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, | |
5c333864 | 243 | HRTIMER_MODE_ABS_PINNED, 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 | 1157 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1158 | HRTIMER_MODE_REL_PINNED, 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 | } | |
b43e3521 | 2195 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2196 | |
2197 | /* | |
2dd73a4f PW |
2198 | * Return a low guess at the load of a migration-source cpu weighted |
2199 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2200 | * |
2201 | * We want to under-estimate the load of migration sources, to | |
2202 | * balance conservatively. | |
2203 | */ | |
a9957449 | 2204 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2205 | { |
70b97a7f | 2206 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2207 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2208 | |
93b75217 | 2209 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2210 | return total; |
b910472d | 2211 | |
dd41f596 | 2212 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2213 | } |
2214 | ||
2215 | /* | |
2dd73a4f PW |
2216 | * Return a high guess at the load of a migration-target cpu weighted |
2217 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2218 | */ |
a9957449 | 2219 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2220 | { |
70b97a7f | 2221 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2222 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2223 | |
93b75217 | 2224 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2225 | return total; |
3b0bd9bc | 2226 | |
dd41f596 | 2227 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2228 | } |
2229 | ||
147cbb4b NP |
2230 | /* |
2231 | * find_idlest_group finds and returns the least busy CPU group within the | |
2232 | * domain. | |
2233 | */ | |
2234 | static struct sched_group * | |
2235 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2236 | { | |
2237 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2238 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2239 | int load_idx = sd->forkexec_idx; | |
2240 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2241 | ||
2242 | do { | |
2243 | unsigned long load, avg_load; | |
2244 | int local_group; | |
2245 | int i; | |
2246 | ||
da5a5522 | 2247 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2248 | if (!cpumask_intersects(sched_group_cpus(group), |
2249 | &p->cpus_allowed)) | |
3a5c359a | 2250 | continue; |
da5a5522 | 2251 | |
758b2cdc RR |
2252 | local_group = cpumask_test_cpu(this_cpu, |
2253 | sched_group_cpus(group)); | |
147cbb4b NP |
2254 | |
2255 | /* Tally up the load of all CPUs in the group */ | |
2256 | avg_load = 0; | |
2257 | ||
758b2cdc | 2258 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2259 | /* Bias balancing toward cpus of our domain */ |
2260 | if (local_group) | |
2261 | load = source_load(i, load_idx); | |
2262 | else | |
2263 | load = target_load(i, load_idx); | |
2264 | ||
2265 | avg_load += load; | |
2266 | } | |
2267 | ||
2268 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2269 | avg_load = sg_div_cpu_power(group, |
2270 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2271 | |
2272 | if (local_group) { | |
2273 | this_load = avg_load; | |
2274 | this = group; | |
2275 | } else if (avg_load < min_load) { | |
2276 | min_load = avg_load; | |
2277 | idlest = group; | |
2278 | } | |
3a5c359a | 2279 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2280 | |
2281 | if (!idlest || 100*this_load < imbalance*min_load) | |
2282 | return NULL; | |
2283 | return idlest; | |
2284 | } | |
2285 | ||
2286 | /* | |
0feaece9 | 2287 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2288 | */ |
95cdf3b7 | 2289 | static int |
758b2cdc | 2290 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2291 | { |
2292 | unsigned long load, min_load = ULONG_MAX; | |
2293 | int idlest = -1; | |
2294 | int i; | |
2295 | ||
da5a5522 | 2296 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2297 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2298 | load = weighted_cpuload(i); |
147cbb4b NP |
2299 | |
2300 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2301 | min_load = load; | |
2302 | idlest = i; | |
2303 | } | |
2304 | } | |
2305 | ||
2306 | return idlest; | |
2307 | } | |
2308 | ||
476d139c NP |
2309 | /* |
2310 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2311 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2312 | * SD_BALANCE_EXEC. | |
2313 | * | |
2314 | * Balance, ie. select the least loaded group. | |
2315 | * | |
2316 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2317 | * | |
2318 | * preempt must be disabled. | |
2319 | */ | |
2320 | static int sched_balance_self(int cpu, int flag) | |
2321 | { | |
2322 | struct task_struct *t = current; | |
2323 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2324 | |
c96d145e | 2325 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2326 | /* |
2327 | * If power savings logic is enabled for a domain, stop there. | |
2328 | */ | |
5c45bf27 SS |
2329 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2330 | break; | |
476d139c NP |
2331 | if (tmp->flags & flag) |
2332 | sd = tmp; | |
c96d145e | 2333 | } |
476d139c | 2334 | |
039a1c41 PZ |
2335 | if (sd) |
2336 | update_shares(sd); | |
2337 | ||
476d139c | 2338 | while (sd) { |
476d139c | 2339 | struct sched_group *group; |
1a848870 SS |
2340 | int new_cpu, weight; |
2341 | ||
2342 | if (!(sd->flags & flag)) { | |
2343 | sd = sd->child; | |
2344 | continue; | |
2345 | } | |
476d139c | 2346 | |
476d139c | 2347 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2348 | if (!group) { |
2349 | sd = sd->child; | |
2350 | continue; | |
2351 | } | |
476d139c | 2352 | |
758b2cdc | 2353 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2354 | if (new_cpu == -1 || new_cpu == cpu) { |
2355 | /* Now try balancing at a lower domain level of cpu */ | |
2356 | sd = sd->child; | |
2357 | continue; | |
2358 | } | |
476d139c | 2359 | |
1a848870 | 2360 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2361 | cpu = new_cpu; |
758b2cdc | 2362 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2363 | sd = NULL; |
476d139c | 2364 | for_each_domain(cpu, tmp) { |
758b2cdc | 2365 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2366 | break; |
2367 | if (tmp->flags & flag) | |
2368 | sd = tmp; | |
2369 | } | |
2370 | /* while loop will break here if sd == NULL */ | |
2371 | } | |
2372 | ||
2373 | return cpu; | |
2374 | } | |
2375 | ||
2376 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2377 | |
0793a61d TG |
2378 | /** |
2379 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2380 | * @p: the task to evaluate | |
2381 | * @func: the function to be called | |
2382 | * @info: the function call argument | |
2383 | * | |
2384 | * Calls the function @func when the task is currently running. This might | |
2385 | * be on the current CPU, which just calls the function directly | |
2386 | */ | |
2387 | void task_oncpu_function_call(struct task_struct *p, | |
2388 | void (*func) (void *info), void *info) | |
2389 | { | |
2390 | int cpu; | |
2391 | ||
2392 | preempt_disable(); | |
2393 | cpu = task_cpu(p); | |
2394 | if (task_curr(p)) | |
2395 | smp_call_function_single(cpu, func, info, 1); | |
2396 | preempt_enable(); | |
2397 | } | |
2398 | ||
1da177e4 LT |
2399 | /*** |
2400 | * try_to_wake_up - wake up a thread | |
2401 | * @p: the to-be-woken-up thread | |
2402 | * @state: the mask of task states that can be woken | |
2403 | * @sync: do a synchronous wakeup? | |
2404 | * | |
2405 | * Put it on the run-queue if it's not already there. The "current" | |
2406 | * thread is always on the run-queue (except when the actual | |
2407 | * re-schedule is in progress), and as such you're allowed to do | |
2408 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2409 | * runnable without the overhead of this. | |
2410 | * | |
2411 | * returns failure only if the task is already active. | |
2412 | */ | |
36c8b586 | 2413 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2414 | { |
cc367732 | 2415 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2416 | unsigned long flags; |
2417 | long old_state; | |
70b97a7f | 2418 | struct rq *rq; |
1da177e4 | 2419 | |
b85d0667 IM |
2420 | if (!sched_feat(SYNC_WAKEUPS)) |
2421 | sync = 0; | |
2422 | ||
2398f2c6 | 2423 | #ifdef CONFIG_SMP |
57310a98 | 2424 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2425 | struct sched_domain *sd; |
2426 | ||
2427 | this_cpu = raw_smp_processor_id(); | |
2428 | cpu = task_cpu(p); | |
2429 | ||
2430 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2431 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2432 | update_shares(sd); |
2433 | break; | |
2434 | } | |
2435 | } | |
2436 | } | |
2437 | #endif | |
2438 | ||
04e2f174 | 2439 | smp_wmb(); |
1da177e4 | 2440 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2441 | update_rq_clock(rq); |
1da177e4 LT |
2442 | old_state = p->state; |
2443 | if (!(old_state & state)) | |
2444 | goto out; | |
2445 | ||
dd41f596 | 2446 | if (p->se.on_rq) |
1da177e4 LT |
2447 | goto out_running; |
2448 | ||
2449 | cpu = task_cpu(p); | |
cc367732 | 2450 | orig_cpu = cpu; |
1da177e4 LT |
2451 | this_cpu = smp_processor_id(); |
2452 | ||
2453 | #ifdef CONFIG_SMP | |
2454 | if (unlikely(task_running(rq, p))) | |
2455 | goto out_activate; | |
2456 | ||
5d2f5a61 DA |
2457 | cpu = p->sched_class->select_task_rq(p, sync); |
2458 | if (cpu != orig_cpu) { | |
2459 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2460 | task_rq_unlock(rq, &flags); |
2461 | /* might preempt at this point */ | |
2462 | rq = task_rq_lock(p, &flags); | |
2463 | old_state = p->state; | |
2464 | if (!(old_state & state)) | |
2465 | goto out; | |
dd41f596 | 2466 | if (p->se.on_rq) |
1da177e4 LT |
2467 | goto out_running; |
2468 | ||
2469 | this_cpu = smp_processor_id(); | |
2470 | cpu = task_cpu(p); | |
2471 | } | |
2472 | ||
e7693a36 GH |
2473 | #ifdef CONFIG_SCHEDSTATS |
2474 | schedstat_inc(rq, ttwu_count); | |
2475 | if (cpu == this_cpu) | |
2476 | schedstat_inc(rq, ttwu_local); | |
2477 | else { | |
2478 | struct sched_domain *sd; | |
2479 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2480 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2481 | schedstat_inc(sd, ttwu_wake_remote); |
2482 | break; | |
2483 | } | |
2484 | } | |
2485 | } | |
6d6bc0ad | 2486 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2487 | |
1da177e4 LT |
2488 | out_activate: |
2489 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2490 | schedstat_inc(p, se.nr_wakeups); |
2491 | if (sync) | |
2492 | schedstat_inc(p, se.nr_wakeups_sync); | |
2493 | if (orig_cpu != cpu) | |
2494 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2495 | if (cpu == this_cpu) | |
2496 | schedstat_inc(p, se.nr_wakeups_local); | |
2497 | else | |
2498 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2499 | activate_task(rq, p, 1); |
1da177e4 LT |
2500 | success = 1; |
2501 | ||
831451ac PZ |
2502 | /* |
2503 | * Only attribute actual wakeups done by this task. | |
2504 | */ | |
2505 | if (!in_interrupt()) { | |
2506 | struct sched_entity *se = ¤t->se; | |
2507 | u64 sample = se->sum_exec_runtime; | |
2508 | ||
2509 | if (se->last_wakeup) | |
2510 | sample -= se->last_wakeup; | |
2511 | else | |
2512 | sample -= se->start_runtime; | |
2513 | update_avg(&se->avg_wakeup, sample); | |
2514 | ||
2515 | se->last_wakeup = se->sum_exec_runtime; | |
2516 | } | |
2517 | ||
1da177e4 | 2518 | out_running: |
468a15bb | 2519 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2520 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2521 | |
1da177e4 | 2522 | p->state = TASK_RUNNING; |
9a897c5a SR |
2523 | #ifdef CONFIG_SMP |
2524 | if (p->sched_class->task_wake_up) | |
2525 | p->sched_class->task_wake_up(rq, p); | |
2526 | #endif | |
1da177e4 LT |
2527 | out: |
2528 | task_rq_unlock(rq, &flags); | |
2529 | ||
2530 | return success; | |
2531 | } | |
2532 | ||
50fa610a DH |
2533 | /** |
2534 | * wake_up_process - Wake up a specific process | |
2535 | * @p: The process to be woken up. | |
2536 | * | |
2537 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2538 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2539 | * running. | |
2540 | * | |
2541 | * It may be assumed that this function implies a write memory barrier before | |
2542 | * changing the task state if and only if any tasks are woken up. | |
2543 | */ | |
7ad5b3a5 | 2544 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2545 | { |
d9514f6c | 2546 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2547 | } |
1da177e4 LT |
2548 | EXPORT_SYMBOL(wake_up_process); |
2549 | ||
7ad5b3a5 | 2550 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2551 | { |
2552 | return try_to_wake_up(p, state, 0); | |
2553 | } | |
2554 | ||
1da177e4 LT |
2555 | /* |
2556 | * Perform scheduler related setup for a newly forked process p. | |
2557 | * p is forked by current. | |
dd41f596 IM |
2558 | * |
2559 | * __sched_fork() is basic setup used by init_idle() too: | |
2560 | */ | |
2561 | static void __sched_fork(struct task_struct *p) | |
2562 | { | |
dd41f596 IM |
2563 | p->se.exec_start = 0; |
2564 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2565 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2566 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2567 | p->se.last_wakeup = 0; |
2568 | p->se.avg_overlap = 0; | |
831451ac PZ |
2569 | p->se.start_runtime = 0; |
2570 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2571 | |
2572 | #ifdef CONFIG_SCHEDSTATS | |
2573 | p->se.wait_start = 0; | |
dd41f596 IM |
2574 | p->se.sum_sleep_runtime = 0; |
2575 | p->se.sleep_start = 0; | |
dd41f596 IM |
2576 | p->se.block_start = 0; |
2577 | p->se.sleep_max = 0; | |
2578 | p->se.block_max = 0; | |
2579 | p->se.exec_max = 0; | |
eba1ed4b | 2580 | p->se.slice_max = 0; |
dd41f596 | 2581 | p->se.wait_max = 0; |
6cfb0d5d | 2582 | #endif |
476d139c | 2583 | |
fa717060 | 2584 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2585 | p->se.on_rq = 0; |
4a55bd5e | 2586 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2587 | |
e107be36 AK |
2588 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2589 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2590 | #endif | |
2591 | ||
1da177e4 LT |
2592 | /* |
2593 | * We mark the process as running here, but have not actually | |
2594 | * inserted it onto the runqueue yet. This guarantees that | |
2595 | * nobody will actually run it, and a signal or other external | |
2596 | * event cannot wake it up and insert it on the runqueue either. | |
2597 | */ | |
2598 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2599 | } |
2600 | ||
2601 | /* | |
2602 | * fork()/clone()-time setup: | |
2603 | */ | |
2604 | void sched_fork(struct task_struct *p, int clone_flags) | |
2605 | { | |
2606 | int cpu = get_cpu(); | |
2607 | ||
2608 | __sched_fork(p); | |
2609 | ||
2610 | #ifdef CONFIG_SMP | |
2611 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2612 | #endif | |
02e4bac2 | 2613 | set_task_cpu(p, cpu); |
b29739f9 IM |
2614 | |
2615 | /* | |
2616 | * Make sure we do not leak PI boosting priority to the child: | |
2617 | */ | |
2618 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2619 | if (!rt_prio(p->prio)) |
2620 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2621 | |
52f17b6c | 2622 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2623 | if (likely(sched_info_on())) |
52f17b6c | 2624 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2625 | #endif |
d6077cb8 | 2626 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2627 | p->oncpu = 0; |
2628 | #endif | |
1da177e4 | 2629 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2630 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2631 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2632 | #endif |
917b627d GH |
2633 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2634 | ||
476d139c | 2635 | put_cpu(); |
1da177e4 LT |
2636 | } |
2637 | ||
2638 | /* | |
2639 | * wake_up_new_task - wake up a newly created task for the first time. | |
2640 | * | |
2641 | * This function will do some initial scheduler statistics housekeeping | |
2642 | * that must be done for every newly created context, then puts the task | |
2643 | * on the runqueue and wakes it. | |
2644 | */ | |
7ad5b3a5 | 2645 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2646 | { |
2647 | unsigned long flags; | |
dd41f596 | 2648 | struct rq *rq; |
1da177e4 LT |
2649 | |
2650 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2651 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2652 | update_rq_clock(rq); |
1da177e4 LT |
2653 | |
2654 | p->prio = effective_prio(p); | |
2655 | ||
b9dca1e0 | 2656 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2657 | activate_task(rq, p, 0); |
1da177e4 | 2658 | } else { |
1da177e4 | 2659 | /* |
dd41f596 IM |
2660 | * Let the scheduling class do new task startup |
2661 | * management (if any): | |
1da177e4 | 2662 | */ |
ee0827d8 | 2663 | p->sched_class->task_new(rq, p); |
c09595f6 | 2664 | inc_nr_running(rq); |
1da177e4 | 2665 | } |
c71dd42d | 2666 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2667 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2668 | #ifdef CONFIG_SMP |
2669 | if (p->sched_class->task_wake_up) | |
2670 | p->sched_class->task_wake_up(rq, p); | |
2671 | #endif | |
dd41f596 | 2672 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2673 | } |
2674 | ||
e107be36 AK |
2675 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2676 | ||
2677 | /** | |
80dd99b3 | 2678 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2679 | * @notifier: notifier struct to register |
e107be36 AK |
2680 | */ |
2681 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2682 | { | |
2683 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2684 | } | |
2685 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2686 | ||
2687 | /** | |
2688 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2689 | * @notifier: notifier struct to unregister |
e107be36 AK |
2690 | * |
2691 | * This is safe to call from within a preemption notifier. | |
2692 | */ | |
2693 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2694 | { | |
2695 | hlist_del(¬ifier->link); | |
2696 | } | |
2697 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2698 | ||
2699 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2700 | { | |
2701 | struct preempt_notifier *notifier; | |
2702 | struct hlist_node *node; | |
2703 | ||
2704 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2705 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2706 | } | |
2707 | ||
2708 | static void | |
2709 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2710 | struct task_struct *next) | |
2711 | { | |
2712 | struct preempt_notifier *notifier; | |
2713 | struct hlist_node *node; | |
2714 | ||
2715 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2716 | notifier->ops->sched_out(notifier, next); | |
2717 | } | |
2718 | ||
6d6bc0ad | 2719 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2720 | |
2721 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2722 | { | |
2723 | } | |
2724 | ||
2725 | static void | |
2726 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2727 | struct task_struct *next) | |
2728 | { | |
2729 | } | |
2730 | ||
6d6bc0ad | 2731 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2732 | |
4866cde0 NP |
2733 | /** |
2734 | * prepare_task_switch - prepare to switch tasks | |
2735 | * @rq: the runqueue preparing to switch | |
421cee29 | 2736 | * @prev: the current task that is being switched out |
4866cde0 NP |
2737 | * @next: the task we are going to switch to. |
2738 | * | |
2739 | * This is called with the rq lock held and interrupts off. It must | |
2740 | * be paired with a subsequent finish_task_switch after the context | |
2741 | * switch. | |
2742 | * | |
2743 | * prepare_task_switch sets up locking and calls architecture specific | |
2744 | * hooks. | |
2745 | */ | |
e107be36 AK |
2746 | static inline void |
2747 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2748 | struct task_struct *next) | |
4866cde0 | 2749 | { |
e107be36 | 2750 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2751 | prepare_lock_switch(rq, next); |
2752 | prepare_arch_switch(next); | |
2753 | } | |
2754 | ||
1da177e4 LT |
2755 | /** |
2756 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2757 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2758 | * @prev: the thread we just switched away from. |
2759 | * | |
4866cde0 NP |
2760 | * finish_task_switch must be called after the context switch, paired |
2761 | * with a prepare_task_switch call before the context switch. | |
2762 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2763 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2764 | * |
2765 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2766 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2767 | * with the lock held can cause deadlocks; see schedule() for |
2768 | * details.) | |
2769 | */ | |
a9957449 | 2770 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2771 | __releases(rq->lock) |
2772 | { | |
1da177e4 | 2773 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2774 | long prev_state; |
967fc046 GH |
2775 | #ifdef CONFIG_SMP |
2776 | int post_schedule = 0; | |
2777 | ||
2778 | if (current->sched_class->needs_post_schedule) | |
2779 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2780 | #endif | |
1da177e4 LT |
2781 | |
2782 | rq->prev_mm = NULL; | |
2783 | ||
2784 | /* | |
2785 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2786 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2787 | * schedule one last time. The schedule call will never return, and |
2788 | * the scheduled task must drop that reference. | |
c394cc9f | 2789 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2790 | * still held, otherwise prev could be scheduled on another cpu, die |
2791 | * there before we look at prev->state, and then the reference would | |
2792 | * be dropped twice. | |
2793 | * Manfred Spraul <manfred@colorfullife.com> | |
2794 | */ | |
55a101f8 | 2795 | prev_state = prev->state; |
4866cde0 | 2796 | finish_arch_switch(prev); |
0793a61d | 2797 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2798 | finish_lock_switch(rq, prev); |
9a897c5a | 2799 | #ifdef CONFIG_SMP |
967fc046 | 2800 | if (post_schedule) |
9a897c5a SR |
2801 | current->sched_class->post_schedule(rq); |
2802 | #endif | |
e8fa1362 | 2803 | |
e107be36 | 2804 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2805 | if (mm) |
2806 | mmdrop(mm); | |
c394cc9f | 2807 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2808 | /* |
2809 | * Remove function-return probe instances associated with this | |
2810 | * task and put them back on the free list. | |
9761eea8 | 2811 | */ |
c6fd91f0 | 2812 | kprobe_flush_task(prev); |
1da177e4 | 2813 | put_task_struct(prev); |
c6fd91f0 | 2814 | } |
1da177e4 LT |
2815 | } |
2816 | ||
2817 | /** | |
2818 | * schedule_tail - first thing a freshly forked thread must call. | |
2819 | * @prev: the thread we just switched away from. | |
2820 | */ | |
36c8b586 | 2821 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2822 | __releases(rq->lock) |
2823 | { | |
70b97a7f IM |
2824 | struct rq *rq = this_rq(); |
2825 | ||
4866cde0 NP |
2826 | finish_task_switch(rq, prev); |
2827 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2828 | /* In this case, finish_task_switch does not reenable preemption */ | |
2829 | preempt_enable(); | |
2830 | #endif | |
1da177e4 | 2831 | if (current->set_child_tid) |
b488893a | 2832 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2833 | } |
2834 | ||
2835 | /* | |
2836 | * context_switch - switch to the new MM and the new | |
2837 | * thread's register state. | |
2838 | */ | |
dd41f596 | 2839 | static inline void |
70b97a7f | 2840 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2841 | struct task_struct *next) |
1da177e4 | 2842 | { |
dd41f596 | 2843 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2844 | |
e107be36 | 2845 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2846 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2847 | mm = next->mm; |
2848 | oldmm = prev->active_mm; | |
9226d125 ZA |
2849 | /* |
2850 | * For paravirt, this is coupled with an exit in switch_to to | |
2851 | * combine the page table reload and the switch backend into | |
2852 | * one hypercall. | |
2853 | */ | |
224101ed | 2854 | arch_start_context_switch(prev); |
9226d125 | 2855 | |
dd41f596 | 2856 | if (unlikely(!mm)) { |
1da177e4 LT |
2857 | next->active_mm = oldmm; |
2858 | atomic_inc(&oldmm->mm_count); | |
2859 | enter_lazy_tlb(oldmm, next); | |
2860 | } else | |
2861 | switch_mm(oldmm, mm, next); | |
2862 | ||
dd41f596 | 2863 | if (unlikely(!prev->mm)) { |
1da177e4 | 2864 | prev->active_mm = NULL; |
1da177e4 LT |
2865 | rq->prev_mm = oldmm; |
2866 | } | |
3a5f5e48 IM |
2867 | /* |
2868 | * Since the runqueue lock will be released by the next | |
2869 | * task (which is an invalid locking op but in the case | |
2870 | * of the scheduler it's an obvious special-case), so we | |
2871 | * do an early lockdep release here: | |
2872 | */ | |
2873 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2874 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2875 | #endif |
1da177e4 LT |
2876 | |
2877 | /* Here we just switch the register state and the stack. */ | |
2878 | switch_to(prev, next, prev); | |
2879 | ||
dd41f596 IM |
2880 | barrier(); |
2881 | /* | |
2882 | * this_rq must be evaluated again because prev may have moved | |
2883 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2884 | * frame will be invalid. | |
2885 | */ | |
2886 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2887 | } |
2888 | ||
2889 | /* | |
2890 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2891 | * | |
2892 | * externally visible scheduler statistics: current number of runnable | |
2893 | * threads, current number of uninterruptible-sleeping threads, total | |
2894 | * number of context switches performed since bootup. | |
2895 | */ | |
2896 | unsigned long nr_running(void) | |
2897 | { | |
2898 | unsigned long i, sum = 0; | |
2899 | ||
2900 | for_each_online_cpu(i) | |
2901 | sum += cpu_rq(i)->nr_running; | |
2902 | ||
2903 | return sum; | |
2904 | } | |
2905 | ||
2906 | unsigned long nr_uninterruptible(void) | |
2907 | { | |
2908 | unsigned long i, sum = 0; | |
2909 | ||
0a945022 | 2910 | for_each_possible_cpu(i) |
1da177e4 LT |
2911 | sum += cpu_rq(i)->nr_uninterruptible; |
2912 | ||
2913 | /* | |
2914 | * Since we read the counters lockless, it might be slightly | |
2915 | * inaccurate. Do not allow it to go below zero though: | |
2916 | */ | |
2917 | if (unlikely((long)sum < 0)) | |
2918 | sum = 0; | |
2919 | ||
2920 | return sum; | |
2921 | } | |
2922 | ||
2923 | unsigned long long nr_context_switches(void) | |
2924 | { | |
cc94abfc SR |
2925 | int i; |
2926 | unsigned long long sum = 0; | |
1da177e4 | 2927 | |
0a945022 | 2928 | for_each_possible_cpu(i) |
1da177e4 LT |
2929 | sum += cpu_rq(i)->nr_switches; |
2930 | ||
2931 | return sum; | |
2932 | } | |
2933 | ||
2934 | unsigned long nr_iowait(void) | |
2935 | { | |
2936 | unsigned long i, sum = 0; | |
2937 | ||
0a945022 | 2938 | for_each_possible_cpu(i) |
1da177e4 LT |
2939 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2940 | ||
2941 | return sum; | |
2942 | } | |
2943 | ||
dce48a84 TG |
2944 | /* Variables and functions for calc_load */ |
2945 | static atomic_long_t calc_load_tasks; | |
2946 | static unsigned long calc_load_update; | |
2947 | unsigned long avenrun[3]; | |
2948 | EXPORT_SYMBOL(avenrun); | |
2949 | ||
2d02494f TG |
2950 | /** |
2951 | * get_avenrun - get the load average array | |
2952 | * @loads: pointer to dest load array | |
2953 | * @offset: offset to add | |
2954 | * @shift: shift count to shift the result left | |
2955 | * | |
2956 | * These values are estimates at best, so no need for locking. | |
2957 | */ | |
2958 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2959 | { | |
2960 | loads[0] = (avenrun[0] + offset) << shift; | |
2961 | loads[1] = (avenrun[1] + offset) << shift; | |
2962 | loads[2] = (avenrun[2] + offset) << shift; | |
2963 | } | |
2964 | ||
dce48a84 TG |
2965 | static unsigned long |
2966 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2967 | { |
dce48a84 TG |
2968 | load *= exp; |
2969 | load += active * (FIXED_1 - exp); | |
2970 | return load >> FSHIFT; | |
2971 | } | |
db1b1fef | 2972 | |
dce48a84 TG |
2973 | /* |
2974 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2975 | * CPUs have updated calc_load_tasks. | |
2976 | */ | |
2977 | void calc_global_load(void) | |
2978 | { | |
2979 | unsigned long upd = calc_load_update + 10; | |
2980 | long active; | |
2981 | ||
2982 | if (time_before(jiffies, upd)) | |
2983 | return; | |
db1b1fef | 2984 | |
dce48a84 TG |
2985 | active = atomic_long_read(&calc_load_tasks); |
2986 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2987 | |
dce48a84 TG |
2988 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2989 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2990 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2991 | ||
2992 | calc_load_update += LOAD_FREQ; | |
2993 | } | |
2994 | ||
2995 | /* | |
2996 | * Either called from update_cpu_load() or from a cpu going idle | |
2997 | */ | |
2998 | static void calc_load_account_active(struct rq *this_rq) | |
2999 | { | |
3000 | long nr_active, delta; | |
3001 | ||
3002 | nr_active = this_rq->nr_running; | |
3003 | nr_active += (long) this_rq->nr_uninterruptible; | |
3004 | ||
3005 | if (nr_active != this_rq->calc_load_active) { | |
3006 | delta = nr_active - this_rq->calc_load_active; | |
3007 | this_rq->calc_load_active = nr_active; | |
3008 | atomic_long_add(delta, &calc_load_tasks); | |
3009 | } | |
db1b1fef JS |
3010 | } |
3011 | ||
23a185ca PM |
3012 | /* |
3013 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3014 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3015 | */ | |
23a185ca PM |
3016 | u64 cpu_nr_migrations(int cpu) |
3017 | { | |
3018 | return cpu_rq(cpu)->nr_migrations_in; | |
3019 | } | |
3020 | ||
48f24c4d | 3021 | /* |
dd41f596 IM |
3022 | * Update rq->cpu_load[] statistics. This function is usually called every |
3023 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3024 | */ |
dd41f596 | 3025 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3026 | { |
495eca49 | 3027 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3028 | int i, scale; |
3029 | ||
3030 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3031 | |
3032 | /* Update our load: */ | |
3033 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3034 | unsigned long old_load, new_load; | |
3035 | ||
3036 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3037 | ||
3038 | old_load = this_rq->cpu_load[i]; | |
3039 | new_load = this_load; | |
a25707f3 IM |
3040 | /* |
3041 | * Round up the averaging division if load is increasing. This | |
3042 | * prevents us from getting stuck on 9 if the load is 10, for | |
3043 | * example. | |
3044 | */ | |
3045 | if (new_load > old_load) | |
3046 | new_load += scale-1; | |
dd41f596 IM |
3047 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3048 | } | |
dce48a84 TG |
3049 | |
3050 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3051 | this_rq->calc_load_update += LOAD_FREQ; | |
3052 | calc_load_account_active(this_rq); | |
3053 | } | |
48f24c4d IM |
3054 | } |
3055 | ||
dd41f596 IM |
3056 | #ifdef CONFIG_SMP |
3057 | ||
1da177e4 LT |
3058 | /* |
3059 | * double_rq_lock - safely lock two runqueues | |
3060 | * | |
3061 | * Note this does not disable interrupts like task_rq_lock, | |
3062 | * you need to do so manually before calling. | |
3063 | */ | |
70b97a7f | 3064 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3065 | __acquires(rq1->lock) |
3066 | __acquires(rq2->lock) | |
3067 | { | |
054b9108 | 3068 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3069 | if (rq1 == rq2) { |
3070 | spin_lock(&rq1->lock); | |
3071 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3072 | } else { | |
c96d145e | 3073 | if (rq1 < rq2) { |
1da177e4 | 3074 | spin_lock(&rq1->lock); |
5e710e37 | 3075 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3076 | } else { |
3077 | spin_lock(&rq2->lock); | |
5e710e37 | 3078 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3079 | } |
3080 | } | |
6e82a3be IM |
3081 | update_rq_clock(rq1); |
3082 | update_rq_clock(rq2); | |
1da177e4 LT |
3083 | } |
3084 | ||
3085 | /* | |
3086 | * double_rq_unlock - safely unlock two runqueues | |
3087 | * | |
3088 | * Note this does not restore interrupts like task_rq_unlock, | |
3089 | * you need to do so manually after calling. | |
3090 | */ | |
70b97a7f | 3091 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3092 | __releases(rq1->lock) |
3093 | __releases(rq2->lock) | |
3094 | { | |
3095 | spin_unlock(&rq1->lock); | |
3096 | if (rq1 != rq2) | |
3097 | spin_unlock(&rq2->lock); | |
3098 | else | |
3099 | __release(rq2->lock); | |
3100 | } | |
3101 | ||
1da177e4 LT |
3102 | /* |
3103 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3104 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3105 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3106 | * the cpu_allowed mask is restored. |
3107 | */ | |
36c8b586 | 3108 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3109 | { |
70b97a7f | 3110 | struct migration_req req; |
1da177e4 | 3111 | unsigned long flags; |
70b97a7f | 3112 | struct rq *rq; |
1da177e4 LT |
3113 | |
3114 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3115 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3116 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3117 | goto out; |
3118 | ||
3119 | /* force the process onto the specified CPU */ | |
3120 | if (migrate_task(p, dest_cpu, &req)) { | |
3121 | /* Need to wait for migration thread (might exit: take ref). */ | |
3122 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3123 | |
1da177e4 LT |
3124 | get_task_struct(mt); |
3125 | task_rq_unlock(rq, &flags); | |
3126 | wake_up_process(mt); | |
3127 | put_task_struct(mt); | |
3128 | wait_for_completion(&req.done); | |
36c8b586 | 3129 | |
1da177e4 LT |
3130 | return; |
3131 | } | |
3132 | out: | |
3133 | task_rq_unlock(rq, &flags); | |
3134 | } | |
3135 | ||
3136 | /* | |
476d139c NP |
3137 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3138 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3139 | */ |
3140 | void sched_exec(void) | |
3141 | { | |
1da177e4 | 3142 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3143 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3144 | put_cpu(); |
476d139c NP |
3145 | if (new_cpu != this_cpu) |
3146 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3147 | } |
3148 | ||
3149 | /* | |
3150 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3151 | * Both runqueues must be locked. | |
3152 | */ | |
dd41f596 IM |
3153 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3154 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3155 | { |
2e1cb74a | 3156 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3157 | set_task_cpu(p, this_cpu); |
dd41f596 | 3158 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3159 | /* |
3160 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3161 | * to be always true for them. | |
3162 | */ | |
15afe09b | 3163 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3164 | } |
3165 | ||
3166 | /* | |
3167 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3168 | */ | |
858119e1 | 3169 | static |
70b97a7f | 3170 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3171 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3172 | int *all_pinned) |
1da177e4 | 3173 | { |
708dc512 | 3174 | int tsk_cache_hot = 0; |
1da177e4 LT |
3175 | /* |
3176 | * We do not migrate tasks that are: | |
3177 | * 1) running (obviously), or | |
3178 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3179 | * 3) are cache-hot on their current CPU. | |
3180 | */ | |
96f874e2 | 3181 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3182 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3183 | return 0; |
cc367732 | 3184 | } |
81026794 NP |
3185 | *all_pinned = 0; |
3186 | ||
cc367732 IM |
3187 | if (task_running(rq, p)) { |
3188 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3189 | return 0; |
cc367732 | 3190 | } |
1da177e4 | 3191 | |
da84d961 IM |
3192 | /* |
3193 | * Aggressive migration if: | |
3194 | * 1) task is cache cold, or | |
3195 | * 2) too many balance attempts have failed. | |
3196 | */ | |
3197 | ||
708dc512 LH |
3198 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3199 | if (!tsk_cache_hot || | |
3200 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3201 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3202 | if (tsk_cache_hot) { |
da84d961 | 3203 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3204 | schedstat_inc(p, se.nr_forced_migrations); |
3205 | } | |
da84d961 IM |
3206 | #endif |
3207 | return 1; | |
3208 | } | |
3209 | ||
708dc512 | 3210 | if (tsk_cache_hot) { |
cc367732 | 3211 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3212 | return 0; |
cc367732 | 3213 | } |
1da177e4 LT |
3214 | return 1; |
3215 | } | |
3216 | ||
e1d1484f PW |
3217 | static unsigned long |
3218 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3219 | unsigned long max_load_move, struct sched_domain *sd, | |
3220 | enum cpu_idle_type idle, int *all_pinned, | |
3221 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3222 | { |
051c6764 | 3223 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3224 | struct task_struct *p; |
3225 | long rem_load_move = max_load_move; | |
1da177e4 | 3226 | |
e1d1484f | 3227 | if (max_load_move == 0) |
1da177e4 LT |
3228 | goto out; |
3229 | ||
81026794 NP |
3230 | pinned = 1; |
3231 | ||
1da177e4 | 3232 | /* |
dd41f596 | 3233 | * Start the load-balancing iterator: |
1da177e4 | 3234 | */ |
dd41f596 IM |
3235 | p = iterator->start(iterator->arg); |
3236 | next: | |
b82d9fdd | 3237 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3238 | goto out; |
051c6764 PZ |
3239 | |
3240 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3241 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3242 | p = iterator->next(iterator->arg); |
3243 | goto next; | |
1da177e4 LT |
3244 | } |
3245 | ||
dd41f596 | 3246 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3247 | pulled++; |
dd41f596 | 3248 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3249 | |
7e96fa58 GH |
3250 | #ifdef CONFIG_PREEMPT |
3251 | /* | |
3252 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3253 | * will stop after the first task is pulled to minimize the critical | |
3254 | * section. | |
3255 | */ | |
3256 | if (idle == CPU_NEWLY_IDLE) | |
3257 | goto out; | |
3258 | #endif | |
3259 | ||
2dd73a4f | 3260 | /* |
b82d9fdd | 3261 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3262 | */ |
e1d1484f | 3263 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3264 | if (p->prio < *this_best_prio) |
3265 | *this_best_prio = p->prio; | |
dd41f596 IM |
3266 | p = iterator->next(iterator->arg); |
3267 | goto next; | |
1da177e4 LT |
3268 | } |
3269 | out: | |
3270 | /* | |
e1d1484f | 3271 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3272 | * so we can safely collect pull_task() stats here rather than |
3273 | * inside pull_task(). | |
3274 | */ | |
3275 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3276 | |
3277 | if (all_pinned) | |
3278 | *all_pinned = pinned; | |
e1d1484f PW |
3279 | |
3280 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3281 | } |
3282 | ||
dd41f596 | 3283 | /* |
43010659 PW |
3284 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3285 | * this_rq, as part of a balancing operation within domain "sd". | |
3286 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3287 | * |
3288 | * Called with both runqueues locked. | |
3289 | */ | |
3290 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3291 | unsigned long max_load_move, |
dd41f596 IM |
3292 | struct sched_domain *sd, enum cpu_idle_type idle, |
3293 | int *all_pinned) | |
3294 | { | |
5522d5d5 | 3295 | const struct sched_class *class = sched_class_highest; |
43010659 | 3296 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3297 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3298 | |
3299 | do { | |
43010659 PW |
3300 | total_load_moved += |
3301 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3302 | max_load_move - total_load_moved, |
a4ac01c3 | 3303 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3304 | class = class->next; |
c4acb2c0 | 3305 | |
7e96fa58 GH |
3306 | #ifdef CONFIG_PREEMPT |
3307 | /* | |
3308 | * NEWIDLE balancing is a source of latency, so preemptible | |
3309 | * kernels will stop after the first task is pulled to minimize | |
3310 | * the critical section. | |
3311 | */ | |
c4acb2c0 GH |
3312 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3313 | break; | |
7e96fa58 | 3314 | #endif |
43010659 | 3315 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3316 | |
43010659 PW |
3317 | return total_load_moved > 0; |
3318 | } | |
3319 | ||
e1d1484f PW |
3320 | static int |
3321 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3322 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3323 | struct rq_iterator *iterator) | |
3324 | { | |
3325 | struct task_struct *p = iterator->start(iterator->arg); | |
3326 | int pinned = 0; | |
3327 | ||
3328 | while (p) { | |
3329 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3330 | pull_task(busiest, p, this_rq, this_cpu); | |
3331 | /* | |
3332 | * Right now, this is only the second place pull_task() | |
3333 | * is called, so we can safely collect pull_task() | |
3334 | * stats here rather than inside pull_task(). | |
3335 | */ | |
3336 | schedstat_inc(sd, lb_gained[idle]); | |
3337 | ||
3338 | return 1; | |
3339 | } | |
3340 | p = iterator->next(iterator->arg); | |
3341 | } | |
3342 | ||
3343 | return 0; | |
3344 | } | |
3345 | ||
43010659 PW |
3346 | /* |
3347 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3348 | * part of active balancing operations within "domain". | |
3349 | * Returns 1 if successful and 0 otherwise. | |
3350 | * | |
3351 | * Called with both runqueues locked. | |
3352 | */ | |
3353 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3354 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3355 | { | |
5522d5d5 | 3356 | const struct sched_class *class; |
43010659 PW |
3357 | |
3358 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3359 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3360 | return 1; |
3361 | ||
3362 | return 0; | |
dd41f596 | 3363 | } |
67bb6c03 | 3364 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3365 | /* |
222d656d GS |
3366 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3367 | * during load balancing. | |
1da177e4 | 3368 | */ |
222d656d GS |
3369 | struct sd_lb_stats { |
3370 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3371 | struct sched_group *this; /* Local group in this sd */ | |
3372 | unsigned long total_load; /* Total load of all groups in sd */ | |
3373 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3374 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3375 | ||
3376 | /** Statistics of this group */ | |
3377 | unsigned long this_load; | |
3378 | unsigned long this_load_per_task; | |
3379 | unsigned long this_nr_running; | |
3380 | ||
3381 | /* Statistics of the busiest group */ | |
3382 | unsigned long max_load; | |
3383 | unsigned long busiest_load_per_task; | |
3384 | unsigned long busiest_nr_running; | |
3385 | ||
3386 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3387 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3388 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3389 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3390 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3391 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3392 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3393 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3394 | #endif |
222d656d | 3395 | }; |
1da177e4 | 3396 | |
d5ac537e | 3397 | /* |
381be78f GS |
3398 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3399 | */ | |
3400 | struct sg_lb_stats { | |
3401 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3402 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3403 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3404 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3405 | unsigned long group_capacity; | |
3406 | int group_imb; /* Is there an imbalance in the group ? */ | |
3407 | }; | |
408ed066 | 3408 | |
67bb6c03 GS |
3409 | /** |
3410 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3411 | * @group: The group whose first cpu is to be returned. | |
3412 | */ | |
3413 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3414 | { | |
3415 | return cpumask_first(sched_group_cpus(group)); | |
3416 | } | |
3417 | ||
3418 | /** | |
3419 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3420 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3421 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3422 | */ | |
3423 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3424 | enum cpu_idle_type idle) | |
3425 | { | |
3426 | int load_idx; | |
3427 | ||
3428 | switch (idle) { | |
3429 | case CPU_NOT_IDLE: | |
7897986b | 3430 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3431 | break; |
3432 | ||
3433 | case CPU_NEWLY_IDLE: | |
7897986b | 3434 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3435 | break; |
3436 | default: | |
7897986b | 3437 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3438 | break; |
3439 | } | |
1da177e4 | 3440 | |
67bb6c03 GS |
3441 | return load_idx; |
3442 | } | |
1da177e4 | 3443 | |
1da177e4 | 3444 | |
c071df18 GS |
3445 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3446 | /** | |
3447 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3448 | * the given sched_domain, during load balancing. | |
3449 | * | |
3450 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3451 | * @sds: Variable containing the statistics for sd. | |
3452 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3453 | */ | |
3454 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3455 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3456 | { | |
3457 | /* | |
3458 | * Busy processors will not participate in power savings | |
3459 | * balance. | |
3460 | */ | |
3461 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3462 | sds->power_savings_balance = 0; | |
3463 | else { | |
3464 | sds->power_savings_balance = 1; | |
3465 | sds->min_nr_running = ULONG_MAX; | |
3466 | sds->leader_nr_running = 0; | |
3467 | } | |
3468 | } | |
783609c6 | 3469 | |
c071df18 GS |
3470 | /** |
3471 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3472 | * sched_domain while performing load balancing. | |
3473 | * | |
3474 | * @group: sched_group belonging to the sched_domain under consideration. | |
3475 | * @sds: Variable containing the statistics of the sched_domain | |
3476 | * @local_group: Does group contain the CPU for which we're performing | |
3477 | * load balancing ? | |
3478 | * @sgs: Variable containing the statistics of the group. | |
3479 | */ | |
3480 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3481 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3482 | { | |
408ed066 | 3483 | |
c071df18 GS |
3484 | if (!sds->power_savings_balance) |
3485 | return; | |
1da177e4 | 3486 | |
c071df18 GS |
3487 | /* |
3488 | * If the local group is idle or completely loaded | |
3489 | * no need to do power savings balance at this domain | |
3490 | */ | |
3491 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3492 | !sds->this_nr_running)) | |
3493 | sds->power_savings_balance = 0; | |
2dd73a4f | 3494 | |
c071df18 GS |
3495 | /* |
3496 | * If a group is already running at full capacity or idle, | |
3497 | * don't include that group in power savings calculations | |
3498 | */ | |
3499 | if (!sds->power_savings_balance || | |
3500 | sgs->sum_nr_running >= sgs->group_capacity || | |
3501 | !sgs->sum_nr_running) | |
3502 | return; | |
5969fe06 | 3503 | |
c071df18 GS |
3504 | /* |
3505 | * Calculate the group which has the least non-idle load. | |
3506 | * This is the group from where we need to pick up the load | |
3507 | * for saving power | |
3508 | */ | |
3509 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3510 | (sgs->sum_nr_running == sds->min_nr_running && | |
3511 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3512 | sds->group_min = group; | |
3513 | sds->min_nr_running = sgs->sum_nr_running; | |
3514 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3515 | sgs->sum_nr_running; | |
3516 | } | |
783609c6 | 3517 | |
c071df18 GS |
3518 | /* |
3519 | * Calculate the group which is almost near its | |
3520 | * capacity but still has some space to pick up some load | |
3521 | * from other group and save more power | |
3522 | */ | |
3523 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3524 | return; | |
1da177e4 | 3525 | |
c071df18 GS |
3526 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3527 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3528 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3529 | sds->group_leader = group; | |
3530 | sds->leader_nr_running = sgs->sum_nr_running; | |
3531 | } | |
3532 | } | |
408ed066 | 3533 | |
c071df18 | 3534 | /** |
d5ac537e | 3535 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3536 | * @sds: Variable containing the statistics of the sched_domain |
3537 | * under consideration. | |
3538 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3539 | * @imbalance: Variable to store the imbalance. | |
3540 | * | |
d5ac537e RD |
3541 | * Description: |
3542 | * Check if we have potential to perform some power-savings balance. | |
3543 | * If yes, set the busiest group to be the least loaded group in the | |
3544 | * sched_domain, so that it's CPUs can be put to idle. | |
3545 | * | |
c071df18 GS |
3546 | * Returns 1 if there is potential to perform power-savings balance. |
3547 | * Else returns 0. | |
3548 | */ | |
3549 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3550 | int this_cpu, unsigned long *imbalance) | |
3551 | { | |
3552 | if (!sds->power_savings_balance) | |
3553 | return 0; | |
1da177e4 | 3554 | |
c071df18 GS |
3555 | if (sds->this != sds->group_leader || |
3556 | sds->group_leader == sds->group_min) | |
3557 | return 0; | |
783609c6 | 3558 | |
c071df18 GS |
3559 | *imbalance = sds->min_load_per_task; |
3560 | sds->busiest = sds->group_min; | |
1da177e4 | 3561 | |
c071df18 GS |
3562 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3563 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3564 | group_first_cpu(sds->group_leader); | |
3565 | } | |
3566 | ||
3567 | return 1; | |
1da177e4 | 3568 | |
c071df18 GS |
3569 | } |
3570 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3571 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3572 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3573 | { | |
3574 | return; | |
3575 | } | |
408ed066 | 3576 | |
c071df18 GS |
3577 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3578 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3579 | { | |
3580 | return; | |
3581 | } | |
3582 | ||
3583 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3584 | int this_cpu, unsigned long *imbalance) | |
3585 | { | |
3586 | return 0; | |
3587 | } | |
3588 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3589 | ||
3590 | ||
1f8c553d GS |
3591 | /** |
3592 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3593 | * @group: sched_group whose statistics are to be updated. | |
3594 | * @this_cpu: Cpu for which load balance is currently performed. | |
3595 | * @idle: Idle status of this_cpu | |
3596 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3597 | * @sd_idle: Idle status of the sched_domain containing group. | |
3598 | * @local_group: Does group contain this_cpu. | |
3599 | * @cpus: Set of cpus considered for load balancing. | |
3600 | * @balance: Should we balance. | |
3601 | * @sgs: variable to hold the statistics for this group. | |
3602 | */ | |
3603 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3604 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3605 | int local_group, const struct cpumask *cpus, | |
3606 | int *balance, struct sg_lb_stats *sgs) | |
3607 | { | |
3608 | unsigned long load, max_cpu_load, min_cpu_load; | |
3609 | int i; | |
3610 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3611 | unsigned long sum_avg_load_per_task; | |
3612 | unsigned long avg_load_per_task; | |
3613 | ||
3614 | if (local_group) | |
3615 | balance_cpu = group_first_cpu(group); | |
3616 | ||
3617 | /* Tally up the load of all CPUs in the group */ | |
3618 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3619 | max_cpu_load = 0; | |
3620 | min_cpu_load = ~0UL; | |
408ed066 | 3621 | |
1f8c553d GS |
3622 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3623 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3624 | |
1f8c553d GS |
3625 | if (*sd_idle && rq->nr_running) |
3626 | *sd_idle = 0; | |
5c45bf27 | 3627 | |
1f8c553d | 3628 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3629 | if (local_group) { |
1f8c553d GS |
3630 | if (idle_cpu(i) && !first_idle_cpu) { |
3631 | first_idle_cpu = 1; | |
3632 | balance_cpu = i; | |
3633 | } | |
3634 | ||
3635 | load = target_load(i, load_idx); | |
3636 | } else { | |
3637 | load = source_load(i, load_idx); | |
3638 | if (load > max_cpu_load) | |
3639 | max_cpu_load = load; | |
3640 | if (min_cpu_load > load) | |
3641 | min_cpu_load = load; | |
1da177e4 | 3642 | } |
5c45bf27 | 3643 | |
1f8c553d GS |
3644 | sgs->group_load += load; |
3645 | sgs->sum_nr_running += rq->nr_running; | |
3646 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3647 | |
1f8c553d GS |
3648 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3649 | } | |
5c45bf27 | 3650 | |
1f8c553d GS |
3651 | /* |
3652 | * First idle cpu or the first cpu(busiest) in this sched group | |
3653 | * is eligible for doing load balancing at this and above | |
3654 | * domains. In the newly idle case, we will allow all the cpu's | |
3655 | * to do the newly idle load balance. | |
3656 | */ | |
3657 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3658 | balance_cpu != this_cpu && balance) { | |
3659 | *balance = 0; | |
3660 | return; | |
3661 | } | |
5c45bf27 | 3662 | |
1f8c553d GS |
3663 | /* Adjust by relative CPU power of the group */ |
3664 | sgs->avg_load = sg_div_cpu_power(group, | |
3665 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3666 | |
1f8c553d GS |
3667 | |
3668 | /* | |
3669 | * Consider the group unbalanced when the imbalance is larger | |
3670 | * than the average weight of two tasks. | |
3671 | * | |
3672 | * APZ: with cgroup the avg task weight can vary wildly and | |
3673 | * might not be a suitable number - should we keep a | |
3674 | * normalized nr_running number somewhere that negates | |
3675 | * the hierarchy? | |
3676 | */ | |
3677 | avg_load_per_task = sg_div_cpu_power(group, | |
3678 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3679 | ||
3680 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3681 | sgs->group_imb = 1; | |
3682 | ||
3683 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3684 | ||
3685 | } | |
dd41f596 | 3686 | |
37abe198 GS |
3687 | /** |
3688 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3689 | * @sd: sched_domain whose statistics are to be updated. | |
3690 | * @this_cpu: Cpu for which load balance is currently performed. | |
3691 | * @idle: Idle status of this_cpu | |
3692 | * @sd_idle: Idle status of the sched_domain containing group. | |
3693 | * @cpus: Set of cpus considered for load balancing. | |
3694 | * @balance: Should we balance. | |
3695 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3696 | */ |
37abe198 GS |
3697 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3698 | enum cpu_idle_type idle, int *sd_idle, | |
3699 | const struct cpumask *cpus, int *balance, | |
3700 | struct sd_lb_stats *sds) | |
1da177e4 | 3701 | { |
222d656d | 3702 | struct sched_group *group = sd->groups; |
37abe198 | 3703 | struct sg_lb_stats sgs; |
222d656d GS |
3704 | int load_idx; |
3705 | ||
c071df18 | 3706 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3707 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3708 | |
3709 | do { | |
1da177e4 | 3710 | int local_group; |
1da177e4 | 3711 | |
758b2cdc RR |
3712 | local_group = cpumask_test_cpu(this_cpu, |
3713 | sched_group_cpus(group)); | |
381be78f | 3714 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3715 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3716 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3717 | |
37abe198 GS |
3718 | if (local_group && balance && !(*balance)) |
3719 | return; | |
783609c6 | 3720 | |
37abe198 GS |
3721 | sds->total_load += sgs.group_load; |
3722 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3723 | |
1da177e4 | 3724 | if (local_group) { |
37abe198 GS |
3725 | sds->this_load = sgs.avg_load; |
3726 | sds->this = group; | |
3727 | sds->this_nr_running = sgs.sum_nr_running; | |
3728 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3729 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3730 | (sgs.sum_nr_running > sgs.group_capacity || |
3731 | sgs.group_imb)) { | |
37abe198 GS |
3732 | sds->max_load = sgs.avg_load; |
3733 | sds->busiest = group; | |
3734 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3735 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3736 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3737 | } |
5c45bf27 | 3738 | |
c071df18 | 3739 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3740 | group = group->next; |
3741 | } while (group != sd->groups); | |
3742 | ||
37abe198 | 3743 | } |
1da177e4 | 3744 | |
2e6f44ae GS |
3745 | /** |
3746 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3747 | * amongst the groups of a sched_domain, during |
3748 | * load balancing. | |
2e6f44ae GS |
3749 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3750 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3751 | * @imbalance: Variable to store the imbalance. | |
3752 | */ | |
3753 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3754 | int this_cpu, unsigned long *imbalance) | |
3755 | { | |
3756 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3757 | unsigned int imbn = 2; | |
3758 | ||
3759 | if (sds->this_nr_running) { | |
3760 | sds->this_load_per_task /= sds->this_nr_running; | |
3761 | if (sds->busiest_load_per_task > | |
3762 | sds->this_load_per_task) | |
3763 | imbn = 1; | |
3764 | } else | |
3765 | sds->this_load_per_task = | |
3766 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3767 | |
2e6f44ae GS |
3768 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3769 | sds->busiest_load_per_task * imbn) { | |
3770 | *imbalance = sds->busiest_load_per_task; | |
3771 | return; | |
3772 | } | |
908a7c1b | 3773 | |
1da177e4 | 3774 | /* |
2e6f44ae GS |
3775 | * OK, we don't have enough imbalance to justify moving tasks, |
3776 | * however we may be able to increase total CPU power used by | |
3777 | * moving them. | |
1da177e4 | 3778 | */ |
2dd73a4f | 3779 | |
2e6f44ae GS |
3780 | pwr_now += sds->busiest->__cpu_power * |
3781 | min(sds->busiest_load_per_task, sds->max_load); | |
3782 | pwr_now += sds->this->__cpu_power * | |
3783 | min(sds->this_load_per_task, sds->this_load); | |
3784 | pwr_now /= SCHED_LOAD_SCALE; | |
3785 | ||
3786 | /* Amount of load we'd subtract */ | |
3787 | tmp = sg_div_cpu_power(sds->busiest, | |
3788 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3789 | if (sds->max_load > tmp) | |
3790 | pwr_move += sds->busiest->__cpu_power * | |
3791 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3792 | ||
3793 | /* Amount of load we'd add */ | |
3794 | if (sds->max_load * sds->busiest->__cpu_power < | |
3795 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3796 | tmp = sg_div_cpu_power(sds->this, | |
3797 | sds->max_load * sds->busiest->__cpu_power); | |
3798 | else | |
3799 | tmp = sg_div_cpu_power(sds->this, | |
3800 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3801 | pwr_move += sds->this->__cpu_power * | |
3802 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3803 | pwr_move /= SCHED_LOAD_SCALE; | |
3804 | ||
3805 | /* Move if we gain throughput */ | |
3806 | if (pwr_move > pwr_now) | |
3807 | *imbalance = sds->busiest_load_per_task; | |
3808 | } | |
dbc523a3 GS |
3809 | |
3810 | /** | |
3811 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3812 | * groups of a given sched_domain during load balance. | |
3813 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3814 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3815 | * @imbalance: The variable to store the imbalance. | |
3816 | */ | |
3817 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3818 | unsigned long *imbalance) | |
3819 | { | |
3820 | unsigned long max_pull; | |
2dd73a4f PW |
3821 | /* |
3822 | * In the presence of smp nice balancing, certain scenarios can have | |
3823 | * max load less than avg load(as we skip the groups at or below | |
3824 | * its cpu_power, while calculating max_load..) | |
3825 | */ | |
dbc523a3 | 3826 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3827 | *imbalance = 0; |
dbc523a3 | 3828 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3829 | } |
0c117f1b SS |
3830 | |
3831 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3832 | max_pull = min(sds->max_load - sds->avg_load, |
3833 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3834 | |
1da177e4 | 3835 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3836 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3837 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3838 | / SCHED_LOAD_SCALE; |
3839 | ||
2dd73a4f PW |
3840 | /* |
3841 | * if *imbalance is less than the average load per runnable task | |
3842 | * there is no gaurantee that any tasks will be moved so we'll have | |
3843 | * a think about bumping its value to force at least one task to be | |
3844 | * moved | |
3845 | */ | |
dbc523a3 GS |
3846 | if (*imbalance < sds->busiest_load_per_task) |
3847 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3848 | |
dbc523a3 | 3849 | } |
37abe198 | 3850 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3851 | |
b7bb4c9b GS |
3852 | /** |
3853 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3854 | * if there is an imbalance. If there isn't an imbalance, and | |
3855 | * the user has opted for power-savings, it returns a group whose | |
3856 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3857 | * such a group exists. | |
3858 | * | |
3859 | * Also calculates the amount of weighted load which should be moved | |
3860 | * to restore balance. | |
3861 | * | |
3862 | * @sd: The sched_domain whose busiest group is to be returned. | |
3863 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3864 | * @imbalance: Variable which stores amount of weighted load which should | |
3865 | * be moved to restore balance/put a group to idle. | |
3866 | * @idle: The idle status of this_cpu. | |
3867 | * @sd_idle: The idleness of sd | |
3868 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3869 | * @balance: Pointer to a variable indicating if this_cpu | |
3870 | * is the appropriate cpu to perform load balancing at this_level. | |
3871 | * | |
3872 | * Returns: - the busiest group if imbalance exists. | |
3873 | * - If no imbalance and user has opted for power-savings balance, | |
3874 | * return the least loaded group whose CPUs can be | |
3875 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3876 | */ |
3877 | static struct sched_group * | |
3878 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3879 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3880 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3881 | { | |
3882 | struct sd_lb_stats sds; | |
1da177e4 | 3883 | |
37abe198 | 3884 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3885 | |
37abe198 GS |
3886 | /* |
3887 | * Compute the various statistics relavent for load balancing at | |
3888 | * this level. | |
3889 | */ | |
3890 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3891 | balance, &sds); | |
3892 | ||
b7bb4c9b GS |
3893 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3894 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3895 | * at this level. | |
3896 | * 2) There is no busy sibling group to pull from. | |
3897 | * 3) This group is the busiest group. | |
3898 | * 4) This group is more busy than the avg busieness at this | |
3899 | * sched_domain. | |
3900 | * 5) The imbalance is within the specified limit. | |
3901 | * 6) Any rebalance would lead to ping-pong | |
3902 | */ | |
37abe198 GS |
3903 | if (balance && !(*balance)) |
3904 | goto ret; | |
1da177e4 | 3905 | |
b7bb4c9b GS |
3906 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3907 | goto out_balanced; | |
1da177e4 | 3908 | |
b7bb4c9b | 3909 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3910 | goto out_balanced; |
1da177e4 | 3911 | |
222d656d | 3912 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3913 | |
b7bb4c9b GS |
3914 | if (sds.this_load >= sds.avg_load) |
3915 | goto out_balanced; | |
3916 | ||
3917 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3918 | goto out_balanced; |
3919 | ||
222d656d GS |
3920 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3921 | if (sds.group_imb) | |
3922 | sds.busiest_load_per_task = | |
3923 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3924 | |
1da177e4 LT |
3925 | /* |
3926 | * We're trying to get all the cpus to the average_load, so we don't | |
3927 | * want to push ourselves above the average load, nor do we wish to | |
3928 | * reduce the max loaded cpu below the average load, as either of these | |
3929 | * actions would just result in more rebalancing later, and ping-pong | |
3930 | * tasks around. Thus we look for the minimum possible imbalance. | |
3931 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3932 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3933 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3934 | * appear as very large values with unsigned longs. |
3935 | */ | |
222d656d | 3936 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3937 | goto out_balanced; |
3938 | ||
dbc523a3 GS |
3939 | /* Looks like there is an imbalance. Compute it */ |
3940 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3941 | return sds.busiest; |
1da177e4 LT |
3942 | |
3943 | out_balanced: | |
c071df18 GS |
3944 | /* |
3945 | * There is no obvious imbalance. But check if we can do some balancing | |
3946 | * to save power. | |
3947 | */ | |
3948 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3949 | return sds.busiest; | |
783609c6 | 3950 | ret: |
1da177e4 LT |
3951 | *imbalance = 0; |
3952 | return NULL; | |
3953 | } | |
3954 | ||
3955 | /* | |
3956 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3957 | */ | |
70b97a7f | 3958 | static struct rq * |
d15bcfdb | 3959 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3960 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3961 | { |
70b97a7f | 3962 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3963 | unsigned long max_load = 0; |
1da177e4 LT |
3964 | int i; |
3965 | ||
758b2cdc | 3966 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3967 | unsigned long wl; |
0a2966b4 | 3968 | |
96f874e2 | 3969 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3970 | continue; |
3971 | ||
48f24c4d | 3972 | rq = cpu_rq(i); |
dd41f596 | 3973 | wl = weighted_cpuload(i); |
2dd73a4f | 3974 | |
dd41f596 | 3975 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3976 | continue; |
1da177e4 | 3977 | |
dd41f596 IM |
3978 | if (wl > max_load) { |
3979 | max_load = wl; | |
48f24c4d | 3980 | busiest = rq; |
1da177e4 LT |
3981 | } |
3982 | } | |
3983 | ||
3984 | return busiest; | |
3985 | } | |
3986 | ||
77391d71 NP |
3987 | /* |
3988 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3989 | * so long as it is large enough. | |
3990 | */ | |
3991 | #define MAX_PINNED_INTERVAL 512 | |
3992 | ||
df7c8e84 RR |
3993 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3994 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3995 | ||
1da177e4 LT |
3996 | /* |
3997 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3998 | * tasks if there is an imbalance. | |
1da177e4 | 3999 | */ |
70b97a7f | 4000 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4001 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4002 | int *balance) |
1da177e4 | 4003 | { |
43010659 | 4004 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4005 | struct sched_group *group; |
1da177e4 | 4006 | unsigned long imbalance; |
70b97a7f | 4007 | struct rq *busiest; |
fe2eea3f | 4008 | unsigned long flags; |
df7c8e84 | 4009 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4010 | |
96f874e2 | 4011 | cpumask_setall(cpus); |
7c16ec58 | 4012 | |
89c4710e SS |
4013 | /* |
4014 | * When power savings policy is enabled for the parent domain, idle | |
4015 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4016 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4017 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4018 | */ |
d15bcfdb | 4019 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4020 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4021 | sd_idle = 1; |
1da177e4 | 4022 | |
2d72376b | 4023 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4024 | |
0a2966b4 | 4025 | redo: |
c8cba857 | 4026 | update_shares(sd); |
0a2966b4 | 4027 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4028 | cpus, balance); |
783609c6 | 4029 | |
06066714 | 4030 | if (*balance == 0) |
783609c6 | 4031 | goto out_balanced; |
783609c6 | 4032 | |
1da177e4 LT |
4033 | if (!group) { |
4034 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4035 | goto out_balanced; | |
4036 | } | |
4037 | ||
7c16ec58 | 4038 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4039 | if (!busiest) { |
4040 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4041 | goto out_balanced; | |
4042 | } | |
4043 | ||
db935dbd | 4044 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4045 | |
4046 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4047 | ||
43010659 | 4048 | ld_moved = 0; |
1da177e4 LT |
4049 | if (busiest->nr_running > 1) { |
4050 | /* | |
4051 | * Attempt to move tasks. If find_busiest_group has found | |
4052 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4053 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4054 | * correctly treated as an imbalance. |
4055 | */ | |
fe2eea3f | 4056 | local_irq_save(flags); |
e17224bf | 4057 | double_rq_lock(this_rq, busiest); |
43010659 | 4058 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4059 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4060 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4061 | local_irq_restore(flags); |
81026794 | 4062 | |
46cb4b7c SS |
4063 | /* |
4064 | * some other cpu did the load balance for us. | |
4065 | */ | |
43010659 | 4066 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4067 | resched_cpu(this_cpu); |
4068 | ||
81026794 | 4069 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4070 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4071 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4072 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4073 | goto redo; |
81026794 | 4074 | goto out_balanced; |
0a2966b4 | 4075 | } |
1da177e4 | 4076 | } |
81026794 | 4077 | |
43010659 | 4078 | if (!ld_moved) { |
1da177e4 LT |
4079 | schedstat_inc(sd, lb_failed[idle]); |
4080 | sd->nr_balance_failed++; | |
4081 | ||
4082 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4083 | |
fe2eea3f | 4084 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4085 | |
4086 | /* don't kick the migration_thread, if the curr | |
4087 | * task on busiest cpu can't be moved to this_cpu | |
4088 | */ | |
96f874e2 RR |
4089 | if (!cpumask_test_cpu(this_cpu, |
4090 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4091 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4092 | all_pinned = 1; |
4093 | goto out_one_pinned; | |
4094 | } | |
4095 | ||
1da177e4 LT |
4096 | if (!busiest->active_balance) { |
4097 | busiest->active_balance = 1; | |
4098 | busiest->push_cpu = this_cpu; | |
81026794 | 4099 | active_balance = 1; |
1da177e4 | 4100 | } |
fe2eea3f | 4101 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4102 | if (active_balance) |
1da177e4 LT |
4103 | wake_up_process(busiest->migration_thread); |
4104 | ||
4105 | /* | |
4106 | * We've kicked active balancing, reset the failure | |
4107 | * counter. | |
4108 | */ | |
39507451 | 4109 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4110 | } |
81026794 | 4111 | } else |
1da177e4 LT |
4112 | sd->nr_balance_failed = 0; |
4113 | ||
81026794 | 4114 | if (likely(!active_balance)) { |
1da177e4 LT |
4115 | /* We were unbalanced, so reset the balancing interval */ |
4116 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4117 | } else { |
4118 | /* | |
4119 | * If we've begun active balancing, start to back off. This | |
4120 | * case may not be covered by the all_pinned logic if there | |
4121 | * is only 1 task on the busy runqueue (because we don't call | |
4122 | * move_tasks). | |
4123 | */ | |
4124 | if (sd->balance_interval < sd->max_interval) | |
4125 | sd->balance_interval *= 2; | |
1da177e4 LT |
4126 | } |
4127 | ||
43010659 | 4128 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4129 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4130 | ld_moved = -1; |
4131 | ||
4132 | goto out; | |
1da177e4 LT |
4133 | |
4134 | out_balanced: | |
1da177e4 LT |
4135 | schedstat_inc(sd, lb_balanced[idle]); |
4136 | ||
16cfb1c0 | 4137 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4138 | |
4139 | out_one_pinned: | |
1da177e4 | 4140 | /* tune up the balancing interval */ |
77391d71 NP |
4141 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4142 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4143 | sd->balance_interval *= 2; |
4144 | ||
48f24c4d | 4145 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4146 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4147 | ld_moved = -1; |
4148 | else | |
4149 | ld_moved = 0; | |
4150 | out: | |
c8cba857 PZ |
4151 | if (ld_moved) |
4152 | update_shares(sd); | |
c09595f6 | 4153 | return ld_moved; |
1da177e4 LT |
4154 | } |
4155 | ||
4156 | /* | |
4157 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4158 | * tasks if there is an imbalance. | |
4159 | * | |
d15bcfdb | 4160 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4161 | * this_rq is locked. |
4162 | */ | |
48f24c4d | 4163 | static int |
df7c8e84 | 4164 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4165 | { |
4166 | struct sched_group *group; | |
70b97a7f | 4167 | struct rq *busiest = NULL; |
1da177e4 | 4168 | unsigned long imbalance; |
43010659 | 4169 | int ld_moved = 0; |
5969fe06 | 4170 | int sd_idle = 0; |
969bb4e4 | 4171 | int all_pinned = 0; |
df7c8e84 | 4172 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4173 | |
96f874e2 | 4174 | cpumask_setall(cpus); |
5969fe06 | 4175 | |
89c4710e SS |
4176 | /* |
4177 | * When power savings policy is enabled for the parent domain, idle | |
4178 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4179 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4180 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4181 | */ |
4182 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4183 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4184 | sd_idle = 1; |
1da177e4 | 4185 | |
2d72376b | 4186 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4187 | redo: |
3e5459b4 | 4188 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4189 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4190 | &sd_idle, cpus, NULL); |
1da177e4 | 4191 | if (!group) { |
d15bcfdb | 4192 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4193 | goto out_balanced; |
1da177e4 LT |
4194 | } |
4195 | ||
7c16ec58 | 4196 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4197 | if (!busiest) { |
d15bcfdb | 4198 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4199 | goto out_balanced; |
1da177e4 LT |
4200 | } |
4201 | ||
db935dbd NP |
4202 | BUG_ON(busiest == this_rq); |
4203 | ||
d15bcfdb | 4204 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4205 | |
43010659 | 4206 | ld_moved = 0; |
d6d5cfaf NP |
4207 | if (busiest->nr_running > 1) { |
4208 | /* Attempt to move tasks */ | |
4209 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4210 | /* this_rq->clock is already updated */ |
4211 | update_rq_clock(busiest); | |
43010659 | 4212 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4213 | imbalance, sd, CPU_NEWLY_IDLE, |
4214 | &all_pinned); | |
1b12bbc7 | 4215 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4216 | |
969bb4e4 | 4217 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4218 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4219 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4220 | goto redo; |
4221 | } | |
d6d5cfaf NP |
4222 | } |
4223 | ||
43010659 | 4224 | if (!ld_moved) { |
36dffab6 | 4225 | int active_balance = 0; |
ad273b32 | 4226 | |
d15bcfdb | 4227 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4228 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4229 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4230 | return -1; |
ad273b32 VS |
4231 | |
4232 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4233 | return -1; | |
4234 | ||
4235 | if (sd->nr_balance_failed++ < 2) | |
4236 | return -1; | |
4237 | ||
4238 | /* | |
4239 | * The only task running in a non-idle cpu can be moved to this | |
4240 | * cpu in an attempt to completely freeup the other CPU | |
4241 | * package. The same method used to move task in load_balance() | |
4242 | * have been extended for load_balance_newidle() to speedup | |
4243 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4244 | * | |
4245 | * The package power saving logic comes from | |
4246 | * find_busiest_group(). If there are no imbalance, then | |
4247 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4248 | * f_b_g() will select a group from which a running task may be | |
4249 | * pulled to this cpu in order to make the other package idle. | |
4250 | * If there is no opportunity to make a package idle and if | |
4251 | * there are no imbalance, then f_b_g() will return NULL and no | |
4252 | * action will be taken in load_balance_newidle(). | |
4253 | * | |
4254 | * Under normal task pull operation due to imbalance, there | |
4255 | * will be more than one task in the source run queue and | |
4256 | * move_tasks() will succeed. ld_moved will be true and this | |
4257 | * active balance code will not be triggered. | |
4258 | */ | |
4259 | ||
4260 | /* Lock busiest in correct order while this_rq is held */ | |
4261 | double_lock_balance(this_rq, busiest); | |
4262 | ||
4263 | /* | |
4264 | * don't kick the migration_thread, if the curr | |
4265 | * task on busiest cpu can't be moved to this_cpu | |
4266 | */ | |
6ca09dfc | 4267 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4268 | double_unlock_balance(this_rq, busiest); |
4269 | all_pinned = 1; | |
4270 | return ld_moved; | |
4271 | } | |
4272 | ||
4273 | if (!busiest->active_balance) { | |
4274 | busiest->active_balance = 1; | |
4275 | busiest->push_cpu = this_cpu; | |
4276 | active_balance = 1; | |
4277 | } | |
4278 | ||
4279 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4280 | /* |
4281 | * Should not call ttwu while holding a rq->lock | |
4282 | */ | |
4283 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4284 | if (active_balance) |
4285 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4286 | spin_lock(&this_rq->lock); |
ad273b32 | 4287 | |
5969fe06 | 4288 | } else |
16cfb1c0 | 4289 | sd->nr_balance_failed = 0; |
1da177e4 | 4290 | |
3e5459b4 | 4291 | update_shares_locked(this_rq, sd); |
43010659 | 4292 | return ld_moved; |
16cfb1c0 NP |
4293 | |
4294 | out_balanced: | |
d15bcfdb | 4295 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4296 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4297 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4298 | return -1; |
16cfb1c0 | 4299 | sd->nr_balance_failed = 0; |
48f24c4d | 4300 | |
16cfb1c0 | 4301 | return 0; |
1da177e4 LT |
4302 | } |
4303 | ||
4304 | /* | |
4305 | * idle_balance is called by schedule() if this_cpu is about to become | |
4306 | * idle. Attempts to pull tasks from other CPUs. | |
4307 | */ | |
70b97a7f | 4308 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4309 | { |
4310 | struct sched_domain *sd; | |
efbe027e | 4311 | int pulled_task = 0; |
dd41f596 | 4312 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4313 | |
4314 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4315 | unsigned long interval; |
4316 | ||
4317 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4318 | continue; | |
4319 | ||
4320 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4321 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4322 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4323 | sd); |
92c4ca5c CL |
4324 | |
4325 | interval = msecs_to_jiffies(sd->balance_interval); | |
4326 | if (time_after(next_balance, sd->last_balance + interval)) | |
4327 | next_balance = sd->last_balance + interval; | |
4328 | if (pulled_task) | |
4329 | break; | |
1da177e4 | 4330 | } |
dd41f596 | 4331 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4332 | /* |
4333 | * We are going idle. next_balance may be set based on | |
4334 | * a busy processor. So reset next_balance. | |
4335 | */ | |
4336 | this_rq->next_balance = next_balance; | |
dd41f596 | 4337 | } |
1da177e4 LT |
4338 | } |
4339 | ||
4340 | /* | |
4341 | * active_load_balance is run by migration threads. It pushes running tasks | |
4342 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4343 | * running on each physical CPU where possible, and avoids physical / | |
4344 | * logical imbalances. | |
4345 | * | |
4346 | * Called with busiest_rq locked. | |
4347 | */ | |
70b97a7f | 4348 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4349 | { |
39507451 | 4350 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4351 | struct sched_domain *sd; |
4352 | struct rq *target_rq; | |
39507451 | 4353 | |
48f24c4d | 4354 | /* Is there any task to move? */ |
39507451 | 4355 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4356 | return; |
4357 | ||
4358 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4359 | |
4360 | /* | |
39507451 | 4361 | * This condition is "impossible", if it occurs |
41a2d6cf | 4362 | * we need to fix it. Originally reported by |
39507451 | 4363 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4364 | */ |
39507451 | 4365 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4366 | |
39507451 NP |
4367 | /* move a task from busiest_rq to target_rq */ |
4368 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4369 | update_rq_clock(busiest_rq); |
4370 | update_rq_clock(target_rq); | |
39507451 NP |
4371 | |
4372 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4373 | for_each_domain(target_cpu, sd) { |
39507451 | 4374 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4375 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4376 | break; |
c96d145e | 4377 | } |
39507451 | 4378 | |
48f24c4d | 4379 | if (likely(sd)) { |
2d72376b | 4380 | schedstat_inc(sd, alb_count); |
39507451 | 4381 | |
43010659 PW |
4382 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4383 | sd, CPU_IDLE)) | |
48f24c4d IM |
4384 | schedstat_inc(sd, alb_pushed); |
4385 | else | |
4386 | schedstat_inc(sd, alb_failed); | |
4387 | } | |
1b12bbc7 | 4388 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4389 | } |
4390 | ||
46cb4b7c SS |
4391 | #ifdef CONFIG_NO_HZ |
4392 | static struct { | |
4393 | atomic_t load_balancer; | |
7d1e6a9b | 4394 | cpumask_var_t cpu_mask; |
f711f609 | 4395 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4396 | } nohz ____cacheline_aligned = { |
4397 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4398 | }; |
4399 | ||
eea08f32 AB |
4400 | int get_nohz_load_balancer(void) |
4401 | { | |
4402 | return atomic_read(&nohz.load_balancer); | |
4403 | } | |
4404 | ||
f711f609 GS |
4405 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4406 | /** | |
4407 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4408 | * @cpu: The cpu whose lowest level of sched domain is to | |
4409 | * be returned. | |
4410 | * @flag: The flag to check for the lowest sched_domain | |
4411 | * for the given cpu. | |
4412 | * | |
4413 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4414 | */ | |
4415 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4416 | { | |
4417 | struct sched_domain *sd; | |
4418 | ||
4419 | for_each_domain(cpu, sd) | |
4420 | if (sd && (sd->flags & flag)) | |
4421 | break; | |
4422 | ||
4423 | return sd; | |
4424 | } | |
4425 | ||
4426 | /** | |
4427 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4428 | * @cpu: The cpu whose domains we're iterating over. | |
4429 | * @sd: variable holding the value of the power_savings_sd | |
4430 | * for cpu. | |
4431 | * @flag: The flag to filter the sched_domains to be iterated. | |
4432 | * | |
4433 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4434 | * set, starting from the lowest sched_domain to the highest. | |
4435 | */ | |
4436 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4437 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4438 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4439 | ||
4440 | /** | |
4441 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4442 | * @ilb_group: group to be checked for semi-idleness | |
4443 | * | |
4444 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4445 | * | |
4446 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4447 | * and atleast one non-idle CPU. This helper function checks if the given | |
4448 | * sched_group is semi-idle or not. | |
4449 | */ | |
4450 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4451 | { | |
4452 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4453 | sched_group_cpus(ilb_group)); | |
4454 | ||
4455 | /* | |
4456 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4457 | * and atleast one idle cpu. | |
4458 | */ | |
4459 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4460 | return 0; | |
4461 | ||
4462 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4463 | return 0; | |
4464 | ||
4465 | return 1; | |
4466 | } | |
4467 | /** | |
4468 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4469 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4470 | * | |
4471 | * Returns: Returns the id of the idle load balancer if it exists, | |
4472 | * Else, returns >= nr_cpu_ids. | |
4473 | * | |
4474 | * This algorithm picks the idle load balancer such that it belongs to a | |
4475 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4476 | * completely idle packages/cores just for the purpose of idle load balancing | |
4477 | * when there are other idle cpu's which are better suited for that job. | |
4478 | */ | |
4479 | static int find_new_ilb(int cpu) | |
4480 | { | |
4481 | struct sched_domain *sd; | |
4482 | struct sched_group *ilb_group; | |
4483 | ||
4484 | /* | |
4485 | * Have idle load balancer selection from semi-idle packages only | |
4486 | * when power-aware load balancing is enabled | |
4487 | */ | |
4488 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4489 | goto out_done; | |
4490 | ||
4491 | /* | |
4492 | * Optimize for the case when we have no idle CPUs or only one | |
4493 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4494 | */ | |
4495 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4496 | goto out_done; | |
4497 | ||
4498 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4499 | ilb_group = sd->groups; | |
4500 | ||
4501 | do { | |
4502 | if (is_semi_idle_group(ilb_group)) | |
4503 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4504 | ||
4505 | ilb_group = ilb_group->next; | |
4506 | ||
4507 | } while (ilb_group != sd->groups); | |
4508 | } | |
4509 | ||
4510 | out_done: | |
4511 | return cpumask_first(nohz.cpu_mask); | |
4512 | } | |
4513 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4514 | static inline int find_new_ilb(int call_cpu) | |
4515 | { | |
6e29ec57 | 4516 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4517 | } |
4518 | #endif | |
4519 | ||
7835b98b | 4520 | /* |
46cb4b7c SS |
4521 | * This routine will try to nominate the ilb (idle load balancing) |
4522 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4523 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4524 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4525 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4526 | * arrives... | |
4527 | * | |
4528 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4529 | * for idle load balancing. ilb owner will still be part of | |
4530 | * nohz.cpu_mask.. | |
7835b98b | 4531 | * |
46cb4b7c SS |
4532 | * While stopping the tick, this cpu will become the ilb owner if there |
4533 | * is no other owner. And will be the owner till that cpu becomes busy | |
4534 | * or if all cpus in the system stop their ticks at which point | |
4535 | * there is no need for ilb owner. | |
4536 | * | |
4537 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4538 | * next busy scheduler_tick() | |
4539 | */ | |
4540 | int select_nohz_load_balancer(int stop_tick) | |
4541 | { | |
4542 | int cpu = smp_processor_id(); | |
4543 | ||
4544 | if (stop_tick) { | |
46cb4b7c SS |
4545 | cpu_rq(cpu)->in_nohz_recently = 1; |
4546 | ||
483b4ee6 SS |
4547 | if (!cpu_active(cpu)) { |
4548 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4549 | return 0; | |
4550 | ||
4551 | /* | |
4552 | * If we are going offline and still the leader, | |
4553 | * give up! | |
4554 | */ | |
46cb4b7c SS |
4555 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4556 | BUG(); | |
483b4ee6 | 4557 | |
46cb4b7c SS |
4558 | return 0; |
4559 | } | |
4560 | ||
483b4ee6 SS |
4561 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4562 | ||
46cb4b7c | 4563 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4564 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4565 | if (atomic_read(&nohz.load_balancer) == cpu) |
4566 | atomic_set(&nohz.load_balancer, -1); | |
4567 | return 0; | |
4568 | } | |
4569 | ||
4570 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4571 | /* make me the ilb owner */ | |
4572 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4573 | return 1; | |
e790fb0b GS |
4574 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4575 | int new_ilb; | |
4576 | ||
4577 | if (!(sched_smt_power_savings || | |
4578 | sched_mc_power_savings)) | |
4579 | return 1; | |
4580 | /* | |
4581 | * Check to see if there is a more power-efficient | |
4582 | * ilb. | |
4583 | */ | |
4584 | new_ilb = find_new_ilb(cpu); | |
4585 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4586 | atomic_set(&nohz.load_balancer, -1); | |
4587 | resched_cpu(new_ilb); | |
4588 | return 0; | |
4589 | } | |
46cb4b7c | 4590 | return 1; |
e790fb0b | 4591 | } |
46cb4b7c | 4592 | } else { |
7d1e6a9b | 4593 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4594 | return 0; |
4595 | ||
7d1e6a9b | 4596 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4597 | |
4598 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4599 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4600 | BUG(); | |
4601 | } | |
4602 | return 0; | |
4603 | } | |
4604 | #endif | |
4605 | ||
4606 | static DEFINE_SPINLOCK(balancing); | |
4607 | ||
4608 | /* | |
7835b98b CL |
4609 | * It checks each scheduling domain to see if it is due to be balanced, |
4610 | * and initiates a balancing operation if so. | |
4611 | * | |
4612 | * Balancing parameters are set up in arch_init_sched_domains. | |
4613 | */ | |
a9957449 | 4614 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4615 | { |
46cb4b7c SS |
4616 | int balance = 1; |
4617 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4618 | unsigned long interval; |
4619 | struct sched_domain *sd; | |
46cb4b7c | 4620 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4621 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4622 | int update_next_balance = 0; |
d07355f5 | 4623 | int need_serialize; |
1da177e4 | 4624 | |
46cb4b7c | 4625 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4626 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4627 | continue; | |
4628 | ||
4629 | interval = sd->balance_interval; | |
d15bcfdb | 4630 | if (idle != CPU_IDLE) |
1da177e4 LT |
4631 | interval *= sd->busy_factor; |
4632 | ||
4633 | /* scale ms to jiffies */ | |
4634 | interval = msecs_to_jiffies(interval); | |
4635 | if (unlikely(!interval)) | |
4636 | interval = 1; | |
dd41f596 IM |
4637 | if (interval > HZ*NR_CPUS/10) |
4638 | interval = HZ*NR_CPUS/10; | |
4639 | ||
d07355f5 | 4640 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4641 | |
d07355f5 | 4642 | if (need_serialize) { |
08c183f3 CL |
4643 | if (!spin_trylock(&balancing)) |
4644 | goto out; | |
4645 | } | |
4646 | ||
c9819f45 | 4647 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4648 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4649 | /* |
4650 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4651 | * longer idle, or one of our SMT siblings is |
4652 | * not idle. | |
4653 | */ | |
d15bcfdb | 4654 | idle = CPU_NOT_IDLE; |
1da177e4 | 4655 | } |
1bd77f2d | 4656 | sd->last_balance = jiffies; |
1da177e4 | 4657 | } |
d07355f5 | 4658 | if (need_serialize) |
08c183f3 CL |
4659 | spin_unlock(&balancing); |
4660 | out: | |
f549da84 | 4661 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4662 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4663 | update_next_balance = 1; |
4664 | } | |
783609c6 SS |
4665 | |
4666 | /* | |
4667 | * Stop the load balance at this level. There is another | |
4668 | * CPU in our sched group which is doing load balancing more | |
4669 | * actively. | |
4670 | */ | |
4671 | if (!balance) | |
4672 | break; | |
1da177e4 | 4673 | } |
f549da84 SS |
4674 | |
4675 | /* | |
4676 | * next_balance will be updated only when there is a need. | |
4677 | * When the cpu is attached to null domain for ex, it will not be | |
4678 | * updated. | |
4679 | */ | |
4680 | if (likely(update_next_balance)) | |
4681 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4682 | } |
4683 | ||
4684 | /* | |
4685 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4686 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4687 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4688 | */ | |
4689 | static void run_rebalance_domains(struct softirq_action *h) | |
4690 | { | |
dd41f596 IM |
4691 | int this_cpu = smp_processor_id(); |
4692 | struct rq *this_rq = cpu_rq(this_cpu); | |
4693 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4694 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4695 | |
dd41f596 | 4696 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4697 | |
4698 | #ifdef CONFIG_NO_HZ | |
4699 | /* | |
4700 | * If this cpu is the owner for idle load balancing, then do the | |
4701 | * balancing on behalf of the other idle cpus whose ticks are | |
4702 | * stopped. | |
4703 | */ | |
dd41f596 IM |
4704 | if (this_rq->idle_at_tick && |
4705 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4706 | struct rq *rq; |
4707 | int balance_cpu; | |
4708 | ||
7d1e6a9b RR |
4709 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4710 | if (balance_cpu == this_cpu) | |
4711 | continue; | |
4712 | ||
46cb4b7c SS |
4713 | /* |
4714 | * If this cpu gets work to do, stop the load balancing | |
4715 | * work being done for other cpus. Next load | |
4716 | * balancing owner will pick it up. | |
4717 | */ | |
4718 | if (need_resched()) | |
4719 | break; | |
4720 | ||
de0cf899 | 4721 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4722 | |
4723 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4724 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4725 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4726 | } |
4727 | } | |
4728 | #endif | |
4729 | } | |
4730 | ||
8a0be9ef FW |
4731 | static inline int on_null_domain(int cpu) |
4732 | { | |
4733 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4734 | } | |
4735 | ||
46cb4b7c SS |
4736 | /* |
4737 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4738 | * | |
4739 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4740 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4741 | * if the whole system is idle. | |
4742 | */ | |
dd41f596 | 4743 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4744 | { |
46cb4b7c SS |
4745 | #ifdef CONFIG_NO_HZ |
4746 | /* | |
4747 | * If we were in the nohz mode recently and busy at the current | |
4748 | * scheduler tick, then check if we need to nominate new idle | |
4749 | * load balancer. | |
4750 | */ | |
4751 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4752 | rq->in_nohz_recently = 0; | |
4753 | ||
4754 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4755 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4756 | atomic_set(&nohz.load_balancer, -1); |
4757 | } | |
4758 | ||
4759 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4760 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4761 | |
434d53b0 | 4762 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4763 | resched_cpu(ilb); |
4764 | } | |
4765 | } | |
4766 | ||
4767 | /* | |
4768 | * If this cpu is idle and doing idle load balancing for all the | |
4769 | * cpus with ticks stopped, is it time for that to stop? | |
4770 | */ | |
4771 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4772 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4773 | resched_cpu(cpu); |
4774 | return; | |
4775 | } | |
4776 | ||
4777 | /* | |
4778 | * If this cpu is idle and the idle load balancing is done by | |
4779 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4780 | */ | |
4781 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4782 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4783 | return; |
4784 | #endif | |
8a0be9ef FW |
4785 | /* Don't need to rebalance while attached to NULL domain */ |
4786 | if (time_after_eq(jiffies, rq->next_balance) && | |
4787 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4788 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4789 | } |
dd41f596 IM |
4790 | |
4791 | #else /* CONFIG_SMP */ | |
4792 | ||
1da177e4 LT |
4793 | /* |
4794 | * on UP we do not need to balance between CPUs: | |
4795 | */ | |
70b97a7f | 4796 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4797 | { |
4798 | } | |
dd41f596 | 4799 | |
1da177e4 LT |
4800 | #endif |
4801 | ||
1da177e4 LT |
4802 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4803 | ||
4804 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4805 | ||
4806 | /* | |
c5f8d995 | 4807 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4808 | * @p in case that task is currently running. |
c5f8d995 HS |
4809 | * |
4810 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4811 | */ |
c5f8d995 HS |
4812 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4813 | { | |
4814 | u64 ns = 0; | |
4815 | ||
4816 | if (task_current(rq, p)) { | |
4817 | update_rq_clock(rq); | |
4818 | ns = rq->clock - p->se.exec_start; | |
4819 | if ((s64)ns < 0) | |
4820 | ns = 0; | |
4821 | } | |
4822 | ||
4823 | return ns; | |
4824 | } | |
4825 | ||
bb34d92f | 4826 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4827 | { |
1da177e4 | 4828 | unsigned long flags; |
41b86e9c | 4829 | struct rq *rq; |
bb34d92f | 4830 | u64 ns = 0; |
48f24c4d | 4831 | |
41b86e9c | 4832 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4833 | ns = do_task_delta_exec(p, rq); |
4834 | task_rq_unlock(rq, &flags); | |
1508487e | 4835 | |
c5f8d995 HS |
4836 | return ns; |
4837 | } | |
f06febc9 | 4838 | |
c5f8d995 HS |
4839 | /* |
4840 | * Return accounted runtime for the task. | |
4841 | * In case the task is currently running, return the runtime plus current's | |
4842 | * pending runtime that have not been accounted yet. | |
4843 | */ | |
4844 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4845 | { | |
4846 | unsigned long flags; | |
4847 | struct rq *rq; | |
4848 | u64 ns = 0; | |
4849 | ||
4850 | rq = task_rq_lock(p, &flags); | |
4851 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4852 | task_rq_unlock(rq, &flags); | |
4853 | ||
4854 | return ns; | |
4855 | } | |
48f24c4d | 4856 | |
c5f8d995 HS |
4857 | /* |
4858 | * Return sum_exec_runtime for the thread group. | |
4859 | * In case the task is currently running, return the sum plus current's | |
4860 | * pending runtime that have not been accounted yet. | |
4861 | * | |
4862 | * Note that the thread group might have other running tasks as well, | |
4863 | * so the return value not includes other pending runtime that other | |
4864 | * running tasks might have. | |
4865 | */ | |
4866 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4867 | { | |
4868 | struct task_cputime totals; | |
4869 | unsigned long flags; | |
4870 | struct rq *rq; | |
4871 | u64 ns; | |
4872 | ||
4873 | rq = task_rq_lock(p, &flags); | |
4874 | thread_group_cputime(p, &totals); | |
4875 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4876 | task_rq_unlock(rq, &flags); |
48f24c4d | 4877 | |
1da177e4 LT |
4878 | return ns; |
4879 | } | |
4880 | ||
1da177e4 LT |
4881 | /* |
4882 | * Account user cpu time to a process. | |
4883 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4884 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4885 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4886 | */ |
457533a7 MS |
4887 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4888 | cputime_t cputime_scaled) | |
1da177e4 LT |
4889 | { |
4890 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4891 | cputime64_t tmp; | |
4892 | ||
457533a7 | 4893 | /* Add user time to process. */ |
1da177e4 | 4894 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4895 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4896 | account_group_user_time(p, cputime); |
1da177e4 LT |
4897 | |
4898 | /* Add user time to cpustat. */ | |
4899 | tmp = cputime_to_cputime64(cputime); | |
4900 | if (TASK_NICE(p) > 0) | |
4901 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4902 | else | |
4903 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4904 | |
4905 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4906 | /* Account for user time used */ |
4907 | acct_update_integrals(p); | |
1da177e4 LT |
4908 | } |
4909 | ||
94886b84 LV |
4910 | /* |
4911 | * Account guest cpu time to a process. | |
4912 | * @p: the process that the cpu time gets accounted to | |
4913 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4914 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4915 | */ |
457533a7 MS |
4916 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4917 | cputime_t cputime_scaled) | |
94886b84 LV |
4918 | { |
4919 | cputime64_t tmp; | |
4920 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4921 | ||
4922 | tmp = cputime_to_cputime64(cputime); | |
4923 | ||
457533a7 | 4924 | /* Add guest time to process. */ |
94886b84 | 4925 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4926 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4927 | account_group_user_time(p, cputime); |
94886b84 LV |
4928 | p->gtime = cputime_add(p->gtime, cputime); |
4929 | ||
457533a7 | 4930 | /* Add guest time to cpustat. */ |
94886b84 LV |
4931 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4932 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4933 | } | |
4934 | ||
1da177e4 LT |
4935 | /* |
4936 | * Account system cpu time to a process. | |
4937 | * @p: the process that the cpu time gets accounted to | |
4938 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4939 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4940 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4941 | */ |
4942 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4943 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4944 | { |
4945 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4946 | cputime64_t tmp; |
4947 | ||
983ed7a6 | 4948 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4949 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4950 | return; |
4951 | } | |
94886b84 | 4952 | |
457533a7 | 4953 | /* Add system time to process. */ |
1da177e4 | 4954 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4955 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4956 | account_group_system_time(p, cputime); |
1da177e4 LT |
4957 | |
4958 | /* Add system time to cpustat. */ | |
4959 | tmp = cputime_to_cputime64(cputime); | |
4960 | if (hardirq_count() - hardirq_offset) | |
4961 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4962 | else if (softirq_count()) | |
4963 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4964 | else |
79741dd3 MS |
4965 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4966 | ||
ef12fefa BR |
4967 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4968 | ||
1da177e4 LT |
4969 | /* Account for system time used */ |
4970 | acct_update_integrals(p); | |
1da177e4 LT |
4971 | } |
4972 | ||
c66f08be | 4973 | /* |
1da177e4 | 4974 | * Account for involuntary wait time. |
1da177e4 | 4975 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4976 | */ |
79741dd3 | 4977 | void account_steal_time(cputime_t cputime) |
c66f08be | 4978 | { |
79741dd3 MS |
4979 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4980 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4981 | ||
4982 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4983 | } |
4984 | ||
1da177e4 | 4985 | /* |
79741dd3 MS |
4986 | * Account for idle time. |
4987 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4988 | */ |
79741dd3 | 4989 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4990 | { |
4991 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4992 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4993 | struct rq *rq = this_rq(); |
1da177e4 | 4994 | |
79741dd3 MS |
4995 | if (atomic_read(&rq->nr_iowait) > 0) |
4996 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4997 | else | |
4998 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4999 | } |
5000 | ||
79741dd3 MS |
5001 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5002 | ||
5003 | /* | |
5004 | * Account a single tick of cpu time. | |
5005 | * @p: the process that the cpu time gets accounted to | |
5006 | * @user_tick: indicates if the tick is a user or a system tick | |
5007 | */ | |
5008 | void account_process_tick(struct task_struct *p, int user_tick) | |
5009 | { | |
5010 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5011 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5012 | struct rq *rq = this_rq(); | |
5013 | ||
5014 | if (user_tick) | |
5015 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5016 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5017 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5018 | one_jiffy_scaled); | |
5019 | else | |
5020 | account_idle_time(one_jiffy); | |
5021 | } | |
5022 | ||
5023 | /* | |
5024 | * Account multiple ticks of steal time. | |
5025 | * @p: the process from which the cpu time has been stolen | |
5026 | * @ticks: number of stolen ticks | |
5027 | */ | |
5028 | void account_steal_ticks(unsigned long ticks) | |
5029 | { | |
5030 | account_steal_time(jiffies_to_cputime(ticks)); | |
5031 | } | |
5032 | ||
5033 | /* | |
5034 | * Account multiple ticks of idle time. | |
5035 | * @ticks: number of stolen ticks | |
5036 | */ | |
5037 | void account_idle_ticks(unsigned long ticks) | |
5038 | { | |
5039 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5040 | } |
5041 | ||
79741dd3 MS |
5042 | #endif |
5043 | ||
49048622 BS |
5044 | /* |
5045 | * Use precise platform statistics if available: | |
5046 | */ | |
5047 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5048 | cputime_t task_utime(struct task_struct *p) | |
5049 | { | |
5050 | return p->utime; | |
5051 | } | |
5052 | ||
5053 | cputime_t task_stime(struct task_struct *p) | |
5054 | { | |
5055 | return p->stime; | |
5056 | } | |
5057 | #else | |
5058 | cputime_t task_utime(struct task_struct *p) | |
5059 | { | |
5060 | clock_t utime = cputime_to_clock_t(p->utime), | |
5061 | total = utime + cputime_to_clock_t(p->stime); | |
5062 | u64 temp; | |
5063 | ||
5064 | /* | |
5065 | * Use CFS's precise accounting: | |
5066 | */ | |
5067 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5068 | ||
5069 | if (total) { | |
5070 | temp *= utime; | |
5071 | do_div(temp, total); | |
5072 | } | |
5073 | utime = (clock_t)temp; | |
5074 | ||
5075 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5076 | return p->prev_utime; | |
5077 | } | |
5078 | ||
5079 | cputime_t task_stime(struct task_struct *p) | |
5080 | { | |
5081 | clock_t stime; | |
5082 | ||
5083 | /* | |
5084 | * Use CFS's precise accounting. (we subtract utime from | |
5085 | * the total, to make sure the total observed by userspace | |
5086 | * grows monotonically - apps rely on that): | |
5087 | */ | |
5088 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5089 | cputime_to_clock_t(task_utime(p)); | |
5090 | ||
5091 | if (stime >= 0) | |
5092 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5093 | ||
5094 | return p->prev_stime; | |
5095 | } | |
5096 | #endif | |
5097 | ||
5098 | inline cputime_t task_gtime(struct task_struct *p) | |
5099 | { | |
5100 | return p->gtime; | |
5101 | } | |
5102 | ||
7835b98b CL |
5103 | /* |
5104 | * This function gets called by the timer code, with HZ frequency. | |
5105 | * We call it with interrupts disabled. | |
5106 | * | |
5107 | * It also gets called by the fork code, when changing the parent's | |
5108 | * timeslices. | |
5109 | */ | |
5110 | void scheduler_tick(void) | |
5111 | { | |
7835b98b CL |
5112 | int cpu = smp_processor_id(); |
5113 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5114 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5115 | |
5116 | sched_clock_tick(); | |
dd41f596 IM |
5117 | |
5118 | spin_lock(&rq->lock); | |
3e51f33f | 5119 | update_rq_clock(rq); |
f1a438d8 | 5120 | update_cpu_load(rq); |
fa85ae24 | 5121 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5122 | spin_unlock(&rq->lock); |
7835b98b | 5123 | |
e220d2dc PZ |
5124 | perf_counter_task_tick(curr, cpu); |
5125 | ||
e418e1c2 | 5126 | #ifdef CONFIG_SMP |
dd41f596 IM |
5127 | rq->idle_at_tick = idle_cpu(cpu); |
5128 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5129 | #endif |
1da177e4 LT |
5130 | } |
5131 | ||
132380a0 | 5132 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5133 | { |
5134 | if (in_lock_functions(addr)) { | |
5135 | addr = CALLER_ADDR2; | |
5136 | if (in_lock_functions(addr)) | |
5137 | addr = CALLER_ADDR3; | |
5138 | } | |
5139 | return addr; | |
5140 | } | |
1da177e4 | 5141 | |
7e49fcce SR |
5142 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5143 | defined(CONFIG_PREEMPT_TRACER)) | |
5144 | ||
43627582 | 5145 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5146 | { |
6cd8a4bb | 5147 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5148 | /* |
5149 | * Underflow? | |
5150 | */ | |
9a11b49a IM |
5151 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5152 | return; | |
6cd8a4bb | 5153 | #endif |
1da177e4 | 5154 | preempt_count() += val; |
6cd8a4bb | 5155 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5156 | /* |
5157 | * Spinlock count overflowing soon? | |
5158 | */ | |
33859f7f MOS |
5159 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5160 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5161 | #endif |
5162 | if (preempt_count() == val) | |
5163 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5164 | } |
5165 | EXPORT_SYMBOL(add_preempt_count); | |
5166 | ||
43627582 | 5167 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5168 | { |
6cd8a4bb | 5169 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5170 | /* |
5171 | * Underflow? | |
5172 | */ | |
01e3eb82 | 5173 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5174 | return; |
1da177e4 LT |
5175 | /* |
5176 | * Is the spinlock portion underflowing? | |
5177 | */ | |
9a11b49a IM |
5178 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5179 | !(preempt_count() & PREEMPT_MASK))) | |
5180 | return; | |
6cd8a4bb | 5181 | #endif |
9a11b49a | 5182 | |
6cd8a4bb SR |
5183 | if (preempt_count() == val) |
5184 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5185 | preempt_count() -= val; |
5186 | } | |
5187 | EXPORT_SYMBOL(sub_preempt_count); | |
5188 | ||
5189 | #endif | |
5190 | ||
5191 | /* | |
dd41f596 | 5192 | * Print scheduling while atomic bug: |
1da177e4 | 5193 | */ |
dd41f596 | 5194 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5195 | { |
838225b4 SS |
5196 | struct pt_regs *regs = get_irq_regs(); |
5197 | ||
5198 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5199 | prev->comm, prev->pid, preempt_count()); | |
5200 | ||
dd41f596 | 5201 | debug_show_held_locks(prev); |
e21f5b15 | 5202 | print_modules(); |
dd41f596 IM |
5203 | if (irqs_disabled()) |
5204 | print_irqtrace_events(prev); | |
838225b4 SS |
5205 | |
5206 | if (regs) | |
5207 | show_regs(regs); | |
5208 | else | |
5209 | dump_stack(); | |
dd41f596 | 5210 | } |
1da177e4 | 5211 | |
dd41f596 IM |
5212 | /* |
5213 | * Various schedule()-time debugging checks and statistics: | |
5214 | */ | |
5215 | static inline void schedule_debug(struct task_struct *prev) | |
5216 | { | |
1da177e4 | 5217 | /* |
41a2d6cf | 5218 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5219 | * schedule() atomically, we ignore that path for now. |
5220 | * Otherwise, whine if we are scheduling when we should not be. | |
5221 | */ | |
3f33a7ce | 5222 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5223 | __schedule_bug(prev); |
5224 | ||
1da177e4 LT |
5225 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5226 | ||
2d72376b | 5227 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5228 | #ifdef CONFIG_SCHEDSTATS |
5229 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5230 | schedstat_inc(this_rq(), bkl_count); |
5231 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5232 | } |
5233 | #endif | |
dd41f596 IM |
5234 | } |
5235 | ||
df1c99d4 MG |
5236 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5237 | { | |
5238 | if (prev->state == TASK_RUNNING) { | |
5239 | u64 runtime = prev->se.sum_exec_runtime; | |
5240 | ||
5241 | runtime -= prev->se.prev_sum_exec_runtime; | |
5242 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5243 | ||
5244 | /* | |
5245 | * In order to avoid avg_overlap growing stale when we are | |
5246 | * indeed overlapping and hence not getting put to sleep, grow | |
5247 | * the avg_overlap on preemption. | |
5248 | * | |
5249 | * We use the average preemption runtime because that | |
5250 | * correlates to the amount of cache footprint a task can | |
5251 | * build up. | |
5252 | */ | |
5253 | update_avg(&prev->se.avg_overlap, runtime); | |
5254 | } | |
5255 | prev->sched_class->put_prev_task(rq, prev); | |
5256 | } | |
5257 | ||
dd41f596 IM |
5258 | /* |
5259 | * Pick up the highest-prio task: | |
5260 | */ | |
5261 | static inline struct task_struct * | |
b67802ea | 5262 | pick_next_task(struct rq *rq) |
dd41f596 | 5263 | { |
5522d5d5 | 5264 | const struct sched_class *class; |
dd41f596 | 5265 | struct task_struct *p; |
1da177e4 LT |
5266 | |
5267 | /* | |
dd41f596 IM |
5268 | * Optimization: we know that if all tasks are in |
5269 | * the fair class we can call that function directly: | |
1da177e4 | 5270 | */ |
dd41f596 | 5271 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5272 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5273 | if (likely(p)) |
5274 | return p; | |
1da177e4 LT |
5275 | } |
5276 | ||
dd41f596 IM |
5277 | class = sched_class_highest; |
5278 | for ( ; ; ) { | |
fb8d4724 | 5279 | p = class->pick_next_task(rq); |
dd41f596 IM |
5280 | if (p) |
5281 | return p; | |
5282 | /* | |
5283 | * Will never be NULL as the idle class always | |
5284 | * returns a non-NULL p: | |
5285 | */ | |
5286 | class = class->next; | |
5287 | } | |
5288 | } | |
1da177e4 | 5289 | |
dd41f596 IM |
5290 | /* |
5291 | * schedule() is the main scheduler function. | |
5292 | */ | |
ff743345 | 5293 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5294 | { |
5295 | struct task_struct *prev, *next; | |
67ca7bde | 5296 | unsigned long *switch_count; |
dd41f596 | 5297 | struct rq *rq; |
31656519 | 5298 | int cpu; |
dd41f596 | 5299 | |
ff743345 PZ |
5300 | need_resched: |
5301 | preempt_disable(); | |
dd41f596 IM |
5302 | cpu = smp_processor_id(); |
5303 | rq = cpu_rq(cpu); | |
5304 | rcu_qsctr_inc(cpu); | |
5305 | prev = rq->curr; | |
5306 | switch_count = &prev->nivcsw; | |
5307 | ||
5308 | release_kernel_lock(prev); | |
5309 | need_resched_nonpreemptible: | |
5310 | ||
5311 | schedule_debug(prev); | |
1da177e4 | 5312 | |
31656519 | 5313 | if (sched_feat(HRTICK)) |
f333fdc9 | 5314 | hrtick_clear(rq); |
8f4d37ec | 5315 | |
8cd162ce | 5316 | spin_lock_irq(&rq->lock); |
3e51f33f | 5317 | update_rq_clock(rq); |
1e819950 | 5318 | clear_tsk_need_resched(prev); |
1da177e4 | 5319 | |
1da177e4 | 5320 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5321 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5322 | prev->state = TASK_RUNNING; |
16882c1e | 5323 | else |
2e1cb74a | 5324 | deactivate_task(rq, prev, 1); |
dd41f596 | 5325 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5326 | } |
5327 | ||
9a897c5a SR |
5328 | #ifdef CONFIG_SMP |
5329 | if (prev->sched_class->pre_schedule) | |
5330 | prev->sched_class->pre_schedule(rq, prev); | |
5331 | #endif | |
f65eda4f | 5332 | |
dd41f596 | 5333 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5334 | idle_balance(cpu, rq); |
1da177e4 | 5335 | |
df1c99d4 | 5336 | put_prev_task(rq, prev); |
b67802ea | 5337 | next = pick_next_task(rq); |
1da177e4 | 5338 | |
1da177e4 | 5339 | if (likely(prev != next)) { |
673a90a1 | 5340 | sched_info_switch(prev, next); |
564c2b21 | 5341 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5342 | |
1da177e4 LT |
5343 | rq->nr_switches++; |
5344 | rq->curr = next; | |
5345 | ++*switch_count; | |
5346 | ||
dd41f596 | 5347 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5348 | /* |
5349 | * the context switch might have flipped the stack from under | |
5350 | * us, hence refresh the local variables. | |
5351 | */ | |
5352 | cpu = smp_processor_id(); | |
5353 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5354 | } else |
5355 | spin_unlock_irq(&rq->lock); | |
5356 | ||
8f4d37ec | 5357 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5358 | goto need_resched_nonpreemptible; |
8f4d37ec | 5359 | |
1da177e4 | 5360 | preempt_enable_no_resched(); |
ff743345 | 5361 | if (need_resched()) |
1da177e4 LT |
5362 | goto need_resched; |
5363 | } | |
1da177e4 LT |
5364 | EXPORT_SYMBOL(schedule); |
5365 | ||
0d66bf6d PZ |
5366 | #ifdef CONFIG_SMP |
5367 | /* | |
5368 | * Look out! "owner" is an entirely speculative pointer | |
5369 | * access and not reliable. | |
5370 | */ | |
5371 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5372 | { | |
5373 | unsigned int cpu; | |
5374 | struct rq *rq; | |
5375 | ||
5376 | if (!sched_feat(OWNER_SPIN)) | |
5377 | return 0; | |
5378 | ||
5379 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5380 | /* | |
5381 | * Need to access the cpu field knowing that | |
5382 | * DEBUG_PAGEALLOC could have unmapped it if | |
5383 | * the mutex owner just released it and exited. | |
5384 | */ | |
5385 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5386 | goto out; | |
5387 | #else | |
5388 | cpu = owner->cpu; | |
5389 | #endif | |
5390 | ||
5391 | /* | |
5392 | * Even if the access succeeded (likely case), | |
5393 | * the cpu field may no longer be valid. | |
5394 | */ | |
5395 | if (cpu >= nr_cpumask_bits) | |
5396 | goto out; | |
5397 | ||
5398 | /* | |
5399 | * We need to validate that we can do a | |
5400 | * get_cpu() and that we have the percpu area. | |
5401 | */ | |
5402 | if (!cpu_online(cpu)) | |
5403 | goto out; | |
5404 | ||
5405 | rq = cpu_rq(cpu); | |
5406 | ||
5407 | for (;;) { | |
5408 | /* | |
5409 | * Owner changed, break to re-assess state. | |
5410 | */ | |
5411 | if (lock->owner != owner) | |
5412 | break; | |
5413 | ||
5414 | /* | |
5415 | * Is that owner really running on that cpu? | |
5416 | */ | |
5417 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5418 | return 0; | |
5419 | ||
5420 | cpu_relax(); | |
5421 | } | |
5422 | out: | |
5423 | return 1; | |
5424 | } | |
5425 | #endif | |
5426 | ||
1da177e4 LT |
5427 | #ifdef CONFIG_PREEMPT |
5428 | /* | |
2ed6e34f | 5429 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5430 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5431 | * occur there and call schedule directly. |
5432 | */ | |
5433 | asmlinkage void __sched preempt_schedule(void) | |
5434 | { | |
5435 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5436 | |
1da177e4 LT |
5437 | /* |
5438 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5439 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5440 | */ |
beed33a8 | 5441 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5442 | return; |
5443 | ||
3a5c359a AK |
5444 | do { |
5445 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5446 | schedule(); |
3a5c359a | 5447 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5448 | |
3a5c359a AK |
5449 | /* |
5450 | * Check again in case we missed a preemption opportunity | |
5451 | * between schedule and now. | |
5452 | */ | |
5453 | barrier(); | |
5ed0cec0 | 5454 | } while (need_resched()); |
1da177e4 | 5455 | } |
1da177e4 LT |
5456 | EXPORT_SYMBOL(preempt_schedule); |
5457 | ||
5458 | /* | |
2ed6e34f | 5459 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5460 | * off of irq context. |
5461 | * Note, that this is called and return with irqs disabled. This will | |
5462 | * protect us against recursive calling from irq. | |
5463 | */ | |
5464 | asmlinkage void __sched preempt_schedule_irq(void) | |
5465 | { | |
5466 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5467 | |
2ed6e34f | 5468 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5469 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5470 | ||
3a5c359a AK |
5471 | do { |
5472 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5473 | local_irq_enable(); |
5474 | schedule(); | |
5475 | local_irq_disable(); | |
3a5c359a | 5476 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5477 | |
3a5c359a AK |
5478 | /* |
5479 | * Check again in case we missed a preemption opportunity | |
5480 | * between schedule and now. | |
5481 | */ | |
5482 | barrier(); | |
5ed0cec0 | 5483 | } while (need_resched()); |
1da177e4 LT |
5484 | } |
5485 | ||
5486 | #endif /* CONFIG_PREEMPT */ | |
5487 | ||
95cdf3b7 IM |
5488 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5489 | void *key) | |
1da177e4 | 5490 | { |
48f24c4d | 5491 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5492 | } |
1da177e4 LT |
5493 | EXPORT_SYMBOL(default_wake_function); |
5494 | ||
5495 | /* | |
41a2d6cf IM |
5496 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5497 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5498 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5499 | * | |
5500 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5501 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5502 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5503 | */ | |
78ddb08f | 5504 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5505 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5506 | { |
2e45874c | 5507 | wait_queue_t *curr, *next; |
1da177e4 | 5508 | |
2e45874c | 5509 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5510 | unsigned flags = curr->flags; |
5511 | ||
1da177e4 | 5512 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5513 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5514 | break; |
5515 | } | |
5516 | } | |
5517 | ||
5518 | /** | |
5519 | * __wake_up - wake up threads blocked on a waitqueue. | |
5520 | * @q: the waitqueue | |
5521 | * @mode: which threads | |
5522 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5523 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5524 | * |
5525 | * It may be assumed that this function implies a write memory barrier before | |
5526 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5527 | */ |
7ad5b3a5 | 5528 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5529 | int nr_exclusive, void *key) |
1da177e4 LT |
5530 | { |
5531 | unsigned long flags; | |
5532 | ||
5533 | spin_lock_irqsave(&q->lock, flags); | |
5534 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5535 | spin_unlock_irqrestore(&q->lock, flags); | |
5536 | } | |
1da177e4 LT |
5537 | EXPORT_SYMBOL(__wake_up); |
5538 | ||
5539 | /* | |
5540 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5541 | */ | |
7ad5b3a5 | 5542 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5543 | { |
5544 | __wake_up_common(q, mode, 1, 0, NULL); | |
5545 | } | |
5546 | ||
4ede816a DL |
5547 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5548 | { | |
5549 | __wake_up_common(q, mode, 1, 0, key); | |
5550 | } | |
5551 | ||
1da177e4 | 5552 | /** |
4ede816a | 5553 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5554 | * @q: the waitqueue |
5555 | * @mode: which threads | |
5556 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5557 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5558 | * |
5559 | * The sync wakeup differs that the waker knows that it will schedule | |
5560 | * away soon, so while the target thread will be woken up, it will not | |
5561 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5562 | * with each other. This can prevent needless bouncing between CPUs. | |
5563 | * | |
5564 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5565 | * |
5566 | * It may be assumed that this function implies a write memory barrier before | |
5567 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5568 | */ |
4ede816a DL |
5569 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5570 | int nr_exclusive, void *key) | |
1da177e4 LT |
5571 | { |
5572 | unsigned long flags; | |
5573 | int sync = 1; | |
5574 | ||
5575 | if (unlikely(!q)) | |
5576 | return; | |
5577 | ||
5578 | if (unlikely(!nr_exclusive)) | |
5579 | sync = 0; | |
5580 | ||
5581 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5582 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5583 | spin_unlock_irqrestore(&q->lock, flags); |
5584 | } | |
4ede816a DL |
5585 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5586 | ||
5587 | /* | |
5588 | * __wake_up_sync - see __wake_up_sync_key() | |
5589 | */ | |
5590 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5591 | { | |
5592 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5593 | } | |
1da177e4 LT |
5594 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5595 | ||
65eb3dc6 KD |
5596 | /** |
5597 | * complete: - signals a single thread waiting on this completion | |
5598 | * @x: holds the state of this particular completion | |
5599 | * | |
5600 | * This will wake up a single thread waiting on this completion. Threads will be | |
5601 | * awakened in the same order in which they were queued. | |
5602 | * | |
5603 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5604 | * |
5605 | * It may be assumed that this function implies a write memory barrier before | |
5606 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5607 | */ |
b15136e9 | 5608 | void complete(struct completion *x) |
1da177e4 LT |
5609 | { |
5610 | unsigned long flags; | |
5611 | ||
5612 | spin_lock_irqsave(&x->wait.lock, flags); | |
5613 | x->done++; | |
d9514f6c | 5614 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5615 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5616 | } | |
5617 | EXPORT_SYMBOL(complete); | |
5618 | ||
65eb3dc6 KD |
5619 | /** |
5620 | * complete_all: - signals all threads waiting on this completion | |
5621 | * @x: holds the state of this particular completion | |
5622 | * | |
5623 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5624 | * |
5625 | * It may be assumed that this function implies a write memory barrier before | |
5626 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5627 | */ |
b15136e9 | 5628 | void complete_all(struct completion *x) |
1da177e4 LT |
5629 | { |
5630 | unsigned long flags; | |
5631 | ||
5632 | spin_lock_irqsave(&x->wait.lock, flags); | |
5633 | x->done += UINT_MAX/2; | |
d9514f6c | 5634 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5635 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5636 | } | |
5637 | EXPORT_SYMBOL(complete_all); | |
5638 | ||
8cbbe86d AK |
5639 | static inline long __sched |
5640 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5641 | { |
1da177e4 LT |
5642 | if (!x->done) { |
5643 | DECLARE_WAITQUEUE(wait, current); | |
5644 | ||
5645 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5646 | __add_wait_queue_tail(&x->wait, &wait); | |
5647 | do { | |
94d3d824 | 5648 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5649 | timeout = -ERESTARTSYS; |
5650 | break; | |
8cbbe86d AK |
5651 | } |
5652 | __set_current_state(state); | |
1da177e4 LT |
5653 | spin_unlock_irq(&x->wait.lock); |
5654 | timeout = schedule_timeout(timeout); | |
5655 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5656 | } while (!x->done && timeout); |
1da177e4 | 5657 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5658 | if (!x->done) |
5659 | return timeout; | |
1da177e4 LT |
5660 | } |
5661 | x->done--; | |
ea71a546 | 5662 | return timeout ?: 1; |
1da177e4 | 5663 | } |
1da177e4 | 5664 | |
8cbbe86d AK |
5665 | static long __sched |
5666 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5667 | { |
1da177e4 LT |
5668 | might_sleep(); |
5669 | ||
5670 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5671 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5672 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5673 | return timeout; |
5674 | } | |
1da177e4 | 5675 | |
65eb3dc6 KD |
5676 | /** |
5677 | * wait_for_completion: - waits for completion of a task | |
5678 | * @x: holds the state of this particular completion | |
5679 | * | |
5680 | * This waits to be signaled for completion of a specific task. It is NOT | |
5681 | * interruptible and there is no timeout. | |
5682 | * | |
5683 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5684 | * and interrupt capability. Also see complete(). | |
5685 | */ | |
b15136e9 | 5686 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5687 | { |
5688 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5689 | } |
8cbbe86d | 5690 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5691 | |
65eb3dc6 KD |
5692 | /** |
5693 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5694 | * @x: holds the state of this particular completion | |
5695 | * @timeout: timeout value in jiffies | |
5696 | * | |
5697 | * This waits for either a completion of a specific task to be signaled or for a | |
5698 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5699 | * interruptible. | |
5700 | */ | |
b15136e9 | 5701 | unsigned long __sched |
8cbbe86d | 5702 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5703 | { |
8cbbe86d | 5704 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5705 | } |
8cbbe86d | 5706 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5707 | |
65eb3dc6 KD |
5708 | /** |
5709 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5710 | * @x: holds the state of this particular completion | |
5711 | * | |
5712 | * This waits for completion of a specific task to be signaled. It is | |
5713 | * interruptible. | |
5714 | */ | |
8cbbe86d | 5715 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5716 | { |
51e97990 AK |
5717 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5718 | if (t == -ERESTARTSYS) | |
5719 | return t; | |
5720 | return 0; | |
0fec171c | 5721 | } |
8cbbe86d | 5722 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5723 | |
65eb3dc6 KD |
5724 | /** |
5725 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5726 | * @x: holds the state of this particular completion | |
5727 | * @timeout: timeout value in jiffies | |
5728 | * | |
5729 | * This waits for either a completion of a specific task to be signaled or for a | |
5730 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5731 | */ | |
b15136e9 | 5732 | unsigned long __sched |
8cbbe86d AK |
5733 | wait_for_completion_interruptible_timeout(struct completion *x, |
5734 | unsigned long timeout) | |
0fec171c | 5735 | { |
8cbbe86d | 5736 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5737 | } |
8cbbe86d | 5738 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5739 | |
65eb3dc6 KD |
5740 | /** |
5741 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5742 | * @x: holds the state of this particular completion | |
5743 | * | |
5744 | * This waits to be signaled for completion of a specific task. It can be | |
5745 | * interrupted by a kill signal. | |
5746 | */ | |
009e577e MW |
5747 | int __sched wait_for_completion_killable(struct completion *x) |
5748 | { | |
5749 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5750 | if (t == -ERESTARTSYS) | |
5751 | return t; | |
5752 | return 0; | |
5753 | } | |
5754 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5755 | ||
be4de352 DC |
5756 | /** |
5757 | * try_wait_for_completion - try to decrement a completion without blocking | |
5758 | * @x: completion structure | |
5759 | * | |
5760 | * Returns: 0 if a decrement cannot be done without blocking | |
5761 | * 1 if a decrement succeeded. | |
5762 | * | |
5763 | * If a completion is being used as a counting completion, | |
5764 | * attempt to decrement the counter without blocking. This | |
5765 | * enables us to avoid waiting if the resource the completion | |
5766 | * is protecting is not available. | |
5767 | */ | |
5768 | bool try_wait_for_completion(struct completion *x) | |
5769 | { | |
5770 | int ret = 1; | |
5771 | ||
5772 | spin_lock_irq(&x->wait.lock); | |
5773 | if (!x->done) | |
5774 | ret = 0; | |
5775 | else | |
5776 | x->done--; | |
5777 | spin_unlock_irq(&x->wait.lock); | |
5778 | return ret; | |
5779 | } | |
5780 | EXPORT_SYMBOL(try_wait_for_completion); | |
5781 | ||
5782 | /** | |
5783 | * completion_done - Test to see if a completion has any waiters | |
5784 | * @x: completion structure | |
5785 | * | |
5786 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5787 | * 1 if there are no waiters. | |
5788 | * | |
5789 | */ | |
5790 | bool completion_done(struct completion *x) | |
5791 | { | |
5792 | int ret = 1; | |
5793 | ||
5794 | spin_lock_irq(&x->wait.lock); | |
5795 | if (!x->done) | |
5796 | ret = 0; | |
5797 | spin_unlock_irq(&x->wait.lock); | |
5798 | return ret; | |
5799 | } | |
5800 | EXPORT_SYMBOL(completion_done); | |
5801 | ||
8cbbe86d AK |
5802 | static long __sched |
5803 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5804 | { |
0fec171c IM |
5805 | unsigned long flags; |
5806 | wait_queue_t wait; | |
5807 | ||
5808 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5809 | |
8cbbe86d | 5810 | __set_current_state(state); |
1da177e4 | 5811 | |
8cbbe86d AK |
5812 | spin_lock_irqsave(&q->lock, flags); |
5813 | __add_wait_queue(q, &wait); | |
5814 | spin_unlock(&q->lock); | |
5815 | timeout = schedule_timeout(timeout); | |
5816 | spin_lock_irq(&q->lock); | |
5817 | __remove_wait_queue(q, &wait); | |
5818 | spin_unlock_irqrestore(&q->lock, flags); | |
5819 | ||
5820 | return timeout; | |
5821 | } | |
5822 | ||
5823 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5824 | { | |
5825 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5826 | } |
1da177e4 LT |
5827 | EXPORT_SYMBOL(interruptible_sleep_on); |
5828 | ||
0fec171c | 5829 | long __sched |
95cdf3b7 | 5830 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5831 | { |
8cbbe86d | 5832 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5833 | } |
1da177e4 LT |
5834 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5835 | ||
0fec171c | 5836 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5837 | { |
8cbbe86d | 5838 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5839 | } |
1da177e4 LT |
5840 | EXPORT_SYMBOL(sleep_on); |
5841 | ||
0fec171c | 5842 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5843 | { |
8cbbe86d | 5844 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5845 | } |
1da177e4 LT |
5846 | EXPORT_SYMBOL(sleep_on_timeout); |
5847 | ||
b29739f9 IM |
5848 | #ifdef CONFIG_RT_MUTEXES |
5849 | ||
5850 | /* | |
5851 | * rt_mutex_setprio - set the current priority of a task | |
5852 | * @p: task | |
5853 | * @prio: prio value (kernel-internal form) | |
5854 | * | |
5855 | * This function changes the 'effective' priority of a task. It does | |
5856 | * not touch ->normal_prio like __setscheduler(). | |
5857 | * | |
5858 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5859 | */ | |
36c8b586 | 5860 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5861 | { |
5862 | unsigned long flags; | |
83b699ed | 5863 | int oldprio, on_rq, running; |
70b97a7f | 5864 | struct rq *rq; |
cb469845 | 5865 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5866 | |
5867 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5868 | ||
5869 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5870 | update_rq_clock(rq); |
b29739f9 | 5871 | |
d5f9f942 | 5872 | oldprio = p->prio; |
dd41f596 | 5873 | on_rq = p->se.on_rq; |
051a1d1a | 5874 | running = task_current(rq, p); |
0e1f3483 | 5875 | if (on_rq) |
69be72c1 | 5876 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5877 | if (running) |
5878 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5879 | |
5880 | if (rt_prio(prio)) | |
5881 | p->sched_class = &rt_sched_class; | |
5882 | else | |
5883 | p->sched_class = &fair_sched_class; | |
5884 | ||
b29739f9 IM |
5885 | p->prio = prio; |
5886 | ||
0e1f3483 HS |
5887 | if (running) |
5888 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5889 | if (on_rq) { |
8159f87e | 5890 | enqueue_task(rq, p, 0); |
cb469845 SR |
5891 | |
5892 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5893 | } |
5894 | task_rq_unlock(rq, &flags); | |
5895 | } | |
5896 | ||
5897 | #endif | |
5898 | ||
36c8b586 | 5899 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5900 | { |
dd41f596 | 5901 | int old_prio, delta, on_rq; |
1da177e4 | 5902 | unsigned long flags; |
70b97a7f | 5903 | struct rq *rq; |
1da177e4 LT |
5904 | |
5905 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5906 | return; | |
5907 | /* | |
5908 | * We have to be careful, if called from sys_setpriority(), | |
5909 | * the task might be in the middle of scheduling on another CPU. | |
5910 | */ | |
5911 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5912 | update_rq_clock(rq); |
1da177e4 LT |
5913 | /* |
5914 | * The RT priorities are set via sched_setscheduler(), but we still | |
5915 | * allow the 'normal' nice value to be set - but as expected | |
5916 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5917 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5918 | */ |
e05606d3 | 5919 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5920 | p->static_prio = NICE_TO_PRIO(nice); |
5921 | goto out_unlock; | |
5922 | } | |
dd41f596 | 5923 | on_rq = p->se.on_rq; |
c09595f6 | 5924 | if (on_rq) |
69be72c1 | 5925 | dequeue_task(rq, p, 0); |
1da177e4 | 5926 | |
1da177e4 | 5927 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5928 | set_load_weight(p); |
b29739f9 IM |
5929 | old_prio = p->prio; |
5930 | p->prio = effective_prio(p); | |
5931 | delta = p->prio - old_prio; | |
1da177e4 | 5932 | |
dd41f596 | 5933 | if (on_rq) { |
8159f87e | 5934 | enqueue_task(rq, p, 0); |
1da177e4 | 5935 | /* |
d5f9f942 AM |
5936 | * If the task increased its priority or is running and |
5937 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5938 | */ |
d5f9f942 | 5939 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5940 | resched_task(rq->curr); |
5941 | } | |
5942 | out_unlock: | |
5943 | task_rq_unlock(rq, &flags); | |
5944 | } | |
1da177e4 LT |
5945 | EXPORT_SYMBOL(set_user_nice); |
5946 | ||
e43379f1 MM |
5947 | /* |
5948 | * can_nice - check if a task can reduce its nice value | |
5949 | * @p: task | |
5950 | * @nice: nice value | |
5951 | */ | |
36c8b586 | 5952 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5953 | { |
024f4747 MM |
5954 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5955 | int nice_rlim = 20 - nice; | |
48f24c4d | 5956 | |
e43379f1 MM |
5957 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5958 | capable(CAP_SYS_NICE)); | |
5959 | } | |
5960 | ||
1da177e4 LT |
5961 | #ifdef __ARCH_WANT_SYS_NICE |
5962 | ||
5963 | /* | |
5964 | * sys_nice - change the priority of the current process. | |
5965 | * @increment: priority increment | |
5966 | * | |
5967 | * sys_setpriority is a more generic, but much slower function that | |
5968 | * does similar things. | |
5969 | */ | |
5add95d4 | 5970 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5971 | { |
48f24c4d | 5972 | long nice, retval; |
1da177e4 LT |
5973 | |
5974 | /* | |
5975 | * Setpriority might change our priority at the same moment. | |
5976 | * We don't have to worry. Conceptually one call occurs first | |
5977 | * and we have a single winner. | |
5978 | */ | |
e43379f1 MM |
5979 | if (increment < -40) |
5980 | increment = -40; | |
1da177e4 LT |
5981 | if (increment > 40) |
5982 | increment = 40; | |
5983 | ||
2b8f836f | 5984 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5985 | if (nice < -20) |
5986 | nice = -20; | |
5987 | if (nice > 19) | |
5988 | nice = 19; | |
5989 | ||
e43379f1 MM |
5990 | if (increment < 0 && !can_nice(current, nice)) |
5991 | return -EPERM; | |
5992 | ||
1da177e4 LT |
5993 | retval = security_task_setnice(current, nice); |
5994 | if (retval) | |
5995 | return retval; | |
5996 | ||
5997 | set_user_nice(current, nice); | |
5998 | return 0; | |
5999 | } | |
6000 | ||
6001 | #endif | |
6002 | ||
6003 | /** | |
6004 | * task_prio - return the priority value of a given task. | |
6005 | * @p: the task in question. | |
6006 | * | |
6007 | * This is the priority value as seen by users in /proc. | |
6008 | * RT tasks are offset by -200. Normal tasks are centered | |
6009 | * around 0, value goes from -16 to +15. | |
6010 | */ | |
36c8b586 | 6011 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6012 | { |
6013 | return p->prio - MAX_RT_PRIO; | |
6014 | } | |
6015 | ||
6016 | /** | |
6017 | * task_nice - return the nice value of a given task. | |
6018 | * @p: the task in question. | |
6019 | */ | |
36c8b586 | 6020 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6021 | { |
6022 | return TASK_NICE(p); | |
6023 | } | |
150d8bed | 6024 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6025 | |
6026 | /** | |
6027 | * idle_cpu - is a given cpu idle currently? | |
6028 | * @cpu: the processor in question. | |
6029 | */ | |
6030 | int idle_cpu(int cpu) | |
6031 | { | |
6032 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6033 | } | |
6034 | ||
1da177e4 LT |
6035 | /** |
6036 | * idle_task - return the idle task for a given cpu. | |
6037 | * @cpu: the processor in question. | |
6038 | */ | |
36c8b586 | 6039 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6040 | { |
6041 | return cpu_rq(cpu)->idle; | |
6042 | } | |
6043 | ||
6044 | /** | |
6045 | * find_process_by_pid - find a process with a matching PID value. | |
6046 | * @pid: the pid in question. | |
6047 | */ | |
a9957449 | 6048 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6049 | { |
228ebcbe | 6050 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6051 | } |
6052 | ||
6053 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6054 | static void |
6055 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6056 | { |
dd41f596 | 6057 | BUG_ON(p->se.on_rq); |
48f24c4d | 6058 | |
1da177e4 | 6059 | p->policy = policy; |
dd41f596 IM |
6060 | switch (p->policy) { |
6061 | case SCHED_NORMAL: | |
6062 | case SCHED_BATCH: | |
6063 | case SCHED_IDLE: | |
6064 | p->sched_class = &fair_sched_class; | |
6065 | break; | |
6066 | case SCHED_FIFO: | |
6067 | case SCHED_RR: | |
6068 | p->sched_class = &rt_sched_class; | |
6069 | break; | |
6070 | } | |
6071 | ||
1da177e4 | 6072 | p->rt_priority = prio; |
b29739f9 IM |
6073 | p->normal_prio = normal_prio(p); |
6074 | /* we are holding p->pi_lock already */ | |
6075 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6076 | set_load_weight(p); |
1da177e4 LT |
6077 | } |
6078 | ||
c69e8d9c DH |
6079 | /* |
6080 | * check the target process has a UID that matches the current process's | |
6081 | */ | |
6082 | static bool check_same_owner(struct task_struct *p) | |
6083 | { | |
6084 | const struct cred *cred = current_cred(), *pcred; | |
6085 | bool match; | |
6086 | ||
6087 | rcu_read_lock(); | |
6088 | pcred = __task_cred(p); | |
6089 | match = (cred->euid == pcred->euid || | |
6090 | cred->euid == pcred->uid); | |
6091 | rcu_read_unlock(); | |
6092 | return match; | |
6093 | } | |
6094 | ||
961ccddd RR |
6095 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6096 | struct sched_param *param, bool user) | |
1da177e4 | 6097 | { |
83b699ed | 6098 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6099 | unsigned long flags; |
cb469845 | 6100 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6101 | struct rq *rq; |
1da177e4 | 6102 | |
66e5393a SR |
6103 | /* may grab non-irq protected spin_locks */ |
6104 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6105 | recheck: |
6106 | /* double check policy once rq lock held */ | |
6107 | if (policy < 0) | |
6108 | policy = oldpolicy = p->policy; | |
6109 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
6110 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
6111 | policy != SCHED_IDLE) | |
b0a9499c | 6112 | return -EINVAL; |
1da177e4 LT |
6113 | /* |
6114 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6115 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6116 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6117 | */ |
6118 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6119 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6120 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6121 | return -EINVAL; |
e05606d3 | 6122 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6123 | return -EINVAL; |
6124 | ||
37e4ab3f OC |
6125 | /* |
6126 | * Allow unprivileged RT tasks to decrease priority: | |
6127 | */ | |
961ccddd | 6128 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6129 | if (rt_policy(policy)) { |
8dc3e909 | 6130 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6131 | |
6132 | if (!lock_task_sighand(p, &flags)) | |
6133 | return -ESRCH; | |
6134 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6135 | unlock_task_sighand(p, &flags); | |
6136 | ||
6137 | /* can't set/change the rt policy */ | |
6138 | if (policy != p->policy && !rlim_rtprio) | |
6139 | return -EPERM; | |
6140 | ||
6141 | /* can't increase priority */ | |
6142 | if (param->sched_priority > p->rt_priority && | |
6143 | param->sched_priority > rlim_rtprio) | |
6144 | return -EPERM; | |
6145 | } | |
dd41f596 IM |
6146 | /* |
6147 | * Like positive nice levels, dont allow tasks to | |
6148 | * move out of SCHED_IDLE either: | |
6149 | */ | |
6150 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6151 | return -EPERM; | |
5fe1d75f | 6152 | |
37e4ab3f | 6153 | /* can't change other user's priorities */ |
c69e8d9c | 6154 | if (!check_same_owner(p)) |
37e4ab3f OC |
6155 | return -EPERM; |
6156 | } | |
1da177e4 | 6157 | |
725aad24 | 6158 | if (user) { |
b68aa230 | 6159 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6160 | /* |
6161 | * Do not allow realtime tasks into groups that have no runtime | |
6162 | * assigned. | |
6163 | */ | |
9a7e0b18 PZ |
6164 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6165 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6166 | return -EPERM; |
b68aa230 PZ |
6167 | #endif |
6168 | ||
725aad24 JF |
6169 | retval = security_task_setscheduler(p, policy, param); |
6170 | if (retval) | |
6171 | return retval; | |
6172 | } | |
6173 | ||
b29739f9 IM |
6174 | /* |
6175 | * make sure no PI-waiters arrive (or leave) while we are | |
6176 | * changing the priority of the task: | |
6177 | */ | |
6178 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6179 | /* |
6180 | * To be able to change p->policy safely, the apropriate | |
6181 | * runqueue lock must be held. | |
6182 | */ | |
b29739f9 | 6183 | rq = __task_rq_lock(p); |
1da177e4 LT |
6184 | /* recheck policy now with rq lock held */ |
6185 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6186 | policy = oldpolicy = -1; | |
b29739f9 IM |
6187 | __task_rq_unlock(rq); |
6188 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6189 | goto recheck; |
6190 | } | |
2daa3577 | 6191 | update_rq_clock(rq); |
dd41f596 | 6192 | on_rq = p->se.on_rq; |
051a1d1a | 6193 | running = task_current(rq, p); |
0e1f3483 | 6194 | if (on_rq) |
2e1cb74a | 6195 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6196 | if (running) |
6197 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6198 | |
1da177e4 | 6199 | oldprio = p->prio; |
dd41f596 | 6200 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6201 | |
0e1f3483 HS |
6202 | if (running) |
6203 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6204 | if (on_rq) { |
6205 | activate_task(rq, p, 0); | |
cb469845 SR |
6206 | |
6207 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6208 | } |
b29739f9 IM |
6209 | __task_rq_unlock(rq); |
6210 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6211 | ||
95e02ca9 TG |
6212 | rt_mutex_adjust_pi(p); |
6213 | ||
1da177e4 LT |
6214 | return 0; |
6215 | } | |
961ccddd RR |
6216 | |
6217 | /** | |
6218 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6219 | * @p: the task in question. | |
6220 | * @policy: new policy. | |
6221 | * @param: structure containing the new RT priority. | |
6222 | * | |
6223 | * NOTE that the task may be already dead. | |
6224 | */ | |
6225 | int sched_setscheduler(struct task_struct *p, int policy, | |
6226 | struct sched_param *param) | |
6227 | { | |
6228 | return __sched_setscheduler(p, policy, param, true); | |
6229 | } | |
1da177e4 LT |
6230 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6231 | ||
961ccddd RR |
6232 | /** |
6233 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6234 | * @p: the task in question. | |
6235 | * @policy: new policy. | |
6236 | * @param: structure containing the new RT priority. | |
6237 | * | |
6238 | * Just like sched_setscheduler, only don't bother checking if the | |
6239 | * current context has permission. For example, this is needed in | |
6240 | * stop_machine(): we create temporary high priority worker threads, | |
6241 | * but our caller might not have that capability. | |
6242 | */ | |
6243 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6244 | struct sched_param *param) | |
6245 | { | |
6246 | return __sched_setscheduler(p, policy, param, false); | |
6247 | } | |
6248 | ||
95cdf3b7 IM |
6249 | static int |
6250 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6251 | { |
1da177e4 LT |
6252 | struct sched_param lparam; |
6253 | struct task_struct *p; | |
36c8b586 | 6254 | int retval; |
1da177e4 LT |
6255 | |
6256 | if (!param || pid < 0) | |
6257 | return -EINVAL; | |
6258 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6259 | return -EFAULT; | |
5fe1d75f ON |
6260 | |
6261 | rcu_read_lock(); | |
6262 | retval = -ESRCH; | |
1da177e4 | 6263 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6264 | if (p != NULL) |
6265 | retval = sched_setscheduler(p, policy, &lparam); | |
6266 | rcu_read_unlock(); | |
36c8b586 | 6267 | |
1da177e4 LT |
6268 | return retval; |
6269 | } | |
6270 | ||
6271 | /** | |
6272 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6273 | * @pid: the pid in question. | |
6274 | * @policy: new policy. | |
6275 | * @param: structure containing the new RT priority. | |
6276 | */ | |
5add95d4 HC |
6277 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6278 | struct sched_param __user *, param) | |
1da177e4 | 6279 | { |
c21761f1 JB |
6280 | /* negative values for policy are not valid */ |
6281 | if (policy < 0) | |
6282 | return -EINVAL; | |
6283 | ||
1da177e4 LT |
6284 | return do_sched_setscheduler(pid, policy, param); |
6285 | } | |
6286 | ||
6287 | /** | |
6288 | * sys_sched_setparam - set/change the RT priority of a thread | |
6289 | * @pid: the pid in question. | |
6290 | * @param: structure containing the new RT priority. | |
6291 | */ | |
5add95d4 | 6292 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6293 | { |
6294 | return do_sched_setscheduler(pid, -1, param); | |
6295 | } | |
6296 | ||
6297 | /** | |
6298 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6299 | * @pid: the pid in question. | |
6300 | */ | |
5add95d4 | 6301 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6302 | { |
36c8b586 | 6303 | struct task_struct *p; |
3a5c359a | 6304 | int retval; |
1da177e4 LT |
6305 | |
6306 | if (pid < 0) | |
3a5c359a | 6307 | return -EINVAL; |
1da177e4 LT |
6308 | |
6309 | retval = -ESRCH; | |
6310 | read_lock(&tasklist_lock); | |
6311 | p = find_process_by_pid(pid); | |
6312 | if (p) { | |
6313 | retval = security_task_getscheduler(p); | |
6314 | if (!retval) | |
6315 | retval = p->policy; | |
6316 | } | |
6317 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6318 | return retval; |
6319 | } | |
6320 | ||
6321 | /** | |
6322 | * sys_sched_getscheduler - get the RT priority of a thread | |
6323 | * @pid: the pid in question. | |
6324 | * @param: structure containing the RT priority. | |
6325 | */ | |
5add95d4 | 6326 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6327 | { |
6328 | struct sched_param lp; | |
36c8b586 | 6329 | struct task_struct *p; |
3a5c359a | 6330 | int retval; |
1da177e4 LT |
6331 | |
6332 | if (!param || pid < 0) | |
3a5c359a | 6333 | return -EINVAL; |
1da177e4 LT |
6334 | |
6335 | read_lock(&tasklist_lock); | |
6336 | p = find_process_by_pid(pid); | |
6337 | retval = -ESRCH; | |
6338 | if (!p) | |
6339 | goto out_unlock; | |
6340 | ||
6341 | retval = security_task_getscheduler(p); | |
6342 | if (retval) | |
6343 | goto out_unlock; | |
6344 | ||
6345 | lp.sched_priority = p->rt_priority; | |
6346 | read_unlock(&tasklist_lock); | |
6347 | ||
6348 | /* | |
6349 | * This one might sleep, we cannot do it with a spinlock held ... | |
6350 | */ | |
6351 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6352 | ||
1da177e4 LT |
6353 | return retval; |
6354 | ||
6355 | out_unlock: | |
6356 | read_unlock(&tasklist_lock); | |
6357 | return retval; | |
6358 | } | |
6359 | ||
96f874e2 | 6360 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6361 | { |
5a16f3d3 | 6362 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6363 | struct task_struct *p; |
6364 | int retval; | |
1da177e4 | 6365 | |
95402b38 | 6366 | get_online_cpus(); |
1da177e4 LT |
6367 | read_lock(&tasklist_lock); |
6368 | ||
6369 | p = find_process_by_pid(pid); | |
6370 | if (!p) { | |
6371 | read_unlock(&tasklist_lock); | |
95402b38 | 6372 | put_online_cpus(); |
1da177e4 LT |
6373 | return -ESRCH; |
6374 | } | |
6375 | ||
6376 | /* | |
6377 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6378 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6379 | * usage count and then drop tasklist_lock. |
6380 | */ | |
6381 | get_task_struct(p); | |
6382 | read_unlock(&tasklist_lock); | |
6383 | ||
5a16f3d3 RR |
6384 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6385 | retval = -ENOMEM; | |
6386 | goto out_put_task; | |
6387 | } | |
6388 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6389 | retval = -ENOMEM; | |
6390 | goto out_free_cpus_allowed; | |
6391 | } | |
1da177e4 | 6392 | retval = -EPERM; |
c69e8d9c | 6393 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6394 | goto out_unlock; |
6395 | ||
e7834f8f DQ |
6396 | retval = security_task_setscheduler(p, 0, NULL); |
6397 | if (retval) | |
6398 | goto out_unlock; | |
6399 | ||
5a16f3d3 RR |
6400 | cpuset_cpus_allowed(p, cpus_allowed); |
6401 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6402 | again: |
5a16f3d3 | 6403 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6404 | |
8707d8b8 | 6405 | if (!retval) { |
5a16f3d3 RR |
6406 | cpuset_cpus_allowed(p, cpus_allowed); |
6407 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6408 | /* |
6409 | * We must have raced with a concurrent cpuset | |
6410 | * update. Just reset the cpus_allowed to the | |
6411 | * cpuset's cpus_allowed | |
6412 | */ | |
5a16f3d3 | 6413 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6414 | goto again; |
6415 | } | |
6416 | } | |
1da177e4 | 6417 | out_unlock: |
5a16f3d3 RR |
6418 | free_cpumask_var(new_mask); |
6419 | out_free_cpus_allowed: | |
6420 | free_cpumask_var(cpus_allowed); | |
6421 | out_put_task: | |
1da177e4 | 6422 | put_task_struct(p); |
95402b38 | 6423 | put_online_cpus(); |
1da177e4 LT |
6424 | return retval; |
6425 | } | |
6426 | ||
6427 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6428 | struct cpumask *new_mask) |
1da177e4 | 6429 | { |
96f874e2 RR |
6430 | if (len < cpumask_size()) |
6431 | cpumask_clear(new_mask); | |
6432 | else if (len > cpumask_size()) | |
6433 | len = cpumask_size(); | |
6434 | ||
1da177e4 LT |
6435 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6436 | } | |
6437 | ||
6438 | /** | |
6439 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6440 | * @pid: pid of the process | |
6441 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6442 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6443 | */ | |
5add95d4 HC |
6444 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6445 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6446 | { |
5a16f3d3 | 6447 | cpumask_var_t new_mask; |
1da177e4 LT |
6448 | int retval; |
6449 | ||
5a16f3d3 RR |
6450 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6451 | return -ENOMEM; | |
1da177e4 | 6452 | |
5a16f3d3 RR |
6453 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6454 | if (retval == 0) | |
6455 | retval = sched_setaffinity(pid, new_mask); | |
6456 | free_cpumask_var(new_mask); | |
6457 | return retval; | |
1da177e4 LT |
6458 | } |
6459 | ||
96f874e2 | 6460 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6461 | { |
36c8b586 | 6462 | struct task_struct *p; |
1da177e4 | 6463 | int retval; |
1da177e4 | 6464 | |
95402b38 | 6465 | get_online_cpus(); |
1da177e4 LT |
6466 | read_lock(&tasklist_lock); |
6467 | ||
6468 | retval = -ESRCH; | |
6469 | p = find_process_by_pid(pid); | |
6470 | if (!p) | |
6471 | goto out_unlock; | |
6472 | ||
e7834f8f DQ |
6473 | retval = security_task_getscheduler(p); |
6474 | if (retval) | |
6475 | goto out_unlock; | |
6476 | ||
96f874e2 | 6477 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6478 | |
6479 | out_unlock: | |
6480 | read_unlock(&tasklist_lock); | |
95402b38 | 6481 | put_online_cpus(); |
1da177e4 | 6482 | |
9531b62f | 6483 | return retval; |
1da177e4 LT |
6484 | } |
6485 | ||
6486 | /** | |
6487 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6488 | * @pid: pid of the process | |
6489 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6490 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6491 | */ | |
5add95d4 HC |
6492 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6493 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6494 | { |
6495 | int ret; | |
f17c8607 | 6496 | cpumask_var_t mask; |
1da177e4 | 6497 | |
f17c8607 | 6498 | if (len < cpumask_size()) |
1da177e4 LT |
6499 | return -EINVAL; |
6500 | ||
f17c8607 RR |
6501 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6502 | return -ENOMEM; | |
1da177e4 | 6503 | |
f17c8607 RR |
6504 | ret = sched_getaffinity(pid, mask); |
6505 | if (ret == 0) { | |
6506 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6507 | ret = -EFAULT; | |
6508 | else | |
6509 | ret = cpumask_size(); | |
6510 | } | |
6511 | free_cpumask_var(mask); | |
1da177e4 | 6512 | |
f17c8607 | 6513 | return ret; |
1da177e4 LT |
6514 | } |
6515 | ||
6516 | /** | |
6517 | * sys_sched_yield - yield the current processor to other threads. | |
6518 | * | |
dd41f596 IM |
6519 | * This function yields the current CPU to other tasks. If there are no |
6520 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6521 | */ |
5add95d4 | 6522 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6523 | { |
70b97a7f | 6524 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6525 | |
2d72376b | 6526 | schedstat_inc(rq, yld_count); |
4530d7ab | 6527 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6528 | |
6529 | /* | |
6530 | * Since we are going to call schedule() anyway, there's | |
6531 | * no need to preempt or enable interrupts: | |
6532 | */ | |
6533 | __release(rq->lock); | |
8a25d5de | 6534 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6535 | _raw_spin_unlock(&rq->lock); |
6536 | preempt_enable_no_resched(); | |
6537 | ||
6538 | schedule(); | |
6539 | ||
6540 | return 0; | |
6541 | } | |
6542 | ||
e7b38404 | 6543 | static void __cond_resched(void) |
1da177e4 | 6544 | { |
8e0a43d8 IM |
6545 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6546 | __might_sleep(__FILE__, __LINE__); | |
6547 | #endif | |
5bbcfd90 IM |
6548 | /* |
6549 | * The BKS might be reacquired before we have dropped | |
6550 | * PREEMPT_ACTIVE, which could trigger a second | |
6551 | * cond_resched() call. | |
6552 | */ | |
1da177e4 LT |
6553 | do { |
6554 | add_preempt_count(PREEMPT_ACTIVE); | |
6555 | schedule(); | |
6556 | sub_preempt_count(PREEMPT_ACTIVE); | |
6557 | } while (need_resched()); | |
6558 | } | |
6559 | ||
02b67cc3 | 6560 | int __sched _cond_resched(void) |
1da177e4 | 6561 | { |
9414232f IM |
6562 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6563 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6564 | __cond_resched(); |
6565 | return 1; | |
6566 | } | |
6567 | return 0; | |
6568 | } | |
02b67cc3 | 6569 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6570 | |
6571 | /* | |
6572 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6573 | * call schedule, and on return reacquire the lock. | |
6574 | * | |
41a2d6cf | 6575 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6576 | * operations here to prevent schedule() from being called twice (once via |
6577 | * spin_unlock(), once by hand). | |
6578 | */ | |
95cdf3b7 | 6579 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6580 | { |
95c354fe | 6581 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6582 | int ret = 0; |
6583 | ||
95c354fe | 6584 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6585 | spin_unlock(lock); |
95c354fe NP |
6586 | if (resched && need_resched()) |
6587 | __cond_resched(); | |
6588 | else | |
6589 | cpu_relax(); | |
6df3cecb | 6590 | ret = 1; |
1da177e4 | 6591 | spin_lock(lock); |
1da177e4 | 6592 | } |
6df3cecb | 6593 | return ret; |
1da177e4 | 6594 | } |
1da177e4 LT |
6595 | EXPORT_SYMBOL(cond_resched_lock); |
6596 | ||
6597 | int __sched cond_resched_softirq(void) | |
6598 | { | |
6599 | BUG_ON(!in_softirq()); | |
6600 | ||
9414232f | 6601 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6602 | local_bh_enable(); |
1da177e4 LT |
6603 | __cond_resched(); |
6604 | local_bh_disable(); | |
6605 | return 1; | |
6606 | } | |
6607 | return 0; | |
6608 | } | |
1da177e4 LT |
6609 | EXPORT_SYMBOL(cond_resched_softirq); |
6610 | ||
1da177e4 LT |
6611 | /** |
6612 | * yield - yield the current processor to other threads. | |
6613 | * | |
72fd4a35 | 6614 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6615 | * thread runnable and calls sys_sched_yield(). |
6616 | */ | |
6617 | void __sched yield(void) | |
6618 | { | |
6619 | set_current_state(TASK_RUNNING); | |
6620 | sys_sched_yield(); | |
6621 | } | |
1da177e4 LT |
6622 | EXPORT_SYMBOL(yield); |
6623 | ||
6624 | /* | |
41a2d6cf | 6625 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6626 | * that process accounting knows that this is a task in IO wait state. |
6627 | * | |
6628 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6629 | * has set its backing_dev_info: the queue against which it should throttle) | |
6630 | */ | |
6631 | void __sched io_schedule(void) | |
6632 | { | |
70b97a7f | 6633 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6634 | |
0ff92245 | 6635 | delayacct_blkio_start(); |
1da177e4 LT |
6636 | atomic_inc(&rq->nr_iowait); |
6637 | schedule(); | |
6638 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6639 | delayacct_blkio_end(); |
1da177e4 | 6640 | } |
1da177e4 LT |
6641 | EXPORT_SYMBOL(io_schedule); |
6642 | ||
6643 | long __sched io_schedule_timeout(long timeout) | |
6644 | { | |
70b97a7f | 6645 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6646 | long ret; |
6647 | ||
0ff92245 | 6648 | delayacct_blkio_start(); |
1da177e4 LT |
6649 | atomic_inc(&rq->nr_iowait); |
6650 | ret = schedule_timeout(timeout); | |
6651 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6652 | delayacct_blkio_end(); |
1da177e4 LT |
6653 | return ret; |
6654 | } | |
6655 | ||
6656 | /** | |
6657 | * sys_sched_get_priority_max - return maximum RT priority. | |
6658 | * @policy: scheduling class. | |
6659 | * | |
6660 | * this syscall returns the maximum rt_priority that can be used | |
6661 | * by a given scheduling class. | |
6662 | */ | |
5add95d4 | 6663 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6664 | { |
6665 | int ret = -EINVAL; | |
6666 | ||
6667 | switch (policy) { | |
6668 | case SCHED_FIFO: | |
6669 | case SCHED_RR: | |
6670 | ret = MAX_USER_RT_PRIO-1; | |
6671 | break; | |
6672 | case SCHED_NORMAL: | |
b0a9499c | 6673 | case SCHED_BATCH: |
dd41f596 | 6674 | case SCHED_IDLE: |
1da177e4 LT |
6675 | ret = 0; |
6676 | break; | |
6677 | } | |
6678 | return ret; | |
6679 | } | |
6680 | ||
6681 | /** | |
6682 | * sys_sched_get_priority_min - return minimum RT priority. | |
6683 | * @policy: scheduling class. | |
6684 | * | |
6685 | * this syscall returns the minimum rt_priority that can be used | |
6686 | * by a given scheduling class. | |
6687 | */ | |
5add95d4 | 6688 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6689 | { |
6690 | int ret = -EINVAL; | |
6691 | ||
6692 | switch (policy) { | |
6693 | case SCHED_FIFO: | |
6694 | case SCHED_RR: | |
6695 | ret = 1; | |
6696 | break; | |
6697 | case SCHED_NORMAL: | |
b0a9499c | 6698 | case SCHED_BATCH: |
dd41f596 | 6699 | case SCHED_IDLE: |
1da177e4 LT |
6700 | ret = 0; |
6701 | } | |
6702 | return ret; | |
6703 | } | |
6704 | ||
6705 | /** | |
6706 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6707 | * @pid: pid of the process. | |
6708 | * @interval: userspace pointer to the timeslice value. | |
6709 | * | |
6710 | * this syscall writes the default timeslice value of a given process | |
6711 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6712 | */ | |
17da2bd9 | 6713 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6714 | struct timespec __user *, interval) |
1da177e4 | 6715 | { |
36c8b586 | 6716 | struct task_struct *p; |
a4ec24b4 | 6717 | unsigned int time_slice; |
3a5c359a | 6718 | int retval; |
1da177e4 | 6719 | struct timespec t; |
1da177e4 LT |
6720 | |
6721 | if (pid < 0) | |
3a5c359a | 6722 | return -EINVAL; |
1da177e4 LT |
6723 | |
6724 | retval = -ESRCH; | |
6725 | read_lock(&tasklist_lock); | |
6726 | p = find_process_by_pid(pid); | |
6727 | if (!p) | |
6728 | goto out_unlock; | |
6729 | ||
6730 | retval = security_task_getscheduler(p); | |
6731 | if (retval) | |
6732 | goto out_unlock; | |
6733 | ||
77034937 IM |
6734 | /* |
6735 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6736 | * tasks that are on an otherwise idle runqueue: | |
6737 | */ | |
6738 | time_slice = 0; | |
6739 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6740 | time_slice = DEF_TIMESLICE; |
1868f958 | 6741 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6742 | struct sched_entity *se = &p->se; |
6743 | unsigned long flags; | |
6744 | struct rq *rq; | |
6745 | ||
6746 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6747 | if (rq->cfs.load.weight) |
6748 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6749 | task_rq_unlock(rq, &flags); |
6750 | } | |
1da177e4 | 6751 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6752 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6753 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6754 | return retval; |
3a5c359a | 6755 | |
1da177e4 LT |
6756 | out_unlock: |
6757 | read_unlock(&tasklist_lock); | |
6758 | return retval; | |
6759 | } | |
6760 | ||
7c731e0a | 6761 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6762 | |
82a1fcb9 | 6763 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6764 | { |
1da177e4 | 6765 | unsigned long free = 0; |
36c8b586 | 6766 | unsigned state; |
1da177e4 | 6767 | |
1da177e4 | 6768 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6769 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6770 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6771 | #if BITS_PER_LONG == 32 |
1da177e4 | 6772 | if (state == TASK_RUNNING) |
cc4ea795 | 6773 | printk(KERN_CONT " running "); |
1da177e4 | 6774 | else |
cc4ea795 | 6775 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6776 | #else |
6777 | if (state == TASK_RUNNING) | |
cc4ea795 | 6778 | printk(KERN_CONT " running task "); |
1da177e4 | 6779 | else |
cc4ea795 | 6780 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6781 | #endif |
6782 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6783 | free = stack_not_used(p); |
1da177e4 | 6784 | #endif |
aa47b7e0 DR |
6785 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6786 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6787 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6788 | |
5fb5e6de | 6789 | show_stack(p, NULL); |
1da177e4 LT |
6790 | } |
6791 | ||
e59e2ae2 | 6792 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6793 | { |
36c8b586 | 6794 | struct task_struct *g, *p; |
1da177e4 | 6795 | |
4bd77321 IM |
6796 | #if BITS_PER_LONG == 32 |
6797 | printk(KERN_INFO | |
6798 | " task PC stack pid father\n"); | |
1da177e4 | 6799 | #else |
4bd77321 IM |
6800 | printk(KERN_INFO |
6801 | " task PC stack pid father\n"); | |
1da177e4 LT |
6802 | #endif |
6803 | read_lock(&tasklist_lock); | |
6804 | do_each_thread(g, p) { | |
6805 | /* | |
6806 | * reset the NMI-timeout, listing all files on a slow | |
6807 | * console might take alot of time: | |
6808 | */ | |
6809 | touch_nmi_watchdog(); | |
39bc89fd | 6810 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6811 | sched_show_task(p); |
1da177e4 LT |
6812 | } while_each_thread(g, p); |
6813 | ||
04c9167f JF |
6814 | touch_all_softlockup_watchdogs(); |
6815 | ||
dd41f596 IM |
6816 | #ifdef CONFIG_SCHED_DEBUG |
6817 | sysrq_sched_debug_show(); | |
6818 | #endif | |
1da177e4 | 6819 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6820 | /* |
6821 | * Only show locks if all tasks are dumped: | |
6822 | */ | |
6823 | if (state_filter == -1) | |
6824 | debug_show_all_locks(); | |
1da177e4 LT |
6825 | } |
6826 | ||
1df21055 IM |
6827 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6828 | { | |
dd41f596 | 6829 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6830 | } |
6831 | ||
f340c0d1 IM |
6832 | /** |
6833 | * init_idle - set up an idle thread for a given CPU | |
6834 | * @idle: task in question | |
6835 | * @cpu: cpu the idle task belongs to | |
6836 | * | |
6837 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6838 | * flag, to make booting more robust. | |
6839 | */ | |
5c1e1767 | 6840 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6841 | { |
70b97a7f | 6842 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6843 | unsigned long flags; |
6844 | ||
5cbd54ef IM |
6845 | spin_lock_irqsave(&rq->lock, flags); |
6846 | ||
dd41f596 IM |
6847 | __sched_fork(idle); |
6848 | idle->se.exec_start = sched_clock(); | |
6849 | ||
b29739f9 | 6850 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6851 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6852 | __set_task_cpu(idle, cpu); |
1da177e4 | 6853 | |
1da177e4 | 6854 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6855 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6856 | idle->oncpu = 1; | |
6857 | #endif | |
1da177e4 LT |
6858 | spin_unlock_irqrestore(&rq->lock, flags); |
6859 | ||
6860 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6861 | #if defined(CONFIG_PREEMPT) |
6862 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6863 | #else | |
a1261f54 | 6864 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6865 | #endif |
dd41f596 IM |
6866 | /* |
6867 | * The idle tasks have their own, simple scheduling class: | |
6868 | */ | |
6869 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6870 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6871 | } |
6872 | ||
6873 | /* | |
6874 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6875 | * indicates which cpus entered this state. This is used | |
6876 | * in the rcu update to wait only for active cpus. For system | |
6877 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6878 | * always be CPU_BITS_NONE. |
1da177e4 | 6879 | */ |
6a7b3dc3 | 6880 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6881 | |
19978ca6 IM |
6882 | /* |
6883 | * Increase the granularity value when there are more CPUs, | |
6884 | * because with more CPUs the 'effective latency' as visible | |
6885 | * to users decreases. But the relationship is not linear, | |
6886 | * so pick a second-best guess by going with the log2 of the | |
6887 | * number of CPUs. | |
6888 | * | |
6889 | * This idea comes from the SD scheduler of Con Kolivas: | |
6890 | */ | |
6891 | static inline void sched_init_granularity(void) | |
6892 | { | |
6893 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6894 | const unsigned long limit = 200000000; | |
6895 | ||
6896 | sysctl_sched_min_granularity *= factor; | |
6897 | if (sysctl_sched_min_granularity > limit) | |
6898 | sysctl_sched_min_granularity = limit; | |
6899 | ||
6900 | sysctl_sched_latency *= factor; | |
6901 | if (sysctl_sched_latency > limit) | |
6902 | sysctl_sched_latency = limit; | |
6903 | ||
6904 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6905 | |
6906 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6907 | } |
6908 | ||
1da177e4 LT |
6909 | #ifdef CONFIG_SMP |
6910 | /* | |
6911 | * This is how migration works: | |
6912 | * | |
70b97a7f | 6913 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6914 | * runqueue and wake up that CPU's migration thread. |
6915 | * 2) we down() the locked semaphore => thread blocks. | |
6916 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6917 | * thread off the CPU) | |
6918 | * 4) it gets the migration request and checks whether the migrated | |
6919 | * task is still in the wrong runqueue. | |
6920 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6921 | * it and puts it into the right queue. | |
6922 | * 6) migration thread up()s the semaphore. | |
6923 | * 7) we wake up and the migration is done. | |
6924 | */ | |
6925 | ||
6926 | /* | |
6927 | * Change a given task's CPU affinity. Migrate the thread to a | |
6928 | * proper CPU and schedule it away if the CPU it's executing on | |
6929 | * is removed from the allowed bitmask. | |
6930 | * | |
6931 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6932 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6933 | * call is not atomic; no spinlocks may be held. |
6934 | */ | |
96f874e2 | 6935 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6936 | { |
70b97a7f | 6937 | struct migration_req req; |
1da177e4 | 6938 | unsigned long flags; |
70b97a7f | 6939 | struct rq *rq; |
48f24c4d | 6940 | int ret = 0; |
1da177e4 LT |
6941 | |
6942 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6943 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6944 | ret = -EINVAL; |
6945 | goto out; | |
6946 | } | |
6947 | ||
9985b0ba | 6948 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6949 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6950 | ret = -EINVAL; |
6951 | goto out; | |
6952 | } | |
6953 | ||
73fe6aae | 6954 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6955 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6956 | else { |
96f874e2 RR |
6957 | cpumask_copy(&p->cpus_allowed, new_mask); |
6958 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6959 | } |
6960 | ||
1da177e4 | 6961 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6962 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6963 | goto out; |
6964 | ||
1e5ce4f4 | 6965 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6966 | /* Need help from migration thread: drop lock and wait. */ |
6967 | task_rq_unlock(rq, &flags); | |
6968 | wake_up_process(rq->migration_thread); | |
6969 | wait_for_completion(&req.done); | |
6970 | tlb_migrate_finish(p->mm); | |
6971 | return 0; | |
6972 | } | |
6973 | out: | |
6974 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6975 | |
1da177e4 LT |
6976 | return ret; |
6977 | } | |
cd8ba7cd | 6978 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6979 | |
6980 | /* | |
41a2d6cf | 6981 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6982 | * this because either it can't run here any more (set_cpus_allowed() |
6983 | * away from this CPU, or CPU going down), or because we're | |
6984 | * attempting to rebalance this task on exec (sched_exec). | |
6985 | * | |
6986 | * So we race with normal scheduler movements, but that's OK, as long | |
6987 | * as the task is no longer on this CPU. | |
efc30814 KK |
6988 | * |
6989 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6990 | */ |
efc30814 | 6991 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6992 | { |
70b97a7f | 6993 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6994 | int ret = 0, on_rq; |
1da177e4 | 6995 | |
e761b772 | 6996 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6997 | return ret; |
1da177e4 LT |
6998 | |
6999 | rq_src = cpu_rq(src_cpu); | |
7000 | rq_dest = cpu_rq(dest_cpu); | |
7001 | ||
7002 | double_rq_lock(rq_src, rq_dest); | |
7003 | /* Already moved. */ | |
7004 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7005 | goto done; |
1da177e4 | 7006 | /* Affinity changed (again). */ |
96f874e2 | 7007 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7008 | goto fail; |
1da177e4 | 7009 | |
dd41f596 | 7010 | on_rq = p->se.on_rq; |
6e82a3be | 7011 | if (on_rq) |
2e1cb74a | 7012 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7013 | |
1da177e4 | 7014 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7015 | if (on_rq) { |
7016 | activate_task(rq_dest, p, 0); | |
15afe09b | 7017 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7018 | } |
b1e38734 | 7019 | done: |
efc30814 | 7020 | ret = 1; |
b1e38734 | 7021 | fail: |
1da177e4 | 7022 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7023 | return ret; |
1da177e4 LT |
7024 | } |
7025 | ||
7026 | /* | |
7027 | * migration_thread - this is a highprio system thread that performs | |
7028 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7029 | * another runqueue. | |
7030 | */ | |
95cdf3b7 | 7031 | static int migration_thread(void *data) |
1da177e4 | 7032 | { |
1da177e4 | 7033 | int cpu = (long)data; |
70b97a7f | 7034 | struct rq *rq; |
1da177e4 LT |
7035 | |
7036 | rq = cpu_rq(cpu); | |
7037 | BUG_ON(rq->migration_thread != current); | |
7038 | ||
7039 | set_current_state(TASK_INTERRUPTIBLE); | |
7040 | while (!kthread_should_stop()) { | |
70b97a7f | 7041 | struct migration_req *req; |
1da177e4 | 7042 | struct list_head *head; |
1da177e4 | 7043 | |
1da177e4 LT |
7044 | spin_lock_irq(&rq->lock); |
7045 | ||
7046 | if (cpu_is_offline(cpu)) { | |
7047 | spin_unlock_irq(&rq->lock); | |
7048 | goto wait_to_die; | |
7049 | } | |
7050 | ||
7051 | if (rq->active_balance) { | |
7052 | active_load_balance(rq, cpu); | |
7053 | rq->active_balance = 0; | |
7054 | } | |
7055 | ||
7056 | head = &rq->migration_queue; | |
7057 | ||
7058 | if (list_empty(head)) { | |
7059 | spin_unlock_irq(&rq->lock); | |
7060 | schedule(); | |
7061 | set_current_state(TASK_INTERRUPTIBLE); | |
7062 | continue; | |
7063 | } | |
70b97a7f | 7064 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7065 | list_del_init(head->next); |
7066 | ||
674311d5 NP |
7067 | spin_unlock(&rq->lock); |
7068 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7069 | local_irq_enable(); | |
1da177e4 LT |
7070 | |
7071 | complete(&req->done); | |
7072 | } | |
7073 | __set_current_state(TASK_RUNNING); | |
7074 | return 0; | |
7075 | ||
7076 | wait_to_die: | |
7077 | /* Wait for kthread_stop */ | |
7078 | set_current_state(TASK_INTERRUPTIBLE); | |
7079 | while (!kthread_should_stop()) { | |
7080 | schedule(); | |
7081 | set_current_state(TASK_INTERRUPTIBLE); | |
7082 | } | |
7083 | __set_current_state(TASK_RUNNING); | |
7084 | return 0; | |
7085 | } | |
7086 | ||
7087 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7088 | |
7089 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7090 | { | |
7091 | int ret; | |
7092 | ||
7093 | local_irq_disable(); | |
7094 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7095 | local_irq_enable(); | |
7096 | return ret; | |
7097 | } | |
7098 | ||
054b9108 | 7099 | /* |
3a4fa0a2 | 7100 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7101 | */ |
48f24c4d | 7102 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7103 | { |
70b97a7f | 7104 | int dest_cpu; |
6ca09dfc | 7105 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7106 | |
7107 | again: | |
7108 | /* Look for allowed, online CPU in same node. */ | |
7109 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7110 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7111 | goto move; | |
7112 | ||
7113 | /* Any allowed, online CPU? */ | |
7114 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7115 | if (dest_cpu < nr_cpu_ids) | |
7116 | goto move; | |
7117 | ||
7118 | /* No more Mr. Nice Guy. */ | |
7119 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7120 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7121 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7122 | |
e76bd8d9 RR |
7123 | /* |
7124 | * Don't tell them about moving exiting tasks or | |
7125 | * kernel threads (both mm NULL), since they never | |
7126 | * leave kernel. | |
7127 | */ | |
7128 | if (p->mm && printk_ratelimit()) { | |
7129 | printk(KERN_INFO "process %d (%s) no " | |
7130 | "longer affine to cpu%d\n", | |
7131 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7132 | } |
e76bd8d9 RR |
7133 | } |
7134 | ||
7135 | move: | |
7136 | /* It can have affinity changed while we were choosing. */ | |
7137 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7138 | goto again; | |
1da177e4 LT |
7139 | } |
7140 | ||
7141 | /* | |
7142 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7143 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7144 | * for performance reasons the counter is not stricly tracking tasks to | |
7145 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7146 | * to keep the global sum constant after CPU-down: | |
7147 | */ | |
70b97a7f | 7148 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7149 | { |
1e5ce4f4 | 7150 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7151 | unsigned long flags; |
7152 | ||
7153 | local_irq_save(flags); | |
7154 | double_rq_lock(rq_src, rq_dest); | |
7155 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7156 | rq_src->nr_uninterruptible = 0; | |
7157 | double_rq_unlock(rq_src, rq_dest); | |
7158 | local_irq_restore(flags); | |
7159 | } | |
7160 | ||
7161 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7162 | static void migrate_live_tasks(int src_cpu) | |
7163 | { | |
48f24c4d | 7164 | struct task_struct *p, *t; |
1da177e4 | 7165 | |
f7b4cddc | 7166 | read_lock(&tasklist_lock); |
1da177e4 | 7167 | |
48f24c4d IM |
7168 | do_each_thread(t, p) { |
7169 | if (p == current) | |
1da177e4 LT |
7170 | continue; |
7171 | ||
48f24c4d IM |
7172 | if (task_cpu(p) == src_cpu) |
7173 | move_task_off_dead_cpu(src_cpu, p); | |
7174 | } while_each_thread(t, p); | |
1da177e4 | 7175 | |
f7b4cddc | 7176 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7177 | } |
7178 | ||
dd41f596 IM |
7179 | /* |
7180 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7181 | * It does so by boosting its priority to highest possible. |
7182 | * Used by CPU offline code. | |
1da177e4 LT |
7183 | */ |
7184 | void sched_idle_next(void) | |
7185 | { | |
48f24c4d | 7186 | int this_cpu = smp_processor_id(); |
70b97a7f | 7187 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7188 | struct task_struct *p = rq->idle; |
7189 | unsigned long flags; | |
7190 | ||
7191 | /* cpu has to be offline */ | |
48f24c4d | 7192 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7193 | |
48f24c4d IM |
7194 | /* |
7195 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7196 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7197 | */ |
7198 | spin_lock_irqsave(&rq->lock, flags); | |
7199 | ||
dd41f596 | 7200 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7201 | |
94bc9a7b DA |
7202 | update_rq_clock(rq); |
7203 | activate_task(rq, p, 0); | |
1da177e4 LT |
7204 | |
7205 | spin_unlock_irqrestore(&rq->lock, flags); | |
7206 | } | |
7207 | ||
48f24c4d IM |
7208 | /* |
7209 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7210 | * offline. |
7211 | */ | |
7212 | void idle_task_exit(void) | |
7213 | { | |
7214 | struct mm_struct *mm = current->active_mm; | |
7215 | ||
7216 | BUG_ON(cpu_online(smp_processor_id())); | |
7217 | ||
7218 | if (mm != &init_mm) | |
7219 | switch_mm(mm, &init_mm, current); | |
7220 | mmdrop(mm); | |
7221 | } | |
7222 | ||
054b9108 | 7223 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7224 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7225 | { |
70b97a7f | 7226 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7227 | |
7228 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7229 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7230 | |
7231 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7232 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7233 | |
48f24c4d | 7234 | get_task_struct(p); |
1da177e4 LT |
7235 | |
7236 | /* | |
7237 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7238 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7239 | * fine. |
7240 | */ | |
f7b4cddc | 7241 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7242 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7243 | spin_lock_irq(&rq->lock); |
1da177e4 | 7244 | |
48f24c4d | 7245 | put_task_struct(p); |
1da177e4 LT |
7246 | } |
7247 | ||
7248 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7249 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7250 | { | |
70b97a7f | 7251 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7252 | struct task_struct *next; |
48f24c4d | 7253 | |
dd41f596 IM |
7254 | for ( ; ; ) { |
7255 | if (!rq->nr_running) | |
7256 | break; | |
a8e504d2 | 7257 | update_rq_clock(rq); |
b67802ea | 7258 | next = pick_next_task(rq); |
dd41f596 IM |
7259 | if (!next) |
7260 | break; | |
79c53799 | 7261 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7262 | migrate_dead(dead_cpu, next); |
e692ab53 | 7263 | |
1da177e4 LT |
7264 | } |
7265 | } | |
dce48a84 TG |
7266 | |
7267 | /* | |
7268 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7269 | */ | |
7270 | static void calc_global_load_remove(struct rq *rq) | |
7271 | { | |
7272 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7273 | } | |
1da177e4 LT |
7274 | #endif /* CONFIG_HOTPLUG_CPU */ |
7275 | ||
e692ab53 NP |
7276 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7277 | ||
7278 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7279 | { |
7280 | .procname = "sched_domain", | |
c57baf1e | 7281 | .mode = 0555, |
e0361851 | 7282 | }, |
38605cae | 7283 | {0, }, |
e692ab53 NP |
7284 | }; |
7285 | ||
7286 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7287 | { |
c57baf1e | 7288 | .ctl_name = CTL_KERN, |
e0361851 | 7289 | .procname = "kernel", |
c57baf1e | 7290 | .mode = 0555, |
e0361851 AD |
7291 | .child = sd_ctl_dir, |
7292 | }, | |
38605cae | 7293 | {0, }, |
e692ab53 NP |
7294 | }; |
7295 | ||
7296 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7297 | { | |
7298 | struct ctl_table *entry = | |
5cf9f062 | 7299 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7300 | |
e692ab53 NP |
7301 | return entry; |
7302 | } | |
7303 | ||
6382bc90 MM |
7304 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7305 | { | |
cd790076 | 7306 | struct ctl_table *entry; |
6382bc90 | 7307 | |
cd790076 MM |
7308 | /* |
7309 | * In the intermediate directories, both the child directory and | |
7310 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7311 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7312 | * static strings and all have proc handlers. |
7313 | */ | |
7314 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7315 | if (entry->child) |
7316 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7317 | if (entry->proc_handler == NULL) |
7318 | kfree(entry->procname); | |
7319 | } | |
6382bc90 MM |
7320 | |
7321 | kfree(*tablep); | |
7322 | *tablep = NULL; | |
7323 | } | |
7324 | ||
e692ab53 | 7325 | static void |
e0361851 | 7326 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7327 | const char *procname, void *data, int maxlen, |
7328 | mode_t mode, proc_handler *proc_handler) | |
7329 | { | |
e692ab53 NP |
7330 | entry->procname = procname; |
7331 | entry->data = data; | |
7332 | entry->maxlen = maxlen; | |
7333 | entry->mode = mode; | |
7334 | entry->proc_handler = proc_handler; | |
7335 | } | |
7336 | ||
7337 | static struct ctl_table * | |
7338 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7339 | { | |
a5d8c348 | 7340 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7341 | |
ad1cdc1d MM |
7342 | if (table == NULL) |
7343 | return NULL; | |
7344 | ||
e0361851 | 7345 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7346 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7347 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7348 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7349 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7350 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7351 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7352 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7353 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7354 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7355 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7356 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7357 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7358 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7359 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7360 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7361 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7362 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7363 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7364 | &sd->cache_nice_tries, |
7365 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7366 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7367 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7368 | set_table_entry(&table[11], "name", sd->name, |
7369 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7370 | /* &table[12] is terminator */ | |
e692ab53 NP |
7371 | |
7372 | return table; | |
7373 | } | |
7374 | ||
9a4e7159 | 7375 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7376 | { |
7377 | struct ctl_table *entry, *table; | |
7378 | struct sched_domain *sd; | |
7379 | int domain_num = 0, i; | |
7380 | char buf[32]; | |
7381 | ||
7382 | for_each_domain(cpu, sd) | |
7383 | domain_num++; | |
7384 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7385 | if (table == NULL) |
7386 | return NULL; | |
e692ab53 NP |
7387 | |
7388 | i = 0; | |
7389 | for_each_domain(cpu, sd) { | |
7390 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7391 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7392 | entry->mode = 0555; |
e692ab53 NP |
7393 | entry->child = sd_alloc_ctl_domain_table(sd); |
7394 | entry++; | |
7395 | i++; | |
7396 | } | |
7397 | return table; | |
7398 | } | |
7399 | ||
7400 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7401 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7402 | { |
7403 | int i, cpu_num = num_online_cpus(); | |
7404 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7405 | char buf[32]; | |
7406 | ||
7378547f MM |
7407 | WARN_ON(sd_ctl_dir[0].child); |
7408 | sd_ctl_dir[0].child = entry; | |
7409 | ||
ad1cdc1d MM |
7410 | if (entry == NULL) |
7411 | return; | |
7412 | ||
97b6ea7b | 7413 | for_each_online_cpu(i) { |
e692ab53 | 7414 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7415 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7416 | entry->mode = 0555; |
e692ab53 | 7417 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7418 | entry++; |
e692ab53 | 7419 | } |
7378547f MM |
7420 | |
7421 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7422 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7423 | } | |
6382bc90 | 7424 | |
7378547f | 7425 | /* may be called multiple times per register */ |
6382bc90 MM |
7426 | static void unregister_sched_domain_sysctl(void) |
7427 | { | |
7378547f MM |
7428 | if (sd_sysctl_header) |
7429 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7430 | sd_sysctl_header = NULL; |
7378547f MM |
7431 | if (sd_ctl_dir[0].child) |
7432 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7433 | } |
e692ab53 | 7434 | #else |
6382bc90 MM |
7435 | static void register_sched_domain_sysctl(void) |
7436 | { | |
7437 | } | |
7438 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7439 | { |
7440 | } | |
7441 | #endif | |
7442 | ||
1f11eb6a GH |
7443 | static void set_rq_online(struct rq *rq) |
7444 | { | |
7445 | if (!rq->online) { | |
7446 | const struct sched_class *class; | |
7447 | ||
c6c4927b | 7448 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7449 | rq->online = 1; |
7450 | ||
7451 | for_each_class(class) { | |
7452 | if (class->rq_online) | |
7453 | class->rq_online(rq); | |
7454 | } | |
7455 | } | |
7456 | } | |
7457 | ||
7458 | static void set_rq_offline(struct rq *rq) | |
7459 | { | |
7460 | if (rq->online) { | |
7461 | const struct sched_class *class; | |
7462 | ||
7463 | for_each_class(class) { | |
7464 | if (class->rq_offline) | |
7465 | class->rq_offline(rq); | |
7466 | } | |
7467 | ||
c6c4927b | 7468 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7469 | rq->online = 0; |
7470 | } | |
7471 | } | |
7472 | ||
1da177e4 LT |
7473 | /* |
7474 | * migration_call - callback that gets triggered when a CPU is added. | |
7475 | * Here we can start up the necessary migration thread for the new CPU. | |
7476 | */ | |
48f24c4d IM |
7477 | static int __cpuinit |
7478 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7479 | { |
1da177e4 | 7480 | struct task_struct *p; |
48f24c4d | 7481 | int cpu = (long)hcpu; |
1da177e4 | 7482 | unsigned long flags; |
70b97a7f | 7483 | struct rq *rq; |
1da177e4 LT |
7484 | |
7485 | switch (action) { | |
5be9361c | 7486 | |
1da177e4 | 7487 | case CPU_UP_PREPARE: |
8bb78442 | 7488 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7489 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7490 | if (IS_ERR(p)) |
7491 | return NOTIFY_BAD; | |
1da177e4 LT |
7492 | kthread_bind(p, cpu); |
7493 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7494 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7495 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7496 | task_rq_unlock(rq, &flags); |
7497 | cpu_rq(cpu)->migration_thread = p; | |
7498 | break; | |
48f24c4d | 7499 | |
1da177e4 | 7500 | case CPU_ONLINE: |
8bb78442 | 7501 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7502 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7503 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7504 | |
7505 | /* Update our root-domain */ | |
7506 | rq = cpu_rq(cpu); | |
7507 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7508 | rq->calc_load_update = calc_load_update; |
7509 | rq->calc_load_active = 0; | |
1f94ef59 | 7510 | if (rq->rd) { |
c6c4927b | 7511 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7512 | |
7513 | set_rq_online(rq); | |
1f94ef59 GH |
7514 | } |
7515 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7516 | break; |
48f24c4d | 7517 | |
1da177e4 LT |
7518 | #ifdef CONFIG_HOTPLUG_CPU |
7519 | case CPU_UP_CANCELED: | |
8bb78442 | 7520 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7521 | if (!cpu_rq(cpu)->migration_thread) |
7522 | break; | |
41a2d6cf | 7523 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7524 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7525 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7526 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7527 | cpu_rq(cpu)->migration_thread = NULL; | |
7528 | break; | |
48f24c4d | 7529 | |
1da177e4 | 7530 | case CPU_DEAD: |
8bb78442 | 7531 | case CPU_DEAD_FROZEN: |
470fd646 | 7532 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7533 | migrate_live_tasks(cpu); |
7534 | rq = cpu_rq(cpu); | |
7535 | kthread_stop(rq->migration_thread); | |
7536 | rq->migration_thread = NULL; | |
7537 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7538 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7539 | update_rq_clock(rq); |
2e1cb74a | 7540 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7541 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7542 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7543 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7544 | migrate_dead_tasks(cpu); |
d2da272a | 7545 | spin_unlock_irq(&rq->lock); |
470fd646 | 7546 | cpuset_unlock(); |
1da177e4 LT |
7547 | migrate_nr_uninterruptible(rq); |
7548 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7549 | calc_global_load_remove(rq); |
41a2d6cf IM |
7550 | /* |
7551 | * No need to migrate the tasks: it was best-effort if | |
7552 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7553 | * the requestors. | |
7554 | */ | |
1da177e4 LT |
7555 | spin_lock_irq(&rq->lock); |
7556 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7557 | struct migration_req *req; |
7558 | ||
1da177e4 | 7559 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7560 | struct migration_req, list); |
1da177e4 | 7561 | list_del_init(&req->list); |
9a2bd244 | 7562 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7563 | complete(&req->done); |
9a2bd244 | 7564 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7565 | } |
7566 | spin_unlock_irq(&rq->lock); | |
7567 | break; | |
57d885fe | 7568 | |
08f503b0 GH |
7569 | case CPU_DYING: |
7570 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7571 | /* Update our root-domain */ |
7572 | rq = cpu_rq(cpu); | |
7573 | spin_lock_irqsave(&rq->lock, flags); | |
7574 | if (rq->rd) { | |
c6c4927b | 7575 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7576 | set_rq_offline(rq); |
57d885fe GH |
7577 | } |
7578 | spin_unlock_irqrestore(&rq->lock, flags); | |
7579 | break; | |
1da177e4 LT |
7580 | #endif |
7581 | } | |
7582 | return NOTIFY_OK; | |
7583 | } | |
7584 | ||
f38b0820 PM |
7585 | /* |
7586 | * Register at high priority so that task migration (migrate_all_tasks) | |
7587 | * happens before everything else. This has to be lower priority than | |
7588 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7589 | */ |
26c2143b | 7590 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7591 | .notifier_call = migration_call, |
7592 | .priority = 10 | |
7593 | }; | |
7594 | ||
7babe8db | 7595 | static int __init migration_init(void) |
1da177e4 LT |
7596 | { |
7597 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7598 | int err; |
48f24c4d IM |
7599 | |
7600 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7601 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7602 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7603 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7604 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7605 | |
7606 | return err; | |
1da177e4 | 7607 | } |
7babe8db | 7608 | early_initcall(migration_init); |
1da177e4 LT |
7609 | #endif |
7610 | ||
7611 | #ifdef CONFIG_SMP | |
476f3534 | 7612 | |
3e9830dc | 7613 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7614 | |
7c16ec58 | 7615 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7616 | struct cpumask *groupmask) |
1da177e4 | 7617 | { |
4dcf6aff | 7618 | struct sched_group *group = sd->groups; |
434d53b0 | 7619 | char str[256]; |
1da177e4 | 7620 | |
968ea6d8 | 7621 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7622 | cpumask_clear(groupmask); |
4dcf6aff IM |
7623 | |
7624 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7625 | ||
7626 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7627 | printk("does not load-balance\n"); | |
7628 | if (sd->parent) | |
7629 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7630 | " has parent"); | |
7631 | return -1; | |
41c7ce9a NP |
7632 | } |
7633 | ||
eefd796a | 7634 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7635 | |
758b2cdc | 7636 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7637 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7638 | "CPU%d\n", cpu); | |
7639 | } | |
758b2cdc | 7640 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7641 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7642 | " CPU%d\n", cpu); | |
7643 | } | |
1da177e4 | 7644 | |
4dcf6aff | 7645 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7646 | do { |
4dcf6aff IM |
7647 | if (!group) { |
7648 | printk("\n"); | |
7649 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7650 | break; |
7651 | } | |
7652 | ||
4dcf6aff IM |
7653 | if (!group->__cpu_power) { |
7654 | printk(KERN_CONT "\n"); | |
7655 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7656 | "set\n"); | |
7657 | break; | |
7658 | } | |
1da177e4 | 7659 | |
758b2cdc | 7660 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7661 | printk(KERN_CONT "\n"); |
7662 | printk(KERN_ERR "ERROR: empty group\n"); | |
7663 | break; | |
7664 | } | |
1da177e4 | 7665 | |
758b2cdc | 7666 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7667 | printk(KERN_CONT "\n"); |
7668 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7669 | break; | |
7670 | } | |
1da177e4 | 7671 | |
758b2cdc | 7672 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7673 | |
968ea6d8 | 7674 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7675 | |
7676 | printk(KERN_CONT " %s", str); | |
7677 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7678 | printk(KERN_CONT " (__cpu_power = %d)", | |
7679 | group->__cpu_power); | |
7680 | } | |
1da177e4 | 7681 | |
4dcf6aff IM |
7682 | group = group->next; |
7683 | } while (group != sd->groups); | |
7684 | printk(KERN_CONT "\n"); | |
1da177e4 | 7685 | |
758b2cdc | 7686 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7687 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7688 | |
758b2cdc RR |
7689 | if (sd->parent && |
7690 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7691 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7692 | "of domain->span\n"); | |
7693 | return 0; | |
7694 | } | |
1da177e4 | 7695 | |
4dcf6aff IM |
7696 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7697 | { | |
d5dd3db1 | 7698 | cpumask_var_t groupmask; |
4dcf6aff | 7699 | int level = 0; |
1da177e4 | 7700 | |
4dcf6aff IM |
7701 | if (!sd) { |
7702 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7703 | return; | |
7704 | } | |
1da177e4 | 7705 | |
4dcf6aff IM |
7706 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7707 | ||
d5dd3db1 | 7708 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7709 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7710 | return; | |
7711 | } | |
7712 | ||
4dcf6aff | 7713 | for (;;) { |
7c16ec58 | 7714 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7715 | break; |
1da177e4 LT |
7716 | level++; |
7717 | sd = sd->parent; | |
33859f7f | 7718 | if (!sd) |
4dcf6aff IM |
7719 | break; |
7720 | } | |
d5dd3db1 | 7721 | free_cpumask_var(groupmask); |
1da177e4 | 7722 | } |
6d6bc0ad | 7723 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7724 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7725 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7726 | |
1a20ff27 | 7727 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7728 | { |
758b2cdc | 7729 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7730 | return 1; |
7731 | ||
7732 | /* Following flags need at least 2 groups */ | |
7733 | if (sd->flags & (SD_LOAD_BALANCE | | |
7734 | SD_BALANCE_NEWIDLE | | |
7735 | SD_BALANCE_FORK | | |
89c4710e SS |
7736 | SD_BALANCE_EXEC | |
7737 | SD_SHARE_CPUPOWER | | |
7738 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7739 | if (sd->groups != sd->groups->next) |
7740 | return 0; | |
7741 | } | |
7742 | ||
7743 | /* Following flags don't use groups */ | |
7744 | if (sd->flags & (SD_WAKE_IDLE | | |
7745 | SD_WAKE_AFFINE | | |
7746 | SD_WAKE_BALANCE)) | |
7747 | return 0; | |
7748 | ||
7749 | return 1; | |
7750 | } | |
7751 | ||
48f24c4d IM |
7752 | static int |
7753 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7754 | { |
7755 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7756 | ||
7757 | if (sd_degenerate(parent)) | |
7758 | return 1; | |
7759 | ||
758b2cdc | 7760 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7761 | return 0; |
7762 | ||
7763 | /* Does parent contain flags not in child? */ | |
7764 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7765 | if (cflags & SD_WAKE_AFFINE) | |
7766 | pflags &= ~SD_WAKE_BALANCE; | |
7767 | /* Flags needing groups don't count if only 1 group in parent */ | |
7768 | if (parent->groups == parent->groups->next) { | |
7769 | pflags &= ~(SD_LOAD_BALANCE | | |
7770 | SD_BALANCE_NEWIDLE | | |
7771 | SD_BALANCE_FORK | | |
89c4710e SS |
7772 | SD_BALANCE_EXEC | |
7773 | SD_SHARE_CPUPOWER | | |
7774 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7775 | if (nr_node_ids == 1) |
7776 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7777 | } |
7778 | if (~cflags & pflags) | |
7779 | return 0; | |
7780 | ||
7781 | return 1; | |
7782 | } | |
7783 | ||
c6c4927b RR |
7784 | static void free_rootdomain(struct root_domain *rd) |
7785 | { | |
68e74568 RR |
7786 | cpupri_cleanup(&rd->cpupri); |
7787 | ||
c6c4927b RR |
7788 | free_cpumask_var(rd->rto_mask); |
7789 | free_cpumask_var(rd->online); | |
7790 | free_cpumask_var(rd->span); | |
7791 | kfree(rd); | |
7792 | } | |
7793 | ||
57d885fe GH |
7794 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7795 | { | |
a0490fa3 | 7796 | struct root_domain *old_rd = NULL; |
57d885fe | 7797 | unsigned long flags; |
57d885fe GH |
7798 | |
7799 | spin_lock_irqsave(&rq->lock, flags); | |
7800 | ||
7801 | if (rq->rd) { | |
a0490fa3 | 7802 | old_rd = rq->rd; |
57d885fe | 7803 | |
c6c4927b | 7804 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7805 | set_rq_offline(rq); |
57d885fe | 7806 | |
c6c4927b | 7807 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7808 | |
a0490fa3 IM |
7809 | /* |
7810 | * If we dont want to free the old_rt yet then | |
7811 | * set old_rd to NULL to skip the freeing later | |
7812 | * in this function: | |
7813 | */ | |
7814 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7815 | old_rd = NULL; | |
57d885fe GH |
7816 | } |
7817 | ||
7818 | atomic_inc(&rd->refcount); | |
7819 | rq->rd = rd; | |
7820 | ||
c6c4927b RR |
7821 | cpumask_set_cpu(rq->cpu, rd->span); |
7822 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7823 | set_rq_online(rq); |
57d885fe GH |
7824 | |
7825 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7826 | |
7827 | if (old_rd) | |
7828 | free_rootdomain(old_rd); | |
57d885fe GH |
7829 | } |
7830 | ||
db2f59c8 | 7831 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7832 | { |
36b7b6d4 PE |
7833 | gfp_t gfp = GFP_KERNEL; |
7834 | ||
57d885fe GH |
7835 | memset(rd, 0, sizeof(*rd)); |
7836 | ||
36b7b6d4 PE |
7837 | if (bootmem) |
7838 | gfp = GFP_NOWAIT; | |
c6c4927b | 7839 | |
36b7b6d4 | 7840 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7841 | goto out; |
36b7b6d4 | 7842 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7843 | goto free_span; |
36b7b6d4 | 7844 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7845 | goto free_online; |
6e0534f2 | 7846 | |
0fb53029 | 7847 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7848 | goto free_rto_mask; |
c6c4927b | 7849 | return 0; |
6e0534f2 | 7850 | |
68e74568 RR |
7851 | free_rto_mask: |
7852 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7853 | free_online: |
7854 | free_cpumask_var(rd->online); | |
7855 | free_span: | |
7856 | free_cpumask_var(rd->span); | |
0c910d28 | 7857 | out: |
c6c4927b | 7858 | return -ENOMEM; |
57d885fe GH |
7859 | } |
7860 | ||
7861 | static void init_defrootdomain(void) | |
7862 | { | |
c6c4927b RR |
7863 | init_rootdomain(&def_root_domain, true); |
7864 | ||
57d885fe GH |
7865 | atomic_set(&def_root_domain.refcount, 1); |
7866 | } | |
7867 | ||
dc938520 | 7868 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7869 | { |
7870 | struct root_domain *rd; | |
7871 | ||
7872 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7873 | if (!rd) | |
7874 | return NULL; | |
7875 | ||
c6c4927b RR |
7876 | if (init_rootdomain(rd, false) != 0) { |
7877 | kfree(rd); | |
7878 | return NULL; | |
7879 | } | |
57d885fe GH |
7880 | |
7881 | return rd; | |
7882 | } | |
7883 | ||
1da177e4 | 7884 | /* |
0eab9146 | 7885 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7886 | * hold the hotplug lock. |
7887 | */ | |
0eab9146 IM |
7888 | static void |
7889 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7890 | { |
70b97a7f | 7891 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7892 | struct sched_domain *tmp; |
7893 | ||
7894 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7895 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7896 | struct sched_domain *parent = tmp->parent; |
7897 | if (!parent) | |
7898 | break; | |
f29c9b1c | 7899 | |
1a848870 | 7900 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7901 | tmp->parent = parent->parent; |
1a848870 SS |
7902 | if (parent->parent) |
7903 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7904 | } else |
7905 | tmp = tmp->parent; | |
245af2c7 SS |
7906 | } |
7907 | ||
1a848870 | 7908 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7909 | sd = sd->parent; |
1a848870 SS |
7910 | if (sd) |
7911 | sd->child = NULL; | |
7912 | } | |
1da177e4 LT |
7913 | |
7914 | sched_domain_debug(sd, cpu); | |
7915 | ||
57d885fe | 7916 | rq_attach_root(rq, rd); |
674311d5 | 7917 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7918 | } |
7919 | ||
7920 | /* cpus with isolated domains */ | |
dcc30a35 | 7921 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7922 | |
7923 | /* Setup the mask of cpus configured for isolated domains */ | |
7924 | static int __init isolated_cpu_setup(char *str) | |
7925 | { | |
968ea6d8 | 7926 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7927 | return 1; |
7928 | } | |
7929 | ||
8927f494 | 7930 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7931 | |
7932 | /* | |
6711cab4 SS |
7933 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7934 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7935 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7936 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7937 | * |
7938 | * init_sched_build_groups will build a circular linked list of the groups | |
7939 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7940 | * and ->cpu_power to 0. | |
7941 | */ | |
a616058b | 7942 | static void |
96f874e2 RR |
7943 | init_sched_build_groups(const struct cpumask *span, |
7944 | const struct cpumask *cpu_map, | |
7945 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7946 | struct sched_group **sg, |
96f874e2 RR |
7947 | struct cpumask *tmpmask), |
7948 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7949 | { |
7950 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7951 | int i; |
7952 | ||
96f874e2 | 7953 | cpumask_clear(covered); |
7c16ec58 | 7954 | |
abcd083a | 7955 | for_each_cpu(i, span) { |
6711cab4 | 7956 | struct sched_group *sg; |
7c16ec58 | 7957 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7958 | int j; |
7959 | ||
758b2cdc | 7960 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7961 | continue; |
7962 | ||
758b2cdc | 7963 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7964 | sg->__cpu_power = 0; |
1da177e4 | 7965 | |
abcd083a | 7966 | for_each_cpu(j, span) { |
7c16ec58 | 7967 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7968 | continue; |
7969 | ||
96f874e2 | 7970 | cpumask_set_cpu(j, covered); |
758b2cdc | 7971 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7972 | } |
7973 | if (!first) | |
7974 | first = sg; | |
7975 | if (last) | |
7976 | last->next = sg; | |
7977 | last = sg; | |
7978 | } | |
7979 | last->next = first; | |
7980 | } | |
7981 | ||
9c1cfda2 | 7982 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7983 | |
9c1cfda2 | 7984 | #ifdef CONFIG_NUMA |
198e2f18 | 7985 | |
9c1cfda2 JH |
7986 | /** |
7987 | * find_next_best_node - find the next node to include in a sched_domain | |
7988 | * @node: node whose sched_domain we're building | |
7989 | * @used_nodes: nodes already in the sched_domain | |
7990 | * | |
41a2d6cf | 7991 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7992 | * finds the closest node not already in the @used_nodes map. |
7993 | * | |
7994 | * Should use nodemask_t. | |
7995 | */ | |
c5f59f08 | 7996 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7997 | { |
7998 | int i, n, val, min_val, best_node = 0; | |
7999 | ||
8000 | min_val = INT_MAX; | |
8001 | ||
076ac2af | 8002 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8003 | /* Start at @node */ |
076ac2af | 8004 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8005 | |
8006 | if (!nr_cpus_node(n)) | |
8007 | continue; | |
8008 | ||
8009 | /* Skip already used nodes */ | |
c5f59f08 | 8010 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8011 | continue; |
8012 | ||
8013 | /* Simple min distance search */ | |
8014 | val = node_distance(node, n); | |
8015 | ||
8016 | if (val < min_val) { | |
8017 | min_val = val; | |
8018 | best_node = n; | |
8019 | } | |
8020 | } | |
8021 | ||
c5f59f08 | 8022 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8023 | return best_node; |
8024 | } | |
8025 | ||
8026 | /** | |
8027 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8028 | * @node: node whose cpumask we're constructing | |
73486722 | 8029 | * @span: resulting cpumask |
9c1cfda2 | 8030 | * |
41a2d6cf | 8031 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8032 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8033 | * out optimally. | |
8034 | */ | |
96f874e2 | 8035 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8036 | { |
c5f59f08 | 8037 | nodemask_t used_nodes; |
48f24c4d | 8038 | int i; |
9c1cfda2 | 8039 | |
6ca09dfc | 8040 | cpumask_clear(span); |
c5f59f08 | 8041 | nodes_clear(used_nodes); |
9c1cfda2 | 8042 | |
6ca09dfc | 8043 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8044 | node_set(node, used_nodes); |
9c1cfda2 JH |
8045 | |
8046 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8047 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8048 | |
6ca09dfc | 8049 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8050 | } |
9c1cfda2 | 8051 | } |
6d6bc0ad | 8052 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8053 | |
5c45bf27 | 8054 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8055 | |
6c99e9ad RR |
8056 | /* |
8057 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8058 | * |
8059 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8060 | * and struct sched_domain. ) | |
6c99e9ad RR |
8061 | */ |
8062 | struct static_sched_group { | |
8063 | struct sched_group sg; | |
8064 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8065 | }; | |
8066 | ||
8067 | struct static_sched_domain { | |
8068 | struct sched_domain sd; | |
8069 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8070 | }; | |
8071 | ||
9c1cfda2 | 8072 | /* |
48f24c4d | 8073 | * SMT sched-domains: |
9c1cfda2 | 8074 | */ |
1da177e4 | 8075 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8076 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8077 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8078 | |
41a2d6cf | 8079 | static int |
96f874e2 RR |
8080 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8081 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8082 | { |
6711cab4 | 8083 | if (sg) |
6c99e9ad | 8084 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8085 | return cpu; |
8086 | } | |
6d6bc0ad | 8087 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8088 | |
48f24c4d IM |
8089 | /* |
8090 | * multi-core sched-domains: | |
8091 | */ | |
1e9f28fa | 8092 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8093 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8094 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8095 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8096 | |
8097 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8098 | static int |
96f874e2 RR |
8099 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8100 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8101 | { |
6711cab4 | 8102 | int group; |
7c16ec58 | 8103 | |
c69fc56d | 8104 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8105 | group = cpumask_first(mask); |
6711cab4 | 8106 | if (sg) |
6c99e9ad | 8107 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8108 | return group; |
1e9f28fa SS |
8109 | } |
8110 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8111 | static int |
96f874e2 RR |
8112 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8113 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8114 | { |
6711cab4 | 8115 | if (sg) |
6c99e9ad | 8116 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8117 | return cpu; |
8118 | } | |
8119 | #endif | |
8120 | ||
6c99e9ad RR |
8121 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8122 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8123 | |
41a2d6cf | 8124 | static int |
96f874e2 RR |
8125 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8126 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8127 | { |
6711cab4 | 8128 | int group; |
48f24c4d | 8129 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8130 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8131 | group = cpumask_first(mask); |
1e9f28fa | 8132 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8133 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8134 | group = cpumask_first(mask); |
1da177e4 | 8135 | #else |
6711cab4 | 8136 | group = cpu; |
1da177e4 | 8137 | #endif |
6711cab4 | 8138 | if (sg) |
6c99e9ad | 8139 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8140 | return group; |
1da177e4 LT |
8141 | } |
8142 | ||
8143 | #ifdef CONFIG_NUMA | |
1da177e4 | 8144 | /* |
9c1cfda2 JH |
8145 | * The init_sched_build_groups can't handle what we want to do with node |
8146 | * groups, so roll our own. Now each node has its own list of groups which | |
8147 | * gets dynamically allocated. | |
1da177e4 | 8148 | */ |
62ea9ceb | 8149 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8150 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8151 | |
62ea9ceb | 8152 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8153 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8154 | |
96f874e2 RR |
8155 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8156 | struct sched_group **sg, | |
8157 | struct cpumask *nodemask) | |
9c1cfda2 | 8158 | { |
6711cab4 SS |
8159 | int group; |
8160 | ||
6ca09dfc | 8161 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8162 | group = cpumask_first(nodemask); |
6711cab4 SS |
8163 | |
8164 | if (sg) | |
6c99e9ad | 8165 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8166 | return group; |
1da177e4 | 8167 | } |
6711cab4 | 8168 | |
08069033 SS |
8169 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8170 | { | |
8171 | struct sched_group *sg = group_head; | |
8172 | int j; | |
8173 | ||
8174 | if (!sg) | |
8175 | return; | |
3a5c359a | 8176 | do { |
758b2cdc | 8177 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8178 | struct sched_domain *sd; |
08069033 | 8179 | |
6c99e9ad | 8180 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8181 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8182 | /* |
8183 | * Only add "power" once for each | |
8184 | * physical package. | |
8185 | */ | |
8186 | continue; | |
8187 | } | |
08069033 | 8188 | |
3a5c359a AK |
8189 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8190 | } | |
8191 | sg = sg->next; | |
8192 | } while (sg != group_head); | |
08069033 | 8193 | } |
6d6bc0ad | 8194 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8195 | |
a616058b | 8196 | #ifdef CONFIG_NUMA |
51888ca2 | 8197 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8198 | static void free_sched_groups(const struct cpumask *cpu_map, |
8199 | struct cpumask *nodemask) | |
51888ca2 | 8200 | { |
a616058b | 8201 | int cpu, i; |
51888ca2 | 8202 | |
abcd083a | 8203 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8204 | struct sched_group **sched_group_nodes |
8205 | = sched_group_nodes_bycpu[cpu]; | |
8206 | ||
51888ca2 SV |
8207 | if (!sched_group_nodes) |
8208 | continue; | |
8209 | ||
076ac2af | 8210 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8211 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8212 | ||
6ca09dfc | 8213 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8214 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8215 | continue; |
8216 | ||
8217 | if (sg == NULL) | |
8218 | continue; | |
8219 | sg = sg->next; | |
8220 | next_sg: | |
8221 | oldsg = sg; | |
8222 | sg = sg->next; | |
8223 | kfree(oldsg); | |
8224 | if (oldsg != sched_group_nodes[i]) | |
8225 | goto next_sg; | |
8226 | } | |
8227 | kfree(sched_group_nodes); | |
8228 | sched_group_nodes_bycpu[cpu] = NULL; | |
8229 | } | |
51888ca2 | 8230 | } |
6d6bc0ad | 8231 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8232 | static void free_sched_groups(const struct cpumask *cpu_map, |
8233 | struct cpumask *nodemask) | |
a616058b SS |
8234 | { |
8235 | } | |
6d6bc0ad | 8236 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8237 | |
89c4710e SS |
8238 | /* |
8239 | * Initialize sched groups cpu_power. | |
8240 | * | |
8241 | * cpu_power indicates the capacity of sched group, which is used while | |
8242 | * distributing the load between different sched groups in a sched domain. | |
8243 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8244 | * there are asymmetries in the topology. If there are asymmetries, group | |
8245 | * having more cpu_power will pickup more load compared to the group having | |
8246 | * less cpu_power. | |
8247 | * | |
8248 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8249 | * the maximum number of tasks a group can handle in the presence of other idle | |
8250 | * or lightly loaded groups in the same sched domain. | |
8251 | */ | |
8252 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8253 | { | |
8254 | struct sched_domain *child; | |
8255 | struct sched_group *group; | |
8256 | ||
8257 | WARN_ON(!sd || !sd->groups); | |
8258 | ||
13318a71 | 8259 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8260 | return; |
8261 | ||
8262 | child = sd->child; | |
8263 | ||
5517d86b ED |
8264 | sd->groups->__cpu_power = 0; |
8265 | ||
89c4710e SS |
8266 | /* |
8267 | * For perf policy, if the groups in child domain share resources | |
8268 | * (for example cores sharing some portions of the cache hierarchy | |
8269 | * or SMT), then set this domain groups cpu_power such that each group | |
8270 | * can handle only one task, when there are other idle groups in the | |
8271 | * same sched domain. | |
8272 | */ | |
8273 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8274 | (child->flags & | |
8275 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8276 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8277 | return; |
8278 | } | |
8279 | ||
89c4710e SS |
8280 | /* |
8281 | * add cpu_power of each child group to this groups cpu_power | |
8282 | */ | |
8283 | group = child->groups; | |
8284 | do { | |
5517d86b | 8285 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8286 | group = group->next; |
8287 | } while (group != child->groups); | |
8288 | } | |
8289 | ||
7c16ec58 MT |
8290 | /* |
8291 | * Initializers for schedule domains | |
8292 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8293 | */ | |
8294 | ||
a5d8c348 IM |
8295 | #ifdef CONFIG_SCHED_DEBUG |
8296 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8297 | #else | |
8298 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8299 | #endif | |
8300 | ||
7c16ec58 | 8301 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8302 | |
7c16ec58 MT |
8303 | #define SD_INIT_FUNC(type) \ |
8304 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8305 | { \ | |
8306 | memset(sd, 0, sizeof(*sd)); \ | |
8307 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8308 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8309 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8310 | } |
8311 | ||
8312 | SD_INIT_FUNC(CPU) | |
8313 | #ifdef CONFIG_NUMA | |
8314 | SD_INIT_FUNC(ALLNODES) | |
8315 | SD_INIT_FUNC(NODE) | |
8316 | #endif | |
8317 | #ifdef CONFIG_SCHED_SMT | |
8318 | SD_INIT_FUNC(SIBLING) | |
8319 | #endif | |
8320 | #ifdef CONFIG_SCHED_MC | |
8321 | SD_INIT_FUNC(MC) | |
8322 | #endif | |
8323 | ||
1d3504fc HS |
8324 | static int default_relax_domain_level = -1; |
8325 | ||
8326 | static int __init setup_relax_domain_level(char *str) | |
8327 | { | |
30e0e178 LZ |
8328 | unsigned long val; |
8329 | ||
8330 | val = simple_strtoul(str, NULL, 0); | |
8331 | if (val < SD_LV_MAX) | |
8332 | default_relax_domain_level = val; | |
8333 | ||
1d3504fc HS |
8334 | return 1; |
8335 | } | |
8336 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8337 | ||
8338 | static void set_domain_attribute(struct sched_domain *sd, | |
8339 | struct sched_domain_attr *attr) | |
8340 | { | |
8341 | int request; | |
8342 | ||
8343 | if (!attr || attr->relax_domain_level < 0) { | |
8344 | if (default_relax_domain_level < 0) | |
8345 | return; | |
8346 | else | |
8347 | request = default_relax_domain_level; | |
8348 | } else | |
8349 | request = attr->relax_domain_level; | |
8350 | if (request < sd->level) { | |
8351 | /* turn off idle balance on this domain */ | |
8352 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8353 | } else { | |
8354 | /* turn on idle balance on this domain */ | |
8355 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8356 | } | |
8357 | } | |
8358 | ||
1da177e4 | 8359 | /* |
1a20ff27 DG |
8360 | * Build sched domains for a given set of cpus and attach the sched domains |
8361 | * to the individual cpus | |
1da177e4 | 8362 | */ |
96f874e2 | 8363 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8364 | struct sched_domain_attr *attr) |
1da177e4 | 8365 | { |
3404c8d9 | 8366 | int i, err = -ENOMEM; |
57d885fe | 8367 | struct root_domain *rd; |
3404c8d9 RR |
8368 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8369 | tmpmask; | |
d1b55138 | 8370 | #ifdef CONFIG_NUMA |
3404c8d9 | 8371 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8372 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8373 | int sd_allnodes = 0; |
d1b55138 | 8374 | |
3404c8d9 RR |
8375 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8376 | goto out; | |
8377 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8378 | goto free_domainspan; | |
8379 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8380 | goto free_covered; | |
8381 | #endif | |
8382 | ||
8383 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8384 | goto free_notcovered; | |
8385 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8386 | goto free_nodemask; | |
8387 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8388 | goto free_this_sibling_map; | |
8389 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8390 | goto free_this_core_map; | |
8391 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8392 | goto free_send_covered; | |
8393 | ||
8394 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8395 | /* |
8396 | * Allocate the per-node list of sched groups | |
8397 | */ | |
076ac2af | 8398 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8399 | GFP_KERNEL); |
d1b55138 JH |
8400 | if (!sched_group_nodes) { |
8401 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8402 | goto free_tmpmask; |
d1b55138 | 8403 | } |
d1b55138 | 8404 | #endif |
1da177e4 | 8405 | |
dc938520 | 8406 | rd = alloc_rootdomain(); |
57d885fe GH |
8407 | if (!rd) { |
8408 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8409 | goto free_sched_groups; |
57d885fe GH |
8410 | } |
8411 | ||
7c16ec58 | 8412 | #ifdef CONFIG_NUMA |
96f874e2 | 8413 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8414 | #endif |
8415 | ||
1da177e4 | 8416 | /* |
1a20ff27 | 8417 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8418 | */ |
abcd083a | 8419 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8420 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8421 | |
6ca09dfc | 8422 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8423 | |
8424 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8425 | if (cpumask_weight(cpu_map) > |
8426 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8427 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8428 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8429 | set_domain_attribute(sd, attr); |
758b2cdc | 8430 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8431 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8432 | p = sd; |
6711cab4 | 8433 | sd_allnodes = 1; |
9c1cfda2 JH |
8434 | } else |
8435 | p = NULL; | |
8436 | ||
62ea9ceb | 8437 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8438 | SD_INIT(sd, NODE); |
1d3504fc | 8439 | set_domain_attribute(sd, attr); |
758b2cdc | 8440 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8441 | sd->parent = p; |
1a848870 SS |
8442 | if (p) |
8443 | p->child = sd; | |
758b2cdc RR |
8444 | cpumask_and(sched_domain_span(sd), |
8445 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8446 | #endif |
8447 | ||
8448 | p = sd; | |
6c99e9ad | 8449 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8450 | SD_INIT(sd, CPU); |
1d3504fc | 8451 | set_domain_attribute(sd, attr); |
758b2cdc | 8452 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8453 | sd->parent = p; |
1a848870 SS |
8454 | if (p) |
8455 | p->child = sd; | |
7c16ec58 | 8456 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8457 | |
1e9f28fa SS |
8458 | #ifdef CONFIG_SCHED_MC |
8459 | p = sd; | |
6c99e9ad | 8460 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8461 | SD_INIT(sd, MC); |
1d3504fc | 8462 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8463 | cpumask_and(sched_domain_span(sd), cpu_map, |
8464 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8465 | sd->parent = p; |
1a848870 | 8466 | p->child = sd; |
7c16ec58 | 8467 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8468 | #endif |
8469 | ||
1da177e4 LT |
8470 | #ifdef CONFIG_SCHED_SMT |
8471 | p = sd; | |
6c99e9ad | 8472 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8473 | SD_INIT(sd, SIBLING); |
1d3504fc | 8474 | set_domain_attribute(sd, attr); |
758b2cdc | 8475 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8476 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8477 | sd->parent = p; |
1a848870 | 8478 | p->child = sd; |
7c16ec58 | 8479 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8480 | #endif |
8481 | } | |
8482 | ||
8483 | #ifdef CONFIG_SCHED_SMT | |
8484 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8485 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8486 | cpumask_and(this_sibling_map, |
c69fc56d | 8487 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8488 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8489 | continue; |
8490 | ||
dd41f596 | 8491 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8492 | &cpu_to_cpu_group, |
8493 | send_covered, tmpmask); | |
1da177e4 LT |
8494 | } |
8495 | #endif | |
8496 | ||
1e9f28fa SS |
8497 | #ifdef CONFIG_SCHED_MC |
8498 | /* Set up multi-core groups */ | |
abcd083a | 8499 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8500 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8501 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8502 | continue; |
7c16ec58 | 8503 | |
dd41f596 | 8504 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8505 | &cpu_to_core_group, |
8506 | send_covered, tmpmask); | |
1e9f28fa SS |
8507 | } |
8508 | #endif | |
8509 | ||
1da177e4 | 8510 | /* Set up physical groups */ |
076ac2af | 8511 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8512 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8513 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8514 | continue; |
8515 | ||
7c16ec58 MT |
8516 | init_sched_build_groups(nodemask, cpu_map, |
8517 | &cpu_to_phys_group, | |
8518 | send_covered, tmpmask); | |
1da177e4 LT |
8519 | } |
8520 | ||
8521 | #ifdef CONFIG_NUMA | |
8522 | /* Set up node groups */ | |
7c16ec58 | 8523 | if (sd_allnodes) { |
7c16ec58 MT |
8524 | init_sched_build_groups(cpu_map, cpu_map, |
8525 | &cpu_to_allnodes_group, | |
8526 | send_covered, tmpmask); | |
8527 | } | |
9c1cfda2 | 8528 | |
076ac2af | 8529 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8530 | /* Set up node groups */ |
8531 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8532 | int j; |
8533 | ||
96f874e2 | 8534 | cpumask_clear(covered); |
6ca09dfc | 8535 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8536 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8537 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8538 | continue; |
d1b55138 | 8539 | } |
9c1cfda2 | 8540 | |
4bdbaad3 | 8541 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8542 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8543 | |
6c99e9ad RR |
8544 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8545 | GFP_KERNEL, i); | |
51888ca2 SV |
8546 | if (!sg) { |
8547 | printk(KERN_WARNING "Can not alloc domain group for " | |
8548 | "node %d\n", i); | |
8549 | goto error; | |
8550 | } | |
9c1cfda2 | 8551 | sched_group_nodes[i] = sg; |
abcd083a | 8552 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8553 | struct sched_domain *sd; |
9761eea8 | 8554 | |
62ea9ceb | 8555 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8556 | sd->groups = sg; |
9c1cfda2 | 8557 | } |
5517d86b | 8558 | sg->__cpu_power = 0; |
758b2cdc | 8559 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8560 | sg->next = sg; |
96f874e2 | 8561 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8562 | prev = sg; |
8563 | ||
076ac2af | 8564 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8565 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8566 | |
96f874e2 RR |
8567 | cpumask_complement(notcovered, covered); |
8568 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8569 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8570 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8571 | break; |
8572 | ||
6ca09dfc | 8573 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8574 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8575 | continue; |
8576 | ||
6c99e9ad RR |
8577 | sg = kmalloc_node(sizeof(struct sched_group) + |
8578 | cpumask_size(), | |
15f0b676 | 8579 | GFP_KERNEL, i); |
9c1cfda2 JH |
8580 | if (!sg) { |
8581 | printk(KERN_WARNING | |
8582 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8583 | goto error; |
9c1cfda2 | 8584 | } |
5517d86b | 8585 | sg->__cpu_power = 0; |
758b2cdc | 8586 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8587 | sg->next = prev->next; |
96f874e2 | 8588 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8589 | prev->next = sg; |
8590 | prev = sg; | |
8591 | } | |
9c1cfda2 | 8592 | } |
1da177e4 LT |
8593 | #endif |
8594 | ||
8595 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8596 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8597 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8598 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8599 | |
89c4710e | 8600 | init_sched_groups_power(i, sd); |
5c45bf27 | 8601 | } |
1da177e4 | 8602 | #endif |
1e9f28fa | 8603 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8604 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8605 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8606 | |
89c4710e | 8607 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8608 | } |
8609 | #endif | |
1e9f28fa | 8610 | |
abcd083a | 8611 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8612 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8613 | |
89c4710e | 8614 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8615 | } |
8616 | ||
9c1cfda2 | 8617 | #ifdef CONFIG_NUMA |
076ac2af | 8618 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8619 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8620 | |
6711cab4 SS |
8621 | if (sd_allnodes) { |
8622 | struct sched_group *sg; | |
f712c0c7 | 8623 | |
96f874e2 | 8624 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8625 | tmpmask); |
f712c0c7 SS |
8626 | init_numa_sched_groups_power(sg); |
8627 | } | |
9c1cfda2 JH |
8628 | #endif |
8629 | ||
1da177e4 | 8630 | /* Attach the domains */ |
abcd083a | 8631 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8632 | struct sched_domain *sd; |
8633 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8634 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8635 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8636 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8637 | #else |
6c99e9ad | 8638 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8639 | #endif |
57d885fe | 8640 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8641 | } |
51888ca2 | 8642 | |
3404c8d9 RR |
8643 | err = 0; |
8644 | ||
8645 | free_tmpmask: | |
8646 | free_cpumask_var(tmpmask); | |
8647 | free_send_covered: | |
8648 | free_cpumask_var(send_covered); | |
8649 | free_this_core_map: | |
8650 | free_cpumask_var(this_core_map); | |
8651 | free_this_sibling_map: | |
8652 | free_cpumask_var(this_sibling_map); | |
8653 | free_nodemask: | |
8654 | free_cpumask_var(nodemask); | |
8655 | free_notcovered: | |
8656 | #ifdef CONFIG_NUMA | |
8657 | free_cpumask_var(notcovered); | |
8658 | free_covered: | |
8659 | free_cpumask_var(covered); | |
8660 | free_domainspan: | |
8661 | free_cpumask_var(domainspan); | |
8662 | out: | |
8663 | #endif | |
8664 | return err; | |
8665 | ||
8666 | free_sched_groups: | |
8667 | #ifdef CONFIG_NUMA | |
8668 | kfree(sched_group_nodes); | |
8669 | #endif | |
8670 | goto free_tmpmask; | |
51888ca2 | 8671 | |
a616058b | 8672 | #ifdef CONFIG_NUMA |
51888ca2 | 8673 | error: |
7c16ec58 | 8674 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8675 | free_rootdomain(rd); |
3404c8d9 | 8676 | goto free_tmpmask; |
a616058b | 8677 | #endif |
1da177e4 | 8678 | } |
029190c5 | 8679 | |
96f874e2 | 8680 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8681 | { |
8682 | return __build_sched_domains(cpu_map, NULL); | |
8683 | } | |
8684 | ||
96f874e2 | 8685 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8686 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8687 | static struct sched_domain_attr *dattr_cur; |
8688 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8689 | |
8690 | /* | |
8691 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8692 | * cpumask) fails, then fallback to a single sched domain, |
8693 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8694 | */ |
4212823f | 8695 | static cpumask_var_t fallback_doms; |
029190c5 | 8696 | |
ee79d1bd HC |
8697 | /* |
8698 | * arch_update_cpu_topology lets virtualized architectures update the | |
8699 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8700 | * or 0 if it stayed the same. | |
8701 | */ | |
8702 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8703 | { |
ee79d1bd | 8704 | return 0; |
22e52b07 HC |
8705 | } |
8706 | ||
1a20ff27 | 8707 | /* |
41a2d6cf | 8708 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8709 | * For now this just excludes isolated cpus, but could be used to |
8710 | * exclude other special cases in the future. | |
1a20ff27 | 8711 | */ |
96f874e2 | 8712 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8713 | { |
7378547f MM |
8714 | int err; |
8715 | ||
22e52b07 | 8716 | arch_update_cpu_topology(); |
029190c5 | 8717 | ndoms_cur = 1; |
96f874e2 | 8718 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8719 | if (!doms_cur) |
4212823f | 8720 | doms_cur = fallback_doms; |
dcc30a35 | 8721 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8722 | dattr_cur = NULL; |
7378547f | 8723 | err = build_sched_domains(doms_cur); |
6382bc90 | 8724 | register_sched_domain_sysctl(); |
7378547f MM |
8725 | |
8726 | return err; | |
1a20ff27 DG |
8727 | } |
8728 | ||
96f874e2 RR |
8729 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8730 | struct cpumask *tmpmask) | |
1da177e4 | 8731 | { |
7c16ec58 | 8732 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8733 | } |
1da177e4 | 8734 | |
1a20ff27 DG |
8735 | /* |
8736 | * Detach sched domains from a group of cpus specified in cpu_map | |
8737 | * These cpus will now be attached to the NULL domain | |
8738 | */ | |
96f874e2 | 8739 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8740 | { |
96f874e2 RR |
8741 | /* Save because hotplug lock held. */ |
8742 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8743 | int i; |
8744 | ||
abcd083a | 8745 | for_each_cpu(i, cpu_map) |
57d885fe | 8746 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8747 | synchronize_sched(); |
96f874e2 | 8748 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8749 | } |
8750 | ||
1d3504fc HS |
8751 | /* handle null as "default" */ |
8752 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8753 | struct sched_domain_attr *new, int idx_new) | |
8754 | { | |
8755 | struct sched_domain_attr tmp; | |
8756 | ||
8757 | /* fast path */ | |
8758 | if (!new && !cur) | |
8759 | return 1; | |
8760 | ||
8761 | tmp = SD_ATTR_INIT; | |
8762 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8763 | new ? (new + idx_new) : &tmp, | |
8764 | sizeof(struct sched_domain_attr)); | |
8765 | } | |
8766 | ||
029190c5 PJ |
8767 | /* |
8768 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8769 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8770 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8771 | * It destroys each deleted domain and builds each new domain. | |
8772 | * | |
96f874e2 | 8773 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8774 | * The masks don't intersect (don't overlap.) We should setup one |
8775 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8776 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8777 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8778 | * it as it is. | |
8779 | * | |
41a2d6cf IM |
8780 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8781 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8782 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8783 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8784 | * the single partition 'fallback_doms', it also forces the domains | |
8785 | * to be rebuilt. | |
029190c5 | 8786 | * |
96f874e2 | 8787 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8788 | * ndoms_new == 0 is a special case for destroying existing domains, |
8789 | * and it will not create the default domain. | |
dfb512ec | 8790 | * |
029190c5 PJ |
8791 | * Call with hotplug lock held |
8792 | */ | |
96f874e2 RR |
8793 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8794 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8795 | struct sched_domain_attr *dattr_new) |
029190c5 | 8796 | { |
dfb512ec | 8797 | int i, j, n; |
d65bd5ec | 8798 | int new_topology; |
029190c5 | 8799 | |
712555ee | 8800 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8801 | |
7378547f MM |
8802 | /* always unregister in case we don't destroy any domains */ |
8803 | unregister_sched_domain_sysctl(); | |
8804 | ||
d65bd5ec HC |
8805 | /* Let architecture update cpu core mappings. */ |
8806 | new_topology = arch_update_cpu_topology(); | |
8807 | ||
dfb512ec | 8808 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8809 | |
8810 | /* Destroy deleted domains */ | |
8811 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8812 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8813 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8814 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8815 | goto match1; |
8816 | } | |
8817 | /* no match - a current sched domain not in new doms_new[] */ | |
8818 | detach_destroy_domains(doms_cur + i); | |
8819 | match1: | |
8820 | ; | |
8821 | } | |
8822 | ||
e761b772 MK |
8823 | if (doms_new == NULL) { |
8824 | ndoms_cur = 0; | |
4212823f | 8825 | doms_new = fallback_doms; |
dcc30a35 | 8826 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8827 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8828 | } |
8829 | ||
029190c5 PJ |
8830 | /* Build new domains */ |
8831 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8832 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8833 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8834 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8835 | goto match2; |
8836 | } | |
8837 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8838 | __build_sched_domains(doms_new + i, |
8839 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8840 | match2: |
8841 | ; | |
8842 | } | |
8843 | ||
8844 | /* Remember the new sched domains */ | |
4212823f | 8845 | if (doms_cur != fallback_doms) |
029190c5 | 8846 | kfree(doms_cur); |
1d3504fc | 8847 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8848 | doms_cur = doms_new; |
1d3504fc | 8849 | dattr_cur = dattr_new; |
029190c5 | 8850 | ndoms_cur = ndoms_new; |
7378547f MM |
8851 | |
8852 | register_sched_domain_sysctl(); | |
a1835615 | 8853 | |
712555ee | 8854 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8855 | } |
8856 | ||
5c45bf27 | 8857 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8858 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8859 | { |
95402b38 | 8860 | get_online_cpus(); |
dfb512ec MK |
8861 | |
8862 | /* Destroy domains first to force the rebuild */ | |
8863 | partition_sched_domains(0, NULL, NULL); | |
8864 | ||
e761b772 | 8865 | rebuild_sched_domains(); |
95402b38 | 8866 | put_online_cpus(); |
5c45bf27 SS |
8867 | } |
8868 | ||
8869 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8870 | { | |
afb8a9b7 | 8871 | unsigned int level = 0; |
5c45bf27 | 8872 | |
afb8a9b7 GS |
8873 | if (sscanf(buf, "%u", &level) != 1) |
8874 | return -EINVAL; | |
8875 | ||
8876 | /* | |
8877 | * level is always be positive so don't check for | |
8878 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8879 | * What happens on 0 or 1 byte write, | |
8880 | * need to check for count as well? | |
8881 | */ | |
8882 | ||
8883 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8884 | return -EINVAL; |
8885 | ||
8886 | if (smt) | |
afb8a9b7 | 8887 | sched_smt_power_savings = level; |
5c45bf27 | 8888 | else |
afb8a9b7 | 8889 | sched_mc_power_savings = level; |
5c45bf27 | 8890 | |
c70f22d2 | 8891 | arch_reinit_sched_domains(); |
5c45bf27 | 8892 | |
c70f22d2 | 8893 | return count; |
5c45bf27 SS |
8894 | } |
8895 | ||
5c45bf27 | 8896 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8897 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8898 | char *page) | |
5c45bf27 SS |
8899 | { |
8900 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8901 | } | |
f718cd4a | 8902 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8903 | const char *buf, size_t count) |
5c45bf27 SS |
8904 | { |
8905 | return sched_power_savings_store(buf, count, 0); | |
8906 | } | |
f718cd4a AK |
8907 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8908 | sched_mc_power_savings_show, | |
8909 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8910 | #endif |
8911 | ||
8912 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8913 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8914 | char *page) | |
5c45bf27 SS |
8915 | { |
8916 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8917 | } | |
f718cd4a | 8918 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8919 | const char *buf, size_t count) |
5c45bf27 SS |
8920 | { |
8921 | return sched_power_savings_store(buf, count, 1); | |
8922 | } | |
f718cd4a AK |
8923 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8924 | sched_smt_power_savings_show, | |
6707de00 AB |
8925 | sched_smt_power_savings_store); |
8926 | #endif | |
8927 | ||
39aac648 | 8928 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8929 | { |
8930 | int err = 0; | |
8931 | ||
8932 | #ifdef CONFIG_SCHED_SMT | |
8933 | if (smt_capable()) | |
8934 | err = sysfs_create_file(&cls->kset.kobj, | |
8935 | &attr_sched_smt_power_savings.attr); | |
8936 | #endif | |
8937 | #ifdef CONFIG_SCHED_MC | |
8938 | if (!err && mc_capable()) | |
8939 | err = sysfs_create_file(&cls->kset.kobj, | |
8940 | &attr_sched_mc_power_savings.attr); | |
8941 | #endif | |
8942 | return err; | |
8943 | } | |
6d6bc0ad | 8944 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8945 | |
e761b772 | 8946 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8947 | /* |
e761b772 MK |
8948 | * Add online and remove offline CPUs from the scheduler domains. |
8949 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8950 | */ |
8951 | static int update_sched_domains(struct notifier_block *nfb, | |
8952 | unsigned long action, void *hcpu) | |
e761b772 MK |
8953 | { |
8954 | switch (action) { | |
8955 | case CPU_ONLINE: | |
8956 | case CPU_ONLINE_FROZEN: | |
8957 | case CPU_DEAD: | |
8958 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8959 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8960 | return NOTIFY_OK; |
8961 | ||
8962 | default: | |
8963 | return NOTIFY_DONE; | |
8964 | } | |
8965 | } | |
8966 | #endif | |
8967 | ||
8968 | static int update_runtime(struct notifier_block *nfb, | |
8969 | unsigned long action, void *hcpu) | |
1da177e4 | 8970 | { |
7def2be1 PZ |
8971 | int cpu = (int)(long)hcpu; |
8972 | ||
1da177e4 | 8973 | switch (action) { |
1da177e4 | 8974 | case CPU_DOWN_PREPARE: |
8bb78442 | 8975 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8976 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8977 | return NOTIFY_OK; |
8978 | ||
1da177e4 | 8979 | case CPU_DOWN_FAILED: |
8bb78442 | 8980 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8981 | case CPU_ONLINE: |
8bb78442 | 8982 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8983 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8984 | return NOTIFY_OK; |
8985 | ||
1da177e4 LT |
8986 | default: |
8987 | return NOTIFY_DONE; | |
8988 | } | |
1da177e4 | 8989 | } |
1da177e4 LT |
8990 | |
8991 | void __init sched_init_smp(void) | |
8992 | { | |
dcc30a35 RR |
8993 | cpumask_var_t non_isolated_cpus; |
8994 | ||
8995 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8996 | |
434d53b0 MT |
8997 | #if defined(CONFIG_NUMA) |
8998 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8999 | GFP_KERNEL); | |
9000 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9001 | #endif | |
95402b38 | 9002 | get_online_cpus(); |
712555ee | 9003 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9004 | arch_init_sched_domains(cpu_online_mask); |
9005 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9006 | if (cpumask_empty(non_isolated_cpus)) | |
9007 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9008 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9009 | put_online_cpus(); |
e761b772 MK |
9010 | |
9011 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9012 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9013 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9014 | #endif |
9015 | ||
9016 | /* RT runtime code needs to handle some hotplug events */ | |
9017 | hotcpu_notifier(update_runtime, 0); | |
9018 | ||
b328ca18 | 9019 | init_hrtick(); |
5c1e1767 NP |
9020 | |
9021 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9022 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9023 | BUG(); |
19978ca6 | 9024 | sched_init_granularity(); |
dcc30a35 | 9025 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9026 | |
9027 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9028 | init_sched_rt_class(); |
1da177e4 LT |
9029 | } |
9030 | #else | |
9031 | void __init sched_init_smp(void) | |
9032 | { | |
19978ca6 | 9033 | sched_init_granularity(); |
1da177e4 LT |
9034 | } |
9035 | #endif /* CONFIG_SMP */ | |
9036 | ||
cd1bb94b AB |
9037 | const_debug unsigned int sysctl_timer_migration = 1; |
9038 | ||
1da177e4 LT |
9039 | int in_sched_functions(unsigned long addr) |
9040 | { | |
1da177e4 LT |
9041 | return in_lock_functions(addr) || |
9042 | (addr >= (unsigned long)__sched_text_start | |
9043 | && addr < (unsigned long)__sched_text_end); | |
9044 | } | |
9045 | ||
a9957449 | 9046 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9047 | { |
9048 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9049 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9050 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9051 | cfs_rq->rq = rq; | |
9052 | #endif | |
67e9fb2a | 9053 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9054 | } |
9055 | ||
fa85ae24 PZ |
9056 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9057 | { | |
9058 | struct rt_prio_array *array; | |
9059 | int i; | |
9060 | ||
9061 | array = &rt_rq->active; | |
9062 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9063 | INIT_LIST_HEAD(array->queue + i); | |
9064 | __clear_bit(i, array->bitmap); | |
9065 | } | |
9066 | /* delimiter for bitsearch: */ | |
9067 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9068 | ||
052f1dc7 | 9069 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9070 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9071 | #ifdef CONFIG_SMP |
e864c499 | 9072 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9073 | #endif |
48d5e258 | 9074 | #endif |
fa85ae24 PZ |
9075 | #ifdef CONFIG_SMP |
9076 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9077 | rt_rq->overloaded = 0; |
917b627d | 9078 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9079 | #endif |
9080 | ||
9081 | rt_rq->rt_time = 0; | |
9082 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9083 | rt_rq->rt_runtime = 0; |
9084 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9085 | |
052f1dc7 | 9086 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9087 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9088 | rt_rq->rq = rq; |
9089 | #endif | |
fa85ae24 PZ |
9090 | } |
9091 | ||
6f505b16 | 9092 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9093 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9094 | struct sched_entity *se, int cpu, int add, | |
9095 | struct sched_entity *parent) | |
6f505b16 | 9096 | { |
ec7dc8ac | 9097 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9098 | tg->cfs_rq[cpu] = cfs_rq; |
9099 | init_cfs_rq(cfs_rq, rq); | |
9100 | cfs_rq->tg = tg; | |
9101 | if (add) | |
9102 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9103 | ||
9104 | tg->se[cpu] = se; | |
354d60c2 DG |
9105 | /* se could be NULL for init_task_group */ |
9106 | if (!se) | |
9107 | return; | |
9108 | ||
ec7dc8ac DG |
9109 | if (!parent) |
9110 | se->cfs_rq = &rq->cfs; | |
9111 | else | |
9112 | se->cfs_rq = parent->my_q; | |
9113 | ||
6f505b16 PZ |
9114 | se->my_q = cfs_rq; |
9115 | se->load.weight = tg->shares; | |
e05510d0 | 9116 | se->load.inv_weight = 0; |
ec7dc8ac | 9117 | se->parent = parent; |
6f505b16 | 9118 | } |
052f1dc7 | 9119 | #endif |
6f505b16 | 9120 | |
052f1dc7 | 9121 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9122 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9123 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9124 | struct sched_rt_entity *parent) | |
6f505b16 | 9125 | { |
ec7dc8ac DG |
9126 | struct rq *rq = cpu_rq(cpu); |
9127 | ||
6f505b16 PZ |
9128 | tg->rt_rq[cpu] = rt_rq; |
9129 | init_rt_rq(rt_rq, rq); | |
9130 | rt_rq->tg = tg; | |
9131 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9132 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9133 | if (add) |
9134 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9135 | ||
9136 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9137 | if (!rt_se) |
9138 | return; | |
9139 | ||
ec7dc8ac DG |
9140 | if (!parent) |
9141 | rt_se->rt_rq = &rq->rt; | |
9142 | else | |
9143 | rt_se->rt_rq = parent->my_q; | |
9144 | ||
6f505b16 | 9145 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9146 | rt_se->parent = parent; |
6f505b16 PZ |
9147 | INIT_LIST_HEAD(&rt_se->run_list); |
9148 | } | |
9149 | #endif | |
9150 | ||
1da177e4 LT |
9151 | void __init sched_init(void) |
9152 | { | |
dd41f596 | 9153 | int i, j; |
434d53b0 MT |
9154 | unsigned long alloc_size = 0, ptr; |
9155 | ||
9156 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9157 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9158 | #endif | |
9159 | #ifdef CONFIG_RT_GROUP_SCHED | |
9160 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9161 | #endif |
9162 | #ifdef CONFIG_USER_SCHED | |
9163 | alloc_size *= 2; | |
df7c8e84 RR |
9164 | #endif |
9165 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9166 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9167 | #endif |
9168 | /* | |
9169 | * As sched_init() is called before page_alloc is setup, | |
9170 | * we use alloc_bootmem(). | |
9171 | */ | |
9172 | if (alloc_size) { | |
36b7b6d4 | 9173 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9174 | |
9175 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9176 | init_task_group.se = (struct sched_entity **)ptr; | |
9177 | ptr += nr_cpu_ids * sizeof(void **); | |
9178 | ||
9179 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9180 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9181 | |
9182 | #ifdef CONFIG_USER_SCHED | |
9183 | root_task_group.se = (struct sched_entity **)ptr; | |
9184 | ptr += nr_cpu_ids * sizeof(void **); | |
9185 | ||
9186 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9187 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9188 | #endif /* CONFIG_USER_SCHED */ |
9189 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9190 | #ifdef CONFIG_RT_GROUP_SCHED |
9191 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9192 | ptr += nr_cpu_ids * sizeof(void **); | |
9193 | ||
9194 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9195 | ptr += nr_cpu_ids * sizeof(void **); |
9196 | ||
9197 | #ifdef CONFIG_USER_SCHED | |
9198 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9199 | ptr += nr_cpu_ids * sizeof(void **); | |
9200 | ||
9201 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9202 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9203 | #endif /* CONFIG_USER_SCHED */ |
9204 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9205 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9206 | for_each_possible_cpu(i) { | |
9207 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9208 | ptr += cpumask_size(); | |
9209 | } | |
9210 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9211 | } |
dd41f596 | 9212 | |
57d885fe GH |
9213 | #ifdef CONFIG_SMP |
9214 | init_defrootdomain(); | |
9215 | #endif | |
9216 | ||
d0b27fa7 PZ |
9217 | init_rt_bandwidth(&def_rt_bandwidth, |
9218 | global_rt_period(), global_rt_runtime()); | |
9219 | ||
9220 | #ifdef CONFIG_RT_GROUP_SCHED | |
9221 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9222 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9223 | #ifdef CONFIG_USER_SCHED |
9224 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9225 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9226 | #endif /* CONFIG_USER_SCHED */ |
9227 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9228 | |
052f1dc7 | 9229 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9230 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9231 | INIT_LIST_HEAD(&init_task_group.children); |
9232 | ||
9233 | #ifdef CONFIG_USER_SCHED | |
9234 | INIT_LIST_HEAD(&root_task_group.children); | |
9235 | init_task_group.parent = &root_task_group; | |
9236 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9237 | #endif /* CONFIG_USER_SCHED */ |
9238 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9239 | |
0a945022 | 9240 | for_each_possible_cpu(i) { |
70b97a7f | 9241 | struct rq *rq; |
1da177e4 LT |
9242 | |
9243 | rq = cpu_rq(i); | |
9244 | spin_lock_init(&rq->lock); | |
7897986b | 9245 | rq->nr_running = 0; |
dce48a84 TG |
9246 | rq->calc_load_active = 0; |
9247 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9248 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9249 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9251 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9252 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9253 | #ifdef CONFIG_CGROUP_SCHED |
9254 | /* | |
9255 | * How much cpu bandwidth does init_task_group get? | |
9256 | * | |
9257 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9258 | * gets 100% of the cpu resources in the system. This overall | |
9259 | * system cpu resource is divided among the tasks of | |
9260 | * init_task_group and its child task-groups in a fair manner, | |
9261 | * based on each entity's (task or task-group's) weight | |
9262 | * (se->load.weight). | |
9263 | * | |
9264 | * In other words, if init_task_group has 10 tasks of weight | |
9265 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9266 | * then A0's share of the cpu resource is: | |
9267 | * | |
0d905bca | 9268 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9269 | * |
9270 | * We achieve this by letting init_task_group's tasks sit | |
9271 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9272 | */ | |
ec7dc8ac | 9273 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9274 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9275 | root_task_group.shares = NICE_0_LOAD; |
9276 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9277 | /* |
9278 | * In case of task-groups formed thr' the user id of tasks, | |
9279 | * init_task_group represents tasks belonging to root user. | |
9280 | * Hence it forms a sibling of all subsequent groups formed. | |
9281 | * In this case, init_task_group gets only a fraction of overall | |
9282 | * system cpu resource, based on the weight assigned to root | |
9283 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9284 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9285 | * (init_cfs_rq) and having one entity represent this group of | |
9286 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9287 | */ | |
ec7dc8ac | 9288 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9289 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9290 | &per_cpu(init_sched_entity, i), i, 1, |
9291 | root_task_group.se[i]); | |
6f505b16 | 9292 | |
052f1dc7 | 9293 | #endif |
354d60c2 DG |
9294 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9295 | ||
9296 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9297 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9298 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9299 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9300 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9301 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9302 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9303 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9304 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9305 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9306 | root_task_group.rt_se[i]); | |
354d60c2 | 9307 | #endif |
dd41f596 | 9308 | #endif |
1da177e4 | 9309 | |
dd41f596 IM |
9310 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9311 | rq->cpu_load[j] = 0; | |
1da177e4 | 9312 | #ifdef CONFIG_SMP |
41c7ce9a | 9313 | rq->sd = NULL; |
57d885fe | 9314 | rq->rd = NULL; |
1da177e4 | 9315 | rq->active_balance = 0; |
dd41f596 | 9316 | rq->next_balance = jiffies; |
1da177e4 | 9317 | rq->push_cpu = 0; |
0a2966b4 | 9318 | rq->cpu = i; |
1f11eb6a | 9319 | rq->online = 0; |
1da177e4 LT |
9320 | rq->migration_thread = NULL; |
9321 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9322 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9323 | #endif |
8f4d37ec | 9324 | init_rq_hrtick(rq); |
1da177e4 | 9325 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9326 | } |
9327 | ||
2dd73a4f | 9328 | set_load_weight(&init_task); |
b50f60ce | 9329 | |
e107be36 AK |
9330 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9331 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9332 | #endif | |
9333 | ||
c9819f45 | 9334 | #ifdef CONFIG_SMP |
962cf36c | 9335 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9336 | #endif |
9337 | ||
b50f60ce HC |
9338 | #ifdef CONFIG_RT_MUTEXES |
9339 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9340 | #endif | |
9341 | ||
1da177e4 LT |
9342 | /* |
9343 | * The boot idle thread does lazy MMU switching as well: | |
9344 | */ | |
9345 | atomic_inc(&init_mm.mm_count); | |
9346 | enter_lazy_tlb(&init_mm, current); | |
9347 | ||
9348 | /* | |
9349 | * Make us the idle thread. Technically, schedule() should not be | |
9350 | * called from this thread, however somewhere below it might be, | |
9351 | * but because we are the idle thread, we just pick up running again | |
9352 | * when this runqueue becomes "idle". | |
9353 | */ | |
9354 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9355 | |
9356 | calc_load_update = jiffies + LOAD_FREQ; | |
9357 | ||
dd41f596 IM |
9358 | /* |
9359 | * During early bootup we pretend to be a normal task: | |
9360 | */ | |
9361 | current->sched_class = &fair_sched_class; | |
6892b75e | 9362 | |
6a7b3dc3 | 9363 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9364 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9365 | #ifdef CONFIG_SMP |
7d1e6a9b | 9366 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9367 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9368 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9369 | #endif |
4bdddf8f | 9370 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9371 | #endif /* SMP */ |
6a7b3dc3 | 9372 | |
0d905bca IM |
9373 | perf_counter_init(); |
9374 | ||
6892b75e | 9375 | scheduler_running = 1; |
1da177e4 LT |
9376 | } |
9377 | ||
9378 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9379 | void __might_sleep(char *file, int line) | |
9380 | { | |
48f24c4d | 9381 | #ifdef in_atomic |
1da177e4 LT |
9382 | static unsigned long prev_jiffy; /* ratelimiting */ |
9383 | ||
aef745fc IM |
9384 | if ((!in_atomic() && !irqs_disabled()) || |
9385 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9386 | return; | |
9387 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9388 | return; | |
9389 | prev_jiffy = jiffies; | |
9390 | ||
9391 | printk(KERN_ERR | |
9392 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9393 | file, line); | |
9394 | printk(KERN_ERR | |
9395 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9396 | in_atomic(), irqs_disabled(), | |
9397 | current->pid, current->comm); | |
9398 | ||
9399 | debug_show_held_locks(current); | |
9400 | if (irqs_disabled()) | |
9401 | print_irqtrace_events(current); | |
9402 | dump_stack(); | |
1da177e4 LT |
9403 | #endif |
9404 | } | |
9405 | EXPORT_SYMBOL(__might_sleep); | |
9406 | #endif | |
9407 | ||
9408 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9409 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9410 | { | |
9411 | int on_rq; | |
3e51f33f | 9412 | |
3a5e4dc1 AK |
9413 | update_rq_clock(rq); |
9414 | on_rq = p->se.on_rq; | |
9415 | if (on_rq) | |
9416 | deactivate_task(rq, p, 0); | |
9417 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9418 | if (on_rq) { | |
9419 | activate_task(rq, p, 0); | |
9420 | resched_task(rq->curr); | |
9421 | } | |
9422 | } | |
9423 | ||
1da177e4 LT |
9424 | void normalize_rt_tasks(void) |
9425 | { | |
a0f98a1c | 9426 | struct task_struct *g, *p; |
1da177e4 | 9427 | unsigned long flags; |
70b97a7f | 9428 | struct rq *rq; |
1da177e4 | 9429 | |
4cf5d77a | 9430 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9431 | do_each_thread(g, p) { |
178be793 IM |
9432 | /* |
9433 | * Only normalize user tasks: | |
9434 | */ | |
9435 | if (!p->mm) | |
9436 | continue; | |
9437 | ||
6cfb0d5d | 9438 | p->se.exec_start = 0; |
6cfb0d5d | 9439 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9440 | p->se.wait_start = 0; |
dd41f596 | 9441 | p->se.sleep_start = 0; |
dd41f596 | 9442 | p->se.block_start = 0; |
6cfb0d5d | 9443 | #endif |
dd41f596 IM |
9444 | |
9445 | if (!rt_task(p)) { | |
9446 | /* | |
9447 | * Renice negative nice level userspace | |
9448 | * tasks back to 0: | |
9449 | */ | |
9450 | if (TASK_NICE(p) < 0 && p->mm) | |
9451 | set_user_nice(p, 0); | |
1da177e4 | 9452 | continue; |
dd41f596 | 9453 | } |
1da177e4 | 9454 | |
4cf5d77a | 9455 | spin_lock(&p->pi_lock); |
b29739f9 | 9456 | rq = __task_rq_lock(p); |
1da177e4 | 9457 | |
178be793 | 9458 | normalize_task(rq, p); |
3a5e4dc1 | 9459 | |
b29739f9 | 9460 | __task_rq_unlock(rq); |
4cf5d77a | 9461 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9462 | } while_each_thread(g, p); |
9463 | ||
4cf5d77a | 9464 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9465 | } |
9466 | ||
9467 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9468 | |
9469 | #ifdef CONFIG_IA64 | |
9470 | /* | |
9471 | * These functions are only useful for the IA64 MCA handling. | |
9472 | * | |
9473 | * They can only be called when the whole system has been | |
9474 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9475 | * activity can take place. Using them for anything else would | |
9476 | * be a serious bug, and as a result, they aren't even visible | |
9477 | * under any other configuration. | |
9478 | */ | |
9479 | ||
9480 | /** | |
9481 | * curr_task - return the current task for a given cpu. | |
9482 | * @cpu: the processor in question. | |
9483 | * | |
9484 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9485 | */ | |
36c8b586 | 9486 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9487 | { |
9488 | return cpu_curr(cpu); | |
9489 | } | |
9490 | ||
9491 | /** | |
9492 | * set_curr_task - set the current task for a given cpu. | |
9493 | * @cpu: the processor in question. | |
9494 | * @p: the task pointer to set. | |
9495 | * | |
9496 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9497 | * are serviced on a separate stack. It allows the architecture to switch the |
9498 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9499 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9500 | * and caller must save the original value of the current task (see | |
9501 | * curr_task() above) and restore that value before reenabling interrupts and | |
9502 | * re-starting the system. | |
9503 | * | |
9504 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9505 | */ | |
36c8b586 | 9506 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9507 | { |
9508 | cpu_curr(cpu) = p; | |
9509 | } | |
9510 | ||
9511 | #endif | |
29f59db3 | 9512 | |
bccbe08a PZ |
9513 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9514 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9515 | { |
9516 | int i; | |
9517 | ||
9518 | for_each_possible_cpu(i) { | |
9519 | if (tg->cfs_rq) | |
9520 | kfree(tg->cfs_rq[i]); | |
9521 | if (tg->se) | |
9522 | kfree(tg->se[i]); | |
6f505b16 PZ |
9523 | } |
9524 | ||
9525 | kfree(tg->cfs_rq); | |
9526 | kfree(tg->se); | |
6f505b16 PZ |
9527 | } |
9528 | ||
ec7dc8ac DG |
9529 | static |
9530 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9531 | { |
29f59db3 | 9532 | struct cfs_rq *cfs_rq; |
eab17229 | 9533 | struct sched_entity *se; |
9b5b7751 | 9534 | struct rq *rq; |
29f59db3 SV |
9535 | int i; |
9536 | ||
434d53b0 | 9537 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9538 | if (!tg->cfs_rq) |
9539 | goto err; | |
434d53b0 | 9540 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9541 | if (!tg->se) |
9542 | goto err; | |
052f1dc7 PZ |
9543 | |
9544 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9545 | |
9546 | for_each_possible_cpu(i) { | |
9b5b7751 | 9547 | rq = cpu_rq(i); |
29f59db3 | 9548 | |
eab17229 LZ |
9549 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9550 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9551 | if (!cfs_rq) |
9552 | goto err; | |
9553 | ||
eab17229 LZ |
9554 | se = kzalloc_node(sizeof(struct sched_entity), |
9555 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9556 | if (!se) |
9557 | goto err; | |
9558 | ||
eab17229 | 9559 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9560 | } |
9561 | ||
9562 | return 1; | |
9563 | ||
9564 | err: | |
9565 | return 0; | |
9566 | } | |
9567 | ||
9568 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9569 | { | |
9570 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9571 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9572 | } | |
9573 | ||
9574 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9575 | { | |
9576 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9577 | } | |
6d6bc0ad | 9578 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9579 | static inline void free_fair_sched_group(struct task_group *tg) |
9580 | { | |
9581 | } | |
9582 | ||
ec7dc8ac DG |
9583 | static inline |
9584 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9585 | { |
9586 | return 1; | |
9587 | } | |
9588 | ||
9589 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9590 | { | |
9591 | } | |
9592 | ||
9593 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9594 | { | |
9595 | } | |
6d6bc0ad | 9596 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9597 | |
9598 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9599 | static void free_rt_sched_group(struct task_group *tg) |
9600 | { | |
9601 | int i; | |
9602 | ||
d0b27fa7 PZ |
9603 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9604 | ||
bccbe08a PZ |
9605 | for_each_possible_cpu(i) { |
9606 | if (tg->rt_rq) | |
9607 | kfree(tg->rt_rq[i]); | |
9608 | if (tg->rt_se) | |
9609 | kfree(tg->rt_se[i]); | |
9610 | } | |
9611 | ||
9612 | kfree(tg->rt_rq); | |
9613 | kfree(tg->rt_se); | |
9614 | } | |
9615 | ||
ec7dc8ac DG |
9616 | static |
9617 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9618 | { |
9619 | struct rt_rq *rt_rq; | |
eab17229 | 9620 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9621 | struct rq *rq; |
9622 | int i; | |
9623 | ||
434d53b0 | 9624 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9625 | if (!tg->rt_rq) |
9626 | goto err; | |
434d53b0 | 9627 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9628 | if (!tg->rt_se) |
9629 | goto err; | |
9630 | ||
d0b27fa7 PZ |
9631 | init_rt_bandwidth(&tg->rt_bandwidth, |
9632 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9633 | |
9634 | for_each_possible_cpu(i) { | |
9635 | rq = cpu_rq(i); | |
9636 | ||
eab17229 LZ |
9637 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9638 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9639 | if (!rt_rq) |
9640 | goto err; | |
29f59db3 | 9641 | |
eab17229 LZ |
9642 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9643 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9644 | if (!rt_se) |
9645 | goto err; | |
29f59db3 | 9646 | |
eab17229 | 9647 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9648 | } |
9649 | ||
bccbe08a PZ |
9650 | return 1; |
9651 | ||
9652 | err: | |
9653 | return 0; | |
9654 | } | |
9655 | ||
9656 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9657 | { | |
9658 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9659 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9660 | } | |
9661 | ||
9662 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9663 | { | |
9664 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9665 | } | |
6d6bc0ad | 9666 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9667 | static inline void free_rt_sched_group(struct task_group *tg) |
9668 | { | |
9669 | } | |
9670 | ||
ec7dc8ac DG |
9671 | static inline |
9672 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9673 | { |
9674 | return 1; | |
9675 | } | |
9676 | ||
9677 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9678 | { | |
9679 | } | |
9680 | ||
9681 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9682 | { | |
9683 | } | |
6d6bc0ad | 9684 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9685 | |
d0b27fa7 | 9686 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9687 | static void free_sched_group(struct task_group *tg) |
9688 | { | |
9689 | free_fair_sched_group(tg); | |
9690 | free_rt_sched_group(tg); | |
9691 | kfree(tg); | |
9692 | } | |
9693 | ||
9694 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9695 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9696 | { |
9697 | struct task_group *tg; | |
9698 | unsigned long flags; | |
9699 | int i; | |
9700 | ||
9701 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9702 | if (!tg) | |
9703 | return ERR_PTR(-ENOMEM); | |
9704 | ||
ec7dc8ac | 9705 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9706 | goto err; |
9707 | ||
ec7dc8ac | 9708 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9709 | goto err; |
9710 | ||
8ed36996 | 9711 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9712 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9713 | register_fair_sched_group(tg, i); |
9714 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9715 | } |
6f505b16 | 9716 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9717 | |
9718 | WARN_ON(!parent); /* root should already exist */ | |
9719 | ||
9720 | tg->parent = parent; | |
f473aa5e | 9721 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9722 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9723 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9724 | |
9b5b7751 | 9725 | return tg; |
29f59db3 SV |
9726 | |
9727 | err: | |
6f505b16 | 9728 | free_sched_group(tg); |
29f59db3 SV |
9729 | return ERR_PTR(-ENOMEM); |
9730 | } | |
9731 | ||
9b5b7751 | 9732 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9733 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9734 | { |
29f59db3 | 9735 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9736 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9737 | } |
9738 | ||
9b5b7751 | 9739 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9740 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9741 | { |
8ed36996 | 9742 | unsigned long flags; |
9b5b7751 | 9743 | int i; |
29f59db3 | 9744 | |
8ed36996 | 9745 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9746 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9747 | unregister_fair_sched_group(tg, i); |
9748 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9749 | } |
6f505b16 | 9750 | list_del_rcu(&tg->list); |
f473aa5e | 9751 | list_del_rcu(&tg->siblings); |
8ed36996 | 9752 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9753 | |
9b5b7751 | 9754 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9755 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9756 | } |
9757 | ||
9b5b7751 | 9758 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9759 | * The caller of this function should have put the task in its new group |
9760 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9761 | * reflect its new group. | |
9b5b7751 SV |
9762 | */ |
9763 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9764 | { |
9765 | int on_rq, running; | |
9766 | unsigned long flags; | |
9767 | struct rq *rq; | |
9768 | ||
9769 | rq = task_rq_lock(tsk, &flags); | |
9770 | ||
29f59db3 SV |
9771 | update_rq_clock(rq); |
9772 | ||
051a1d1a | 9773 | running = task_current(rq, tsk); |
29f59db3 SV |
9774 | on_rq = tsk->se.on_rq; |
9775 | ||
0e1f3483 | 9776 | if (on_rq) |
29f59db3 | 9777 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9778 | if (unlikely(running)) |
9779 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9780 | |
6f505b16 | 9781 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9782 | |
810b3817 PZ |
9783 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9784 | if (tsk->sched_class->moved_group) | |
9785 | tsk->sched_class->moved_group(tsk); | |
9786 | #endif | |
9787 | ||
0e1f3483 HS |
9788 | if (unlikely(running)) |
9789 | tsk->sched_class->set_curr_task(rq); | |
9790 | if (on_rq) | |
7074badb | 9791 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9792 | |
29f59db3 SV |
9793 | task_rq_unlock(rq, &flags); |
9794 | } | |
6d6bc0ad | 9795 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9796 | |
052f1dc7 | 9797 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9798 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9799 | { |
9800 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9801 | int on_rq; |
9802 | ||
29f59db3 | 9803 | on_rq = se->on_rq; |
62fb1851 | 9804 | if (on_rq) |
29f59db3 SV |
9805 | dequeue_entity(cfs_rq, se, 0); |
9806 | ||
9807 | se->load.weight = shares; | |
e05510d0 | 9808 | se->load.inv_weight = 0; |
29f59db3 | 9809 | |
62fb1851 | 9810 | if (on_rq) |
29f59db3 | 9811 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9812 | } |
62fb1851 | 9813 | |
c09595f6 PZ |
9814 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9815 | { | |
9816 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9817 | struct rq *rq = cfs_rq->rq; | |
9818 | unsigned long flags; | |
9819 | ||
9820 | spin_lock_irqsave(&rq->lock, flags); | |
9821 | __set_se_shares(se, shares); | |
9822 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9823 | } |
9824 | ||
8ed36996 PZ |
9825 | static DEFINE_MUTEX(shares_mutex); |
9826 | ||
4cf86d77 | 9827 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9828 | { |
9829 | int i; | |
8ed36996 | 9830 | unsigned long flags; |
c61935fd | 9831 | |
ec7dc8ac DG |
9832 | /* |
9833 | * We can't change the weight of the root cgroup. | |
9834 | */ | |
9835 | if (!tg->se[0]) | |
9836 | return -EINVAL; | |
9837 | ||
18d95a28 PZ |
9838 | if (shares < MIN_SHARES) |
9839 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9840 | else if (shares > MAX_SHARES) |
9841 | shares = MAX_SHARES; | |
62fb1851 | 9842 | |
8ed36996 | 9843 | mutex_lock(&shares_mutex); |
9b5b7751 | 9844 | if (tg->shares == shares) |
5cb350ba | 9845 | goto done; |
29f59db3 | 9846 | |
8ed36996 | 9847 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9848 | for_each_possible_cpu(i) |
9849 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9850 | list_del_rcu(&tg->siblings); |
8ed36996 | 9851 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9852 | |
9853 | /* wait for any ongoing reference to this group to finish */ | |
9854 | synchronize_sched(); | |
9855 | ||
9856 | /* | |
9857 | * Now we are free to modify the group's share on each cpu | |
9858 | * w/o tripping rebalance_share or load_balance_fair. | |
9859 | */ | |
9b5b7751 | 9860 | tg->shares = shares; |
c09595f6 PZ |
9861 | for_each_possible_cpu(i) { |
9862 | /* | |
9863 | * force a rebalance | |
9864 | */ | |
9865 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9866 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9867 | } |
29f59db3 | 9868 | |
6b2d7700 SV |
9869 | /* |
9870 | * Enable load balance activity on this group, by inserting it back on | |
9871 | * each cpu's rq->leaf_cfs_rq_list. | |
9872 | */ | |
8ed36996 | 9873 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9874 | for_each_possible_cpu(i) |
9875 | register_fair_sched_group(tg, i); | |
f473aa5e | 9876 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9877 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9878 | done: |
8ed36996 | 9879 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9880 | return 0; |
29f59db3 SV |
9881 | } |
9882 | ||
5cb350ba DG |
9883 | unsigned long sched_group_shares(struct task_group *tg) |
9884 | { | |
9885 | return tg->shares; | |
9886 | } | |
052f1dc7 | 9887 | #endif |
5cb350ba | 9888 | |
052f1dc7 | 9889 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9890 | /* |
9f0c1e56 | 9891 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9892 | */ |
9f0c1e56 PZ |
9893 | static DEFINE_MUTEX(rt_constraints_mutex); |
9894 | ||
9895 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9896 | { | |
9897 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9898 | return 1ULL << 20; |
9f0c1e56 | 9899 | |
9a7e0b18 | 9900 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9901 | } |
9902 | ||
9a7e0b18 PZ |
9903 | /* Must be called with tasklist_lock held */ |
9904 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9905 | { |
9a7e0b18 | 9906 | struct task_struct *g, *p; |
b40b2e8e | 9907 | |
9a7e0b18 PZ |
9908 | do_each_thread(g, p) { |
9909 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9910 | return 1; | |
9911 | } while_each_thread(g, p); | |
b40b2e8e | 9912 | |
9a7e0b18 PZ |
9913 | return 0; |
9914 | } | |
b40b2e8e | 9915 | |
9a7e0b18 PZ |
9916 | struct rt_schedulable_data { |
9917 | struct task_group *tg; | |
9918 | u64 rt_period; | |
9919 | u64 rt_runtime; | |
9920 | }; | |
b40b2e8e | 9921 | |
9a7e0b18 PZ |
9922 | static int tg_schedulable(struct task_group *tg, void *data) |
9923 | { | |
9924 | struct rt_schedulable_data *d = data; | |
9925 | struct task_group *child; | |
9926 | unsigned long total, sum = 0; | |
9927 | u64 period, runtime; | |
b40b2e8e | 9928 | |
9a7e0b18 PZ |
9929 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9930 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9931 | |
9a7e0b18 PZ |
9932 | if (tg == d->tg) { |
9933 | period = d->rt_period; | |
9934 | runtime = d->rt_runtime; | |
b40b2e8e | 9935 | } |
b40b2e8e | 9936 | |
98a4826b PZ |
9937 | #ifdef CONFIG_USER_SCHED |
9938 | if (tg == &root_task_group) { | |
9939 | period = global_rt_period(); | |
9940 | runtime = global_rt_runtime(); | |
9941 | } | |
9942 | #endif | |
9943 | ||
4653f803 PZ |
9944 | /* |
9945 | * Cannot have more runtime than the period. | |
9946 | */ | |
9947 | if (runtime > period && runtime != RUNTIME_INF) | |
9948 | return -EINVAL; | |
6f505b16 | 9949 | |
4653f803 PZ |
9950 | /* |
9951 | * Ensure we don't starve existing RT tasks. | |
9952 | */ | |
9a7e0b18 PZ |
9953 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9954 | return -EBUSY; | |
6f505b16 | 9955 | |
9a7e0b18 | 9956 | total = to_ratio(period, runtime); |
6f505b16 | 9957 | |
4653f803 PZ |
9958 | /* |
9959 | * Nobody can have more than the global setting allows. | |
9960 | */ | |
9961 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9962 | return -EINVAL; | |
6f505b16 | 9963 | |
4653f803 PZ |
9964 | /* |
9965 | * The sum of our children's runtime should not exceed our own. | |
9966 | */ | |
9a7e0b18 PZ |
9967 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9968 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9969 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9970 | |
9a7e0b18 PZ |
9971 | if (child == d->tg) { |
9972 | period = d->rt_period; | |
9973 | runtime = d->rt_runtime; | |
9974 | } | |
6f505b16 | 9975 | |
9a7e0b18 | 9976 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9977 | } |
6f505b16 | 9978 | |
9a7e0b18 PZ |
9979 | if (sum > total) |
9980 | return -EINVAL; | |
9981 | ||
9982 | return 0; | |
6f505b16 PZ |
9983 | } |
9984 | ||
9a7e0b18 | 9985 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9986 | { |
9a7e0b18 PZ |
9987 | struct rt_schedulable_data data = { |
9988 | .tg = tg, | |
9989 | .rt_period = period, | |
9990 | .rt_runtime = runtime, | |
9991 | }; | |
9992 | ||
9993 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9994 | } |
9995 | ||
d0b27fa7 PZ |
9996 | static int tg_set_bandwidth(struct task_group *tg, |
9997 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9998 | { |
ac086bc2 | 9999 | int i, err = 0; |
9f0c1e56 | 10000 | |
9f0c1e56 | 10001 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10002 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10003 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10004 | if (err) | |
9f0c1e56 | 10005 | goto unlock; |
ac086bc2 PZ |
10006 | |
10007 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10008 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10009 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10010 | |
10011 | for_each_possible_cpu(i) { | |
10012 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10013 | ||
10014 | spin_lock(&rt_rq->rt_runtime_lock); | |
10015 | rt_rq->rt_runtime = rt_runtime; | |
10016 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10017 | } | |
10018 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10019 | unlock: |
521f1a24 | 10020 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10021 | mutex_unlock(&rt_constraints_mutex); |
10022 | ||
10023 | return err; | |
6f505b16 PZ |
10024 | } |
10025 | ||
d0b27fa7 PZ |
10026 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10027 | { | |
10028 | u64 rt_runtime, rt_period; | |
10029 | ||
10030 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10031 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10032 | if (rt_runtime_us < 0) | |
10033 | rt_runtime = RUNTIME_INF; | |
10034 | ||
10035 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10036 | } | |
10037 | ||
9f0c1e56 PZ |
10038 | long sched_group_rt_runtime(struct task_group *tg) |
10039 | { | |
10040 | u64 rt_runtime_us; | |
10041 | ||
d0b27fa7 | 10042 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10043 | return -1; |
10044 | ||
d0b27fa7 | 10045 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10046 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10047 | return rt_runtime_us; | |
10048 | } | |
d0b27fa7 PZ |
10049 | |
10050 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10051 | { | |
10052 | u64 rt_runtime, rt_period; | |
10053 | ||
10054 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10055 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10056 | ||
619b0488 R |
10057 | if (rt_period == 0) |
10058 | return -EINVAL; | |
10059 | ||
d0b27fa7 PZ |
10060 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10061 | } | |
10062 | ||
10063 | long sched_group_rt_period(struct task_group *tg) | |
10064 | { | |
10065 | u64 rt_period_us; | |
10066 | ||
10067 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10068 | do_div(rt_period_us, NSEC_PER_USEC); | |
10069 | return rt_period_us; | |
10070 | } | |
10071 | ||
10072 | static int sched_rt_global_constraints(void) | |
10073 | { | |
4653f803 | 10074 | u64 runtime, period; |
d0b27fa7 PZ |
10075 | int ret = 0; |
10076 | ||
ec5d4989 HS |
10077 | if (sysctl_sched_rt_period <= 0) |
10078 | return -EINVAL; | |
10079 | ||
4653f803 PZ |
10080 | runtime = global_rt_runtime(); |
10081 | period = global_rt_period(); | |
10082 | ||
10083 | /* | |
10084 | * Sanity check on the sysctl variables. | |
10085 | */ | |
10086 | if (runtime > period && runtime != RUNTIME_INF) | |
10087 | return -EINVAL; | |
10b612f4 | 10088 | |
d0b27fa7 | 10089 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10090 | read_lock(&tasklist_lock); |
4653f803 | 10091 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10092 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10093 | mutex_unlock(&rt_constraints_mutex); |
10094 | ||
10095 | return ret; | |
10096 | } | |
54e99124 DG |
10097 | |
10098 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10099 | { | |
10100 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10101 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10102 | return 0; | |
10103 | ||
10104 | return 1; | |
10105 | } | |
10106 | ||
6d6bc0ad | 10107 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10108 | static int sched_rt_global_constraints(void) |
10109 | { | |
ac086bc2 PZ |
10110 | unsigned long flags; |
10111 | int i; | |
10112 | ||
ec5d4989 HS |
10113 | if (sysctl_sched_rt_period <= 0) |
10114 | return -EINVAL; | |
10115 | ||
60aa605d PZ |
10116 | /* |
10117 | * There's always some RT tasks in the root group | |
10118 | * -- migration, kstopmachine etc.. | |
10119 | */ | |
10120 | if (sysctl_sched_rt_runtime == 0) | |
10121 | return -EBUSY; | |
10122 | ||
ac086bc2 PZ |
10123 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10124 | for_each_possible_cpu(i) { | |
10125 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10126 | ||
10127 | spin_lock(&rt_rq->rt_runtime_lock); | |
10128 | rt_rq->rt_runtime = global_rt_runtime(); | |
10129 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10130 | } | |
10131 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10132 | ||
d0b27fa7 PZ |
10133 | return 0; |
10134 | } | |
6d6bc0ad | 10135 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10136 | |
10137 | int sched_rt_handler(struct ctl_table *table, int write, | |
10138 | struct file *filp, void __user *buffer, size_t *lenp, | |
10139 | loff_t *ppos) | |
10140 | { | |
10141 | int ret; | |
10142 | int old_period, old_runtime; | |
10143 | static DEFINE_MUTEX(mutex); | |
10144 | ||
10145 | mutex_lock(&mutex); | |
10146 | old_period = sysctl_sched_rt_period; | |
10147 | old_runtime = sysctl_sched_rt_runtime; | |
10148 | ||
10149 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10150 | ||
10151 | if (!ret && write) { | |
10152 | ret = sched_rt_global_constraints(); | |
10153 | if (ret) { | |
10154 | sysctl_sched_rt_period = old_period; | |
10155 | sysctl_sched_rt_runtime = old_runtime; | |
10156 | } else { | |
10157 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10158 | def_rt_bandwidth.rt_period = | |
10159 | ns_to_ktime(global_rt_period()); | |
10160 | } | |
10161 | } | |
10162 | mutex_unlock(&mutex); | |
10163 | ||
10164 | return ret; | |
10165 | } | |
68318b8e | 10166 | |
052f1dc7 | 10167 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10168 | |
10169 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10170 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10171 | { |
2b01dfe3 PM |
10172 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10173 | struct task_group, css); | |
68318b8e SV |
10174 | } |
10175 | ||
10176 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10177 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10178 | { |
ec7dc8ac | 10179 | struct task_group *tg, *parent; |
68318b8e | 10180 | |
2b01dfe3 | 10181 | if (!cgrp->parent) { |
68318b8e | 10182 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10183 | return &init_task_group.css; |
10184 | } | |
10185 | ||
ec7dc8ac DG |
10186 | parent = cgroup_tg(cgrp->parent); |
10187 | tg = sched_create_group(parent); | |
68318b8e SV |
10188 | if (IS_ERR(tg)) |
10189 | return ERR_PTR(-ENOMEM); | |
10190 | ||
68318b8e SV |
10191 | return &tg->css; |
10192 | } | |
10193 | ||
41a2d6cf IM |
10194 | static void |
10195 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10196 | { |
2b01dfe3 | 10197 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10198 | |
10199 | sched_destroy_group(tg); | |
10200 | } | |
10201 | ||
41a2d6cf IM |
10202 | static int |
10203 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10204 | struct task_struct *tsk) | |
68318b8e | 10205 | { |
b68aa230 | 10206 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10207 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10208 | return -EINVAL; |
10209 | #else | |
68318b8e SV |
10210 | /* We don't support RT-tasks being in separate groups */ |
10211 | if (tsk->sched_class != &fair_sched_class) | |
10212 | return -EINVAL; | |
b68aa230 | 10213 | #endif |
68318b8e SV |
10214 | |
10215 | return 0; | |
10216 | } | |
10217 | ||
10218 | static void | |
2b01dfe3 | 10219 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10220 | struct cgroup *old_cont, struct task_struct *tsk) |
10221 | { | |
10222 | sched_move_task(tsk); | |
10223 | } | |
10224 | ||
052f1dc7 | 10225 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10226 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10227 | u64 shareval) |
68318b8e | 10228 | { |
2b01dfe3 | 10229 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10230 | } |
10231 | ||
f4c753b7 | 10232 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10233 | { |
2b01dfe3 | 10234 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10235 | |
10236 | return (u64) tg->shares; | |
10237 | } | |
6d6bc0ad | 10238 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10239 | |
052f1dc7 | 10240 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10241 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10242 | s64 val) |
6f505b16 | 10243 | { |
06ecb27c | 10244 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10245 | } |
10246 | ||
06ecb27c | 10247 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10248 | { |
06ecb27c | 10249 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10250 | } |
d0b27fa7 PZ |
10251 | |
10252 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10253 | u64 rt_period_us) | |
10254 | { | |
10255 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10256 | } | |
10257 | ||
10258 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10259 | { | |
10260 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10261 | } | |
6d6bc0ad | 10262 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10263 | |
fe5c7cc2 | 10264 | static struct cftype cpu_files[] = { |
052f1dc7 | 10265 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10266 | { |
10267 | .name = "shares", | |
f4c753b7 PM |
10268 | .read_u64 = cpu_shares_read_u64, |
10269 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10270 | }, |
052f1dc7 PZ |
10271 | #endif |
10272 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10273 | { |
9f0c1e56 | 10274 | .name = "rt_runtime_us", |
06ecb27c PM |
10275 | .read_s64 = cpu_rt_runtime_read, |
10276 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10277 | }, |
d0b27fa7 PZ |
10278 | { |
10279 | .name = "rt_period_us", | |
f4c753b7 PM |
10280 | .read_u64 = cpu_rt_period_read_uint, |
10281 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10282 | }, |
052f1dc7 | 10283 | #endif |
68318b8e SV |
10284 | }; |
10285 | ||
10286 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10287 | { | |
fe5c7cc2 | 10288 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10289 | } |
10290 | ||
10291 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10292 | .name = "cpu", |
10293 | .create = cpu_cgroup_create, | |
10294 | .destroy = cpu_cgroup_destroy, | |
10295 | .can_attach = cpu_cgroup_can_attach, | |
10296 | .attach = cpu_cgroup_attach, | |
10297 | .populate = cpu_cgroup_populate, | |
10298 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10299 | .early_init = 1, |
10300 | }; | |
10301 | ||
052f1dc7 | 10302 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10303 | |
10304 | #ifdef CONFIG_CGROUP_CPUACCT | |
10305 | ||
10306 | /* | |
10307 | * CPU accounting code for task groups. | |
10308 | * | |
10309 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10310 | * (balbir@in.ibm.com). | |
10311 | */ | |
10312 | ||
934352f2 | 10313 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10314 | struct cpuacct { |
10315 | struct cgroup_subsys_state css; | |
10316 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10317 | u64 *cpuusage; | |
ef12fefa | 10318 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10319 | struct cpuacct *parent; |
d842de87 SV |
10320 | }; |
10321 | ||
10322 | struct cgroup_subsys cpuacct_subsys; | |
10323 | ||
10324 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10325 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10326 | { |
32cd756a | 10327 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10328 | struct cpuacct, css); |
10329 | } | |
10330 | ||
10331 | /* return cpu accounting group to which this task belongs */ | |
10332 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10333 | { | |
10334 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10335 | struct cpuacct, css); | |
10336 | } | |
10337 | ||
10338 | /* create a new cpu accounting group */ | |
10339 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10340 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10341 | { |
10342 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10343 | int i; |
d842de87 SV |
10344 | |
10345 | if (!ca) | |
ef12fefa | 10346 | goto out; |
d842de87 SV |
10347 | |
10348 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10349 | if (!ca->cpuusage) |
10350 | goto out_free_ca; | |
10351 | ||
10352 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10353 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10354 | goto out_free_counters; | |
d842de87 | 10355 | |
934352f2 BR |
10356 | if (cgrp->parent) |
10357 | ca->parent = cgroup_ca(cgrp->parent); | |
10358 | ||
d842de87 | 10359 | return &ca->css; |
ef12fefa BR |
10360 | |
10361 | out_free_counters: | |
10362 | while (--i >= 0) | |
10363 | percpu_counter_destroy(&ca->cpustat[i]); | |
10364 | free_percpu(ca->cpuusage); | |
10365 | out_free_ca: | |
10366 | kfree(ca); | |
10367 | out: | |
10368 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10369 | } |
10370 | ||
10371 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10372 | static void |
32cd756a | 10373 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10374 | { |
32cd756a | 10375 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10376 | int i; |
d842de87 | 10377 | |
ef12fefa BR |
10378 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10379 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10380 | free_percpu(ca->cpuusage); |
10381 | kfree(ca); | |
10382 | } | |
10383 | ||
720f5498 KC |
10384 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10385 | { | |
b36128c8 | 10386 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10387 | u64 data; |
10388 | ||
10389 | #ifndef CONFIG_64BIT | |
10390 | /* | |
10391 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10392 | */ | |
10393 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10394 | data = *cpuusage; | |
10395 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10396 | #else | |
10397 | data = *cpuusage; | |
10398 | #endif | |
10399 | ||
10400 | return data; | |
10401 | } | |
10402 | ||
10403 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10404 | { | |
b36128c8 | 10405 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10406 | |
10407 | #ifndef CONFIG_64BIT | |
10408 | /* | |
10409 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10410 | */ | |
10411 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10412 | *cpuusage = val; | |
10413 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10414 | #else | |
10415 | *cpuusage = val; | |
10416 | #endif | |
10417 | } | |
10418 | ||
d842de87 | 10419 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10420 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10421 | { |
32cd756a | 10422 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10423 | u64 totalcpuusage = 0; |
10424 | int i; | |
10425 | ||
720f5498 KC |
10426 | for_each_present_cpu(i) |
10427 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10428 | |
10429 | return totalcpuusage; | |
10430 | } | |
10431 | ||
0297b803 DG |
10432 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10433 | u64 reset) | |
10434 | { | |
10435 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10436 | int err = 0; | |
10437 | int i; | |
10438 | ||
10439 | if (reset) { | |
10440 | err = -EINVAL; | |
10441 | goto out; | |
10442 | } | |
10443 | ||
720f5498 KC |
10444 | for_each_present_cpu(i) |
10445 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10446 | |
0297b803 DG |
10447 | out: |
10448 | return err; | |
10449 | } | |
10450 | ||
e9515c3c KC |
10451 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10452 | struct seq_file *m) | |
10453 | { | |
10454 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10455 | u64 percpu; | |
10456 | int i; | |
10457 | ||
10458 | for_each_present_cpu(i) { | |
10459 | percpu = cpuacct_cpuusage_read(ca, i); | |
10460 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10461 | } | |
10462 | seq_printf(m, "\n"); | |
10463 | return 0; | |
10464 | } | |
10465 | ||
ef12fefa BR |
10466 | static const char *cpuacct_stat_desc[] = { |
10467 | [CPUACCT_STAT_USER] = "user", | |
10468 | [CPUACCT_STAT_SYSTEM] = "system", | |
10469 | }; | |
10470 | ||
10471 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10472 | struct cgroup_map_cb *cb) | |
10473 | { | |
10474 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10475 | int i; | |
10476 | ||
10477 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10478 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10479 | val = cputime64_to_clock_t(val); | |
10480 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10481 | } | |
10482 | return 0; | |
10483 | } | |
10484 | ||
d842de87 SV |
10485 | static struct cftype files[] = { |
10486 | { | |
10487 | .name = "usage", | |
f4c753b7 PM |
10488 | .read_u64 = cpuusage_read, |
10489 | .write_u64 = cpuusage_write, | |
d842de87 | 10490 | }, |
e9515c3c KC |
10491 | { |
10492 | .name = "usage_percpu", | |
10493 | .read_seq_string = cpuacct_percpu_seq_read, | |
10494 | }, | |
ef12fefa BR |
10495 | { |
10496 | .name = "stat", | |
10497 | .read_map = cpuacct_stats_show, | |
10498 | }, | |
d842de87 SV |
10499 | }; |
10500 | ||
32cd756a | 10501 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10502 | { |
32cd756a | 10503 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10504 | } |
10505 | ||
10506 | /* | |
10507 | * charge this task's execution time to its accounting group. | |
10508 | * | |
10509 | * called with rq->lock held. | |
10510 | */ | |
10511 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10512 | { | |
10513 | struct cpuacct *ca; | |
934352f2 | 10514 | int cpu; |
d842de87 | 10515 | |
c40c6f85 | 10516 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10517 | return; |
10518 | ||
934352f2 | 10519 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10520 | |
10521 | rcu_read_lock(); | |
10522 | ||
d842de87 | 10523 | ca = task_ca(tsk); |
d842de87 | 10524 | |
934352f2 | 10525 | for (; ca; ca = ca->parent) { |
b36128c8 | 10526 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10527 | *cpuusage += cputime; |
10528 | } | |
a18b83b7 BR |
10529 | |
10530 | rcu_read_unlock(); | |
d842de87 SV |
10531 | } |
10532 | ||
ef12fefa BR |
10533 | /* |
10534 | * Charge the system/user time to the task's accounting group. | |
10535 | */ | |
10536 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10537 | enum cpuacct_stat_index idx, cputime_t val) | |
10538 | { | |
10539 | struct cpuacct *ca; | |
10540 | ||
10541 | if (unlikely(!cpuacct_subsys.active)) | |
10542 | return; | |
10543 | ||
10544 | rcu_read_lock(); | |
10545 | ca = task_ca(tsk); | |
10546 | ||
10547 | do { | |
10548 | percpu_counter_add(&ca->cpustat[idx], val); | |
10549 | ca = ca->parent; | |
10550 | } while (ca); | |
10551 | rcu_read_unlock(); | |
10552 | } | |
10553 | ||
d842de87 SV |
10554 | struct cgroup_subsys cpuacct_subsys = { |
10555 | .name = "cpuacct", | |
10556 | .create = cpuacct_create, | |
10557 | .destroy = cpuacct_destroy, | |
10558 | .populate = cpuacct_populate, | |
10559 | .subsys_id = cpuacct_subsys_id, | |
10560 | }; | |
10561 | #endif /* CONFIG_CGROUP_CPUACCT */ |