Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
0d905bca | 42 | #include <linux/perf_counter.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
5517d86b | 67 | #include <linux/reciprocal_div.h> |
dff06c15 | 68 | #include <linux/unistd.h> |
f5ff8422 | 69 | #include <linux/pagemap.h> |
8f4d37ec | 70 | #include <linux/hrtimer.h> |
30914a58 | 71 | #include <linux/tick.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
5517d86b | 123 | #ifdef CONFIG_SMP |
fd2ab30b SN |
124 | |
125 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
126 | ||
5517d86b ED |
127 | /* |
128 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
129 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
130 | */ | |
131 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
132 | { | |
133 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Each time a sched group cpu_power is changed, | |
138 | * we must compute its reciprocal value | |
139 | */ | |
140 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
141 | { | |
142 | sg->__cpu_power += val; | |
143 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
144 | } | |
145 | #endif | |
146 | ||
e05606d3 IM |
147 | static inline int rt_policy(int policy) |
148 | { | |
3f33a7ce | 149 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
150 | return 1; |
151 | return 0; | |
152 | } | |
153 | ||
154 | static inline int task_has_rt_policy(struct task_struct *p) | |
155 | { | |
156 | return rt_policy(p->policy); | |
157 | } | |
158 | ||
1da177e4 | 159 | /* |
6aa645ea | 160 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 161 | */ |
6aa645ea IM |
162 | struct rt_prio_array { |
163 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
164 | struct list_head queue[MAX_RT_PRIO]; | |
165 | }; | |
166 | ||
d0b27fa7 | 167 | struct rt_bandwidth { |
ea736ed5 IM |
168 | /* nests inside the rq lock: */ |
169 | spinlock_t rt_runtime_lock; | |
170 | ktime_t rt_period; | |
171 | u64 rt_runtime; | |
172 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
173 | }; |
174 | ||
175 | static struct rt_bandwidth def_rt_bandwidth; | |
176 | ||
177 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
178 | ||
179 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
180 | { | |
181 | struct rt_bandwidth *rt_b = | |
182 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
183 | ktime_t now; | |
184 | int overrun; | |
185 | int idle = 0; | |
186 | ||
187 | for (;;) { | |
188 | now = hrtimer_cb_get_time(timer); | |
189 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
190 | ||
191 | if (!overrun) | |
192 | break; | |
193 | ||
194 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
195 | } | |
196 | ||
197 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
198 | } | |
199 | ||
200 | static | |
201 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
202 | { | |
203 | rt_b->rt_period = ns_to_ktime(period); | |
204 | rt_b->rt_runtime = runtime; | |
205 | ||
ac086bc2 PZ |
206 | spin_lock_init(&rt_b->rt_runtime_lock); |
207 | ||
d0b27fa7 PZ |
208 | hrtimer_init(&rt_b->rt_period_timer, |
209 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
210 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
211 | } |
212 | ||
c8bfff6d KH |
213 | static inline int rt_bandwidth_enabled(void) |
214 | { | |
215 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
216 | } |
217 | ||
218 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
219 | { | |
220 | ktime_t now; | |
221 | ||
cac64d00 | 222 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
223 | return; |
224 | ||
225 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
226 | return; | |
227 | ||
228 | spin_lock(&rt_b->rt_runtime_lock); | |
229 | for (;;) { | |
7f1e2ca9 PZ |
230 | unsigned long delta; |
231 | ktime_t soft, hard; | |
232 | ||
d0b27fa7 PZ |
233 | if (hrtimer_active(&rt_b->rt_period_timer)) |
234 | break; | |
235 | ||
236 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
237 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
238 | |
239 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
240 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
241 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
242 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
243 | HRTIMER_MODE_ABS, 0); | |
d0b27fa7 PZ |
244 | } |
245 | spin_unlock(&rt_b->rt_runtime_lock); | |
246 | } | |
247 | ||
248 | #ifdef CONFIG_RT_GROUP_SCHED | |
249 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
250 | { | |
251 | hrtimer_cancel(&rt_b->rt_period_timer); | |
252 | } | |
253 | #endif | |
254 | ||
712555ee HC |
255 | /* |
256 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
257 | * detach_destroy_domains and partition_sched_domains. | |
258 | */ | |
259 | static DEFINE_MUTEX(sched_domains_mutex); | |
260 | ||
052f1dc7 | 261 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 262 | |
68318b8e SV |
263 | #include <linux/cgroup.h> |
264 | ||
29f59db3 SV |
265 | struct cfs_rq; |
266 | ||
6f505b16 PZ |
267 | static LIST_HEAD(task_groups); |
268 | ||
29f59db3 | 269 | /* task group related information */ |
4cf86d77 | 270 | struct task_group { |
052f1dc7 | 271 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
272 | struct cgroup_subsys_state css; |
273 | #endif | |
052f1dc7 | 274 | |
6c415b92 AB |
275 | #ifdef CONFIG_USER_SCHED |
276 | uid_t uid; | |
277 | #endif | |
278 | ||
052f1dc7 | 279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
280 | /* schedulable entities of this group on each cpu */ |
281 | struct sched_entity **se; | |
282 | /* runqueue "owned" by this group on each cpu */ | |
283 | struct cfs_rq **cfs_rq; | |
284 | unsigned long shares; | |
052f1dc7 PZ |
285 | #endif |
286 | ||
287 | #ifdef CONFIG_RT_GROUP_SCHED | |
288 | struct sched_rt_entity **rt_se; | |
289 | struct rt_rq **rt_rq; | |
290 | ||
d0b27fa7 | 291 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 292 | #endif |
6b2d7700 | 293 | |
ae8393e5 | 294 | struct rcu_head rcu; |
6f505b16 | 295 | struct list_head list; |
f473aa5e PZ |
296 | |
297 | struct task_group *parent; | |
298 | struct list_head siblings; | |
299 | struct list_head children; | |
29f59db3 SV |
300 | }; |
301 | ||
354d60c2 | 302 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 303 | |
6c415b92 AB |
304 | /* Helper function to pass uid information to create_sched_user() */ |
305 | void set_tg_uid(struct user_struct *user) | |
306 | { | |
307 | user->tg->uid = user->uid; | |
308 | } | |
309 | ||
eff766a6 PZ |
310 | /* |
311 | * Root task group. | |
312 | * Every UID task group (including init_task_group aka UID-0) will | |
313 | * be a child to this group. | |
314 | */ | |
315 | struct task_group root_task_group; | |
316 | ||
052f1dc7 | 317 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
318 | /* Default task group's sched entity on each cpu */ |
319 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
320 | /* Default task group's cfs_rq on each cpu */ | |
321 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 322 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
323 | |
324 | #ifdef CONFIG_RT_GROUP_SCHED | |
325 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
326 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 327 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 328 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 329 | #define root_task_group init_task_group |
9a7e0b18 | 330 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 331 | |
8ed36996 | 332 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
333 | * a task group's cpu shares. |
334 | */ | |
8ed36996 | 335 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 336 | |
57310a98 PZ |
337 | #ifdef CONFIG_SMP |
338 | static int root_task_group_empty(void) | |
339 | { | |
340 | return list_empty(&root_task_group.children); | |
341 | } | |
342 | #endif | |
343 | ||
052f1dc7 | 344 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
345 | #ifdef CONFIG_USER_SCHED |
346 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 347 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 348 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 349 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 350 | |
cb4ad1ff | 351 | /* |
2e084786 LJ |
352 | * A weight of 0 or 1 can cause arithmetics problems. |
353 | * A weight of a cfs_rq is the sum of weights of which entities | |
354 | * are queued on this cfs_rq, so a weight of a entity should not be | |
355 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
356 | * (The default weight is 1024 - so there's no practical |
357 | * limitation from this.) | |
358 | */ | |
18d95a28 | 359 | #define MIN_SHARES 2 |
2e084786 | 360 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 361 | |
052f1dc7 PZ |
362 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
363 | #endif | |
364 | ||
29f59db3 | 365 | /* Default task group. |
3a252015 | 366 | * Every task in system belong to this group at bootup. |
29f59db3 | 367 | */ |
434d53b0 | 368 | struct task_group init_task_group; |
29f59db3 SV |
369 | |
370 | /* return group to which a task belongs */ | |
4cf86d77 | 371 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 372 | { |
4cf86d77 | 373 | struct task_group *tg; |
9b5b7751 | 374 | |
052f1dc7 | 375 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
376 | rcu_read_lock(); |
377 | tg = __task_cred(p)->user->tg; | |
378 | rcu_read_unlock(); | |
052f1dc7 | 379 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
380 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
381 | struct task_group, css); | |
24e377a8 | 382 | #else |
41a2d6cf | 383 | tg = &init_task_group; |
24e377a8 | 384 | #endif |
9b5b7751 | 385 | return tg; |
29f59db3 SV |
386 | } |
387 | ||
388 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 389 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 390 | { |
052f1dc7 | 391 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
392 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
393 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 394 | #endif |
6f505b16 | 395 | |
052f1dc7 | 396 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
398 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 399 | #endif |
29f59db3 SV |
400 | } |
401 | ||
402 | #else | |
403 | ||
57310a98 PZ |
404 | #ifdef CONFIG_SMP |
405 | static int root_task_group_empty(void) | |
406 | { | |
407 | return 1; | |
408 | } | |
409 | #endif | |
410 | ||
6f505b16 | 411 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
412 | static inline struct task_group *task_group(struct task_struct *p) |
413 | { | |
414 | return NULL; | |
415 | } | |
29f59db3 | 416 | |
052f1dc7 | 417 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 418 | |
6aa645ea IM |
419 | /* CFS-related fields in a runqueue */ |
420 | struct cfs_rq { | |
421 | struct load_weight load; | |
422 | unsigned long nr_running; | |
423 | ||
6aa645ea | 424 | u64 exec_clock; |
e9acbff6 | 425 | u64 min_vruntime; |
6aa645ea IM |
426 | |
427 | struct rb_root tasks_timeline; | |
428 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
429 | |
430 | struct list_head tasks; | |
431 | struct list_head *balance_iterator; | |
432 | ||
433 | /* | |
434 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
435 | * It is set to NULL otherwise (i.e when none are currently running). |
436 | */ | |
4793241b | 437 | struct sched_entity *curr, *next, *last; |
ddc97297 | 438 | |
5ac5c4d6 | 439 | unsigned int nr_spread_over; |
ddc97297 | 440 | |
62160e3f | 441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
442 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
443 | ||
41a2d6cf IM |
444 | /* |
445 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
446 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
447 | * (like users, containers etc.) | |
448 | * | |
449 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
450 | * list is used during load balance. | |
451 | */ | |
41a2d6cf IM |
452 | struct list_head leaf_cfs_rq_list; |
453 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
454 | |
455 | #ifdef CONFIG_SMP | |
c09595f6 | 456 | /* |
c8cba857 | 457 | * the part of load.weight contributed by tasks |
c09595f6 | 458 | */ |
c8cba857 | 459 | unsigned long task_weight; |
c09595f6 | 460 | |
c8cba857 PZ |
461 | /* |
462 | * h_load = weight * f(tg) | |
463 | * | |
464 | * Where f(tg) is the recursive weight fraction assigned to | |
465 | * this group. | |
466 | */ | |
467 | unsigned long h_load; | |
c09595f6 | 468 | |
c8cba857 PZ |
469 | /* |
470 | * this cpu's part of tg->shares | |
471 | */ | |
472 | unsigned long shares; | |
f1d239f7 PZ |
473 | |
474 | /* | |
475 | * load.weight at the time we set shares | |
476 | */ | |
477 | unsigned long rq_weight; | |
c09595f6 | 478 | #endif |
6aa645ea IM |
479 | #endif |
480 | }; | |
1da177e4 | 481 | |
6aa645ea IM |
482 | /* Real-Time classes' related field in a runqueue: */ |
483 | struct rt_rq { | |
484 | struct rt_prio_array active; | |
63489e45 | 485 | unsigned long rt_nr_running; |
052f1dc7 | 486 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
487 | struct { |
488 | int curr; /* highest queued rt task prio */ | |
398a153b | 489 | #ifdef CONFIG_SMP |
e864c499 | 490 | int next; /* next highest */ |
398a153b | 491 | #endif |
e864c499 | 492 | } highest_prio; |
6f505b16 | 493 | #endif |
fa85ae24 | 494 | #ifdef CONFIG_SMP |
73fe6aae | 495 | unsigned long rt_nr_migratory; |
a22d7fc1 | 496 | int overloaded; |
917b627d | 497 | struct plist_head pushable_tasks; |
fa85ae24 | 498 | #endif |
6f505b16 | 499 | int rt_throttled; |
fa85ae24 | 500 | u64 rt_time; |
ac086bc2 | 501 | u64 rt_runtime; |
ea736ed5 | 502 | /* Nests inside the rq lock: */ |
ac086bc2 | 503 | spinlock_t rt_runtime_lock; |
6f505b16 | 504 | |
052f1dc7 | 505 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
506 | unsigned long rt_nr_boosted; |
507 | ||
6f505b16 PZ |
508 | struct rq *rq; |
509 | struct list_head leaf_rt_rq_list; | |
510 | struct task_group *tg; | |
511 | struct sched_rt_entity *rt_se; | |
512 | #endif | |
6aa645ea IM |
513 | }; |
514 | ||
57d885fe GH |
515 | #ifdef CONFIG_SMP |
516 | ||
517 | /* | |
518 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
519 | * variables. Each exclusive cpuset essentially defines an island domain by |
520 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
521 | * exclusive cpuset is created, we also create and attach a new root-domain |
522 | * object. | |
523 | * | |
57d885fe GH |
524 | */ |
525 | struct root_domain { | |
526 | atomic_t refcount; | |
c6c4927b RR |
527 | cpumask_var_t span; |
528 | cpumask_var_t online; | |
637f5085 | 529 | |
0eab9146 | 530 | /* |
637f5085 GH |
531 | * The "RT overload" flag: it gets set if a CPU has more than |
532 | * one runnable RT task. | |
533 | */ | |
c6c4927b | 534 | cpumask_var_t rto_mask; |
0eab9146 | 535 | atomic_t rto_count; |
6e0534f2 GH |
536 | #ifdef CONFIG_SMP |
537 | struct cpupri cpupri; | |
538 | #endif | |
7a09b1a2 VS |
539 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
540 | /* | |
541 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
542 | * used when most cpus are idle in the system indicating overall very | |
543 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
544 | */ | |
545 | unsigned int sched_mc_preferred_wakeup_cpu; | |
546 | #endif | |
57d885fe GH |
547 | }; |
548 | ||
dc938520 GH |
549 | /* |
550 | * By default the system creates a single root-domain with all cpus as | |
551 | * members (mimicking the global state we have today). | |
552 | */ | |
57d885fe GH |
553 | static struct root_domain def_root_domain; |
554 | ||
555 | #endif | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * This is the main, per-CPU runqueue data structure. | |
559 | * | |
560 | * Locking rule: those places that want to lock multiple runqueues | |
561 | * (such as the load balancing or the thread migration code), lock | |
562 | * acquire operations must be ordered by ascending &runqueue. | |
563 | */ | |
70b97a7f | 564 | struct rq { |
d8016491 IM |
565 | /* runqueue lock: */ |
566 | spinlock_t lock; | |
1da177e4 LT |
567 | |
568 | /* | |
569 | * nr_running and cpu_load should be in the same cacheline because | |
570 | * remote CPUs use both these fields when doing load calculation. | |
571 | */ | |
572 | unsigned long nr_running; | |
6aa645ea IM |
573 | #define CPU_LOAD_IDX_MAX 5 |
574 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 575 | #ifdef CONFIG_NO_HZ |
15934a37 | 576 | unsigned long last_tick_seen; |
46cb4b7c SS |
577 | unsigned char in_nohz_recently; |
578 | #endif | |
d8016491 IM |
579 | /* capture load from *all* tasks on this cpu: */ |
580 | struct load_weight load; | |
6aa645ea IM |
581 | unsigned long nr_load_updates; |
582 | u64 nr_switches; | |
23a185ca | 583 | u64 nr_migrations_in; |
6aa645ea IM |
584 | |
585 | struct cfs_rq cfs; | |
6f505b16 | 586 | struct rt_rq rt; |
6f505b16 | 587 | |
6aa645ea | 588 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
589 | /* list of leaf cfs_rq on this cpu: */ |
590 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
591 | #endif |
592 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 593 | struct list_head leaf_rt_rq_list; |
1da177e4 | 594 | #endif |
1da177e4 LT |
595 | |
596 | /* | |
597 | * This is part of a global counter where only the total sum | |
598 | * over all CPUs matters. A task can increase this counter on | |
599 | * one CPU and if it got migrated afterwards it may decrease | |
600 | * it on another CPU. Always updated under the runqueue lock: | |
601 | */ | |
602 | unsigned long nr_uninterruptible; | |
603 | ||
36c8b586 | 604 | struct task_struct *curr, *idle; |
c9819f45 | 605 | unsigned long next_balance; |
1da177e4 | 606 | struct mm_struct *prev_mm; |
6aa645ea | 607 | |
3e51f33f | 608 | u64 clock; |
6aa645ea | 609 | |
1da177e4 LT |
610 | atomic_t nr_iowait; |
611 | ||
612 | #ifdef CONFIG_SMP | |
0eab9146 | 613 | struct root_domain *rd; |
1da177e4 LT |
614 | struct sched_domain *sd; |
615 | ||
a0a522ce | 616 | unsigned char idle_at_tick; |
1da177e4 LT |
617 | /* For active balancing */ |
618 | int active_balance; | |
619 | int push_cpu; | |
d8016491 IM |
620 | /* cpu of this runqueue: */ |
621 | int cpu; | |
1f11eb6a | 622 | int online; |
1da177e4 | 623 | |
a8a51d5e | 624 | unsigned long avg_load_per_task; |
1da177e4 | 625 | |
36c8b586 | 626 | struct task_struct *migration_thread; |
1da177e4 LT |
627 | struct list_head migration_queue; |
628 | #endif | |
629 | ||
dce48a84 TG |
630 | /* calc_load related fields */ |
631 | unsigned long calc_load_update; | |
632 | long calc_load_active; | |
633 | ||
8f4d37ec | 634 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
635 | #ifdef CONFIG_SMP |
636 | int hrtick_csd_pending; | |
637 | struct call_single_data hrtick_csd; | |
638 | #endif | |
8f4d37ec PZ |
639 | struct hrtimer hrtick_timer; |
640 | #endif | |
641 | ||
1da177e4 LT |
642 | #ifdef CONFIG_SCHEDSTATS |
643 | /* latency stats */ | |
644 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
645 | unsigned long long rq_cpu_time; |
646 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
647 | |
648 | /* sys_sched_yield() stats */ | |
480b9434 | 649 | unsigned int yld_count; |
1da177e4 LT |
650 | |
651 | /* schedule() stats */ | |
480b9434 KC |
652 | unsigned int sched_switch; |
653 | unsigned int sched_count; | |
654 | unsigned int sched_goidle; | |
1da177e4 LT |
655 | |
656 | /* try_to_wake_up() stats */ | |
480b9434 KC |
657 | unsigned int ttwu_count; |
658 | unsigned int ttwu_local; | |
b8efb561 IM |
659 | |
660 | /* BKL stats */ | |
480b9434 | 661 | unsigned int bkl_count; |
1da177e4 LT |
662 | #endif |
663 | }; | |
664 | ||
f34e3b61 | 665 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 666 | |
15afe09b | 667 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 668 | { |
15afe09b | 669 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
670 | } |
671 | ||
0a2966b4 CL |
672 | static inline int cpu_of(struct rq *rq) |
673 | { | |
674 | #ifdef CONFIG_SMP | |
675 | return rq->cpu; | |
676 | #else | |
677 | return 0; | |
678 | #endif | |
679 | } | |
680 | ||
674311d5 NP |
681 | /* |
682 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 683 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
684 | * |
685 | * The domain tree of any CPU may only be accessed from within | |
686 | * preempt-disabled sections. | |
687 | */ | |
48f24c4d IM |
688 | #define for_each_domain(cpu, __sd) \ |
689 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
690 | |
691 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
692 | #define this_rq() (&__get_cpu_var(runqueues)) | |
693 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
694 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
695 | ||
aa9c4c0f | 696 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
697 | { |
698 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
699 | } | |
700 | ||
bf5c91ba IM |
701 | /* |
702 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
703 | */ | |
704 | #ifdef CONFIG_SCHED_DEBUG | |
705 | # define const_debug __read_mostly | |
706 | #else | |
707 | # define const_debug static const | |
708 | #endif | |
709 | ||
017730c1 IM |
710 | /** |
711 | * runqueue_is_locked | |
712 | * | |
713 | * Returns true if the current cpu runqueue is locked. | |
714 | * This interface allows printk to be called with the runqueue lock | |
715 | * held and know whether or not it is OK to wake up the klogd. | |
716 | */ | |
717 | int runqueue_is_locked(void) | |
718 | { | |
719 | int cpu = get_cpu(); | |
720 | struct rq *rq = cpu_rq(cpu); | |
721 | int ret; | |
722 | ||
723 | ret = spin_is_locked(&rq->lock); | |
724 | put_cpu(); | |
725 | return ret; | |
726 | } | |
727 | ||
bf5c91ba IM |
728 | /* |
729 | * Debugging: various feature bits | |
730 | */ | |
f00b45c1 PZ |
731 | |
732 | #define SCHED_FEAT(name, enabled) \ | |
733 | __SCHED_FEAT_##name , | |
734 | ||
bf5c91ba | 735 | enum { |
f00b45c1 | 736 | #include "sched_features.h" |
bf5c91ba IM |
737 | }; |
738 | ||
f00b45c1 PZ |
739 | #undef SCHED_FEAT |
740 | ||
741 | #define SCHED_FEAT(name, enabled) \ | |
742 | (1UL << __SCHED_FEAT_##name) * enabled | | |
743 | ||
bf5c91ba | 744 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
745 | #include "sched_features.h" |
746 | 0; | |
747 | ||
748 | #undef SCHED_FEAT | |
749 | ||
750 | #ifdef CONFIG_SCHED_DEBUG | |
751 | #define SCHED_FEAT(name, enabled) \ | |
752 | #name , | |
753 | ||
983ed7a6 | 754 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
755 | #include "sched_features.h" |
756 | NULL | |
757 | }; | |
758 | ||
759 | #undef SCHED_FEAT | |
760 | ||
34f3a814 | 761 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 762 | { |
f00b45c1 PZ |
763 | int i; |
764 | ||
765 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
766 | if (!(sysctl_sched_features & (1UL << i))) |
767 | seq_puts(m, "NO_"); | |
768 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 769 | } |
34f3a814 | 770 | seq_puts(m, "\n"); |
f00b45c1 | 771 | |
34f3a814 | 772 | return 0; |
f00b45c1 PZ |
773 | } |
774 | ||
775 | static ssize_t | |
776 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
777 | size_t cnt, loff_t *ppos) | |
778 | { | |
779 | char buf[64]; | |
780 | char *cmp = buf; | |
781 | int neg = 0; | |
782 | int i; | |
783 | ||
784 | if (cnt > 63) | |
785 | cnt = 63; | |
786 | ||
787 | if (copy_from_user(&buf, ubuf, cnt)) | |
788 | return -EFAULT; | |
789 | ||
790 | buf[cnt] = 0; | |
791 | ||
c24b7c52 | 792 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
793 | neg = 1; |
794 | cmp += 3; | |
795 | } | |
796 | ||
797 | for (i = 0; sched_feat_names[i]; i++) { | |
798 | int len = strlen(sched_feat_names[i]); | |
799 | ||
800 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
801 | if (neg) | |
802 | sysctl_sched_features &= ~(1UL << i); | |
803 | else | |
804 | sysctl_sched_features |= (1UL << i); | |
805 | break; | |
806 | } | |
807 | } | |
808 | ||
809 | if (!sched_feat_names[i]) | |
810 | return -EINVAL; | |
811 | ||
812 | filp->f_pos += cnt; | |
813 | ||
814 | return cnt; | |
815 | } | |
816 | ||
34f3a814 LZ |
817 | static int sched_feat_open(struct inode *inode, struct file *filp) |
818 | { | |
819 | return single_open(filp, sched_feat_show, NULL); | |
820 | } | |
821 | ||
f00b45c1 | 822 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
823 | .open = sched_feat_open, |
824 | .write = sched_feat_write, | |
825 | .read = seq_read, | |
826 | .llseek = seq_lseek, | |
827 | .release = single_release, | |
f00b45c1 PZ |
828 | }; |
829 | ||
830 | static __init int sched_init_debug(void) | |
831 | { | |
f00b45c1 PZ |
832 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
833 | &sched_feat_fops); | |
834 | ||
835 | return 0; | |
836 | } | |
837 | late_initcall(sched_init_debug); | |
838 | ||
839 | #endif | |
840 | ||
841 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 842 | |
b82d9fdd PZ |
843 | /* |
844 | * Number of tasks to iterate in a single balance run. | |
845 | * Limited because this is done with IRQs disabled. | |
846 | */ | |
847 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
848 | ||
2398f2c6 PZ |
849 | /* |
850 | * ratelimit for updating the group shares. | |
55cd5340 | 851 | * default: 0.25ms |
2398f2c6 | 852 | */ |
55cd5340 | 853 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 854 | |
ffda12a1 PZ |
855 | /* |
856 | * Inject some fuzzyness into changing the per-cpu group shares | |
857 | * this avoids remote rq-locks at the expense of fairness. | |
858 | * default: 4 | |
859 | */ | |
860 | unsigned int sysctl_sched_shares_thresh = 4; | |
861 | ||
fa85ae24 | 862 | /* |
9f0c1e56 | 863 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
864 | * default: 1s |
865 | */ | |
9f0c1e56 | 866 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 867 | |
6892b75e IM |
868 | static __read_mostly int scheduler_running; |
869 | ||
9f0c1e56 PZ |
870 | /* |
871 | * part of the period that we allow rt tasks to run in us. | |
872 | * default: 0.95s | |
873 | */ | |
874 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 875 | |
d0b27fa7 PZ |
876 | static inline u64 global_rt_period(void) |
877 | { | |
878 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
879 | } | |
880 | ||
881 | static inline u64 global_rt_runtime(void) | |
882 | { | |
e26873bb | 883 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
884 | return RUNTIME_INF; |
885 | ||
886 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
887 | } | |
fa85ae24 | 888 | |
1da177e4 | 889 | #ifndef prepare_arch_switch |
4866cde0 NP |
890 | # define prepare_arch_switch(next) do { } while (0) |
891 | #endif | |
892 | #ifndef finish_arch_switch | |
893 | # define finish_arch_switch(prev) do { } while (0) | |
894 | #endif | |
895 | ||
051a1d1a DA |
896 | static inline int task_current(struct rq *rq, struct task_struct *p) |
897 | { | |
898 | return rq->curr == p; | |
899 | } | |
900 | ||
4866cde0 | 901 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 902 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 903 | { |
051a1d1a | 904 | return task_current(rq, p); |
4866cde0 NP |
905 | } |
906 | ||
70b97a7f | 907 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
908 | { |
909 | } | |
910 | ||
70b97a7f | 911 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 912 | { |
da04c035 IM |
913 | #ifdef CONFIG_DEBUG_SPINLOCK |
914 | /* this is a valid case when another task releases the spinlock */ | |
915 | rq->lock.owner = current; | |
916 | #endif | |
8a25d5de IM |
917 | /* |
918 | * If we are tracking spinlock dependencies then we have to | |
919 | * fix up the runqueue lock - which gets 'carried over' from | |
920 | * prev into current: | |
921 | */ | |
922 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
923 | ||
4866cde0 NP |
924 | spin_unlock_irq(&rq->lock); |
925 | } | |
926 | ||
927 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 928 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
929 | { |
930 | #ifdef CONFIG_SMP | |
931 | return p->oncpu; | |
932 | #else | |
051a1d1a | 933 | return task_current(rq, p); |
4866cde0 NP |
934 | #endif |
935 | } | |
936 | ||
70b97a7f | 937 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
938 | { |
939 | #ifdef CONFIG_SMP | |
940 | /* | |
941 | * We can optimise this out completely for !SMP, because the | |
942 | * SMP rebalancing from interrupt is the only thing that cares | |
943 | * here. | |
944 | */ | |
945 | next->oncpu = 1; | |
946 | #endif | |
947 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
948 | spin_unlock_irq(&rq->lock); | |
949 | #else | |
950 | spin_unlock(&rq->lock); | |
951 | #endif | |
952 | } | |
953 | ||
70b97a7f | 954 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
955 | { |
956 | #ifdef CONFIG_SMP | |
957 | /* | |
958 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
959 | * We must ensure this doesn't happen until the switch is completely | |
960 | * finished. | |
961 | */ | |
962 | smp_wmb(); | |
963 | prev->oncpu = 0; | |
964 | #endif | |
965 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
966 | local_irq_enable(); | |
1da177e4 | 967 | #endif |
4866cde0 NP |
968 | } |
969 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 970 | |
b29739f9 IM |
971 | /* |
972 | * __task_rq_lock - lock the runqueue a given task resides on. | |
973 | * Must be called interrupts disabled. | |
974 | */ | |
70b97a7f | 975 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
976 | __acquires(rq->lock) |
977 | { | |
3a5c359a AK |
978 | for (;;) { |
979 | struct rq *rq = task_rq(p); | |
980 | spin_lock(&rq->lock); | |
981 | if (likely(rq == task_rq(p))) | |
982 | return rq; | |
b29739f9 | 983 | spin_unlock(&rq->lock); |
b29739f9 | 984 | } |
b29739f9 IM |
985 | } |
986 | ||
1da177e4 LT |
987 | /* |
988 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 989 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
990 | * explicitly disabling preemption. |
991 | */ | |
70b97a7f | 992 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
993 | __acquires(rq->lock) |
994 | { | |
70b97a7f | 995 | struct rq *rq; |
1da177e4 | 996 | |
3a5c359a AK |
997 | for (;;) { |
998 | local_irq_save(*flags); | |
999 | rq = task_rq(p); | |
1000 | spin_lock(&rq->lock); | |
1001 | if (likely(rq == task_rq(p))) | |
1002 | return rq; | |
1da177e4 | 1003 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1004 | } |
1da177e4 LT |
1005 | } |
1006 | ||
ad474cac ON |
1007 | void task_rq_unlock_wait(struct task_struct *p) |
1008 | { | |
1009 | struct rq *rq = task_rq(p); | |
1010 | ||
1011 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1012 | spin_unlock_wait(&rq->lock); | |
1013 | } | |
1014 | ||
a9957449 | 1015 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1016 | __releases(rq->lock) |
1017 | { | |
1018 | spin_unlock(&rq->lock); | |
1019 | } | |
1020 | ||
70b97a7f | 1021 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1022 | __releases(rq->lock) |
1023 | { | |
1024 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1025 | } | |
1026 | ||
1da177e4 | 1027 | /* |
cc2a73b5 | 1028 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1029 | */ |
a9957449 | 1030 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1031 | __acquires(rq->lock) |
1032 | { | |
70b97a7f | 1033 | struct rq *rq; |
1da177e4 LT |
1034 | |
1035 | local_irq_disable(); | |
1036 | rq = this_rq(); | |
1037 | spin_lock(&rq->lock); | |
1038 | ||
1039 | return rq; | |
1040 | } | |
1041 | ||
8f4d37ec PZ |
1042 | #ifdef CONFIG_SCHED_HRTICK |
1043 | /* | |
1044 | * Use HR-timers to deliver accurate preemption points. | |
1045 | * | |
1046 | * Its all a bit involved since we cannot program an hrt while holding the | |
1047 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1048 | * reschedule event. | |
1049 | * | |
1050 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1051 | * rq->lock. | |
1052 | */ | |
8f4d37ec PZ |
1053 | |
1054 | /* | |
1055 | * Use hrtick when: | |
1056 | * - enabled by features | |
1057 | * - hrtimer is actually high res | |
1058 | */ | |
1059 | static inline int hrtick_enabled(struct rq *rq) | |
1060 | { | |
1061 | if (!sched_feat(HRTICK)) | |
1062 | return 0; | |
ba42059f | 1063 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1064 | return 0; |
8f4d37ec PZ |
1065 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1066 | } | |
1067 | ||
8f4d37ec PZ |
1068 | static void hrtick_clear(struct rq *rq) |
1069 | { | |
1070 | if (hrtimer_active(&rq->hrtick_timer)) | |
1071 | hrtimer_cancel(&rq->hrtick_timer); | |
1072 | } | |
1073 | ||
8f4d37ec PZ |
1074 | /* |
1075 | * High-resolution timer tick. | |
1076 | * Runs from hardirq context with interrupts disabled. | |
1077 | */ | |
1078 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1079 | { | |
1080 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1081 | ||
1082 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1083 | ||
1084 | spin_lock(&rq->lock); | |
3e51f33f | 1085 | update_rq_clock(rq); |
8f4d37ec PZ |
1086 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1087 | spin_unlock(&rq->lock); | |
1088 | ||
1089 | return HRTIMER_NORESTART; | |
1090 | } | |
1091 | ||
95e904c7 | 1092 | #ifdef CONFIG_SMP |
31656519 PZ |
1093 | /* |
1094 | * called from hardirq (IPI) context | |
1095 | */ | |
1096 | static void __hrtick_start(void *arg) | |
b328ca18 | 1097 | { |
31656519 | 1098 | struct rq *rq = arg; |
b328ca18 | 1099 | |
31656519 PZ |
1100 | spin_lock(&rq->lock); |
1101 | hrtimer_restart(&rq->hrtick_timer); | |
1102 | rq->hrtick_csd_pending = 0; | |
1103 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1104 | } |
1105 | ||
31656519 PZ |
1106 | /* |
1107 | * Called to set the hrtick timer state. | |
1108 | * | |
1109 | * called with rq->lock held and irqs disabled | |
1110 | */ | |
1111 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1112 | { |
31656519 PZ |
1113 | struct hrtimer *timer = &rq->hrtick_timer; |
1114 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1115 | |
cc584b21 | 1116 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1117 | |
1118 | if (rq == this_rq()) { | |
1119 | hrtimer_restart(timer); | |
1120 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1121 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1122 | rq->hrtick_csd_pending = 1; |
1123 | } | |
b328ca18 PZ |
1124 | } |
1125 | ||
1126 | static int | |
1127 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1128 | { | |
1129 | int cpu = (int)(long)hcpu; | |
1130 | ||
1131 | switch (action) { | |
1132 | case CPU_UP_CANCELED: | |
1133 | case CPU_UP_CANCELED_FROZEN: | |
1134 | case CPU_DOWN_PREPARE: | |
1135 | case CPU_DOWN_PREPARE_FROZEN: | |
1136 | case CPU_DEAD: | |
1137 | case CPU_DEAD_FROZEN: | |
31656519 | 1138 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1139 | return NOTIFY_OK; |
1140 | } | |
1141 | ||
1142 | return NOTIFY_DONE; | |
1143 | } | |
1144 | ||
fa748203 | 1145 | static __init void init_hrtick(void) |
b328ca18 PZ |
1146 | { |
1147 | hotcpu_notifier(hotplug_hrtick, 0); | |
1148 | } | |
31656519 PZ |
1149 | #else |
1150 | /* | |
1151 | * Called to set the hrtick timer state. | |
1152 | * | |
1153 | * called with rq->lock held and irqs disabled | |
1154 | */ | |
1155 | static void hrtick_start(struct rq *rq, u64 delay) | |
1156 | { | |
7f1e2ca9 PZ |
1157 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
1158 | HRTIMER_MODE_REL, 0); | |
31656519 | 1159 | } |
b328ca18 | 1160 | |
006c75f1 | 1161 | static inline void init_hrtick(void) |
8f4d37ec | 1162 | { |
8f4d37ec | 1163 | } |
31656519 | 1164 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1165 | |
31656519 | 1166 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1167 | { |
31656519 PZ |
1168 | #ifdef CONFIG_SMP |
1169 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1170 | |
31656519 PZ |
1171 | rq->hrtick_csd.flags = 0; |
1172 | rq->hrtick_csd.func = __hrtick_start; | |
1173 | rq->hrtick_csd.info = rq; | |
1174 | #endif | |
8f4d37ec | 1175 | |
31656519 PZ |
1176 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1177 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1178 | } |
006c75f1 | 1179 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1180 | static inline void hrtick_clear(struct rq *rq) |
1181 | { | |
1182 | } | |
1183 | ||
8f4d37ec PZ |
1184 | static inline void init_rq_hrtick(struct rq *rq) |
1185 | { | |
1186 | } | |
1187 | ||
b328ca18 PZ |
1188 | static inline void init_hrtick(void) |
1189 | { | |
1190 | } | |
006c75f1 | 1191 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1192 | |
c24d20db IM |
1193 | /* |
1194 | * resched_task - mark a task 'to be rescheduled now'. | |
1195 | * | |
1196 | * On UP this means the setting of the need_resched flag, on SMP it | |
1197 | * might also involve a cross-CPU call to trigger the scheduler on | |
1198 | * the target CPU. | |
1199 | */ | |
1200 | #ifdef CONFIG_SMP | |
1201 | ||
1202 | #ifndef tsk_is_polling | |
1203 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1204 | #endif | |
1205 | ||
31656519 | 1206 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1207 | { |
1208 | int cpu; | |
1209 | ||
1210 | assert_spin_locked(&task_rq(p)->lock); | |
1211 | ||
5ed0cec0 | 1212 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1213 | return; |
1214 | ||
5ed0cec0 | 1215 | set_tsk_need_resched(p); |
c24d20db IM |
1216 | |
1217 | cpu = task_cpu(p); | |
1218 | if (cpu == smp_processor_id()) | |
1219 | return; | |
1220 | ||
1221 | /* NEED_RESCHED must be visible before we test polling */ | |
1222 | smp_mb(); | |
1223 | if (!tsk_is_polling(p)) | |
1224 | smp_send_reschedule(cpu); | |
1225 | } | |
1226 | ||
1227 | static void resched_cpu(int cpu) | |
1228 | { | |
1229 | struct rq *rq = cpu_rq(cpu); | |
1230 | unsigned long flags; | |
1231 | ||
1232 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1233 | return; | |
1234 | resched_task(cpu_curr(cpu)); | |
1235 | spin_unlock_irqrestore(&rq->lock, flags); | |
1236 | } | |
06d8308c TG |
1237 | |
1238 | #ifdef CONFIG_NO_HZ | |
1239 | /* | |
1240 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1241 | * idle CPU then this timer might expire before the next timer event | |
1242 | * which is scheduled to wake up that CPU. In case of a completely | |
1243 | * idle system the next event might even be infinite time into the | |
1244 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1245 | * leaves the inner idle loop so the newly added timer is taken into | |
1246 | * account when the CPU goes back to idle and evaluates the timer | |
1247 | * wheel for the next timer event. | |
1248 | */ | |
1249 | void wake_up_idle_cpu(int cpu) | |
1250 | { | |
1251 | struct rq *rq = cpu_rq(cpu); | |
1252 | ||
1253 | if (cpu == smp_processor_id()) | |
1254 | return; | |
1255 | ||
1256 | /* | |
1257 | * This is safe, as this function is called with the timer | |
1258 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1259 | * to idle and has not yet set rq->curr to idle then it will | |
1260 | * be serialized on the timer wheel base lock and take the new | |
1261 | * timer into account automatically. | |
1262 | */ | |
1263 | if (rq->curr != rq->idle) | |
1264 | return; | |
1265 | ||
1266 | /* | |
1267 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1268 | * lockless. The worst case is that the other CPU runs the | |
1269 | * idle task through an additional NOOP schedule() | |
1270 | */ | |
5ed0cec0 | 1271 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1272 | |
1273 | /* NEED_RESCHED must be visible before we test polling */ | |
1274 | smp_mb(); | |
1275 | if (!tsk_is_polling(rq->idle)) | |
1276 | smp_send_reschedule(cpu); | |
1277 | } | |
6d6bc0ad | 1278 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1279 | |
6d6bc0ad | 1280 | #else /* !CONFIG_SMP */ |
31656519 | 1281 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1282 | { |
1283 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1284 | set_tsk_need_resched(p); |
c24d20db | 1285 | } |
6d6bc0ad | 1286 | #endif /* CONFIG_SMP */ |
c24d20db | 1287 | |
45bf76df IM |
1288 | #if BITS_PER_LONG == 32 |
1289 | # define WMULT_CONST (~0UL) | |
1290 | #else | |
1291 | # define WMULT_CONST (1UL << 32) | |
1292 | #endif | |
1293 | ||
1294 | #define WMULT_SHIFT 32 | |
1295 | ||
194081eb IM |
1296 | /* |
1297 | * Shift right and round: | |
1298 | */ | |
cf2ab469 | 1299 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1300 | |
a7be37ac PZ |
1301 | /* |
1302 | * delta *= weight / lw | |
1303 | */ | |
cb1c4fc9 | 1304 | static unsigned long |
45bf76df IM |
1305 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1306 | struct load_weight *lw) | |
1307 | { | |
1308 | u64 tmp; | |
1309 | ||
7a232e03 LJ |
1310 | if (!lw->inv_weight) { |
1311 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1312 | lw->inv_weight = 1; | |
1313 | else | |
1314 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1315 | / (lw->weight+1); | |
1316 | } | |
45bf76df IM |
1317 | |
1318 | tmp = (u64)delta_exec * weight; | |
1319 | /* | |
1320 | * Check whether we'd overflow the 64-bit multiplication: | |
1321 | */ | |
194081eb | 1322 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1323 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1324 | WMULT_SHIFT/2); |
1325 | else | |
cf2ab469 | 1326 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1327 | |
ecf691da | 1328 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1329 | } |
1330 | ||
1091985b | 1331 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1332 | { |
1333 | lw->weight += inc; | |
e89996ae | 1334 | lw->inv_weight = 0; |
45bf76df IM |
1335 | } |
1336 | ||
1091985b | 1337 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1338 | { |
1339 | lw->weight -= dec; | |
e89996ae | 1340 | lw->inv_weight = 0; |
45bf76df IM |
1341 | } |
1342 | ||
2dd73a4f PW |
1343 | /* |
1344 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1345 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1346 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1347 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1348 | * scaled version of the new time slice allocation that they receive on time |
1349 | * slice expiry etc. | |
1350 | */ | |
1351 | ||
cce7ade8 PZ |
1352 | #define WEIGHT_IDLEPRIO 3 |
1353 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1354 | |
1355 | /* | |
1356 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1357 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1358 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1359 | * that remained on nice 0. | |
1360 | * | |
1361 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1362 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1363 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1364 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1365 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1366 | */ |
1367 | static const int prio_to_weight[40] = { | |
254753dc IM |
1368 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1369 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1370 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1371 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1372 | /* 0 */ 1024, 820, 655, 526, 423, | |
1373 | /* 5 */ 335, 272, 215, 172, 137, | |
1374 | /* 10 */ 110, 87, 70, 56, 45, | |
1375 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1376 | }; |
1377 | ||
5714d2de IM |
1378 | /* |
1379 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1380 | * | |
1381 | * In cases where the weight does not change often, we can use the | |
1382 | * precalculated inverse to speed up arithmetics by turning divisions | |
1383 | * into multiplications: | |
1384 | */ | |
dd41f596 | 1385 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1386 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1387 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1388 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1389 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1390 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1391 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1392 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1393 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1394 | }; |
2dd73a4f | 1395 | |
dd41f596 IM |
1396 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1397 | ||
1398 | /* | |
1399 | * runqueue iterator, to support SMP load-balancing between different | |
1400 | * scheduling classes, without having to expose their internal data | |
1401 | * structures to the load-balancing proper: | |
1402 | */ | |
1403 | struct rq_iterator { | |
1404 | void *arg; | |
1405 | struct task_struct *(*start)(void *); | |
1406 | struct task_struct *(*next)(void *); | |
1407 | }; | |
1408 | ||
e1d1484f PW |
1409 | #ifdef CONFIG_SMP |
1410 | static unsigned long | |
1411 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1412 | unsigned long max_load_move, struct sched_domain *sd, | |
1413 | enum cpu_idle_type idle, int *all_pinned, | |
1414 | int *this_best_prio, struct rq_iterator *iterator); | |
1415 | ||
1416 | static int | |
1417 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1418 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1419 | struct rq_iterator *iterator); | |
e1d1484f | 1420 | #endif |
dd41f596 | 1421 | |
ef12fefa BR |
1422 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1423 | enum cpuacct_stat_index { | |
1424 | CPUACCT_STAT_USER, /* ... user mode */ | |
1425 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1426 | ||
1427 | CPUACCT_STAT_NSTATS, | |
1428 | }; | |
1429 | ||
d842de87 SV |
1430 | #ifdef CONFIG_CGROUP_CPUACCT |
1431 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1432 | static void cpuacct_update_stats(struct task_struct *tsk, |
1433 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1434 | #else |
1435 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1436 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1437 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1438 | #endif |
1439 | ||
18d95a28 PZ |
1440 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1441 | { | |
1442 | update_load_add(&rq->load, load); | |
1443 | } | |
1444 | ||
1445 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1446 | { | |
1447 | update_load_sub(&rq->load, load); | |
1448 | } | |
1449 | ||
7940ca36 | 1450 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1451 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1452 | |
1453 | /* | |
1454 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1455 | * leaving it for the final time. | |
1456 | */ | |
eb755805 | 1457 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1458 | { |
1459 | struct task_group *parent, *child; | |
eb755805 | 1460 | int ret; |
c09595f6 PZ |
1461 | |
1462 | rcu_read_lock(); | |
1463 | parent = &root_task_group; | |
1464 | down: | |
eb755805 PZ |
1465 | ret = (*down)(parent, data); |
1466 | if (ret) | |
1467 | goto out_unlock; | |
c09595f6 PZ |
1468 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1469 | parent = child; | |
1470 | goto down; | |
1471 | ||
1472 | up: | |
1473 | continue; | |
1474 | } | |
eb755805 PZ |
1475 | ret = (*up)(parent, data); |
1476 | if (ret) | |
1477 | goto out_unlock; | |
c09595f6 PZ |
1478 | |
1479 | child = parent; | |
1480 | parent = parent->parent; | |
1481 | if (parent) | |
1482 | goto up; | |
eb755805 | 1483 | out_unlock: |
c09595f6 | 1484 | rcu_read_unlock(); |
eb755805 PZ |
1485 | |
1486 | return ret; | |
c09595f6 PZ |
1487 | } |
1488 | ||
eb755805 PZ |
1489 | static int tg_nop(struct task_group *tg, void *data) |
1490 | { | |
1491 | return 0; | |
c09595f6 | 1492 | } |
eb755805 PZ |
1493 | #endif |
1494 | ||
1495 | #ifdef CONFIG_SMP | |
1496 | static unsigned long source_load(int cpu, int type); | |
1497 | static unsigned long target_load(int cpu, int type); | |
1498 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1499 | ||
1500 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1501 | { | |
1502 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1503 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1504 | |
4cd42620 SR |
1505 | if (nr_running) |
1506 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1507 | else |
1508 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1509 | |
1510 | return rq->avg_load_per_task; | |
1511 | } | |
1512 | ||
1513 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1514 | |
c09595f6 PZ |
1515 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1516 | ||
1517 | /* | |
1518 | * Calculate and set the cpu's group shares. | |
1519 | */ | |
1520 | static void | |
ffda12a1 PZ |
1521 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1522 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1523 | { |
c09595f6 PZ |
1524 | unsigned long shares; |
1525 | unsigned long rq_weight; | |
1526 | ||
c8cba857 | 1527 | if (!tg->se[cpu]) |
c09595f6 PZ |
1528 | return; |
1529 | ||
ec4e0e2f | 1530 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1531 | |
c09595f6 PZ |
1532 | /* |
1533 | * \Sum shares * rq_weight | |
1534 | * shares = ----------------------- | |
1535 | * \Sum rq_weight | |
1536 | * | |
1537 | */ | |
ec4e0e2f | 1538 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1539 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1540 | |
ffda12a1 PZ |
1541 | if (abs(shares - tg->se[cpu]->load.weight) > |
1542 | sysctl_sched_shares_thresh) { | |
1543 | struct rq *rq = cpu_rq(cpu); | |
1544 | unsigned long flags; | |
c09595f6 | 1545 | |
ffda12a1 | 1546 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1547 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1548 | |
ffda12a1 PZ |
1549 | __set_se_shares(tg->se[cpu], shares); |
1550 | spin_unlock_irqrestore(&rq->lock, flags); | |
1551 | } | |
18d95a28 | 1552 | } |
c09595f6 PZ |
1553 | |
1554 | /* | |
c8cba857 PZ |
1555 | * Re-compute the task group their per cpu shares over the given domain. |
1556 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1557 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1558 | */ |
eb755805 | 1559 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1560 | { |
ec4e0e2f | 1561 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1562 | unsigned long shares = 0; |
eb755805 | 1563 | struct sched_domain *sd = data; |
c8cba857 | 1564 | int i; |
c09595f6 | 1565 | |
758b2cdc | 1566 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1567 | /* |
1568 | * If there are currently no tasks on the cpu pretend there | |
1569 | * is one of average load so that when a new task gets to | |
1570 | * run here it will not get delayed by group starvation. | |
1571 | */ | |
1572 | weight = tg->cfs_rq[i]->load.weight; | |
1573 | if (!weight) | |
1574 | weight = NICE_0_LOAD; | |
1575 | ||
1576 | tg->cfs_rq[i]->rq_weight = weight; | |
1577 | rq_weight += weight; | |
c8cba857 | 1578 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1579 | } |
c09595f6 | 1580 | |
c8cba857 PZ |
1581 | if ((!shares && rq_weight) || shares > tg->shares) |
1582 | shares = tg->shares; | |
1583 | ||
1584 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1585 | shares = tg->shares; | |
c09595f6 | 1586 | |
758b2cdc | 1587 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1588 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1589 | |
1590 | return 0; | |
c09595f6 PZ |
1591 | } |
1592 | ||
1593 | /* | |
c8cba857 PZ |
1594 | * Compute the cpu's hierarchical load factor for each task group. |
1595 | * This needs to be done in a top-down fashion because the load of a child | |
1596 | * group is a fraction of its parents load. | |
c09595f6 | 1597 | */ |
eb755805 | 1598 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1599 | { |
c8cba857 | 1600 | unsigned long load; |
eb755805 | 1601 | long cpu = (long)data; |
c09595f6 | 1602 | |
c8cba857 PZ |
1603 | if (!tg->parent) { |
1604 | load = cpu_rq(cpu)->load.weight; | |
1605 | } else { | |
1606 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1607 | load *= tg->cfs_rq[cpu]->shares; | |
1608 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1609 | } | |
c09595f6 | 1610 | |
c8cba857 | 1611 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1612 | |
eb755805 | 1613 | return 0; |
c09595f6 PZ |
1614 | } |
1615 | ||
c8cba857 | 1616 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1617 | { |
2398f2c6 PZ |
1618 | u64 now = cpu_clock(raw_smp_processor_id()); |
1619 | s64 elapsed = now - sd->last_update; | |
1620 | ||
1621 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1622 | sd->last_update = now; | |
eb755805 | 1623 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1624 | } |
4d8d595d PZ |
1625 | } |
1626 | ||
3e5459b4 PZ |
1627 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1628 | { | |
1629 | spin_unlock(&rq->lock); | |
1630 | update_shares(sd); | |
1631 | spin_lock(&rq->lock); | |
1632 | } | |
1633 | ||
eb755805 | 1634 | static void update_h_load(long cpu) |
c09595f6 | 1635 | { |
eb755805 | 1636 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1637 | } |
1638 | ||
c09595f6 PZ |
1639 | #else |
1640 | ||
c8cba857 | 1641 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1642 | { |
1643 | } | |
1644 | ||
3e5459b4 PZ |
1645 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1646 | { | |
1647 | } | |
1648 | ||
18d95a28 PZ |
1649 | #endif |
1650 | ||
8f45e2b5 GH |
1651 | #ifdef CONFIG_PREEMPT |
1652 | ||
70574a99 | 1653 | /* |
8f45e2b5 GH |
1654 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1655 | * way at the expense of forcing extra atomic operations in all | |
1656 | * invocations. This assures that the double_lock is acquired using the | |
1657 | * same underlying policy as the spinlock_t on this architecture, which | |
1658 | * reduces latency compared to the unfair variant below. However, it | |
1659 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1660 | */ |
8f45e2b5 GH |
1661 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1662 | __releases(this_rq->lock) | |
1663 | __acquires(busiest->lock) | |
1664 | __acquires(this_rq->lock) | |
1665 | { | |
1666 | spin_unlock(&this_rq->lock); | |
1667 | double_rq_lock(this_rq, busiest); | |
1668 | ||
1669 | return 1; | |
1670 | } | |
1671 | ||
1672 | #else | |
1673 | /* | |
1674 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1675 | * latency by eliminating extra atomic operations when the locks are | |
1676 | * already in proper order on entry. This favors lower cpu-ids and will | |
1677 | * grant the double lock to lower cpus over higher ids under contention, | |
1678 | * regardless of entry order into the function. | |
1679 | */ | |
1680 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1681 | __releases(this_rq->lock) |
1682 | __acquires(busiest->lock) | |
1683 | __acquires(this_rq->lock) | |
1684 | { | |
1685 | int ret = 0; | |
1686 | ||
70574a99 AD |
1687 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1688 | if (busiest < this_rq) { | |
1689 | spin_unlock(&this_rq->lock); | |
1690 | spin_lock(&busiest->lock); | |
1691 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1692 | ret = 1; | |
1693 | } else | |
1694 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1695 | } | |
1696 | return ret; | |
1697 | } | |
1698 | ||
8f45e2b5 GH |
1699 | #endif /* CONFIG_PREEMPT */ |
1700 | ||
1701 | /* | |
1702 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1703 | */ | |
1704 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1705 | { | |
1706 | if (unlikely(!irqs_disabled())) { | |
1707 | /* printk() doesn't work good under rq->lock */ | |
1708 | spin_unlock(&this_rq->lock); | |
1709 | BUG_ON(1); | |
1710 | } | |
1711 | ||
1712 | return _double_lock_balance(this_rq, busiest); | |
1713 | } | |
1714 | ||
70574a99 AD |
1715 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1716 | __releases(busiest->lock) | |
1717 | { | |
1718 | spin_unlock(&busiest->lock); | |
1719 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1720 | } | |
18d95a28 PZ |
1721 | #endif |
1722 | ||
30432094 | 1723 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1724 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1725 | { | |
30432094 | 1726 | #ifdef CONFIG_SMP |
34e83e85 IM |
1727 | cfs_rq->shares = shares; |
1728 | #endif | |
1729 | } | |
30432094 | 1730 | #endif |
e7693a36 | 1731 | |
dce48a84 TG |
1732 | static void calc_load_account_active(struct rq *this_rq); |
1733 | ||
dd41f596 | 1734 | #include "sched_stats.h" |
dd41f596 | 1735 | #include "sched_idletask.c" |
5522d5d5 IM |
1736 | #include "sched_fair.c" |
1737 | #include "sched_rt.c" | |
dd41f596 IM |
1738 | #ifdef CONFIG_SCHED_DEBUG |
1739 | # include "sched_debug.c" | |
1740 | #endif | |
1741 | ||
1742 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1743 | #define for_each_class(class) \ |
1744 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1745 | |
c09595f6 | 1746 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1747 | { |
1748 | rq->nr_running++; | |
9c217245 IM |
1749 | } |
1750 | ||
c09595f6 | 1751 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1752 | { |
1753 | rq->nr_running--; | |
9c217245 IM |
1754 | } |
1755 | ||
45bf76df IM |
1756 | static void set_load_weight(struct task_struct *p) |
1757 | { | |
1758 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1759 | p->se.load.weight = prio_to_weight[0] * 2; |
1760 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1761 | return; | |
1762 | } | |
45bf76df | 1763 | |
dd41f596 IM |
1764 | /* |
1765 | * SCHED_IDLE tasks get minimal weight: | |
1766 | */ | |
1767 | if (p->policy == SCHED_IDLE) { | |
1768 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1769 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1770 | return; | |
1771 | } | |
71f8bd46 | 1772 | |
dd41f596 IM |
1773 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1774 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1775 | } |
1776 | ||
2087a1ad GH |
1777 | static void update_avg(u64 *avg, u64 sample) |
1778 | { | |
1779 | s64 diff = sample - *avg; | |
1780 | *avg += diff >> 3; | |
1781 | } | |
1782 | ||
8159f87e | 1783 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1784 | { |
831451ac PZ |
1785 | if (wakeup) |
1786 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1787 | ||
dd41f596 | 1788 | sched_info_queued(p); |
fd390f6a | 1789 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1790 | p->se.on_rq = 1; |
71f8bd46 IM |
1791 | } |
1792 | ||
69be72c1 | 1793 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1794 | { |
831451ac PZ |
1795 | if (sleep) { |
1796 | if (p->se.last_wakeup) { | |
1797 | update_avg(&p->se.avg_overlap, | |
1798 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1799 | p->se.last_wakeup = 0; | |
1800 | } else { | |
1801 | update_avg(&p->se.avg_wakeup, | |
1802 | sysctl_sched_wakeup_granularity); | |
1803 | } | |
2087a1ad GH |
1804 | } |
1805 | ||
46ac22ba | 1806 | sched_info_dequeued(p); |
f02231e5 | 1807 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1808 | p->se.on_rq = 0; |
71f8bd46 IM |
1809 | } |
1810 | ||
14531189 | 1811 | /* |
dd41f596 | 1812 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1813 | */ |
14531189 IM |
1814 | static inline int __normal_prio(struct task_struct *p) |
1815 | { | |
dd41f596 | 1816 | return p->static_prio; |
14531189 IM |
1817 | } |
1818 | ||
b29739f9 IM |
1819 | /* |
1820 | * Calculate the expected normal priority: i.e. priority | |
1821 | * without taking RT-inheritance into account. Might be | |
1822 | * boosted by interactivity modifiers. Changes upon fork, | |
1823 | * setprio syscalls, and whenever the interactivity | |
1824 | * estimator recalculates. | |
1825 | */ | |
36c8b586 | 1826 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1827 | { |
1828 | int prio; | |
1829 | ||
e05606d3 | 1830 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1831 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1832 | else | |
1833 | prio = __normal_prio(p); | |
1834 | return prio; | |
1835 | } | |
1836 | ||
1837 | /* | |
1838 | * Calculate the current priority, i.e. the priority | |
1839 | * taken into account by the scheduler. This value might | |
1840 | * be boosted by RT tasks, or might be boosted by | |
1841 | * interactivity modifiers. Will be RT if the task got | |
1842 | * RT-boosted. If not then it returns p->normal_prio. | |
1843 | */ | |
36c8b586 | 1844 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1845 | { |
1846 | p->normal_prio = normal_prio(p); | |
1847 | /* | |
1848 | * If we are RT tasks or we were boosted to RT priority, | |
1849 | * keep the priority unchanged. Otherwise, update priority | |
1850 | * to the normal priority: | |
1851 | */ | |
1852 | if (!rt_prio(p->prio)) | |
1853 | return p->normal_prio; | |
1854 | return p->prio; | |
1855 | } | |
1856 | ||
1da177e4 | 1857 | /* |
dd41f596 | 1858 | * activate_task - move a task to the runqueue. |
1da177e4 | 1859 | */ |
dd41f596 | 1860 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1861 | { |
d9514f6c | 1862 | if (task_contributes_to_load(p)) |
dd41f596 | 1863 | rq->nr_uninterruptible--; |
1da177e4 | 1864 | |
8159f87e | 1865 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1866 | inc_nr_running(rq); |
1da177e4 LT |
1867 | } |
1868 | ||
1da177e4 LT |
1869 | /* |
1870 | * deactivate_task - remove a task from the runqueue. | |
1871 | */ | |
2e1cb74a | 1872 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1873 | { |
d9514f6c | 1874 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1875 | rq->nr_uninterruptible++; |
1876 | ||
69be72c1 | 1877 | dequeue_task(rq, p, sleep); |
c09595f6 | 1878 | dec_nr_running(rq); |
1da177e4 LT |
1879 | } |
1880 | ||
1da177e4 LT |
1881 | /** |
1882 | * task_curr - is this task currently executing on a CPU? | |
1883 | * @p: the task in question. | |
1884 | */ | |
36c8b586 | 1885 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1886 | { |
1887 | return cpu_curr(task_cpu(p)) == p; | |
1888 | } | |
1889 | ||
dd41f596 IM |
1890 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1891 | { | |
6f505b16 | 1892 | set_task_rq(p, cpu); |
dd41f596 | 1893 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1894 | /* |
1895 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1896 | * successfuly executed on another CPU. We must ensure that updates of | |
1897 | * per-task data have been completed by this moment. | |
1898 | */ | |
1899 | smp_wmb(); | |
dd41f596 | 1900 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1901 | #endif |
2dd73a4f PW |
1902 | } |
1903 | ||
cb469845 SR |
1904 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1905 | const struct sched_class *prev_class, | |
1906 | int oldprio, int running) | |
1907 | { | |
1908 | if (prev_class != p->sched_class) { | |
1909 | if (prev_class->switched_from) | |
1910 | prev_class->switched_from(rq, p, running); | |
1911 | p->sched_class->switched_to(rq, p, running); | |
1912 | } else | |
1913 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1914 | } | |
1915 | ||
1da177e4 | 1916 | #ifdef CONFIG_SMP |
c65cc870 | 1917 | |
e958b360 TG |
1918 | /* Used instead of source_load when we know the type == 0 */ |
1919 | static unsigned long weighted_cpuload(const int cpu) | |
1920 | { | |
1921 | return cpu_rq(cpu)->load.weight; | |
1922 | } | |
1923 | ||
cc367732 IM |
1924 | /* |
1925 | * Is this task likely cache-hot: | |
1926 | */ | |
e7693a36 | 1927 | static int |
cc367732 IM |
1928 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1929 | { | |
1930 | s64 delta; | |
1931 | ||
f540a608 IM |
1932 | /* |
1933 | * Buddy candidates are cache hot: | |
1934 | */ | |
4793241b PZ |
1935 | if (sched_feat(CACHE_HOT_BUDDY) && |
1936 | (&p->se == cfs_rq_of(&p->se)->next || | |
1937 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1938 | return 1; |
1939 | ||
cc367732 IM |
1940 | if (p->sched_class != &fair_sched_class) |
1941 | return 0; | |
1942 | ||
6bc1665b IM |
1943 | if (sysctl_sched_migration_cost == -1) |
1944 | return 1; | |
1945 | if (sysctl_sched_migration_cost == 0) | |
1946 | return 0; | |
1947 | ||
cc367732 IM |
1948 | delta = now - p->se.exec_start; |
1949 | ||
1950 | return delta < (s64)sysctl_sched_migration_cost; | |
1951 | } | |
1952 | ||
1953 | ||
dd41f596 | 1954 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1955 | { |
dd41f596 IM |
1956 | int old_cpu = task_cpu(p); |
1957 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1958 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1959 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1960 | u64 clock_offset; |
dd41f596 IM |
1961 | |
1962 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1963 | |
de1d7286 | 1964 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1965 | |
6cfb0d5d IM |
1966 | #ifdef CONFIG_SCHEDSTATS |
1967 | if (p->se.wait_start) | |
1968 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1969 | if (p->se.sleep_start) |
1970 | p->se.sleep_start -= clock_offset; | |
1971 | if (p->se.block_start) | |
1972 | p->se.block_start -= clock_offset; | |
6c594c21 | 1973 | #endif |
cc367732 | 1974 | if (old_cpu != new_cpu) { |
6c594c21 | 1975 | p->se.nr_migrations++; |
23a185ca | 1976 | new_rq->nr_migrations_in++; |
6c594c21 | 1977 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
1978 | if (task_hot(p, old_rq->clock, NULL)) |
1979 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 1980 | #endif |
3f731ca6 | 1981 | perf_counter_task_migration(p, new_cpu); |
6c594c21 | 1982 | } |
2830cf8c SV |
1983 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1984 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1985 | |
1986 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1987 | } |
1988 | ||
70b97a7f | 1989 | struct migration_req { |
1da177e4 | 1990 | struct list_head list; |
1da177e4 | 1991 | |
36c8b586 | 1992 | struct task_struct *task; |
1da177e4 LT |
1993 | int dest_cpu; |
1994 | ||
1da177e4 | 1995 | struct completion done; |
70b97a7f | 1996 | }; |
1da177e4 LT |
1997 | |
1998 | /* | |
1999 | * The task's runqueue lock must be held. | |
2000 | * Returns true if you have to wait for migration thread. | |
2001 | */ | |
36c8b586 | 2002 | static int |
70b97a7f | 2003 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2004 | { |
70b97a7f | 2005 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2006 | |
2007 | /* | |
2008 | * If the task is not on a runqueue (and not running), then | |
2009 | * it is sufficient to simply update the task's cpu field. | |
2010 | */ | |
dd41f596 | 2011 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2012 | set_task_cpu(p, dest_cpu); |
2013 | return 0; | |
2014 | } | |
2015 | ||
2016 | init_completion(&req->done); | |
1da177e4 LT |
2017 | req->task = p; |
2018 | req->dest_cpu = dest_cpu; | |
2019 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2020 | |
1da177e4 LT |
2021 | return 1; |
2022 | } | |
2023 | ||
a26b89f0 MM |
2024 | /* |
2025 | * wait_task_context_switch - wait for a thread to complete at least one | |
2026 | * context switch. | |
2027 | * | |
2028 | * @p must not be current. | |
2029 | */ | |
2030 | void wait_task_context_switch(struct task_struct *p) | |
2031 | { | |
2032 | unsigned long nvcsw, nivcsw, flags; | |
2033 | int running; | |
2034 | struct rq *rq; | |
2035 | ||
2036 | nvcsw = p->nvcsw; | |
2037 | nivcsw = p->nivcsw; | |
2038 | for (;;) { | |
2039 | /* | |
2040 | * The runqueue is assigned before the actual context | |
2041 | * switch. We need to take the runqueue lock. | |
2042 | * | |
2043 | * We could check initially without the lock but it is | |
2044 | * very likely that we need to take the lock in every | |
2045 | * iteration. | |
2046 | */ | |
2047 | rq = task_rq_lock(p, &flags); | |
2048 | running = task_running(rq, p); | |
2049 | task_rq_unlock(rq, &flags); | |
2050 | ||
2051 | if (likely(!running)) | |
2052 | break; | |
2053 | /* | |
2054 | * The switch count is incremented before the actual | |
2055 | * context switch. We thus wait for two switches to be | |
2056 | * sure at least one completed. | |
2057 | */ | |
2058 | if ((p->nvcsw - nvcsw) > 1) | |
2059 | break; | |
2060 | if ((p->nivcsw - nivcsw) > 1) | |
2061 | break; | |
2062 | ||
2063 | cpu_relax(); | |
2064 | } | |
2065 | } | |
2066 | ||
1da177e4 LT |
2067 | /* |
2068 | * wait_task_inactive - wait for a thread to unschedule. | |
2069 | * | |
85ba2d86 RM |
2070 | * If @match_state is nonzero, it's the @p->state value just checked and |
2071 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2072 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2073 | * we return a positive number (its total switch count). If a second call | |
2074 | * a short while later returns the same number, the caller can be sure that | |
2075 | * @p has remained unscheduled the whole time. | |
2076 | * | |
1da177e4 LT |
2077 | * The caller must ensure that the task *will* unschedule sometime soon, |
2078 | * else this function might spin for a *long* time. This function can't | |
2079 | * be called with interrupts off, or it may introduce deadlock with | |
2080 | * smp_call_function() if an IPI is sent by the same process we are | |
2081 | * waiting to become inactive. | |
2082 | */ | |
85ba2d86 | 2083 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2084 | { |
2085 | unsigned long flags; | |
dd41f596 | 2086 | int running, on_rq; |
85ba2d86 | 2087 | unsigned long ncsw; |
70b97a7f | 2088 | struct rq *rq; |
1da177e4 | 2089 | |
3a5c359a AK |
2090 | for (;;) { |
2091 | /* | |
2092 | * We do the initial early heuristics without holding | |
2093 | * any task-queue locks at all. We'll only try to get | |
2094 | * the runqueue lock when things look like they will | |
2095 | * work out! | |
2096 | */ | |
2097 | rq = task_rq(p); | |
fa490cfd | 2098 | |
3a5c359a AK |
2099 | /* |
2100 | * If the task is actively running on another CPU | |
2101 | * still, just relax and busy-wait without holding | |
2102 | * any locks. | |
2103 | * | |
2104 | * NOTE! Since we don't hold any locks, it's not | |
2105 | * even sure that "rq" stays as the right runqueue! | |
2106 | * But we don't care, since "task_running()" will | |
2107 | * return false if the runqueue has changed and p | |
2108 | * is actually now running somewhere else! | |
2109 | */ | |
85ba2d86 RM |
2110 | while (task_running(rq, p)) { |
2111 | if (match_state && unlikely(p->state != match_state)) | |
2112 | return 0; | |
3a5c359a | 2113 | cpu_relax(); |
85ba2d86 | 2114 | } |
fa490cfd | 2115 | |
3a5c359a AK |
2116 | /* |
2117 | * Ok, time to look more closely! We need the rq | |
2118 | * lock now, to be *sure*. If we're wrong, we'll | |
2119 | * just go back and repeat. | |
2120 | */ | |
2121 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2122 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2123 | running = task_running(rq, p); |
2124 | on_rq = p->se.on_rq; | |
85ba2d86 | 2125 | ncsw = 0; |
f31e11d8 | 2126 | if (!match_state || p->state == match_state) |
93dcf55f | 2127 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2128 | task_rq_unlock(rq, &flags); |
fa490cfd | 2129 | |
85ba2d86 RM |
2130 | /* |
2131 | * If it changed from the expected state, bail out now. | |
2132 | */ | |
2133 | if (unlikely(!ncsw)) | |
2134 | break; | |
2135 | ||
3a5c359a AK |
2136 | /* |
2137 | * Was it really running after all now that we | |
2138 | * checked with the proper locks actually held? | |
2139 | * | |
2140 | * Oops. Go back and try again.. | |
2141 | */ | |
2142 | if (unlikely(running)) { | |
2143 | cpu_relax(); | |
2144 | continue; | |
2145 | } | |
fa490cfd | 2146 | |
3a5c359a AK |
2147 | /* |
2148 | * It's not enough that it's not actively running, | |
2149 | * it must be off the runqueue _entirely_, and not | |
2150 | * preempted! | |
2151 | * | |
80dd99b3 | 2152 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2153 | * running right now), it's preempted, and we should |
2154 | * yield - it could be a while. | |
2155 | */ | |
2156 | if (unlikely(on_rq)) { | |
2157 | schedule_timeout_uninterruptible(1); | |
2158 | continue; | |
2159 | } | |
fa490cfd | 2160 | |
3a5c359a AK |
2161 | /* |
2162 | * Ahh, all good. It wasn't running, and it wasn't | |
2163 | * runnable, which means that it will never become | |
2164 | * running in the future either. We're all done! | |
2165 | */ | |
2166 | break; | |
2167 | } | |
85ba2d86 RM |
2168 | |
2169 | return ncsw; | |
1da177e4 LT |
2170 | } |
2171 | ||
2172 | /*** | |
2173 | * kick_process - kick a running thread to enter/exit the kernel | |
2174 | * @p: the to-be-kicked thread | |
2175 | * | |
2176 | * Cause a process which is running on another CPU to enter | |
2177 | * kernel-mode, without any delay. (to get signals handled.) | |
2178 | * | |
2179 | * NOTE: this function doesnt have to take the runqueue lock, | |
2180 | * because all it wants to ensure is that the remote task enters | |
2181 | * the kernel. If the IPI races and the task has been migrated | |
2182 | * to another CPU then no harm is done and the purpose has been | |
2183 | * achieved as well. | |
2184 | */ | |
36c8b586 | 2185 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2186 | { |
2187 | int cpu; | |
2188 | ||
2189 | preempt_disable(); | |
2190 | cpu = task_cpu(p); | |
2191 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2192 | smp_send_reschedule(cpu); | |
2193 | preempt_enable(); | |
2194 | } | |
2195 | ||
2196 | /* | |
2dd73a4f PW |
2197 | * Return a low guess at the load of a migration-source cpu weighted |
2198 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2199 | * |
2200 | * We want to under-estimate the load of migration sources, to | |
2201 | * balance conservatively. | |
2202 | */ | |
a9957449 | 2203 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2204 | { |
70b97a7f | 2205 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2206 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2207 | |
93b75217 | 2208 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2209 | return total; |
b910472d | 2210 | |
dd41f596 | 2211 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2212 | } |
2213 | ||
2214 | /* | |
2dd73a4f PW |
2215 | * Return a high guess at the load of a migration-target cpu weighted |
2216 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2217 | */ |
a9957449 | 2218 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2219 | { |
70b97a7f | 2220 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2221 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2222 | |
93b75217 | 2223 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2224 | return total; |
3b0bd9bc | 2225 | |
dd41f596 | 2226 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2227 | } |
2228 | ||
147cbb4b NP |
2229 | /* |
2230 | * find_idlest_group finds and returns the least busy CPU group within the | |
2231 | * domain. | |
2232 | */ | |
2233 | static struct sched_group * | |
2234 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2235 | { | |
2236 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2237 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2238 | int load_idx = sd->forkexec_idx; | |
2239 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2240 | ||
2241 | do { | |
2242 | unsigned long load, avg_load; | |
2243 | int local_group; | |
2244 | int i; | |
2245 | ||
da5a5522 | 2246 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2247 | if (!cpumask_intersects(sched_group_cpus(group), |
2248 | &p->cpus_allowed)) | |
3a5c359a | 2249 | continue; |
da5a5522 | 2250 | |
758b2cdc RR |
2251 | local_group = cpumask_test_cpu(this_cpu, |
2252 | sched_group_cpus(group)); | |
147cbb4b NP |
2253 | |
2254 | /* Tally up the load of all CPUs in the group */ | |
2255 | avg_load = 0; | |
2256 | ||
758b2cdc | 2257 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2258 | /* Bias balancing toward cpus of our domain */ |
2259 | if (local_group) | |
2260 | load = source_load(i, load_idx); | |
2261 | else | |
2262 | load = target_load(i, load_idx); | |
2263 | ||
2264 | avg_load += load; | |
2265 | } | |
2266 | ||
2267 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2268 | avg_load = sg_div_cpu_power(group, |
2269 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2270 | |
2271 | if (local_group) { | |
2272 | this_load = avg_load; | |
2273 | this = group; | |
2274 | } else if (avg_load < min_load) { | |
2275 | min_load = avg_load; | |
2276 | idlest = group; | |
2277 | } | |
3a5c359a | 2278 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2279 | |
2280 | if (!idlest || 100*this_load < imbalance*min_load) | |
2281 | return NULL; | |
2282 | return idlest; | |
2283 | } | |
2284 | ||
2285 | /* | |
0feaece9 | 2286 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2287 | */ |
95cdf3b7 | 2288 | static int |
758b2cdc | 2289 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2290 | { |
2291 | unsigned long load, min_load = ULONG_MAX; | |
2292 | int idlest = -1; | |
2293 | int i; | |
2294 | ||
da5a5522 | 2295 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2296 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2297 | load = weighted_cpuload(i); |
147cbb4b NP |
2298 | |
2299 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2300 | min_load = load; | |
2301 | idlest = i; | |
2302 | } | |
2303 | } | |
2304 | ||
2305 | return idlest; | |
2306 | } | |
2307 | ||
476d139c NP |
2308 | /* |
2309 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2310 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2311 | * SD_BALANCE_EXEC. | |
2312 | * | |
2313 | * Balance, ie. select the least loaded group. | |
2314 | * | |
2315 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2316 | * | |
2317 | * preempt must be disabled. | |
2318 | */ | |
2319 | static int sched_balance_self(int cpu, int flag) | |
2320 | { | |
2321 | struct task_struct *t = current; | |
2322 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2323 | |
c96d145e | 2324 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2325 | /* |
2326 | * If power savings logic is enabled for a domain, stop there. | |
2327 | */ | |
5c45bf27 SS |
2328 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2329 | break; | |
476d139c NP |
2330 | if (tmp->flags & flag) |
2331 | sd = tmp; | |
c96d145e | 2332 | } |
476d139c | 2333 | |
039a1c41 PZ |
2334 | if (sd) |
2335 | update_shares(sd); | |
2336 | ||
476d139c | 2337 | while (sd) { |
476d139c | 2338 | struct sched_group *group; |
1a848870 SS |
2339 | int new_cpu, weight; |
2340 | ||
2341 | if (!(sd->flags & flag)) { | |
2342 | sd = sd->child; | |
2343 | continue; | |
2344 | } | |
476d139c | 2345 | |
476d139c | 2346 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2347 | if (!group) { |
2348 | sd = sd->child; | |
2349 | continue; | |
2350 | } | |
476d139c | 2351 | |
758b2cdc | 2352 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2353 | if (new_cpu == -1 || new_cpu == cpu) { |
2354 | /* Now try balancing at a lower domain level of cpu */ | |
2355 | sd = sd->child; | |
2356 | continue; | |
2357 | } | |
476d139c | 2358 | |
1a848870 | 2359 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2360 | cpu = new_cpu; |
758b2cdc | 2361 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2362 | sd = NULL; |
476d139c | 2363 | for_each_domain(cpu, tmp) { |
758b2cdc | 2364 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2365 | break; |
2366 | if (tmp->flags & flag) | |
2367 | sd = tmp; | |
2368 | } | |
2369 | /* while loop will break here if sd == NULL */ | |
2370 | } | |
2371 | ||
2372 | return cpu; | |
2373 | } | |
2374 | ||
2375 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2376 | |
0793a61d TG |
2377 | /** |
2378 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2379 | * @p: the task to evaluate | |
2380 | * @func: the function to be called | |
2381 | * @info: the function call argument | |
2382 | * | |
2383 | * Calls the function @func when the task is currently running. This might | |
2384 | * be on the current CPU, which just calls the function directly | |
2385 | */ | |
2386 | void task_oncpu_function_call(struct task_struct *p, | |
2387 | void (*func) (void *info), void *info) | |
2388 | { | |
2389 | int cpu; | |
2390 | ||
2391 | preempt_disable(); | |
2392 | cpu = task_cpu(p); | |
2393 | if (task_curr(p)) | |
2394 | smp_call_function_single(cpu, func, info, 1); | |
2395 | preempt_enable(); | |
2396 | } | |
2397 | ||
1da177e4 LT |
2398 | /*** |
2399 | * try_to_wake_up - wake up a thread | |
2400 | * @p: the to-be-woken-up thread | |
2401 | * @state: the mask of task states that can be woken | |
2402 | * @sync: do a synchronous wakeup? | |
2403 | * | |
2404 | * Put it on the run-queue if it's not already there. The "current" | |
2405 | * thread is always on the run-queue (except when the actual | |
2406 | * re-schedule is in progress), and as such you're allowed to do | |
2407 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2408 | * runnable without the overhead of this. | |
2409 | * | |
2410 | * returns failure only if the task is already active. | |
2411 | */ | |
36c8b586 | 2412 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2413 | { |
cc367732 | 2414 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2415 | unsigned long flags; |
2416 | long old_state; | |
70b97a7f | 2417 | struct rq *rq; |
1da177e4 | 2418 | |
b85d0667 IM |
2419 | if (!sched_feat(SYNC_WAKEUPS)) |
2420 | sync = 0; | |
2421 | ||
2398f2c6 | 2422 | #ifdef CONFIG_SMP |
57310a98 | 2423 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2424 | struct sched_domain *sd; |
2425 | ||
2426 | this_cpu = raw_smp_processor_id(); | |
2427 | cpu = task_cpu(p); | |
2428 | ||
2429 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2430 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2431 | update_shares(sd); |
2432 | break; | |
2433 | } | |
2434 | } | |
2435 | } | |
2436 | #endif | |
2437 | ||
04e2f174 | 2438 | smp_wmb(); |
1da177e4 | 2439 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2440 | update_rq_clock(rq); |
1da177e4 LT |
2441 | old_state = p->state; |
2442 | if (!(old_state & state)) | |
2443 | goto out; | |
2444 | ||
dd41f596 | 2445 | if (p->se.on_rq) |
1da177e4 LT |
2446 | goto out_running; |
2447 | ||
2448 | cpu = task_cpu(p); | |
cc367732 | 2449 | orig_cpu = cpu; |
1da177e4 LT |
2450 | this_cpu = smp_processor_id(); |
2451 | ||
2452 | #ifdef CONFIG_SMP | |
2453 | if (unlikely(task_running(rq, p))) | |
2454 | goto out_activate; | |
2455 | ||
5d2f5a61 DA |
2456 | cpu = p->sched_class->select_task_rq(p, sync); |
2457 | if (cpu != orig_cpu) { | |
2458 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2459 | task_rq_unlock(rq, &flags); |
2460 | /* might preempt at this point */ | |
2461 | rq = task_rq_lock(p, &flags); | |
2462 | old_state = p->state; | |
2463 | if (!(old_state & state)) | |
2464 | goto out; | |
dd41f596 | 2465 | if (p->se.on_rq) |
1da177e4 LT |
2466 | goto out_running; |
2467 | ||
2468 | this_cpu = smp_processor_id(); | |
2469 | cpu = task_cpu(p); | |
2470 | } | |
2471 | ||
e7693a36 GH |
2472 | #ifdef CONFIG_SCHEDSTATS |
2473 | schedstat_inc(rq, ttwu_count); | |
2474 | if (cpu == this_cpu) | |
2475 | schedstat_inc(rq, ttwu_local); | |
2476 | else { | |
2477 | struct sched_domain *sd; | |
2478 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2479 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2480 | schedstat_inc(sd, ttwu_wake_remote); |
2481 | break; | |
2482 | } | |
2483 | } | |
2484 | } | |
6d6bc0ad | 2485 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2486 | |
1da177e4 LT |
2487 | out_activate: |
2488 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2489 | schedstat_inc(p, se.nr_wakeups); |
2490 | if (sync) | |
2491 | schedstat_inc(p, se.nr_wakeups_sync); | |
2492 | if (orig_cpu != cpu) | |
2493 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2494 | if (cpu == this_cpu) | |
2495 | schedstat_inc(p, se.nr_wakeups_local); | |
2496 | else | |
2497 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2498 | activate_task(rq, p, 1); |
1da177e4 LT |
2499 | success = 1; |
2500 | ||
831451ac PZ |
2501 | /* |
2502 | * Only attribute actual wakeups done by this task. | |
2503 | */ | |
2504 | if (!in_interrupt()) { | |
2505 | struct sched_entity *se = ¤t->se; | |
2506 | u64 sample = se->sum_exec_runtime; | |
2507 | ||
2508 | if (se->last_wakeup) | |
2509 | sample -= se->last_wakeup; | |
2510 | else | |
2511 | sample -= se->start_runtime; | |
2512 | update_avg(&se->avg_wakeup, sample); | |
2513 | ||
2514 | se->last_wakeup = se->sum_exec_runtime; | |
2515 | } | |
2516 | ||
1da177e4 | 2517 | out_running: |
468a15bb | 2518 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2519 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2520 | |
1da177e4 | 2521 | p->state = TASK_RUNNING; |
9a897c5a SR |
2522 | #ifdef CONFIG_SMP |
2523 | if (p->sched_class->task_wake_up) | |
2524 | p->sched_class->task_wake_up(rq, p); | |
2525 | #endif | |
1da177e4 LT |
2526 | out: |
2527 | task_rq_unlock(rq, &flags); | |
2528 | ||
2529 | return success; | |
2530 | } | |
2531 | ||
50fa610a DH |
2532 | /** |
2533 | * wake_up_process - Wake up a specific process | |
2534 | * @p: The process to be woken up. | |
2535 | * | |
2536 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2537 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2538 | * running. | |
2539 | * | |
2540 | * It may be assumed that this function implies a write memory barrier before | |
2541 | * changing the task state if and only if any tasks are woken up. | |
2542 | */ | |
7ad5b3a5 | 2543 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2544 | { |
d9514f6c | 2545 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2546 | } |
1da177e4 LT |
2547 | EXPORT_SYMBOL(wake_up_process); |
2548 | ||
7ad5b3a5 | 2549 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2550 | { |
2551 | return try_to_wake_up(p, state, 0); | |
2552 | } | |
2553 | ||
1da177e4 LT |
2554 | /* |
2555 | * Perform scheduler related setup for a newly forked process p. | |
2556 | * p is forked by current. | |
dd41f596 IM |
2557 | * |
2558 | * __sched_fork() is basic setup used by init_idle() too: | |
2559 | */ | |
2560 | static void __sched_fork(struct task_struct *p) | |
2561 | { | |
dd41f596 IM |
2562 | p->se.exec_start = 0; |
2563 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2564 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2565 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2566 | p->se.last_wakeup = 0; |
2567 | p->se.avg_overlap = 0; | |
831451ac PZ |
2568 | p->se.start_runtime = 0; |
2569 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2570 | |
2571 | #ifdef CONFIG_SCHEDSTATS | |
2572 | p->se.wait_start = 0; | |
dd41f596 IM |
2573 | p->se.sum_sleep_runtime = 0; |
2574 | p->se.sleep_start = 0; | |
dd41f596 IM |
2575 | p->se.block_start = 0; |
2576 | p->se.sleep_max = 0; | |
2577 | p->se.block_max = 0; | |
2578 | p->se.exec_max = 0; | |
eba1ed4b | 2579 | p->se.slice_max = 0; |
dd41f596 | 2580 | p->se.wait_max = 0; |
6cfb0d5d | 2581 | #endif |
476d139c | 2582 | |
fa717060 | 2583 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2584 | p->se.on_rq = 0; |
4a55bd5e | 2585 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2586 | |
e107be36 AK |
2587 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2588 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2589 | #endif | |
2590 | ||
1da177e4 LT |
2591 | /* |
2592 | * We mark the process as running here, but have not actually | |
2593 | * inserted it onto the runqueue yet. This guarantees that | |
2594 | * nobody will actually run it, and a signal or other external | |
2595 | * event cannot wake it up and insert it on the runqueue either. | |
2596 | */ | |
2597 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2598 | } |
2599 | ||
2600 | /* | |
2601 | * fork()/clone()-time setup: | |
2602 | */ | |
2603 | void sched_fork(struct task_struct *p, int clone_flags) | |
2604 | { | |
2605 | int cpu = get_cpu(); | |
2606 | ||
2607 | __sched_fork(p); | |
2608 | ||
2609 | #ifdef CONFIG_SMP | |
2610 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2611 | #endif | |
02e4bac2 | 2612 | set_task_cpu(p, cpu); |
b29739f9 IM |
2613 | |
2614 | /* | |
2615 | * Make sure we do not leak PI boosting priority to the child: | |
2616 | */ | |
2617 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2618 | if (!rt_prio(p->prio)) |
2619 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2620 | |
52f17b6c | 2621 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2622 | if (likely(sched_info_on())) |
52f17b6c | 2623 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2624 | #endif |
d6077cb8 | 2625 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2626 | p->oncpu = 0; |
2627 | #endif | |
1da177e4 | 2628 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2629 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2630 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2631 | #endif |
917b627d GH |
2632 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2633 | ||
476d139c | 2634 | put_cpu(); |
1da177e4 LT |
2635 | } |
2636 | ||
2637 | /* | |
2638 | * wake_up_new_task - wake up a newly created task for the first time. | |
2639 | * | |
2640 | * This function will do some initial scheduler statistics housekeeping | |
2641 | * that must be done for every newly created context, then puts the task | |
2642 | * on the runqueue and wakes it. | |
2643 | */ | |
7ad5b3a5 | 2644 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2645 | { |
2646 | unsigned long flags; | |
dd41f596 | 2647 | struct rq *rq; |
1da177e4 LT |
2648 | |
2649 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2650 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2651 | update_rq_clock(rq); |
1da177e4 LT |
2652 | |
2653 | p->prio = effective_prio(p); | |
2654 | ||
b9dca1e0 | 2655 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2656 | activate_task(rq, p, 0); |
1da177e4 | 2657 | } else { |
1da177e4 | 2658 | /* |
dd41f596 IM |
2659 | * Let the scheduling class do new task startup |
2660 | * management (if any): | |
1da177e4 | 2661 | */ |
ee0827d8 | 2662 | p->sched_class->task_new(rq, p); |
c09595f6 | 2663 | inc_nr_running(rq); |
1da177e4 | 2664 | } |
c71dd42d | 2665 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2666 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2667 | #ifdef CONFIG_SMP |
2668 | if (p->sched_class->task_wake_up) | |
2669 | p->sched_class->task_wake_up(rq, p); | |
2670 | #endif | |
dd41f596 | 2671 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2672 | } |
2673 | ||
e107be36 AK |
2674 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2675 | ||
2676 | /** | |
80dd99b3 | 2677 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2678 | * @notifier: notifier struct to register |
e107be36 AK |
2679 | */ |
2680 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2681 | { | |
2682 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2683 | } | |
2684 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2685 | ||
2686 | /** | |
2687 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2688 | * @notifier: notifier struct to unregister |
e107be36 AK |
2689 | * |
2690 | * This is safe to call from within a preemption notifier. | |
2691 | */ | |
2692 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2693 | { | |
2694 | hlist_del(¬ifier->link); | |
2695 | } | |
2696 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2697 | ||
2698 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2699 | { | |
2700 | struct preempt_notifier *notifier; | |
2701 | struct hlist_node *node; | |
2702 | ||
2703 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2704 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2705 | } | |
2706 | ||
2707 | static void | |
2708 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2709 | struct task_struct *next) | |
2710 | { | |
2711 | struct preempt_notifier *notifier; | |
2712 | struct hlist_node *node; | |
2713 | ||
2714 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2715 | notifier->ops->sched_out(notifier, next); | |
2716 | } | |
2717 | ||
6d6bc0ad | 2718 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2719 | |
2720 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2721 | { | |
2722 | } | |
2723 | ||
2724 | static void | |
2725 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2726 | struct task_struct *next) | |
2727 | { | |
2728 | } | |
2729 | ||
6d6bc0ad | 2730 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2731 | |
4866cde0 NP |
2732 | /** |
2733 | * prepare_task_switch - prepare to switch tasks | |
2734 | * @rq: the runqueue preparing to switch | |
421cee29 | 2735 | * @prev: the current task that is being switched out |
4866cde0 NP |
2736 | * @next: the task we are going to switch to. |
2737 | * | |
2738 | * This is called with the rq lock held and interrupts off. It must | |
2739 | * be paired with a subsequent finish_task_switch after the context | |
2740 | * switch. | |
2741 | * | |
2742 | * prepare_task_switch sets up locking and calls architecture specific | |
2743 | * hooks. | |
2744 | */ | |
e107be36 AK |
2745 | static inline void |
2746 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2747 | struct task_struct *next) | |
4866cde0 | 2748 | { |
e107be36 | 2749 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2750 | prepare_lock_switch(rq, next); |
2751 | prepare_arch_switch(next); | |
2752 | } | |
2753 | ||
1da177e4 LT |
2754 | /** |
2755 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2756 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2757 | * @prev: the thread we just switched away from. |
2758 | * | |
4866cde0 NP |
2759 | * finish_task_switch must be called after the context switch, paired |
2760 | * with a prepare_task_switch call before the context switch. | |
2761 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2762 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2763 | * |
2764 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2765 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2766 | * with the lock held can cause deadlocks; see schedule() for |
2767 | * details.) | |
2768 | */ | |
a9957449 | 2769 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2770 | __releases(rq->lock) |
2771 | { | |
1da177e4 | 2772 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2773 | long prev_state; |
967fc046 GH |
2774 | #ifdef CONFIG_SMP |
2775 | int post_schedule = 0; | |
2776 | ||
2777 | if (current->sched_class->needs_post_schedule) | |
2778 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2779 | #endif | |
1da177e4 LT |
2780 | |
2781 | rq->prev_mm = NULL; | |
2782 | ||
2783 | /* | |
2784 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2785 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2786 | * schedule one last time. The schedule call will never return, and |
2787 | * the scheduled task must drop that reference. | |
c394cc9f | 2788 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2789 | * still held, otherwise prev could be scheduled on another cpu, die |
2790 | * there before we look at prev->state, and then the reference would | |
2791 | * be dropped twice. | |
2792 | * Manfred Spraul <manfred@colorfullife.com> | |
2793 | */ | |
55a101f8 | 2794 | prev_state = prev->state; |
4866cde0 | 2795 | finish_arch_switch(prev); |
0793a61d | 2796 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2797 | finish_lock_switch(rq, prev); |
9a897c5a | 2798 | #ifdef CONFIG_SMP |
967fc046 | 2799 | if (post_schedule) |
9a897c5a SR |
2800 | current->sched_class->post_schedule(rq); |
2801 | #endif | |
e8fa1362 | 2802 | |
e107be36 | 2803 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2804 | if (mm) |
2805 | mmdrop(mm); | |
c394cc9f | 2806 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2807 | /* |
2808 | * Remove function-return probe instances associated with this | |
2809 | * task and put them back on the free list. | |
9761eea8 | 2810 | */ |
c6fd91f0 | 2811 | kprobe_flush_task(prev); |
1da177e4 | 2812 | put_task_struct(prev); |
c6fd91f0 | 2813 | } |
1da177e4 LT |
2814 | } |
2815 | ||
2816 | /** | |
2817 | * schedule_tail - first thing a freshly forked thread must call. | |
2818 | * @prev: the thread we just switched away from. | |
2819 | */ | |
36c8b586 | 2820 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2821 | __releases(rq->lock) |
2822 | { | |
70b97a7f IM |
2823 | struct rq *rq = this_rq(); |
2824 | ||
4866cde0 NP |
2825 | finish_task_switch(rq, prev); |
2826 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2827 | /* In this case, finish_task_switch does not reenable preemption */ | |
2828 | preempt_enable(); | |
2829 | #endif | |
1da177e4 | 2830 | if (current->set_child_tid) |
b488893a | 2831 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2832 | } |
2833 | ||
2834 | /* | |
2835 | * context_switch - switch to the new MM and the new | |
2836 | * thread's register state. | |
2837 | */ | |
dd41f596 | 2838 | static inline void |
70b97a7f | 2839 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2840 | struct task_struct *next) |
1da177e4 | 2841 | { |
dd41f596 | 2842 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2843 | |
e107be36 | 2844 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2845 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2846 | mm = next->mm; |
2847 | oldmm = prev->active_mm; | |
9226d125 ZA |
2848 | /* |
2849 | * For paravirt, this is coupled with an exit in switch_to to | |
2850 | * combine the page table reload and the switch backend into | |
2851 | * one hypercall. | |
2852 | */ | |
224101ed | 2853 | arch_start_context_switch(prev); |
9226d125 | 2854 | |
dd41f596 | 2855 | if (unlikely(!mm)) { |
1da177e4 LT |
2856 | next->active_mm = oldmm; |
2857 | atomic_inc(&oldmm->mm_count); | |
2858 | enter_lazy_tlb(oldmm, next); | |
2859 | } else | |
2860 | switch_mm(oldmm, mm, next); | |
2861 | ||
dd41f596 | 2862 | if (unlikely(!prev->mm)) { |
1da177e4 | 2863 | prev->active_mm = NULL; |
1da177e4 LT |
2864 | rq->prev_mm = oldmm; |
2865 | } | |
3a5f5e48 IM |
2866 | /* |
2867 | * Since the runqueue lock will be released by the next | |
2868 | * task (which is an invalid locking op but in the case | |
2869 | * of the scheduler it's an obvious special-case), so we | |
2870 | * do an early lockdep release here: | |
2871 | */ | |
2872 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2873 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2874 | #endif |
1da177e4 LT |
2875 | |
2876 | /* Here we just switch the register state and the stack. */ | |
2877 | switch_to(prev, next, prev); | |
2878 | ||
dd41f596 IM |
2879 | barrier(); |
2880 | /* | |
2881 | * this_rq must be evaluated again because prev may have moved | |
2882 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2883 | * frame will be invalid. | |
2884 | */ | |
2885 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2886 | } |
2887 | ||
2888 | /* | |
2889 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2890 | * | |
2891 | * externally visible scheduler statistics: current number of runnable | |
2892 | * threads, current number of uninterruptible-sleeping threads, total | |
2893 | * number of context switches performed since bootup. | |
2894 | */ | |
2895 | unsigned long nr_running(void) | |
2896 | { | |
2897 | unsigned long i, sum = 0; | |
2898 | ||
2899 | for_each_online_cpu(i) | |
2900 | sum += cpu_rq(i)->nr_running; | |
2901 | ||
2902 | return sum; | |
2903 | } | |
2904 | ||
2905 | unsigned long nr_uninterruptible(void) | |
2906 | { | |
2907 | unsigned long i, sum = 0; | |
2908 | ||
0a945022 | 2909 | for_each_possible_cpu(i) |
1da177e4 LT |
2910 | sum += cpu_rq(i)->nr_uninterruptible; |
2911 | ||
2912 | /* | |
2913 | * Since we read the counters lockless, it might be slightly | |
2914 | * inaccurate. Do not allow it to go below zero though: | |
2915 | */ | |
2916 | if (unlikely((long)sum < 0)) | |
2917 | sum = 0; | |
2918 | ||
2919 | return sum; | |
2920 | } | |
2921 | ||
2922 | unsigned long long nr_context_switches(void) | |
2923 | { | |
cc94abfc SR |
2924 | int i; |
2925 | unsigned long long sum = 0; | |
1da177e4 | 2926 | |
0a945022 | 2927 | for_each_possible_cpu(i) |
1da177e4 LT |
2928 | sum += cpu_rq(i)->nr_switches; |
2929 | ||
2930 | return sum; | |
2931 | } | |
2932 | ||
2933 | unsigned long nr_iowait(void) | |
2934 | { | |
2935 | unsigned long i, sum = 0; | |
2936 | ||
0a945022 | 2937 | for_each_possible_cpu(i) |
1da177e4 LT |
2938 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2939 | ||
2940 | return sum; | |
2941 | } | |
2942 | ||
dce48a84 TG |
2943 | /* Variables and functions for calc_load */ |
2944 | static atomic_long_t calc_load_tasks; | |
2945 | static unsigned long calc_load_update; | |
2946 | unsigned long avenrun[3]; | |
2947 | EXPORT_SYMBOL(avenrun); | |
2948 | ||
2d02494f TG |
2949 | /** |
2950 | * get_avenrun - get the load average array | |
2951 | * @loads: pointer to dest load array | |
2952 | * @offset: offset to add | |
2953 | * @shift: shift count to shift the result left | |
2954 | * | |
2955 | * These values are estimates at best, so no need for locking. | |
2956 | */ | |
2957 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2958 | { | |
2959 | loads[0] = (avenrun[0] + offset) << shift; | |
2960 | loads[1] = (avenrun[1] + offset) << shift; | |
2961 | loads[2] = (avenrun[2] + offset) << shift; | |
2962 | } | |
2963 | ||
dce48a84 TG |
2964 | static unsigned long |
2965 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2966 | { |
dce48a84 TG |
2967 | load *= exp; |
2968 | load += active * (FIXED_1 - exp); | |
2969 | return load >> FSHIFT; | |
2970 | } | |
db1b1fef | 2971 | |
dce48a84 TG |
2972 | /* |
2973 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2974 | * CPUs have updated calc_load_tasks. | |
2975 | */ | |
2976 | void calc_global_load(void) | |
2977 | { | |
2978 | unsigned long upd = calc_load_update + 10; | |
2979 | long active; | |
2980 | ||
2981 | if (time_before(jiffies, upd)) | |
2982 | return; | |
db1b1fef | 2983 | |
dce48a84 TG |
2984 | active = atomic_long_read(&calc_load_tasks); |
2985 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2986 | |
dce48a84 TG |
2987 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2988 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2989 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2990 | ||
2991 | calc_load_update += LOAD_FREQ; | |
2992 | } | |
2993 | ||
2994 | /* | |
2995 | * Either called from update_cpu_load() or from a cpu going idle | |
2996 | */ | |
2997 | static void calc_load_account_active(struct rq *this_rq) | |
2998 | { | |
2999 | long nr_active, delta; | |
3000 | ||
3001 | nr_active = this_rq->nr_running; | |
3002 | nr_active += (long) this_rq->nr_uninterruptible; | |
3003 | ||
3004 | if (nr_active != this_rq->calc_load_active) { | |
3005 | delta = nr_active - this_rq->calc_load_active; | |
3006 | this_rq->calc_load_active = nr_active; | |
3007 | atomic_long_add(delta, &calc_load_tasks); | |
3008 | } | |
db1b1fef JS |
3009 | } |
3010 | ||
23a185ca PM |
3011 | /* |
3012 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3013 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3014 | */ | |
23a185ca PM |
3015 | u64 cpu_nr_migrations(int cpu) |
3016 | { | |
3017 | return cpu_rq(cpu)->nr_migrations_in; | |
3018 | } | |
3019 | ||
48f24c4d | 3020 | /* |
dd41f596 IM |
3021 | * Update rq->cpu_load[] statistics. This function is usually called every |
3022 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3023 | */ |
dd41f596 | 3024 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3025 | { |
495eca49 | 3026 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3027 | int i, scale; |
3028 | ||
3029 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3030 | |
3031 | /* Update our load: */ | |
3032 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3033 | unsigned long old_load, new_load; | |
3034 | ||
3035 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3036 | ||
3037 | old_load = this_rq->cpu_load[i]; | |
3038 | new_load = this_load; | |
a25707f3 IM |
3039 | /* |
3040 | * Round up the averaging division if load is increasing. This | |
3041 | * prevents us from getting stuck on 9 if the load is 10, for | |
3042 | * example. | |
3043 | */ | |
3044 | if (new_load > old_load) | |
3045 | new_load += scale-1; | |
dd41f596 IM |
3046 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3047 | } | |
dce48a84 TG |
3048 | |
3049 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3050 | this_rq->calc_load_update += LOAD_FREQ; | |
3051 | calc_load_account_active(this_rq); | |
3052 | } | |
48f24c4d IM |
3053 | } |
3054 | ||
dd41f596 IM |
3055 | #ifdef CONFIG_SMP |
3056 | ||
1da177e4 LT |
3057 | /* |
3058 | * double_rq_lock - safely lock two runqueues | |
3059 | * | |
3060 | * Note this does not disable interrupts like task_rq_lock, | |
3061 | * you need to do so manually before calling. | |
3062 | */ | |
70b97a7f | 3063 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3064 | __acquires(rq1->lock) |
3065 | __acquires(rq2->lock) | |
3066 | { | |
054b9108 | 3067 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3068 | if (rq1 == rq2) { |
3069 | spin_lock(&rq1->lock); | |
3070 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3071 | } else { | |
c96d145e | 3072 | if (rq1 < rq2) { |
1da177e4 | 3073 | spin_lock(&rq1->lock); |
5e710e37 | 3074 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3075 | } else { |
3076 | spin_lock(&rq2->lock); | |
5e710e37 | 3077 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3078 | } |
3079 | } | |
6e82a3be IM |
3080 | update_rq_clock(rq1); |
3081 | update_rq_clock(rq2); | |
1da177e4 LT |
3082 | } |
3083 | ||
3084 | /* | |
3085 | * double_rq_unlock - safely unlock two runqueues | |
3086 | * | |
3087 | * Note this does not restore interrupts like task_rq_unlock, | |
3088 | * you need to do so manually after calling. | |
3089 | */ | |
70b97a7f | 3090 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3091 | __releases(rq1->lock) |
3092 | __releases(rq2->lock) | |
3093 | { | |
3094 | spin_unlock(&rq1->lock); | |
3095 | if (rq1 != rq2) | |
3096 | spin_unlock(&rq2->lock); | |
3097 | else | |
3098 | __release(rq2->lock); | |
3099 | } | |
3100 | ||
1da177e4 LT |
3101 | /* |
3102 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3103 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3104 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3105 | * the cpu_allowed mask is restored. |
3106 | */ | |
36c8b586 | 3107 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3108 | { |
70b97a7f | 3109 | struct migration_req req; |
1da177e4 | 3110 | unsigned long flags; |
70b97a7f | 3111 | struct rq *rq; |
1da177e4 LT |
3112 | |
3113 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3114 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3115 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3116 | goto out; |
3117 | ||
3118 | /* force the process onto the specified CPU */ | |
3119 | if (migrate_task(p, dest_cpu, &req)) { | |
3120 | /* Need to wait for migration thread (might exit: take ref). */ | |
3121 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3122 | |
1da177e4 LT |
3123 | get_task_struct(mt); |
3124 | task_rq_unlock(rq, &flags); | |
3125 | wake_up_process(mt); | |
3126 | put_task_struct(mt); | |
3127 | wait_for_completion(&req.done); | |
36c8b586 | 3128 | |
1da177e4 LT |
3129 | return; |
3130 | } | |
3131 | out: | |
3132 | task_rq_unlock(rq, &flags); | |
3133 | } | |
3134 | ||
3135 | /* | |
476d139c NP |
3136 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3137 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3138 | */ |
3139 | void sched_exec(void) | |
3140 | { | |
1da177e4 | 3141 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3142 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3143 | put_cpu(); |
476d139c NP |
3144 | if (new_cpu != this_cpu) |
3145 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3146 | } |
3147 | ||
3148 | /* | |
3149 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3150 | * Both runqueues must be locked. | |
3151 | */ | |
dd41f596 IM |
3152 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3153 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3154 | { |
2e1cb74a | 3155 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3156 | set_task_cpu(p, this_cpu); |
dd41f596 | 3157 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3158 | /* |
3159 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3160 | * to be always true for them. | |
3161 | */ | |
15afe09b | 3162 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3163 | } |
3164 | ||
3165 | /* | |
3166 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3167 | */ | |
858119e1 | 3168 | static |
70b97a7f | 3169 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3170 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3171 | int *all_pinned) |
1da177e4 | 3172 | { |
708dc512 | 3173 | int tsk_cache_hot = 0; |
1da177e4 LT |
3174 | /* |
3175 | * We do not migrate tasks that are: | |
3176 | * 1) running (obviously), or | |
3177 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3178 | * 3) are cache-hot on their current CPU. | |
3179 | */ | |
96f874e2 | 3180 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3181 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3182 | return 0; |
cc367732 | 3183 | } |
81026794 NP |
3184 | *all_pinned = 0; |
3185 | ||
cc367732 IM |
3186 | if (task_running(rq, p)) { |
3187 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3188 | return 0; |
cc367732 | 3189 | } |
1da177e4 | 3190 | |
da84d961 IM |
3191 | /* |
3192 | * Aggressive migration if: | |
3193 | * 1) task is cache cold, or | |
3194 | * 2) too many balance attempts have failed. | |
3195 | */ | |
3196 | ||
708dc512 LH |
3197 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3198 | if (!tsk_cache_hot || | |
3199 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3200 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3201 | if (tsk_cache_hot) { |
da84d961 | 3202 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3203 | schedstat_inc(p, se.nr_forced_migrations); |
3204 | } | |
da84d961 IM |
3205 | #endif |
3206 | return 1; | |
3207 | } | |
3208 | ||
708dc512 | 3209 | if (tsk_cache_hot) { |
cc367732 | 3210 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3211 | return 0; |
cc367732 | 3212 | } |
1da177e4 LT |
3213 | return 1; |
3214 | } | |
3215 | ||
e1d1484f PW |
3216 | static unsigned long |
3217 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3218 | unsigned long max_load_move, struct sched_domain *sd, | |
3219 | enum cpu_idle_type idle, int *all_pinned, | |
3220 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3221 | { |
051c6764 | 3222 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3223 | struct task_struct *p; |
3224 | long rem_load_move = max_load_move; | |
1da177e4 | 3225 | |
e1d1484f | 3226 | if (max_load_move == 0) |
1da177e4 LT |
3227 | goto out; |
3228 | ||
81026794 NP |
3229 | pinned = 1; |
3230 | ||
1da177e4 | 3231 | /* |
dd41f596 | 3232 | * Start the load-balancing iterator: |
1da177e4 | 3233 | */ |
dd41f596 IM |
3234 | p = iterator->start(iterator->arg); |
3235 | next: | |
b82d9fdd | 3236 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3237 | goto out; |
051c6764 PZ |
3238 | |
3239 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3240 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3241 | p = iterator->next(iterator->arg); |
3242 | goto next; | |
1da177e4 LT |
3243 | } |
3244 | ||
dd41f596 | 3245 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3246 | pulled++; |
dd41f596 | 3247 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3248 | |
7e96fa58 GH |
3249 | #ifdef CONFIG_PREEMPT |
3250 | /* | |
3251 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3252 | * will stop after the first task is pulled to minimize the critical | |
3253 | * section. | |
3254 | */ | |
3255 | if (idle == CPU_NEWLY_IDLE) | |
3256 | goto out; | |
3257 | #endif | |
3258 | ||
2dd73a4f | 3259 | /* |
b82d9fdd | 3260 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3261 | */ |
e1d1484f | 3262 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3263 | if (p->prio < *this_best_prio) |
3264 | *this_best_prio = p->prio; | |
dd41f596 IM |
3265 | p = iterator->next(iterator->arg); |
3266 | goto next; | |
1da177e4 LT |
3267 | } |
3268 | out: | |
3269 | /* | |
e1d1484f | 3270 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3271 | * so we can safely collect pull_task() stats here rather than |
3272 | * inside pull_task(). | |
3273 | */ | |
3274 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3275 | |
3276 | if (all_pinned) | |
3277 | *all_pinned = pinned; | |
e1d1484f PW |
3278 | |
3279 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3280 | } |
3281 | ||
dd41f596 | 3282 | /* |
43010659 PW |
3283 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3284 | * this_rq, as part of a balancing operation within domain "sd". | |
3285 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3286 | * |
3287 | * Called with both runqueues locked. | |
3288 | */ | |
3289 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3290 | unsigned long max_load_move, |
dd41f596 IM |
3291 | struct sched_domain *sd, enum cpu_idle_type idle, |
3292 | int *all_pinned) | |
3293 | { | |
5522d5d5 | 3294 | const struct sched_class *class = sched_class_highest; |
43010659 | 3295 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3296 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3297 | |
3298 | do { | |
43010659 PW |
3299 | total_load_moved += |
3300 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3301 | max_load_move - total_load_moved, |
a4ac01c3 | 3302 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3303 | class = class->next; |
c4acb2c0 | 3304 | |
7e96fa58 GH |
3305 | #ifdef CONFIG_PREEMPT |
3306 | /* | |
3307 | * NEWIDLE balancing is a source of latency, so preemptible | |
3308 | * kernels will stop after the first task is pulled to minimize | |
3309 | * the critical section. | |
3310 | */ | |
c4acb2c0 GH |
3311 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3312 | break; | |
7e96fa58 | 3313 | #endif |
43010659 | 3314 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3315 | |
43010659 PW |
3316 | return total_load_moved > 0; |
3317 | } | |
3318 | ||
e1d1484f PW |
3319 | static int |
3320 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3321 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3322 | struct rq_iterator *iterator) | |
3323 | { | |
3324 | struct task_struct *p = iterator->start(iterator->arg); | |
3325 | int pinned = 0; | |
3326 | ||
3327 | while (p) { | |
3328 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3329 | pull_task(busiest, p, this_rq, this_cpu); | |
3330 | /* | |
3331 | * Right now, this is only the second place pull_task() | |
3332 | * is called, so we can safely collect pull_task() | |
3333 | * stats here rather than inside pull_task(). | |
3334 | */ | |
3335 | schedstat_inc(sd, lb_gained[idle]); | |
3336 | ||
3337 | return 1; | |
3338 | } | |
3339 | p = iterator->next(iterator->arg); | |
3340 | } | |
3341 | ||
3342 | return 0; | |
3343 | } | |
3344 | ||
43010659 PW |
3345 | /* |
3346 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3347 | * part of active balancing operations within "domain". | |
3348 | * Returns 1 if successful and 0 otherwise. | |
3349 | * | |
3350 | * Called with both runqueues locked. | |
3351 | */ | |
3352 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3353 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3354 | { | |
5522d5d5 | 3355 | const struct sched_class *class; |
43010659 PW |
3356 | |
3357 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3358 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3359 | return 1; |
3360 | ||
3361 | return 0; | |
dd41f596 | 3362 | } |
67bb6c03 | 3363 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3364 | /* |
222d656d GS |
3365 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3366 | * during load balancing. | |
1da177e4 | 3367 | */ |
222d656d GS |
3368 | struct sd_lb_stats { |
3369 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3370 | struct sched_group *this; /* Local group in this sd */ | |
3371 | unsigned long total_load; /* Total load of all groups in sd */ | |
3372 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3373 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3374 | ||
3375 | /** Statistics of this group */ | |
3376 | unsigned long this_load; | |
3377 | unsigned long this_load_per_task; | |
3378 | unsigned long this_nr_running; | |
3379 | ||
3380 | /* Statistics of the busiest group */ | |
3381 | unsigned long max_load; | |
3382 | unsigned long busiest_load_per_task; | |
3383 | unsigned long busiest_nr_running; | |
3384 | ||
3385 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3386 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3387 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3388 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3389 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3390 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3391 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3392 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3393 | #endif |
222d656d | 3394 | }; |
1da177e4 | 3395 | |
d5ac537e | 3396 | /* |
381be78f GS |
3397 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3398 | */ | |
3399 | struct sg_lb_stats { | |
3400 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3401 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3402 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3403 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3404 | unsigned long group_capacity; | |
3405 | int group_imb; /* Is there an imbalance in the group ? */ | |
3406 | }; | |
408ed066 | 3407 | |
67bb6c03 GS |
3408 | /** |
3409 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3410 | * @group: The group whose first cpu is to be returned. | |
3411 | */ | |
3412 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3413 | { | |
3414 | return cpumask_first(sched_group_cpus(group)); | |
3415 | } | |
3416 | ||
3417 | /** | |
3418 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3419 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3420 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3421 | */ | |
3422 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3423 | enum cpu_idle_type idle) | |
3424 | { | |
3425 | int load_idx; | |
3426 | ||
3427 | switch (idle) { | |
3428 | case CPU_NOT_IDLE: | |
7897986b | 3429 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3430 | break; |
3431 | ||
3432 | case CPU_NEWLY_IDLE: | |
7897986b | 3433 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3434 | break; |
3435 | default: | |
7897986b | 3436 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3437 | break; |
3438 | } | |
1da177e4 | 3439 | |
67bb6c03 GS |
3440 | return load_idx; |
3441 | } | |
1da177e4 | 3442 | |
1da177e4 | 3443 | |
c071df18 GS |
3444 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3445 | /** | |
3446 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3447 | * the given sched_domain, during load balancing. | |
3448 | * | |
3449 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3450 | * @sds: Variable containing the statistics for sd. | |
3451 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3452 | */ | |
3453 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3454 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3455 | { | |
3456 | /* | |
3457 | * Busy processors will not participate in power savings | |
3458 | * balance. | |
3459 | */ | |
3460 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3461 | sds->power_savings_balance = 0; | |
3462 | else { | |
3463 | sds->power_savings_balance = 1; | |
3464 | sds->min_nr_running = ULONG_MAX; | |
3465 | sds->leader_nr_running = 0; | |
3466 | } | |
3467 | } | |
783609c6 | 3468 | |
c071df18 GS |
3469 | /** |
3470 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3471 | * sched_domain while performing load balancing. | |
3472 | * | |
3473 | * @group: sched_group belonging to the sched_domain under consideration. | |
3474 | * @sds: Variable containing the statistics of the sched_domain | |
3475 | * @local_group: Does group contain the CPU for which we're performing | |
3476 | * load balancing ? | |
3477 | * @sgs: Variable containing the statistics of the group. | |
3478 | */ | |
3479 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3480 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3481 | { | |
408ed066 | 3482 | |
c071df18 GS |
3483 | if (!sds->power_savings_balance) |
3484 | return; | |
1da177e4 | 3485 | |
c071df18 GS |
3486 | /* |
3487 | * If the local group is idle or completely loaded | |
3488 | * no need to do power savings balance at this domain | |
3489 | */ | |
3490 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3491 | !sds->this_nr_running)) | |
3492 | sds->power_savings_balance = 0; | |
2dd73a4f | 3493 | |
c071df18 GS |
3494 | /* |
3495 | * If a group is already running at full capacity or idle, | |
3496 | * don't include that group in power savings calculations | |
3497 | */ | |
3498 | if (!sds->power_savings_balance || | |
3499 | sgs->sum_nr_running >= sgs->group_capacity || | |
3500 | !sgs->sum_nr_running) | |
3501 | return; | |
5969fe06 | 3502 | |
c071df18 GS |
3503 | /* |
3504 | * Calculate the group which has the least non-idle load. | |
3505 | * This is the group from where we need to pick up the load | |
3506 | * for saving power | |
3507 | */ | |
3508 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3509 | (sgs->sum_nr_running == sds->min_nr_running && | |
3510 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3511 | sds->group_min = group; | |
3512 | sds->min_nr_running = sgs->sum_nr_running; | |
3513 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3514 | sgs->sum_nr_running; | |
3515 | } | |
783609c6 | 3516 | |
c071df18 GS |
3517 | /* |
3518 | * Calculate the group which is almost near its | |
3519 | * capacity but still has some space to pick up some load | |
3520 | * from other group and save more power | |
3521 | */ | |
3522 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3523 | return; | |
1da177e4 | 3524 | |
c071df18 GS |
3525 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3526 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3527 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3528 | sds->group_leader = group; | |
3529 | sds->leader_nr_running = sgs->sum_nr_running; | |
3530 | } | |
3531 | } | |
408ed066 | 3532 | |
c071df18 | 3533 | /** |
d5ac537e | 3534 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3535 | * @sds: Variable containing the statistics of the sched_domain |
3536 | * under consideration. | |
3537 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3538 | * @imbalance: Variable to store the imbalance. | |
3539 | * | |
d5ac537e RD |
3540 | * Description: |
3541 | * Check if we have potential to perform some power-savings balance. | |
3542 | * If yes, set the busiest group to be the least loaded group in the | |
3543 | * sched_domain, so that it's CPUs can be put to idle. | |
3544 | * | |
c071df18 GS |
3545 | * Returns 1 if there is potential to perform power-savings balance. |
3546 | * Else returns 0. | |
3547 | */ | |
3548 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3549 | int this_cpu, unsigned long *imbalance) | |
3550 | { | |
3551 | if (!sds->power_savings_balance) | |
3552 | return 0; | |
1da177e4 | 3553 | |
c071df18 GS |
3554 | if (sds->this != sds->group_leader || |
3555 | sds->group_leader == sds->group_min) | |
3556 | return 0; | |
783609c6 | 3557 | |
c071df18 GS |
3558 | *imbalance = sds->min_load_per_task; |
3559 | sds->busiest = sds->group_min; | |
1da177e4 | 3560 | |
c071df18 GS |
3561 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3562 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3563 | group_first_cpu(sds->group_leader); | |
3564 | } | |
3565 | ||
3566 | return 1; | |
1da177e4 | 3567 | |
c071df18 GS |
3568 | } |
3569 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3570 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3571 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3572 | { | |
3573 | return; | |
3574 | } | |
408ed066 | 3575 | |
c071df18 GS |
3576 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3577 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3578 | { | |
3579 | return; | |
3580 | } | |
3581 | ||
3582 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3583 | int this_cpu, unsigned long *imbalance) | |
3584 | { | |
3585 | return 0; | |
3586 | } | |
3587 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3588 | ||
3589 | ||
1f8c553d GS |
3590 | /** |
3591 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3592 | * @group: sched_group whose statistics are to be updated. | |
3593 | * @this_cpu: Cpu for which load balance is currently performed. | |
3594 | * @idle: Idle status of this_cpu | |
3595 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3596 | * @sd_idle: Idle status of the sched_domain containing group. | |
3597 | * @local_group: Does group contain this_cpu. | |
3598 | * @cpus: Set of cpus considered for load balancing. | |
3599 | * @balance: Should we balance. | |
3600 | * @sgs: variable to hold the statistics for this group. | |
3601 | */ | |
3602 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3603 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3604 | int local_group, const struct cpumask *cpus, | |
3605 | int *balance, struct sg_lb_stats *sgs) | |
3606 | { | |
3607 | unsigned long load, max_cpu_load, min_cpu_load; | |
3608 | int i; | |
3609 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3610 | unsigned long sum_avg_load_per_task; | |
3611 | unsigned long avg_load_per_task; | |
3612 | ||
3613 | if (local_group) | |
3614 | balance_cpu = group_first_cpu(group); | |
3615 | ||
3616 | /* Tally up the load of all CPUs in the group */ | |
3617 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3618 | max_cpu_load = 0; | |
3619 | min_cpu_load = ~0UL; | |
408ed066 | 3620 | |
1f8c553d GS |
3621 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3622 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3623 | |
1f8c553d GS |
3624 | if (*sd_idle && rq->nr_running) |
3625 | *sd_idle = 0; | |
5c45bf27 | 3626 | |
1f8c553d | 3627 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3628 | if (local_group) { |
1f8c553d GS |
3629 | if (idle_cpu(i) && !first_idle_cpu) { |
3630 | first_idle_cpu = 1; | |
3631 | balance_cpu = i; | |
3632 | } | |
3633 | ||
3634 | load = target_load(i, load_idx); | |
3635 | } else { | |
3636 | load = source_load(i, load_idx); | |
3637 | if (load > max_cpu_load) | |
3638 | max_cpu_load = load; | |
3639 | if (min_cpu_load > load) | |
3640 | min_cpu_load = load; | |
1da177e4 | 3641 | } |
5c45bf27 | 3642 | |
1f8c553d GS |
3643 | sgs->group_load += load; |
3644 | sgs->sum_nr_running += rq->nr_running; | |
3645 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3646 | |
1f8c553d GS |
3647 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3648 | } | |
5c45bf27 | 3649 | |
1f8c553d GS |
3650 | /* |
3651 | * First idle cpu or the first cpu(busiest) in this sched group | |
3652 | * is eligible for doing load balancing at this and above | |
3653 | * domains. In the newly idle case, we will allow all the cpu's | |
3654 | * to do the newly idle load balance. | |
3655 | */ | |
3656 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3657 | balance_cpu != this_cpu && balance) { | |
3658 | *balance = 0; | |
3659 | return; | |
3660 | } | |
5c45bf27 | 3661 | |
1f8c553d GS |
3662 | /* Adjust by relative CPU power of the group */ |
3663 | sgs->avg_load = sg_div_cpu_power(group, | |
3664 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3665 | |
1f8c553d GS |
3666 | |
3667 | /* | |
3668 | * Consider the group unbalanced when the imbalance is larger | |
3669 | * than the average weight of two tasks. | |
3670 | * | |
3671 | * APZ: with cgroup the avg task weight can vary wildly and | |
3672 | * might not be a suitable number - should we keep a | |
3673 | * normalized nr_running number somewhere that negates | |
3674 | * the hierarchy? | |
3675 | */ | |
3676 | avg_load_per_task = sg_div_cpu_power(group, | |
3677 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3678 | ||
3679 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3680 | sgs->group_imb = 1; | |
3681 | ||
3682 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3683 | ||
3684 | } | |
dd41f596 | 3685 | |
37abe198 GS |
3686 | /** |
3687 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3688 | * @sd: sched_domain whose statistics are to be updated. | |
3689 | * @this_cpu: Cpu for which load balance is currently performed. | |
3690 | * @idle: Idle status of this_cpu | |
3691 | * @sd_idle: Idle status of the sched_domain containing group. | |
3692 | * @cpus: Set of cpus considered for load balancing. | |
3693 | * @balance: Should we balance. | |
3694 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3695 | */ |
37abe198 GS |
3696 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3697 | enum cpu_idle_type idle, int *sd_idle, | |
3698 | const struct cpumask *cpus, int *balance, | |
3699 | struct sd_lb_stats *sds) | |
1da177e4 | 3700 | { |
222d656d | 3701 | struct sched_group *group = sd->groups; |
37abe198 | 3702 | struct sg_lb_stats sgs; |
222d656d GS |
3703 | int load_idx; |
3704 | ||
c071df18 | 3705 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3706 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3707 | |
3708 | do { | |
1da177e4 | 3709 | int local_group; |
1da177e4 | 3710 | |
758b2cdc RR |
3711 | local_group = cpumask_test_cpu(this_cpu, |
3712 | sched_group_cpus(group)); | |
381be78f | 3713 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3714 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3715 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3716 | |
37abe198 GS |
3717 | if (local_group && balance && !(*balance)) |
3718 | return; | |
783609c6 | 3719 | |
37abe198 GS |
3720 | sds->total_load += sgs.group_load; |
3721 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3722 | |
1da177e4 | 3723 | if (local_group) { |
37abe198 GS |
3724 | sds->this_load = sgs.avg_load; |
3725 | sds->this = group; | |
3726 | sds->this_nr_running = sgs.sum_nr_running; | |
3727 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3728 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3729 | (sgs.sum_nr_running > sgs.group_capacity || |
3730 | sgs.group_imb)) { | |
37abe198 GS |
3731 | sds->max_load = sgs.avg_load; |
3732 | sds->busiest = group; | |
3733 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3734 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3735 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3736 | } |
5c45bf27 | 3737 | |
c071df18 | 3738 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3739 | group = group->next; |
3740 | } while (group != sd->groups); | |
3741 | ||
37abe198 | 3742 | } |
1da177e4 | 3743 | |
2e6f44ae GS |
3744 | /** |
3745 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3746 | * amongst the groups of a sched_domain, during |
3747 | * load balancing. | |
2e6f44ae GS |
3748 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3749 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3750 | * @imbalance: Variable to store the imbalance. | |
3751 | */ | |
3752 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3753 | int this_cpu, unsigned long *imbalance) | |
3754 | { | |
3755 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3756 | unsigned int imbn = 2; | |
3757 | ||
3758 | if (sds->this_nr_running) { | |
3759 | sds->this_load_per_task /= sds->this_nr_running; | |
3760 | if (sds->busiest_load_per_task > | |
3761 | sds->this_load_per_task) | |
3762 | imbn = 1; | |
3763 | } else | |
3764 | sds->this_load_per_task = | |
3765 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3766 | |
2e6f44ae GS |
3767 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3768 | sds->busiest_load_per_task * imbn) { | |
3769 | *imbalance = sds->busiest_load_per_task; | |
3770 | return; | |
3771 | } | |
908a7c1b | 3772 | |
1da177e4 | 3773 | /* |
2e6f44ae GS |
3774 | * OK, we don't have enough imbalance to justify moving tasks, |
3775 | * however we may be able to increase total CPU power used by | |
3776 | * moving them. | |
1da177e4 | 3777 | */ |
2dd73a4f | 3778 | |
2e6f44ae GS |
3779 | pwr_now += sds->busiest->__cpu_power * |
3780 | min(sds->busiest_load_per_task, sds->max_load); | |
3781 | pwr_now += sds->this->__cpu_power * | |
3782 | min(sds->this_load_per_task, sds->this_load); | |
3783 | pwr_now /= SCHED_LOAD_SCALE; | |
3784 | ||
3785 | /* Amount of load we'd subtract */ | |
3786 | tmp = sg_div_cpu_power(sds->busiest, | |
3787 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3788 | if (sds->max_load > tmp) | |
3789 | pwr_move += sds->busiest->__cpu_power * | |
3790 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3791 | ||
3792 | /* Amount of load we'd add */ | |
3793 | if (sds->max_load * sds->busiest->__cpu_power < | |
3794 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3795 | tmp = sg_div_cpu_power(sds->this, | |
3796 | sds->max_load * sds->busiest->__cpu_power); | |
3797 | else | |
3798 | tmp = sg_div_cpu_power(sds->this, | |
3799 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3800 | pwr_move += sds->this->__cpu_power * | |
3801 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3802 | pwr_move /= SCHED_LOAD_SCALE; | |
3803 | ||
3804 | /* Move if we gain throughput */ | |
3805 | if (pwr_move > pwr_now) | |
3806 | *imbalance = sds->busiest_load_per_task; | |
3807 | } | |
dbc523a3 GS |
3808 | |
3809 | /** | |
3810 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3811 | * groups of a given sched_domain during load balance. | |
3812 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3813 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3814 | * @imbalance: The variable to store the imbalance. | |
3815 | */ | |
3816 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3817 | unsigned long *imbalance) | |
3818 | { | |
3819 | unsigned long max_pull; | |
2dd73a4f PW |
3820 | /* |
3821 | * In the presence of smp nice balancing, certain scenarios can have | |
3822 | * max load less than avg load(as we skip the groups at or below | |
3823 | * its cpu_power, while calculating max_load..) | |
3824 | */ | |
dbc523a3 | 3825 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3826 | *imbalance = 0; |
dbc523a3 | 3827 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3828 | } |
0c117f1b SS |
3829 | |
3830 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3831 | max_pull = min(sds->max_load - sds->avg_load, |
3832 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3833 | |
1da177e4 | 3834 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3835 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3836 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3837 | / SCHED_LOAD_SCALE; |
3838 | ||
2dd73a4f PW |
3839 | /* |
3840 | * if *imbalance is less than the average load per runnable task | |
3841 | * there is no gaurantee that any tasks will be moved so we'll have | |
3842 | * a think about bumping its value to force at least one task to be | |
3843 | * moved | |
3844 | */ | |
dbc523a3 GS |
3845 | if (*imbalance < sds->busiest_load_per_task) |
3846 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3847 | |
dbc523a3 | 3848 | } |
37abe198 | 3849 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3850 | |
b7bb4c9b GS |
3851 | /** |
3852 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3853 | * if there is an imbalance. If there isn't an imbalance, and | |
3854 | * the user has opted for power-savings, it returns a group whose | |
3855 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3856 | * such a group exists. | |
3857 | * | |
3858 | * Also calculates the amount of weighted load which should be moved | |
3859 | * to restore balance. | |
3860 | * | |
3861 | * @sd: The sched_domain whose busiest group is to be returned. | |
3862 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3863 | * @imbalance: Variable which stores amount of weighted load which should | |
3864 | * be moved to restore balance/put a group to idle. | |
3865 | * @idle: The idle status of this_cpu. | |
3866 | * @sd_idle: The idleness of sd | |
3867 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3868 | * @balance: Pointer to a variable indicating if this_cpu | |
3869 | * is the appropriate cpu to perform load balancing at this_level. | |
3870 | * | |
3871 | * Returns: - the busiest group if imbalance exists. | |
3872 | * - If no imbalance and user has opted for power-savings balance, | |
3873 | * return the least loaded group whose CPUs can be | |
3874 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3875 | */ |
3876 | static struct sched_group * | |
3877 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3878 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3879 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3880 | { | |
3881 | struct sd_lb_stats sds; | |
1da177e4 | 3882 | |
37abe198 | 3883 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3884 | |
37abe198 GS |
3885 | /* |
3886 | * Compute the various statistics relavent for load balancing at | |
3887 | * this level. | |
3888 | */ | |
3889 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3890 | balance, &sds); | |
3891 | ||
b7bb4c9b GS |
3892 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3893 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3894 | * at this level. | |
3895 | * 2) There is no busy sibling group to pull from. | |
3896 | * 3) This group is the busiest group. | |
3897 | * 4) This group is more busy than the avg busieness at this | |
3898 | * sched_domain. | |
3899 | * 5) The imbalance is within the specified limit. | |
3900 | * 6) Any rebalance would lead to ping-pong | |
3901 | */ | |
37abe198 GS |
3902 | if (balance && !(*balance)) |
3903 | goto ret; | |
1da177e4 | 3904 | |
b7bb4c9b GS |
3905 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3906 | goto out_balanced; | |
1da177e4 | 3907 | |
b7bb4c9b | 3908 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3909 | goto out_balanced; |
1da177e4 | 3910 | |
222d656d | 3911 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3912 | |
b7bb4c9b GS |
3913 | if (sds.this_load >= sds.avg_load) |
3914 | goto out_balanced; | |
3915 | ||
3916 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3917 | goto out_balanced; |
3918 | ||
222d656d GS |
3919 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3920 | if (sds.group_imb) | |
3921 | sds.busiest_load_per_task = | |
3922 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3923 | |
1da177e4 LT |
3924 | /* |
3925 | * We're trying to get all the cpus to the average_load, so we don't | |
3926 | * want to push ourselves above the average load, nor do we wish to | |
3927 | * reduce the max loaded cpu below the average load, as either of these | |
3928 | * actions would just result in more rebalancing later, and ping-pong | |
3929 | * tasks around. Thus we look for the minimum possible imbalance. | |
3930 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3931 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3932 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3933 | * appear as very large values with unsigned longs. |
3934 | */ | |
222d656d | 3935 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3936 | goto out_balanced; |
3937 | ||
dbc523a3 GS |
3938 | /* Looks like there is an imbalance. Compute it */ |
3939 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3940 | return sds.busiest; |
1da177e4 LT |
3941 | |
3942 | out_balanced: | |
c071df18 GS |
3943 | /* |
3944 | * There is no obvious imbalance. But check if we can do some balancing | |
3945 | * to save power. | |
3946 | */ | |
3947 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3948 | return sds.busiest; | |
783609c6 | 3949 | ret: |
1da177e4 LT |
3950 | *imbalance = 0; |
3951 | return NULL; | |
3952 | } | |
3953 | ||
3954 | /* | |
3955 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3956 | */ | |
70b97a7f | 3957 | static struct rq * |
d15bcfdb | 3958 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3959 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3960 | { |
70b97a7f | 3961 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3962 | unsigned long max_load = 0; |
1da177e4 LT |
3963 | int i; |
3964 | ||
758b2cdc | 3965 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3966 | unsigned long wl; |
0a2966b4 | 3967 | |
96f874e2 | 3968 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3969 | continue; |
3970 | ||
48f24c4d | 3971 | rq = cpu_rq(i); |
dd41f596 | 3972 | wl = weighted_cpuload(i); |
2dd73a4f | 3973 | |
dd41f596 | 3974 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3975 | continue; |
1da177e4 | 3976 | |
dd41f596 IM |
3977 | if (wl > max_load) { |
3978 | max_load = wl; | |
48f24c4d | 3979 | busiest = rq; |
1da177e4 LT |
3980 | } |
3981 | } | |
3982 | ||
3983 | return busiest; | |
3984 | } | |
3985 | ||
77391d71 NP |
3986 | /* |
3987 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3988 | * so long as it is large enough. | |
3989 | */ | |
3990 | #define MAX_PINNED_INTERVAL 512 | |
3991 | ||
df7c8e84 RR |
3992 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3993 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3994 | ||
1da177e4 LT |
3995 | /* |
3996 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3997 | * tasks if there is an imbalance. | |
1da177e4 | 3998 | */ |
70b97a7f | 3999 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4000 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4001 | int *balance) |
1da177e4 | 4002 | { |
43010659 | 4003 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4004 | struct sched_group *group; |
1da177e4 | 4005 | unsigned long imbalance; |
70b97a7f | 4006 | struct rq *busiest; |
fe2eea3f | 4007 | unsigned long flags; |
df7c8e84 | 4008 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4009 | |
96f874e2 | 4010 | cpumask_setall(cpus); |
7c16ec58 | 4011 | |
89c4710e SS |
4012 | /* |
4013 | * When power savings policy is enabled for the parent domain, idle | |
4014 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4015 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4016 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4017 | */ |
d15bcfdb | 4018 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4019 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4020 | sd_idle = 1; |
1da177e4 | 4021 | |
2d72376b | 4022 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4023 | |
0a2966b4 | 4024 | redo: |
c8cba857 | 4025 | update_shares(sd); |
0a2966b4 | 4026 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4027 | cpus, balance); |
783609c6 | 4028 | |
06066714 | 4029 | if (*balance == 0) |
783609c6 | 4030 | goto out_balanced; |
783609c6 | 4031 | |
1da177e4 LT |
4032 | if (!group) { |
4033 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4034 | goto out_balanced; | |
4035 | } | |
4036 | ||
7c16ec58 | 4037 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4038 | if (!busiest) { |
4039 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4040 | goto out_balanced; | |
4041 | } | |
4042 | ||
db935dbd | 4043 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4044 | |
4045 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4046 | ||
43010659 | 4047 | ld_moved = 0; |
1da177e4 LT |
4048 | if (busiest->nr_running > 1) { |
4049 | /* | |
4050 | * Attempt to move tasks. If find_busiest_group has found | |
4051 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4052 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4053 | * correctly treated as an imbalance. |
4054 | */ | |
fe2eea3f | 4055 | local_irq_save(flags); |
e17224bf | 4056 | double_rq_lock(this_rq, busiest); |
43010659 | 4057 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4058 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4059 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4060 | local_irq_restore(flags); |
81026794 | 4061 | |
46cb4b7c SS |
4062 | /* |
4063 | * some other cpu did the load balance for us. | |
4064 | */ | |
43010659 | 4065 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4066 | resched_cpu(this_cpu); |
4067 | ||
81026794 | 4068 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4069 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4070 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4071 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4072 | goto redo; |
81026794 | 4073 | goto out_balanced; |
0a2966b4 | 4074 | } |
1da177e4 | 4075 | } |
81026794 | 4076 | |
43010659 | 4077 | if (!ld_moved) { |
1da177e4 LT |
4078 | schedstat_inc(sd, lb_failed[idle]); |
4079 | sd->nr_balance_failed++; | |
4080 | ||
4081 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4082 | |
fe2eea3f | 4083 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4084 | |
4085 | /* don't kick the migration_thread, if the curr | |
4086 | * task on busiest cpu can't be moved to this_cpu | |
4087 | */ | |
96f874e2 RR |
4088 | if (!cpumask_test_cpu(this_cpu, |
4089 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4090 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4091 | all_pinned = 1; |
4092 | goto out_one_pinned; | |
4093 | } | |
4094 | ||
1da177e4 LT |
4095 | if (!busiest->active_balance) { |
4096 | busiest->active_balance = 1; | |
4097 | busiest->push_cpu = this_cpu; | |
81026794 | 4098 | active_balance = 1; |
1da177e4 | 4099 | } |
fe2eea3f | 4100 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4101 | if (active_balance) |
1da177e4 LT |
4102 | wake_up_process(busiest->migration_thread); |
4103 | ||
4104 | /* | |
4105 | * We've kicked active balancing, reset the failure | |
4106 | * counter. | |
4107 | */ | |
39507451 | 4108 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4109 | } |
81026794 | 4110 | } else |
1da177e4 LT |
4111 | sd->nr_balance_failed = 0; |
4112 | ||
81026794 | 4113 | if (likely(!active_balance)) { |
1da177e4 LT |
4114 | /* We were unbalanced, so reset the balancing interval */ |
4115 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4116 | } else { |
4117 | /* | |
4118 | * If we've begun active balancing, start to back off. This | |
4119 | * case may not be covered by the all_pinned logic if there | |
4120 | * is only 1 task on the busy runqueue (because we don't call | |
4121 | * move_tasks). | |
4122 | */ | |
4123 | if (sd->balance_interval < sd->max_interval) | |
4124 | sd->balance_interval *= 2; | |
1da177e4 LT |
4125 | } |
4126 | ||
43010659 | 4127 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4128 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4129 | ld_moved = -1; |
4130 | ||
4131 | goto out; | |
1da177e4 LT |
4132 | |
4133 | out_balanced: | |
1da177e4 LT |
4134 | schedstat_inc(sd, lb_balanced[idle]); |
4135 | ||
16cfb1c0 | 4136 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4137 | |
4138 | out_one_pinned: | |
1da177e4 | 4139 | /* tune up the balancing interval */ |
77391d71 NP |
4140 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4141 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4142 | sd->balance_interval *= 2; |
4143 | ||
48f24c4d | 4144 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4145 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4146 | ld_moved = -1; |
4147 | else | |
4148 | ld_moved = 0; | |
4149 | out: | |
c8cba857 PZ |
4150 | if (ld_moved) |
4151 | update_shares(sd); | |
c09595f6 | 4152 | return ld_moved; |
1da177e4 LT |
4153 | } |
4154 | ||
4155 | /* | |
4156 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4157 | * tasks if there is an imbalance. | |
4158 | * | |
d15bcfdb | 4159 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4160 | * this_rq is locked. |
4161 | */ | |
48f24c4d | 4162 | static int |
df7c8e84 | 4163 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4164 | { |
4165 | struct sched_group *group; | |
70b97a7f | 4166 | struct rq *busiest = NULL; |
1da177e4 | 4167 | unsigned long imbalance; |
43010659 | 4168 | int ld_moved = 0; |
5969fe06 | 4169 | int sd_idle = 0; |
969bb4e4 | 4170 | int all_pinned = 0; |
df7c8e84 | 4171 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4172 | |
96f874e2 | 4173 | cpumask_setall(cpus); |
5969fe06 | 4174 | |
89c4710e SS |
4175 | /* |
4176 | * When power savings policy is enabled for the parent domain, idle | |
4177 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4178 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4179 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4180 | */ |
4181 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4182 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4183 | sd_idle = 1; |
1da177e4 | 4184 | |
2d72376b | 4185 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4186 | redo: |
3e5459b4 | 4187 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4188 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4189 | &sd_idle, cpus, NULL); |
1da177e4 | 4190 | if (!group) { |
d15bcfdb | 4191 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4192 | goto out_balanced; |
1da177e4 LT |
4193 | } |
4194 | ||
7c16ec58 | 4195 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4196 | if (!busiest) { |
d15bcfdb | 4197 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4198 | goto out_balanced; |
1da177e4 LT |
4199 | } |
4200 | ||
db935dbd NP |
4201 | BUG_ON(busiest == this_rq); |
4202 | ||
d15bcfdb | 4203 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4204 | |
43010659 | 4205 | ld_moved = 0; |
d6d5cfaf NP |
4206 | if (busiest->nr_running > 1) { |
4207 | /* Attempt to move tasks */ | |
4208 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4209 | /* this_rq->clock is already updated */ |
4210 | update_rq_clock(busiest); | |
43010659 | 4211 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4212 | imbalance, sd, CPU_NEWLY_IDLE, |
4213 | &all_pinned); | |
1b12bbc7 | 4214 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4215 | |
969bb4e4 | 4216 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4217 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4218 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4219 | goto redo; |
4220 | } | |
d6d5cfaf NP |
4221 | } |
4222 | ||
43010659 | 4223 | if (!ld_moved) { |
36dffab6 | 4224 | int active_balance = 0; |
ad273b32 | 4225 | |
d15bcfdb | 4226 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4227 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4228 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4229 | return -1; |
ad273b32 VS |
4230 | |
4231 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4232 | return -1; | |
4233 | ||
4234 | if (sd->nr_balance_failed++ < 2) | |
4235 | return -1; | |
4236 | ||
4237 | /* | |
4238 | * The only task running in a non-idle cpu can be moved to this | |
4239 | * cpu in an attempt to completely freeup the other CPU | |
4240 | * package. The same method used to move task in load_balance() | |
4241 | * have been extended for load_balance_newidle() to speedup | |
4242 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4243 | * | |
4244 | * The package power saving logic comes from | |
4245 | * find_busiest_group(). If there are no imbalance, then | |
4246 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4247 | * f_b_g() will select a group from which a running task may be | |
4248 | * pulled to this cpu in order to make the other package idle. | |
4249 | * If there is no opportunity to make a package idle and if | |
4250 | * there are no imbalance, then f_b_g() will return NULL and no | |
4251 | * action will be taken in load_balance_newidle(). | |
4252 | * | |
4253 | * Under normal task pull operation due to imbalance, there | |
4254 | * will be more than one task in the source run queue and | |
4255 | * move_tasks() will succeed. ld_moved will be true and this | |
4256 | * active balance code will not be triggered. | |
4257 | */ | |
4258 | ||
4259 | /* Lock busiest in correct order while this_rq is held */ | |
4260 | double_lock_balance(this_rq, busiest); | |
4261 | ||
4262 | /* | |
4263 | * don't kick the migration_thread, if the curr | |
4264 | * task on busiest cpu can't be moved to this_cpu | |
4265 | */ | |
6ca09dfc | 4266 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4267 | double_unlock_balance(this_rq, busiest); |
4268 | all_pinned = 1; | |
4269 | return ld_moved; | |
4270 | } | |
4271 | ||
4272 | if (!busiest->active_balance) { | |
4273 | busiest->active_balance = 1; | |
4274 | busiest->push_cpu = this_cpu; | |
4275 | active_balance = 1; | |
4276 | } | |
4277 | ||
4278 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4279 | /* |
4280 | * Should not call ttwu while holding a rq->lock | |
4281 | */ | |
4282 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4283 | if (active_balance) |
4284 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4285 | spin_lock(&this_rq->lock); |
ad273b32 | 4286 | |
5969fe06 | 4287 | } else |
16cfb1c0 | 4288 | sd->nr_balance_failed = 0; |
1da177e4 | 4289 | |
3e5459b4 | 4290 | update_shares_locked(this_rq, sd); |
43010659 | 4291 | return ld_moved; |
16cfb1c0 NP |
4292 | |
4293 | out_balanced: | |
d15bcfdb | 4294 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4295 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4296 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4297 | return -1; |
16cfb1c0 | 4298 | sd->nr_balance_failed = 0; |
48f24c4d | 4299 | |
16cfb1c0 | 4300 | return 0; |
1da177e4 LT |
4301 | } |
4302 | ||
4303 | /* | |
4304 | * idle_balance is called by schedule() if this_cpu is about to become | |
4305 | * idle. Attempts to pull tasks from other CPUs. | |
4306 | */ | |
70b97a7f | 4307 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4308 | { |
4309 | struct sched_domain *sd; | |
efbe027e | 4310 | int pulled_task = 0; |
dd41f596 | 4311 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4312 | |
4313 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4314 | unsigned long interval; |
4315 | ||
4316 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4317 | continue; | |
4318 | ||
4319 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4320 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4321 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4322 | sd); |
92c4ca5c CL |
4323 | |
4324 | interval = msecs_to_jiffies(sd->balance_interval); | |
4325 | if (time_after(next_balance, sd->last_balance + interval)) | |
4326 | next_balance = sd->last_balance + interval; | |
4327 | if (pulled_task) | |
4328 | break; | |
1da177e4 | 4329 | } |
dd41f596 | 4330 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4331 | /* |
4332 | * We are going idle. next_balance may be set based on | |
4333 | * a busy processor. So reset next_balance. | |
4334 | */ | |
4335 | this_rq->next_balance = next_balance; | |
dd41f596 | 4336 | } |
1da177e4 LT |
4337 | } |
4338 | ||
4339 | /* | |
4340 | * active_load_balance is run by migration threads. It pushes running tasks | |
4341 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4342 | * running on each physical CPU where possible, and avoids physical / | |
4343 | * logical imbalances. | |
4344 | * | |
4345 | * Called with busiest_rq locked. | |
4346 | */ | |
70b97a7f | 4347 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4348 | { |
39507451 | 4349 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4350 | struct sched_domain *sd; |
4351 | struct rq *target_rq; | |
39507451 | 4352 | |
48f24c4d | 4353 | /* Is there any task to move? */ |
39507451 | 4354 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4355 | return; |
4356 | ||
4357 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4358 | |
4359 | /* | |
39507451 | 4360 | * This condition is "impossible", if it occurs |
41a2d6cf | 4361 | * we need to fix it. Originally reported by |
39507451 | 4362 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4363 | */ |
39507451 | 4364 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4365 | |
39507451 NP |
4366 | /* move a task from busiest_rq to target_rq */ |
4367 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4368 | update_rq_clock(busiest_rq); |
4369 | update_rq_clock(target_rq); | |
39507451 NP |
4370 | |
4371 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4372 | for_each_domain(target_cpu, sd) { |
39507451 | 4373 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4374 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4375 | break; |
c96d145e | 4376 | } |
39507451 | 4377 | |
48f24c4d | 4378 | if (likely(sd)) { |
2d72376b | 4379 | schedstat_inc(sd, alb_count); |
39507451 | 4380 | |
43010659 PW |
4381 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4382 | sd, CPU_IDLE)) | |
48f24c4d IM |
4383 | schedstat_inc(sd, alb_pushed); |
4384 | else | |
4385 | schedstat_inc(sd, alb_failed); | |
4386 | } | |
1b12bbc7 | 4387 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4388 | } |
4389 | ||
46cb4b7c SS |
4390 | #ifdef CONFIG_NO_HZ |
4391 | static struct { | |
4392 | atomic_t load_balancer; | |
7d1e6a9b | 4393 | cpumask_var_t cpu_mask; |
f711f609 | 4394 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4395 | } nohz ____cacheline_aligned = { |
4396 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4397 | }; |
4398 | ||
f711f609 GS |
4399 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4400 | /** | |
4401 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4402 | * @cpu: The cpu whose lowest level of sched domain is to | |
4403 | * be returned. | |
4404 | * @flag: The flag to check for the lowest sched_domain | |
4405 | * for the given cpu. | |
4406 | * | |
4407 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4408 | */ | |
4409 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4410 | { | |
4411 | struct sched_domain *sd; | |
4412 | ||
4413 | for_each_domain(cpu, sd) | |
4414 | if (sd && (sd->flags & flag)) | |
4415 | break; | |
4416 | ||
4417 | return sd; | |
4418 | } | |
4419 | ||
4420 | /** | |
4421 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4422 | * @cpu: The cpu whose domains we're iterating over. | |
4423 | * @sd: variable holding the value of the power_savings_sd | |
4424 | * for cpu. | |
4425 | * @flag: The flag to filter the sched_domains to be iterated. | |
4426 | * | |
4427 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4428 | * set, starting from the lowest sched_domain to the highest. | |
4429 | */ | |
4430 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4431 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4432 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4433 | ||
4434 | /** | |
4435 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4436 | * @ilb_group: group to be checked for semi-idleness | |
4437 | * | |
4438 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4439 | * | |
4440 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4441 | * and atleast one non-idle CPU. This helper function checks if the given | |
4442 | * sched_group is semi-idle or not. | |
4443 | */ | |
4444 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4445 | { | |
4446 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4447 | sched_group_cpus(ilb_group)); | |
4448 | ||
4449 | /* | |
4450 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4451 | * and atleast one idle cpu. | |
4452 | */ | |
4453 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4454 | return 0; | |
4455 | ||
4456 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4457 | return 0; | |
4458 | ||
4459 | return 1; | |
4460 | } | |
4461 | /** | |
4462 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4463 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4464 | * | |
4465 | * Returns: Returns the id of the idle load balancer if it exists, | |
4466 | * Else, returns >= nr_cpu_ids. | |
4467 | * | |
4468 | * This algorithm picks the idle load balancer such that it belongs to a | |
4469 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4470 | * completely idle packages/cores just for the purpose of idle load balancing | |
4471 | * when there are other idle cpu's which are better suited for that job. | |
4472 | */ | |
4473 | static int find_new_ilb(int cpu) | |
4474 | { | |
4475 | struct sched_domain *sd; | |
4476 | struct sched_group *ilb_group; | |
4477 | ||
4478 | /* | |
4479 | * Have idle load balancer selection from semi-idle packages only | |
4480 | * when power-aware load balancing is enabled | |
4481 | */ | |
4482 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4483 | goto out_done; | |
4484 | ||
4485 | /* | |
4486 | * Optimize for the case when we have no idle CPUs or only one | |
4487 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4488 | */ | |
4489 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4490 | goto out_done; | |
4491 | ||
4492 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4493 | ilb_group = sd->groups; | |
4494 | ||
4495 | do { | |
4496 | if (is_semi_idle_group(ilb_group)) | |
4497 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4498 | ||
4499 | ilb_group = ilb_group->next; | |
4500 | ||
4501 | } while (ilb_group != sd->groups); | |
4502 | } | |
4503 | ||
4504 | out_done: | |
4505 | return cpumask_first(nohz.cpu_mask); | |
4506 | } | |
4507 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4508 | static inline int find_new_ilb(int call_cpu) | |
4509 | { | |
6e29ec57 | 4510 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4511 | } |
4512 | #endif | |
4513 | ||
7835b98b | 4514 | /* |
46cb4b7c SS |
4515 | * This routine will try to nominate the ilb (idle load balancing) |
4516 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4517 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4518 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4519 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4520 | * arrives... | |
4521 | * | |
4522 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4523 | * for idle load balancing. ilb owner will still be part of | |
4524 | * nohz.cpu_mask.. | |
7835b98b | 4525 | * |
46cb4b7c SS |
4526 | * While stopping the tick, this cpu will become the ilb owner if there |
4527 | * is no other owner. And will be the owner till that cpu becomes busy | |
4528 | * or if all cpus in the system stop their ticks at which point | |
4529 | * there is no need for ilb owner. | |
4530 | * | |
4531 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4532 | * next busy scheduler_tick() | |
4533 | */ | |
4534 | int select_nohz_load_balancer(int stop_tick) | |
4535 | { | |
4536 | int cpu = smp_processor_id(); | |
4537 | ||
4538 | if (stop_tick) { | |
46cb4b7c SS |
4539 | cpu_rq(cpu)->in_nohz_recently = 1; |
4540 | ||
483b4ee6 SS |
4541 | if (!cpu_active(cpu)) { |
4542 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4543 | return 0; | |
4544 | ||
4545 | /* | |
4546 | * If we are going offline and still the leader, | |
4547 | * give up! | |
4548 | */ | |
46cb4b7c SS |
4549 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4550 | BUG(); | |
483b4ee6 | 4551 | |
46cb4b7c SS |
4552 | return 0; |
4553 | } | |
4554 | ||
483b4ee6 SS |
4555 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4556 | ||
46cb4b7c | 4557 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4558 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4559 | if (atomic_read(&nohz.load_balancer) == cpu) |
4560 | atomic_set(&nohz.load_balancer, -1); | |
4561 | return 0; | |
4562 | } | |
4563 | ||
4564 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4565 | /* make me the ilb owner */ | |
4566 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4567 | return 1; | |
e790fb0b GS |
4568 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4569 | int new_ilb; | |
4570 | ||
4571 | if (!(sched_smt_power_savings || | |
4572 | sched_mc_power_savings)) | |
4573 | return 1; | |
4574 | /* | |
4575 | * Check to see if there is a more power-efficient | |
4576 | * ilb. | |
4577 | */ | |
4578 | new_ilb = find_new_ilb(cpu); | |
4579 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4580 | atomic_set(&nohz.load_balancer, -1); | |
4581 | resched_cpu(new_ilb); | |
4582 | return 0; | |
4583 | } | |
46cb4b7c | 4584 | return 1; |
e790fb0b | 4585 | } |
46cb4b7c | 4586 | } else { |
7d1e6a9b | 4587 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4588 | return 0; |
4589 | ||
7d1e6a9b | 4590 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4591 | |
4592 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4593 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4594 | BUG(); | |
4595 | } | |
4596 | return 0; | |
4597 | } | |
4598 | #endif | |
4599 | ||
4600 | static DEFINE_SPINLOCK(balancing); | |
4601 | ||
4602 | /* | |
7835b98b CL |
4603 | * It checks each scheduling domain to see if it is due to be balanced, |
4604 | * and initiates a balancing operation if so. | |
4605 | * | |
4606 | * Balancing parameters are set up in arch_init_sched_domains. | |
4607 | */ | |
a9957449 | 4608 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4609 | { |
46cb4b7c SS |
4610 | int balance = 1; |
4611 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4612 | unsigned long interval; |
4613 | struct sched_domain *sd; | |
46cb4b7c | 4614 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4615 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4616 | int update_next_balance = 0; |
d07355f5 | 4617 | int need_serialize; |
1da177e4 | 4618 | |
46cb4b7c | 4619 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4620 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4621 | continue; | |
4622 | ||
4623 | interval = sd->balance_interval; | |
d15bcfdb | 4624 | if (idle != CPU_IDLE) |
1da177e4 LT |
4625 | interval *= sd->busy_factor; |
4626 | ||
4627 | /* scale ms to jiffies */ | |
4628 | interval = msecs_to_jiffies(interval); | |
4629 | if (unlikely(!interval)) | |
4630 | interval = 1; | |
dd41f596 IM |
4631 | if (interval > HZ*NR_CPUS/10) |
4632 | interval = HZ*NR_CPUS/10; | |
4633 | ||
d07355f5 | 4634 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4635 | |
d07355f5 | 4636 | if (need_serialize) { |
08c183f3 CL |
4637 | if (!spin_trylock(&balancing)) |
4638 | goto out; | |
4639 | } | |
4640 | ||
c9819f45 | 4641 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4642 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4643 | /* |
4644 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4645 | * longer idle, or one of our SMT siblings is |
4646 | * not idle. | |
4647 | */ | |
d15bcfdb | 4648 | idle = CPU_NOT_IDLE; |
1da177e4 | 4649 | } |
1bd77f2d | 4650 | sd->last_balance = jiffies; |
1da177e4 | 4651 | } |
d07355f5 | 4652 | if (need_serialize) |
08c183f3 CL |
4653 | spin_unlock(&balancing); |
4654 | out: | |
f549da84 | 4655 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4656 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4657 | update_next_balance = 1; |
4658 | } | |
783609c6 SS |
4659 | |
4660 | /* | |
4661 | * Stop the load balance at this level. There is another | |
4662 | * CPU in our sched group which is doing load balancing more | |
4663 | * actively. | |
4664 | */ | |
4665 | if (!balance) | |
4666 | break; | |
1da177e4 | 4667 | } |
f549da84 SS |
4668 | |
4669 | /* | |
4670 | * next_balance will be updated only when there is a need. | |
4671 | * When the cpu is attached to null domain for ex, it will not be | |
4672 | * updated. | |
4673 | */ | |
4674 | if (likely(update_next_balance)) | |
4675 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4676 | } |
4677 | ||
4678 | /* | |
4679 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4680 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4681 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4682 | */ | |
4683 | static void run_rebalance_domains(struct softirq_action *h) | |
4684 | { | |
dd41f596 IM |
4685 | int this_cpu = smp_processor_id(); |
4686 | struct rq *this_rq = cpu_rq(this_cpu); | |
4687 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4688 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4689 | |
dd41f596 | 4690 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4691 | |
4692 | #ifdef CONFIG_NO_HZ | |
4693 | /* | |
4694 | * If this cpu is the owner for idle load balancing, then do the | |
4695 | * balancing on behalf of the other idle cpus whose ticks are | |
4696 | * stopped. | |
4697 | */ | |
dd41f596 IM |
4698 | if (this_rq->idle_at_tick && |
4699 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4700 | struct rq *rq; |
4701 | int balance_cpu; | |
4702 | ||
7d1e6a9b RR |
4703 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4704 | if (balance_cpu == this_cpu) | |
4705 | continue; | |
4706 | ||
46cb4b7c SS |
4707 | /* |
4708 | * If this cpu gets work to do, stop the load balancing | |
4709 | * work being done for other cpus. Next load | |
4710 | * balancing owner will pick it up. | |
4711 | */ | |
4712 | if (need_resched()) | |
4713 | break; | |
4714 | ||
de0cf899 | 4715 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4716 | |
4717 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4718 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4719 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4720 | } |
4721 | } | |
4722 | #endif | |
4723 | } | |
4724 | ||
8a0be9ef FW |
4725 | static inline int on_null_domain(int cpu) |
4726 | { | |
4727 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4728 | } | |
4729 | ||
46cb4b7c SS |
4730 | /* |
4731 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4732 | * | |
4733 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4734 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4735 | * if the whole system is idle. | |
4736 | */ | |
dd41f596 | 4737 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4738 | { |
46cb4b7c SS |
4739 | #ifdef CONFIG_NO_HZ |
4740 | /* | |
4741 | * If we were in the nohz mode recently and busy at the current | |
4742 | * scheduler tick, then check if we need to nominate new idle | |
4743 | * load balancer. | |
4744 | */ | |
4745 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4746 | rq->in_nohz_recently = 0; | |
4747 | ||
4748 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4749 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4750 | atomic_set(&nohz.load_balancer, -1); |
4751 | } | |
4752 | ||
4753 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4754 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4755 | |
434d53b0 | 4756 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4757 | resched_cpu(ilb); |
4758 | } | |
4759 | } | |
4760 | ||
4761 | /* | |
4762 | * If this cpu is idle and doing idle load balancing for all the | |
4763 | * cpus with ticks stopped, is it time for that to stop? | |
4764 | */ | |
4765 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4766 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4767 | resched_cpu(cpu); |
4768 | return; | |
4769 | } | |
4770 | ||
4771 | /* | |
4772 | * If this cpu is idle and the idle load balancing is done by | |
4773 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4774 | */ | |
4775 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4776 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4777 | return; |
4778 | #endif | |
8a0be9ef FW |
4779 | /* Don't need to rebalance while attached to NULL domain */ |
4780 | if (time_after_eq(jiffies, rq->next_balance) && | |
4781 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4782 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4783 | } |
dd41f596 IM |
4784 | |
4785 | #else /* CONFIG_SMP */ | |
4786 | ||
1da177e4 LT |
4787 | /* |
4788 | * on UP we do not need to balance between CPUs: | |
4789 | */ | |
70b97a7f | 4790 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4791 | { |
4792 | } | |
dd41f596 | 4793 | |
1da177e4 LT |
4794 | #endif |
4795 | ||
1da177e4 LT |
4796 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4797 | ||
4798 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4799 | ||
4800 | /* | |
c5f8d995 | 4801 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4802 | * @p in case that task is currently running. |
c5f8d995 HS |
4803 | * |
4804 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4805 | */ |
c5f8d995 HS |
4806 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4807 | { | |
4808 | u64 ns = 0; | |
4809 | ||
4810 | if (task_current(rq, p)) { | |
4811 | update_rq_clock(rq); | |
4812 | ns = rq->clock - p->se.exec_start; | |
4813 | if ((s64)ns < 0) | |
4814 | ns = 0; | |
4815 | } | |
4816 | ||
4817 | return ns; | |
4818 | } | |
4819 | ||
bb34d92f | 4820 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4821 | { |
1da177e4 | 4822 | unsigned long flags; |
41b86e9c | 4823 | struct rq *rq; |
bb34d92f | 4824 | u64 ns = 0; |
48f24c4d | 4825 | |
41b86e9c | 4826 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4827 | ns = do_task_delta_exec(p, rq); |
4828 | task_rq_unlock(rq, &flags); | |
1508487e | 4829 | |
c5f8d995 HS |
4830 | return ns; |
4831 | } | |
f06febc9 | 4832 | |
c5f8d995 HS |
4833 | /* |
4834 | * Return accounted runtime for the task. | |
4835 | * In case the task is currently running, return the runtime plus current's | |
4836 | * pending runtime that have not been accounted yet. | |
4837 | */ | |
4838 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4839 | { | |
4840 | unsigned long flags; | |
4841 | struct rq *rq; | |
4842 | u64 ns = 0; | |
4843 | ||
4844 | rq = task_rq_lock(p, &flags); | |
4845 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4846 | task_rq_unlock(rq, &flags); | |
4847 | ||
4848 | return ns; | |
4849 | } | |
48f24c4d | 4850 | |
c5f8d995 HS |
4851 | /* |
4852 | * Return sum_exec_runtime for the thread group. | |
4853 | * In case the task is currently running, return the sum plus current's | |
4854 | * pending runtime that have not been accounted yet. | |
4855 | * | |
4856 | * Note that the thread group might have other running tasks as well, | |
4857 | * so the return value not includes other pending runtime that other | |
4858 | * running tasks might have. | |
4859 | */ | |
4860 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4861 | { | |
4862 | struct task_cputime totals; | |
4863 | unsigned long flags; | |
4864 | struct rq *rq; | |
4865 | u64 ns; | |
4866 | ||
4867 | rq = task_rq_lock(p, &flags); | |
4868 | thread_group_cputime(p, &totals); | |
4869 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4870 | task_rq_unlock(rq, &flags); |
48f24c4d | 4871 | |
1da177e4 LT |
4872 | return ns; |
4873 | } | |
4874 | ||
1da177e4 LT |
4875 | /* |
4876 | * Account user cpu time to a process. | |
4877 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4878 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4879 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4880 | */ |
457533a7 MS |
4881 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4882 | cputime_t cputime_scaled) | |
1da177e4 LT |
4883 | { |
4884 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4885 | cputime64_t tmp; | |
4886 | ||
457533a7 | 4887 | /* Add user time to process. */ |
1da177e4 | 4888 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4889 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4890 | account_group_user_time(p, cputime); |
1da177e4 LT |
4891 | |
4892 | /* Add user time to cpustat. */ | |
4893 | tmp = cputime_to_cputime64(cputime); | |
4894 | if (TASK_NICE(p) > 0) | |
4895 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4896 | else | |
4897 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4898 | |
4899 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4900 | /* Account for user time used */ |
4901 | acct_update_integrals(p); | |
1da177e4 LT |
4902 | } |
4903 | ||
94886b84 LV |
4904 | /* |
4905 | * Account guest cpu time to a process. | |
4906 | * @p: the process that the cpu time gets accounted to | |
4907 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4908 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4909 | */ |
457533a7 MS |
4910 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4911 | cputime_t cputime_scaled) | |
94886b84 LV |
4912 | { |
4913 | cputime64_t tmp; | |
4914 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4915 | ||
4916 | tmp = cputime_to_cputime64(cputime); | |
4917 | ||
457533a7 | 4918 | /* Add guest time to process. */ |
94886b84 | 4919 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4920 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4921 | account_group_user_time(p, cputime); |
94886b84 LV |
4922 | p->gtime = cputime_add(p->gtime, cputime); |
4923 | ||
457533a7 | 4924 | /* Add guest time to cpustat. */ |
94886b84 LV |
4925 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4926 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4927 | } | |
4928 | ||
1da177e4 LT |
4929 | /* |
4930 | * Account system cpu time to a process. | |
4931 | * @p: the process that the cpu time gets accounted to | |
4932 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4933 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4934 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4935 | */ |
4936 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4937 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4938 | { |
4939 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4940 | cputime64_t tmp; |
4941 | ||
983ed7a6 | 4942 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4943 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4944 | return; |
4945 | } | |
94886b84 | 4946 | |
457533a7 | 4947 | /* Add system time to process. */ |
1da177e4 | 4948 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4949 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4950 | account_group_system_time(p, cputime); |
1da177e4 LT |
4951 | |
4952 | /* Add system time to cpustat. */ | |
4953 | tmp = cputime_to_cputime64(cputime); | |
4954 | if (hardirq_count() - hardirq_offset) | |
4955 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4956 | else if (softirq_count()) | |
4957 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4958 | else |
79741dd3 MS |
4959 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4960 | ||
ef12fefa BR |
4961 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4962 | ||
1da177e4 LT |
4963 | /* Account for system time used */ |
4964 | acct_update_integrals(p); | |
1da177e4 LT |
4965 | } |
4966 | ||
c66f08be | 4967 | /* |
1da177e4 | 4968 | * Account for involuntary wait time. |
1da177e4 | 4969 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4970 | */ |
79741dd3 | 4971 | void account_steal_time(cputime_t cputime) |
c66f08be | 4972 | { |
79741dd3 MS |
4973 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4974 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4975 | ||
4976 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4977 | } |
4978 | ||
1da177e4 | 4979 | /* |
79741dd3 MS |
4980 | * Account for idle time. |
4981 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4982 | */ |
79741dd3 | 4983 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4984 | { |
4985 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4986 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4987 | struct rq *rq = this_rq(); |
1da177e4 | 4988 | |
79741dd3 MS |
4989 | if (atomic_read(&rq->nr_iowait) > 0) |
4990 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4991 | else | |
4992 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4993 | } |
4994 | ||
79741dd3 MS |
4995 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4996 | ||
4997 | /* | |
4998 | * Account a single tick of cpu time. | |
4999 | * @p: the process that the cpu time gets accounted to | |
5000 | * @user_tick: indicates if the tick is a user or a system tick | |
5001 | */ | |
5002 | void account_process_tick(struct task_struct *p, int user_tick) | |
5003 | { | |
5004 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5005 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5006 | struct rq *rq = this_rq(); | |
5007 | ||
5008 | if (user_tick) | |
5009 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5010 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5011 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5012 | one_jiffy_scaled); | |
5013 | else | |
5014 | account_idle_time(one_jiffy); | |
5015 | } | |
5016 | ||
5017 | /* | |
5018 | * Account multiple ticks of steal time. | |
5019 | * @p: the process from which the cpu time has been stolen | |
5020 | * @ticks: number of stolen ticks | |
5021 | */ | |
5022 | void account_steal_ticks(unsigned long ticks) | |
5023 | { | |
5024 | account_steal_time(jiffies_to_cputime(ticks)); | |
5025 | } | |
5026 | ||
5027 | /* | |
5028 | * Account multiple ticks of idle time. | |
5029 | * @ticks: number of stolen ticks | |
5030 | */ | |
5031 | void account_idle_ticks(unsigned long ticks) | |
5032 | { | |
5033 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5034 | } |
5035 | ||
79741dd3 MS |
5036 | #endif |
5037 | ||
49048622 BS |
5038 | /* |
5039 | * Use precise platform statistics if available: | |
5040 | */ | |
5041 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5042 | cputime_t task_utime(struct task_struct *p) | |
5043 | { | |
5044 | return p->utime; | |
5045 | } | |
5046 | ||
5047 | cputime_t task_stime(struct task_struct *p) | |
5048 | { | |
5049 | return p->stime; | |
5050 | } | |
5051 | #else | |
5052 | cputime_t task_utime(struct task_struct *p) | |
5053 | { | |
5054 | clock_t utime = cputime_to_clock_t(p->utime), | |
5055 | total = utime + cputime_to_clock_t(p->stime); | |
5056 | u64 temp; | |
5057 | ||
5058 | /* | |
5059 | * Use CFS's precise accounting: | |
5060 | */ | |
5061 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5062 | ||
5063 | if (total) { | |
5064 | temp *= utime; | |
5065 | do_div(temp, total); | |
5066 | } | |
5067 | utime = (clock_t)temp; | |
5068 | ||
5069 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5070 | return p->prev_utime; | |
5071 | } | |
5072 | ||
5073 | cputime_t task_stime(struct task_struct *p) | |
5074 | { | |
5075 | clock_t stime; | |
5076 | ||
5077 | /* | |
5078 | * Use CFS's precise accounting. (we subtract utime from | |
5079 | * the total, to make sure the total observed by userspace | |
5080 | * grows monotonically - apps rely on that): | |
5081 | */ | |
5082 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5083 | cputime_to_clock_t(task_utime(p)); | |
5084 | ||
5085 | if (stime >= 0) | |
5086 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5087 | ||
5088 | return p->prev_stime; | |
5089 | } | |
5090 | #endif | |
5091 | ||
5092 | inline cputime_t task_gtime(struct task_struct *p) | |
5093 | { | |
5094 | return p->gtime; | |
5095 | } | |
5096 | ||
7835b98b CL |
5097 | /* |
5098 | * This function gets called by the timer code, with HZ frequency. | |
5099 | * We call it with interrupts disabled. | |
5100 | * | |
5101 | * It also gets called by the fork code, when changing the parent's | |
5102 | * timeslices. | |
5103 | */ | |
5104 | void scheduler_tick(void) | |
5105 | { | |
7835b98b CL |
5106 | int cpu = smp_processor_id(); |
5107 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5108 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5109 | |
5110 | sched_clock_tick(); | |
dd41f596 IM |
5111 | |
5112 | spin_lock(&rq->lock); | |
3e51f33f | 5113 | update_rq_clock(rq); |
f1a438d8 | 5114 | update_cpu_load(rq); |
fa85ae24 | 5115 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5116 | spin_unlock(&rq->lock); |
7835b98b | 5117 | |
e220d2dc PZ |
5118 | perf_counter_task_tick(curr, cpu); |
5119 | ||
e418e1c2 | 5120 | #ifdef CONFIG_SMP |
dd41f596 IM |
5121 | rq->idle_at_tick = idle_cpu(cpu); |
5122 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5123 | #endif |
1da177e4 LT |
5124 | } |
5125 | ||
132380a0 | 5126 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5127 | { |
5128 | if (in_lock_functions(addr)) { | |
5129 | addr = CALLER_ADDR2; | |
5130 | if (in_lock_functions(addr)) | |
5131 | addr = CALLER_ADDR3; | |
5132 | } | |
5133 | return addr; | |
5134 | } | |
1da177e4 | 5135 | |
7e49fcce SR |
5136 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5137 | defined(CONFIG_PREEMPT_TRACER)) | |
5138 | ||
43627582 | 5139 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5140 | { |
6cd8a4bb | 5141 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5142 | /* |
5143 | * Underflow? | |
5144 | */ | |
9a11b49a IM |
5145 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5146 | return; | |
6cd8a4bb | 5147 | #endif |
1da177e4 | 5148 | preempt_count() += val; |
6cd8a4bb | 5149 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5150 | /* |
5151 | * Spinlock count overflowing soon? | |
5152 | */ | |
33859f7f MOS |
5153 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5154 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5155 | #endif |
5156 | if (preempt_count() == val) | |
5157 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5158 | } |
5159 | EXPORT_SYMBOL(add_preempt_count); | |
5160 | ||
43627582 | 5161 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5162 | { |
6cd8a4bb | 5163 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5164 | /* |
5165 | * Underflow? | |
5166 | */ | |
01e3eb82 | 5167 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5168 | return; |
1da177e4 LT |
5169 | /* |
5170 | * Is the spinlock portion underflowing? | |
5171 | */ | |
9a11b49a IM |
5172 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5173 | !(preempt_count() & PREEMPT_MASK))) | |
5174 | return; | |
6cd8a4bb | 5175 | #endif |
9a11b49a | 5176 | |
6cd8a4bb SR |
5177 | if (preempt_count() == val) |
5178 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5179 | preempt_count() -= val; |
5180 | } | |
5181 | EXPORT_SYMBOL(sub_preempt_count); | |
5182 | ||
5183 | #endif | |
5184 | ||
5185 | /* | |
dd41f596 | 5186 | * Print scheduling while atomic bug: |
1da177e4 | 5187 | */ |
dd41f596 | 5188 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5189 | { |
838225b4 SS |
5190 | struct pt_regs *regs = get_irq_regs(); |
5191 | ||
5192 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5193 | prev->comm, prev->pid, preempt_count()); | |
5194 | ||
dd41f596 | 5195 | debug_show_held_locks(prev); |
e21f5b15 | 5196 | print_modules(); |
dd41f596 IM |
5197 | if (irqs_disabled()) |
5198 | print_irqtrace_events(prev); | |
838225b4 SS |
5199 | |
5200 | if (regs) | |
5201 | show_regs(regs); | |
5202 | else | |
5203 | dump_stack(); | |
dd41f596 | 5204 | } |
1da177e4 | 5205 | |
dd41f596 IM |
5206 | /* |
5207 | * Various schedule()-time debugging checks and statistics: | |
5208 | */ | |
5209 | static inline void schedule_debug(struct task_struct *prev) | |
5210 | { | |
1da177e4 | 5211 | /* |
41a2d6cf | 5212 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5213 | * schedule() atomically, we ignore that path for now. |
5214 | * Otherwise, whine if we are scheduling when we should not be. | |
5215 | */ | |
3f33a7ce | 5216 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5217 | __schedule_bug(prev); |
5218 | ||
1da177e4 LT |
5219 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5220 | ||
2d72376b | 5221 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5222 | #ifdef CONFIG_SCHEDSTATS |
5223 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5224 | schedstat_inc(this_rq(), bkl_count); |
5225 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5226 | } |
5227 | #endif | |
dd41f596 IM |
5228 | } |
5229 | ||
df1c99d4 MG |
5230 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5231 | { | |
5232 | if (prev->state == TASK_RUNNING) { | |
5233 | u64 runtime = prev->se.sum_exec_runtime; | |
5234 | ||
5235 | runtime -= prev->se.prev_sum_exec_runtime; | |
5236 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5237 | ||
5238 | /* | |
5239 | * In order to avoid avg_overlap growing stale when we are | |
5240 | * indeed overlapping and hence not getting put to sleep, grow | |
5241 | * the avg_overlap on preemption. | |
5242 | * | |
5243 | * We use the average preemption runtime because that | |
5244 | * correlates to the amount of cache footprint a task can | |
5245 | * build up. | |
5246 | */ | |
5247 | update_avg(&prev->se.avg_overlap, runtime); | |
5248 | } | |
5249 | prev->sched_class->put_prev_task(rq, prev); | |
5250 | } | |
5251 | ||
dd41f596 IM |
5252 | /* |
5253 | * Pick up the highest-prio task: | |
5254 | */ | |
5255 | static inline struct task_struct * | |
b67802ea | 5256 | pick_next_task(struct rq *rq) |
dd41f596 | 5257 | { |
5522d5d5 | 5258 | const struct sched_class *class; |
dd41f596 | 5259 | struct task_struct *p; |
1da177e4 LT |
5260 | |
5261 | /* | |
dd41f596 IM |
5262 | * Optimization: we know that if all tasks are in |
5263 | * the fair class we can call that function directly: | |
1da177e4 | 5264 | */ |
dd41f596 | 5265 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5266 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5267 | if (likely(p)) |
5268 | return p; | |
1da177e4 LT |
5269 | } |
5270 | ||
dd41f596 IM |
5271 | class = sched_class_highest; |
5272 | for ( ; ; ) { | |
fb8d4724 | 5273 | p = class->pick_next_task(rq); |
dd41f596 IM |
5274 | if (p) |
5275 | return p; | |
5276 | /* | |
5277 | * Will never be NULL as the idle class always | |
5278 | * returns a non-NULL p: | |
5279 | */ | |
5280 | class = class->next; | |
5281 | } | |
5282 | } | |
1da177e4 | 5283 | |
dd41f596 IM |
5284 | /* |
5285 | * schedule() is the main scheduler function. | |
5286 | */ | |
ff743345 | 5287 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5288 | { |
5289 | struct task_struct *prev, *next; | |
67ca7bde | 5290 | unsigned long *switch_count; |
dd41f596 | 5291 | struct rq *rq; |
31656519 | 5292 | int cpu; |
dd41f596 | 5293 | |
ff743345 PZ |
5294 | need_resched: |
5295 | preempt_disable(); | |
dd41f596 IM |
5296 | cpu = smp_processor_id(); |
5297 | rq = cpu_rq(cpu); | |
5298 | rcu_qsctr_inc(cpu); | |
5299 | prev = rq->curr; | |
5300 | switch_count = &prev->nivcsw; | |
5301 | ||
5302 | release_kernel_lock(prev); | |
5303 | need_resched_nonpreemptible: | |
5304 | ||
5305 | schedule_debug(prev); | |
1da177e4 | 5306 | |
31656519 | 5307 | if (sched_feat(HRTICK)) |
f333fdc9 | 5308 | hrtick_clear(rq); |
8f4d37ec | 5309 | |
8cd162ce | 5310 | spin_lock_irq(&rq->lock); |
3e51f33f | 5311 | update_rq_clock(rq); |
1e819950 | 5312 | clear_tsk_need_resched(prev); |
1da177e4 | 5313 | |
1da177e4 | 5314 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5315 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5316 | prev->state = TASK_RUNNING; |
16882c1e | 5317 | else |
2e1cb74a | 5318 | deactivate_task(rq, prev, 1); |
dd41f596 | 5319 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5320 | } |
5321 | ||
9a897c5a SR |
5322 | #ifdef CONFIG_SMP |
5323 | if (prev->sched_class->pre_schedule) | |
5324 | prev->sched_class->pre_schedule(rq, prev); | |
5325 | #endif | |
f65eda4f | 5326 | |
dd41f596 | 5327 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5328 | idle_balance(cpu, rq); |
1da177e4 | 5329 | |
df1c99d4 | 5330 | put_prev_task(rq, prev); |
b67802ea | 5331 | next = pick_next_task(rq); |
1da177e4 | 5332 | |
1da177e4 | 5333 | if (likely(prev != next)) { |
673a90a1 | 5334 | sched_info_switch(prev, next); |
564c2b21 | 5335 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5336 | |
1da177e4 LT |
5337 | rq->nr_switches++; |
5338 | rq->curr = next; | |
5339 | ++*switch_count; | |
5340 | ||
dd41f596 | 5341 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5342 | /* |
5343 | * the context switch might have flipped the stack from under | |
5344 | * us, hence refresh the local variables. | |
5345 | */ | |
5346 | cpu = smp_processor_id(); | |
5347 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5348 | } else |
5349 | spin_unlock_irq(&rq->lock); | |
5350 | ||
8f4d37ec | 5351 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5352 | goto need_resched_nonpreemptible; |
8f4d37ec | 5353 | |
1da177e4 | 5354 | preempt_enable_no_resched(); |
ff743345 | 5355 | if (need_resched()) |
1da177e4 LT |
5356 | goto need_resched; |
5357 | } | |
1da177e4 LT |
5358 | EXPORT_SYMBOL(schedule); |
5359 | ||
0d66bf6d PZ |
5360 | #ifdef CONFIG_SMP |
5361 | /* | |
5362 | * Look out! "owner" is an entirely speculative pointer | |
5363 | * access and not reliable. | |
5364 | */ | |
5365 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5366 | { | |
5367 | unsigned int cpu; | |
5368 | struct rq *rq; | |
5369 | ||
5370 | if (!sched_feat(OWNER_SPIN)) | |
5371 | return 0; | |
5372 | ||
5373 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5374 | /* | |
5375 | * Need to access the cpu field knowing that | |
5376 | * DEBUG_PAGEALLOC could have unmapped it if | |
5377 | * the mutex owner just released it and exited. | |
5378 | */ | |
5379 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5380 | goto out; | |
5381 | #else | |
5382 | cpu = owner->cpu; | |
5383 | #endif | |
5384 | ||
5385 | /* | |
5386 | * Even if the access succeeded (likely case), | |
5387 | * the cpu field may no longer be valid. | |
5388 | */ | |
5389 | if (cpu >= nr_cpumask_bits) | |
5390 | goto out; | |
5391 | ||
5392 | /* | |
5393 | * We need to validate that we can do a | |
5394 | * get_cpu() and that we have the percpu area. | |
5395 | */ | |
5396 | if (!cpu_online(cpu)) | |
5397 | goto out; | |
5398 | ||
5399 | rq = cpu_rq(cpu); | |
5400 | ||
5401 | for (;;) { | |
5402 | /* | |
5403 | * Owner changed, break to re-assess state. | |
5404 | */ | |
5405 | if (lock->owner != owner) | |
5406 | break; | |
5407 | ||
5408 | /* | |
5409 | * Is that owner really running on that cpu? | |
5410 | */ | |
5411 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5412 | return 0; | |
5413 | ||
5414 | cpu_relax(); | |
5415 | } | |
5416 | out: | |
5417 | return 1; | |
5418 | } | |
5419 | #endif | |
5420 | ||
1da177e4 LT |
5421 | #ifdef CONFIG_PREEMPT |
5422 | /* | |
2ed6e34f | 5423 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5424 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5425 | * occur there and call schedule directly. |
5426 | */ | |
5427 | asmlinkage void __sched preempt_schedule(void) | |
5428 | { | |
5429 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5430 | |
1da177e4 LT |
5431 | /* |
5432 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5433 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5434 | */ |
beed33a8 | 5435 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5436 | return; |
5437 | ||
3a5c359a AK |
5438 | do { |
5439 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5440 | schedule(); |
3a5c359a | 5441 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5442 | |
3a5c359a AK |
5443 | /* |
5444 | * Check again in case we missed a preemption opportunity | |
5445 | * between schedule and now. | |
5446 | */ | |
5447 | barrier(); | |
5ed0cec0 | 5448 | } while (need_resched()); |
1da177e4 | 5449 | } |
1da177e4 LT |
5450 | EXPORT_SYMBOL(preempt_schedule); |
5451 | ||
5452 | /* | |
2ed6e34f | 5453 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5454 | * off of irq context. |
5455 | * Note, that this is called and return with irqs disabled. This will | |
5456 | * protect us against recursive calling from irq. | |
5457 | */ | |
5458 | asmlinkage void __sched preempt_schedule_irq(void) | |
5459 | { | |
5460 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5461 | |
2ed6e34f | 5462 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5463 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5464 | ||
3a5c359a AK |
5465 | do { |
5466 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5467 | local_irq_enable(); |
5468 | schedule(); | |
5469 | local_irq_disable(); | |
3a5c359a | 5470 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5471 | |
3a5c359a AK |
5472 | /* |
5473 | * Check again in case we missed a preemption opportunity | |
5474 | * between schedule and now. | |
5475 | */ | |
5476 | barrier(); | |
5ed0cec0 | 5477 | } while (need_resched()); |
1da177e4 LT |
5478 | } |
5479 | ||
5480 | #endif /* CONFIG_PREEMPT */ | |
5481 | ||
95cdf3b7 IM |
5482 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5483 | void *key) | |
1da177e4 | 5484 | { |
48f24c4d | 5485 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5486 | } |
1da177e4 LT |
5487 | EXPORT_SYMBOL(default_wake_function); |
5488 | ||
5489 | /* | |
41a2d6cf IM |
5490 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5491 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5492 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5493 | * | |
5494 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5495 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5496 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5497 | */ | |
78ddb08f | 5498 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5499 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5500 | { |
2e45874c | 5501 | wait_queue_t *curr, *next; |
1da177e4 | 5502 | |
2e45874c | 5503 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5504 | unsigned flags = curr->flags; |
5505 | ||
1da177e4 | 5506 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5507 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5508 | break; |
5509 | } | |
5510 | } | |
5511 | ||
5512 | /** | |
5513 | * __wake_up - wake up threads blocked on a waitqueue. | |
5514 | * @q: the waitqueue | |
5515 | * @mode: which threads | |
5516 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5517 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5518 | * |
5519 | * It may be assumed that this function implies a write memory barrier before | |
5520 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5521 | */ |
7ad5b3a5 | 5522 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5523 | int nr_exclusive, void *key) |
1da177e4 LT |
5524 | { |
5525 | unsigned long flags; | |
5526 | ||
5527 | spin_lock_irqsave(&q->lock, flags); | |
5528 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5529 | spin_unlock_irqrestore(&q->lock, flags); | |
5530 | } | |
1da177e4 LT |
5531 | EXPORT_SYMBOL(__wake_up); |
5532 | ||
5533 | /* | |
5534 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5535 | */ | |
7ad5b3a5 | 5536 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5537 | { |
5538 | __wake_up_common(q, mode, 1, 0, NULL); | |
5539 | } | |
5540 | ||
4ede816a DL |
5541 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5542 | { | |
5543 | __wake_up_common(q, mode, 1, 0, key); | |
5544 | } | |
5545 | ||
1da177e4 | 5546 | /** |
4ede816a | 5547 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5548 | * @q: the waitqueue |
5549 | * @mode: which threads | |
5550 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5551 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5552 | * |
5553 | * The sync wakeup differs that the waker knows that it will schedule | |
5554 | * away soon, so while the target thread will be woken up, it will not | |
5555 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5556 | * with each other. This can prevent needless bouncing between CPUs. | |
5557 | * | |
5558 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5559 | * |
5560 | * It may be assumed that this function implies a write memory barrier before | |
5561 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5562 | */ |
4ede816a DL |
5563 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5564 | int nr_exclusive, void *key) | |
1da177e4 LT |
5565 | { |
5566 | unsigned long flags; | |
5567 | int sync = 1; | |
5568 | ||
5569 | if (unlikely(!q)) | |
5570 | return; | |
5571 | ||
5572 | if (unlikely(!nr_exclusive)) | |
5573 | sync = 0; | |
5574 | ||
5575 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5576 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5577 | spin_unlock_irqrestore(&q->lock, flags); |
5578 | } | |
4ede816a DL |
5579 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5580 | ||
5581 | /* | |
5582 | * __wake_up_sync - see __wake_up_sync_key() | |
5583 | */ | |
5584 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5585 | { | |
5586 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5587 | } | |
1da177e4 LT |
5588 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5589 | ||
65eb3dc6 KD |
5590 | /** |
5591 | * complete: - signals a single thread waiting on this completion | |
5592 | * @x: holds the state of this particular completion | |
5593 | * | |
5594 | * This will wake up a single thread waiting on this completion. Threads will be | |
5595 | * awakened in the same order in which they were queued. | |
5596 | * | |
5597 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5598 | * |
5599 | * It may be assumed that this function implies a write memory barrier before | |
5600 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5601 | */ |
b15136e9 | 5602 | void complete(struct completion *x) |
1da177e4 LT |
5603 | { |
5604 | unsigned long flags; | |
5605 | ||
5606 | spin_lock_irqsave(&x->wait.lock, flags); | |
5607 | x->done++; | |
d9514f6c | 5608 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5609 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5610 | } | |
5611 | EXPORT_SYMBOL(complete); | |
5612 | ||
65eb3dc6 KD |
5613 | /** |
5614 | * complete_all: - signals all threads waiting on this completion | |
5615 | * @x: holds the state of this particular completion | |
5616 | * | |
5617 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5618 | * |
5619 | * It may be assumed that this function implies a write memory barrier before | |
5620 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5621 | */ |
b15136e9 | 5622 | void complete_all(struct completion *x) |
1da177e4 LT |
5623 | { |
5624 | unsigned long flags; | |
5625 | ||
5626 | spin_lock_irqsave(&x->wait.lock, flags); | |
5627 | x->done += UINT_MAX/2; | |
d9514f6c | 5628 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5629 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5630 | } | |
5631 | EXPORT_SYMBOL(complete_all); | |
5632 | ||
8cbbe86d AK |
5633 | static inline long __sched |
5634 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5635 | { |
1da177e4 LT |
5636 | if (!x->done) { |
5637 | DECLARE_WAITQUEUE(wait, current); | |
5638 | ||
5639 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5640 | __add_wait_queue_tail(&x->wait, &wait); | |
5641 | do { | |
94d3d824 | 5642 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5643 | timeout = -ERESTARTSYS; |
5644 | break; | |
8cbbe86d AK |
5645 | } |
5646 | __set_current_state(state); | |
1da177e4 LT |
5647 | spin_unlock_irq(&x->wait.lock); |
5648 | timeout = schedule_timeout(timeout); | |
5649 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5650 | } while (!x->done && timeout); |
1da177e4 | 5651 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5652 | if (!x->done) |
5653 | return timeout; | |
1da177e4 LT |
5654 | } |
5655 | x->done--; | |
ea71a546 | 5656 | return timeout ?: 1; |
1da177e4 | 5657 | } |
1da177e4 | 5658 | |
8cbbe86d AK |
5659 | static long __sched |
5660 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5661 | { |
1da177e4 LT |
5662 | might_sleep(); |
5663 | ||
5664 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5665 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5666 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5667 | return timeout; |
5668 | } | |
1da177e4 | 5669 | |
65eb3dc6 KD |
5670 | /** |
5671 | * wait_for_completion: - waits for completion of a task | |
5672 | * @x: holds the state of this particular completion | |
5673 | * | |
5674 | * This waits to be signaled for completion of a specific task. It is NOT | |
5675 | * interruptible and there is no timeout. | |
5676 | * | |
5677 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5678 | * and interrupt capability. Also see complete(). | |
5679 | */ | |
b15136e9 | 5680 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5681 | { |
5682 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5683 | } |
8cbbe86d | 5684 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5685 | |
65eb3dc6 KD |
5686 | /** |
5687 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5688 | * @x: holds the state of this particular completion | |
5689 | * @timeout: timeout value in jiffies | |
5690 | * | |
5691 | * This waits for either a completion of a specific task to be signaled or for a | |
5692 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5693 | * interruptible. | |
5694 | */ | |
b15136e9 | 5695 | unsigned long __sched |
8cbbe86d | 5696 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5697 | { |
8cbbe86d | 5698 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5699 | } |
8cbbe86d | 5700 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5701 | |
65eb3dc6 KD |
5702 | /** |
5703 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5704 | * @x: holds the state of this particular completion | |
5705 | * | |
5706 | * This waits for completion of a specific task to be signaled. It is | |
5707 | * interruptible. | |
5708 | */ | |
8cbbe86d | 5709 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5710 | { |
51e97990 AK |
5711 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5712 | if (t == -ERESTARTSYS) | |
5713 | return t; | |
5714 | return 0; | |
0fec171c | 5715 | } |
8cbbe86d | 5716 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5717 | |
65eb3dc6 KD |
5718 | /** |
5719 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5720 | * @x: holds the state of this particular completion | |
5721 | * @timeout: timeout value in jiffies | |
5722 | * | |
5723 | * This waits for either a completion of a specific task to be signaled or for a | |
5724 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5725 | */ | |
b15136e9 | 5726 | unsigned long __sched |
8cbbe86d AK |
5727 | wait_for_completion_interruptible_timeout(struct completion *x, |
5728 | unsigned long timeout) | |
0fec171c | 5729 | { |
8cbbe86d | 5730 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5731 | } |
8cbbe86d | 5732 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5733 | |
65eb3dc6 KD |
5734 | /** |
5735 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5736 | * @x: holds the state of this particular completion | |
5737 | * | |
5738 | * This waits to be signaled for completion of a specific task. It can be | |
5739 | * interrupted by a kill signal. | |
5740 | */ | |
009e577e MW |
5741 | int __sched wait_for_completion_killable(struct completion *x) |
5742 | { | |
5743 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5744 | if (t == -ERESTARTSYS) | |
5745 | return t; | |
5746 | return 0; | |
5747 | } | |
5748 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5749 | ||
be4de352 DC |
5750 | /** |
5751 | * try_wait_for_completion - try to decrement a completion without blocking | |
5752 | * @x: completion structure | |
5753 | * | |
5754 | * Returns: 0 if a decrement cannot be done without blocking | |
5755 | * 1 if a decrement succeeded. | |
5756 | * | |
5757 | * If a completion is being used as a counting completion, | |
5758 | * attempt to decrement the counter without blocking. This | |
5759 | * enables us to avoid waiting if the resource the completion | |
5760 | * is protecting is not available. | |
5761 | */ | |
5762 | bool try_wait_for_completion(struct completion *x) | |
5763 | { | |
5764 | int ret = 1; | |
5765 | ||
5766 | spin_lock_irq(&x->wait.lock); | |
5767 | if (!x->done) | |
5768 | ret = 0; | |
5769 | else | |
5770 | x->done--; | |
5771 | spin_unlock_irq(&x->wait.lock); | |
5772 | return ret; | |
5773 | } | |
5774 | EXPORT_SYMBOL(try_wait_for_completion); | |
5775 | ||
5776 | /** | |
5777 | * completion_done - Test to see if a completion has any waiters | |
5778 | * @x: completion structure | |
5779 | * | |
5780 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5781 | * 1 if there are no waiters. | |
5782 | * | |
5783 | */ | |
5784 | bool completion_done(struct completion *x) | |
5785 | { | |
5786 | int ret = 1; | |
5787 | ||
5788 | spin_lock_irq(&x->wait.lock); | |
5789 | if (!x->done) | |
5790 | ret = 0; | |
5791 | spin_unlock_irq(&x->wait.lock); | |
5792 | return ret; | |
5793 | } | |
5794 | EXPORT_SYMBOL(completion_done); | |
5795 | ||
8cbbe86d AK |
5796 | static long __sched |
5797 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5798 | { |
0fec171c IM |
5799 | unsigned long flags; |
5800 | wait_queue_t wait; | |
5801 | ||
5802 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5803 | |
8cbbe86d | 5804 | __set_current_state(state); |
1da177e4 | 5805 | |
8cbbe86d AK |
5806 | spin_lock_irqsave(&q->lock, flags); |
5807 | __add_wait_queue(q, &wait); | |
5808 | spin_unlock(&q->lock); | |
5809 | timeout = schedule_timeout(timeout); | |
5810 | spin_lock_irq(&q->lock); | |
5811 | __remove_wait_queue(q, &wait); | |
5812 | spin_unlock_irqrestore(&q->lock, flags); | |
5813 | ||
5814 | return timeout; | |
5815 | } | |
5816 | ||
5817 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5818 | { | |
5819 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5820 | } |
1da177e4 LT |
5821 | EXPORT_SYMBOL(interruptible_sleep_on); |
5822 | ||
0fec171c | 5823 | long __sched |
95cdf3b7 | 5824 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5825 | { |
8cbbe86d | 5826 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5827 | } |
1da177e4 LT |
5828 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5829 | ||
0fec171c | 5830 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5831 | { |
8cbbe86d | 5832 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5833 | } |
1da177e4 LT |
5834 | EXPORT_SYMBOL(sleep_on); |
5835 | ||
0fec171c | 5836 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5837 | { |
8cbbe86d | 5838 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5839 | } |
1da177e4 LT |
5840 | EXPORT_SYMBOL(sleep_on_timeout); |
5841 | ||
b29739f9 IM |
5842 | #ifdef CONFIG_RT_MUTEXES |
5843 | ||
5844 | /* | |
5845 | * rt_mutex_setprio - set the current priority of a task | |
5846 | * @p: task | |
5847 | * @prio: prio value (kernel-internal form) | |
5848 | * | |
5849 | * This function changes the 'effective' priority of a task. It does | |
5850 | * not touch ->normal_prio like __setscheduler(). | |
5851 | * | |
5852 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5853 | */ | |
36c8b586 | 5854 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5855 | { |
5856 | unsigned long flags; | |
83b699ed | 5857 | int oldprio, on_rq, running; |
70b97a7f | 5858 | struct rq *rq; |
cb469845 | 5859 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5860 | |
5861 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5862 | ||
5863 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5864 | update_rq_clock(rq); |
b29739f9 | 5865 | |
d5f9f942 | 5866 | oldprio = p->prio; |
dd41f596 | 5867 | on_rq = p->se.on_rq; |
051a1d1a | 5868 | running = task_current(rq, p); |
0e1f3483 | 5869 | if (on_rq) |
69be72c1 | 5870 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5871 | if (running) |
5872 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5873 | |
5874 | if (rt_prio(prio)) | |
5875 | p->sched_class = &rt_sched_class; | |
5876 | else | |
5877 | p->sched_class = &fair_sched_class; | |
5878 | ||
b29739f9 IM |
5879 | p->prio = prio; |
5880 | ||
0e1f3483 HS |
5881 | if (running) |
5882 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5883 | if (on_rq) { |
8159f87e | 5884 | enqueue_task(rq, p, 0); |
cb469845 SR |
5885 | |
5886 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5887 | } |
5888 | task_rq_unlock(rq, &flags); | |
5889 | } | |
5890 | ||
5891 | #endif | |
5892 | ||
36c8b586 | 5893 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5894 | { |
dd41f596 | 5895 | int old_prio, delta, on_rq; |
1da177e4 | 5896 | unsigned long flags; |
70b97a7f | 5897 | struct rq *rq; |
1da177e4 LT |
5898 | |
5899 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5900 | return; | |
5901 | /* | |
5902 | * We have to be careful, if called from sys_setpriority(), | |
5903 | * the task might be in the middle of scheduling on another CPU. | |
5904 | */ | |
5905 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5906 | update_rq_clock(rq); |
1da177e4 LT |
5907 | /* |
5908 | * The RT priorities are set via sched_setscheduler(), but we still | |
5909 | * allow the 'normal' nice value to be set - but as expected | |
5910 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5911 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5912 | */ |
e05606d3 | 5913 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5914 | p->static_prio = NICE_TO_PRIO(nice); |
5915 | goto out_unlock; | |
5916 | } | |
dd41f596 | 5917 | on_rq = p->se.on_rq; |
c09595f6 | 5918 | if (on_rq) |
69be72c1 | 5919 | dequeue_task(rq, p, 0); |
1da177e4 | 5920 | |
1da177e4 | 5921 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5922 | set_load_weight(p); |
b29739f9 IM |
5923 | old_prio = p->prio; |
5924 | p->prio = effective_prio(p); | |
5925 | delta = p->prio - old_prio; | |
1da177e4 | 5926 | |
dd41f596 | 5927 | if (on_rq) { |
8159f87e | 5928 | enqueue_task(rq, p, 0); |
1da177e4 | 5929 | /* |
d5f9f942 AM |
5930 | * If the task increased its priority or is running and |
5931 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5932 | */ |
d5f9f942 | 5933 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5934 | resched_task(rq->curr); |
5935 | } | |
5936 | out_unlock: | |
5937 | task_rq_unlock(rq, &flags); | |
5938 | } | |
1da177e4 LT |
5939 | EXPORT_SYMBOL(set_user_nice); |
5940 | ||
e43379f1 MM |
5941 | /* |
5942 | * can_nice - check if a task can reduce its nice value | |
5943 | * @p: task | |
5944 | * @nice: nice value | |
5945 | */ | |
36c8b586 | 5946 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5947 | { |
024f4747 MM |
5948 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5949 | int nice_rlim = 20 - nice; | |
48f24c4d | 5950 | |
e43379f1 MM |
5951 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5952 | capable(CAP_SYS_NICE)); | |
5953 | } | |
5954 | ||
1da177e4 LT |
5955 | #ifdef __ARCH_WANT_SYS_NICE |
5956 | ||
5957 | /* | |
5958 | * sys_nice - change the priority of the current process. | |
5959 | * @increment: priority increment | |
5960 | * | |
5961 | * sys_setpriority is a more generic, but much slower function that | |
5962 | * does similar things. | |
5963 | */ | |
5add95d4 | 5964 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5965 | { |
48f24c4d | 5966 | long nice, retval; |
1da177e4 LT |
5967 | |
5968 | /* | |
5969 | * Setpriority might change our priority at the same moment. | |
5970 | * We don't have to worry. Conceptually one call occurs first | |
5971 | * and we have a single winner. | |
5972 | */ | |
e43379f1 MM |
5973 | if (increment < -40) |
5974 | increment = -40; | |
1da177e4 LT |
5975 | if (increment > 40) |
5976 | increment = 40; | |
5977 | ||
2b8f836f | 5978 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5979 | if (nice < -20) |
5980 | nice = -20; | |
5981 | if (nice > 19) | |
5982 | nice = 19; | |
5983 | ||
e43379f1 MM |
5984 | if (increment < 0 && !can_nice(current, nice)) |
5985 | return -EPERM; | |
5986 | ||
1da177e4 LT |
5987 | retval = security_task_setnice(current, nice); |
5988 | if (retval) | |
5989 | return retval; | |
5990 | ||
5991 | set_user_nice(current, nice); | |
5992 | return 0; | |
5993 | } | |
5994 | ||
5995 | #endif | |
5996 | ||
5997 | /** | |
5998 | * task_prio - return the priority value of a given task. | |
5999 | * @p: the task in question. | |
6000 | * | |
6001 | * This is the priority value as seen by users in /proc. | |
6002 | * RT tasks are offset by -200. Normal tasks are centered | |
6003 | * around 0, value goes from -16 to +15. | |
6004 | */ | |
36c8b586 | 6005 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6006 | { |
6007 | return p->prio - MAX_RT_PRIO; | |
6008 | } | |
6009 | ||
6010 | /** | |
6011 | * task_nice - return the nice value of a given task. | |
6012 | * @p: the task in question. | |
6013 | */ | |
36c8b586 | 6014 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6015 | { |
6016 | return TASK_NICE(p); | |
6017 | } | |
150d8bed | 6018 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6019 | |
6020 | /** | |
6021 | * idle_cpu - is a given cpu idle currently? | |
6022 | * @cpu: the processor in question. | |
6023 | */ | |
6024 | int idle_cpu(int cpu) | |
6025 | { | |
6026 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6027 | } | |
6028 | ||
1da177e4 LT |
6029 | /** |
6030 | * idle_task - return the idle task for a given cpu. | |
6031 | * @cpu: the processor in question. | |
6032 | */ | |
36c8b586 | 6033 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6034 | { |
6035 | return cpu_rq(cpu)->idle; | |
6036 | } | |
6037 | ||
6038 | /** | |
6039 | * find_process_by_pid - find a process with a matching PID value. | |
6040 | * @pid: the pid in question. | |
6041 | */ | |
a9957449 | 6042 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6043 | { |
228ebcbe | 6044 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6045 | } |
6046 | ||
6047 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6048 | static void |
6049 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6050 | { |
dd41f596 | 6051 | BUG_ON(p->se.on_rq); |
48f24c4d | 6052 | |
1da177e4 | 6053 | p->policy = policy; |
dd41f596 IM |
6054 | switch (p->policy) { |
6055 | case SCHED_NORMAL: | |
6056 | case SCHED_BATCH: | |
6057 | case SCHED_IDLE: | |
6058 | p->sched_class = &fair_sched_class; | |
6059 | break; | |
6060 | case SCHED_FIFO: | |
6061 | case SCHED_RR: | |
6062 | p->sched_class = &rt_sched_class; | |
6063 | break; | |
6064 | } | |
6065 | ||
1da177e4 | 6066 | p->rt_priority = prio; |
b29739f9 IM |
6067 | p->normal_prio = normal_prio(p); |
6068 | /* we are holding p->pi_lock already */ | |
6069 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6070 | set_load_weight(p); |
1da177e4 LT |
6071 | } |
6072 | ||
c69e8d9c DH |
6073 | /* |
6074 | * check the target process has a UID that matches the current process's | |
6075 | */ | |
6076 | static bool check_same_owner(struct task_struct *p) | |
6077 | { | |
6078 | const struct cred *cred = current_cred(), *pcred; | |
6079 | bool match; | |
6080 | ||
6081 | rcu_read_lock(); | |
6082 | pcred = __task_cred(p); | |
6083 | match = (cred->euid == pcred->euid || | |
6084 | cred->euid == pcred->uid); | |
6085 | rcu_read_unlock(); | |
6086 | return match; | |
6087 | } | |
6088 | ||
961ccddd RR |
6089 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6090 | struct sched_param *param, bool user) | |
1da177e4 | 6091 | { |
83b699ed | 6092 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6093 | unsigned long flags; |
cb469845 | 6094 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6095 | struct rq *rq; |
1da177e4 | 6096 | |
66e5393a SR |
6097 | /* may grab non-irq protected spin_locks */ |
6098 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6099 | recheck: |
6100 | /* double check policy once rq lock held */ | |
6101 | if (policy < 0) | |
6102 | policy = oldpolicy = p->policy; | |
6103 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
6104 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
6105 | policy != SCHED_IDLE) | |
b0a9499c | 6106 | return -EINVAL; |
1da177e4 LT |
6107 | /* |
6108 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6109 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6110 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6111 | */ |
6112 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6113 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6114 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6115 | return -EINVAL; |
e05606d3 | 6116 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6117 | return -EINVAL; |
6118 | ||
37e4ab3f OC |
6119 | /* |
6120 | * Allow unprivileged RT tasks to decrease priority: | |
6121 | */ | |
961ccddd | 6122 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6123 | if (rt_policy(policy)) { |
8dc3e909 | 6124 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6125 | |
6126 | if (!lock_task_sighand(p, &flags)) | |
6127 | return -ESRCH; | |
6128 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6129 | unlock_task_sighand(p, &flags); | |
6130 | ||
6131 | /* can't set/change the rt policy */ | |
6132 | if (policy != p->policy && !rlim_rtprio) | |
6133 | return -EPERM; | |
6134 | ||
6135 | /* can't increase priority */ | |
6136 | if (param->sched_priority > p->rt_priority && | |
6137 | param->sched_priority > rlim_rtprio) | |
6138 | return -EPERM; | |
6139 | } | |
dd41f596 IM |
6140 | /* |
6141 | * Like positive nice levels, dont allow tasks to | |
6142 | * move out of SCHED_IDLE either: | |
6143 | */ | |
6144 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6145 | return -EPERM; | |
5fe1d75f | 6146 | |
37e4ab3f | 6147 | /* can't change other user's priorities */ |
c69e8d9c | 6148 | if (!check_same_owner(p)) |
37e4ab3f OC |
6149 | return -EPERM; |
6150 | } | |
1da177e4 | 6151 | |
725aad24 | 6152 | if (user) { |
b68aa230 | 6153 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6154 | /* |
6155 | * Do not allow realtime tasks into groups that have no runtime | |
6156 | * assigned. | |
6157 | */ | |
9a7e0b18 PZ |
6158 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6159 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6160 | return -EPERM; |
b68aa230 PZ |
6161 | #endif |
6162 | ||
725aad24 JF |
6163 | retval = security_task_setscheduler(p, policy, param); |
6164 | if (retval) | |
6165 | return retval; | |
6166 | } | |
6167 | ||
b29739f9 IM |
6168 | /* |
6169 | * make sure no PI-waiters arrive (or leave) while we are | |
6170 | * changing the priority of the task: | |
6171 | */ | |
6172 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6173 | /* |
6174 | * To be able to change p->policy safely, the apropriate | |
6175 | * runqueue lock must be held. | |
6176 | */ | |
b29739f9 | 6177 | rq = __task_rq_lock(p); |
1da177e4 LT |
6178 | /* recheck policy now with rq lock held */ |
6179 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6180 | policy = oldpolicy = -1; | |
b29739f9 IM |
6181 | __task_rq_unlock(rq); |
6182 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6183 | goto recheck; |
6184 | } | |
2daa3577 | 6185 | update_rq_clock(rq); |
dd41f596 | 6186 | on_rq = p->se.on_rq; |
051a1d1a | 6187 | running = task_current(rq, p); |
0e1f3483 | 6188 | if (on_rq) |
2e1cb74a | 6189 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6190 | if (running) |
6191 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6192 | |
1da177e4 | 6193 | oldprio = p->prio; |
dd41f596 | 6194 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6195 | |
0e1f3483 HS |
6196 | if (running) |
6197 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6198 | if (on_rq) { |
6199 | activate_task(rq, p, 0); | |
cb469845 SR |
6200 | |
6201 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6202 | } |
b29739f9 IM |
6203 | __task_rq_unlock(rq); |
6204 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6205 | ||
95e02ca9 TG |
6206 | rt_mutex_adjust_pi(p); |
6207 | ||
1da177e4 LT |
6208 | return 0; |
6209 | } | |
961ccddd RR |
6210 | |
6211 | /** | |
6212 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6213 | * @p: the task in question. | |
6214 | * @policy: new policy. | |
6215 | * @param: structure containing the new RT priority. | |
6216 | * | |
6217 | * NOTE that the task may be already dead. | |
6218 | */ | |
6219 | int sched_setscheduler(struct task_struct *p, int policy, | |
6220 | struct sched_param *param) | |
6221 | { | |
6222 | return __sched_setscheduler(p, policy, param, true); | |
6223 | } | |
1da177e4 LT |
6224 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6225 | ||
961ccddd RR |
6226 | /** |
6227 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6228 | * @p: the task in question. | |
6229 | * @policy: new policy. | |
6230 | * @param: structure containing the new RT priority. | |
6231 | * | |
6232 | * Just like sched_setscheduler, only don't bother checking if the | |
6233 | * current context has permission. For example, this is needed in | |
6234 | * stop_machine(): we create temporary high priority worker threads, | |
6235 | * but our caller might not have that capability. | |
6236 | */ | |
6237 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6238 | struct sched_param *param) | |
6239 | { | |
6240 | return __sched_setscheduler(p, policy, param, false); | |
6241 | } | |
6242 | ||
95cdf3b7 IM |
6243 | static int |
6244 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6245 | { |
1da177e4 LT |
6246 | struct sched_param lparam; |
6247 | struct task_struct *p; | |
36c8b586 | 6248 | int retval; |
1da177e4 LT |
6249 | |
6250 | if (!param || pid < 0) | |
6251 | return -EINVAL; | |
6252 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6253 | return -EFAULT; | |
5fe1d75f ON |
6254 | |
6255 | rcu_read_lock(); | |
6256 | retval = -ESRCH; | |
1da177e4 | 6257 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6258 | if (p != NULL) |
6259 | retval = sched_setscheduler(p, policy, &lparam); | |
6260 | rcu_read_unlock(); | |
36c8b586 | 6261 | |
1da177e4 LT |
6262 | return retval; |
6263 | } | |
6264 | ||
6265 | /** | |
6266 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6267 | * @pid: the pid in question. | |
6268 | * @policy: new policy. | |
6269 | * @param: structure containing the new RT priority. | |
6270 | */ | |
5add95d4 HC |
6271 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6272 | struct sched_param __user *, param) | |
1da177e4 | 6273 | { |
c21761f1 JB |
6274 | /* negative values for policy are not valid */ |
6275 | if (policy < 0) | |
6276 | return -EINVAL; | |
6277 | ||
1da177e4 LT |
6278 | return do_sched_setscheduler(pid, policy, param); |
6279 | } | |
6280 | ||
6281 | /** | |
6282 | * sys_sched_setparam - set/change the RT priority of a thread | |
6283 | * @pid: the pid in question. | |
6284 | * @param: structure containing the new RT priority. | |
6285 | */ | |
5add95d4 | 6286 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6287 | { |
6288 | return do_sched_setscheduler(pid, -1, param); | |
6289 | } | |
6290 | ||
6291 | /** | |
6292 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6293 | * @pid: the pid in question. | |
6294 | */ | |
5add95d4 | 6295 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6296 | { |
36c8b586 | 6297 | struct task_struct *p; |
3a5c359a | 6298 | int retval; |
1da177e4 LT |
6299 | |
6300 | if (pid < 0) | |
3a5c359a | 6301 | return -EINVAL; |
1da177e4 LT |
6302 | |
6303 | retval = -ESRCH; | |
6304 | read_lock(&tasklist_lock); | |
6305 | p = find_process_by_pid(pid); | |
6306 | if (p) { | |
6307 | retval = security_task_getscheduler(p); | |
6308 | if (!retval) | |
6309 | retval = p->policy; | |
6310 | } | |
6311 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6312 | return retval; |
6313 | } | |
6314 | ||
6315 | /** | |
6316 | * sys_sched_getscheduler - get the RT priority of a thread | |
6317 | * @pid: the pid in question. | |
6318 | * @param: structure containing the RT priority. | |
6319 | */ | |
5add95d4 | 6320 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6321 | { |
6322 | struct sched_param lp; | |
36c8b586 | 6323 | struct task_struct *p; |
3a5c359a | 6324 | int retval; |
1da177e4 LT |
6325 | |
6326 | if (!param || pid < 0) | |
3a5c359a | 6327 | return -EINVAL; |
1da177e4 LT |
6328 | |
6329 | read_lock(&tasklist_lock); | |
6330 | p = find_process_by_pid(pid); | |
6331 | retval = -ESRCH; | |
6332 | if (!p) | |
6333 | goto out_unlock; | |
6334 | ||
6335 | retval = security_task_getscheduler(p); | |
6336 | if (retval) | |
6337 | goto out_unlock; | |
6338 | ||
6339 | lp.sched_priority = p->rt_priority; | |
6340 | read_unlock(&tasklist_lock); | |
6341 | ||
6342 | /* | |
6343 | * This one might sleep, we cannot do it with a spinlock held ... | |
6344 | */ | |
6345 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6346 | ||
1da177e4 LT |
6347 | return retval; |
6348 | ||
6349 | out_unlock: | |
6350 | read_unlock(&tasklist_lock); | |
6351 | return retval; | |
6352 | } | |
6353 | ||
96f874e2 | 6354 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6355 | { |
5a16f3d3 | 6356 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6357 | struct task_struct *p; |
6358 | int retval; | |
1da177e4 | 6359 | |
95402b38 | 6360 | get_online_cpus(); |
1da177e4 LT |
6361 | read_lock(&tasklist_lock); |
6362 | ||
6363 | p = find_process_by_pid(pid); | |
6364 | if (!p) { | |
6365 | read_unlock(&tasklist_lock); | |
95402b38 | 6366 | put_online_cpus(); |
1da177e4 LT |
6367 | return -ESRCH; |
6368 | } | |
6369 | ||
6370 | /* | |
6371 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6372 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6373 | * usage count and then drop tasklist_lock. |
6374 | */ | |
6375 | get_task_struct(p); | |
6376 | read_unlock(&tasklist_lock); | |
6377 | ||
5a16f3d3 RR |
6378 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6379 | retval = -ENOMEM; | |
6380 | goto out_put_task; | |
6381 | } | |
6382 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6383 | retval = -ENOMEM; | |
6384 | goto out_free_cpus_allowed; | |
6385 | } | |
1da177e4 | 6386 | retval = -EPERM; |
c69e8d9c | 6387 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6388 | goto out_unlock; |
6389 | ||
e7834f8f DQ |
6390 | retval = security_task_setscheduler(p, 0, NULL); |
6391 | if (retval) | |
6392 | goto out_unlock; | |
6393 | ||
5a16f3d3 RR |
6394 | cpuset_cpus_allowed(p, cpus_allowed); |
6395 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6396 | again: |
5a16f3d3 | 6397 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6398 | |
8707d8b8 | 6399 | if (!retval) { |
5a16f3d3 RR |
6400 | cpuset_cpus_allowed(p, cpus_allowed); |
6401 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6402 | /* |
6403 | * We must have raced with a concurrent cpuset | |
6404 | * update. Just reset the cpus_allowed to the | |
6405 | * cpuset's cpus_allowed | |
6406 | */ | |
5a16f3d3 | 6407 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6408 | goto again; |
6409 | } | |
6410 | } | |
1da177e4 | 6411 | out_unlock: |
5a16f3d3 RR |
6412 | free_cpumask_var(new_mask); |
6413 | out_free_cpus_allowed: | |
6414 | free_cpumask_var(cpus_allowed); | |
6415 | out_put_task: | |
1da177e4 | 6416 | put_task_struct(p); |
95402b38 | 6417 | put_online_cpus(); |
1da177e4 LT |
6418 | return retval; |
6419 | } | |
6420 | ||
6421 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6422 | struct cpumask *new_mask) |
1da177e4 | 6423 | { |
96f874e2 RR |
6424 | if (len < cpumask_size()) |
6425 | cpumask_clear(new_mask); | |
6426 | else if (len > cpumask_size()) | |
6427 | len = cpumask_size(); | |
6428 | ||
1da177e4 LT |
6429 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6430 | } | |
6431 | ||
6432 | /** | |
6433 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6434 | * @pid: pid of the process | |
6435 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6436 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6437 | */ | |
5add95d4 HC |
6438 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6439 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6440 | { |
5a16f3d3 | 6441 | cpumask_var_t new_mask; |
1da177e4 LT |
6442 | int retval; |
6443 | ||
5a16f3d3 RR |
6444 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6445 | return -ENOMEM; | |
1da177e4 | 6446 | |
5a16f3d3 RR |
6447 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6448 | if (retval == 0) | |
6449 | retval = sched_setaffinity(pid, new_mask); | |
6450 | free_cpumask_var(new_mask); | |
6451 | return retval; | |
1da177e4 LT |
6452 | } |
6453 | ||
96f874e2 | 6454 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6455 | { |
36c8b586 | 6456 | struct task_struct *p; |
1da177e4 | 6457 | int retval; |
1da177e4 | 6458 | |
95402b38 | 6459 | get_online_cpus(); |
1da177e4 LT |
6460 | read_lock(&tasklist_lock); |
6461 | ||
6462 | retval = -ESRCH; | |
6463 | p = find_process_by_pid(pid); | |
6464 | if (!p) | |
6465 | goto out_unlock; | |
6466 | ||
e7834f8f DQ |
6467 | retval = security_task_getscheduler(p); |
6468 | if (retval) | |
6469 | goto out_unlock; | |
6470 | ||
96f874e2 | 6471 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6472 | |
6473 | out_unlock: | |
6474 | read_unlock(&tasklist_lock); | |
95402b38 | 6475 | put_online_cpus(); |
1da177e4 | 6476 | |
9531b62f | 6477 | return retval; |
1da177e4 LT |
6478 | } |
6479 | ||
6480 | /** | |
6481 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6482 | * @pid: pid of the process | |
6483 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6484 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6485 | */ | |
5add95d4 HC |
6486 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6487 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6488 | { |
6489 | int ret; | |
f17c8607 | 6490 | cpumask_var_t mask; |
1da177e4 | 6491 | |
f17c8607 | 6492 | if (len < cpumask_size()) |
1da177e4 LT |
6493 | return -EINVAL; |
6494 | ||
f17c8607 RR |
6495 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6496 | return -ENOMEM; | |
1da177e4 | 6497 | |
f17c8607 RR |
6498 | ret = sched_getaffinity(pid, mask); |
6499 | if (ret == 0) { | |
6500 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6501 | ret = -EFAULT; | |
6502 | else | |
6503 | ret = cpumask_size(); | |
6504 | } | |
6505 | free_cpumask_var(mask); | |
1da177e4 | 6506 | |
f17c8607 | 6507 | return ret; |
1da177e4 LT |
6508 | } |
6509 | ||
6510 | /** | |
6511 | * sys_sched_yield - yield the current processor to other threads. | |
6512 | * | |
dd41f596 IM |
6513 | * This function yields the current CPU to other tasks. If there are no |
6514 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6515 | */ |
5add95d4 | 6516 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6517 | { |
70b97a7f | 6518 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6519 | |
2d72376b | 6520 | schedstat_inc(rq, yld_count); |
4530d7ab | 6521 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6522 | |
6523 | /* | |
6524 | * Since we are going to call schedule() anyway, there's | |
6525 | * no need to preempt or enable interrupts: | |
6526 | */ | |
6527 | __release(rq->lock); | |
8a25d5de | 6528 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6529 | _raw_spin_unlock(&rq->lock); |
6530 | preempt_enable_no_resched(); | |
6531 | ||
6532 | schedule(); | |
6533 | ||
6534 | return 0; | |
6535 | } | |
6536 | ||
e7b38404 | 6537 | static void __cond_resched(void) |
1da177e4 | 6538 | { |
8e0a43d8 IM |
6539 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6540 | __might_sleep(__FILE__, __LINE__); | |
6541 | #endif | |
5bbcfd90 IM |
6542 | /* |
6543 | * The BKS might be reacquired before we have dropped | |
6544 | * PREEMPT_ACTIVE, which could trigger a second | |
6545 | * cond_resched() call. | |
6546 | */ | |
1da177e4 LT |
6547 | do { |
6548 | add_preempt_count(PREEMPT_ACTIVE); | |
6549 | schedule(); | |
6550 | sub_preempt_count(PREEMPT_ACTIVE); | |
6551 | } while (need_resched()); | |
6552 | } | |
6553 | ||
02b67cc3 | 6554 | int __sched _cond_resched(void) |
1da177e4 | 6555 | { |
9414232f IM |
6556 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6557 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6558 | __cond_resched(); |
6559 | return 1; | |
6560 | } | |
6561 | return 0; | |
6562 | } | |
02b67cc3 | 6563 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6564 | |
6565 | /* | |
6566 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6567 | * call schedule, and on return reacquire the lock. | |
6568 | * | |
41a2d6cf | 6569 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6570 | * operations here to prevent schedule() from being called twice (once via |
6571 | * spin_unlock(), once by hand). | |
6572 | */ | |
95cdf3b7 | 6573 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6574 | { |
95c354fe | 6575 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6576 | int ret = 0; |
6577 | ||
95c354fe | 6578 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6579 | spin_unlock(lock); |
95c354fe NP |
6580 | if (resched && need_resched()) |
6581 | __cond_resched(); | |
6582 | else | |
6583 | cpu_relax(); | |
6df3cecb | 6584 | ret = 1; |
1da177e4 | 6585 | spin_lock(lock); |
1da177e4 | 6586 | } |
6df3cecb | 6587 | return ret; |
1da177e4 | 6588 | } |
1da177e4 LT |
6589 | EXPORT_SYMBOL(cond_resched_lock); |
6590 | ||
6591 | int __sched cond_resched_softirq(void) | |
6592 | { | |
6593 | BUG_ON(!in_softirq()); | |
6594 | ||
9414232f | 6595 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6596 | local_bh_enable(); |
1da177e4 LT |
6597 | __cond_resched(); |
6598 | local_bh_disable(); | |
6599 | return 1; | |
6600 | } | |
6601 | return 0; | |
6602 | } | |
1da177e4 LT |
6603 | EXPORT_SYMBOL(cond_resched_softirq); |
6604 | ||
1da177e4 LT |
6605 | /** |
6606 | * yield - yield the current processor to other threads. | |
6607 | * | |
72fd4a35 | 6608 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6609 | * thread runnable and calls sys_sched_yield(). |
6610 | */ | |
6611 | void __sched yield(void) | |
6612 | { | |
6613 | set_current_state(TASK_RUNNING); | |
6614 | sys_sched_yield(); | |
6615 | } | |
1da177e4 LT |
6616 | EXPORT_SYMBOL(yield); |
6617 | ||
6618 | /* | |
41a2d6cf | 6619 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6620 | * that process accounting knows that this is a task in IO wait state. |
6621 | * | |
6622 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6623 | * has set its backing_dev_info: the queue against which it should throttle) | |
6624 | */ | |
6625 | void __sched io_schedule(void) | |
6626 | { | |
70b97a7f | 6627 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6628 | |
0ff92245 | 6629 | delayacct_blkio_start(); |
1da177e4 LT |
6630 | atomic_inc(&rq->nr_iowait); |
6631 | schedule(); | |
6632 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6633 | delayacct_blkio_end(); |
1da177e4 | 6634 | } |
1da177e4 LT |
6635 | EXPORT_SYMBOL(io_schedule); |
6636 | ||
6637 | long __sched io_schedule_timeout(long timeout) | |
6638 | { | |
70b97a7f | 6639 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6640 | long ret; |
6641 | ||
0ff92245 | 6642 | delayacct_blkio_start(); |
1da177e4 LT |
6643 | atomic_inc(&rq->nr_iowait); |
6644 | ret = schedule_timeout(timeout); | |
6645 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6646 | delayacct_blkio_end(); |
1da177e4 LT |
6647 | return ret; |
6648 | } | |
6649 | ||
6650 | /** | |
6651 | * sys_sched_get_priority_max - return maximum RT priority. | |
6652 | * @policy: scheduling class. | |
6653 | * | |
6654 | * this syscall returns the maximum rt_priority that can be used | |
6655 | * by a given scheduling class. | |
6656 | */ | |
5add95d4 | 6657 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6658 | { |
6659 | int ret = -EINVAL; | |
6660 | ||
6661 | switch (policy) { | |
6662 | case SCHED_FIFO: | |
6663 | case SCHED_RR: | |
6664 | ret = MAX_USER_RT_PRIO-1; | |
6665 | break; | |
6666 | case SCHED_NORMAL: | |
b0a9499c | 6667 | case SCHED_BATCH: |
dd41f596 | 6668 | case SCHED_IDLE: |
1da177e4 LT |
6669 | ret = 0; |
6670 | break; | |
6671 | } | |
6672 | return ret; | |
6673 | } | |
6674 | ||
6675 | /** | |
6676 | * sys_sched_get_priority_min - return minimum RT priority. | |
6677 | * @policy: scheduling class. | |
6678 | * | |
6679 | * this syscall returns the minimum rt_priority that can be used | |
6680 | * by a given scheduling class. | |
6681 | */ | |
5add95d4 | 6682 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6683 | { |
6684 | int ret = -EINVAL; | |
6685 | ||
6686 | switch (policy) { | |
6687 | case SCHED_FIFO: | |
6688 | case SCHED_RR: | |
6689 | ret = 1; | |
6690 | break; | |
6691 | case SCHED_NORMAL: | |
b0a9499c | 6692 | case SCHED_BATCH: |
dd41f596 | 6693 | case SCHED_IDLE: |
1da177e4 LT |
6694 | ret = 0; |
6695 | } | |
6696 | return ret; | |
6697 | } | |
6698 | ||
6699 | /** | |
6700 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6701 | * @pid: pid of the process. | |
6702 | * @interval: userspace pointer to the timeslice value. | |
6703 | * | |
6704 | * this syscall writes the default timeslice value of a given process | |
6705 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6706 | */ | |
17da2bd9 | 6707 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6708 | struct timespec __user *, interval) |
1da177e4 | 6709 | { |
36c8b586 | 6710 | struct task_struct *p; |
a4ec24b4 | 6711 | unsigned int time_slice; |
3a5c359a | 6712 | int retval; |
1da177e4 | 6713 | struct timespec t; |
1da177e4 LT |
6714 | |
6715 | if (pid < 0) | |
3a5c359a | 6716 | return -EINVAL; |
1da177e4 LT |
6717 | |
6718 | retval = -ESRCH; | |
6719 | read_lock(&tasklist_lock); | |
6720 | p = find_process_by_pid(pid); | |
6721 | if (!p) | |
6722 | goto out_unlock; | |
6723 | ||
6724 | retval = security_task_getscheduler(p); | |
6725 | if (retval) | |
6726 | goto out_unlock; | |
6727 | ||
77034937 IM |
6728 | /* |
6729 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6730 | * tasks that are on an otherwise idle runqueue: | |
6731 | */ | |
6732 | time_slice = 0; | |
6733 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6734 | time_slice = DEF_TIMESLICE; |
1868f958 | 6735 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6736 | struct sched_entity *se = &p->se; |
6737 | unsigned long flags; | |
6738 | struct rq *rq; | |
6739 | ||
6740 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6741 | if (rq->cfs.load.weight) |
6742 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6743 | task_rq_unlock(rq, &flags); |
6744 | } | |
1da177e4 | 6745 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6746 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6747 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6748 | return retval; |
3a5c359a | 6749 | |
1da177e4 LT |
6750 | out_unlock: |
6751 | read_unlock(&tasklist_lock); | |
6752 | return retval; | |
6753 | } | |
6754 | ||
7c731e0a | 6755 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6756 | |
82a1fcb9 | 6757 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6758 | { |
1da177e4 | 6759 | unsigned long free = 0; |
36c8b586 | 6760 | unsigned state; |
1da177e4 | 6761 | |
1da177e4 | 6762 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6763 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6764 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6765 | #if BITS_PER_LONG == 32 |
1da177e4 | 6766 | if (state == TASK_RUNNING) |
cc4ea795 | 6767 | printk(KERN_CONT " running "); |
1da177e4 | 6768 | else |
cc4ea795 | 6769 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6770 | #else |
6771 | if (state == TASK_RUNNING) | |
cc4ea795 | 6772 | printk(KERN_CONT " running task "); |
1da177e4 | 6773 | else |
cc4ea795 | 6774 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6775 | #endif |
6776 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6777 | free = stack_not_used(p); |
1da177e4 | 6778 | #endif |
aa47b7e0 DR |
6779 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6780 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6781 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6782 | |
5fb5e6de | 6783 | show_stack(p, NULL); |
1da177e4 LT |
6784 | } |
6785 | ||
e59e2ae2 | 6786 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6787 | { |
36c8b586 | 6788 | struct task_struct *g, *p; |
1da177e4 | 6789 | |
4bd77321 IM |
6790 | #if BITS_PER_LONG == 32 |
6791 | printk(KERN_INFO | |
6792 | " task PC stack pid father\n"); | |
1da177e4 | 6793 | #else |
4bd77321 IM |
6794 | printk(KERN_INFO |
6795 | " task PC stack pid father\n"); | |
1da177e4 LT |
6796 | #endif |
6797 | read_lock(&tasklist_lock); | |
6798 | do_each_thread(g, p) { | |
6799 | /* | |
6800 | * reset the NMI-timeout, listing all files on a slow | |
6801 | * console might take alot of time: | |
6802 | */ | |
6803 | touch_nmi_watchdog(); | |
39bc89fd | 6804 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6805 | sched_show_task(p); |
1da177e4 LT |
6806 | } while_each_thread(g, p); |
6807 | ||
04c9167f JF |
6808 | touch_all_softlockup_watchdogs(); |
6809 | ||
dd41f596 IM |
6810 | #ifdef CONFIG_SCHED_DEBUG |
6811 | sysrq_sched_debug_show(); | |
6812 | #endif | |
1da177e4 | 6813 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6814 | /* |
6815 | * Only show locks if all tasks are dumped: | |
6816 | */ | |
6817 | if (state_filter == -1) | |
6818 | debug_show_all_locks(); | |
1da177e4 LT |
6819 | } |
6820 | ||
1df21055 IM |
6821 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6822 | { | |
dd41f596 | 6823 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6824 | } |
6825 | ||
f340c0d1 IM |
6826 | /** |
6827 | * init_idle - set up an idle thread for a given CPU | |
6828 | * @idle: task in question | |
6829 | * @cpu: cpu the idle task belongs to | |
6830 | * | |
6831 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6832 | * flag, to make booting more robust. | |
6833 | */ | |
5c1e1767 | 6834 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6835 | { |
70b97a7f | 6836 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6837 | unsigned long flags; |
6838 | ||
5cbd54ef IM |
6839 | spin_lock_irqsave(&rq->lock, flags); |
6840 | ||
dd41f596 IM |
6841 | __sched_fork(idle); |
6842 | idle->se.exec_start = sched_clock(); | |
6843 | ||
b29739f9 | 6844 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6845 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6846 | __set_task_cpu(idle, cpu); |
1da177e4 | 6847 | |
1da177e4 | 6848 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6849 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6850 | idle->oncpu = 1; | |
6851 | #endif | |
1da177e4 LT |
6852 | spin_unlock_irqrestore(&rq->lock, flags); |
6853 | ||
6854 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6855 | #if defined(CONFIG_PREEMPT) |
6856 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6857 | #else | |
a1261f54 | 6858 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6859 | #endif |
dd41f596 IM |
6860 | /* |
6861 | * The idle tasks have their own, simple scheduling class: | |
6862 | */ | |
6863 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6864 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6865 | } |
6866 | ||
6867 | /* | |
6868 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6869 | * indicates which cpus entered this state. This is used | |
6870 | * in the rcu update to wait only for active cpus. For system | |
6871 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6872 | * always be CPU_BITS_NONE. |
1da177e4 | 6873 | */ |
6a7b3dc3 | 6874 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6875 | |
19978ca6 IM |
6876 | /* |
6877 | * Increase the granularity value when there are more CPUs, | |
6878 | * because with more CPUs the 'effective latency' as visible | |
6879 | * to users decreases. But the relationship is not linear, | |
6880 | * so pick a second-best guess by going with the log2 of the | |
6881 | * number of CPUs. | |
6882 | * | |
6883 | * This idea comes from the SD scheduler of Con Kolivas: | |
6884 | */ | |
6885 | static inline void sched_init_granularity(void) | |
6886 | { | |
6887 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6888 | const unsigned long limit = 200000000; | |
6889 | ||
6890 | sysctl_sched_min_granularity *= factor; | |
6891 | if (sysctl_sched_min_granularity > limit) | |
6892 | sysctl_sched_min_granularity = limit; | |
6893 | ||
6894 | sysctl_sched_latency *= factor; | |
6895 | if (sysctl_sched_latency > limit) | |
6896 | sysctl_sched_latency = limit; | |
6897 | ||
6898 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6899 | |
6900 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6901 | } |
6902 | ||
1da177e4 LT |
6903 | #ifdef CONFIG_SMP |
6904 | /* | |
6905 | * This is how migration works: | |
6906 | * | |
70b97a7f | 6907 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6908 | * runqueue and wake up that CPU's migration thread. |
6909 | * 2) we down() the locked semaphore => thread blocks. | |
6910 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6911 | * thread off the CPU) | |
6912 | * 4) it gets the migration request and checks whether the migrated | |
6913 | * task is still in the wrong runqueue. | |
6914 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6915 | * it and puts it into the right queue. | |
6916 | * 6) migration thread up()s the semaphore. | |
6917 | * 7) we wake up and the migration is done. | |
6918 | */ | |
6919 | ||
6920 | /* | |
6921 | * Change a given task's CPU affinity. Migrate the thread to a | |
6922 | * proper CPU and schedule it away if the CPU it's executing on | |
6923 | * is removed from the allowed bitmask. | |
6924 | * | |
6925 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6926 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6927 | * call is not atomic; no spinlocks may be held. |
6928 | */ | |
96f874e2 | 6929 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6930 | { |
70b97a7f | 6931 | struct migration_req req; |
1da177e4 | 6932 | unsigned long flags; |
70b97a7f | 6933 | struct rq *rq; |
48f24c4d | 6934 | int ret = 0; |
1da177e4 LT |
6935 | |
6936 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6937 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6938 | ret = -EINVAL; |
6939 | goto out; | |
6940 | } | |
6941 | ||
9985b0ba | 6942 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6943 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6944 | ret = -EINVAL; |
6945 | goto out; | |
6946 | } | |
6947 | ||
73fe6aae | 6948 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6949 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6950 | else { |
96f874e2 RR |
6951 | cpumask_copy(&p->cpus_allowed, new_mask); |
6952 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6953 | } |
6954 | ||
1da177e4 | 6955 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6956 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6957 | goto out; |
6958 | ||
1e5ce4f4 | 6959 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6960 | /* Need help from migration thread: drop lock and wait. */ |
6961 | task_rq_unlock(rq, &flags); | |
6962 | wake_up_process(rq->migration_thread); | |
6963 | wait_for_completion(&req.done); | |
6964 | tlb_migrate_finish(p->mm); | |
6965 | return 0; | |
6966 | } | |
6967 | out: | |
6968 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6969 | |
1da177e4 LT |
6970 | return ret; |
6971 | } | |
cd8ba7cd | 6972 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6973 | |
6974 | /* | |
41a2d6cf | 6975 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6976 | * this because either it can't run here any more (set_cpus_allowed() |
6977 | * away from this CPU, or CPU going down), or because we're | |
6978 | * attempting to rebalance this task on exec (sched_exec). | |
6979 | * | |
6980 | * So we race with normal scheduler movements, but that's OK, as long | |
6981 | * as the task is no longer on this CPU. | |
efc30814 KK |
6982 | * |
6983 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6984 | */ |
efc30814 | 6985 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6986 | { |
70b97a7f | 6987 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6988 | int ret = 0, on_rq; |
1da177e4 | 6989 | |
e761b772 | 6990 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6991 | return ret; |
1da177e4 LT |
6992 | |
6993 | rq_src = cpu_rq(src_cpu); | |
6994 | rq_dest = cpu_rq(dest_cpu); | |
6995 | ||
6996 | double_rq_lock(rq_src, rq_dest); | |
6997 | /* Already moved. */ | |
6998 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6999 | goto done; |
1da177e4 | 7000 | /* Affinity changed (again). */ |
96f874e2 | 7001 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7002 | goto fail; |
1da177e4 | 7003 | |
dd41f596 | 7004 | on_rq = p->se.on_rq; |
6e82a3be | 7005 | if (on_rq) |
2e1cb74a | 7006 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7007 | |
1da177e4 | 7008 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7009 | if (on_rq) { |
7010 | activate_task(rq_dest, p, 0); | |
15afe09b | 7011 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7012 | } |
b1e38734 | 7013 | done: |
efc30814 | 7014 | ret = 1; |
b1e38734 | 7015 | fail: |
1da177e4 | 7016 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7017 | return ret; |
1da177e4 LT |
7018 | } |
7019 | ||
7020 | /* | |
7021 | * migration_thread - this is a highprio system thread that performs | |
7022 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7023 | * another runqueue. | |
7024 | */ | |
95cdf3b7 | 7025 | static int migration_thread(void *data) |
1da177e4 | 7026 | { |
1da177e4 | 7027 | int cpu = (long)data; |
70b97a7f | 7028 | struct rq *rq; |
1da177e4 LT |
7029 | |
7030 | rq = cpu_rq(cpu); | |
7031 | BUG_ON(rq->migration_thread != current); | |
7032 | ||
7033 | set_current_state(TASK_INTERRUPTIBLE); | |
7034 | while (!kthread_should_stop()) { | |
70b97a7f | 7035 | struct migration_req *req; |
1da177e4 | 7036 | struct list_head *head; |
1da177e4 | 7037 | |
1da177e4 LT |
7038 | spin_lock_irq(&rq->lock); |
7039 | ||
7040 | if (cpu_is_offline(cpu)) { | |
7041 | spin_unlock_irq(&rq->lock); | |
7042 | goto wait_to_die; | |
7043 | } | |
7044 | ||
7045 | if (rq->active_balance) { | |
7046 | active_load_balance(rq, cpu); | |
7047 | rq->active_balance = 0; | |
7048 | } | |
7049 | ||
7050 | head = &rq->migration_queue; | |
7051 | ||
7052 | if (list_empty(head)) { | |
7053 | spin_unlock_irq(&rq->lock); | |
7054 | schedule(); | |
7055 | set_current_state(TASK_INTERRUPTIBLE); | |
7056 | continue; | |
7057 | } | |
70b97a7f | 7058 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7059 | list_del_init(head->next); |
7060 | ||
674311d5 NP |
7061 | spin_unlock(&rq->lock); |
7062 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7063 | local_irq_enable(); | |
1da177e4 LT |
7064 | |
7065 | complete(&req->done); | |
7066 | } | |
7067 | __set_current_state(TASK_RUNNING); | |
7068 | return 0; | |
7069 | ||
7070 | wait_to_die: | |
7071 | /* Wait for kthread_stop */ | |
7072 | set_current_state(TASK_INTERRUPTIBLE); | |
7073 | while (!kthread_should_stop()) { | |
7074 | schedule(); | |
7075 | set_current_state(TASK_INTERRUPTIBLE); | |
7076 | } | |
7077 | __set_current_state(TASK_RUNNING); | |
7078 | return 0; | |
7079 | } | |
7080 | ||
7081 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7082 | |
7083 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7084 | { | |
7085 | int ret; | |
7086 | ||
7087 | local_irq_disable(); | |
7088 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7089 | local_irq_enable(); | |
7090 | return ret; | |
7091 | } | |
7092 | ||
054b9108 | 7093 | /* |
3a4fa0a2 | 7094 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7095 | */ |
48f24c4d | 7096 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7097 | { |
70b97a7f | 7098 | int dest_cpu; |
6ca09dfc | 7099 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7100 | |
7101 | again: | |
7102 | /* Look for allowed, online CPU in same node. */ | |
7103 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7104 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7105 | goto move; | |
7106 | ||
7107 | /* Any allowed, online CPU? */ | |
7108 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7109 | if (dest_cpu < nr_cpu_ids) | |
7110 | goto move; | |
7111 | ||
7112 | /* No more Mr. Nice Guy. */ | |
7113 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7114 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7115 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7116 | |
e76bd8d9 RR |
7117 | /* |
7118 | * Don't tell them about moving exiting tasks or | |
7119 | * kernel threads (both mm NULL), since they never | |
7120 | * leave kernel. | |
7121 | */ | |
7122 | if (p->mm && printk_ratelimit()) { | |
7123 | printk(KERN_INFO "process %d (%s) no " | |
7124 | "longer affine to cpu%d\n", | |
7125 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7126 | } |
e76bd8d9 RR |
7127 | } |
7128 | ||
7129 | move: | |
7130 | /* It can have affinity changed while we were choosing. */ | |
7131 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7132 | goto again; | |
1da177e4 LT |
7133 | } |
7134 | ||
7135 | /* | |
7136 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7137 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7138 | * for performance reasons the counter is not stricly tracking tasks to | |
7139 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7140 | * to keep the global sum constant after CPU-down: | |
7141 | */ | |
70b97a7f | 7142 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7143 | { |
1e5ce4f4 | 7144 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7145 | unsigned long flags; |
7146 | ||
7147 | local_irq_save(flags); | |
7148 | double_rq_lock(rq_src, rq_dest); | |
7149 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7150 | rq_src->nr_uninterruptible = 0; | |
7151 | double_rq_unlock(rq_src, rq_dest); | |
7152 | local_irq_restore(flags); | |
7153 | } | |
7154 | ||
7155 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7156 | static void migrate_live_tasks(int src_cpu) | |
7157 | { | |
48f24c4d | 7158 | struct task_struct *p, *t; |
1da177e4 | 7159 | |
f7b4cddc | 7160 | read_lock(&tasklist_lock); |
1da177e4 | 7161 | |
48f24c4d IM |
7162 | do_each_thread(t, p) { |
7163 | if (p == current) | |
1da177e4 LT |
7164 | continue; |
7165 | ||
48f24c4d IM |
7166 | if (task_cpu(p) == src_cpu) |
7167 | move_task_off_dead_cpu(src_cpu, p); | |
7168 | } while_each_thread(t, p); | |
1da177e4 | 7169 | |
f7b4cddc | 7170 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7171 | } |
7172 | ||
dd41f596 IM |
7173 | /* |
7174 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7175 | * It does so by boosting its priority to highest possible. |
7176 | * Used by CPU offline code. | |
1da177e4 LT |
7177 | */ |
7178 | void sched_idle_next(void) | |
7179 | { | |
48f24c4d | 7180 | int this_cpu = smp_processor_id(); |
70b97a7f | 7181 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7182 | struct task_struct *p = rq->idle; |
7183 | unsigned long flags; | |
7184 | ||
7185 | /* cpu has to be offline */ | |
48f24c4d | 7186 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7187 | |
48f24c4d IM |
7188 | /* |
7189 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7190 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7191 | */ |
7192 | spin_lock_irqsave(&rq->lock, flags); | |
7193 | ||
dd41f596 | 7194 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7195 | |
94bc9a7b DA |
7196 | update_rq_clock(rq); |
7197 | activate_task(rq, p, 0); | |
1da177e4 LT |
7198 | |
7199 | spin_unlock_irqrestore(&rq->lock, flags); | |
7200 | } | |
7201 | ||
48f24c4d IM |
7202 | /* |
7203 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7204 | * offline. |
7205 | */ | |
7206 | void idle_task_exit(void) | |
7207 | { | |
7208 | struct mm_struct *mm = current->active_mm; | |
7209 | ||
7210 | BUG_ON(cpu_online(smp_processor_id())); | |
7211 | ||
7212 | if (mm != &init_mm) | |
7213 | switch_mm(mm, &init_mm, current); | |
7214 | mmdrop(mm); | |
7215 | } | |
7216 | ||
054b9108 | 7217 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7218 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7219 | { |
70b97a7f | 7220 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7221 | |
7222 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7223 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7224 | |
7225 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7226 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7227 | |
48f24c4d | 7228 | get_task_struct(p); |
1da177e4 LT |
7229 | |
7230 | /* | |
7231 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7232 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7233 | * fine. |
7234 | */ | |
f7b4cddc | 7235 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7236 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7237 | spin_lock_irq(&rq->lock); |
1da177e4 | 7238 | |
48f24c4d | 7239 | put_task_struct(p); |
1da177e4 LT |
7240 | } |
7241 | ||
7242 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7243 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7244 | { | |
70b97a7f | 7245 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7246 | struct task_struct *next; |
48f24c4d | 7247 | |
dd41f596 IM |
7248 | for ( ; ; ) { |
7249 | if (!rq->nr_running) | |
7250 | break; | |
a8e504d2 | 7251 | update_rq_clock(rq); |
b67802ea | 7252 | next = pick_next_task(rq); |
dd41f596 IM |
7253 | if (!next) |
7254 | break; | |
79c53799 | 7255 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7256 | migrate_dead(dead_cpu, next); |
e692ab53 | 7257 | |
1da177e4 LT |
7258 | } |
7259 | } | |
dce48a84 TG |
7260 | |
7261 | /* | |
7262 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7263 | */ | |
7264 | static void calc_global_load_remove(struct rq *rq) | |
7265 | { | |
7266 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7267 | } | |
1da177e4 LT |
7268 | #endif /* CONFIG_HOTPLUG_CPU */ |
7269 | ||
e692ab53 NP |
7270 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7271 | ||
7272 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7273 | { |
7274 | .procname = "sched_domain", | |
c57baf1e | 7275 | .mode = 0555, |
e0361851 | 7276 | }, |
38605cae | 7277 | {0, }, |
e692ab53 NP |
7278 | }; |
7279 | ||
7280 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7281 | { |
c57baf1e | 7282 | .ctl_name = CTL_KERN, |
e0361851 | 7283 | .procname = "kernel", |
c57baf1e | 7284 | .mode = 0555, |
e0361851 AD |
7285 | .child = sd_ctl_dir, |
7286 | }, | |
38605cae | 7287 | {0, }, |
e692ab53 NP |
7288 | }; |
7289 | ||
7290 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7291 | { | |
7292 | struct ctl_table *entry = | |
5cf9f062 | 7293 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7294 | |
e692ab53 NP |
7295 | return entry; |
7296 | } | |
7297 | ||
6382bc90 MM |
7298 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7299 | { | |
cd790076 | 7300 | struct ctl_table *entry; |
6382bc90 | 7301 | |
cd790076 MM |
7302 | /* |
7303 | * In the intermediate directories, both the child directory and | |
7304 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7305 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7306 | * static strings and all have proc handlers. |
7307 | */ | |
7308 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7309 | if (entry->child) |
7310 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7311 | if (entry->proc_handler == NULL) |
7312 | kfree(entry->procname); | |
7313 | } | |
6382bc90 MM |
7314 | |
7315 | kfree(*tablep); | |
7316 | *tablep = NULL; | |
7317 | } | |
7318 | ||
e692ab53 | 7319 | static void |
e0361851 | 7320 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7321 | const char *procname, void *data, int maxlen, |
7322 | mode_t mode, proc_handler *proc_handler) | |
7323 | { | |
e692ab53 NP |
7324 | entry->procname = procname; |
7325 | entry->data = data; | |
7326 | entry->maxlen = maxlen; | |
7327 | entry->mode = mode; | |
7328 | entry->proc_handler = proc_handler; | |
7329 | } | |
7330 | ||
7331 | static struct ctl_table * | |
7332 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7333 | { | |
a5d8c348 | 7334 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7335 | |
ad1cdc1d MM |
7336 | if (table == NULL) |
7337 | return NULL; | |
7338 | ||
e0361851 | 7339 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7340 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7341 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7342 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7343 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7344 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7345 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7346 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7347 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7348 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7349 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7350 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7351 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7352 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7353 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7354 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7355 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7356 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7357 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7358 | &sd->cache_nice_tries, |
7359 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7360 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7361 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7362 | set_table_entry(&table[11], "name", sd->name, |
7363 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7364 | /* &table[12] is terminator */ | |
e692ab53 NP |
7365 | |
7366 | return table; | |
7367 | } | |
7368 | ||
9a4e7159 | 7369 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7370 | { |
7371 | struct ctl_table *entry, *table; | |
7372 | struct sched_domain *sd; | |
7373 | int domain_num = 0, i; | |
7374 | char buf[32]; | |
7375 | ||
7376 | for_each_domain(cpu, sd) | |
7377 | domain_num++; | |
7378 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7379 | if (table == NULL) |
7380 | return NULL; | |
e692ab53 NP |
7381 | |
7382 | i = 0; | |
7383 | for_each_domain(cpu, sd) { | |
7384 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7385 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7386 | entry->mode = 0555; |
e692ab53 NP |
7387 | entry->child = sd_alloc_ctl_domain_table(sd); |
7388 | entry++; | |
7389 | i++; | |
7390 | } | |
7391 | return table; | |
7392 | } | |
7393 | ||
7394 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7395 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7396 | { |
7397 | int i, cpu_num = num_online_cpus(); | |
7398 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7399 | char buf[32]; | |
7400 | ||
7378547f MM |
7401 | WARN_ON(sd_ctl_dir[0].child); |
7402 | sd_ctl_dir[0].child = entry; | |
7403 | ||
ad1cdc1d MM |
7404 | if (entry == NULL) |
7405 | return; | |
7406 | ||
97b6ea7b | 7407 | for_each_online_cpu(i) { |
e692ab53 | 7408 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7409 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7410 | entry->mode = 0555; |
e692ab53 | 7411 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7412 | entry++; |
e692ab53 | 7413 | } |
7378547f MM |
7414 | |
7415 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7416 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7417 | } | |
6382bc90 | 7418 | |
7378547f | 7419 | /* may be called multiple times per register */ |
6382bc90 MM |
7420 | static void unregister_sched_domain_sysctl(void) |
7421 | { | |
7378547f MM |
7422 | if (sd_sysctl_header) |
7423 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7424 | sd_sysctl_header = NULL; |
7378547f MM |
7425 | if (sd_ctl_dir[0].child) |
7426 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7427 | } |
e692ab53 | 7428 | #else |
6382bc90 MM |
7429 | static void register_sched_domain_sysctl(void) |
7430 | { | |
7431 | } | |
7432 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7433 | { |
7434 | } | |
7435 | #endif | |
7436 | ||
1f11eb6a GH |
7437 | static void set_rq_online(struct rq *rq) |
7438 | { | |
7439 | if (!rq->online) { | |
7440 | const struct sched_class *class; | |
7441 | ||
c6c4927b | 7442 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7443 | rq->online = 1; |
7444 | ||
7445 | for_each_class(class) { | |
7446 | if (class->rq_online) | |
7447 | class->rq_online(rq); | |
7448 | } | |
7449 | } | |
7450 | } | |
7451 | ||
7452 | static void set_rq_offline(struct rq *rq) | |
7453 | { | |
7454 | if (rq->online) { | |
7455 | const struct sched_class *class; | |
7456 | ||
7457 | for_each_class(class) { | |
7458 | if (class->rq_offline) | |
7459 | class->rq_offline(rq); | |
7460 | } | |
7461 | ||
c6c4927b | 7462 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7463 | rq->online = 0; |
7464 | } | |
7465 | } | |
7466 | ||
1da177e4 LT |
7467 | /* |
7468 | * migration_call - callback that gets triggered when a CPU is added. | |
7469 | * Here we can start up the necessary migration thread for the new CPU. | |
7470 | */ | |
48f24c4d IM |
7471 | static int __cpuinit |
7472 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7473 | { |
1da177e4 | 7474 | struct task_struct *p; |
48f24c4d | 7475 | int cpu = (long)hcpu; |
1da177e4 | 7476 | unsigned long flags; |
70b97a7f | 7477 | struct rq *rq; |
1da177e4 LT |
7478 | |
7479 | switch (action) { | |
5be9361c | 7480 | |
1da177e4 | 7481 | case CPU_UP_PREPARE: |
8bb78442 | 7482 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7483 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7484 | if (IS_ERR(p)) |
7485 | return NOTIFY_BAD; | |
1da177e4 LT |
7486 | kthread_bind(p, cpu); |
7487 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7488 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7489 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7490 | task_rq_unlock(rq, &flags); |
7491 | cpu_rq(cpu)->migration_thread = p; | |
7492 | break; | |
48f24c4d | 7493 | |
1da177e4 | 7494 | case CPU_ONLINE: |
8bb78442 | 7495 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7496 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7497 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7498 | |
7499 | /* Update our root-domain */ | |
7500 | rq = cpu_rq(cpu); | |
7501 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7502 | rq->calc_load_update = calc_load_update; |
7503 | rq->calc_load_active = 0; | |
1f94ef59 | 7504 | if (rq->rd) { |
c6c4927b | 7505 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7506 | |
7507 | set_rq_online(rq); | |
1f94ef59 GH |
7508 | } |
7509 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7510 | break; |
48f24c4d | 7511 | |
1da177e4 LT |
7512 | #ifdef CONFIG_HOTPLUG_CPU |
7513 | case CPU_UP_CANCELED: | |
8bb78442 | 7514 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7515 | if (!cpu_rq(cpu)->migration_thread) |
7516 | break; | |
41a2d6cf | 7517 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7518 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7519 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7520 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7521 | cpu_rq(cpu)->migration_thread = NULL; | |
7522 | break; | |
48f24c4d | 7523 | |
1da177e4 | 7524 | case CPU_DEAD: |
8bb78442 | 7525 | case CPU_DEAD_FROZEN: |
470fd646 | 7526 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7527 | migrate_live_tasks(cpu); |
7528 | rq = cpu_rq(cpu); | |
7529 | kthread_stop(rq->migration_thread); | |
7530 | rq->migration_thread = NULL; | |
7531 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7532 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7533 | update_rq_clock(rq); |
2e1cb74a | 7534 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7535 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7536 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7537 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7538 | migrate_dead_tasks(cpu); |
d2da272a | 7539 | spin_unlock_irq(&rq->lock); |
470fd646 | 7540 | cpuset_unlock(); |
1da177e4 LT |
7541 | migrate_nr_uninterruptible(rq); |
7542 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7543 | calc_global_load_remove(rq); |
41a2d6cf IM |
7544 | /* |
7545 | * No need to migrate the tasks: it was best-effort if | |
7546 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7547 | * the requestors. | |
7548 | */ | |
1da177e4 LT |
7549 | spin_lock_irq(&rq->lock); |
7550 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7551 | struct migration_req *req; |
7552 | ||
1da177e4 | 7553 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7554 | struct migration_req, list); |
1da177e4 | 7555 | list_del_init(&req->list); |
9a2bd244 | 7556 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7557 | complete(&req->done); |
9a2bd244 | 7558 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7559 | } |
7560 | spin_unlock_irq(&rq->lock); | |
7561 | break; | |
57d885fe | 7562 | |
08f503b0 GH |
7563 | case CPU_DYING: |
7564 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7565 | /* Update our root-domain */ |
7566 | rq = cpu_rq(cpu); | |
7567 | spin_lock_irqsave(&rq->lock, flags); | |
7568 | if (rq->rd) { | |
c6c4927b | 7569 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7570 | set_rq_offline(rq); |
57d885fe GH |
7571 | } |
7572 | spin_unlock_irqrestore(&rq->lock, flags); | |
7573 | break; | |
1da177e4 LT |
7574 | #endif |
7575 | } | |
7576 | return NOTIFY_OK; | |
7577 | } | |
7578 | ||
f38b0820 PM |
7579 | /* |
7580 | * Register at high priority so that task migration (migrate_all_tasks) | |
7581 | * happens before everything else. This has to be lower priority than | |
7582 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7583 | */ |
26c2143b | 7584 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7585 | .notifier_call = migration_call, |
7586 | .priority = 10 | |
7587 | }; | |
7588 | ||
7babe8db | 7589 | static int __init migration_init(void) |
1da177e4 LT |
7590 | { |
7591 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7592 | int err; |
48f24c4d IM |
7593 | |
7594 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7595 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7596 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7597 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7598 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7599 | |
7600 | return err; | |
1da177e4 | 7601 | } |
7babe8db | 7602 | early_initcall(migration_init); |
1da177e4 LT |
7603 | #endif |
7604 | ||
7605 | #ifdef CONFIG_SMP | |
476f3534 | 7606 | |
3e9830dc | 7607 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7608 | |
7c16ec58 | 7609 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7610 | struct cpumask *groupmask) |
1da177e4 | 7611 | { |
4dcf6aff | 7612 | struct sched_group *group = sd->groups; |
434d53b0 | 7613 | char str[256]; |
1da177e4 | 7614 | |
968ea6d8 | 7615 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7616 | cpumask_clear(groupmask); |
4dcf6aff IM |
7617 | |
7618 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7619 | ||
7620 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7621 | printk("does not load-balance\n"); | |
7622 | if (sd->parent) | |
7623 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7624 | " has parent"); | |
7625 | return -1; | |
41c7ce9a NP |
7626 | } |
7627 | ||
eefd796a | 7628 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7629 | |
758b2cdc | 7630 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7631 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7632 | "CPU%d\n", cpu); | |
7633 | } | |
758b2cdc | 7634 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7635 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7636 | " CPU%d\n", cpu); | |
7637 | } | |
1da177e4 | 7638 | |
4dcf6aff | 7639 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7640 | do { |
4dcf6aff IM |
7641 | if (!group) { |
7642 | printk("\n"); | |
7643 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7644 | break; |
7645 | } | |
7646 | ||
4dcf6aff IM |
7647 | if (!group->__cpu_power) { |
7648 | printk(KERN_CONT "\n"); | |
7649 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7650 | "set\n"); | |
7651 | break; | |
7652 | } | |
1da177e4 | 7653 | |
758b2cdc | 7654 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7655 | printk(KERN_CONT "\n"); |
7656 | printk(KERN_ERR "ERROR: empty group\n"); | |
7657 | break; | |
7658 | } | |
1da177e4 | 7659 | |
758b2cdc | 7660 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7661 | printk(KERN_CONT "\n"); |
7662 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7663 | break; | |
7664 | } | |
1da177e4 | 7665 | |
758b2cdc | 7666 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7667 | |
968ea6d8 | 7668 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7669 | |
7670 | printk(KERN_CONT " %s", str); | |
7671 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7672 | printk(KERN_CONT " (__cpu_power = %d)", | |
7673 | group->__cpu_power); | |
7674 | } | |
1da177e4 | 7675 | |
4dcf6aff IM |
7676 | group = group->next; |
7677 | } while (group != sd->groups); | |
7678 | printk(KERN_CONT "\n"); | |
1da177e4 | 7679 | |
758b2cdc | 7680 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7681 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7682 | |
758b2cdc RR |
7683 | if (sd->parent && |
7684 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7685 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7686 | "of domain->span\n"); | |
7687 | return 0; | |
7688 | } | |
1da177e4 | 7689 | |
4dcf6aff IM |
7690 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7691 | { | |
d5dd3db1 | 7692 | cpumask_var_t groupmask; |
4dcf6aff | 7693 | int level = 0; |
1da177e4 | 7694 | |
4dcf6aff IM |
7695 | if (!sd) { |
7696 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7697 | return; | |
7698 | } | |
1da177e4 | 7699 | |
4dcf6aff IM |
7700 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7701 | ||
d5dd3db1 | 7702 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7703 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7704 | return; | |
7705 | } | |
7706 | ||
4dcf6aff | 7707 | for (;;) { |
7c16ec58 | 7708 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7709 | break; |
1da177e4 LT |
7710 | level++; |
7711 | sd = sd->parent; | |
33859f7f | 7712 | if (!sd) |
4dcf6aff IM |
7713 | break; |
7714 | } | |
d5dd3db1 | 7715 | free_cpumask_var(groupmask); |
1da177e4 | 7716 | } |
6d6bc0ad | 7717 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7718 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7719 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7720 | |
1a20ff27 | 7721 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7722 | { |
758b2cdc | 7723 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7724 | return 1; |
7725 | ||
7726 | /* Following flags need at least 2 groups */ | |
7727 | if (sd->flags & (SD_LOAD_BALANCE | | |
7728 | SD_BALANCE_NEWIDLE | | |
7729 | SD_BALANCE_FORK | | |
89c4710e SS |
7730 | SD_BALANCE_EXEC | |
7731 | SD_SHARE_CPUPOWER | | |
7732 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7733 | if (sd->groups != sd->groups->next) |
7734 | return 0; | |
7735 | } | |
7736 | ||
7737 | /* Following flags don't use groups */ | |
7738 | if (sd->flags & (SD_WAKE_IDLE | | |
7739 | SD_WAKE_AFFINE | | |
7740 | SD_WAKE_BALANCE)) | |
7741 | return 0; | |
7742 | ||
7743 | return 1; | |
7744 | } | |
7745 | ||
48f24c4d IM |
7746 | static int |
7747 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7748 | { |
7749 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7750 | ||
7751 | if (sd_degenerate(parent)) | |
7752 | return 1; | |
7753 | ||
758b2cdc | 7754 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7755 | return 0; |
7756 | ||
7757 | /* Does parent contain flags not in child? */ | |
7758 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7759 | if (cflags & SD_WAKE_AFFINE) | |
7760 | pflags &= ~SD_WAKE_BALANCE; | |
7761 | /* Flags needing groups don't count if only 1 group in parent */ | |
7762 | if (parent->groups == parent->groups->next) { | |
7763 | pflags &= ~(SD_LOAD_BALANCE | | |
7764 | SD_BALANCE_NEWIDLE | | |
7765 | SD_BALANCE_FORK | | |
89c4710e SS |
7766 | SD_BALANCE_EXEC | |
7767 | SD_SHARE_CPUPOWER | | |
7768 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7769 | if (nr_node_ids == 1) |
7770 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7771 | } |
7772 | if (~cflags & pflags) | |
7773 | return 0; | |
7774 | ||
7775 | return 1; | |
7776 | } | |
7777 | ||
c6c4927b RR |
7778 | static void free_rootdomain(struct root_domain *rd) |
7779 | { | |
68e74568 RR |
7780 | cpupri_cleanup(&rd->cpupri); |
7781 | ||
c6c4927b RR |
7782 | free_cpumask_var(rd->rto_mask); |
7783 | free_cpumask_var(rd->online); | |
7784 | free_cpumask_var(rd->span); | |
7785 | kfree(rd); | |
7786 | } | |
7787 | ||
57d885fe GH |
7788 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7789 | { | |
a0490fa3 | 7790 | struct root_domain *old_rd = NULL; |
57d885fe | 7791 | unsigned long flags; |
57d885fe GH |
7792 | |
7793 | spin_lock_irqsave(&rq->lock, flags); | |
7794 | ||
7795 | if (rq->rd) { | |
a0490fa3 | 7796 | old_rd = rq->rd; |
57d885fe | 7797 | |
c6c4927b | 7798 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7799 | set_rq_offline(rq); |
57d885fe | 7800 | |
c6c4927b | 7801 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7802 | |
a0490fa3 IM |
7803 | /* |
7804 | * If we dont want to free the old_rt yet then | |
7805 | * set old_rd to NULL to skip the freeing later | |
7806 | * in this function: | |
7807 | */ | |
7808 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7809 | old_rd = NULL; | |
57d885fe GH |
7810 | } |
7811 | ||
7812 | atomic_inc(&rd->refcount); | |
7813 | rq->rd = rd; | |
7814 | ||
c6c4927b RR |
7815 | cpumask_set_cpu(rq->cpu, rd->span); |
7816 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7817 | set_rq_online(rq); |
57d885fe GH |
7818 | |
7819 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7820 | |
7821 | if (old_rd) | |
7822 | free_rootdomain(old_rd); | |
57d885fe GH |
7823 | } |
7824 | ||
db2f59c8 | 7825 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7826 | { |
36b7b6d4 PE |
7827 | gfp_t gfp = GFP_KERNEL; |
7828 | ||
57d885fe GH |
7829 | memset(rd, 0, sizeof(*rd)); |
7830 | ||
36b7b6d4 PE |
7831 | if (bootmem) |
7832 | gfp = GFP_NOWAIT; | |
c6c4927b | 7833 | |
36b7b6d4 | 7834 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7835 | goto out; |
36b7b6d4 | 7836 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7837 | goto free_span; |
36b7b6d4 | 7838 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7839 | goto free_online; |
6e0534f2 | 7840 | |
0fb53029 | 7841 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7842 | goto free_rto_mask; |
c6c4927b | 7843 | return 0; |
6e0534f2 | 7844 | |
68e74568 RR |
7845 | free_rto_mask: |
7846 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7847 | free_online: |
7848 | free_cpumask_var(rd->online); | |
7849 | free_span: | |
7850 | free_cpumask_var(rd->span); | |
0c910d28 | 7851 | out: |
c6c4927b | 7852 | return -ENOMEM; |
57d885fe GH |
7853 | } |
7854 | ||
7855 | static void init_defrootdomain(void) | |
7856 | { | |
c6c4927b RR |
7857 | init_rootdomain(&def_root_domain, true); |
7858 | ||
57d885fe GH |
7859 | atomic_set(&def_root_domain.refcount, 1); |
7860 | } | |
7861 | ||
dc938520 | 7862 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7863 | { |
7864 | struct root_domain *rd; | |
7865 | ||
7866 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7867 | if (!rd) | |
7868 | return NULL; | |
7869 | ||
c6c4927b RR |
7870 | if (init_rootdomain(rd, false) != 0) { |
7871 | kfree(rd); | |
7872 | return NULL; | |
7873 | } | |
57d885fe GH |
7874 | |
7875 | return rd; | |
7876 | } | |
7877 | ||
1da177e4 | 7878 | /* |
0eab9146 | 7879 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7880 | * hold the hotplug lock. |
7881 | */ | |
0eab9146 IM |
7882 | static void |
7883 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7884 | { |
70b97a7f | 7885 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7886 | struct sched_domain *tmp; |
7887 | ||
7888 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7889 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7890 | struct sched_domain *parent = tmp->parent; |
7891 | if (!parent) | |
7892 | break; | |
f29c9b1c | 7893 | |
1a848870 | 7894 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7895 | tmp->parent = parent->parent; |
1a848870 SS |
7896 | if (parent->parent) |
7897 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7898 | } else |
7899 | tmp = tmp->parent; | |
245af2c7 SS |
7900 | } |
7901 | ||
1a848870 | 7902 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7903 | sd = sd->parent; |
1a848870 SS |
7904 | if (sd) |
7905 | sd->child = NULL; | |
7906 | } | |
1da177e4 LT |
7907 | |
7908 | sched_domain_debug(sd, cpu); | |
7909 | ||
57d885fe | 7910 | rq_attach_root(rq, rd); |
674311d5 | 7911 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7912 | } |
7913 | ||
7914 | /* cpus with isolated domains */ | |
dcc30a35 | 7915 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7916 | |
7917 | /* Setup the mask of cpus configured for isolated domains */ | |
7918 | static int __init isolated_cpu_setup(char *str) | |
7919 | { | |
968ea6d8 | 7920 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7921 | return 1; |
7922 | } | |
7923 | ||
8927f494 | 7924 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7925 | |
7926 | /* | |
6711cab4 SS |
7927 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7928 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7929 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7930 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7931 | * |
7932 | * init_sched_build_groups will build a circular linked list of the groups | |
7933 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7934 | * and ->cpu_power to 0. | |
7935 | */ | |
a616058b | 7936 | static void |
96f874e2 RR |
7937 | init_sched_build_groups(const struct cpumask *span, |
7938 | const struct cpumask *cpu_map, | |
7939 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7940 | struct sched_group **sg, |
96f874e2 RR |
7941 | struct cpumask *tmpmask), |
7942 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7943 | { |
7944 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7945 | int i; |
7946 | ||
96f874e2 | 7947 | cpumask_clear(covered); |
7c16ec58 | 7948 | |
abcd083a | 7949 | for_each_cpu(i, span) { |
6711cab4 | 7950 | struct sched_group *sg; |
7c16ec58 | 7951 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7952 | int j; |
7953 | ||
758b2cdc | 7954 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7955 | continue; |
7956 | ||
758b2cdc | 7957 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7958 | sg->__cpu_power = 0; |
1da177e4 | 7959 | |
abcd083a | 7960 | for_each_cpu(j, span) { |
7c16ec58 | 7961 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7962 | continue; |
7963 | ||
96f874e2 | 7964 | cpumask_set_cpu(j, covered); |
758b2cdc | 7965 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7966 | } |
7967 | if (!first) | |
7968 | first = sg; | |
7969 | if (last) | |
7970 | last->next = sg; | |
7971 | last = sg; | |
7972 | } | |
7973 | last->next = first; | |
7974 | } | |
7975 | ||
9c1cfda2 | 7976 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7977 | |
9c1cfda2 | 7978 | #ifdef CONFIG_NUMA |
198e2f18 | 7979 | |
9c1cfda2 JH |
7980 | /** |
7981 | * find_next_best_node - find the next node to include in a sched_domain | |
7982 | * @node: node whose sched_domain we're building | |
7983 | * @used_nodes: nodes already in the sched_domain | |
7984 | * | |
41a2d6cf | 7985 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7986 | * finds the closest node not already in the @used_nodes map. |
7987 | * | |
7988 | * Should use nodemask_t. | |
7989 | */ | |
c5f59f08 | 7990 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7991 | { |
7992 | int i, n, val, min_val, best_node = 0; | |
7993 | ||
7994 | min_val = INT_MAX; | |
7995 | ||
076ac2af | 7996 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7997 | /* Start at @node */ |
076ac2af | 7998 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7999 | |
8000 | if (!nr_cpus_node(n)) | |
8001 | continue; | |
8002 | ||
8003 | /* Skip already used nodes */ | |
c5f59f08 | 8004 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8005 | continue; |
8006 | ||
8007 | /* Simple min distance search */ | |
8008 | val = node_distance(node, n); | |
8009 | ||
8010 | if (val < min_val) { | |
8011 | min_val = val; | |
8012 | best_node = n; | |
8013 | } | |
8014 | } | |
8015 | ||
c5f59f08 | 8016 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8017 | return best_node; |
8018 | } | |
8019 | ||
8020 | /** | |
8021 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8022 | * @node: node whose cpumask we're constructing | |
73486722 | 8023 | * @span: resulting cpumask |
9c1cfda2 | 8024 | * |
41a2d6cf | 8025 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8026 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8027 | * out optimally. | |
8028 | */ | |
96f874e2 | 8029 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8030 | { |
c5f59f08 | 8031 | nodemask_t used_nodes; |
48f24c4d | 8032 | int i; |
9c1cfda2 | 8033 | |
6ca09dfc | 8034 | cpumask_clear(span); |
c5f59f08 | 8035 | nodes_clear(used_nodes); |
9c1cfda2 | 8036 | |
6ca09dfc | 8037 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8038 | node_set(node, used_nodes); |
9c1cfda2 JH |
8039 | |
8040 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8041 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8042 | |
6ca09dfc | 8043 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8044 | } |
9c1cfda2 | 8045 | } |
6d6bc0ad | 8046 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8047 | |
5c45bf27 | 8048 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8049 | |
6c99e9ad RR |
8050 | /* |
8051 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8052 | * |
8053 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8054 | * and struct sched_domain. ) | |
6c99e9ad RR |
8055 | */ |
8056 | struct static_sched_group { | |
8057 | struct sched_group sg; | |
8058 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8059 | }; | |
8060 | ||
8061 | struct static_sched_domain { | |
8062 | struct sched_domain sd; | |
8063 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8064 | }; | |
8065 | ||
9c1cfda2 | 8066 | /* |
48f24c4d | 8067 | * SMT sched-domains: |
9c1cfda2 | 8068 | */ |
1da177e4 | 8069 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8070 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8071 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8072 | |
41a2d6cf | 8073 | static int |
96f874e2 RR |
8074 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8075 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8076 | { |
6711cab4 | 8077 | if (sg) |
6c99e9ad | 8078 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8079 | return cpu; |
8080 | } | |
6d6bc0ad | 8081 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8082 | |
48f24c4d IM |
8083 | /* |
8084 | * multi-core sched-domains: | |
8085 | */ | |
1e9f28fa | 8086 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8087 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8088 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8089 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8090 | |
8091 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8092 | static int |
96f874e2 RR |
8093 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8094 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8095 | { |
6711cab4 | 8096 | int group; |
7c16ec58 | 8097 | |
c69fc56d | 8098 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8099 | group = cpumask_first(mask); |
6711cab4 | 8100 | if (sg) |
6c99e9ad | 8101 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8102 | return group; |
1e9f28fa SS |
8103 | } |
8104 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8105 | static int |
96f874e2 RR |
8106 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8107 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8108 | { |
6711cab4 | 8109 | if (sg) |
6c99e9ad | 8110 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8111 | return cpu; |
8112 | } | |
8113 | #endif | |
8114 | ||
6c99e9ad RR |
8115 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8116 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8117 | |
41a2d6cf | 8118 | static int |
96f874e2 RR |
8119 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8120 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8121 | { |
6711cab4 | 8122 | int group; |
48f24c4d | 8123 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8124 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8125 | group = cpumask_first(mask); |
1e9f28fa | 8126 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8127 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8128 | group = cpumask_first(mask); |
1da177e4 | 8129 | #else |
6711cab4 | 8130 | group = cpu; |
1da177e4 | 8131 | #endif |
6711cab4 | 8132 | if (sg) |
6c99e9ad | 8133 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8134 | return group; |
1da177e4 LT |
8135 | } |
8136 | ||
8137 | #ifdef CONFIG_NUMA | |
1da177e4 | 8138 | /* |
9c1cfda2 JH |
8139 | * The init_sched_build_groups can't handle what we want to do with node |
8140 | * groups, so roll our own. Now each node has its own list of groups which | |
8141 | * gets dynamically allocated. | |
1da177e4 | 8142 | */ |
62ea9ceb | 8143 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8144 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8145 | |
62ea9ceb | 8146 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8147 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8148 | |
96f874e2 RR |
8149 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8150 | struct sched_group **sg, | |
8151 | struct cpumask *nodemask) | |
9c1cfda2 | 8152 | { |
6711cab4 SS |
8153 | int group; |
8154 | ||
6ca09dfc | 8155 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8156 | group = cpumask_first(nodemask); |
6711cab4 SS |
8157 | |
8158 | if (sg) | |
6c99e9ad | 8159 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8160 | return group; |
1da177e4 | 8161 | } |
6711cab4 | 8162 | |
08069033 SS |
8163 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8164 | { | |
8165 | struct sched_group *sg = group_head; | |
8166 | int j; | |
8167 | ||
8168 | if (!sg) | |
8169 | return; | |
3a5c359a | 8170 | do { |
758b2cdc | 8171 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8172 | struct sched_domain *sd; |
08069033 | 8173 | |
6c99e9ad | 8174 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8175 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8176 | /* |
8177 | * Only add "power" once for each | |
8178 | * physical package. | |
8179 | */ | |
8180 | continue; | |
8181 | } | |
08069033 | 8182 | |
3a5c359a AK |
8183 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8184 | } | |
8185 | sg = sg->next; | |
8186 | } while (sg != group_head); | |
08069033 | 8187 | } |
6d6bc0ad | 8188 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8189 | |
a616058b | 8190 | #ifdef CONFIG_NUMA |
51888ca2 | 8191 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8192 | static void free_sched_groups(const struct cpumask *cpu_map, |
8193 | struct cpumask *nodemask) | |
51888ca2 | 8194 | { |
a616058b | 8195 | int cpu, i; |
51888ca2 | 8196 | |
abcd083a | 8197 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8198 | struct sched_group **sched_group_nodes |
8199 | = sched_group_nodes_bycpu[cpu]; | |
8200 | ||
51888ca2 SV |
8201 | if (!sched_group_nodes) |
8202 | continue; | |
8203 | ||
076ac2af | 8204 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8205 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8206 | ||
6ca09dfc | 8207 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8208 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8209 | continue; |
8210 | ||
8211 | if (sg == NULL) | |
8212 | continue; | |
8213 | sg = sg->next; | |
8214 | next_sg: | |
8215 | oldsg = sg; | |
8216 | sg = sg->next; | |
8217 | kfree(oldsg); | |
8218 | if (oldsg != sched_group_nodes[i]) | |
8219 | goto next_sg; | |
8220 | } | |
8221 | kfree(sched_group_nodes); | |
8222 | sched_group_nodes_bycpu[cpu] = NULL; | |
8223 | } | |
51888ca2 | 8224 | } |
6d6bc0ad | 8225 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8226 | static void free_sched_groups(const struct cpumask *cpu_map, |
8227 | struct cpumask *nodemask) | |
a616058b SS |
8228 | { |
8229 | } | |
6d6bc0ad | 8230 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8231 | |
89c4710e SS |
8232 | /* |
8233 | * Initialize sched groups cpu_power. | |
8234 | * | |
8235 | * cpu_power indicates the capacity of sched group, which is used while | |
8236 | * distributing the load between different sched groups in a sched domain. | |
8237 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8238 | * there are asymmetries in the topology. If there are asymmetries, group | |
8239 | * having more cpu_power will pickup more load compared to the group having | |
8240 | * less cpu_power. | |
8241 | * | |
8242 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8243 | * the maximum number of tasks a group can handle in the presence of other idle | |
8244 | * or lightly loaded groups in the same sched domain. | |
8245 | */ | |
8246 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8247 | { | |
8248 | struct sched_domain *child; | |
8249 | struct sched_group *group; | |
8250 | ||
8251 | WARN_ON(!sd || !sd->groups); | |
8252 | ||
13318a71 | 8253 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8254 | return; |
8255 | ||
8256 | child = sd->child; | |
8257 | ||
5517d86b ED |
8258 | sd->groups->__cpu_power = 0; |
8259 | ||
89c4710e SS |
8260 | /* |
8261 | * For perf policy, if the groups in child domain share resources | |
8262 | * (for example cores sharing some portions of the cache hierarchy | |
8263 | * or SMT), then set this domain groups cpu_power such that each group | |
8264 | * can handle only one task, when there are other idle groups in the | |
8265 | * same sched domain. | |
8266 | */ | |
8267 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8268 | (child->flags & | |
8269 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8270 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8271 | return; |
8272 | } | |
8273 | ||
89c4710e SS |
8274 | /* |
8275 | * add cpu_power of each child group to this groups cpu_power | |
8276 | */ | |
8277 | group = child->groups; | |
8278 | do { | |
5517d86b | 8279 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8280 | group = group->next; |
8281 | } while (group != child->groups); | |
8282 | } | |
8283 | ||
7c16ec58 MT |
8284 | /* |
8285 | * Initializers for schedule domains | |
8286 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8287 | */ | |
8288 | ||
a5d8c348 IM |
8289 | #ifdef CONFIG_SCHED_DEBUG |
8290 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8291 | #else | |
8292 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8293 | #endif | |
8294 | ||
7c16ec58 | 8295 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8296 | |
7c16ec58 MT |
8297 | #define SD_INIT_FUNC(type) \ |
8298 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8299 | { \ | |
8300 | memset(sd, 0, sizeof(*sd)); \ | |
8301 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8302 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8303 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8304 | } |
8305 | ||
8306 | SD_INIT_FUNC(CPU) | |
8307 | #ifdef CONFIG_NUMA | |
8308 | SD_INIT_FUNC(ALLNODES) | |
8309 | SD_INIT_FUNC(NODE) | |
8310 | #endif | |
8311 | #ifdef CONFIG_SCHED_SMT | |
8312 | SD_INIT_FUNC(SIBLING) | |
8313 | #endif | |
8314 | #ifdef CONFIG_SCHED_MC | |
8315 | SD_INIT_FUNC(MC) | |
8316 | #endif | |
8317 | ||
1d3504fc HS |
8318 | static int default_relax_domain_level = -1; |
8319 | ||
8320 | static int __init setup_relax_domain_level(char *str) | |
8321 | { | |
30e0e178 LZ |
8322 | unsigned long val; |
8323 | ||
8324 | val = simple_strtoul(str, NULL, 0); | |
8325 | if (val < SD_LV_MAX) | |
8326 | default_relax_domain_level = val; | |
8327 | ||
1d3504fc HS |
8328 | return 1; |
8329 | } | |
8330 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8331 | ||
8332 | static void set_domain_attribute(struct sched_domain *sd, | |
8333 | struct sched_domain_attr *attr) | |
8334 | { | |
8335 | int request; | |
8336 | ||
8337 | if (!attr || attr->relax_domain_level < 0) { | |
8338 | if (default_relax_domain_level < 0) | |
8339 | return; | |
8340 | else | |
8341 | request = default_relax_domain_level; | |
8342 | } else | |
8343 | request = attr->relax_domain_level; | |
8344 | if (request < sd->level) { | |
8345 | /* turn off idle balance on this domain */ | |
8346 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8347 | } else { | |
8348 | /* turn on idle balance on this domain */ | |
8349 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8350 | } | |
8351 | } | |
8352 | ||
1da177e4 | 8353 | /* |
1a20ff27 DG |
8354 | * Build sched domains for a given set of cpus and attach the sched domains |
8355 | * to the individual cpus | |
1da177e4 | 8356 | */ |
96f874e2 | 8357 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8358 | struct sched_domain_attr *attr) |
1da177e4 | 8359 | { |
3404c8d9 | 8360 | int i, err = -ENOMEM; |
57d885fe | 8361 | struct root_domain *rd; |
3404c8d9 RR |
8362 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8363 | tmpmask; | |
d1b55138 | 8364 | #ifdef CONFIG_NUMA |
3404c8d9 | 8365 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8366 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8367 | int sd_allnodes = 0; |
d1b55138 | 8368 | |
3404c8d9 RR |
8369 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8370 | goto out; | |
8371 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8372 | goto free_domainspan; | |
8373 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8374 | goto free_covered; | |
8375 | #endif | |
8376 | ||
8377 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8378 | goto free_notcovered; | |
8379 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8380 | goto free_nodemask; | |
8381 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8382 | goto free_this_sibling_map; | |
8383 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8384 | goto free_this_core_map; | |
8385 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8386 | goto free_send_covered; | |
8387 | ||
8388 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8389 | /* |
8390 | * Allocate the per-node list of sched groups | |
8391 | */ | |
076ac2af | 8392 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8393 | GFP_KERNEL); |
d1b55138 JH |
8394 | if (!sched_group_nodes) { |
8395 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8396 | goto free_tmpmask; |
d1b55138 | 8397 | } |
d1b55138 | 8398 | #endif |
1da177e4 | 8399 | |
dc938520 | 8400 | rd = alloc_rootdomain(); |
57d885fe GH |
8401 | if (!rd) { |
8402 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8403 | goto free_sched_groups; |
57d885fe GH |
8404 | } |
8405 | ||
7c16ec58 | 8406 | #ifdef CONFIG_NUMA |
96f874e2 | 8407 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8408 | #endif |
8409 | ||
1da177e4 | 8410 | /* |
1a20ff27 | 8411 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8412 | */ |
abcd083a | 8413 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8414 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8415 | |
6ca09dfc | 8416 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8417 | |
8418 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8419 | if (cpumask_weight(cpu_map) > |
8420 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8421 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8422 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8423 | set_domain_attribute(sd, attr); |
758b2cdc | 8424 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8425 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8426 | p = sd; |
6711cab4 | 8427 | sd_allnodes = 1; |
9c1cfda2 JH |
8428 | } else |
8429 | p = NULL; | |
8430 | ||
62ea9ceb | 8431 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8432 | SD_INIT(sd, NODE); |
1d3504fc | 8433 | set_domain_attribute(sd, attr); |
758b2cdc | 8434 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8435 | sd->parent = p; |
1a848870 SS |
8436 | if (p) |
8437 | p->child = sd; | |
758b2cdc RR |
8438 | cpumask_and(sched_domain_span(sd), |
8439 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8440 | #endif |
8441 | ||
8442 | p = sd; | |
6c99e9ad | 8443 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8444 | SD_INIT(sd, CPU); |
1d3504fc | 8445 | set_domain_attribute(sd, attr); |
758b2cdc | 8446 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8447 | sd->parent = p; |
1a848870 SS |
8448 | if (p) |
8449 | p->child = sd; | |
7c16ec58 | 8450 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8451 | |
1e9f28fa SS |
8452 | #ifdef CONFIG_SCHED_MC |
8453 | p = sd; | |
6c99e9ad | 8454 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8455 | SD_INIT(sd, MC); |
1d3504fc | 8456 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8457 | cpumask_and(sched_domain_span(sd), cpu_map, |
8458 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8459 | sd->parent = p; |
1a848870 | 8460 | p->child = sd; |
7c16ec58 | 8461 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8462 | #endif |
8463 | ||
1da177e4 LT |
8464 | #ifdef CONFIG_SCHED_SMT |
8465 | p = sd; | |
6c99e9ad | 8466 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8467 | SD_INIT(sd, SIBLING); |
1d3504fc | 8468 | set_domain_attribute(sd, attr); |
758b2cdc | 8469 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8470 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8471 | sd->parent = p; |
1a848870 | 8472 | p->child = sd; |
7c16ec58 | 8473 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8474 | #endif |
8475 | } | |
8476 | ||
8477 | #ifdef CONFIG_SCHED_SMT | |
8478 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8479 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8480 | cpumask_and(this_sibling_map, |
c69fc56d | 8481 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8482 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8483 | continue; |
8484 | ||
dd41f596 | 8485 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8486 | &cpu_to_cpu_group, |
8487 | send_covered, tmpmask); | |
1da177e4 LT |
8488 | } |
8489 | #endif | |
8490 | ||
1e9f28fa SS |
8491 | #ifdef CONFIG_SCHED_MC |
8492 | /* Set up multi-core groups */ | |
abcd083a | 8493 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8494 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8495 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8496 | continue; |
7c16ec58 | 8497 | |
dd41f596 | 8498 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8499 | &cpu_to_core_group, |
8500 | send_covered, tmpmask); | |
1e9f28fa SS |
8501 | } |
8502 | #endif | |
8503 | ||
1da177e4 | 8504 | /* Set up physical groups */ |
076ac2af | 8505 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8506 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8507 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8508 | continue; |
8509 | ||
7c16ec58 MT |
8510 | init_sched_build_groups(nodemask, cpu_map, |
8511 | &cpu_to_phys_group, | |
8512 | send_covered, tmpmask); | |
1da177e4 LT |
8513 | } |
8514 | ||
8515 | #ifdef CONFIG_NUMA | |
8516 | /* Set up node groups */ | |
7c16ec58 | 8517 | if (sd_allnodes) { |
7c16ec58 MT |
8518 | init_sched_build_groups(cpu_map, cpu_map, |
8519 | &cpu_to_allnodes_group, | |
8520 | send_covered, tmpmask); | |
8521 | } | |
9c1cfda2 | 8522 | |
076ac2af | 8523 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8524 | /* Set up node groups */ |
8525 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8526 | int j; |
8527 | ||
96f874e2 | 8528 | cpumask_clear(covered); |
6ca09dfc | 8529 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8530 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8531 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8532 | continue; |
d1b55138 | 8533 | } |
9c1cfda2 | 8534 | |
4bdbaad3 | 8535 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8536 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8537 | |
6c99e9ad RR |
8538 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8539 | GFP_KERNEL, i); | |
51888ca2 SV |
8540 | if (!sg) { |
8541 | printk(KERN_WARNING "Can not alloc domain group for " | |
8542 | "node %d\n", i); | |
8543 | goto error; | |
8544 | } | |
9c1cfda2 | 8545 | sched_group_nodes[i] = sg; |
abcd083a | 8546 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8547 | struct sched_domain *sd; |
9761eea8 | 8548 | |
62ea9ceb | 8549 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8550 | sd->groups = sg; |
9c1cfda2 | 8551 | } |
5517d86b | 8552 | sg->__cpu_power = 0; |
758b2cdc | 8553 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8554 | sg->next = sg; |
96f874e2 | 8555 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8556 | prev = sg; |
8557 | ||
076ac2af | 8558 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8559 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8560 | |
96f874e2 RR |
8561 | cpumask_complement(notcovered, covered); |
8562 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8563 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8564 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8565 | break; |
8566 | ||
6ca09dfc | 8567 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8568 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8569 | continue; |
8570 | ||
6c99e9ad RR |
8571 | sg = kmalloc_node(sizeof(struct sched_group) + |
8572 | cpumask_size(), | |
15f0b676 | 8573 | GFP_KERNEL, i); |
9c1cfda2 JH |
8574 | if (!sg) { |
8575 | printk(KERN_WARNING | |
8576 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8577 | goto error; |
9c1cfda2 | 8578 | } |
5517d86b | 8579 | sg->__cpu_power = 0; |
758b2cdc | 8580 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8581 | sg->next = prev->next; |
96f874e2 | 8582 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8583 | prev->next = sg; |
8584 | prev = sg; | |
8585 | } | |
9c1cfda2 | 8586 | } |
1da177e4 LT |
8587 | #endif |
8588 | ||
8589 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8590 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8591 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8592 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8593 | |
89c4710e | 8594 | init_sched_groups_power(i, sd); |
5c45bf27 | 8595 | } |
1da177e4 | 8596 | #endif |
1e9f28fa | 8597 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8598 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8599 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8600 | |
89c4710e | 8601 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8602 | } |
8603 | #endif | |
1e9f28fa | 8604 | |
abcd083a | 8605 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8606 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8607 | |
89c4710e | 8608 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8609 | } |
8610 | ||
9c1cfda2 | 8611 | #ifdef CONFIG_NUMA |
076ac2af | 8612 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8613 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8614 | |
6711cab4 SS |
8615 | if (sd_allnodes) { |
8616 | struct sched_group *sg; | |
f712c0c7 | 8617 | |
96f874e2 | 8618 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8619 | tmpmask); |
f712c0c7 SS |
8620 | init_numa_sched_groups_power(sg); |
8621 | } | |
9c1cfda2 JH |
8622 | #endif |
8623 | ||
1da177e4 | 8624 | /* Attach the domains */ |
abcd083a | 8625 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8626 | struct sched_domain *sd; |
8627 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8628 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8629 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8630 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8631 | #else |
6c99e9ad | 8632 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8633 | #endif |
57d885fe | 8634 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8635 | } |
51888ca2 | 8636 | |
3404c8d9 RR |
8637 | err = 0; |
8638 | ||
8639 | free_tmpmask: | |
8640 | free_cpumask_var(tmpmask); | |
8641 | free_send_covered: | |
8642 | free_cpumask_var(send_covered); | |
8643 | free_this_core_map: | |
8644 | free_cpumask_var(this_core_map); | |
8645 | free_this_sibling_map: | |
8646 | free_cpumask_var(this_sibling_map); | |
8647 | free_nodemask: | |
8648 | free_cpumask_var(nodemask); | |
8649 | free_notcovered: | |
8650 | #ifdef CONFIG_NUMA | |
8651 | free_cpumask_var(notcovered); | |
8652 | free_covered: | |
8653 | free_cpumask_var(covered); | |
8654 | free_domainspan: | |
8655 | free_cpumask_var(domainspan); | |
8656 | out: | |
8657 | #endif | |
8658 | return err; | |
8659 | ||
8660 | free_sched_groups: | |
8661 | #ifdef CONFIG_NUMA | |
8662 | kfree(sched_group_nodes); | |
8663 | #endif | |
8664 | goto free_tmpmask; | |
51888ca2 | 8665 | |
a616058b | 8666 | #ifdef CONFIG_NUMA |
51888ca2 | 8667 | error: |
7c16ec58 | 8668 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8669 | free_rootdomain(rd); |
3404c8d9 | 8670 | goto free_tmpmask; |
a616058b | 8671 | #endif |
1da177e4 | 8672 | } |
029190c5 | 8673 | |
96f874e2 | 8674 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8675 | { |
8676 | return __build_sched_domains(cpu_map, NULL); | |
8677 | } | |
8678 | ||
96f874e2 | 8679 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8680 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8681 | static struct sched_domain_attr *dattr_cur; |
8682 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8683 | |
8684 | /* | |
8685 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8686 | * cpumask) fails, then fallback to a single sched domain, |
8687 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8688 | */ |
4212823f | 8689 | static cpumask_var_t fallback_doms; |
029190c5 | 8690 | |
ee79d1bd HC |
8691 | /* |
8692 | * arch_update_cpu_topology lets virtualized architectures update the | |
8693 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8694 | * or 0 if it stayed the same. | |
8695 | */ | |
8696 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8697 | { |
ee79d1bd | 8698 | return 0; |
22e52b07 HC |
8699 | } |
8700 | ||
1a20ff27 | 8701 | /* |
41a2d6cf | 8702 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8703 | * For now this just excludes isolated cpus, but could be used to |
8704 | * exclude other special cases in the future. | |
1a20ff27 | 8705 | */ |
96f874e2 | 8706 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8707 | { |
7378547f MM |
8708 | int err; |
8709 | ||
22e52b07 | 8710 | arch_update_cpu_topology(); |
029190c5 | 8711 | ndoms_cur = 1; |
96f874e2 | 8712 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8713 | if (!doms_cur) |
4212823f | 8714 | doms_cur = fallback_doms; |
dcc30a35 | 8715 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8716 | dattr_cur = NULL; |
7378547f | 8717 | err = build_sched_domains(doms_cur); |
6382bc90 | 8718 | register_sched_domain_sysctl(); |
7378547f MM |
8719 | |
8720 | return err; | |
1a20ff27 DG |
8721 | } |
8722 | ||
96f874e2 RR |
8723 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8724 | struct cpumask *tmpmask) | |
1da177e4 | 8725 | { |
7c16ec58 | 8726 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8727 | } |
1da177e4 | 8728 | |
1a20ff27 DG |
8729 | /* |
8730 | * Detach sched domains from a group of cpus specified in cpu_map | |
8731 | * These cpus will now be attached to the NULL domain | |
8732 | */ | |
96f874e2 | 8733 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8734 | { |
96f874e2 RR |
8735 | /* Save because hotplug lock held. */ |
8736 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8737 | int i; |
8738 | ||
abcd083a | 8739 | for_each_cpu(i, cpu_map) |
57d885fe | 8740 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8741 | synchronize_sched(); |
96f874e2 | 8742 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8743 | } |
8744 | ||
1d3504fc HS |
8745 | /* handle null as "default" */ |
8746 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8747 | struct sched_domain_attr *new, int idx_new) | |
8748 | { | |
8749 | struct sched_domain_attr tmp; | |
8750 | ||
8751 | /* fast path */ | |
8752 | if (!new && !cur) | |
8753 | return 1; | |
8754 | ||
8755 | tmp = SD_ATTR_INIT; | |
8756 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8757 | new ? (new + idx_new) : &tmp, | |
8758 | sizeof(struct sched_domain_attr)); | |
8759 | } | |
8760 | ||
029190c5 PJ |
8761 | /* |
8762 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8763 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8764 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8765 | * It destroys each deleted domain and builds each new domain. | |
8766 | * | |
96f874e2 | 8767 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8768 | * The masks don't intersect (don't overlap.) We should setup one |
8769 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8770 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8771 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8772 | * it as it is. | |
8773 | * | |
41a2d6cf IM |
8774 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8775 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8776 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8777 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8778 | * the single partition 'fallback_doms', it also forces the domains | |
8779 | * to be rebuilt. | |
029190c5 | 8780 | * |
96f874e2 | 8781 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8782 | * ndoms_new == 0 is a special case for destroying existing domains, |
8783 | * and it will not create the default domain. | |
dfb512ec | 8784 | * |
029190c5 PJ |
8785 | * Call with hotplug lock held |
8786 | */ | |
96f874e2 RR |
8787 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8788 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8789 | struct sched_domain_attr *dattr_new) |
029190c5 | 8790 | { |
dfb512ec | 8791 | int i, j, n; |
d65bd5ec | 8792 | int new_topology; |
029190c5 | 8793 | |
712555ee | 8794 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8795 | |
7378547f MM |
8796 | /* always unregister in case we don't destroy any domains */ |
8797 | unregister_sched_domain_sysctl(); | |
8798 | ||
d65bd5ec HC |
8799 | /* Let architecture update cpu core mappings. */ |
8800 | new_topology = arch_update_cpu_topology(); | |
8801 | ||
dfb512ec | 8802 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8803 | |
8804 | /* Destroy deleted domains */ | |
8805 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8806 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8807 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8808 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8809 | goto match1; |
8810 | } | |
8811 | /* no match - a current sched domain not in new doms_new[] */ | |
8812 | detach_destroy_domains(doms_cur + i); | |
8813 | match1: | |
8814 | ; | |
8815 | } | |
8816 | ||
e761b772 MK |
8817 | if (doms_new == NULL) { |
8818 | ndoms_cur = 0; | |
4212823f | 8819 | doms_new = fallback_doms; |
dcc30a35 | 8820 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8821 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8822 | } |
8823 | ||
029190c5 PJ |
8824 | /* Build new domains */ |
8825 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8826 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8827 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8828 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8829 | goto match2; |
8830 | } | |
8831 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8832 | __build_sched_domains(doms_new + i, |
8833 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8834 | match2: |
8835 | ; | |
8836 | } | |
8837 | ||
8838 | /* Remember the new sched domains */ | |
4212823f | 8839 | if (doms_cur != fallback_doms) |
029190c5 | 8840 | kfree(doms_cur); |
1d3504fc | 8841 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8842 | doms_cur = doms_new; |
1d3504fc | 8843 | dattr_cur = dattr_new; |
029190c5 | 8844 | ndoms_cur = ndoms_new; |
7378547f MM |
8845 | |
8846 | register_sched_domain_sysctl(); | |
a1835615 | 8847 | |
712555ee | 8848 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8849 | } |
8850 | ||
5c45bf27 | 8851 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8852 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8853 | { |
95402b38 | 8854 | get_online_cpus(); |
dfb512ec MK |
8855 | |
8856 | /* Destroy domains first to force the rebuild */ | |
8857 | partition_sched_domains(0, NULL, NULL); | |
8858 | ||
e761b772 | 8859 | rebuild_sched_domains(); |
95402b38 | 8860 | put_online_cpus(); |
5c45bf27 SS |
8861 | } |
8862 | ||
8863 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8864 | { | |
afb8a9b7 | 8865 | unsigned int level = 0; |
5c45bf27 | 8866 | |
afb8a9b7 GS |
8867 | if (sscanf(buf, "%u", &level) != 1) |
8868 | return -EINVAL; | |
8869 | ||
8870 | /* | |
8871 | * level is always be positive so don't check for | |
8872 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8873 | * What happens on 0 or 1 byte write, | |
8874 | * need to check for count as well? | |
8875 | */ | |
8876 | ||
8877 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8878 | return -EINVAL; |
8879 | ||
8880 | if (smt) | |
afb8a9b7 | 8881 | sched_smt_power_savings = level; |
5c45bf27 | 8882 | else |
afb8a9b7 | 8883 | sched_mc_power_savings = level; |
5c45bf27 | 8884 | |
c70f22d2 | 8885 | arch_reinit_sched_domains(); |
5c45bf27 | 8886 | |
c70f22d2 | 8887 | return count; |
5c45bf27 SS |
8888 | } |
8889 | ||
5c45bf27 | 8890 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8891 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8892 | char *page) | |
5c45bf27 SS |
8893 | { |
8894 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8895 | } | |
f718cd4a | 8896 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8897 | const char *buf, size_t count) |
5c45bf27 SS |
8898 | { |
8899 | return sched_power_savings_store(buf, count, 0); | |
8900 | } | |
f718cd4a AK |
8901 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8902 | sched_mc_power_savings_show, | |
8903 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8904 | #endif |
8905 | ||
8906 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8907 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8908 | char *page) | |
5c45bf27 SS |
8909 | { |
8910 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8911 | } | |
f718cd4a | 8912 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8913 | const char *buf, size_t count) |
5c45bf27 SS |
8914 | { |
8915 | return sched_power_savings_store(buf, count, 1); | |
8916 | } | |
f718cd4a AK |
8917 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8918 | sched_smt_power_savings_show, | |
6707de00 AB |
8919 | sched_smt_power_savings_store); |
8920 | #endif | |
8921 | ||
39aac648 | 8922 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8923 | { |
8924 | int err = 0; | |
8925 | ||
8926 | #ifdef CONFIG_SCHED_SMT | |
8927 | if (smt_capable()) | |
8928 | err = sysfs_create_file(&cls->kset.kobj, | |
8929 | &attr_sched_smt_power_savings.attr); | |
8930 | #endif | |
8931 | #ifdef CONFIG_SCHED_MC | |
8932 | if (!err && mc_capable()) | |
8933 | err = sysfs_create_file(&cls->kset.kobj, | |
8934 | &attr_sched_mc_power_savings.attr); | |
8935 | #endif | |
8936 | return err; | |
8937 | } | |
6d6bc0ad | 8938 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8939 | |
e761b772 | 8940 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8941 | /* |
e761b772 MK |
8942 | * Add online and remove offline CPUs from the scheduler domains. |
8943 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8944 | */ |
8945 | static int update_sched_domains(struct notifier_block *nfb, | |
8946 | unsigned long action, void *hcpu) | |
e761b772 MK |
8947 | { |
8948 | switch (action) { | |
8949 | case CPU_ONLINE: | |
8950 | case CPU_ONLINE_FROZEN: | |
8951 | case CPU_DEAD: | |
8952 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8953 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8954 | return NOTIFY_OK; |
8955 | ||
8956 | default: | |
8957 | return NOTIFY_DONE; | |
8958 | } | |
8959 | } | |
8960 | #endif | |
8961 | ||
8962 | static int update_runtime(struct notifier_block *nfb, | |
8963 | unsigned long action, void *hcpu) | |
1da177e4 | 8964 | { |
7def2be1 PZ |
8965 | int cpu = (int)(long)hcpu; |
8966 | ||
1da177e4 | 8967 | switch (action) { |
1da177e4 | 8968 | case CPU_DOWN_PREPARE: |
8bb78442 | 8969 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8970 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8971 | return NOTIFY_OK; |
8972 | ||
1da177e4 | 8973 | case CPU_DOWN_FAILED: |
8bb78442 | 8974 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8975 | case CPU_ONLINE: |
8bb78442 | 8976 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8977 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8978 | return NOTIFY_OK; |
8979 | ||
1da177e4 LT |
8980 | default: |
8981 | return NOTIFY_DONE; | |
8982 | } | |
1da177e4 | 8983 | } |
1da177e4 LT |
8984 | |
8985 | void __init sched_init_smp(void) | |
8986 | { | |
dcc30a35 RR |
8987 | cpumask_var_t non_isolated_cpus; |
8988 | ||
8989 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8990 | |
434d53b0 MT |
8991 | #if defined(CONFIG_NUMA) |
8992 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8993 | GFP_KERNEL); | |
8994 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8995 | #endif | |
95402b38 | 8996 | get_online_cpus(); |
712555ee | 8997 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8998 | arch_init_sched_domains(cpu_online_mask); |
8999 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9000 | if (cpumask_empty(non_isolated_cpus)) | |
9001 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9002 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9003 | put_online_cpus(); |
e761b772 MK |
9004 | |
9005 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9006 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9007 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9008 | #endif |
9009 | ||
9010 | /* RT runtime code needs to handle some hotplug events */ | |
9011 | hotcpu_notifier(update_runtime, 0); | |
9012 | ||
b328ca18 | 9013 | init_hrtick(); |
5c1e1767 NP |
9014 | |
9015 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9016 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9017 | BUG(); |
19978ca6 | 9018 | sched_init_granularity(); |
dcc30a35 | 9019 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9020 | |
9021 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9022 | init_sched_rt_class(); |
1da177e4 LT |
9023 | } |
9024 | #else | |
9025 | void __init sched_init_smp(void) | |
9026 | { | |
19978ca6 | 9027 | sched_init_granularity(); |
1da177e4 LT |
9028 | } |
9029 | #endif /* CONFIG_SMP */ | |
9030 | ||
9031 | int in_sched_functions(unsigned long addr) | |
9032 | { | |
1da177e4 LT |
9033 | return in_lock_functions(addr) || |
9034 | (addr >= (unsigned long)__sched_text_start | |
9035 | && addr < (unsigned long)__sched_text_end); | |
9036 | } | |
9037 | ||
a9957449 | 9038 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9039 | { |
9040 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9041 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9042 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9043 | cfs_rq->rq = rq; | |
9044 | #endif | |
67e9fb2a | 9045 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9046 | } |
9047 | ||
fa85ae24 PZ |
9048 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9049 | { | |
9050 | struct rt_prio_array *array; | |
9051 | int i; | |
9052 | ||
9053 | array = &rt_rq->active; | |
9054 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9055 | INIT_LIST_HEAD(array->queue + i); | |
9056 | __clear_bit(i, array->bitmap); | |
9057 | } | |
9058 | /* delimiter for bitsearch: */ | |
9059 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9060 | ||
052f1dc7 | 9061 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9062 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9063 | #ifdef CONFIG_SMP |
e864c499 | 9064 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9065 | #endif |
48d5e258 | 9066 | #endif |
fa85ae24 PZ |
9067 | #ifdef CONFIG_SMP |
9068 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9069 | rt_rq->overloaded = 0; |
917b627d | 9070 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9071 | #endif |
9072 | ||
9073 | rt_rq->rt_time = 0; | |
9074 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9075 | rt_rq->rt_runtime = 0; |
9076 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9077 | |
052f1dc7 | 9078 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9079 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9080 | rt_rq->rq = rq; |
9081 | #endif | |
fa85ae24 PZ |
9082 | } |
9083 | ||
6f505b16 | 9084 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9085 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9086 | struct sched_entity *se, int cpu, int add, | |
9087 | struct sched_entity *parent) | |
6f505b16 | 9088 | { |
ec7dc8ac | 9089 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9090 | tg->cfs_rq[cpu] = cfs_rq; |
9091 | init_cfs_rq(cfs_rq, rq); | |
9092 | cfs_rq->tg = tg; | |
9093 | if (add) | |
9094 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9095 | ||
9096 | tg->se[cpu] = se; | |
354d60c2 DG |
9097 | /* se could be NULL for init_task_group */ |
9098 | if (!se) | |
9099 | return; | |
9100 | ||
ec7dc8ac DG |
9101 | if (!parent) |
9102 | se->cfs_rq = &rq->cfs; | |
9103 | else | |
9104 | se->cfs_rq = parent->my_q; | |
9105 | ||
6f505b16 PZ |
9106 | se->my_q = cfs_rq; |
9107 | se->load.weight = tg->shares; | |
e05510d0 | 9108 | se->load.inv_weight = 0; |
ec7dc8ac | 9109 | se->parent = parent; |
6f505b16 | 9110 | } |
052f1dc7 | 9111 | #endif |
6f505b16 | 9112 | |
052f1dc7 | 9113 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9114 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9115 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9116 | struct sched_rt_entity *parent) | |
6f505b16 | 9117 | { |
ec7dc8ac DG |
9118 | struct rq *rq = cpu_rq(cpu); |
9119 | ||
6f505b16 PZ |
9120 | tg->rt_rq[cpu] = rt_rq; |
9121 | init_rt_rq(rt_rq, rq); | |
9122 | rt_rq->tg = tg; | |
9123 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9124 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9125 | if (add) |
9126 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9127 | ||
9128 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9129 | if (!rt_se) |
9130 | return; | |
9131 | ||
ec7dc8ac DG |
9132 | if (!parent) |
9133 | rt_se->rt_rq = &rq->rt; | |
9134 | else | |
9135 | rt_se->rt_rq = parent->my_q; | |
9136 | ||
6f505b16 | 9137 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9138 | rt_se->parent = parent; |
6f505b16 PZ |
9139 | INIT_LIST_HEAD(&rt_se->run_list); |
9140 | } | |
9141 | #endif | |
9142 | ||
1da177e4 LT |
9143 | void __init sched_init(void) |
9144 | { | |
dd41f596 | 9145 | int i, j; |
434d53b0 MT |
9146 | unsigned long alloc_size = 0, ptr; |
9147 | ||
9148 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9149 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9150 | #endif | |
9151 | #ifdef CONFIG_RT_GROUP_SCHED | |
9152 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9153 | #endif |
9154 | #ifdef CONFIG_USER_SCHED | |
9155 | alloc_size *= 2; | |
df7c8e84 RR |
9156 | #endif |
9157 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9158 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9159 | #endif |
9160 | /* | |
9161 | * As sched_init() is called before page_alloc is setup, | |
9162 | * we use alloc_bootmem(). | |
9163 | */ | |
9164 | if (alloc_size) { | |
36b7b6d4 | 9165 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9166 | |
9167 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9168 | init_task_group.se = (struct sched_entity **)ptr; | |
9169 | ptr += nr_cpu_ids * sizeof(void **); | |
9170 | ||
9171 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9172 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9173 | |
9174 | #ifdef CONFIG_USER_SCHED | |
9175 | root_task_group.se = (struct sched_entity **)ptr; | |
9176 | ptr += nr_cpu_ids * sizeof(void **); | |
9177 | ||
9178 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9179 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9180 | #endif /* CONFIG_USER_SCHED */ |
9181 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9182 | #ifdef CONFIG_RT_GROUP_SCHED |
9183 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9184 | ptr += nr_cpu_ids * sizeof(void **); | |
9185 | ||
9186 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9187 | ptr += nr_cpu_ids * sizeof(void **); |
9188 | ||
9189 | #ifdef CONFIG_USER_SCHED | |
9190 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9191 | ptr += nr_cpu_ids * sizeof(void **); | |
9192 | ||
9193 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9194 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9195 | #endif /* CONFIG_USER_SCHED */ |
9196 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9197 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9198 | for_each_possible_cpu(i) { | |
9199 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9200 | ptr += cpumask_size(); | |
9201 | } | |
9202 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9203 | } |
dd41f596 | 9204 | |
57d885fe GH |
9205 | #ifdef CONFIG_SMP |
9206 | init_defrootdomain(); | |
9207 | #endif | |
9208 | ||
d0b27fa7 PZ |
9209 | init_rt_bandwidth(&def_rt_bandwidth, |
9210 | global_rt_period(), global_rt_runtime()); | |
9211 | ||
9212 | #ifdef CONFIG_RT_GROUP_SCHED | |
9213 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9214 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9215 | #ifdef CONFIG_USER_SCHED |
9216 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9217 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9218 | #endif /* CONFIG_USER_SCHED */ |
9219 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9220 | |
052f1dc7 | 9221 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9222 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9223 | INIT_LIST_HEAD(&init_task_group.children); |
9224 | ||
9225 | #ifdef CONFIG_USER_SCHED | |
9226 | INIT_LIST_HEAD(&root_task_group.children); | |
9227 | init_task_group.parent = &root_task_group; | |
9228 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9229 | #endif /* CONFIG_USER_SCHED */ |
9230 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9231 | |
0a945022 | 9232 | for_each_possible_cpu(i) { |
70b97a7f | 9233 | struct rq *rq; |
1da177e4 LT |
9234 | |
9235 | rq = cpu_rq(i); | |
9236 | spin_lock_init(&rq->lock); | |
7897986b | 9237 | rq->nr_running = 0; |
dce48a84 TG |
9238 | rq->calc_load_active = 0; |
9239 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9240 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9241 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9242 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9243 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9244 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9245 | #ifdef CONFIG_CGROUP_SCHED |
9246 | /* | |
9247 | * How much cpu bandwidth does init_task_group get? | |
9248 | * | |
9249 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9250 | * gets 100% of the cpu resources in the system. This overall | |
9251 | * system cpu resource is divided among the tasks of | |
9252 | * init_task_group and its child task-groups in a fair manner, | |
9253 | * based on each entity's (task or task-group's) weight | |
9254 | * (se->load.weight). | |
9255 | * | |
9256 | * In other words, if init_task_group has 10 tasks of weight | |
9257 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9258 | * then A0's share of the cpu resource is: | |
9259 | * | |
0d905bca | 9260 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9261 | * |
9262 | * We achieve this by letting init_task_group's tasks sit | |
9263 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9264 | */ | |
ec7dc8ac | 9265 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9266 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9267 | root_task_group.shares = NICE_0_LOAD; |
9268 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9269 | /* |
9270 | * In case of task-groups formed thr' the user id of tasks, | |
9271 | * init_task_group represents tasks belonging to root user. | |
9272 | * Hence it forms a sibling of all subsequent groups formed. | |
9273 | * In this case, init_task_group gets only a fraction of overall | |
9274 | * system cpu resource, based on the weight assigned to root | |
9275 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9276 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9277 | * (init_cfs_rq) and having one entity represent this group of | |
9278 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9279 | */ | |
ec7dc8ac | 9280 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9281 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9282 | &per_cpu(init_sched_entity, i), i, 1, |
9283 | root_task_group.se[i]); | |
6f505b16 | 9284 | |
052f1dc7 | 9285 | #endif |
354d60c2 DG |
9286 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9287 | ||
9288 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9289 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9290 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9291 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9292 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9293 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9294 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9295 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9296 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9297 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9298 | root_task_group.rt_se[i]); | |
354d60c2 | 9299 | #endif |
dd41f596 | 9300 | #endif |
1da177e4 | 9301 | |
dd41f596 IM |
9302 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9303 | rq->cpu_load[j] = 0; | |
1da177e4 | 9304 | #ifdef CONFIG_SMP |
41c7ce9a | 9305 | rq->sd = NULL; |
57d885fe | 9306 | rq->rd = NULL; |
1da177e4 | 9307 | rq->active_balance = 0; |
dd41f596 | 9308 | rq->next_balance = jiffies; |
1da177e4 | 9309 | rq->push_cpu = 0; |
0a2966b4 | 9310 | rq->cpu = i; |
1f11eb6a | 9311 | rq->online = 0; |
1da177e4 LT |
9312 | rq->migration_thread = NULL; |
9313 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9314 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9315 | #endif |
8f4d37ec | 9316 | init_rq_hrtick(rq); |
1da177e4 | 9317 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9318 | } |
9319 | ||
2dd73a4f | 9320 | set_load_weight(&init_task); |
b50f60ce | 9321 | |
e107be36 AK |
9322 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9323 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9324 | #endif | |
9325 | ||
c9819f45 | 9326 | #ifdef CONFIG_SMP |
962cf36c | 9327 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9328 | #endif |
9329 | ||
b50f60ce HC |
9330 | #ifdef CONFIG_RT_MUTEXES |
9331 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9332 | #endif | |
9333 | ||
1da177e4 LT |
9334 | /* |
9335 | * The boot idle thread does lazy MMU switching as well: | |
9336 | */ | |
9337 | atomic_inc(&init_mm.mm_count); | |
9338 | enter_lazy_tlb(&init_mm, current); | |
9339 | ||
9340 | /* | |
9341 | * Make us the idle thread. Technically, schedule() should not be | |
9342 | * called from this thread, however somewhere below it might be, | |
9343 | * but because we are the idle thread, we just pick up running again | |
9344 | * when this runqueue becomes "idle". | |
9345 | */ | |
9346 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9347 | |
9348 | calc_load_update = jiffies + LOAD_FREQ; | |
9349 | ||
dd41f596 IM |
9350 | /* |
9351 | * During early bootup we pretend to be a normal task: | |
9352 | */ | |
9353 | current->sched_class = &fair_sched_class; | |
6892b75e | 9354 | |
6a7b3dc3 | 9355 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9356 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9357 | #ifdef CONFIG_SMP |
7d1e6a9b | 9358 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9359 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9360 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9361 | #endif |
4bdddf8f | 9362 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9363 | #endif /* SMP */ |
6a7b3dc3 | 9364 | |
0d905bca IM |
9365 | perf_counter_init(); |
9366 | ||
6892b75e | 9367 | scheduler_running = 1; |
1da177e4 LT |
9368 | } |
9369 | ||
9370 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9371 | void __might_sleep(char *file, int line) | |
9372 | { | |
48f24c4d | 9373 | #ifdef in_atomic |
1da177e4 LT |
9374 | static unsigned long prev_jiffy; /* ratelimiting */ |
9375 | ||
aef745fc IM |
9376 | if ((!in_atomic() && !irqs_disabled()) || |
9377 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9378 | return; | |
9379 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9380 | return; | |
9381 | prev_jiffy = jiffies; | |
9382 | ||
9383 | printk(KERN_ERR | |
9384 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9385 | file, line); | |
9386 | printk(KERN_ERR | |
9387 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9388 | in_atomic(), irqs_disabled(), | |
9389 | current->pid, current->comm); | |
9390 | ||
9391 | debug_show_held_locks(current); | |
9392 | if (irqs_disabled()) | |
9393 | print_irqtrace_events(current); | |
9394 | dump_stack(); | |
1da177e4 LT |
9395 | #endif |
9396 | } | |
9397 | EXPORT_SYMBOL(__might_sleep); | |
9398 | #endif | |
9399 | ||
9400 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9401 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9402 | { | |
9403 | int on_rq; | |
3e51f33f | 9404 | |
3a5e4dc1 AK |
9405 | update_rq_clock(rq); |
9406 | on_rq = p->se.on_rq; | |
9407 | if (on_rq) | |
9408 | deactivate_task(rq, p, 0); | |
9409 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9410 | if (on_rq) { | |
9411 | activate_task(rq, p, 0); | |
9412 | resched_task(rq->curr); | |
9413 | } | |
9414 | } | |
9415 | ||
1da177e4 LT |
9416 | void normalize_rt_tasks(void) |
9417 | { | |
a0f98a1c | 9418 | struct task_struct *g, *p; |
1da177e4 | 9419 | unsigned long flags; |
70b97a7f | 9420 | struct rq *rq; |
1da177e4 | 9421 | |
4cf5d77a | 9422 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9423 | do_each_thread(g, p) { |
178be793 IM |
9424 | /* |
9425 | * Only normalize user tasks: | |
9426 | */ | |
9427 | if (!p->mm) | |
9428 | continue; | |
9429 | ||
6cfb0d5d | 9430 | p->se.exec_start = 0; |
6cfb0d5d | 9431 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9432 | p->se.wait_start = 0; |
dd41f596 | 9433 | p->se.sleep_start = 0; |
dd41f596 | 9434 | p->se.block_start = 0; |
6cfb0d5d | 9435 | #endif |
dd41f596 IM |
9436 | |
9437 | if (!rt_task(p)) { | |
9438 | /* | |
9439 | * Renice negative nice level userspace | |
9440 | * tasks back to 0: | |
9441 | */ | |
9442 | if (TASK_NICE(p) < 0 && p->mm) | |
9443 | set_user_nice(p, 0); | |
1da177e4 | 9444 | continue; |
dd41f596 | 9445 | } |
1da177e4 | 9446 | |
4cf5d77a | 9447 | spin_lock(&p->pi_lock); |
b29739f9 | 9448 | rq = __task_rq_lock(p); |
1da177e4 | 9449 | |
178be793 | 9450 | normalize_task(rq, p); |
3a5e4dc1 | 9451 | |
b29739f9 | 9452 | __task_rq_unlock(rq); |
4cf5d77a | 9453 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9454 | } while_each_thread(g, p); |
9455 | ||
4cf5d77a | 9456 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9457 | } |
9458 | ||
9459 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9460 | |
9461 | #ifdef CONFIG_IA64 | |
9462 | /* | |
9463 | * These functions are only useful for the IA64 MCA handling. | |
9464 | * | |
9465 | * They can only be called when the whole system has been | |
9466 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9467 | * activity can take place. Using them for anything else would | |
9468 | * be a serious bug, and as a result, they aren't even visible | |
9469 | * under any other configuration. | |
9470 | */ | |
9471 | ||
9472 | /** | |
9473 | * curr_task - return the current task for a given cpu. | |
9474 | * @cpu: the processor in question. | |
9475 | * | |
9476 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9477 | */ | |
36c8b586 | 9478 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9479 | { |
9480 | return cpu_curr(cpu); | |
9481 | } | |
9482 | ||
9483 | /** | |
9484 | * set_curr_task - set the current task for a given cpu. | |
9485 | * @cpu: the processor in question. | |
9486 | * @p: the task pointer to set. | |
9487 | * | |
9488 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9489 | * are serviced on a separate stack. It allows the architecture to switch the |
9490 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9491 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9492 | * and caller must save the original value of the current task (see | |
9493 | * curr_task() above) and restore that value before reenabling interrupts and | |
9494 | * re-starting the system. | |
9495 | * | |
9496 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9497 | */ | |
36c8b586 | 9498 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9499 | { |
9500 | cpu_curr(cpu) = p; | |
9501 | } | |
9502 | ||
9503 | #endif | |
29f59db3 | 9504 | |
bccbe08a PZ |
9505 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9506 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9507 | { |
9508 | int i; | |
9509 | ||
9510 | for_each_possible_cpu(i) { | |
9511 | if (tg->cfs_rq) | |
9512 | kfree(tg->cfs_rq[i]); | |
9513 | if (tg->se) | |
9514 | kfree(tg->se[i]); | |
6f505b16 PZ |
9515 | } |
9516 | ||
9517 | kfree(tg->cfs_rq); | |
9518 | kfree(tg->se); | |
6f505b16 PZ |
9519 | } |
9520 | ||
ec7dc8ac DG |
9521 | static |
9522 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9523 | { |
29f59db3 | 9524 | struct cfs_rq *cfs_rq; |
eab17229 | 9525 | struct sched_entity *se; |
9b5b7751 | 9526 | struct rq *rq; |
29f59db3 SV |
9527 | int i; |
9528 | ||
434d53b0 | 9529 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9530 | if (!tg->cfs_rq) |
9531 | goto err; | |
434d53b0 | 9532 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9533 | if (!tg->se) |
9534 | goto err; | |
052f1dc7 PZ |
9535 | |
9536 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9537 | |
9538 | for_each_possible_cpu(i) { | |
9b5b7751 | 9539 | rq = cpu_rq(i); |
29f59db3 | 9540 | |
eab17229 LZ |
9541 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9542 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9543 | if (!cfs_rq) |
9544 | goto err; | |
9545 | ||
eab17229 LZ |
9546 | se = kzalloc_node(sizeof(struct sched_entity), |
9547 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9548 | if (!se) |
9549 | goto err; | |
9550 | ||
eab17229 | 9551 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9552 | } |
9553 | ||
9554 | return 1; | |
9555 | ||
9556 | err: | |
9557 | return 0; | |
9558 | } | |
9559 | ||
9560 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9561 | { | |
9562 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9563 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9564 | } | |
9565 | ||
9566 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9567 | { | |
9568 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9569 | } | |
6d6bc0ad | 9570 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9571 | static inline void free_fair_sched_group(struct task_group *tg) |
9572 | { | |
9573 | } | |
9574 | ||
ec7dc8ac DG |
9575 | static inline |
9576 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9577 | { |
9578 | return 1; | |
9579 | } | |
9580 | ||
9581 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9582 | { | |
9583 | } | |
9584 | ||
9585 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9586 | { | |
9587 | } | |
6d6bc0ad | 9588 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9589 | |
9590 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9591 | static void free_rt_sched_group(struct task_group *tg) |
9592 | { | |
9593 | int i; | |
9594 | ||
d0b27fa7 PZ |
9595 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9596 | ||
bccbe08a PZ |
9597 | for_each_possible_cpu(i) { |
9598 | if (tg->rt_rq) | |
9599 | kfree(tg->rt_rq[i]); | |
9600 | if (tg->rt_se) | |
9601 | kfree(tg->rt_se[i]); | |
9602 | } | |
9603 | ||
9604 | kfree(tg->rt_rq); | |
9605 | kfree(tg->rt_se); | |
9606 | } | |
9607 | ||
ec7dc8ac DG |
9608 | static |
9609 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9610 | { |
9611 | struct rt_rq *rt_rq; | |
eab17229 | 9612 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9613 | struct rq *rq; |
9614 | int i; | |
9615 | ||
434d53b0 | 9616 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9617 | if (!tg->rt_rq) |
9618 | goto err; | |
434d53b0 | 9619 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9620 | if (!tg->rt_se) |
9621 | goto err; | |
9622 | ||
d0b27fa7 PZ |
9623 | init_rt_bandwidth(&tg->rt_bandwidth, |
9624 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9625 | |
9626 | for_each_possible_cpu(i) { | |
9627 | rq = cpu_rq(i); | |
9628 | ||
eab17229 LZ |
9629 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9630 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9631 | if (!rt_rq) |
9632 | goto err; | |
29f59db3 | 9633 | |
eab17229 LZ |
9634 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9635 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9636 | if (!rt_se) |
9637 | goto err; | |
29f59db3 | 9638 | |
eab17229 | 9639 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9640 | } |
9641 | ||
bccbe08a PZ |
9642 | return 1; |
9643 | ||
9644 | err: | |
9645 | return 0; | |
9646 | } | |
9647 | ||
9648 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9649 | { | |
9650 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9651 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9652 | } | |
9653 | ||
9654 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9655 | { | |
9656 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9657 | } | |
6d6bc0ad | 9658 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9659 | static inline void free_rt_sched_group(struct task_group *tg) |
9660 | { | |
9661 | } | |
9662 | ||
ec7dc8ac DG |
9663 | static inline |
9664 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9665 | { |
9666 | return 1; | |
9667 | } | |
9668 | ||
9669 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9670 | { | |
9671 | } | |
9672 | ||
9673 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9674 | { | |
9675 | } | |
6d6bc0ad | 9676 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9677 | |
d0b27fa7 | 9678 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9679 | static void free_sched_group(struct task_group *tg) |
9680 | { | |
9681 | free_fair_sched_group(tg); | |
9682 | free_rt_sched_group(tg); | |
9683 | kfree(tg); | |
9684 | } | |
9685 | ||
9686 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9687 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9688 | { |
9689 | struct task_group *tg; | |
9690 | unsigned long flags; | |
9691 | int i; | |
9692 | ||
9693 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9694 | if (!tg) | |
9695 | return ERR_PTR(-ENOMEM); | |
9696 | ||
ec7dc8ac | 9697 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9698 | goto err; |
9699 | ||
ec7dc8ac | 9700 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9701 | goto err; |
9702 | ||
8ed36996 | 9703 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9704 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9705 | register_fair_sched_group(tg, i); |
9706 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9707 | } |
6f505b16 | 9708 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9709 | |
9710 | WARN_ON(!parent); /* root should already exist */ | |
9711 | ||
9712 | tg->parent = parent; | |
f473aa5e | 9713 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9714 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9715 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9716 | |
9b5b7751 | 9717 | return tg; |
29f59db3 SV |
9718 | |
9719 | err: | |
6f505b16 | 9720 | free_sched_group(tg); |
29f59db3 SV |
9721 | return ERR_PTR(-ENOMEM); |
9722 | } | |
9723 | ||
9b5b7751 | 9724 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9725 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9726 | { |
29f59db3 | 9727 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9728 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9729 | } |
9730 | ||
9b5b7751 | 9731 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9732 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9733 | { |
8ed36996 | 9734 | unsigned long flags; |
9b5b7751 | 9735 | int i; |
29f59db3 | 9736 | |
8ed36996 | 9737 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9738 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9739 | unregister_fair_sched_group(tg, i); |
9740 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9741 | } |
6f505b16 | 9742 | list_del_rcu(&tg->list); |
f473aa5e | 9743 | list_del_rcu(&tg->siblings); |
8ed36996 | 9744 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9745 | |
9b5b7751 | 9746 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9747 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9748 | } |
9749 | ||
9b5b7751 | 9750 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9751 | * The caller of this function should have put the task in its new group |
9752 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9753 | * reflect its new group. | |
9b5b7751 SV |
9754 | */ |
9755 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9756 | { |
9757 | int on_rq, running; | |
9758 | unsigned long flags; | |
9759 | struct rq *rq; | |
9760 | ||
9761 | rq = task_rq_lock(tsk, &flags); | |
9762 | ||
29f59db3 SV |
9763 | update_rq_clock(rq); |
9764 | ||
051a1d1a | 9765 | running = task_current(rq, tsk); |
29f59db3 SV |
9766 | on_rq = tsk->se.on_rq; |
9767 | ||
0e1f3483 | 9768 | if (on_rq) |
29f59db3 | 9769 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9770 | if (unlikely(running)) |
9771 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9772 | |
6f505b16 | 9773 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9774 | |
810b3817 PZ |
9775 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9776 | if (tsk->sched_class->moved_group) | |
9777 | tsk->sched_class->moved_group(tsk); | |
9778 | #endif | |
9779 | ||
0e1f3483 HS |
9780 | if (unlikely(running)) |
9781 | tsk->sched_class->set_curr_task(rq); | |
9782 | if (on_rq) | |
7074badb | 9783 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9784 | |
29f59db3 SV |
9785 | task_rq_unlock(rq, &flags); |
9786 | } | |
6d6bc0ad | 9787 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9788 | |
052f1dc7 | 9789 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9790 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9791 | { |
9792 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9793 | int on_rq; |
9794 | ||
29f59db3 | 9795 | on_rq = se->on_rq; |
62fb1851 | 9796 | if (on_rq) |
29f59db3 SV |
9797 | dequeue_entity(cfs_rq, se, 0); |
9798 | ||
9799 | se->load.weight = shares; | |
e05510d0 | 9800 | se->load.inv_weight = 0; |
29f59db3 | 9801 | |
62fb1851 | 9802 | if (on_rq) |
29f59db3 | 9803 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9804 | } |
62fb1851 | 9805 | |
c09595f6 PZ |
9806 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9807 | { | |
9808 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9809 | struct rq *rq = cfs_rq->rq; | |
9810 | unsigned long flags; | |
9811 | ||
9812 | spin_lock_irqsave(&rq->lock, flags); | |
9813 | __set_se_shares(se, shares); | |
9814 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9815 | } |
9816 | ||
8ed36996 PZ |
9817 | static DEFINE_MUTEX(shares_mutex); |
9818 | ||
4cf86d77 | 9819 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9820 | { |
9821 | int i; | |
8ed36996 | 9822 | unsigned long flags; |
c61935fd | 9823 | |
ec7dc8ac DG |
9824 | /* |
9825 | * We can't change the weight of the root cgroup. | |
9826 | */ | |
9827 | if (!tg->se[0]) | |
9828 | return -EINVAL; | |
9829 | ||
18d95a28 PZ |
9830 | if (shares < MIN_SHARES) |
9831 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9832 | else if (shares > MAX_SHARES) |
9833 | shares = MAX_SHARES; | |
62fb1851 | 9834 | |
8ed36996 | 9835 | mutex_lock(&shares_mutex); |
9b5b7751 | 9836 | if (tg->shares == shares) |
5cb350ba | 9837 | goto done; |
29f59db3 | 9838 | |
8ed36996 | 9839 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9840 | for_each_possible_cpu(i) |
9841 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9842 | list_del_rcu(&tg->siblings); |
8ed36996 | 9843 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9844 | |
9845 | /* wait for any ongoing reference to this group to finish */ | |
9846 | synchronize_sched(); | |
9847 | ||
9848 | /* | |
9849 | * Now we are free to modify the group's share on each cpu | |
9850 | * w/o tripping rebalance_share or load_balance_fair. | |
9851 | */ | |
9b5b7751 | 9852 | tg->shares = shares; |
c09595f6 PZ |
9853 | for_each_possible_cpu(i) { |
9854 | /* | |
9855 | * force a rebalance | |
9856 | */ | |
9857 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9858 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9859 | } |
29f59db3 | 9860 | |
6b2d7700 SV |
9861 | /* |
9862 | * Enable load balance activity on this group, by inserting it back on | |
9863 | * each cpu's rq->leaf_cfs_rq_list. | |
9864 | */ | |
8ed36996 | 9865 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9866 | for_each_possible_cpu(i) |
9867 | register_fair_sched_group(tg, i); | |
f473aa5e | 9868 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9869 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9870 | done: |
8ed36996 | 9871 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9872 | return 0; |
29f59db3 SV |
9873 | } |
9874 | ||
5cb350ba DG |
9875 | unsigned long sched_group_shares(struct task_group *tg) |
9876 | { | |
9877 | return tg->shares; | |
9878 | } | |
052f1dc7 | 9879 | #endif |
5cb350ba | 9880 | |
052f1dc7 | 9881 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9882 | /* |
9f0c1e56 | 9883 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9884 | */ |
9f0c1e56 PZ |
9885 | static DEFINE_MUTEX(rt_constraints_mutex); |
9886 | ||
9887 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9888 | { | |
9889 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9890 | return 1ULL << 20; |
9f0c1e56 | 9891 | |
9a7e0b18 | 9892 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9893 | } |
9894 | ||
9a7e0b18 PZ |
9895 | /* Must be called with tasklist_lock held */ |
9896 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9897 | { |
9a7e0b18 | 9898 | struct task_struct *g, *p; |
b40b2e8e | 9899 | |
9a7e0b18 PZ |
9900 | do_each_thread(g, p) { |
9901 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9902 | return 1; | |
9903 | } while_each_thread(g, p); | |
b40b2e8e | 9904 | |
9a7e0b18 PZ |
9905 | return 0; |
9906 | } | |
b40b2e8e | 9907 | |
9a7e0b18 PZ |
9908 | struct rt_schedulable_data { |
9909 | struct task_group *tg; | |
9910 | u64 rt_period; | |
9911 | u64 rt_runtime; | |
9912 | }; | |
b40b2e8e | 9913 | |
9a7e0b18 PZ |
9914 | static int tg_schedulable(struct task_group *tg, void *data) |
9915 | { | |
9916 | struct rt_schedulable_data *d = data; | |
9917 | struct task_group *child; | |
9918 | unsigned long total, sum = 0; | |
9919 | u64 period, runtime; | |
b40b2e8e | 9920 | |
9a7e0b18 PZ |
9921 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9922 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9923 | |
9a7e0b18 PZ |
9924 | if (tg == d->tg) { |
9925 | period = d->rt_period; | |
9926 | runtime = d->rt_runtime; | |
b40b2e8e | 9927 | } |
b40b2e8e | 9928 | |
98a4826b PZ |
9929 | #ifdef CONFIG_USER_SCHED |
9930 | if (tg == &root_task_group) { | |
9931 | period = global_rt_period(); | |
9932 | runtime = global_rt_runtime(); | |
9933 | } | |
9934 | #endif | |
9935 | ||
4653f803 PZ |
9936 | /* |
9937 | * Cannot have more runtime than the period. | |
9938 | */ | |
9939 | if (runtime > period && runtime != RUNTIME_INF) | |
9940 | return -EINVAL; | |
6f505b16 | 9941 | |
4653f803 PZ |
9942 | /* |
9943 | * Ensure we don't starve existing RT tasks. | |
9944 | */ | |
9a7e0b18 PZ |
9945 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9946 | return -EBUSY; | |
6f505b16 | 9947 | |
9a7e0b18 | 9948 | total = to_ratio(period, runtime); |
6f505b16 | 9949 | |
4653f803 PZ |
9950 | /* |
9951 | * Nobody can have more than the global setting allows. | |
9952 | */ | |
9953 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9954 | return -EINVAL; | |
6f505b16 | 9955 | |
4653f803 PZ |
9956 | /* |
9957 | * The sum of our children's runtime should not exceed our own. | |
9958 | */ | |
9a7e0b18 PZ |
9959 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9960 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9961 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9962 | |
9a7e0b18 PZ |
9963 | if (child == d->tg) { |
9964 | period = d->rt_period; | |
9965 | runtime = d->rt_runtime; | |
9966 | } | |
6f505b16 | 9967 | |
9a7e0b18 | 9968 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9969 | } |
6f505b16 | 9970 | |
9a7e0b18 PZ |
9971 | if (sum > total) |
9972 | return -EINVAL; | |
9973 | ||
9974 | return 0; | |
6f505b16 PZ |
9975 | } |
9976 | ||
9a7e0b18 | 9977 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9978 | { |
9a7e0b18 PZ |
9979 | struct rt_schedulable_data data = { |
9980 | .tg = tg, | |
9981 | .rt_period = period, | |
9982 | .rt_runtime = runtime, | |
9983 | }; | |
9984 | ||
9985 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9986 | } |
9987 | ||
d0b27fa7 PZ |
9988 | static int tg_set_bandwidth(struct task_group *tg, |
9989 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9990 | { |
ac086bc2 | 9991 | int i, err = 0; |
9f0c1e56 | 9992 | |
9f0c1e56 | 9993 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9994 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9995 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9996 | if (err) | |
9f0c1e56 | 9997 | goto unlock; |
ac086bc2 PZ |
9998 | |
9999 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10000 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10001 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10002 | |
10003 | for_each_possible_cpu(i) { | |
10004 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10005 | ||
10006 | spin_lock(&rt_rq->rt_runtime_lock); | |
10007 | rt_rq->rt_runtime = rt_runtime; | |
10008 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10009 | } | |
10010 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10011 | unlock: |
521f1a24 | 10012 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10013 | mutex_unlock(&rt_constraints_mutex); |
10014 | ||
10015 | return err; | |
6f505b16 PZ |
10016 | } |
10017 | ||
d0b27fa7 PZ |
10018 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10019 | { | |
10020 | u64 rt_runtime, rt_period; | |
10021 | ||
10022 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10023 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10024 | if (rt_runtime_us < 0) | |
10025 | rt_runtime = RUNTIME_INF; | |
10026 | ||
10027 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10028 | } | |
10029 | ||
9f0c1e56 PZ |
10030 | long sched_group_rt_runtime(struct task_group *tg) |
10031 | { | |
10032 | u64 rt_runtime_us; | |
10033 | ||
d0b27fa7 | 10034 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10035 | return -1; |
10036 | ||
d0b27fa7 | 10037 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10038 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10039 | return rt_runtime_us; | |
10040 | } | |
d0b27fa7 PZ |
10041 | |
10042 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10043 | { | |
10044 | u64 rt_runtime, rt_period; | |
10045 | ||
10046 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10047 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10048 | ||
619b0488 R |
10049 | if (rt_period == 0) |
10050 | return -EINVAL; | |
10051 | ||
d0b27fa7 PZ |
10052 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10053 | } | |
10054 | ||
10055 | long sched_group_rt_period(struct task_group *tg) | |
10056 | { | |
10057 | u64 rt_period_us; | |
10058 | ||
10059 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10060 | do_div(rt_period_us, NSEC_PER_USEC); | |
10061 | return rt_period_us; | |
10062 | } | |
10063 | ||
10064 | static int sched_rt_global_constraints(void) | |
10065 | { | |
4653f803 | 10066 | u64 runtime, period; |
d0b27fa7 PZ |
10067 | int ret = 0; |
10068 | ||
ec5d4989 HS |
10069 | if (sysctl_sched_rt_period <= 0) |
10070 | return -EINVAL; | |
10071 | ||
4653f803 PZ |
10072 | runtime = global_rt_runtime(); |
10073 | period = global_rt_period(); | |
10074 | ||
10075 | /* | |
10076 | * Sanity check on the sysctl variables. | |
10077 | */ | |
10078 | if (runtime > period && runtime != RUNTIME_INF) | |
10079 | return -EINVAL; | |
10b612f4 | 10080 | |
d0b27fa7 | 10081 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10082 | read_lock(&tasklist_lock); |
4653f803 | 10083 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10084 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10085 | mutex_unlock(&rt_constraints_mutex); |
10086 | ||
10087 | return ret; | |
10088 | } | |
54e99124 DG |
10089 | |
10090 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10091 | { | |
10092 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10093 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10094 | return 0; | |
10095 | ||
10096 | return 1; | |
10097 | } | |
10098 | ||
6d6bc0ad | 10099 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10100 | static int sched_rt_global_constraints(void) |
10101 | { | |
ac086bc2 PZ |
10102 | unsigned long flags; |
10103 | int i; | |
10104 | ||
ec5d4989 HS |
10105 | if (sysctl_sched_rt_period <= 0) |
10106 | return -EINVAL; | |
10107 | ||
60aa605d PZ |
10108 | /* |
10109 | * There's always some RT tasks in the root group | |
10110 | * -- migration, kstopmachine etc.. | |
10111 | */ | |
10112 | if (sysctl_sched_rt_runtime == 0) | |
10113 | return -EBUSY; | |
10114 | ||
ac086bc2 PZ |
10115 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10116 | for_each_possible_cpu(i) { | |
10117 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10118 | ||
10119 | spin_lock(&rt_rq->rt_runtime_lock); | |
10120 | rt_rq->rt_runtime = global_rt_runtime(); | |
10121 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10122 | } | |
10123 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10124 | ||
d0b27fa7 PZ |
10125 | return 0; |
10126 | } | |
6d6bc0ad | 10127 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10128 | |
10129 | int sched_rt_handler(struct ctl_table *table, int write, | |
10130 | struct file *filp, void __user *buffer, size_t *lenp, | |
10131 | loff_t *ppos) | |
10132 | { | |
10133 | int ret; | |
10134 | int old_period, old_runtime; | |
10135 | static DEFINE_MUTEX(mutex); | |
10136 | ||
10137 | mutex_lock(&mutex); | |
10138 | old_period = sysctl_sched_rt_period; | |
10139 | old_runtime = sysctl_sched_rt_runtime; | |
10140 | ||
10141 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10142 | ||
10143 | if (!ret && write) { | |
10144 | ret = sched_rt_global_constraints(); | |
10145 | if (ret) { | |
10146 | sysctl_sched_rt_period = old_period; | |
10147 | sysctl_sched_rt_runtime = old_runtime; | |
10148 | } else { | |
10149 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10150 | def_rt_bandwidth.rt_period = | |
10151 | ns_to_ktime(global_rt_period()); | |
10152 | } | |
10153 | } | |
10154 | mutex_unlock(&mutex); | |
10155 | ||
10156 | return ret; | |
10157 | } | |
68318b8e | 10158 | |
052f1dc7 | 10159 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10160 | |
10161 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10162 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10163 | { |
2b01dfe3 PM |
10164 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10165 | struct task_group, css); | |
68318b8e SV |
10166 | } |
10167 | ||
10168 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10169 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10170 | { |
ec7dc8ac | 10171 | struct task_group *tg, *parent; |
68318b8e | 10172 | |
2b01dfe3 | 10173 | if (!cgrp->parent) { |
68318b8e | 10174 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10175 | return &init_task_group.css; |
10176 | } | |
10177 | ||
ec7dc8ac DG |
10178 | parent = cgroup_tg(cgrp->parent); |
10179 | tg = sched_create_group(parent); | |
68318b8e SV |
10180 | if (IS_ERR(tg)) |
10181 | return ERR_PTR(-ENOMEM); | |
10182 | ||
68318b8e SV |
10183 | return &tg->css; |
10184 | } | |
10185 | ||
41a2d6cf IM |
10186 | static void |
10187 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10188 | { |
2b01dfe3 | 10189 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10190 | |
10191 | sched_destroy_group(tg); | |
10192 | } | |
10193 | ||
41a2d6cf IM |
10194 | static int |
10195 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10196 | struct task_struct *tsk) | |
68318b8e | 10197 | { |
b68aa230 | 10198 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10199 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10200 | return -EINVAL; |
10201 | #else | |
68318b8e SV |
10202 | /* We don't support RT-tasks being in separate groups */ |
10203 | if (tsk->sched_class != &fair_sched_class) | |
10204 | return -EINVAL; | |
b68aa230 | 10205 | #endif |
68318b8e SV |
10206 | |
10207 | return 0; | |
10208 | } | |
10209 | ||
10210 | static void | |
2b01dfe3 | 10211 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10212 | struct cgroup *old_cont, struct task_struct *tsk) |
10213 | { | |
10214 | sched_move_task(tsk); | |
10215 | } | |
10216 | ||
052f1dc7 | 10217 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10218 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10219 | u64 shareval) |
68318b8e | 10220 | { |
2b01dfe3 | 10221 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10222 | } |
10223 | ||
f4c753b7 | 10224 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10225 | { |
2b01dfe3 | 10226 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10227 | |
10228 | return (u64) tg->shares; | |
10229 | } | |
6d6bc0ad | 10230 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10231 | |
052f1dc7 | 10232 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10233 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10234 | s64 val) |
6f505b16 | 10235 | { |
06ecb27c | 10236 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10237 | } |
10238 | ||
06ecb27c | 10239 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10240 | { |
06ecb27c | 10241 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10242 | } |
d0b27fa7 PZ |
10243 | |
10244 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10245 | u64 rt_period_us) | |
10246 | { | |
10247 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10248 | } | |
10249 | ||
10250 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10251 | { | |
10252 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10253 | } | |
6d6bc0ad | 10254 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10255 | |
fe5c7cc2 | 10256 | static struct cftype cpu_files[] = { |
052f1dc7 | 10257 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10258 | { |
10259 | .name = "shares", | |
f4c753b7 PM |
10260 | .read_u64 = cpu_shares_read_u64, |
10261 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10262 | }, |
052f1dc7 PZ |
10263 | #endif |
10264 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10265 | { |
9f0c1e56 | 10266 | .name = "rt_runtime_us", |
06ecb27c PM |
10267 | .read_s64 = cpu_rt_runtime_read, |
10268 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10269 | }, |
d0b27fa7 PZ |
10270 | { |
10271 | .name = "rt_period_us", | |
f4c753b7 PM |
10272 | .read_u64 = cpu_rt_period_read_uint, |
10273 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10274 | }, |
052f1dc7 | 10275 | #endif |
68318b8e SV |
10276 | }; |
10277 | ||
10278 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10279 | { | |
fe5c7cc2 | 10280 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10281 | } |
10282 | ||
10283 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10284 | .name = "cpu", |
10285 | .create = cpu_cgroup_create, | |
10286 | .destroy = cpu_cgroup_destroy, | |
10287 | .can_attach = cpu_cgroup_can_attach, | |
10288 | .attach = cpu_cgroup_attach, | |
10289 | .populate = cpu_cgroup_populate, | |
10290 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10291 | .early_init = 1, |
10292 | }; | |
10293 | ||
052f1dc7 | 10294 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10295 | |
10296 | #ifdef CONFIG_CGROUP_CPUACCT | |
10297 | ||
10298 | /* | |
10299 | * CPU accounting code for task groups. | |
10300 | * | |
10301 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10302 | * (balbir@in.ibm.com). | |
10303 | */ | |
10304 | ||
934352f2 | 10305 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10306 | struct cpuacct { |
10307 | struct cgroup_subsys_state css; | |
10308 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10309 | u64 *cpuusage; | |
ef12fefa | 10310 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10311 | struct cpuacct *parent; |
d842de87 SV |
10312 | }; |
10313 | ||
10314 | struct cgroup_subsys cpuacct_subsys; | |
10315 | ||
10316 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10317 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10318 | { |
32cd756a | 10319 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10320 | struct cpuacct, css); |
10321 | } | |
10322 | ||
10323 | /* return cpu accounting group to which this task belongs */ | |
10324 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10325 | { | |
10326 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10327 | struct cpuacct, css); | |
10328 | } | |
10329 | ||
10330 | /* create a new cpu accounting group */ | |
10331 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10332 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10333 | { |
10334 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10335 | int i; |
d842de87 SV |
10336 | |
10337 | if (!ca) | |
ef12fefa | 10338 | goto out; |
d842de87 SV |
10339 | |
10340 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10341 | if (!ca->cpuusage) |
10342 | goto out_free_ca; | |
10343 | ||
10344 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10345 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10346 | goto out_free_counters; | |
d842de87 | 10347 | |
934352f2 BR |
10348 | if (cgrp->parent) |
10349 | ca->parent = cgroup_ca(cgrp->parent); | |
10350 | ||
d842de87 | 10351 | return &ca->css; |
ef12fefa BR |
10352 | |
10353 | out_free_counters: | |
10354 | while (--i >= 0) | |
10355 | percpu_counter_destroy(&ca->cpustat[i]); | |
10356 | free_percpu(ca->cpuusage); | |
10357 | out_free_ca: | |
10358 | kfree(ca); | |
10359 | out: | |
10360 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10361 | } |
10362 | ||
10363 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10364 | static void |
32cd756a | 10365 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10366 | { |
32cd756a | 10367 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10368 | int i; |
d842de87 | 10369 | |
ef12fefa BR |
10370 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10371 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10372 | free_percpu(ca->cpuusage); |
10373 | kfree(ca); | |
10374 | } | |
10375 | ||
720f5498 KC |
10376 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10377 | { | |
b36128c8 | 10378 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10379 | u64 data; |
10380 | ||
10381 | #ifndef CONFIG_64BIT | |
10382 | /* | |
10383 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10384 | */ | |
10385 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10386 | data = *cpuusage; | |
10387 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10388 | #else | |
10389 | data = *cpuusage; | |
10390 | #endif | |
10391 | ||
10392 | return data; | |
10393 | } | |
10394 | ||
10395 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10396 | { | |
b36128c8 | 10397 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10398 | |
10399 | #ifndef CONFIG_64BIT | |
10400 | /* | |
10401 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10402 | */ | |
10403 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10404 | *cpuusage = val; | |
10405 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10406 | #else | |
10407 | *cpuusage = val; | |
10408 | #endif | |
10409 | } | |
10410 | ||
d842de87 | 10411 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10412 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10413 | { |
32cd756a | 10414 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10415 | u64 totalcpuusage = 0; |
10416 | int i; | |
10417 | ||
720f5498 KC |
10418 | for_each_present_cpu(i) |
10419 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10420 | |
10421 | return totalcpuusage; | |
10422 | } | |
10423 | ||
0297b803 DG |
10424 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10425 | u64 reset) | |
10426 | { | |
10427 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10428 | int err = 0; | |
10429 | int i; | |
10430 | ||
10431 | if (reset) { | |
10432 | err = -EINVAL; | |
10433 | goto out; | |
10434 | } | |
10435 | ||
720f5498 KC |
10436 | for_each_present_cpu(i) |
10437 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10438 | |
0297b803 DG |
10439 | out: |
10440 | return err; | |
10441 | } | |
10442 | ||
e9515c3c KC |
10443 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10444 | struct seq_file *m) | |
10445 | { | |
10446 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10447 | u64 percpu; | |
10448 | int i; | |
10449 | ||
10450 | for_each_present_cpu(i) { | |
10451 | percpu = cpuacct_cpuusage_read(ca, i); | |
10452 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10453 | } | |
10454 | seq_printf(m, "\n"); | |
10455 | return 0; | |
10456 | } | |
10457 | ||
ef12fefa BR |
10458 | static const char *cpuacct_stat_desc[] = { |
10459 | [CPUACCT_STAT_USER] = "user", | |
10460 | [CPUACCT_STAT_SYSTEM] = "system", | |
10461 | }; | |
10462 | ||
10463 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10464 | struct cgroup_map_cb *cb) | |
10465 | { | |
10466 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10467 | int i; | |
10468 | ||
10469 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10470 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10471 | val = cputime64_to_clock_t(val); | |
10472 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10473 | } | |
10474 | return 0; | |
10475 | } | |
10476 | ||
d842de87 SV |
10477 | static struct cftype files[] = { |
10478 | { | |
10479 | .name = "usage", | |
f4c753b7 PM |
10480 | .read_u64 = cpuusage_read, |
10481 | .write_u64 = cpuusage_write, | |
d842de87 | 10482 | }, |
e9515c3c KC |
10483 | { |
10484 | .name = "usage_percpu", | |
10485 | .read_seq_string = cpuacct_percpu_seq_read, | |
10486 | }, | |
ef12fefa BR |
10487 | { |
10488 | .name = "stat", | |
10489 | .read_map = cpuacct_stats_show, | |
10490 | }, | |
d842de87 SV |
10491 | }; |
10492 | ||
32cd756a | 10493 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10494 | { |
32cd756a | 10495 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10496 | } |
10497 | ||
10498 | /* | |
10499 | * charge this task's execution time to its accounting group. | |
10500 | * | |
10501 | * called with rq->lock held. | |
10502 | */ | |
10503 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10504 | { | |
10505 | struct cpuacct *ca; | |
934352f2 | 10506 | int cpu; |
d842de87 | 10507 | |
c40c6f85 | 10508 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10509 | return; |
10510 | ||
934352f2 | 10511 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10512 | |
10513 | rcu_read_lock(); | |
10514 | ||
d842de87 | 10515 | ca = task_ca(tsk); |
d842de87 | 10516 | |
934352f2 | 10517 | for (; ca; ca = ca->parent) { |
b36128c8 | 10518 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10519 | *cpuusage += cputime; |
10520 | } | |
a18b83b7 BR |
10521 | |
10522 | rcu_read_unlock(); | |
d842de87 SV |
10523 | } |
10524 | ||
ef12fefa BR |
10525 | /* |
10526 | * Charge the system/user time to the task's accounting group. | |
10527 | */ | |
10528 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10529 | enum cpuacct_stat_index idx, cputime_t val) | |
10530 | { | |
10531 | struct cpuacct *ca; | |
10532 | ||
10533 | if (unlikely(!cpuacct_subsys.active)) | |
10534 | return; | |
10535 | ||
10536 | rcu_read_lock(); | |
10537 | ca = task_ca(tsk); | |
10538 | ||
10539 | do { | |
10540 | percpu_counter_add(&ca->cpustat[idx], val); | |
10541 | ca = ca->parent; | |
10542 | } while (ca); | |
10543 | rcu_read_unlock(); | |
10544 | } | |
10545 | ||
d842de87 SV |
10546 | struct cgroup_subsys cpuacct_subsys = { |
10547 | .name = "cpuacct", | |
10548 | .create = cpuacct_create, | |
10549 | .destroy = cpuacct_destroy, | |
10550 | .populate = cpuacct_populate, | |
10551 | .subsys_id = cpuacct_subsys_id, | |
10552 | }; | |
10553 | #endif /* CONFIG_CGROUP_CPUACCT */ |