]>
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 | ||
8f4d37ec | 633 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
634 | #ifdef CONFIG_SMP |
635 | int hrtick_csd_pending; | |
636 | struct call_single_data hrtick_csd; | |
637 | #endif | |
8f4d37ec PZ |
638 | struct hrtimer hrtick_timer; |
639 | #endif | |
640 | ||
1da177e4 LT |
641 | #ifdef CONFIG_SCHEDSTATS |
642 | /* latency stats */ | |
643 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
644 | unsigned long long rq_cpu_time; |
645 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
646 | |
647 | /* sys_sched_yield() stats */ | |
480b9434 | 648 | unsigned int yld_count; |
1da177e4 LT |
649 | |
650 | /* schedule() stats */ | |
480b9434 KC |
651 | unsigned int sched_switch; |
652 | unsigned int sched_count; | |
653 | unsigned int sched_goidle; | |
1da177e4 LT |
654 | |
655 | /* try_to_wake_up() stats */ | |
480b9434 KC |
656 | unsigned int ttwu_count; |
657 | unsigned int ttwu_local; | |
b8efb561 IM |
658 | |
659 | /* BKL stats */ | |
480b9434 | 660 | unsigned int bkl_count; |
1da177e4 LT |
661 | #endif |
662 | }; | |
663 | ||
f34e3b61 | 664 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 665 | |
15afe09b | 666 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 667 | { |
15afe09b | 668 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
669 | } |
670 | ||
0a2966b4 CL |
671 | static inline int cpu_of(struct rq *rq) |
672 | { | |
673 | #ifdef CONFIG_SMP | |
674 | return rq->cpu; | |
675 | #else | |
676 | return 0; | |
677 | #endif | |
678 | } | |
679 | ||
674311d5 NP |
680 | /* |
681 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 682 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
683 | * |
684 | * The domain tree of any CPU may only be accessed from within | |
685 | * preempt-disabled sections. | |
686 | */ | |
48f24c4d IM |
687 | #define for_each_domain(cpu, __sd) \ |
688 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
689 | |
690 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
691 | #define this_rq() (&__get_cpu_var(runqueues)) | |
692 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
693 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
694 | ||
3e51f33f PZ |
695 | static inline void update_rq_clock(struct rq *rq) |
696 | { | |
697 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
698 | } | |
699 | ||
bf5c91ba IM |
700 | /* |
701 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
702 | */ | |
703 | #ifdef CONFIG_SCHED_DEBUG | |
704 | # define const_debug __read_mostly | |
705 | #else | |
706 | # define const_debug static const | |
707 | #endif | |
708 | ||
017730c1 IM |
709 | /** |
710 | * runqueue_is_locked | |
711 | * | |
712 | * Returns true if the current cpu runqueue is locked. | |
713 | * This interface allows printk to be called with the runqueue lock | |
714 | * held and know whether or not it is OK to wake up the klogd. | |
715 | */ | |
716 | int runqueue_is_locked(void) | |
717 | { | |
718 | int cpu = get_cpu(); | |
719 | struct rq *rq = cpu_rq(cpu); | |
720 | int ret; | |
721 | ||
722 | ret = spin_is_locked(&rq->lock); | |
723 | put_cpu(); | |
724 | return ret; | |
725 | } | |
726 | ||
bf5c91ba IM |
727 | /* |
728 | * Debugging: various feature bits | |
729 | */ | |
f00b45c1 PZ |
730 | |
731 | #define SCHED_FEAT(name, enabled) \ | |
732 | __SCHED_FEAT_##name , | |
733 | ||
bf5c91ba | 734 | enum { |
f00b45c1 | 735 | #include "sched_features.h" |
bf5c91ba IM |
736 | }; |
737 | ||
f00b45c1 PZ |
738 | #undef SCHED_FEAT |
739 | ||
740 | #define SCHED_FEAT(name, enabled) \ | |
741 | (1UL << __SCHED_FEAT_##name) * enabled | | |
742 | ||
bf5c91ba | 743 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
744 | #include "sched_features.h" |
745 | 0; | |
746 | ||
747 | #undef SCHED_FEAT | |
748 | ||
749 | #ifdef CONFIG_SCHED_DEBUG | |
750 | #define SCHED_FEAT(name, enabled) \ | |
751 | #name , | |
752 | ||
983ed7a6 | 753 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
754 | #include "sched_features.h" |
755 | NULL | |
756 | }; | |
757 | ||
758 | #undef SCHED_FEAT | |
759 | ||
34f3a814 | 760 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 761 | { |
f00b45c1 PZ |
762 | int i; |
763 | ||
764 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
765 | if (!(sysctl_sched_features & (1UL << i))) |
766 | seq_puts(m, "NO_"); | |
767 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 768 | } |
34f3a814 | 769 | seq_puts(m, "\n"); |
f00b45c1 | 770 | |
34f3a814 | 771 | return 0; |
f00b45c1 PZ |
772 | } |
773 | ||
774 | static ssize_t | |
775 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
776 | size_t cnt, loff_t *ppos) | |
777 | { | |
778 | char buf[64]; | |
779 | char *cmp = buf; | |
780 | int neg = 0; | |
781 | int i; | |
782 | ||
783 | if (cnt > 63) | |
784 | cnt = 63; | |
785 | ||
786 | if (copy_from_user(&buf, ubuf, cnt)) | |
787 | return -EFAULT; | |
788 | ||
789 | buf[cnt] = 0; | |
790 | ||
c24b7c52 | 791 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
792 | neg = 1; |
793 | cmp += 3; | |
794 | } | |
795 | ||
796 | for (i = 0; sched_feat_names[i]; i++) { | |
797 | int len = strlen(sched_feat_names[i]); | |
798 | ||
799 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
800 | if (neg) | |
801 | sysctl_sched_features &= ~(1UL << i); | |
802 | else | |
803 | sysctl_sched_features |= (1UL << i); | |
804 | break; | |
805 | } | |
806 | } | |
807 | ||
808 | if (!sched_feat_names[i]) | |
809 | return -EINVAL; | |
810 | ||
811 | filp->f_pos += cnt; | |
812 | ||
813 | return cnt; | |
814 | } | |
815 | ||
34f3a814 LZ |
816 | static int sched_feat_open(struct inode *inode, struct file *filp) |
817 | { | |
818 | return single_open(filp, sched_feat_show, NULL); | |
819 | } | |
820 | ||
f00b45c1 | 821 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
822 | .open = sched_feat_open, |
823 | .write = sched_feat_write, | |
824 | .read = seq_read, | |
825 | .llseek = seq_lseek, | |
826 | .release = single_release, | |
f00b45c1 PZ |
827 | }; |
828 | ||
829 | static __init int sched_init_debug(void) | |
830 | { | |
f00b45c1 PZ |
831 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
832 | &sched_feat_fops); | |
833 | ||
834 | return 0; | |
835 | } | |
836 | late_initcall(sched_init_debug); | |
837 | ||
838 | #endif | |
839 | ||
840 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 841 | |
b82d9fdd PZ |
842 | /* |
843 | * Number of tasks to iterate in a single balance run. | |
844 | * Limited because this is done with IRQs disabled. | |
845 | */ | |
846 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
847 | ||
2398f2c6 PZ |
848 | /* |
849 | * ratelimit for updating the group shares. | |
55cd5340 | 850 | * default: 0.25ms |
2398f2c6 | 851 | */ |
55cd5340 | 852 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 853 | |
ffda12a1 PZ |
854 | /* |
855 | * Inject some fuzzyness into changing the per-cpu group shares | |
856 | * this avoids remote rq-locks at the expense of fairness. | |
857 | * default: 4 | |
858 | */ | |
859 | unsigned int sysctl_sched_shares_thresh = 4; | |
860 | ||
fa85ae24 | 861 | /* |
9f0c1e56 | 862 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
863 | * default: 1s |
864 | */ | |
9f0c1e56 | 865 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 866 | |
6892b75e IM |
867 | static __read_mostly int scheduler_running; |
868 | ||
9f0c1e56 PZ |
869 | /* |
870 | * part of the period that we allow rt tasks to run in us. | |
871 | * default: 0.95s | |
872 | */ | |
873 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 874 | |
d0b27fa7 PZ |
875 | static inline u64 global_rt_period(void) |
876 | { | |
877 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
878 | } | |
879 | ||
880 | static inline u64 global_rt_runtime(void) | |
881 | { | |
e26873bb | 882 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
883 | return RUNTIME_INF; |
884 | ||
885 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
886 | } | |
fa85ae24 | 887 | |
1da177e4 | 888 | #ifndef prepare_arch_switch |
4866cde0 NP |
889 | # define prepare_arch_switch(next) do { } while (0) |
890 | #endif | |
891 | #ifndef finish_arch_switch | |
892 | # define finish_arch_switch(prev) do { } while (0) | |
893 | #endif | |
894 | ||
051a1d1a DA |
895 | static inline int task_current(struct rq *rq, struct task_struct *p) |
896 | { | |
897 | return rq->curr == p; | |
898 | } | |
899 | ||
4866cde0 | 900 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 901 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 902 | { |
051a1d1a | 903 | return task_current(rq, p); |
4866cde0 NP |
904 | } |
905 | ||
70b97a7f | 906 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
907 | { |
908 | } | |
909 | ||
70b97a7f | 910 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 911 | { |
da04c035 IM |
912 | #ifdef CONFIG_DEBUG_SPINLOCK |
913 | /* this is a valid case when another task releases the spinlock */ | |
914 | rq->lock.owner = current; | |
915 | #endif | |
8a25d5de IM |
916 | /* |
917 | * If we are tracking spinlock dependencies then we have to | |
918 | * fix up the runqueue lock - which gets 'carried over' from | |
919 | * prev into current: | |
920 | */ | |
921 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
922 | ||
4866cde0 NP |
923 | spin_unlock_irq(&rq->lock); |
924 | } | |
925 | ||
926 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 927 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
928 | { |
929 | #ifdef CONFIG_SMP | |
930 | return p->oncpu; | |
931 | #else | |
051a1d1a | 932 | return task_current(rq, p); |
4866cde0 NP |
933 | #endif |
934 | } | |
935 | ||
70b97a7f | 936 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
937 | { |
938 | #ifdef CONFIG_SMP | |
939 | /* | |
940 | * We can optimise this out completely for !SMP, because the | |
941 | * SMP rebalancing from interrupt is the only thing that cares | |
942 | * here. | |
943 | */ | |
944 | next->oncpu = 1; | |
945 | #endif | |
946 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
947 | spin_unlock_irq(&rq->lock); | |
948 | #else | |
949 | spin_unlock(&rq->lock); | |
950 | #endif | |
951 | } | |
952 | ||
70b97a7f | 953 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
954 | { |
955 | #ifdef CONFIG_SMP | |
956 | /* | |
957 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
958 | * We must ensure this doesn't happen until the switch is completely | |
959 | * finished. | |
960 | */ | |
961 | smp_wmb(); | |
962 | prev->oncpu = 0; | |
963 | #endif | |
964 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
965 | local_irq_enable(); | |
1da177e4 | 966 | #endif |
4866cde0 NP |
967 | } |
968 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 969 | |
b29739f9 IM |
970 | /* |
971 | * __task_rq_lock - lock the runqueue a given task resides on. | |
972 | * Must be called interrupts disabled. | |
973 | */ | |
70b97a7f | 974 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
975 | __acquires(rq->lock) |
976 | { | |
3a5c359a AK |
977 | for (;;) { |
978 | struct rq *rq = task_rq(p); | |
979 | spin_lock(&rq->lock); | |
980 | if (likely(rq == task_rq(p))) | |
981 | return rq; | |
b29739f9 | 982 | spin_unlock(&rq->lock); |
b29739f9 | 983 | } |
b29739f9 IM |
984 | } |
985 | ||
1da177e4 LT |
986 | /* |
987 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 988 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
989 | * explicitly disabling preemption. |
990 | */ | |
70b97a7f | 991 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
992 | __acquires(rq->lock) |
993 | { | |
70b97a7f | 994 | struct rq *rq; |
1da177e4 | 995 | |
3a5c359a AK |
996 | for (;;) { |
997 | local_irq_save(*flags); | |
998 | rq = task_rq(p); | |
999 | spin_lock(&rq->lock); | |
1000 | if (likely(rq == task_rq(p))) | |
1001 | return rq; | |
1da177e4 | 1002 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1003 | } |
1da177e4 LT |
1004 | } |
1005 | ||
ad474cac ON |
1006 | void task_rq_unlock_wait(struct task_struct *p) |
1007 | { | |
1008 | struct rq *rq = task_rq(p); | |
1009 | ||
1010 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1011 | spin_unlock_wait(&rq->lock); | |
1012 | } | |
1013 | ||
a9957449 | 1014 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1015 | __releases(rq->lock) |
1016 | { | |
1017 | spin_unlock(&rq->lock); | |
1018 | } | |
1019 | ||
70b97a7f | 1020 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1021 | __releases(rq->lock) |
1022 | { | |
1023 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1024 | } | |
1025 | ||
1da177e4 | 1026 | /* |
cc2a73b5 | 1027 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1028 | */ |
a9957449 | 1029 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1030 | __acquires(rq->lock) |
1031 | { | |
70b97a7f | 1032 | struct rq *rq; |
1da177e4 LT |
1033 | |
1034 | local_irq_disable(); | |
1035 | rq = this_rq(); | |
1036 | spin_lock(&rq->lock); | |
1037 | ||
1038 | return rq; | |
1039 | } | |
1040 | ||
8f4d37ec PZ |
1041 | #ifdef CONFIG_SCHED_HRTICK |
1042 | /* | |
1043 | * Use HR-timers to deliver accurate preemption points. | |
1044 | * | |
1045 | * Its all a bit involved since we cannot program an hrt while holding the | |
1046 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1047 | * reschedule event. | |
1048 | * | |
1049 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1050 | * rq->lock. | |
1051 | */ | |
8f4d37ec PZ |
1052 | |
1053 | /* | |
1054 | * Use hrtick when: | |
1055 | * - enabled by features | |
1056 | * - hrtimer is actually high res | |
1057 | */ | |
1058 | static inline int hrtick_enabled(struct rq *rq) | |
1059 | { | |
1060 | if (!sched_feat(HRTICK)) | |
1061 | return 0; | |
ba42059f | 1062 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1063 | return 0; |
8f4d37ec PZ |
1064 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1065 | } | |
1066 | ||
8f4d37ec PZ |
1067 | static void hrtick_clear(struct rq *rq) |
1068 | { | |
1069 | if (hrtimer_active(&rq->hrtick_timer)) | |
1070 | hrtimer_cancel(&rq->hrtick_timer); | |
1071 | } | |
1072 | ||
8f4d37ec PZ |
1073 | /* |
1074 | * High-resolution timer tick. | |
1075 | * Runs from hardirq context with interrupts disabled. | |
1076 | */ | |
1077 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1078 | { | |
1079 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1080 | ||
1081 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1082 | ||
1083 | spin_lock(&rq->lock); | |
3e51f33f | 1084 | update_rq_clock(rq); |
8f4d37ec PZ |
1085 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1086 | spin_unlock(&rq->lock); | |
1087 | ||
1088 | return HRTIMER_NORESTART; | |
1089 | } | |
1090 | ||
95e904c7 | 1091 | #ifdef CONFIG_SMP |
31656519 PZ |
1092 | /* |
1093 | * called from hardirq (IPI) context | |
1094 | */ | |
1095 | static void __hrtick_start(void *arg) | |
b328ca18 | 1096 | { |
31656519 | 1097 | struct rq *rq = arg; |
b328ca18 | 1098 | |
31656519 PZ |
1099 | spin_lock(&rq->lock); |
1100 | hrtimer_restart(&rq->hrtick_timer); | |
1101 | rq->hrtick_csd_pending = 0; | |
1102 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1103 | } |
1104 | ||
31656519 PZ |
1105 | /* |
1106 | * Called to set the hrtick timer state. | |
1107 | * | |
1108 | * called with rq->lock held and irqs disabled | |
1109 | */ | |
1110 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1111 | { |
31656519 PZ |
1112 | struct hrtimer *timer = &rq->hrtick_timer; |
1113 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1114 | |
cc584b21 | 1115 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1116 | |
1117 | if (rq == this_rq()) { | |
1118 | hrtimer_restart(timer); | |
1119 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1120 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1121 | rq->hrtick_csd_pending = 1; |
1122 | } | |
b328ca18 PZ |
1123 | } |
1124 | ||
1125 | static int | |
1126 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1127 | { | |
1128 | int cpu = (int)(long)hcpu; | |
1129 | ||
1130 | switch (action) { | |
1131 | case CPU_UP_CANCELED: | |
1132 | case CPU_UP_CANCELED_FROZEN: | |
1133 | case CPU_DOWN_PREPARE: | |
1134 | case CPU_DOWN_PREPARE_FROZEN: | |
1135 | case CPU_DEAD: | |
1136 | case CPU_DEAD_FROZEN: | |
31656519 | 1137 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1138 | return NOTIFY_OK; |
1139 | } | |
1140 | ||
1141 | return NOTIFY_DONE; | |
1142 | } | |
1143 | ||
fa748203 | 1144 | static __init void init_hrtick(void) |
b328ca18 PZ |
1145 | { |
1146 | hotcpu_notifier(hotplug_hrtick, 0); | |
1147 | } | |
31656519 PZ |
1148 | #else |
1149 | /* | |
1150 | * Called to set the hrtick timer state. | |
1151 | * | |
1152 | * called with rq->lock held and irqs disabled | |
1153 | */ | |
1154 | static void hrtick_start(struct rq *rq, u64 delay) | |
1155 | { | |
7f1e2ca9 PZ |
1156 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
1157 | HRTIMER_MODE_REL, 0); | |
31656519 | 1158 | } |
b328ca18 | 1159 | |
006c75f1 | 1160 | static inline void init_hrtick(void) |
8f4d37ec | 1161 | { |
8f4d37ec | 1162 | } |
31656519 | 1163 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1164 | |
31656519 | 1165 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1166 | { |
31656519 PZ |
1167 | #ifdef CONFIG_SMP |
1168 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1169 | |
31656519 PZ |
1170 | rq->hrtick_csd.flags = 0; |
1171 | rq->hrtick_csd.func = __hrtick_start; | |
1172 | rq->hrtick_csd.info = rq; | |
1173 | #endif | |
8f4d37ec | 1174 | |
31656519 PZ |
1175 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1176 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1177 | } |
006c75f1 | 1178 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1179 | static inline void hrtick_clear(struct rq *rq) |
1180 | { | |
1181 | } | |
1182 | ||
8f4d37ec PZ |
1183 | static inline void init_rq_hrtick(struct rq *rq) |
1184 | { | |
1185 | } | |
1186 | ||
b328ca18 PZ |
1187 | static inline void init_hrtick(void) |
1188 | { | |
1189 | } | |
006c75f1 | 1190 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1191 | |
c24d20db IM |
1192 | /* |
1193 | * resched_task - mark a task 'to be rescheduled now'. | |
1194 | * | |
1195 | * On UP this means the setting of the need_resched flag, on SMP it | |
1196 | * might also involve a cross-CPU call to trigger the scheduler on | |
1197 | * the target CPU. | |
1198 | */ | |
1199 | #ifdef CONFIG_SMP | |
1200 | ||
1201 | #ifndef tsk_is_polling | |
1202 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1203 | #endif | |
1204 | ||
31656519 | 1205 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1206 | { |
1207 | int cpu; | |
1208 | ||
1209 | assert_spin_locked(&task_rq(p)->lock); | |
1210 | ||
5ed0cec0 | 1211 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1212 | return; |
1213 | ||
5ed0cec0 | 1214 | set_tsk_need_resched(p); |
c24d20db IM |
1215 | |
1216 | cpu = task_cpu(p); | |
1217 | if (cpu == smp_processor_id()) | |
1218 | return; | |
1219 | ||
1220 | /* NEED_RESCHED must be visible before we test polling */ | |
1221 | smp_mb(); | |
1222 | if (!tsk_is_polling(p)) | |
1223 | smp_send_reschedule(cpu); | |
1224 | } | |
1225 | ||
1226 | static void resched_cpu(int cpu) | |
1227 | { | |
1228 | struct rq *rq = cpu_rq(cpu); | |
1229 | unsigned long flags; | |
1230 | ||
1231 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1232 | return; | |
1233 | resched_task(cpu_curr(cpu)); | |
1234 | spin_unlock_irqrestore(&rq->lock, flags); | |
1235 | } | |
06d8308c TG |
1236 | |
1237 | #ifdef CONFIG_NO_HZ | |
1238 | /* | |
1239 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1240 | * idle CPU then this timer might expire before the next timer event | |
1241 | * which is scheduled to wake up that CPU. In case of a completely | |
1242 | * idle system the next event might even be infinite time into the | |
1243 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1244 | * leaves the inner idle loop so the newly added timer is taken into | |
1245 | * account when the CPU goes back to idle and evaluates the timer | |
1246 | * wheel for the next timer event. | |
1247 | */ | |
1248 | void wake_up_idle_cpu(int cpu) | |
1249 | { | |
1250 | struct rq *rq = cpu_rq(cpu); | |
1251 | ||
1252 | if (cpu == smp_processor_id()) | |
1253 | return; | |
1254 | ||
1255 | /* | |
1256 | * This is safe, as this function is called with the timer | |
1257 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1258 | * to idle and has not yet set rq->curr to idle then it will | |
1259 | * be serialized on the timer wheel base lock and take the new | |
1260 | * timer into account automatically. | |
1261 | */ | |
1262 | if (rq->curr != rq->idle) | |
1263 | return; | |
1264 | ||
1265 | /* | |
1266 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1267 | * lockless. The worst case is that the other CPU runs the | |
1268 | * idle task through an additional NOOP schedule() | |
1269 | */ | |
5ed0cec0 | 1270 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1271 | |
1272 | /* NEED_RESCHED must be visible before we test polling */ | |
1273 | smp_mb(); | |
1274 | if (!tsk_is_polling(rq->idle)) | |
1275 | smp_send_reschedule(cpu); | |
1276 | } | |
6d6bc0ad | 1277 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1278 | |
6d6bc0ad | 1279 | #else /* !CONFIG_SMP */ |
31656519 | 1280 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1281 | { |
1282 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1283 | set_tsk_need_resched(p); |
c24d20db | 1284 | } |
6d6bc0ad | 1285 | #endif /* CONFIG_SMP */ |
c24d20db | 1286 | |
45bf76df IM |
1287 | #if BITS_PER_LONG == 32 |
1288 | # define WMULT_CONST (~0UL) | |
1289 | #else | |
1290 | # define WMULT_CONST (1UL << 32) | |
1291 | #endif | |
1292 | ||
1293 | #define WMULT_SHIFT 32 | |
1294 | ||
194081eb IM |
1295 | /* |
1296 | * Shift right and round: | |
1297 | */ | |
cf2ab469 | 1298 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1299 | |
a7be37ac PZ |
1300 | /* |
1301 | * delta *= weight / lw | |
1302 | */ | |
cb1c4fc9 | 1303 | static unsigned long |
45bf76df IM |
1304 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1305 | struct load_weight *lw) | |
1306 | { | |
1307 | u64 tmp; | |
1308 | ||
7a232e03 LJ |
1309 | if (!lw->inv_weight) { |
1310 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1311 | lw->inv_weight = 1; | |
1312 | else | |
1313 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1314 | / (lw->weight+1); | |
1315 | } | |
45bf76df IM |
1316 | |
1317 | tmp = (u64)delta_exec * weight; | |
1318 | /* | |
1319 | * Check whether we'd overflow the 64-bit multiplication: | |
1320 | */ | |
194081eb | 1321 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1322 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1323 | WMULT_SHIFT/2); |
1324 | else | |
cf2ab469 | 1325 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1326 | |
ecf691da | 1327 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1331 | { |
1332 | lw->weight += inc; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1337 | { |
1338 | lw->weight -= dec; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
2dd73a4f PW |
1342 | /* |
1343 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1344 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1345 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1346 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1347 | * scaled version of the new time slice allocation that they receive on time |
1348 | * slice expiry etc. | |
1349 | */ | |
1350 | ||
cce7ade8 PZ |
1351 | #define WEIGHT_IDLEPRIO 3 |
1352 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1353 | |
1354 | /* | |
1355 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1356 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1357 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1358 | * that remained on nice 0. | |
1359 | * | |
1360 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1361 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1362 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1363 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1364 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1365 | */ |
1366 | static const int prio_to_weight[40] = { | |
254753dc IM |
1367 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1368 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1369 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1370 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1371 | /* 0 */ 1024, 820, 655, 526, 423, | |
1372 | /* 5 */ 335, 272, 215, 172, 137, | |
1373 | /* 10 */ 110, 87, 70, 56, 45, | |
1374 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1375 | }; |
1376 | ||
5714d2de IM |
1377 | /* |
1378 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1379 | * | |
1380 | * In cases where the weight does not change often, we can use the | |
1381 | * precalculated inverse to speed up arithmetics by turning divisions | |
1382 | * into multiplications: | |
1383 | */ | |
dd41f596 | 1384 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1385 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1386 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1387 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1388 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1389 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1390 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1391 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1392 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1393 | }; |
2dd73a4f | 1394 | |
dd41f596 IM |
1395 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1396 | ||
1397 | /* | |
1398 | * runqueue iterator, to support SMP load-balancing between different | |
1399 | * scheduling classes, without having to expose their internal data | |
1400 | * structures to the load-balancing proper: | |
1401 | */ | |
1402 | struct rq_iterator { | |
1403 | void *arg; | |
1404 | struct task_struct *(*start)(void *); | |
1405 | struct task_struct *(*next)(void *); | |
1406 | }; | |
1407 | ||
e1d1484f PW |
1408 | #ifdef CONFIG_SMP |
1409 | static unsigned long | |
1410 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1411 | unsigned long max_load_move, struct sched_domain *sd, | |
1412 | enum cpu_idle_type idle, int *all_pinned, | |
1413 | int *this_best_prio, struct rq_iterator *iterator); | |
1414 | ||
1415 | static int | |
1416 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1417 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1418 | struct rq_iterator *iterator); | |
e1d1484f | 1419 | #endif |
dd41f596 | 1420 | |
ef12fefa BR |
1421 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1422 | enum cpuacct_stat_index { | |
1423 | CPUACCT_STAT_USER, /* ... user mode */ | |
1424 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1425 | ||
1426 | CPUACCT_STAT_NSTATS, | |
1427 | }; | |
1428 | ||
d842de87 SV |
1429 | #ifdef CONFIG_CGROUP_CPUACCT |
1430 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1431 | static void cpuacct_update_stats(struct task_struct *tsk, |
1432 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1433 | #else |
1434 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1435 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1437 | #endif |
1438 | ||
18d95a28 PZ |
1439 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1440 | { | |
1441 | update_load_add(&rq->load, load); | |
1442 | } | |
1443 | ||
1444 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1445 | { | |
1446 | update_load_sub(&rq->load, load); | |
1447 | } | |
1448 | ||
7940ca36 | 1449 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1450 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1451 | |
1452 | /* | |
1453 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1454 | * leaving it for the final time. | |
1455 | */ | |
eb755805 | 1456 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1457 | { |
1458 | struct task_group *parent, *child; | |
eb755805 | 1459 | int ret; |
c09595f6 PZ |
1460 | |
1461 | rcu_read_lock(); | |
1462 | parent = &root_task_group; | |
1463 | down: | |
eb755805 PZ |
1464 | ret = (*down)(parent, data); |
1465 | if (ret) | |
1466 | goto out_unlock; | |
c09595f6 PZ |
1467 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1468 | parent = child; | |
1469 | goto down; | |
1470 | ||
1471 | up: | |
1472 | continue; | |
1473 | } | |
eb755805 PZ |
1474 | ret = (*up)(parent, data); |
1475 | if (ret) | |
1476 | goto out_unlock; | |
c09595f6 PZ |
1477 | |
1478 | child = parent; | |
1479 | parent = parent->parent; | |
1480 | if (parent) | |
1481 | goto up; | |
eb755805 | 1482 | out_unlock: |
c09595f6 | 1483 | rcu_read_unlock(); |
eb755805 PZ |
1484 | |
1485 | return ret; | |
c09595f6 PZ |
1486 | } |
1487 | ||
eb755805 PZ |
1488 | static int tg_nop(struct task_group *tg, void *data) |
1489 | { | |
1490 | return 0; | |
c09595f6 | 1491 | } |
eb755805 PZ |
1492 | #endif |
1493 | ||
1494 | #ifdef CONFIG_SMP | |
1495 | static unsigned long source_load(int cpu, int type); | |
1496 | static unsigned long target_load(int cpu, int type); | |
1497 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1498 | ||
1499 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1500 | { | |
1501 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1502 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1503 | |
4cd42620 SR |
1504 | if (nr_running) |
1505 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1506 | else |
1507 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1508 | |
1509 | return rq->avg_load_per_task; | |
1510 | } | |
1511 | ||
1512 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1513 | |
c09595f6 PZ |
1514 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1515 | ||
1516 | /* | |
1517 | * Calculate and set the cpu's group shares. | |
1518 | */ | |
1519 | static void | |
ffda12a1 PZ |
1520 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1521 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1522 | { |
c09595f6 PZ |
1523 | unsigned long shares; |
1524 | unsigned long rq_weight; | |
1525 | ||
c8cba857 | 1526 | if (!tg->se[cpu]) |
c09595f6 PZ |
1527 | return; |
1528 | ||
ec4e0e2f | 1529 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1530 | |
c09595f6 PZ |
1531 | /* |
1532 | * \Sum shares * rq_weight | |
1533 | * shares = ----------------------- | |
1534 | * \Sum rq_weight | |
1535 | * | |
1536 | */ | |
ec4e0e2f | 1537 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1538 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1539 | |
ffda12a1 PZ |
1540 | if (abs(shares - tg->se[cpu]->load.weight) > |
1541 | sysctl_sched_shares_thresh) { | |
1542 | struct rq *rq = cpu_rq(cpu); | |
1543 | unsigned long flags; | |
c09595f6 | 1544 | |
ffda12a1 | 1545 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1546 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1547 | |
ffda12a1 PZ |
1548 | __set_se_shares(tg->se[cpu], shares); |
1549 | spin_unlock_irqrestore(&rq->lock, flags); | |
1550 | } | |
18d95a28 | 1551 | } |
c09595f6 PZ |
1552 | |
1553 | /* | |
c8cba857 PZ |
1554 | * Re-compute the task group their per cpu shares over the given domain. |
1555 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1556 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1557 | */ |
eb755805 | 1558 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1559 | { |
ec4e0e2f | 1560 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1561 | unsigned long shares = 0; |
eb755805 | 1562 | struct sched_domain *sd = data; |
c8cba857 | 1563 | int i; |
c09595f6 | 1564 | |
758b2cdc | 1565 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1566 | /* |
1567 | * If there are currently no tasks on the cpu pretend there | |
1568 | * is one of average load so that when a new task gets to | |
1569 | * run here it will not get delayed by group starvation. | |
1570 | */ | |
1571 | weight = tg->cfs_rq[i]->load.weight; | |
1572 | if (!weight) | |
1573 | weight = NICE_0_LOAD; | |
1574 | ||
1575 | tg->cfs_rq[i]->rq_weight = weight; | |
1576 | rq_weight += weight; | |
c8cba857 | 1577 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1578 | } |
c09595f6 | 1579 | |
c8cba857 PZ |
1580 | if ((!shares && rq_weight) || shares > tg->shares) |
1581 | shares = tg->shares; | |
1582 | ||
1583 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1584 | shares = tg->shares; | |
c09595f6 | 1585 | |
758b2cdc | 1586 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1587 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1588 | |
1589 | return 0; | |
c09595f6 PZ |
1590 | } |
1591 | ||
1592 | /* | |
c8cba857 PZ |
1593 | * Compute the cpu's hierarchical load factor for each task group. |
1594 | * This needs to be done in a top-down fashion because the load of a child | |
1595 | * group is a fraction of its parents load. | |
c09595f6 | 1596 | */ |
eb755805 | 1597 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1598 | { |
c8cba857 | 1599 | unsigned long load; |
eb755805 | 1600 | long cpu = (long)data; |
c09595f6 | 1601 | |
c8cba857 PZ |
1602 | if (!tg->parent) { |
1603 | load = cpu_rq(cpu)->load.weight; | |
1604 | } else { | |
1605 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1606 | load *= tg->cfs_rq[cpu]->shares; | |
1607 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1608 | } | |
c09595f6 | 1609 | |
c8cba857 | 1610 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1611 | |
eb755805 | 1612 | return 0; |
c09595f6 PZ |
1613 | } |
1614 | ||
c8cba857 | 1615 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1616 | { |
2398f2c6 PZ |
1617 | u64 now = cpu_clock(raw_smp_processor_id()); |
1618 | s64 elapsed = now - sd->last_update; | |
1619 | ||
1620 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1621 | sd->last_update = now; | |
eb755805 | 1622 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1623 | } |
4d8d595d PZ |
1624 | } |
1625 | ||
3e5459b4 PZ |
1626 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1627 | { | |
1628 | spin_unlock(&rq->lock); | |
1629 | update_shares(sd); | |
1630 | spin_lock(&rq->lock); | |
1631 | } | |
1632 | ||
eb755805 | 1633 | static void update_h_load(long cpu) |
c09595f6 | 1634 | { |
eb755805 | 1635 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1636 | } |
1637 | ||
c09595f6 PZ |
1638 | #else |
1639 | ||
c8cba857 | 1640 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1641 | { |
1642 | } | |
1643 | ||
3e5459b4 PZ |
1644 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1645 | { | |
1646 | } | |
1647 | ||
18d95a28 PZ |
1648 | #endif |
1649 | ||
8f45e2b5 GH |
1650 | #ifdef CONFIG_PREEMPT |
1651 | ||
70574a99 | 1652 | /* |
8f45e2b5 GH |
1653 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1654 | * way at the expense of forcing extra atomic operations in all | |
1655 | * invocations. This assures that the double_lock is acquired using the | |
1656 | * same underlying policy as the spinlock_t on this architecture, which | |
1657 | * reduces latency compared to the unfair variant below. However, it | |
1658 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1659 | */ |
8f45e2b5 GH |
1660 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1661 | __releases(this_rq->lock) | |
1662 | __acquires(busiest->lock) | |
1663 | __acquires(this_rq->lock) | |
1664 | { | |
1665 | spin_unlock(&this_rq->lock); | |
1666 | double_rq_lock(this_rq, busiest); | |
1667 | ||
1668 | return 1; | |
1669 | } | |
1670 | ||
1671 | #else | |
1672 | /* | |
1673 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1674 | * latency by eliminating extra atomic operations when the locks are | |
1675 | * already in proper order on entry. This favors lower cpu-ids and will | |
1676 | * grant the double lock to lower cpus over higher ids under contention, | |
1677 | * regardless of entry order into the function. | |
1678 | */ | |
1679 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1680 | __releases(this_rq->lock) |
1681 | __acquires(busiest->lock) | |
1682 | __acquires(this_rq->lock) | |
1683 | { | |
1684 | int ret = 0; | |
1685 | ||
70574a99 AD |
1686 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1687 | if (busiest < this_rq) { | |
1688 | spin_unlock(&this_rq->lock); | |
1689 | spin_lock(&busiest->lock); | |
1690 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1691 | ret = 1; | |
1692 | } else | |
1693 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1694 | } | |
1695 | return ret; | |
1696 | } | |
1697 | ||
8f45e2b5 GH |
1698 | #endif /* CONFIG_PREEMPT */ |
1699 | ||
1700 | /* | |
1701 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1702 | */ | |
1703 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1704 | { | |
1705 | if (unlikely(!irqs_disabled())) { | |
1706 | /* printk() doesn't work good under rq->lock */ | |
1707 | spin_unlock(&this_rq->lock); | |
1708 | BUG_ON(1); | |
1709 | } | |
1710 | ||
1711 | return _double_lock_balance(this_rq, busiest); | |
1712 | } | |
1713 | ||
70574a99 AD |
1714 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1715 | __releases(busiest->lock) | |
1716 | { | |
1717 | spin_unlock(&busiest->lock); | |
1718 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1719 | } | |
18d95a28 PZ |
1720 | #endif |
1721 | ||
30432094 | 1722 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1723 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1724 | { | |
30432094 | 1725 | #ifdef CONFIG_SMP |
34e83e85 IM |
1726 | cfs_rq->shares = shares; |
1727 | #endif | |
1728 | } | |
30432094 | 1729 | #endif |
e7693a36 | 1730 | |
dd41f596 | 1731 | #include "sched_stats.h" |
dd41f596 | 1732 | #include "sched_idletask.c" |
5522d5d5 IM |
1733 | #include "sched_fair.c" |
1734 | #include "sched_rt.c" | |
dd41f596 IM |
1735 | #ifdef CONFIG_SCHED_DEBUG |
1736 | # include "sched_debug.c" | |
1737 | #endif | |
1738 | ||
1739 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1740 | #define for_each_class(class) \ |
1741 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1742 | |
c09595f6 | 1743 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1744 | { |
1745 | rq->nr_running++; | |
9c217245 IM |
1746 | } |
1747 | ||
c09595f6 | 1748 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1749 | { |
1750 | rq->nr_running--; | |
9c217245 IM |
1751 | } |
1752 | ||
45bf76df IM |
1753 | static void set_load_weight(struct task_struct *p) |
1754 | { | |
1755 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1756 | p->se.load.weight = prio_to_weight[0] * 2; |
1757 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1758 | return; | |
1759 | } | |
45bf76df | 1760 | |
dd41f596 IM |
1761 | /* |
1762 | * SCHED_IDLE tasks get minimal weight: | |
1763 | */ | |
1764 | if (p->policy == SCHED_IDLE) { | |
1765 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1766 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1767 | return; | |
1768 | } | |
71f8bd46 | 1769 | |
dd41f596 IM |
1770 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1771 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1772 | } |
1773 | ||
2087a1ad GH |
1774 | static void update_avg(u64 *avg, u64 sample) |
1775 | { | |
1776 | s64 diff = sample - *avg; | |
1777 | *avg += diff >> 3; | |
1778 | } | |
1779 | ||
8159f87e | 1780 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1781 | { |
831451ac PZ |
1782 | if (wakeup) |
1783 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1784 | ||
dd41f596 | 1785 | sched_info_queued(p); |
fd390f6a | 1786 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1787 | p->se.on_rq = 1; |
71f8bd46 IM |
1788 | } |
1789 | ||
69be72c1 | 1790 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1791 | { |
831451ac PZ |
1792 | if (sleep) { |
1793 | if (p->se.last_wakeup) { | |
1794 | update_avg(&p->se.avg_overlap, | |
1795 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1796 | p->se.last_wakeup = 0; | |
1797 | } else { | |
1798 | update_avg(&p->se.avg_wakeup, | |
1799 | sysctl_sched_wakeup_granularity); | |
1800 | } | |
2087a1ad GH |
1801 | } |
1802 | ||
46ac22ba | 1803 | sched_info_dequeued(p); |
f02231e5 | 1804 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1805 | p->se.on_rq = 0; |
71f8bd46 IM |
1806 | } |
1807 | ||
14531189 | 1808 | /* |
dd41f596 | 1809 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1810 | */ |
14531189 IM |
1811 | static inline int __normal_prio(struct task_struct *p) |
1812 | { | |
dd41f596 | 1813 | return p->static_prio; |
14531189 IM |
1814 | } |
1815 | ||
b29739f9 IM |
1816 | /* |
1817 | * Calculate the expected normal priority: i.e. priority | |
1818 | * without taking RT-inheritance into account. Might be | |
1819 | * boosted by interactivity modifiers. Changes upon fork, | |
1820 | * setprio syscalls, and whenever the interactivity | |
1821 | * estimator recalculates. | |
1822 | */ | |
36c8b586 | 1823 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1824 | { |
1825 | int prio; | |
1826 | ||
e05606d3 | 1827 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1828 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1829 | else | |
1830 | prio = __normal_prio(p); | |
1831 | return prio; | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * Calculate the current priority, i.e. the priority | |
1836 | * taken into account by the scheduler. This value might | |
1837 | * be boosted by RT tasks, or might be boosted by | |
1838 | * interactivity modifiers. Will be RT if the task got | |
1839 | * RT-boosted. If not then it returns p->normal_prio. | |
1840 | */ | |
36c8b586 | 1841 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1842 | { |
1843 | p->normal_prio = normal_prio(p); | |
1844 | /* | |
1845 | * If we are RT tasks or we were boosted to RT priority, | |
1846 | * keep the priority unchanged. Otherwise, update priority | |
1847 | * to the normal priority: | |
1848 | */ | |
1849 | if (!rt_prio(p->prio)) | |
1850 | return p->normal_prio; | |
1851 | return p->prio; | |
1852 | } | |
1853 | ||
1da177e4 | 1854 | /* |
dd41f596 | 1855 | * activate_task - move a task to the runqueue. |
1da177e4 | 1856 | */ |
dd41f596 | 1857 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1858 | { |
d9514f6c | 1859 | if (task_contributes_to_load(p)) |
dd41f596 | 1860 | rq->nr_uninterruptible--; |
1da177e4 | 1861 | |
8159f87e | 1862 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1863 | inc_nr_running(rq); |
1da177e4 LT |
1864 | } |
1865 | ||
1da177e4 LT |
1866 | /* |
1867 | * deactivate_task - remove a task from the runqueue. | |
1868 | */ | |
2e1cb74a | 1869 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1870 | { |
d9514f6c | 1871 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1872 | rq->nr_uninterruptible++; |
1873 | ||
69be72c1 | 1874 | dequeue_task(rq, p, sleep); |
c09595f6 | 1875 | dec_nr_running(rq); |
1da177e4 LT |
1876 | } |
1877 | ||
1da177e4 LT |
1878 | /** |
1879 | * task_curr - is this task currently executing on a CPU? | |
1880 | * @p: the task in question. | |
1881 | */ | |
36c8b586 | 1882 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1883 | { |
1884 | return cpu_curr(task_cpu(p)) == p; | |
1885 | } | |
1886 | ||
dd41f596 IM |
1887 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1888 | { | |
6f505b16 | 1889 | set_task_rq(p, cpu); |
dd41f596 | 1890 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1891 | /* |
1892 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1893 | * successfuly executed on another CPU. We must ensure that updates of | |
1894 | * per-task data have been completed by this moment. | |
1895 | */ | |
1896 | smp_wmb(); | |
dd41f596 | 1897 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1898 | #endif |
2dd73a4f PW |
1899 | } |
1900 | ||
cb469845 SR |
1901 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1902 | const struct sched_class *prev_class, | |
1903 | int oldprio, int running) | |
1904 | { | |
1905 | if (prev_class != p->sched_class) { | |
1906 | if (prev_class->switched_from) | |
1907 | prev_class->switched_from(rq, p, running); | |
1908 | p->sched_class->switched_to(rq, p, running); | |
1909 | } else | |
1910 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1911 | } | |
1912 | ||
1da177e4 | 1913 | #ifdef CONFIG_SMP |
c65cc870 | 1914 | |
e958b360 TG |
1915 | /* Used instead of source_load when we know the type == 0 */ |
1916 | static unsigned long weighted_cpuload(const int cpu) | |
1917 | { | |
1918 | return cpu_rq(cpu)->load.weight; | |
1919 | } | |
1920 | ||
cc367732 IM |
1921 | /* |
1922 | * Is this task likely cache-hot: | |
1923 | */ | |
e7693a36 | 1924 | static int |
cc367732 IM |
1925 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1926 | { | |
1927 | s64 delta; | |
1928 | ||
f540a608 IM |
1929 | /* |
1930 | * Buddy candidates are cache hot: | |
1931 | */ | |
4793241b PZ |
1932 | if (sched_feat(CACHE_HOT_BUDDY) && |
1933 | (&p->se == cfs_rq_of(&p->se)->next || | |
1934 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1935 | return 1; |
1936 | ||
cc367732 IM |
1937 | if (p->sched_class != &fair_sched_class) |
1938 | return 0; | |
1939 | ||
6bc1665b IM |
1940 | if (sysctl_sched_migration_cost == -1) |
1941 | return 1; | |
1942 | if (sysctl_sched_migration_cost == 0) | |
1943 | return 0; | |
1944 | ||
cc367732 IM |
1945 | delta = now - p->se.exec_start; |
1946 | ||
1947 | return delta < (s64)sysctl_sched_migration_cost; | |
1948 | } | |
1949 | ||
1950 | ||
dd41f596 | 1951 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1952 | { |
dd41f596 IM |
1953 | int old_cpu = task_cpu(p); |
1954 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1955 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1956 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1957 | u64 clock_offset; |
dd41f596 IM |
1958 | |
1959 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1960 | |
cbc34ed1 PZ |
1961 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1962 | ||
6cfb0d5d IM |
1963 | #ifdef CONFIG_SCHEDSTATS |
1964 | if (p->se.wait_start) | |
1965 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1966 | if (p->se.sleep_start) |
1967 | p->se.sleep_start -= clock_offset; | |
1968 | if (p->se.block_start) | |
1969 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1970 | if (old_cpu != new_cpu) { |
1971 | schedstat_inc(p, se.nr_migrations); | |
1972 | if (task_hot(p, old_rq->clock, NULL)) | |
1973 | schedstat_inc(p, se.nr_forced2_migrations); | |
1974 | } | |
6cfb0d5d | 1975 | #endif |
2830cf8c SV |
1976 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1977 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1978 | |
1979 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1980 | } |
1981 | ||
70b97a7f | 1982 | struct migration_req { |
1da177e4 | 1983 | struct list_head list; |
1da177e4 | 1984 | |
36c8b586 | 1985 | struct task_struct *task; |
1da177e4 LT |
1986 | int dest_cpu; |
1987 | ||
1da177e4 | 1988 | struct completion done; |
70b97a7f | 1989 | }; |
1da177e4 LT |
1990 | |
1991 | /* | |
1992 | * The task's runqueue lock must be held. | |
1993 | * Returns true if you have to wait for migration thread. | |
1994 | */ | |
36c8b586 | 1995 | static int |
70b97a7f | 1996 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1997 | { |
70b97a7f | 1998 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1999 | |
2000 | /* | |
2001 | * If the task is not on a runqueue (and not running), then | |
2002 | * it is sufficient to simply update the task's cpu field. | |
2003 | */ | |
dd41f596 | 2004 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2005 | set_task_cpu(p, dest_cpu); |
2006 | return 0; | |
2007 | } | |
2008 | ||
2009 | init_completion(&req->done); | |
1da177e4 LT |
2010 | req->task = p; |
2011 | req->dest_cpu = dest_cpu; | |
2012 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2013 | |
1da177e4 LT |
2014 | return 1; |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * wait_task_inactive - wait for a thread to unschedule. | |
2019 | * | |
85ba2d86 RM |
2020 | * If @match_state is nonzero, it's the @p->state value just checked and |
2021 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2022 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2023 | * we return a positive number (its total switch count). If a second call | |
2024 | * a short while later returns the same number, the caller can be sure that | |
2025 | * @p has remained unscheduled the whole time. | |
2026 | * | |
1da177e4 LT |
2027 | * The caller must ensure that the task *will* unschedule sometime soon, |
2028 | * else this function might spin for a *long* time. This function can't | |
2029 | * be called with interrupts off, or it may introduce deadlock with | |
2030 | * smp_call_function() if an IPI is sent by the same process we are | |
2031 | * waiting to become inactive. | |
2032 | */ | |
85ba2d86 | 2033 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2034 | { |
2035 | unsigned long flags; | |
dd41f596 | 2036 | int running, on_rq; |
85ba2d86 | 2037 | unsigned long ncsw; |
70b97a7f | 2038 | struct rq *rq; |
1da177e4 | 2039 | |
3a5c359a AK |
2040 | for (;;) { |
2041 | /* | |
2042 | * We do the initial early heuristics without holding | |
2043 | * any task-queue locks at all. We'll only try to get | |
2044 | * the runqueue lock when things look like they will | |
2045 | * work out! | |
2046 | */ | |
2047 | rq = task_rq(p); | |
fa490cfd | 2048 | |
3a5c359a AK |
2049 | /* |
2050 | * If the task is actively running on another CPU | |
2051 | * still, just relax and busy-wait without holding | |
2052 | * any locks. | |
2053 | * | |
2054 | * NOTE! Since we don't hold any locks, it's not | |
2055 | * even sure that "rq" stays as the right runqueue! | |
2056 | * But we don't care, since "task_running()" will | |
2057 | * return false if the runqueue has changed and p | |
2058 | * is actually now running somewhere else! | |
2059 | */ | |
85ba2d86 RM |
2060 | while (task_running(rq, p)) { |
2061 | if (match_state && unlikely(p->state != match_state)) | |
2062 | return 0; | |
3a5c359a | 2063 | cpu_relax(); |
85ba2d86 | 2064 | } |
fa490cfd | 2065 | |
3a5c359a AK |
2066 | /* |
2067 | * Ok, time to look more closely! We need the rq | |
2068 | * lock now, to be *sure*. If we're wrong, we'll | |
2069 | * just go back and repeat. | |
2070 | */ | |
2071 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2072 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2073 | running = task_running(rq, p); |
2074 | on_rq = p->se.on_rq; | |
85ba2d86 | 2075 | ncsw = 0; |
f31e11d8 | 2076 | if (!match_state || p->state == match_state) |
93dcf55f | 2077 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2078 | task_rq_unlock(rq, &flags); |
fa490cfd | 2079 | |
85ba2d86 RM |
2080 | /* |
2081 | * If it changed from the expected state, bail out now. | |
2082 | */ | |
2083 | if (unlikely(!ncsw)) | |
2084 | break; | |
2085 | ||
3a5c359a AK |
2086 | /* |
2087 | * Was it really running after all now that we | |
2088 | * checked with the proper locks actually held? | |
2089 | * | |
2090 | * Oops. Go back and try again.. | |
2091 | */ | |
2092 | if (unlikely(running)) { | |
2093 | cpu_relax(); | |
2094 | continue; | |
2095 | } | |
fa490cfd | 2096 | |
3a5c359a AK |
2097 | /* |
2098 | * It's not enough that it's not actively running, | |
2099 | * it must be off the runqueue _entirely_, and not | |
2100 | * preempted! | |
2101 | * | |
80dd99b3 | 2102 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2103 | * running right now), it's preempted, and we should |
2104 | * yield - it could be a while. | |
2105 | */ | |
2106 | if (unlikely(on_rq)) { | |
2107 | schedule_timeout_uninterruptible(1); | |
2108 | continue; | |
2109 | } | |
fa490cfd | 2110 | |
3a5c359a AK |
2111 | /* |
2112 | * Ahh, all good. It wasn't running, and it wasn't | |
2113 | * runnable, which means that it will never become | |
2114 | * running in the future either. We're all done! | |
2115 | */ | |
2116 | break; | |
2117 | } | |
85ba2d86 RM |
2118 | |
2119 | return ncsw; | |
1da177e4 LT |
2120 | } |
2121 | ||
2122 | /*** | |
2123 | * kick_process - kick a running thread to enter/exit the kernel | |
2124 | * @p: the to-be-kicked thread | |
2125 | * | |
2126 | * Cause a process which is running on another CPU to enter | |
2127 | * kernel-mode, without any delay. (to get signals handled.) | |
2128 | * | |
2129 | * NOTE: this function doesnt have to take the runqueue lock, | |
2130 | * because all it wants to ensure is that the remote task enters | |
2131 | * the kernel. If the IPI races and the task has been migrated | |
2132 | * to another CPU then no harm is done and the purpose has been | |
2133 | * achieved as well. | |
2134 | */ | |
36c8b586 | 2135 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2136 | { |
2137 | int cpu; | |
2138 | ||
2139 | preempt_disable(); | |
2140 | cpu = task_cpu(p); | |
2141 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2142 | smp_send_reschedule(cpu); | |
2143 | preempt_enable(); | |
2144 | } | |
2145 | ||
2146 | /* | |
2dd73a4f PW |
2147 | * Return a low guess at the load of a migration-source cpu weighted |
2148 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2149 | * |
2150 | * We want to under-estimate the load of migration sources, to | |
2151 | * balance conservatively. | |
2152 | */ | |
a9957449 | 2153 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2154 | { |
70b97a7f | 2155 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2156 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2157 | |
93b75217 | 2158 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2159 | return total; |
b910472d | 2160 | |
dd41f596 | 2161 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2162 | } |
2163 | ||
2164 | /* | |
2dd73a4f PW |
2165 | * Return a high guess at the load of a migration-target cpu weighted |
2166 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2167 | */ |
a9957449 | 2168 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2169 | { |
70b97a7f | 2170 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2171 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2172 | |
93b75217 | 2173 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2174 | return total; |
3b0bd9bc | 2175 | |
dd41f596 | 2176 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2177 | } |
2178 | ||
147cbb4b NP |
2179 | /* |
2180 | * find_idlest_group finds and returns the least busy CPU group within the | |
2181 | * domain. | |
2182 | */ | |
2183 | static struct sched_group * | |
2184 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2185 | { | |
2186 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2187 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2188 | int load_idx = sd->forkexec_idx; | |
2189 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2190 | ||
2191 | do { | |
2192 | unsigned long load, avg_load; | |
2193 | int local_group; | |
2194 | int i; | |
2195 | ||
da5a5522 | 2196 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2197 | if (!cpumask_intersects(sched_group_cpus(group), |
2198 | &p->cpus_allowed)) | |
3a5c359a | 2199 | continue; |
da5a5522 | 2200 | |
758b2cdc RR |
2201 | local_group = cpumask_test_cpu(this_cpu, |
2202 | sched_group_cpus(group)); | |
147cbb4b NP |
2203 | |
2204 | /* Tally up the load of all CPUs in the group */ | |
2205 | avg_load = 0; | |
2206 | ||
758b2cdc | 2207 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2208 | /* Bias balancing toward cpus of our domain */ |
2209 | if (local_group) | |
2210 | load = source_load(i, load_idx); | |
2211 | else | |
2212 | load = target_load(i, load_idx); | |
2213 | ||
2214 | avg_load += load; | |
2215 | } | |
2216 | ||
2217 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2218 | avg_load = sg_div_cpu_power(group, |
2219 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2220 | |
2221 | if (local_group) { | |
2222 | this_load = avg_load; | |
2223 | this = group; | |
2224 | } else if (avg_load < min_load) { | |
2225 | min_load = avg_load; | |
2226 | idlest = group; | |
2227 | } | |
3a5c359a | 2228 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2229 | |
2230 | if (!idlest || 100*this_load < imbalance*min_load) | |
2231 | return NULL; | |
2232 | return idlest; | |
2233 | } | |
2234 | ||
2235 | /* | |
0feaece9 | 2236 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2237 | */ |
95cdf3b7 | 2238 | static int |
758b2cdc | 2239 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2240 | { |
2241 | unsigned long load, min_load = ULONG_MAX; | |
2242 | int idlest = -1; | |
2243 | int i; | |
2244 | ||
da5a5522 | 2245 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2246 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2247 | load = weighted_cpuload(i); |
147cbb4b NP |
2248 | |
2249 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2250 | min_load = load; | |
2251 | idlest = i; | |
2252 | } | |
2253 | } | |
2254 | ||
2255 | return idlest; | |
2256 | } | |
2257 | ||
476d139c NP |
2258 | /* |
2259 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2260 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2261 | * SD_BALANCE_EXEC. | |
2262 | * | |
2263 | * Balance, ie. select the least loaded group. | |
2264 | * | |
2265 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2266 | * | |
2267 | * preempt must be disabled. | |
2268 | */ | |
2269 | static int sched_balance_self(int cpu, int flag) | |
2270 | { | |
2271 | struct task_struct *t = current; | |
2272 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2273 | |
c96d145e | 2274 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2275 | /* |
2276 | * If power savings logic is enabled for a domain, stop there. | |
2277 | */ | |
5c45bf27 SS |
2278 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2279 | break; | |
476d139c NP |
2280 | if (tmp->flags & flag) |
2281 | sd = tmp; | |
c96d145e | 2282 | } |
476d139c | 2283 | |
039a1c41 PZ |
2284 | if (sd) |
2285 | update_shares(sd); | |
2286 | ||
476d139c | 2287 | while (sd) { |
476d139c | 2288 | struct sched_group *group; |
1a848870 SS |
2289 | int new_cpu, weight; |
2290 | ||
2291 | if (!(sd->flags & flag)) { | |
2292 | sd = sd->child; | |
2293 | continue; | |
2294 | } | |
476d139c | 2295 | |
476d139c | 2296 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2297 | if (!group) { |
2298 | sd = sd->child; | |
2299 | continue; | |
2300 | } | |
476d139c | 2301 | |
758b2cdc | 2302 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2303 | if (new_cpu == -1 || new_cpu == cpu) { |
2304 | /* Now try balancing at a lower domain level of cpu */ | |
2305 | sd = sd->child; | |
2306 | continue; | |
2307 | } | |
476d139c | 2308 | |
1a848870 | 2309 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2310 | cpu = new_cpu; |
758b2cdc | 2311 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2312 | sd = NULL; |
476d139c | 2313 | for_each_domain(cpu, tmp) { |
758b2cdc | 2314 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2315 | break; |
2316 | if (tmp->flags & flag) | |
2317 | sd = tmp; | |
2318 | } | |
2319 | /* while loop will break here if sd == NULL */ | |
2320 | } | |
2321 | ||
2322 | return cpu; | |
2323 | } | |
2324 | ||
2325 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2326 | |
1da177e4 LT |
2327 | /*** |
2328 | * try_to_wake_up - wake up a thread | |
2329 | * @p: the to-be-woken-up thread | |
2330 | * @state: the mask of task states that can be woken | |
2331 | * @sync: do a synchronous wakeup? | |
2332 | * | |
2333 | * Put it on the run-queue if it's not already there. The "current" | |
2334 | * thread is always on the run-queue (except when the actual | |
2335 | * re-schedule is in progress), and as such you're allowed to do | |
2336 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2337 | * runnable without the overhead of this. | |
2338 | * | |
2339 | * returns failure only if the task is already active. | |
2340 | */ | |
36c8b586 | 2341 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2342 | { |
cc367732 | 2343 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2344 | unsigned long flags; |
2345 | long old_state; | |
70b97a7f | 2346 | struct rq *rq; |
1da177e4 | 2347 | |
b85d0667 IM |
2348 | if (!sched_feat(SYNC_WAKEUPS)) |
2349 | sync = 0; | |
2350 | ||
2398f2c6 | 2351 | #ifdef CONFIG_SMP |
57310a98 | 2352 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2353 | struct sched_domain *sd; |
2354 | ||
2355 | this_cpu = raw_smp_processor_id(); | |
2356 | cpu = task_cpu(p); | |
2357 | ||
2358 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2359 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2360 | update_shares(sd); |
2361 | break; | |
2362 | } | |
2363 | } | |
2364 | } | |
2365 | #endif | |
2366 | ||
04e2f174 | 2367 | smp_wmb(); |
1da177e4 | 2368 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2369 | update_rq_clock(rq); |
1da177e4 LT |
2370 | old_state = p->state; |
2371 | if (!(old_state & state)) | |
2372 | goto out; | |
2373 | ||
dd41f596 | 2374 | if (p->se.on_rq) |
1da177e4 LT |
2375 | goto out_running; |
2376 | ||
2377 | cpu = task_cpu(p); | |
cc367732 | 2378 | orig_cpu = cpu; |
1da177e4 LT |
2379 | this_cpu = smp_processor_id(); |
2380 | ||
2381 | #ifdef CONFIG_SMP | |
2382 | if (unlikely(task_running(rq, p))) | |
2383 | goto out_activate; | |
2384 | ||
5d2f5a61 DA |
2385 | cpu = p->sched_class->select_task_rq(p, sync); |
2386 | if (cpu != orig_cpu) { | |
2387 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2388 | task_rq_unlock(rq, &flags); |
2389 | /* might preempt at this point */ | |
2390 | rq = task_rq_lock(p, &flags); | |
2391 | old_state = p->state; | |
2392 | if (!(old_state & state)) | |
2393 | goto out; | |
dd41f596 | 2394 | if (p->se.on_rq) |
1da177e4 LT |
2395 | goto out_running; |
2396 | ||
2397 | this_cpu = smp_processor_id(); | |
2398 | cpu = task_cpu(p); | |
2399 | } | |
2400 | ||
e7693a36 GH |
2401 | #ifdef CONFIG_SCHEDSTATS |
2402 | schedstat_inc(rq, ttwu_count); | |
2403 | if (cpu == this_cpu) | |
2404 | schedstat_inc(rq, ttwu_local); | |
2405 | else { | |
2406 | struct sched_domain *sd; | |
2407 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2408 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2409 | schedstat_inc(sd, ttwu_wake_remote); |
2410 | break; | |
2411 | } | |
2412 | } | |
2413 | } | |
6d6bc0ad | 2414 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2415 | |
1da177e4 LT |
2416 | out_activate: |
2417 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2418 | schedstat_inc(p, se.nr_wakeups); |
2419 | if (sync) | |
2420 | schedstat_inc(p, se.nr_wakeups_sync); | |
2421 | if (orig_cpu != cpu) | |
2422 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2423 | if (cpu == this_cpu) | |
2424 | schedstat_inc(p, se.nr_wakeups_local); | |
2425 | else | |
2426 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2427 | activate_task(rq, p, 1); |
1da177e4 LT |
2428 | success = 1; |
2429 | ||
831451ac PZ |
2430 | /* |
2431 | * Only attribute actual wakeups done by this task. | |
2432 | */ | |
2433 | if (!in_interrupt()) { | |
2434 | struct sched_entity *se = ¤t->se; | |
2435 | u64 sample = se->sum_exec_runtime; | |
2436 | ||
2437 | if (se->last_wakeup) | |
2438 | sample -= se->last_wakeup; | |
2439 | else | |
2440 | sample -= se->start_runtime; | |
2441 | update_avg(&se->avg_wakeup, sample); | |
2442 | ||
2443 | se->last_wakeup = se->sum_exec_runtime; | |
2444 | } | |
2445 | ||
1da177e4 | 2446 | out_running: |
468a15bb | 2447 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2448 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2449 | |
1da177e4 | 2450 | p->state = TASK_RUNNING; |
9a897c5a SR |
2451 | #ifdef CONFIG_SMP |
2452 | if (p->sched_class->task_wake_up) | |
2453 | p->sched_class->task_wake_up(rq, p); | |
2454 | #endif | |
1da177e4 LT |
2455 | out: |
2456 | task_rq_unlock(rq, &flags); | |
2457 | ||
2458 | return success; | |
2459 | } | |
2460 | ||
7ad5b3a5 | 2461 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2462 | { |
d9514f6c | 2463 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2464 | } |
1da177e4 LT |
2465 | EXPORT_SYMBOL(wake_up_process); |
2466 | ||
7ad5b3a5 | 2467 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2468 | { |
2469 | return try_to_wake_up(p, state, 0); | |
2470 | } | |
2471 | ||
1da177e4 LT |
2472 | /* |
2473 | * Perform scheduler related setup for a newly forked process p. | |
2474 | * p is forked by current. | |
dd41f596 IM |
2475 | * |
2476 | * __sched_fork() is basic setup used by init_idle() too: | |
2477 | */ | |
2478 | static void __sched_fork(struct task_struct *p) | |
2479 | { | |
dd41f596 IM |
2480 | p->se.exec_start = 0; |
2481 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2482 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2483 | p->se.last_wakeup = 0; |
2484 | p->se.avg_overlap = 0; | |
831451ac PZ |
2485 | p->se.start_runtime = 0; |
2486 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2487 | |
2488 | #ifdef CONFIG_SCHEDSTATS | |
2489 | p->se.wait_start = 0; | |
dd41f596 IM |
2490 | p->se.sum_sleep_runtime = 0; |
2491 | p->se.sleep_start = 0; | |
dd41f596 IM |
2492 | p->se.block_start = 0; |
2493 | p->se.sleep_max = 0; | |
2494 | p->se.block_max = 0; | |
2495 | p->se.exec_max = 0; | |
eba1ed4b | 2496 | p->se.slice_max = 0; |
dd41f596 | 2497 | p->se.wait_max = 0; |
6cfb0d5d | 2498 | #endif |
476d139c | 2499 | |
fa717060 | 2500 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2501 | p->se.on_rq = 0; |
4a55bd5e | 2502 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2503 | |
e107be36 AK |
2504 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2505 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2506 | #endif | |
2507 | ||
1da177e4 LT |
2508 | /* |
2509 | * We mark the process as running here, but have not actually | |
2510 | * inserted it onto the runqueue yet. This guarantees that | |
2511 | * nobody will actually run it, and a signal or other external | |
2512 | * event cannot wake it up and insert it on the runqueue either. | |
2513 | */ | |
2514 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2515 | } |
2516 | ||
2517 | /* | |
2518 | * fork()/clone()-time setup: | |
2519 | */ | |
2520 | void sched_fork(struct task_struct *p, int clone_flags) | |
2521 | { | |
2522 | int cpu = get_cpu(); | |
2523 | ||
2524 | __sched_fork(p); | |
2525 | ||
2526 | #ifdef CONFIG_SMP | |
2527 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2528 | #endif | |
02e4bac2 | 2529 | set_task_cpu(p, cpu); |
b29739f9 IM |
2530 | |
2531 | /* | |
2532 | * Make sure we do not leak PI boosting priority to the child: | |
2533 | */ | |
2534 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2535 | if (!rt_prio(p->prio)) |
2536 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2537 | |
52f17b6c | 2538 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2539 | if (likely(sched_info_on())) |
52f17b6c | 2540 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2541 | #endif |
d6077cb8 | 2542 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2543 | p->oncpu = 0; |
2544 | #endif | |
1da177e4 | 2545 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2546 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2547 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2548 | #endif |
917b627d GH |
2549 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2550 | ||
476d139c | 2551 | put_cpu(); |
1da177e4 LT |
2552 | } |
2553 | ||
2554 | /* | |
2555 | * wake_up_new_task - wake up a newly created task for the first time. | |
2556 | * | |
2557 | * This function will do some initial scheduler statistics housekeeping | |
2558 | * that must be done for every newly created context, then puts the task | |
2559 | * on the runqueue and wakes it. | |
2560 | */ | |
7ad5b3a5 | 2561 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2562 | { |
2563 | unsigned long flags; | |
dd41f596 | 2564 | struct rq *rq; |
1da177e4 LT |
2565 | |
2566 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2567 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2568 | update_rq_clock(rq); |
1da177e4 LT |
2569 | |
2570 | p->prio = effective_prio(p); | |
2571 | ||
b9dca1e0 | 2572 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2573 | activate_task(rq, p, 0); |
1da177e4 | 2574 | } else { |
1da177e4 | 2575 | /* |
dd41f596 IM |
2576 | * Let the scheduling class do new task startup |
2577 | * management (if any): | |
1da177e4 | 2578 | */ |
ee0827d8 | 2579 | p->sched_class->task_new(rq, p); |
c09595f6 | 2580 | inc_nr_running(rq); |
1da177e4 | 2581 | } |
c71dd42d | 2582 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2583 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2584 | #ifdef CONFIG_SMP |
2585 | if (p->sched_class->task_wake_up) | |
2586 | p->sched_class->task_wake_up(rq, p); | |
2587 | #endif | |
dd41f596 | 2588 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2589 | } |
2590 | ||
e107be36 AK |
2591 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2592 | ||
2593 | /** | |
80dd99b3 | 2594 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2595 | * @notifier: notifier struct to register |
e107be36 AK |
2596 | */ |
2597 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2598 | { | |
2599 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2600 | } | |
2601 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2602 | ||
2603 | /** | |
2604 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2605 | * @notifier: notifier struct to unregister |
e107be36 AK |
2606 | * |
2607 | * This is safe to call from within a preemption notifier. | |
2608 | */ | |
2609 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2610 | { | |
2611 | hlist_del(¬ifier->link); | |
2612 | } | |
2613 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2614 | ||
2615 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2616 | { | |
2617 | struct preempt_notifier *notifier; | |
2618 | struct hlist_node *node; | |
2619 | ||
2620 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2621 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2622 | } | |
2623 | ||
2624 | static void | |
2625 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2626 | struct task_struct *next) | |
2627 | { | |
2628 | struct preempt_notifier *notifier; | |
2629 | struct hlist_node *node; | |
2630 | ||
2631 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2632 | notifier->ops->sched_out(notifier, next); | |
2633 | } | |
2634 | ||
6d6bc0ad | 2635 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2636 | |
2637 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2638 | { | |
2639 | } | |
2640 | ||
2641 | static void | |
2642 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2643 | struct task_struct *next) | |
2644 | { | |
2645 | } | |
2646 | ||
6d6bc0ad | 2647 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2648 | |
4866cde0 NP |
2649 | /** |
2650 | * prepare_task_switch - prepare to switch tasks | |
2651 | * @rq: the runqueue preparing to switch | |
421cee29 | 2652 | * @prev: the current task that is being switched out |
4866cde0 NP |
2653 | * @next: the task we are going to switch to. |
2654 | * | |
2655 | * This is called with the rq lock held and interrupts off. It must | |
2656 | * be paired with a subsequent finish_task_switch after the context | |
2657 | * switch. | |
2658 | * | |
2659 | * prepare_task_switch sets up locking and calls architecture specific | |
2660 | * hooks. | |
2661 | */ | |
e107be36 AK |
2662 | static inline void |
2663 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2664 | struct task_struct *next) | |
4866cde0 | 2665 | { |
e107be36 | 2666 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2667 | prepare_lock_switch(rq, next); |
2668 | prepare_arch_switch(next); | |
2669 | } | |
2670 | ||
1da177e4 LT |
2671 | /** |
2672 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2673 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2674 | * @prev: the thread we just switched away from. |
2675 | * | |
4866cde0 NP |
2676 | * finish_task_switch must be called after the context switch, paired |
2677 | * with a prepare_task_switch call before the context switch. | |
2678 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2679 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2680 | * |
2681 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2682 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2683 | * with the lock held can cause deadlocks; see schedule() for |
2684 | * details.) | |
2685 | */ | |
a9957449 | 2686 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2687 | __releases(rq->lock) |
2688 | { | |
1da177e4 | 2689 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2690 | long prev_state; |
967fc046 GH |
2691 | #ifdef CONFIG_SMP |
2692 | int post_schedule = 0; | |
2693 | ||
2694 | if (current->sched_class->needs_post_schedule) | |
2695 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2696 | #endif | |
1da177e4 LT |
2697 | |
2698 | rq->prev_mm = NULL; | |
2699 | ||
2700 | /* | |
2701 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2702 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2703 | * schedule one last time. The schedule call will never return, and |
2704 | * the scheduled task must drop that reference. | |
c394cc9f | 2705 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2706 | * still held, otherwise prev could be scheduled on another cpu, die |
2707 | * there before we look at prev->state, and then the reference would | |
2708 | * be dropped twice. | |
2709 | * Manfred Spraul <manfred@colorfullife.com> | |
2710 | */ | |
55a101f8 | 2711 | prev_state = prev->state; |
4866cde0 NP |
2712 | finish_arch_switch(prev); |
2713 | finish_lock_switch(rq, prev); | |
9a897c5a | 2714 | #ifdef CONFIG_SMP |
967fc046 | 2715 | if (post_schedule) |
9a897c5a SR |
2716 | current->sched_class->post_schedule(rq); |
2717 | #endif | |
e8fa1362 | 2718 | |
e107be36 | 2719 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2720 | if (mm) |
2721 | mmdrop(mm); | |
c394cc9f | 2722 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2723 | /* |
2724 | * Remove function-return probe instances associated with this | |
2725 | * task and put them back on the free list. | |
9761eea8 | 2726 | */ |
c6fd91f0 | 2727 | kprobe_flush_task(prev); |
1da177e4 | 2728 | put_task_struct(prev); |
c6fd91f0 | 2729 | } |
1da177e4 LT |
2730 | } |
2731 | ||
2732 | /** | |
2733 | * schedule_tail - first thing a freshly forked thread must call. | |
2734 | * @prev: the thread we just switched away from. | |
2735 | */ | |
36c8b586 | 2736 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2737 | __releases(rq->lock) |
2738 | { | |
70b97a7f IM |
2739 | struct rq *rq = this_rq(); |
2740 | ||
4866cde0 NP |
2741 | finish_task_switch(rq, prev); |
2742 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2743 | /* In this case, finish_task_switch does not reenable preemption */ | |
2744 | preempt_enable(); | |
2745 | #endif | |
1da177e4 | 2746 | if (current->set_child_tid) |
b488893a | 2747 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2748 | } |
2749 | ||
2750 | /* | |
2751 | * context_switch - switch to the new MM and the new | |
2752 | * thread's register state. | |
2753 | */ | |
dd41f596 | 2754 | static inline void |
70b97a7f | 2755 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2756 | struct task_struct *next) |
1da177e4 | 2757 | { |
dd41f596 | 2758 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2759 | |
e107be36 | 2760 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2761 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2762 | mm = next->mm; |
2763 | oldmm = prev->active_mm; | |
9226d125 ZA |
2764 | /* |
2765 | * For paravirt, this is coupled with an exit in switch_to to | |
2766 | * combine the page table reload and the switch backend into | |
2767 | * one hypercall. | |
2768 | */ | |
2769 | arch_enter_lazy_cpu_mode(); | |
2770 | ||
dd41f596 | 2771 | if (unlikely(!mm)) { |
1da177e4 LT |
2772 | next->active_mm = oldmm; |
2773 | atomic_inc(&oldmm->mm_count); | |
2774 | enter_lazy_tlb(oldmm, next); | |
2775 | } else | |
2776 | switch_mm(oldmm, mm, next); | |
2777 | ||
dd41f596 | 2778 | if (unlikely(!prev->mm)) { |
1da177e4 | 2779 | prev->active_mm = NULL; |
1da177e4 LT |
2780 | rq->prev_mm = oldmm; |
2781 | } | |
3a5f5e48 IM |
2782 | /* |
2783 | * Since the runqueue lock will be released by the next | |
2784 | * task (which is an invalid locking op but in the case | |
2785 | * of the scheduler it's an obvious special-case), so we | |
2786 | * do an early lockdep release here: | |
2787 | */ | |
2788 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2789 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2790 | #endif |
1da177e4 LT |
2791 | |
2792 | /* Here we just switch the register state and the stack. */ | |
2793 | switch_to(prev, next, prev); | |
2794 | ||
dd41f596 IM |
2795 | barrier(); |
2796 | /* | |
2797 | * this_rq must be evaluated again because prev may have moved | |
2798 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2799 | * frame will be invalid. | |
2800 | */ | |
2801 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2802 | } |
2803 | ||
2804 | /* | |
2805 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2806 | * | |
2807 | * externally visible scheduler statistics: current number of runnable | |
2808 | * threads, current number of uninterruptible-sleeping threads, total | |
2809 | * number of context switches performed since bootup. | |
2810 | */ | |
2811 | unsigned long nr_running(void) | |
2812 | { | |
2813 | unsigned long i, sum = 0; | |
2814 | ||
2815 | for_each_online_cpu(i) | |
2816 | sum += cpu_rq(i)->nr_running; | |
2817 | ||
2818 | return sum; | |
2819 | } | |
2820 | ||
2821 | unsigned long nr_uninterruptible(void) | |
2822 | { | |
2823 | unsigned long i, sum = 0; | |
2824 | ||
0a945022 | 2825 | for_each_possible_cpu(i) |
1da177e4 LT |
2826 | sum += cpu_rq(i)->nr_uninterruptible; |
2827 | ||
2828 | /* | |
2829 | * Since we read the counters lockless, it might be slightly | |
2830 | * inaccurate. Do not allow it to go below zero though: | |
2831 | */ | |
2832 | if (unlikely((long)sum < 0)) | |
2833 | sum = 0; | |
2834 | ||
2835 | return sum; | |
2836 | } | |
2837 | ||
2838 | unsigned long long nr_context_switches(void) | |
2839 | { | |
cc94abfc SR |
2840 | int i; |
2841 | unsigned long long sum = 0; | |
1da177e4 | 2842 | |
0a945022 | 2843 | for_each_possible_cpu(i) |
1da177e4 LT |
2844 | sum += cpu_rq(i)->nr_switches; |
2845 | ||
2846 | return sum; | |
2847 | } | |
2848 | ||
2849 | unsigned long nr_iowait(void) | |
2850 | { | |
2851 | unsigned long i, sum = 0; | |
2852 | ||
0a945022 | 2853 | for_each_possible_cpu(i) |
1da177e4 LT |
2854 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2855 | ||
2856 | return sum; | |
2857 | } | |
2858 | ||
db1b1fef JS |
2859 | unsigned long nr_active(void) |
2860 | { | |
2861 | unsigned long i, running = 0, uninterruptible = 0; | |
2862 | ||
2863 | for_each_online_cpu(i) { | |
2864 | running += cpu_rq(i)->nr_running; | |
2865 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2866 | } | |
2867 | ||
2868 | if (unlikely((long)uninterruptible < 0)) | |
2869 | uninterruptible = 0; | |
2870 | ||
2871 | return running + uninterruptible; | |
2872 | } | |
2873 | ||
48f24c4d | 2874 | /* |
dd41f596 IM |
2875 | * Update rq->cpu_load[] statistics. This function is usually called every |
2876 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2877 | */ |
dd41f596 | 2878 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2879 | { |
495eca49 | 2880 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2881 | int i, scale; |
2882 | ||
2883 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2884 | |
2885 | /* Update our load: */ | |
2886 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2887 | unsigned long old_load, new_load; | |
2888 | ||
2889 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2890 | ||
2891 | old_load = this_rq->cpu_load[i]; | |
2892 | new_load = this_load; | |
a25707f3 IM |
2893 | /* |
2894 | * Round up the averaging division if load is increasing. This | |
2895 | * prevents us from getting stuck on 9 if the load is 10, for | |
2896 | * example. | |
2897 | */ | |
2898 | if (new_load > old_load) | |
2899 | new_load += scale-1; | |
dd41f596 IM |
2900 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2901 | } | |
48f24c4d IM |
2902 | } |
2903 | ||
dd41f596 IM |
2904 | #ifdef CONFIG_SMP |
2905 | ||
1da177e4 LT |
2906 | /* |
2907 | * double_rq_lock - safely lock two runqueues | |
2908 | * | |
2909 | * Note this does not disable interrupts like task_rq_lock, | |
2910 | * you need to do so manually before calling. | |
2911 | */ | |
70b97a7f | 2912 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2913 | __acquires(rq1->lock) |
2914 | __acquires(rq2->lock) | |
2915 | { | |
054b9108 | 2916 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2917 | if (rq1 == rq2) { |
2918 | spin_lock(&rq1->lock); | |
2919 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2920 | } else { | |
c96d145e | 2921 | if (rq1 < rq2) { |
1da177e4 | 2922 | spin_lock(&rq1->lock); |
5e710e37 | 2923 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2924 | } else { |
2925 | spin_lock(&rq2->lock); | |
5e710e37 | 2926 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2927 | } |
2928 | } | |
6e82a3be IM |
2929 | update_rq_clock(rq1); |
2930 | update_rq_clock(rq2); | |
1da177e4 LT |
2931 | } |
2932 | ||
2933 | /* | |
2934 | * double_rq_unlock - safely unlock two runqueues | |
2935 | * | |
2936 | * Note this does not restore interrupts like task_rq_unlock, | |
2937 | * you need to do so manually after calling. | |
2938 | */ | |
70b97a7f | 2939 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2940 | __releases(rq1->lock) |
2941 | __releases(rq2->lock) | |
2942 | { | |
2943 | spin_unlock(&rq1->lock); | |
2944 | if (rq1 != rq2) | |
2945 | spin_unlock(&rq2->lock); | |
2946 | else | |
2947 | __release(rq2->lock); | |
2948 | } | |
2949 | ||
1da177e4 LT |
2950 | /* |
2951 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2952 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2953 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2954 | * the cpu_allowed mask is restored. |
2955 | */ | |
36c8b586 | 2956 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2957 | { |
70b97a7f | 2958 | struct migration_req req; |
1da177e4 | 2959 | unsigned long flags; |
70b97a7f | 2960 | struct rq *rq; |
1da177e4 LT |
2961 | |
2962 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2963 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2964 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2965 | goto out; |
2966 | ||
2967 | /* force the process onto the specified CPU */ | |
2968 | if (migrate_task(p, dest_cpu, &req)) { | |
2969 | /* Need to wait for migration thread (might exit: take ref). */ | |
2970 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2971 | |
1da177e4 LT |
2972 | get_task_struct(mt); |
2973 | task_rq_unlock(rq, &flags); | |
2974 | wake_up_process(mt); | |
2975 | put_task_struct(mt); | |
2976 | wait_for_completion(&req.done); | |
36c8b586 | 2977 | |
1da177e4 LT |
2978 | return; |
2979 | } | |
2980 | out: | |
2981 | task_rq_unlock(rq, &flags); | |
2982 | } | |
2983 | ||
2984 | /* | |
476d139c NP |
2985 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2986 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2987 | */ |
2988 | void sched_exec(void) | |
2989 | { | |
1da177e4 | 2990 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2991 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2992 | put_cpu(); |
476d139c NP |
2993 | if (new_cpu != this_cpu) |
2994 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2995 | } |
2996 | ||
2997 | /* | |
2998 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2999 | * Both runqueues must be locked. | |
3000 | */ | |
dd41f596 IM |
3001 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3002 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3003 | { |
2e1cb74a | 3004 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3005 | set_task_cpu(p, this_cpu); |
dd41f596 | 3006 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3007 | /* |
3008 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3009 | * to be always true for them. | |
3010 | */ | |
15afe09b | 3011 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3012 | } |
3013 | ||
3014 | /* | |
3015 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3016 | */ | |
858119e1 | 3017 | static |
70b97a7f | 3018 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3019 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3020 | int *all_pinned) |
1da177e4 | 3021 | { |
708dc512 | 3022 | int tsk_cache_hot = 0; |
1da177e4 LT |
3023 | /* |
3024 | * We do not migrate tasks that are: | |
3025 | * 1) running (obviously), or | |
3026 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3027 | * 3) are cache-hot on their current CPU. | |
3028 | */ | |
96f874e2 | 3029 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3030 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3031 | return 0; |
cc367732 | 3032 | } |
81026794 NP |
3033 | *all_pinned = 0; |
3034 | ||
cc367732 IM |
3035 | if (task_running(rq, p)) { |
3036 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3037 | return 0; |
cc367732 | 3038 | } |
1da177e4 | 3039 | |
da84d961 IM |
3040 | /* |
3041 | * Aggressive migration if: | |
3042 | * 1) task is cache cold, or | |
3043 | * 2) too many balance attempts have failed. | |
3044 | */ | |
3045 | ||
708dc512 LH |
3046 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3047 | if (!tsk_cache_hot || | |
3048 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3049 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3050 | if (tsk_cache_hot) { |
da84d961 | 3051 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3052 | schedstat_inc(p, se.nr_forced_migrations); |
3053 | } | |
da84d961 IM |
3054 | #endif |
3055 | return 1; | |
3056 | } | |
3057 | ||
708dc512 | 3058 | if (tsk_cache_hot) { |
cc367732 | 3059 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3060 | return 0; |
cc367732 | 3061 | } |
1da177e4 LT |
3062 | return 1; |
3063 | } | |
3064 | ||
e1d1484f PW |
3065 | static unsigned long |
3066 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3067 | unsigned long max_load_move, struct sched_domain *sd, | |
3068 | enum cpu_idle_type idle, int *all_pinned, | |
3069 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3070 | { |
051c6764 | 3071 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3072 | struct task_struct *p; |
3073 | long rem_load_move = max_load_move; | |
1da177e4 | 3074 | |
e1d1484f | 3075 | if (max_load_move == 0) |
1da177e4 LT |
3076 | goto out; |
3077 | ||
81026794 NP |
3078 | pinned = 1; |
3079 | ||
1da177e4 | 3080 | /* |
dd41f596 | 3081 | * Start the load-balancing iterator: |
1da177e4 | 3082 | */ |
dd41f596 IM |
3083 | p = iterator->start(iterator->arg); |
3084 | next: | |
b82d9fdd | 3085 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3086 | goto out; |
051c6764 PZ |
3087 | |
3088 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3089 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3090 | p = iterator->next(iterator->arg); |
3091 | goto next; | |
1da177e4 LT |
3092 | } |
3093 | ||
dd41f596 | 3094 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3095 | pulled++; |
dd41f596 | 3096 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3097 | |
7e96fa58 GH |
3098 | #ifdef CONFIG_PREEMPT |
3099 | /* | |
3100 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3101 | * will stop after the first task is pulled to minimize the critical | |
3102 | * section. | |
3103 | */ | |
3104 | if (idle == CPU_NEWLY_IDLE) | |
3105 | goto out; | |
3106 | #endif | |
3107 | ||
2dd73a4f | 3108 | /* |
b82d9fdd | 3109 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3110 | */ |
e1d1484f | 3111 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3112 | if (p->prio < *this_best_prio) |
3113 | *this_best_prio = p->prio; | |
dd41f596 IM |
3114 | p = iterator->next(iterator->arg); |
3115 | goto next; | |
1da177e4 LT |
3116 | } |
3117 | out: | |
3118 | /* | |
e1d1484f | 3119 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3120 | * so we can safely collect pull_task() stats here rather than |
3121 | * inside pull_task(). | |
3122 | */ | |
3123 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3124 | |
3125 | if (all_pinned) | |
3126 | *all_pinned = pinned; | |
e1d1484f PW |
3127 | |
3128 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3129 | } |
3130 | ||
dd41f596 | 3131 | /* |
43010659 PW |
3132 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3133 | * this_rq, as part of a balancing operation within domain "sd". | |
3134 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3135 | * |
3136 | * Called with both runqueues locked. | |
3137 | */ | |
3138 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3139 | unsigned long max_load_move, |
dd41f596 IM |
3140 | struct sched_domain *sd, enum cpu_idle_type idle, |
3141 | int *all_pinned) | |
3142 | { | |
5522d5d5 | 3143 | const struct sched_class *class = sched_class_highest; |
43010659 | 3144 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3145 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3146 | |
3147 | do { | |
43010659 PW |
3148 | total_load_moved += |
3149 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3150 | max_load_move - total_load_moved, |
a4ac01c3 | 3151 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3152 | class = class->next; |
c4acb2c0 | 3153 | |
7e96fa58 GH |
3154 | #ifdef CONFIG_PREEMPT |
3155 | /* | |
3156 | * NEWIDLE balancing is a source of latency, so preemptible | |
3157 | * kernels will stop after the first task is pulled to minimize | |
3158 | * the critical section. | |
3159 | */ | |
c4acb2c0 GH |
3160 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3161 | break; | |
7e96fa58 | 3162 | #endif |
43010659 | 3163 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3164 | |
43010659 PW |
3165 | return total_load_moved > 0; |
3166 | } | |
3167 | ||
e1d1484f PW |
3168 | static int |
3169 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3170 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3171 | struct rq_iterator *iterator) | |
3172 | { | |
3173 | struct task_struct *p = iterator->start(iterator->arg); | |
3174 | int pinned = 0; | |
3175 | ||
3176 | while (p) { | |
3177 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3178 | pull_task(busiest, p, this_rq, this_cpu); | |
3179 | /* | |
3180 | * Right now, this is only the second place pull_task() | |
3181 | * is called, so we can safely collect pull_task() | |
3182 | * stats here rather than inside pull_task(). | |
3183 | */ | |
3184 | schedstat_inc(sd, lb_gained[idle]); | |
3185 | ||
3186 | return 1; | |
3187 | } | |
3188 | p = iterator->next(iterator->arg); | |
3189 | } | |
3190 | ||
3191 | return 0; | |
3192 | } | |
3193 | ||
43010659 PW |
3194 | /* |
3195 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3196 | * part of active balancing operations within "domain". | |
3197 | * Returns 1 if successful and 0 otherwise. | |
3198 | * | |
3199 | * Called with both runqueues locked. | |
3200 | */ | |
3201 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3202 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3203 | { | |
5522d5d5 | 3204 | const struct sched_class *class; |
43010659 PW |
3205 | |
3206 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3207 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3208 | return 1; |
3209 | ||
3210 | return 0; | |
dd41f596 | 3211 | } |
67bb6c03 | 3212 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3213 | /* |
222d656d GS |
3214 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3215 | * during load balancing. | |
1da177e4 | 3216 | */ |
222d656d GS |
3217 | struct sd_lb_stats { |
3218 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3219 | struct sched_group *this; /* Local group in this sd */ | |
3220 | unsigned long total_load; /* Total load of all groups in sd */ | |
3221 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3222 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3223 | ||
3224 | /** Statistics of this group */ | |
3225 | unsigned long this_load; | |
3226 | unsigned long this_load_per_task; | |
3227 | unsigned long this_nr_running; | |
3228 | ||
3229 | /* Statistics of the busiest group */ | |
3230 | unsigned long max_load; | |
3231 | unsigned long busiest_load_per_task; | |
3232 | unsigned long busiest_nr_running; | |
3233 | ||
3234 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3235 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3236 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3237 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3238 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3239 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3240 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3241 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3242 | #endif |
222d656d | 3243 | }; |
1da177e4 | 3244 | |
d5ac537e | 3245 | /* |
381be78f GS |
3246 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3247 | */ | |
3248 | struct sg_lb_stats { | |
3249 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3250 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3251 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3252 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3253 | unsigned long group_capacity; | |
3254 | int group_imb; /* Is there an imbalance in the group ? */ | |
3255 | }; | |
408ed066 | 3256 | |
67bb6c03 GS |
3257 | /** |
3258 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3259 | * @group: The group whose first cpu is to be returned. | |
3260 | */ | |
3261 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3262 | { | |
3263 | return cpumask_first(sched_group_cpus(group)); | |
3264 | } | |
3265 | ||
3266 | /** | |
3267 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3268 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3269 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3270 | */ | |
3271 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3272 | enum cpu_idle_type idle) | |
3273 | { | |
3274 | int load_idx; | |
3275 | ||
3276 | switch (idle) { | |
3277 | case CPU_NOT_IDLE: | |
7897986b | 3278 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3279 | break; |
3280 | ||
3281 | case CPU_NEWLY_IDLE: | |
7897986b | 3282 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3283 | break; |
3284 | default: | |
7897986b | 3285 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3286 | break; |
3287 | } | |
1da177e4 | 3288 | |
67bb6c03 GS |
3289 | return load_idx; |
3290 | } | |
1da177e4 | 3291 | |
1da177e4 | 3292 | |
c071df18 GS |
3293 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3294 | /** | |
3295 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3296 | * the given sched_domain, during load balancing. | |
3297 | * | |
3298 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3299 | * @sds: Variable containing the statistics for sd. | |
3300 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3301 | */ | |
3302 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3303 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3304 | { | |
3305 | /* | |
3306 | * Busy processors will not participate in power savings | |
3307 | * balance. | |
3308 | */ | |
3309 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3310 | sds->power_savings_balance = 0; | |
3311 | else { | |
3312 | sds->power_savings_balance = 1; | |
3313 | sds->min_nr_running = ULONG_MAX; | |
3314 | sds->leader_nr_running = 0; | |
3315 | } | |
3316 | } | |
783609c6 | 3317 | |
c071df18 GS |
3318 | /** |
3319 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3320 | * sched_domain while performing load balancing. | |
3321 | * | |
3322 | * @group: sched_group belonging to the sched_domain under consideration. | |
3323 | * @sds: Variable containing the statistics of the sched_domain | |
3324 | * @local_group: Does group contain the CPU for which we're performing | |
3325 | * load balancing ? | |
3326 | * @sgs: Variable containing the statistics of the group. | |
3327 | */ | |
3328 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3329 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3330 | { | |
408ed066 | 3331 | |
c071df18 GS |
3332 | if (!sds->power_savings_balance) |
3333 | return; | |
1da177e4 | 3334 | |
c071df18 GS |
3335 | /* |
3336 | * If the local group is idle or completely loaded | |
3337 | * no need to do power savings balance at this domain | |
3338 | */ | |
3339 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3340 | !sds->this_nr_running)) | |
3341 | sds->power_savings_balance = 0; | |
2dd73a4f | 3342 | |
c071df18 GS |
3343 | /* |
3344 | * If a group is already running at full capacity or idle, | |
3345 | * don't include that group in power savings calculations | |
3346 | */ | |
3347 | if (!sds->power_savings_balance || | |
3348 | sgs->sum_nr_running >= sgs->group_capacity || | |
3349 | !sgs->sum_nr_running) | |
3350 | return; | |
5969fe06 | 3351 | |
c071df18 GS |
3352 | /* |
3353 | * Calculate the group which has the least non-idle load. | |
3354 | * This is the group from where we need to pick up the load | |
3355 | * for saving power | |
3356 | */ | |
3357 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3358 | (sgs->sum_nr_running == sds->min_nr_running && | |
3359 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3360 | sds->group_min = group; | |
3361 | sds->min_nr_running = sgs->sum_nr_running; | |
3362 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3363 | sgs->sum_nr_running; | |
3364 | } | |
783609c6 | 3365 | |
c071df18 GS |
3366 | /* |
3367 | * Calculate the group which is almost near its | |
3368 | * capacity but still has some space to pick up some load | |
3369 | * from other group and save more power | |
3370 | */ | |
3371 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3372 | return; | |
1da177e4 | 3373 | |
c071df18 GS |
3374 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3375 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3376 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3377 | sds->group_leader = group; | |
3378 | sds->leader_nr_running = sgs->sum_nr_running; | |
3379 | } | |
3380 | } | |
408ed066 | 3381 | |
c071df18 | 3382 | /** |
d5ac537e | 3383 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3384 | * @sds: Variable containing the statistics of the sched_domain |
3385 | * under consideration. | |
3386 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3387 | * @imbalance: Variable to store the imbalance. | |
3388 | * | |
d5ac537e RD |
3389 | * Description: |
3390 | * Check if we have potential to perform some power-savings balance. | |
3391 | * If yes, set the busiest group to be the least loaded group in the | |
3392 | * sched_domain, so that it's CPUs can be put to idle. | |
3393 | * | |
c071df18 GS |
3394 | * Returns 1 if there is potential to perform power-savings balance. |
3395 | * Else returns 0. | |
3396 | */ | |
3397 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3398 | int this_cpu, unsigned long *imbalance) | |
3399 | { | |
3400 | if (!sds->power_savings_balance) | |
3401 | return 0; | |
1da177e4 | 3402 | |
c071df18 GS |
3403 | if (sds->this != sds->group_leader || |
3404 | sds->group_leader == sds->group_min) | |
3405 | return 0; | |
783609c6 | 3406 | |
c071df18 GS |
3407 | *imbalance = sds->min_load_per_task; |
3408 | sds->busiest = sds->group_min; | |
1da177e4 | 3409 | |
c071df18 GS |
3410 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3411 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3412 | group_first_cpu(sds->group_leader); | |
3413 | } | |
3414 | ||
3415 | return 1; | |
1da177e4 | 3416 | |
c071df18 GS |
3417 | } |
3418 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3419 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3420 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3421 | { | |
3422 | return; | |
3423 | } | |
408ed066 | 3424 | |
c071df18 GS |
3425 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3426 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3427 | { | |
3428 | return; | |
3429 | } | |
3430 | ||
3431 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3432 | int this_cpu, unsigned long *imbalance) | |
3433 | { | |
3434 | return 0; | |
3435 | } | |
3436 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3437 | ||
3438 | ||
1f8c553d GS |
3439 | /** |
3440 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3441 | * @group: sched_group whose statistics are to be updated. | |
3442 | * @this_cpu: Cpu for which load balance is currently performed. | |
3443 | * @idle: Idle status of this_cpu | |
3444 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3445 | * @sd_idle: Idle status of the sched_domain containing group. | |
3446 | * @local_group: Does group contain this_cpu. | |
3447 | * @cpus: Set of cpus considered for load balancing. | |
3448 | * @balance: Should we balance. | |
3449 | * @sgs: variable to hold the statistics for this group. | |
3450 | */ | |
3451 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3452 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3453 | int local_group, const struct cpumask *cpus, | |
3454 | int *balance, struct sg_lb_stats *sgs) | |
3455 | { | |
3456 | unsigned long load, max_cpu_load, min_cpu_load; | |
3457 | int i; | |
3458 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3459 | unsigned long sum_avg_load_per_task; | |
3460 | unsigned long avg_load_per_task; | |
3461 | ||
3462 | if (local_group) | |
3463 | balance_cpu = group_first_cpu(group); | |
3464 | ||
3465 | /* Tally up the load of all CPUs in the group */ | |
3466 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3467 | max_cpu_load = 0; | |
3468 | min_cpu_load = ~0UL; | |
408ed066 | 3469 | |
1f8c553d GS |
3470 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3471 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3472 | |
1f8c553d GS |
3473 | if (*sd_idle && rq->nr_running) |
3474 | *sd_idle = 0; | |
5c45bf27 | 3475 | |
1f8c553d | 3476 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3477 | if (local_group) { |
1f8c553d GS |
3478 | if (idle_cpu(i) && !first_idle_cpu) { |
3479 | first_idle_cpu = 1; | |
3480 | balance_cpu = i; | |
3481 | } | |
3482 | ||
3483 | load = target_load(i, load_idx); | |
3484 | } else { | |
3485 | load = source_load(i, load_idx); | |
3486 | if (load > max_cpu_load) | |
3487 | max_cpu_load = load; | |
3488 | if (min_cpu_load > load) | |
3489 | min_cpu_load = load; | |
1da177e4 | 3490 | } |
5c45bf27 | 3491 | |
1f8c553d GS |
3492 | sgs->group_load += load; |
3493 | sgs->sum_nr_running += rq->nr_running; | |
3494 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3495 | |
1f8c553d GS |
3496 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3497 | } | |
5c45bf27 | 3498 | |
1f8c553d GS |
3499 | /* |
3500 | * First idle cpu or the first cpu(busiest) in this sched group | |
3501 | * is eligible for doing load balancing at this and above | |
3502 | * domains. In the newly idle case, we will allow all the cpu's | |
3503 | * to do the newly idle load balance. | |
3504 | */ | |
3505 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3506 | balance_cpu != this_cpu && balance) { | |
3507 | *balance = 0; | |
3508 | return; | |
3509 | } | |
5c45bf27 | 3510 | |
1f8c553d GS |
3511 | /* Adjust by relative CPU power of the group */ |
3512 | sgs->avg_load = sg_div_cpu_power(group, | |
3513 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3514 | |
1f8c553d GS |
3515 | |
3516 | /* | |
3517 | * Consider the group unbalanced when the imbalance is larger | |
3518 | * than the average weight of two tasks. | |
3519 | * | |
3520 | * APZ: with cgroup the avg task weight can vary wildly and | |
3521 | * might not be a suitable number - should we keep a | |
3522 | * normalized nr_running number somewhere that negates | |
3523 | * the hierarchy? | |
3524 | */ | |
3525 | avg_load_per_task = sg_div_cpu_power(group, | |
3526 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3527 | ||
3528 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3529 | sgs->group_imb = 1; | |
3530 | ||
3531 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3532 | ||
3533 | } | |
dd41f596 | 3534 | |
37abe198 GS |
3535 | /** |
3536 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3537 | * @sd: sched_domain whose statistics are to be updated. | |
3538 | * @this_cpu: Cpu for which load balance is currently performed. | |
3539 | * @idle: Idle status of this_cpu | |
3540 | * @sd_idle: Idle status of the sched_domain containing group. | |
3541 | * @cpus: Set of cpus considered for load balancing. | |
3542 | * @balance: Should we balance. | |
3543 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3544 | */ |
37abe198 GS |
3545 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3546 | enum cpu_idle_type idle, int *sd_idle, | |
3547 | const struct cpumask *cpus, int *balance, | |
3548 | struct sd_lb_stats *sds) | |
1da177e4 | 3549 | { |
222d656d | 3550 | struct sched_group *group = sd->groups; |
37abe198 | 3551 | struct sg_lb_stats sgs; |
222d656d GS |
3552 | int load_idx; |
3553 | ||
c071df18 | 3554 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3555 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3556 | |
3557 | do { | |
1da177e4 | 3558 | int local_group; |
1da177e4 | 3559 | |
758b2cdc RR |
3560 | local_group = cpumask_test_cpu(this_cpu, |
3561 | sched_group_cpus(group)); | |
381be78f | 3562 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3563 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3564 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3565 | |
37abe198 GS |
3566 | if (local_group && balance && !(*balance)) |
3567 | return; | |
783609c6 | 3568 | |
37abe198 GS |
3569 | sds->total_load += sgs.group_load; |
3570 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3571 | |
1da177e4 | 3572 | if (local_group) { |
37abe198 GS |
3573 | sds->this_load = sgs.avg_load; |
3574 | sds->this = group; | |
3575 | sds->this_nr_running = sgs.sum_nr_running; | |
3576 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3577 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3578 | (sgs.sum_nr_running > sgs.group_capacity || |
3579 | sgs.group_imb)) { | |
37abe198 GS |
3580 | sds->max_load = sgs.avg_load; |
3581 | sds->busiest = group; | |
3582 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3583 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3584 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3585 | } |
5c45bf27 | 3586 | |
c071df18 | 3587 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3588 | group = group->next; |
3589 | } while (group != sd->groups); | |
3590 | ||
37abe198 | 3591 | } |
1da177e4 | 3592 | |
2e6f44ae GS |
3593 | /** |
3594 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3595 | * amongst the groups of a sched_domain, during |
3596 | * load balancing. | |
2e6f44ae GS |
3597 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3598 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3599 | * @imbalance: Variable to store the imbalance. | |
3600 | */ | |
3601 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3602 | int this_cpu, unsigned long *imbalance) | |
3603 | { | |
3604 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3605 | unsigned int imbn = 2; | |
3606 | ||
3607 | if (sds->this_nr_running) { | |
3608 | sds->this_load_per_task /= sds->this_nr_running; | |
3609 | if (sds->busiest_load_per_task > | |
3610 | sds->this_load_per_task) | |
3611 | imbn = 1; | |
3612 | } else | |
3613 | sds->this_load_per_task = | |
3614 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3615 | |
2e6f44ae GS |
3616 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3617 | sds->busiest_load_per_task * imbn) { | |
3618 | *imbalance = sds->busiest_load_per_task; | |
3619 | return; | |
3620 | } | |
908a7c1b | 3621 | |
1da177e4 | 3622 | /* |
2e6f44ae GS |
3623 | * OK, we don't have enough imbalance to justify moving tasks, |
3624 | * however we may be able to increase total CPU power used by | |
3625 | * moving them. | |
1da177e4 | 3626 | */ |
2dd73a4f | 3627 | |
2e6f44ae GS |
3628 | pwr_now += sds->busiest->__cpu_power * |
3629 | min(sds->busiest_load_per_task, sds->max_load); | |
3630 | pwr_now += sds->this->__cpu_power * | |
3631 | min(sds->this_load_per_task, sds->this_load); | |
3632 | pwr_now /= SCHED_LOAD_SCALE; | |
3633 | ||
3634 | /* Amount of load we'd subtract */ | |
3635 | tmp = sg_div_cpu_power(sds->busiest, | |
3636 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3637 | if (sds->max_load > tmp) | |
3638 | pwr_move += sds->busiest->__cpu_power * | |
3639 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3640 | ||
3641 | /* Amount of load we'd add */ | |
3642 | if (sds->max_load * sds->busiest->__cpu_power < | |
3643 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3644 | tmp = sg_div_cpu_power(sds->this, | |
3645 | sds->max_load * sds->busiest->__cpu_power); | |
3646 | else | |
3647 | tmp = sg_div_cpu_power(sds->this, | |
3648 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3649 | pwr_move += sds->this->__cpu_power * | |
3650 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3651 | pwr_move /= SCHED_LOAD_SCALE; | |
3652 | ||
3653 | /* Move if we gain throughput */ | |
3654 | if (pwr_move > pwr_now) | |
3655 | *imbalance = sds->busiest_load_per_task; | |
3656 | } | |
dbc523a3 GS |
3657 | |
3658 | /** | |
3659 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3660 | * groups of a given sched_domain during load balance. | |
3661 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3662 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3663 | * @imbalance: The variable to store the imbalance. | |
3664 | */ | |
3665 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3666 | unsigned long *imbalance) | |
3667 | { | |
3668 | unsigned long max_pull; | |
2dd73a4f PW |
3669 | /* |
3670 | * In the presence of smp nice balancing, certain scenarios can have | |
3671 | * max load less than avg load(as we skip the groups at or below | |
3672 | * its cpu_power, while calculating max_load..) | |
3673 | */ | |
dbc523a3 | 3674 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3675 | *imbalance = 0; |
dbc523a3 | 3676 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3677 | } |
0c117f1b SS |
3678 | |
3679 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3680 | max_pull = min(sds->max_load - sds->avg_load, |
3681 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3682 | |
1da177e4 | 3683 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3684 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3685 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3686 | / SCHED_LOAD_SCALE; |
3687 | ||
2dd73a4f PW |
3688 | /* |
3689 | * if *imbalance is less than the average load per runnable task | |
3690 | * there is no gaurantee that any tasks will be moved so we'll have | |
3691 | * a think about bumping its value to force at least one task to be | |
3692 | * moved | |
3693 | */ | |
dbc523a3 GS |
3694 | if (*imbalance < sds->busiest_load_per_task) |
3695 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3696 | |
dbc523a3 | 3697 | } |
37abe198 | 3698 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3699 | |
b7bb4c9b GS |
3700 | /** |
3701 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3702 | * if there is an imbalance. If there isn't an imbalance, and | |
3703 | * the user has opted for power-savings, it returns a group whose | |
3704 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3705 | * such a group exists. | |
3706 | * | |
3707 | * Also calculates the amount of weighted load which should be moved | |
3708 | * to restore balance. | |
3709 | * | |
3710 | * @sd: The sched_domain whose busiest group is to be returned. | |
3711 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3712 | * @imbalance: Variable which stores amount of weighted load which should | |
3713 | * be moved to restore balance/put a group to idle. | |
3714 | * @idle: The idle status of this_cpu. | |
3715 | * @sd_idle: The idleness of sd | |
3716 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3717 | * @balance: Pointer to a variable indicating if this_cpu | |
3718 | * is the appropriate cpu to perform load balancing at this_level. | |
3719 | * | |
3720 | * Returns: - the busiest group if imbalance exists. | |
3721 | * - If no imbalance and user has opted for power-savings balance, | |
3722 | * return the least loaded group whose CPUs can be | |
3723 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3724 | */ |
3725 | static struct sched_group * | |
3726 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3727 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3728 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3729 | { | |
3730 | struct sd_lb_stats sds; | |
1da177e4 | 3731 | |
37abe198 | 3732 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3733 | |
37abe198 GS |
3734 | /* |
3735 | * Compute the various statistics relavent for load balancing at | |
3736 | * this level. | |
3737 | */ | |
3738 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3739 | balance, &sds); | |
3740 | ||
b7bb4c9b GS |
3741 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3742 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3743 | * at this level. | |
3744 | * 2) There is no busy sibling group to pull from. | |
3745 | * 3) This group is the busiest group. | |
3746 | * 4) This group is more busy than the avg busieness at this | |
3747 | * sched_domain. | |
3748 | * 5) The imbalance is within the specified limit. | |
3749 | * 6) Any rebalance would lead to ping-pong | |
3750 | */ | |
37abe198 GS |
3751 | if (balance && !(*balance)) |
3752 | goto ret; | |
1da177e4 | 3753 | |
b7bb4c9b GS |
3754 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3755 | goto out_balanced; | |
1da177e4 | 3756 | |
b7bb4c9b | 3757 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3758 | goto out_balanced; |
1da177e4 | 3759 | |
222d656d | 3760 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3761 | |
b7bb4c9b GS |
3762 | if (sds.this_load >= sds.avg_load) |
3763 | goto out_balanced; | |
3764 | ||
3765 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3766 | goto out_balanced; |
3767 | ||
222d656d GS |
3768 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3769 | if (sds.group_imb) | |
3770 | sds.busiest_load_per_task = | |
3771 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3772 | |
1da177e4 LT |
3773 | /* |
3774 | * We're trying to get all the cpus to the average_load, so we don't | |
3775 | * want to push ourselves above the average load, nor do we wish to | |
3776 | * reduce the max loaded cpu below the average load, as either of these | |
3777 | * actions would just result in more rebalancing later, and ping-pong | |
3778 | * tasks around. Thus we look for the minimum possible imbalance. | |
3779 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3780 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3781 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3782 | * appear as very large values with unsigned longs. |
3783 | */ | |
222d656d | 3784 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3785 | goto out_balanced; |
3786 | ||
dbc523a3 GS |
3787 | /* Looks like there is an imbalance. Compute it */ |
3788 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3789 | return sds.busiest; |
1da177e4 LT |
3790 | |
3791 | out_balanced: | |
c071df18 GS |
3792 | /* |
3793 | * There is no obvious imbalance. But check if we can do some balancing | |
3794 | * to save power. | |
3795 | */ | |
3796 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3797 | return sds.busiest; | |
783609c6 | 3798 | ret: |
1da177e4 LT |
3799 | *imbalance = 0; |
3800 | return NULL; | |
3801 | } | |
3802 | ||
3803 | /* | |
3804 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3805 | */ | |
70b97a7f | 3806 | static struct rq * |
d15bcfdb | 3807 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3808 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3809 | { |
70b97a7f | 3810 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3811 | unsigned long max_load = 0; |
1da177e4 LT |
3812 | int i; |
3813 | ||
758b2cdc | 3814 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3815 | unsigned long wl; |
0a2966b4 | 3816 | |
96f874e2 | 3817 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3818 | continue; |
3819 | ||
48f24c4d | 3820 | rq = cpu_rq(i); |
dd41f596 | 3821 | wl = weighted_cpuload(i); |
2dd73a4f | 3822 | |
dd41f596 | 3823 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3824 | continue; |
1da177e4 | 3825 | |
dd41f596 IM |
3826 | if (wl > max_load) { |
3827 | max_load = wl; | |
48f24c4d | 3828 | busiest = rq; |
1da177e4 LT |
3829 | } |
3830 | } | |
3831 | ||
3832 | return busiest; | |
3833 | } | |
3834 | ||
77391d71 NP |
3835 | /* |
3836 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3837 | * so long as it is large enough. | |
3838 | */ | |
3839 | #define MAX_PINNED_INTERVAL 512 | |
3840 | ||
df7c8e84 RR |
3841 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3842 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3843 | ||
1da177e4 LT |
3844 | /* |
3845 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3846 | * tasks if there is an imbalance. | |
1da177e4 | 3847 | */ |
70b97a7f | 3848 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3849 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 3850 | int *balance) |
1da177e4 | 3851 | { |
43010659 | 3852 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3853 | struct sched_group *group; |
1da177e4 | 3854 | unsigned long imbalance; |
70b97a7f | 3855 | struct rq *busiest; |
fe2eea3f | 3856 | unsigned long flags; |
df7c8e84 | 3857 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 3858 | |
96f874e2 | 3859 | cpumask_setall(cpus); |
7c16ec58 | 3860 | |
89c4710e SS |
3861 | /* |
3862 | * When power savings policy is enabled for the parent domain, idle | |
3863 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3864 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3865 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3866 | */ |
d15bcfdb | 3867 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3868 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3869 | sd_idle = 1; |
1da177e4 | 3870 | |
2d72376b | 3871 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3872 | |
0a2966b4 | 3873 | redo: |
c8cba857 | 3874 | update_shares(sd); |
0a2966b4 | 3875 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3876 | cpus, balance); |
783609c6 | 3877 | |
06066714 | 3878 | if (*balance == 0) |
783609c6 | 3879 | goto out_balanced; |
783609c6 | 3880 | |
1da177e4 LT |
3881 | if (!group) { |
3882 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3883 | goto out_balanced; | |
3884 | } | |
3885 | ||
7c16ec58 | 3886 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3887 | if (!busiest) { |
3888 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3889 | goto out_balanced; | |
3890 | } | |
3891 | ||
db935dbd | 3892 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3893 | |
3894 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3895 | ||
43010659 | 3896 | ld_moved = 0; |
1da177e4 LT |
3897 | if (busiest->nr_running > 1) { |
3898 | /* | |
3899 | * Attempt to move tasks. If find_busiest_group has found | |
3900 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3901 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3902 | * correctly treated as an imbalance. |
3903 | */ | |
fe2eea3f | 3904 | local_irq_save(flags); |
e17224bf | 3905 | double_rq_lock(this_rq, busiest); |
43010659 | 3906 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3907 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3908 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3909 | local_irq_restore(flags); |
81026794 | 3910 | |
46cb4b7c SS |
3911 | /* |
3912 | * some other cpu did the load balance for us. | |
3913 | */ | |
43010659 | 3914 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3915 | resched_cpu(this_cpu); |
3916 | ||
81026794 | 3917 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3918 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3919 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3920 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3921 | goto redo; |
81026794 | 3922 | goto out_balanced; |
0a2966b4 | 3923 | } |
1da177e4 | 3924 | } |
81026794 | 3925 | |
43010659 | 3926 | if (!ld_moved) { |
1da177e4 LT |
3927 | schedstat_inc(sd, lb_failed[idle]); |
3928 | sd->nr_balance_failed++; | |
3929 | ||
3930 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3931 | |
fe2eea3f | 3932 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3933 | |
3934 | /* don't kick the migration_thread, if the curr | |
3935 | * task on busiest cpu can't be moved to this_cpu | |
3936 | */ | |
96f874e2 RR |
3937 | if (!cpumask_test_cpu(this_cpu, |
3938 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3939 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3940 | all_pinned = 1; |
3941 | goto out_one_pinned; | |
3942 | } | |
3943 | ||
1da177e4 LT |
3944 | if (!busiest->active_balance) { |
3945 | busiest->active_balance = 1; | |
3946 | busiest->push_cpu = this_cpu; | |
81026794 | 3947 | active_balance = 1; |
1da177e4 | 3948 | } |
fe2eea3f | 3949 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3950 | if (active_balance) |
1da177e4 LT |
3951 | wake_up_process(busiest->migration_thread); |
3952 | ||
3953 | /* | |
3954 | * We've kicked active balancing, reset the failure | |
3955 | * counter. | |
3956 | */ | |
39507451 | 3957 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3958 | } |
81026794 | 3959 | } else |
1da177e4 LT |
3960 | sd->nr_balance_failed = 0; |
3961 | ||
81026794 | 3962 | if (likely(!active_balance)) { |
1da177e4 LT |
3963 | /* We were unbalanced, so reset the balancing interval */ |
3964 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3965 | } else { |
3966 | /* | |
3967 | * If we've begun active balancing, start to back off. This | |
3968 | * case may not be covered by the all_pinned logic if there | |
3969 | * is only 1 task on the busy runqueue (because we don't call | |
3970 | * move_tasks). | |
3971 | */ | |
3972 | if (sd->balance_interval < sd->max_interval) | |
3973 | sd->balance_interval *= 2; | |
1da177e4 LT |
3974 | } |
3975 | ||
43010659 | 3976 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3977 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3978 | ld_moved = -1; |
3979 | ||
3980 | goto out; | |
1da177e4 LT |
3981 | |
3982 | out_balanced: | |
1da177e4 LT |
3983 | schedstat_inc(sd, lb_balanced[idle]); |
3984 | ||
16cfb1c0 | 3985 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3986 | |
3987 | out_one_pinned: | |
1da177e4 | 3988 | /* tune up the balancing interval */ |
77391d71 NP |
3989 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3990 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3991 | sd->balance_interval *= 2; |
3992 | ||
48f24c4d | 3993 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3994 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3995 | ld_moved = -1; |
3996 | else | |
3997 | ld_moved = 0; | |
3998 | out: | |
c8cba857 PZ |
3999 | if (ld_moved) |
4000 | update_shares(sd); | |
c09595f6 | 4001 | return ld_moved; |
1da177e4 LT |
4002 | } |
4003 | ||
4004 | /* | |
4005 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4006 | * tasks if there is an imbalance. | |
4007 | * | |
d15bcfdb | 4008 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4009 | * this_rq is locked. |
4010 | */ | |
48f24c4d | 4011 | static int |
df7c8e84 | 4012 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4013 | { |
4014 | struct sched_group *group; | |
70b97a7f | 4015 | struct rq *busiest = NULL; |
1da177e4 | 4016 | unsigned long imbalance; |
43010659 | 4017 | int ld_moved = 0; |
5969fe06 | 4018 | int sd_idle = 0; |
969bb4e4 | 4019 | int all_pinned = 0; |
df7c8e84 | 4020 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4021 | |
96f874e2 | 4022 | cpumask_setall(cpus); |
5969fe06 | 4023 | |
89c4710e SS |
4024 | /* |
4025 | * When power savings policy is enabled for the parent domain, idle | |
4026 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4027 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4028 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4029 | */ |
4030 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4031 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4032 | sd_idle = 1; |
1da177e4 | 4033 | |
2d72376b | 4034 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4035 | redo: |
3e5459b4 | 4036 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4037 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4038 | &sd_idle, cpus, NULL); |
1da177e4 | 4039 | if (!group) { |
d15bcfdb | 4040 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4041 | goto out_balanced; |
1da177e4 LT |
4042 | } |
4043 | ||
7c16ec58 | 4044 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4045 | if (!busiest) { |
d15bcfdb | 4046 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4047 | goto out_balanced; |
1da177e4 LT |
4048 | } |
4049 | ||
db935dbd NP |
4050 | BUG_ON(busiest == this_rq); |
4051 | ||
d15bcfdb | 4052 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4053 | |
43010659 | 4054 | ld_moved = 0; |
d6d5cfaf NP |
4055 | if (busiest->nr_running > 1) { |
4056 | /* Attempt to move tasks */ | |
4057 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4058 | /* this_rq->clock is already updated */ |
4059 | update_rq_clock(busiest); | |
43010659 | 4060 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4061 | imbalance, sd, CPU_NEWLY_IDLE, |
4062 | &all_pinned); | |
1b12bbc7 | 4063 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4064 | |
969bb4e4 | 4065 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4066 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4067 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4068 | goto redo; |
4069 | } | |
d6d5cfaf NP |
4070 | } |
4071 | ||
43010659 | 4072 | if (!ld_moved) { |
36dffab6 | 4073 | int active_balance = 0; |
ad273b32 | 4074 | |
d15bcfdb | 4075 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4076 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4077 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4078 | return -1; |
ad273b32 VS |
4079 | |
4080 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4081 | return -1; | |
4082 | ||
4083 | if (sd->nr_balance_failed++ < 2) | |
4084 | return -1; | |
4085 | ||
4086 | /* | |
4087 | * The only task running in a non-idle cpu can be moved to this | |
4088 | * cpu in an attempt to completely freeup the other CPU | |
4089 | * package. The same method used to move task in load_balance() | |
4090 | * have been extended for load_balance_newidle() to speedup | |
4091 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4092 | * | |
4093 | * The package power saving logic comes from | |
4094 | * find_busiest_group(). If there are no imbalance, then | |
4095 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4096 | * f_b_g() will select a group from which a running task may be | |
4097 | * pulled to this cpu in order to make the other package idle. | |
4098 | * If there is no opportunity to make a package idle and if | |
4099 | * there are no imbalance, then f_b_g() will return NULL and no | |
4100 | * action will be taken in load_balance_newidle(). | |
4101 | * | |
4102 | * Under normal task pull operation due to imbalance, there | |
4103 | * will be more than one task in the source run queue and | |
4104 | * move_tasks() will succeed. ld_moved will be true and this | |
4105 | * active balance code will not be triggered. | |
4106 | */ | |
4107 | ||
4108 | /* Lock busiest in correct order while this_rq is held */ | |
4109 | double_lock_balance(this_rq, busiest); | |
4110 | ||
4111 | /* | |
4112 | * don't kick the migration_thread, if the curr | |
4113 | * task on busiest cpu can't be moved to this_cpu | |
4114 | */ | |
6ca09dfc | 4115 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4116 | double_unlock_balance(this_rq, busiest); |
4117 | all_pinned = 1; | |
4118 | return ld_moved; | |
4119 | } | |
4120 | ||
4121 | if (!busiest->active_balance) { | |
4122 | busiest->active_balance = 1; | |
4123 | busiest->push_cpu = this_cpu; | |
4124 | active_balance = 1; | |
4125 | } | |
4126 | ||
4127 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4128 | /* |
4129 | * Should not call ttwu while holding a rq->lock | |
4130 | */ | |
4131 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4132 | if (active_balance) |
4133 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4134 | spin_lock(&this_rq->lock); |
ad273b32 | 4135 | |
5969fe06 | 4136 | } else |
16cfb1c0 | 4137 | sd->nr_balance_failed = 0; |
1da177e4 | 4138 | |
3e5459b4 | 4139 | update_shares_locked(this_rq, sd); |
43010659 | 4140 | return ld_moved; |
16cfb1c0 NP |
4141 | |
4142 | out_balanced: | |
d15bcfdb | 4143 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4144 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4145 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4146 | return -1; |
16cfb1c0 | 4147 | sd->nr_balance_failed = 0; |
48f24c4d | 4148 | |
16cfb1c0 | 4149 | return 0; |
1da177e4 LT |
4150 | } |
4151 | ||
4152 | /* | |
4153 | * idle_balance is called by schedule() if this_cpu is about to become | |
4154 | * idle. Attempts to pull tasks from other CPUs. | |
4155 | */ | |
70b97a7f | 4156 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4157 | { |
4158 | struct sched_domain *sd; | |
efbe027e | 4159 | int pulled_task = 0; |
dd41f596 | 4160 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4161 | |
4162 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4163 | unsigned long interval; |
4164 | ||
4165 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4166 | continue; | |
4167 | ||
4168 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4169 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4170 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4171 | sd); |
92c4ca5c CL |
4172 | |
4173 | interval = msecs_to_jiffies(sd->balance_interval); | |
4174 | if (time_after(next_balance, sd->last_balance + interval)) | |
4175 | next_balance = sd->last_balance + interval; | |
4176 | if (pulled_task) | |
4177 | break; | |
1da177e4 | 4178 | } |
dd41f596 | 4179 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4180 | /* |
4181 | * We are going idle. next_balance may be set based on | |
4182 | * a busy processor. So reset next_balance. | |
4183 | */ | |
4184 | this_rq->next_balance = next_balance; | |
dd41f596 | 4185 | } |
1da177e4 LT |
4186 | } |
4187 | ||
4188 | /* | |
4189 | * active_load_balance is run by migration threads. It pushes running tasks | |
4190 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4191 | * running on each physical CPU where possible, and avoids physical / | |
4192 | * logical imbalances. | |
4193 | * | |
4194 | * Called with busiest_rq locked. | |
4195 | */ | |
70b97a7f | 4196 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4197 | { |
39507451 | 4198 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4199 | struct sched_domain *sd; |
4200 | struct rq *target_rq; | |
39507451 | 4201 | |
48f24c4d | 4202 | /* Is there any task to move? */ |
39507451 | 4203 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4204 | return; |
4205 | ||
4206 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4207 | |
4208 | /* | |
39507451 | 4209 | * This condition is "impossible", if it occurs |
41a2d6cf | 4210 | * we need to fix it. Originally reported by |
39507451 | 4211 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4212 | */ |
39507451 | 4213 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4214 | |
39507451 NP |
4215 | /* move a task from busiest_rq to target_rq */ |
4216 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4217 | update_rq_clock(busiest_rq); |
4218 | update_rq_clock(target_rq); | |
39507451 NP |
4219 | |
4220 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4221 | for_each_domain(target_cpu, sd) { |
39507451 | 4222 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4223 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4224 | break; |
c96d145e | 4225 | } |
39507451 | 4226 | |
48f24c4d | 4227 | if (likely(sd)) { |
2d72376b | 4228 | schedstat_inc(sd, alb_count); |
39507451 | 4229 | |
43010659 PW |
4230 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4231 | sd, CPU_IDLE)) | |
48f24c4d IM |
4232 | schedstat_inc(sd, alb_pushed); |
4233 | else | |
4234 | schedstat_inc(sd, alb_failed); | |
4235 | } | |
1b12bbc7 | 4236 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4237 | } |
4238 | ||
46cb4b7c SS |
4239 | #ifdef CONFIG_NO_HZ |
4240 | static struct { | |
4241 | atomic_t load_balancer; | |
7d1e6a9b | 4242 | cpumask_var_t cpu_mask; |
f711f609 | 4243 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4244 | } nohz ____cacheline_aligned = { |
4245 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4246 | }; |
4247 | ||
f711f609 GS |
4248 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4249 | /** | |
4250 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4251 | * @cpu: The cpu whose lowest level of sched domain is to | |
4252 | * be returned. | |
4253 | * @flag: The flag to check for the lowest sched_domain | |
4254 | * for the given cpu. | |
4255 | * | |
4256 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4257 | */ | |
4258 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4259 | { | |
4260 | struct sched_domain *sd; | |
4261 | ||
4262 | for_each_domain(cpu, sd) | |
4263 | if (sd && (sd->flags & flag)) | |
4264 | break; | |
4265 | ||
4266 | return sd; | |
4267 | } | |
4268 | ||
4269 | /** | |
4270 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4271 | * @cpu: The cpu whose domains we're iterating over. | |
4272 | * @sd: variable holding the value of the power_savings_sd | |
4273 | * for cpu. | |
4274 | * @flag: The flag to filter the sched_domains to be iterated. | |
4275 | * | |
4276 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4277 | * set, starting from the lowest sched_domain to the highest. | |
4278 | */ | |
4279 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4280 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4281 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4282 | ||
4283 | /** | |
4284 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4285 | * @ilb_group: group to be checked for semi-idleness | |
4286 | * | |
4287 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4288 | * | |
4289 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4290 | * and atleast one non-idle CPU. This helper function checks if the given | |
4291 | * sched_group is semi-idle or not. | |
4292 | */ | |
4293 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4294 | { | |
4295 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4296 | sched_group_cpus(ilb_group)); | |
4297 | ||
4298 | /* | |
4299 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4300 | * and atleast one idle cpu. | |
4301 | */ | |
4302 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4303 | return 0; | |
4304 | ||
4305 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4306 | return 0; | |
4307 | ||
4308 | return 1; | |
4309 | } | |
4310 | /** | |
4311 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4312 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4313 | * | |
4314 | * Returns: Returns the id of the idle load balancer if it exists, | |
4315 | * Else, returns >= nr_cpu_ids. | |
4316 | * | |
4317 | * This algorithm picks the idle load balancer such that it belongs to a | |
4318 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4319 | * completely idle packages/cores just for the purpose of idle load balancing | |
4320 | * when there are other idle cpu's which are better suited for that job. | |
4321 | */ | |
4322 | static int find_new_ilb(int cpu) | |
4323 | { | |
4324 | struct sched_domain *sd; | |
4325 | struct sched_group *ilb_group; | |
4326 | ||
4327 | /* | |
4328 | * Have idle load balancer selection from semi-idle packages only | |
4329 | * when power-aware load balancing is enabled | |
4330 | */ | |
4331 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4332 | goto out_done; | |
4333 | ||
4334 | /* | |
4335 | * Optimize for the case when we have no idle CPUs or only one | |
4336 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4337 | */ | |
4338 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4339 | goto out_done; | |
4340 | ||
4341 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4342 | ilb_group = sd->groups; | |
4343 | ||
4344 | do { | |
4345 | if (is_semi_idle_group(ilb_group)) | |
4346 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4347 | ||
4348 | ilb_group = ilb_group->next; | |
4349 | ||
4350 | } while (ilb_group != sd->groups); | |
4351 | } | |
4352 | ||
4353 | out_done: | |
4354 | return cpumask_first(nohz.cpu_mask); | |
4355 | } | |
4356 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4357 | static inline int find_new_ilb(int call_cpu) | |
4358 | { | |
4359 | return first_cpu(nohz.cpu_mask); | |
4360 | } | |
4361 | #endif | |
4362 | ||
7835b98b | 4363 | /* |
46cb4b7c SS |
4364 | * This routine will try to nominate the ilb (idle load balancing) |
4365 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4366 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4367 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4368 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4369 | * arrives... | |
4370 | * | |
4371 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4372 | * for idle load balancing. ilb owner will still be part of | |
4373 | * nohz.cpu_mask.. | |
7835b98b | 4374 | * |
46cb4b7c SS |
4375 | * While stopping the tick, this cpu will become the ilb owner if there |
4376 | * is no other owner. And will be the owner till that cpu becomes busy | |
4377 | * or if all cpus in the system stop their ticks at which point | |
4378 | * there is no need for ilb owner. | |
4379 | * | |
4380 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4381 | * next busy scheduler_tick() | |
4382 | */ | |
4383 | int select_nohz_load_balancer(int stop_tick) | |
4384 | { | |
4385 | int cpu = smp_processor_id(); | |
4386 | ||
4387 | if (stop_tick) { | |
46cb4b7c SS |
4388 | cpu_rq(cpu)->in_nohz_recently = 1; |
4389 | ||
483b4ee6 SS |
4390 | if (!cpu_active(cpu)) { |
4391 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4392 | return 0; | |
4393 | ||
4394 | /* | |
4395 | * If we are going offline and still the leader, | |
4396 | * give up! | |
4397 | */ | |
46cb4b7c SS |
4398 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4399 | BUG(); | |
483b4ee6 | 4400 | |
46cb4b7c SS |
4401 | return 0; |
4402 | } | |
4403 | ||
483b4ee6 SS |
4404 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4405 | ||
46cb4b7c | 4406 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4407 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4408 | if (atomic_read(&nohz.load_balancer) == cpu) |
4409 | atomic_set(&nohz.load_balancer, -1); | |
4410 | return 0; | |
4411 | } | |
4412 | ||
4413 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4414 | /* make me the ilb owner */ | |
4415 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4416 | return 1; | |
4417 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
4418 | return 1; | |
4419 | } else { | |
7d1e6a9b | 4420 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4421 | return 0; |
4422 | ||
7d1e6a9b | 4423 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4424 | |
4425 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4426 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4427 | BUG(); | |
4428 | } | |
4429 | return 0; | |
4430 | } | |
4431 | #endif | |
4432 | ||
4433 | static DEFINE_SPINLOCK(balancing); | |
4434 | ||
4435 | /* | |
7835b98b CL |
4436 | * It checks each scheduling domain to see if it is due to be balanced, |
4437 | * and initiates a balancing operation if so. | |
4438 | * | |
4439 | * Balancing parameters are set up in arch_init_sched_domains. | |
4440 | */ | |
a9957449 | 4441 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4442 | { |
46cb4b7c SS |
4443 | int balance = 1; |
4444 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4445 | unsigned long interval; |
4446 | struct sched_domain *sd; | |
46cb4b7c | 4447 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4448 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4449 | int update_next_balance = 0; |
d07355f5 | 4450 | int need_serialize; |
1da177e4 | 4451 | |
46cb4b7c | 4452 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4453 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4454 | continue; | |
4455 | ||
4456 | interval = sd->balance_interval; | |
d15bcfdb | 4457 | if (idle != CPU_IDLE) |
1da177e4 LT |
4458 | interval *= sd->busy_factor; |
4459 | ||
4460 | /* scale ms to jiffies */ | |
4461 | interval = msecs_to_jiffies(interval); | |
4462 | if (unlikely(!interval)) | |
4463 | interval = 1; | |
dd41f596 IM |
4464 | if (interval > HZ*NR_CPUS/10) |
4465 | interval = HZ*NR_CPUS/10; | |
4466 | ||
d07355f5 | 4467 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4468 | |
d07355f5 | 4469 | if (need_serialize) { |
08c183f3 CL |
4470 | if (!spin_trylock(&balancing)) |
4471 | goto out; | |
4472 | } | |
4473 | ||
c9819f45 | 4474 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4475 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4476 | /* |
4477 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4478 | * longer idle, or one of our SMT siblings is |
4479 | * not idle. | |
4480 | */ | |
d15bcfdb | 4481 | idle = CPU_NOT_IDLE; |
1da177e4 | 4482 | } |
1bd77f2d | 4483 | sd->last_balance = jiffies; |
1da177e4 | 4484 | } |
d07355f5 | 4485 | if (need_serialize) |
08c183f3 CL |
4486 | spin_unlock(&balancing); |
4487 | out: | |
f549da84 | 4488 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4489 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4490 | update_next_balance = 1; |
4491 | } | |
783609c6 SS |
4492 | |
4493 | /* | |
4494 | * Stop the load balance at this level. There is another | |
4495 | * CPU in our sched group which is doing load balancing more | |
4496 | * actively. | |
4497 | */ | |
4498 | if (!balance) | |
4499 | break; | |
1da177e4 | 4500 | } |
f549da84 SS |
4501 | |
4502 | /* | |
4503 | * next_balance will be updated only when there is a need. | |
4504 | * When the cpu is attached to null domain for ex, it will not be | |
4505 | * updated. | |
4506 | */ | |
4507 | if (likely(update_next_balance)) | |
4508 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4509 | } |
4510 | ||
4511 | /* | |
4512 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4513 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4514 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4515 | */ | |
4516 | static void run_rebalance_domains(struct softirq_action *h) | |
4517 | { | |
dd41f596 IM |
4518 | int this_cpu = smp_processor_id(); |
4519 | struct rq *this_rq = cpu_rq(this_cpu); | |
4520 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4521 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4522 | |
dd41f596 | 4523 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4524 | |
4525 | #ifdef CONFIG_NO_HZ | |
4526 | /* | |
4527 | * If this cpu is the owner for idle load balancing, then do the | |
4528 | * balancing on behalf of the other idle cpus whose ticks are | |
4529 | * stopped. | |
4530 | */ | |
dd41f596 IM |
4531 | if (this_rq->idle_at_tick && |
4532 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4533 | struct rq *rq; |
4534 | int balance_cpu; | |
4535 | ||
7d1e6a9b RR |
4536 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4537 | if (balance_cpu == this_cpu) | |
4538 | continue; | |
4539 | ||
46cb4b7c SS |
4540 | /* |
4541 | * If this cpu gets work to do, stop the load balancing | |
4542 | * work being done for other cpus. Next load | |
4543 | * balancing owner will pick it up. | |
4544 | */ | |
4545 | if (need_resched()) | |
4546 | break; | |
4547 | ||
de0cf899 | 4548 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4549 | |
4550 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4551 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4552 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4553 | } |
4554 | } | |
4555 | #endif | |
4556 | } | |
4557 | ||
8a0be9ef FW |
4558 | static inline int on_null_domain(int cpu) |
4559 | { | |
4560 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4561 | } | |
4562 | ||
46cb4b7c SS |
4563 | /* |
4564 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4565 | * | |
4566 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4567 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4568 | * if the whole system is idle. | |
4569 | */ | |
dd41f596 | 4570 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4571 | { |
46cb4b7c SS |
4572 | #ifdef CONFIG_NO_HZ |
4573 | /* | |
4574 | * If we were in the nohz mode recently and busy at the current | |
4575 | * scheduler tick, then check if we need to nominate new idle | |
4576 | * load balancer. | |
4577 | */ | |
4578 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4579 | rq->in_nohz_recently = 0; | |
4580 | ||
4581 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4582 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4583 | atomic_set(&nohz.load_balancer, -1); |
4584 | } | |
4585 | ||
4586 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4587 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4588 | |
434d53b0 | 4589 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4590 | resched_cpu(ilb); |
4591 | } | |
4592 | } | |
4593 | ||
4594 | /* | |
4595 | * If this cpu is idle and doing idle load balancing for all the | |
4596 | * cpus with ticks stopped, is it time for that to stop? | |
4597 | */ | |
4598 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4599 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4600 | resched_cpu(cpu); |
4601 | return; | |
4602 | } | |
4603 | ||
4604 | /* | |
4605 | * If this cpu is idle and the idle load balancing is done by | |
4606 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4607 | */ | |
4608 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4609 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4610 | return; |
4611 | #endif | |
8a0be9ef FW |
4612 | /* Don't need to rebalance while attached to NULL domain */ |
4613 | if (time_after_eq(jiffies, rq->next_balance) && | |
4614 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4615 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4616 | } |
dd41f596 IM |
4617 | |
4618 | #else /* CONFIG_SMP */ | |
4619 | ||
1da177e4 LT |
4620 | /* |
4621 | * on UP we do not need to balance between CPUs: | |
4622 | */ | |
70b97a7f | 4623 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4624 | { |
4625 | } | |
dd41f596 | 4626 | |
1da177e4 LT |
4627 | #endif |
4628 | ||
1da177e4 LT |
4629 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4630 | ||
4631 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4632 | ||
4633 | /* | |
c5f8d995 | 4634 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4635 | * @p in case that task is currently running. |
c5f8d995 HS |
4636 | * |
4637 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4638 | */ |
c5f8d995 HS |
4639 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4640 | { | |
4641 | u64 ns = 0; | |
4642 | ||
4643 | if (task_current(rq, p)) { | |
4644 | update_rq_clock(rq); | |
4645 | ns = rq->clock - p->se.exec_start; | |
4646 | if ((s64)ns < 0) | |
4647 | ns = 0; | |
4648 | } | |
4649 | ||
4650 | return ns; | |
4651 | } | |
4652 | ||
bb34d92f | 4653 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4654 | { |
1da177e4 | 4655 | unsigned long flags; |
41b86e9c | 4656 | struct rq *rq; |
bb34d92f | 4657 | u64 ns = 0; |
48f24c4d | 4658 | |
41b86e9c | 4659 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4660 | ns = do_task_delta_exec(p, rq); |
4661 | task_rq_unlock(rq, &flags); | |
1508487e | 4662 | |
c5f8d995 HS |
4663 | return ns; |
4664 | } | |
f06febc9 | 4665 | |
c5f8d995 HS |
4666 | /* |
4667 | * Return accounted runtime for the task. | |
4668 | * In case the task is currently running, return the runtime plus current's | |
4669 | * pending runtime that have not been accounted yet. | |
4670 | */ | |
4671 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4672 | { | |
4673 | unsigned long flags; | |
4674 | struct rq *rq; | |
4675 | u64 ns = 0; | |
4676 | ||
4677 | rq = task_rq_lock(p, &flags); | |
4678 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4679 | task_rq_unlock(rq, &flags); | |
4680 | ||
4681 | return ns; | |
4682 | } | |
48f24c4d | 4683 | |
c5f8d995 HS |
4684 | /* |
4685 | * Return sum_exec_runtime for the thread group. | |
4686 | * In case the task is currently running, return the sum plus current's | |
4687 | * pending runtime that have not been accounted yet. | |
4688 | * | |
4689 | * Note that the thread group might have other running tasks as well, | |
4690 | * so the return value not includes other pending runtime that other | |
4691 | * running tasks might have. | |
4692 | */ | |
4693 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4694 | { | |
4695 | struct task_cputime totals; | |
4696 | unsigned long flags; | |
4697 | struct rq *rq; | |
4698 | u64 ns; | |
4699 | ||
4700 | rq = task_rq_lock(p, &flags); | |
4701 | thread_group_cputime(p, &totals); | |
4702 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4703 | task_rq_unlock(rq, &flags); |
48f24c4d | 4704 | |
1da177e4 LT |
4705 | return ns; |
4706 | } | |
4707 | ||
1da177e4 LT |
4708 | /* |
4709 | * Account user cpu time to a process. | |
4710 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4711 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4712 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4713 | */ |
457533a7 MS |
4714 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4715 | cputime_t cputime_scaled) | |
1da177e4 LT |
4716 | { |
4717 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4718 | cputime64_t tmp; | |
4719 | ||
457533a7 | 4720 | /* Add user time to process. */ |
1da177e4 | 4721 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4722 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4723 | account_group_user_time(p, cputime); |
1da177e4 LT |
4724 | |
4725 | /* Add user time to cpustat. */ | |
4726 | tmp = cputime_to_cputime64(cputime); | |
4727 | if (TASK_NICE(p) > 0) | |
4728 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4729 | else | |
4730 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4731 | |
4732 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4733 | /* Account for user time used */ |
4734 | acct_update_integrals(p); | |
1da177e4 LT |
4735 | } |
4736 | ||
94886b84 LV |
4737 | /* |
4738 | * Account guest cpu time to a process. | |
4739 | * @p: the process that the cpu time gets accounted to | |
4740 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4741 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4742 | */ |
457533a7 MS |
4743 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4744 | cputime_t cputime_scaled) | |
94886b84 LV |
4745 | { |
4746 | cputime64_t tmp; | |
4747 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4748 | ||
4749 | tmp = cputime_to_cputime64(cputime); | |
4750 | ||
457533a7 | 4751 | /* Add guest time to process. */ |
94886b84 | 4752 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4753 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4754 | account_group_user_time(p, cputime); |
94886b84 LV |
4755 | p->gtime = cputime_add(p->gtime, cputime); |
4756 | ||
457533a7 | 4757 | /* Add guest time to cpustat. */ |
94886b84 LV |
4758 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4759 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4760 | } | |
4761 | ||
1da177e4 LT |
4762 | /* |
4763 | * Account system cpu time to a process. | |
4764 | * @p: the process that the cpu time gets accounted to | |
4765 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4766 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4767 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4768 | */ |
4769 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4770 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4771 | { |
4772 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4773 | cputime64_t tmp; |
4774 | ||
983ed7a6 | 4775 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4776 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4777 | return; |
4778 | } | |
94886b84 | 4779 | |
457533a7 | 4780 | /* Add system time to process. */ |
1da177e4 | 4781 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4782 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4783 | account_group_system_time(p, cputime); |
1da177e4 LT |
4784 | |
4785 | /* Add system time to cpustat. */ | |
4786 | tmp = cputime_to_cputime64(cputime); | |
4787 | if (hardirq_count() - hardirq_offset) | |
4788 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4789 | else if (softirq_count()) | |
4790 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4791 | else |
79741dd3 MS |
4792 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4793 | ||
ef12fefa BR |
4794 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4795 | ||
1da177e4 LT |
4796 | /* Account for system time used */ |
4797 | acct_update_integrals(p); | |
1da177e4 LT |
4798 | } |
4799 | ||
c66f08be | 4800 | /* |
1da177e4 | 4801 | * Account for involuntary wait time. |
1da177e4 | 4802 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4803 | */ |
79741dd3 | 4804 | void account_steal_time(cputime_t cputime) |
c66f08be | 4805 | { |
79741dd3 MS |
4806 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4807 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4808 | ||
4809 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4810 | } |
4811 | ||
1da177e4 | 4812 | /* |
79741dd3 MS |
4813 | * Account for idle time. |
4814 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4815 | */ |
79741dd3 | 4816 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4817 | { |
4818 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4819 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4820 | struct rq *rq = this_rq(); |
1da177e4 | 4821 | |
79741dd3 MS |
4822 | if (atomic_read(&rq->nr_iowait) > 0) |
4823 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4824 | else | |
4825 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4826 | } |
4827 | ||
79741dd3 MS |
4828 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4829 | ||
4830 | /* | |
4831 | * Account a single tick of cpu time. | |
4832 | * @p: the process that the cpu time gets accounted to | |
4833 | * @user_tick: indicates if the tick is a user or a system tick | |
4834 | */ | |
4835 | void account_process_tick(struct task_struct *p, int user_tick) | |
4836 | { | |
4837 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4838 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4839 | struct rq *rq = this_rq(); | |
4840 | ||
4841 | if (user_tick) | |
4842 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
4843 | else if (p != rq->idle) | |
4844 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, | |
4845 | one_jiffy_scaled); | |
4846 | else | |
4847 | account_idle_time(one_jiffy); | |
4848 | } | |
4849 | ||
4850 | /* | |
4851 | * Account multiple ticks of steal time. | |
4852 | * @p: the process from which the cpu time has been stolen | |
4853 | * @ticks: number of stolen ticks | |
4854 | */ | |
4855 | void account_steal_ticks(unsigned long ticks) | |
4856 | { | |
4857 | account_steal_time(jiffies_to_cputime(ticks)); | |
4858 | } | |
4859 | ||
4860 | /* | |
4861 | * Account multiple ticks of idle time. | |
4862 | * @ticks: number of stolen ticks | |
4863 | */ | |
4864 | void account_idle_ticks(unsigned long ticks) | |
4865 | { | |
4866 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4867 | } |
4868 | ||
79741dd3 MS |
4869 | #endif |
4870 | ||
49048622 BS |
4871 | /* |
4872 | * Use precise platform statistics if available: | |
4873 | */ | |
4874 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4875 | cputime_t task_utime(struct task_struct *p) | |
4876 | { | |
4877 | return p->utime; | |
4878 | } | |
4879 | ||
4880 | cputime_t task_stime(struct task_struct *p) | |
4881 | { | |
4882 | return p->stime; | |
4883 | } | |
4884 | #else | |
4885 | cputime_t task_utime(struct task_struct *p) | |
4886 | { | |
4887 | clock_t utime = cputime_to_clock_t(p->utime), | |
4888 | total = utime + cputime_to_clock_t(p->stime); | |
4889 | u64 temp; | |
4890 | ||
4891 | /* | |
4892 | * Use CFS's precise accounting: | |
4893 | */ | |
4894 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4895 | ||
4896 | if (total) { | |
4897 | temp *= utime; | |
4898 | do_div(temp, total); | |
4899 | } | |
4900 | utime = (clock_t)temp; | |
4901 | ||
4902 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4903 | return p->prev_utime; | |
4904 | } | |
4905 | ||
4906 | cputime_t task_stime(struct task_struct *p) | |
4907 | { | |
4908 | clock_t stime; | |
4909 | ||
4910 | /* | |
4911 | * Use CFS's precise accounting. (we subtract utime from | |
4912 | * the total, to make sure the total observed by userspace | |
4913 | * grows monotonically - apps rely on that): | |
4914 | */ | |
4915 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4916 | cputime_to_clock_t(task_utime(p)); | |
4917 | ||
4918 | if (stime >= 0) | |
4919 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4920 | ||
4921 | return p->prev_stime; | |
4922 | } | |
4923 | #endif | |
4924 | ||
4925 | inline cputime_t task_gtime(struct task_struct *p) | |
4926 | { | |
4927 | return p->gtime; | |
4928 | } | |
4929 | ||
7835b98b CL |
4930 | /* |
4931 | * This function gets called by the timer code, with HZ frequency. | |
4932 | * We call it with interrupts disabled. | |
4933 | * | |
4934 | * It also gets called by the fork code, when changing the parent's | |
4935 | * timeslices. | |
4936 | */ | |
4937 | void scheduler_tick(void) | |
4938 | { | |
7835b98b CL |
4939 | int cpu = smp_processor_id(); |
4940 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4941 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4942 | |
4943 | sched_clock_tick(); | |
dd41f596 IM |
4944 | |
4945 | spin_lock(&rq->lock); | |
3e51f33f | 4946 | update_rq_clock(rq); |
f1a438d8 | 4947 | update_cpu_load(rq); |
fa85ae24 | 4948 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4949 | spin_unlock(&rq->lock); |
7835b98b | 4950 | |
e418e1c2 | 4951 | #ifdef CONFIG_SMP |
dd41f596 IM |
4952 | rq->idle_at_tick = idle_cpu(cpu); |
4953 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4954 | #endif |
1da177e4 LT |
4955 | } |
4956 | ||
7e49fcce | 4957 | unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4958 | { |
4959 | if (in_lock_functions(addr)) { | |
4960 | addr = CALLER_ADDR2; | |
4961 | if (in_lock_functions(addr)) | |
4962 | addr = CALLER_ADDR3; | |
4963 | } | |
4964 | return addr; | |
4965 | } | |
1da177e4 | 4966 | |
7e49fcce SR |
4967 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4968 | defined(CONFIG_PREEMPT_TRACER)) | |
4969 | ||
43627582 | 4970 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4971 | { |
6cd8a4bb | 4972 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4973 | /* |
4974 | * Underflow? | |
4975 | */ | |
9a11b49a IM |
4976 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4977 | return; | |
6cd8a4bb | 4978 | #endif |
1da177e4 | 4979 | preempt_count() += val; |
6cd8a4bb | 4980 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4981 | /* |
4982 | * Spinlock count overflowing soon? | |
4983 | */ | |
33859f7f MOS |
4984 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4985 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4986 | #endif |
4987 | if (preempt_count() == val) | |
4988 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4989 | } |
4990 | EXPORT_SYMBOL(add_preempt_count); | |
4991 | ||
43627582 | 4992 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4993 | { |
6cd8a4bb | 4994 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4995 | /* |
4996 | * Underflow? | |
4997 | */ | |
01e3eb82 | 4998 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4999 | return; |
1da177e4 LT |
5000 | /* |
5001 | * Is the spinlock portion underflowing? | |
5002 | */ | |
9a11b49a IM |
5003 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5004 | !(preempt_count() & PREEMPT_MASK))) | |
5005 | return; | |
6cd8a4bb | 5006 | #endif |
9a11b49a | 5007 | |
6cd8a4bb SR |
5008 | if (preempt_count() == val) |
5009 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5010 | preempt_count() -= val; |
5011 | } | |
5012 | EXPORT_SYMBOL(sub_preempt_count); | |
5013 | ||
5014 | #endif | |
5015 | ||
5016 | /* | |
dd41f596 | 5017 | * Print scheduling while atomic bug: |
1da177e4 | 5018 | */ |
dd41f596 | 5019 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5020 | { |
838225b4 SS |
5021 | struct pt_regs *regs = get_irq_regs(); |
5022 | ||
5023 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5024 | prev->comm, prev->pid, preempt_count()); | |
5025 | ||
dd41f596 | 5026 | debug_show_held_locks(prev); |
e21f5b15 | 5027 | print_modules(); |
dd41f596 IM |
5028 | if (irqs_disabled()) |
5029 | print_irqtrace_events(prev); | |
838225b4 SS |
5030 | |
5031 | if (regs) | |
5032 | show_regs(regs); | |
5033 | else | |
5034 | dump_stack(); | |
dd41f596 | 5035 | } |
1da177e4 | 5036 | |
dd41f596 IM |
5037 | /* |
5038 | * Various schedule()-time debugging checks and statistics: | |
5039 | */ | |
5040 | static inline void schedule_debug(struct task_struct *prev) | |
5041 | { | |
1da177e4 | 5042 | /* |
41a2d6cf | 5043 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5044 | * schedule() atomically, we ignore that path for now. |
5045 | * Otherwise, whine if we are scheduling when we should not be. | |
5046 | */ | |
3f33a7ce | 5047 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5048 | __schedule_bug(prev); |
5049 | ||
1da177e4 LT |
5050 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5051 | ||
2d72376b | 5052 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5053 | #ifdef CONFIG_SCHEDSTATS |
5054 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5055 | schedstat_inc(this_rq(), bkl_count); |
5056 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5057 | } |
5058 | #endif | |
dd41f596 IM |
5059 | } |
5060 | ||
df1c99d4 MG |
5061 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5062 | { | |
5063 | if (prev->state == TASK_RUNNING) { | |
5064 | u64 runtime = prev->se.sum_exec_runtime; | |
5065 | ||
5066 | runtime -= prev->se.prev_sum_exec_runtime; | |
5067 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5068 | ||
5069 | /* | |
5070 | * In order to avoid avg_overlap growing stale when we are | |
5071 | * indeed overlapping and hence not getting put to sleep, grow | |
5072 | * the avg_overlap on preemption. | |
5073 | * | |
5074 | * We use the average preemption runtime because that | |
5075 | * correlates to the amount of cache footprint a task can | |
5076 | * build up. | |
5077 | */ | |
5078 | update_avg(&prev->se.avg_overlap, runtime); | |
5079 | } | |
5080 | prev->sched_class->put_prev_task(rq, prev); | |
5081 | } | |
5082 | ||
dd41f596 IM |
5083 | /* |
5084 | * Pick up the highest-prio task: | |
5085 | */ | |
5086 | static inline struct task_struct * | |
b67802ea | 5087 | pick_next_task(struct rq *rq) |
dd41f596 | 5088 | { |
5522d5d5 | 5089 | const struct sched_class *class; |
dd41f596 | 5090 | struct task_struct *p; |
1da177e4 LT |
5091 | |
5092 | /* | |
dd41f596 IM |
5093 | * Optimization: we know that if all tasks are in |
5094 | * the fair class we can call that function directly: | |
1da177e4 | 5095 | */ |
dd41f596 | 5096 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5097 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5098 | if (likely(p)) |
5099 | return p; | |
1da177e4 LT |
5100 | } |
5101 | ||
dd41f596 IM |
5102 | class = sched_class_highest; |
5103 | for ( ; ; ) { | |
fb8d4724 | 5104 | p = class->pick_next_task(rq); |
dd41f596 IM |
5105 | if (p) |
5106 | return p; | |
5107 | /* | |
5108 | * Will never be NULL as the idle class always | |
5109 | * returns a non-NULL p: | |
5110 | */ | |
5111 | class = class->next; | |
5112 | } | |
5113 | } | |
1da177e4 | 5114 | |
dd41f596 IM |
5115 | /* |
5116 | * schedule() is the main scheduler function. | |
5117 | */ | |
41719b03 | 5118 | asmlinkage void __sched __schedule(void) |
dd41f596 IM |
5119 | { |
5120 | struct task_struct *prev, *next; | |
67ca7bde | 5121 | unsigned long *switch_count; |
dd41f596 | 5122 | struct rq *rq; |
31656519 | 5123 | int cpu; |
dd41f596 | 5124 | |
dd41f596 IM |
5125 | cpu = smp_processor_id(); |
5126 | rq = cpu_rq(cpu); | |
5127 | rcu_qsctr_inc(cpu); | |
5128 | prev = rq->curr; | |
5129 | switch_count = &prev->nivcsw; | |
5130 | ||
5131 | release_kernel_lock(prev); | |
5132 | need_resched_nonpreemptible: | |
5133 | ||
5134 | schedule_debug(prev); | |
1da177e4 | 5135 | |
31656519 | 5136 | if (sched_feat(HRTICK)) |
f333fdc9 | 5137 | hrtick_clear(rq); |
8f4d37ec | 5138 | |
8cd162ce | 5139 | spin_lock_irq(&rq->lock); |
3e51f33f | 5140 | update_rq_clock(rq); |
1e819950 | 5141 | clear_tsk_need_resched(prev); |
1da177e4 | 5142 | |
1da177e4 | 5143 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5144 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5145 | prev->state = TASK_RUNNING; |
16882c1e | 5146 | else |
2e1cb74a | 5147 | deactivate_task(rq, prev, 1); |
dd41f596 | 5148 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5149 | } |
5150 | ||
9a897c5a SR |
5151 | #ifdef CONFIG_SMP |
5152 | if (prev->sched_class->pre_schedule) | |
5153 | prev->sched_class->pre_schedule(rq, prev); | |
5154 | #endif | |
f65eda4f | 5155 | |
dd41f596 | 5156 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5157 | idle_balance(cpu, rq); |
1da177e4 | 5158 | |
df1c99d4 | 5159 | put_prev_task(rq, prev); |
b67802ea | 5160 | next = pick_next_task(rq); |
1da177e4 | 5161 | |
1da177e4 | 5162 | if (likely(prev != next)) { |
673a90a1 DS |
5163 | sched_info_switch(prev, next); |
5164 | ||
1da177e4 LT |
5165 | rq->nr_switches++; |
5166 | rq->curr = next; | |
5167 | ++*switch_count; | |
5168 | ||
dd41f596 | 5169 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5170 | /* |
5171 | * the context switch might have flipped the stack from under | |
5172 | * us, hence refresh the local variables. | |
5173 | */ | |
5174 | cpu = smp_processor_id(); | |
5175 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5176 | } else |
5177 | spin_unlock_irq(&rq->lock); | |
5178 | ||
8f4d37ec | 5179 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5180 | goto need_resched_nonpreemptible; |
41719b03 | 5181 | } |
8f4d37ec | 5182 | |
41719b03 PZ |
5183 | asmlinkage void __sched schedule(void) |
5184 | { | |
5185 | need_resched: | |
5186 | preempt_disable(); | |
5187 | __schedule(); | |
1da177e4 LT |
5188 | preempt_enable_no_resched(); |
5189 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
5190 | goto need_resched; | |
5191 | } | |
1da177e4 LT |
5192 | EXPORT_SYMBOL(schedule); |
5193 | ||
0d66bf6d PZ |
5194 | #ifdef CONFIG_SMP |
5195 | /* | |
5196 | * Look out! "owner" is an entirely speculative pointer | |
5197 | * access and not reliable. | |
5198 | */ | |
5199 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5200 | { | |
5201 | unsigned int cpu; | |
5202 | struct rq *rq; | |
5203 | ||
5204 | if (!sched_feat(OWNER_SPIN)) | |
5205 | return 0; | |
5206 | ||
5207 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5208 | /* | |
5209 | * Need to access the cpu field knowing that | |
5210 | * DEBUG_PAGEALLOC could have unmapped it if | |
5211 | * the mutex owner just released it and exited. | |
5212 | */ | |
5213 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5214 | goto out; | |
5215 | #else | |
5216 | cpu = owner->cpu; | |
5217 | #endif | |
5218 | ||
5219 | /* | |
5220 | * Even if the access succeeded (likely case), | |
5221 | * the cpu field may no longer be valid. | |
5222 | */ | |
5223 | if (cpu >= nr_cpumask_bits) | |
5224 | goto out; | |
5225 | ||
5226 | /* | |
5227 | * We need to validate that we can do a | |
5228 | * get_cpu() and that we have the percpu area. | |
5229 | */ | |
5230 | if (!cpu_online(cpu)) | |
5231 | goto out; | |
5232 | ||
5233 | rq = cpu_rq(cpu); | |
5234 | ||
5235 | for (;;) { | |
5236 | /* | |
5237 | * Owner changed, break to re-assess state. | |
5238 | */ | |
5239 | if (lock->owner != owner) | |
5240 | break; | |
5241 | ||
5242 | /* | |
5243 | * Is that owner really running on that cpu? | |
5244 | */ | |
5245 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5246 | return 0; | |
5247 | ||
5248 | cpu_relax(); | |
5249 | } | |
5250 | out: | |
5251 | return 1; | |
5252 | } | |
5253 | #endif | |
5254 | ||
1da177e4 LT |
5255 | #ifdef CONFIG_PREEMPT |
5256 | /* | |
2ed6e34f | 5257 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5258 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5259 | * occur there and call schedule directly. |
5260 | */ | |
5261 | asmlinkage void __sched preempt_schedule(void) | |
5262 | { | |
5263 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5264 | |
1da177e4 LT |
5265 | /* |
5266 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5267 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5268 | */ |
beed33a8 | 5269 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5270 | return; |
5271 | ||
3a5c359a AK |
5272 | do { |
5273 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5274 | schedule(); |
3a5c359a | 5275 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5276 | |
3a5c359a AK |
5277 | /* |
5278 | * Check again in case we missed a preemption opportunity | |
5279 | * between schedule and now. | |
5280 | */ | |
5281 | barrier(); | |
5ed0cec0 | 5282 | } while (need_resched()); |
1da177e4 | 5283 | } |
1da177e4 LT |
5284 | EXPORT_SYMBOL(preempt_schedule); |
5285 | ||
5286 | /* | |
2ed6e34f | 5287 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5288 | * off of irq context. |
5289 | * Note, that this is called and return with irqs disabled. This will | |
5290 | * protect us against recursive calling from irq. | |
5291 | */ | |
5292 | asmlinkage void __sched preempt_schedule_irq(void) | |
5293 | { | |
5294 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5295 | |
2ed6e34f | 5296 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5297 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5298 | ||
3a5c359a AK |
5299 | do { |
5300 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5301 | local_irq_enable(); |
5302 | schedule(); | |
5303 | local_irq_disable(); | |
3a5c359a | 5304 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5305 | |
3a5c359a AK |
5306 | /* |
5307 | * Check again in case we missed a preemption opportunity | |
5308 | * between schedule and now. | |
5309 | */ | |
5310 | barrier(); | |
5ed0cec0 | 5311 | } while (need_resched()); |
1da177e4 LT |
5312 | } |
5313 | ||
5314 | #endif /* CONFIG_PREEMPT */ | |
5315 | ||
95cdf3b7 IM |
5316 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5317 | void *key) | |
1da177e4 | 5318 | { |
48f24c4d | 5319 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5320 | } |
1da177e4 LT |
5321 | EXPORT_SYMBOL(default_wake_function); |
5322 | ||
5323 | /* | |
41a2d6cf IM |
5324 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5325 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5326 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5327 | * | |
5328 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5329 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5330 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5331 | */ | |
777c6c5f JW |
5332 | void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
5333 | int nr_exclusive, int sync, void *key) | |
1da177e4 | 5334 | { |
2e45874c | 5335 | wait_queue_t *curr, *next; |
1da177e4 | 5336 | |
2e45874c | 5337 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5338 | unsigned flags = curr->flags; |
5339 | ||
1da177e4 | 5340 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5341 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5342 | break; |
5343 | } | |
5344 | } | |
5345 | ||
5346 | /** | |
5347 | * __wake_up - wake up threads blocked on a waitqueue. | |
5348 | * @q: the waitqueue | |
5349 | * @mode: which threads | |
5350 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5351 | * @key: is directly passed to the wakeup function |
1da177e4 | 5352 | */ |
7ad5b3a5 | 5353 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5354 | int nr_exclusive, void *key) |
1da177e4 LT |
5355 | { |
5356 | unsigned long flags; | |
5357 | ||
5358 | spin_lock_irqsave(&q->lock, flags); | |
5359 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5360 | spin_unlock_irqrestore(&q->lock, flags); | |
5361 | } | |
1da177e4 LT |
5362 | EXPORT_SYMBOL(__wake_up); |
5363 | ||
5364 | /* | |
5365 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5366 | */ | |
7ad5b3a5 | 5367 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5368 | { |
5369 | __wake_up_common(q, mode, 1, 0, NULL); | |
5370 | } | |
5371 | ||
4ede816a DL |
5372 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5373 | { | |
5374 | __wake_up_common(q, mode, 1, 0, key); | |
5375 | } | |
5376 | ||
1da177e4 | 5377 | /** |
4ede816a | 5378 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5379 | * @q: the waitqueue |
5380 | * @mode: which threads | |
5381 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5382 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5383 | * |
5384 | * The sync wakeup differs that the waker knows that it will schedule | |
5385 | * away soon, so while the target thread will be woken up, it will not | |
5386 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5387 | * with each other. This can prevent needless bouncing between CPUs. | |
5388 | * | |
5389 | * On UP it can prevent extra preemption. | |
5390 | */ | |
4ede816a DL |
5391 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5392 | int nr_exclusive, void *key) | |
1da177e4 LT |
5393 | { |
5394 | unsigned long flags; | |
5395 | int sync = 1; | |
5396 | ||
5397 | if (unlikely(!q)) | |
5398 | return; | |
5399 | ||
5400 | if (unlikely(!nr_exclusive)) | |
5401 | sync = 0; | |
5402 | ||
5403 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5404 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5405 | spin_unlock_irqrestore(&q->lock, flags); |
5406 | } | |
4ede816a DL |
5407 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5408 | ||
5409 | /* | |
5410 | * __wake_up_sync - see __wake_up_sync_key() | |
5411 | */ | |
5412 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5413 | { | |
5414 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5415 | } | |
1da177e4 LT |
5416 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5417 | ||
65eb3dc6 KD |
5418 | /** |
5419 | * complete: - signals a single thread waiting on this completion | |
5420 | * @x: holds the state of this particular completion | |
5421 | * | |
5422 | * This will wake up a single thread waiting on this completion. Threads will be | |
5423 | * awakened in the same order in which they were queued. | |
5424 | * | |
5425 | * See also complete_all(), wait_for_completion() and related routines. | |
5426 | */ | |
b15136e9 | 5427 | void complete(struct completion *x) |
1da177e4 LT |
5428 | { |
5429 | unsigned long flags; | |
5430 | ||
5431 | spin_lock_irqsave(&x->wait.lock, flags); | |
5432 | x->done++; | |
d9514f6c | 5433 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5434 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5435 | } | |
5436 | EXPORT_SYMBOL(complete); | |
5437 | ||
65eb3dc6 KD |
5438 | /** |
5439 | * complete_all: - signals all threads waiting on this completion | |
5440 | * @x: holds the state of this particular completion | |
5441 | * | |
5442 | * This will wake up all threads waiting on this particular completion event. | |
5443 | */ | |
b15136e9 | 5444 | void complete_all(struct completion *x) |
1da177e4 LT |
5445 | { |
5446 | unsigned long flags; | |
5447 | ||
5448 | spin_lock_irqsave(&x->wait.lock, flags); | |
5449 | x->done += UINT_MAX/2; | |
d9514f6c | 5450 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5451 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5452 | } | |
5453 | EXPORT_SYMBOL(complete_all); | |
5454 | ||
8cbbe86d AK |
5455 | static inline long __sched |
5456 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5457 | { |
1da177e4 LT |
5458 | if (!x->done) { |
5459 | DECLARE_WAITQUEUE(wait, current); | |
5460 | ||
5461 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5462 | __add_wait_queue_tail(&x->wait, &wait); | |
5463 | do { | |
94d3d824 | 5464 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5465 | timeout = -ERESTARTSYS; |
5466 | break; | |
8cbbe86d AK |
5467 | } |
5468 | __set_current_state(state); | |
1da177e4 LT |
5469 | spin_unlock_irq(&x->wait.lock); |
5470 | timeout = schedule_timeout(timeout); | |
5471 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5472 | } while (!x->done && timeout); |
1da177e4 | 5473 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5474 | if (!x->done) |
5475 | return timeout; | |
1da177e4 LT |
5476 | } |
5477 | x->done--; | |
ea71a546 | 5478 | return timeout ?: 1; |
1da177e4 | 5479 | } |
1da177e4 | 5480 | |
8cbbe86d AK |
5481 | static long __sched |
5482 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5483 | { |
1da177e4 LT |
5484 | might_sleep(); |
5485 | ||
5486 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5487 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5488 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5489 | return timeout; |
5490 | } | |
1da177e4 | 5491 | |
65eb3dc6 KD |
5492 | /** |
5493 | * wait_for_completion: - waits for completion of a task | |
5494 | * @x: holds the state of this particular completion | |
5495 | * | |
5496 | * This waits to be signaled for completion of a specific task. It is NOT | |
5497 | * interruptible and there is no timeout. | |
5498 | * | |
5499 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5500 | * and interrupt capability. Also see complete(). | |
5501 | */ | |
b15136e9 | 5502 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5503 | { |
5504 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5505 | } |
8cbbe86d | 5506 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5507 | |
65eb3dc6 KD |
5508 | /** |
5509 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5510 | * @x: holds the state of this particular completion | |
5511 | * @timeout: timeout value in jiffies | |
5512 | * | |
5513 | * This waits for either a completion of a specific task to be signaled or for a | |
5514 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5515 | * interruptible. | |
5516 | */ | |
b15136e9 | 5517 | unsigned long __sched |
8cbbe86d | 5518 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5519 | { |
8cbbe86d | 5520 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5521 | } |
8cbbe86d | 5522 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5523 | |
65eb3dc6 KD |
5524 | /** |
5525 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5526 | * @x: holds the state of this particular completion | |
5527 | * | |
5528 | * This waits for completion of a specific task to be signaled. It is | |
5529 | * interruptible. | |
5530 | */ | |
8cbbe86d | 5531 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5532 | { |
51e97990 AK |
5533 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5534 | if (t == -ERESTARTSYS) | |
5535 | return t; | |
5536 | return 0; | |
0fec171c | 5537 | } |
8cbbe86d | 5538 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5539 | |
65eb3dc6 KD |
5540 | /** |
5541 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5542 | * @x: holds the state of this particular completion | |
5543 | * @timeout: timeout value in jiffies | |
5544 | * | |
5545 | * This waits for either a completion of a specific task to be signaled or for a | |
5546 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5547 | */ | |
b15136e9 | 5548 | unsigned long __sched |
8cbbe86d AK |
5549 | wait_for_completion_interruptible_timeout(struct completion *x, |
5550 | unsigned long timeout) | |
0fec171c | 5551 | { |
8cbbe86d | 5552 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5553 | } |
8cbbe86d | 5554 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5555 | |
65eb3dc6 KD |
5556 | /** |
5557 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5558 | * @x: holds the state of this particular completion | |
5559 | * | |
5560 | * This waits to be signaled for completion of a specific task. It can be | |
5561 | * interrupted by a kill signal. | |
5562 | */ | |
009e577e MW |
5563 | int __sched wait_for_completion_killable(struct completion *x) |
5564 | { | |
5565 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5566 | if (t == -ERESTARTSYS) | |
5567 | return t; | |
5568 | return 0; | |
5569 | } | |
5570 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5571 | ||
be4de352 DC |
5572 | /** |
5573 | * try_wait_for_completion - try to decrement a completion without blocking | |
5574 | * @x: completion structure | |
5575 | * | |
5576 | * Returns: 0 if a decrement cannot be done without blocking | |
5577 | * 1 if a decrement succeeded. | |
5578 | * | |
5579 | * If a completion is being used as a counting completion, | |
5580 | * attempt to decrement the counter without blocking. This | |
5581 | * enables us to avoid waiting if the resource the completion | |
5582 | * is protecting is not available. | |
5583 | */ | |
5584 | bool try_wait_for_completion(struct completion *x) | |
5585 | { | |
5586 | int ret = 1; | |
5587 | ||
5588 | spin_lock_irq(&x->wait.lock); | |
5589 | if (!x->done) | |
5590 | ret = 0; | |
5591 | else | |
5592 | x->done--; | |
5593 | spin_unlock_irq(&x->wait.lock); | |
5594 | return ret; | |
5595 | } | |
5596 | EXPORT_SYMBOL(try_wait_for_completion); | |
5597 | ||
5598 | /** | |
5599 | * completion_done - Test to see if a completion has any waiters | |
5600 | * @x: completion structure | |
5601 | * | |
5602 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5603 | * 1 if there are no waiters. | |
5604 | * | |
5605 | */ | |
5606 | bool completion_done(struct completion *x) | |
5607 | { | |
5608 | int ret = 1; | |
5609 | ||
5610 | spin_lock_irq(&x->wait.lock); | |
5611 | if (!x->done) | |
5612 | ret = 0; | |
5613 | spin_unlock_irq(&x->wait.lock); | |
5614 | return ret; | |
5615 | } | |
5616 | EXPORT_SYMBOL(completion_done); | |
5617 | ||
8cbbe86d AK |
5618 | static long __sched |
5619 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5620 | { |
0fec171c IM |
5621 | unsigned long flags; |
5622 | wait_queue_t wait; | |
5623 | ||
5624 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5625 | |
8cbbe86d | 5626 | __set_current_state(state); |
1da177e4 | 5627 | |
8cbbe86d AK |
5628 | spin_lock_irqsave(&q->lock, flags); |
5629 | __add_wait_queue(q, &wait); | |
5630 | spin_unlock(&q->lock); | |
5631 | timeout = schedule_timeout(timeout); | |
5632 | spin_lock_irq(&q->lock); | |
5633 | __remove_wait_queue(q, &wait); | |
5634 | spin_unlock_irqrestore(&q->lock, flags); | |
5635 | ||
5636 | return timeout; | |
5637 | } | |
5638 | ||
5639 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5640 | { | |
5641 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5642 | } |
1da177e4 LT |
5643 | EXPORT_SYMBOL(interruptible_sleep_on); |
5644 | ||
0fec171c | 5645 | long __sched |
95cdf3b7 | 5646 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5647 | { |
8cbbe86d | 5648 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5649 | } |
1da177e4 LT |
5650 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5651 | ||
0fec171c | 5652 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5653 | { |
8cbbe86d | 5654 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5655 | } |
1da177e4 LT |
5656 | EXPORT_SYMBOL(sleep_on); |
5657 | ||
0fec171c | 5658 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5659 | { |
8cbbe86d | 5660 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5661 | } |
1da177e4 LT |
5662 | EXPORT_SYMBOL(sleep_on_timeout); |
5663 | ||
b29739f9 IM |
5664 | #ifdef CONFIG_RT_MUTEXES |
5665 | ||
5666 | /* | |
5667 | * rt_mutex_setprio - set the current priority of a task | |
5668 | * @p: task | |
5669 | * @prio: prio value (kernel-internal form) | |
5670 | * | |
5671 | * This function changes the 'effective' priority of a task. It does | |
5672 | * not touch ->normal_prio like __setscheduler(). | |
5673 | * | |
5674 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5675 | */ | |
36c8b586 | 5676 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5677 | { |
5678 | unsigned long flags; | |
83b699ed | 5679 | int oldprio, on_rq, running; |
70b97a7f | 5680 | struct rq *rq; |
cb469845 | 5681 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5682 | |
5683 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5684 | ||
5685 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5686 | update_rq_clock(rq); |
b29739f9 | 5687 | |
d5f9f942 | 5688 | oldprio = p->prio; |
dd41f596 | 5689 | on_rq = p->se.on_rq; |
051a1d1a | 5690 | running = task_current(rq, p); |
0e1f3483 | 5691 | if (on_rq) |
69be72c1 | 5692 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5693 | if (running) |
5694 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5695 | |
5696 | if (rt_prio(prio)) | |
5697 | p->sched_class = &rt_sched_class; | |
5698 | else | |
5699 | p->sched_class = &fair_sched_class; | |
5700 | ||
b29739f9 IM |
5701 | p->prio = prio; |
5702 | ||
0e1f3483 HS |
5703 | if (running) |
5704 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5705 | if (on_rq) { |
8159f87e | 5706 | enqueue_task(rq, p, 0); |
cb469845 SR |
5707 | |
5708 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5709 | } |
5710 | task_rq_unlock(rq, &flags); | |
5711 | } | |
5712 | ||
5713 | #endif | |
5714 | ||
36c8b586 | 5715 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5716 | { |
dd41f596 | 5717 | int old_prio, delta, on_rq; |
1da177e4 | 5718 | unsigned long flags; |
70b97a7f | 5719 | struct rq *rq; |
1da177e4 LT |
5720 | |
5721 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5722 | return; | |
5723 | /* | |
5724 | * We have to be careful, if called from sys_setpriority(), | |
5725 | * the task might be in the middle of scheduling on another CPU. | |
5726 | */ | |
5727 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5728 | update_rq_clock(rq); |
1da177e4 LT |
5729 | /* |
5730 | * The RT priorities are set via sched_setscheduler(), but we still | |
5731 | * allow the 'normal' nice value to be set - but as expected | |
5732 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5733 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5734 | */ |
e05606d3 | 5735 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5736 | p->static_prio = NICE_TO_PRIO(nice); |
5737 | goto out_unlock; | |
5738 | } | |
dd41f596 | 5739 | on_rq = p->se.on_rq; |
c09595f6 | 5740 | if (on_rq) |
69be72c1 | 5741 | dequeue_task(rq, p, 0); |
1da177e4 | 5742 | |
1da177e4 | 5743 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5744 | set_load_weight(p); |
b29739f9 IM |
5745 | old_prio = p->prio; |
5746 | p->prio = effective_prio(p); | |
5747 | delta = p->prio - old_prio; | |
1da177e4 | 5748 | |
dd41f596 | 5749 | if (on_rq) { |
8159f87e | 5750 | enqueue_task(rq, p, 0); |
1da177e4 | 5751 | /* |
d5f9f942 AM |
5752 | * If the task increased its priority or is running and |
5753 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5754 | */ |
d5f9f942 | 5755 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5756 | resched_task(rq->curr); |
5757 | } | |
5758 | out_unlock: | |
5759 | task_rq_unlock(rq, &flags); | |
5760 | } | |
1da177e4 LT |
5761 | EXPORT_SYMBOL(set_user_nice); |
5762 | ||
e43379f1 MM |
5763 | /* |
5764 | * can_nice - check if a task can reduce its nice value | |
5765 | * @p: task | |
5766 | * @nice: nice value | |
5767 | */ | |
36c8b586 | 5768 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5769 | { |
024f4747 MM |
5770 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5771 | int nice_rlim = 20 - nice; | |
48f24c4d | 5772 | |
e43379f1 MM |
5773 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5774 | capable(CAP_SYS_NICE)); | |
5775 | } | |
5776 | ||
1da177e4 LT |
5777 | #ifdef __ARCH_WANT_SYS_NICE |
5778 | ||
5779 | /* | |
5780 | * sys_nice - change the priority of the current process. | |
5781 | * @increment: priority increment | |
5782 | * | |
5783 | * sys_setpriority is a more generic, but much slower function that | |
5784 | * does similar things. | |
5785 | */ | |
5add95d4 | 5786 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5787 | { |
48f24c4d | 5788 | long nice, retval; |
1da177e4 LT |
5789 | |
5790 | /* | |
5791 | * Setpriority might change our priority at the same moment. | |
5792 | * We don't have to worry. Conceptually one call occurs first | |
5793 | * and we have a single winner. | |
5794 | */ | |
e43379f1 MM |
5795 | if (increment < -40) |
5796 | increment = -40; | |
1da177e4 LT |
5797 | if (increment > 40) |
5798 | increment = 40; | |
5799 | ||
2b8f836f | 5800 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5801 | if (nice < -20) |
5802 | nice = -20; | |
5803 | if (nice > 19) | |
5804 | nice = 19; | |
5805 | ||
e43379f1 MM |
5806 | if (increment < 0 && !can_nice(current, nice)) |
5807 | return -EPERM; | |
5808 | ||
1da177e4 LT |
5809 | retval = security_task_setnice(current, nice); |
5810 | if (retval) | |
5811 | return retval; | |
5812 | ||
5813 | set_user_nice(current, nice); | |
5814 | return 0; | |
5815 | } | |
5816 | ||
5817 | #endif | |
5818 | ||
5819 | /** | |
5820 | * task_prio - return the priority value of a given task. | |
5821 | * @p: the task in question. | |
5822 | * | |
5823 | * This is the priority value as seen by users in /proc. | |
5824 | * RT tasks are offset by -200. Normal tasks are centered | |
5825 | * around 0, value goes from -16 to +15. | |
5826 | */ | |
36c8b586 | 5827 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5828 | { |
5829 | return p->prio - MAX_RT_PRIO; | |
5830 | } | |
5831 | ||
5832 | /** | |
5833 | * task_nice - return the nice value of a given task. | |
5834 | * @p: the task in question. | |
5835 | */ | |
36c8b586 | 5836 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5837 | { |
5838 | return TASK_NICE(p); | |
5839 | } | |
150d8bed | 5840 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5841 | |
5842 | /** | |
5843 | * idle_cpu - is a given cpu idle currently? | |
5844 | * @cpu: the processor in question. | |
5845 | */ | |
5846 | int idle_cpu(int cpu) | |
5847 | { | |
5848 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5849 | } | |
5850 | ||
1da177e4 LT |
5851 | /** |
5852 | * idle_task - return the idle task for a given cpu. | |
5853 | * @cpu: the processor in question. | |
5854 | */ | |
36c8b586 | 5855 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5856 | { |
5857 | return cpu_rq(cpu)->idle; | |
5858 | } | |
5859 | ||
5860 | /** | |
5861 | * find_process_by_pid - find a process with a matching PID value. | |
5862 | * @pid: the pid in question. | |
5863 | */ | |
a9957449 | 5864 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5865 | { |
228ebcbe | 5866 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5867 | } |
5868 | ||
5869 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5870 | static void |
5871 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5872 | { |
dd41f596 | 5873 | BUG_ON(p->se.on_rq); |
48f24c4d | 5874 | |
1da177e4 | 5875 | p->policy = policy; |
dd41f596 IM |
5876 | switch (p->policy) { |
5877 | case SCHED_NORMAL: | |
5878 | case SCHED_BATCH: | |
5879 | case SCHED_IDLE: | |
5880 | p->sched_class = &fair_sched_class; | |
5881 | break; | |
5882 | case SCHED_FIFO: | |
5883 | case SCHED_RR: | |
5884 | p->sched_class = &rt_sched_class; | |
5885 | break; | |
5886 | } | |
5887 | ||
1da177e4 | 5888 | p->rt_priority = prio; |
b29739f9 IM |
5889 | p->normal_prio = normal_prio(p); |
5890 | /* we are holding p->pi_lock already */ | |
5891 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5892 | set_load_weight(p); |
1da177e4 LT |
5893 | } |
5894 | ||
c69e8d9c DH |
5895 | /* |
5896 | * check the target process has a UID that matches the current process's | |
5897 | */ | |
5898 | static bool check_same_owner(struct task_struct *p) | |
5899 | { | |
5900 | const struct cred *cred = current_cred(), *pcred; | |
5901 | bool match; | |
5902 | ||
5903 | rcu_read_lock(); | |
5904 | pcred = __task_cred(p); | |
5905 | match = (cred->euid == pcred->euid || | |
5906 | cred->euid == pcred->uid); | |
5907 | rcu_read_unlock(); | |
5908 | return match; | |
5909 | } | |
5910 | ||
961ccddd RR |
5911 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5912 | struct sched_param *param, bool user) | |
1da177e4 | 5913 | { |
83b699ed | 5914 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5915 | unsigned long flags; |
cb469845 | 5916 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5917 | struct rq *rq; |
1da177e4 | 5918 | |
66e5393a SR |
5919 | /* may grab non-irq protected spin_locks */ |
5920 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5921 | recheck: |
5922 | /* double check policy once rq lock held */ | |
5923 | if (policy < 0) | |
5924 | policy = oldpolicy = p->policy; | |
5925 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5926 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5927 | policy != SCHED_IDLE) | |
b0a9499c | 5928 | return -EINVAL; |
1da177e4 LT |
5929 | /* |
5930 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5931 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5932 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5933 | */ |
5934 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5935 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5936 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5937 | return -EINVAL; |
e05606d3 | 5938 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5939 | return -EINVAL; |
5940 | ||
37e4ab3f OC |
5941 | /* |
5942 | * Allow unprivileged RT tasks to decrease priority: | |
5943 | */ | |
961ccddd | 5944 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5945 | if (rt_policy(policy)) { |
8dc3e909 | 5946 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5947 | |
5948 | if (!lock_task_sighand(p, &flags)) | |
5949 | return -ESRCH; | |
5950 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5951 | unlock_task_sighand(p, &flags); | |
5952 | ||
5953 | /* can't set/change the rt policy */ | |
5954 | if (policy != p->policy && !rlim_rtprio) | |
5955 | return -EPERM; | |
5956 | ||
5957 | /* can't increase priority */ | |
5958 | if (param->sched_priority > p->rt_priority && | |
5959 | param->sched_priority > rlim_rtprio) | |
5960 | return -EPERM; | |
5961 | } | |
dd41f596 IM |
5962 | /* |
5963 | * Like positive nice levels, dont allow tasks to | |
5964 | * move out of SCHED_IDLE either: | |
5965 | */ | |
5966 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5967 | return -EPERM; | |
5fe1d75f | 5968 | |
37e4ab3f | 5969 | /* can't change other user's priorities */ |
c69e8d9c | 5970 | if (!check_same_owner(p)) |
37e4ab3f OC |
5971 | return -EPERM; |
5972 | } | |
1da177e4 | 5973 | |
725aad24 | 5974 | if (user) { |
b68aa230 | 5975 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5976 | /* |
5977 | * Do not allow realtime tasks into groups that have no runtime | |
5978 | * assigned. | |
5979 | */ | |
9a7e0b18 PZ |
5980 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5981 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5982 | return -EPERM; |
b68aa230 PZ |
5983 | #endif |
5984 | ||
725aad24 JF |
5985 | retval = security_task_setscheduler(p, policy, param); |
5986 | if (retval) | |
5987 | return retval; | |
5988 | } | |
5989 | ||
b29739f9 IM |
5990 | /* |
5991 | * make sure no PI-waiters arrive (or leave) while we are | |
5992 | * changing the priority of the task: | |
5993 | */ | |
5994 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5995 | /* |
5996 | * To be able to change p->policy safely, the apropriate | |
5997 | * runqueue lock must be held. | |
5998 | */ | |
b29739f9 | 5999 | rq = __task_rq_lock(p); |
1da177e4 LT |
6000 | /* recheck policy now with rq lock held */ |
6001 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6002 | policy = oldpolicy = -1; | |
b29739f9 IM |
6003 | __task_rq_unlock(rq); |
6004 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6005 | goto recheck; |
6006 | } | |
2daa3577 | 6007 | update_rq_clock(rq); |
dd41f596 | 6008 | on_rq = p->se.on_rq; |
051a1d1a | 6009 | running = task_current(rq, p); |
0e1f3483 | 6010 | if (on_rq) |
2e1cb74a | 6011 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6012 | if (running) |
6013 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6014 | |
1da177e4 | 6015 | oldprio = p->prio; |
dd41f596 | 6016 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6017 | |
0e1f3483 HS |
6018 | if (running) |
6019 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6020 | if (on_rq) { |
6021 | activate_task(rq, p, 0); | |
cb469845 SR |
6022 | |
6023 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6024 | } |
b29739f9 IM |
6025 | __task_rq_unlock(rq); |
6026 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6027 | ||
95e02ca9 TG |
6028 | rt_mutex_adjust_pi(p); |
6029 | ||
1da177e4 LT |
6030 | return 0; |
6031 | } | |
961ccddd RR |
6032 | |
6033 | /** | |
6034 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6035 | * @p: the task in question. | |
6036 | * @policy: new policy. | |
6037 | * @param: structure containing the new RT priority. | |
6038 | * | |
6039 | * NOTE that the task may be already dead. | |
6040 | */ | |
6041 | int sched_setscheduler(struct task_struct *p, int policy, | |
6042 | struct sched_param *param) | |
6043 | { | |
6044 | return __sched_setscheduler(p, policy, param, true); | |
6045 | } | |
1da177e4 LT |
6046 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6047 | ||
961ccddd RR |
6048 | /** |
6049 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6050 | * @p: the task in question. | |
6051 | * @policy: new policy. | |
6052 | * @param: structure containing the new RT priority. | |
6053 | * | |
6054 | * Just like sched_setscheduler, only don't bother checking if the | |
6055 | * current context has permission. For example, this is needed in | |
6056 | * stop_machine(): we create temporary high priority worker threads, | |
6057 | * but our caller might not have that capability. | |
6058 | */ | |
6059 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6060 | struct sched_param *param) | |
6061 | { | |
6062 | return __sched_setscheduler(p, policy, param, false); | |
6063 | } | |
6064 | ||
95cdf3b7 IM |
6065 | static int |
6066 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6067 | { |
1da177e4 LT |
6068 | struct sched_param lparam; |
6069 | struct task_struct *p; | |
36c8b586 | 6070 | int retval; |
1da177e4 LT |
6071 | |
6072 | if (!param || pid < 0) | |
6073 | return -EINVAL; | |
6074 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6075 | return -EFAULT; | |
5fe1d75f ON |
6076 | |
6077 | rcu_read_lock(); | |
6078 | retval = -ESRCH; | |
1da177e4 | 6079 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6080 | if (p != NULL) |
6081 | retval = sched_setscheduler(p, policy, &lparam); | |
6082 | rcu_read_unlock(); | |
36c8b586 | 6083 | |
1da177e4 LT |
6084 | return retval; |
6085 | } | |
6086 | ||
6087 | /** | |
6088 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6089 | * @pid: the pid in question. | |
6090 | * @policy: new policy. | |
6091 | * @param: structure containing the new RT priority. | |
6092 | */ | |
5add95d4 HC |
6093 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6094 | struct sched_param __user *, param) | |
1da177e4 | 6095 | { |
c21761f1 JB |
6096 | /* negative values for policy are not valid */ |
6097 | if (policy < 0) | |
6098 | return -EINVAL; | |
6099 | ||
1da177e4 LT |
6100 | return do_sched_setscheduler(pid, policy, param); |
6101 | } | |
6102 | ||
6103 | /** | |
6104 | * sys_sched_setparam - set/change the RT priority of a thread | |
6105 | * @pid: the pid in question. | |
6106 | * @param: structure containing the new RT priority. | |
6107 | */ | |
5add95d4 | 6108 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6109 | { |
6110 | return do_sched_setscheduler(pid, -1, param); | |
6111 | } | |
6112 | ||
6113 | /** | |
6114 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6115 | * @pid: the pid in question. | |
6116 | */ | |
5add95d4 | 6117 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6118 | { |
36c8b586 | 6119 | struct task_struct *p; |
3a5c359a | 6120 | int retval; |
1da177e4 LT |
6121 | |
6122 | if (pid < 0) | |
3a5c359a | 6123 | return -EINVAL; |
1da177e4 LT |
6124 | |
6125 | retval = -ESRCH; | |
6126 | read_lock(&tasklist_lock); | |
6127 | p = find_process_by_pid(pid); | |
6128 | if (p) { | |
6129 | retval = security_task_getscheduler(p); | |
6130 | if (!retval) | |
6131 | retval = p->policy; | |
6132 | } | |
6133 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6134 | return retval; |
6135 | } | |
6136 | ||
6137 | /** | |
6138 | * sys_sched_getscheduler - get the RT priority of a thread | |
6139 | * @pid: the pid in question. | |
6140 | * @param: structure containing the RT priority. | |
6141 | */ | |
5add95d4 | 6142 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6143 | { |
6144 | struct sched_param lp; | |
36c8b586 | 6145 | struct task_struct *p; |
3a5c359a | 6146 | int retval; |
1da177e4 LT |
6147 | |
6148 | if (!param || pid < 0) | |
3a5c359a | 6149 | return -EINVAL; |
1da177e4 LT |
6150 | |
6151 | read_lock(&tasklist_lock); | |
6152 | p = find_process_by_pid(pid); | |
6153 | retval = -ESRCH; | |
6154 | if (!p) | |
6155 | goto out_unlock; | |
6156 | ||
6157 | retval = security_task_getscheduler(p); | |
6158 | if (retval) | |
6159 | goto out_unlock; | |
6160 | ||
6161 | lp.sched_priority = p->rt_priority; | |
6162 | read_unlock(&tasklist_lock); | |
6163 | ||
6164 | /* | |
6165 | * This one might sleep, we cannot do it with a spinlock held ... | |
6166 | */ | |
6167 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6168 | ||
1da177e4 LT |
6169 | return retval; |
6170 | ||
6171 | out_unlock: | |
6172 | read_unlock(&tasklist_lock); | |
6173 | return retval; | |
6174 | } | |
6175 | ||
96f874e2 | 6176 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6177 | { |
5a16f3d3 | 6178 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6179 | struct task_struct *p; |
6180 | int retval; | |
1da177e4 | 6181 | |
95402b38 | 6182 | get_online_cpus(); |
1da177e4 LT |
6183 | read_lock(&tasklist_lock); |
6184 | ||
6185 | p = find_process_by_pid(pid); | |
6186 | if (!p) { | |
6187 | read_unlock(&tasklist_lock); | |
95402b38 | 6188 | put_online_cpus(); |
1da177e4 LT |
6189 | return -ESRCH; |
6190 | } | |
6191 | ||
6192 | /* | |
6193 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6194 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6195 | * usage count and then drop tasklist_lock. |
6196 | */ | |
6197 | get_task_struct(p); | |
6198 | read_unlock(&tasklist_lock); | |
6199 | ||
5a16f3d3 RR |
6200 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6201 | retval = -ENOMEM; | |
6202 | goto out_put_task; | |
6203 | } | |
6204 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6205 | retval = -ENOMEM; | |
6206 | goto out_free_cpus_allowed; | |
6207 | } | |
1da177e4 | 6208 | retval = -EPERM; |
c69e8d9c | 6209 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6210 | goto out_unlock; |
6211 | ||
e7834f8f DQ |
6212 | retval = security_task_setscheduler(p, 0, NULL); |
6213 | if (retval) | |
6214 | goto out_unlock; | |
6215 | ||
5a16f3d3 RR |
6216 | cpuset_cpus_allowed(p, cpus_allowed); |
6217 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6218 | again: |
5a16f3d3 | 6219 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6220 | |
8707d8b8 | 6221 | if (!retval) { |
5a16f3d3 RR |
6222 | cpuset_cpus_allowed(p, cpus_allowed); |
6223 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6224 | /* |
6225 | * We must have raced with a concurrent cpuset | |
6226 | * update. Just reset the cpus_allowed to the | |
6227 | * cpuset's cpus_allowed | |
6228 | */ | |
5a16f3d3 | 6229 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6230 | goto again; |
6231 | } | |
6232 | } | |
1da177e4 | 6233 | out_unlock: |
5a16f3d3 RR |
6234 | free_cpumask_var(new_mask); |
6235 | out_free_cpus_allowed: | |
6236 | free_cpumask_var(cpus_allowed); | |
6237 | out_put_task: | |
1da177e4 | 6238 | put_task_struct(p); |
95402b38 | 6239 | put_online_cpus(); |
1da177e4 LT |
6240 | return retval; |
6241 | } | |
6242 | ||
6243 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6244 | struct cpumask *new_mask) |
1da177e4 | 6245 | { |
96f874e2 RR |
6246 | if (len < cpumask_size()) |
6247 | cpumask_clear(new_mask); | |
6248 | else if (len > cpumask_size()) | |
6249 | len = cpumask_size(); | |
6250 | ||
1da177e4 LT |
6251 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6252 | } | |
6253 | ||
6254 | /** | |
6255 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6256 | * @pid: pid of the process | |
6257 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6258 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6259 | */ | |
5add95d4 HC |
6260 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6261 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6262 | { |
5a16f3d3 | 6263 | cpumask_var_t new_mask; |
1da177e4 LT |
6264 | int retval; |
6265 | ||
5a16f3d3 RR |
6266 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6267 | return -ENOMEM; | |
1da177e4 | 6268 | |
5a16f3d3 RR |
6269 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6270 | if (retval == 0) | |
6271 | retval = sched_setaffinity(pid, new_mask); | |
6272 | free_cpumask_var(new_mask); | |
6273 | return retval; | |
1da177e4 LT |
6274 | } |
6275 | ||
96f874e2 | 6276 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6277 | { |
36c8b586 | 6278 | struct task_struct *p; |
1da177e4 | 6279 | int retval; |
1da177e4 | 6280 | |
95402b38 | 6281 | get_online_cpus(); |
1da177e4 LT |
6282 | read_lock(&tasklist_lock); |
6283 | ||
6284 | retval = -ESRCH; | |
6285 | p = find_process_by_pid(pid); | |
6286 | if (!p) | |
6287 | goto out_unlock; | |
6288 | ||
e7834f8f DQ |
6289 | retval = security_task_getscheduler(p); |
6290 | if (retval) | |
6291 | goto out_unlock; | |
6292 | ||
96f874e2 | 6293 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6294 | |
6295 | out_unlock: | |
6296 | read_unlock(&tasklist_lock); | |
95402b38 | 6297 | put_online_cpus(); |
1da177e4 | 6298 | |
9531b62f | 6299 | return retval; |
1da177e4 LT |
6300 | } |
6301 | ||
6302 | /** | |
6303 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6304 | * @pid: pid of the process | |
6305 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6306 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6307 | */ | |
5add95d4 HC |
6308 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6309 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6310 | { |
6311 | int ret; | |
f17c8607 | 6312 | cpumask_var_t mask; |
1da177e4 | 6313 | |
f17c8607 | 6314 | if (len < cpumask_size()) |
1da177e4 LT |
6315 | return -EINVAL; |
6316 | ||
f17c8607 RR |
6317 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6318 | return -ENOMEM; | |
1da177e4 | 6319 | |
f17c8607 RR |
6320 | ret = sched_getaffinity(pid, mask); |
6321 | if (ret == 0) { | |
6322 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6323 | ret = -EFAULT; | |
6324 | else | |
6325 | ret = cpumask_size(); | |
6326 | } | |
6327 | free_cpumask_var(mask); | |
1da177e4 | 6328 | |
f17c8607 | 6329 | return ret; |
1da177e4 LT |
6330 | } |
6331 | ||
6332 | /** | |
6333 | * sys_sched_yield - yield the current processor to other threads. | |
6334 | * | |
dd41f596 IM |
6335 | * This function yields the current CPU to other tasks. If there are no |
6336 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6337 | */ |
5add95d4 | 6338 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6339 | { |
70b97a7f | 6340 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6341 | |
2d72376b | 6342 | schedstat_inc(rq, yld_count); |
4530d7ab | 6343 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6344 | |
6345 | /* | |
6346 | * Since we are going to call schedule() anyway, there's | |
6347 | * no need to preempt or enable interrupts: | |
6348 | */ | |
6349 | __release(rq->lock); | |
8a25d5de | 6350 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6351 | _raw_spin_unlock(&rq->lock); |
6352 | preempt_enable_no_resched(); | |
6353 | ||
6354 | schedule(); | |
6355 | ||
6356 | return 0; | |
6357 | } | |
6358 | ||
e7b38404 | 6359 | static void __cond_resched(void) |
1da177e4 | 6360 | { |
8e0a43d8 IM |
6361 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6362 | __might_sleep(__FILE__, __LINE__); | |
6363 | #endif | |
5bbcfd90 IM |
6364 | /* |
6365 | * The BKS might be reacquired before we have dropped | |
6366 | * PREEMPT_ACTIVE, which could trigger a second | |
6367 | * cond_resched() call. | |
6368 | */ | |
1da177e4 LT |
6369 | do { |
6370 | add_preempt_count(PREEMPT_ACTIVE); | |
6371 | schedule(); | |
6372 | sub_preempt_count(PREEMPT_ACTIVE); | |
6373 | } while (need_resched()); | |
6374 | } | |
6375 | ||
02b67cc3 | 6376 | int __sched _cond_resched(void) |
1da177e4 | 6377 | { |
9414232f IM |
6378 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6379 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6380 | __cond_resched(); |
6381 | return 1; | |
6382 | } | |
6383 | return 0; | |
6384 | } | |
02b67cc3 | 6385 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6386 | |
6387 | /* | |
6388 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6389 | * call schedule, and on return reacquire the lock. | |
6390 | * | |
41a2d6cf | 6391 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6392 | * operations here to prevent schedule() from being called twice (once via |
6393 | * spin_unlock(), once by hand). | |
6394 | */ | |
95cdf3b7 | 6395 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6396 | { |
95c354fe | 6397 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6398 | int ret = 0; |
6399 | ||
95c354fe | 6400 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6401 | spin_unlock(lock); |
95c354fe NP |
6402 | if (resched && need_resched()) |
6403 | __cond_resched(); | |
6404 | else | |
6405 | cpu_relax(); | |
6df3cecb | 6406 | ret = 1; |
1da177e4 | 6407 | spin_lock(lock); |
1da177e4 | 6408 | } |
6df3cecb | 6409 | return ret; |
1da177e4 | 6410 | } |
1da177e4 LT |
6411 | EXPORT_SYMBOL(cond_resched_lock); |
6412 | ||
6413 | int __sched cond_resched_softirq(void) | |
6414 | { | |
6415 | BUG_ON(!in_softirq()); | |
6416 | ||
9414232f | 6417 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6418 | local_bh_enable(); |
1da177e4 LT |
6419 | __cond_resched(); |
6420 | local_bh_disable(); | |
6421 | return 1; | |
6422 | } | |
6423 | return 0; | |
6424 | } | |
1da177e4 LT |
6425 | EXPORT_SYMBOL(cond_resched_softirq); |
6426 | ||
1da177e4 LT |
6427 | /** |
6428 | * yield - yield the current processor to other threads. | |
6429 | * | |
72fd4a35 | 6430 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6431 | * thread runnable and calls sys_sched_yield(). |
6432 | */ | |
6433 | void __sched yield(void) | |
6434 | { | |
6435 | set_current_state(TASK_RUNNING); | |
6436 | sys_sched_yield(); | |
6437 | } | |
1da177e4 LT |
6438 | EXPORT_SYMBOL(yield); |
6439 | ||
6440 | /* | |
41a2d6cf | 6441 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6442 | * that process accounting knows that this is a task in IO wait state. |
6443 | * | |
6444 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6445 | * has set its backing_dev_info: the queue against which it should throttle) | |
6446 | */ | |
6447 | void __sched io_schedule(void) | |
6448 | { | |
70b97a7f | 6449 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6450 | |
0ff92245 | 6451 | delayacct_blkio_start(); |
1da177e4 LT |
6452 | atomic_inc(&rq->nr_iowait); |
6453 | schedule(); | |
6454 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6455 | delayacct_blkio_end(); |
1da177e4 | 6456 | } |
1da177e4 LT |
6457 | EXPORT_SYMBOL(io_schedule); |
6458 | ||
6459 | long __sched io_schedule_timeout(long timeout) | |
6460 | { | |
70b97a7f | 6461 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6462 | long ret; |
6463 | ||
0ff92245 | 6464 | delayacct_blkio_start(); |
1da177e4 LT |
6465 | atomic_inc(&rq->nr_iowait); |
6466 | ret = schedule_timeout(timeout); | |
6467 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6468 | delayacct_blkio_end(); |
1da177e4 LT |
6469 | return ret; |
6470 | } | |
6471 | ||
6472 | /** | |
6473 | * sys_sched_get_priority_max - return maximum RT priority. | |
6474 | * @policy: scheduling class. | |
6475 | * | |
6476 | * this syscall returns the maximum rt_priority that can be used | |
6477 | * by a given scheduling class. | |
6478 | */ | |
5add95d4 | 6479 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6480 | { |
6481 | int ret = -EINVAL; | |
6482 | ||
6483 | switch (policy) { | |
6484 | case SCHED_FIFO: | |
6485 | case SCHED_RR: | |
6486 | ret = MAX_USER_RT_PRIO-1; | |
6487 | break; | |
6488 | case SCHED_NORMAL: | |
b0a9499c | 6489 | case SCHED_BATCH: |
dd41f596 | 6490 | case SCHED_IDLE: |
1da177e4 LT |
6491 | ret = 0; |
6492 | break; | |
6493 | } | |
6494 | return ret; | |
6495 | } | |
6496 | ||
6497 | /** | |
6498 | * sys_sched_get_priority_min - return minimum RT priority. | |
6499 | * @policy: scheduling class. | |
6500 | * | |
6501 | * this syscall returns the minimum rt_priority that can be used | |
6502 | * by a given scheduling class. | |
6503 | */ | |
5add95d4 | 6504 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6505 | { |
6506 | int ret = -EINVAL; | |
6507 | ||
6508 | switch (policy) { | |
6509 | case SCHED_FIFO: | |
6510 | case SCHED_RR: | |
6511 | ret = 1; | |
6512 | break; | |
6513 | case SCHED_NORMAL: | |
b0a9499c | 6514 | case SCHED_BATCH: |
dd41f596 | 6515 | case SCHED_IDLE: |
1da177e4 LT |
6516 | ret = 0; |
6517 | } | |
6518 | return ret; | |
6519 | } | |
6520 | ||
6521 | /** | |
6522 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6523 | * @pid: pid of the process. | |
6524 | * @interval: userspace pointer to the timeslice value. | |
6525 | * | |
6526 | * this syscall writes the default timeslice value of a given process | |
6527 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6528 | */ | |
17da2bd9 | 6529 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6530 | struct timespec __user *, interval) |
1da177e4 | 6531 | { |
36c8b586 | 6532 | struct task_struct *p; |
a4ec24b4 | 6533 | unsigned int time_slice; |
3a5c359a | 6534 | int retval; |
1da177e4 | 6535 | struct timespec t; |
1da177e4 LT |
6536 | |
6537 | if (pid < 0) | |
3a5c359a | 6538 | return -EINVAL; |
1da177e4 LT |
6539 | |
6540 | retval = -ESRCH; | |
6541 | read_lock(&tasklist_lock); | |
6542 | p = find_process_by_pid(pid); | |
6543 | if (!p) | |
6544 | goto out_unlock; | |
6545 | ||
6546 | retval = security_task_getscheduler(p); | |
6547 | if (retval) | |
6548 | goto out_unlock; | |
6549 | ||
77034937 IM |
6550 | /* |
6551 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6552 | * tasks that are on an otherwise idle runqueue: | |
6553 | */ | |
6554 | time_slice = 0; | |
6555 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6556 | time_slice = DEF_TIMESLICE; |
1868f958 | 6557 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6558 | struct sched_entity *se = &p->se; |
6559 | unsigned long flags; | |
6560 | struct rq *rq; | |
6561 | ||
6562 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6563 | if (rq->cfs.load.weight) |
6564 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6565 | task_rq_unlock(rq, &flags); |
6566 | } | |
1da177e4 | 6567 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6568 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6569 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6570 | return retval; |
3a5c359a | 6571 | |
1da177e4 LT |
6572 | out_unlock: |
6573 | read_unlock(&tasklist_lock); | |
6574 | return retval; | |
6575 | } | |
6576 | ||
7c731e0a | 6577 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6578 | |
82a1fcb9 | 6579 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6580 | { |
1da177e4 | 6581 | unsigned long free = 0; |
36c8b586 | 6582 | unsigned state; |
1da177e4 | 6583 | |
1da177e4 | 6584 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6585 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6586 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6587 | #if BITS_PER_LONG == 32 |
1da177e4 | 6588 | if (state == TASK_RUNNING) |
cc4ea795 | 6589 | printk(KERN_CONT " running "); |
1da177e4 | 6590 | else |
cc4ea795 | 6591 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6592 | #else |
6593 | if (state == TASK_RUNNING) | |
cc4ea795 | 6594 | printk(KERN_CONT " running task "); |
1da177e4 | 6595 | else |
cc4ea795 | 6596 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6597 | #endif |
6598 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6599 | free = stack_not_used(p); |
1da177e4 | 6600 | #endif |
ba25f9dc | 6601 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 6602 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 6603 | |
5fb5e6de | 6604 | show_stack(p, NULL); |
1da177e4 LT |
6605 | } |
6606 | ||
e59e2ae2 | 6607 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6608 | { |
36c8b586 | 6609 | struct task_struct *g, *p; |
1da177e4 | 6610 | |
4bd77321 IM |
6611 | #if BITS_PER_LONG == 32 |
6612 | printk(KERN_INFO | |
6613 | " task PC stack pid father\n"); | |
1da177e4 | 6614 | #else |
4bd77321 IM |
6615 | printk(KERN_INFO |
6616 | " task PC stack pid father\n"); | |
1da177e4 LT |
6617 | #endif |
6618 | read_lock(&tasklist_lock); | |
6619 | do_each_thread(g, p) { | |
6620 | /* | |
6621 | * reset the NMI-timeout, listing all files on a slow | |
6622 | * console might take alot of time: | |
6623 | */ | |
6624 | touch_nmi_watchdog(); | |
39bc89fd | 6625 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6626 | sched_show_task(p); |
1da177e4 LT |
6627 | } while_each_thread(g, p); |
6628 | ||
04c9167f JF |
6629 | touch_all_softlockup_watchdogs(); |
6630 | ||
dd41f596 IM |
6631 | #ifdef CONFIG_SCHED_DEBUG |
6632 | sysrq_sched_debug_show(); | |
6633 | #endif | |
1da177e4 | 6634 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6635 | /* |
6636 | * Only show locks if all tasks are dumped: | |
6637 | */ | |
6638 | if (state_filter == -1) | |
6639 | debug_show_all_locks(); | |
1da177e4 LT |
6640 | } |
6641 | ||
1df21055 IM |
6642 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6643 | { | |
dd41f596 | 6644 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6645 | } |
6646 | ||
f340c0d1 IM |
6647 | /** |
6648 | * init_idle - set up an idle thread for a given CPU | |
6649 | * @idle: task in question | |
6650 | * @cpu: cpu the idle task belongs to | |
6651 | * | |
6652 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6653 | * flag, to make booting more robust. | |
6654 | */ | |
5c1e1767 | 6655 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6656 | { |
70b97a7f | 6657 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6658 | unsigned long flags; |
6659 | ||
5cbd54ef IM |
6660 | spin_lock_irqsave(&rq->lock, flags); |
6661 | ||
dd41f596 IM |
6662 | __sched_fork(idle); |
6663 | idle->se.exec_start = sched_clock(); | |
6664 | ||
b29739f9 | 6665 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6666 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6667 | __set_task_cpu(idle, cpu); |
1da177e4 | 6668 | |
1da177e4 | 6669 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6670 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6671 | idle->oncpu = 1; | |
6672 | #endif | |
1da177e4 LT |
6673 | spin_unlock_irqrestore(&rq->lock, flags); |
6674 | ||
6675 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6676 | #if defined(CONFIG_PREEMPT) |
6677 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6678 | #else | |
a1261f54 | 6679 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6680 | #endif |
dd41f596 IM |
6681 | /* |
6682 | * The idle tasks have their own, simple scheduling class: | |
6683 | */ | |
6684 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6685 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6686 | } |
6687 | ||
6688 | /* | |
6689 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6690 | * indicates which cpus entered this state. This is used | |
6691 | * in the rcu update to wait only for active cpus. For system | |
6692 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6693 | * always be CPU_BITS_NONE. |
1da177e4 | 6694 | */ |
6a7b3dc3 | 6695 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6696 | |
19978ca6 IM |
6697 | /* |
6698 | * Increase the granularity value when there are more CPUs, | |
6699 | * because with more CPUs the 'effective latency' as visible | |
6700 | * to users decreases. But the relationship is not linear, | |
6701 | * so pick a second-best guess by going with the log2 of the | |
6702 | * number of CPUs. | |
6703 | * | |
6704 | * This idea comes from the SD scheduler of Con Kolivas: | |
6705 | */ | |
6706 | static inline void sched_init_granularity(void) | |
6707 | { | |
6708 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6709 | const unsigned long limit = 200000000; | |
6710 | ||
6711 | sysctl_sched_min_granularity *= factor; | |
6712 | if (sysctl_sched_min_granularity > limit) | |
6713 | sysctl_sched_min_granularity = limit; | |
6714 | ||
6715 | sysctl_sched_latency *= factor; | |
6716 | if (sysctl_sched_latency > limit) | |
6717 | sysctl_sched_latency = limit; | |
6718 | ||
6719 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6720 | |
6721 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6722 | } |
6723 | ||
1da177e4 LT |
6724 | #ifdef CONFIG_SMP |
6725 | /* | |
6726 | * This is how migration works: | |
6727 | * | |
70b97a7f | 6728 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6729 | * runqueue and wake up that CPU's migration thread. |
6730 | * 2) we down() the locked semaphore => thread blocks. | |
6731 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6732 | * thread off the CPU) | |
6733 | * 4) it gets the migration request and checks whether the migrated | |
6734 | * task is still in the wrong runqueue. | |
6735 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6736 | * it and puts it into the right queue. | |
6737 | * 6) migration thread up()s the semaphore. | |
6738 | * 7) we wake up and the migration is done. | |
6739 | */ | |
6740 | ||
6741 | /* | |
6742 | * Change a given task's CPU affinity. Migrate the thread to a | |
6743 | * proper CPU and schedule it away if the CPU it's executing on | |
6744 | * is removed from the allowed bitmask. | |
6745 | * | |
6746 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6747 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6748 | * call is not atomic; no spinlocks may be held. |
6749 | */ | |
96f874e2 | 6750 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6751 | { |
70b97a7f | 6752 | struct migration_req req; |
1da177e4 | 6753 | unsigned long flags; |
70b97a7f | 6754 | struct rq *rq; |
48f24c4d | 6755 | int ret = 0; |
1da177e4 LT |
6756 | |
6757 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6758 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6759 | ret = -EINVAL; |
6760 | goto out; | |
6761 | } | |
6762 | ||
9985b0ba | 6763 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6764 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6765 | ret = -EINVAL; |
6766 | goto out; | |
6767 | } | |
6768 | ||
73fe6aae | 6769 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6770 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6771 | else { |
96f874e2 RR |
6772 | cpumask_copy(&p->cpus_allowed, new_mask); |
6773 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6774 | } |
6775 | ||
1da177e4 | 6776 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6777 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6778 | goto out; |
6779 | ||
1e5ce4f4 | 6780 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6781 | /* Need help from migration thread: drop lock and wait. */ |
6782 | task_rq_unlock(rq, &flags); | |
6783 | wake_up_process(rq->migration_thread); | |
6784 | wait_for_completion(&req.done); | |
6785 | tlb_migrate_finish(p->mm); | |
6786 | return 0; | |
6787 | } | |
6788 | out: | |
6789 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6790 | |
1da177e4 LT |
6791 | return ret; |
6792 | } | |
cd8ba7cd | 6793 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6794 | |
6795 | /* | |
41a2d6cf | 6796 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6797 | * this because either it can't run here any more (set_cpus_allowed() |
6798 | * away from this CPU, or CPU going down), or because we're | |
6799 | * attempting to rebalance this task on exec (sched_exec). | |
6800 | * | |
6801 | * So we race with normal scheduler movements, but that's OK, as long | |
6802 | * as the task is no longer on this CPU. | |
efc30814 KK |
6803 | * |
6804 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6805 | */ |
efc30814 | 6806 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6807 | { |
70b97a7f | 6808 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6809 | int ret = 0, on_rq; |
1da177e4 | 6810 | |
e761b772 | 6811 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6812 | return ret; |
1da177e4 LT |
6813 | |
6814 | rq_src = cpu_rq(src_cpu); | |
6815 | rq_dest = cpu_rq(dest_cpu); | |
6816 | ||
6817 | double_rq_lock(rq_src, rq_dest); | |
6818 | /* Already moved. */ | |
6819 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6820 | goto done; |
1da177e4 | 6821 | /* Affinity changed (again). */ |
96f874e2 | 6822 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6823 | goto fail; |
1da177e4 | 6824 | |
dd41f596 | 6825 | on_rq = p->se.on_rq; |
6e82a3be | 6826 | if (on_rq) |
2e1cb74a | 6827 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6828 | |
1da177e4 | 6829 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6830 | if (on_rq) { |
6831 | activate_task(rq_dest, p, 0); | |
15afe09b | 6832 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6833 | } |
b1e38734 | 6834 | done: |
efc30814 | 6835 | ret = 1; |
b1e38734 | 6836 | fail: |
1da177e4 | 6837 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6838 | return ret; |
1da177e4 LT |
6839 | } |
6840 | ||
6841 | /* | |
6842 | * migration_thread - this is a highprio system thread that performs | |
6843 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6844 | * another runqueue. | |
6845 | */ | |
95cdf3b7 | 6846 | static int migration_thread(void *data) |
1da177e4 | 6847 | { |
1da177e4 | 6848 | int cpu = (long)data; |
70b97a7f | 6849 | struct rq *rq; |
1da177e4 LT |
6850 | |
6851 | rq = cpu_rq(cpu); | |
6852 | BUG_ON(rq->migration_thread != current); | |
6853 | ||
6854 | set_current_state(TASK_INTERRUPTIBLE); | |
6855 | while (!kthread_should_stop()) { | |
70b97a7f | 6856 | struct migration_req *req; |
1da177e4 | 6857 | struct list_head *head; |
1da177e4 | 6858 | |
1da177e4 LT |
6859 | spin_lock_irq(&rq->lock); |
6860 | ||
6861 | if (cpu_is_offline(cpu)) { | |
6862 | spin_unlock_irq(&rq->lock); | |
6863 | goto wait_to_die; | |
6864 | } | |
6865 | ||
6866 | if (rq->active_balance) { | |
6867 | active_load_balance(rq, cpu); | |
6868 | rq->active_balance = 0; | |
6869 | } | |
6870 | ||
6871 | head = &rq->migration_queue; | |
6872 | ||
6873 | if (list_empty(head)) { | |
6874 | spin_unlock_irq(&rq->lock); | |
6875 | schedule(); | |
6876 | set_current_state(TASK_INTERRUPTIBLE); | |
6877 | continue; | |
6878 | } | |
70b97a7f | 6879 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6880 | list_del_init(head->next); |
6881 | ||
674311d5 NP |
6882 | spin_unlock(&rq->lock); |
6883 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6884 | local_irq_enable(); | |
1da177e4 LT |
6885 | |
6886 | complete(&req->done); | |
6887 | } | |
6888 | __set_current_state(TASK_RUNNING); | |
6889 | return 0; | |
6890 | ||
6891 | wait_to_die: | |
6892 | /* Wait for kthread_stop */ | |
6893 | set_current_state(TASK_INTERRUPTIBLE); | |
6894 | while (!kthread_should_stop()) { | |
6895 | schedule(); | |
6896 | set_current_state(TASK_INTERRUPTIBLE); | |
6897 | } | |
6898 | __set_current_state(TASK_RUNNING); | |
6899 | return 0; | |
6900 | } | |
6901 | ||
6902 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6903 | |
6904 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6905 | { | |
6906 | int ret; | |
6907 | ||
6908 | local_irq_disable(); | |
6909 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6910 | local_irq_enable(); | |
6911 | return ret; | |
6912 | } | |
6913 | ||
054b9108 | 6914 | /* |
3a4fa0a2 | 6915 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6916 | */ |
48f24c4d | 6917 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6918 | { |
70b97a7f | 6919 | int dest_cpu; |
6ca09dfc | 6920 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
6921 | |
6922 | again: | |
6923 | /* Look for allowed, online CPU in same node. */ | |
6924 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6925 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6926 | goto move; | |
6927 | ||
6928 | /* Any allowed, online CPU? */ | |
6929 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6930 | if (dest_cpu < nr_cpu_ids) | |
6931 | goto move; | |
6932 | ||
6933 | /* No more Mr. Nice Guy. */ | |
6934 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6935 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6936 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6937 | |
e76bd8d9 RR |
6938 | /* |
6939 | * Don't tell them about moving exiting tasks or | |
6940 | * kernel threads (both mm NULL), since they never | |
6941 | * leave kernel. | |
6942 | */ | |
6943 | if (p->mm && printk_ratelimit()) { | |
6944 | printk(KERN_INFO "process %d (%s) no " | |
6945 | "longer affine to cpu%d\n", | |
6946 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6947 | } |
e76bd8d9 RR |
6948 | } |
6949 | ||
6950 | move: | |
6951 | /* It can have affinity changed while we were choosing. */ | |
6952 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6953 | goto again; | |
1da177e4 LT |
6954 | } |
6955 | ||
6956 | /* | |
6957 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6958 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6959 | * for performance reasons the counter is not stricly tracking tasks to | |
6960 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6961 | * to keep the global sum constant after CPU-down: | |
6962 | */ | |
70b97a7f | 6963 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6964 | { |
1e5ce4f4 | 6965 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6966 | unsigned long flags; |
6967 | ||
6968 | local_irq_save(flags); | |
6969 | double_rq_lock(rq_src, rq_dest); | |
6970 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6971 | rq_src->nr_uninterruptible = 0; | |
6972 | double_rq_unlock(rq_src, rq_dest); | |
6973 | local_irq_restore(flags); | |
6974 | } | |
6975 | ||
6976 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6977 | static void migrate_live_tasks(int src_cpu) | |
6978 | { | |
48f24c4d | 6979 | struct task_struct *p, *t; |
1da177e4 | 6980 | |
f7b4cddc | 6981 | read_lock(&tasklist_lock); |
1da177e4 | 6982 | |
48f24c4d IM |
6983 | do_each_thread(t, p) { |
6984 | if (p == current) | |
1da177e4 LT |
6985 | continue; |
6986 | ||
48f24c4d IM |
6987 | if (task_cpu(p) == src_cpu) |
6988 | move_task_off_dead_cpu(src_cpu, p); | |
6989 | } while_each_thread(t, p); | |
1da177e4 | 6990 | |
f7b4cddc | 6991 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6992 | } |
6993 | ||
dd41f596 IM |
6994 | /* |
6995 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6996 | * It does so by boosting its priority to highest possible. |
6997 | * Used by CPU offline code. | |
1da177e4 LT |
6998 | */ |
6999 | void sched_idle_next(void) | |
7000 | { | |
48f24c4d | 7001 | int this_cpu = smp_processor_id(); |
70b97a7f | 7002 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7003 | struct task_struct *p = rq->idle; |
7004 | unsigned long flags; | |
7005 | ||
7006 | /* cpu has to be offline */ | |
48f24c4d | 7007 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7008 | |
48f24c4d IM |
7009 | /* |
7010 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7011 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7012 | */ |
7013 | spin_lock_irqsave(&rq->lock, flags); | |
7014 | ||
dd41f596 | 7015 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7016 | |
94bc9a7b DA |
7017 | update_rq_clock(rq); |
7018 | activate_task(rq, p, 0); | |
1da177e4 LT |
7019 | |
7020 | spin_unlock_irqrestore(&rq->lock, flags); | |
7021 | } | |
7022 | ||
48f24c4d IM |
7023 | /* |
7024 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7025 | * offline. |
7026 | */ | |
7027 | void idle_task_exit(void) | |
7028 | { | |
7029 | struct mm_struct *mm = current->active_mm; | |
7030 | ||
7031 | BUG_ON(cpu_online(smp_processor_id())); | |
7032 | ||
7033 | if (mm != &init_mm) | |
7034 | switch_mm(mm, &init_mm, current); | |
7035 | mmdrop(mm); | |
7036 | } | |
7037 | ||
054b9108 | 7038 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7039 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7040 | { |
70b97a7f | 7041 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7042 | |
7043 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7044 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7045 | |
7046 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7047 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7048 | |
48f24c4d | 7049 | get_task_struct(p); |
1da177e4 LT |
7050 | |
7051 | /* | |
7052 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7053 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7054 | * fine. |
7055 | */ | |
f7b4cddc | 7056 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7057 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7058 | spin_lock_irq(&rq->lock); |
1da177e4 | 7059 | |
48f24c4d | 7060 | put_task_struct(p); |
1da177e4 LT |
7061 | } |
7062 | ||
7063 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7064 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7065 | { | |
70b97a7f | 7066 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7067 | struct task_struct *next; |
48f24c4d | 7068 | |
dd41f596 IM |
7069 | for ( ; ; ) { |
7070 | if (!rq->nr_running) | |
7071 | break; | |
a8e504d2 | 7072 | update_rq_clock(rq); |
b67802ea | 7073 | next = pick_next_task(rq); |
dd41f596 IM |
7074 | if (!next) |
7075 | break; | |
79c53799 | 7076 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7077 | migrate_dead(dead_cpu, next); |
e692ab53 | 7078 | |
1da177e4 LT |
7079 | } |
7080 | } | |
7081 | #endif /* CONFIG_HOTPLUG_CPU */ | |
7082 | ||
e692ab53 NP |
7083 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7084 | ||
7085 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7086 | { |
7087 | .procname = "sched_domain", | |
c57baf1e | 7088 | .mode = 0555, |
e0361851 | 7089 | }, |
38605cae | 7090 | {0, }, |
e692ab53 NP |
7091 | }; |
7092 | ||
7093 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7094 | { |
c57baf1e | 7095 | .ctl_name = CTL_KERN, |
e0361851 | 7096 | .procname = "kernel", |
c57baf1e | 7097 | .mode = 0555, |
e0361851 AD |
7098 | .child = sd_ctl_dir, |
7099 | }, | |
38605cae | 7100 | {0, }, |
e692ab53 NP |
7101 | }; |
7102 | ||
7103 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7104 | { | |
7105 | struct ctl_table *entry = | |
5cf9f062 | 7106 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7107 | |
e692ab53 NP |
7108 | return entry; |
7109 | } | |
7110 | ||
6382bc90 MM |
7111 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7112 | { | |
cd790076 | 7113 | struct ctl_table *entry; |
6382bc90 | 7114 | |
cd790076 MM |
7115 | /* |
7116 | * In the intermediate directories, both the child directory and | |
7117 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7118 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7119 | * static strings and all have proc handlers. |
7120 | */ | |
7121 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7122 | if (entry->child) |
7123 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7124 | if (entry->proc_handler == NULL) |
7125 | kfree(entry->procname); | |
7126 | } | |
6382bc90 MM |
7127 | |
7128 | kfree(*tablep); | |
7129 | *tablep = NULL; | |
7130 | } | |
7131 | ||
e692ab53 | 7132 | static void |
e0361851 | 7133 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7134 | const char *procname, void *data, int maxlen, |
7135 | mode_t mode, proc_handler *proc_handler) | |
7136 | { | |
e692ab53 NP |
7137 | entry->procname = procname; |
7138 | entry->data = data; | |
7139 | entry->maxlen = maxlen; | |
7140 | entry->mode = mode; | |
7141 | entry->proc_handler = proc_handler; | |
7142 | } | |
7143 | ||
7144 | static struct ctl_table * | |
7145 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7146 | { | |
a5d8c348 | 7147 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7148 | |
ad1cdc1d MM |
7149 | if (table == NULL) |
7150 | return NULL; | |
7151 | ||
e0361851 | 7152 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7153 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7154 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7155 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7156 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7157 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7158 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7159 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7160 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7161 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7162 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7163 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7164 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7165 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7166 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7167 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7168 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7169 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7170 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7171 | &sd->cache_nice_tries, |
7172 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7173 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7174 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7175 | set_table_entry(&table[11], "name", sd->name, |
7176 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7177 | /* &table[12] is terminator */ | |
e692ab53 NP |
7178 | |
7179 | return table; | |
7180 | } | |
7181 | ||
9a4e7159 | 7182 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7183 | { |
7184 | struct ctl_table *entry, *table; | |
7185 | struct sched_domain *sd; | |
7186 | int domain_num = 0, i; | |
7187 | char buf[32]; | |
7188 | ||
7189 | for_each_domain(cpu, sd) | |
7190 | domain_num++; | |
7191 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7192 | if (table == NULL) |
7193 | return NULL; | |
e692ab53 NP |
7194 | |
7195 | i = 0; | |
7196 | for_each_domain(cpu, sd) { | |
7197 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7198 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7199 | entry->mode = 0555; |
e692ab53 NP |
7200 | entry->child = sd_alloc_ctl_domain_table(sd); |
7201 | entry++; | |
7202 | i++; | |
7203 | } | |
7204 | return table; | |
7205 | } | |
7206 | ||
7207 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7208 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7209 | { |
7210 | int i, cpu_num = num_online_cpus(); | |
7211 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7212 | char buf[32]; | |
7213 | ||
7378547f MM |
7214 | WARN_ON(sd_ctl_dir[0].child); |
7215 | sd_ctl_dir[0].child = entry; | |
7216 | ||
ad1cdc1d MM |
7217 | if (entry == NULL) |
7218 | return; | |
7219 | ||
97b6ea7b | 7220 | for_each_online_cpu(i) { |
e692ab53 | 7221 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7222 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7223 | entry->mode = 0555; |
e692ab53 | 7224 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7225 | entry++; |
e692ab53 | 7226 | } |
7378547f MM |
7227 | |
7228 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7229 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7230 | } | |
6382bc90 | 7231 | |
7378547f | 7232 | /* may be called multiple times per register */ |
6382bc90 MM |
7233 | static void unregister_sched_domain_sysctl(void) |
7234 | { | |
7378547f MM |
7235 | if (sd_sysctl_header) |
7236 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7237 | sd_sysctl_header = NULL; |
7378547f MM |
7238 | if (sd_ctl_dir[0].child) |
7239 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7240 | } |
e692ab53 | 7241 | #else |
6382bc90 MM |
7242 | static void register_sched_domain_sysctl(void) |
7243 | { | |
7244 | } | |
7245 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7246 | { |
7247 | } | |
7248 | #endif | |
7249 | ||
1f11eb6a GH |
7250 | static void set_rq_online(struct rq *rq) |
7251 | { | |
7252 | if (!rq->online) { | |
7253 | const struct sched_class *class; | |
7254 | ||
c6c4927b | 7255 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7256 | rq->online = 1; |
7257 | ||
7258 | for_each_class(class) { | |
7259 | if (class->rq_online) | |
7260 | class->rq_online(rq); | |
7261 | } | |
7262 | } | |
7263 | } | |
7264 | ||
7265 | static void set_rq_offline(struct rq *rq) | |
7266 | { | |
7267 | if (rq->online) { | |
7268 | const struct sched_class *class; | |
7269 | ||
7270 | for_each_class(class) { | |
7271 | if (class->rq_offline) | |
7272 | class->rq_offline(rq); | |
7273 | } | |
7274 | ||
c6c4927b | 7275 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7276 | rq->online = 0; |
7277 | } | |
7278 | } | |
7279 | ||
1da177e4 LT |
7280 | /* |
7281 | * migration_call - callback that gets triggered when a CPU is added. | |
7282 | * Here we can start up the necessary migration thread for the new CPU. | |
7283 | */ | |
48f24c4d IM |
7284 | static int __cpuinit |
7285 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7286 | { |
1da177e4 | 7287 | struct task_struct *p; |
48f24c4d | 7288 | int cpu = (long)hcpu; |
1da177e4 | 7289 | unsigned long flags; |
70b97a7f | 7290 | struct rq *rq; |
1da177e4 LT |
7291 | |
7292 | switch (action) { | |
5be9361c | 7293 | |
1da177e4 | 7294 | case CPU_UP_PREPARE: |
8bb78442 | 7295 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7296 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7297 | if (IS_ERR(p)) |
7298 | return NOTIFY_BAD; | |
1da177e4 LT |
7299 | kthread_bind(p, cpu); |
7300 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7301 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7302 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7303 | task_rq_unlock(rq, &flags); |
7304 | cpu_rq(cpu)->migration_thread = p; | |
7305 | break; | |
48f24c4d | 7306 | |
1da177e4 | 7307 | case CPU_ONLINE: |
8bb78442 | 7308 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7309 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7310 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7311 | |
7312 | /* Update our root-domain */ | |
7313 | rq = cpu_rq(cpu); | |
7314 | spin_lock_irqsave(&rq->lock, flags); | |
7315 | if (rq->rd) { | |
c6c4927b | 7316 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7317 | |
7318 | set_rq_online(rq); | |
1f94ef59 GH |
7319 | } |
7320 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7321 | break; |
48f24c4d | 7322 | |
1da177e4 LT |
7323 | #ifdef CONFIG_HOTPLUG_CPU |
7324 | case CPU_UP_CANCELED: | |
8bb78442 | 7325 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7326 | if (!cpu_rq(cpu)->migration_thread) |
7327 | break; | |
41a2d6cf | 7328 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7329 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7330 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7331 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7332 | cpu_rq(cpu)->migration_thread = NULL; | |
7333 | break; | |
48f24c4d | 7334 | |
1da177e4 | 7335 | case CPU_DEAD: |
8bb78442 | 7336 | case CPU_DEAD_FROZEN: |
470fd646 | 7337 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7338 | migrate_live_tasks(cpu); |
7339 | rq = cpu_rq(cpu); | |
7340 | kthread_stop(rq->migration_thread); | |
7341 | rq->migration_thread = NULL; | |
7342 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7343 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7344 | update_rq_clock(rq); |
2e1cb74a | 7345 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7346 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7347 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7348 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7349 | migrate_dead_tasks(cpu); |
d2da272a | 7350 | spin_unlock_irq(&rq->lock); |
470fd646 | 7351 | cpuset_unlock(); |
1da177e4 LT |
7352 | migrate_nr_uninterruptible(rq); |
7353 | BUG_ON(rq->nr_running != 0); | |
7354 | ||
41a2d6cf IM |
7355 | /* |
7356 | * No need to migrate the tasks: it was best-effort if | |
7357 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7358 | * the requestors. | |
7359 | */ | |
1da177e4 LT |
7360 | spin_lock_irq(&rq->lock); |
7361 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7362 | struct migration_req *req; |
7363 | ||
1da177e4 | 7364 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7365 | struct migration_req, list); |
1da177e4 | 7366 | list_del_init(&req->list); |
9a2bd244 | 7367 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7368 | complete(&req->done); |
9a2bd244 | 7369 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7370 | } |
7371 | spin_unlock_irq(&rq->lock); | |
7372 | break; | |
57d885fe | 7373 | |
08f503b0 GH |
7374 | case CPU_DYING: |
7375 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7376 | /* Update our root-domain */ |
7377 | rq = cpu_rq(cpu); | |
7378 | spin_lock_irqsave(&rq->lock, flags); | |
7379 | if (rq->rd) { | |
c6c4927b | 7380 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7381 | set_rq_offline(rq); |
57d885fe GH |
7382 | } |
7383 | spin_unlock_irqrestore(&rq->lock, flags); | |
7384 | break; | |
1da177e4 LT |
7385 | #endif |
7386 | } | |
7387 | return NOTIFY_OK; | |
7388 | } | |
7389 | ||
7390 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
7391 | * happens before everything else. | |
7392 | */ | |
26c2143b | 7393 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7394 | .notifier_call = migration_call, |
7395 | .priority = 10 | |
7396 | }; | |
7397 | ||
7babe8db | 7398 | static int __init migration_init(void) |
1da177e4 LT |
7399 | { |
7400 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7401 | int err; |
48f24c4d IM |
7402 | |
7403 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7404 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7405 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7406 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7407 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7408 | |
7409 | return err; | |
1da177e4 | 7410 | } |
7babe8db | 7411 | early_initcall(migration_init); |
1da177e4 LT |
7412 | #endif |
7413 | ||
7414 | #ifdef CONFIG_SMP | |
476f3534 | 7415 | |
3e9830dc | 7416 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7417 | |
7c16ec58 | 7418 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7419 | struct cpumask *groupmask) |
1da177e4 | 7420 | { |
4dcf6aff | 7421 | struct sched_group *group = sd->groups; |
434d53b0 | 7422 | char str[256]; |
1da177e4 | 7423 | |
968ea6d8 | 7424 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7425 | cpumask_clear(groupmask); |
4dcf6aff IM |
7426 | |
7427 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7428 | ||
7429 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7430 | printk("does not load-balance\n"); | |
7431 | if (sd->parent) | |
7432 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7433 | " has parent"); | |
7434 | return -1; | |
41c7ce9a NP |
7435 | } |
7436 | ||
eefd796a | 7437 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7438 | |
758b2cdc | 7439 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7440 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7441 | "CPU%d\n", cpu); | |
7442 | } | |
758b2cdc | 7443 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7444 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7445 | " CPU%d\n", cpu); | |
7446 | } | |
1da177e4 | 7447 | |
4dcf6aff | 7448 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7449 | do { |
4dcf6aff IM |
7450 | if (!group) { |
7451 | printk("\n"); | |
7452 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7453 | break; |
7454 | } | |
7455 | ||
4dcf6aff IM |
7456 | if (!group->__cpu_power) { |
7457 | printk(KERN_CONT "\n"); | |
7458 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7459 | "set\n"); | |
7460 | break; | |
7461 | } | |
1da177e4 | 7462 | |
758b2cdc | 7463 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7464 | printk(KERN_CONT "\n"); |
7465 | printk(KERN_ERR "ERROR: empty group\n"); | |
7466 | break; | |
7467 | } | |
1da177e4 | 7468 | |
758b2cdc | 7469 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7470 | printk(KERN_CONT "\n"); |
7471 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7472 | break; | |
7473 | } | |
1da177e4 | 7474 | |
758b2cdc | 7475 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7476 | |
968ea6d8 | 7477 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
46e0bb9c GS |
7478 | printk(KERN_CONT " %s (__cpu_power = %d)", str, |
7479 | group->__cpu_power); | |
1da177e4 | 7480 | |
4dcf6aff IM |
7481 | group = group->next; |
7482 | } while (group != sd->groups); | |
7483 | printk(KERN_CONT "\n"); | |
1da177e4 | 7484 | |
758b2cdc | 7485 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7486 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7487 | |
758b2cdc RR |
7488 | if (sd->parent && |
7489 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7490 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7491 | "of domain->span\n"); | |
7492 | return 0; | |
7493 | } | |
1da177e4 | 7494 | |
4dcf6aff IM |
7495 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7496 | { | |
d5dd3db1 | 7497 | cpumask_var_t groupmask; |
4dcf6aff | 7498 | int level = 0; |
1da177e4 | 7499 | |
4dcf6aff IM |
7500 | if (!sd) { |
7501 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7502 | return; | |
7503 | } | |
1da177e4 | 7504 | |
4dcf6aff IM |
7505 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7506 | ||
d5dd3db1 | 7507 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7508 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7509 | return; | |
7510 | } | |
7511 | ||
4dcf6aff | 7512 | for (;;) { |
7c16ec58 | 7513 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7514 | break; |
1da177e4 LT |
7515 | level++; |
7516 | sd = sd->parent; | |
33859f7f | 7517 | if (!sd) |
4dcf6aff IM |
7518 | break; |
7519 | } | |
d5dd3db1 | 7520 | free_cpumask_var(groupmask); |
1da177e4 | 7521 | } |
6d6bc0ad | 7522 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7523 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7524 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7525 | |
1a20ff27 | 7526 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7527 | { |
758b2cdc | 7528 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7529 | return 1; |
7530 | ||
7531 | /* Following flags need at least 2 groups */ | |
7532 | if (sd->flags & (SD_LOAD_BALANCE | | |
7533 | SD_BALANCE_NEWIDLE | | |
7534 | SD_BALANCE_FORK | | |
89c4710e SS |
7535 | SD_BALANCE_EXEC | |
7536 | SD_SHARE_CPUPOWER | | |
7537 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7538 | if (sd->groups != sd->groups->next) |
7539 | return 0; | |
7540 | } | |
7541 | ||
7542 | /* Following flags don't use groups */ | |
7543 | if (sd->flags & (SD_WAKE_IDLE | | |
7544 | SD_WAKE_AFFINE | | |
7545 | SD_WAKE_BALANCE)) | |
7546 | return 0; | |
7547 | ||
7548 | return 1; | |
7549 | } | |
7550 | ||
48f24c4d IM |
7551 | static int |
7552 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7553 | { |
7554 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7555 | ||
7556 | if (sd_degenerate(parent)) | |
7557 | return 1; | |
7558 | ||
758b2cdc | 7559 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7560 | return 0; |
7561 | ||
7562 | /* Does parent contain flags not in child? */ | |
7563 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7564 | if (cflags & SD_WAKE_AFFINE) | |
7565 | pflags &= ~SD_WAKE_BALANCE; | |
7566 | /* Flags needing groups don't count if only 1 group in parent */ | |
7567 | if (parent->groups == parent->groups->next) { | |
7568 | pflags &= ~(SD_LOAD_BALANCE | | |
7569 | SD_BALANCE_NEWIDLE | | |
7570 | SD_BALANCE_FORK | | |
89c4710e SS |
7571 | SD_BALANCE_EXEC | |
7572 | SD_SHARE_CPUPOWER | | |
7573 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7574 | if (nr_node_ids == 1) |
7575 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7576 | } |
7577 | if (~cflags & pflags) | |
7578 | return 0; | |
7579 | ||
7580 | return 1; | |
7581 | } | |
7582 | ||
c6c4927b RR |
7583 | static void free_rootdomain(struct root_domain *rd) |
7584 | { | |
68e74568 RR |
7585 | cpupri_cleanup(&rd->cpupri); |
7586 | ||
c6c4927b RR |
7587 | free_cpumask_var(rd->rto_mask); |
7588 | free_cpumask_var(rd->online); | |
7589 | free_cpumask_var(rd->span); | |
7590 | kfree(rd); | |
7591 | } | |
7592 | ||
57d885fe GH |
7593 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7594 | { | |
a0490fa3 | 7595 | struct root_domain *old_rd = NULL; |
57d885fe | 7596 | unsigned long flags; |
57d885fe GH |
7597 | |
7598 | spin_lock_irqsave(&rq->lock, flags); | |
7599 | ||
7600 | if (rq->rd) { | |
a0490fa3 | 7601 | old_rd = rq->rd; |
57d885fe | 7602 | |
c6c4927b | 7603 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7604 | set_rq_offline(rq); |
57d885fe | 7605 | |
c6c4927b | 7606 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7607 | |
a0490fa3 IM |
7608 | /* |
7609 | * If we dont want to free the old_rt yet then | |
7610 | * set old_rd to NULL to skip the freeing later | |
7611 | * in this function: | |
7612 | */ | |
7613 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7614 | old_rd = NULL; | |
57d885fe GH |
7615 | } |
7616 | ||
7617 | atomic_inc(&rd->refcount); | |
7618 | rq->rd = rd; | |
7619 | ||
c6c4927b RR |
7620 | cpumask_set_cpu(rq->cpu, rd->span); |
7621 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7622 | set_rq_online(rq); |
57d885fe GH |
7623 | |
7624 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7625 | |
7626 | if (old_rd) | |
7627 | free_rootdomain(old_rd); | |
57d885fe GH |
7628 | } |
7629 | ||
db2f59c8 | 7630 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7631 | { |
7632 | memset(rd, 0, sizeof(*rd)); | |
7633 | ||
c6c4927b RR |
7634 | if (bootmem) { |
7635 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7636 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7637 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7638 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7639 | return 0; |
7640 | } | |
7641 | ||
7642 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7643 | goto out; |
c6c4927b RR |
7644 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7645 | goto free_span; | |
7646 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7647 | goto free_online; | |
6e0534f2 | 7648 | |
68e74568 RR |
7649 | if (cpupri_init(&rd->cpupri, false) != 0) |
7650 | goto free_rto_mask; | |
c6c4927b | 7651 | return 0; |
6e0534f2 | 7652 | |
68e74568 RR |
7653 | free_rto_mask: |
7654 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7655 | free_online: |
7656 | free_cpumask_var(rd->online); | |
7657 | free_span: | |
7658 | free_cpumask_var(rd->span); | |
0c910d28 | 7659 | out: |
c6c4927b | 7660 | return -ENOMEM; |
57d885fe GH |
7661 | } |
7662 | ||
7663 | static void init_defrootdomain(void) | |
7664 | { | |
c6c4927b RR |
7665 | init_rootdomain(&def_root_domain, true); |
7666 | ||
57d885fe GH |
7667 | atomic_set(&def_root_domain.refcount, 1); |
7668 | } | |
7669 | ||
dc938520 | 7670 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7671 | { |
7672 | struct root_domain *rd; | |
7673 | ||
7674 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7675 | if (!rd) | |
7676 | return NULL; | |
7677 | ||
c6c4927b RR |
7678 | if (init_rootdomain(rd, false) != 0) { |
7679 | kfree(rd); | |
7680 | return NULL; | |
7681 | } | |
57d885fe GH |
7682 | |
7683 | return rd; | |
7684 | } | |
7685 | ||
1da177e4 | 7686 | /* |
0eab9146 | 7687 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7688 | * hold the hotplug lock. |
7689 | */ | |
0eab9146 IM |
7690 | static void |
7691 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7692 | { |
70b97a7f | 7693 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7694 | struct sched_domain *tmp; |
7695 | ||
7696 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7697 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7698 | struct sched_domain *parent = tmp->parent; |
7699 | if (!parent) | |
7700 | break; | |
f29c9b1c | 7701 | |
1a848870 | 7702 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7703 | tmp->parent = parent->parent; |
1a848870 SS |
7704 | if (parent->parent) |
7705 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7706 | } else |
7707 | tmp = tmp->parent; | |
245af2c7 SS |
7708 | } |
7709 | ||
1a848870 | 7710 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7711 | sd = sd->parent; |
1a848870 SS |
7712 | if (sd) |
7713 | sd->child = NULL; | |
7714 | } | |
1da177e4 LT |
7715 | |
7716 | sched_domain_debug(sd, cpu); | |
7717 | ||
57d885fe | 7718 | rq_attach_root(rq, rd); |
674311d5 | 7719 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7720 | } |
7721 | ||
7722 | /* cpus with isolated domains */ | |
dcc30a35 | 7723 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7724 | |
7725 | /* Setup the mask of cpus configured for isolated domains */ | |
7726 | static int __init isolated_cpu_setup(char *str) | |
7727 | { | |
968ea6d8 | 7728 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7729 | return 1; |
7730 | } | |
7731 | ||
8927f494 | 7732 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7733 | |
7734 | /* | |
6711cab4 SS |
7735 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7736 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7737 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7738 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7739 | * |
7740 | * init_sched_build_groups will build a circular linked list of the groups | |
7741 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7742 | * and ->cpu_power to 0. | |
7743 | */ | |
a616058b | 7744 | static void |
96f874e2 RR |
7745 | init_sched_build_groups(const struct cpumask *span, |
7746 | const struct cpumask *cpu_map, | |
7747 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7748 | struct sched_group **sg, |
96f874e2 RR |
7749 | struct cpumask *tmpmask), |
7750 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7751 | { |
7752 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7753 | int i; |
7754 | ||
96f874e2 | 7755 | cpumask_clear(covered); |
7c16ec58 | 7756 | |
abcd083a | 7757 | for_each_cpu(i, span) { |
6711cab4 | 7758 | struct sched_group *sg; |
7c16ec58 | 7759 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7760 | int j; |
7761 | ||
758b2cdc | 7762 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7763 | continue; |
7764 | ||
758b2cdc | 7765 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7766 | sg->__cpu_power = 0; |
1da177e4 | 7767 | |
abcd083a | 7768 | for_each_cpu(j, span) { |
7c16ec58 | 7769 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7770 | continue; |
7771 | ||
96f874e2 | 7772 | cpumask_set_cpu(j, covered); |
758b2cdc | 7773 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7774 | } |
7775 | if (!first) | |
7776 | first = sg; | |
7777 | if (last) | |
7778 | last->next = sg; | |
7779 | last = sg; | |
7780 | } | |
7781 | last->next = first; | |
7782 | } | |
7783 | ||
9c1cfda2 | 7784 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7785 | |
9c1cfda2 | 7786 | #ifdef CONFIG_NUMA |
198e2f18 | 7787 | |
9c1cfda2 JH |
7788 | /** |
7789 | * find_next_best_node - find the next node to include in a sched_domain | |
7790 | * @node: node whose sched_domain we're building | |
7791 | * @used_nodes: nodes already in the sched_domain | |
7792 | * | |
41a2d6cf | 7793 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7794 | * finds the closest node not already in the @used_nodes map. |
7795 | * | |
7796 | * Should use nodemask_t. | |
7797 | */ | |
c5f59f08 | 7798 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7799 | { |
7800 | int i, n, val, min_val, best_node = 0; | |
7801 | ||
7802 | min_val = INT_MAX; | |
7803 | ||
076ac2af | 7804 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7805 | /* Start at @node */ |
076ac2af | 7806 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7807 | |
7808 | if (!nr_cpus_node(n)) | |
7809 | continue; | |
7810 | ||
7811 | /* Skip already used nodes */ | |
c5f59f08 | 7812 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7813 | continue; |
7814 | ||
7815 | /* Simple min distance search */ | |
7816 | val = node_distance(node, n); | |
7817 | ||
7818 | if (val < min_val) { | |
7819 | min_val = val; | |
7820 | best_node = n; | |
7821 | } | |
7822 | } | |
7823 | ||
c5f59f08 | 7824 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7825 | return best_node; |
7826 | } | |
7827 | ||
7828 | /** | |
7829 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7830 | * @node: node whose cpumask we're constructing | |
73486722 | 7831 | * @span: resulting cpumask |
9c1cfda2 | 7832 | * |
41a2d6cf | 7833 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7834 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7835 | * out optimally. | |
7836 | */ | |
96f874e2 | 7837 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7838 | { |
c5f59f08 | 7839 | nodemask_t used_nodes; |
48f24c4d | 7840 | int i; |
9c1cfda2 | 7841 | |
6ca09dfc | 7842 | cpumask_clear(span); |
c5f59f08 | 7843 | nodes_clear(used_nodes); |
9c1cfda2 | 7844 | |
6ca09dfc | 7845 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7846 | node_set(node, used_nodes); |
9c1cfda2 JH |
7847 | |
7848 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7849 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7850 | |
6ca09dfc | 7851 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7852 | } |
9c1cfda2 | 7853 | } |
6d6bc0ad | 7854 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7855 | |
5c45bf27 | 7856 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7857 | |
6c99e9ad RR |
7858 | /* |
7859 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7860 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7861 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7862 | */ | |
7863 | struct static_sched_group { | |
7864 | struct sched_group sg; | |
7865 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7866 | }; | |
7867 | ||
7868 | struct static_sched_domain { | |
7869 | struct sched_domain sd; | |
7870 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7871 | }; | |
7872 | ||
9c1cfda2 | 7873 | /* |
48f24c4d | 7874 | * SMT sched-domains: |
9c1cfda2 | 7875 | */ |
1da177e4 | 7876 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7877 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7878 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7879 | |
41a2d6cf | 7880 | static int |
96f874e2 RR |
7881 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7882 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7883 | { |
6711cab4 | 7884 | if (sg) |
6c99e9ad | 7885 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7886 | return cpu; |
7887 | } | |
6d6bc0ad | 7888 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7889 | |
48f24c4d IM |
7890 | /* |
7891 | * multi-core sched-domains: | |
7892 | */ | |
1e9f28fa | 7893 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7894 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7895 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7896 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7897 | |
7898 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7899 | static int |
96f874e2 RR |
7900 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7901 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7902 | { |
6711cab4 | 7903 | int group; |
7c16ec58 | 7904 | |
c69fc56d | 7905 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7906 | group = cpumask_first(mask); |
6711cab4 | 7907 | if (sg) |
6c99e9ad | 7908 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7909 | return group; |
1e9f28fa SS |
7910 | } |
7911 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7912 | static int |
96f874e2 RR |
7913 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7914 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7915 | { |
6711cab4 | 7916 | if (sg) |
6c99e9ad | 7917 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7918 | return cpu; |
7919 | } | |
7920 | #endif | |
7921 | ||
6c99e9ad RR |
7922 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7923 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7924 | |
41a2d6cf | 7925 | static int |
96f874e2 RR |
7926 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7927 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7928 | { |
6711cab4 | 7929 | int group; |
48f24c4d | 7930 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7931 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7932 | group = cpumask_first(mask); |
1e9f28fa | 7933 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 7934 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7935 | group = cpumask_first(mask); |
1da177e4 | 7936 | #else |
6711cab4 | 7937 | group = cpu; |
1da177e4 | 7938 | #endif |
6711cab4 | 7939 | if (sg) |
6c99e9ad | 7940 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7941 | return group; |
1da177e4 LT |
7942 | } |
7943 | ||
7944 | #ifdef CONFIG_NUMA | |
1da177e4 | 7945 | /* |
9c1cfda2 JH |
7946 | * The init_sched_build_groups can't handle what we want to do with node |
7947 | * groups, so roll our own. Now each node has its own list of groups which | |
7948 | * gets dynamically allocated. | |
1da177e4 | 7949 | */ |
62ea9ceb | 7950 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7951 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7952 | |
62ea9ceb | 7953 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7954 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7955 | |
96f874e2 RR |
7956 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7957 | struct sched_group **sg, | |
7958 | struct cpumask *nodemask) | |
9c1cfda2 | 7959 | { |
6711cab4 SS |
7960 | int group; |
7961 | ||
6ca09dfc | 7962 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7963 | group = cpumask_first(nodemask); |
6711cab4 SS |
7964 | |
7965 | if (sg) | |
6c99e9ad | 7966 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7967 | return group; |
1da177e4 | 7968 | } |
6711cab4 | 7969 | |
08069033 SS |
7970 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7971 | { | |
7972 | struct sched_group *sg = group_head; | |
7973 | int j; | |
7974 | ||
7975 | if (!sg) | |
7976 | return; | |
3a5c359a | 7977 | do { |
758b2cdc | 7978 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7979 | struct sched_domain *sd; |
08069033 | 7980 | |
6c99e9ad | 7981 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7982 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7983 | /* |
7984 | * Only add "power" once for each | |
7985 | * physical package. | |
7986 | */ | |
7987 | continue; | |
7988 | } | |
08069033 | 7989 | |
3a5c359a AK |
7990 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7991 | } | |
7992 | sg = sg->next; | |
7993 | } while (sg != group_head); | |
08069033 | 7994 | } |
6d6bc0ad | 7995 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7996 | |
a616058b | 7997 | #ifdef CONFIG_NUMA |
51888ca2 | 7998 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7999 | static void free_sched_groups(const struct cpumask *cpu_map, |
8000 | struct cpumask *nodemask) | |
51888ca2 | 8001 | { |
a616058b | 8002 | int cpu, i; |
51888ca2 | 8003 | |
abcd083a | 8004 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8005 | struct sched_group **sched_group_nodes |
8006 | = sched_group_nodes_bycpu[cpu]; | |
8007 | ||
51888ca2 SV |
8008 | if (!sched_group_nodes) |
8009 | continue; | |
8010 | ||
076ac2af | 8011 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8012 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8013 | ||
6ca09dfc | 8014 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8015 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8016 | continue; |
8017 | ||
8018 | if (sg == NULL) | |
8019 | continue; | |
8020 | sg = sg->next; | |
8021 | next_sg: | |
8022 | oldsg = sg; | |
8023 | sg = sg->next; | |
8024 | kfree(oldsg); | |
8025 | if (oldsg != sched_group_nodes[i]) | |
8026 | goto next_sg; | |
8027 | } | |
8028 | kfree(sched_group_nodes); | |
8029 | sched_group_nodes_bycpu[cpu] = NULL; | |
8030 | } | |
51888ca2 | 8031 | } |
6d6bc0ad | 8032 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8033 | static void free_sched_groups(const struct cpumask *cpu_map, |
8034 | struct cpumask *nodemask) | |
a616058b SS |
8035 | { |
8036 | } | |
6d6bc0ad | 8037 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8038 | |
89c4710e SS |
8039 | /* |
8040 | * Initialize sched groups cpu_power. | |
8041 | * | |
8042 | * cpu_power indicates the capacity of sched group, which is used while | |
8043 | * distributing the load between different sched groups in a sched domain. | |
8044 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8045 | * there are asymmetries in the topology. If there are asymmetries, group | |
8046 | * having more cpu_power will pickup more load compared to the group having | |
8047 | * less cpu_power. | |
8048 | * | |
8049 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8050 | * the maximum number of tasks a group can handle in the presence of other idle | |
8051 | * or lightly loaded groups in the same sched domain. | |
8052 | */ | |
8053 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8054 | { | |
8055 | struct sched_domain *child; | |
8056 | struct sched_group *group; | |
8057 | ||
8058 | WARN_ON(!sd || !sd->groups); | |
8059 | ||
758b2cdc | 8060 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
8061 | return; |
8062 | ||
8063 | child = sd->child; | |
8064 | ||
5517d86b ED |
8065 | sd->groups->__cpu_power = 0; |
8066 | ||
89c4710e SS |
8067 | /* |
8068 | * For perf policy, if the groups in child domain share resources | |
8069 | * (for example cores sharing some portions of the cache hierarchy | |
8070 | * or SMT), then set this domain groups cpu_power such that each group | |
8071 | * can handle only one task, when there are other idle groups in the | |
8072 | * same sched domain. | |
8073 | */ | |
8074 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8075 | (child->flags & | |
8076 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8077 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8078 | return; |
8079 | } | |
8080 | ||
89c4710e SS |
8081 | /* |
8082 | * add cpu_power of each child group to this groups cpu_power | |
8083 | */ | |
8084 | group = child->groups; | |
8085 | do { | |
5517d86b | 8086 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8087 | group = group->next; |
8088 | } while (group != child->groups); | |
8089 | } | |
8090 | ||
7c16ec58 MT |
8091 | /* |
8092 | * Initializers for schedule domains | |
8093 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8094 | */ | |
8095 | ||
a5d8c348 IM |
8096 | #ifdef CONFIG_SCHED_DEBUG |
8097 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8098 | #else | |
8099 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8100 | #endif | |
8101 | ||
7c16ec58 | 8102 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8103 | |
7c16ec58 MT |
8104 | #define SD_INIT_FUNC(type) \ |
8105 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8106 | { \ | |
8107 | memset(sd, 0, sizeof(*sd)); \ | |
8108 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8109 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8110 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8111 | } |
8112 | ||
8113 | SD_INIT_FUNC(CPU) | |
8114 | #ifdef CONFIG_NUMA | |
8115 | SD_INIT_FUNC(ALLNODES) | |
8116 | SD_INIT_FUNC(NODE) | |
8117 | #endif | |
8118 | #ifdef CONFIG_SCHED_SMT | |
8119 | SD_INIT_FUNC(SIBLING) | |
8120 | #endif | |
8121 | #ifdef CONFIG_SCHED_MC | |
8122 | SD_INIT_FUNC(MC) | |
8123 | #endif | |
8124 | ||
1d3504fc HS |
8125 | static int default_relax_domain_level = -1; |
8126 | ||
8127 | static int __init setup_relax_domain_level(char *str) | |
8128 | { | |
30e0e178 LZ |
8129 | unsigned long val; |
8130 | ||
8131 | val = simple_strtoul(str, NULL, 0); | |
8132 | if (val < SD_LV_MAX) | |
8133 | default_relax_domain_level = val; | |
8134 | ||
1d3504fc HS |
8135 | return 1; |
8136 | } | |
8137 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8138 | ||
8139 | static void set_domain_attribute(struct sched_domain *sd, | |
8140 | struct sched_domain_attr *attr) | |
8141 | { | |
8142 | int request; | |
8143 | ||
8144 | if (!attr || attr->relax_domain_level < 0) { | |
8145 | if (default_relax_domain_level < 0) | |
8146 | return; | |
8147 | else | |
8148 | request = default_relax_domain_level; | |
8149 | } else | |
8150 | request = attr->relax_domain_level; | |
8151 | if (request < sd->level) { | |
8152 | /* turn off idle balance on this domain */ | |
8153 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8154 | } else { | |
8155 | /* turn on idle balance on this domain */ | |
8156 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8157 | } | |
8158 | } | |
8159 | ||
1da177e4 | 8160 | /* |
1a20ff27 DG |
8161 | * Build sched domains for a given set of cpus and attach the sched domains |
8162 | * to the individual cpus | |
1da177e4 | 8163 | */ |
96f874e2 | 8164 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8165 | struct sched_domain_attr *attr) |
1da177e4 | 8166 | { |
3404c8d9 | 8167 | int i, err = -ENOMEM; |
57d885fe | 8168 | struct root_domain *rd; |
3404c8d9 RR |
8169 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8170 | tmpmask; | |
d1b55138 | 8171 | #ifdef CONFIG_NUMA |
3404c8d9 | 8172 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8173 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8174 | int sd_allnodes = 0; |
d1b55138 | 8175 | |
3404c8d9 RR |
8176 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8177 | goto out; | |
8178 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8179 | goto free_domainspan; | |
8180 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8181 | goto free_covered; | |
8182 | #endif | |
8183 | ||
8184 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8185 | goto free_notcovered; | |
8186 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8187 | goto free_nodemask; | |
8188 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8189 | goto free_this_sibling_map; | |
8190 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8191 | goto free_this_core_map; | |
8192 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8193 | goto free_send_covered; | |
8194 | ||
8195 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8196 | /* |
8197 | * Allocate the per-node list of sched groups | |
8198 | */ | |
076ac2af | 8199 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8200 | GFP_KERNEL); |
d1b55138 JH |
8201 | if (!sched_group_nodes) { |
8202 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8203 | goto free_tmpmask; |
d1b55138 | 8204 | } |
d1b55138 | 8205 | #endif |
1da177e4 | 8206 | |
dc938520 | 8207 | rd = alloc_rootdomain(); |
57d885fe GH |
8208 | if (!rd) { |
8209 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8210 | goto free_sched_groups; |
57d885fe GH |
8211 | } |
8212 | ||
7c16ec58 | 8213 | #ifdef CONFIG_NUMA |
96f874e2 | 8214 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8215 | #endif |
8216 | ||
1da177e4 | 8217 | /* |
1a20ff27 | 8218 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8219 | */ |
abcd083a | 8220 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8221 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8222 | |
6ca09dfc | 8223 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8224 | |
8225 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8226 | if (cpumask_weight(cpu_map) > |
8227 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8228 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8229 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8230 | set_domain_attribute(sd, attr); |
758b2cdc | 8231 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8232 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8233 | p = sd; |
6711cab4 | 8234 | sd_allnodes = 1; |
9c1cfda2 JH |
8235 | } else |
8236 | p = NULL; | |
8237 | ||
62ea9ceb | 8238 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8239 | SD_INIT(sd, NODE); |
1d3504fc | 8240 | set_domain_attribute(sd, attr); |
758b2cdc | 8241 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8242 | sd->parent = p; |
1a848870 SS |
8243 | if (p) |
8244 | p->child = sd; | |
758b2cdc RR |
8245 | cpumask_and(sched_domain_span(sd), |
8246 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8247 | #endif |
8248 | ||
8249 | p = sd; | |
6c99e9ad | 8250 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8251 | SD_INIT(sd, CPU); |
1d3504fc | 8252 | set_domain_attribute(sd, attr); |
758b2cdc | 8253 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8254 | sd->parent = p; |
1a848870 SS |
8255 | if (p) |
8256 | p->child = sd; | |
7c16ec58 | 8257 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8258 | |
1e9f28fa SS |
8259 | #ifdef CONFIG_SCHED_MC |
8260 | p = sd; | |
6c99e9ad | 8261 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8262 | SD_INIT(sd, MC); |
1d3504fc | 8263 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8264 | cpumask_and(sched_domain_span(sd), cpu_map, |
8265 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8266 | sd->parent = p; |
1a848870 | 8267 | p->child = sd; |
7c16ec58 | 8268 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8269 | #endif |
8270 | ||
1da177e4 LT |
8271 | #ifdef CONFIG_SCHED_SMT |
8272 | p = sd; | |
6c99e9ad | 8273 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8274 | SD_INIT(sd, SIBLING); |
1d3504fc | 8275 | set_domain_attribute(sd, attr); |
758b2cdc | 8276 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8277 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8278 | sd->parent = p; |
1a848870 | 8279 | p->child = sd; |
7c16ec58 | 8280 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8281 | #endif |
8282 | } | |
8283 | ||
8284 | #ifdef CONFIG_SCHED_SMT | |
8285 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8286 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8287 | cpumask_and(this_sibling_map, |
c69fc56d | 8288 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8289 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8290 | continue; |
8291 | ||
dd41f596 | 8292 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8293 | &cpu_to_cpu_group, |
8294 | send_covered, tmpmask); | |
1da177e4 LT |
8295 | } |
8296 | #endif | |
8297 | ||
1e9f28fa SS |
8298 | #ifdef CONFIG_SCHED_MC |
8299 | /* Set up multi-core groups */ | |
abcd083a | 8300 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8301 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8302 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8303 | continue; |
7c16ec58 | 8304 | |
dd41f596 | 8305 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8306 | &cpu_to_core_group, |
8307 | send_covered, tmpmask); | |
1e9f28fa SS |
8308 | } |
8309 | #endif | |
8310 | ||
1da177e4 | 8311 | /* Set up physical groups */ |
076ac2af | 8312 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8313 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8314 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8315 | continue; |
8316 | ||
7c16ec58 MT |
8317 | init_sched_build_groups(nodemask, cpu_map, |
8318 | &cpu_to_phys_group, | |
8319 | send_covered, tmpmask); | |
1da177e4 LT |
8320 | } |
8321 | ||
8322 | #ifdef CONFIG_NUMA | |
8323 | /* Set up node groups */ | |
7c16ec58 | 8324 | if (sd_allnodes) { |
7c16ec58 MT |
8325 | init_sched_build_groups(cpu_map, cpu_map, |
8326 | &cpu_to_allnodes_group, | |
8327 | send_covered, tmpmask); | |
8328 | } | |
9c1cfda2 | 8329 | |
076ac2af | 8330 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8331 | /* Set up node groups */ |
8332 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8333 | int j; |
8334 | ||
96f874e2 | 8335 | cpumask_clear(covered); |
6ca09dfc | 8336 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8337 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8338 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8339 | continue; |
d1b55138 | 8340 | } |
9c1cfda2 | 8341 | |
4bdbaad3 | 8342 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8343 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8344 | |
6c99e9ad RR |
8345 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8346 | GFP_KERNEL, i); | |
51888ca2 SV |
8347 | if (!sg) { |
8348 | printk(KERN_WARNING "Can not alloc domain group for " | |
8349 | "node %d\n", i); | |
8350 | goto error; | |
8351 | } | |
9c1cfda2 | 8352 | sched_group_nodes[i] = sg; |
abcd083a | 8353 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8354 | struct sched_domain *sd; |
9761eea8 | 8355 | |
62ea9ceb | 8356 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8357 | sd->groups = sg; |
9c1cfda2 | 8358 | } |
5517d86b | 8359 | sg->__cpu_power = 0; |
758b2cdc | 8360 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8361 | sg->next = sg; |
96f874e2 | 8362 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8363 | prev = sg; |
8364 | ||
076ac2af | 8365 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8366 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8367 | |
96f874e2 RR |
8368 | cpumask_complement(notcovered, covered); |
8369 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8370 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8371 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8372 | break; |
8373 | ||
6ca09dfc | 8374 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8375 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8376 | continue; |
8377 | ||
6c99e9ad RR |
8378 | sg = kmalloc_node(sizeof(struct sched_group) + |
8379 | cpumask_size(), | |
15f0b676 | 8380 | GFP_KERNEL, i); |
9c1cfda2 JH |
8381 | if (!sg) { |
8382 | printk(KERN_WARNING | |
8383 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8384 | goto error; |
9c1cfda2 | 8385 | } |
5517d86b | 8386 | sg->__cpu_power = 0; |
758b2cdc | 8387 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8388 | sg->next = prev->next; |
96f874e2 | 8389 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8390 | prev->next = sg; |
8391 | prev = sg; | |
8392 | } | |
9c1cfda2 | 8393 | } |
1da177e4 LT |
8394 | #endif |
8395 | ||
8396 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8397 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8398 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8399 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8400 | |
89c4710e | 8401 | init_sched_groups_power(i, sd); |
5c45bf27 | 8402 | } |
1da177e4 | 8403 | #endif |
1e9f28fa | 8404 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8405 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8406 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8407 | |
89c4710e | 8408 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8409 | } |
8410 | #endif | |
1e9f28fa | 8411 | |
abcd083a | 8412 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8413 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8414 | |
89c4710e | 8415 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8416 | } |
8417 | ||
9c1cfda2 | 8418 | #ifdef CONFIG_NUMA |
076ac2af | 8419 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8420 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8421 | |
6711cab4 SS |
8422 | if (sd_allnodes) { |
8423 | struct sched_group *sg; | |
f712c0c7 | 8424 | |
96f874e2 | 8425 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8426 | tmpmask); |
f712c0c7 SS |
8427 | init_numa_sched_groups_power(sg); |
8428 | } | |
9c1cfda2 JH |
8429 | #endif |
8430 | ||
1da177e4 | 8431 | /* Attach the domains */ |
abcd083a | 8432 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8433 | struct sched_domain *sd; |
8434 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8435 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8436 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8437 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8438 | #else |
6c99e9ad | 8439 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8440 | #endif |
57d885fe | 8441 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8442 | } |
51888ca2 | 8443 | |
3404c8d9 RR |
8444 | err = 0; |
8445 | ||
8446 | free_tmpmask: | |
8447 | free_cpumask_var(tmpmask); | |
8448 | free_send_covered: | |
8449 | free_cpumask_var(send_covered); | |
8450 | free_this_core_map: | |
8451 | free_cpumask_var(this_core_map); | |
8452 | free_this_sibling_map: | |
8453 | free_cpumask_var(this_sibling_map); | |
8454 | free_nodemask: | |
8455 | free_cpumask_var(nodemask); | |
8456 | free_notcovered: | |
8457 | #ifdef CONFIG_NUMA | |
8458 | free_cpumask_var(notcovered); | |
8459 | free_covered: | |
8460 | free_cpumask_var(covered); | |
8461 | free_domainspan: | |
8462 | free_cpumask_var(domainspan); | |
8463 | out: | |
8464 | #endif | |
8465 | return err; | |
8466 | ||
8467 | free_sched_groups: | |
8468 | #ifdef CONFIG_NUMA | |
8469 | kfree(sched_group_nodes); | |
8470 | #endif | |
8471 | goto free_tmpmask; | |
51888ca2 | 8472 | |
a616058b | 8473 | #ifdef CONFIG_NUMA |
51888ca2 | 8474 | error: |
7c16ec58 | 8475 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8476 | free_rootdomain(rd); |
3404c8d9 | 8477 | goto free_tmpmask; |
a616058b | 8478 | #endif |
1da177e4 | 8479 | } |
029190c5 | 8480 | |
96f874e2 | 8481 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8482 | { |
8483 | return __build_sched_domains(cpu_map, NULL); | |
8484 | } | |
8485 | ||
96f874e2 | 8486 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8487 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8488 | static struct sched_domain_attr *dattr_cur; |
8489 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8490 | |
8491 | /* | |
8492 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8493 | * cpumask) fails, then fallback to a single sched domain, |
8494 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8495 | */ |
4212823f | 8496 | static cpumask_var_t fallback_doms; |
029190c5 | 8497 | |
ee79d1bd HC |
8498 | /* |
8499 | * arch_update_cpu_topology lets virtualized architectures update the | |
8500 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8501 | * or 0 if it stayed the same. | |
8502 | */ | |
8503 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8504 | { |
ee79d1bd | 8505 | return 0; |
22e52b07 HC |
8506 | } |
8507 | ||
1a20ff27 | 8508 | /* |
41a2d6cf | 8509 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8510 | * For now this just excludes isolated cpus, but could be used to |
8511 | * exclude other special cases in the future. | |
1a20ff27 | 8512 | */ |
96f874e2 | 8513 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8514 | { |
7378547f MM |
8515 | int err; |
8516 | ||
22e52b07 | 8517 | arch_update_cpu_topology(); |
029190c5 | 8518 | ndoms_cur = 1; |
96f874e2 | 8519 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8520 | if (!doms_cur) |
4212823f | 8521 | doms_cur = fallback_doms; |
dcc30a35 | 8522 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8523 | dattr_cur = NULL; |
7378547f | 8524 | err = build_sched_domains(doms_cur); |
6382bc90 | 8525 | register_sched_domain_sysctl(); |
7378547f MM |
8526 | |
8527 | return err; | |
1a20ff27 DG |
8528 | } |
8529 | ||
96f874e2 RR |
8530 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8531 | struct cpumask *tmpmask) | |
1da177e4 | 8532 | { |
7c16ec58 | 8533 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8534 | } |
1da177e4 | 8535 | |
1a20ff27 DG |
8536 | /* |
8537 | * Detach sched domains from a group of cpus specified in cpu_map | |
8538 | * These cpus will now be attached to the NULL domain | |
8539 | */ | |
96f874e2 | 8540 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8541 | { |
96f874e2 RR |
8542 | /* Save because hotplug lock held. */ |
8543 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8544 | int i; |
8545 | ||
abcd083a | 8546 | for_each_cpu(i, cpu_map) |
57d885fe | 8547 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8548 | synchronize_sched(); |
96f874e2 | 8549 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8550 | } |
8551 | ||
1d3504fc HS |
8552 | /* handle null as "default" */ |
8553 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8554 | struct sched_domain_attr *new, int idx_new) | |
8555 | { | |
8556 | struct sched_domain_attr tmp; | |
8557 | ||
8558 | /* fast path */ | |
8559 | if (!new && !cur) | |
8560 | return 1; | |
8561 | ||
8562 | tmp = SD_ATTR_INIT; | |
8563 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8564 | new ? (new + idx_new) : &tmp, | |
8565 | sizeof(struct sched_domain_attr)); | |
8566 | } | |
8567 | ||
029190c5 PJ |
8568 | /* |
8569 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8570 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8571 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8572 | * It destroys each deleted domain and builds each new domain. | |
8573 | * | |
96f874e2 | 8574 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8575 | * The masks don't intersect (don't overlap.) We should setup one |
8576 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8577 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8578 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8579 | * it as it is. | |
8580 | * | |
41a2d6cf IM |
8581 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8582 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8583 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8584 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8585 | * the single partition 'fallback_doms', it also forces the domains | |
8586 | * to be rebuilt. | |
029190c5 | 8587 | * |
96f874e2 | 8588 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8589 | * ndoms_new == 0 is a special case for destroying existing domains, |
8590 | * and it will not create the default domain. | |
dfb512ec | 8591 | * |
029190c5 PJ |
8592 | * Call with hotplug lock held |
8593 | */ | |
96f874e2 RR |
8594 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8595 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8596 | struct sched_domain_attr *dattr_new) |
029190c5 | 8597 | { |
dfb512ec | 8598 | int i, j, n; |
d65bd5ec | 8599 | int new_topology; |
029190c5 | 8600 | |
712555ee | 8601 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8602 | |
7378547f MM |
8603 | /* always unregister in case we don't destroy any domains */ |
8604 | unregister_sched_domain_sysctl(); | |
8605 | ||
d65bd5ec HC |
8606 | /* Let architecture update cpu core mappings. */ |
8607 | new_topology = arch_update_cpu_topology(); | |
8608 | ||
dfb512ec | 8609 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8610 | |
8611 | /* Destroy deleted domains */ | |
8612 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8613 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8614 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8615 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8616 | goto match1; |
8617 | } | |
8618 | /* no match - a current sched domain not in new doms_new[] */ | |
8619 | detach_destroy_domains(doms_cur + i); | |
8620 | match1: | |
8621 | ; | |
8622 | } | |
8623 | ||
e761b772 MK |
8624 | if (doms_new == NULL) { |
8625 | ndoms_cur = 0; | |
4212823f | 8626 | doms_new = fallback_doms; |
dcc30a35 | 8627 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8628 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8629 | } |
8630 | ||
029190c5 PJ |
8631 | /* Build new domains */ |
8632 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8633 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8634 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8635 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8636 | goto match2; |
8637 | } | |
8638 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8639 | __build_sched_domains(doms_new + i, |
8640 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8641 | match2: |
8642 | ; | |
8643 | } | |
8644 | ||
8645 | /* Remember the new sched domains */ | |
4212823f | 8646 | if (doms_cur != fallback_doms) |
029190c5 | 8647 | kfree(doms_cur); |
1d3504fc | 8648 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8649 | doms_cur = doms_new; |
1d3504fc | 8650 | dattr_cur = dattr_new; |
029190c5 | 8651 | ndoms_cur = ndoms_new; |
7378547f MM |
8652 | |
8653 | register_sched_domain_sysctl(); | |
a1835615 | 8654 | |
712555ee | 8655 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8656 | } |
8657 | ||
5c45bf27 | 8658 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8659 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8660 | { |
95402b38 | 8661 | get_online_cpus(); |
dfb512ec MK |
8662 | |
8663 | /* Destroy domains first to force the rebuild */ | |
8664 | partition_sched_domains(0, NULL, NULL); | |
8665 | ||
e761b772 | 8666 | rebuild_sched_domains(); |
95402b38 | 8667 | put_online_cpus(); |
5c45bf27 SS |
8668 | } |
8669 | ||
8670 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8671 | { | |
afb8a9b7 | 8672 | unsigned int level = 0; |
5c45bf27 | 8673 | |
afb8a9b7 GS |
8674 | if (sscanf(buf, "%u", &level) != 1) |
8675 | return -EINVAL; | |
8676 | ||
8677 | /* | |
8678 | * level is always be positive so don't check for | |
8679 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8680 | * What happens on 0 or 1 byte write, | |
8681 | * need to check for count as well? | |
8682 | */ | |
8683 | ||
8684 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8685 | return -EINVAL; |
8686 | ||
8687 | if (smt) | |
afb8a9b7 | 8688 | sched_smt_power_savings = level; |
5c45bf27 | 8689 | else |
afb8a9b7 | 8690 | sched_mc_power_savings = level; |
5c45bf27 | 8691 | |
c70f22d2 | 8692 | arch_reinit_sched_domains(); |
5c45bf27 | 8693 | |
c70f22d2 | 8694 | return count; |
5c45bf27 SS |
8695 | } |
8696 | ||
5c45bf27 | 8697 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8698 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8699 | char *page) | |
5c45bf27 SS |
8700 | { |
8701 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8702 | } | |
f718cd4a | 8703 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8704 | const char *buf, size_t count) |
5c45bf27 SS |
8705 | { |
8706 | return sched_power_savings_store(buf, count, 0); | |
8707 | } | |
f718cd4a AK |
8708 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8709 | sched_mc_power_savings_show, | |
8710 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8711 | #endif |
8712 | ||
8713 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8714 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8715 | char *page) | |
5c45bf27 SS |
8716 | { |
8717 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8718 | } | |
f718cd4a | 8719 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8720 | const char *buf, size_t count) |
5c45bf27 SS |
8721 | { |
8722 | return sched_power_savings_store(buf, count, 1); | |
8723 | } | |
f718cd4a AK |
8724 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8725 | sched_smt_power_savings_show, | |
6707de00 AB |
8726 | sched_smt_power_savings_store); |
8727 | #endif | |
8728 | ||
39aac648 | 8729 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8730 | { |
8731 | int err = 0; | |
8732 | ||
8733 | #ifdef CONFIG_SCHED_SMT | |
8734 | if (smt_capable()) | |
8735 | err = sysfs_create_file(&cls->kset.kobj, | |
8736 | &attr_sched_smt_power_savings.attr); | |
8737 | #endif | |
8738 | #ifdef CONFIG_SCHED_MC | |
8739 | if (!err && mc_capable()) | |
8740 | err = sysfs_create_file(&cls->kset.kobj, | |
8741 | &attr_sched_mc_power_savings.attr); | |
8742 | #endif | |
8743 | return err; | |
8744 | } | |
6d6bc0ad | 8745 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8746 | |
e761b772 | 8747 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8748 | /* |
e761b772 MK |
8749 | * Add online and remove offline CPUs from the scheduler domains. |
8750 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8751 | */ |
8752 | static int update_sched_domains(struct notifier_block *nfb, | |
8753 | unsigned long action, void *hcpu) | |
e761b772 MK |
8754 | { |
8755 | switch (action) { | |
8756 | case CPU_ONLINE: | |
8757 | case CPU_ONLINE_FROZEN: | |
8758 | case CPU_DEAD: | |
8759 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8760 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8761 | return NOTIFY_OK; |
8762 | ||
8763 | default: | |
8764 | return NOTIFY_DONE; | |
8765 | } | |
8766 | } | |
8767 | #endif | |
8768 | ||
8769 | static int update_runtime(struct notifier_block *nfb, | |
8770 | unsigned long action, void *hcpu) | |
1da177e4 | 8771 | { |
7def2be1 PZ |
8772 | int cpu = (int)(long)hcpu; |
8773 | ||
1da177e4 | 8774 | switch (action) { |
1da177e4 | 8775 | case CPU_DOWN_PREPARE: |
8bb78442 | 8776 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8777 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8778 | return NOTIFY_OK; |
8779 | ||
1da177e4 | 8780 | case CPU_DOWN_FAILED: |
8bb78442 | 8781 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8782 | case CPU_ONLINE: |
8bb78442 | 8783 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8784 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8785 | return NOTIFY_OK; |
8786 | ||
1da177e4 LT |
8787 | default: |
8788 | return NOTIFY_DONE; | |
8789 | } | |
1da177e4 | 8790 | } |
1da177e4 LT |
8791 | |
8792 | void __init sched_init_smp(void) | |
8793 | { | |
dcc30a35 RR |
8794 | cpumask_var_t non_isolated_cpus; |
8795 | ||
8796 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8797 | |
434d53b0 MT |
8798 | #if defined(CONFIG_NUMA) |
8799 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8800 | GFP_KERNEL); | |
8801 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8802 | #endif | |
95402b38 | 8803 | get_online_cpus(); |
712555ee | 8804 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8805 | arch_init_sched_domains(cpu_online_mask); |
8806 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8807 | if (cpumask_empty(non_isolated_cpus)) | |
8808 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8809 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8810 | put_online_cpus(); |
e761b772 MK |
8811 | |
8812 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8813 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8814 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8815 | #endif |
8816 | ||
8817 | /* RT runtime code needs to handle some hotplug events */ | |
8818 | hotcpu_notifier(update_runtime, 0); | |
8819 | ||
b328ca18 | 8820 | init_hrtick(); |
5c1e1767 NP |
8821 | |
8822 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8823 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8824 | BUG(); |
19978ca6 | 8825 | sched_init_granularity(); |
dcc30a35 | 8826 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8827 | |
8828 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8829 | init_sched_rt_class(); |
1da177e4 LT |
8830 | } |
8831 | #else | |
8832 | void __init sched_init_smp(void) | |
8833 | { | |
19978ca6 | 8834 | sched_init_granularity(); |
1da177e4 LT |
8835 | } |
8836 | #endif /* CONFIG_SMP */ | |
8837 | ||
8838 | int in_sched_functions(unsigned long addr) | |
8839 | { | |
1da177e4 LT |
8840 | return in_lock_functions(addr) || |
8841 | (addr >= (unsigned long)__sched_text_start | |
8842 | && addr < (unsigned long)__sched_text_end); | |
8843 | } | |
8844 | ||
a9957449 | 8845 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8846 | { |
8847 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8848 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8849 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8850 | cfs_rq->rq = rq; | |
8851 | #endif | |
67e9fb2a | 8852 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8853 | } |
8854 | ||
fa85ae24 PZ |
8855 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8856 | { | |
8857 | struct rt_prio_array *array; | |
8858 | int i; | |
8859 | ||
8860 | array = &rt_rq->active; | |
8861 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8862 | INIT_LIST_HEAD(array->queue + i); | |
8863 | __clear_bit(i, array->bitmap); | |
8864 | } | |
8865 | /* delimiter for bitsearch: */ | |
8866 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8867 | ||
052f1dc7 | 8868 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8869 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8870 | #ifdef CONFIG_SMP |
e864c499 | 8871 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8872 | #endif |
48d5e258 | 8873 | #endif |
fa85ae24 PZ |
8874 | #ifdef CONFIG_SMP |
8875 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8876 | rt_rq->overloaded = 0; |
917b627d | 8877 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8878 | #endif |
8879 | ||
8880 | rt_rq->rt_time = 0; | |
8881 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8882 | rt_rq->rt_runtime = 0; |
8883 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8884 | |
052f1dc7 | 8885 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8886 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8887 | rt_rq->rq = rq; |
8888 | #endif | |
fa85ae24 PZ |
8889 | } |
8890 | ||
6f505b16 | 8891 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8892 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8893 | struct sched_entity *se, int cpu, int add, | |
8894 | struct sched_entity *parent) | |
6f505b16 | 8895 | { |
ec7dc8ac | 8896 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8897 | tg->cfs_rq[cpu] = cfs_rq; |
8898 | init_cfs_rq(cfs_rq, rq); | |
8899 | cfs_rq->tg = tg; | |
8900 | if (add) | |
8901 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8902 | ||
8903 | tg->se[cpu] = se; | |
354d60c2 DG |
8904 | /* se could be NULL for init_task_group */ |
8905 | if (!se) | |
8906 | return; | |
8907 | ||
ec7dc8ac DG |
8908 | if (!parent) |
8909 | se->cfs_rq = &rq->cfs; | |
8910 | else | |
8911 | se->cfs_rq = parent->my_q; | |
8912 | ||
6f505b16 PZ |
8913 | se->my_q = cfs_rq; |
8914 | se->load.weight = tg->shares; | |
e05510d0 | 8915 | se->load.inv_weight = 0; |
ec7dc8ac | 8916 | se->parent = parent; |
6f505b16 | 8917 | } |
052f1dc7 | 8918 | #endif |
6f505b16 | 8919 | |
052f1dc7 | 8920 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8921 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8922 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8923 | struct sched_rt_entity *parent) | |
6f505b16 | 8924 | { |
ec7dc8ac DG |
8925 | struct rq *rq = cpu_rq(cpu); |
8926 | ||
6f505b16 PZ |
8927 | tg->rt_rq[cpu] = rt_rq; |
8928 | init_rt_rq(rt_rq, rq); | |
8929 | rt_rq->tg = tg; | |
8930 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8931 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8932 | if (add) |
8933 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8934 | ||
8935 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8936 | if (!rt_se) |
8937 | return; | |
8938 | ||
ec7dc8ac DG |
8939 | if (!parent) |
8940 | rt_se->rt_rq = &rq->rt; | |
8941 | else | |
8942 | rt_se->rt_rq = parent->my_q; | |
8943 | ||
6f505b16 | 8944 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8945 | rt_se->parent = parent; |
6f505b16 PZ |
8946 | INIT_LIST_HEAD(&rt_se->run_list); |
8947 | } | |
8948 | #endif | |
8949 | ||
1da177e4 LT |
8950 | void __init sched_init(void) |
8951 | { | |
dd41f596 | 8952 | int i, j; |
434d53b0 MT |
8953 | unsigned long alloc_size = 0, ptr; |
8954 | ||
8955 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8956 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8957 | #endif | |
8958 | #ifdef CONFIG_RT_GROUP_SCHED | |
8959 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8960 | #endif |
8961 | #ifdef CONFIG_USER_SCHED | |
8962 | alloc_size *= 2; | |
df7c8e84 RR |
8963 | #endif |
8964 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 8965 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
8966 | #endif |
8967 | /* | |
8968 | * As sched_init() is called before page_alloc is setup, | |
8969 | * we use alloc_bootmem(). | |
8970 | */ | |
8971 | if (alloc_size) { | |
5a9d3225 | 8972 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8973 | |
8974 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8975 | init_task_group.se = (struct sched_entity **)ptr; | |
8976 | ptr += nr_cpu_ids * sizeof(void **); | |
8977 | ||
8978 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8979 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8980 | |
8981 | #ifdef CONFIG_USER_SCHED | |
8982 | root_task_group.se = (struct sched_entity **)ptr; | |
8983 | ptr += nr_cpu_ids * sizeof(void **); | |
8984 | ||
8985 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8986 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8987 | #endif /* CONFIG_USER_SCHED */ |
8988 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8989 | #ifdef CONFIG_RT_GROUP_SCHED |
8990 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8991 | ptr += nr_cpu_ids * sizeof(void **); | |
8992 | ||
8993 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8994 | ptr += nr_cpu_ids * sizeof(void **); |
8995 | ||
8996 | #ifdef CONFIG_USER_SCHED | |
8997 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8998 | ptr += nr_cpu_ids * sizeof(void **); | |
8999 | ||
9000 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9001 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9002 | #endif /* CONFIG_USER_SCHED */ |
9003 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9004 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9005 | for_each_possible_cpu(i) { | |
9006 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9007 | ptr += cpumask_size(); | |
9008 | } | |
9009 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9010 | } |
dd41f596 | 9011 | |
57d885fe GH |
9012 | #ifdef CONFIG_SMP |
9013 | init_defrootdomain(); | |
9014 | #endif | |
9015 | ||
d0b27fa7 PZ |
9016 | init_rt_bandwidth(&def_rt_bandwidth, |
9017 | global_rt_period(), global_rt_runtime()); | |
9018 | ||
9019 | #ifdef CONFIG_RT_GROUP_SCHED | |
9020 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9021 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9022 | #ifdef CONFIG_USER_SCHED |
9023 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9024 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9025 | #endif /* CONFIG_USER_SCHED */ |
9026 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9027 | |
052f1dc7 | 9028 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9029 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9030 | INIT_LIST_HEAD(&init_task_group.children); |
9031 | ||
9032 | #ifdef CONFIG_USER_SCHED | |
9033 | INIT_LIST_HEAD(&root_task_group.children); | |
9034 | init_task_group.parent = &root_task_group; | |
9035 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9036 | #endif /* CONFIG_USER_SCHED */ |
9037 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9038 | |
0a945022 | 9039 | for_each_possible_cpu(i) { |
70b97a7f | 9040 | struct rq *rq; |
1da177e4 LT |
9041 | |
9042 | rq = cpu_rq(i); | |
9043 | spin_lock_init(&rq->lock); | |
7897986b | 9044 | rq->nr_running = 0; |
dd41f596 | 9045 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9046 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9047 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9048 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9049 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9050 | #ifdef CONFIG_CGROUP_SCHED |
9051 | /* | |
9052 | * How much cpu bandwidth does init_task_group get? | |
9053 | * | |
9054 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9055 | * gets 100% of the cpu resources in the system. This overall | |
9056 | * system cpu resource is divided among the tasks of | |
9057 | * init_task_group and its child task-groups in a fair manner, | |
9058 | * based on each entity's (task or task-group's) weight | |
9059 | * (se->load.weight). | |
9060 | * | |
9061 | * In other words, if init_task_group has 10 tasks of weight | |
9062 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9063 | * then A0's share of the cpu resource is: | |
9064 | * | |
9065 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
9066 | * | |
9067 | * We achieve this by letting init_task_group's tasks sit | |
9068 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9069 | */ | |
ec7dc8ac | 9070 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9071 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9072 | root_task_group.shares = NICE_0_LOAD; |
9073 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9074 | /* |
9075 | * In case of task-groups formed thr' the user id of tasks, | |
9076 | * init_task_group represents tasks belonging to root user. | |
9077 | * Hence it forms a sibling of all subsequent groups formed. | |
9078 | * In this case, init_task_group gets only a fraction of overall | |
9079 | * system cpu resource, based on the weight assigned to root | |
9080 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9081 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9082 | * (init_cfs_rq) and having one entity represent this group of | |
9083 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9084 | */ | |
ec7dc8ac | 9085 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9086 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9087 | &per_cpu(init_sched_entity, i), i, 1, |
9088 | root_task_group.se[i]); | |
6f505b16 | 9089 | |
052f1dc7 | 9090 | #endif |
354d60c2 DG |
9091 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9092 | ||
9093 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9094 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9095 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9096 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9097 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9098 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9099 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9100 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9101 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9102 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9103 | root_task_group.rt_se[i]); | |
354d60c2 | 9104 | #endif |
dd41f596 | 9105 | #endif |
1da177e4 | 9106 | |
dd41f596 IM |
9107 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9108 | rq->cpu_load[j] = 0; | |
1da177e4 | 9109 | #ifdef CONFIG_SMP |
41c7ce9a | 9110 | rq->sd = NULL; |
57d885fe | 9111 | rq->rd = NULL; |
1da177e4 | 9112 | rq->active_balance = 0; |
dd41f596 | 9113 | rq->next_balance = jiffies; |
1da177e4 | 9114 | rq->push_cpu = 0; |
0a2966b4 | 9115 | rq->cpu = i; |
1f11eb6a | 9116 | rq->online = 0; |
1da177e4 LT |
9117 | rq->migration_thread = NULL; |
9118 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9119 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9120 | #endif |
8f4d37ec | 9121 | init_rq_hrtick(rq); |
1da177e4 | 9122 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9123 | } |
9124 | ||
2dd73a4f | 9125 | set_load_weight(&init_task); |
b50f60ce | 9126 | |
e107be36 AK |
9127 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9128 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9129 | #endif | |
9130 | ||
c9819f45 | 9131 | #ifdef CONFIG_SMP |
962cf36c | 9132 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9133 | #endif |
9134 | ||
b50f60ce HC |
9135 | #ifdef CONFIG_RT_MUTEXES |
9136 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9137 | #endif | |
9138 | ||
1da177e4 LT |
9139 | /* |
9140 | * The boot idle thread does lazy MMU switching as well: | |
9141 | */ | |
9142 | atomic_inc(&init_mm.mm_count); | |
9143 | enter_lazy_tlb(&init_mm, current); | |
9144 | ||
9145 | /* | |
9146 | * Make us the idle thread. Technically, schedule() should not be | |
9147 | * called from this thread, however somewhere below it might be, | |
9148 | * but because we are the idle thread, we just pick up running again | |
9149 | * when this runqueue becomes "idle". | |
9150 | */ | |
9151 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
9152 | /* |
9153 | * During early bootup we pretend to be a normal task: | |
9154 | */ | |
9155 | current->sched_class = &fair_sched_class; | |
6892b75e | 9156 | |
6a7b3dc3 RR |
9157 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
9158 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 9159 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
9160 | #ifdef CONFIG_NO_HZ |
9161 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
f711f609 | 9162 | alloc_bootmem_cpumask_var(&nohz.ilb_grp_nohz_mask); |
7d1e6a9b | 9163 | #endif |
dcc30a35 | 9164 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 9165 | #endif /* SMP */ |
6a7b3dc3 | 9166 | |
6892b75e | 9167 | scheduler_running = 1; |
1da177e4 LT |
9168 | } |
9169 | ||
9170 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9171 | void __might_sleep(char *file, int line) | |
9172 | { | |
48f24c4d | 9173 | #ifdef in_atomic |
1da177e4 LT |
9174 | static unsigned long prev_jiffy; /* ratelimiting */ |
9175 | ||
aef745fc IM |
9176 | if ((!in_atomic() && !irqs_disabled()) || |
9177 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9178 | return; | |
9179 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9180 | return; | |
9181 | prev_jiffy = jiffies; | |
9182 | ||
9183 | printk(KERN_ERR | |
9184 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9185 | file, line); | |
9186 | printk(KERN_ERR | |
9187 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9188 | in_atomic(), irqs_disabled(), | |
9189 | current->pid, current->comm); | |
9190 | ||
9191 | debug_show_held_locks(current); | |
9192 | if (irqs_disabled()) | |
9193 | print_irqtrace_events(current); | |
9194 | dump_stack(); | |
1da177e4 LT |
9195 | #endif |
9196 | } | |
9197 | EXPORT_SYMBOL(__might_sleep); | |
9198 | #endif | |
9199 | ||
9200 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9201 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9202 | { | |
9203 | int on_rq; | |
3e51f33f | 9204 | |
3a5e4dc1 AK |
9205 | update_rq_clock(rq); |
9206 | on_rq = p->se.on_rq; | |
9207 | if (on_rq) | |
9208 | deactivate_task(rq, p, 0); | |
9209 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9210 | if (on_rq) { | |
9211 | activate_task(rq, p, 0); | |
9212 | resched_task(rq->curr); | |
9213 | } | |
9214 | } | |
9215 | ||
1da177e4 LT |
9216 | void normalize_rt_tasks(void) |
9217 | { | |
a0f98a1c | 9218 | struct task_struct *g, *p; |
1da177e4 | 9219 | unsigned long flags; |
70b97a7f | 9220 | struct rq *rq; |
1da177e4 | 9221 | |
4cf5d77a | 9222 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9223 | do_each_thread(g, p) { |
178be793 IM |
9224 | /* |
9225 | * Only normalize user tasks: | |
9226 | */ | |
9227 | if (!p->mm) | |
9228 | continue; | |
9229 | ||
6cfb0d5d | 9230 | p->se.exec_start = 0; |
6cfb0d5d | 9231 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9232 | p->se.wait_start = 0; |
dd41f596 | 9233 | p->se.sleep_start = 0; |
dd41f596 | 9234 | p->se.block_start = 0; |
6cfb0d5d | 9235 | #endif |
dd41f596 IM |
9236 | |
9237 | if (!rt_task(p)) { | |
9238 | /* | |
9239 | * Renice negative nice level userspace | |
9240 | * tasks back to 0: | |
9241 | */ | |
9242 | if (TASK_NICE(p) < 0 && p->mm) | |
9243 | set_user_nice(p, 0); | |
1da177e4 | 9244 | continue; |
dd41f596 | 9245 | } |
1da177e4 | 9246 | |
4cf5d77a | 9247 | spin_lock(&p->pi_lock); |
b29739f9 | 9248 | rq = __task_rq_lock(p); |
1da177e4 | 9249 | |
178be793 | 9250 | normalize_task(rq, p); |
3a5e4dc1 | 9251 | |
b29739f9 | 9252 | __task_rq_unlock(rq); |
4cf5d77a | 9253 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9254 | } while_each_thread(g, p); |
9255 | ||
4cf5d77a | 9256 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9257 | } |
9258 | ||
9259 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9260 | |
9261 | #ifdef CONFIG_IA64 | |
9262 | /* | |
9263 | * These functions are only useful for the IA64 MCA handling. | |
9264 | * | |
9265 | * They can only be called when the whole system has been | |
9266 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9267 | * activity can take place. Using them for anything else would | |
9268 | * be a serious bug, and as a result, they aren't even visible | |
9269 | * under any other configuration. | |
9270 | */ | |
9271 | ||
9272 | /** | |
9273 | * curr_task - return the current task for a given cpu. | |
9274 | * @cpu: the processor in question. | |
9275 | * | |
9276 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9277 | */ | |
36c8b586 | 9278 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9279 | { |
9280 | return cpu_curr(cpu); | |
9281 | } | |
9282 | ||
9283 | /** | |
9284 | * set_curr_task - set the current task for a given cpu. | |
9285 | * @cpu: the processor in question. | |
9286 | * @p: the task pointer to set. | |
9287 | * | |
9288 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9289 | * are serviced on a separate stack. It allows the architecture to switch the |
9290 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9291 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9292 | * and caller must save the original value of the current task (see | |
9293 | * curr_task() above) and restore that value before reenabling interrupts and | |
9294 | * re-starting the system. | |
9295 | * | |
9296 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9297 | */ | |
36c8b586 | 9298 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9299 | { |
9300 | cpu_curr(cpu) = p; | |
9301 | } | |
9302 | ||
9303 | #endif | |
29f59db3 | 9304 | |
bccbe08a PZ |
9305 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9306 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9307 | { |
9308 | int i; | |
9309 | ||
9310 | for_each_possible_cpu(i) { | |
9311 | if (tg->cfs_rq) | |
9312 | kfree(tg->cfs_rq[i]); | |
9313 | if (tg->se) | |
9314 | kfree(tg->se[i]); | |
6f505b16 PZ |
9315 | } |
9316 | ||
9317 | kfree(tg->cfs_rq); | |
9318 | kfree(tg->se); | |
6f505b16 PZ |
9319 | } |
9320 | ||
ec7dc8ac DG |
9321 | static |
9322 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9323 | { |
29f59db3 | 9324 | struct cfs_rq *cfs_rq; |
eab17229 | 9325 | struct sched_entity *se; |
9b5b7751 | 9326 | struct rq *rq; |
29f59db3 SV |
9327 | int i; |
9328 | ||
434d53b0 | 9329 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9330 | if (!tg->cfs_rq) |
9331 | goto err; | |
434d53b0 | 9332 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9333 | if (!tg->se) |
9334 | goto err; | |
052f1dc7 PZ |
9335 | |
9336 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9337 | |
9338 | for_each_possible_cpu(i) { | |
9b5b7751 | 9339 | rq = cpu_rq(i); |
29f59db3 | 9340 | |
eab17229 LZ |
9341 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9342 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9343 | if (!cfs_rq) |
9344 | goto err; | |
9345 | ||
eab17229 LZ |
9346 | se = kzalloc_node(sizeof(struct sched_entity), |
9347 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9348 | if (!se) |
9349 | goto err; | |
9350 | ||
eab17229 | 9351 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9352 | } |
9353 | ||
9354 | return 1; | |
9355 | ||
9356 | err: | |
9357 | return 0; | |
9358 | } | |
9359 | ||
9360 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9361 | { | |
9362 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9363 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9364 | } | |
9365 | ||
9366 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9367 | { | |
9368 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9369 | } | |
6d6bc0ad | 9370 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9371 | static inline void free_fair_sched_group(struct task_group *tg) |
9372 | { | |
9373 | } | |
9374 | ||
ec7dc8ac DG |
9375 | static inline |
9376 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9377 | { |
9378 | return 1; | |
9379 | } | |
9380 | ||
9381 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9382 | { | |
9383 | } | |
9384 | ||
9385 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9386 | { | |
9387 | } | |
6d6bc0ad | 9388 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9389 | |
9390 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9391 | static void free_rt_sched_group(struct task_group *tg) |
9392 | { | |
9393 | int i; | |
9394 | ||
d0b27fa7 PZ |
9395 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9396 | ||
bccbe08a PZ |
9397 | for_each_possible_cpu(i) { |
9398 | if (tg->rt_rq) | |
9399 | kfree(tg->rt_rq[i]); | |
9400 | if (tg->rt_se) | |
9401 | kfree(tg->rt_se[i]); | |
9402 | } | |
9403 | ||
9404 | kfree(tg->rt_rq); | |
9405 | kfree(tg->rt_se); | |
9406 | } | |
9407 | ||
ec7dc8ac DG |
9408 | static |
9409 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9410 | { |
9411 | struct rt_rq *rt_rq; | |
eab17229 | 9412 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9413 | struct rq *rq; |
9414 | int i; | |
9415 | ||
434d53b0 | 9416 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9417 | if (!tg->rt_rq) |
9418 | goto err; | |
434d53b0 | 9419 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9420 | if (!tg->rt_se) |
9421 | goto err; | |
9422 | ||
d0b27fa7 PZ |
9423 | init_rt_bandwidth(&tg->rt_bandwidth, |
9424 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9425 | |
9426 | for_each_possible_cpu(i) { | |
9427 | rq = cpu_rq(i); | |
9428 | ||
eab17229 LZ |
9429 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9430 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9431 | if (!rt_rq) |
9432 | goto err; | |
29f59db3 | 9433 | |
eab17229 LZ |
9434 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9435 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9436 | if (!rt_se) |
9437 | goto err; | |
29f59db3 | 9438 | |
eab17229 | 9439 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9440 | } |
9441 | ||
bccbe08a PZ |
9442 | return 1; |
9443 | ||
9444 | err: | |
9445 | return 0; | |
9446 | } | |
9447 | ||
9448 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9449 | { | |
9450 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9451 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9452 | } | |
9453 | ||
9454 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9455 | { | |
9456 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9457 | } | |
6d6bc0ad | 9458 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9459 | static inline void free_rt_sched_group(struct task_group *tg) |
9460 | { | |
9461 | } | |
9462 | ||
ec7dc8ac DG |
9463 | static inline |
9464 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9465 | { |
9466 | return 1; | |
9467 | } | |
9468 | ||
9469 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9470 | { | |
9471 | } | |
9472 | ||
9473 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9474 | { | |
9475 | } | |
6d6bc0ad | 9476 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9477 | |
d0b27fa7 | 9478 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9479 | static void free_sched_group(struct task_group *tg) |
9480 | { | |
9481 | free_fair_sched_group(tg); | |
9482 | free_rt_sched_group(tg); | |
9483 | kfree(tg); | |
9484 | } | |
9485 | ||
9486 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9487 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9488 | { |
9489 | struct task_group *tg; | |
9490 | unsigned long flags; | |
9491 | int i; | |
9492 | ||
9493 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9494 | if (!tg) | |
9495 | return ERR_PTR(-ENOMEM); | |
9496 | ||
ec7dc8ac | 9497 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9498 | goto err; |
9499 | ||
ec7dc8ac | 9500 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9501 | goto err; |
9502 | ||
8ed36996 | 9503 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9504 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9505 | register_fair_sched_group(tg, i); |
9506 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9507 | } |
6f505b16 | 9508 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9509 | |
9510 | WARN_ON(!parent); /* root should already exist */ | |
9511 | ||
9512 | tg->parent = parent; | |
f473aa5e | 9513 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9514 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9515 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9516 | |
9b5b7751 | 9517 | return tg; |
29f59db3 SV |
9518 | |
9519 | err: | |
6f505b16 | 9520 | free_sched_group(tg); |
29f59db3 SV |
9521 | return ERR_PTR(-ENOMEM); |
9522 | } | |
9523 | ||
9b5b7751 | 9524 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9525 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9526 | { |
29f59db3 | 9527 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9528 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9529 | } |
9530 | ||
9b5b7751 | 9531 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9532 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9533 | { |
8ed36996 | 9534 | unsigned long flags; |
9b5b7751 | 9535 | int i; |
29f59db3 | 9536 | |
8ed36996 | 9537 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9538 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9539 | unregister_fair_sched_group(tg, i); |
9540 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9541 | } |
6f505b16 | 9542 | list_del_rcu(&tg->list); |
f473aa5e | 9543 | list_del_rcu(&tg->siblings); |
8ed36996 | 9544 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9545 | |
9b5b7751 | 9546 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9547 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9548 | } |
9549 | ||
9b5b7751 | 9550 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9551 | * The caller of this function should have put the task in its new group |
9552 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9553 | * reflect its new group. | |
9b5b7751 SV |
9554 | */ |
9555 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9556 | { |
9557 | int on_rq, running; | |
9558 | unsigned long flags; | |
9559 | struct rq *rq; | |
9560 | ||
9561 | rq = task_rq_lock(tsk, &flags); | |
9562 | ||
29f59db3 SV |
9563 | update_rq_clock(rq); |
9564 | ||
051a1d1a | 9565 | running = task_current(rq, tsk); |
29f59db3 SV |
9566 | on_rq = tsk->se.on_rq; |
9567 | ||
0e1f3483 | 9568 | if (on_rq) |
29f59db3 | 9569 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9570 | if (unlikely(running)) |
9571 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9572 | |
6f505b16 | 9573 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9574 | |
810b3817 PZ |
9575 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9576 | if (tsk->sched_class->moved_group) | |
9577 | tsk->sched_class->moved_group(tsk); | |
9578 | #endif | |
9579 | ||
0e1f3483 HS |
9580 | if (unlikely(running)) |
9581 | tsk->sched_class->set_curr_task(rq); | |
9582 | if (on_rq) | |
7074badb | 9583 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9584 | |
29f59db3 SV |
9585 | task_rq_unlock(rq, &flags); |
9586 | } | |
6d6bc0ad | 9587 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9588 | |
052f1dc7 | 9589 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9590 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9591 | { |
9592 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9593 | int on_rq; |
9594 | ||
29f59db3 | 9595 | on_rq = se->on_rq; |
62fb1851 | 9596 | if (on_rq) |
29f59db3 SV |
9597 | dequeue_entity(cfs_rq, se, 0); |
9598 | ||
9599 | se->load.weight = shares; | |
e05510d0 | 9600 | se->load.inv_weight = 0; |
29f59db3 | 9601 | |
62fb1851 | 9602 | if (on_rq) |
29f59db3 | 9603 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9604 | } |
62fb1851 | 9605 | |
c09595f6 PZ |
9606 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9607 | { | |
9608 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9609 | struct rq *rq = cfs_rq->rq; | |
9610 | unsigned long flags; | |
9611 | ||
9612 | spin_lock_irqsave(&rq->lock, flags); | |
9613 | __set_se_shares(se, shares); | |
9614 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9615 | } |
9616 | ||
8ed36996 PZ |
9617 | static DEFINE_MUTEX(shares_mutex); |
9618 | ||
4cf86d77 | 9619 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9620 | { |
9621 | int i; | |
8ed36996 | 9622 | unsigned long flags; |
c61935fd | 9623 | |
ec7dc8ac DG |
9624 | /* |
9625 | * We can't change the weight of the root cgroup. | |
9626 | */ | |
9627 | if (!tg->se[0]) | |
9628 | return -EINVAL; | |
9629 | ||
18d95a28 PZ |
9630 | if (shares < MIN_SHARES) |
9631 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9632 | else if (shares > MAX_SHARES) |
9633 | shares = MAX_SHARES; | |
62fb1851 | 9634 | |
8ed36996 | 9635 | mutex_lock(&shares_mutex); |
9b5b7751 | 9636 | if (tg->shares == shares) |
5cb350ba | 9637 | goto done; |
29f59db3 | 9638 | |
8ed36996 | 9639 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9640 | for_each_possible_cpu(i) |
9641 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9642 | list_del_rcu(&tg->siblings); |
8ed36996 | 9643 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9644 | |
9645 | /* wait for any ongoing reference to this group to finish */ | |
9646 | synchronize_sched(); | |
9647 | ||
9648 | /* | |
9649 | * Now we are free to modify the group's share on each cpu | |
9650 | * w/o tripping rebalance_share or load_balance_fair. | |
9651 | */ | |
9b5b7751 | 9652 | tg->shares = shares; |
c09595f6 PZ |
9653 | for_each_possible_cpu(i) { |
9654 | /* | |
9655 | * force a rebalance | |
9656 | */ | |
9657 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9658 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9659 | } |
29f59db3 | 9660 | |
6b2d7700 SV |
9661 | /* |
9662 | * Enable load balance activity on this group, by inserting it back on | |
9663 | * each cpu's rq->leaf_cfs_rq_list. | |
9664 | */ | |
8ed36996 | 9665 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9666 | for_each_possible_cpu(i) |
9667 | register_fair_sched_group(tg, i); | |
f473aa5e | 9668 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9669 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9670 | done: |
8ed36996 | 9671 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9672 | return 0; |
29f59db3 SV |
9673 | } |
9674 | ||
5cb350ba DG |
9675 | unsigned long sched_group_shares(struct task_group *tg) |
9676 | { | |
9677 | return tg->shares; | |
9678 | } | |
052f1dc7 | 9679 | #endif |
5cb350ba | 9680 | |
052f1dc7 | 9681 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9682 | /* |
9f0c1e56 | 9683 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9684 | */ |
9f0c1e56 PZ |
9685 | static DEFINE_MUTEX(rt_constraints_mutex); |
9686 | ||
9687 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9688 | { | |
9689 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9690 | return 1ULL << 20; |
9f0c1e56 | 9691 | |
9a7e0b18 | 9692 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9693 | } |
9694 | ||
9a7e0b18 PZ |
9695 | /* Must be called with tasklist_lock held */ |
9696 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9697 | { |
9a7e0b18 | 9698 | struct task_struct *g, *p; |
b40b2e8e | 9699 | |
9a7e0b18 PZ |
9700 | do_each_thread(g, p) { |
9701 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9702 | return 1; | |
9703 | } while_each_thread(g, p); | |
b40b2e8e | 9704 | |
9a7e0b18 PZ |
9705 | return 0; |
9706 | } | |
b40b2e8e | 9707 | |
9a7e0b18 PZ |
9708 | struct rt_schedulable_data { |
9709 | struct task_group *tg; | |
9710 | u64 rt_period; | |
9711 | u64 rt_runtime; | |
9712 | }; | |
b40b2e8e | 9713 | |
9a7e0b18 PZ |
9714 | static int tg_schedulable(struct task_group *tg, void *data) |
9715 | { | |
9716 | struct rt_schedulable_data *d = data; | |
9717 | struct task_group *child; | |
9718 | unsigned long total, sum = 0; | |
9719 | u64 period, runtime; | |
b40b2e8e | 9720 | |
9a7e0b18 PZ |
9721 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9722 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9723 | |
9a7e0b18 PZ |
9724 | if (tg == d->tg) { |
9725 | period = d->rt_period; | |
9726 | runtime = d->rt_runtime; | |
b40b2e8e | 9727 | } |
b40b2e8e | 9728 | |
98a4826b PZ |
9729 | #ifdef CONFIG_USER_SCHED |
9730 | if (tg == &root_task_group) { | |
9731 | period = global_rt_period(); | |
9732 | runtime = global_rt_runtime(); | |
9733 | } | |
9734 | #endif | |
9735 | ||
4653f803 PZ |
9736 | /* |
9737 | * Cannot have more runtime than the period. | |
9738 | */ | |
9739 | if (runtime > period && runtime != RUNTIME_INF) | |
9740 | return -EINVAL; | |
6f505b16 | 9741 | |
4653f803 PZ |
9742 | /* |
9743 | * Ensure we don't starve existing RT tasks. | |
9744 | */ | |
9a7e0b18 PZ |
9745 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9746 | return -EBUSY; | |
6f505b16 | 9747 | |
9a7e0b18 | 9748 | total = to_ratio(period, runtime); |
6f505b16 | 9749 | |
4653f803 PZ |
9750 | /* |
9751 | * Nobody can have more than the global setting allows. | |
9752 | */ | |
9753 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9754 | return -EINVAL; | |
6f505b16 | 9755 | |
4653f803 PZ |
9756 | /* |
9757 | * The sum of our children's runtime should not exceed our own. | |
9758 | */ | |
9a7e0b18 PZ |
9759 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9760 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9761 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9762 | |
9a7e0b18 PZ |
9763 | if (child == d->tg) { |
9764 | period = d->rt_period; | |
9765 | runtime = d->rt_runtime; | |
9766 | } | |
6f505b16 | 9767 | |
9a7e0b18 | 9768 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9769 | } |
6f505b16 | 9770 | |
9a7e0b18 PZ |
9771 | if (sum > total) |
9772 | return -EINVAL; | |
9773 | ||
9774 | return 0; | |
6f505b16 PZ |
9775 | } |
9776 | ||
9a7e0b18 | 9777 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9778 | { |
9a7e0b18 PZ |
9779 | struct rt_schedulable_data data = { |
9780 | .tg = tg, | |
9781 | .rt_period = period, | |
9782 | .rt_runtime = runtime, | |
9783 | }; | |
9784 | ||
9785 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9786 | } |
9787 | ||
d0b27fa7 PZ |
9788 | static int tg_set_bandwidth(struct task_group *tg, |
9789 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9790 | { |
ac086bc2 | 9791 | int i, err = 0; |
9f0c1e56 | 9792 | |
9f0c1e56 | 9793 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9794 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9795 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9796 | if (err) | |
9f0c1e56 | 9797 | goto unlock; |
ac086bc2 PZ |
9798 | |
9799 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9800 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9801 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9802 | |
9803 | for_each_possible_cpu(i) { | |
9804 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9805 | ||
9806 | spin_lock(&rt_rq->rt_runtime_lock); | |
9807 | rt_rq->rt_runtime = rt_runtime; | |
9808 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9809 | } | |
9810 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9811 | unlock: |
521f1a24 | 9812 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9813 | mutex_unlock(&rt_constraints_mutex); |
9814 | ||
9815 | return err; | |
6f505b16 PZ |
9816 | } |
9817 | ||
d0b27fa7 PZ |
9818 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9819 | { | |
9820 | u64 rt_runtime, rt_period; | |
9821 | ||
9822 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9823 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9824 | if (rt_runtime_us < 0) | |
9825 | rt_runtime = RUNTIME_INF; | |
9826 | ||
9827 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9828 | } | |
9829 | ||
9f0c1e56 PZ |
9830 | long sched_group_rt_runtime(struct task_group *tg) |
9831 | { | |
9832 | u64 rt_runtime_us; | |
9833 | ||
d0b27fa7 | 9834 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9835 | return -1; |
9836 | ||
d0b27fa7 | 9837 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9838 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9839 | return rt_runtime_us; | |
9840 | } | |
d0b27fa7 PZ |
9841 | |
9842 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9843 | { | |
9844 | u64 rt_runtime, rt_period; | |
9845 | ||
9846 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9847 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9848 | ||
619b0488 R |
9849 | if (rt_period == 0) |
9850 | return -EINVAL; | |
9851 | ||
d0b27fa7 PZ |
9852 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9853 | } | |
9854 | ||
9855 | long sched_group_rt_period(struct task_group *tg) | |
9856 | { | |
9857 | u64 rt_period_us; | |
9858 | ||
9859 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9860 | do_div(rt_period_us, NSEC_PER_USEC); | |
9861 | return rt_period_us; | |
9862 | } | |
9863 | ||
9864 | static int sched_rt_global_constraints(void) | |
9865 | { | |
4653f803 | 9866 | u64 runtime, period; |
d0b27fa7 PZ |
9867 | int ret = 0; |
9868 | ||
ec5d4989 HS |
9869 | if (sysctl_sched_rt_period <= 0) |
9870 | return -EINVAL; | |
9871 | ||
4653f803 PZ |
9872 | runtime = global_rt_runtime(); |
9873 | period = global_rt_period(); | |
9874 | ||
9875 | /* | |
9876 | * Sanity check on the sysctl variables. | |
9877 | */ | |
9878 | if (runtime > period && runtime != RUNTIME_INF) | |
9879 | return -EINVAL; | |
10b612f4 | 9880 | |
d0b27fa7 | 9881 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9882 | read_lock(&tasklist_lock); |
4653f803 | 9883 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9884 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9885 | mutex_unlock(&rt_constraints_mutex); |
9886 | ||
9887 | return ret; | |
9888 | } | |
54e99124 DG |
9889 | |
9890 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9891 | { | |
9892 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9893 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9894 | return 0; | |
9895 | ||
9896 | return 1; | |
9897 | } | |
9898 | ||
6d6bc0ad | 9899 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9900 | static int sched_rt_global_constraints(void) |
9901 | { | |
ac086bc2 PZ |
9902 | unsigned long flags; |
9903 | int i; | |
9904 | ||
ec5d4989 HS |
9905 | if (sysctl_sched_rt_period <= 0) |
9906 | return -EINVAL; | |
9907 | ||
ac086bc2 PZ |
9908 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9909 | for_each_possible_cpu(i) { | |
9910 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9911 | ||
9912 | spin_lock(&rt_rq->rt_runtime_lock); | |
9913 | rt_rq->rt_runtime = global_rt_runtime(); | |
9914 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9915 | } | |
9916 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9917 | ||
d0b27fa7 PZ |
9918 | return 0; |
9919 | } | |
6d6bc0ad | 9920 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9921 | |
9922 | int sched_rt_handler(struct ctl_table *table, int write, | |
9923 | struct file *filp, void __user *buffer, size_t *lenp, | |
9924 | loff_t *ppos) | |
9925 | { | |
9926 | int ret; | |
9927 | int old_period, old_runtime; | |
9928 | static DEFINE_MUTEX(mutex); | |
9929 | ||
9930 | mutex_lock(&mutex); | |
9931 | old_period = sysctl_sched_rt_period; | |
9932 | old_runtime = sysctl_sched_rt_runtime; | |
9933 | ||
9934 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9935 | ||
9936 | if (!ret && write) { | |
9937 | ret = sched_rt_global_constraints(); | |
9938 | if (ret) { | |
9939 | sysctl_sched_rt_period = old_period; | |
9940 | sysctl_sched_rt_runtime = old_runtime; | |
9941 | } else { | |
9942 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9943 | def_rt_bandwidth.rt_period = | |
9944 | ns_to_ktime(global_rt_period()); | |
9945 | } | |
9946 | } | |
9947 | mutex_unlock(&mutex); | |
9948 | ||
9949 | return ret; | |
9950 | } | |
68318b8e | 9951 | |
052f1dc7 | 9952 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9953 | |
9954 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9955 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9956 | { |
2b01dfe3 PM |
9957 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9958 | struct task_group, css); | |
68318b8e SV |
9959 | } |
9960 | ||
9961 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9962 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9963 | { |
ec7dc8ac | 9964 | struct task_group *tg, *parent; |
68318b8e | 9965 | |
2b01dfe3 | 9966 | if (!cgrp->parent) { |
68318b8e | 9967 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9968 | return &init_task_group.css; |
9969 | } | |
9970 | ||
ec7dc8ac DG |
9971 | parent = cgroup_tg(cgrp->parent); |
9972 | tg = sched_create_group(parent); | |
68318b8e SV |
9973 | if (IS_ERR(tg)) |
9974 | return ERR_PTR(-ENOMEM); | |
9975 | ||
68318b8e SV |
9976 | return &tg->css; |
9977 | } | |
9978 | ||
41a2d6cf IM |
9979 | static void |
9980 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9981 | { |
2b01dfe3 | 9982 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9983 | |
9984 | sched_destroy_group(tg); | |
9985 | } | |
9986 | ||
41a2d6cf IM |
9987 | static int |
9988 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9989 | struct task_struct *tsk) | |
68318b8e | 9990 | { |
b68aa230 | 9991 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9992 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9993 | return -EINVAL; |
9994 | #else | |
68318b8e SV |
9995 | /* We don't support RT-tasks being in separate groups */ |
9996 | if (tsk->sched_class != &fair_sched_class) | |
9997 | return -EINVAL; | |
b68aa230 | 9998 | #endif |
68318b8e SV |
9999 | |
10000 | return 0; | |
10001 | } | |
10002 | ||
10003 | static void | |
2b01dfe3 | 10004 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10005 | struct cgroup *old_cont, struct task_struct *tsk) |
10006 | { | |
10007 | sched_move_task(tsk); | |
10008 | } | |
10009 | ||
052f1dc7 | 10010 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10011 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10012 | u64 shareval) |
68318b8e | 10013 | { |
2b01dfe3 | 10014 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10015 | } |
10016 | ||
f4c753b7 | 10017 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10018 | { |
2b01dfe3 | 10019 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10020 | |
10021 | return (u64) tg->shares; | |
10022 | } | |
6d6bc0ad | 10023 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10024 | |
052f1dc7 | 10025 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10026 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10027 | s64 val) |
6f505b16 | 10028 | { |
06ecb27c | 10029 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10030 | } |
10031 | ||
06ecb27c | 10032 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10033 | { |
06ecb27c | 10034 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10035 | } |
d0b27fa7 PZ |
10036 | |
10037 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10038 | u64 rt_period_us) | |
10039 | { | |
10040 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10041 | } | |
10042 | ||
10043 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10044 | { | |
10045 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10046 | } | |
6d6bc0ad | 10047 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10048 | |
fe5c7cc2 | 10049 | static struct cftype cpu_files[] = { |
052f1dc7 | 10050 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10051 | { |
10052 | .name = "shares", | |
f4c753b7 PM |
10053 | .read_u64 = cpu_shares_read_u64, |
10054 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10055 | }, |
052f1dc7 PZ |
10056 | #endif |
10057 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10058 | { |
9f0c1e56 | 10059 | .name = "rt_runtime_us", |
06ecb27c PM |
10060 | .read_s64 = cpu_rt_runtime_read, |
10061 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10062 | }, |
d0b27fa7 PZ |
10063 | { |
10064 | .name = "rt_period_us", | |
f4c753b7 PM |
10065 | .read_u64 = cpu_rt_period_read_uint, |
10066 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10067 | }, |
052f1dc7 | 10068 | #endif |
68318b8e SV |
10069 | }; |
10070 | ||
10071 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10072 | { | |
fe5c7cc2 | 10073 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10074 | } |
10075 | ||
10076 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10077 | .name = "cpu", |
10078 | .create = cpu_cgroup_create, | |
10079 | .destroy = cpu_cgroup_destroy, | |
10080 | .can_attach = cpu_cgroup_can_attach, | |
10081 | .attach = cpu_cgroup_attach, | |
10082 | .populate = cpu_cgroup_populate, | |
10083 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10084 | .early_init = 1, |
10085 | }; | |
10086 | ||
052f1dc7 | 10087 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10088 | |
10089 | #ifdef CONFIG_CGROUP_CPUACCT | |
10090 | ||
10091 | /* | |
10092 | * CPU accounting code for task groups. | |
10093 | * | |
10094 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10095 | * (balbir@in.ibm.com). | |
10096 | */ | |
10097 | ||
934352f2 | 10098 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10099 | struct cpuacct { |
10100 | struct cgroup_subsys_state css; | |
10101 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10102 | u64 *cpuusage; | |
ef12fefa | 10103 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10104 | struct cpuacct *parent; |
d842de87 SV |
10105 | }; |
10106 | ||
10107 | struct cgroup_subsys cpuacct_subsys; | |
10108 | ||
10109 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10110 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10111 | { |
32cd756a | 10112 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10113 | struct cpuacct, css); |
10114 | } | |
10115 | ||
10116 | /* return cpu accounting group to which this task belongs */ | |
10117 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10118 | { | |
10119 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10120 | struct cpuacct, css); | |
10121 | } | |
10122 | ||
10123 | /* create a new cpu accounting group */ | |
10124 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10125 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10126 | { |
10127 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10128 | int i; |
d842de87 SV |
10129 | |
10130 | if (!ca) | |
ef12fefa | 10131 | goto out; |
d842de87 SV |
10132 | |
10133 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10134 | if (!ca->cpuusage) |
10135 | goto out_free_ca; | |
10136 | ||
10137 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10138 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10139 | goto out_free_counters; | |
d842de87 | 10140 | |
934352f2 BR |
10141 | if (cgrp->parent) |
10142 | ca->parent = cgroup_ca(cgrp->parent); | |
10143 | ||
d842de87 | 10144 | return &ca->css; |
ef12fefa BR |
10145 | |
10146 | out_free_counters: | |
10147 | while (--i >= 0) | |
10148 | percpu_counter_destroy(&ca->cpustat[i]); | |
10149 | free_percpu(ca->cpuusage); | |
10150 | out_free_ca: | |
10151 | kfree(ca); | |
10152 | out: | |
10153 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10154 | } |
10155 | ||
10156 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10157 | static void |
32cd756a | 10158 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10159 | { |
32cd756a | 10160 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10161 | int i; |
d842de87 | 10162 | |
ef12fefa BR |
10163 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10164 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10165 | free_percpu(ca->cpuusage); |
10166 | kfree(ca); | |
10167 | } | |
10168 | ||
720f5498 KC |
10169 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10170 | { | |
b36128c8 | 10171 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10172 | u64 data; |
10173 | ||
10174 | #ifndef CONFIG_64BIT | |
10175 | /* | |
10176 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10177 | */ | |
10178 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10179 | data = *cpuusage; | |
10180 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10181 | #else | |
10182 | data = *cpuusage; | |
10183 | #endif | |
10184 | ||
10185 | return data; | |
10186 | } | |
10187 | ||
10188 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10189 | { | |
b36128c8 | 10190 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10191 | |
10192 | #ifndef CONFIG_64BIT | |
10193 | /* | |
10194 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10195 | */ | |
10196 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10197 | *cpuusage = val; | |
10198 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10199 | #else | |
10200 | *cpuusage = val; | |
10201 | #endif | |
10202 | } | |
10203 | ||
d842de87 | 10204 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10205 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10206 | { |
32cd756a | 10207 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10208 | u64 totalcpuusage = 0; |
10209 | int i; | |
10210 | ||
720f5498 KC |
10211 | for_each_present_cpu(i) |
10212 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10213 | |
10214 | return totalcpuusage; | |
10215 | } | |
10216 | ||
0297b803 DG |
10217 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10218 | u64 reset) | |
10219 | { | |
10220 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10221 | int err = 0; | |
10222 | int i; | |
10223 | ||
10224 | if (reset) { | |
10225 | err = -EINVAL; | |
10226 | goto out; | |
10227 | } | |
10228 | ||
720f5498 KC |
10229 | for_each_present_cpu(i) |
10230 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10231 | |
0297b803 DG |
10232 | out: |
10233 | return err; | |
10234 | } | |
10235 | ||
e9515c3c KC |
10236 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10237 | struct seq_file *m) | |
10238 | { | |
10239 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10240 | u64 percpu; | |
10241 | int i; | |
10242 | ||
10243 | for_each_present_cpu(i) { | |
10244 | percpu = cpuacct_cpuusage_read(ca, i); | |
10245 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10246 | } | |
10247 | seq_printf(m, "\n"); | |
10248 | return 0; | |
10249 | } | |
10250 | ||
ef12fefa BR |
10251 | static const char *cpuacct_stat_desc[] = { |
10252 | [CPUACCT_STAT_USER] = "user", | |
10253 | [CPUACCT_STAT_SYSTEM] = "system", | |
10254 | }; | |
10255 | ||
10256 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10257 | struct cgroup_map_cb *cb) | |
10258 | { | |
10259 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10260 | int i; | |
10261 | ||
10262 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10263 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10264 | val = cputime64_to_clock_t(val); | |
10265 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10266 | } | |
10267 | return 0; | |
10268 | } | |
10269 | ||
d842de87 SV |
10270 | static struct cftype files[] = { |
10271 | { | |
10272 | .name = "usage", | |
f4c753b7 PM |
10273 | .read_u64 = cpuusage_read, |
10274 | .write_u64 = cpuusage_write, | |
d842de87 | 10275 | }, |
e9515c3c KC |
10276 | { |
10277 | .name = "usage_percpu", | |
10278 | .read_seq_string = cpuacct_percpu_seq_read, | |
10279 | }, | |
ef12fefa BR |
10280 | { |
10281 | .name = "stat", | |
10282 | .read_map = cpuacct_stats_show, | |
10283 | }, | |
d842de87 SV |
10284 | }; |
10285 | ||
32cd756a | 10286 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10287 | { |
32cd756a | 10288 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10289 | } |
10290 | ||
10291 | /* | |
10292 | * charge this task's execution time to its accounting group. | |
10293 | * | |
10294 | * called with rq->lock held. | |
10295 | */ | |
10296 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10297 | { | |
10298 | struct cpuacct *ca; | |
934352f2 | 10299 | int cpu; |
d842de87 | 10300 | |
c40c6f85 | 10301 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10302 | return; |
10303 | ||
934352f2 | 10304 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10305 | |
10306 | rcu_read_lock(); | |
10307 | ||
d842de87 | 10308 | ca = task_ca(tsk); |
d842de87 | 10309 | |
934352f2 | 10310 | for (; ca; ca = ca->parent) { |
b36128c8 | 10311 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10312 | *cpuusage += cputime; |
10313 | } | |
a18b83b7 BR |
10314 | |
10315 | rcu_read_unlock(); | |
d842de87 SV |
10316 | } |
10317 | ||
ef12fefa BR |
10318 | /* |
10319 | * Charge the system/user time to the task's accounting group. | |
10320 | */ | |
10321 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10322 | enum cpuacct_stat_index idx, cputime_t val) | |
10323 | { | |
10324 | struct cpuacct *ca; | |
10325 | ||
10326 | if (unlikely(!cpuacct_subsys.active)) | |
10327 | return; | |
10328 | ||
10329 | rcu_read_lock(); | |
10330 | ca = task_ca(tsk); | |
10331 | ||
10332 | do { | |
10333 | percpu_counter_add(&ca->cpustat[idx], val); | |
10334 | ca = ca->parent; | |
10335 | } while (ca); | |
10336 | rcu_read_unlock(); | |
10337 | } | |
10338 | ||
d842de87 SV |
10339 | struct cgroup_subsys cpuacct_subsys = { |
10340 | .name = "cpuacct", | |
10341 | .create = cpuacct_create, | |
10342 | .destroy = cpuacct_destroy, | |
10343 | .populate = cpuacct_populate, | |
10344 | .subsys_id = cpuacct_subsys_id, | |
10345 | }; | |
10346 | #endif /* CONFIG_CGROUP_CPUACCT */ |