | } |
dd41f596 IM |
4708 | |
4709 | #else /* CONFIG_SMP */ | |
4710 | ||
1da177e4 LT |
4711 | /* |
4712 | * on UP we do not need to balance between CPUs: | |
4713 | */ | |
70b97a7f | 4714 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4715 | { |
4716 | } | |
dd41f596 | 4717 | |
1da177e4 LT |
4718 | #endif |
4719 | ||
1da177e4 LT |
4720 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4721 | ||
4722 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4723 | ||
4724 | /* | |
c5f8d995 | 4725 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4726 | * @p in case that task is currently running. |
c5f8d995 HS |
4727 | * |
4728 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4729 | */ |
c5f8d995 HS |
4730 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4731 | { | |
4732 | u64 ns = 0; | |
4733 | ||
4734 | if (task_current(rq, p)) { | |
4735 | update_rq_clock(rq); | |
4736 | ns = rq->clock - p->se.exec_start; | |
4737 | if ((s64)ns < 0) | |
4738 | ns = 0; | |
4739 | } | |
4740 | ||
4741 | return ns; | |
4742 | } | |
4743 | ||
bb34d92f | 4744 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4745 | { |
1da177e4 | 4746 | unsigned long flags; |
41b86e9c | 4747 | struct rq *rq; |
bb34d92f | 4748 | u64 ns = 0; |
48f24c4d | 4749 | |
41b86e9c | 4750 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4751 | ns = do_task_delta_exec(p, rq); |
4752 | task_rq_unlock(rq, &flags); | |
1508487e | 4753 | |
c5f8d995 HS |
4754 | return ns; |
4755 | } | |
f06febc9 | 4756 | |
c5f8d995 HS |
4757 | /* |
4758 | * Return accounted runtime for the task. | |
4759 | * In case the task is currently running, return the runtime plus current's | |
4760 | * pending runtime that have not been accounted yet. | |
4761 | */ | |
4762 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4763 | { | |
4764 | unsigned long flags; | |
4765 | struct rq *rq; | |
4766 | u64 ns = 0; | |
4767 | ||
4768 | rq = task_rq_lock(p, &flags); | |
4769 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4770 | task_rq_unlock(rq, &flags); | |
4771 | ||
4772 | return ns; | |
4773 | } | |
48f24c4d | 4774 | |
c5f8d995 HS |
4775 | /* |
4776 | * Return sum_exec_runtime for the thread group. | |
4777 | * In case the task is currently running, return the sum plus current's | |
4778 | * pending runtime that have not been accounted yet. | |
4779 | * | |
4780 | * Note that the thread group might have other running tasks as well, | |
4781 | * so the return value not includes other pending runtime that other | |
4782 | * running tasks might have. | |
4783 | */ | |
4784 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4785 | { | |
4786 | struct task_cputime totals; | |
4787 | unsigned long flags; | |
4788 | struct rq *rq; | |
4789 | u64 ns; | |
4790 | ||
4791 | rq = task_rq_lock(p, &flags); | |
4792 | thread_group_cputime(p, &totals); | |
4793 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4794 | task_rq_unlock(rq, &flags); |
48f24c4d | 4795 | |
1da177e4 LT |
4796 | return ns; |
4797 | } | |
4798 | ||
1da177e4 LT |
4799 | /* |
4800 | * Account user cpu time to a process. | |
4801 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4802 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4803 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4804 | */ |
457533a7 MS |
4805 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4806 | cputime_t cputime_scaled) | |
1da177e4 LT |
4807 | { |
4808 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4809 | cputime64_t tmp; | |
4810 | ||
457533a7 | 4811 | /* Add user time to process. */ |
1da177e4 | 4812 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4813 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4814 | account_group_user_time(p, cputime); |
1da177e4 LT |
4815 | |
4816 | /* Add user time to cpustat. */ | |
4817 | tmp = cputime_to_cputime64(cputime); | |
4818 | if (TASK_NICE(p) > 0) | |
4819 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4820 | else | |
4821 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4822 | |
4823 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4824 | /* Account for user time used */ |
4825 | acct_update_integrals(p); | |
1da177e4 LT |
4826 | } |
4827 | ||
94886b84 LV |
4828 | /* |
4829 | * Account guest cpu time to a process. | |
4830 | * @p: the process that the cpu time gets accounted to | |
4831 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4832 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4833 | */ |
457533a7 MS |
4834 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4835 | cputime_t cputime_scaled) | |
94886b84 LV |
4836 | { |
4837 | cputime64_t tmp; | |
4838 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4839 | ||
4840 | tmp = cputime_to_cputime64(cputime); | |
4841 | ||
457533a7 | 4842 | /* Add guest time to process. */ |
94886b84 | 4843 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4844 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4845 | account_group_user_time(p, cputime); |
94886b84 LV |
4846 | p->gtime = cputime_add(p->gtime, cputime); |
4847 | ||
457533a7 | 4848 | /* Add guest time to cpustat. */ |
94886b84 LV |
4849 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4850 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4851 | } | |
4852 | ||
1da177e4 LT |
4853 | /* |
4854 | * Account system cpu time to a process. | |
4855 | * @p: the process that the cpu time gets accounted to | |
4856 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4857 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4858 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4859 | */ |
4860 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4861 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4862 | { |
4863 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4864 | cputime64_t tmp; |
4865 | ||
983ed7a6 | 4866 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4867 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4868 | return; |
4869 | } | |
94886b84 | 4870 | |
457533a7 | 4871 | /* Add system time to process. */ |
1da177e4 | 4872 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4873 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4874 | account_group_system_time(p, cputime); |
1da177e4 LT |
4875 | |
4876 | /* Add system time to cpustat. */ | |
4877 | tmp = cputime_to_cputime64(cputime); | |
4878 | if (hardirq_count() - hardirq_offset) | |
4879 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4880 | else if (softirq_count()) | |
4881 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4882 | else |
79741dd3 MS |
4883 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4884 | ||
ef12fefa BR |
4885 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4886 | ||
1da177e4 LT |
4887 | /* Account for system time used */ |
4888 | acct_update_integrals(p); | |
1da177e4 LT |
4889 | } |
4890 | ||
c66f08be | 4891 | /* |
1da177e4 | 4892 | * Account for involuntary wait time. |
1da177e4 | 4893 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4894 | */ |
79741dd3 | 4895 | void account_steal_time(cputime_t cputime) |
c66f08be | 4896 | { |
79741dd3 MS |
4897 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4898 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4899 | ||
4900 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4901 | } |
4902 | ||
1da177e4 | 4903 | /* |
79741dd3 MS |
4904 | * Account for idle time. |
4905 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4906 | */ |
79741dd3 | 4907 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4908 | { |
4909 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4910 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4911 | struct rq *rq = this_rq(); |
1da177e4 | 4912 | |
79741dd3 MS |
4913 | if (atomic_read(&rq->nr_iowait) > 0) |
4914 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4915 | else | |
4916 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4917 | } |
4918 | ||
79741dd3 MS |
4919 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4920 | ||
4921 | /* | |
4922 | * Account a single tick of cpu time. | |
4923 | * @p: the process that the cpu time gets accounted to | |
4924 | * @user_tick: indicates if the tick is a user or a system tick | |
4925 | */ | |
4926 | void account_process_tick(struct task_struct *p, int user_tick) | |
4927 | { | |
4928 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4929 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4930 | struct rq *rq = this_rq(); | |
4931 | ||
4932 | if (user_tick) | |
4933 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 4934 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
4935 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
4936 | one_jiffy_scaled); | |
4937 | else | |
4938 | account_idle_time(one_jiffy); | |
4939 | } | |
4940 | ||
4941 | /* | |
4942 | * Account multiple ticks of steal time. | |
4943 | * @p: the process from which the cpu time has been stolen | |
4944 | * @ticks: number of stolen ticks | |
4945 | */ | |
4946 | void account_steal_ticks(unsigned long ticks) | |
4947 | { | |
4948 | account_steal_time(jiffies_to_cputime(ticks)); | |
4949 | } | |
4950 | ||
4951 | /* | |
4952 | * Account multiple ticks of idle time. | |
4953 | * @ticks: number of stolen ticks | |
4954 | */ | |
4955 | void account_idle_ticks(unsigned long ticks) | |
4956 | { | |
4957 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4958 | } |
4959 | ||
79741dd3 MS |
4960 | #endif |
4961 | ||
49048622 BS |
4962 | /* |
4963 | * Use precise platform statistics if available: | |
4964 | */ | |
4965 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4966 | cputime_t task_utime(struct task_struct *p) | |
4967 | { | |
4968 | return p->utime; | |
4969 | } | |
4970 | ||
4971 | cputime_t task_stime(struct task_struct *p) | |
4972 | { | |
4973 | return p->stime; | |
4974 | } | |
4975 | #else | |
4976 | cputime_t task_utime(struct task_struct *p) | |
4977 | { | |
4978 | clock_t utime = cputime_to_clock_t(p->utime), | |
4979 | total = utime + cputime_to_clock_t(p->stime); | |
4980 | u64 temp; | |
4981 | ||
4982 | /* | |
4983 | * Use CFS's precise accounting: | |
4984 | */ | |
4985 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4986 | ||
4987 | if (total) { | |
4988 | temp *= utime; | |
4989 | do_div(temp, total); | |
4990 | } | |
4991 | utime = (clock_t)temp; | |
4992 | ||
4993 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4994 | return p->prev_utime; | |
4995 | } | |
4996 | ||
4997 | cputime_t task_stime(struct task_struct *p) | |
4998 | { | |
4999 | clock_t stime; | |
5000 | ||
5001 | /* | |
5002 | * Use CFS's precise accounting. (we subtract utime from | |
5003 | * the total, to make sure the total observed by userspace | |
5004 | * grows monotonically - apps rely on that): | |
5005 | */ | |
5006 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5007 | cputime_to_clock_t(task_utime(p)); | |
5008 | ||
5009 | if (stime >= 0) | |
5010 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5011 | ||
5012 | return p->prev_stime; | |
5013 | } | |
5014 | #endif | |
5015 | ||
5016 | inline cputime_t task_gtime(struct task_struct *p) | |
5017 | { | |
5018 | return p->gtime; | |
5019 | } | |
5020 | ||
7835b98b CL |
5021 | /* |
5022 | * This function gets called by the timer code, with HZ frequency. | |
5023 | * We call it with interrupts disabled. | |
5024 | * | |
5025 | * It also gets called by the fork code, when changing the parent's | |
5026 | * timeslices. | |
5027 | */ | |
5028 | void scheduler_tick(void) | |
5029 | { | |
7835b98b CL |
5030 | int cpu = smp_processor_id(); |
5031 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5032 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5033 | |
5034 | sched_clock_tick(); | |
dd41f596 IM |
5035 | |
5036 | spin_lock(&rq->lock); | |
3e51f33f | 5037 | update_rq_clock(rq); |
f1a438d8 | 5038 | update_cpu_load(rq); |
fa85ae24 | 5039 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5040 | spin_unlock(&rq->lock); |
7835b98b | 5041 | |
e418e1c2 | 5042 | #ifdef CONFIG_SMP |
dd41f596 IM |
5043 | rq->idle_at_tick = idle_cpu(cpu); |
5044 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5045 | #endif |
1da177e4 LT |
5046 | } |
5047 | ||
132380a0 | 5048 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5049 | { |
5050 | if (in_lock_functions(addr)) { | |
5051 | addr = CALLER_ADDR2; | |
5052 | if (in_lock_functions(addr)) | |
5053 | addr = CALLER_ADDR3; | |
5054 | } | |
5055 | return addr; | |
5056 | } | |
1da177e4 | 5057 | |
7e49fcce SR |
5058 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5059 | defined(CONFIG_PREEMPT_TRACER)) | |
5060 | ||
43627582 | 5061 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5062 | { |
6cd8a4bb | 5063 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5064 | /* |
5065 | * Underflow? | |
5066 | */ | |
9a11b49a IM |
5067 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5068 | return; | |
6cd8a4bb | 5069 | #endif |
1da177e4 | 5070 | preempt_count() += val; |
6cd8a4bb | 5071 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5072 | /* |
5073 | * Spinlock count overflowing soon? | |
5074 | */ | |
33859f7f MOS |
5075 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5076 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5077 | #endif |
5078 | if (preempt_count() == val) | |
5079 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5080 | } |
5081 | EXPORT_SYMBOL(add_preempt_count); | |
5082 | ||
43627582 | 5083 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5084 | { |
6cd8a4bb | 5085 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5086 | /* |
5087 | * Underflow? | |
5088 | */ | |
01e3eb82 | 5089 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5090 | return; |
1da177e4 LT |
5091 | /* |
5092 | * Is the spinlock portion underflowing? | |
5093 | */ | |
9a11b49a IM |
5094 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5095 | !(preempt_count() & PREEMPT_MASK))) | |
5096 | return; | |
6cd8a4bb | 5097 | #endif |
9a11b49a | 5098 | |
6cd8a4bb SR |
5099 | if (preempt_count() == val) |
5100 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5101 | preempt_count() -= val; |
5102 | } | |
5103 | EXPORT_SYMBOL(sub_preempt_count); | |
5104 | ||
5105 | #endif | |
5106 | ||
5107 | /* | |
dd41f596 | 5108 | * Print scheduling while atomic bug: |
1da177e4 | 5109 | */ |
dd41f596 | 5110 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5111 | { |
838225b4 SS |
5112 | struct pt_regs *regs = get_irq_regs(); |
5113 | ||
5114 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5115 | prev->comm, prev->pid, preempt_count()); | |
5116 | ||
dd41f596 | 5117 | debug_show_held_locks(prev); |
e21f5b15 | 5118 | print_modules(); |
dd41f596 IM |
5119 | if (irqs_disabled()) |
5120 | print_irqtrace_events(prev); | |
838225b4 SS |
5121 | |
5122 | if (regs) | |
5123 | show_regs(regs); | |
5124 | else | |
5125 | dump_stack(); | |
dd41f596 | 5126 | } |
1da177e4 | 5127 | |
dd41f596 IM |
5128 | /* |
5129 | * Various schedule()-time debugging checks and statistics: | |
5130 | */ | |
5131 | static inline void schedule_debug(struct task_struct *prev) | |
5132 | { | |
1da177e4 | 5133 | /* |
41a2d6cf | 5134 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5135 | * schedule() atomically, we ignore that path for now. |
5136 | * Otherwise, whine if we are scheduling when we should not be. | |
5137 | */ | |
3f33a7ce | 5138 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5139 | __schedule_bug(prev); |
5140 | ||
1da177e4 LT |
5141 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5142 | ||
2d72376b | 5143 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5144 | #ifdef CONFIG_SCHEDSTATS |
5145 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5146 | schedstat_inc(this_rq(), bkl_count); |
5147 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5148 | } |
5149 | #endif | |
dd41f596 IM |
5150 | } |
5151 | ||
df1c99d4 MG |
5152 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5153 | { | |
5154 | if (prev->state == TASK_RUNNING) { | |
5155 | u64 runtime = prev->se.sum_exec_runtime; | |
5156 | ||
5157 | runtime -= prev->se.prev_sum_exec_runtime; | |
5158 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5159 | ||
5160 | /* | |
5161 | * In order to avoid avg_overlap growing stale when we are | |
5162 | * indeed overlapping and hence not getting put to sleep, grow | |
5163 | * the avg_overlap on preemption. | |
5164 | * | |
5165 | * We use the average preemption runtime because that | |
5166 | * correlates to the amount of cache footprint a task can | |
5167 | * build up. | |
5168 | */ | |
5169 | update_avg(&prev->se.avg_overlap, runtime); | |
5170 | } | |
5171 | prev->sched_class->put_prev_task(rq, prev); | |
5172 | } | |
5173 | ||
dd41f596 IM |
5174 | /* |
5175 | * Pick up the highest-prio task: | |
5176 | */ | |
5177 | static inline struct task_struct * | |
b67802ea | 5178 | pick_next_task(struct rq *rq) |
dd41f596 | 5179 | { |
5522d5d5 | 5180 | const struct sched_class *class; |
dd41f596 | 5181 | struct task_struct *p; |
1da177e4 LT |
5182 | |
5183 | /* | |
dd41f596 IM |
5184 | * Optimization: we know that if all tasks are in |
5185 | * the fair class we can call that function directly: | |
1da177e4 | 5186 | */ |
dd41f596 | 5187 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5188 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5189 | if (likely(p)) |
5190 | return p; | |
1da177e4 LT |
5191 | } |
5192 | ||
dd41f596 IM |
5193 | class = sched_class_highest; |
5194 | for ( ; ; ) { | |
fb8d4724 | 5195 | p = class->pick_next_task(rq); |
dd41f596 IM |
5196 | if (p) |
5197 | return p; | |
5198 | /* | |
5199 | * Will never be NULL as the idle class always | |
5200 | * returns a non-NULL p: | |
5201 | */ | |
5202 | class = class->next; | |
5203 | } | |
5204 | } | |
1da177e4 | 5205 | |
dd41f596 IM |
5206 | /* |
5207 | * schedule() is the main scheduler function. | |
5208 | */ | |
ff743345 | 5209 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5210 | { |
5211 | struct task_struct *prev, *next; | |
67ca7bde | 5212 | unsigned long *switch_count; |
dd41f596 | 5213 | struct rq *rq; |
31656519 | 5214 | int cpu; |
dd41f596 | 5215 | |
ff743345 PZ |
5216 | need_resched: |
5217 | preempt_disable(); | |
dd41f596 IM |
5218 | cpu = smp_processor_id(); |
5219 | rq = cpu_rq(cpu); | |
5220 | rcu_qsctr_inc(cpu); | |
5221 | prev = rq->curr; | |
5222 | switch_count = &prev->nivcsw; | |
5223 | ||
5224 | release_kernel_lock(prev); | |
5225 | need_resched_nonpreemptible: | |
5226 | ||
5227 | schedule_debug(prev); | |
1da177e4 | 5228 | |
31656519 | 5229 | if (sched_feat(HRTICK)) |
f333fdc9 | 5230 | hrtick_clear(rq); |
8f4d37ec | 5231 | |
8cd162ce | 5232 | spin_lock_irq(&rq->lock); |
3e51f33f | 5233 | update_rq_clock(rq); |
1e819950 | 5234 | clear_tsk_need_resched(prev); |
1da177e4 | 5235 | |
1da177e4 | 5236 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5237 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5238 | prev->state = TASK_RUNNING; |
16882c1e | 5239 | else |
2e1cb74a | 5240 | deactivate_task(rq, prev, 1); |
dd41f596 | 5241 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5242 | } |
5243 | ||
9a897c5a SR |
5244 | #ifdef CONFIG_SMP |
5245 | if (prev->sched_class->pre_schedule) | |
5246 | prev->sched_class->pre_schedule(rq, prev); | |
5247 | #endif | |
f65eda4f | 5248 | |
dd41f596 | 5249 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5250 | idle_balance(cpu, rq); |
1da177e4 | 5251 | |
df1c99d4 | 5252 | put_prev_task(rq, prev); |
b67802ea | 5253 | next = pick_next_task(rq); |
1da177e4 | 5254 | |
1da177e4 | 5255 | if (likely(prev != next)) { |
673a90a1 DS |
5256 | sched_info_switch(prev, next); |
5257 | ||
1da177e4 LT |
5258 | rq->nr_switches++; |
5259 | rq->curr = next; | |
5260 | ++*switch_count; | |
5261 | ||
dd41f596 | 5262 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5263 | /* |
5264 | * the context switch might have flipped the stack from under | |
5265 | * us, hence refresh the local variables. | |
5266 | */ | |
5267 | cpu = smp_processor_id(); | |
5268 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5269 | } else |
5270 | spin_unlock_irq(&rq->lock); | |
5271 | ||
8f4d37ec | 5272 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5273 | goto need_resched_nonpreemptible; |
8f4d37ec | 5274 | |
1da177e4 | 5275 | preempt_enable_no_resched(); |
ff743345 | 5276 | if (need_resched()) |
1da177e4 LT |
5277 | goto need_resched; |
5278 | } | |
1da177e4 LT |
5279 | EXPORT_SYMBOL(schedule); |
5280 | ||
0d66bf6d PZ |
5281 | #ifdef CONFIG_SMP |
5282 | /* | |
5283 | * Look out! "owner" is an entirely speculative pointer | |
5284 | * access and not reliable. | |
5285 | */ | |
5286 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5287 | { | |
5288 | unsigned int cpu; | |
5289 | struct rq *rq; | |
5290 | ||
5291 | if (!sched_feat(OWNER_SPIN)) | |
5292 | return 0; | |
5293 | ||
5294 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5295 | /* | |
5296 | * Need to access the cpu field knowing that | |
5297 | * DEBUG_PAGEALLOC could have unmapped it if | |
5298 | * the mutex owner just released it and exited. | |
5299 | */ | |
5300 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5301 | goto out; | |
5302 | #else | |
5303 | cpu = owner->cpu; | |
5304 | #endif | |
5305 | ||
5306 | /* | |
5307 | * Even if the access succeeded (likely case), | |
5308 | * the cpu field may no longer be valid. | |
5309 | */ | |
5310 | if (cpu >= nr_cpumask_bits) | |
5311 | goto out; | |
5312 | ||
5313 | /* | |
5314 | * We need to validate that we can do a | |
5315 | * get_cpu() and that we have the percpu area. | |
5316 | */ | |
5317 | if (!cpu_online(cpu)) | |
5318 | goto out; | |
5319 | ||
5320 | rq = cpu_rq(cpu); | |
5321 | ||
5322 | for (;;) { | |
5323 | /* | |
5324 | * Owner changed, break to re-assess state. | |
5325 | */ | |
5326 | if (lock->owner != owner) | |
5327 | break; | |
5328 | ||
5329 | /* | |
5330 | * Is that owner really running on that cpu? | |
5331 | */ | |
5332 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5333 | return 0; | |
5334 | ||
5335 | cpu_relax(); | |
5336 | } | |
5337 | out: | |
5338 | return 1; | |
5339 | } | |
5340 | #endif | |
5341 | ||
1da177e4 LT |
5342 | #ifdef CONFIG_PREEMPT |
5343 | /* | |
2ed6e34f | 5344 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5345 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5346 | * occur there and call schedule directly. |
5347 | */ | |
5348 | asmlinkage void __sched preempt_schedule(void) | |
5349 | { | |
5350 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5351 | |
1da177e4 LT |
5352 | /* |
5353 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5354 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5355 | */ |
beed33a8 | 5356 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5357 | return; |
5358 | ||
3a5c359a AK |
5359 | do { |
5360 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5361 | schedule(); |
3a5c359a | 5362 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5363 | |
3a5c359a AK |
5364 | /* |
5365 | * Check again in case we missed a preemption opportunity | |
5366 | * between schedule and now. | |
5367 | */ | |
5368 | barrier(); | |
5ed0cec0 | 5369 | } while (need_resched()); |
1da177e4 | 5370 | } |
1da177e4 LT |
5371 | EXPORT_SYMBOL(preempt_schedule); |
5372 | ||
5373 | /* | |
2ed6e34f | 5374 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5375 | * off of irq context. |
5376 | * Note, that this is called and return with irqs disabled. This will | |
5377 | * protect us against recursive calling from irq. | |
5378 | */ | |
5379 | asmlinkage void __sched preempt_schedule_irq(void) | |
5380 | { | |
5381 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5382 | |
2ed6e34f | 5383 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5384 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5385 | ||
3a5c359a AK |
5386 | do { |
5387 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5388 | local_irq_enable(); |
5389 | schedule(); | |
5390 | local_irq_disable(); | |
3a5c359a | 5391 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5392 | |
3a5c359a AK |
5393 | /* |
5394 | * Check again in case we missed a preemption opportunity | |
5395 | * between schedule and now. | |
5396 | */ | |
5397 | barrier(); | |
5ed0cec0 | 5398 | } while (need_resched()); |
1da177e4 LT |
5399 | } |
5400 | ||
5401 | #endif /* CONFIG_PREEMPT */ | |
5402 | ||
95cdf3b7 IM |
5403 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5404 | void *key) | |
1da177e4 | 5405 | { |
48f24c4d | 5406 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5407 | } |
1da177e4 LT |
5408 | EXPORT_SYMBOL(default_wake_function); |
5409 | ||
5410 | /* | |
41a2d6cf IM |
5411 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5412 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5413 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5414 | * | |
5415 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5416 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5417 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5418 | */ | |
78ddb08f | 5419 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5420 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5421 | { |
2e45874c | 5422 | wait_queue_t *curr, *next; |
1da177e4 | 5423 | |
2e45874c | 5424 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5425 | unsigned flags = curr->flags; |
5426 | ||
1da177e4 | 5427 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5428 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5429 | break; |
5430 | } | |
5431 | } | |
5432 | ||
5433 | /** | |
5434 | * __wake_up - wake up threads blocked on a waitqueue. | |
5435 | * @q: the waitqueue | |
5436 | * @mode: which threads | |
5437 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5438 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5439 | * |
5440 | * It may be assumed that this function implies a write memory barrier before | |
5441 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5442 | */ |
7ad5b3a5 | 5443 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5444 | int nr_exclusive, void *key) |
1da177e4 LT |
5445 | { |
5446 | unsigned long flags; | |
5447 | ||
5448 | spin_lock_irqsave(&q->lock, flags); | |
5449 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5450 | spin_unlock_irqrestore(&q->lock, flags); | |
5451 | } | |
1da177e4 LT |
5452 | EXPORT_SYMBOL(__wake_up); |
5453 | ||
5454 | /* | |
5455 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5456 | */ | |
7ad5b3a5 | 5457 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5458 | { |
5459 | __wake_up_common(q, mode, 1, 0, NULL); | |
5460 | } | |
5461 | ||
4ede816a DL |
5462 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5463 | { | |
5464 | __wake_up_common(q, mode, 1, 0, key); | |
5465 | } | |
5466 | ||
1da177e4 | 5467 | /** |
4ede816a | 5468 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5469 | * @q: the waitqueue |
5470 | * @mode: which threads | |
5471 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5472 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5473 | * |
5474 | * The sync wakeup differs that the waker knows that it will schedule | |
5475 | * away soon, so while the target thread will be woken up, it will not | |
5476 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5477 | * with each other. This can prevent needless bouncing between CPUs. | |
5478 | * | |
5479 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5480 | * |
5481 | * It may be assumed that this function implies a write memory barrier before | |
5482 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5483 | */ |
4ede816a DL |
5484 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5485 | int nr_exclusive, void *key) | |
1da177e4 LT |
5486 | { |
5487 | unsigned long flags; | |
5488 | int sync = 1; | |
5489 | ||
5490 | if (unlikely(!q)) | |
5491 | return; | |
5492 | ||
5493 | if (unlikely(!nr_exclusive)) | |
5494 | sync = 0; | |
5495 | ||
5496 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5497 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5498 | spin_unlock_irqrestore(&q->lock, flags); |
5499 | } | |
4ede816a DL |
5500 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5501 | ||
5502 | /* | |
5503 | * __wake_up_sync - see __wake_up_sync_key() | |
5504 | */ | |
5505 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5506 | { | |
5507 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5508 | } | |
1da177e4 LT |
5509 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5510 | ||
65eb3dc6 KD |
5511 | /** |
5512 | * complete: - signals a single thread waiting on this completion | |
5513 | * @x: holds the state of this particular completion | |
5514 | * | |
5515 | * This will wake up a single thread waiting on this completion. Threads will be | |
5516 | * awakened in the same order in which they were queued. | |
5517 | * | |
5518 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5519 | * |
5520 | * It may be assumed that this function implies a write memory barrier before | |
5521 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5522 | */ |
b15136e9 | 5523 | void complete(struct completion *x) |
1da177e4 LT |
5524 | { |
5525 | unsigned long flags; | |
5526 | ||
5527 | spin_lock_irqsave(&x->wait.lock, flags); | |
5528 | x->done++; | |
d9514f6c | 5529 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5530 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5531 | } | |
5532 | EXPORT_SYMBOL(complete); | |
5533 | ||
65eb3dc6 KD |
5534 | /** |
5535 | * complete_all: - signals all threads waiting on this completion | |
5536 | * @x: holds the state of this particular completion | |
5537 | * | |
5538 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5539 | * |
5540 | * It may be assumed that this function implies a write memory barrier before | |
5541 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5542 | */ |
b15136e9 | 5543 | void complete_all(struct completion *x) |
1da177e4 LT |
5544 | { |
5545 | unsigned long flags; | |
5546 | ||
5547 | spin_lock_irqsave(&x->wait.lock, flags); | |
5548 | x->done += UINT_MAX/2; | |
d9514f6c | 5549 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5550 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5551 | } | |
5552 | EXPORT_SYMBOL(complete_all); | |
5553 | ||
8cbbe86d AK |
5554 | static inline long __sched |
5555 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5556 | { |
1da177e4 LT |
5557 | if (!x->done) { |
5558 | DECLARE_WAITQUEUE(wait, current); | |
5559 | ||
5560 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5561 | __add_wait_queue_tail(&x->wait, &wait); | |
5562 | do { | |
94d3d824 | 5563 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5564 | timeout = -ERESTARTSYS; |
5565 | break; | |
8cbbe86d AK |
5566 | } |
5567 | __set_current_state(state); | |
1da177e4 LT |
5568 | spin_unlock_irq(&x->wait.lock); |
5569 | timeout = schedule_timeout(timeout); | |
5570 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5571 | } while (!x->done && timeout); |
1da177e4 | 5572 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5573 | if (!x->done) |
5574 | return timeout; | |
1da177e4 LT |
5575 | } |
5576 | x->done--; | |
ea71a546 | 5577 | return timeout ?: 1; |
1da177e4 | 5578 | } |
1da177e4 | 5579 | |
8cbbe86d AK |
5580 | static long __sched |
5581 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5582 | { |
1da177e4 LT |
5583 | might_sleep(); |
5584 | ||
5585 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5586 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5587 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5588 | return timeout; |
5589 | } | |
1da177e4 | 5590 | |
65eb3dc6 KD |
5591 | /** |
5592 | * wait_for_completion: - waits for completion of a task | |
5593 | * @x: holds the state of this particular completion | |
5594 | * | |
5595 | * This waits to be signaled for completion of a specific task. It is NOT | |
5596 | * interruptible and there is no timeout. | |
5597 | * | |
5598 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5599 | * and interrupt capability. Also see complete(). | |
5600 | */ | |
b15136e9 | 5601 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5602 | { |
5603 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5604 | } |
8cbbe86d | 5605 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5606 | |
65eb3dc6 KD |
5607 | /** |
5608 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5609 | * @x: holds the state of this particular completion | |
5610 | * @timeout: timeout value in jiffies | |
5611 | * | |
5612 | * This waits for either a completion of a specific task to be signaled or for a | |
5613 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5614 | * interruptible. | |
5615 | */ | |
b15136e9 | 5616 | unsigned long __sched |
8cbbe86d | 5617 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5618 | { |
8cbbe86d | 5619 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5620 | } |
8cbbe86d | 5621 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5622 | |
65eb3dc6 KD |
5623 | /** |
5624 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5625 | * @x: holds the state of this particular completion | |
5626 | * | |
5627 | * This waits for completion of a specific task to be signaled. It is | |
5628 | * interruptible. | |
5629 | */ | |
8cbbe86d | 5630 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5631 | { |
51e97990 AK |
5632 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5633 | if (t == -ERESTARTSYS) | |
5634 | return t; | |
5635 | return 0; | |
0fec171c | 5636 | } |
8cbbe86d | 5637 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5638 | |
65eb3dc6 KD |
5639 | /** |
5640 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5641 | * @x: holds the state of this particular completion | |
5642 | * @timeout: timeout value in jiffies | |
5643 | * | |
5644 | * This waits for either a completion of a specific task to be signaled or for a | |
5645 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5646 | */ | |
b15136e9 | 5647 | unsigned long __sched |
8cbbe86d AK |
5648 | wait_for_completion_interruptible_timeout(struct completion *x, |
5649 | unsigned long timeout) | |
0fec171c | 5650 | { |
8cbbe86d | 5651 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5652 | } |
8cbbe86d | 5653 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5654 | |
65eb3dc6 KD |
5655 | /** |
5656 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5657 | * @x: holds the state of this particular completion | |
5658 | * | |
5659 | * This waits to be signaled for completion of a specific task. It can be | |
5660 | * interrupted by a kill signal. | |
5661 | */ | |
009e577e MW |
5662 | int __sched wait_for_completion_killable(struct completion *x) |
5663 | { | |
5664 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5665 | if (t == -ERESTARTSYS) | |
5666 | return t; | |
5667 | return 0; | |
5668 | } | |
5669 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5670 | ||
be4de352 DC |
5671 | /** |
5672 | * try_wait_for_completion - try to decrement a completion without blocking | |
5673 | * @x: completion structure | |
5674 | * | |
5675 | * Returns: 0 if a decrement cannot be done without blocking | |
5676 | * 1 if a decrement succeeded. | |
5677 | * | |
5678 | * If a completion is being used as a counting completion, | |
5679 | * attempt to decrement the counter without blocking. This | |
5680 | * enables us to avoid waiting if the resource the completion | |
5681 | * is protecting is not available. | |
5682 | */ | |
5683 | bool try_wait_for_completion(struct completion *x) | |
5684 | { | |
5685 | int ret = 1; | |
5686 | ||
5687 | spin_lock_irq(&x->wait.lock); | |
5688 | if (!x->done) | |
5689 | ret = 0; | |
5690 | else | |
5691 | x->done--; | |
5692 | spin_unlock_irq(&x->wait.lock); | |
5693 | return ret; | |
5694 | } | |
5695 | EXPORT_SYMBOL(try_wait_for_completion); | |
5696 | ||
5697 | /** | |
5698 | * completion_done - Test to see if a completion has any waiters | |
5699 | * @x: completion structure | |
5700 | * | |
5701 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5702 | * 1 if there are no waiters. | |
5703 | * | |
5704 | */ | |
5705 | bool completion_done(struct completion *x) | |
5706 | { | |
5707 | int ret = 1; | |
5708 | ||
5709 | spin_lock_irq(&x->wait.lock); | |
5710 | if (!x->done) | |
5711 | ret = 0; | |
5712 | spin_unlock_irq(&x->wait.lock); | |
5713 | return ret; | |
5714 | } | |
5715 | EXPORT_SYMBOL(completion_done); | |
5716 | ||
8cbbe86d AK |
5717 | static long __sched |
5718 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5719 | { |
0fec171c IM |
5720 | unsigned long flags; |
5721 | wait_queue_t wait; | |
5722 | ||
5723 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5724 | |
8cbbe86d | 5725 | __set_current_state(state); |
1da177e4 | 5726 | |
8cbbe86d AK |
5727 | spin_lock_irqsave(&q->lock, flags); |
5728 | __add_wait_queue(q, &wait); | |
5729 | spin_unlock(&q->lock); | |
5730 | timeout = schedule_timeout(timeout); | |
5731 | spin_lock_irq(&q->lock); | |
5732 | __remove_wait_queue(q, &wait); | |
5733 | spin_unlock_irqrestore(&q->lock, flags); | |
5734 | ||
5735 | return timeout; | |
5736 | } | |
5737 | ||
5738 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5739 | { | |
5740 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5741 | } |
1da177e4 LT |
5742 | EXPORT_SYMBOL(interruptible_sleep_on); |
5743 | ||
0fec171c | 5744 | long __sched |
95cdf3b7 | 5745 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5746 | { |
8cbbe86d | 5747 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5748 | } |
1da177e4 LT |
5749 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5750 | ||
0fec171c | 5751 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5752 | { |
8cbbe86d | 5753 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5754 | } |
1da177e4 LT |
5755 | EXPORT_SYMBOL(sleep_on); |
5756 | ||
0fec171c | 5757 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5758 | { |
8cbbe86d | 5759 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5760 | } |
1da177e4 LT |
5761 | EXPORT_SYMBOL(sleep_on_timeout); |
5762 | ||
b29739f9 IM |
5763 | #ifdef CONFIG_RT_MUTEXES |
5764 | ||
5765 | /* | |
5766 | * rt_mutex_setprio - set the current priority of a task | |
5767 | * @p: task | |
5768 | * @prio: prio value (kernel-internal form) | |
5769 | * | |
5770 | * This function changes the 'effective' priority of a task. It does | |
5771 | * not touch ->normal_prio like __setscheduler(). | |
5772 | * | |
5773 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5774 | */ | |
36c8b586 | 5775 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5776 | { |
5777 | unsigned long flags; | |
83b699ed | 5778 | int oldprio, on_rq, running; |
70b97a7f | 5779 | struct rq *rq; |
cb469845 | 5780 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5781 | |
5782 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5783 | ||
5784 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5785 | update_rq_clock(rq); |
b29739f9 | 5786 | |
d5f9f942 | 5787 | oldprio = p->prio; |
dd41f596 | 5788 | on_rq = p->se.on_rq; |
051a1d1a | 5789 | running = task_current(rq, p); |
0e1f3483 | 5790 | if (on_rq) |
69be72c1 | 5791 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5792 | if (running) |
5793 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5794 | |
5795 | if (rt_prio(prio)) | |
5796 | p->sched_class = &rt_sched_class; | |
5797 | else | |
5798 | p->sched_class = &fair_sched_class; | |
5799 | ||
b29739f9 IM |
5800 | p->prio = prio; |
5801 | ||
0e1f3483 HS |
5802 | if (running) |
5803 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5804 | if (on_rq) { |
8159f87e | 5805 | enqueue_task(rq, p, 0); |
cb469845 SR |
5806 | |
5807 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5808 | } |
5809 | task_rq_unlock(rq, &flags); | |
5810 | } | |
5811 | ||
5812 | #endif | |
5813 | ||
36c8b586 | 5814 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5815 | { |
dd41f596 | 5816 | int old_prio, delta, on_rq; |
1da177e4 | 5817 | unsigned long flags; |
70b97a7f | 5818 | struct rq *rq; |
1da177e4 LT |
5819 | |
5820 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5821 | return; | |
5822 | /* | |
5823 | * We have to be careful, if called from sys_setpriority(), | |
5824 | * the task might be in the middle of scheduling on another CPU. | |
5825 | */ | |
5826 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5827 | update_rq_clock(rq); |
1da177e4 LT |
5828 | /* |
5829 | * The RT priorities are set via sched_setscheduler(), but we still | |
5830 | * allow the 'normal' nice value to be set - but as expected | |
5831 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5832 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5833 | */ |
e05606d3 | 5834 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5835 | p->static_prio = NICE_TO_PRIO(nice); |
5836 | goto out_unlock; | |
5837 | } | |
dd41f596 | 5838 | on_rq = p->se.on_rq; |
c09595f6 | 5839 | if (on_rq) |
69be72c1 | 5840 | dequeue_task(rq, p, 0); |
1da177e4 | 5841 | |
1da177e4 | 5842 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5843 | set_load_weight(p); |
b29739f9 IM |
5844 | old_prio = p->prio; |
5845 | p->prio = effective_prio(p); | |
5846 | delta = p->prio - old_prio; | |
1da177e4 | 5847 | |
dd41f596 | 5848 | if (on_rq) { |
8159f87e | 5849 | enqueue_task(rq, p, 0); |
1da177e4 | 5850 | /* |
d5f9f942 AM |
5851 | * If the task increased its priority or is running and |
5852 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5853 | */ |
d5f9f942 | 5854 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5855 | resched_task(rq->curr); |
5856 | } | |
5857 | out_unlock: | |
5858 | task_rq_unlock(rq, &flags); | |
5859 | } | |
1da177e4 LT |
5860 | EXPORT_SYMBOL(set_user_nice); |
5861 | ||
e43379f1 MM |
5862 | /* |
5863 | * can_nice - check if a task can reduce its nice value | |
5864 | * @p: task | |
5865 | * @nice: nice value | |
5866 | */ | |
36c8b586 | 5867 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5868 | { |
024f4747 MM |
5869 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5870 | int nice_rlim = 20 - nice; | |
48f24c4d | 5871 | |
e43379f1 MM |
5872 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5873 | capable(CAP_SYS_NICE)); | |
5874 | } | |
5875 | ||
1da177e4 LT |
5876 | #ifdef __ARCH_WANT_SYS_NICE |
5877 | ||
5878 | /* | |
5879 | * sys_nice - change the priority of the current process. | |
5880 | * @increment: priority increment | |
5881 | * | |
5882 | * sys_setpriority is a more generic, but much slower function that | |
5883 | * does similar things. | |
5884 | */ | |
5add95d4 | 5885 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5886 | { |
48f24c4d | 5887 | long nice, retval; |
1da177e4 LT |
5888 | |
5889 | /* | |
5890 | * Setpriority might change our priority at the same moment. | |
5891 | * We don't have to worry. Conceptually one call occurs first | |
5892 | * and we have a single winner. | |
5893 | */ | |
e43379f1 MM |
5894 | if (increment < -40) |
5895 | increment = -40; | |
1da177e4 LT |
5896 | if (increment > 40) |
5897 | increment = 40; | |
5898 | ||
2b8f836f | 5899 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5900 | if (nice < -20) |
5901 | nice = -20; | |
5902 | if (nice > 19) | |
5903 | nice = 19; | |
5904 | ||
e43379f1 MM |
5905 | if (increment < 0 && !can_nice(current, nice)) |
5906 | return -EPERM; | |
5907 | ||
1da177e4 LT |
5908 | retval = security_task_setnice(current, nice); |
5909 | if (retval) | |
5910 | return retval; | |
5911 | ||
5912 | set_user_nice(current, nice); | |
5913 | return 0; | |
5914 | } | |
5915 | ||
5916 | #endif | |
5917 | ||
5918 | /** | |
5919 | * task_prio - return the priority value of a given task. | |
5920 | * @p: the task in question. | |
5921 | * | |
5922 | * This is the priority value as seen by users in /proc. | |
5923 | * RT tasks are offset by -200. Normal tasks are centered | |
5924 | * around 0, value goes from -16 to +15. | |
5925 | */ | |
36c8b586 | 5926 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5927 | { |
5928 | return p->prio - MAX_RT_PRIO; | |
5929 | } | |
5930 | ||
5931 | /** | |
5932 | * task_nice - return the nice value of a given task. | |
5933 | * @p: the task in question. | |
5934 | */ | |
36c8b586 | 5935 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5936 | { |
5937 | return TASK_NICE(p); | |
5938 | } | |
150d8bed | 5939 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5940 | |
5941 | /** | |
5942 | * idle_cpu - is a given cpu idle currently? | |
5943 | * @cpu: the processor in question. | |
5944 | */ | |
5945 | int idle_cpu(int cpu) | |
5946 | { | |
5947 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5948 | } | |
5949 | ||
1da177e4 LT |
5950 | /** |
5951 | * idle_task - return the idle task for a given cpu. | |
5952 | * @cpu: the processor in question. | |
5953 | */ | |
36c8b586 | 5954 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5955 | { |
5956 | return cpu_rq(cpu)->idle; | |
5957 | } | |
5958 | ||
5959 | /** | |
5960 | * find_process_by_pid - find a process with a matching PID value. | |
5961 | * @pid: the pid in question. | |
5962 | */ | |
a9957449 | 5963 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5964 | { |
228ebcbe | 5965 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5966 | } |
5967 | ||
5968 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5969 | static void |
5970 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5971 | { |
dd41f596 | 5972 | BUG_ON(p->se.on_rq); |
48f24c4d | 5973 | |
1da177e4 | 5974 | p->policy = policy; |
dd41f596 IM |
5975 | switch (p->policy) { |
5976 | case SCHED_NORMAL: | |
5977 | case SCHED_BATCH: | |
5978 | case SCHED_IDLE: | |
5979 | p->sched_class = &fair_sched_class; | |
5980 | break; | |
5981 | case SCHED_FIFO: | |
5982 | case SCHED_RR: | |
5983 | p->sched_class = &rt_sched_class; | |
5984 | break; | |
5985 | } | |
5986 | ||
1da177e4 | 5987 | p->rt_priority = prio; |
b29739f9 IM |
5988 | p->normal_prio = normal_prio(p); |
5989 | /* we are holding p->pi_lock already */ | |
5990 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5991 | set_load_weight(p); |
1da177e4 LT |
5992 | } |
5993 | ||
c69e8d9c DH |
5994 | /* |
5995 | * check the target process has a UID that matches the current process's | |
5996 | */ | |
5997 | static bool check_same_owner(struct task_struct *p) | |
5998 | { | |
5999 | const struct cred *cred = current_cred(), *pcred; | |
6000 | bool match; | |
6001 | ||
6002 | rcu_read_lock(); | |
6003 | pcred = __task_cred(p); | |
6004 | match = (cred->euid == pcred->euid || | |
6005 | cred->euid == pcred->uid); | |
6006 | rcu_read_unlock(); | |
6007 | return match; | |
6008 | } | |
6009 | ||
961ccddd RR |
6010 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6011 | struct sched_param *param, bool user) | |
1da177e4 | 6012 | { |
83b699ed | 6013 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6014 | unsigned long flags; |
cb469845 | 6015 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6016 | struct rq *rq; |
1da177e4 | 6017 | |
66e5393a SR |
6018 | /* may grab non-irq protected spin_locks */ |
6019 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6020 | recheck: |
6021 | /* double check policy once rq lock held */ | |
6022 | if (policy < 0) | |
6023 | policy = oldpolicy = p->policy; | |
6024 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
6025 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
6026 | policy != SCHED_IDLE) | |
b0a9499c | 6027 | return -EINVAL; |
1da177e4 LT |
6028 | /* |
6029 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6030 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6031 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6032 | */ |
6033 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6034 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6035 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6036 | return -EINVAL; |
e05606d3 | 6037 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6038 | return -EINVAL; |
6039 | ||
37e4ab3f OC |
6040 | /* |
6041 | * Allow unprivileged RT tasks to decrease priority: | |
6042 | */ | |
961ccddd | 6043 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6044 | if (rt_policy(policy)) { |
8dc3e909 | 6045 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6046 | |
6047 | if (!lock_task_sighand(p, &flags)) | |
6048 | return -ESRCH; | |
6049 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6050 | unlock_task_sighand(p, &flags); | |
6051 | ||
6052 | /* can't set/change the rt policy */ | |
6053 | if (policy != p->policy && !rlim_rtprio) | |
6054 | return -EPERM; | |
6055 | ||
6056 | /* can't increase priority */ | |
6057 | if (param->sched_priority > p->rt_priority && | |
6058 | param->sched_priority > rlim_rtprio) | |
6059 | return -EPERM; | |
6060 | } | |
dd41f596 IM |
6061 | /* |
6062 | * Like positive nice levels, dont allow tasks to | |
6063 | * move out of SCHED_IDLE either: | |
6064 | */ | |
6065 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6066 | return -EPERM; | |
5fe1d75f | 6067 | |
37e4ab3f | 6068 | /* can't change other user's priorities */ |
c69e8d9c | 6069 | if (!check_same_owner(p)) |
37e4ab3f OC |
6070 | return -EPERM; |
6071 | } | |
1da177e4 | 6072 | |
725aad24 | 6073 | if (user) { |
b68aa230 | 6074 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6075 | /* |
6076 | * Do not allow realtime tasks into groups that have no runtime | |
6077 | * assigned. | |
6078 | */ | |
9a7e0b18 PZ |
6079 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6080 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6081 | return -EPERM; |
b68aa230 PZ |
6082 | #endif |
6083 | ||
725aad24 JF |
6084 | retval = security_task_setscheduler(p, policy, param); |
6085 | if (retval) | |
6086 | return retval; | |
6087 | } | |
6088 | ||
b29739f9 IM |
6089 | /* |
6090 | * make sure no PI-waiters arrive (or leave) while we are | |
6091 | * changing the priority of the task: | |
6092 | */ | |
6093 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6094 | /* |
6095 | * To be able to change p->policy safely, the apropriate | |
6096 | * runqueue lock must be held. | |
6097 | */ | |
b29739f9 | 6098 | rq = __task_rq_lock(p); |
1da177e4 LT |
6099 | /* recheck policy now with rq lock held */ |
6100 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6101 | policy = oldpolicy = -1; | |
b29739f9 IM |
6102 | __task_rq_unlock(rq); |
6103 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6104 | goto recheck; |
6105 | } | |
2daa3577 | 6106 | update_rq_clock(rq); |
dd41f596 | 6107 | on_rq = p->se.on_rq; |
051a1d1a | 6108 | running = task_current(rq, p); |
0e1f3483 | 6109 | if (on_rq) |
2e1cb74a | 6110 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6111 | if (running) |
6112 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6113 | |
1da177e4 | 6114 | oldprio = p->prio; |
dd41f596 | 6115 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6116 | |
0e1f3483 HS |
6117 | if (running) |
6118 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6119 | if (on_rq) { |
6120 | activate_task(rq, p, 0); | |
cb469845 SR |
6121 | |
6122 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6123 | } |
b29739f9 IM |
6124 | __task_rq_unlock(rq); |
6125 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6126 | ||
95e02ca9 TG |
6127 | rt_mutex_adjust_pi(p); |
6128 | ||
1da177e4 LT |
6129 | return 0; |
6130 | } | |
961ccddd RR |
6131 | |
6132 | /** | |
6133 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6134 | * @p: the task in question. | |
6135 | * @policy: new policy. | |
6136 | * @param: structure containing the new RT priority. | |
6137 | * | |
6138 | * NOTE that the task may be already dead. | |
6139 | */ | |
6140 | int sched_setscheduler(struct task_struct *p, int policy, | |
6141 | struct sched_param *param) | |
6142 | { | |
6143 | return __sched_setscheduler(p, policy, param, true); | |
6144 | } | |
1da177e4 LT |
6145 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6146 | ||
961ccddd RR |
6147 | /** |
6148 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6149 | * @p: the task in question. | |
6150 | * @policy: new policy. | |
6151 | * @param: structure containing the new RT priority. | |
6152 | * | |
6153 | * Just like sched_setscheduler, only don't bother checking if the | |
6154 | * current context has permission. For example, this is needed in | |
6155 | * stop_machine(): we create temporary high priority worker threads, | |
6156 | * but our caller might not have that capability. | |
6157 | */ | |
6158 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6159 | struct sched_param *param) | |
6160 | { | |
6161 | return __sched_setscheduler(p, policy, param, false); | |
6162 | } | |
6163 | ||
95cdf3b7 IM |
6164 | static int |
6165 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6166 | { |
1da177e4 LT |
6167 | struct sched_param lparam; |
6168 | struct task_struct *p; | |
36c8b586 | 6169 | int retval; |
1da177e4 LT |
6170 | |
6171 | if (!param || pid < 0) | |
6172 | return -EINVAL; | |
6173 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6174 | return -EFAULT; | |
5fe1d75f ON |
6175 | |
6176 | rcu_read_lock(); | |
6177 | retval = -ESRCH; | |
1da177e4 | 6178 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6179 | if (p != NULL) |
6180 | retval = sched_setscheduler(p, policy, &lparam); | |
6181 | rcu_read_unlock(); | |
36c8b586 | 6182 | |
1da177e4 LT |
6183 | return retval; |
6184 | } | |
6185 | ||
6186 | /** | |
6187 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6188 | * @pid: the pid in question. | |
6189 | * @policy: new policy. | |
6190 | * @param: structure containing the new RT priority. | |
6191 | */ | |
5add95d4 HC |
6192 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6193 | struct sched_param __user *, param) | |
1da177e4 | 6194 | { |
c21761f1 JB |
6195 | /* negative values for policy are not valid */ |
6196 | if (policy < 0) | |
6197 | return -EINVAL; | |
6198 | ||
1da177e4 LT |
6199 | return do_sched_setscheduler(pid, policy, param); |
6200 | } | |
6201 | ||
6202 | /** | |
6203 | * sys_sched_setparam - set/change the RT priority of a thread | |
6204 | * @pid: the pid in question. | |
6205 | * @param: structure containing the new RT priority. | |
6206 | */ | |
5add95d4 | 6207 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6208 | { |
6209 | return do_sched_setscheduler(pid, -1, param); | |
6210 | } | |
6211 | ||
6212 | /** | |
6213 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6214 | * @pid: the pid in question. | |
6215 | */ | |
5add95d4 | 6216 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6217 | { |
36c8b586 | 6218 | struct task_struct *p; |
3a5c359a | 6219 | int retval; |
1da177e4 LT |
6220 | |
6221 | if (pid < 0) | |
3a5c359a | 6222 | return -EINVAL; |
1da177e4 LT |
6223 | |
6224 | retval = -ESRCH; | |
6225 | read_lock(&tasklist_lock); | |
6226 | p = find_process_by_pid(pid); | |
6227 | if (p) { | |
6228 | retval = security_task_getscheduler(p); | |
6229 | if (!retval) | |
6230 | retval = p->policy; | |
6231 | } | |
6232 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6233 | return retval; |
6234 | } | |
6235 | ||
6236 | /** | |
6237 | * sys_sched_getscheduler - get the RT priority of a thread | |
6238 | * @pid: the pid in question. | |
6239 | * @param: structure containing the RT priority. | |
6240 | */ | |
5add95d4 | 6241 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6242 | { |
6243 | struct sched_param lp; | |
36c8b586 | 6244 | struct task_struct *p; |
3a5c359a | 6245 | int retval; |
1da177e4 LT |
6246 | |
6247 | if (!param || pid < 0) | |
3a5c359a | 6248 | return -EINVAL; |
1da177e4 LT |
6249 | |
6250 | read_lock(&tasklist_lock); | |
6251 | p = find_process_by_pid(pid); | |
6252 | retval = -ESRCH; | |
6253 | if (!p) | |
6254 | goto out_unlock; | |
6255 | ||
6256 | retval = security_task_getscheduler(p); | |
6257 | if (retval) | |
6258 | goto out_unlock; | |
6259 | ||
6260 | lp.sched_priority = p->rt_priority; | |
6261 | read_unlock(&tasklist_lock); | |
6262 | ||
6263 | /* | |
6264 | * This one might sleep, we cannot do it with a spinlock held ... | |
6265 | */ | |
6266 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6267 | ||
1da177e4 LT |
6268 | return retval; |
6269 | ||
6270 | out_unlock: | |
6271 | read_unlock(&tasklist_lock); | |
6272 | return retval; | |
6273 | } | |
6274 | ||
96f874e2 | 6275 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6276 | { |
5a16f3d3 | 6277 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6278 | struct task_struct *p; |
6279 | int retval; | |
1da177e4 | 6280 | |
95402b38 | 6281 | get_online_cpus(); |
1da177e4 LT |
6282 | read_lock(&tasklist_lock); |
6283 | ||
6284 | p = find_process_by_pid(pid); | |
6285 | if (!p) { | |
6286 | read_unlock(&tasklist_lock); | |
95402b38 | 6287 | put_online_cpus(); |
1da177e4 LT |
6288 | return -ESRCH; |
6289 | } | |
6290 | ||
6291 | /* | |
6292 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6293 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6294 | * usage count and then drop tasklist_lock. |
6295 | */ | |
6296 | get_task_struct(p); | |
6297 | read_unlock(&tasklist_lock); | |
6298 | ||
5a16f3d3 RR |
6299 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6300 | retval = -ENOMEM; | |
6301 | goto out_put_task; | |
6302 | } | |
6303 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6304 | retval = -ENOMEM; | |
6305 | goto out_free_cpus_allowed; | |
6306 | } | |
1da177e4 | 6307 | retval = -EPERM; |
c69e8d9c | 6308 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6309 | goto out_unlock; |
6310 | ||
e7834f8f DQ |
6311 | retval = security_task_setscheduler(p, 0, NULL); |
6312 | if (retval) | |
6313 | goto out_unlock; | |
6314 | ||
5a16f3d3 RR |
6315 | cpuset_cpus_allowed(p, cpus_allowed); |
6316 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6317 | again: |
5a16f3d3 | 6318 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6319 | |
8707d8b8 | 6320 | if (!retval) { |
5a16f3d3 RR |
6321 | cpuset_cpus_allowed(p, cpus_allowed); |
6322 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6323 | /* |
6324 | * We must have raced with a concurrent cpuset | |
6325 | * update. Just reset the cpus_allowed to the | |
6326 | * cpuset's cpus_allowed | |
6327 | */ | |
5a16f3d3 | 6328 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6329 | goto again; |
6330 | } | |
6331 | } | |
1da177e4 | 6332 | out_unlock: |
5a16f3d3 RR |
6333 | free_cpumask_var(new_mask); |
6334 | out_free_cpus_allowed: | |
6335 | free_cpumask_var(cpus_allowed); | |
6336 | out_put_task: | |
1da177e4 | 6337 | put_task_struct(p); |
95402b38 | 6338 | put_online_cpus(); |
1da177e4 LT |
6339 | return retval; |
6340 | } | |
6341 | ||
6342 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6343 | struct cpumask *new_mask) |
1da177e4 | 6344 | { |
96f874e2 RR |
6345 | if (len < cpumask_size()) |
6346 | cpumask_clear(new_mask); | |
6347 | else if (len > cpumask_size()) | |
6348 | len = cpumask_size(); | |
6349 | ||
1da177e4 LT |
6350 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6351 | } | |
6352 | ||
6353 | /** | |
6354 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6355 | * @pid: pid of the process | |
6356 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6357 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6358 | */ | |
5add95d4 HC |
6359 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6360 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6361 | { |
5a16f3d3 | 6362 | cpumask_var_t new_mask; |
1da177e4 LT |
6363 | int retval; |
6364 | ||
5a16f3d3 RR |
6365 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6366 | return -ENOMEM; | |
1da177e4 | 6367 | |
5a16f3d3 RR |
6368 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6369 | if (retval == 0) | |
6370 | retval = sched_setaffinity(pid, new_mask); | |
6371 | free_cpumask_var(new_mask); | |
6372 | return retval; | |
1da177e4 LT |
6373 | } |
6374 | ||
96f874e2 | 6375 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6376 | { |
36c8b586 | 6377 | struct task_struct *p; |
1da177e4 | 6378 | int retval; |
1da177e4 | 6379 | |
95402b38 | 6380 | get_online_cpus(); |
1da177e4 LT |
6381 | read_lock(&tasklist_lock); |
6382 | ||
6383 | retval = -ESRCH; | |
6384 | p = find_process_by_pid(pid); | |
6385 | if (!p) | |
6386 | goto out_unlock; | |
6387 | ||
e7834f8f DQ |
6388 | retval = security_task_getscheduler(p); |
6389 | if (retval) | |
6390 | goto out_unlock; | |
6391 | ||
96f874e2 | 6392 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6393 | |
6394 | out_unlock: | |
6395 | read_unlock(&tasklist_lock); | |
95402b38 | 6396 | put_online_cpus(); |
1da177e4 | 6397 | |
9531b62f | 6398 | return retval; |
1da177e4 LT |
6399 | } |
6400 | ||
6401 | /** | |
6402 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6403 | * @pid: pid of the process | |
6404 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6405 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6406 | */ | |
5add95d4 HC |
6407 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6408 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6409 | { |
6410 | int ret; | |
f17c8607 | 6411 | cpumask_var_t mask; |
1da177e4 | 6412 | |
f17c8607 | 6413 | if (len < cpumask_size()) |
1da177e4 LT |
6414 | return -EINVAL; |
6415 | ||
f17c8607 RR |
6416 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6417 | return -ENOMEM; | |
1da177e4 | 6418 | |
f17c8607 RR |
6419 | ret = sched_getaffinity(pid, mask); |
6420 | if (ret == 0) { | |
6421 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6422 | ret = -EFAULT; | |
6423 | else | |
6424 | ret = cpumask_size(); | |
6425 | } | |
6426 | free_cpumask_var(mask); | |
1da177e4 | 6427 | |
f17c8607 | 6428 | return ret; |
1da177e4 LT |
6429 | } |
6430 | ||
6431 | /** | |
6432 | * sys_sched_yield - yield the current processor to other threads. | |
6433 | * | |
dd41f596 IM |
6434 | * This function yields the current CPU to other tasks. If there are no |
6435 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6436 | */ |
5add95d4 | 6437 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6438 | { |
70b97a7f | 6439 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6440 | |
2d72376b | 6441 | schedstat_inc(rq, yld_count); |
4530d7ab | 6442 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6443 | |
6444 | /* | |
6445 | * Since we are going to call schedule() anyway, there's | |
6446 | * no need to preempt or enable interrupts: | |
6447 | */ | |
6448 | __release(rq->lock); | |
8a25d5de | 6449 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6450 | _raw_spin_unlock(&rq->lock); |
6451 | preempt_enable_no_resched(); | |
6452 | ||
6453 | schedule(); | |
6454 | ||
6455 | return 0; | |
6456 | } | |
6457 | ||
e7b38404 | 6458 | static void __cond_resched(void) |
1da177e4 | 6459 | { |
8e0a43d8 IM |
6460 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6461 | __might_sleep(__FILE__, __LINE__); | |
6462 | #endif | |
5bbcfd90 IM |
6463 | /* |
6464 | * The BKS might be reacquired before we have dropped | |
6465 | * PREEMPT_ACTIVE, which could trigger a second | |
6466 | * cond_resched() call. | |
6467 | */ | |
1da177e4 LT |
6468 | do { |
6469 | add_preempt_count(PREEMPT_ACTIVE); | |
6470 | schedule(); | |
6471 | sub_preempt_count(PREEMPT_ACTIVE); | |
6472 | } while (need_resched()); | |
6473 | } | |
6474 | ||
02b67cc3 | 6475 | int __sched _cond_resched(void) |
1da177e4 | 6476 | { |
9414232f IM |
6477 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6478 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6479 | __cond_resched(); |
6480 | return 1; | |
6481 | } | |
6482 | return 0; | |
6483 | } | |
02b67cc3 | 6484 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6485 | |
6486 | /* | |
6487 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6488 | * call schedule, and on return reacquire the lock. | |
6489 | * | |
41a2d6cf | 6490 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6491 | * operations here to prevent schedule() from being called twice (once via |
6492 | * spin_unlock(), once by hand). | |
6493 | */ | |
95cdf3b7 | 6494 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6495 | { |
95c354fe | 6496 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6497 | int ret = 0; |
6498 | ||
95c354fe | 6499 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6500 | spin_unlock(lock); |
95c354fe NP |
6501 | if (resched && need_resched()) |
6502 | __cond_resched(); | |
6503 | else | |
6504 | cpu_relax(); | |
6df3cecb | 6505 | ret = 1; |
1da177e4 | 6506 | spin_lock(lock); |
1da177e4 | 6507 | } |
6df3cecb | 6508 | return ret; |
1da177e4 | 6509 | } |
1da177e4 LT |
6510 | EXPORT_SYMBOL(cond_resched_lock); |
6511 | ||
6512 | int __sched cond_resched_softirq(void) | |
6513 | { | |
6514 | BUG_ON(!in_softirq()); | |
6515 | ||
9414232f | 6516 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6517 | local_bh_enable(); |
1da177e4 LT |
6518 | __cond_resched(); |
6519 | local_bh_disable(); | |
6520 | return 1; | |
6521 | } | |
6522 | return 0; | |
6523 | } | |
1da177e4 LT |
6524 | EXPORT_SYMBOL(cond_resched_softirq); |
6525 | ||
1da177e4 LT |
6526 | /** |
6527 | * yield - yield the current processor to other threads. | |
6528 | * | |
72fd4a35 | 6529 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6530 | * thread runnable and calls sys_sched_yield(). |
6531 | */ | |
6532 | void __sched yield(void) | |
6533 | { | |
6534 | set_current_state(TASK_RUNNING); | |
6535 | sys_sched_yield(); | |
6536 | } | |
1da177e4 LT |
6537 | EXPORT_SYMBOL(yield); |
6538 | ||
6539 | /* | |
41a2d6cf | 6540 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6541 | * that process accounting knows that this is a task in IO wait state. |
6542 | * | |
6543 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6544 | * has set its backing_dev_info: the queue against which it should throttle) | |
6545 | */ | |
6546 | void __sched io_schedule(void) | |
6547 | { | |
70b97a7f | 6548 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6549 | |
0ff92245 | 6550 | delayacct_blkio_start(); |
1da177e4 LT |
6551 | atomic_inc(&rq->nr_iowait); |
6552 | schedule(); | |
6553 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6554 | delayacct_blkio_end(); |
1da177e4 | 6555 | } |
1da177e4 LT |
6556 | EXPORT_SYMBOL(io_schedule); |
6557 | ||
6558 | long __sched io_schedule_timeout(long timeout) | |
6559 | { | |
70b97a7f | 6560 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6561 | long ret; |
6562 | ||
0ff92245 | 6563 | delayacct_blkio_start(); |
1da177e4 LT |
6564 | atomic_inc(&rq->nr_iowait); |
6565 | ret = schedule_timeout(timeout); | |
6566 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6567 | delayacct_blkio_end(); |
1da177e4 LT |
6568 | return ret; |
6569 | } | |
6570 | ||
6571 | /** | |
6572 | * sys_sched_get_priority_max - return maximum RT priority. | |
6573 | * @policy: scheduling class. | |
6574 | * | |
6575 | * this syscall returns the maximum rt_priority that can be used | |
6576 | * by a given scheduling class. | |
6577 | */ | |
5add95d4 | 6578 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6579 | { |
6580 | int ret = -EINVAL; | |
6581 | ||
6582 | switch (policy) { | |
6583 | case SCHED_FIFO: | |
6584 | case SCHED_RR: | |
6585 | ret = MAX_USER_RT_PRIO-1; | |
6586 | break; | |
6587 | case SCHED_NORMAL: | |
b0a9499c | 6588 | case SCHED_BATCH: |
dd41f596 | 6589 | case SCHED_IDLE: |
1da177e4 LT |
6590 | ret = 0; |
6591 | break; | |
6592 | } | |
6593 | return ret; | |
6594 | } | |
6595 | ||
6596 | /** | |
6597 | * sys_sched_get_priority_min - return minimum RT priority. | |
6598 | * @policy: scheduling class. | |
6599 | * | |
6600 | * this syscall returns the minimum rt_priority that can be used | |
6601 | * by a given scheduling class. | |
6602 | */ | |
5add95d4 | 6603 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6604 | { |
6605 | int ret = -EINVAL; | |
6606 | ||
6607 | switch (policy) { | |
6608 | case SCHED_FIFO: | |
6609 | case SCHED_RR: | |
6610 | ret = 1; | |
6611 | break; | |
6612 | case SCHED_NORMAL: | |
b0a9499c | 6613 | case SCHED_BATCH: |
dd41f596 | 6614 | case SCHED_IDLE: |
1da177e4 LT |
6615 | ret = 0; |
6616 | } | |
6617 | return ret; | |
6618 | } | |
6619 | ||
6620 | /** | |
6621 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6622 | * @pid: pid of the process. | |
6623 | * @interval: userspace pointer to the timeslice value. | |
6624 | * | |
6625 | * this syscall writes the default timeslice value of a given process | |
6626 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6627 | */ | |
17da2bd9 | 6628 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6629 | struct timespec __user *, interval) |
1da177e4 | 6630 | { |
36c8b586 | 6631 | struct task_struct *p; |
a4ec24b4 | 6632 | unsigned int time_slice; |
3a5c359a | 6633 | int retval; |
1da177e4 | 6634 | struct timespec t; |
1da177e4 LT |
6635 | |
6636 | if (pid < 0) | |
3a5c359a | 6637 | return -EINVAL; |
1da177e4 LT |
6638 | |
6639 | retval = -ESRCH; | |
6640 | read_lock(&tasklist_lock); | |
6641 | p = find_process_by_pid(pid); | |
6642 | if (!p) | |
6643 | goto out_unlock; | |
6644 | ||
6645 | retval = security_task_getscheduler(p); | |
6646 | if (retval) | |
6647 | goto out_unlock; | |
6648 | ||
77034937 IM |
6649 | /* |
6650 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6651 | * tasks that are on an otherwise idle runqueue: | |
6652 | */ | |
6653 | time_slice = 0; | |
6654 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6655 | time_slice = DEF_TIMESLICE; |
1868f958 | 6656 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6657 | struct sched_entity *se = &p->se; |
6658 | unsigned long flags; | |
6659 | struct rq *rq; | |
6660 | ||
6661 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6662 | if (rq->cfs.load.weight) |
6663 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6664 | task_rq_unlock(rq, &flags); |
6665 | } | |
1da177e4 | 6666 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6667 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6668 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6669 | return retval; |
3a5c359a | 6670 | |
1da177e4 LT |
6671 | out_unlock: |
6672 | read_unlock(&tasklist_lock); | |
6673 | return retval; | |
6674 | } | |
6675 | ||
7c731e0a | 6676 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6677 | |
82a1fcb9 | 6678 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6679 | { |
1da177e4 | 6680 | unsigned long free = 0; |
36c8b586 | 6681 | unsigned state; |
1da177e4 | 6682 | |
1da177e4 | 6683 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6684 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6685 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6686 | #if BITS_PER_LONG == 32 |
1da177e4 | 6687 | if (state == TASK_RUNNING) |
cc4ea795 | 6688 | printk(KERN_CONT " running "); |
1da177e4 | 6689 | else |
cc4ea795 | 6690 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6691 | #else |
6692 | if (state == TASK_RUNNING) | |
cc4ea795 | 6693 | printk(KERN_CONT " running task "); |
1da177e4 | 6694 | else |
cc4ea795 | 6695 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6696 | #endif |
6697 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6698 | free = stack_not_used(p); |
1da177e4 | 6699 | #endif |
aa47b7e0 DR |
6700 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6701 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6702 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6703 | |
5fb5e6de | 6704 | show_stack(p, NULL); |
1da177e4 LT |
6705 | } |
6706 | ||
e59e2ae2 | 6707 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6708 | { |
36c8b586 | 6709 | struct task_struct *g, *p; |
1da177e4 | 6710 | |
4bd77321 IM |
6711 | #if BITS_PER_LONG == 32 |
6712 | printk(KERN_INFO | |
6713 | " task PC stack pid father\n"); | |
1da177e4 | 6714 | #else |
4bd77321 IM |
6715 | printk(KERN_INFO |
6716 | " task PC stack pid father\n"); | |
1da177e4 LT |
6717 | #endif |
6718 | read_lock(&tasklist_lock); | |
6719 | do_each_thread(g, p) { | |
6720 | /* | |
6721 | * reset the NMI-timeout, listing all files on a slow | |
6722 | * console might take alot of time: | |
6723 | */ | |
6724 | touch_nmi_watchdog(); | |
39bc89fd | 6725 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6726 | sched_show_task(p); |
1da177e4 LT |
6727 | } while_each_thread(g, p); |
6728 | ||
04c9167f JF |
6729 | touch_all_softlockup_watchdogs(); |
6730 | ||
dd41f596 IM |
6731 | #ifdef CONFIG_SCHED_DEBUG |
6732 | sysrq_sched_debug_show(); | |
6733 | #endif | |
1da177e4 | 6734 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6735 | /* |
6736 | * Only show locks if all tasks are dumped: | |
6737 | */ | |
6738 | if (state_filter == -1) | |
6739 | debug_show_all_locks(); | |
1da177e4 LT |
6740 | } |
6741 | ||
1df21055 IM |
6742 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6743 | { | |
dd41f596 | 6744 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6745 | } |
6746 | ||
f340c0d1 IM |
6747 | /** |
6748 | * init_idle - set up an idle thread for a given CPU | |
6749 | * @idle: task in question | |
6750 | * @cpu: cpu the idle task belongs to | |
6751 | * | |
6752 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6753 | * flag, to make booting more robust. | |
6754 | */ | |
5c1e1767 | 6755 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6756 | { |
70b97a7f | 6757 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6758 | unsigned long flags; |
6759 | ||
5cbd54ef IM |
6760 | spin_lock_irqsave(&rq->lock, flags); |
6761 | ||
dd41f596 IM |
6762 | __sched_fork(idle); |
6763 | idle->se.exec_start = sched_clock(); | |
6764 | ||
b29739f9 | 6765 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6766 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6767 | __set_task_cpu(idle, cpu); |
1da177e4 | 6768 | |
1da177e4 | 6769 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6770 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6771 | idle->oncpu = 1; | |
6772 | #endif | |
1da177e4 LT |
6773 | spin_unlock_irqrestore(&rq->lock, flags); |
6774 | ||
6775 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6776 | #if defined(CONFIG_PREEMPT) |
6777 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6778 | #else | |
a1261f54 | 6779 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6780 | #endif |
dd41f596 IM |
6781 | /* |
6782 | * The idle tasks have their own, simple scheduling class: | |
6783 | */ | |
6784 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6785 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6786 | } |
6787 | ||
6788 | /* | |
6789 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6790 | * indicates which cpus entered this state. This is used | |
6791 | * in the rcu update to wait only for active cpus. For system | |
6792 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6793 | * always be CPU_BITS_NONE. |
1da177e4 | 6794 | */ |
6a7b3dc3 | 6795 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6796 | |
19978ca6 IM |
6797 | /* |
6798 | * Increase the granularity value when there are more CPUs, | |
6799 | * because with more CPUs the 'effective latency' as visible | |
6800 | * to users decreases. But the relationship is not linear, | |
6801 | * so pick a second-best guess by going with the log2 of the | |
6802 | * number of CPUs. | |
6803 | * | |
6804 | * This idea comes from the SD scheduler of Con Kolivas: | |
6805 | */ | |
6806 | static inline void sched_init_granularity(void) | |
6807 | { | |
6808 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6809 | const unsigned long limit = 200000000; | |
6810 | ||
6811 | sysctl_sched_min_granularity *= factor; | |
6812 | if (sysctl_sched_min_granularity > limit) | |
6813 | sysctl_sched_min_granularity = limit; | |
6814 | ||
6815 | sysctl_sched_latency *= factor; | |
6816 | if (sysctl_sched_latency > limit) | |
6817 | sysctl_sched_latency = limit; | |
6818 | ||
6819 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6820 | |
6821 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6822 | } |
6823 | ||
1da177e4 LT |
6824 | #ifdef CONFIG_SMP |
6825 | /* | |
6826 | * This is how migration works: | |
6827 | * | |
70b97a7f | 6828 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6829 | * runqueue and wake up that CPU's migration thread. |
6830 | * 2) we down() the locked semaphore => thread blocks. | |
6831 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6832 | * thread off the CPU) | |
6833 | * 4) it gets the migration request and checks whether the migrated | |
6834 | * task is still in the wrong runqueue. | |
6835 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6836 | * it and puts it into the right queue. | |
6837 | * 6) migration thread up()s the semaphore. | |
6838 | * 7) we wake up and the migration is done. | |
6839 | */ | |
6840 | ||
6841 | /* | |
6842 | * Change a given task's CPU affinity. Migrate the thread to a | |
6843 | * proper CPU and schedule it away if the CPU it's executing on | |
6844 | * is removed from the allowed bitmask. | |
6845 | * | |
6846 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6847 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6848 | * call is not atomic; no spinlocks may be held. |
6849 | */ | |
96f874e2 | 6850 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6851 | { |
70b97a7f | 6852 | struct migration_req req; |
1da177e4 | 6853 | unsigned long flags; |
70b97a7f | 6854 | struct rq *rq; |
48f24c4d | 6855 | int ret = 0; |
1da177e4 LT |
6856 | |
6857 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6858 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6859 | ret = -EINVAL; |
6860 | goto out; | |
6861 | } | |
6862 | ||
9985b0ba | 6863 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6864 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6865 | ret = -EINVAL; |
6866 | goto out; | |
6867 | } | |
6868 | ||
73fe6aae | 6869 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6870 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6871 | else { |
96f874e2 RR |
6872 | cpumask_copy(&p->cpus_allowed, new_mask); |
6873 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6874 | } |
6875 | ||
1da177e4 | 6876 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6877 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6878 | goto out; |
6879 | ||
1e5ce4f4 | 6880 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6881 | /* Need help from migration thread: drop lock and wait. */ |
6882 | task_rq_unlock(rq, &flags); | |
6883 | wake_up_process(rq->migration_thread); | |
6884 | wait_for_completion(&req.done); | |
6885 | tlb_migrate_finish(p->mm); | |
6886 | return 0; | |
6887 | } | |
6888 | out: | |
6889 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6890 | |
1da177e4 LT |
6891 | return ret; |
6892 | } | |
cd8ba7cd | 6893 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6894 | |
6895 | /* | |
41a2d6cf | 6896 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6897 | * this because either it can't run here any more (set_cpus_allowed() |
6898 | * away from this CPU, or CPU going down), or because we're | |
6899 | * attempting to rebalance this task on exec (sched_exec). | |
6900 | * | |
6901 | * So we race with normal scheduler movements, but that's OK, as long | |
6902 | * as the task is no longer on this CPU. | |
efc30814 KK |
6903 | * |
6904 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6905 | */ |
efc30814 | 6906 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6907 | { |
70b97a7f | 6908 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6909 | int ret = 0, on_rq; |
1da177e4 | 6910 | |
e761b772 | 6911 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6912 | return ret; |
1da177e4 LT |
6913 | |
6914 | rq_src = cpu_rq(src_cpu); | |
6915 | rq_dest = cpu_rq(dest_cpu); | |
6916 | ||
6917 | double_rq_lock(rq_src, rq_dest); | |
6918 | /* Already moved. */ | |
6919 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6920 | goto done; |
1da177e4 | 6921 | /* Affinity changed (again). */ |
96f874e2 | 6922 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6923 | goto fail; |
1da177e4 | 6924 | |
dd41f596 | 6925 | on_rq = p->se.on_rq; |
6e82a3be | 6926 | if (on_rq) |
2e1cb74a | 6927 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6928 | |
1da177e4 | 6929 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6930 | if (on_rq) { |
6931 | activate_task(rq_dest, p, 0); | |
15afe09b | 6932 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6933 | } |
b1e38734 | 6934 | done: |
efc30814 | 6935 | ret = 1; |
b1e38734 | 6936 | fail: |
1da177e4 | 6937 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6938 | return ret; |
1da177e4 LT |
6939 | } |
6940 | ||
6941 | /* | |
6942 | * migration_thread - this is a highprio system thread that performs | |
6943 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6944 | * another runqueue. | |
6945 | */ | |
95cdf3b7 | 6946 | static int migration_thread(void *data) |
1da177e4 | 6947 | { |
1da177e4 | 6948 | int cpu = (long)data; |
70b97a7f | 6949 | struct rq *rq; |
1da177e4 LT |
6950 | |
6951 | rq = cpu_rq(cpu); | |
6952 | BUG_ON(rq->migration_thread != current); | |
6953 | ||
6954 | set_current_state(TASK_INTERRUPTIBLE); | |
6955 | while (!kthread_should_stop()) { | |
70b97a7f | 6956 | struct migration_req *req; |
1da177e4 | 6957 | struct list_head *head; |
1da177e4 | 6958 | |
1da177e4 LT |
6959 | spin_lock_irq(&rq->lock); |
6960 | ||
6961 | if (cpu_is_offline(cpu)) { | |
6962 | spin_unlock_irq(&rq->lock); | |
6963 | goto wait_to_die; | |
6964 | } | |
6965 | ||
6966 | if (rq->active_balance) { | |
6967 | active_load_balance(rq, cpu); | |
6968 | rq->active_balance = 0; | |
6969 | } | |
6970 | ||
6971 | head = &rq->migration_queue; | |
6972 | ||
6973 | if (list_empty(head)) { | |
6974 | spin_unlock_irq(&rq->lock); | |
6975 | schedule(); | |
6976 | set_current_state(TASK_INTERRUPTIBLE); | |
6977 | continue; | |
6978 | } | |
70b97a7f | 6979 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6980 | list_del_init(head->next); |
6981 | ||
674311d5 NP |
6982 | spin_unlock(&rq->lock); |
6983 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6984 | local_irq_enable(); | |
1da177e4 LT |
6985 | |
6986 | complete(&req->done); | |
6987 | } | |
6988 | __set_current_state(TASK_RUNNING); | |
6989 | return 0; | |
6990 | ||
6991 | wait_to_die: | |
6992 | /* Wait for kthread_stop */ | |
6993 | set_current_state(TASK_INTERRUPTIBLE); | |
6994 | while (!kthread_should_stop()) { | |
6995 | schedule(); | |
6996 | set_current_state(TASK_INTERRUPTIBLE); | |
6997 | } | |
6998 | __set_current_state(TASK_RUNNING); | |
6999 | return 0; | |
7000 | } | |
7001 | ||
7002 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7003 | |
7004 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7005 | { | |
7006 | int ret; | |
7007 | ||
7008 | local_irq_disable(); | |
7009 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7010 | local_irq_enable(); | |
7011 | return ret; | |
7012 | } | |
7013 | ||
054b9108 | 7014 | /* |
3a4fa0a2 | 7015 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7016 | */ |
48f24c4d | 7017 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7018 | { |
70b97a7f | 7019 | int dest_cpu; |
6ca09dfc | 7020 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7021 | |
7022 | again: | |
7023 | /* Look for allowed, online CPU in same node. */ | |
7024 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7025 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7026 | goto move; | |
7027 | ||
7028 | /* Any allowed, online CPU? */ | |
7029 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7030 | if (dest_cpu < nr_cpu_ids) | |
7031 | goto move; | |
7032 | ||
7033 | /* No more Mr. Nice Guy. */ | |
7034 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7035 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7036 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7037 | |
e76bd8d9 RR |
7038 | /* |
7039 | * Don't tell them about moving exiting tasks or | |
7040 | * kernel threads (both mm NULL), since they never | |
7041 | * leave kernel. | |
7042 | */ | |
7043 | if (p->mm && printk_ratelimit()) { | |
7044 | printk(KERN_INFO "process %d (%s) no " | |
7045 | "longer affine to cpu%d\n", | |
7046 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7047 | } |
e76bd8d9 RR |
7048 | } |
7049 | ||
7050 | move: | |
7051 | /* It can have affinity changed while we were choosing. */ | |
7052 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7053 | goto again; | |
1da177e4 LT |
7054 | } |
7055 | ||
7056 | /* | |
7057 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7058 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7059 | * for performance reasons the counter is not stricly tracking tasks to | |
7060 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7061 | * to keep the global sum constant after CPU-down: | |
7062 | */ | |
70b97a7f | 7063 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7064 | { |
1e5ce4f4 | 7065 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7066 | unsigned long flags; |
7067 | ||
7068 | local_irq_save(flags); | |
7069 | double_rq_lock(rq_src, rq_dest); | |
7070 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7071 | rq_src->nr_uninterruptible = 0; | |
7072 | double_rq_unlock(rq_src, rq_dest); | |
7073 | local_irq_restore(flags); | |
7074 | } | |
7075 | ||
7076 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7077 | static void migrate_live_tasks(int src_cpu) | |
7078 | { | |
48f24c4d | 7079 | struct task_struct *p, *t; |
1da177e4 | 7080 | |
f7b4cddc | 7081 | read_lock(&tasklist_lock); |
1da177e4 | 7082 | |
48f24c4d IM |
7083 | do_each_thread(t, p) { |
7084 | if (p == current) | |
1da177e4 LT |
7085 | continue; |
7086 | ||
48f24c4d IM |
7087 | if (task_cpu(p) == src_cpu) |
7088 | move_task_off_dead_cpu(src_cpu, p); | |
7089 | } while_each_thread(t, p); | |
1da177e4 | 7090 | |
f7b4cddc | 7091 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7092 | } |
7093 | ||
dd41f596 IM |
7094 | /* |
7095 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7096 | * It does so by boosting its priority to highest possible. |
7097 | * Used by CPU offline code. | |
1da177e4 LT |
7098 | */ |
7099 | void sched_idle_next(void) | |
7100 | { | |
48f24c4d | 7101 | int this_cpu = smp_processor_id(); |
70b97a7f | 7102 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7103 | struct task_struct *p = rq->idle; |
7104 | unsigned long flags; | |
7105 | ||
7106 | /* cpu has to be offline */ | |
48f24c4d | 7107 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7108 | |
48f24c4d IM |
7109 | /* |
7110 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7111 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7112 | */ |
7113 | spin_lock_irqsave(&rq->lock, flags); | |
7114 | ||
dd41f596 | 7115 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7116 | |
94bc9a7b DA |
7117 | update_rq_clock(rq); |
7118 | activate_task(rq, p, 0); | |
1da177e4 LT |
7119 | |
7120 | spin_unlock_irqrestore(&rq->lock, flags); | |
7121 | } | |
7122 | ||
48f24c4d IM |
7123 | /* |
7124 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7125 | * offline. |
7126 | */ | |
7127 | void idle_task_exit(void) | |
7128 | { | |
7129 | struct mm_struct *mm = current->active_mm; | |
7130 | ||
7131 | BUG_ON(cpu_online(smp_processor_id())); | |
7132 | ||
7133 | if (mm != &init_mm) | |
7134 | switch_mm(mm, &init_mm, current); | |
7135 | mmdrop(mm); | |
7136 | } | |
7137 | ||
054b9108 | 7138 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7139 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7140 | { |
70b97a7f | 7141 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7142 | |
7143 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7144 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7145 | |
7146 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7147 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7148 | |
48f24c4d | 7149 | get_task_struct(p); |
1da177e4 LT |
7150 | |
7151 | /* | |
7152 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7153 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7154 | * fine. |
7155 | */ | |
f7b4cddc | 7156 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7157 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7158 | spin_lock_irq(&rq->lock); |
1da177e4 | 7159 | |
48f24c4d | 7160 | put_task_struct(p); |
1da177e4 LT |
7161 | } |
7162 | ||
7163 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7164 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7165 | { | |
70b97a7f | 7166 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7167 | struct task_struct *next; |
48f24c4d | 7168 | |
dd41f596 IM |
7169 | for ( ; ; ) { |
7170 | if (!rq->nr_running) | |
7171 | break; | |
a8e504d2 | 7172 | update_rq_clock(rq); |
b67802ea | 7173 | next = pick_next_task(rq); |
dd41f596 IM |
7174 | if (!next) |
7175 | break; | |
79c53799 | 7176 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7177 | migrate_dead(dead_cpu, next); |
e692ab53 | 7178 | |
1da177e4 LT |
7179 | } |
7180 | } | |
dce48a84 TG |
7181 | |
7182 | /* | |
7183 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7184 | */ | |
7185 | static void calc_global_load_remove(struct rq *rq) | |
7186 | { | |
7187 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7188 | } | |
1da177e4 LT |
7189 | #endif /* CONFIG_HOTPLUG_CPU */ |
7190 | ||
e692ab53 NP |
7191 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7192 | ||
7193 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7194 | { |
7195 | .procname = "sched_domain", | |
c57baf1e | 7196 | .mode = 0555, |
e0361851 | 7197 | }, |
38605cae | 7198 | {0, }, |
e692ab53 NP |
7199 | }; |
7200 | ||
7201 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7202 | { |
c57baf1e | 7203 | .ctl_name = CTL_KERN, |
e0361851 | 7204 | .procname = "kernel", |
c57baf1e | 7205 | .mode = 0555, |
e0361851 AD |
7206 | .child = sd_ctl_dir, |
7207 | }, | |
38605cae | 7208 | {0, }, |
e692ab53 NP |
7209 | }; |
7210 | ||
7211 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7212 | { | |
7213 | struct ctl_table *entry = | |
5cf9f062 | 7214 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7215 | |
e692ab53 NP |
7216 | return entry; |
7217 | } | |
7218 | ||
6382bc90 MM |
7219 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7220 | { | |
cd790076 | 7221 | struct ctl_table *entry; |
6382bc90 | 7222 | |
cd790076 MM |
7223 | /* |
7224 | * In the intermediate directories, both the child directory and | |
7225 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7226 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7227 | * static strings and all have proc handlers. |
7228 | */ | |
7229 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7230 | if (entry->child) |
7231 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7232 | if (entry->proc_handler == NULL) |
7233 | kfree(entry->procname); | |
7234 | } | |
6382bc90 MM |
7235 | |
7236 | kfree(*tablep); | |
7237 | *tablep = NULL; | |
7238 | } | |
7239 | ||
e692ab53 | 7240 | static void |
e0361851 | 7241 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7242 | const char *procname, void *data, int maxlen, |
7243 | mode_t mode, proc_handler *proc_handler) | |
7244 | { | |
e692ab53 NP |
7245 | entry->procname = procname; |
7246 | entry->data = data; | |
7247 | entry->maxlen = maxlen; | |
7248 | entry->mode = mode; | |
7249 | entry->proc_handler = proc_handler; | |
7250 | } | |
7251 | ||
7252 | static struct ctl_table * | |
7253 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7254 | { | |
a5d8c348 | 7255 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7256 | |
ad1cdc1d MM |
7257 | if (table == NULL) |
7258 | return NULL; | |
7259 | ||
e0361851 | 7260 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7261 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7262 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7263 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7264 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7265 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7266 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7267 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7268 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7269 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7270 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7271 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7272 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7273 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7274 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7275 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7276 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7277 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7278 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7279 | &sd->cache_nice_tries, |
7280 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7281 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7282 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7283 | set_table_entry(&table[11], "name", sd->name, |
7284 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7285 | /* &table[12] is terminator */ | |
e692ab53 NP |
7286 | |
7287 | return table; | |
7288 | } | |
7289 | ||
9a4e7159 | 7290 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7291 | { |
7292 | struct ctl_table *entry, *table; | |
7293 | struct sched_domain *sd; | |
7294 | int domain_num = 0, i; | |
7295 | char buf[32]; | |
7296 | ||
7297 | for_each_domain(cpu, sd) | |
7298 | domain_num++; | |
7299 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7300 | if (table == NULL) |
7301 | return NULL; | |
e692ab53 NP |
7302 | |
7303 | i = 0; | |
7304 | for_each_domain(cpu, sd) { | |
7305 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7306 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7307 | entry->mode = 0555; |
e692ab53 NP |
7308 | entry->child = sd_alloc_ctl_domain_table(sd); |
7309 | entry++; | |
7310 | i++; | |
7311 | } | |
7312 | return table; | |
7313 | } | |
7314 | ||
7315 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7316 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7317 | { |
7318 | int i, cpu_num = num_online_cpus(); | |
7319 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7320 | char buf[32]; | |
7321 | ||
7378547f MM |
7322 | WARN_ON(sd_ctl_dir[0].child); |
7323 | sd_ctl_dir[0].child = entry; | |
7324 | ||
ad1cdc1d MM |
7325 | if (entry == NULL) |
7326 | return; | |
7327 | ||
97b6ea7b | 7328 | for_each_online_cpu(i) { |
e692ab53 | 7329 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7330 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7331 | entry->mode = 0555; |
e692ab53 | 7332 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7333 | entry++; |
e692ab53 | 7334 | } |
7378547f MM |
7335 | |
7336 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7337 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7338 | } | |
6382bc90 | 7339 | |
7378547f | 7340 | /* may be called multiple times per register */ |
6382bc90 MM |
7341 | static void unregister_sched_domain_sysctl(void) |
7342 | { | |
7378547f MM |
7343 | if (sd_sysctl_header) |
7344 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7345 | sd_sysctl_header = NULL; |
7378547f MM |
7346 | if (sd_ctl_dir[0].child) |
7347 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7348 | } |
e692ab53 | 7349 | #else |
6382bc90 MM |
7350 | static void register_sched_domain_sysctl(void) |
7351 | { | |
7352 | } | |
7353 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7354 | { |
7355 | } | |
7356 | #endif | |
7357 | ||
1f11eb6a GH |
7358 | static void set_rq_online(struct rq *rq) |
7359 | { | |
7360 | if (!rq->online) { | |
7361 | const struct sched_class *class; | |
7362 | ||
c6c4927b | 7363 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7364 | rq->online = 1; |
7365 | ||
7366 | for_each_class(class) { | |
7367 | if (class->rq_online) | |
7368 | class->rq_online(rq); | |
7369 | } | |
7370 | } | |
7371 | } | |
7372 | ||
7373 | static void set_rq_offline(struct rq *rq) | |
7374 | { | |
7375 | if (rq->online) { | |
7376 | const struct sched_class *class; | |
7377 | ||
7378 | for_each_class(class) { | |
7379 | if (class->rq_offline) | |
7380 | class->rq_offline(rq); | |
7381 | } | |
7382 | ||
c6c4927b | 7383 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7384 | rq->online = 0; |
7385 | } | |
7386 | } | |
7387 | ||
1da177e4 LT |
7388 | /* |
7389 | * migration_call - callback that gets triggered when a CPU is added. | |
7390 | * Here we can start up the necessary migration thread for the new CPU. | |
7391 | */ | |
48f24c4d IM |
7392 | static int __cpuinit |
7393 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7394 | { |
1da177e4 | 7395 | struct task_struct *p; |
48f24c4d | 7396 | int cpu = (long)hcpu; |
1da177e4 | 7397 | unsigned long flags; |
70b97a7f | 7398 | struct rq *rq; |
1da177e4 LT |
7399 | |
7400 | switch (action) { | |
5be9361c | 7401 | |
1da177e4 | 7402 | case CPU_UP_PREPARE: |
8bb78442 | 7403 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7404 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7405 | if (IS_ERR(p)) |
7406 | return NOTIFY_BAD; | |
1da177e4 LT |
7407 | kthread_bind(p, cpu); |
7408 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7409 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7410 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7411 | task_rq_unlock(rq, &flags); |
7412 | cpu_rq(cpu)->migration_thread = p; | |
7413 | break; | |
48f24c4d | 7414 | |
1da177e4 | 7415 | case CPU_ONLINE: |
8bb78442 | 7416 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7417 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7418 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7419 | |
7420 | /* Update our root-domain */ | |
7421 | rq = cpu_rq(cpu); | |
7422 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7423 | rq->calc_load_update = calc_load_update; |
7424 | rq->calc_load_active = 0; | |
1f94ef59 | 7425 | if (rq->rd) { |
c6c4927b | 7426 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7427 | |
7428 | set_rq_online(rq); | |
1f94ef59 GH |
7429 | } |
7430 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7431 | break; |
48f24c4d | 7432 | |
1da177e4 LT |
7433 | #ifdef CONFIG_HOTPLUG_CPU |
7434 | case CPU_UP_CANCELED: | |
8bb78442 | 7435 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7436 | if (!cpu_rq(cpu)->migration_thread) |
7437 | break; | |
41a2d6cf | 7438 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7439 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7440 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7441 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7442 | cpu_rq(cpu)->migration_thread = NULL; | |
7443 | break; | |
48f24c4d | 7444 | |
1da177e4 | 7445 | case CPU_DEAD: |
8bb78442 | 7446 | case CPU_DEAD_FROZEN: |
470fd646 | 7447 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7448 | migrate_live_tasks(cpu); |
7449 | rq = cpu_rq(cpu); | |
7450 | kthread_stop(rq->migration_thread); | |
7451 | rq->migration_thread = NULL; | |
7452 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7453 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7454 | update_rq_clock(rq); |
2e1cb74a | 7455 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7456 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7457 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7458 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7459 | migrate_dead_tasks(cpu); |
d2da272a | 7460 | spin_unlock_irq(&rq->lock); |
470fd646 | 7461 | cpuset_unlock(); |
1da177e4 LT |
7462 | migrate_nr_uninterruptible(rq); |
7463 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7464 | calc_global_load_remove(rq); |
41a2d6cf IM |
7465 | /* |
7466 | * No need to migrate the tasks: it was best-effort if | |
7467 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7468 | * the requestors. | |
7469 | */ | |
1da177e4 LT |
7470 | spin_lock_irq(&rq->lock); |
7471 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7472 | struct migration_req *req; |
7473 | ||
1da177e4 | 7474 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7475 | struct migration_req, list); |
1da177e4 | 7476 | list_del_init(&req->list); |
9a2bd244 | 7477 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7478 | complete(&req->done); |
9a2bd244 | 7479 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7480 | } |
7481 | spin_unlock_irq(&rq->lock); | |
7482 | break; | |
57d885fe | 7483 | |
08f503b0 GH |
7484 | case CPU_DYING: |
7485 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7486 | /* Update our root-domain */ |
7487 | rq = cpu_rq(cpu); | |
7488 | spin_lock_irqsave(&rq->lock, flags); | |
7489 | if (rq->rd) { | |
c6c4927b | 7490 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7491 | set_rq_offline(rq); |
57d885fe GH |
7492 | } |
7493 | spin_unlock_irqrestore(&rq->lock, flags); | |
7494 | break; | |
1da177e4 LT |
7495 | #endif |
7496 | } | |
7497 | return NOTIFY_OK; | |
7498 | } | |
7499 | ||
7500 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
7501 | * happens before everything else. | |
7502 | */ | |
26c2143b | 7503 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7504 | .notifier_call = migration_call, |
7505 | .priority = 10 | |
7506 | }; | |
7507 | ||
7babe8db | 7508 | static int __init migration_init(void) |
1da177e4 LT |
7509 | { |
7510 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7511 | int err; |
48f24c4d IM |
7512 | |
7513 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7514 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7515 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7516 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7517 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7518 | |
7519 | return err; | |
1da177e4 | 7520 | } |
7babe8db | 7521 | early_initcall(migration_init); |
1da177e4 LT |
7522 | #endif |
7523 | ||
7524 | #ifdef CONFIG_SMP | |
476f3534 | 7525 | |
3e9830dc | 7526 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7527 | |
7c16ec58 | 7528 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7529 | struct cpumask *groupmask) |
1da177e4 | 7530 | { |
4dcf6aff | 7531 | struct sched_group *group = sd->groups; |
434d53b0 | 7532 | char str[256]; |
1da177e4 | 7533 | |
968ea6d8 | 7534 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7535 | cpumask_clear(groupmask); |
4dcf6aff IM |
7536 | |
7537 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7538 | ||
7539 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7540 | printk("does not load-balance\n"); | |
7541 | if (sd->parent) | |
7542 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7543 | " has parent"); | |
7544 | return -1; | |
41c7ce9a NP |
7545 | } |
7546 | ||
eefd796a | 7547 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7548 | |
758b2cdc | 7549 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7550 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7551 | "CPU%d\n", cpu); | |
7552 | } | |
758b2cdc | 7553 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7554 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7555 | " CPU%d\n", cpu); | |
7556 | } | |
1da177e4 | 7557 | |
4dcf6aff | 7558 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7559 | do { |
4dcf6aff IM |
7560 | if (!group) { |
7561 | printk("\n"); | |
7562 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7563 | break; |
7564 | } | |
7565 | ||
4dcf6aff IM |
7566 | if (!group->__cpu_power) { |
7567 | printk(KERN_CONT "\n"); | |
7568 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7569 | "set\n"); | |
7570 | break; | |
7571 | } | |
1da177e4 | 7572 | |
758b2cdc | 7573 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7574 | printk(KERN_CONT "\n"); |
7575 | printk(KERN_ERR "ERROR: empty group\n"); | |
7576 | break; | |
7577 | } | |
1da177e4 | 7578 | |
758b2cdc | 7579 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7580 | printk(KERN_CONT "\n"); |
7581 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7582 | break; | |
7583 | } | |
1da177e4 | 7584 | |
758b2cdc | 7585 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7586 | |
968ea6d8 | 7587 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7588 | |
7589 | printk(KERN_CONT " %s", str); | |
7590 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7591 | printk(KERN_CONT " (__cpu_power = %d)", | |
7592 | group->__cpu_power); | |
7593 | } | |
1da177e4 | 7594 | |
4dcf6aff IM |
7595 | group = group->next; |
7596 | } while (group != sd->groups); | |
7597 | printk(KERN_CONT "\n"); | |
1da177e4 | 7598 | |
758b2cdc | 7599 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7600 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7601 | |
758b2cdc RR |
7602 | if (sd->parent && |
7603 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7604 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7605 | "of domain->span\n"); | |
7606 | return 0; | |
7607 | } | |
1da177e4 | 7608 | |
4dcf6aff IM |
7609 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7610 | { | |
d5dd3db1 | 7611 | cpumask_var_t groupmask; |
4dcf6aff | 7612 | int level = 0; |
1da177e4 | 7613 | |
4dcf6aff IM |
7614 | if (!sd) { |
7615 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7616 | return; | |
7617 | } | |
1da177e4 | 7618 | |
4dcf6aff IM |
7619 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7620 | ||
d5dd3db1 | 7621 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7622 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7623 | return; | |
7624 | } | |
7625 | ||
4dcf6aff | 7626 | for (;;) { |
7c16ec58 | 7627 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7628 | break; |
1da177e4 LT |
7629 | level++; |
7630 | sd = sd->parent; | |
33859f7f | 7631 | if (!sd) |
4dcf6aff IM |
7632 | break; |
7633 | } | |
d5dd3db1 | 7634 | free_cpumask_var(groupmask); |
1da177e4 | 7635 | } |
6d6bc0ad | 7636 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7637 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7638 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7639 | |
1a20ff27 | 7640 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7641 | { |
758b2cdc | 7642 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7643 | return 1; |
7644 | ||
7645 | /* Following flags need at least 2 groups */ | |
7646 | if (sd->flags & (SD_LOAD_BALANCE | | |
7647 | SD_BALANCE_NEWIDLE | | |
7648 | SD_BALANCE_FORK | | |
89c4710e SS |
7649 | SD_BALANCE_EXEC | |
7650 | SD_SHARE_CPUPOWER | | |
7651 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7652 | if (sd->groups != sd->groups->next) |
7653 | return 0; | |
7654 | } | |
7655 | ||
7656 | /* Following flags don't use groups */ | |
7657 | if (sd->flags & (SD_WAKE_IDLE | | |
7658 | SD_WAKE_AFFINE | | |
7659 | SD_WAKE_BALANCE)) | |
7660 | return 0; | |
7661 | ||
7662 | return 1; | |
7663 | } | |
7664 | ||
48f24c4d IM |
7665 | static int |
7666 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7667 | { |
7668 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7669 | ||
7670 | if (sd_degenerate(parent)) | |
7671 | return 1; | |
7672 | ||
758b2cdc | 7673 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7674 | return 0; |
7675 | ||
7676 | /* Does parent contain flags not in child? */ | |
7677 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7678 | if (cflags & SD_WAKE_AFFINE) | |
7679 | pflags &= ~SD_WAKE_BALANCE; | |
7680 | /* Flags needing groups don't count if only 1 group in parent */ | |
7681 | if (parent->groups == parent->groups->next) { | |
7682 | pflags &= ~(SD_LOAD_BALANCE | | |
7683 | SD_BALANCE_NEWIDLE | | |
7684 | SD_BALANCE_FORK | | |
89c4710e SS |
7685 | SD_BALANCE_EXEC | |
7686 | SD_SHARE_CPUPOWER | | |
7687 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7688 | if (nr_node_ids == 1) |
7689 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7690 | } |
7691 | if (~cflags & pflags) | |
7692 | return 0; | |
7693 | ||
7694 | return 1; | |
7695 | } | |
7696 | ||
c6c4927b RR |
7697 | static void free_rootdomain(struct root_domain *rd) |
7698 | { | |
68e74568 RR |
7699 | cpupri_cleanup(&rd->cpupri); |
7700 | ||
c6c4927b RR |
7701 | free_cpumask_var(rd->rto_mask); |
7702 | free_cpumask_var(rd->online); | |
7703 | free_cpumask_var(rd->span); | |
7704 | kfree(rd); | |
7705 | } | |
7706 | ||
57d885fe GH |
7707 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7708 | { | |
a0490fa3 | 7709 | struct root_domain *old_rd = NULL; |
57d885fe | 7710 | unsigned long flags; |
57d885fe GH |
7711 | |
7712 | spin_lock_irqsave(&rq->lock, flags); | |
7713 | ||
7714 | if (rq->rd) { | |
a0490fa3 | 7715 | old_rd = rq->rd; |
57d885fe | 7716 | |
c6c4927b | 7717 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7718 | set_rq_offline(rq); |
57d885fe | 7719 | |
c6c4927b | 7720 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7721 | |
a0490fa3 IM |
7722 | /* |
7723 | * If we dont want to free the old_rt yet then | |
7724 | * set old_rd to NULL to skip the freeing later | |
7725 | * in this function: | |
7726 | */ | |
7727 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7728 | old_rd = NULL; | |
57d885fe GH |
7729 | } |
7730 | ||
7731 | atomic_inc(&rd->refcount); | |
7732 | rq->rd = rd; | |
7733 | ||
c6c4927b RR |
7734 | cpumask_set_cpu(rq->cpu, rd->span); |
7735 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7736 | set_rq_online(rq); |
57d885fe GH |
7737 | |
7738 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7739 | |
7740 | if (old_rd) | |
7741 | free_rootdomain(old_rd); | |
57d885fe GH |
7742 | } |
7743 | ||
db2f59c8 | 7744 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7745 | { |
7746 | memset(rd, 0, sizeof(*rd)); | |
7747 | ||
c6c4927b RR |
7748 | if (bootmem) { |
7749 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7750 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7751 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7752 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7753 | return 0; |
7754 | } | |
7755 | ||
7756 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7757 | goto out; |
c6c4927b RR |
7758 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7759 | goto free_span; | |
7760 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7761 | goto free_online; | |
6e0534f2 | 7762 | |
68e74568 RR |
7763 | if (cpupri_init(&rd->cpupri, false) != 0) |
7764 | goto free_rto_mask; | |
c6c4927b | 7765 | return 0; |
6e0534f2 | 7766 | |
68e74568 RR |
7767 | free_rto_mask: |
7768 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7769 | free_online: |
7770 | free_cpumask_var(rd->online); | |
7771 | free_span: | |
7772 | free_cpumask_var(rd->span); | |
0c910d28 | 7773 | out: |
c6c4927b | 7774 | return -ENOMEM; |
57d885fe GH |
7775 | } |
7776 | ||
7777 | static void init_defrootdomain(void) | |
7778 | { | |
c6c4927b RR |
7779 | init_rootdomain(&def_root_domain, true); |
7780 | ||
57d885fe GH |
7781 | atomic_set(&def_root_domain.refcount, 1); |
7782 | } | |
7783 | ||
dc938520 | 7784 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7785 | { |
7786 | struct root_domain *rd; | |
7787 | ||
7788 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7789 | if (!rd) | |
7790 | return NULL; | |
7791 | ||
c6c4927b RR |
7792 | if (init_rootdomain(rd, false) != 0) { |
7793 | kfree(rd); | |
7794 | return NULL; | |
7795 | } | |
57d885fe GH |
7796 | |
7797 | return rd; | |
7798 | } | |
7799 | ||
1da177e4 | 7800 | /* |
0eab9146 | 7801 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7802 | * hold the hotplug lock. |
7803 | */ | |
0eab9146 IM |
7804 | static void |
7805 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7806 | { |
70b97a7f | 7807 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7808 | struct sched_domain *tmp; |
7809 | ||
7810 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7811 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7812 | struct sched_domain *parent = tmp->parent; |
7813 | if (!parent) | |
7814 | break; | |
f29c9b1c | 7815 | |
1a848870 | 7816 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7817 | tmp->parent = parent->parent; |
1a848870 SS |
7818 | if (parent->parent) |
7819 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7820 | } else |
7821 | tmp = tmp->parent; | |
245af2c7 SS |
7822 | } |
7823 | ||
1a848870 | 7824 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7825 | sd = sd->parent; |
1a848870 SS |
7826 | if (sd) |
7827 | sd->child = NULL; | |
7828 | } | |
1da177e4 LT |
7829 | |
7830 | sched_domain_debug(sd, cpu); | |
7831 | ||
57d885fe | 7832 | rq_attach_root(rq, rd); |
674311d5 | 7833 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7834 | } |
7835 | ||
7836 | /* cpus with isolated domains */ | |
dcc30a35 | 7837 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7838 | |
7839 | /* Setup the mask of cpus configured for isolated domains */ | |
7840 | static int __init isolated_cpu_setup(char *str) | |
7841 | { | |
968ea6d8 | 7842 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7843 | return 1; |
7844 | } | |
7845 | ||
8927f494 | 7846 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7847 | |
7848 | /* | |
6711cab4 SS |
7849 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7850 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7851 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7852 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7853 | * |
7854 | * init_sched_build_groups will build a circular linked list of the groups | |
7855 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7856 | * and ->cpu_power to 0. | |
7857 | */ | |
a616058b | 7858 | static void |
96f874e2 RR |
7859 | init_sched_build_groups(const struct cpumask *span, |
7860 | const struct cpumask *cpu_map, | |
7861 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7862 | struct sched_group **sg, |
96f874e2 RR |
7863 | struct cpumask *tmpmask), |
7864 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7865 | { |
7866 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7867 | int i; |
7868 | ||
96f874e2 | 7869 | cpumask_clear(covered); |
7c16ec58 | 7870 | |
abcd083a | 7871 | for_each_cpu(i, span) { |
6711cab4 | 7872 | struct sched_group *sg; |
7c16ec58 | 7873 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7874 | int j; |
7875 | ||
758b2cdc | 7876 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7877 | continue; |
7878 | ||
758b2cdc | 7879 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7880 | sg->__cpu_power = 0; |
1da177e4 | 7881 | |
abcd083a | 7882 | for_each_cpu(j, span) { |
7c16ec58 | 7883 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7884 | continue; |
7885 | ||
96f874e2 | 7886 | cpumask_set_cpu(j, covered); |
758b2cdc | 7887 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7888 | } |
7889 | if (!first) | |
7890 | first = sg; | |
7891 | if (last) | |
7892 | last->next = sg; | |
7893 | last = sg; | |
7894 | } | |
7895 | last->next = first; | |
7896 | } | |
7897 | ||
9c1cfda2 | 7898 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7899 | |
9c1cfda2 | 7900 | #ifdef CONFIG_NUMA |
198e2f18 | 7901 | |
9c1cfda2 JH |
7902 | /** |
7903 | * find_next_best_node - find the next node to include in a sched_domain | |
7904 | * @node: node whose sched_domain we're building | |
7905 | * @used_nodes: nodes already in the sched_domain | |
7906 | * | |
41a2d6cf | 7907 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7908 | * finds the closest node not already in the @used_nodes map. |
7909 | * | |
7910 | * Should use nodemask_t. | |
7911 | */ | |
c5f59f08 | 7912 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7913 | { |
7914 | int i, n, val, min_val, best_node = 0; | |
7915 | ||
7916 | min_val = INT_MAX; | |
7917 | ||
076ac2af | 7918 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7919 | /* Start at @node */ |
076ac2af | 7920 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7921 | |
7922 | if (!nr_cpus_node(n)) | |
7923 | continue; | |
7924 | ||
7925 | /* Skip already used nodes */ | |
c5f59f08 | 7926 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7927 | continue; |
7928 | ||
7929 | /* Simple min distance search */ | |
7930 | val = node_distance(node, n); | |
7931 | ||
7932 | if (val < min_val) { | |
7933 | min_val = val; | |
7934 | best_node = n; | |
7935 | } | |
7936 | } | |
7937 | ||
c5f59f08 | 7938 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7939 | return best_node; |
7940 | } | |
7941 | ||
7942 | /** | |
7943 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7944 | * @node: node whose cpumask we're constructing | |
73486722 | 7945 | * @span: resulting cpumask |
9c1cfda2 | 7946 | * |
41a2d6cf | 7947 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7948 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7949 | * out optimally. | |
7950 | */ | |
96f874e2 | 7951 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7952 | { |
c5f59f08 | 7953 | nodemask_t used_nodes; |
48f24c4d | 7954 | int i; |
9c1cfda2 | 7955 | |
6ca09dfc | 7956 | cpumask_clear(span); |
c5f59f08 | 7957 | nodes_clear(used_nodes); |
9c1cfda2 | 7958 | |
6ca09dfc | 7959 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7960 | node_set(node, used_nodes); |
9c1cfda2 JH |
7961 | |
7962 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7963 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7964 | |
6ca09dfc | 7965 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7966 | } |
9c1cfda2 | 7967 | } |
6d6bc0ad | 7968 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7969 | |
5c45bf27 | 7970 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7971 | |
6c99e9ad RR |
7972 | /* |
7973 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
7974 | * |
7975 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
7976 | * and struct sched_domain. ) | |
6c99e9ad RR |
7977 | */ |
7978 | struct static_sched_group { | |
7979 | struct sched_group sg; | |
7980 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7981 | }; | |
7982 | ||
7983 | struct static_sched_domain { | |
7984 | struct sched_domain sd; | |
7985 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7986 | }; | |
7987 | ||
9c1cfda2 | 7988 | /* |
48f24c4d | 7989 | * SMT sched-domains: |
9c1cfda2 | 7990 | */ |
1da177e4 | 7991 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7992 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7993 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7994 | |
41a2d6cf | 7995 | static int |
96f874e2 RR |
7996 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7997 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7998 | { |
6711cab4 | 7999 | if (sg) |
6c99e9ad | 8000 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8001 | return cpu; |
8002 | } | |
6d6bc0ad | 8003 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8004 | |
48f24c4d IM |
8005 | /* |
8006 | * multi-core sched-domains: | |
8007 | */ | |
1e9f28fa | 8008 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8009 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8010 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8011 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8012 | |
8013 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8014 | static int |
96f874e2 RR |
8015 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8016 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8017 | { |
6711cab4 | 8018 | int group; |
7c16ec58 | 8019 | |
c69fc56d | 8020 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8021 | group = cpumask_first(mask); |
6711cab4 | 8022 | if (sg) |
6c99e9ad | 8023 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8024 | return group; |
1e9f28fa SS |
8025 | } |
8026 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8027 | static int |
96f874e2 RR |
8028 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8029 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8030 | { |
6711cab4 | 8031 | if (sg) |
6c99e9ad | 8032 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8033 | return cpu; |
8034 | } | |
8035 | #endif | |
8036 | ||
6c99e9ad RR |
8037 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8038 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8039 | |
41a2d6cf | 8040 | static int |
96f874e2 RR |
8041 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8042 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8043 | { |
6711cab4 | 8044 | int group; |
48f24c4d | 8045 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8046 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8047 | group = cpumask_first(mask); |
1e9f28fa | 8048 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8049 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8050 | group = cpumask_first(mask); |
1da177e4 | 8051 | #else |
6711cab4 | 8052 | group = cpu; |
1da177e4 | 8053 | #endif |
6711cab4 | 8054 | if (sg) |
6c99e9ad | 8055 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8056 | return group; |
1da177e4 LT |
8057 | } |
8058 | ||
8059 | #ifdef CONFIG_NUMA | |
1da177e4 | 8060 | /* |
9c1cfda2 JH |
8061 | * The init_sched_build_groups can't handle what we want to do with node |
8062 | * groups, so roll our own. Now each node has its own list of groups which | |
8063 | * gets dynamically allocated. | |
1da177e4 | 8064 | */ |
62ea9ceb | 8065 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8066 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8067 | |
62ea9ceb | 8068 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8069 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8070 | |
96f874e2 RR |
8071 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8072 | struct sched_group **sg, | |
8073 | struct cpumask *nodemask) | |
9c1cfda2 | 8074 | { |
6711cab4 SS |
8075 | int group; |
8076 | ||
6ca09dfc | 8077 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8078 | group = cpumask_first(nodemask); |
6711cab4 SS |
8079 | |
8080 | if (sg) | |
6c99e9ad | 8081 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8082 | return group; |
1da177e4 | 8083 | } |
6711cab4 | 8084 | |
08069033 SS |
8085 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8086 | { | |
8087 | struct sched_group *sg = group_head; | |
8088 | int j; | |
8089 | ||
8090 | if (!sg) | |
8091 | return; | |
3a5c359a | 8092 | do { |
758b2cdc | 8093 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8094 | struct sched_domain *sd; |
08069033 | 8095 | |
6c99e9ad | 8096 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8097 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8098 | /* |
8099 | * Only add "power" once for each | |
8100 | * physical package. | |
8101 | */ | |
8102 | continue; | |
8103 | } | |
08069033 | 8104 | |
3a5c359a AK |
8105 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8106 | } | |
8107 | sg = sg->next; | |
8108 | } while (sg != group_head); | |
08069033 | 8109 | } |
6d6bc0ad | 8110 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8111 | |
a616058b | 8112 | #ifdef CONFIG_NUMA |
51888ca2 | 8113 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8114 | static void free_sched_groups(const struct cpumask *cpu_map, |
8115 | struct cpumask *nodemask) | |
51888ca2 | 8116 | { |
a616058b | 8117 | int cpu, i; |
51888ca2 | 8118 | |
abcd083a | 8119 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8120 | struct sched_group **sched_group_nodes |
8121 | = sched_group_nodes_bycpu[cpu]; | |
8122 | ||
51888ca2 SV |
8123 | if (!sched_group_nodes) |
8124 | continue; | |
8125 | ||
076ac2af | 8126 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8127 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8128 | ||
6ca09dfc | 8129 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8130 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8131 | continue; |
8132 | ||
8133 | if (sg == NULL) | |
8134 | continue; | |
8135 | sg = sg->next; | |
8136 | next_sg: | |
8137 | oldsg = sg; | |
8138 | sg = sg->next; | |
8139 | kfree(oldsg); | |
8140 | if (oldsg != sched_group_nodes[i]) | |
8141 | goto next_sg; | |
8142 | } | |
8143 | kfree(sched_group_nodes); | |
8144 | sched_group_nodes_bycpu[cpu] = NULL; | |
8145 | } | |
51888ca2 | 8146 | } |
6d6bc0ad | 8147 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8148 | static void free_sched_groups(const struct cpumask *cpu_map, |
8149 | struct cpumask *nodemask) | |
a616058b SS |
8150 | { |
8151 | } | |
6d6bc0ad | 8152 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8153 | |
89c4710e SS |
8154 | /* |
8155 | * Initialize sched groups cpu_power. | |
8156 | * | |
8157 | * cpu_power indicates the capacity of sched group, which is used while | |
8158 | * distributing the load between different sched groups in a sched domain. | |
8159 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8160 | * there are asymmetries in the topology. If there are asymmetries, group | |
8161 | * having more cpu_power will pickup more load compared to the group having | |
8162 | * less cpu_power. | |
8163 | * | |
8164 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8165 | * the maximum number of tasks a group can handle in the presence of other idle | |
8166 | * or lightly loaded groups in the same sched domain. | |
8167 | */ | |
8168 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8169 | { | |
8170 | struct sched_domain *child; | |
8171 | struct sched_group *group; | |
8172 | ||
8173 | WARN_ON(!sd || !sd->groups); | |
8174 | ||
13318a71 | 8175 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8176 | return; |
8177 | ||
8178 | child = sd->child; | |
8179 | ||
5517d86b ED |
8180 | sd->groups->__cpu_power = 0; |
8181 | ||
89c4710e SS |
8182 | /* |
8183 | * For perf policy, if the groups in child domain share resources | |
8184 | * (for example cores sharing some portions of the cache hierarchy | |
8185 | * or SMT), then set this domain groups cpu_power such that each group | |
8186 | * can handle only one task, when there are other idle groups in the | |
8187 | * same sched domain. | |
8188 | */ | |
8189 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8190 | (child->flags & | |
8191 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8192 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8193 | return; |
8194 | } | |
8195 | ||
89c4710e SS |
8196 | /* |
8197 | * add cpu_power of each child group to this groups cpu_power | |
8198 | */ | |
8199 | group = child->groups; | |
8200 | do { | |
5517d86b | 8201 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8202 | group = group->next; |
8203 | } while (group != child->groups); | |
8204 | } | |
8205 | ||
7c16ec58 MT |
8206 | /* |
8207 | * Initializers for schedule domains | |
8208 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8209 | */ | |
8210 | ||
a5d8c348 IM |
8211 | #ifdef CONFIG_SCHED_DEBUG |
8212 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8213 | #else | |
8214 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8215 | #endif | |
8216 | ||
7c16ec58 | 8217 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8218 | |
7c16ec58 MT |
8219 | #define SD_INIT_FUNC(type) \ |
8220 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8221 | { \ | |
8222 | memset(sd, 0, sizeof(*sd)); \ | |
8223 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8224 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8225 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8226 | } |
8227 | ||
8228 | SD_INIT_FUNC(CPU) | |
8229 | #ifdef CONFIG_NUMA | |
8230 | SD_INIT_FUNC(ALLNODES) | |
8231 | SD_INIT_FUNC(NODE) | |
8232 | #endif | |
8233 | #ifdef CONFIG_SCHED_SMT | |
8234 | SD_INIT_FUNC(SIBLING) | |
8235 | #endif | |
8236 | #ifdef CONFIG_SCHED_MC | |
8237 | SD_INIT_FUNC(MC) | |
8238 | #endif | |
8239 | ||
1d3504fc HS |
8240 | static int default_relax_domain_level = -1; |
8241 | ||
8242 | static int __init setup_relax_domain_level(char *str) | |
8243 | { | |
30e0e178 LZ |
8244 | unsigned long val; |
8245 | ||
8246 | val = simple_strtoul(str, NULL, 0); | |
8247 | if (val < SD_LV_MAX) | |
8248 | default_relax_domain_level = val; | |
8249 | ||
1d3504fc HS |
8250 | return 1; |
8251 | } | |
8252 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8253 | ||
8254 | static void set_domain_attribute(struct sched_domain *sd, | |
8255 | struct sched_domain_attr *attr) | |
8256 | { | |
8257 | int request; | |
8258 | ||
8259 | if (!attr || attr->relax_domain_level < 0) { | |
8260 | if (default_relax_domain_level < 0) | |
8261 | return; | |
8262 | else | |
8263 | request = default_relax_domain_level; | |
8264 | } else | |
8265 | request = attr->relax_domain_level; | |
8266 | if (request < sd->level) { | |
8267 | /* turn off idle balance on this domain */ | |
8268 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8269 | } else { | |
8270 | /* turn on idle balance on this domain */ | |
8271 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8272 | } | |
8273 | } | |
8274 | ||
1da177e4 | 8275 | /* |
1a20ff27 DG |
8276 | * Build sched domains for a given set of cpus and attach the sched domains |
8277 | * to the individual cpus | |
1da177e4 | 8278 | */ |
96f874e2 | 8279 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8280 | struct sched_domain_attr *attr) |
1da177e4 | 8281 | { |
3404c8d9 | 8282 | int i, err = -ENOMEM; |
57d885fe | 8283 | struct root_domain *rd; |
3404c8d9 RR |
8284 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8285 | tmpmask; | |
d1b55138 | 8286 | #ifdef CONFIG_NUMA |
3404c8d9 | 8287 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8288 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8289 | int sd_allnodes = 0; |
d1b55138 | 8290 | |
3404c8d9 RR |
8291 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8292 | goto out; | |
8293 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8294 | goto free_domainspan; | |
8295 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8296 | goto free_covered; | |
8297 | #endif | |
8298 | ||
8299 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8300 | goto free_notcovered; | |
8301 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8302 | goto free_nodemask; | |
8303 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8304 | goto free_this_sibling_map; | |
8305 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8306 | goto free_this_core_map; | |
8307 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8308 | goto free_send_covered; | |
8309 | ||
8310 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8311 | /* |
8312 | * Allocate the per-node list of sched groups | |
8313 | */ | |
076ac2af | 8314 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8315 | GFP_KERNEL); |
d1b55138 JH |
8316 | if (!sched_group_nodes) { |
8317 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8318 | goto free_tmpmask; |
d1b55138 | 8319 | } |
d1b55138 | 8320 | #endif |
1da177e4 | 8321 | |
dc938520 | 8322 | rd = alloc_rootdomain(); |
57d885fe GH |
8323 | if (!rd) { |
8324 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8325 | goto free_sched_groups; |
57d885fe GH |
8326 | } |
8327 | ||
7c16ec58 | 8328 | #ifdef CONFIG_NUMA |
96f874e2 | 8329 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8330 | #endif |
8331 | ||
1da177e4 | 8332 | /* |
1a20ff27 | 8333 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8334 | */ |
abcd083a | 8335 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8336 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8337 | |
6ca09dfc | 8338 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8339 | |
8340 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8341 | if (cpumask_weight(cpu_map) > |
8342 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8343 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8344 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8345 | set_domain_attribute(sd, attr); |
758b2cdc | 8346 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8347 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8348 | p = sd; |
6711cab4 | 8349 | sd_allnodes = 1; |
9c1cfda2 JH |
8350 | } else |
8351 | p = NULL; | |
8352 | ||
62ea9ceb | 8353 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8354 | SD_INIT(sd, NODE); |
1d3504fc | 8355 | set_domain_attribute(sd, attr); |
758b2cdc | 8356 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8357 | sd->parent = p; |
1a848870 SS |
8358 | if (p) |
8359 | p->child = sd; | |
758b2cdc RR |
8360 | cpumask_and(sched_domain_span(sd), |
8361 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8362 | #endif |
8363 | ||
8364 | p = sd; | |
6c99e9ad | 8365 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8366 | SD_INIT(sd, CPU); |
1d3504fc | 8367 | set_domain_attribute(sd, attr); |
758b2cdc | 8368 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8369 | sd->parent = p; |
1a848870 SS |
8370 | if (p) |
8371 | p->child = sd; | |
7c16ec58 | 8372 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8373 | |
1e9f28fa SS |
8374 | #ifdef CONFIG_SCHED_MC |
8375 | p = sd; | |
6c99e9ad | 8376 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8377 | SD_INIT(sd, MC); |
1d3504fc | 8378 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8379 | cpumask_and(sched_domain_span(sd), cpu_map, |
8380 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8381 | sd->parent = p; |
1a848870 | 8382 | p->child = sd; |
7c16ec58 | 8383 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8384 | #endif |
8385 | ||
1da177e4 LT |
8386 | #ifdef CONFIG_SCHED_SMT |
8387 | p = sd; | |
6c99e9ad | 8388 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8389 | SD_INIT(sd, SIBLING); |
1d3504fc | 8390 | set_domain_attribute(sd, attr); |
758b2cdc | 8391 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8392 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8393 | sd->parent = p; |
1a848870 | 8394 | p->child = sd; |
7c16ec58 | 8395 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8396 | #endif |
8397 | } | |
8398 | ||
8399 | #ifdef CONFIG_SCHED_SMT | |
8400 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8401 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8402 | cpumask_and(this_sibling_map, |
c69fc56d | 8403 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8404 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8405 | continue; |
8406 | ||
dd41f596 | 8407 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8408 | &cpu_to_cpu_group, |
8409 | send_covered, tmpmask); | |
1da177e4 LT |
8410 | } |
8411 | #endif | |
8412 | ||
1e9f28fa SS |
8413 | #ifdef CONFIG_SCHED_MC |
8414 | /* Set up multi-core groups */ | |
abcd083a | 8415 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8416 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8417 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8418 | continue; |
7c16ec58 | 8419 | |
dd41f596 | 8420 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8421 | &cpu_to_core_group, |
8422 | send_covered, tmpmask); | |
1e9f28fa SS |
8423 | } |
8424 | #endif | |
8425 | ||
1da177e4 | 8426 | /* Set up physical groups */ |
076ac2af | 8427 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8428 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8429 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8430 | continue; |
8431 | ||
7c16ec58 MT |
8432 | init_sched_build_groups(nodemask, cpu_map, |
8433 | &cpu_to_phys_group, | |
8434 | send_covered, tmpmask); | |
1da177e4 LT |
8435 | } |
8436 | ||
8437 | #ifdef CONFIG_NUMA | |
8438 | /* Set up node groups */ | |
7c16ec58 | 8439 | if (sd_allnodes) { |
7c16ec58 MT |
8440 | init_sched_build_groups(cpu_map, cpu_map, |
8441 | &cpu_to_allnodes_group, | |
8442 | send_covered, tmpmask); | |
8443 | } | |
9c1cfda2 | 8444 | |
076ac2af | 8445 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8446 | /* Set up node groups */ |
8447 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8448 | int j; |
8449 | ||
96f874e2 | 8450 | cpumask_clear(covered); |
6ca09dfc | 8451 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8452 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8453 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8454 | continue; |
d1b55138 | 8455 | } |
9c1cfda2 | 8456 | |
4bdbaad3 | 8457 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8458 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8459 | |
6c99e9ad RR |
8460 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8461 | GFP_KERNEL, i); | |
51888ca2 SV |
8462 | if (!sg) { |
8463 | printk(KERN_WARNING "Can not alloc domain group for " | |
8464 | "node %d\n", i); | |
8465 | goto error; | |
8466 | } | |
9c1cfda2 | 8467 | sched_group_nodes[i] = sg; |
abcd083a | 8468 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8469 | struct sched_domain *sd; |
9761eea8 | 8470 | |
62ea9ceb | 8471 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8472 | sd->groups = sg; |
9c1cfda2 | 8473 | } |
5517d86b | 8474 | sg->__cpu_power = 0; |
758b2cdc | 8475 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8476 | sg->next = sg; |
96f874e2 | 8477 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8478 | prev = sg; |
8479 | ||
076ac2af | 8480 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8481 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8482 | |
96f874e2 RR |
8483 | cpumask_complement(notcovered, covered); |
8484 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8485 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8486 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8487 | break; |
8488 | ||
6ca09dfc | 8489 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8490 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8491 | continue; |
8492 | ||
6c99e9ad RR |
8493 | sg = kmalloc_node(sizeof(struct sched_group) + |
8494 | cpumask_size(), | |
15f0b676 | 8495 | GFP_KERNEL, i); |
9c1cfda2 JH |
8496 | if (!sg) { |
8497 | printk(KERN_WARNING | |
8498 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8499 | goto error; |
9c1cfda2 | 8500 | } |
5517d86b | 8501 | sg->__cpu_power = 0; |
758b2cdc | 8502 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8503 | sg->next = prev->next; |
96f874e2 | 8504 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8505 | prev->next = sg; |
8506 | prev = sg; | |
8507 | } | |
9c1cfda2 | 8508 | } |
1da177e4 LT |
8509 | #endif |
8510 | ||
8511 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8512 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8513 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8514 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8515 | |
89c4710e | 8516 | init_sched_groups_power(i, sd); |
5c45bf27 | 8517 | } |
1da177e4 | 8518 | #endif |
1e9f28fa | 8519 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8520 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8521 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8522 | |
89c4710e | 8523 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8524 | } |
8525 | #endif | |
1e9f28fa | 8526 | |
abcd083a | 8527 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8528 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8529 | |
89c4710e | 8530 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8531 | } |
8532 | ||
9c1cfda2 | 8533 | #ifdef CONFIG_NUMA |
076ac2af | 8534 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8535 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8536 | |
6711cab4 SS |
8537 | if (sd_allnodes) { |
8538 | struct sched_group *sg; | |
f712c0c7 | 8539 | |
96f874e2 | 8540 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8541 | tmpmask); |
f712c0c7 SS |
8542 | init_numa_sched_groups_power(sg); |
8543 | } | |
9c1cfda2 JH |
8544 | #endif |
8545 | ||
1da177e4 | 8546 | /* Attach the domains */ |
abcd083a | 8547 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8548 | struct sched_domain *sd; |
8549 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8550 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8551 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8552 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8553 | #else |
6c99e9ad | 8554 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8555 | #endif |
57d885fe | 8556 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8557 | } |
51888ca2 | 8558 | |
3404c8d9 RR |
8559 | err = 0; |
8560 | ||
8561 | free_tmpmask: | |
8562 | free_cpumask_var(tmpmask); | |
8563 | free_send_covered: | |
8564 | free_cpumask_var(send_covered); | |
8565 | free_this_core_map: | |
8566 | free_cpumask_var(this_core_map); | |
8567 | free_this_sibling_map: | |
8568 | free_cpumask_var(this_sibling_map); | |
8569 | free_nodemask: | |
8570 | free_cpumask_var(nodemask); | |
8571 | free_notcovered: | |
8572 | #ifdef CONFIG_NUMA | |
8573 | free_cpumask_var(notcovered); | |
8574 | free_covered: | |
8575 | free_cpumask_var(covered); | |
8576 | free_domainspan: | |
8577 | free_cpumask_var(domainspan); | |
8578 | out: | |
8579 | #endif | |
8580 | return err; | |
8581 | ||
8582 | free_sched_groups: | |
8583 | #ifdef CONFIG_NUMA | |
8584 | kfree(sched_group_nodes); | |
8585 | #endif | |
8586 | goto free_tmpmask; | |
51888ca2 | 8587 | |
a616058b | 8588 | #ifdef CONFIG_NUMA |
51888ca2 | 8589 | error: |
7c16ec58 | 8590 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8591 | free_rootdomain(rd); |
3404c8d9 | 8592 | goto free_tmpmask; |
a616058b | 8593 | #endif |
1da177e4 | 8594 | } |
029190c5 | 8595 | |
96f874e2 | 8596 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8597 | { |
8598 | return __build_sched_domains(cpu_map, NULL); | |
8599 | } | |
8600 | ||
96f874e2 | 8601 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8602 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8603 | static struct sched_domain_attr *dattr_cur; |
8604 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8605 | |
8606 | /* | |
8607 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8608 | * cpumask) fails, then fallback to a single sched domain, |
8609 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8610 | */ |
4212823f | 8611 | static cpumask_var_t fallback_doms; |
029190c5 | 8612 | |
ee79d1bd HC |
8613 | /* |
8614 | * arch_update_cpu_topology lets virtualized architectures update the | |
8615 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8616 | * or 0 if it stayed the same. | |
8617 | */ | |
8618 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8619 | { |
ee79d1bd | 8620 | return 0; |
22e52b07 HC |
8621 | } |
8622 | ||
1a20ff27 | 8623 | /* |
41a2d6cf | 8624 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8625 | * For now this just excludes isolated cpus, but could be used to |
8626 | * exclude other special cases in the future. | |
1a20ff27 | 8627 | */ |
96f874e2 | 8628 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8629 | { |
7378547f MM |
8630 | int err; |
8631 | ||
22e52b07 | 8632 | arch_update_cpu_topology(); |
029190c5 | 8633 | ndoms_cur = 1; |
96f874e2 | 8634 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8635 | if (!doms_cur) |
4212823f | 8636 | doms_cur = fallback_doms; |
dcc30a35 | 8637 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8638 | dattr_cur = NULL; |
7378547f | 8639 | err = build_sched_domains(doms_cur); |
6382bc90 | 8640 | register_sched_domain_sysctl(); |
7378547f MM |
8641 | |
8642 | return err; | |
1a20ff27 DG |
8643 | } |
8644 | ||
96f874e2 RR |
8645 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8646 | struct cpumask *tmpmask) | |
1da177e4 | 8647 | { |
7c16ec58 | 8648 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8649 | } |
1da177e4 | 8650 | |
1a20ff27 DG |
8651 | /* |
8652 | * Detach sched domains from a group of cpus specified in cpu_map | |
8653 | * These cpus will now be attached to the NULL domain | |
8654 | */ | |
96f874e2 | 8655 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8656 | { |
96f874e2 RR |
8657 | /* Save because hotplug lock held. */ |
8658 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8659 | int i; |
8660 | ||
abcd083a | 8661 | for_each_cpu(i, cpu_map) |
57d885fe | 8662 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8663 | synchronize_sched(); |
96f874e2 | 8664 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8665 | } |
8666 | ||
1d3504fc HS |
8667 | /* handle null as "default" */ |
8668 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8669 | struct sched_domain_attr *new, int idx_new) | |
8670 | { | |
8671 | struct sched_domain_attr tmp; | |
8672 | ||
8673 | /* fast path */ | |
8674 | if (!new && !cur) | |
8675 | return 1; | |
8676 | ||
8677 | tmp = SD_ATTR_INIT; | |
8678 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8679 | new ? (new + idx_new) : &tmp, | |
8680 | sizeof(struct sched_domain_attr)); | |
8681 | } | |
8682 | ||
029190c5 PJ |
8683 | /* |
8684 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8685 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8686 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8687 | * It destroys each deleted domain and builds each new domain. | |
8688 | * | |
96f874e2 | 8689 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8690 | * The masks don't intersect (don't overlap.) We should setup one |
8691 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8692 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8693 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8694 | * it as it is. | |
8695 | * | |
41a2d6cf IM |
8696 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8697 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8698 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8699 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8700 | * the single partition 'fallback_doms', it also forces the domains | |
8701 | * to be rebuilt. | |
029190c5 | 8702 | * |
96f874e2 | 8703 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8704 | * ndoms_new == 0 is a special case for destroying existing domains, |
8705 | * and it will not create the default domain. | |
dfb512ec | 8706 | * |
029190c5 PJ |
8707 | * Call with hotplug lock held |
8708 | */ | |
96f874e2 RR |
8709 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8710 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8711 | struct sched_domain_attr *dattr_new) |
029190c5 | 8712 | { |
dfb512ec | 8713 | int i, j, n; |
d65bd5ec | 8714 | int new_topology; |
029190c5 | 8715 | |
712555ee | 8716 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8717 | |
7378547f MM |
8718 | /* always unregister in case we don't destroy any domains */ |
8719 | unregister_sched_domain_sysctl(); | |
8720 | ||
d65bd5ec HC |
8721 | /* Let architecture update cpu core mappings. */ |
8722 | new_topology = arch_update_cpu_topology(); | |
8723 | ||
dfb512ec | 8724 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8725 | |
8726 | /* Destroy deleted domains */ | |
8727 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8728 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8729 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8730 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8731 | goto match1; |
8732 | } | |
8733 | /* no match - a current sched domain not in new doms_new[] */ | |
8734 | detach_destroy_domains(doms_cur + i); | |
8735 | match1: | |
8736 | ; | |
8737 | } | |
8738 | ||
e761b772 MK |
8739 | if (doms_new == NULL) { |
8740 | ndoms_cur = 0; | |
4212823f | 8741 | doms_new = fallback_doms; |
dcc30a35 | 8742 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8743 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8744 | } |
8745 | ||
029190c5 PJ |
8746 | /* Build new domains */ |
8747 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8748 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8749 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8750 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8751 | goto match2; |
8752 | } | |
8753 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8754 | __build_sched_domains(doms_new + i, |
8755 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8756 | match2: |
8757 | ; | |
8758 | } | |
8759 | ||
8760 | /* Remember the new sched domains */ | |
4212823f | 8761 | if (doms_cur != fallback_doms) |
029190c5 | 8762 | kfree(doms_cur); |
1d3504fc | 8763 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8764 | doms_cur = doms_new; |
1d3504fc | 8765 | dattr_cur = dattr_new; |
029190c5 | 8766 | ndoms_cur = ndoms_new; |
7378547f MM |
8767 | |
8768 | register_sched_domain_sysctl(); | |
a1835615 | 8769 | |
712555ee | 8770 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8771 | } |
8772 | ||
5c45bf27 | 8773 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8774 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8775 | { |
95402b38 | 8776 | get_online_cpus(); |
dfb512ec MK |
8777 | |
8778 | /* Destroy domains first to force the rebuild */ | |
8779 | partition_sched_domains(0, NULL, NULL); | |
8780 | ||
e761b772 | 8781 | rebuild_sched_domains(); |
95402b38 | 8782 | put_online_cpus(); |
5c45bf27 SS |
8783 | } |
8784 | ||
8785 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8786 | { | |
afb8a9b7 | 8787 | unsigned int level = 0; |
5c45bf27 | 8788 | |
afb8a9b7 GS |
8789 | if (sscanf(buf, "%u", &level) != 1) |
8790 | return -EINVAL; | |
8791 | ||
8792 | /* | |
8793 | * level is always be positive so don't check for | |
8794 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8795 | * What happens on 0 or 1 byte write, | |
8796 | * need to check for count as well? | |
8797 | */ | |
8798 | ||
8799 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8800 | return -EINVAL; |
8801 | ||
8802 | if (smt) | |
afb8a9b7 | 8803 | sched_smt_power_savings = level; |
5c45bf27 | 8804 | else |
afb8a9b7 | 8805 | sched_mc_power_savings = level; |
5c45bf27 | 8806 | |
c70f22d2 | 8807 | arch_reinit_sched_domains(); |
5c45bf27 | 8808 | |
c70f22d2 | 8809 | return count; |
5c45bf27 SS |
8810 | } |
8811 | ||
5c45bf27 | 8812 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8813 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8814 | char *page) | |
5c45bf27 SS |
8815 | { |
8816 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8817 | } | |
f718cd4a | 8818 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8819 | const char *buf, size_t count) |
5c45bf27 SS |
8820 | { |
8821 | return sched_power_savings_store(buf, count, 0); | |
8822 | } | |
f718cd4a AK |
8823 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8824 | sched_mc_power_savings_show, | |
8825 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8826 | #endif |
8827 | ||
8828 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8829 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8830 | char *page) | |
5c45bf27 SS |
8831 | { |
8832 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8833 | } | |
f718cd4a | 8834 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8835 | const char *buf, size_t count) |
5c45bf27 SS |
8836 | { |
8837 | return sched_power_savings_store(buf, count, 1); | |
8838 | } | |
f718cd4a AK |
8839 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8840 | sched_smt_power_savings_show, | |
6707de00 AB |
8841 | sched_smt_power_savings_store); |
8842 | #endif | |
8843 | ||
39aac648 | 8844 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8845 | { |
8846 | int err = 0; | |
8847 | ||
8848 | #ifdef CONFIG_SCHED_SMT | |
8849 | if (smt_capable()) | |
8850 | err = sysfs_create_file(&cls->kset.kobj, | |
8851 | &attr_sched_smt_power_savings.attr); | |
8852 | #endif | |
8853 | #ifdef CONFIG_SCHED_MC | |
8854 | if (!err && mc_capable()) | |
8855 | err = sysfs_create_file(&cls->kset.kobj, | |
8856 | &attr_sched_mc_power_savings.attr); | |
8857 | #endif | |
8858 | return err; | |
8859 | } | |
6d6bc0ad | 8860 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8861 | |
e761b772 | 8862 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8863 | /* |
e761b772 MK |
8864 | * Add online and remove offline CPUs from the scheduler domains. |
8865 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8866 | */ |
8867 | static int update_sched_domains(struct notifier_block *nfb, | |
8868 | unsigned long action, void *hcpu) | |
e761b772 MK |
8869 | { |
8870 | switch (action) { | |
8871 | case CPU_ONLINE: | |
8872 | case CPU_ONLINE_FROZEN: | |
8873 | case CPU_DEAD: | |
8874 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8875 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8876 | return NOTIFY_OK; |
8877 | ||
8878 | default: | |
8879 | return NOTIFY_DONE; | |
8880 | } | |
8881 | } | |
8882 | #endif | |
8883 | ||
8884 | static int update_runtime(struct notifier_block *nfb, | |
8885 | unsigned long action, void *hcpu) | |
1da177e4 | 8886 | { |
7def2be1 PZ |
8887 | int cpu = (int)(long)hcpu; |
8888 | ||
1da177e4 | 8889 | switch (action) { |
1da177e4 | 8890 | case CPU_DOWN_PREPARE: |
8bb78442 | 8891 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8892 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8893 | return NOTIFY_OK; |
8894 | ||
1da177e4 | 8895 | case CPU_DOWN_FAILED: |
8bb78442 | 8896 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8897 | case CPU_ONLINE: |
8bb78442 | 8898 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8899 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8900 | return NOTIFY_OK; |
8901 | ||
1da177e4 LT |
8902 | default: |
8903 | return NOTIFY_DONE; | |
8904 | } | |
1da177e4 | 8905 | } |
1da177e4 LT |
8906 | |
8907 | void __init sched_init_smp(void) | |
8908 | { | |
dcc30a35 RR |
8909 | cpumask_var_t non_isolated_cpus; |
8910 | ||
8911 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8912 | |
434d53b0 MT |
8913 | #if defined(CONFIG_NUMA) |
8914 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8915 | GFP_KERNEL); | |
8916 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8917 | #endif | |
95402b38 | 8918 | get_online_cpus(); |
712555ee | 8919 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8920 | arch_init_sched_domains(cpu_online_mask); |
8921 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8922 | if (cpumask_empty(non_isolated_cpus)) | |
8923 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8924 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8925 | put_online_cpus(); |
e761b772 MK |
8926 | |
8927 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8928 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8929 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8930 | #endif |
8931 | ||
8932 | /* RT runtime code needs to handle some hotplug events */ | |
8933 | hotcpu_notifier(update_runtime, 0); | |
8934 | ||
b328ca18 | 8935 | init_hrtick(); |
5c1e1767 NP |
8936 | |
8937 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8938 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8939 | BUG(); |
19978ca6 | 8940 | sched_init_granularity(); |
dcc30a35 | 8941 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8942 | |
8943 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8944 | init_sched_rt_class(); |
1da177e4 LT |
8945 | } |
8946 | #else | |
8947 | void __init sched_init_smp(void) | |
8948 | { | |
19978ca6 | 8949 | sched_init_granularity(); |
1da177e4 LT |
8950 | } |
8951 | #endif /* CONFIG_SMP */ | |
8952 | ||
8953 | int in_sched_functions(unsigned long addr) | |
8954 | { | |
1da177e4 LT |
8955 | return in_lock_functions(addr) || |
8956 | (addr >= (unsigned long)__sched_text_start | |
8957 | && addr < (unsigned long)__sched_text_end); | |
8958 | } | |
8959 | ||
a9957449 | 8960 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8961 | { |
8962 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8963 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8964 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8965 | cfs_rq->rq = rq; | |
8966 | #endif | |
67e9fb2a | 8967 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8968 | } |
8969 | ||
fa85ae24 PZ |
8970 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8971 | { | |
8972 | struct rt_prio_array *array; | |
8973 | int i; | |
8974 | ||
8975 | array = &rt_rq->active; | |
8976 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8977 | INIT_LIST_HEAD(array->queue + i); | |
8978 | __clear_bit(i, array->bitmap); | |
8979 | } | |
8980 | /* delimiter for bitsearch: */ | |
8981 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8982 | ||
052f1dc7 | 8983 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8984 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8985 | #ifdef CONFIG_SMP |
e864c499 | 8986 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8987 | #endif |
48d5e258 | 8988 | #endif |
fa85ae24 PZ |
8989 | #ifdef CONFIG_SMP |
8990 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8991 | rt_rq->overloaded = 0; |
917b627d | 8992 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8993 | #endif |
8994 | ||
8995 | rt_rq->rt_time = 0; | |
8996 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8997 | rt_rq->rt_runtime = 0; |
8998 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8999 | |
052f1dc7 | 9000 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9001 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9002 | rt_rq->rq = rq; |
9003 | #endif | |
fa85ae24 PZ |
9004 | } |
9005 | ||
6f505b16 | 9006 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9007 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9008 | struct sched_entity *se, int cpu, int add, | |
9009 | struct sched_entity *parent) | |
6f505b16 | 9010 | { |
ec7dc8ac | 9011 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9012 | tg->cfs_rq[cpu] = cfs_rq; |
9013 | init_cfs_rq(cfs_rq, rq); | |
9014 | cfs_rq->tg = tg; | |
9015 | if (add) | |
9016 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9017 | ||
9018 | tg->se[cpu] = se; | |
354d60c2 DG |
9019 | /* se could be NULL for init_task_group */ |
9020 | if (!se) | |
9021 | return; | |
9022 | ||
ec7dc8ac DG |
9023 | if (!parent) |
9024 | se->cfs_rq = &rq->cfs; | |
9025 | else | |
9026 | se->cfs_rq = parent->my_q; | |
9027 | ||
6f505b16 PZ |
9028 | se->my_q = cfs_rq; |
9029 | se->load.weight = tg->shares; | |
e05510d0 | 9030 | se->load.inv_weight = 0; |
ec7dc8ac | 9031 | se->parent = parent; |
6f505b16 | 9032 | } |
052f1dc7 | 9033 | #endif |
6f505b16 | 9034 | |
052f1dc7 | 9035 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9036 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9037 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9038 | struct sched_rt_entity *parent) | |
6f505b16 | 9039 | { |
ec7dc8ac DG |
9040 | struct rq *rq = cpu_rq(cpu); |
9041 | ||
6f505b16 PZ |
9042 | tg->rt_rq[cpu] = rt_rq; |
9043 | init_rt_rq(rt_rq, rq); | |
9044 | rt_rq->tg = tg; | |
9045 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9046 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9047 | if (add) |
9048 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9049 | ||
9050 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9051 | if (!rt_se) |
9052 | return; | |
9053 | ||
ec7dc8ac DG |
9054 | if (!parent) |
9055 | rt_se->rt_rq = &rq->rt; | |
9056 | else | |
9057 | rt_se->rt_rq = parent->my_q; | |
9058 | ||
6f505b16 | 9059 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9060 | rt_se->parent = parent; |
6f505b16 PZ |
9061 | INIT_LIST_HEAD(&rt_se->run_list); |
9062 | } | |
9063 | #endif | |
9064 | ||
1da177e4 LT |
9065 | void __init sched_init(void) |
9066 | { | |
dd41f596 | 9067 | int i, j; |
434d53b0 MT |
9068 | unsigned long alloc_size = 0, ptr; |
9069 | ||
9070 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9071 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9072 | #endif | |
9073 | #ifdef CONFIG_RT_GROUP_SCHED | |
9074 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9075 | #endif |
9076 | #ifdef CONFIG_USER_SCHED | |
9077 | alloc_size *= 2; | |
df7c8e84 RR |
9078 | #endif |
9079 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9080 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9081 | #endif |
9082 | /* | |
9083 | * As sched_init() is called before page_alloc is setup, | |
9084 | * we use alloc_bootmem(). | |
9085 | */ | |
9086 | if (alloc_size) { | |
5a9d3225 | 9087 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
9088 | |
9089 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9090 | init_task_group.se = (struct sched_entity **)ptr; | |
9091 | ptr += nr_cpu_ids * sizeof(void **); | |
9092 | ||
9093 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9094 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9095 | |
9096 | #ifdef CONFIG_USER_SCHED | |
9097 | root_task_group.se = (struct sched_entity **)ptr; | |
9098 | ptr += nr_cpu_ids * sizeof(void **); | |
9099 | ||
9100 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9101 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9102 | #endif /* CONFIG_USER_SCHED */ |
9103 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9104 | #ifdef CONFIG_RT_GROUP_SCHED |
9105 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9106 | ptr += nr_cpu_ids * sizeof(void **); | |
9107 | ||
9108 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9109 | ptr += nr_cpu_ids * sizeof(void **); |
9110 | ||
9111 | #ifdef CONFIG_USER_SCHED | |
9112 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9113 | ptr += nr_cpu_ids * sizeof(void **); | |
9114 | ||
9115 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9116 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9117 | #endif /* CONFIG_USER_SCHED */ |
9118 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9119 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9120 | for_each_possible_cpu(i) { | |
9121 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9122 | ptr += cpumask_size(); | |
9123 | } | |
9124 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9125 | } |
dd41f596 | 9126 | |
57d885fe GH |
9127 | #ifdef CONFIG_SMP |
9128 | init_defrootdomain(); | |
9129 | #endif | |
9130 | ||
d0b27fa7 PZ |
9131 | init_rt_bandwidth(&def_rt_bandwidth, |
9132 | global_rt_period(), global_rt_runtime()); | |
9133 | ||
9134 | #ifdef CONFIG_RT_GROUP_SCHED | |
9135 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9136 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9137 | #ifdef CONFIG_USER_SCHED |
9138 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9139 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9140 | #endif /* CONFIG_USER_SCHED */ |
9141 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9142 | |
052f1dc7 | 9143 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9144 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9145 | INIT_LIST_HEAD(&init_task_group.children); |
9146 | ||
9147 | #ifdef CONFIG_USER_SCHED | |
9148 | INIT_LIST_HEAD(&root_task_group.children); | |
9149 | init_task_group.parent = &root_task_group; | |
9150 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9151 | #endif /* CONFIG_USER_SCHED */ |
9152 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9153 | |
0a945022 | 9154 | for_each_possible_cpu(i) { |
70b97a7f | 9155 | struct rq *rq; |
1da177e4 LT |
9156 | |
9157 | rq = cpu_rq(i); | |
9158 | spin_lock_init(&rq->lock); | |
7897986b | 9159 | rq->nr_running = 0; |
dce48a84 TG |
9160 | rq->calc_load_active = 0; |
9161 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9162 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9163 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9164 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9165 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9166 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9167 | #ifdef CONFIG_CGROUP_SCHED |
9168 | /* | |
9169 | * How much cpu bandwidth does init_task_group get? | |
9170 | * | |
9171 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9172 | * gets 100% of the cpu resources in the system. This overall | |
9173 | * system cpu resource is divided among the tasks of | |
9174 | * init_task_group and its child task-groups in a fair manner, | |
9175 | * based on each entity's (task or task-group's) weight | |
9176 | * (se->load.weight). | |
9177 | * | |
9178 | * In other words, if init_task_group has 10 tasks of weight | |
9179 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9180 | * then A0's share of the cpu resource is: | |
9181 | * | |
9182 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
9183 | * | |
9184 | * We achieve this by letting init_task_group's tasks sit | |
9185 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9186 | */ | |
ec7dc8ac | 9187 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9188 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9189 | root_task_group.shares = NICE_0_LOAD; |
9190 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9191 | /* |
9192 | * In case of task-groups formed thr' the user id of tasks, | |
9193 | * init_task_group represents tasks belonging to root user. | |
9194 | * Hence it forms a sibling of all subsequent groups formed. | |
9195 | * In this case, init_task_group gets only a fraction of overall | |
9196 | * system cpu resource, based on the weight assigned to root | |
9197 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9198 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9199 | * (init_cfs_rq) and having one entity represent this group of | |
9200 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9201 | */ | |
ec7dc8ac | 9202 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9203 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9204 | &per_cpu(init_sched_entity, i), i, 1, |
9205 | root_task_group.se[i]); | |
6f505b16 | 9206 | |
052f1dc7 | 9207 | #endif |
354d60c2 DG |
9208 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9209 | ||
9210 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9211 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9212 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9213 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9214 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9215 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9216 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9217 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9218 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9219 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9220 | root_task_group.rt_se[i]); | |
354d60c2 | 9221 | #endif |
dd41f596 | 9222 | #endif |
1da177e4 | 9223 | |
dd41f596 IM |
9224 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9225 | rq->cpu_load[j] = 0; | |
1da177e4 | 9226 | #ifdef CONFIG_SMP |
41c7ce9a | 9227 | rq->sd = NULL; |
57d885fe | 9228 | rq->rd = NULL; |
1da177e4 | 9229 | rq->active_balance = 0; |
dd41f596 | 9230 | rq->next_balance = jiffies; |
1da177e4 | 9231 | rq->push_cpu = 0; |
0a2966b4 | 9232 | rq->cpu = i; |
1f11eb6a | 9233 | rq->online = 0; |
1da177e4 LT |
9234 | rq->migration_thread = NULL; |
9235 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9236 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9237 | #endif |
8f4d37ec | 9238 | init_rq_hrtick(rq); |
1da177e4 | 9239 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9240 | } |
9241 | ||
2dd73a4f | 9242 | set_load_weight(&init_task); |
b50f60ce | 9243 | |
e107be36 AK |
9244 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9245 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9246 | #endif | |
9247 | ||
c9819f45 | 9248 | #ifdef CONFIG_SMP |
962cf36c | 9249 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9250 | #endif |
9251 | ||
b50f60ce HC |
9252 | #ifdef CONFIG_RT_MUTEXES |
9253 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9254 | #endif | |
9255 | ||
1da177e4 LT |
9256 | /* |
9257 | * The boot idle thread does lazy MMU switching as well: | |
9258 | */ | |
9259 | atomic_inc(&init_mm.mm_count); | |
9260 | enter_lazy_tlb(&init_mm, current); | |
9261 | ||
9262 | /* | |
9263 | * Make us the idle thread. Technically, schedule() should not be | |
9264 | * called from this thread, however somewhere below it might be, | |
9265 | * but because we are the idle thread, we just pick up running again | |
9266 | * when this runqueue becomes "idle". | |
9267 | */ | |
9268 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9269 | |
9270 | calc_load_update = jiffies + LOAD_FREQ; | |
9271 | ||
dd41f596 IM |
9272 | /* |
9273 | * During early bootup we pretend to be a normal task: | |
9274 | */ | |
9275 | current->sched_class = &fair_sched_class; | |
6892b75e | 9276 | |
6a7b3dc3 RR |
9277 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
9278 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 9279 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
9280 | #ifdef CONFIG_NO_HZ |
9281 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
f711f609 | 9282 | alloc_bootmem_cpumask_var(&nohz.ilb_grp_nohz_mask); |
7d1e6a9b | 9283 | #endif |
dcc30a35 | 9284 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 9285 | #endif /* SMP */ |
6a7b3dc3 | 9286 | |
6892b75e | 9287 | scheduler_running = 1; |
1da177e4 LT |
9288 | } |
9289 | ||
9290 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9291 | void __might_sleep(char *file, int line) | |
9292 | { | |
48f24c4d | 9293 | #ifdef in_atomic |
1da177e4 LT |
9294 | static unsigned long prev_jiffy; /* ratelimiting */ |
9295 | ||
aef745fc IM |
9296 | if ((!in_atomic() && !irqs_disabled()) || |
9297 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9298 | return; | |
9299 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9300 | return; | |
9301 | prev_jiffy = jiffies; | |
9302 | ||
9303 | printk(KERN_ERR | |
9304 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9305 | file, line); | |
9306 | printk(KERN_ERR | |
9307 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9308 | in_atomic(), irqs_disabled(), | |
9309 | current->pid, current->comm); | |
9310 | ||
9311 | debug_show_held_locks(current); | |
9312 | if (irqs_disabled()) | |
9313 | print_irqtrace_events(current); | |
9314 | dump_stack(); | |
1da177e4 LT |
9315 | #endif |
9316 | } | |
9317 | EXPORT_SYMBOL(__might_sleep); | |
9318 | #endif | |
9319 | ||
9320 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9321 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9322 | { | |
9323 | int on_rq; | |
3e51f33f | 9324 | |
3a5e4dc1 AK |
9325 | update_rq_clock(rq); |
9326 | on_rq = p->se.on_rq; | |
9327 | if (on_rq) | |
9328 | deactivate_task(rq, p, 0); | |
9329 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9330 | if (on_rq) { | |
9331 | activate_task(rq, p, 0); | |
9332 | resched_task(rq->curr); | |
9333 | } | |
9334 | } | |
9335 | ||
1da177e4 LT |
9336 | void normalize_rt_tasks(void) |
9337 | { | |
a0f98a1c | 9338 | struct task_struct *g, *p; |
1da177e4 | 9339 | unsigned long flags; |
70b97a7f | 9340 | struct rq *rq; |
1da177e4 | 9341 | |
4cf5d77a | 9342 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9343 | do_each_thread(g, p) { |
178be793 IM |
9344 | /* |
9345 | * Only normalize user tasks: | |
9346 | */ | |
9347 | if (!p->mm) | |
9348 | continue; | |
9349 | ||
6cfb0d5d | 9350 | p->se.exec_start = 0; |
6cfb0d5d | 9351 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9352 | p->se.wait_start = 0; |
dd41f596 | 9353 | p->se.sleep_start = 0; |
dd41f596 | 9354 | p->se.block_start = 0; |
6cfb0d5d | 9355 | #endif |
dd41f596 IM |
9356 | |
9357 | if (!rt_task(p)) { | |
9358 | /* | |
9359 | * Renice negative nice level userspace | |
9360 | * tasks back to 0: | |
9361 | */ | |
9362 | if (TASK_NICE(p) < 0 && p->mm) | |
9363 | set_user_nice(p, 0); | |
1da177e4 | 9364 | continue; |
dd41f596 | 9365 | } |
1da177e4 | 9366 | |
4cf5d77a | 9367 | spin_lock(&p->pi_lock); |
b29739f9 | 9368 | rq = __task_rq_lock(p); |
1da177e4 | 9369 | |
178be793 | 9370 | normalize_task(rq, p); |
3a5e4dc1 | 9371 | |
b29739f9 | 9372 | __task_rq_unlock(rq); |
4cf5d77a | 9373 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9374 | } while_each_thread(g, p); |
9375 | ||
4cf5d77a | 9376 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9377 | } |
9378 | ||
9379 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9380 | |
9381 | #ifdef CONFIG_IA64 | |
9382 | /* | |
9383 | * These functions are only useful for the IA64 MCA handling. | |
9384 | * | |
9385 | * They can only be called when the whole system has been | |
9386 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9387 | * activity can take place. Using them for anything else would | |
9388 | * be a serious bug, and as a result, they aren't even visible | |
9389 | * under any other configuration. | |
9390 | */ | |
9391 | ||
9392 | /** | |
9393 | * curr_task - return the current task for a given cpu. | |
9394 | * @cpu: the processor in question. | |
9395 | * | |
9396 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9397 | */ | |
36c8b586 | 9398 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9399 | { |
9400 | return cpu_curr(cpu); | |
9401 | } | |
9402 | ||
9403 | /** | |
9404 | * set_curr_task - set the current task for a given cpu. | |
9405 | * @cpu: the processor in question. | |
9406 | * @p: the task pointer to set. | |
9407 | * | |
9408 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9409 | * are serviced on a separate stack. It allows the architecture to switch the |
9410 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9411 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9412 | * and caller must save the original value of the current task (see | |
9413 | * curr_task() above) and restore that value before reenabling interrupts and | |
9414 | * re-starting the system. | |
9415 | * | |
9416 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9417 | */ | |
36c8b586 | 9418 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9419 | { |
9420 | cpu_curr(cpu) = p; | |
9421 | } | |
9422 | ||
9423 | #endif | |
29f59db3 | 9424 | |
bccbe08a PZ |
9425 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9426 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9427 | { |
9428 | int i; | |
9429 | ||
9430 | for_each_possible_cpu(i) { | |
9431 | if (tg->cfs_rq) | |
9432 | kfree(tg->cfs_rq[i]); | |
9433 | if (tg->se) | |
9434 | kfree(tg->se[i]); | |
6f505b16 PZ |
9435 | } |
9436 | ||
9437 | kfree(tg->cfs_rq); | |
9438 | kfree(tg->se); | |
6f505b16 PZ |
9439 | } |
9440 | ||
ec7dc8ac DG |
9441 | static |
9442 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9443 | { |
29f59db3 | 9444 | struct cfs_rq *cfs_rq; |
eab17229 | 9445 | struct sched_entity *se; |
9b5b7751 | 9446 | struct rq *rq; |
29f59db3 SV |
9447 | int i; |
9448 | ||
434d53b0 | 9449 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9450 | if (!tg->cfs_rq) |
9451 | goto err; | |
434d53b0 | 9452 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9453 | if (!tg->se) |
9454 | goto err; | |
052f1dc7 PZ |
9455 | |
9456 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9457 | |
9458 | for_each_possible_cpu(i) { | |
9b5b7751 | 9459 | rq = cpu_rq(i); |
29f59db3 | 9460 | |
eab17229 LZ |
9461 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9462 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9463 | if (!cfs_rq) |
9464 | goto err; | |
9465 | ||
eab17229 LZ |
9466 | se = kzalloc_node(sizeof(struct sched_entity), |
9467 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9468 | if (!se) |
9469 | goto err; | |
9470 | ||
eab17229 | 9471 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9472 | } |
9473 | ||
9474 | return 1; | |
9475 | ||
9476 | err: | |
9477 | return 0; | |
9478 | } | |
9479 | ||
9480 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9481 | { | |
9482 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9483 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9484 | } | |
9485 | ||
9486 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9487 | { | |
9488 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9489 | } | |
6d6bc0ad | 9490 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9491 | static inline void free_fair_sched_group(struct task_group *tg) |
9492 | { | |
9493 | } | |
9494 | ||
ec7dc8ac DG |
9495 | static inline |
9496 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9497 | { |
9498 | return 1; | |
9499 | } | |
9500 | ||
9501 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9502 | { | |
9503 | } | |
9504 | ||
9505 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9506 | { | |
9507 | } | |
6d6bc0ad | 9508 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9509 | |
9510 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9511 | static void free_rt_sched_group(struct task_group *tg) |
9512 | { | |
9513 | int i; | |
9514 | ||
d0b27fa7 PZ |
9515 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9516 | ||
bccbe08a PZ |
9517 | for_each_possible_cpu(i) { |
9518 | if (tg->rt_rq) | |
9519 | kfree(tg->rt_rq[i]); | |
9520 | if (tg->rt_se) | |
9521 | kfree(tg->rt_se[i]); | |
9522 | } | |
9523 | ||
9524 | kfree(tg->rt_rq); | |
9525 | kfree(tg->rt_se); | |
9526 | } | |
9527 | ||
ec7dc8ac DG |
9528 | static |
9529 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9530 | { |
9531 | struct rt_rq *rt_rq; | |
eab17229 | 9532 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9533 | struct rq *rq; |
9534 | int i; | |
9535 | ||
434d53b0 | 9536 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9537 | if (!tg->rt_rq) |
9538 | goto err; | |
434d53b0 | 9539 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9540 | if (!tg->rt_se) |
9541 | goto err; | |
9542 | ||
d0b27fa7 PZ |
9543 | init_rt_bandwidth(&tg->rt_bandwidth, |
9544 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9545 | |
9546 | for_each_possible_cpu(i) { | |
9547 | rq = cpu_rq(i); | |
9548 | ||
eab17229 LZ |
9549 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9550 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9551 | if (!rt_rq) |
9552 | goto err; | |
29f59db3 | 9553 | |
eab17229 LZ |
9554 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9555 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9556 | if (!rt_se) |
9557 | goto err; | |
29f59db3 | 9558 | |
eab17229 | 9559 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9560 | } |
9561 | ||
bccbe08a PZ |
9562 | return 1; |
9563 | ||
9564 | err: | |
9565 | return 0; | |
9566 | } | |
9567 | ||
9568 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9569 | { | |
9570 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9571 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9572 | } | |
9573 | ||
9574 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9575 | { | |
9576 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9577 | } | |
6d6bc0ad | 9578 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9579 | static inline void free_rt_sched_group(struct task_group *tg) |
9580 | { | |
9581 | } | |
9582 | ||
ec7dc8ac DG |
9583 | static inline |
9584 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9585 | { |
9586 | return 1; | |
9587 | } | |
9588 | ||
9589 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9590 | { | |
9591 | } | |
9592 | ||
9593 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9594 | { | |
9595 | } | |
6d6bc0ad | 9596 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9597 | |
d0b27fa7 | 9598 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9599 | static void free_sched_group(struct task_group *tg) |
9600 | { | |
9601 | free_fair_sched_group(tg); | |
9602 | free_rt_sched_group(tg); | |
9603 | kfree(tg); | |
9604 | } | |
9605 | ||
9606 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9607 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9608 | { |
9609 | struct task_group *tg; | |
9610 | unsigned long flags; | |
9611 | int i; | |
9612 | ||
9613 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9614 | if (!tg) | |
9615 | return ERR_PTR(-ENOMEM); | |
9616 | ||
ec7dc8ac | 9617 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9618 | goto err; |
9619 | ||
ec7dc8ac | 9620 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9621 | goto err; |
9622 | ||
8ed36996 | 9623 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9624 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9625 | register_fair_sched_group(tg, i); |
9626 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9627 | } |
6f505b16 | 9628 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9629 | |
9630 | WARN_ON(!parent); /* root should already exist */ | |
9631 | ||
9632 | tg->parent = parent; | |
f473aa5e | 9633 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9634 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9635 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9636 | |
9b5b7751 | 9637 | return tg; |
29f59db3 SV |
9638 | |
9639 | err: | |
6f505b16 | 9640 | free_sched_group(tg); |
29f59db3 SV |
9641 | return ERR_PTR(-ENOMEM); |
9642 | } | |
9643 | ||
9b5b7751 | 9644 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9645 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9646 | { |
29f59db3 | 9647 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9648 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9649 | } |
9650 | ||
9b5b7751 | 9651 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9652 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9653 | { |
8ed36996 | 9654 | unsigned long flags; |
9b5b7751 | 9655 | int i; |
29f59db3 | 9656 | |
8ed36996 | 9657 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9658 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9659 | unregister_fair_sched_group(tg, i); |
9660 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9661 | } |
6f505b16 | 9662 | list_del_rcu(&tg->list); |
f473aa5e | 9663 | list_del_rcu(&tg->siblings); |
8ed36996 | 9664 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9665 | |
9b5b7751 | 9666 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9667 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9668 | } |
9669 | ||
9b5b7751 | 9670 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9671 | * The caller of this function should have put the task in its new group |
9672 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9673 | * reflect its new group. | |
9b5b7751 SV |
9674 | */ |
9675 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9676 | { |
9677 | int on_rq, running; | |
9678 | unsigned long flags; | |
9679 | struct rq *rq; | |
9680 | ||
9681 | rq = task_rq_lock(tsk, &flags); | |
9682 | ||
29f59db3 SV |
9683 | update_rq_clock(rq); |
9684 | ||
051a1d1a | 9685 | running = task_current(rq, tsk); |
29f59db3 SV |
9686 | on_rq = tsk->se.on_rq; |
9687 | ||
0e1f3483 | 9688 | if (on_rq) |
29f59db3 | 9689 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9690 | if (unlikely(running)) |
9691 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9692 | |
6f505b16 | 9693 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9694 | |
810b3817 PZ |
9695 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9696 | if (tsk->sched_class->moved_group) | |
9697 | tsk->sched_class->moved_group(tsk); | |
9698 | #endif | |
9699 | ||
0e1f3483 HS |
9700 | if (unlikely(running)) |
9701 | tsk->sched_class->set_curr_task(rq); | |
9702 | if (on_rq) | |
7074badb | 9703 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9704 | |
29f59db3 SV |
9705 | task_rq_unlock(rq, &flags); |
9706 | } | |
6d6bc0ad | 9707 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9708 | |
052f1dc7 | 9709 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9710 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9711 | { |
9712 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9713 | int on_rq; |
9714 | ||
29f59db3 | 9715 | on_rq = se->on_rq; |
62fb1851 | 9716 | if (on_rq) |
29f59db3 SV |
9717 | dequeue_entity(cfs_rq, se, 0); |
9718 | ||
9719 | se->load.weight = shares; | |
e05510d0 | 9720 | se->load.inv_weight = 0; |
29f59db3 | 9721 | |
62fb1851 | 9722 | if (on_rq) |
29f59db3 | 9723 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9724 | } |
62fb1851 | 9725 | |
c09595f6 PZ |
9726 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9727 | { | |
9728 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9729 | struct rq *rq = cfs_rq->rq; | |
9730 | unsigned long flags; | |
9731 | ||
9732 | spin_lock_irqsave(&rq->lock, flags); | |
9733 | __set_se_shares(se, shares); | |
9734 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9735 | } |
9736 | ||
8ed36996 PZ |
9737 | static DEFINE_MUTEX(shares_mutex); |
9738 | ||
4cf86d77 | 9739 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9740 | { |
9741 | int i; | |
8ed36996 | 9742 | unsigned long flags; |
c61935fd | 9743 | |
ec7dc8ac DG |
9744 | /* |
9745 | * We can't change the weight of the root cgroup. | |
9746 | */ | |
9747 | if (!tg->se[0]) | |
9748 | return -EINVAL; | |
9749 | ||
18d95a28 PZ |
9750 | if (shares < MIN_SHARES) |
9751 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9752 | else if (shares > MAX_SHARES) |
9753 | shares = MAX_SHARES; | |
62fb1851 | 9754 | |
8ed36996 | 9755 | mutex_lock(&shares_mutex); |
9b5b7751 | 9756 | if (tg->shares == shares) |
5cb350ba | 9757 | goto done; |
29f59db3 | 9758 | |
8ed36996 | 9759 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9760 | for_each_possible_cpu(i) |
9761 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9762 | list_del_rcu(&tg->siblings); |
8ed36996 | 9763 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9764 | |
9765 | /* wait for any ongoing reference to this group to finish */ | |
9766 | synchronize_sched(); | |
9767 | ||
9768 | /* | |
9769 | * Now we are free to modify the group's share on each cpu | |
9770 | * w/o tripping rebalance_share or load_balance_fair. | |
9771 | */ | |
9b5b7751 | 9772 | tg->shares = shares; |
c09595f6 PZ |
9773 | for_each_possible_cpu(i) { |
9774 | /* | |
9775 | * force a rebalance | |
9776 | */ | |
9777 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9778 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9779 | } |
29f59db3 | 9780 | |
6b2d7700 SV |
9781 | /* |
9782 | * Enable load balance activity on this group, by inserting it back on | |
9783 | * each cpu's rq->leaf_cfs_rq_list. | |
9784 | */ | |
8ed36996 | 9785 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9786 | for_each_possible_cpu(i) |
9787 | register_fair_sched_group(tg, i); | |
f473aa5e | 9788 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9789 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9790 | done: |
8ed36996 | 9791 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9792 | return 0; |
29f59db3 SV |
9793 | } |
9794 | ||
5cb350ba DG |
9795 | unsigned long sched_group_shares(struct task_group *tg) |
9796 | { | |
9797 | return tg->shares; | |
9798 | } | |
052f1dc7 | 9799 | #endif |
5cb350ba | 9800 | |
052f1dc7 | 9801 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9802 | /* |
9f0c1e56 | 9803 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9804 | */ |
9f0c1e56 PZ |
9805 | static DEFINE_MUTEX(rt_constraints_mutex); |
9806 | ||
9807 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9808 | { | |
9809 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9810 | return 1ULL << 20; |
9f0c1e56 | 9811 | |
9a7e0b18 | 9812 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9813 | } |
9814 | ||
9a7e0b18 PZ |
9815 | /* Must be called with tasklist_lock held */ |
9816 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9817 | { |
9a7e0b18 | 9818 | struct task_struct *g, *p; |
b40b2e8e | 9819 | |
9a7e0b18 PZ |
9820 | do_each_thread(g, p) { |
9821 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9822 | return 1; | |
9823 | } while_each_thread(g, p); | |
b40b2e8e | 9824 | |
9a7e0b18 PZ |
9825 | return 0; |
9826 | } | |
b40b2e8e | 9827 | |
9a7e0b18 PZ |
9828 | struct rt_schedulable_data { |
9829 | struct task_group *tg; | |
9830 | u64 rt_period; | |
9831 | u64 rt_runtime; | |
9832 | }; | |
b40b2e8e | 9833 | |
9a7e0b18 PZ |
9834 | static int tg_schedulable(struct task_group *tg, void *data) |
9835 | { | |
9836 | struct rt_schedulable_data *d = data; | |
9837 | struct task_group *child; | |
9838 | unsigned long total, sum = 0; | |
9839 | u64 period, runtime; | |
b40b2e8e | 9840 | |
9a7e0b18 PZ |
9841 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9842 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9843 | |
9a7e0b18 PZ |
9844 | if (tg == d->tg) { |
9845 | period = d->rt_period; | |
9846 | runtime = d->rt_runtime; | |
b40b2e8e | 9847 | } |
b40b2e8e | 9848 | |
98a4826b PZ |
9849 | #ifdef CONFIG_USER_SCHED |
9850 | if (tg == &root_task_group) { | |
9851 | period = global_rt_period(); | |
9852 | runtime = global_rt_runtime(); | |
9853 | } | |
9854 | #endif | |
9855 | ||
4653f803 PZ |
9856 | /* |
9857 | * Cannot have more runtime than the period. | |
9858 | */ | |
9859 | if (runtime > period && runtime != RUNTIME_INF) | |
9860 | return -EINVAL; | |
6f505b16 | 9861 | |
4653f803 PZ |
9862 | /* |
9863 | * Ensure we don't starve existing RT tasks. | |
9864 | */ | |
9a7e0b18 PZ |
9865 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9866 | return -EBUSY; | |
6f505b16 | 9867 | |
9a7e0b18 | 9868 | total = to_ratio(period, runtime); |
6f505b16 | 9869 | |
4653f803 PZ |
9870 | /* |
9871 | * Nobody can have more than the global setting allows. | |
9872 | */ | |
9873 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9874 | return -EINVAL; | |
6f505b16 | 9875 | |
4653f803 PZ |
9876 | /* |
9877 | * The sum of our children's runtime should not exceed our own. | |
9878 | */ | |
9a7e0b18 PZ |
9879 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9880 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9881 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9882 | |
9a7e0b18 PZ |
9883 | if (child == d->tg) { |
9884 | period = d->rt_period; | |
9885 | runtime = d->rt_runtime; | |
9886 | } | |
6f505b16 | 9887 | |
9a7e0b18 | 9888 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9889 | } |
6f505b16 | 9890 | |
9a7e0b18 PZ |
9891 | if (sum > total) |
9892 | return -EINVAL; | |
9893 | ||
9894 | return 0; | |
6f505b16 PZ |
9895 | } |
9896 | ||
9a7e0b18 | 9897 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9898 | { |
9a7e0b18 PZ |
9899 | struct rt_schedulable_data data = { |
9900 | .tg = tg, | |
9901 | .rt_period = period, | |
9902 | .rt_runtime = runtime, | |
9903 | }; | |
9904 | ||
9905 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9906 | } |
9907 | ||
d0b27fa7 PZ |
9908 | static int tg_set_bandwidth(struct task_group *tg, |
9909 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9910 | { |
ac086bc2 | 9911 | int i, err = 0; |
9f0c1e56 | 9912 | |
9f0c1e56 | 9913 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9914 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9915 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9916 | if (err) | |
9f0c1e56 | 9917 | goto unlock; |
ac086bc2 PZ |
9918 | |
9919 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9920 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9921 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9922 | |
9923 | for_each_possible_cpu(i) { | |
9924 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9925 | ||
9926 | spin_lock(&rt_rq->rt_runtime_lock); | |
9927 | rt_rq->rt_runtime = rt_runtime; | |
9928 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9929 | } | |
9930 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9931 | unlock: |
521f1a24 | 9932 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9933 | mutex_unlock(&rt_constraints_mutex); |
9934 | ||
9935 | return err; | |
6f505b16 PZ |
9936 | } |
9937 | ||
d0b27fa7 PZ |
9938 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9939 | { | |
9940 | u64 rt_runtime, rt_period; | |
9941 | ||
9942 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9943 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9944 | if (rt_runtime_us < 0) | |
9945 | rt_runtime = RUNTIME_INF; | |
9946 | ||
9947 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9948 | } | |
9949 | ||
9f0c1e56 PZ |
9950 | long sched_group_rt_runtime(struct task_group *tg) |
9951 | { | |
9952 | u64 rt_runtime_us; | |
9953 | ||
d0b27fa7 | 9954 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9955 | return -1; |
9956 | ||
d0b27fa7 | 9957 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9958 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9959 | return rt_runtime_us; | |
9960 | } | |
d0b27fa7 PZ |
9961 | |
9962 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9963 | { | |
9964 | u64 rt_runtime, rt_period; | |
9965 | ||
9966 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9967 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9968 | ||
619b0488 R |
9969 | if (rt_period == 0) |
9970 | return -EINVAL; | |
9971 | ||
d0b27fa7 PZ |
9972 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9973 | } | |
9974 | ||
9975 | long sched_group_rt_period(struct task_group *tg) | |
9976 | { | |
9977 | u64 rt_period_us; | |
9978 | ||
9979 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9980 | do_div(rt_period_us, NSEC_PER_USEC); | |
9981 | return rt_period_us; | |
9982 | } | |
9983 | ||
9984 | static int sched_rt_global_constraints(void) | |
9985 | { | |
4653f803 | 9986 | u64 runtime, period; |
d0b27fa7 PZ |
9987 | int ret = 0; |
9988 | ||
ec5d4989 HS |
9989 | if (sysctl_sched_rt_period <= 0) |
9990 | return -EINVAL; | |
9991 | ||
4653f803 PZ |
9992 | runtime = global_rt_runtime(); |
9993 | period = global_rt_period(); | |
9994 | ||
9995 | /* | |
9996 | * Sanity check on the sysctl variables. | |
9997 | */ | |
9998 | if (runtime > period && runtime != RUNTIME_INF) | |
9999 | return -EINVAL; | |
10b612f4 | 10000 | |
d0b27fa7 | 10001 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10002 | read_lock(&tasklist_lock); |
4653f803 | 10003 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10004 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10005 | mutex_unlock(&rt_constraints_mutex); |
10006 | ||
10007 | return ret; | |
10008 | } | |
54e99124 DG |
10009 | |
10010 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10011 | { | |
10012 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10013 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10014 | return 0; | |
10015 | ||
10016 | return 1; | |
10017 | } | |
10018 | ||
6d6bc0ad | 10019 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10020 | static int sched_rt_global_constraints(void) |
10021 | { | |
ac086bc2 PZ |
10022 | unsigned long flags; |
10023 | int i; | |
10024 | ||
ec5d4989 HS |
10025 | if (sysctl_sched_rt_period <= 0) |
10026 | return -EINVAL; | |
10027 | ||
60aa605d PZ |
10028 | /* |
10029 | * There's always some RT tasks in the root group | |
10030 | * -- migration, kstopmachine etc.. | |
10031 | */ | |
10032 | if (sysctl_sched_rt_runtime == 0) | |
10033 | return -EBUSY; | |
10034 | ||
ac086bc2 PZ |
10035 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10036 | for_each_possible_cpu(i) { | |
10037 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10038 | ||
10039 | spin_lock(&rt_rq->rt_runtime_lock); | |
10040 | rt_rq->rt_runtime = global_rt_runtime(); | |
10041 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10042 | } | |
10043 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10044 | ||
d0b27fa7 PZ |
10045 | return 0; |
10046 | } | |
6d6bc0ad | 10047 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10048 | |
10049 | int sched_rt_handler(struct ctl_table *table, int write, | |
10050 | struct file *filp, void __user *buffer, size_t *lenp, | |
10051 | loff_t *ppos) | |
10052 | { | |
10053 | int ret; | |
10054 | int old_period, old_runtime; | |
10055 | static DEFINE_MUTEX(mutex); | |
10056 | ||
10057 | mutex_lock(&mutex); | |
10058 | old_period = sysctl_sched_rt_period; | |
10059 | old_runtime = sysctl_sched_rt_runtime; | |
10060 | ||
10061 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10062 | ||
10063 | if (!ret && write) { | |
10064 | ret = sched_rt_global_constraints(); | |
10065 | if (ret) { | |
10066 | sysctl_sched_rt_period = old_period; | |
10067 | sysctl_sched_rt_runtime = old_runtime; | |
10068 | } else { | |
10069 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10070 | def_rt_bandwidth.rt_period = | |
10071 | ns_to_ktime(global_rt_period()); | |
10072 | } | |
10073 | } | |
10074 | mutex_unlock(&mutex); | |
10075 | ||
10076 | return ret; | |
10077 | } | |
68318b8e | 10078 | |
052f1dc7 | 10079 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10080 | |
10081 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10082 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10083 | { |
2b01dfe3 PM |
10084 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10085 | struct task_group, css); | |
68318b8e SV |
10086 | } |
10087 | ||
10088 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10089 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10090 | { |
ec7dc8ac | 10091 | struct task_group *tg, *parent; |
68318b8e | 10092 | |
2b01dfe3 | 10093 | if (!cgrp->parent) { |
68318b8e | 10094 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10095 | return &init_task_group.css; |
10096 | } | |
10097 | ||
ec7dc8ac DG |
10098 | parent = cgroup_tg(cgrp->parent); |
10099 | tg = sched_create_group(parent); | |
68318b8e SV |
10100 | if (IS_ERR(tg)) |
10101 | return ERR_PTR(-ENOMEM); | |
10102 | ||
68318b8e SV |
10103 | return &tg->css; |
10104 | } | |
10105 | ||
41a2d6cf IM |
10106 | static void |
10107 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10108 | { |
2b01dfe3 | 10109 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10110 | |
10111 | sched_destroy_group(tg); | |
10112 | } | |
10113 | ||
41a2d6cf IM |
10114 | static int |
10115 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10116 | struct task_struct *tsk) | |
68318b8e | 10117 | { |
b68aa230 | 10118 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10119 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10120 | return -EINVAL; |
10121 | #else | |
68318b8e SV |
10122 | /* We don't support RT-tasks being in separate groups */ |
10123 | if (tsk->sched_class != &fair_sched_class) | |
10124 | return -EINVAL; | |
b68aa230 | 10125 | #endif |
68318b8e SV |
10126 | |
10127 | return 0; | |
10128 | } | |
10129 | ||
10130 | static void | |
2b01dfe3 | 10131 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10132 | struct cgroup *old_cont, struct task_struct *tsk) |
10133 | { | |
10134 | sched_move_task(tsk); | |
10135 | } | |
10136 | ||
052f1dc7 | 10137 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10138 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10139 | u64 shareval) |
68318b8e | 10140 | { |
2b01dfe3 | 10141 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10142 | } |
10143 | ||
f4c753b7 | 10144 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10145 | { |
2b01dfe3 | 10146 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10147 | |
10148 | return (u64) tg->shares; | |
10149 | } | |
6d6bc0ad | 10150 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10151 | |
052f1dc7 | 10152 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10153 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10154 | s64 val) |
6f505b16 | 10155 | { |
06ecb27c | 10156 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10157 | } |
10158 | ||
06ecb27c | 10159 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10160 | { |
06ecb27c | 10161 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10162 | } |
d0b27fa7 PZ |
10163 | |
10164 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10165 | u64 rt_period_us) | |
10166 | { | |
10167 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10168 | } | |
10169 | ||
10170 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10171 | { | |
10172 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10173 | } | |
6d6bc0ad | 10174 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10175 | |
fe5c7cc2 | 10176 | static struct cftype cpu_files[] = { |
052f1dc7 | 10177 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10178 | { |
10179 | .name = "shares", | |
f4c753b7 PM |
10180 | .read_u64 = cpu_shares_read_u64, |
10181 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10182 | }, |
052f1dc7 PZ |
10183 | #endif |
10184 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10185 | { |
9f0c1e56 | 10186 | .name = "rt_runtime_us", |
06ecb27c PM |
10187 | .read_s64 = cpu_rt_runtime_read, |
10188 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10189 | }, |
d0b27fa7 PZ |
10190 | { |
10191 | .name = "rt_period_us", | |
f4c753b7 PM |
10192 | .read_u64 = cpu_rt_period_read_uint, |
10193 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10194 | }, |
052f1dc7 | 10195 | #endif |
68318b8e SV |
10196 | }; |
10197 | ||
10198 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10199 | { | |
fe5c7cc2 | 10200 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10201 | } |
10202 | ||
10203 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10204 | .name = "cpu", |
10205 | .create = cpu_cgroup_create, | |
10206 | .destroy = cpu_cgroup_destroy, | |
10207 | .can_attach = cpu_cgroup_can_attach, | |
10208 | .attach = cpu_cgroup_attach, | |
10209 | .populate = cpu_cgroup_populate, | |
10210 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10211 | .early_init = 1, |
10212 | }; | |
10213 | ||
052f1dc7 | 10214 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10215 | |
10216 | #ifdef CONFIG_CGROUP_CPUACCT | |
10217 | ||
10218 | /* | |
10219 | * CPU accounting code for task groups. | |
10220 | * | |
10221 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10222 | * (balbir@in.ibm.com). | |
10223 | */ | |
10224 | ||
934352f2 | 10225 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10226 | struct cpuacct { |
10227 | struct cgroup_subsys_state css; | |
10228 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10229 | u64 *cpuusage; | |
ef12fefa | 10230 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10231 | struct cpuacct *parent; |
d842de87 SV |
10232 | }; |
10233 | ||
10234 | struct cgroup_subsys cpuacct_subsys; | |
10235 | ||
10236 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10237 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10238 | { |
32cd756a | 10239 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10240 | struct cpuacct, css); |
10241 | } | |
10242 | ||
10243 | /* return cpu accounting group to which this task belongs */ | |
10244 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10245 | { | |
10246 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10247 | struct cpuacct, css); | |
10248 | } | |
10249 | ||
10250 | /* create a new cpu accounting group */ | |
10251 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10252 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10253 | { |
10254 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10255 | int i; |
d842de87 SV |
10256 | |
10257 | if (!ca) | |
ef12fefa | 10258 | goto out; |
d842de87 SV |
10259 | |
10260 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10261 | if (!ca->cpuusage) |
10262 | goto out_free_ca; | |
10263 | ||
10264 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10265 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10266 | goto out_free_counters; | |
d842de87 | 10267 | |
934352f2 BR |
10268 | if (cgrp->parent) |
10269 | ca->parent = cgroup_ca(cgrp->parent); | |
10270 | ||
d842de87 | 10271 | return &ca->css; |
ef12fefa BR |
10272 | |
10273 | out_free_counters: | |
10274 | while (--i >= 0) | |
10275 | percpu_counter_destroy(&ca->cpustat[i]); | |
10276 | free_percpu(ca->cpuusage); | |
10277 | out_free_ca: | |
10278 | kfree(ca); | |
10279 | out: | |
10280 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10281 | } |
10282 | ||
10283 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10284 | static void |
32cd756a | 10285 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10286 | { |
32cd756a | 10287 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10288 | int i; |
d842de87 | 10289 | |
ef12fefa BR |
10290 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10291 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10292 | free_percpu(ca->cpuusage); |
10293 | kfree(ca); | |
10294 | } | |
10295 | ||
720f5498 KC |
10296 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10297 | { | |
b36128c8 | 10298 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10299 | u64 data; |
10300 | ||
10301 | #ifndef CONFIG_64BIT | |
10302 | /* | |
10303 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10304 | */ | |
10305 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10306 | data = *cpuusage; | |
10307 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10308 | #else | |
10309 | data = *cpuusage; | |
10310 | #endif | |
10311 | ||
10312 | return data; | |
10313 | } | |
10314 | ||
10315 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10316 | { | |
b36128c8 | 10317 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10318 | |
10319 | #ifndef CONFIG_64BIT | |
10320 | /* | |
10321 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10322 | */ | |
10323 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10324 | *cpuusage = val; | |
10325 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10326 | #else | |
10327 | *cpuusage = val; | |
10328 | #endif | |
10329 | } | |
10330 | ||
d842de87 | 10331 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10332 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10333 | { |
32cd756a | 10334 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10335 | u64 totalcpuusage = 0; |
10336 | int i; | |
10337 | ||
720f5498 KC |
10338 | for_each_present_cpu(i) |
10339 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10340 | |
10341 | return totalcpuusage; | |
10342 | } | |
10343 | ||
0297b803 DG |
10344 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10345 | u64 reset) | |
10346 | { | |
10347 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10348 | int err = 0; | |
10349 | int i; | |
10350 | ||
10351 | if (reset) { | |
10352 | err = -EINVAL; | |
10353 | goto out; | |
10354 | } | |
10355 | ||
720f5498 KC |
10356 | for_each_present_cpu(i) |
10357 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10358 | |
0297b803 DG |
10359 | out: |
10360 | return err; | |
10361 | } | |
10362 | ||
e9515c3c KC |
10363 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10364 | struct seq_file *m) | |
10365 | { | |
10366 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10367 | u64 percpu; | |
10368 | int i; | |
10369 | ||
10370 | for_each_present_cpu(i) { | |
10371 | percpu = cpuacct_cpuusage_read(ca, i); | |
10372 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10373 | } | |
10374 | seq_printf(m, "\n"); | |
10375 | return 0; | |
10376 | } | |
10377 | ||
ef12fefa BR |
10378 | static const char *cpuacct_stat_desc[] = { |
10379 | [CPUACCT_STAT_USER] = "user", | |
10380 | [CPUACCT_STAT_SYSTEM] = "system", | |
10381 | }; | |
10382 | ||
10383 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10384 | struct cgroup_map_cb *cb) | |
10385 | { | |
10386 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10387 | int i; | |
10388 | ||
10389 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10390 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10391 | val = cputime64_to_clock_t(val); | |
10392 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10393 | } | |
10394 | return 0; | |
10395 | } | |
10396 | ||
d842de87 SV |
10397 | static struct cftype files[] = { |
10398 | { | |
10399 | .name = "usage", | |
f4c753b7 PM |
10400 | .read_u64 = cpuusage_read, |
10401 | .write_u64 = cpuusage_write, | |
d842de87 | 10402 | }, |
e9515c3c KC |
10403 | { |
10404 | .name = "usage_percpu", | |
10405 | .read_seq_string = cpuacct_percpu_seq_read, | |
10406 | }, | |
ef12fefa BR |
10407 | { |
10408 | .name = "stat", | |
10409 | .read_map = cpuacct_stats_show, | |
10410 | }, | |
d842de87 SV |
10411 | }; |
10412 | ||
32cd756a | 10413 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10414 | { |
32cd756a | 10415 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10416 | } |
10417 | ||
10418 | /* | |
10419 | * charge this task's execution time to its accounting group. | |
10420 | * | |
10421 | * called with rq->lock held. | |
10422 | */ | |
10423 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10424 | { | |
10425 | struct cpuacct *ca; | |
934352f2 | 10426 | int cpu; |
d842de87 | 10427 | |
c40c6f85 | 10428 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10429 | return; |
10430 | ||
934352f2 | 10431 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10432 | |
10433 | rcu_read_lock(); | |
10434 | ||
d842de87 | 10435 | ca = task_ca(tsk); |
d842de87 | 10436 | |
934352f2 | 10437 | for (; ca; ca = ca->parent) { |
b36128c8 | 10438 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10439 | *cpuusage += cputime; |
10440 | } | |
a18b83b7 BR |
10441 | |
10442 | rcu_read_unlock(); | |
d842de87 SV |
10443 | } |
10444 | ||
ef12fefa BR |
10445 | /* |
10446 | * Charge the system/user time to the task's accounting group. | |
10447 | */ | |
10448 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10449 | enum cpuacct_stat_index idx, cputime_t val) | |
10450 | { | |
10451 | struct cpuacct *ca; | |
10452 | ||
10453 | if (unlikely(!cpuacct_subsys.active)) | |
10454 | return; | |
10455 | ||
10456 | rcu_read_lock(); | |
10457 | ca = task_ca(tsk); | |
10458 | ||
10459 | do { | |
10460 | percpu_counter_add(&ca->cpustat[idx], val); | |
10461 | ca = ca->parent; | |
10462 | } while (ca); | |
10463 | rcu_read_unlock(); | |
10464 | } | |
10465 | ||
d842de87 SV |
10466 | struct cgroup_subsys cpuacct_subsys = { |
10467 | .name = "cpuacct", | |
10468 | .create = cpuacct_create, | |
10469 | .destroy = cpuacct_destroy, | |
10470 | .populate = cpuacct_populate, | |
10471 | .subsys_id = cpuacct_subsys_id, | |
10472 | }; | |
10473 | #endif /* CONFIG_CGROUP_CPUACCT */ |