]>
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; |
a1ba4d8b | 496 | unsigned long rt_nr_total; |
a22d7fc1 | 497 | int overloaded; |
917b627d | 498 | struct plist_head pushable_tasks; |
fa85ae24 | 499 | #endif |
6f505b16 | 500 | int rt_throttled; |
fa85ae24 | 501 | u64 rt_time; |
ac086bc2 | 502 | u64 rt_runtime; |
ea736ed5 | 503 | /* Nests inside the rq lock: */ |
ac086bc2 | 504 | spinlock_t rt_runtime_lock; |
6f505b16 | 505 | |
052f1dc7 | 506 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
507 | unsigned long rt_nr_boosted; |
508 | ||
6f505b16 PZ |
509 | struct rq *rq; |
510 | struct list_head leaf_rt_rq_list; | |
511 | struct task_group *tg; | |
512 | struct sched_rt_entity *rt_se; | |
513 | #endif | |
6aa645ea IM |
514 | }; |
515 | ||
57d885fe GH |
516 | #ifdef CONFIG_SMP |
517 | ||
518 | /* | |
519 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
520 | * variables. Each exclusive cpuset essentially defines an island domain by |
521 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
522 | * exclusive cpuset is created, we also create and attach a new root-domain |
523 | * object. | |
524 | * | |
57d885fe GH |
525 | */ |
526 | struct root_domain { | |
527 | atomic_t refcount; | |
c6c4927b RR |
528 | cpumask_var_t span; |
529 | cpumask_var_t online; | |
637f5085 | 530 | |
0eab9146 | 531 | /* |
637f5085 GH |
532 | * The "RT overload" flag: it gets set if a CPU has more than |
533 | * one runnable RT task. | |
534 | */ | |
c6c4927b | 535 | cpumask_var_t rto_mask; |
0eab9146 | 536 | atomic_t rto_count; |
6e0534f2 GH |
537 | #ifdef CONFIG_SMP |
538 | struct cpupri cpupri; | |
539 | #endif | |
7a09b1a2 VS |
540 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
541 | /* | |
542 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
543 | * used when most cpus are idle in the system indicating overall very | |
544 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
545 | */ | |
546 | unsigned int sched_mc_preferred_wakeup_cpu; | |
547 | #endif | |
57d885fe GH |
548 | }; |
549 | ||
dc938520 GH |
550 | /* |
551 | * By default the system creates a single root-domain with all cpus as | |
552 | * members (mimicking the global state we have today). | |
553 | */ | |
57d885fe GH |
554 | static struct root_domain def_root_domain; |
555 | ||
556 | #endif | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * This is the main, per-CPU runqueue data structure. | |
560 | * | |
561 | * Locking rule: those places that want to lock multiple runqueues | |
562 | * (such as the load balancing or the thread migration code), lock | |
563 | * acquire operations must be ordered by ascending &runqueue. | |
564 | */ | |
70b97a7f | 565 | struct rq { |
d8016491 IM |
566 | /* runqueue lock: */ |
567 | spinlock_t lock; | |
1da177e4 LT |
568 | |
569 | /* | |
570 | * nr_running and cpu_load should be in the same cacheline because | |
571 | * remote CPUs use both these fields when doing load calculation. | |
572 | */ | |
573 | unsigned long nr_running; | |
6aa645ea IM |
574 | #define CPU_LOAD_IDX_MAX 5 |
575 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 576 | #ifdef CONFIG_NO_HZ |
15934a37 | 577 | unsigned long last_tick_seen; |
46cb4b7c SS |
578 | unsigned char in_nohz_recently; |
579 | #endif | |
d8016491 IM |
580 | /* capture load from *all* tasks on this cpu: */ |
581 | struct load_weight load; | |
6aa645ea IM |
582 | unsigned long nr_load_updates; |
583 | u64 nr_switches; | |
23a185ca | 584 | u64 nr_migrations_in; |
6aa645ea IM |
585 | |
586 | struct cfs_rq cfs; | |
6f505b16 | 587 | struct rt_rq rt; |
6f505b16 | 588 | |
6aa645ea | 589 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
590 | /* list of leaf cfs_rq on this cpu: */ |
591 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
592 | #endif |
593 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 594 | struct list_head leaf_rt_rq_list; |
1da177e4 | 595 | #endif |
1da177e4 LT |
596 | |
597 | /* | |
598 | * This is part of a global counter where only the total sum | |
599 | * over all CPUs matters. A task can increase this counter on | |
600 | * one CPU and if it got migrated afterwards it may decrease | |
601 | * it on another CPU. Always updated under the runqueue lock: | |
602 | */ | |
603 | unsigned long nr_uninterruptible; | |
604 | ||
36c8b586 | 605 | struct task_struct *curr, *idle; |
c9819f45 | 606 | unsigned long next_balance; |
1da177e4 | 607 | struct mm_struct *prev_mm; |
6aa645ea | 608 | |
3e51f33f | 609 | u64 clock; |
6aa645ea | 610 | |
1da177e4 LT |
611 | atomic_t nr_iowait; |
612 | ||
613 | #ifdef CONFIG_SMP | |
0eab9146 | 614 | struct root_domain *rd; |
1da177e4 LT |
615 | struct sched_domain *sd; |
616 | ||
a0a522ce | 617 | unsigned char idle_at_tick; |
1da177e4 LT |
618 | /* For active balancing */ |
619 | int active_balance; | |
620 | int push_cpu; | |
d8016491 IM |
621 | /* cpu of this runqueue: */ |
622 | int cpu; | |
1f11eb6a | 623 | int online; |
1da177e4 | 624 | |
a8a51d5e | 625 | unsigned long avg_load_per_task; |
1da177e4 | 626 | |
36c8b586 | 627 | struct task_struct *migration_thread; |
1da177e4 LT |
628 | struct list_head migration_queue; |
629 | #endif | |
630 | ||
dce48a84 TG |
631 | /* calc_load related fields */ |
632 | unsigned long calc_load_update; | |
633 | long calc_load_active; | |
634 | ||
8f4d37ec | 635 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
636 | #ifdef CONFIG_SMP |
637 | int hrtick_csd_pending; | |
638 | struct call_single_data hrtick_csd; | |
639 | #endif | |
8f4d37ec PZ |
640 | struct hrtimer hrtick_timer; |
641 | #endif | |
642 | ||
1da177e4 LT |
643 | #ifdef CONFIG_SCHEDSTATS |
644 | /* latency stats */ | |
645 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
646 | unsigned long long rq_cpu_time; |
647 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
648 | |
649 | /* sys_sched_yield() stats */ | |
480b9434 | 650 | unsigned int yld_count; |
1da177e4 LT |
651 | |
652 | /* schedule() stats */ | |
480b9434 KC |
653 | unsigned int sched_switch; |
654 | unsigned int sched_count; | |
655 | unsigned int sched_goidle; | |
1da177e4 LT |
656 | |
657 | /* try_to_wake_up() stats */ | |
480b9434 KC |
658 | unsigned int ttwu_count; |
659 | unsigned int ttwu_local; | |
b8efb561 IM |
660 | |
661 | /* BKL stats */ | |
480b9434 | 662 | unsigned int bkl_count; |
1da177e4 LT |
663 | #endif |
664 | }; | |
665 | ||
f34e3b61 | 666 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 667 | |
15afe09b | 668 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 669 | { |
15afe09b | 670 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
671 | } |
672 | ||
0a2966b4 CL |
673 | static inline int cpu_of(struct rq *rq) |
674 | { | |
675 | #ifdef CONFIG_SMP | |
676 | return rq->cpu; | |
677 | #else | |
678 | return 0; | |
679 | #endif | |
680 | } | |
681 | ||
674311d5 NP |
682 | /* |
683 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 684 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
685 | * |
686 | * The domain tree of any CPU may only be accessed from within | |
687 | * preempt-disabled sections. | |
688 | */ | |
48f24c4d IM |
689 | #define for_each_domain(cpu, __sd) \ |
690 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
691 | |
692 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
693 | #define this_rq() (&__get_cpu_var(runqueues)) | |
694 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
695 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
696 | ||
aa9c4c0f | 697 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
698 | { |
699 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
700 | } | |
701 | ||
bf5c91ba IM |
702 | /* |
703 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
704 | */ | |
705 | #ifdef CONFIG_SCHED_DEBUG | |
706 | # define const_debug __read_mostly | |
707 | #else | |
708 | # define const_debug static const | |
709 | #endif | |
710 | ||
017730c1 IM |
711 | /** |
712 | * runqueue_is_locked | |
713 | * | |
714 | * Returns true if the current cpu runqueue is locked. | |
715 | * This interface allows printk to be called with the runqueue lock | |
716 | * held and know whether or not it is OK to wake up the klogd. | |
717 | */ | |
718 | int runqueue_is_locked(void) | |
719 | { | |
720 | int cpu = get_cpu(); | |
721 | struct rq *rq = cpu_rq(cpu); | |
722 | int ret; | |
723 | ||
724 | ret = spin_is_locked(&rq->lock); | |
725 | put_cpu(); | |
726 | return ret; | |
727 | } | |
728 | ||
bf5c91ba IM |
729 | /* |
730 | * Debugging: various feature bits | |
731 | */ | |
f00b45c1 PZ |
732 | |
733 | #define SCHED_FEAT(name, enabled) \ | |
734 | __SCHED_FEAT_##name , | |
735 | ||
bf5c91ba | 736 | enum { |
f00b45c1 | 737 | #include "sched_features.h" |
bf5c91ba IM |
738 | }; |
739 | ||
f00b45c1 PZ |
740 | #undef SCHED_FEAT |
741 | ||
742 | #define SCHED_FEAT(name, enabled) \ | |
743 | (1UL << __SCHED_FEAT_##name) * enabled | | |
744 | ||
bf5c91ba | 745 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
746 | #include "sched_features.h" |
747 | 0; | |
748 | ||
749 | #undef SCHED_FEAT | |
750 | ||
751 | #ifdef CONFIG_SCHED_DEBUG | |
752 | #define SCHED_FEAT(name, enabled) \ | |
753 | #name , | |
754 | ||
983ed7a6 | 755 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
756 | #include "sched_features.h" |
757 | NULL | |
758 | }; | |
759 | ||
760 | #undef SCHED_FEAT | |
761 | ||
34f3a814 | 762 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 763 | { |
f00b45c1 PZ |
764 | int i; |
765 | ||
766 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
767 | if (!(sysctl_sched_features & (1UL << i))) |
768 | seq_puts(m, "NO_"); | |
769 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 770 | } |
34f3a814 | 771 | seq_puts(m, "\n"); |
f00b45c1 | 772 | |
34f3a814 | 773 | return 0; |
f00b45c1 PZ |
774 | } |
775 | ||
776 | static ssize_t | |
777 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
778 | size_t cnt, loff_t *ppos) | |
779 | { | |
780 | char buf[64]; | |
781 | char *cmp = buf; | |
782 | int neg = 0; | |
783 | int i; | |
784 | ||
785 | if (cnt > 63) | |
786 | cnt = 63; | |
787 | ||
788 | if (copy_from_user(&buf, ubuf, cnt)) | |
789 | return -EFAULT; | |
790 | ||
791 | buf[cnt] = 0; | |
792 | ||
c24b7c52 | 793 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
794 | neg = 1; |
795 | cmp += 3; | |
796 | } | |
797 | ||
798 | for (i = 0; sched_feat_names[i]; i++) { | |
799 | int len = strlen(sched_feat_names[i]); | |
800 | ||
801 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
802 | if (neg) | |
803 | sysctl_sched_features &= ~(1UL << i); | |
804 | else | |
805 | sysctl_sched_features |= (1UL << i); | |
806 | break; | |
807 | } | |
808 | } | |
809 | ||
810 | if (!sched_feat_names[i]) | |
811 | return -EINVAL; | |
812 | ||
813 | filp->f_pos += cnt; | |
814 | ||
815 | return cnt; | |
816 | } | |
817 | ||
34f3a814 LZ |
818 | static int sched_feat_open(struct inode *inode, struct file *filp) |
819 | { | |
820 | return single_open(filp, sched_feat_show, NULL); | |
821 | } | |
822 | ||
f00b45c1 | 823 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
824 | .open = sched_feat_open, |
825 | .write = sched_feat_write, | |
826 | .read = seq_read, | |
827 | .llseek = seq_lseek, | |
828 | .release = single_release, | |
f00b45c1 PZ |
829 | }; |
830 | ||
831 | static __init int sched_init_debug(void) | |
832 | { | |
f00b45c1 PZ |
833 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
834 | &sched_feat_fops); | |
835 | ||
836 | return 0; | |
837 | } | |
838 | late_initcall(sched_init_debug); | |
839 | ||
840 | #endif | |
841 | ||
842 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 843 | |
b82d9fdd PZ |
844 | /* |
845 | * Number of tasks to iterate in a single balance run. | |
846 | * Limited because this is done with IRQs disabled. | |
847 | */ | |
848 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
849 | ||
2398f2c6 PZ |
850 | /* |
851 | * ratelimit for updating the group shares. | |
55cd5340 | 852 | * default: 0.25ms |
2398f2c6 | 853 | */ |
55cd5340 | 854 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 855 | |
ffda12a1 PZ |
856 | /* |
857 | * Inject some fuzzyness into changing the per-cpu group shares | |
858 | * this avoids remote rq-locks at the expense of fairness. | |
859 | * default: 4 | |
860 | */ | |
861 | unsigned int sysctl_sched_shares_thresh = 4; | |
862 | ||
fa85ae24 | 863 | /* |
9f0c1e56 | 864 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
865 | * default: 1s |
866 | */ | |
9f0c1e56 | 867 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 868 | |
6892b75e IM |
869 | static __read_mostly int scheduler_running; |
870 | ||
9f0c1e56 PZ |
871 | /* |
872 | * part of the period that we allow rt tasks to run in us. | |
873 | * default: 0.95s | |
874 | */ | |
875 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 876 | |
d0b27fa7 PZ |
877 | static inline u64 global_rt_period(void) |
878 | { | |
879 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
880 | } | |
881 | ||
882 | static inline u64 global_rt_runtime(void) | |
883 | { | |
e26873bb | 884 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
885 | return RUNTIME_INF; |
886 | ||
887 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
888 | } | |
fa85ae24 | 889 | |
1da177e4 | 890 | #ifndef prepare_arch_switch |
4866cde0 NP |
891 | # define prepare_arch_switch(next) do { } while (0) |
892 | #endif | |
893 | #ifndef finish_arch_switch | |
894 | # define finish_arch_switch(prev) do { } while (0) | |
895 | #endif | |
896 | ||
051a1d1a DA |
897 | static inline int task_current(struct rq *rq, struct task_struct *p) |
898 | { | |
899 | return rq->curr == p; | |
900 | } | |
901 | ||
4866cde0 | 902 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 903 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 904 | { |
051a1d1a | 905 | return task_current(rq, p); |
4866cde0 NP |
906 | } |
907 | ||
70b97a7f | 908 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
909 | { |
910 | } | |
911 | ||
70b97a7f | 912 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 913 | { |
da04c035 IM |
914 | #ifdef CONFIG_DEBUG_SPINLOCK |
915 | /* this is a valid case when another task releases the spinlock */ | |
916 | rq->lock.owner = current; | |
917 | #endif | |
8a25d5de IM |
918 | /* |
919 | * If we are tracking spinlock dependencies then we have to | |
920 | * fix up the runqueue lock - which gets 'carried over' from | |
921 | * prev into current: | |
922 | */ | |
923 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
924 | ||
4866cde0 NP |
925 | spin_unlock_irq(&rq->lock); |
926 | } | |
927 | ||
928 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 929 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
930 | { |
931 | #ifdef CONFIG_SMP | |
932 | return p->oncpu; | |
933 | #else | |
051a1d1a | 934 | return task_current(rq, p); |
4866cde0 NP |
935 | #endif |
936 | } | |
937 | ||
70b97a7f | 938 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
939 | { |
940 | #ifdef CONFIG_SMP | |
941 | /* | |
942 | * We can optimise this out completely for !SMP, because the | |
943 | * SMP rebalancing from interrupt is the only thing that cares | |
944 | * here. | |
945 | */ | |
946 | next->oncpu = 1; | |
947 | #endif | |
948 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
949 | spin_unlock_irq(&rq->lock); | |
950 | #else | |
951 | spin_unlock(&rq->lock); | |
952 | #endif | |
953 | } | |
954 | ||
70b97a7f | 955 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
956 | { |
957 | #ifdef CONFIG_SMP | |
958 | /* | |
959 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
960 | * We must ensure this doesn't happen until the switch is completely | |
961 | * finished. | |
962 | */ | |
963 | smp_wmb(); | |
964 | prev->oncpu = 0; | |
965 | #endif | |
966 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
967 | local_irq_enable(); | |
1da177e4 | 968 | #endif |
4866cde0 NP |
969 | } |
970 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 971 | |
b29739f9 IM |
972 | /* |
973 | * __task_rq_lock - lock the runqueue a given task resides on. | |
974 | * Must be called interrupts disabled. | |
975 | */ | |
70b97a7f | 976 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
977 | __acquires(rq->lock) |
978 | { | |
3a5c359a AK |
979 | for (;;) { |
980 | struct rq *rq = task_rq(p); | |
981 | spin_lock(&rq->lock); | |
982 | if (likely(rq == task_rq(p))) | |
983 | return rq; | |
b29739f9 | 984 | spin_unlock(&rq->lock); |
b29739f9 | 985 | } |
b29739f9 IM |
986 | } |
987 | ||
1da177e4 LT |
988 | /* |
989 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 990 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
991 | * explicitly disabling preemption. |
992 | */ | |
70b97a7f | 993 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
994 | __acquires(rq->lock) |
995 | { | |
70b97a7f | 996 | struct rq *rq; |
1da177e4 | 997 | |
3a5c359a AK |
998 | for (;;) { |
999 | local_irq_save(*flags); | |
1000 | rq = task_rq(p); | |
1001 | spin_lock(&rq->lock); | |
1002 | if (likely(rq == task_rq(p))) | |
1003 | return rq; | |
1da177e4 | 1004 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1005 | } |
1da177e4 LT |
1006 | } |
1007 | ||
ad474cac ON |
1008 | void task_rq_unlock_wait(struct task_struct *p) |
1009 | { | |
1010 | struct rq *rq = task_rq(p); | |
1011 | ||
1012 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1013 | spin_unlock_wait(&rq->lock); | |
1014 | } | |
1015 | ||
a9957449 | 1016 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1017 | __releases(rq->lock) |
1018 | { | |
1019 | spin_unlock(&rq->lock); | |
1020 | } | |
1021 | ||
70b97a7f | 1022 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1023 | __releases(rq->lock) |
1024 | { | |
1025 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1026 | } | |
1027 | ||
1da177e4 | 1028 | /* |
cc2a73b5 | 1029 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1030 | */ |
a9957449 | 1031 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1032 | __acquires(rq->lock) |
1033 | { | |
70b97a7f | 1034 | struct rq *rq; |
1da177e4 LT |
1035 | |
1036 | local_irq_disable(); | |
1037 | rq = this_rq(); | |
1038 | spin_lock(&rq->lock); | |
1039 | ||
1040 | return rq; | |
1041 | } | |
1042 | ||
8f4d37ec PZ |
1043 | #ifdef CONFIG_SCHED_HRTICK |
1044 | /* | |
1045 | * Use HR-timers to deliver accurate preemption points. | |
1046 | * | |
1047 | * Its all a bit involved since we cannot program an hrt while holding the | |
1048 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1049 | * reschedule event. | |
1050 | * | |
1051 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1052 | * rq->lock. | |
1053 | */ | |
8f4d37ec PZ |
1054 | |
1055 | /* | |
1056 | * Use hrtick when: | |
1057 | * - enabled by features | |
1058 | * - hrtimer is actually high res | |
1059 | */ | |
1060 | static inline int hrtick_enabled(struct rq *rq) | |
1061 | { | |
1062 | if (!sched_feat(HRTICK)) | |
1063 | return 0; | |
ba42059f | 1064 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1065 | return 0; |
8f4d37ec PZ |
1066 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1067 | } | |
1068 | ||
8f4d37ec PZ |
1069 | static void hrtick_clear(struct rq *rq) |
1070 | { | |
1071 | if (hrtimer_active(&rq->hrtick_timer)) | |
1072 | hrtimer_cancel(&rq->hrtick_timer); | |
1073 | } | |
1074 | ||
8f4d37ec PZ |
1075 | /* |
1076 | * High-resolution timer tick. | |
1077 | * Runs from hardirq context with interrupts disabled. | |
1078 | */ | |
1079 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1080 | { | |
1081 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1082 | ||
1083 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1084 | ||
1085 | spin_lock(&rq->lock); | |
3e51f33f | 1086 | update_rq_clock(rq); |
8f4d37ec PZ |
1087 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1088 | spin_unlock(&rq->lock); | |
1089 | ||
1090 | return HRTIMER_NORESTART; | |
1091 | } | |
1092 | ||
95e904c7 | 1093 | #ifdef CONFIG_SMP |
31656519 PZ |
1094 | /* |
1095 | * called from hardirq (IPI) context | |
1096 | */ | |
1097 | static void __hrtick_start(void *arg) | |
b328ca18 | 1098 | { |
31656519 | 1099 | struct rq *rq = arg; |
b328ca18 | 1100 | |
31656519 PZ |
1101 | spin_lock(&rq->lock); |
1102 | hrtimer_restart(&rq->hrtick_timer); | |
1103 | rq->hrtick_csd_pending = 0; | |
1104 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1105 | } |
1106 | ||
31656519 PZ |
1107 | /* |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1113 | { |
31656519 PZ |
1114 | struct hrtimer *timer = &rq->hrtick_timer; |
1115 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1116 | |
cc584b21 | 1117 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1118 | |
1119 | if (rq == this_rq()) { | |
1120 | hrtimer_restart(timer); | |
1121 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1122 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1123 | rq->hrtick_csd_pending = 1; |
1124 | } | |
b328ca18 PZ |
1125 | } |
1126 | ||
1127 | static int | |
1128 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1129 | { | |
1130 | int cpu = (int)(long)hcpu; | |
1131 | ||
1132 | switch (action) { | |
1133 | case CPU_UP_CANCELED: | |
1134 | case CPU_UP_CANCELED_FROZEN: | |
1135 | case CPU_DOWN_PREPARE: | |
1136 | case CPU_DOWN_PREPARE_FROZEN: | |
1137 | case CPU_DEAD: | |
1138 | case CPU_DEAD_FROZEN: | |
31656519 | 1139 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1140 | return NOTIFY_OK; |
1141 | } | |
1142 | ||
1143 | return NOTIFY_DONE; | |
1144 | } | |
1145 | ||
fa748203 | 1146 | static __init void init_hrtick(void) |
b328ca18 PZ |
1147 | { |
1148 | hotcpu_notifier(hotplug_hrtick, 0); | |
1149 | } | |
31656519 PZ |
1150 | #else |
1151 | /* | |
1152 | * Called to set the hrtick timer state. | |
1153 | * | |
1154 | * called with rq->lock held and irqs disabled | |
1155 | */ | |
1156 | static void hrtick_start(struct rq *rq, u64 delay) | |
1157 | { | |
7f1e2ca9 | 1158 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1159 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1160 | } |
b328ca18 | 1161 | |
006c75f1 | 1162 | static inline void init_hrtick(void) |
8f4d37ec | 1163 | { |
8f4d37ec | 1164 | } |
31656519 | 1165 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1166 | |
31656519 | 1167 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1168 | { |
31656519 PZ |
1169 | #ifdef CONFIG_SMP |
1170 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1171 | |
31656519 PZ |
1172 | rq->hrtick_csd.flags = 0; |
1173 | rq->hrtick_csd.func = __hrtick_start; | |
1174 | rq->hrtick_csd.info = rq; | |
1175 | #endif | |
8f4d37ec | 1176 | |
31656519 PZ |
1177 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1178 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1179 | } |
006c75f1 | 1180 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1181 | static inline void hrtick_clear(struct rq *rq) |
1182 | { | |
1183 | } | |
1184 | ||
8f4d37ec PZ |
1185 | static inline void init_rq_hrtick(struct rq *rq) |
1186 | { | |
1187 | } | |
1188 | ||
b328ca18 PZ |
1189 | static inline void init_hrtick(void) |
1190 | { | |
1191 | } | |
006c75f1 | 1192 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1193 | |
c24d20db IM |
1194 | /* |
1195 | * resched_task - mark a task 'to be rescheduled now'. | |
1196 | * | |
1197 | * On UP this means the setting of the need_resched flag, on SMP it | |
1198 | * might also involve a cross-CPU call to trigger the scheduler on | |
1199 | * the target CPU. | |
1200 | */ | |
1201 | #ifdef CONFIG_SMP | |
1202 | ||
1203 | #ifndef tsk_is_polling | |
1204 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1205 | #endif | |
1206 | ||
31656519 | 1207 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1208 | { |
1209 | int cpu; | |
1210 | ||
1211 | assert_spin_locked(&task_rq(p)->lock); | |
1212 | ||
5ed0cec0 | 1213 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1214 | return; |
1215 | ||
5ed0cec0 | 1216 | set_tsk_need_resched(p); |
c24d20db IM |
1217 | |
1218 | cpu = task_cpu(p); | |
1219 | if (cpu == smp_processor_id()) | |
1220 | return; | |
1221 | ||
1222 | /* NEED_RESCHED must be visible before we test polling */ | |
1223 | smp_mb(); | |
1224 | if (!tsk_is_polling(p)) | |
1225 | smp_send_reschedule(cpu); | |
1226 | } | |
1227 | ||
1228 | static void resched_cpu(int cpu) | |
1229 | { | |
1230 | struct rq *rq = cpu_rq(cpu); | |
1231 | unsigned long flags; | |
1232 | ||
1233 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1234 | return; | |
1235 | resched_task(cpu_curr(cpu)); | |
1236 | spin_unlock_irqrestore(&rq->lock, flags); | |
1237 | } | |
06d8308c TG |
1238 | |
1239 | #ifdef CONFIG_NO_HZ | |
1240 | /* | |
1241 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1242 | * idle CPU then this timer might expire before the next timer event | |
1243 | * which is scheduled to wake up that CPU. In case of a completely | |
1244 | * idle system the next event might even be infinite time into the | |
1245 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1246 | * leaves the inner idle loop so the newly added timer is taken into | |
1247 | * account when the CPU goes back to idle and evaluates the timer | |
1248 | * wheel for the next timer event. | |
1249 | */ | |
1250 | void wake_up_idle_cpu(int cpu) | |
1251 | { | |
1252 | struct rq *rq = cpu_rq(cpu); | |
1253 | ||
1254 | if (cpu == smp_processor_id()) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * This is safe, as this function is called with the timer | |
1259 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1260 | * to idle and has not yet set rq->curr to idle then it will | |
1261 | * be serialized on the timer wheel base lock and take the new | |
1262 | * timer into account automatically. | |
1263 | */ | |
1264 | if (rq->curr != rq->idle) | |
1265 | return; | |
1266 | ||
1267 | /* | |
1268 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1269 | * lockless. The worst case is that the other CPU runs the | |
1270 | * idle task through an additional NOOP schedule() | |
1271 | */ | |
5ed0cec0 | 1272 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1273 | |
1274 | /* NEED_RESCHED must be visible before we test polling */ | |
1275 | smp_mb(); | |
1276 | if (!tsk_is_polling(rq->idle)) | |
1277 | smp_send_reschedule(cpu); | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1280 | |
6d6bc0ad | 1281 | #else /* !CONFIG_SMP */ |
31656519 | 1282 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1283 | { |
1284 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1285 | set_tsk_need_resched(p); |
c24d20db | 1286 | } |
6d6bc0ad | 1287 | #endif /* CONFIG_SMP */ |
c24d20db | 1288 | |
45bf76df IM |
1289 | #if BITS_PER_LONG == 32 |
1290 | # define WMULT_CONST (~0UL) | |
1291 | #else | |
1292 | # define WMULT_CONST (1UL << 32) | |
1293 | #endif | |
1294 | ||
1295 | #define WMULT_SHIFT 32 | |
1296 | ||
194081eb IM |
1297 | /* |
1298 | * Shift right and round: | |
1299 | */ | |
cf2ab469 | 1300 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1301 | |
a7be37ac PZ |
1302 | /* |
1303 | * delta *= weight / lw | |
1304 | */ | |
cb1c4fc9 | 1305 | static unsigned long |
45bf76df IM |
1306 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1307 | struct load_weight *lw) | |
1308 | { | |
1309 | u64 tmp; | |
1310 | ||
7a232e03 LJ |
1311 | if (!lw->inv_weight) { |
1312 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1313 | lw->inv_weight = 1; | |
1314 | else | |
1315 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1316 | / (lw->weight+1); | |
1317 | } | |
45bf76df IM |
1318 | |
1319 | tmp = (u64)delta_exec * weight; | |
1320 | /* | |
1321 | * Check whether we'd overflow the 64-bit multiplication: | |
1322 | */ | |
194081eb | 1323 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1324 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1325 | WMULT_SHIFT/2); |
1326 | else | |
cf2ab469 | 1327 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1328 | |
ecf691da | 1329 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1333 | { |
1334 | lw->weight += inc; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
1091985b | 1338 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1339 | { |
1340 | lw->weight -= dec; | |
e89996ae | 1341 | lw->inv_weight = 0; |
45bf76df IM |
1342 | } |
1343 | ||
2dd73a4f PW |
1344 | /* |
1345 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1346 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1347 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1348 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1349 | * scaled version of the new time slice allocation that they receive on time |
1350 | * slice expiry etc. | |
1351 | */ | |
1352 | ||
cce7ade8 PZ |
1353 | #define WEIGHT_IDLEPRIO 3 |
1354 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1355 | |
1356 | /* | |
1357 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1358 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1359 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1360 | * that remained on nice 0. | |
1361 | * | |
1362 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1363 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1364 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1365 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1366 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1367 | */ |
1368 | static const int prio_to_weight[40] = { | |
254753dc IM |
1369 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1370 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1371 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1372 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1373 | /* 0 */ 1024, 820, 655, 526, 423, | |
1374 | /* 5 */ 335, 272, 215, 172, 137, | |
1375 | /* 10 */ 110, 87, 70, 56, 45, | |
1376 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1377 | }; |
1378 | ||
5714d2de IM |
1379 | /* |
1380 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1381 | * | |
1382 | * In cases where the weight does not change often, we can use the | |
1383 | * precalculated inverse to speed up arithmetics by turning divisions | |
1384 | * into multiplications: | |
1385 | */ | |
dd41f596 | 1386 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1387 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1388 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1389 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1390 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1391 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1392 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1393 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1394 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1395 | }; |
2dd73a4f | 1396 | |
dd41f596 IM |
1397 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1398 | ||
1399 | /* | |
1400 | * runqueue iterator, to support SMP load-balancing between different | |
1401 | * scheduling classes, without having to expose their internal data | |
1402 | * structures to the load-balancing proper: | |
1403 | */ | |
1404 | struct rq_iterator { | |
1405 | void *arg; | |
1406 | struct task_struct *(*start)(void *); | |
1407 | struct task_struct *(*next)(void *); | |
1408 | }; | |
1409 | ||
e1d1484f PW |
1410 | #ifdef CONFIG_SMP |
1411 | static unsigned long | |
1412 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1413 | unsigned long max_load_move, struct sched_domain *sd, | |
1414 | enum cpu_idle_type idle, int *all_pinned, | |
1415 | int *this_best_prio, struct rq_iterator *iterator); | |
1416 | ||
1417 | static int | |
1418 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1419 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1420 | struct rq_iterator *iterator); | |
e1d1484f | 1421 | #endif |
dd41f596 | 1422 | |
ef12fefa BR |
1423 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1424 | enum cpuacct_stat_index { | |
1425 | CPUACCT_STAT_USER, /* ... user mode */ | |
1426 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1427 | ||
1428 | CPUACCT_STAT_NSTATS, | |
1429 | }; | |
1430 | ||
d842de87 SV |
1431 | #ifdef CONFIG_CGROUP_CPUACCT |
1432 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1433 | static void cpuacct_update_stats(struct task_struct *tsk, |
1434 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1435 | #else |
1436 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1437 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1438 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1439 | #endif |
1440 | ||
18d95a28 PZ |
1441 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1442 | { | |
1443 | update_load_add(&rq->load, load); | |
1444 | } | |
1445 | ||
1446 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1447 | { | |
1448 | update_load_sub(&rq->load, load); | |
1449 | } | |
1450 | ||
7940ca36 | 1451 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1452 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1453 | |
1454 | /* | |
1455 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1456 | * leaving it for the final time. | |
1457 | */ | |
eb755805 | 1458 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1459 | { |
1460 | struct task_group *parent, *child; | |
eb755805 | 1461 | int ret; |
c09595f6 PZ |
1462 | |
1463 | rcu_read_lock(); | |
1464 | parent = &root_task_group; | |
1465 | down: | |
eb755805 PZ |
1466 | ret = (*down)(parent, data); |
1467 | if (ret) | |
1468 | goto out_unlock; | |
c09595f6 PZ |
1469 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1470 | parent = child; | |
1471 | goto down; | |
1472 | ||
1473 | up: | |
1474 | continue; | |
1475 | } | |
eb755805 PZ |
1476 | ret = (*up)(parent, data); |
1477 | if (ret) | |
1478 | goto out_unlock; | |
c09595f6 PZ |
1479 | |
1480 | child = parent; | |
1481 | parent = parent->parent; | |
1482 | if (parent) | |
1483 | goto up; | |
eb755805 | 1484 | out_unlock: |
c09595f6 | 1485 | rcu_read_unlock(); |
eb755805 PZ |
1486 | |
1487 | return ret; | |
c09595f6 PZ |
1488 | } |
1489 | ||
eb755805 PZ |
1490 | static int tg_nop(struct task_group *tg, void *data) |
1491 | { | |
1492 | return 0; | |
c09595f6 | 1493 | } |
eb755805 PZ |
1494 | #endif |
1495 | ||
1496 | #ifdef CONFIG_SMP | |
1497 | static unsigned long source_load(int cpu, int type); | |
1498 | static unsigned long target_load(int cpu, int type); | |
1499 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1500 | ||
1501 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1502 | { | |
1503 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1504 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1505 | |
4cd42620 SR |
1506 | if (nr_running) |
1507 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1508 | else |
1509 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1510 | |
1511 | return rq->avg_load_per_task; | |
1512 | } | |
1513 | ||
1514 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1515 | |
c09595f6 PZ |
1516 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1517 | ||
1518 | /* | |
1519 | * Calculate and set the cpu's group shares. | |
1520 | */ | |
1521 | static void | |
ffda12a1 PZ |
1522 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1523 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1524 | { |
c09595f6 PZ |
1525 | unsigned long shares; |
1526 | unsigned long rq_weight; | |
1527 | ||
c8cba857 | 1528 | if (!tg->se[cpu]) |
c09595f6 PZ |
1529 | return; |
1530 | ||
ec4e0e2f | 1531 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1532 | |
c09595f6 PZ |
1533 | /* |
1534 | * \Sum shares * rq_weight | |
1535 | * shares = ----------------------- | |
1536 | * \Sum rq_weight | |
1537 | * | |
1538 | */ | |
ec4e0e2f | 1539 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1540 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1541 | |
ffda12a1 PZ |
1542 | if (abs(shares - tg->se[cpu]->load.weight) > |
1543 | sysctl_sched_shares_thresh) { | |
1544 | struct rq *rq = cpu_rq(cpu); | |
1545 | unsigned long flags; | |
c09595f6 | 1546 | |
ffda12a1 | 1547 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1548 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1549 | |
ffda12a1 PZ |
1550 | __set_se_shares(tg->se[cpu], shares); |
1551 | spin_unlock_irqrestore(&rq->lock, flags); | |
1552 | } | |
18d95a28 | 1553 | } |
c09595f6 PZ |
1554 | |
1555 | /* | |
c8cba857 PZ |
1556 | * Re-compute the task group their per cpu shares over the given domain. |
1557 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1558 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1559 | */ |
eb755805 | 1560 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1561 | { |
ec4e0e2f | 1562 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1563 | unsigned long shares = 0; |
eb755805 | 1564 | struct sched_domain *sd = data; |
c8cba857 | 1565 | int i; |
c09595f6 | 1566 | |
758b2cdc | 1567 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1568 | /* |
1569 | * If there are currently no tasks on the cpu pretend there | |
1570 | * is one of average load so that when a new task gets to | |
1571 | * run here it will not get delayed by group starvation. | |
1572 | */ | |
1573 | weight = tg->cfs_rq[i]->load.weight; | |
1574 | if (!weight) | |
1575 | weight = NICE_0_LOAD; | |
1576 | ||
1577 | tg->cfs_rq[i]->rq_weight = weight; | |
1578 | rq_weight += weight; | |
c8cba857 | 1579 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1580 | } |
c09595f6 | 1581 | |
c8cba857 PZ |
1582 | if ((!shares && rq_weight) || shares > tg->shares) |
1583 | shares = tg->shares; | |
1584 | ||
1585 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1586 | shares = tg->shares; | |
c09595f6 | 1587 | |
758b2cdc | 1588 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1589 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1590 | |
1591 | return 0; | |
c09595f6 PZ |
1592 | } |
1593 | ||
1594 | /* | |
c8cba857 PZ |
1595 | * Compute the cpu's hierarchical load factor for each task group. |
1596 | * This needs to be done in a top-down fashion because the load of a child | |
1597 | * group is a fraction of its parents load. | |
c09595f6 | 1598 | */ |
eb755805 | 1599 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1600 | { |
c8cba857 | 1601 | unsigned long load; |
eb755805 | 1602 | long cpu = (long)data; |
c09595f6 | 1603 | |
c8cba857 PZ |
1604 | if (!tg->parent) { |
1605 | load = cpu_rq(cpu)->load.weight; | |
1606 | } else { | |
1607 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1608 | load *= tg->cfs_rq[cpu]->shares; | |
1609 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1610 | } | |
c09595f6 | 1611 | |
c8cba857 | 1612 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1613 | |
eb755805 | 1614 | return 0; |
c09595f6 PZ |
1615 | } |
1616 | ||
c8cba857 | 1617 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1618 | { |
2398f2c6 PZ |
1619 | u64 now = cpu_clock(raw_smp_processor_id()); |
1620 | s64 elapsed = now - sd->last_update; | |
1621 | ||
1622 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1623 | sd->last_update = now; | |
eb755805 | 1624 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1625 | } |
4d8d595d PZ |
1626 | } |
1627 | ||
3e5459b4 PZ |
1628 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1629 | { | |
1630 | spin_unlock(&rq->lock); | |
1631 | update_shares(sd); | |
1632 | spin_lock(&rq->lock); | |
1633 | } | |
1634 | ||
eb755805 | 1635 | static void update_h_load(long cpu) |
c09595f6 | 1636 | { |
eb755805 | 1637 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1638 | } |
1639 | ||
c09595f6 PZ |
1640 | #else |
1641 | ||
c8cba857 | 1642 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1643 | { |
1644 | } | |
1645 | ||
3e5459b4 PZ |
1646 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1647 | { | |
1648 | } | |
1649 | ||
18d95a28 PZ |
1650 | #endif |
1651 | ||
8f45e2b5 GH |
1652 | #ifdef CONFIG_PREEMPT |
1653 | ||
70574a99 | 1654 | /* |
8f45e2b5 GH |
1655 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1656 | * way at the expense of forcing extra atomic operations in all | |
1657 | * invocations. This assures that the double_lock is acquired using the | |
1658 | * same underlying policy as the spinlock_t on this architecture, which | |
1659 | * reduces latency compared to the unfair variant below. However, it | |
1660 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1661 | */ |
8f45e2b5 GH |
1662 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1663 | __releases(this_rq->lock) | |
1664 | __acquires(busiest->lock) | |
1665 | __acquires(this_rq->lock) | |
1666 | { | |
1667 | spin_unlock(&this_rq->lock); | |
1668 | double_rq_lock(this_rq, busiest); | |
1669 | ||
1670 | return 1; | |
1671 | } | |
1672 | ||
1673 | #else | |
1674 | /* | |
1675 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1676 | * latency by eliminating extra atomic operations when the locks are | |
1677 | * already in proper order on entry. This favors lower cpu-ids and will | |
1678 | * grant the double lock to lower cpus over higher ids under contention, | |
1679 | * regardless of entry order into the function. | |
1680 | */ | |
1681 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1682 | __releases(this_rq->lock) |
1683 | __acquires(busiest->lock) | |
1684 | __acquires(this_rq->lock) | |
1685 | { | |
1686 | int ret = 0; | |
1687 | ||
70574a99 AD |
1688 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1689 | if (busiest < this_rq) { | |
1690 | spin_unlock(&this_rq->lock); | |
1691 | spin_lock(&busiest->lock); | |
1692 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1693 | ret = 1; | |
1694 | } else | |
1695 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1696 | } | |
1697 | return ret; | |
1698 | } | |
1699 | ||
8f45e2b5 GH |
1700 | #endif /* CONFIG_PREEMPT */ |
1701 | ||
1702 | /* | |
1703 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1704 | */ | |
1705 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1706 | { | |
1707 | if (unlikely(!irqs_disabled())) { | |
1708 | /* printk() doesn't work good under rq->lock */ | |
1709 | spin_unlock(&this_rq->lock); | |
1710 | BUG_ON(1); | |
1711 | } | |
1712 | ||
1713 | return _double_lock_balance(this_rq, busiest); | |
1714 | } | |
1715 | ||
70574a99 AD |
1716 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1717 | __releases(busiest->lock) | |
1718 | { | |
1719 | spin_unlock(&busiest->lock); | |
1720 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1721 | } | |
18d95a28 PZ |
1722 | #endif |
1723 | ||
30432094 | 1724 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1725 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1726 | { | |
30432094 | 1727 | #ifdef CONFIG_SMP |
34e83e85 IM |
1728 | cfs_rq->shares = shares; |
1729 | #endif | |
1730 | } | |
30432094 | 1731 | #endif |
e7693a36 | 1732 | |
dce48a84 TG |
1733 | static void calc_load_account_active(struct rq *this_rq); |
1734 | ||
dd41f596 | 1735 | #include "sched_stats.h" |
dd41f596 | 1736 | #include "sched_idletask.c" |
5522d5d5 IM |
1737 | #include "sched_fair.c" |
1738 | #include "sched_rt.c" | |
dd41f596 IM |
1739 | #ifdef CONFIG_SCHED_DEBUG |
1740 | # include "sched_debug.c" | |
1741 | #endif | |
1742 | ||
1743 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1744 | #define for_each_class(class) \ |
1745 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1746 | |
c09595f6 | 1747 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1748 | { |
1749 | rq->nr_running++; | |
9c217245 IM |
1750 | } |
1751 | ||
c09595f6 | 1752 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1753 | { |
1754 | rq->nr_running--; | |
9c217245 IM |
1755 | } |
1756 | ||
45bf76df IM |
1757 | static void set_load_weight(struct task_struct *p) |
1758 | { | |
1759 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1760 | p->se.load.weight = prio_to_weight[0] * 2; |
1761 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1762 | return; | |
1763 | } | |
45bf76df | 1764 | |
dd41f596 IM |
1765 | /* |
1766 | * SCHED_IDLE tasks get minimal weight: | |
1767 | */ | |
1768 | if (p->policy == SCHED_IDLE) { | |
1769 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1770 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1771 | return; | |
1772 | } | |
71f8bd46 | 1773 | |
dd41f596 IM |
1774 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1775 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1776 | } |
1777 | ||
2087a1ad GH |
1778 | static void update_avg(u64 *avg, u64 sample) |
1779 | { | |
1780 | s64 diff = sample - *avg; | |
1781 | *avg += diff >> 3; | |
1782 | } | |
1783 | ||
8159f87e | 1784 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1785 | { |
831451ac PZ |
1786 | if (wakeup) |
1787 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1788 | ||
dd41f596 | 1789 | sched_info_queued(p); |
fd390f6a | 1790 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1791 | p->se.on_rq = 1; |
71f8bd46 IM |
1792 | } |
1793 | ||
69be72c1 | 1794 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1795 | { |
831451ac PZ |
1796 | if (sleep) { |
1797 | if (p->se.last_wakeup) { | |
1798 | update_avg(&p->se.avg_overlap, | |
1799 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1800 | p->se.last_wakeup = 0; | |
1801 | } else { | |
1802 | update_avg(&p->se.avg_wakeup, | |
1803 | sysctl_sched_wakeup_granularity); | |
1804 | } | |
2087a1ad GH |
1805 | } |
1806 | ||
46ac22ba | 1807 | sched_info_dequeued(p); |
f02231e5 | 1808 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1809 | p->se.on_rq = 0; |
71f8bd46 IM |
1810 | } |
1811 | ||
14531189 | 1812 | /* |
dd41f596 | 1813 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1814 | */ |
14531189 IM |
1815 | static inline int __normal_prio(struct task_struct *p) |
1816 | { | |
dd41f596 | 1817 | return p->static_prio; |
14531189 IM |
1818 | } |
1819 | ||
b29739f9 IM |
1820 | /* |
1821 | * Calculate the expected normal priority: i.e. priority | |
1822 | * without taking RT-inheritance into account. Might be | |
1823 | * boosted by interactivity modifiers. Changes upon fork, | |
1824 | * setprio syscalls, and whenever the interactivity | |
1825 | * estimator recalculates. | |
1826 | */ | |
36c8b586 | 1827 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1828 | { |
1829 | int prio; | |
1830 | ||
e05606d3 | 1831 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1832 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1833 | else | |
1834 | prio = __normal_prio(p); | |
1835 | return prio; | |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * Calculate the current priority, i.e. the priority | |
1840 | * taken into account by the scheduler. This value might | |
1841 | * be boosted by RT tasks, or might be boosted by | |
1842 | * interactivity modifiers. Will be RT if the task got | |
1843 | * RT-boosted. If not then it returns p->normal_prio. | |
1844 | */ | |
36c8b586 | 1845 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1846 | { |
1847 | p->normal_prio = normal_prio(p); | |
1848 | /* | |
1849 | * If we are RT tasks or we were boosted to RT priority, | |
1850 | * keep the priority unchanged. Otherwise, update priority | |
1851 | * to the normal priority: | |
1852 | */ | |
1853 | if (!rt_prio(p->prio)) | |
1854 | return p->normal_prio; | |
1855 | return p->prio; | |
1856 | } | |
1857 | ||
1da177e4 | 1858 | /* |
dd41f596 | 1859 | * activate_task - move a task to the runqueue. |
1da177e4 | 1860 | */ |
dd41f596 | 1861 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1862 | { |
d9514f6c | 1863 | if (task_contributes_to_load(p)) |
dd41f596 | 1864 | rq->nr_uninterruptible--; |
1da177e4 | 1865 | |
8159f87e | 1866 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1867 | inc_nr_running(rq); |
1da177e4 LT |
1868 | } |
1869 | ||
1da177e4 LT |
1870 | /* |
1871 | * deactivate_task - remove a task from the runqueue. | |
1872 | */ | |
2e1cb74a | 1873 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1874 | { |
d9514f6c | 1875 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1876 | rq->nr_uninterruptible++; |
1877 | ||
69be72c1 | 1878 | dequeue_task(rq, p, sleep); |
c09595f6 | 1879 | dec_nr_running(rq); |
1da177e4 LT |
1880 | } |
1881 | ||
1da177e4 LT |
1882 | /** |
1883 | * task_curr - is this task currently executing on a CPU? | |
1884 | * @p: the task in question. | |
1885 | */ | |
36c8b586 | 1886 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1887 | { |
1888 | return cpu_curr(task_cpu(p)) == p; | |
1889 | } | |
1890 | ||
dd41f596 IM |
1891 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1892 | { | |
6f505b16 | 1893 | set_task_rq(p, cpu); |
dd41f596 | 1894 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1895 | /* |
1896 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1897 | * successfuly executed on another CPU. We must ensure that updates of | |
1898 | * per-task data have been completed by this moment. | |
1899 | */ | |
1900 | smp_wmb(); | |
dd41f596 | 1901 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1902 | #endif |
2dd73a4f PW |
1903 | } |
1904 | ||
cb469845 SR |
1905 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1906 | const struct sched_class *prev_class, | |
1907 | int oldprio, int running) | |
1908 | { | |
1909 | if (prev_class != p->sched_class) { | |
1910 | if (prev_class->switched_from) | |
1911 | prev_class->switched_from(rq, p, running); | |
1912 | p->sched_class->switched_to(rq, p, running); | |
1913 | } else | |
1914 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1915 | } | |
1916 | ||
1da177e4 | 1917 | #ifdef CONFIG_SMP |
c65cc870 | 1918 | |
e958b360 TG |
1919 | /* Used instead of source_load when we know the type == 0 */ |
1920 | static unsigned long weighted_cpuload(const int cpu) | |
1921 | { | |
1922 | return cpu_rq(cpu)->load.weight; | |
1923 | } | |
1924 | ||
cc367732 IM |
1925 | /* |
1926 | * Is this task likely cache-hot: | |
1927 | */ | |
e7693a36 | 1928 | static int |
cc367732 IM |
1929 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1930 | { | |
1931 | s64 delta; | |
1932 | ||
f540a608 IM |
1933 | /* |
1934 | * Buddy candidates are cache hot: | |
1935 | */ | |
4793241b PZ |
1936 | if (sched_feat(CACHE_HOT_BUDDY) && |
1937 | (&p->se == cfs_rq_of(&p->se)->next || | |
1938 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1939 | return 1; |
1940 | ||
cc367732 IM |
1941 | if (p->sched_class != &fair_sched_class) |
1942 | return 0; | |
1943 | ||
6bc1665b IM |
1944 | if (sysctl_sched_migration_cost == -1) |
1945 | return 1; | |
1946 | if (sysctl_sched_migration_cost == 0) | |
1947 | return 0; | |
1948 | ||
cc367732 IM |
1949 | delta = now - p->se.exec_start; |
1950 | ||
1951 | return delta < (s64)sysctl_sched_migration_cost; | |
1952 | } | |
1953 | ||
1954 | ||
dd41f596 | 1955 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1956 | { |
dd41f596 IM |
1957 | int old_cpu = task_cpu(p); |
1958 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1959 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1960 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1961 | u64 clock_offset; |
dd41f596 IM |
1962 | |
1963 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1964 | |
de1d7286 | 1965 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1966 | |
6cfb0d5d IM |
1967 | #ifdef CONFIG_SCHEDSTATS |
1968 | if (p->se.wait_start) | |
1969 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1970 | if (p->se.sleep_start) |
1971 | p->se.sleep_start -= clock_offset; | |
1972 | if (p->se.block_start) | |
1973 | p->se.block_start -= clock_offset; | |
6c594c21 | 1974 | #endif |
cc367732 | 1975 | if (old_cpu != new_cpu) { |
6c594c21 | 1976 | p->se.nr_migrations++; |
23a185ca | 1977 | new_rq->nr_migrations_in++; |
6c594c21 | 1978 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
1979 | if (task_hot(p, old_rq->clock, NULL)) |
1980 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 1981 | #endif |
e5289d4a PZ |
1982 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
1983 | 1, 1, NULL, 0); | |
6c594c21 | 1984 | } |
2830cf8c SV |
1985 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1986 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1987 | |
1988 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1989 | } |
1990 | ||
70b97a7f | 1991 | struct migration_req { |
1da177e4 | 1992 | struct list_head list; |
1da177e4 | 1993 | |
36c8b586 | 1994 | struct task_struct *task; |
1da177e4 LT |
1995 | int dest_cpu; |
1996 | ||
1da177e4 | 1997 | struct completion done; |
70b97a7f | 1998 | }; |
1da177e4 LT |
1999 | |
2000 | /* | |
2001 | * The task's runqueue lock must be held. | |
2002 | * Returns true if you have to wait for migration thread. | |
2003 | */ | |
36c8b586 | 2004 | static int |
70b97a7f | 2005 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2006 | { |
70b97a7f | 2007 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2008 | |
2009 | /* | |
2010 | * If the task is not on a runqueue (and not running), then | |
2011 | * it is sufficient to simply update the task's cpu field. | |
2012 | */ | |
dd41f596 | 2013 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2014 | set_task_cpu(p, dest_cpu); |
2015 | return 0; | |
2016 | } | |
2017 | ||
2018 | init_completion(&req->done); | |
1da177e4 LT |
2019 | req->task = p; |
2020 | req->dest_cpu = dest_cpu; | |
2021 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2022 | |
1da177e4 LT |
2023 | return 1; |
2024 | } | |
2025 | ||
a26b89f0 MM |
2026 | /* |
2027 | * wait_task_context_switch - wait for a thread to complete at least one | |
2028 | * context switch. | |
2029 | * | |
2030 | * @p must not be current. | |
2031 | */ | |
2032 | void wait_task_context_switch(struct task_struct *p) | |
2033 | { | |
2034 | unsigned long nvcsw, nivcsw, flags; | |
2035 | int running; | |
2036 | struct rq *rq; | |
2037 | ||
2038 | nvcsw = p->nvcsw; | |
2039 | nivcsw = p->nivcsw; | |
2040 | for (;;) { | |
2041 | /* | |
2042 | * The runqueue is assigned before the actual context | |
2043 | * switch. We need to take the runqueue lock. | |
2044 | * | |
2045 | * We could check initially without the lock but it is | |
2046 | * very likely that we need to take the lock in every | |
2047 | * iteration. | |
2048 | */ | |
2049 | rq = task_rq_lock(p, &flags); | |
2050 | running = task_running(rq, p); | |
2051 | task_rq_unlock(rq, &flags); | |
2052 | ||
2053 | if (likely(!running)) | |
2054 | break; | |
2055 | /* | |
2056 | * The switch count is incremented before the actual | |
2057 | * context switch. We thus wait for two switches to be | |
2058 | * sure at least one completed. | |
2059 | */ | |
2060 | if ((p->nvcsw - nvcsw) > 1) | |
2061 | break; | |
2062 | if ((p->nivcsw - nivcsw) > 1) | |
2063 | break; | |
2064 | ||
2065 | cpu_relax(); | |
2066 | } | |
2067 | } | |
2068 | ||
1da177e4 LT |
2069 | /* |
2070 | * wait_task_inactive - wait for a thread to unschedule. | |
2071 | * | |
85ba2d86 RM |
2072 | * If @match_state is nonzero, it's the @p->state value just checked and |
2073 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2074 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2075 | * we return a positive number (its total switch count). If a second call | |
2076 | * a short while later returns the same number, the caller can be sure that | |
2077 | * @p has remained unscheduled the whole time. | |
2078 | * | |
1da177e4 LT |
2079 | * The caller must ensure that the task *will* unschedule sometime soon, |
2080 | * else this function might spin for a *long* time. This function can't | |
2081 | * be called with interrupts off, or it may introduce deadlock with | |
2082 | * smp_call_function() if an IPI is sent by the same process we are | |
2083 | * waiting to become inactive. | |
2084 | */ | |
85ba2d86 | 2085 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2086 | { |
2087 | unsigned long flags; | |
dd41f596 | 2088 | int running, on_rq; |
85ba2d86 | 2089 | unsigned long ncsw; |
70b97a7f | 2090 | struct rq *rq; |
1da177e4 | 2091 | |
3a5c359a AK |
2092 | for (;;) { |
2093 | /* | |
2094 | * We do the initial early heuristics without holding | |
2095 | * any task-queue locks at all. We'll only try to get | |
2096 | * the runqueue lock when things look like they will | |
2097 | * work out! | |
2098 | */ | |
2099 | rq = task_rq(p); | |
fa490cfd | 2100 | |
3a5c359a AK |
2101 | /* |
2102 | * If the task is actively running on another CPU | |
2103 | * still, just relax and busy-wait without holding | |
2104 | * any locks. | |
2105 | * | |
2106 | * NOTE! Since we don't hold any locks, it's not | |
2107 | * even sure that "rq" stays as the right runqueue! | |
2108 | * But we don't care, since "task_running()" will | |
2109 | * return false if the runqueue has changed and p | |
2110 | * is actually now running somewhere else! | |
2111 | */ | |
85ba2d86 RM |
2112 | while (task_running(rq, p)) { |
2113 | if (match_state && unlikely(p->state != match_state)) | |
2114 | return 0; | |
3a5c359a | 2115 | cpu_relax(); |
85ba2d86 | 2116 | } |
fa490cfd | 2117 | |
3a5c359a AK |
2118 | /* |
2119 | * Ok, time to look more closely! We need the rq | |
2120 | * lock now, to be *sure*. If we're wrong, we'll | |
2121 | * just go back and repeat. | |
2122 | */ | |
2123 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2124 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2125 | running = task_running(rq, p); |
2126 | on_rq = p->se.on_rq; | |
85ba2d86 | 2127 | ncsw = 0; |
f31e11d8 | 2128 | if (!match_state || p->state == match_state) |
93dcf55f | 2129 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2130 | task_rq_unlock(rq, &flags); |
fa490cfd | 2131 | |
85ba2d86 RM |
2132 | /* |
2133 | * If it changed from the expected state, bail out now. | |
2134 | */ | |
2135 | if (unlikely(!ncsw)) | |
2136 | break; | |
2137 | ||
3a5c359a AK |
2138 | /* |
2139 | * Was it really running after all now that we | |
2140 | * checked with the proper locks actually held? | |
2141 | * | |
2142 | * Oops. Go back and try again.. | |
2143 | */ | |
2144 | if (unlikely(running)) { | |
2145 | cpu_relax(); | |
2146 | continue; | |
2147 | } | |
fa490cfd | 2148 | |
3a5c359a AK |
2149 | /* |
2150 | * It's not enough that it's not actively running, | |
2151 | * it must be off the runqueue _entirely_, and not | |
2152 | * preempted! | |
2153 | * | |
80dd99b3 | 2154 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2155 | * running right now), it's preempted, and we should |
2156 | * yield - it could be a while. | |
2157 | */ | |
2158 | if (unlikely(on_rq)) { | |
2159 | schedule_timeout_uninterruptible(1); | |
2160 | continue; | |
2161 | } | |
fa490cfd | 2162 | |
3a5c359a AK |
2163 | /* |
2164 | * Ahh, all good. It wasn't running, and it wasn't | |
2165 | * runnable, which means that it will never become | |
2166 | * running in the future either. We're all done! | |
2167 | */ | |
2168 | break; | |
2169 | } | |
85ba2d86 RM |
2170 | |
2171 | return ncsw; | |
1da177e4 LT |
2172 | } |
2173 | ||
2174 | /*** | |
2175 | * kick_process - kick a running thread to enter/exit the kernel | |
2176 | * @p: the to-be-kicked thread | |
2177 | * | |
2178 | * Cause a process which is running on another CPU to enter | |
2179 | * kernel-mode, without any delay. (to get signals handled.) | |
2180 | * | |
2181 | * NOTE: this function doesnt have to take the runqueue lock, | |
2182 | * because all it wants to ensure is that the remote task enters | |
2183 | * the kernel. If the IPI races and the task has been migrated | |
2184 | * to another CPU then no harm is done and the purpose has been | |
2185 | * achieved as well. | |
2186 | */ | |
36c8b586 | 2187 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2188 | { |
2189 | int cpu; | |
2190 | ||
2191 | preempt_disable(); | |
2192 | cpu = task_cpu(p); | |
2193 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2194 | smp_send_reschedule(cpu); | |
2195 | preempt_enable(); | |
2196 | } | |
b43e3521 | 2197 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2198 | |
2199 | /* | |
2dd73a4f PW |
2200 | * Return a low guess at the load of a migration-source cpu weighted |
2201 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2202 | * |
2203 | * We want to under-estimate the load of migration sources, to | |
2204 | * balance conservatively. | |
2205 | */ | |
a9957449 | 2206 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2207 | { |
70b97a7f | 2208 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2209 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2210 | |
93b75217 | 2211 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2212 | return total; |
b910472d | 2213 | |
dd41f596 | 2214 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2215 | } |
2216 | ||
2217 | /* | |
2dd73a4f PW |
2218 | * Return a high guess at the load of a migration-target cpu weighted |
2219 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2220 | */ |
a9957449 | 2221 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2222 | { |
70b97a7f | 2223 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2224 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2225 | |
93b75217 | 2226 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2227 | return total; |
3b0bd9bc | 2228 | |
dd41f596 | 2229 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2230 | } |
2231 | ||
147cbb4b NP |
2232 | /* |
2233 | * find_idlest_group finds and returns the least busy CPU group within the | |
2234 | * domain. | |
2235 | */ | |
2236 | static struct sched_group * | |
2237 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2238 | { | |
2239 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2240 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2241 | int load_idx = sd->forkexec_idx; | |
2242 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2243 | ||
2244 | do { | |
2245 | unsigned long load, avg_load; | |
2246 | int local_group; | |
2247 | int i; | |
2248 | ||
da5a5522 | 2249 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2250 | if (!cpumask_intersects(sched_group_cpus(group), |
2251 | &p->cpus_allowed)) | |
3a5c359a | 2252 | continue; |
da5a5522 | 2253 | |
758b2cdc RR |
2254 | local_group = cpumask_test_cpu(this_cpu, |
2255 | sched_group_cpus(group)); | |
147cbb4b NP |
2256 | |
2257 | /* Tally up the load of all CPUs in the group */ | |
2258 | avg_load = 0; | |
2259 | ||
758b2cdc | 2260 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2261 | /* Bias balancing toward cpus of our domain */ |
2262 | if (local_group) | |
2263 | load = source_load(i, load_idx); | |
2264 | else | |
2265 | load = target_load(i, load_idx); | |
2266 | ||
2267 | avg_load += load; | |
2268 | } | |
2269 | ||
2270 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2271 | avg_load = sg_div_cpu_power(group, |
2272 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2273 | |
2274 | if (local_group) { | |
2275 | this_load = avg_load; | |
2276 | this = group; | |
2277 | } else if (avg_load < min_load) { | |
2278 | min_load = avg_load; | |
2279 | idlest = group; | |
2280 | } | |
3a5c359a | 2281 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2282 | |
2283 | if (!idlest || 100*this_load < imbalance*min_load) | |
2284 | return NULL; | |
2285 | return idlest; | |
2286 | } | |
2287 | ||
2288 | /* | |
0feaece9 | 2289 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2290 | */ |
95cdf3b7 | 2291 | static int |
758b2cdc | 2292 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2293 | { |
2294 | unsigned long load, min_load = ULONG_MAX; | |
2295 | int idlest = -1; | |
2296 | int i; | |
2297 | ||
da5a5522 | 2298 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2299 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2300 | load = weighted_cpuload(i); |
147cbb4b NP |
2301 | |
2302 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2303 | min_load = load; | |
2304 | idlest = i; | |
2305 | } | |
2306 | } | |
2307 | ||
2308 | return idlest; | |
2309 | } | |
2310 | ||
476d139c NP |
2311 | /* |
2312 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2313 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2314 | * SD_BALANCE_EXEC. | |
2315 | * | |
2316 | * Balance, ie. select the least loaded group. | |
2317 | * | |
2318 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2319 | * | |
2320 | * preempt must be disabled. | |
2321 | */ | |
2322 | static int sched_balance_self(int cpu, int flag) | |
2323 | { | |
2324 | struct task_struct *t = current; | |
2325 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2326 | |
c96d145e | 2327 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2328 | /* |
2329 | * If power savings logic is enabled for a domain, stop there. | |
2330 | */ | |
5c45bf27 SS |
2331 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2332 | break; | |
476d139c NP |
2333 | if (tmp->flags & flag) |
2334 | sd = tmp; | |
c96d145e | 2335 | } |
476d139c | 2336 | |
039a1c41 PZ |
2337 | if (sd) |
2338 | update_shares(sd); | |
2339 | ||
476d139c | 2340 | while (sd) { |
476d139c | 2341 | struct sched_group *group; |
1a848870 SS |
2342 | int new_cpu, weight; |
2343 | ||
2344 | if (!(sd->flags & flag)) { | |
2345 | sd = sd->child; | |
2346 | continue; | |
2347 | } | |
476d139c | 2348 | |
476d139c | 2349 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2350 | if (!group) { |
2351 | sd = sd->child; | |
2352 | continue; | |
2353 | } | |
476d139c | 2354 | |
758b2cdc | 2355 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2356 | if (new_cpu == -1 || new_cpu == cpu) { |
2357 | /* Now try balancing at a lower domain level of cpu */ | |
2358 | sd = sd->child; | |
2359 | continue; | |
2360 | } | |
476d139c | 2361 | |
1a848870 | 2362 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2363 | cpu = new_cpu; |
758b2cdc | 2364 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2365 | sd = NULL; |
476d139c | 2366 | for_each_domain(cpu, tmp) { |
758b2cdc | 2367 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2368 | break; |
2369 | if (tmp->flags & flag) | |
2370 | sd = tmp; | |
2371 | } | |
2372 | /* while loop will break here if sd == NULL */ | |
2373 | } | |
2374 | ||
2375 | return cpu; | |
2376 | } | |
2377 | ||
2378 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2379 | |
0793a61d TG |
2380 | /** |
2381 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2382 | * @p: the task to evaluate | |
2383 | * @func: the function to be called | |
2384 | * @info: the function call argument | |
2385 | * | |
2386 | * Calls the function @func when the task is currently running. This might | |
2387 | * be on the current CPU, which just calls the function directly | |
2388 | */ | |
2389 | void task_oncpu_function_call(struct task_struct *p, | |
2390 | void (*func) (void *info), void *info) | |
2391 | { | |
2392 | int cpu; | |
2393 | ||
2394 | preempt_disable(); | |
2395 | cpu = task_cpu(p); | |
2396 | if (task_curr(p)) | |
2397 | smp_call_function_single(cpu, func, info, 1); | |
2398 | preempt_enable(); | |
2399 | } | |
2400 | ||
1da177e4 LT |
2401 | /*** |
2402 | * try_to_wake_up - wake up a thread | |
2403 | * @p: the to-be-woken-up thread | |
2404 | * @state: the mask of task states that can be woken | |
2405 | * @sync: do a synchronous wakeup? | |
2406 | * | |
2407 | * Put it on the run-queue if it's not already there. The "current" | |
2408 | * thread is always on the run-queue (except when the actual | |
2409 | * re-schedule is in progress), and as such you're allowed to do | |
2410 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2411 | * runnable without the overhead of this. | |
2412 | * | |
2413 | * returns failure only if the task is already active. | |
2414 | */ | |
36c8b586 | 2415 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2416 | { |
cc367732 | 2417 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2418 | unsigned long flags; |
2419 | long old_state; | |
70b97a7f | 2420 | struct rq *rq; |
1da177e4 | 2421 | |
b85d0667 IM |
2422 | if (!sched_feat(SYNC_WAKEUPS)) |
2423 | sync = 0; | |
2424 | ||
2398f2c6 | 2425 | #ifdef CONFIG_SMP |
57310a98 | 2426 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2427 | struct sched_domain *sd; |
2428 | ||
2429 | this_cpu = raw_smp_processor_id(); | |
2430 | cpu = task_cpu(p); | |
2431 | ||
2432 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2433 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2434 | update_shares(sd); |
2435 | break; | |
2436 | } | |
2437 | } | |
2438 | } | |
2439 | #endif | |
2440 | ||
04e2f174 | 2441 | smp_wmb(); |
1da177e4 | 2442 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2443 | update_rq_clock(rq); |
1da177e4 LT |
2444 | old_state = p->state; |
2445 | if (!(old_state & state)) | |
2446 | goto out; | |
2447 | ||
dd41f596 | 2448 | if (p->se.on_rq) |
1da177e4 LT |
2449 | goto out_running; |
2450 | ||
2451 | cpu = task_cpu(p); | |
cc367732 | 2452 | orig_cpu = cpu; |
1da177e4 LT |
2453 | this_cpu = smp_processor_id(); |
2454 | ||
2455 | #ifdef CONFIG_SMP | |
2456 | if (unlikely(task_running(rq, p))) | |
2457 | goto out_activate; | |
2458 | ||
5d2f5a61 DA |
2459 | cpu = p->sched_class->select_task_rq(p, sync); |
2460 | if (cpu != orig_cpu) { | |
2461 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2462 | task_rq_unlock(rq, &flags); |
2463 | /* might preempt at this point */ | |
2464 | rq = task_rq_lock(p, &flags); | |
2465 | old_state = p->state; | |
2466 | if (!(old_state & state)) | |
2467 | goto out; | |
dd41f596 | 2468 | if (p->se.on_rq) |
1da177e4 LT |
2469 | goto out_running; |
2470 | ||
2471 | this_cpu = smp_processor_id(); | |
2472 | cpu = task_cpu(p); | |
2473 | } | |
2474 | ||
e7693a36 GH |
2475 | #ifdef CONFIG_SCHEDSTATS |
2476 | schedstat_inc(rq, ttwu_count); | |
2477 | if (cpu == this_cpu) | |
2478 | schedstat_inc(rq, ttwu_local); | |
2479 | else { | |
2480 | struct sched_domain *sd; | |
2481 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2482 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2483 | schedstat_inc(sd, ttwu_wake_remote); |
2484 | break; | |
2485 | } | |
2486 | } | |
2487 | } | |
6d6bc0ad | 2488 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2489 | |
1da177e4 LT |
2490 | out_activate: |
2491 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2492 | schedstat_inc(p, se.nr_wakeups); |
2493 | if (sync) | |
2494 | schedstat_inc(p, se.nr_wakeups_sync); | |
2495 | if (orig_cpu != cpu) | |
2496 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2497 | if (cpu == this_cpu) | |
2498 | schedstat_inc(p, se.nr_wakeups_local); | |
2499 | else | |
2500 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2501 | activate_task(rq, p, 1); |
1da177e4 LT |
2502 | success = 1; |
2503 | ||
831451ac PZ |
2504 | /* |
2505 | * Only attribute actual wakeups done by this task. | |
2506 | */ | |
2507 | if (!in_interrupt()) { | |
2508 | struct sched_entity *se = ¤t->se; | |
2509 | u64 sample = se->sum_exec_runtime; | |
2510 | ||
2511 | if (se->last_wakeup) | |
2512 | sample -= se->last_wakeup; | |
2513 | else | |
2514 | sample -= se->start_runtime; | |
2515 | update_avg(&se->avg_wakeup, sample); | |
2516 | ||
2517 | se->last_wakeup = se->sum_exec_runtime; | |
2518 | } | |
2519 | ||
1da177e4 | 2520 | out_running: |
468a15bb | 2521 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2522 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2523 | |
1da177e4 | 2524 | p->state = TASK_RUNNING; |
9a897c5a SR |
2525 | #ifdef CONFIG_SMP |
2526 | if (p->sched_class->task_wake_up) | |
2527 | p->sched_class->task_wake_up(rq, p); | |
2528 | #endif | |
1da177e4 LT |
2529 | out: |
2530 | task_rq_unlock(rq, &flags); | |
2531 | ||
2532 | return success; | |
2533 | } | |
2534 | ||
50fa610a DH |
2535 | /** |
2536 | * wake_up_process - Wake up a specific process | |
2537 | * @p: The process to be woken up. | |
2538 | * | |
2539 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2540 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2541 | * running. | |
2542 | * | |
2543 | * It may be assumed that this function implies a write memory barrier before | |
2544 | * changing the task state if and only if any tasks are woken up. | |
2545 | */ | |
7ad5b3a5 | 2546 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2547 | { |
d9514f6c | 2548 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2549 | } |
1da177e4 LT |
2550 | EXPORT_SYMBOL(wake_up_process); |
2551 | ||
7ad5b3a5 | 2552 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2553 | { |
2554 | return try_to_wake_up(p, state, 0); | |
2555 | } | |
2556 | ||
1da177e4 LT |
2557 | /* |
2558 | * Perform scheduler related setup for a newly forked process p. | |
2559 | * p is forked by current. | |
dd41f596 IM |
2560 | * |
2561 | * __sched_fork() is basic setup used by init_idle() too: | |
2562 | */ | |
2563 | static void __sched_fork(struct task_struct *p) | |
2564 | { | |
dd41f596 IM |
2565 | p->se.exec_start = 0; |
2566 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2567 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2568 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2569 | p->se.last_wakeup = 0; |
2570 | p->se.avg_overlap = 0; | |
831451ac PZ |
2571 | p->se.start_runtime = 0; |
2572 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2573 | |
2574 | #ifdef CONFIG_SCHEDSTATS | |
2575 | p->se.wait_start = 0; | |
dd41f596 IM |
2576 | p->se.sum_sleep_runtime = 0; |
2577 | p->se.sleep_start = 0; | |
dd41f596 IM |
2578 | p->se.block_start = 0; |
2579 | p->se.sleep_max = 0; | |
2580 | p->se.block_max = 0; | |
2581 | p->se.exec_max = 0; | |
eba1ed4b | 2582 | p->se.slice_max = 0; |
dd41f596 | 2583 | p->se.wait_max = 0; |
6cfb0d5d | 2584 | #endif |
476d139c | 2585 | |
fa717060 | 2586 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2587 | p->se.on_rq = 0; |
4a55bd5e | 2588 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2589 | |
e107be36 AK |
2590 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2591 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2592 | #endif | |
2593 | ||
1da177e4 LT |
2594 | /* |
2595 | * We mark the process as running here, but have not actually | |
2596 | * inserted it onto the runqueue yet. This guarantees that | |
2597 | * nobody will actually run it, and a signal or other external | |
2598 | * event cannot wake it up and insert it on the runqueue either. | |
2599 | */ | |
2600 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2601 | } |
2602 | ||
2603 | /* | |
2604 | * fork()/clone()-time setup: | |
2605 | */ | |
2606 | void sched_fork(struct task_struct *p, int clone_flags) | |
2607 | { | |
2608 | int cpu = get_cpu(); | |
2609 | ||
2610 | __sched_fork(p); | |
2611 | ||
2612 | #ifdef CONFIG_SMP | |
2613 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2614 | #endif | |
02e4bac2 | 2615 | set_task_cpu(p, cpu); |
b29739f9 IM |
2616 | |
2617 | /* | |
2618 | * Make sure we do not leak PI boosting priority to the child: | |
2619 | */ | |
2620 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2621 | if (!rt_prio(p->prio)) |
2622 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2623 | |
52f17b6c | 2624 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2625 | if (likely(sched_info_on())) |
52f17b6c | 2626 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2627 | #endif |
d6077cb8 | 2628 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2629 | p->oncpu = 0; |
2630 | #endif | |
1da177e4 | 2631 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2632 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2633 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2634 | #endif |
917b627d GH |
2635 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2636 | ||
476d139c | 2637 | put_cpu(); |
1da177e4 LT |
2638 | } |
2639 | ||
2640 | /* | |
2641 | * wake_up_new_task - wake up a newly created task for the first time. | |
2642 | * | |
2643 | * This function will do some initial scheduler statistics housekeeping | |
2644 | * that must be done for every newly created context, then puts the task | |
2645 | * on the runqueue and wakes it. | |
2646 | */ | |
7ad5b3a5 | 2647 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2648 | { |
2649 | unsigned long flags; | |
dd41f596 | 2650 | struct rq *rq; |
1da177e4 LT |
2651 | |
2652 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2653 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2654 | update_rq_clock(rq); |
1da177e4 LT |
2655 | |
2656 | p->prio = effective_prio(p); | |
2657 | ||
b9dca1e0 | 2658 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2659 | activate_task(rq, p, 0); |
1da177e4 | 2660 | } else { |
1da177e4 | 2661 | /* |
dd41f596 IM |
2662 | * Let the scheduling class do new task startup |
2663 | * management (if any): | |
1da177e4 | 2664 | */ |
ee0827d8 | 2665 | p->sched_class->task_new(rq, p); |
c09595f6 | 2666 | inc_nr_running(rq); |
1da177e4 | 2667 | } |
c71dd42d | 2668 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2669 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2670 | #ifdef CONFIG_SMP |
2671 | if (p->sched_class->task_wake_up) | |
2672 | p->sched_class->task_wake_up(rq, p); | |
2673 | #endif | |
dd41f596 | 2674 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2675 | } |
2676 | ||
e107be36 AK |
2677 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2678 | ||
2679 | /** | |
80dd99b3 | 2680 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2681 | * @notifier: notifier struct to register |
e107be36 AK |
2682 | */ |
2683 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2684 | { | |
2685 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2686 | } | |
2687 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2688 | ||
2689 | /** | |
2690 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2691 | * @notifier: notifier struct to unregister |
e107be36 AK |
2692 | * |
2693 | * This is safe to call from within a preemption notifier. | |
2694 | */ | |
2695 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2696 | { | |
2697 | hlist_del(¬ifier->link); | |
2698 | } | |
2699 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2700 | ||
2701 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2702 | { | |
2703 | struct preempt_notifier *notifier; | |
2704 | struct hlist_node *node; | |
2705 | ||
2706 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2707 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2708 | } | |
2709 | ||
2710 | static void | |
2711 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2712 | struct task_struct *next) | |
2713 | { | |
2714 | struct preempt_notifier *notifier; | |
2715 | struct hlist_node *node; | |
2716 | ||
2717 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2718 | notifier->ops->sched_out(notifier, next); | |
2719 | } | |
2720 | ||
6d6bc0ad | 2721 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2722 | |
2723 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2724 | { | |
2725 | } | |
2726 | ||
2727 | static void | |
2728 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2729 | struct task_struct *next) | |
2730 | { | |
2731 | } | |
2732 | ||
6d6bc0ad | 2733 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2734 | |
4866cde0 NP |
2735 | /** |
2736 | * prepare_task_switch - prepare to switch tasks | |
2737 | * @rq: the runqueue preparing to switch | |
421cee29 | 2738 | * @prev: the current task that is being switched out |
4866cde0 NP |
2739 | * @next: the task we are going to switch to. |
2740 | * | |
2741 | * This is called with the rq lock held and interrupts off. It must | |
2742 | * be paired with a subsequent finish_task_switch after the context | |
2743 | * switch. | |
2744 | * | |
2745 | * prepare_task_switch sets up locking and calls architecture specific | |
2746 | * hooks. | |
2747 | */ | |
e107be36 AK |
2748 | static inline void |
2749 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2750 | struct task_struct *next) | |
4866cde0 | 2751 | { |
e107be36 | 2752 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2753 | prepare_lock_switch(rq, next); |
2754 | prepare_arch_switch(next); | |
2755 | } | |
2756 | ||
1da177e4 LT |
2757 | /** |
2758 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2759 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2760 | * @prev: the thread we just switched away from. |
2761 | * | |
4866cde0 NP |
2762 | * finish_task_switch must be called after the context switch, paired |
2763 | * with a prepare_task_switch call before the context switch. | |
2764 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2765 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2766 | * |
2767 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2768 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2769 | * with the lock held can cause deadlocks; see schedule() for |
2770 | * details.) | |
2771 | */ | |
a9957449 | 2772 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2773 | __releases(rq->lock) |
2774 | { | |
1da177e4 | 2775 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2776 | long prev_state; |
967fc046 GH |
2777 | #ifdef CONFIG_SMP |
2778 | int post_schedule = 0; | |
2779 | ||
2780 | if (current->sched_class->needs_post_schedule) | |
2781 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2782 | #endif | |
1da177e4 LT |
2783 | |
2784 | rq->prev_mm = NULL; | |
2785 | ||
2786 | /* | |
2787 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2788 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2789 | * schedule one last time. The schedule call will never return, and |
2790 | * the scheduled task must drop that reference. | |
c394cc9f | 2791 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2792 | * still held, otherwise prev could be scheduled on another cpu, die |
2793 | * there before we look at prev->state, and then the reference would | |
2794 | * be dropped twice. | |
2795 | * Manfred Spraul <manfred@colorfullife.com> | |
2796 | */ | |
55a101f8 | 2797 | prev_state = prev->state; |
4866cde0 | 2798 | finish_arch_switch(prev); |
0793a61d | 2799 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2800 | finish_lock_switch(rq, prev); |
9a897c5a | 2801 | #ifdef CONFIG_SMP |
967fc046 | 2802 | if (post_schedule) |
9a897c5a SR |
2803 | current->sched_class->post_schedule(rq); |
2804 | #endif | |
e8fa1362 | 2805 | |
e107be36 | 2806 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2807 | if (mm) |
2808 | mmdrop(mm); | |
c394cc9f | 2809 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2810 | /* |
2811 | * Remove function-return probe instances associated with this | |
2812 | * task and put them back on the free list. | |
9761eea8 | 2813 | */ |
c6fd91f0 | 2814 | kprobe_flush_task(prev); |
1da177e4 | 2815 | put_task_struct(prev); |
c6fd91f0 | 2816 | } |
1da177e4 LT |
2817 | } |
2818 | ||
2819 | /** | |
2820 | * schedule_tail - first thing a freshly forked thread must call. | |
2821 | * @prev: the thread we just switched away from. | |
2822 | */ | |
36c8b586 | 2823 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2824 | __releases(rq->lock) |
2825 | { | |
70b97a7f IM |
2826 | struct rq *rq = this_rq(); |
2827 | ||
4866cde0 NP |
2828 | finish_task_switch(rq, prev); |
2829 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2830 | /* In this case, finish_task_switch does not reenable preemption */ | |
2831 | preempt_enable(); | |
2832 | #endif | |
1da177e4 | 2833 | if (current->set_child_tid) |
b488893a | 2834 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2835 | } |
2836 | ||
2837 | /* | |
2838 | * context_switch - switch to the new MM and the new | |
2839 | * thread's register state. | |
2840 | */ | |
dd41f596 | 2841 | static inline void |
70b97a7f | 2842 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2843 | struct task_struct *next) |
1da177e4 | 2844 | { |
dd41f596 | 2845 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2846 | |
e107be36 | 2847 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2848 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2849 | mm = next->mm; |
2850 | oldmm = prev->active_mm; | |
9226d125 ZA |
2851 | /* |
2852 | * For paravirt, this is coupled with an exit in switch_to to | |
2853 | * combine the page table reload and the switch backend into | |
2854 | * one hypercall. | |
2855 | */ | |
224101ed | 2856 | arch_start_context_switch(prev); |
9226d125 | 2857 | |
dd41f596 | 2858 | if (unlikely(!mm)) { |
1da177e4 LT |
2859 | next->active_mm = oldmm; |
2860 | atomic_inc(&oldmm->mm_count); | |
2861 | enter_lazy_tlb(oldmm, next); | |
2862 | } else | |
2863 | switch_mm(oldmm, mm, next); | |
2864 | ||
dd41f596 | 2865 | if (unlikely(!prev->mm)) { |
1da177e4 | 2866 | prev->active_mm = NULL; |
1da177e4 LT |
2867 | rq->prev_mm = oldmm; |
2868 | } | |
3a5f5e48 IM |
2869 | /* |
2870 | * Since the runqueue lock will be released by the next | |
2871 | * task (which is an invalid locking op but in the case | |
2872 | * of the scheduler it's an obvious special-case), so we | |
2873 | * do an early lockdep release here: | |
2874 | */ | |
2875 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2876 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2877 | #endif |
1da177e4 LT |
2878 | |
2879 | /* Here we just switch the register state and the stack. */ | |
2880 | switch_to(prev, next, prev); | |
2881 | ||
dd41f596 IM |
2882 | barrier(); |
2883 | /* | |
2884 | * this_rq must be evaluated again because prev may have moved | |
2885 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2886 | * frame will be invalid. | |
2887 | */ | |
2888 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2889 | } |
2890 | ||
2891 | /* | |
2892 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2893 | * | |
2894 | * externally visible scheduler statistics: current number of runnable | |
2895 | * threads, current number of uninterruptible-sleeping threads, total | |
2896 | * number of context switches performed since bootup. | |
2897 | */ | |
2898 | unsigned long nr_running(void) | |
2899 | { | |
2900 | unsigned long i, sum = 0; | |
2901 | ||
2902 | for_each_online_cpu(i) | |
2903 | sum += cpu_rq(i)->nr_running; | |
2904 | ||
2905 | return sum; | |
2906 | } | |
2907 | ||
2908 | unsigned long nr_uninterruptible(void) | |
2909 | { | |
2910 | unsigned long i, sum = 0; | |
2911 | ||
0a945022 | 2912 | for_each_possible_cpu(i) |
1da177e4 LT |
2913 | sum += cpu_rq(i)->nr_uninterruptible; |
2914 | ||
2915 | /* | |
2916 | * Since we read the counters lockless, it might be slightly | |
2917 | * inaccurate. Do not allow it to go below zero though: | |
2918 | */ | |
2919 | if (unlikely((long)sum < 0)) | |
2920 | sum = 0; | |
2921 | ||
2922 | return sum; | |
2923 | } | |
2924 | ||
2925 | unsigned long long nr_context_switches(void) | |
2926 | { | |
cc94abfc SR |
2927 | int i; |
2928 | unsigned long long sum = 0; | |
1da177e4 | 2929 | |
0a945022 | 2930 | for_each_possible_cpu(i) |
1da177e4 LT |
2931 | sum += cpu_rq(i)->nr_switches; |
2932 | ||
2933 | return sum; | |
2934 | } | |
2935 | ||
2936 | unsigned long nr_iowait(void) | |
2937 | { | |
2938 | unsigned long i, sum = 0; | |
2939 | ||
0a945022 | 2940 | for_each_possible_cpu(i) |
1da177e4 LT |
2941 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2942 | ||
2943 | return sum; | |
2944 | } | |
2945 | ||
dce48a84 TG |
2946 | /* Variables and functions for calc_load */ |
2947 | static atomic_long_t calc_load_tasks; | |
2948 | static unsigned long calc_load_update; | |
2949 | unsigned long avenrun[3]; | |
2950 | EXPORT_SYMBOL(avenrun); | |
2951 | ||
2d02494f TG |
2952 | /** |
2953 | * get_avenrun - get the load average array | |
2954 | * @loads: pointer to dest load array | |
2955 | * @offset: offset to add | |
2956 | * @shift: shift count to shift the result left | |
2957 | * | |
2958 | * These values are estimates at best, so no need for locking. | |
2959 | */ | |
2960 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2961 | { | |
2962 | loads[0] = (avenrun[0] + offset) << shift; | |
2963 | loads[1] = (avenrun[1] + offset) << shift; | |
2964 | loads[2] = (avenrun[2] + offset) << shift; | |
2965 | } | |
2966 | ||
dce48a84 TG |
2967 | static unsigned long |
2968 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2969 | { |
dce48a84 TG |
2970 | load *= exp; |
2971 | load += active * (FIXED_1 - exp); | |
2972 | return load >> FSHIFT; | |
2973 | } | |
db1b1fef | 2974 | |
dce48a84 TG |
2975 | /* |
2976 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2977 | * CPUs have updated calc_load_tasks. | |
2978 | */ | |
2979 | void calc_global_load(void) | |
2980 | { | |
2981 | unsigned long upd = calc_load_update + 10; | |
2982 | long active; | |
2983 | ||
2984 | if (time_before(jiffies, upd)) | |
2985 | return; | |
db1b1fef | 2986 | |
dce48a84 TG |
2987 | active = atomic_long_read(&calc_load_tasks); |
2988 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2989 | |
dce48a84 TG |
2990 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2991 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2992 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2993 | ||
2994 | calc_load_update += LOAD_FREQ; | |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * Either called from update_cpu_load() or from a cpu going idle | |
2999 | */ | |
3000 | static void calc_load_account_active(struct rq *this_rq) | |
3001 | { | |
3002 | long nr_active, delta; | |
3003 | ||
3004 | nr_active = this_rq->nr_running; | |
3005 | nr_active += (long) this_rq->nr_uninterruptible; | |
3006 | ||
3007 | if (nr_active != this_rq->calc_load_active) { | |
3008 | delta = nr_active - this_rq->calc_load_active; | |
3009 | this_rq->calc_load_active = nr_active; | |
3010 | atomic_long_add(delta, &calc_load_tasks); | |
3011 | } | |
db1b1fef JS |
3012 | } |
3013 | ||
23a185ca PM |
3014 | /* |
3015 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3016 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3017 | */ | |
23a185ca PM |
3018 | u64 cpu_nr_migrations(int cpu) |
3019 | { | |
3020 | return cpu_rq(cpu)->nr_migrations_in; | |
3021 | } | |
3022 | ||
48f24c4d | 3023 | /* |
dd41f596 IM |
3024 | * Update rq->cpu_load[] statistics. This function is usually called every |
3025 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3026 | */ |
dd41f596 | 3027 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3028 | { |
495eca49 | 3029 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3030 | int i, scale; |
3031 | ||
3032 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3033 | |
3034 | /* Update our load: */ | |
3035 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3036 | unsigned long old_load, new_load; | |
3037 | ||
3038 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3039 | ||
3040 | old_load = this_rq->cpu_load[i]; | |
3041 | new_load = this_load; | |
a25707f3 IM |
3042 | /* |
3043 | * Round up the averaging division if load is increasing. This | |
3044 | * prevents us from getting stuck on 9 if the load is 10, for | |
3045 | * example. | |
3046 | */ | |
3047 | if (new_load > old_load) | |
3048 | new_load += scale-1; | |
dd41f596 IM |
3049 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3050 | } | |
dce48a84 TG |
3051 | |
3052 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3053 | this_rq->calc_load_update += LOAD_FREQ; | |
3054 | calc_load_account_active(this_rq); | |
3055 | } | |
48f24c4d IM |
3056 | } |
3057 | ||
dd41f596 IM |
3058 | #ifdef CONFIG_SMP |
3059 | ||
1da177e4 LT |
3060 | /* |
3061 | * double_rq_lock - safely lock two runqueues | |
3062 | * | |
3063 | * Note this does not disable interrupts like task_rq_lock, | |
3064 | * you need to do so manually before calling. | |
3065 | */ | |
70b97a7f | 3066 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3067 | __acquires(rq1->lock) |
3068 | __acquires(rq2->lock) | |
3069 | { | |
054b9108 | 3070 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3071 | if (rq1 == rq2) { |
3072 | spin_lock(&rq1->lock); | |
3073 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3074 | } else { | |
c96d145e | 3075 | if (rq1 < rq2) { |
1da177e4 | 3076 | spin_lock(&rq1->lock); |
5e710e37 | 3077 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3078 | } else { |
3079 | spin_lock(&rq2->lock); | |
5e710e37 | 3080 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3081 | } |
3082 | } | |
6e82a3be IM |
3083 | update_rq_clock(rq1); |
3084 | update_rq_clock(rq2); | |
1da177e4 LT |
3085 | } |
3086 | ||
3087 | /* | |
3088 | * double_rq_unlock - safely unlock two runqueues | |
3089 | * | |
3090 | * Note this does not restore interrupts like task_rq_unlock, | |
3091 | * you need to do so manually after calling. | |
3092 | */ | |
70b97a7f | 3093 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3094 | __releases(rq1->lock) |
3095 | __releases(rq2->lock) | |
3096 | { | |
3097 | spin_unlock(&rq1->lock); | |
3098 | if (rq1 != rq2) | |
3099 | spin_unlock(&rq2->lock); | |
3100 | else | |
3101 | __release(rq2->lock); | |
3102 | } | |
3103 | ||
1da177e4 LT |
3104 | /* |
3105 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3106 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3107 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3108 | * the cpu_allowed mask is restored. |
3109 | */ | |
36c8b586 | 3110 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3111 | { |
70b97a7f | 3112 | struct migration_req req; |
1da177e4 | 3113 | unsigned long flags; |
70b97a7f | 3114 | struct rq *rq; |
1da177e4 LT |
3115 | |
3116 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3117 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3118 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3119 | goto out; |
3120 | ||
3121 | /* force the process onto the specified CPU */ | |
3122 | if (migrate_task(p, dest_cpu, &req)) { | |
3123 | /* Need to wait for migration thread (might exit: take ref). */ | |
3124 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3125 | |
1da177e4 LT |
3126 | get_task_struct(mt); |
3127 | task_rq_unlock(rq, &flags); | |
3128 | wake_up_process(mt); | |
3129 | put_task_struct(mt); | |
3130 | wait_for_completion(&req.done); | |
36c8b586 | 3131 | |
1da177e4 LT |
3132 | return; |
3133 | } | |
3134 | out: | |
3135 | task_rq_unlock(rq, &flags); | |
3136 | } | |
3137 | ||
3138 | /* | |
476d139c NP |
3139 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3140 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3141 | */ |
3142 | void sched_exec(void) | |
3143 | { | |
1da177e4 | 3144 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3145 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3146 | put_cpu(); |
476d139c NP |
3147 | if (new_cpu != this_cpu) |
3148 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3149 | } |
3150 | ||
3151 | /* | |
3152 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3153 | * Both runqueues must be locked. | |
3154 | */ | |
dd41f596 IM |
3155 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3156 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3157 | { |
2e1cb74a | 3158 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3159 | set_task_cpu(p, this_cpu); |
dd41f596 | 3160 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3161 | /* |
3162 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3163 | * to be always true for them. | |
3164 | */ | |
15afe09b | 3165 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3166 | } |
3167 | ||
3168 | /* | |
3169 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3170 | */ | |
858119e1 | 3171 | static |
70b97a7f | 3172 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3173 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3174 | int *all_pinned) |
1da177e4 | 3175 | { |
708dc512 | 3176 | int tsk_cache_hot = 0; |
1da177e4 LT |
3177 | /* |
3178 | * We do not migrate tasks that are: | |
3179 | * 1) running (obviously), or | |
3180 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3181 | * 3) are cache-hot on their current CPU. | |
3182 | */ | |
96f874e2 | 3183 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3184 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3185 | return 0; |
cc367732 | 3186 | } |
81026794 NP |
3187 | *all_pinned = 0; |
3188 | ||
cc367732 IM |
3189 | if (task_running(rq, p)) { |
3190 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3191 | return 0; |
cc367732 | 3192 | } |
1da177e4 | 3193 | |
da84d961 IM |
3194 | /* |
3195 | * Aggressive migration if: | |
3196 | * 1) task is cache cold, or | |
3197 | * 2) too many balance attempts have failed. | |
3198 | */ | |
3199 | ||
708dc512 LH |
3200 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3201 | if (!tsk_cache_hot || | |
3202 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3203 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3204 | if (tsk_cache_hot) { |
da84d961 | 3205 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3206 | schedstat_inc(p, se.nr_forced_migrations); |
3207 | } | |
da84d961 IM |
3208 | #endif |
3209 | return 1; | |
3210 | } | |
3211 | ||
708dc512 | 3212 | if (tsk_cache_hot) { |
cc367732 | 3213 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3214 | return 0; |
cc367732 | 3215 | } |
1da177e4 LT |
3216 | return 1; |
3217 | } | |
3218 | ||
e1d1484f PW |
3219 | static unsigned long |
3220 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3221 | unsigned long max_load_move, struct sched_domain *sd, | |
3222 | enum cpu_idle_type idle, int *all_pinned, | |
3223 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3224 | { |
051c6764 | 3225 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3226 | struct task_struct *p; |
3227 | long rem_load_move = max_load_move; | |
1da177e4 | 3228 | |
e1d1484f | 3229 | if (max_load_move == 0) |
1da177e4 LT |
3230 | goto out; |
3231 | ||
81026794 NP |
3232 | pinned = 1; |
3233 | ||
1da177e4 | 3234 | /* |
dd41f596 | 3235 | * Start the load-balancing iterator: |
1da177e4 | 3236 | */ |
dd41f596 IM |
3237 | p = iterator->start(iterator->arg); |
3238 | next: | |
b82d9fdd | 3239 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3240 | goto out; |
051c6764 PZ |
3241 | |
3242 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3243 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3244 | p = iterator->next(iterator->arg); |
3245 | goto next; | |
1da177e4 LT |
3246 | } |
3247 | ||
dd41f596 | 3248 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3249 | pulled++; |
dd41f596 | 3250 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3251 | |
7e96fa58 GH |
3252 | #ifdef CONFIG_PREEMPT |
3253 | /* | |
3254 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3255 | * will stop after the first task is pulled to minimize the critical | |
3256 | * section. | |
3257 | */ | |
3258 | if (idle == CPU_NEWLY_IDLE) | |
3259 | goto out; | |
3260 | #endif | |
3261 | ||
2dd73a4f | 3262 | /* |
b82d9fdd | 3263 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3264 | */ |
e1d1484f | 3265 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3266 | if (p->prio < *this_best_prio) |
3267 | *this_best_prio = p->prio; | |
dd41f596 IM |
3268 | p = iterator->next(iterator->arg); |
3269 | goto next; | |
1da177e4 LT |
3270 | } |
3271 | out: | |
3272 | /* | |
e1d1484f | 3273 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3274 | * so we can safely collect pull_task() stats here rather than |
3275 | * inside pull_task(). | |
3276 | */ | |
3277 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3278 | |
3279 | if (all_pinned) | |
3280 | *all_pinned = pinned; | |
e1d1484f PW |
3281 | |
3282 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3283 | } |
3284 | ||
dd41f596 | 3285 | /* |
43010659 PW |
3286 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3287 | * this_rq, as part of a balancing operation within domain "sd". | |
3288 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3289 | * |
3290 | * Called with both runqueues locked. | |
3291 | */ | |
3292 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3293 | unsigned long max_load_move, |
dd41f596 IM |
3294 | struct sched_domain *sd, enum cpu_idle_type idle, |
3295 | int *all_pinned) | |
3296 | { | |
5522d5d5 | 3297 | const struct sched_class *class = sched_class_highest; |
43010659 | 3298 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3299 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3300 | |
3301 | do { | |
43010659 PW |
3302 | total_load_moved += |
3303 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3304 | max_load_move - total_load_moved, |
a4ac01c3 | 3305 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3306 | class = class->next; |
c4acb2c0 | 3307 | |
7e96fa58 GH |
3308 | #ifdef CONFIG_PREEMPT |
3309 | /* | |
3310 | * NEWIDLE balancing is a source of latency, so preemptible | |
3311 | * kernels will stop after the first task is pulled to minimize | |
3312 | * the critical section. | |
3313 | */ | |
c4acb2c0 GH |
3314 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3315 | break; | |
7e96fa58 | 3316 | #endif |
43010659 | 3317 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3318 | |
43010659 PW |
3319 | return total_load_moved > 0; |
3320 | } | |
3321 | ||
e1d1484f PW |
3322 | static int |
3323 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3324 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3325 | struct rq_iterator *iterator) | |
3326 | { | |
3327 | struct task_struct *p = iterator->start(iterator->arg); | |
3328 | int pinned = 0; | |
3329 | ||
3330 | while (p) { | |
3331 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3332 | pull_task(busiest, p, this_rq, this_cpu); | |
3333 | /* | |
3334 | * Right now, this is only the second place pull_task() | |
3335 | * is called, so we can safely collect pull_task() | |
3336 | * stats here rather than inside pull_task(). | |
3337 | */ | |
3338 | schedstat_inc(sd, lb_gained[idle]); | |
3339 | ||
3340 | return 1; | |
3341 | } | |
3342 | p = iterator->next(iterator->arg); | |
3343 | } | |
3344 | ||
3345 | return 0; | |
3346 | } | |
3347 | ||
43010659 PW |
3348 | /* |
3349 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3350 | * part of active balancing operations within "domain". | |
3351 | * Returns 1 if successful and 0 otherwise. | |
3352 | * | |
3353 | * Called with both runqueues locked. | |
3354 | */ | |
3355 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3356 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3357 | { | |
5522d5d5 | 3358 | const struct sched_class *class; |
43010659 PW |
3359 | |
3360 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3361 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3362 | return 1; |
3363 | ||
3364 | return 0; | |
dd41f596 | 3365 | } |
67bb6c03 | 3366 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3367 | /* |
222d656d GS |
3368 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3369 | * during load balancing. | |
1da177e4 | 3370 | */ |
222d656d GS |
3371 | struct sd_lb_stats { |
3372 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3373 | struct sched_group *this; /* Local group in this sd */ | |
3374 | unsigned long total_load; /* Total load of all groups in sd */ | |
3375 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3376 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3377 | ||
3378 | /** Statistics of this group */ | |
3379 | unsigned long this_load; | |
3380 | unsigned long this_load_per_task; | |
3381 | unsigned long this_nr_running; | |
3382 | ||
3383 | /* Statistics of the busiest group */ | |
3384 | unsigned long max_load; | |
3385 | unsigned long busiest_load_per_task; | |
3386 | unsigned long busiest_nr_running; | |
3387 | ||
3388 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3389 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3390 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3391 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3392 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3393 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3394 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3395 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3396 | #endif |
222d656d | 3397 | }; |
1da177e4 | 3398 | |
d5ac537e | 3399 | /* |
381be78f GS |
3400 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3401 | */ | |
3402 | struct sg_lb_stats { | |
3403 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3404 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3405 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3406 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3407 | unsigned long group_capacity; | |
3408 | int group_imb; /* Is there an imbalance in the group ? */ | |
3409 | }; | |
408ed066 | 3410 | |
67bb6c03 GS |
3411 | /** |
3412 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3413 | * @group: The group whose first cpu is to be returned. | |
3414 | */ | |
3415 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3416 | { | |
3417 | return cpumask_first(sched_group_cpus(group)); | |
3418 | } | |
3419 | ||
3420 | /** | |
3421 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3422 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3423 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3424 | */ | |
3425 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3426 | enum cpu_idle_type idle) | |
3427 | { | |
3428 | int load_idx; | |
3429 | ||
3430 | switch (idle) { | |
3431 | case CPU_NOT_IDLE: | |
7897986b | 3432 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3433 | break; |
3434 | ||
3435 | case CPU_NEWLY_IDLE: | |
7897986b | 3436 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3437 | break; |
3438 | default: | |
7897986b | 3439 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3440 | break; |
3441 | } | |
1da177e4 | 3442 | |
67bb6c03 GS |
3443 | return load_idx; |
3444 | } | |
1da177e4 | 3445 | |
1da177e4 | 3446 | |
c071df18 GS |
3447 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3448 | /** | |
3449 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3450 | * the given sched_domain, during load balancing. | |
3451 | * | |
3452 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3453 | * @sds: Variable containing the statistics for sd. | |
3454 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3455 | */ | |
3456 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3457 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3458 | { | |
3459 | /* | |
3460 | * Busy processors will not participate in power savings | |
3461 | * balance. | |
3462 | */ | |
3463 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3464 | sds->power_savings_balance = 0; | |
3465 | else { | |
3466 | sds->power_savings_balance = 1; | |
3467 | sds->min_nr_running = ULONG_MAX; | |
3468 | sds->leader_nr_running = 0; | |
3469 | } | |
3470 | } | |
783609c6 | 3471 | |
c071df18 GS |
3472 | /** |
3473 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3474 | * sched_domain while performing load balancing. | |
3475 | * | |
3476 | * @group: sched_group belonging to the sched_domain under consideration. | |
3477 | * @sds: Variable containing the statistics of the sched_domain | |
3478 | * @local_group: Does group contain the CPU for which we're performing | |
3479 | * load balancing ? | |
3480 | * @sgs: Variable containing the statistics of the group. | |
3481 | */ | |
3482 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3483 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3484 | { | |
408ed066 | 3485 | |
c071df18 GS |
3486 | if (!sds->power_savings_balance) |
3487 | return; | |
1da177e4 | 3488 | |
c071df18 GS |
3489 | /* |
3490 | * If the local group is idle or completely loaded | |
3491 | * no need to do power savings balance at this domain | |
3492 | */ | |
3493 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3494 | !sds->this_nr_running)) | |
3495 | sds->power_savings_balance = 0; | |
2dd73a4f | 3496 | |
c071df18 GS |
3497 | /* |
3498 | * If a group is already running at full capacity or idle, | |
3499 | * don't include that group in power savings calculations | |
3500 | */ | |
3501 | if (!sds->power_savings_balance || | |
3502 | sgs->sum_nr_running >= sgs->group_capacity || | |
3503 | !sgs->sum_nr_running) | |
3504 | return; | |
5969fe06 | 3505 | |
c071df18 GS |
3506 | /* |
3507 | * Calculate the group which has the least non-idle load. | |
3508 | * This is the group from where we need to pick up the load | |
3509 | * for saving power | |
3510 | */ | |
3511 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3512 | (sgs->sum_nr_running == sds->min_nr_running && | |
3513 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3514 | sds->group_min = group; | |
3515 | sds->min_nr_running = sgs->sum_nr_running; | |
3516 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3517 | sgs->sum_nr_running; | |
3518 | } | |
783609c6 | 3519 | |
c071df18 GS |
3520 | /* |
3521 | * Calculate the group which is almost near its | |
3522 | * capacity but still has some space to pick up some load | |
3523 | * from other group and save more power | |
3524 | */ | |
3525 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3526 | return; | |
1da177e4 | 3527 | |
c071df18 GS |
3528 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3529 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3530 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3531 | sds->group_leader = group; | |
3532 | sds->leader_nr_running = sgs->sum_nr_running; | |
3533 | } | |
3534 | } | |
408ed066 | 3535 | |
c071df18 | 3536 | /** |
d5ac537e | 3537 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3538 | * @sds: Variable containing the statistics of the sched_domain |
3539 | * under consideration. | |
3540 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3541 | * @imbalance: Variable to store the imbalance. | |
3542 | * | |
d5ac537e RD |
3543 | * Description: |
3544 | * Check if we have potential to perform some power-savings balance. | |
3545 | * If yes, set the busiest group to be the least loaded group in the | |
3546 | * sched_domain, so that it's CPUs can be put to idle. | |
3547 | * | |
c071df18 GS |
3548 | * Returns 1 if there is potential to perform power-savings balance. |
3549 | * Else returns 0. | |
3550 | */ | |
3551 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3552 | int this_cpu, unsigned long *imbalance) | |
3553 | { | |
3554 | if (!sds->power_savings_balance) | |
3555 | return 0; | |
1da177e4 | 3556 | |
c071df18 GS |
3557 | if (sds->this != sds->group_leader || |
3558 | sds->group_leader == sds->group_min) | |
3559 | return 0; | |
783609c6 | 3560 | |
c071df18 GS |
3561 | *imbalance = sds->min_load_per_task; |
3562 | sds->busiest = sds->group_min; | |
1da177e4 | 3563 | |
c071df18 GS |
3564 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3565 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3566 | group_first_cpu(sds->group_leader); | |
3567 | } | |
3568 | ||
3569 | return 1; | |
1da177e4 | 3570 | |
c071df18 GS |
3571 | } |
3572 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3573 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3574 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3575 | { | |
3576 | return; | |
3577 | } | |
408ed066 | 3578 | |
c071df18 GS |
3579 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3580 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3581 | { | |
3582 | return; | |
3583 | } | |
3584 | ||
3585 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3586 | int this_cpu, unsigned long *imbalance) | |
3587 | { | |
3588 | return 0; | |
3589 | } | |
3590 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3591 | ||
3592 | ||
1f8c553d GS |
3593 | /** |
3594 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3595 | * @group: sched_group whose statistics are to be updated. | |
3596 | * @this_cpu: Cpu for which load balance is currently performed. | |
3597 | * @idle: Idle status of this_cpu | |
3598 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3599 | * @sd_idle: Idle status of the sched_domain containing group. | |
3600 | * @local_group: Does group contain this_cpu. | |
3601 | * @cpus: Set of cpus considered for load balancing. | |
3602 | * @balance: Should we balance. | |
3603 | * @sgs: variable to hold the statistics for this group. | |
3604 | */ | |
3605 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3606 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3607 | int local_group, const struct cpumask *cpus, | |
3608 | int *balance, struct sg_lb_stats *sgs) | |
3609 | { | |
3610 | unsigned long load, max_cpu_load, min_cpu_load; | |
3611 | int i; | |
3612 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3613 | unsigned long sum_avg_load_per_task; | |
3614 | unsigned long avg_load_per_task; | |
3615 | ||
3616 | if (local_group) | |
3617 | balance_cpu = group_first_cpu(group); | |
3618 | ||
3619 | /* Tally up the load of all CPUs in the group */ | |
3620 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3621 | max_cpu_load = 0; | |
3622 | min_cpu_load = ~0UL; | |
408ed066 | 3623 | |
1f8c553d GS |
3624 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3625 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3626 | |
1f8c553d GS |
3627 | if (*sd_idle && rq->nr_running) |
3628 | *sd_idle = 0; | |
5c45bf27 | 3629 | |
1f8c553d | 3630 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3631 | if (local_group) { |
1f8c553d GS |
3632 | if (idle_cpu(i) && !first_idle_cpu) { |
3633 | first_idle_cpu = 1; | |
3634 | balance_cpu = i; | |
3635 | } | |
3636 | ||
3637 | load = target_load(i, load_idx); | |
3638 | } else { | |
3639 | load = source_load(i, load_idx); | |
3640 | if (load > max_cpu_load) | |
3641 | max_cpu_load = load; | |
3642 | if (min_cpu_load > load) | |
3643 | min_cpu_load = load; | |
1da177e4 | 3644 | } |
5c45bf27 | 3645 | |
1f8c553d GS |
3646 | sgs->group_load += load; |
3647 | sgs->sum_nr_running += rq->nr_running; | |
3648 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3649 | |
1f8c553d GS |
3650 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3651 | } | |
5c45bf27 | 3652 | |
1f8c553d GS |
3653 | /* |
3654 | * First idle cpu or the first cpu(busiest) in this sched group | |
3655 | * is eligible for doing load balancing at this and above | |
3656 | * domains. In the newly idle case, we will allow all the cpu's | |
3657 | * to do the newly idle load balance. | |
3658 | */ | |
3659 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3660 | balance_cpu != this_cpu && balance) { | |
3661 | *balance = 0; | |
3662 | return; | |
3663 | } | |
5c45bf27 | 3664 | |
1f8c553d GS |
3665 | /* Adjust by relative CPU power of the group */ |
3666 | sgs->avg_load = sg_div_cpu_power(group, | |
3667 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3668 | |
1f8c553d GS |
3669 | |
3670 | /* | |
3671 | * Consider the group unbalanced when the imbalance is larger | |
3672 | * than the average weight of two tasks. | |
3673 | * | |
3674 | * APZ: with cgroup the avg task weight can vary wildly and | |
3675 | * might not be a suitable number - should we keep a | |
3676 | * normalized nr_running number somewhere that negates | |
3677 | * the hierarchy? | |
3678 | */ | |
3679 | avg_load_per_task = sg_div_cpu_power(group, | |
3680 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3681 | ||
3682 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3683 | sgs->group_imb = 1; | |
3684 | ||
3685 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3686 | ||
3687 | } | |
dd41f596 | 3688 | |
37abe198 GS |
3689 | /** |
3690 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3691 | * @sd: sched_domain whose statistics are to be updated. | |
3692 | * @this_cpu: Cpu for which load balance is currently performed. | |
3693 | * @idle: Idle status of this_cpu | |
3694 | * @sd_idle: Idle status of the sched_domain containing group. | |
3695 | * @cpus: Set of cpus considered for load balancing. | |
3696 | * @balance: Should we balance. | |
3697 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3698 | */ |
37abe198 GS |
3699 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3700 | enum cpu_idle_type idle, int *sd_idle, | |
3701 | const struct cpumask *cpus, int *balance, | |
3702 | struct sd_lb_stats *sds) | |
1da177e4 | 3703 | { |
222d656d | 3704 | struct sched_group *group = sd->groups; |
37abe198 | 3705 | struct sg_lb_stats sgs; |
222d656d GS |
3706 | int load_idx; |
3707 | ||
c071df18 | 3708 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3709 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3710 | |
3711 | do { | |
1da177e4 | 3712 | int local_group; |
1da177e4 | 3713 | |
758b2cdc RR |
3714 | local_group = cpumask_test_cpu(this_cpu, |
3715 | sched_group_cpus(group)); | |
381be78f | 3716 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3717 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3718 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3719 | |
37abe198 GS |
3720 | if (local_group && balance && !(*balance)) |
3721 | return; | |
783609c6 | 3722 | |
37abe198 GS |
3723 | sds->total_load += sgs.group_load; |
3724 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3725 | |
1da177e4 | 3726 | if (local_group) { |
37abe198 GS |
3727 | sds->this_load = sgs.avg_load; |
3728 | sds->this = group; | |
3729 | sds->this_nr_running = sgs.sum_nr_running; | |
3730 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3731 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3732 | (sgs.sum_nr_running > sgs.group_capacity || |
3733 | sgs.group_imb)) { | |
37abe198 GS |
3734 | sds->max_load = sgs.avg_load; |
3735 | sds->busiest = group; | |
3736 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3737 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3738 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3739 | } |
5c45bf27 | 3740 | |
c071df18 | 3741 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3742 | group = group->next; |
3743 | } while (group != sd->groups); | |
3744 | ||
37abe198 | 3745 | } |
1da177e4 | 3746 | |
2e6f44ae GS |
3747 | /** |
3748 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3749 | * amongst the groups of a sched_domain, during |
3750 | * load balancing. | |
2e6f44ae GS |
3751 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3752 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3753 | * @imbalance: Variable to store the imbalance. | |
3754 | */ | |
3755 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3756 | int this_cpu, unsigned long *imbalance) | |
3757 | { | |
3758 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3759 | unsigned int imbn = 2; | |
3760 | ||
3761 | if (sds->this_nr_running) { | |
3762 | sds->this_load_per_task /= sds->this_nr_running; | |
3763 | if (sds->busiest_load_per_task > | |
3764 | sds->this_load_per_task) | |
3765 | imbn = 1; | |
3766 | } else | |
3767 | sds->this_load_per_task = | |
3768 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3769 | |
2e6f44ae GS |
3770 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3771 | sds->busiest_load_per_task * imbn) { | |
3772 | *imbalance = sds->busiest_load_per_task; | |
3773 | return; | |
3774 | } | |
908a7c1b | 3775 | |
1da177e4 | 3776 | /* |
2e6f44ae GS |
3777 | * OK, we don't have enough imbalance to justify moving tasks, |
3778 | * however we may be able to increase total CPU power used by | |
3779 | * moving them. | |
1da177e4 | 3780 | */ |
2dd73a4f | 3781 | |
2e6f44ae GS |
3782 | pwr_now += sds->busiest->__cpu_power * |
3783 | min(sds->busiest_load_per_task, sds->max_load); | |
3784 | pwr_now += sds->this->__cpu_power * | |
3785 | min(sds->this_load_per_task, sds->this_load); | |
3786 | pwr_now /= SCHED_LOAD_SCALE; | |
3787 | ||
3788 | /* Amount of load we'd subtract */ | |
3789 | tmp = sg_div_cpu_power(sds->busiest, | |
3790 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3791 | if (sds->max_load > tmp) | |
3792 | pwr_move += sds->busiest->__cpu_power * | |
3793 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3794 | ||
3795 | /* Amount of load we'd add */ | |
3796 | if (sds->max_load * sds->busiest->__cpu_power < | |
3797 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3798 | tmp = sg_div_cpu_power(sds->this, | |
3799 | sds->max_load * sds->busiest->__cpu_power); | |
3800 | else | |
3801 | tmp = sg_div_cpu_power(sds->this, | |
3802 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3803 | pwr_move += sds->this->__cpu_power * | |
3804 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3805 | pwr_move /= SCHED_LOAD_SCALE; | |
3806 | ||
3807 | /* Move if we gain throughput */ | |
3808 | if (pwr_move > pwr_now) | |
3809 | *imbalance = sds->busiest_load_per_task; | |
3810 | } | |
dbc523a3 GS |
3811 | |
3812 | /** | |
3813 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3814 | * groups of a given sched_domain during load balance. | |
3815 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3816 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3817 | * @imbalance: The variable to store the imbalance. | |
3818 | */ | |
3819 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3820 | unsigned long *imbalance) | |
3821 | { | |
3822 | unsigned long max_pull; | |
2dd73a4f PW |
3823 | /* |
3824 | * In the presence of smp nice balancing, certain scenarios can have | |
3825 | * max load less than avg load(as we skip the groups at or below | |
3826 | * its cpu_power, while calculating max_load..) | |
3827 | */ | |
dbc523a3 | 3828 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3829 | *imbalance = 0; |
dbc523a3 | 3830 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3831 | } |
0c117f1b SS |
3832 | |
3833 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3834 | max_pull = min(sds->max_load - sds->avg_load, |
3835 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3836 | |
1da177e4 | 3837 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3838 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3839 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3840 | / SCHED_LOAD_SCALE; |
3841 | ||
2dd73a4f PW |
3842 | /* |
3843 | * if *imbalance is less than the average load per runnable task | |
3844 | * there is no gaurantee that any tasks will be moved so we'll have | |
3845 | * a think about bumping its value to force at least one task to be | |
3846 | * moved | |
3847 | */ | |
dbc523a3 GS |
3848 | if (*imbalance < sds->busiest_load_per_task) |
3849 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3850 | |
dbc523a3 | 3851 | } |
37abe198 | 3852 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3853 | |
b7bb4c9b GS |
3854 | /** |
3855 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3856 | * if there is an imbalance. If there isn't an imbalance, and | |
3857 | * the user has opted for power-savings, it returns a group whose | |
3858 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3859 | * such a group exists. | |
3860 | * | |
3861 | * Also calculates the amount of weighted load which should be moved | |
3862 | * to restore balance. | |
3863 | * | |
3864 | * @sd: The sched_domain whose busiest group is to be returned. | |
3865 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3866 | * @imbalance: Variable which stores amount of weighted load which should | |
3867 | * be moved to restore balance/put a group to idle. | |
3868 | * @idle: The idle status of this_cpu. | |
3869 | * @sd_idle: The idleness of sd | |
3870 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3871 | * @balance: Pointer to a variable indicating if this_cpu | |
3872 | * is the appropriate cpu to perform load balancing at this_level. | |
3873 | * | |
3874 | * Returns: - the busiest group if imbalance exists. | |
3875 | * - If no imbalance and user has opted for power-savings balance, | |
3876 | * return the least loaded group whose CPUs can be | |
3877 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3878 | */ |
3879 | static struct sched_group * | |
3880 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3881 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3882 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3883 | { | |
3884 | struct sd_lb_stats sds; | |
1da177e4 | 3885 | |
37abe198 | 3886 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3887 | |
37abe198 GS |
3888 | /* |
3889 | * Compute the various statistics relavent for load balancing at | |
3890 | * this level. | |
3891 | */ | |
3892 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3893 | balance, &sds); | |
3894 | ||
b7bb4c9b GS |
3895 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3896 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3897 | * at this level. | |
3898 | * 2) There is no busy sibling group to pull from. | |
3899 | * 3) This group is the busiest group. | |
3900 | * 4) This group is more busy than the avg busieness at this | |
3901 | * sched_domain. | |
3902 | * 5) The imbalance is within the specified limit. | |
3903 | * 6) Any rebalance would lead to ping-pong | |
3904 | */ | |
37abe198 GS |
3905 | if (balance && !(*balance)) |
3906 | goto ret; | |
1da177e4 | 3907 | |
b7bb4c9b GS |
3908 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3909 | goto out_balanced; | |
1da177e4 | 3910 | |
b7bb4c9b | 3911 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3912 | goto out_balanced; |
1da177e4 | 3913 | |
222d656d | 3914 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3915 | |
b7bb4c9b GS |
3916 | if (sds.this_load >= sds.avg_load) |
3917 | goto out_balanced; | |
3918 | ||
3919 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3920 | goto out_balanced; |
3921 | ||
222d656d GS |
3922 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3923 | if (sds.group_imb) | |
3924 | sds.busiest_load_per_task = | |
3925 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3926 | |
1da177e4 LT |
3927 | /* |
3928 | * We're trying to get all the cpus to the average_load, so we don't | |
3929 | * want to push ourselves above the average load, nor do we wish to | |
3930 | * reduce the max loaded cpu below the average load, as either of these | |
3931 | * actions would just result in more rebalancing later, and ping-pong | |
3932 | * tasks around. Thus we look for the minimum possible imbalance. | |
3933 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3934 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3935 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3936 | * appear as very large values with unsigned longs. |
3937 | */ | |
222d656d | 3938 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3939 | goto out_balanced; |
3940 | ||
dbc523a3 GS |
3941 | /* Looks like there is an imbalance. Compute it */ |
3942 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3943 | return sds.busiest; |
1da177e4 LT |
3944 | |
3945 | out_balanced: | |
c071df18 GS |
3946 | /* |
3947 | * There is no obvious imbalance. But check if we can do some balancing | |
3948 | * to save power. | |
3949 | */ | |
3950 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3951 | return sds.busiest; | |
783609c6 | 3952 | ret: |
1da177e4 LT |
3953 | *imbalance = 0; |
3954 | return NULL; | |
3955 | } | |
3956 | ||
3957 | /* | |
3958 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3959 | */ | |
70b97a7f | 3960 | static struct rq * |
d15bcfdb | 3961 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3962 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3963 | { |
70b97a7f | 3964 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3965 | unsigned long max_load = 0; |
1da177e4 LT |
3966 | int i; |
3967 | ||
758b2cdc | 3968 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3969 | unsigned long wl; |
0a2966b4 | 3970 | |
96f874e2 | 3971 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3972 | continue; |
3973 | ||
48f24c4d | 3974 | rq = cpu_rq(i); |
dd41f596 | 3975 | wl = weighted_cpuload(i); |
2dd73a4f | 3976 | |
dd41f596 | 3977 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3978 | continue; |
1da177e4 | 3979 | |
dd41f596 IM |
3980 | if (wl > max_load) { |
3981 | max_load = wl; | |
48f24c4d | 3982 | busiest = rq; |
1da177e4 LT |
3983 | } |
3984 | } | |
3985 | ||
3986 | return busiest; | |
3987 | } | |
3988 | ||
77391d71 NP |
3989 | /* |
3990 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3991 | * so long as it is large enough. | |
3992 | */ | |
3993 | #define MAX_PINNED_INTERVAL 512 | |
3994 | ||
df7c8e84 RR |
3995 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3996 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3997 | ||
1da177e4 LT |
3998 | /* |
3999 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4000 | * tasks if there is an imbalance. | |
1da177e4 | 4001 | */ |
70b97a7f | 4002 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4003 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4004 | int *balance) |
1da177e4 | 4005 | { |
43010659 | 4006 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4007 | struct sched_group *group; |
1da177e4 | 4008 | unsigned long imbalance; |
70b97a7f | 4009 | struct rq *busiest; |
fe2eea3f | 4010 | unsigned long flags; |
df7c8e84 | 4011 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4012 | |
96f874e2 | 4013 | cpumask_setall(cpus); |
7c16ec58 | 4014 | |
89c4710e SS |
4015 | /* |
4016 | * When power savings policy is enabled for the parent domain, idle | |
4017 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4018 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4019 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4020 | */ |
d15bcfdb | 4021 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4022 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4023 | sd_idle = 1; |
1da177e4 | 4024 | |
2d72376b | 4025 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4026 | |
0a2966b4 | 4027 | redo: |
c8cba857 | 4028 | update_shares(sd); |
0a2966b4 | 4029 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4030 | cpus, balance); |
783609c6 | 4031 | |
06066714 | 4032 | if (*balance == 0) |
783609c6 | 4033 | goto out_balanced; |
783609c6 | 4034 | |
1da177e4 LT |
4035 | if (!group) { |
4036 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4037 | goto out_balanced; | |
4038 | } | |
4039 | ||
7c16ec58 | 4040 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4041 | if (!busiest) { |
4042 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4043 | goto out_balanced; | |
4044 | } | |
4045 | ||
db935dbd | 4046 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4047 | |
4048 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4049 | ||
43010659 | 4050 | ld_moved = 0; |
1da177e4 LT |
4051 | if (busiest->nr_running > 1) { |
4052 | /* | |
4053 | * Attempt to move tasks. If find_busiest_group has found | |
4054 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4055 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4056 | * correctly treated as an imbalance. |
4057 | */ | |
fe2eea3f | 4058 | local_irq_save(flags); |
e17224bf | 4059 | double_rq_lock(this_rq, busiest); |
43010659 | 4060 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4061 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4062 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4063 | local_irq_restore(flags); |
81026794 | 4064 | |
46cb4b7c SS |
4065 | /* |
4066 | * some other cpu did the load balance for us. | |
4067 | */ | |
43010659 | 4068 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4069 | resched_cpu(this_cpu); |
4070 | ||
81026794 | 4071 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4072 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4073 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4074 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4075 | goto redo; |
81026794 | 4076 | goto out_balanced; |
0a2966b4 | 4077 | } |
1da177e4 | 4078 | } |
81026794 | 4079 | |
43010659 | 4080 | if (!ld_moved) { |
1da177e4 LT |
4081 | schedstat_inc(sd, lb_failed[idle]); |
4082 | sd->nr_balance_failed++; | |
4083 | ||
4084 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4085 | |
fe2eea3f | 4086 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4087 | |
4088 | /* don't kick the migration_thread, if the curr | |
4089 | * task on busiest cpu can't be moved to this_cpu | |
4090 | */ | |
96f874e2 RR |
4091 | if (!cpumask_test_cpu(this_cpu, |
4092 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4093 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4094 | all_pinned = 1; |
4095 | goto out_one_pinned; | |
4096 | } | |
4097 | ||
1da177e4 LT |
4098 | if (!busiest->active_balance) { |
4099 | busiest->active_balance = 1; | |
4100 | busiest->push_cpu = this_cpu; | |
81026794 | 4101 | active_balance = 1; |
1da177e4 | 4102 | } |
fe2eea3f | 4103 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4104 | if (active_balance) |
1da177e4 LT |
4105 | wake_up_process(busiest->migration_thread); |
4106 | ||
4107 | /* | |
4108 | * We've kicked active balancing, reset the failure | |
4109 | * counter. | |
4110 | */ | |
39507451 | 4111 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4112 | } |
81026794 | 4113 | } else |
1da177e4 LT |
4114 | sd->nr_balance_failed = 0; |
4115 | ||
81026794 | 4116 | if (likely(!active_balance)) { |
1da177e4 LT |
4117 | /* We were unbalanced, so reset the balancing interval */ |
4118 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4119 | } else { |
4120 | /* | |
4121 | * If we've begun active balancing, start to back off. This | |
4122 | * case may not be covered by the all_pinned logic if there | |
4123 | * is only 1 task on the busy runqueue (because we don't call | |
4124 | * move_tasks). | |
4125 | */ | |
4126 | if (sd->balance_interval < sd->max_interval) | |
4127 | sd->balance_interval *= 2; | |
1da177e4 LT |
4128 | } |
4129 | ||
43010659 | 4130 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4131 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4132 | ld_moved = -1; |
4133 | ||
4134 | goto out; | |
1da177e4 LT |
4135 | |
4136 | out_balanced: | |
1da177e4 LT |
4137 | schedstat_inc(sd, lb_balanced[idle]); |
4138 | ||
16cfb1c0 | 4139 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4140 | |
4141 | out_one_pinned: | |
1da177e4 | 4142 | /* tune up the balancing interval */ |
77391d71 NP |
4143 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4144 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4145 | sd->balance_interval *= 2; |
4146 | ||
48f24c4d | 4147 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4148 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4149 | ld_moved = -1; |
4150 | else | |
4151 | ld_moved = 0; | |
4152 | out: | |
c8cba857 PZ |
4153 | if (ld_moved) |
4154 | update_shares(sd); | |
c09595f6 | 4155 | return ld_moved; |
1da177e4 LT |
4156 | } |
4157 | ||
4158 | /* | |
4159 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4160 | * tasks if there is an imbalance. | |
4161 | * | |
d15bcfdb | 4162 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4163 | * this_rq is locked. |
4164 | */ | |
48f24c4d | 4165 | static int |
df7c8e84 | 4166 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4167 | { |
4168 | struct sched_group *group; | |
70b97a7f | 4169 | struct rq *busiest = NULL; |
1da177e4 | 4170 | unsigned long imbalance; |
43010659 | 4171 | int ld_moved = 0; |
5969fe06 | 4172 | int sd_idle = 0; |
969bb4e4 | 4173 | int all_pinned = 0; |
df7c8e84 | 4174 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4175 | |
96f874e2 | 4176 | cpumask_setall(cpus); |
5969fe06 | 4177 | |
89c4710e SS |
4178 | /* |
4179 | * When power savings policy is enabled for the parent domain, idle | |
4180 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4181 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4182 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4183 | */ |
4184 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4185 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4186 | sd_idle = 1; |
1da177e4 | 4187 | |
2d72376b | 4188 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4189 | redo: |
3e5459b4 | 4190 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4191 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4192 | &sd_idle, cpus, NULL); |
1da177e4 | 4193 | if (!group) { |
d15bcfdb | 4194 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4195 | goto out_balanced; |
1da177e4 LT |
4196 | } |
4197 | ||
7c16ec58 | 4198 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4199 | if (!busiest) { |
d15bcfdb | 4200 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4201 | goto out_balanced; |
1da177e4 LT |
4202 | } |
4203 | ||
db935dbd NP |
4204 | BUG_ON(busiest == this_rq); |
4205 | ||
d15bcfdb | 4206 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4207 | |
43010659 | 4208 | ld_moved = 0; |
d6d5cfaf NP |
4209 | if (busiest->nr_running > 1) { |
4210 | /* Attempt to move tasks */ | |
4211 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4212 | /* this_rq->clock is already updated */ |
4213 | update_rq_clock(busiest); | |
43010659 | 4214 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4215 | imbalance, sd, CPU_NEWLY_IDLE, |
4216 | &all_pinned); | |
1b12bbc7 | 4217 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4218 | |
969bb4e4 | 4219 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4220 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4221 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4222 | goto redo; |
4223 | } | |
d6d5cfaf NP |
4224 | } |
4225 | ||
43010659 | 4226 | if (!ld_moved) { |
36dffab6 | 4227 | int active_balance = 0; |
ad273b32 | 4228 | |
d15bcfdb | 4229 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4230 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4231 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4232 | return -1; |
ad273b32 VS |
4233 | |
4234 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4235 | return -1; | |
4236 | ||
4237 | if (sd->nr_balance_failed++ < 2) | |
4238 | return -1; | |
4239 | ||
4240 | /* | |
4241 | * The only task running in a non-idle cpu can be moved to this | |
4242 | * cpu in an attempt to completely freeup the other CPU | |
4243 | * package. The same method used to move task in load_balance() | |
4244 | * have been extended for load_balance_newidle() to speedup | |
4245 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4246 | * | |
4247 | * The package power saving logic comes from | |
4248 | * find_busiest_group(). If there are no imbalance, then | |
4249 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4250 | * f_b_g() will select a group from which a running task may be | |
4251 | * pulled to this cpu in order to make the other package idle. | |
4252 | * If there is no opportunity to make a package idle and if | |
4253 | * there are no imbalance, then f_b_g() will return NULL and no | |
4254 | * action will be taken in load_balance_newidle(). | |
4255 | * | |
4256 | * Under normal task pull operation due to imbalance, there | |
4257 | * will be more than one task in the source run queue and | |
4258 | * move_tasks() will succeed. ld_moved will be true and this | |
4259 | * active balance code will not be triggered. | |
4260 | */ | |
4261 | ||
4262 | /* Lock busiest in correct order while this_rq is held */ | |
4263 | double_lock_balance(this_rq, busiest); | |
4264 | ||
4265 | /* | |
4266 | * don't kick the migration_thread, if the curr | |
4267 | * task on busiest cpu can't be moved to this_cpu | |
4268 | */ | |
6ca09dfc | 4269 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4270 | double_unlock_balance(this_rq, busiest); |
4271 | all_pinned = 1; | |
4272 | return ld_moved; | |
4273 | } | |
4274 | ||
4275 | if (!busiest->active_balance) { | |
4276 | busiest->active_balance = 1; | |
4277 | busiest->push_cpu = this_cpu; | |
4278 | active_balance = 1; | |
4279 | } | |
4280 | ||
4281 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4282 | /* |
4283 | * Should not call ttwu while holding a rq->lock | |
4284 | */ | |
4285 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4286 | if (active_balance) |
4287 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4288 | spin_lock(&this_rq->lock); |
ad273b32 | 4289 | |
5969fe06 | 4290 | } else |
16cfb1c0 | 4291 | sd->nr_balance_failed = 0; |
1da177e4 | 4292 | |
3e5459b4 | 4293 | update_shares_locked(this_rq, sd); |
43010659 | 4294 | return ld_moved; |
16cfb1c0 NP |
4295 | |
4296 | out_balanced: | |
d15bcfdb | 4297 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4298 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4299 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4300 | return -1; |
16cfb1c0 | 4301 | sd->nr_balance_failed = 0; |
48f24c4d | 4302 | |
16cfb1c0 | 4303 | return 0; |
1da177e4 LT |
4304 | } |
4305 | ||
4306 | /* | |
4307 | * idle_balance is called by schedule() if this_cpu is about to become | |
4308 | * idle. Attempts to pull tasks from other CPUs. | |
4309 | */ | |
70b97a7f | 4310 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4311 | { |
4312 | struct sched_domain *sd; | |
efbe027e | 4313 | int pulled_task = 0; |
dd41f596 | 4314 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4315 | |
4316 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4317 | unsigned long interval; |
4318 | ||
4319 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4320 | continue; | |
4321 | ||
4322 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4323 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4324 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4325 | sd); |
92c4ca5c CL |
4326 | |
4327 | interval = msecs_to_jiffies(sd->balance_interval); | |
4328 | if (time_after(next_balance, sd->last_balance + interval)) | |
4329 | next_balance = sd->last_balance + interval; | |
4330 | if (pulled_task) | |
4331 | break; | |
1da177e4 | 4332 | } |
dd41f596 | 4333 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4334 | /* |
4335 | * We are going idle. next_balance may be set based on | |
4336 | * a busy processor. So reset next_balance. | |
4337 | */ | |
4338 | this_rq->next_balance = next_balance; | |
dd41f596 | 4339 | } |
1da177e4 LT |
4340 | } |
4341 | ||
4342 | /* | |
4343 | * active_load_balance is run by migration threads. It pushes running tasks | |
4344 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4345 | * running on each physical CPU where possible, and avoids physical / | |
4346 | * logical imbalances. | |
4347 | * | |
4348 | * Called with busiest_rq locked. | |
4349 | */ | |
70b97a7f | 4350 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4351 | { |
39507451 | 4352 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4353 | struct sched_domain *sd; |
4354 | struct rq *target_rq; | |
39507451 | 4355 | |
48f24c4d | 4356 | /* Is there any task to move? */ |
39507451 | 4357 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4358 | return; |
4359 | ||
4360 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4361 | |
4362 | /* | |
39507451 | 4363 | * This condition is "impossible", if it occurs |
41a2d6cf | 4364 | * we need to fix it. Originally reported by |
39507451 | 4365 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4366 | */ |
39507451 | 4367 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4368 | |
39507451 NP |
4369 | /* move a task from busiest_rq to target_rq */ |
4370 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4371 | update_rq_clock(busiest_rq); |
4372 | update_rq_clock(target_rq); | |
39507451 NP |
4373 | |
4374 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4375 | for_each_domain(target_cpu, sd) { |
39507451 | 4376 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4377 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4378 | break; |
c96d145e | 4379 | } |
39507451 | 4380 | |
48f24c4d | 4381 | if (likely(sd)) { |
2d72376b | 4382 | schedstat_inc(sd, alb_count); |
39507451 | 4383 | |
43010659 PW |
4384 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4385 | sd, CPU_IDLE)) | |
48f24c4d IM |
4386 | schedstat_inc(sd, alb_pushed); |
4387 | else | |
4388 | schedstat_inc(sd, alb_failed); | |
4389 | } | |
1b12bbc7 | 4390 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4391 | } |
4392 | ||
46cb4b7c SS |
4393 | #ifdef CONFIG_NO_HZ |
4394 | static struct { | |
4395 | atomic_t load_balancer; | |
7d1e6a9b | 4396 | cpumask_var_t cpu_mask; |
f711f609 | 4397 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4398 | } nohz ____cacheline_aligned = { |
4399 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4400 | }; |
4401 | ||
eea08f32 AB |
4402 | int get_nohz_load_balancer(void) |
4403 | { | |
4404 | return atomic_read(&nohz.load_balancer); | |
4405 | } | |
4406 | ||
f711f609 GS |
4407 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4408 | /** | |
4409 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4410 | * @cpu: The cpu whose lowest level of sched domain is to | |
4411 | * be returned. | |
4412 | * @flag: The flag to check for the lowest sched_domain | |
4413 | * for the given cpu. | |
4414 | * | |
4415 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4416 | */ | |
4417 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4418 | { | |
4419 | struct sched_domain *sd; | |
4420 | ||
4421 | for_each_domain(cpu, sd) | |
4422 | if (sd && (sd->flags & flag)) | |
4423 | break; | |
4424 | ||
4425 | return sd; | |
4426 | } | |
4427 | ||
4428 | /** | |
4429 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4430 | * @cpu: The cpu whose domains we're iterating over. | |
4431 | * @sd: variable holding the value of the power_savings_sd | |
4432 | * for cpu. | |
4433 | * @flag: The flag to filter the sched_domains to be iterated. | |
4434 | * | |
4435 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4436 | * set, starting from the lowest sched_domain to the highest. | |
4437 | */ | |
4438 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4439 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4440 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4441 | ||
4442 | /** | |
4443 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4444 | * @ilb_group: group to be checked for semi-idleness | |
4445 | * | |
4446 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4447 | * | |
4448 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4449 | * and atleast one non-idle CPU. This helper function checks if the given | |
4450 | * sched_group is semi-idle or not. | |
4451 | */ | |
4452 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4453 | { | |
4454 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4455 | sched_group_cpus(ilb_group)); | |
4456 | ||
4457 | /* | |
4458 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4459 | * and atleast one idle cpu. | |
4460 | */ | |
4461 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4462 | return 0; | |
4463 | ||
4464 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4465 | return 0; | |
4466 | ||
4467 | return 1; | |
4468 | } | |
4469 | /** | |
4470 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4471 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4472 | * | |
4473 | * Returns: Returns the id of the idle load balancer if it exists, | |
4474 | * Else, returns >= nr_cpu_ids. | |
4475 | * | |
4476 | * This algorithm picks the idle load balancer such that it belongs to a | |
4477 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4478 | * completely idle packages/cores just for the purpose of idle load balancing | |
4479 | * when there are other idle cpu's which are better suited for that job. | |
4480 | */ | |
4481 | static int find_new_ilb(int cpu) | |
4482 | { | |
4483 | struct sched_domain *sd; | |
4484 | struct sched_group *ilb_group; | |
4485 | ||
4486 | /* | |
4487 | * Have idle load balancer selection from semi-idle packages only | |
4488 | * when power-aware load balancing is enabled | |
4489 | */ | |
4490 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4491 | goto out_done; | |
4492 | ||
4493 | /* | |
4494 | * Optimize for the case when we have no idle CPUs or only one | |
4495 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4496 | */ | |
4497 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4498 | goto out_done; | |
4499 | ||
4500 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4501 | ilb_group = sd->groups; | |
4502 | ||
4503 | do { | |
4504 | if (is_semi_idle_group(ilb_group)) | |
4505 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4506 | ||
4507 | ilb_group = ilb_group->next; | |
4508 | ||
4509 | } while (ilb_group != sd->groups); | |
4510 | } | |
4511 | ||
4512 | out_done: | |
4513 | return cpumask_first(nohz.cpu_mask); | |
4514 | } | |
4515 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4516 | static inline int find_new_ilb(int call_cpu) | |
4517 | { | |
6e29ec57 | 4518 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4519 | } |
4520 | #endif | |
4521 | ||
7835b98b | 4522 | /* |
46cb4b7c SS |
4523 | * This routine will try to nominate the ilb (idle load balancing) |
4524 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4525 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4526 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4527 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4528 | * arrives... | |
4529 | * | |
4530 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4531 | * for idle load balancing. ilb owner will still be part of | |
4532 | * nohz.cpu_mask.. | |
7835b98b | 4533 | * |
46cb4b7c SS |
4534 | * While stopping the tick, this cpu will become the ilb owner if there |
4535 | * is no other owner. And will be the owner till that cpu becomes busy | |
4536 | * or if all cpus in the system stop their ticks at which point | |
4537 | * there is no need for ilb owner. | |
4538 | * | |
4539 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4540 | * next busy scheduler_tick() | |
4541 | */ | |
4542 | int select_nohz_load_balancer(int stop_tick) | |
4543 | { | |
4544 | int cpu = smp_processor_id(); | |
4545 | ||
4546 | if (stop_tick) { | |
46cb4b7c SS |
4547 | cpu_rq(cpu)->in_nohz_recently = 1; |
4548 | ||
483b4ee6 SS |
4549 | if (!cpu_active(cpu)) { |
4550 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4551 | return 0; | |
4552 | ||
4553 | /* | |
4554 | * If we are going offline and still the leader, | |
4555 | * give up! | |
4556 | */ | |
46cb4b7c SS |
4557 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4558 | BUG(); | |
483b4ee6 | 4559 | |
46cb4b7c SS |
4560 | return 0; |
4561 | } | |
4562 | ||
483b4ee6 SS |
4563 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4564 | ||
46cb4b7c | 4565 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4566 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4567 | if (atomic_read(&nohz.load_balancer) == cpu) |
4568 | atomic_set(&nohz.load_balancer, -1); | |
4569 | return 0; | |
4570 | } | |
4571 | ||
4572 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4573 | /* make me the ilb owner */ | |
4574 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4575 | return 1; | |
e790fb0b GS |
4576 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4577 | int new_ilb; | |
4578 | ||
4579 | if (!(sched_smt_power_savings || | |
4580 | sched_mc_power_savings)) | |
4581 | return 1; | |
4582 | /* | |
4583 | * Check to see if there is a more power-efficient | |
4584 | * ilb. | |
4585 | */ | |
4586 | new_ilb = find_new_ilb(cpu); | |
4587 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4588 | atomic_set(&nohz.load_balancer, -1); | |
4589 | resched_cpu(new_ilb); | |
4590 | return 0; | |
4591 | } | |
46cb4b7c | 4592 | return 1; |
e790fb0b | 4593 | } |
46cb4b7c | 4594 | } else { |
7d1e6a9b | 4595 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4596 | return 0; |
4597 | ||
7d1e6a9b | 4598 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4599 | |
4600 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4601 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4602 | BUG(); | |
4603 | } | |
4604 | return 0; | |
4605 | } | |
4606 | #endif | |
4607 | ||
4608 | static DEFINE_SPINLOCK(balancing); | |
4609 | ||
4610 | /* | |
7835b98b CL |
4611 | * It checks each scheduling domain to see if it is due to be balanced, |
4612 | * and initiates a balancing operation if so. | |
4613 | * | |
4614 | * Balancing parameters are set up in arch_init_sched_domains. | |
4615 | */ | |
a9957449 | 4616 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4617 | { |
46cb4b7c SS |
4618 | int balance = 1; |
4619 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4620 | unsigned long interval; |
4621 | struct sched_domain *sd; | |
46cb4b7c | 4622 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4623 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4624 | int update_next_balance = 0; |
d07355f5 | 4625 | int need_serialize; |
1da177e4 | 4626 | |
46cb4b7c | 4627 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4628 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4629 | continue; | |
4630 | ||
4631 | interval = sd->balance_interval; | |
d15bcfdb | 4632 | if (idle != CPU_IDLE) |
1da177e4 LT |
4633 | interval *= sd->busy_factor; |
4634 | ||
4635 | /* scale ms to jiffies */ | |
4636 | interval = msecs_to_jiffies(interval); | |
4637 | if (unlikely(!interval)) | |
4638 | interval = 1; | |
dd41f596 IM |
4639 | if (interval > HZ*NR_CPUS/10) |
4640 | interval = HZ*NR_CPUS/10; | |
4641 | ||
d07355f5 | 4642 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4643 | |
d07355f5 | 4644 | if (need_serialize) { |
08c183f3 CL |
4645 | if (!spin_trylock(&balancing)) |
4646 | goto out; | |
4647 | } | |
4648 | ||
c9819f45 | 4649 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4650 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4651 | /* |
4652 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4653 | * longer idle, or one of our SMT siblings is |
4654 | * not idle. | |
4655 | */ | |
d15bcfdb | 4656 | idle = CPU_NOT_IDLE; |
1da177e4 | 4657 | } |
1bd77f2d | 4658 | sd->last_balance = jiffies; |
1da177e4 | 4659 | } |
d07355f5 | 4660 | if (need_serialize) |
08c183f3 CL |
4661 | spin_unlock(&balancing); |
4662 | out: | |
f549da84 | 4663 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4664 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4665 | update_next_balance = 1; |
4666 | } | |
783609c6 SS |
4667 | |
4668 | /* | |
4669 | * Stop the load balance at this level. There is another | |
4670 | * CPU in our sched group which is doing load balancing more | |
4671 | * actively. | |
4672 | */ | |
4673 | if (!balance) | |
4674 | break; | |
1da177e4 | 4675 | } |
f549da84 SS |
4676 | |
4677 | /* | |
4678 | * next_balance will be updated only when there is a need. | |
4679 | * When the cpu is attached to null domain for ex, it will not be | |
4680 | * updated. | |
4681 | */ | |
4682 | if (likely(update_next_balance)) | |
4683 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4684 | } |
4685 | ||
4686 | /* | |
4687 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4688 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4689 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4690 | */ | |
4691 | static void run_rebalance_domains(struct softirq_action *h) | |
4692 | { | |
dd41f596 IM |
4693 | int this_cpu = smp_processor_id(); |
4694 | struct rq *this_rq = cpu_rq(this_cpu); | |
4695 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4696 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4697 | |
dd41f596 | 4698 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4699 | |
4700 | #ifdef CONFIG_NO_HZ | |
4701 | /* | |
4702 | * If this cpu is the owner for idle load balancing, then do the | |
4703 | * balancing on behalf of the other idle cpus whose ticks are | |
4704 | * stopped. | |
4705 | */ | |
dd41f596 IM |
4706 | if (this_rq->idle_at_tick && |
4707 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4708 | struct rq *rq; |
4709 | int balance_cpu; | |
4710 | ||
7d1e6a9b RR |
4711 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4712 | if (balance_cpu == this_cpu) | |
4713 | continue; | |
4714 | ||
46cb4b7c SS |
4715 | /* |
4716 | * If this cpu gets work to do, stop the load balancing | |
4717 | * work being done for other cpus. Next load | |
4718 | * balancing owner will pick it up. | |
4719 | */ | |
4720 | if (need_resched()) | |
4721 | break; | |
4722 | ||
de0cf899 | 4723 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4724 | |
4725 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4726 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4727 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4728 | } |
4729 | } | |
4730 | #endif | |
4731 | } | |
4732 | ||
8a0be9ef FW |
4733 | static inline int on_null_domain(int cpu) |
4734 | { | |
4735 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4736 | } | |
4737 | ||
46cb4b7c SS |
4738 | /* |
4739 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4740 | * | |
4741 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4742 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4743 | * if the whole system is idle. | |
4744 | */ | |
dd41f596 | 4745 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4746 | { |
46cb4b7c SS |
4747 | #ifdef CONFIG_NO_HZ |
4748 | /* | |
4749 | * If we were in the nohz mode recently and busy at the current | |
4750 | * scheduler tick, then check if we need to nominate new idle | |
4751 | * load balancer. | |
4752 | */ | |
4753 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4754 | rq->in_nohz_recently = 0; | |
4755 | ||
4756 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4757 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4758 | atomic_set(&nohz.load_balancer, -1); |
4759 | } | |
4760 | ||
4761 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4762 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4763 | |
434d53b0 | 4764 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4765 | resched_cpu(ilb); |
4766 | } | |
4767 | } | |
4768 | ||
4769 | /* | |
4770 | * If this cpu is idle and doing idle load balancing for all the | |
4771 | * cpus with ticks stopped, is it time for that to stop? | |
4772 | */ | |
4773 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4774 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4775 | resched_cpu(cpu); |
4776 | return; | |
4777 | } | |
4778 | ||
4779 | /* | |
4780 | * If this cpu is idle and the idle load balancing is done by | |
4781 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4782 | */ | |
4783 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4784 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4785 | return; |
4786 | #endif | |
8a0be9ef FW |
4787 | /* Don't need to rebalance while attached to NULL domain */ |
4788 | if (time_after_eq(jiffies, rq->next_balance) && | |
4789 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4790 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4791 | } |
dd41f596 IM |
4792 | |
4793 | #else /* CONFIG_SMP */ | |
4794 | ||
1da177e4 LT |
4795 | /* |
4796 | * on UP we do not need to balance between CPUs: | |
4797 | */ | |
70b97a7f | 4798 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4799 | { |
4800 | } | |
dd41f596 | 4801 | |
1da177e4 LT |
4802 | #endif |
4803 | ||
1da177e4 LT |
4804 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4805 | ||
4806 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4807 | ||
4808 | /* | |
c5f8d995 | 4809 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4810 | * @p in case that task is currently running. |
c5f8d995 HS |
4811 | * |
4812 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4813 | */ |
c5f8d995 HS |
4814 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4815 | { | |
4816 | u64 ns = 0; | |
4817 | ||
4818 | if (task_current(rq, p)) { | |
4819 | update_rq_clock(rq); | |
4820 | ns = rq->clock - p->se.exec_start; | |
4821 | if ((s64)ns < 0) | |
4822 | ns = 0; | |
4823 | } | |
4824 | ||
4825 | return ns; | |
4826 | } | |
4827 | ||
bb34d92f | 4828 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4829 | { |
1da177e4 | 4830 | unsigned long flags; |
41b86e9c | 4831 | struct rq *rq; |
bb34d92f | 4832 | u64 ns = 0; |
48f24c4d | 4833 | |
41b86e9c | 4834 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4835 | ns = do_task_delta_exec(p, rq); |
4836 | task_rq_unlock(rq, &flags); | |
1508487e | 4837 | |
c5f8d995 HS |
4838 | return ns; |
4839 | } | |
f06febc9 | 4840 | |
c5f8d995 HS |
4841 | /* |
4842 | * Return accounted runtime for the task. | |
4843 | * In case the task is currently running, return the runtime plus current's | |
4844 | * pending runtime that have not been accounted yet. | |
4845 | */ | |
4846 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4847 | { | |
4848 | unsigned long flags; | |
4849 | struct rq *rq; | |
4850 | u64 ns = 0; | |
4851 | ||
4852 | rq = task_rq_lock(p, &flags); | |
4853 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4854 | task_rq_unlock(rq, &flags); | |
4855 | ||
4856 | return ns; | |
4857 | } | |
48f24c4d | 4858 | |
c5f8d995 HS |
4859 | /* |
4860 | * Return sum_exec_runtime for the thread group. | |
4861 | * In case the task is currently running, return the sum plus current's | |
4862 | * pending runtime that have not been accounted yet. | |
4863 | * | |
4864 | * Note that the thread group might have other running tasks as well, | |
4865 | * so the return value not includes other pending runtime that other | |
4866 | * running tasks might have. | |
4867 | */ | |
4868 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4869 | { | |
4870 | struct task_cputime totals; | |
4871 | unsigned long flags; | |
4872 | struct rq *rq; | |
4873 | u64 ns; | |
4874 | ||
4875 | rq = task_rq_lock(p, &flags); | |
4876 | thread_group_cputime(p, &totals); | |
4877 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4878 | task_rq_unlock(rq, &flags); |
48f24c4d | 4879 | |
1da177e4 LT |
4880 | return ns; |
4881 | } | |
4882 | ||
1da177e4 LT |
4883 | /* |
4884 | * Account user cpu time to a process. | |
4885 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4886 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4887 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4888 | */ |
457533a7 MS |
4889 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4890 | cputime_t cputime_scaled) | |
1da177e4 LT |
4891 | { |
4892 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4893 | cputime64_t tmp; | |
4894 | ||
457533a7 | 4895 | /* Add user time to process. */ |
1da177e4 | 4896 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4897 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4898 | account_group_user_time(p, cputime); |
1da177e4 LT |
4899 | |
4900 | /* Add user time to cpustat. */ | |
4901 | tmp = cputime_to_cputime64(cputime); | |
4902 | if (TASK_NICE(p) > 0) | |
4903 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4904 | else | |
4905 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4906 | |
4907 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4908 | /* Account for user time used */ |
4909 | acct_update_integrals(p); | |
1da177e4 LT |
4910 | } |
4911 | ||
94886b84 LV |
4912 | /* |
4913 | * Account guest cpu time to a process. | |
4914 | * @p: the process that the cpu time gets accounted to | |
4915 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4916 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4917 | */ |
457533a7 MS |
4918 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4919 | cputime_t cputime_scaled) | |
94886b84 LV |
4920 | { |
4921 | cputime64_t tmp; | |
4922 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4923 | ||
4924 | tmp = cputime_to_cputime64(cputime); | |
4925 | ||
457533a7 | 4926 | /* Add guest time to process. */ |
94886b84 | 4927 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4928 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4929 | account_group_user_time(p, cputime); |
94886b84 LV |
4930 | p->gtime = cputime_add(p->gtime, cputime); |
4931 | ||
457533a7 | 4932 | /* Add guest time to cpustat. */ |
94886b84 LV |
4933 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4934 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4935 | } | |
4936 | ||
1da177e4 LT |
4937 | /* |
4938 | * Account system cpu time to a process. | |
4939 | * @p: the process that the cpu time gets accounted to | |
4940 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4941 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4942 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4943 | */ |
4944 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4945 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4946 | { |
4947 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4948 | cputime64_t tmp; |
4949 | ||
983ed7a6 | 4950 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4951 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4952 | return; |
4953 | } | |
94886b84 | 4954 | |
457533a7 | 4955 | /* Add system time to process. */ |
1da177e4 | 4956 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4957 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4958 | account_group_system_time(p, cputime); |
1da177e4 LT |
4959 | |
4960 | /* Add system time to cpustat. */ | |
4961 | tmp = cputime_to_cputime64(cputime); | |
4962 | if (hardirq_count() - hardirq_offset) | |
4963 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4964 | else if (softirq_count()) | |
4965 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4966 | else |
79741dd3 MS |
4967 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4968 | ||
ef12fefa BR |
4969 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4970 | ||
1da177e4 LT |
4971 | /* Account for system time used */ |
4972 | acct_update_integrals(p); | |
1da177e4 LT |
4973 | } |
4974 | ||
c66f08be | 4975 | /* |
1da177e4 | 4976 | * Account for involuntary wait time. |
1da177e4 | 4977 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4978 | */ |
79741dd3 | 4979 | void account_steal_time(cputime_t cputime) |
c66f08be | 4980 | { |
79741dd3 MS |
4981 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4982 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4983 | ||
4984 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4985 | } |
4986 | ||
1da177e4 | 4987 | /* |
79741dd3 MS |
4988 | * Account for idle time. |
4989 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4990 | */ |
79741dd3 | 4991 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4992 | { |
4993 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4994 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4995 | struct rq *rq = this_rq(); |
1da177e4 | 4996 | |
79741dd3 MS |
4997 | if (atomic_read(&rq->nr_iowait) > 0) |
4998 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4999 | else | |
5000 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5001 | } |
5002 | ||
79741dd3 MS |
5003 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5004 | ||
5005 | /* | |
5006 | * Account a single tick of cpu time. | |
5007 | * @p: the process that the cpu time gets accounted to | |
5008 | * @user_tick: indicates if the tick is a user or a system tick | |
5009 | */ | |
5010 | void account_process_tick(struct task_struct *p, int user_tick) | |
5011 | { | |
5012 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5013 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5014 | struct rq *rq = this_rq(); | |
5015 | ||
5016 | if (user_tick) | |
5017 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5018 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5019 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5020 | one_jiffy_scaled); | |
5021 | else | |
5022 | account_idle_time(one_jiffy); | |
5023 | } | |
5024 | ||
5025 | /* | |
5026 | * Account multiple ticks of steal time. | |
5027 | * @p: the process from which the cpu time has been stolen | |
5028 | * @ticks: number of stolen ticks | |
5029 | */ | |
5030 | void account_steal_ticks(unsigned long ticks) | |
5031 | { | |
5032 | account_steal_time(jiffies_to_cputime(ticks)); | |
5033 | } | |
5034 | ||
5035 | /* | |
5036 | * Account multiple ticks of idle time. | |
5037 | * @ticks: number of stolen ticks | |
5038 | */ | |
5039 | void account_idle_ticks(unsigned long ticks) | |
5040 | { | |
5041 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5042 | } |
5043 | ||
79741dd3 MS |
5044 | #endif |
5045 | ||
49048622 BS |
5046 | /* |
5047 | * Use precise platform statistics if available: | |
5048 | */ | |
5049 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5050 | cputime_t task_utime(struct task_struct *p) | |
5051 | { | |
5052 | return p->utime; | |
5053 | } | |
5054 | ||
5055 | cputime_t task_stime(struct task_struct *p) | |
5056 | { | |
5057 | return p->stime; | |
5058 | } | |
5059 | #else | |
5060 | cputime_t task_utime(struct task_struct *p) | |
5061 | { | |
5062 | clock_t utime = cputime_to_clock_t(p->utime), | |
5063 | total = utime + cputime_to_clock_t(p->stime); | |
5064 | u64 temp; | |
5065 | ||
5066 | /* | |
5067 | * Use CFS's precise accounting: | |
5068 | */ | |
5069 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5070 | ||
5071 | if (total) { | |
5072 | temp *= utime; | |
5073 | do_div(temp, total); | |
5074 | } | |
5075 | utime = (clock_t)temp; | |
5076 | ||
5077 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5078 | return p->prev_utime; | |
5079 | } | |
5080 | ||
5081 | cputime_t task_stime(struct task_struct *p) | |
5082 | { | |
5083 | clock_t stime; | |
5084 | ||
5085 | /* | |
5086 | * Use CFS's precise accounting. (we subtract utime from | |
5087 | * the total, to make sure the total observed by userspace | |
5088 | * grows monotonically - apps rely on that): | |
5089 | */ | |
5090 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5091 | cputime_to_clock_t(task_utime(p)); | |
5092 | ||
5093 | if (stime >= 0) | |
5094 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5095 | ||
5096 | return p->prev_stime; | |
5097 | } | |
5098 | #endif | |
5099 | ||
5100 | inline cputime_t task_gtime(struct task_struct *p) | |
5101 | { | |
5102 | return p->gtime; | |
5103 | } | |
5104 | ||
7835b98b CL |
5105 | /* |
5106 | * This function gets called by the timer code, with HZ frequency. | |
5107 | * We call it with interrupts disabled. | |
5108 | * | |
5109 | * It also gets called by the fork code, when changing the parent's | |
5110 | * timeslices. | |
5111 | */ | |
5112 | void scheduler_tick(void) | |
5113 | { | |
7835b98b CL |
5114 | int cpu = smp_processor_id(); |
5115 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5116 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5117 | |
5118 | sched_clock_tick(); | |
dd41f596 IM |
5119 | |
5120 | spin_lock(&rq->lock); | |
3e51f33f | 5121 | update_rq_clock(rq); |
f1a438d8 | 5122 | update_cpu_load(rq); |
fa85ae24 | 5123 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5124 | spin_unlock(&rq->lock); |
7835b98b | 5125 | |
e220d2dc PZ |
5126 | perf_counter_task_tick(curr, cpu); |
5127 | ||
e418e1c2 | 5128 | #ifdef CONFIG_SMP |
dd41f596 IM |
5129 | rq->idle_at_tick = idle_cpu(cpu); |
5130 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5131 | #endif |
1da177e4 LT |
5132 | } |
5133 | ||
132380a0 | 5134 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5135 | { |
5136 | if (in_lock_functions(addr)) { | |
5137 | addr = CALLER_ADDR2; | |
5138 | if (in_lock_functions(addr)) | |
5139 | addr = CALLER_ADDR3; | |
5140 | } | |
5141 | return addr; | |
5142 | } | |
1da177e4 | 5143 | |
7e49fcce SR |
5144 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5145 | defined(CONFIG_PREEMPT_TRACER)) | |
5146 | ||
43627582 | 5147 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5148 | { |
6cd8a4bb | 5149 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5150 | /* |
5151 | * Underflow? | |
5152 | */ | |
9a11b49a IM |
5153 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5154 | return; | |
6cd8a4bb | 5155 | #endif |
1da177e4 | 5156 | preempt_count() += val; |
6cd8a4bb | 5157 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5158 | /* |
5159 | * Spinlock count overflowing soon? | |
5160 | */ | |
33859f7f MOS |
5161 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5162 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5163 | #endif |
5164 | if (preempt_count() == val) | |
5165 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5166 | } |
5167 | EXPORT_SYMBOL(add_preempt_count); | |
5168 | ||
43627582 | 5169 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5170 | { |
6cd8a4bb | 5171 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5172 | /* |
5173 | * Underflow? | |
5174 | */ | |
01e3eb82 | 5175 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5176 | return; |
1da177e4 LT |
5177 | /* |
5178 | * Is the spinlock portion underflowing? | |
5179 | */ | |
9a11b49a IM |
5180 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5181 | !(preempt_count() & PREEMPT_MASK))) | |
5182 | return; | |
6cd8a4bb | 5183 | #endif |
9a11b49a | 5184 | |
6cd8a4bb SR |
5185 | if (preempt_count() == val) |
5186 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5187 | preempt_count() -= val; |
5188 | } | |
5189 | EXPORT_SYMBOL(sub_preempt_count); | |
5190 | ||
5191 | #endif | |
5192 | ||
5193 | /* | |
dd41f596 | 5194 | * Print scheduling while atomic bug: |
1da177e4 | 5195 | */ |
dd41f596 | 5196 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5197 | { |
838225b4 SS |
5198 | struct pt_regs *regs = get_irq_regs(); |
5199 | ||
5200 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5201 | prev->comm, prev->pid, preempt_count()); | |
5202 | ||
dd41f596 | 5203 | debug_show_held_locks(prev); |
e21f5b15 | 5204 | print_modules(); |
dd41f596 IM |
5205 | if (irqs_disabled()) |
5206 | print_irqtrace_events(prev); | |
838225b4 SS |
5207 | |
5208 | if (regs) | |
5209 | show_regs(regs); | |
5210 | else | |
5211 | dump_stack(); | |
dd41f596 | 5212 | } |
1da177e4 | 5213 | |
dd41f596 IM |
5214 | /* |
5215 | * Various schedule()-time debugging checks and statistics: | |
5216 | */ | |
5217 | static inline void schedule_debug(struct task_struct *prev) | |
5218 | { | |
1da177e4 | 5219 | /* |
41a2d6cf | 5220 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5221 | * schedule() atomically, we ignore that path for now. |
5222 | * Otherwise, whine if we are scheduling when we should not be. | |
5223 | */ | |
3f33a7ce | 5224 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5225 | __schedule_bug(prev); |
5226 | ||
1da177e4 LT |
5227 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5228 | ||
2d72376b | 5229 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5230 | #ifdef CONFIG_SCHEDSTATS |
5231 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5232 | schedstat_inc(this_rq(), bkl_count); |
5233 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5234 | } |
5235 | #endif | |
dd41f596 IM |
5236 | } |
5237 | ||
df1c99d4 MG |
5238 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5239 | { | |
5240 | if (prev->state == TASK_RUNNING) { | |
5241 | u64 runtime = prev->se.sum_exec_runtime; | |
5242 | ||
5243 | runtime -= prev->se.prev_sum_exec_runtime; | |
5244 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5245 | ||
5246 | /* | |
5247 | * In order to avoid avg_overlap growing stale when we are | |
5248 | * indeed overlapping and hence not getting put to sleep, grow | |
5249 | * the avg_overlap on preemption. | |
5250 | * | |
5251 | * We use the average preemption runtime because that | |
5252 | * correlates to the amount of cache footprint a task can | |
5253 | * build up. | |
5254 | */ | |
5255 | update_avg(&prev->se.avg_overlap, runtime); | |
5256 | } | |
5257 | prev->sched_class->put_prev_task(rq, prev); | |
5258 | } | |
5259 | ||
dd41f596 IM |
5260 | /* |
5261 | * Pick up the highest-prio task: | |
5262 | */ | |
5263 | static inline struct task_struct * | |
b67802ea | 5264 | pick_next_task(struct rq *rq) |
dd41f596 | 5265 | { |
5522d5d5 | 5266 | const struct sched_class *class; |
dd41f596 | 5267 | struct task_struct *p; |
1da177e4 LT |
5268 | |
5269 | /* | |
dd41f596 IM |
5270 | * Optimization: we know that if all tasks are in |
5271 | * the fair class we can call that function directly: | |
1da177e4 | 5272 | */ |
dd41f596 | 5273 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5274 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5275 | if (likely(p)) |
5276 | return p; | |
1da177e4 LT |
5277 | } |
5278 | ||
dd41f596 IM |
5279 | class = sched_class_highest; |
5280 | for ( ; ; ) { | |
fb8d4724 | 5281 | p = class->pick_next_task(rq); |
dd41f596 IM |
5282 | if (p) |
5283 | return p; | |
5284 | /* | |
5285 | * Will never be NULL as the idle class always | |
5286 | * returns a non-NULL p: | |
5287 | */ | |
5288 | class = class->next; | |
5289 | } | |
5290 | } | |
1da177e4 | 5291 | |
dd41f596 IM |
5292 | /* |
5293 | * schedule() is the main scheduler function. | |
5294 | */ | |
ff743345 | 5295 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5296 | { |
5297 | struct task_struct *prev, *next; | |
67ca7bde | 5298 | unsigned long *switch_count; |
dd41f596 | 5299 | struct rq *rq; |
31656519 | 5300 | int cpu; |
dd41f596 | 5301 | |
ff743345 PZ |
5302 | need_resched: |
5303 | preempt_disable(); | |
dd41f596 IM |
5304 | cpu = smp_processor_id(); |
5305 | rq = cpu_rq(cpu); | |
5306 | rcu_qsctr_inc(cpu); | |
5307 | prev = rq->curr; | |
5308 | switch_count = &prev->nivcsw; | |
5309 | ||
5310 | release_kernel_lock(prev); | |
5311 | need_resched_nonpreemptible: | |
5312 | ||
5313 | schedule_debug(prev); | |
1da177e4 | 5314 | |
31656519 | 5315 | if (sched_feat(HRTICK)) |
f333fdc9 | 5316 | hrtick_clear(rq); |
8f4d37ec | 5317 | |
8cd162ce | 5318 | spin_lock_irq(&rq->lock); |
3e51f33f | 5319 | update_rq_clock(rq); |
1e819950 | 5320 | clear_tsk_need_resched(prev); |
1da177e4 | 5321 | |
1da177e4 | 5322 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5323 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5324 | prev->state = TASK_RUNNING; |
16882c1e | 5325 | else |
2e1cb74a | 5326 | deactivate_task(rq, prev, 1); |
dd41f596 | 5327 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5328 | } |
5329 | ||
9a897c5a SR |
5330 | #ifdef CONFIG_SMP |
5331 | if (prev->sched_class->pre_schedule) | |
5332 | prev->sched_class->pre_schedule(rq, prev); | |
5333 | #endif | |
f65eda4f | 5334 | |
dd41f596 | 5335 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5336 | idle_balance(cpu, rq); |
1da177e4 | 5337 | |
df1c99d4 | 5338 | put_prev_task(rq, prev); |
b67802ea | 5339 | next = pick_next_task(rq); |
1da177e4 | 5340 | |
1da177e4 | 5341 | if (likely(prev != next)) { |
673a90a1 | 5342 | sched_info_switch(prev, next); |
564c2b21 | 5343 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5344 | |
1da177e4 LT |
5345 | rq->nr_switches++; |
5346 | rq->curr = next; | |
5347 | ++*switch_count; | |
5348 | ||
dd41f596 | 5349 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5350 | /* |
5351 | * the context switch might have flipped the stack from under | |
5352 | * us, hence refresh the local variables. | |
5353 | */ | |
5354 | cpu = smp_processor_id(); | |
5355 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5356 | } else |
5357 | spin_unlock_irq(&rq->lock); | |
5358 | ||
8f4d37ec | 5359 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5360 | goto need_resched_nonpreemptible; |
8f4d37ec | 5361 | |
1da177e4 | 5362 | preempt_enable_no_resched(); |
ff743345 | 5363 | if (need_resched()) |
1da177e4 LT |
5364 | goto need_resched; |
5365 | } | |
1da177e4 LT |
5366 | EXPORT_SYMBOL(schedule); |
5367 | ||
0d66bf6d PZ |
5368 | #ifdef CONFIG_SMP |
5369 | /* | |
5370 | * Look out! "owner" is an entirely speculative pointer | |
5371 | * access and not reliable. | |
5372 | */ | |
5373 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5374 | { | |
5375 | unsigned int cpu; | |
5376 | struct rq *rq; | |
5377 | ||
5378 | if (!sched_feat(OWNER_SPIN)) | |
5379 | return 0; | |
5380 | ||
5381 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5382 | /* | |
5383 | * Need to access the cpu field knowing that | |
5384 | * DEBUG_PAGEALLOC could have unmapped it if | |
5385 | * the mutex owner just released it and exited. | |
5386 | */ | |
5387 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5388 | goto out; | |
5389 | #else | |
5390 | cpu = owner->cpu; | |
5391 | #endif | |
5392 | ||
5393 | /* | |
5394 | * Even if the access succeeded (likely case), | |
5395 | * the cpu field may no longer be valid. | |
5396 | */ | |
5397 | if (cpu >= nr_cpumask_bits) | |
5398 | goto out; | |
5399 | ||
5400 | /* | |
5401 | * We need to validate that we can do a | |
5402 | * get_cpu() and that we have the percpu area. | |
5403 | */ | |
5404 | if (!cpu_online(cpu)) | |
5405 | goto out; | |
5406 | ||
5407 | rq = cpu_rq(cpu); | |
5408 | ||
5409 | for (;;) { | |
5410 | /* | |
5411 | * Owner changed, break to re-assess state. | |
5412 | */ | |
5413 | if (lock->owner != owner) | |
5414 | break; | |
5415 | ||
5416 | /* | |
5417 | * Is that owner really running on that cpu? | |
5418 | */ | |
5419 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5420 | return 0; | |
5421 | ||
5422 | cpu_relax(); | |
5423 | } | |
5424 | out: | |
5425 | return 1; | |
5426 | } | |
5427 | #endif | |
5428 | ||
1da177e4 LT |
5429 | #ifdef CONFIG_PREEMPT |
5430 | /* | |
2ed6e34f | 5431 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5432 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5433 | * occur there and call schedule directly. |
5434 | */ | |
5435 | asmlinkage void __sched preempt_schedule(void) | |
5436 | { | |
5437 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5438 | |
1da177e4 LT |
5439 | /* |
5440 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5441 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5442 | */ |
beed33a8 | 5443 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5444 | return; |
5445 | ||
3a5c359a AK |
5446 | do { |
5447 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5448 | schedule(); |
3a5c359a | 5449 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5450 | |
3a5c359a AK |
5451 | /* |
5452 | * Check again in case we missed a preemption opportunity | |
5453 | * between schedule and now. | |
5454 | */ | |
5455 | barrier(); | |
5ed0cec0 | 5456 | } while (need_resched()); |
1da177e4 | 5457 | } |
1da177e4 LT |
5458 | EXPORT_SYMBOL(preempt_schedule); |
5459 | ||
5460 | /* | |
2ed6e34f | 5461 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5462 | * off of irq context. |
5463 | * Note, that this is called and return with irqs disabled. This will | |
5464 | * protect us against recursive calling from irq. | |
5465 | */ | |
5466 | asmlinkage void __sched preempt_schedule_irq(void) | |
5467 | { | |
5468 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5469 | |
2ed6e34f | 5470 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5471 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5472 | ||
3a5c359a AK |
5473 | do { |
5474 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5475 | local_irq_enable(); |
5476 | schedule(); | |
5477 | local_irq_disable(); | |
3a5c359a | 5478 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5479 | |
3a5c359a AK |
5480 | /* |
5481 | * Check again in case we missed a preemption opportunity | |
5482 | * between schedule and now. | |
5483 | */ | |
5484 | barrier(); | |
5ed0cec0 | 5485 | } while (need_resched()); |
1da177e4 LT |
5486 | } |
5487 | ||
5488 | #endif /* CONFIG_PREEMPT */ | |
5489 | ||
95cdf3b7 IM |
5490 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5491 | void *key) | |
1da177e4 | 5492 | { |
48f24c4d | 5493 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5494 | } |
1da177e4 LT |
5495 | EXPORT_SYMBOL(default_wake_function); |
5496 | ||
5497 | /* | |
41a2d6cf IM |
5498 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5499 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5500 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5501 | * | |
5502 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5503 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5504 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5505 | */ | |
78ddb08f | 5506 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5507 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5508 | { |
2e45874c | 5509 | wait_queue_t *curr, *next; |
1da177e4 | 5510 | |
2e45874c | 5511 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5512 | unsigned flags = curr->flags; |
5513 | ||
1da177e4 | 5514 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5515 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5516 | break; |
5517 | } | |
5518 | } | |
5519 | ||
5520 | /** | |
5521 | * __wake_up - wake up threads blocked on a waitqueue. | |
5522 | * @q: the waitqueue | |
5523 | * @mode: which threads | |
5524 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5525 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5526 | * |
5527 | * It may be assumed that this function implies a write memory barrier before | |
5528 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5529 | */ |
7ad5b3a5 | 5530 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5531 | int nr_exclusive, void *key) |
1da177e4 LT |
5532 | { |
5533 | unsigned long flags; | |
5534 | ||
5535 | spin_lock_irqsave(&q->lock, flags); | |
5536 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5537 | spin_unlock_irqrestore(&q->lock, flags); | |
5538 | } | |
1da177e4 LT |
5539 | EXPORT_SYMBOL(__wake_up); |
5540 | ||
5541 | /* | |
5542 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5543 | */ | |
7ad5b3a5 | 5544 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5545 | { |
5546 | __wake_up_common(q, mode, 1, 0, NULL); | |
5547 | } | |
5548 | ||
4ede816a DL |
5549 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5550 | { | |
5551 | __wake_up_common(q, mode, 1, 0, key); | |
5552 | } | |
5553 | ||
1da177e4 | 5554 | /** |
4ede816a | 5555 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5556 | * @q: the waitqueue |
5557 | * @mode: which threads | |
5558 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5559 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5560 | * |
5561 | * The sync wakeup differs that the waker knows that it will schedule | |
5562 | * away soon, so while the target thread will be woken up, it will not | |
5563 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5564 | * with each other. This can prevent needless bouncing between CPUs. | |
5565 | * | |
5566 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5567 | * |
5568 | * It may be assumed that this function implies a write memory barrier before | |
5569 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5570 | */ |
4ede816a DL |
5571 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5572 | int nr_exclusive, void *key) | |
1da177e4 LT |
5573 | { |
5574 | unsigned long flags; | |
5575 | int sync = 1; | |
5576 | ||
5577 | if (unlikely(!q)) | |
5578 | return; | |
5579 | ||
5580 | if (unlikely(!nr_exclusive)) | |
5581 | sync = 0; | |
5582 | ||
5583 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5584 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5585 | spin_unlock_irqrestore(&q->lock, flags); |
5586 | } | |
4ede816a DL |
5587 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5588 | ||
5589 | /* | |
5590 | * __wake_up_sync - see __wake_up_sync_key() | |
5591 | */ | |
5592 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5593 | { | |
5594 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5595 | } | |
1da177e4 LT |
5596 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5597 | ||
65eb3dc6 KD |
5598 | /** |
5599 | * complete: - signals a single thread waiting on this completion | |
5600 | * @x: holds the state of this particular completion | |
5601 | * | |
5602 | * This will wake up a single thread waiting on this completion. Threads will be | |
5603 | * awakened in the same order in which they were queued. | |
5604 | * | |
5605 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5606 | * |
5607 | * It may be assumed that this function implies a write memory barrier before | |
5608 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5609 | */ |
b15136e9 | 5610 | void complete(struct completion *x) |
1da177e4 LT |
5611 | { |
5612 | unsigned long flags; | |
5613 | ||
5614 | spin_lock_irqsave(&x->wait.lock, flags); | |
5615 | x->done++; | |
d9514f6c | 5616 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5617 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5618 | } | |
5619 | EXPORT_SYMBOL(complete); | |
5620 | ||
65eb3dc6 KD |
5621 | /** |
5622 | * complete_all: - signals all threads waiting on this completion | |
5623 | * @x: holds the state of this particular completion | |
5624 | * | |
5625 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5626 | * |
5627 | * It may be assumed that this function implies a write memory barrier before | |
5628 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5629 | */ |
b15136e9 | 5630 | void complete_all(struct completion *x) |
1da177e4 LT |
5631 | { |
5632 | unsigned long flags; | |
5633 | ||
5634 | spin_lock_irqsave(&x->wait.lock, flags); | |
5635 | x->done += UINT_MAX/2; | |
d9514f6c | 5636 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5637 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5638 | } | |
5639 | EXPORT_SYMBOL(complete_all); | |
5640 | ||
8cbbe86d AK |
5641 | static inline long __sched |
5642 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5643 | { |
1da177e4 LT |
5644 | if (!x->done) { |
5645 | DECLARE_WAITQUEUE(wait, current); | |
5646 | ||
5647 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5648 | __add_wait_queue_tail(&x->wait, &wait); | |
5649 | do { | |
94d3d824 | 5650 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5651 | timeout = -ERESTARTSYS; |
5652 | break; | |
8cbbe86d AK |
5653 | } |
5654 | __set_current_state(state); | |
1da177e4 LT |
5655 | spin_unlock_irq(&x->wait.lock); |
5656 | timeout = schedule_timeout(timeout); | |
5657 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5658 | } while (!x->done && timeout); |
1da177e4 | 5659 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5660 | if (!x->done) |
5661 | return timeout; | |
1da177e4 LT |
5662 | } |
5663 | x->done--; | |
ea71a546 | 5664 | return timeout ?: 1; |
1da177e4 | 5665 | } |
1da177e4 | 5666 | |
8cbbe86d AK |
5667 | static long __sched |
5668 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5669 | { |
1da177e4 LT |
5670 | might_sleep(); |
5671 | ||
5672 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5673 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5674 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5675 | return timeout; |
5676 | } | |
1da177e4 | 5677 | |
65eb3dc6 KD |
5678 | /** |
5679 | * wait_for_completion: - waits for completion of a task | |
5680 | * @x: holds the state of this particular completion | |
5681 | * | |
5682 | * This waits to be signaled for completion of a specific task. It is NOT | |
5683 | * interruptible and there is no timeout. | |
5684 | * | |
5685 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5686 | * and interrupt capability. Also see complete(). | |
5687 | */ | |
b15136e9 | 5688 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5689 | { |
5690 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5691 | } |
8cbbe86d | 5692 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5693 | |
65eb3dc6 KD |
5694 | /** |
5695 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5696 | * @x: holds the state of this particular completion | |
5697 | * @timeout: timeout value in jiffies | |
5698 | * | |
5699 | * This waits for either a completion of a specific task to be signaled or for a | |
5700 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5701 | * interruptible. | |
5702 | */ | |
b15136e9 | 5703 | unsigned long __sched |
8cbbe86d | 5704 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5705 | { |
8cbbe86d | 5706 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5707 | } |
8cbbe86d | 5708 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5709 | |
65eb3dc6 KD |
5710 | /** |
5711 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5712 | * @x: holds the state of this particular completion | |
5713 | * | |
5714 | * This waits for completion of a specific task to be signaled. It is | |
5715 | * interruptible. | |
5716 | */ | |
8cbbe86d | 5717 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5718 | { |
51e97990 AK |
5719 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5720 | if (t == -ERESTARTSYS) | |
5721 | return t; | |
5722 | return 0; | |
0fec171c | 5723 | } |
8cbbe86d | 5724 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5725 | |
65eb3dc6 KD |
5726 | /** |
5727 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5728 | * @x: holds the state of this particular completion | |
5729 | * @timeout: timeout value in jiffies | |
5730 | * | |
5731 | * This waits for either a completion of a specific task to be signaled or for a | |
5732 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5733 | */ | |
b15136e9 | 5734 | unsigned long __sched |
8cbbe86d AK |
5735 | wait_for_completion_interruptible_timeout(struct completion *x, |
5736 | unsigned long timeout) | |
0fec171c | 5737 | { |
8cbbe86d | 5738 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5739 | } |
8cbbe86d | 5740 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5741 | |
65eb3dc6 KD |
5742 | /** |
5743 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5744 | * @x: holds the state of this particular completion | |
5745 | * | |
5746 | * This waits to be signaled for completion of a specific task. It can be | |
5747 | * interrupted by a kill signal. | |
5748 | */ | |
009e577e MW |
5749 | int __sched wait_for_completion_killable(struct completion *x) |
5750 | { | |
5751 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5752 | if (t == -ERESTARTSYS) | |
5753 | return t; | |
5754 | return 0; | |
5755 | } | |
5756 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5757 | ||
be4de352 DC |
5758 | /** |
5759 | * try_wait_for_completion - try to decrement a completion without blocking | |
5760 | * @x: completion structure | |
5761 | * | |
5762 | * Returns: 0 if a decrement cannot be done without blocking | |
5763 | * 1 if a decrement succeeded. | |
5764 | * | |
5765 | * If a completion is being used as a counting completion, | |
5766 | * attempt to decrement the counter without blocking. This | |
5767 | * enables us to avoid waiting if the resource the completion | |
5768 | * is protecting is not available. | |
5769 | */ | |
5770 | bool try_wait_for_completion(struct completion *x) | |
5771 | { | |
5772 | int ret = 1; | |
5773 | ||
5774 | spin_lock_irq(&x->wait.lock); | |
5775 | if (!x->done) | |
5776 | ret = 0; | |
5777 | else | |
5778 | x->done--; | |
5779 | spin_unlock_irq(&x->wait.lock); | |
5780 | return ret; | |
5781 | } | |
5782 | EXPORT_SYMBOL(try_wait_for_completion); | |
5783 | ||
5784 | /** | |
5785 | * completion_done - Test to see if a completion has any waiters | |
5786 | * @x: completion structure | |
5787 | * | |
5788 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5789 | * 1 if there are no waiters. | |
5790 | * | |
5791 | */ | |
5792 | bool completion_done(struct completion *x) | |
5793 | { | |
5794 | int ret = 1; | |
5795 | ||
5796 | spin_lock_irq(&x->wait.lock); | |
5797 | if (!x->done) | |
5798 | ret = 0; | |
5799 | spin_unlock_irq(&x->wait.lock); | |
5800 | return ret; | |
5801 | } | |
5802 | EXPORT_SYMBOL(completion_done); | |
5803 | ||
8cbbe86d AK |
5804 | static long __sched |
5805 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5806 | { |
0fec171c IM |
5807 | unsigned long flags; |
5808 | wait_queue_t wait; | |
5809 | ||
5810 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5811 | |
8cbbe86d | 5812 | __set_current_state(state); |
1da177e4 | 5813 | |
8cbbe86d AK |
5814 | spin_lock_irqsave(&q->lock, flags); |
5815 | __add_wait_queue(q, &wait); | |
5816 | spin_unlock(&q->lock); | |
5817 | timeout = schedule_timeout(timeout); | |
5818 | spin_lock_irq(&q->lock); | |
5819 | __remove_wait_queue(q, &wait); | |
5820 | spin_unlock_irqrestore(&q->lock, flags); | |
5821 | ||
5822 | return timeout; | |
5823 | } | |
5824 | ||
5825 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5826 | { | |
5827 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5828 | } |
1da177e4 LT |
5829 | EXPORT_SYMBOL(interruptible_sleep_on); |
5830 | ||
0fec171c | 5831 | long __sched |
95cdf3b7 | 5832 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5833 | { |
8cbbe86d | 5834 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5835 | } |
1da177e4 LT |
5836 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5837 | ||
0fec171c | 5838 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5839 | { |
8cbbe86d | 5840 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5841 | } |
1da177e4 LT |
5842 | EXPORT_SYMBOL(sleep_on); |
5843 | ||
0fec171c | 5844 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5845 | { |
8cbbe86d | 5846 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5847 | } |
1da177e4 LT |
5848 | EXPORT_SYMBOL(sleep_on_timeout); |
5849 | ||
b29739f9 IM |
5850 | #ifdef CONFIG_RT_MUTEXES |
5851 | ||
5852 | /* | |
5853 | * rt_mutex_setprio - set the current priority of a task | |
5854 | * @p: task | |
5855 | * @prio: prio value (kernel-internal form) | |
5856 | * | |
5857 | * This function changes the 'effective' priority of a task. It does | |
5858 | * not touch ->normal_prio like __setscheduler(). | |
5859 | * | |
5860 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5861 | */ | |
36c8b586 | 5862 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5863 | { |
5864 | unsigned long flags; | |
83b699ed | 5865 | int oldprio, on_rq, running; |
70b97a7f | 5866 | struct rq *rq; |
cb469845 | 5867 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5868 | |
5869 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5870 | ||
5871 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5872 | update_rq_clock(rq); |
b29739f9 | 5873 | |
d5f9f942 | 5874 | oldprio = p->prio; |
dd41f596 | 5875 | on_rq = p->se.on_rq; |
051a1d1a | 5876 | running = task_current(rq, p); |
0e1f3483 | 5877 | if (on_rq) |
69be72c1 | 5878 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5879 | if (running) |
5880 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5881 | |
5882 | if (rt_prio(prio)) | |
5883 | p->sched_class = &rt_sched_class; | |
5884 | else | |
5885 | p->sched_class = &fair_sched_class; | |
5886 | ||
b29739f9 IM |
5887 | p->prio = prio; |
5888 | ||
0e1f3483 HS |
5889 | if (running) |
5890 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5891 | if (on_rq) { |
8159f87e | 5892 | enqueue_task(rq, p, 0); |
cb469845 SR |
5893 | |
5894 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5895 | } |
5896 | task_rq_unlock(rq, &flags); | |
5897 | } | |
5898 | ||
5899 | #endif | |
5900 | ||
36c8b586 | 5901 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5902 | { |
dd41f596 | 5903 | int old_prio, delta, on_rq; |
1da177e4 | 5904 | unsigned long flags; |
70b97a7f | 5905 | struct rq *rq; |
1da177e4 LT |
5906 | |
5907 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5908 | return; | |
5909 | /* | |
5910 | * We have to be careful, if called from sys_setpriority(), | |
5911 | * the task might be in the middle of scheduling on another CPU. | |
5912 | */ | |
5913 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5914 | update_rq_clock(rq); |
1da177e4 LT |
5915 | /* |
5916 | * The RT priorities are set via sched_setscheduler(), but we still | |
5917 | * allow the 'normal' nice value to be set - but as expected | |
5918 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5919 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5920 | */ |
e05606d3 | 5921 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5922 | p->static_prio = NICE_TO_PRIO(nice); |
5923 | goto out_unlock; | |
5924 | } | |
dd41f596 | 5925 | on_rq = p->se.on_rq; |
c09595f6 | 5926 | if (on_rq) |
69be72c1 | 5927 | dequeue_task(rq, p, 0); |
1da177e4 | 5928 | |
1da177e4 | 5929 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5930 | set_load_weight(p); |
b29739f9 IM |
5931 | old_prio = p->prio; |
5932 | p->prio = effective_prio(p); | |
5933 | delta = p->prio - old_prio; | |
1da177e4 | 5934 | |
dd41f596 | 5935 | if (on_rq) { |
8159f87e | 5936 | enqueue_task(rq, p, 0); |
1da177e4 | 5937 | /* |
d5f9f942 AM |
5938 | * If the task increased its priority or is running and |
5939 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5940 | */ |
d5f9f942 | 5941 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5942 | resched_task(rq->curr); |
5943 | } | |
5944 | out_unlock: | |
5945 | task_rq_unlock(rq, &flags); | |
5946 | } | |
1da177e4 LT |
5947 | EXPORT_SYMBOL(set_user_nice); |
5948 | ||
e43379f1 MM |
5949 | /* |
5950 | * can_nice - check if a task can reduce its nice value | |
5951 | * @p: task | |
5952 | * @nice: nice value | |
5953 | */ | |
36c8b586 | 5954 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5955 | { |
024f4747 MM |
5956 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5957 | int nice_rlim = 20 - nice; | |
48f24c4d | 5958 | |
e43379f1 MM |
5959 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5960 | capable(CAP_SYS_NICE)); | |
5961 | } | |
5962 | ||
1da177e4 LT |
5963 | #ifdef __ARCH_WANT_SYS_NICE |
5964 | ||
5965 | /* | |
5966 | * sys_nice - change the priority of the current process. | |
5967 | * @increment: priority increment | |
5968 | * | |
5969 | * sys_setpriority is a more generic, but much slower function that | |
5970 | * does similar things. | |
5971 | */ | |
5add95d4 | 5972 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5973 | { |
48f24c4d | 5974 | long nice, retval; |
1da177e4 LT |
5975 | |
5976 | /* | |
5977 | * Setpriority might change our priority at the same moment. | |
5978 | * We don't have to worry. Conceptually one call occurs first | |
5979 | * and we have a single winner. | |
5980 | */ | |
e43379f1 MM |
5981 | if (increment < -40) |
5982 | increment = -40; | |
1da177e4 LT |
5983 | if (increment > 40) |
5984 | increment = 40; | |
5985 | ||
2b8f836f | 5986 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5987 | if (nice < -20) |
5988 | nice = -20; | |
5989 | if (nice > 19) | |
5990 | nice = 19; | |
5991 | ||
e43379f1 MM |
5992 | if (increment < 0 && !can_nice(current, nice)) |
5993 | return -EPERM; | |
5994 | ||
1da177e4 LT |
5995 | retval = security_task_setnice(current, nice); |
5996 | if (retval) | |
5997 | return retval; | |
5998 | ||
5999 | set_user_nice(current, nice); | |
6000 | return 0; | |
6001 | } | |
6002 | ||
6003 | #endif | |
6004 | ||
6005 | /** | |
6006 | * task_prio - return the priority value of a given task. | |
6007 | * @p: the task in question. | |
6008 | * | |
6009 | * This is the priority value as seen by users in /proc. | |
6010 | * RT tasks are offset by -200. Normal tasks are centered | |
6011 | * around 0, value goes from -16 to +15. | |
6012 | */ | |
36c8b586 | 6013 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6014 | { |
6015 | return p->prio - MAX_RT_PRIO; | |
6016 | } | |
6017 | ||
6018 | /** | |
6019 | * task_nice - return the nice value of a given task. | |
6020 | * @p: the task in question. | |
6021 | */ | |
36c8b586 | 6022 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6023 | { |
6024 | return TASK_NICE(p); | |
6025 | } | |
150d8bed | 6026 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6027 | |
6028 | /** | |
6029 | * idle_cpu - is a given cpu idle currently? | |
6030 | * @cpu: the processor in question. | |
6031 | */ | |
6032 | int idle_cpu(int cpu) | |
6033 | { | |
6034 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6035 | } | |
6036 | ||
1da177e4 LT |
6037 | /** |
6038 | * idle_task - return the idle task for a given cpu. | |
6039 | * @cpu: the processor in question. | |
6040 | */ | |
36c8b586 | 6041 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6042 | { |
6043 | return cpu_rq(cpu)->idle; | |
6044 | } | |
6045 | ||
6046 | /** | |
6047 | * find_process_by_pid - find a process with a matching PID value. | |
6048 | * @pid: the pid in question. | |
6049 | */ | |
a9957449 | 6050 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6051 | { |
228ebcbe | 6052 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6053 | } |
6054 | ||
6055 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6056 | static void |
6057 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6058 | { |
dd41f596 | 6059 | BUG_ON(p->se.on_rq); |
48f24c4d | 6060 | |
1da177e4 | 6061 | p->policy = policy; |
dd41f596 IM |
6062 | switch (p->policy) { |
6063 | case SCHED_NORMAL: | |
6064 | case SCHED_BATCH: | |
6065 | case SCHED_IDLE: | |
6066 | p->sched_class = &fair_sched_class; | |
6067 | break; | |
6068 | case SCHED_FIFO: | |
6069 | case SCHED_RR: | |
6070 | p->sched_class = &rt_sched_class; | |
6071 | break; | |
6072 | } | |
6073 | ||
1da177e4 | 6074 | p->rt_priority = prio; |
b29739f9 IM |
6075 | p->normal_prio = normal_prio(p); |
6076 | /* we are holding p->pi_lock already */ | |
6077 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6078 | set_load_weight(p); |
1da177e4 LT |
6079 | } |
6080 | ||
c69e8d9c DH |
6081 | /* |
6082 | * check the target process has a UID that matches the current process's | |
6083 | */ | |
6084 | static bool check_same_owner(struct task_struct *p) | |
6085 | { | |
6086 | const struct cred *cred = current_cred(), *pcred; | |
6087 | bool match; | |
6088 | ||
6089 | rcu_read_lock(); | |
6090 | pcred = __task_cred(p); | |
6091 | match = (cred->euid == pcred->euid || | |
6092 | cred->euid == pcred->uid); | |
6093 | rcu_read_unlock(); | |
6094 | return match; | |
6095 | } | |
6096 | ||
961ccddd RR |
6097 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6098 | struct sched_param *param, bool user) | |
1da177e4 | 6099 | { |
83b699ed | 6100 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6101 | unsigned long flags; |
cb469845 | 6102 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6103 | struct rq *rq; |
1da177e4 | 6104 | |
66e5393a SR |
6105 | /* may grab non-irq protected spin_locks */ |
6106 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6107 | recheck: |
6108 | /* double check policy once rq lock held */ | |
6109 | if (policy < 0) | |
6110 | policy = oldpolicy = p->policy; | |
6111 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
6112 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
6113 | policy != SCHED_IDLE) | |
b0a9499c | 6114 | return -EINVAL; |
1da177e4 LT |
6115 | /* |
6116 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6117 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6118 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6119 | */ |
6120 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6121 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6122 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6123 | return -EINVAL; |
e05606d3 | 6124 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6125 | return -EINVAL; |
6126 | ||
37e4ab3f OC |
6127 | /* |
6128 | * Allow unprivileged RT tasks to decrease priority: | |
6129 | */ | |
961ccddd | 6130 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6131 | if (rt_policy(policy)) { |
8dc3e909 | 6132 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6133 | |
6134 | if (!lock_task_sighand(p, &flags)) | |
6135 | return -ESRCH; | |
6136 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6137 | unlock_task_sighand(p, &flags); | |
6138 | ||
6139 | /* can't set/change the rt policy */ | |
6140 | if (policy != p->policy && !rlim_rtprio) | |
6141 | return -EPERM; | |
6142 | ||
6143 | /* can't increase priority */ | |
6144 | if (param->sched_priority > p->rt_priority && | |
6145 | param->sched_priority > rlim_rtprio) | |
6146 | return -EPERM; | |
6147 | } | |
dd41f596 IM |
6148 | /* |
6149 | * Like positive nice levels, dont allow tasks to | |
6150 | * move out of SCHED_IDLE either: | |
6151 | */ | |
6152 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6153 | return -EPERM; | |
5fe1d75f | 6154 | |
37e4ab3f | 6155 | /* can't change other user's priorities */ |
c69e8d9c | 6156 | if (!check_same_owner(p)) |
37e4ab3f OC |
6157 | return -EPERM; |
6158 | } | |
1da177e4 | 6159 | |
725aad24 | 6160 | if (user) { |
b68aa230 | 6161 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6162 | /* |
6163 | * Do not allow realtime tasks into groups that have no runtime | |
6164 | * assigned. | |
6165 | */ | |
9a7e0b18 PZ |
6166 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6167 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6168 | return -EPERM; |
b68aa230 PZ |
6169 | #endif |
6170 | ||
725aad24 JF |
6171 | retval = security_task_setscheduler(p, policy, param); |
6172 | if (retval) | |
6173 | return retval; | |
6174 | } | |
6175 | ||
b29739f9 IM |
6176 | /* |
6177 | * make sure no PI-waiters arrive (or leave) while we are | |
6178 | * changing the priority of the task: | |
6179 | */ | |
6180 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6181 | /* |
6182 | * To be able to change p->policy safely, the apropriate | |
6183 | * runqueue lock must be held. | |
6184 | */ | |
b29739f9 | 6185 | rq = __task_rq_lock(p); |
1da177e4 LT |
6186 | /* recheck policy now with rq lock held */ |
6187 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6188 | policy = oldpolicy = -1; | |
b29739f9 IM |
6189 | __task_rq_unlock(rq); |
6190 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6191 | goto recheck; |
6192 | } | |
2daa3577 | 6193 | update_rq_clock(rq); |
dd41f596 | 6194 | on_rq = p->se.on_rq; |
051a1d1a | 6195 | running = task_current(rq, p); |
0e1f3483 | 6196 | if (on_rq) |
2e1cb74a | 6197 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6198 | if (running) |
6199 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6200 | |
1da177e4 | 6201 | oldprio = p->prio; |
dd41f596 | 6202 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6203 | |
0e1f3483 HS |
6204 | if (running) |
6205 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6206 | if (on_rq) { |
6207 | activate_task(rq, p, 0); | |
cb469845 SR |
6208 | |
6209 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6210 | } |
b29739f9 IM |
6211 | __task_rq_unlock(rq); |
6212 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6213 | ||
95e02ca9 TG |
6214 | rt_mutex_adjust_pi(p); |
6215 | ||
1da177e4 LT |
6216 | return 0; |
6217 | } | |
961ccddd RR |
6218 | |
6219 | /** | |
6220 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6221 | * @p: the task in question. | |
6222 | * @policy: new policy. | |
6223 | * @param: structure containing the new RT priority. | |
6224 | * | |
6225 | * NOTE that the task may be already dead. | |
6226 | */ | |
6227 | int sched_setscheduler(struct task_struct *p, int policy, | |
6228 | struct sched_param *param) | |
6229 | { | |
6230 | return __sched_setscheduler(p, policy, param, true); | |
6231 | } | |
1da177e4 LT |
6232 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6233 | ||
961ccddd RR |
6234 | /** |
6235 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6236 | * @p: the task in question. | |
6237 | * @policy: new policy. | |
6238 | * @param: structure containing the new RT priority. | |
6239 | * | |
6240 | * Just like sched_setscheduler, only don't bother checking if the | |
6241 | * current context has permission. For example, this is needed in | |
6242 | * stop_machine(): we create temporary high priority worker threads, | |
6243 | * but our caller might not have that capability. | |
6244 | */ | |
6245 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6246 | struct sched_param *param) | |
6247 | { | |
6248 | return __sched_setscheduler(p, policy, param, false); | |
6249 | } | |
6250 | ||
95cdf3b7 IM |
6251 | static int |
6252 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6253 | { |
1da177e4 LT |
6254 | struct sched_param lparam; |
6255 | struct task_struct *p; | |
36c8b586 | 6256 | int retval; |
1da177e4 LT |
6257 | |
6258 | if (!param || pid < 0) | |
6259 | return -EINVAL; | |
6260 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6261 | return -EFAULT; | |
5fe1d75f ON |
6262 | |
6263 | rcu_read_lock(); | |
6264 | retval = -ESRCH; | |
1da177e4 | 6265 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6266 | if (p != NULL) |
6267 | retval = sched_setscheduler(p, policy, &lparam); | |
6268 | rcu_read_unlock(); | |
36c8b586 | 6269 | |
1da177e4 LT |
6270 | return retval; |
6271 | } | |
6272 | ||
6273 | /** | |
6274 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6275 | * @pid: the pid in question. | |
6276 | * @policy: new policy. | |
6277 | * @param: structure containing the new RT priority. | |
6278 | */ | |
5add95d4 HC |
6279 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6280 | struct sched_param __user *, param) | |
1da177e4 | 6281 | { |
c21761f1 JB |
6282 | /* negative values for policy are not valid */ |
6283 | if (policy < 0) | |
6284 | return -EINVAL; | |
6285 | ||
1da177e4 LT |
6286 | return do_sched_setscheduler(pid, policy, param); |
6287 | } | |
6288 | ||
6289 | /** | |
6290 | * sys_sched_setparam - set/change the RT priority of a thread | |
6291 | * @pid: the pid in question. | |
6292 | * @param: structure containing the new RT priority. | |
6293 | */ | |
5add95d4 | 6294 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6295 | { |
6296 | return do_sched_setscheduler(pid, -1, param); | |
6297 | } | |
6298 | ||
6299 | /** | |
6300 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6301 | * @pid: the pid in question. | |
6302 | */ | |
5add95d4 | 6303 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6304 | { |
36c8b586 | 6305 | struct task_struct *p; |
3a5c359a | 6306 | int retval; |
1da177e4 LT |
6307 | |
6308 | if (pid < 0) | |
3a5c359a | 6309 | return -EINVAL; |
1da177e4 LT |
6310 | |
6311 | retval = -ESRCH; | |
6312 | read_lock(&tasklist_lock); | |
6313 | p = find_process_by_pid(pid); | |
6314 | if (p) { | |
6315 | retval = security_task_getscheduler(p); | |
6316 | if (!retval) | |
6317 | retval = p->policy; | |
6318 | } | |
6319 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6320 | return retval; |
6321 | } | |
6322 | ||
6323 | /** | |
6324 | * sys_sched_getscheduler - get the RT priority of a thread | |
6325 | * @pid: the pid in question. | |
6326 | * @param: structure containing the RT priority. | |
6327 | */ | |
5add95d4 | 6328 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6329 | { |
6330 | struct sched_param lp; | |
36c8b586 | 6331 | struct task_struct *p; |
3a5c359a | 6332 | int retval; |
1da177e4 LT |
6333 | |
6334 | if (!param || pid < 0) | |
3a5c359a | 6335 | return -EINVAL; |
1da177e4 LT |
6336 | |
6337 | read_lock(&tasklist_lock); | |
6338 | p = find_process_by_pid(pid); | |
6339 | retval = -ESRCH; | |
6340 | if (!p) | |
6341 | goto out_unlock; | |
6342 | ||
6343 | retval = security_task_getscheduler(p); | |
6344 | if (retval) | |
6345 | goto out_unlock; | |
6346 | ||
6347 | lp.sched_priority = p->rt_priority; | |
6348 | read_unlock(&tasklist_lock); | |
6349 | ||
6350 | /* | |
6351 | * This one might sleep, we cannot do it with a spinlock held ... | |
6352 | */ | |
6353 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6354 | ||
1da177e4 LT |
6355 | return retval; |
6356 | ||
6357 | out_unlock: | |
6358 | read_unlock(&tasklist_lock); | |
6359 | return retval; | |
6360 | } | |
6361 | ||
96f874e2 | 6362 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6363 | { |
5a16f3d3 | 6364 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6365 | struct task_struct *p; |
6366 | int retval; | |
1da177e4 | 6367 | |
95402b38 | 6368 | get_online_cpus(); |
1da177e4 LT |
6369 | read_lock(&tasklist_lock); |
6370 | ||
6371 | p = find_process_by_pid(pid); | |
6372 | if (!p) { | |
6373 | read_unlock(&tasklist_lock); | |
95402b38 | 6374 | put_online_cpus(); |
1da177e4 LT |
6375 | return -ESRCH; |
6376 | } | |
6377 | ||
6378 | /* | |
6379 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6380 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6381 | * usage count and then drop tasklist_lock. |
6382 | */ | |
6383 | get_task_struct(p); | |
6384 | read_unlock(&tasklist_lock); | |
6385 | ||
5a16f3d3 RR |
6386 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6387 | retval = -ENOMEM; | |
6388 | goto out_put_task; | |
6389 | } | |
6390 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6391 | retval = -ENOMEM; | |
6392 | goto out_free_cpus_allowed; | |
6393 | } | |
1da177e4 | 6394 | retval = -EPERM; |
c69e8d9c | 6395 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6396 | goto out_unlock; |
6397 | ||
e7834f8f DQ |
6398 | retval = security_task_setscheduler(p, 0, NULL); |
6399 | if (retval) | |
6400 | goto out_unlock; | |
6401 | ||
5a16f3d3 RR |
6402 | cpuset_cpus_allowed(p, cpus_allowed); |
6403 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6404 | again: |
5a16f3d3 | 6405 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6406 | |
8707d8b8 | 6407 | if (!retval) { |
5a16f3d3 RR |
6408 | cpuset_cpus_allowed(p, cpus_allowed); |
6409 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6410 | /* |
6411 | * We must have raced with a concurrent cpuset | |
6412 | * update. Just reset the cpus_allowed to the | |
6413 | * cpuset's cpus_allowed | |
6414 | */ | |
5a16f3d3 | 6415 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6416 | goto again; |
6417 | } | |
6418 | } | |
1da177e4 | 6419 | out_unlock: |
5a16f3d3 RR |
6420 | free_cpumask_var(new_mask); |
6421 | out_free_cpus_allowed: | |
6422 | free_cpumask_var(cpus_allowed); | |
6423 | out_put_task: | |
1da177e4 | 6424 | put_task_struct(p); |
95402b38 | 6425 | put_online_cpus(); |
1da177e4 LT |
6426 | return retval; |
6427 | } | |
6428 | ||
6429 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6430 | struct cpumask *new_mask) |
1da177e4 | 6431 | { |
96f874e2 RR |
6432 | if (len < cpumask_size()) |
6433 | cpumask_clear(new_mask); | |
6434 | else if (len > cpumask_size()) | |
6435 | len = cpumask_size(); | |
6436 | ||
1da177e4 LT |
6437 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6438 | } | |
6439 | ||
6440 | /** | |
6441 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6442 | * @pid: pid of the process | |
6443 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6444 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6445 | */ | |
5add95d4 HC |
6446 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6447 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6448 | { |
5a16f3d3 | 6449 | cpumask_var_t new_mask; |
1da177e4 LT |
6450 | int retval; |
6451 | ||
5a16f3d3 RR |
6452 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6453 | return -ENOMEM; | |
1da177e4 | 6454 | |
5a16f3d3 RR |
6455 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6456 | if (retval == 0) | |
6457 | retval = sched_setaffinity(pid, new_mask); | |
6458 | free_cpumask_var(new_mask); | |
6459 | return retval; | |
1da177e4 LT |
6460 | } |
6461 | ||
96f874e2 | 6462 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6463 | { |
36c8b586 | 6464 | struct task_struct *p; |
1da177e4 | 6465 | int retval; |
1da177e4 | 6466 | |
95402b38 | 6467 | get_online_cpus(); |
1da177e4 LT |
6468 | read_lock(&tasklist_lock); |
6469 | ||
6470 | retval = -ESRCH; | |
6471 | p = find_process_by_pid(pid); | |
6472 | if (!p) | |
6473 | goto out_unlock; | |
6474 | ||
e7834f8f DQ |
6475 | retval = security_task_getscheduler(p); |
6476 | if (retval) | |
6477 | goto out_unlock; | |
6478 | ||
96f874e2 | 6479 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6480 | |
6481 | out_unlock: | |
6482 | read_unlock(&tasklist_lock); | |
95402b38 | 6483 | put_online_cpus(); |
1da177e4 | 6484 | |
9531b62f | 6485 | return retval; |
1da177e4 LT |
6486 | } |
6487 | ||
6488 | /** | |
6489 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6490 | * @pid: pid of the process | |
6491 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6492 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6493 | */ | |
5add95d4 HC |
6494 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6495 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6496 | { |
6497 | int ret; | |
f17c8607 | 6498 | cpumask_var_t mask; |
1da177e4 | 6499 | |
f17c8607 | 6500 | if (len < cpumask_size()) |
1da177e4 LT |
6501 | return -EINVAL; |
6502 | ||
f17c8607 RR |
6503 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6504 | return -ENOMEM; | |
1da177e4 | 6505 | |
f17c8607 RR |
6506 | ret = sched_getaffinity(pid, mask); |
6507 | if (ret == 0) { | |
6508 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6509 | ret = -EFAULT; | |
6510 | else | |
6511 | ret = cpumask_size(); | |
6512 | } | |
6513 | free_cpumask_var(mask); | |
1da177e4 | 6514 | |
f17c8607 | 6515 | return ret; |
1da177e4 LT |
6516 | } |
6517 | ||
6518 | /** | |
6519 | * sys_sched_yield - yield the current processor to other threads. | |
6520 | * | |
dd41f596 IM |
6521 | * This function yields the current CPU to other tasks. If there are no |
6522 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6523 | */ |
5add95d4 | 6524 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6525 | { |
70b97a7f | 6526 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6527 | |
2d72376b | 6528 | schedstat_inc(rq, yld_count); |
4530d7ab | 6529 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6530 | |
6531 | /* | |
6532 | * Since we are going to call schedule() anyway, there's | |
6533 | * no need to preempt or enable interrupts: | |
6534 | */ | |
6535 | __release(rq->lock); | |
8a25d5de | 6536 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6537 | _raw_spin_unlock(&rq->lock); |
6538 | preempt_enable_no_resched(); | |
6539 | ||
6540 | schedule(); | |
6541 | ||
6542 | return 0; | |
6543 | } | |
6544 | ||
e7b38404 | 6545 | static void __cond_resched(void) |
1da177e4 | 6546 | { |
8e0a43d8 IM |
6547 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6548 | __might_sleep(__FILE__, __LINE__); | |
6549 | #endif | |
5bbcfd90 IM |
6550 | /* |
6551 | * The BKS might be reacquired before we have dropped | |
6552 | * PREEMPT_ACTIVE, which could trigger a second | |
6553 | * cond_resched() call. | |
6554 | */ | |
1da177e4 LT |
6555 | do { |
6556 | add_preempt_count(PREEMPT_ACTIVE); | |
6557 | schedule(); | |
6558 | sub_preempt_count(PREEMPT_ACTIVE); | |
6559 | } while (need_resched()); | |
6560 | } | |
6561 | ||
02b67cc3 | 6562 | int __sched _cond_resched(void) |
1da177e4 | 6563 | { |
9414232f IM |
6564 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6565 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6566 | __cond_resched(); |
6567 | return 1; | |
6568 | } | |
6569 | return 0; | |
6570 | } | |
02b67cc3 | 6571 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6572 | |
6573 | /* | |
6574 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6575 | * call schedule, and on return reacquire the lock. | |
6576 | * | |
41a2d6cf | 6577 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6578 | * operations here to prevent schedule() from being called twice (once via |
6579 | * spin_unlock(), once by hand). | |
6580 | */ | |
95cdf3b7 | 6581 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6582 | { |
95c354fe | 6583 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6584 | int ret = 0; |
6585 | ||
95c354fe | 6586 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6587 | spin_unlock(lock); |
95c354fe NP |
6588 | if (resched && need_resched()) |
6589 | __cond_resched(); | |
6590 | else | |
6591 | cpu_relax(); | |
6df3cecb | 6592 | ret = 1; |
1da177e4 | 6593 | spin_lock(lock); |
1da177e4 | 6594 | } |
6df3cecb | 6595 | return ret; |
1da177e4 | 6596 | } |
1da177e4 LT |
6597 | EXPORT_SYMBOL(cond_resched_lock); |
6598 | ||
6599 | int __sched cond_resched_softirq(void) | |
6600 | { | |
6601 | BUG_ON(!in_softirq()); | |
6602 | ||
9414232f | 6603 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6604 | local_bh_enable(); |
1da177e4 LT |
6605 | __cond_resched(); |
6606 | local_bh_disable(); | |
6607 | return 1; | |
6608 | } | |
6609 | return 0; | |
6610 | } | |
1da177e4 LT |
6611 | EXPORT_SYMBOL(cond_resched_softirq); |
6612 | ||
1da177e4 LT |
6613 | /** |
6614 | * yield - yield the current processor to other threads. | |
6615 | * | |
72fd4a35 | 6616 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6617 | * thread runnable and calls sys_sched_yield(). |
6618 | */ | |
6619 | void __sched yield(void) | |
6620 | { | |
6621 | set_current_state(TASK_RUNNING); | |
6622 | sys_sched_yield(); | |
6623 | } | |
1da177e4 LT |
6624 | EXPORT_SYMBOL(yield); |
6625 | ||
6626 | /* | |
41a2d6cf | 6627 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6628 | * that process accounting knows that this is a task in IO wait state. |
6629 | * | |
6630 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6631 | * has set its backing_dev_info: the queue against which it should throttle) | |
6632 | */ | |
6633 | void __sched io_schedule(void) | |
6634 | { | |
70b97a7f | 6635 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6636 | |
0ff92245 | 6637 | delayacct_blkio_start(); |
1da177e4 LT |
6638 | atomic_inc(&rq->nr_iowait); |
6639 | schedule(); | |
6640 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6641 | delayacct_blkio_end(); |
1da177e4 | 6642 | } |
1da177e4 LT |
6643 | EXPORT_SYMBOL(io_schedule); |
6644 | ||
6645 | long __sched io_schedule_timeout(long timeout) | |
6646 | { | |
70b97a7f | 6647 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6648 | long ret; |
6649 | ||
0ff92245 | 6650 | delayacct_blkio_start(); |
1da177e4 LT |
6651 | atomic_inc(&rq->nr_iowait); |
6652 | ret = schedule_timeout(timeout); | |
6653 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6654 | delayacct_blkio_end(); |
1da177e4 LT |
6655 | return ret; |
6656 | } | |
6657 | ||
6658 | /** | |
6659 | * sys_sched_get_priority_max - return maximum RT priority. | |
6660 | * @policy: scheduling class. | |
6661 | * | |
6662 | * this syscall returns the maximum rt_priority that can be used | |
6663 | * by a given scheduling class. | |
6664 | */ | |
5add95d4 | 6665 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6666 | { |
6667 | int ret = -EINVAL; | |
6668 | ||
6669 | switch (policy) { | |
6670 | case SCHED_FIFO: | |
6671 | case SCHED_RR: | |
6672 | ret = MAX_USER_RT_PRIO-1; | |
6673 | break; | |
6674 | case SCHED_NORMAL: | |
b0a9499c | 6675 | case SCHED_BATCH: |
dd41f596 | 6676 | case SCHED_IDLE: |
1da177e4 LT |
6677 | ret = 0; |
6678 | break; | |
6679 | } | |
6680 | return ret; | |
6681 | } | |
6682 | ||
6683 | /** | |
6684 | * sys_sched_get_priority_min - return minimum RT priority. | |
6685 | * @policy: scheduling class. | |
6686 | * | |
6687 | * this syscall returns the minimum rt_priority that can be used | |
6688 | * by a given scheduling class. | |
6689 | */ | |
5add95d4 | 6690 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6691 | { |
6692 | int ret = -EINVAL; | |
6693 | ||
6694 | switch (policy) { | |
6695 | case SCHED_FIFO: | |
6696 | case SCHED_RR: | |
6697 | ret = 1; | |
6698 | break; | |
6699 | case SCHED_NORMAL: | |
b0a9499c | 6700 | case SCHED_BATCH: |
dd41f596 | 6701 | case SCHED_IDLE: |
1da177e4 LT |
6702 | ret = 0; |
6703 | } | |
6704 | return ret; | |
6705 | } | |
6706 | ||
6707 | /** | |
6708 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6709 | * @pid: pid of the process. | |
6710 | * @interval: userspace pointer to the timeslice value. | |
6711 | * | |
6712 | * this syscall writes the default timeslice value of a given process | |
6713 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6714 | */ | |
17da2bd9 | 6715 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6716 | struct timespec __user *, interval) |
1da177e4 | 6717 | { |
36c8b586 | 6718 | struct task_struct *p; |
a4ec24b4 | 6719 | unsigned int time_slice; |
3a5c359a | 6720 | int retval; |
1da177e4 | 6721 | struct timespec t; |
1da177e4 LT |
6722 | |
6723 | if (pid < 0) | |
3a5c359a | 6724 | return -EINVAL; |
1da177e4 LT |
6725 | |
6726 | retval = -ESRCH; | |
6727 | read_lock(&tasklist_lock); | |
6728 | p = find_process_by_pid(pid); | |
6729 | if (!p) | |
6730 | goto out_unlock; | |
6731 | ||
6732 | retval = security_task_getscheduler(p); | |
6733 | if (retval) | |
6734 | goto out_unlock; | |
6735 | ||
77034937 IM |
6736 | /* |
6737 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6738 | * tasks that are on an otherwise idle runqueue: | |
6739 | */ | |
6740 | time_slice = 0; | |
6741 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6742 | time_slice = DEF_TIMESLICE; |
1868f958 | 6743 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6744 | struct sched_entity *se = &p->se; |
6745 | unsigned long flags; | |
6746 | struct rq *rq; | |
6747 | ||
6748 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6749 | if (rq->cfs.load.weight) |
6750 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6751 | task_rq_unlock(rq, &flags); |
6752 | } | |
1da177e4 | 6753 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6754 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6755 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6756 | return retval; |
3a5c359a | 6757 | |
1da177e4 LT |
6758 | out_unlock: |
6759 | read_unlock(&tasklist_lock); | |
6760 | return retval; | |
6761 | } | |
6762 | ||
7c731e0a | 6763 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6764 | |
82a1fcb9 | 6765 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6766 | { |
1da177e4 | 6767 | unsigned long free = 0; |
36c8b586 | 6768 | unsigned state; |
1da177e4 | 6769 | |
1da177e4 | 6770 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6771 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6772 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6773 | #if BITS_PER_LONG == 32 |
1da177e4 | 6774 | if (state == TASK_RUNNING) |
cc4ea795 | 6775 | printk(KERN_CONT " running "); |
1da177e4 | 6776 | else |
cc4ea795 | 6777 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6778 | #else |
6779 | if (state == TASK_RUNNING) | |
cc4ea795 | 6780 | printk(KERN_CONT " running task "); |
1da177e4 | 6781 | else |
cc4ea795 | 6782 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6783 | #endif |
6784 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6785 | free = stack_not_used(p); |
1da177e4 | 6786 | #endif |
aa47b7e0 DR |
6787 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6788 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6789 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6790 | |
5fb5e6de | 6791 | show_stack(p, NULL); |
1da177e4 LT |
6792 | } |
6793 | ||
e59e2ae2 | 6794 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6795 | { |
36c8b586 | 6796 | struct task_struct *g, *p; |
1da177e4 | 6797 | |
4bd77321 IM |
6798 | #if BITS_PER_LONG == 32 |
6799 | printk(KERN_INFO | |
6800 | " task PC stack pid father\n"); | |
1da177e4 | 6801 | #else |
4bd77321 IM |
6802 | printk(KERN_INFO |
6803 | " task PC stack pid father\n"); | |
1da177e4 LT |
6804 | #endif |
6805 | read_lock(&tasklist_lock); | |
6806 | do_each_thread(g, p) { | |
6807 | /* | |
6808 | * reset the NMI-timeout, listing all files on a slow | |
6809 | * console might take alot of time: | |
6810 | */ | |
6811 | touch_nmi_watchdog(); | |
39bc89fd | 6812 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6813 | sched_show_task(p); |
1da177e4 LT |
6814 | } while_each_thread(g, p); |
6815 | ||
04c9167f JF |
6816 | touch_all_softlockup_watchdogs(); |
6817 | ||
dd41f596 IM |
6818 | #ifdef CONFIG_SCHED_DEBUG |
6819 | sysrq_sched_debug_show(); | |
6820 | #endif | |
1da177e4 | 6821 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6822 | /* |
6823 | * Only show locks if all tasks are dumped: | |
6824 | */ | |
6825 | if (state_filter == -1) | |
6826 | debug_show_all_locks(); | |
1da177e4 LT |
6827 | } |
6828 | ||
1df21055 IM |
6829 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6830 | { | |
dd41f596 | 6831 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6832 | } |
6833 | ||
f340c0d1 IM |
6834 | /** |
6835 | * init_idle - set up an idle thread for a given CPU | |
6836 | * @idle: task in question | |
6837 | * @cpu: cpu the idle task belongs to | |
6838 | * | |
6839 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6840 | * flag, to make booting more robust. | |
6841 | */ | |
5c1e1767 | 6842 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6843 | { |
70b97a7f | 6844 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6845 | unsigned long flags; |
6846 | ||
5cbd54ef IM |
6847 | spin_lock_irqsave(&rq->lock, flags); |
6848 | ||
dd41f596 IM |
6849 | __sched_fork(idle); |
6850 | idle->se.exec_start = sched_clock(); | |
6851 | ||
b29739f9 | 6852 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6853 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6854 | __set_task_cpu(idle, cpu); |
1da177e4 | 6855 | |
1da177e4 | 6856 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6857 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6858 | idle->oncpu = 1; | |
6859 | #endif | |
1da177e4 LT |
6860 | spin_unlock_irqrestore(&rq->lock, flags); |
6861 | ||
6862 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6863 | #if defined(CONFIG_PREEMPT) |
6864 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6865 | #else | |
a1261f54 | 6866 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6867 | #endif |
dd41f596 IM |
6868 | /* |
6869 | * The idle tasks have their own, simple scheduling class: | |
6870 | */ | |
6871 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6872 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6873 | } |
6874 | ||
6875 | /* | |
6876 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6877 | * indicates which cpus entered this state. This is used | |
6878 | * in the rcu update to wait only for active cpus. For system | |
6879 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6880 | * always be CPU_BITS_NONE. |
1da177e4 | 6881 | */ |
6a7b3dc3 | 6882 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6883 | |
19978ca6 IM |
6884 | /* |
6885 | * Increase the granularity value when there are more CPUs, | |
6886 | * because with more CPUs the 'effective latency' as visible | |
6887 | * to users decreases. But the relationship is not linear, | |
6888 | * so pick a second-best guess by going with the log2 of the | |
6889 | * number of CPUs. | |
6890 | * | |
6891 | * This idea comes from the SD scheduler of Con Kolivas: | |
6892 | */ | |
6893 | static inline void sched_init_granularity(void) | |
6894 | { | |
6895 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6896 | const unsigned long limit = 200000000; | |
6897 | ||
6898 | sysctl_sched_min_granularity *= factor; | |
6899 | if (sysctl_sched_min_granularity > limit) | |
6900 | sysctl_sched_min_granularity = limit; | |
6901 | ||
6902 | sysctl_sched_latency *= factor; | |
6903 | if (sysctl_sched_latency > limit) | |
6904 | sysctl_sched_latency = limit; | |
6905 | ||
6906 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6907 | |
6908 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6909 | } |
6910 | ||
1da177e4 LT |
6911 | #ifdef CONFIG_SMP |
6912 | /* | |
6913 | * This is how migration works: | |
6914 | * | |
70b97a7f | 6915 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6916 | * runqueue and wake up that CPU's migration thread. |
6917 | * 2) we down() the locked semaphore => thread blocks. | |
6918 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6919 | * thread off the CPU) | |
6920 | * 4) it gets the migration request and checks whether the migrated | |
6921 | * task is still in the wrong runqueue. | |
6922 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6923 | * it and puts it into the right queue. | |
6924 | * 6) migration thread up()s the semaphore. | |
6925 | * 7) we wake up and the migration is done. | |
6926 | */ | |
6927 | ||
6928 | /* | |
6929 | * Change a given task's CPU affinity. Migrate the thread to a | |
6930 | * proper CPU and schedule it away if the CPU it's executing on | |
6931 | * is removed from the allowed bitmask. | |
6932 | * | |
6933 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6934 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6935 | * call is not atomic; no spinlocks may be held. |
6936 | */ | |
96f874e2 | 6937 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6938 | { |
70b97a7f | 6939 | struct migration_req req; |
1da177e4 | 6940 | unsigned long flags; |
70b97a7f | 6941 | struct rq *rq; |
48f24c4d | 6942 | int ret = 0; |
1da177e4 LT |
6943 | |
6944 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6945 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6946 | ret = -EINVAL; |
6947 | goto out; | |
6948 | } | |
6949 | ||
9985b0ba | 6950 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6951 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6952 | ret = -EINVAL; |
6953 | goto out; | |
6954 | } | |
6955 | ||
73fe6aae | 6956 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6957 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6958 | else { |
96f874e2 RR |
6959 | cpumask_copy(&p->cpus_allowed, new_mask); |
6960 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6961 | } |
6962 | ||
1da177e4 | 6963 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6964 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6965 | goto out; |
6966 | ||
1e5ce4f4 | 6967 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6968 | /* Need help from migration thread: drop lock and wait. */ |
6969 | task_rq_unlock(rq, &flags); | |
6970 | wake_up_process(rq->migration_thread); | |
6971 | wait_for_completion(&req.done); | |
6972 | tlb_migrate_finish(p->mm); | |
6973 | return 0; | |
6974 | } | |
6975 | out: | |
6976 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6977 | |
1da177e4 LT |
6978 | return ret; |
6979 | } | |
cd8ba7cd | 6980 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6981 | |
6982 | /* | |
41a2d6cf | 6983 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6984 | * this because either it can't run here any more (set_cpus_allowed() |
6985 | * away from this CPU, or CPU going down), or because we're | |
6986 | * attempting to rebalance this task on exec (sched_exec). | |
6987 | * | |
6988 | * So we race with normal scheduler movements, but that's OK, as long | |
6989 | * as the task is no longer on this CPU. | |
efc30814 KK |
6990 | * |
6991 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6992 | */ |
efc30814 | 6993 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6994 | { |
70b97a7f | 6995 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6996 | int ret = 0, on_rq; |
1da177e4 | 6997 | |
e761b772 | 6998 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6999 | return ret; |
1da177e4 LT |
7000 | |
7001 | rq_src = cpu_rq(src_cpu); | |
7002 | rq_dest = cpu_rq(dest_cpu); | |
7003 | ||
7004 | double_rq_lock(rq_src, rq_dest); | |
7005 | /* Already moved. */ | |
7006 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7007 | goto done; |
1da177e4 | 7008 | /* Affinity changed (again). */ |
96f874e2 | 7009 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7010 | goto fail; |
1da177e4 | 7011 | |
dd41f596 | 7012 | on_rq = p->se.on_rq; |
6e82a3be | 7013 | if (on_rq) |
2e1cb74a | 7014 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7015 | |
1da177e4 | 7016 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7017 | if (on_rq) { |
7018 | activate_task(rq_dest, p, 0); | |
15afe09b | 7019 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7020 | } |
b1e38734 | 7021 | done: |
efc30814 | 7022 | ret = 1; |
b1e38734 | 7023 | fail: |
1da177e4 | 7024 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7025 | return ret; |
1da177e4 LT |
7026 | } |
7027 | ||
7028 | /* | |
7029 | * migration_thread - this is a highprio system thread that performs | |
7030 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7031 | * another runqueue. | |
7032 | */ | |
95cdf3b7 | 7033 | static int migration_thread(void *data) |
1da177e4 | 7034 | { |
1da177e4 | 7035 | int cpu = (long)data; |
70b97a7f | 7036 | struct rq *rq; |
1da177e4 LT |
7037 | |
7038 | rq = cpu_rq(cpu); | |
7039 | BUG_ON(rq->migration_thread != current); | |
7040 | ||
7041 | set_current_state(TASK_INTERRUPTIBLE); | |
7042 | while (!kthread_should_stop()) { | |
70b97a7f | 7043 | struct migration_req *req; |
1da177e4 | 7044 | struct list_head *head; |
1da177e4 | 7045 | |
1da177e4 LT |
7046 | spin_lock_irq(&rq->lock); |
7047 | ||
7048 | if (cpu_is_offline(cpu)) { | |
7049 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7050 | break; |
1da177e4 LT |
7051 | } |
7052 | ||
7053 | if (rq->active_balance) { | |
7054 | active_load_balance(rq, cpu); | |
7055 | rq->active_balance = 0; | |
7056 | } | |
7057 | ||
7058 | head = &rq->migration_queue; | |
7059 | ||
7060 | if (list_empty(head)) { | |
7061 | spin_unlock_irq(&rq->lock); | |
7062 | schedule(); | |
7063 | set_current_state(TASK_INTERRUPTIBLE); | |
7064 | continue; | |
7065 | } | |
70b97a7f | 7066 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7067 | list_del_init(head->next); |
7068 | ||
674311d5 NP |
7069 | spin_unlock(&rq->lock); |
7070 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7071 | local_irq_enable(); | |
1da177e4 LT |
7072 | |
7073 | complete(&req->done); | |
7074 | } | |
7075 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7076 | |
1da177e4 LT |
7077 | return 0; |
7078 | } | |
7079 | ||
7080 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7081 | |
7082 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7083 | { | |
7084 | int ret; | |
7085 | ||
7086 | local_irq_disable(); | |
7087 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7088 | local_irq_enable(); | |
7089 | return ret; | |
7090 | } | |
7091 | ||
054b9108 | 7092 | /* |
3a4fa0a2 | 7093 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7094 | */ |
48f24c4d | 7095 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7096 | { |
70b97a7f | 7097 | int dest_cpu; |
6ca09dfc | 7098 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7099 | |
7100 | again: | |
7101 | /* Look for allowed, online CPU in same node. */ | |
7102 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7103 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7104 | goto move; | |
7105 | ||
7106 | /* Any allowed, online CPU? */ | |
7107 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7108 | if (dest_cpu < nr_cpu_ids) | |
7109 | goto move; | |
7110 | ||
7111 | /* No more Mr. Nice Guy. */ | |
7112 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7113 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7114 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7115 | |
e76bd8d9 RR |
7116 | /* |
7117 | * Don't tell them about moving exiting tasks or | |
7118 | * kernel threads (both mm NULL), since they never | |
7119 | * leave kernel. | |
7120 | */ | |
7121 | if (p->mm && printk_ratelimit()) { | |
7122 | printk(KERN_INFO "process %d (%s) no " | |
7123 | "longer affine to cpu%d\n", | |
7124 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7125 | } |
e76bd8d9 RR |
7126 | } |
7127 | ||
7128 | move: | |
7129 | /* It can have affinity changed while we were choosing. */ | |
7130 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7131 | goto again; | |
1da177e4 LT |
7132 | } |
7133 | ||
7134 | /* | |
7135 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7136 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7137 | * for performance reasons the counter is not stricly tracking tasks to | |
7138 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7139 | * to keep the global sum constant after CPU-down: | |
7140 | */ | |
70b97a7f | 7141 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7142 | { |
1e5ce4f4 | 7143 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7144 | unsigned long flags; |
7145 | ||
7146 | local_irq_save(flags); | |
7147 | double_rq_lock(rq_src, rq_dest); | |
7148 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7149 | rq_src->nr_uninterruptible = 0; | |
7150 | double_rq_unlock(rq_src, rq_dest); | |
7151 | local_irq_restore(flags); | |
7152 | } | |
7153 | ||
7154 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7155 | static void migrate_live_tasks(int src_cpu) | |
7156 | { | |
48f24c4d | 7157 | struct task_struct *p, *t; |
1da177e4 | 7158 | |
f7b4cddc | 7159 | read_lock(&tasklist_lock); |
1da177e4 | 7160 | |
48f24c4d IM |
7161 | do_each_thread(t, p) { |
7162 | if (p == current) | |
1da177e4 LT |
7163 | continue; |
7164 | ||
48f24c4d IM |
7165 | if (task_cpu(p) == src_cpu) |
7166 | move_task_off_dead_cpu(src_cpu, p); | |
7167 | } while_each_thread(t, p); | |
1da177e4 | 7168 | |
f7b4cddc | 7169 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7170 | } |
7171 | ||
dd41f596 IM |
7172 | /* |
7173 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7174 | * It does so by boosting its priority to highest possible. |
7175 | * Used by CPU offline code. | |
1da177e4 LT |
7176 | */ |
7177 | void sched_idle_next(void) | |
7178 | { | |
48f24c4d | 7179 | int this_cpu = smp_processor_id(); |
70b97a7f | 7180 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7181 | struct task_struct *p = rq->idle; |
7182 | unsigned long flags; | |
7183 | ||
7184 | /* cpu has to be offline */ | |
48f24c4d | 7185 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7186 | |
48f24c4d IM |
7187 | /* |
7188 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7189 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7190 | */ |
7191 | spin_lock_irqsave(&rq->lock, flags); | |
7192 | ||
dd41f596 | 7193 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7194 | |
94bc9a7b DA |
7195 | update_rq_clock(rq); |
7196 | activate_task(rq, p, 0); | |
1da177e4 LT |
7197 | |
7198 | spin_unlock_irqrestore(&rq->lock, flags); | |
7199 | } | |
7200 | ||
48f24c4d IM |
7201 | /* |
7202 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7203 | * offline. |
7204 | */ | |
7205 | void idle_task_exit(void) | |
7206 | { | |
7207 | struct mm_struct *mm = current->active_mm; | |
7208 | ||
7209 | BUG_ON(cpu_online(smp_processor_id())); | |
7210 | ||
7211 | if (mm != &init_mm) | |
7212 | switch_mm(mm, &init_mm, current); | |
7213 | mmdrop(mm); | |
7214 | } | |
7215 | ||
054b9108 | 7216 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7217 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7218 | { |
70b97a7f | 7219 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7220 | |
7221 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7222 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7223 | |
7224 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7225 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7226 | |
48f24c4d | 7227 | get_task_struct(p); |
1da177e4 LT |
7228 | |
7229 | /* | |
7230 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7231 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7232 | * fine. |
7233 | */ | |
f7b4cddc | 7234 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7235 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7236 | spin_lock_irq(&rq->lock); |
1da177e4 | 7237 | |
48f24c4d | 7238 | put_task_struct(p); |
1da177e4 LT |
7239 | } |
7240 | ||
7241 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7242 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7243 | { | |
70b97a7f | 7244 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7245 | struct task_struct *next; |
48f24c4d | 7246 | |
dd41f596 IM |
7247 | for ( ; ; ) { |
7248 | if (!rq->nr_running) | |
7249 | break; | |
a8e504d2 | 7250 | update_rq_clock(rq); |
b67802ea | 7251 | next = pick_next_task(rq); |
dd41f596 IM |
7252 | if (!next) |
7253 | break; | |
79c53799 | 7254 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7255 | migrate_dead(dead_cpu, next); |
e692ab53 | 7256 | |
1da177e4 LT |
7257 | } |
7258 | } | |
dce48a84 TG |
7259 | |
7260 | /* | |
7261 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7262 | */ | |
7263 | static void calc_global_load_remove(struct rq *rq) | |
7264 | { | |
7265 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7266 | } | |
1da177e4 LT |
7267 | #endif /* CONFIG_HOTPLUG_CPU */ |
7268 | ||
e692ab53 NP |
7269 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7270 | ||
7271 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7272 | { |
7273 | .procname = "sched_domain", | |
c57baf1e | 7274 | .mode = 0555, |
e0361851 | 7275 | }, |
38605cae | 7276 | {0, }, |
e692ab53 NP |
7277 | }; |
7278 | ||
7279 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7280 | { |
c57baf1e | 7281 | .ctl_name = CTL_KERN, |
e0361851 | 7282 | .procname = "kernel", |
c57baf1e | 7283 | .mode = 0555, |
e0361851 AD |
7284 | .child = sd_ctl_dir, |
7285 | }, | |
38605cae | 7286 | {0, }, |
e692ab53 NP |
7287 | }; |
7288 | ||
7289 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7290 | { | |
7291 | struct ctl_table *entry = | |
5cf9f062 | 7292 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7293 | |
e692ab53 NP |
7294 | return entry; |
7295 | } | |
7296 | ||
6382bc90 MM |
7297 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7298 | { | |
cd790076 | 7299 | struct ctl_table *entry; |
6382bc90 | 7300 | |
cd790076 MM |
7301 | /* |
7302 | * In the intermediate directories, both the child directory and | |
7303 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7304 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7305 | * static strings and all have proc handlers. |
7306 | */ | |
7307 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7308 | if (entry->child) |
7309 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7310 | if (entry->proc_handler == NULL) |
7311 | kfree(entry->procname); | |
7312 | } | |
6382bc90 MM |
7313 | |
7314 | kfree(*tablep); | |
7315 | *tablep = NULL; | |
7316 | } | |
7317 | ||
e692ab53 | 7318 | static void |
e0361851 | 7319 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7320 | const char *procname, void *data, int maxlen, |
7321 | mode_t mode, proc_handler *proc_handler) | |
7322 | { | |
e692ab53 NP |
7323 | entry->procname = procname; |
7324 | entry->data = data; | |
7325 | entry->maxlen = maxlen; | |
7326 | entry->mode = mode; | |
7327 | entry->proc_handler = proc_handler; | |
7328 | } | |
7329 | ||
7330 | static struct ctl_table * | |
7331 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7332 | { | |
a5d8c348 | 7333 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7334 | |
ad1cdc1d MM |
7335 | if (table == NULL) |
7336 | return NULL; | |
7337 | ||
e0361851 | 7338 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7339 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7340 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7341 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7342 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7343 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7344 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7345 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7346 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7347 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7348 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7349 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7350 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7351 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7352 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7353 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7354 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7355 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7356 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7357 | &sd->cache_nice_tries, |
7358 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7359 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7360 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7361 | set_table_entry(&table[11], "name", sd->name, |
7362 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7363 | /* &table[12] is terminator */ | |
e692ab53 NP |
7364 | |
7365 | return table; | |
7366 | } | |
7367 | ||
9a4e7159 | 7368 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7369 | { |
7370 | struct ctl_table *entry, *table; | |
7371 | struct sched_domain *sd; | |
7372 | int domain_num = 0, i; | |
7373 | char buf[32]; | |
7374 | ||
7375 | for_each_domain(cpu, sd) | |
7376 | domain_num++; | |
7377 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7378 | if (table == NULL) |
7379 | return NULL; | |
e692ab53 NP |
7380 | |
7381 | i = 0; | |
7382 | for_each_domain(cpu, sd) { | |
7383 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7384 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7385 | entry->mode = 0555; |
e692ab53 NP |
7386 | entry->child = sd_alloc_ctl_domain_table(sd); |
7387 | entry++; | |
7388 | i++; | |
7389 | } | |
7390 | return table; | |
7391 | } | |
7392 | ||
7393 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7394 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7395 | { |
7396 | int i, cpu_num = num_online_cpus(); | |
7397 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7398 | char buf[32]; | |
7399 | ||
7378547f MM |
7400 | WARN_ON(sd_ctl_dir[0].child); |
7401 | sd_ctl_dir[0].child = entry; | |
7402 | ||
ad1cdc1d MM |
7403 | if (entry == NULL) |
7404 | return; | |
7405 | ||
97b6ea7b | 7406 | for_each_online_cpu(i) { |
e692ab53 | 7407 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7408 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7409 | entry->mode = 0555; |
e692ab53 | 7410 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7411 | entry++; |
e692ab53 | 7412 | } |
7378547f MM |
7413 | |
7414 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7415 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7416 | } | |
6382bc90 | 7417 | |
7378547f | 7418 | /* may be called multiple times per register */ |
6382bc90 MM |
7419 | static void unregister_sched_domain_sysctl(void) |
7420 | { | |
7378547f MM |
7421 | if (sd_sysctl_header) |
7422 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7423 | sd_sysctl_header = NULL; |
7378547f MM |
7424 | if (sd_ctl_dir[0].child) |
7425 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7426 | } |
e692ab53 | 7427 | #else |
6382bc90 MM |
7428 | static void register_sched_domain_sysctl(void) |
7429 | { | |
7430 | } | |
7431 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7432 | { |
7433 | } | |
7434 | #endif | |
7435 | ||
1f11eb6a GH |
7436 | static void set_rq_online(struct rq *rq) |
7437 | { | |
7438 | if (!rq->online) { | |
7439 | const struct sched_class *class; | |
7440 | ||
c6c4927b | 7441 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7442 | rq->online = 1; |
7443 | ||
7444 | for_each_class(class) { | |
7445 | if (class->rq_online) | |
7446 | class->rq_online(rq); | |
7447 | } | |
7448 | } | |
7449 | } | |
7450 | ||
7451 | static void set_rq_offline(struct rq *rq) | |
7452 | { | |
7453 | if (rq->online) { | |
7454 | const struct sched_class *class; | |
7455 | ||
7456 | for_each_class(class) { | |
7457 | if (class->rq_offline) | |
7458 | class->rq_offline(rq); | |
7459 | } | |
7460 | ||
c6c4927b | 7461 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7462 | rq->online = 0; |
7463 | } | |
7464 | } | |
7465 | ||
1da177e4 LT |
7466 | /* |
7467 | * migration_call - callback that gets triggered when a CPU is added. | |
7468 | * Here we can start up the necessary migration thread for the new CPU. | |
7469 | */ | |
48f24c4d IM |
7470 | static int __cpuinit |
7471 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7472 | { |
1da177e4 | 7473 | struct task_struct *p; |
48f24c4d | 7474 | int cpu = (long)hcpu; |
1da177e4 | 7475 | unsigned long flags; |
70b97a7f | 7476 | struct rq *rq; |
1da177e4 LT |
7477 | |
7478 | switch (action) { | |
5be9361c | 7479 | |
1da177e4 | 7480 | case CPU_UP_PREPARE: |
8bb78442 | 7481 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7482 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7483 | if (IS_ERR(p)) |
7484 | return NOTIFY_BAD; | |
1da177e4 LT |
7485 | kthread_bind(p, cpu); |
7486 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7487 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7488 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7489 | task_rq_unlock(rq, &flags); |
371cbb38 | 7490 | get_task_struct(p); |
1da177e4 LT |
7491 | cpu_rq(cpu)->migration_thread = p; |
7492 | break; | |
48f24c4d | 7493 | |
1da177e4 | 7494 | case CPU_ONLINE: |
8bb78442 | 7495 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7496 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7497 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7498 | |
7499 | /* Update our root-domain */ | |
7500 | rq = cpu_rq(cpu); | |
7501 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7502 | rq->calc_load_update = calc_load_update; |
7503 | rq->calc_load_active = 0; | |
1f94ef59 | 7504 | if (rq->rd) { |
c6c4927b | 7505 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7506 | |
7507 | set_rq_online(rq); | |
1f94ef59 GH |
7508 | } |
7509 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7510 | break; |
48f24c4d | 7511 | |
1da177e4 LT |
7512 | #ifdef CONFIG_HOTPLUG_CPU |
7513 | case CPU_UP_CANCELED: | |
8bb78442 | 7514 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7515 | if (!cpu_rq(cpu)->migration_thread) |
7516 | break; | |
41a2d6cf | 7517 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7518 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7519 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7520 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7521 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7522 | cpu_rq(cpu)->migration_thread = NULL; |
7523 | break; | |
48f24c4d | 7524 | |
1da177e4 | 7525 | case CPU_DEAD: |
8bb78442 | 7526 | case CPU_DEAD_FROZEN: |
470fd646 | 7527 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7528 | migrate_live_tasks(cpu); |
7529 | rq = cpu_rq(cpu); | |
7530 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7531 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7532 | rq->migration_thread = NULL; |
7533 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7534 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7535 | update_rq_clock(rq); |
2e1cb74a | 7536 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7537 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7538 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7539 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7540 | migrate_dead_tasks(cpu); |
d2da272a | 7541 | spin_unlock_irq(&rq->lock); |
470fd646 | 7542 | cpuset_unlock(); |
1da177e4 LT |
7543 | migrate_nr_uninterruptible(rq); |
7544 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7545 | calc_global_load_remove(rq); |
41a2d6cf IM |
7546 | /* |
7547 | * No need to migrate the tasks: it was best-effort if | |
7548 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7549 | * the requestors. | |
7550 | */ | |
1da177e4 LT |
7551 | spin_lock_irq(&rq->lock); |
7552 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7553 | struct migration_req *req; |
7554 | ||
1da177e4 | 7555 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7556 | struct migration_req, list); |
1da177e4 | 7557 | list_del_init(&req->list); |
9a2bd244 | 7558 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7559 | complete(&req->done); |
9a2bd244 | 7560 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7561 | } |
7562 | spin_unlock_irq(&rq->lock); | |
7563 | break; | |
57d885fe | 7564 | |
08f503b0 GH |
7565 | case CPU_DYING: |
7566 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7567 | /* Update our root-domain */ |
7568 | rq = cpu_rq(cpu); | |
7569 | spin_lock_irqsave(&rq->lock, flags); | |
7570 | if (rq->rd) { | |
c6c4927b | 7571 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7572 | set_rq_offline(rq); |
57d885fe GH |
7573 | } |
7574 | spin_unlock_irqrestore(&rq->lock, flags); | |
7575 | break; | |
1da177e4 LT |
7576 | #endif |
7577 | } | |
7578 | return NOTIFY_OK; | |
7579 | } | |
7580 | ||
f38b0820 PM |
7581 | /* |
7582 | * Register at high priority so that task migration (migrate_all_tasks) | |
7583 | * happens before everything else. This has to be lower priority than | |
7584 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7585 | */ |
26c2143b | 7586 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7587 | .notifier_call = migration_call, |
7588 | .priority = 10 | |
7589 | }; | |
7590 | ||
7babe8db | 7591 | static int __init migration_init(void) |
1da177e4 LT |
7592 | { |
7593 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7594 | int err; |
48f24c4d IM |
7595 | |
7596 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7597 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7598 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7599 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7600 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7601 | |
7602 | return err; | |
1da177e4 | 7603 | } |
7babe8db | 7604 | early_initcall(migration_init); |
1da177e4 LT |
7605 | #endif |
7606 | ||
7607 | #ifdef CONFIG_SMP | |
476f3534 | 7608 | |
3e9830dc | 7609 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7610 | |
7c16ec58 | 7611 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7612 | struct cpumask *groupmask) |
1da177e4 | 7613 | { |
4dcf6aff | 7614 | struct sched_group *group = sd->groups; |
434d53b0 | 7615 | char str[256]; |
1da177e4 | 7616 | |
968ea6d8 | 7617 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7618 | cpumask_clear(groupmask); |
4dcf6aff IM |
7619 | |
7620 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7621 | ||
7622 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7623 | printk("does not load-balance\n"); | |
7624 | if (sd->parent) | |
7625 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7626 | " has parent"); | |
7627 | return -1; | |
41c7ce9a NP |
7628 | } |
7629 | ||
eefd796a | 7630 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7631 | |
758b2cdc | 7632 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7633 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7634 | "CPU%d\n", cpu); | |
7635 | } | |
758b2cdc | 7636 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7637 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7638 | " CPU%d\n", cpu); | |
7639 | } | |
1da177e4 | 7640 | |
4dcf6aff | 7641 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7642 | do { |
4dcf6aff IM |
7643 | if (!group) { |
7644 | printk("\n"); | |
7645 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7646 | break; |
7647 | } | |
7648 | ||
4dcf6aff IM |
7649 | if (!group->__cpu_power) { |
7650 | printk(KERN_CONT "\n"); | |
7651 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7652 | "set\n"); | |
7653 | break; | |
7654 | } | |
1da177e4 | 7655 | |
758b2cdc | 7656 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7657 | printk(KERN_CONT "\n"); |
7658 | printk(KERN_ERR "ERROR: empty group\n"); | |
7659 | break; | |
7660 | } | |
1da177e4 | 7661 | |
758b2cdc | 7662 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7663 | printk(KERN_CONT "\n"); |
7664 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7665 | break; | |
7666 | } | |
1da177e4 | 7667 | |
758b2cdc | 7668 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7669 | |
968ea6d8 | 7670 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7671 | |
7672 | printk(KERN_CONT " %s", str); | |
7673 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7674 | printk(KERN_CONT " (__cpu_power = %d)", | |
7675 | group->__cpu_power); | |
7676 | } | |
1da177e4 | 7677 | |
4dcf6aff IM |
7678 | group = group->next; |
7679 | } while (group != sd->groups); | |
7680 | printk(KERN_CONT "\n"); | |
1da177e4 | 7681 | |
758b2cdc | 7682 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7683 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7684 | |
758b2cdc RR |
7685 | if (sd->parent && |
7686 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7687 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7688 | "of domain->span\n"); | |
7689 | return 0; | |
7690 | } | |
1da177e4 | 7691 | |
4dcf6aff IM |
7692 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7693 | { | |
d5dd3db1 | 7694 | cpumask_var_t groupmask; |
4dcf6aff | 7695 | int level = 0; |
1da177e4 | 7696 | |
4dcf6aff IM |
7697 | if (!sd) { |
7698 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7699 | return; | |
7700 | } | |
1da177e4 | 7701 | |
4dcf6aff IM |
7702 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7703 | ||
d5dd3db1 | 7704 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7705 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7706 | return; | |
7707 | } | |
7708 | ||
4dcf6aff | 7709 | for (;;) { |
7c16ec58 | 7710 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7711 | break; |
1da177e4 LT |
7712 | level++; |
7713 | sd = sd->parent; | |
33859f7f | 7714 | if (!sd) |
4dcf6aff IM |
7715 | break; |
7716 | } | |
d5dd3db1 | 7717 | free_cpumask_var(groupmask); |
1da177e4 | 7718 | } |
6d6bc0ad | 7719 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7720 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7721 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7722 | |
1a20ff27 | 7723 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7724 | { |
758b2cdc | 7725 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7726 | return 1; |
7727 | ||
7728 | /* Following flags need at least 2 groups */ | |
7729 | if (sd->flags & (SD_LOAD_BALANCE | | |
7730 | SD_BALANCE_NEWIDLE | | |
7731 | SD_BALANCE_FORK | | |
89c4710e SS |
7732 | SD_BALANCE_EXEC | |
7733 | SD_SHARE_CPUPOWER | | |
7734 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7735 | if (sd->groups != sd->groups->next) |
7736 | return 0; | |
7737 | } | |
7738 | ||
7739 | /* Following flags don't use groups */ | |
7740 | if (sd->flags & (SD_WAKE_IDLE | | |
7741 | SD_WAKE_AFFINE | | |
7742 | SD_WAKE_BALANCE)) | |
7743 | return 0; | |
7744 | ||
7745 | return 1; | |
7746 | } | |
7747 | ||
48f24c4d IM |
7748 | static int |
7749 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7750 | { |
7751 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7752 | ||
7753 | if (sd_degenerate(parent)) | |
7754 | return 1; | |
7755 | ||
758b2cdc | 7756 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7757 | return 0; |
7758 | ||
7759 | /* Does parent contain flags not in child? */ | |
7760 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7761 | if (cflags & SD_WAKE_AFFINE) | |
7762 | pflags &= ~SD_WAKE_BALANCE; | |
7763 | /* Flags needing groups don't count if only 1 group in parent */ | |
7764 | if (parent->groups == parent->groups->next) { | |
7765 | pflags &= ~(SD_LOAD_BALANCE | | |
7766 | SD_BALANCE_NEWIDLE | | |
7767 | SD_BALANCE_FORK | | |
89c4710e SS |
7768 | SD_BALANCE_EXEC | |
7769 | SD_SHARE_CPUPOWER | | |
7770 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7771 | if (nr_node_ids == 1) |
7772 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7773 | } |
7774 | if (~cflags & pflags) | |
7775 | return 0; | |
7776 | ||
7777 | return 1; | |
7778 | } | |
7779 | ||
c6c4927b RR |
7780 | static void free_rootdomain(struct root_domain *rd) |
7781 | { | |
68e74568 RR |
7782 | cpupri_cleanup(&rd->cpupri); |
7783 | ||
c6c4927b RR |
7784 | free_cpumask_var(rd->rto_mask); |
7785 | free_cpumask_var(rd->online); | |
7786 | free_cpumask_var(rd->span); | |
7787 | kfree(rd); | |
7788 | } | |
7789 | ||
57d885fe GH |
7790 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7791 | { | |
a0490fa3 | 7792 | struct root_domain *old_rd = NULL; |
57d885fe | 7793 | unsigned long flags; |
57d885fe GH |
7794 | |
7795 | spin_lock_irqsave(&rq->lock, flags); | |
7796 | ||
7797 | if (rq->rd) { | |
a0490fa3 | 7798 | old_rd = rq->rd; |
57d885fe | 7799 | |
c6c4927b | 7800 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7801 | set_rq_offline(rq); |
57d885fe | 7802 | |
c6c4927b | 7803 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7804 | |
a0490fa3 IM |
7805 | /* |
7806 | * If we dont want to free the old_rt yet then | |
7807 | * set old_rd to NULL to skip the freeing later | |
7808 | * in this function: | |
7809 | */ | |
7810 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7811 | old_rd = NULL; | |
57d885fe GH |
7812 | } |
7813 | ||
7814 | atomic_inc(&rd->refcount); | |
7815 | rq->rd = rd; | |
7816 | ||
c6c4927b RR |
7817 | cpumask_set_cpu(rq->cpu, rd->span); |
7818 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7819 | set_rq_online(rq); |
57d885fe GH |
7820 | |
7821 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7822 | |
7823 | if (old_rd) | |
7824 | free_rootdomain(old_rd); | |
57d885fe GH |
7825 | } |
7826 | ||
fd5e1b5d | 7827 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7828 | { |
36b7b6d4 PE |
7829 | gfp_t gfp = GFP_KERNEL; |
7830 | ||
57d885fe GH |
7831 | memset(rd, 0, sizeof(*rd)); |
7832 | ||
36b7b6d4 PE |
7833 | if (bootmem) |
7834 | gfp = GFP_NOWAIT; | |
c6c4927b | 7835 | |
36b7b6d4 | 7836 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7837 | goto out; |
36b7b6d4 | 7838 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7839 | goto free_span; |
36b7b6d4 | 7840 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7841 | goto free_online; |
6e0534f2 | 7842 | |
0fb53029 | 7843 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7844 | goto free_rto_mask; |
c6c4927b | 7845 | return 0; |
6e0534f2 | 7846 | |
68e74568 RR |
7847 | free_rto_mask: |
7848 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7849 | free_online: |
7850 | free_cpumask_var(rd->online); | |
7851 | free_span: | |
7852 | free_cpumask_var(rd->span); | |
0c910d28 | 7853 | out: |
c6c4927b | 7854 | return -ENOMEM; |
57d885fe GH |
7855 | } |
7856 | ||
7857 | static void init_defrootdomain(void) | |
7858 | { | |
c6c4927b RR |
7859 | init_rootdomain(&def_root_domain, true); |
7860 | ||
57d885fe GH |
7861 | atomic_set(&def_root_domain.refcount, 1); |
7862 | } | |
7863 | ||
dc938520 | 7864 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7865 | { |
7866 | struct root_domain *rd; | |
7867 | ||
7868 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7869 | if (!rd) | |
7870 | return NULL; | |
7871 | ||
c6c4927b RR |
7872 | if (init_rootdomain(rd, false) != 0) { |
7873 | kfree(rd); | |
7874 | return NULL; | |
7875 | } | |
57d885fe GH |
7876 | |
7877 | return rd; | |
7878 | } | |
7879 | ||
1da177e4 | 7880 | /* |
0eab9146 | 7881 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7882 | * hold the hotplug lock. |
7883 | */ | |
0eab9146 IM |
7884 | static void |
7885 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7886 | { |
70b97a7f | 7887 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7888 | struct sched_domain *tmp; |
7889 | ||
7890 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7891 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7892 | struct sched_domain *parent = tmp->parent; |
7893 | if (!parent) | |
7894 | break; | |
f29c9b1c | 7895 | |
1a848870 | 7896 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7897 | tmp->parent = parent->parent; |
1a848870 SS |
7898 | if (parent->parent) |
7899 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7900 | } else |
7901 | tmp = tmp->parent; | |
245af2c7 SS |
7902 | } |
7903 | ||
1a848870 | 7904 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7905 | sd = sd->parent; |
1a848870 SS |
7906 | if (sd) |
7907 | sd->child = NULL; | |
7908 | } | |
1da177e4 LT |
7909 | |
7910 | sched_domain_debug(sd, cpu); | |
7911 | ||
57d885fe | 7912 | rq_attach_root(rq, rd); |
674311d5 | 7913 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7914 | } |
7915 | ||
7916 | /* cpus with isolated domains */ | |
dcc30a35 | 7917 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7918 | |
7919 | /* Setup the mask of cpus configured for isolated domains */ | |
7920 | static int __init isolated_cpu_setup(char *str) | |
7921 | { | |
968ea6d8 | 7922 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7923 | return 1; |
7924 | } | |
7925 | ||
8927f494 | 7926 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7927 | |
7928 | /* | |
6711cab4 SS |
7929 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7930 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7931 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7932 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7933 | * |
7934 | * init_sched_build_groups will build a circular linked list of the groups | |
7935 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7936 | * and ->cpu_power to 0. | |
7937 | */ | |
a616058b | 7938 | static void |
96f874e2 RR |
7939 | init_sched_build_groups(const struct cpumask *span, |
7940 | const struct cpumask *cpu_map, | |
7941 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7942 | struct sched_group **sg, |
96f874e2 RR |
7943 | struct cpumask *tmpmask), |
7944 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7945 | { |
7946 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7947 | int i; |
7948 | ||
96f874e2 | 7949 | cpumask_clear(covered); |
7c16ec58 | 7950 | |
abcd083a | 7951 | for_each_cpu(i, span) { |
6711cab4 | 7952 | struct sched_group *sg; |
7c16ec58 | 7953 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7954 | int j; |
7955 | ||
758b2cdc | 7956 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7957 | continue; |
7958 | ||
758b2cdc | 7959 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7960 | sg->__cpu_power = 0; |
1da177e4 | 7961 | |
abcd083a | 7962 | for_each_cpu(j, span) { |
7c16ec58 | 7963 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7964 | continue; |
7965 | ||
96f874e2 | 7966 | cpumask_set_cpu(j, covered); |
758b2cdc | 7967 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7968 | } |
7969 | if (!first) | |
7970 | first = sg; | |
7971 | if (last) | |
7972 | last->next = sg; | |
7973 | last = sg; | |
7974 | } | |
7975 | last->next = first; | |
7976 | } | |
7977 | ||
9c1cfda2 | 7978 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7979 | |
9c1cfda2 | 7980 | #ifdef CONFIG_NUMA |
198e2f18 | 7981 | |
9c1cfda2 JH |
7982 | /** |
7983 | * find_next_best_node - find the next node to include in a sched_domain | |
7984 | * @node: node whose sched_domain we're building | |
7985 | * @used_nodes: nodes already in the sched_domain | |
7986 | * | |
41a2d6cf | 7987 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7988 | * finds the closest node not already in the @used_nodes map. |
7989 | * | |
7990 | * Should use nodemask_t. | |
7991 | */ | |
c5f59f08 | 7992 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7993 | { |
7994 | int i, n, val, min_val, best_node = 0; | |
7995 | ||
7996 | min_val = INT_MAX; | |
7997 | ||
076ac2af | 7998 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7999 | /* Start at @node */ |
076ac2af | 8000 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8001 | |
8002 | if (!nr_cpus_node(n)) | |
8003 | continue; | |
8004 | ||
8005 | /* Skip already used nodes */ | |
c5f59f08 | 8006 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8007 | continue; |
8008 | ||
8009 | /* Simple min distance search */ | |
8010 | val = node_distance(node, n); | |
8011 | ||
8012 | if (val < min_val) { | |
8013 | min_val = val; | |
8014 | best_node = n; | |
8015 | } | |
8016 | } | |
8017 | ||
c5f59f08 | 8018 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8019 | return best_node; |
8020 | } | |
8021 | ||
8022 | /** | |
8023 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8024 | * @node: node whose cpumask we're constructing | |
73486722 | 8025 | * @span: resulting cpumask |
9c1cfda2 | 8026 | * |
41a2d6cf | 8027 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8028 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8029 | * out optimally. | |
8030 | */ | |
96f874e2 | 8031 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8032 | { |
c5f59f08 | 8033 | nodemask_t used_nodes; |
48f24c4d | 8034 | int i; |
9c1cfda2 | 8035 | |
6ca09dfc | 8036 | cpumask_clear(span); |
c5f59f08 | 8037 | nodes_clear(used_nodes); |
9c1cfda2 | 8038 | |
6ca09dfc | 8039 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8040 | node_set(node, used_nodes); |
9c1cfda2 JH |
8041 | |
8042 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8043 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8044 | |
6ca09dfc | 8045 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8046 | } |
9c1cfda2 | 8047 | } |
6d6bc0ad | 8048 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8049 | |
5c45bf27 | 8050 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8051 | |
6c99e9ad RR |
8052 | /* |
8053 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8054 | * |
8055 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8056 | * and struct sched_domain. ) | |
6c99e9ad RR |
8057 | */ |
8058 | struct static_sched_group { | |
8059 | struct sched_group sg; | |
8060 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8061 | }; | |
8062 | ||
8063 | struct static_sched_domain { | |
8064 | struct sched_domain sd; | |
8065 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8066 | }; | |
8067 | ||
9c1cfda2 | 8068 | /* |
48f24c4d | 8069 | * SMT sched-domains: |
9c1cfda2 | 8070 | */ |
1da177e4 | 8071 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8072 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8073 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8074 | |
41a2d6cf | 8075 | static int |
96f874e2 RR |
8076 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8077 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8078 | { |
6711cab4 | 8079 | if (sg) |
6c99e9ad | 8080 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8081 | return cpu; |
8082 | } | |
6d6bc0ad | 8083 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8084 | |
48f24c4d IM |
8085 | /* |
8086 | * multi-core sched-domains: | |
8087 | */ | |
1e9f28fa | 8088 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8089 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8090 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8091 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8092 | |
8093 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8094 | static int |
96f874e2 RR |
8095 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8096 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8097 | { |
6711cab4 | 8098 | int group; |
7c16ec58 | 8099 | |
c69fc56d | 8100 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8101 | group = cpumask_first(mask); |
6711cab4 | 8102 | if (sg) |
6c99e9ad | 8103 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8104 | return group; |
1e9f28fa SS |
8105 | } |
8106 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8107 | static int |
96f874e2 RR |
8108 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8109 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8110 | { |
6711cab4 | 8111 | if (sg) |
6c99e9ad | 8112 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8113 | return cpu; |
8114 | } | |
8115 | #endif | |
8116 | ||
6c99e9ad RR |
8117 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8118 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8119 | |
41a2d6cf | 8120 | static int |
96f874e2 RR |
8121 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8122 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8123 | { |
6711cab4 | 8124 | int group; |
48f24c4d | 8125 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8126 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8127 | group = cpumask_first(mask); |
1e9f28fa | 8128 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8129 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8130 | group = cpumask_first(mask); |
1da177e4 | 8131 | #else |
6711cab4 | 8132 | group = cpu; |
1da177e4 | 8133 | #endif |
6711cab4 | 8134 | if (sg) |
6c99e9ad | 8135 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8136 | return group; |
1da177e4 LT |
8137 | } |
8138 | ||
8139 | #ifdef CONFIG_NUMA | |
1da177e4 | 8140 | /* |
9c1cfda2 JH |
8141 | * The init_sched_build_groups can't handle what we want to do with node |
8142 | * groups, so roll our own. Now each node has its own list of groups which | |
8143 | * gets dynamically allocated. | |
1da177e4 | 8144 | */ |
62ea9ceb | 8145 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8146 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8147 | |
62ea9ceb | 8148 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8149 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8150 | |
96f874e2 RR |
8151 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8152 | struct sched_group **sg, | |
8153 | struct cpumask *nodemask) | |
9c1cfda2 | 8154 | { |
6711cab4 SS |
8155 | int group; |
8156 | ||
6ca09dfc | 8157 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8158 | group = cpumask_first(nodemask); |
6711cab4 SS |
8159 | |
8160 | if (sg) | |
6c99e9ad | 8161 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8162 | return group; |
1da177e4 | 8163 | } |
6711cab4 | 8164 | |
08069033 SS |
8165 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8166 | { | |
8167 | struct sched_group *sg = group_head; | |
8168 | int j; | |
8169 | ||
8170 | if (!sg) | |
8171 | return; | |
3a5c359a | 8172 | do { |
758b2cdc | 8173 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8174 | struct sched_domain *sd; |
08069033 | 8175 | |
6c99e9ad | 8176 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8177 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8178 | /* |
8179 | * Only add "power" once for each | |
8180 | * physical package. | |
8181 | */ | |
8182 | continue; | |
8183 | } | |
08069033 | 8184 | |
3a5c359a AK |
8185 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8186 | } | |
8187 | sg = sg->next; | |
8188 | } while (sg != group_head); | |
08069033 | 8189 | } |
6d6bc0ad | 8190 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8191 | |
a616058b | 8192 | #ifdef CONFIG_NUMA |
51888ca2 | 8193 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8194 | static void free_sched_groups(const struct cpumask *cpu_map, |
8195 | struct cpumask *nodemask) | |
51888ca2 | 8196 | { |
a616058b | 8197 | int cpu, i; |
51888ca2 | 8198 | |
abcd083a | 8199 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8200 | struct sched_group **sched_group_nodes |
8201 | = sched_group_nodes_bycpu[cpu]; | |
8202 | ||
51888ca2 SV |
8203 | if (!sched_group_nodes) |
8204 | continue; | |
8205 | ||
076ac2af | 8206 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8207 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8208 | ||
6ca09dfc | 8209 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8210 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8211 | continue; |
8212 | ||
8213 | if (sg == NULL) | |
8214 | continue; | |
8215 | sg = sg->next; | |
8216 | next_sg: | |
8217 | oldsg = sg; | |
8218 | sg = sg->next; | |
8219 | kfree(oldsg); | |
8220 | if (oldsg != sched_group_nodes[i]) | |
8221 | goto next_sg; | |
8222 | } | |
8223 | kfree(sched_group_nodes); | |
8224 | sched_group_nodes_bycpu[cpu] = NULL; | |
8225 | } | |
51888ca2 | 8226 | } |
6d6bc0ad | 8227 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8228 | static void free_sched_groups(const struct cpumask *cpu_map, |
8229 | struct cpumask *nodemask) | |
a616058b SS |
8230 | { |
8231 | } | |
6d6bc0ad | 8232 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8233 | |
89c4710e SS |
8234 | /* |
8235 | * Initialize sched groups cpu_power. | |
8236 | * | |
8237 | * cpu_power indicates the capacity of sched group, which is used while | |
8238 | * distributing the load between different sched groups in a sched domain. | |
8239 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8240 | * there are asymmetries in the topology. If there are asymmetries, group | |
8241 | * having more cpu_power will pickup more load compared to the group having | |
8242 | * less cpu_power. | |
8243 | * | |
8244 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8245 | * the maximum number of tasks a group can handle in the presence of other idle | |
8246 | * or lightly loaded groups in the same sched domain. | |
8247 | */ | |
8248 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8249 | { | |
8250 | struct sched_domain *child; | |
8251 | struct sched_group *group; | |
8252 | ||
8253 | WARN_ON(!sd || !sd->groups); | |
8254 | ||
13318a71 | 8255 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8256 | return; |
8257 | ||
8258 | child = sd->child; | |
8259 | ||
5517d86b ED |
8260 | sd->groups->__cpu_power = 0; |
8261 | ||
89c4710e SS |
8262 | /* |
8263 | * For perf policy, if the groups in child domain share resources | |
8264 | * (for example cores sharing some portions of the cache hierarchy | |
8265 | * or SMT), then set this domain groups cpu_power such that each group | |
8266 | * can handle only one task, when there are other idle groups in the | |
8267 | * same sched domain. | |
8268 | */ | |
8269 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8270 | (child->flags & | |
8271 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8272 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8273 | return; |
8274 | } | |
8275 | ||
89c4710e SS |
8276 | /* |
8277 | * add cpu_power of each child group to this groups cpu_power | |
8278 | */ | |
8279 | group = child->groups; | |
8280 | do { | |
5517d86b | 8281 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8282 | group = group->next; |
8283 | } while (group != child->groups); | |
8284 | } | |
8285 | ||
7c16ec58 MT |
8286 | /* |
8287 | * Initializers for schedule domains | |
8288 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8289 | */ | |
8290 | ||
a5d8c348 IM |
8291 | #ifdef CONFIG_SCHED_DEBUG |
8292 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8293 | #else | |
8294 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8295 | #endif | |
8296 | ||
7c16ec58 | 8297 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8298 | |
7c16ec58 MT |
8299 | #define SD_INIT_FUNC(type) \ |
8300 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8301 | { \ | |
8302 | memset(sd, 0, sizeof(*sd)); \ | |
8303 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8304 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8305 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8306 | } |
8307 | ||
8308 | SD_INIT_FUNC(CPU) | |
8309 | #ifdef CONFIG_NUMA | |
8310 | SD_INIT_FUNC(ALLNODES) | |
8311 | SD_INIT_FUNC(NODE) | |
8312 | #endif | |
8313 | #ifdef CONFIG_SCHED_SMT | |
8314 | SD_INIT_FUNC(SIBLING) | |
8315 | #endif | |
8316 | #ifdef CONFIG_SCHED_MC | |
8317 | SD_INIT_FUNC(MC) | |
8318 | #endif | |
8319 | ||
1d3504fc HS |
8320 | static int default_relax_domain_level = -1; |
8321 | ||
8322 | static int __init setup_relax_domain_level(char *str) | |
8323 | { | |
30e0e178 LZ |
8324 | unsigned long val; |
8325 | ||
8326 | val = simple_strtoul(str, NULL, 0); | |
8327 | if (val < SD_LV_MAX) | |
8328 | default_relax_domain_level = val; | |
8329 | ||
1d3504fc HS |
8330 | return 1; |
8331 | } | |
8332 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8333 | ||
8334 | static void set_domain_attribute(struct sched_domain *sd, | |
8335 | struct sched_domain_attr *attr) | |
8336 | { | |
8337 | int request; | |
8338 | ||
8339 | if (!attr || attr->relax_domain_level < 0) { | |
8340 | if (default_relax_domain_level < 0) | |
8341 | return; | |
8342 | else | |
8343 | request = default_relax_domain_level; | |
8344 | } else | |
8345 | request = attr->relax_domain_level; | |
8346 | if (request < sd->level) { | |
8347 | /* turn off idle balance on this domain */ | |
8348 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8349 | } else { | |
8350 | /* turn on idle balance on this domain */ | |
8351 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8352 | } | |
8353 | } | |
8354 | ||
1da177e4 | 8355 | /* |
1a20ff27 DG |
8356 | * Build sched domains for a given set of cpus and attach the sched domains |
8357 | * to the individual cpus | |
1da177e4 | 8358 | */ |
96f874e2 | 8359 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8360 | struct sched_domain_attr *attr) |
1da177e4 | 8361 | { |
3404c8d9 | 8362 | int i, err = -ENOMEM; |
57d885fe | 8363 | struct root_domain *rd; |
3404c8d9 RR |
8364 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8365 | tmpmask; | |
d1b55138 | 8366 | #ifdef CONFIG_NUMA |
3404c8d9 | 8367 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8368 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8369 | int sd_allnodes = 0; |
d1b55138 | 8370 | |
3404c8d9 RR |
8371 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8372 | goto out; | |
8373 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8374 | goto free_domainspan; | |
8375 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8376 | goto free_covered; | |
8377 | #endif | |
8378 | ||
8379 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8380 | goto free_notcovered; | |
8381 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8382 | goto free_nodemask; | |
8383 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8384 | goto free_this_sibling_map; | |
8385 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8386 | goto free_this_core_map; | |
8387 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8388 | goto free_send_covered; | |
8389 | ||
8390 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8391 | /* |
8392 | * Allocate the per-node list of sched groups | |
8393 | */ | |
076ac2af | 8394 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8395 | GFP_KERNEL); |
d1b55138 JH |
8396 | if (!sched_group_nodes) { |
8397 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8398 | goto free_tmpmask; |
d1b55138 | 8399 | } |
d1b55138 | 8400 | #endif |
1da177e4 | 8401 | |
dc938520 | 8402 | rd = alloc_rootdomain(); |
57d885fe GH |
8403 | if (!rd) { |
8404 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8405 | goto free_sched_groups; |
57d885fe GH |
8406 | } |
8407 | ||
7c16ec58 | 8408 | #ifdef CONFIG_NUMA |
96f874e2 | 8409 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8410 | #endif |
8411 | ||
1da177e4 | 8412 | /* |
1a20ff27 | 8413 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8414 | */ |
abcd083a | 8415 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8416 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8417 | |
6ca09dfc | 8418 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8419 | |
8420 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8421 | if (cpumask_weight(cpu_map) > |
8422 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8423 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8424 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8425 | set_domain_attribute(sd, attr); |
758b2cdc | 8426 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8427 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8428 | p = sd; |
6711cab4 | 8429 | sd_allnodes = 1; |
9c1cfda2 JH |
8430 | } else |
8431 | p = NULL; | |
8432 | ||
62ea9ceb | 8433 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8434 | SD_INIT(sd, NODE); |
1d3504fc | 8435 | set_domain_attribute(sd, attr); |
758b2cdc | 8436 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8437 | sd->parent = p; |
1a848870 SS |
8438 | if (p) |
8439 | p->child = sd; | |
758b2cdc RR |
8440 | cpumask_and(sched_domain_span(sd), |
8441 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8442 | #endif |
8443 | ||
8444 | p = sd; | |
6c99e9ad | 8445 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8446 | SD_INIT(sd, CPU); |
1d3504fc | 8447 | set_domain_attribute(sd, attr); |
758b2cdc | 8448 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8449 | sd->parent = p; |
1a848870 SS |
8450 | if (p) |
8451 | p->child = sd; | |
7c16ec58 | 8452 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8453 | |
1e9f28fa SS |
8454 | #ifdef CONFIG_SCHED_MC |
8455 | p = sd; | |
6c99e9ad | 8456 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8457 | SD_INIT(sd, MC); |
1d3504fc | 8458 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8459 | cpumask_and(sched_domain_span(sd), cpu_map, |
8460 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8461 | sd->parent = p; |
1a848870 | 8462 | p->child = sd; |
7c16ec58 | 8463 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8464 | #endif |
8465 | ||
1da177e4 LT |
8466 | #ifdef CONFIG_SCHED_SMT |
8467 | p = sd; | |
6c99e9ad | 8468 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8469 | SD_INIT(sd, SIBLING); |
1d3504fc | 8470 | set_domain_attribute(sd, attr); |
758b2cdc | 8471 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8472 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8473 | sd->parent = p; |
1a848870 | 8474 | p->child = sd; |
7c16ec58 | 8475 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8476 | #endif |
8477 | } | |
8478 | ||
8479 | #ifdef CONFIG_SCHED_SMT | |
8480 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8481 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8482 | cpumask_and(this_sibling_map, |
c69fc56d | 8483 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8484 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8485 | continue; |
8486 | ||
dd41f596 | 8487 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8488 | &cpu_to_cpu_group, |
8489 | send_covered, tmpmask); | |
1da177e4 LT |
8490 | } |
8491 | #endif | |
8492 | ||
1e9f28fa SS |
8493 | #ifdef CONFIG_SCHED_MC |
8494 | /* Set up multi-core groups */ | |
abcd083a | 8495 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8496 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8497 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8498 | continue; |
7c16ec58 | 8499 | |
dd41f596 | 8500 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8501 | &cpu_to_core_group, |
8502 | send_covered, tmpmask); | |
1e9f28fa SS |
8503 | } |
8504 | #endif | |
8505 | ||
1da177e4 | 8506 | /* Set up physical groups */ |
076ac2af | 8507 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8508 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8509 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8510 | continue; |
8511 | ||
7c16ec58 MT |
8512 | init_sched_build_groups(nodemask, cpu_map, |
8513 | &cpu_to_phys_group, | |
8514 | send_covered, tmpmask); | |
1da177e4 LT |
8515 | } |
8516 | ||
8517 | #ifdef CONFIG_NUMA | |
8518 | /* Set up node groups */ | |
7c16ec58 | 8519 | if (sd_allnodes) { |
7c16ec58 MT |
8520 | init_sched_build_groups(cpu_map, cpu_map, |
8521 | &cpu_to_allnodes_group, | |
8522 | send_covered, tmpmask); | |
8523 | } | |
9c1cfda2 | 8524 | |
076ac2af | 8525 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8526 | /* Set up node groups */ |
8527 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8528 | int j; |
8529 | ||
96f874e2 | 8530 | cpumask_clear(covered); |
6ca09dfc | 8531 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8532 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8533 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8534 | continue; |
d1b55138 | 8535 | } |
9c1cfda2 | 8536 | |
4bdbaad3 | 8537 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8538 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8539 | |
6c99e9ad RR |
8540 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8541 | GFP_KERNEL, i); | |
51888ca2 SV |
8542 | if (!sg) { |
8543 | printk(KERN_WARNING "Can not alloc domain group for " | |
8544 | "node %d\n", i); | |
8545 | goto error; | |
8546 | } | |
9c1cfda2 | 8547 | sched_group_nodes[i] = sg; |
abcd083a | 8548 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8549 | struct sched_domain *sd; |
9761eea8 | 8550 | |
62ea9ceb | 8551 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8552 | sd->groups = sg; |
9c1cfda2 | 8553 | } |
5517d86b | 8554 | sg->__cpu_power = 0; |
758b2cdc | 8555 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8556 | sg->next = sg; |
96f874e2 | 8557 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8558 | prev = sg; |
8559 | ||
076ac2af | 8560 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8561 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8562 | |
96f874e2 RR |
8563 | cpumask_complement(notcovered, covered); |
8564 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8565 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8566 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8567 | break; |
8568 | ||
6ca09dfc | 8569 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8570 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8571 | continue; |
8572 | ||
6c99e9ad RR |
8573 | sg = kmalloc_node(sizeof(struct sched_group) + |
8574 | cpumask_size(), | |
15f0b676 | 8575 | GFP_KERNEL, i); |
9c1cfda2 JH |
8576 | if (!sg) { |
8577 | printk(KERN_WARNING | |
8578 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8579 | goto error; |
9c1cfda2 | 8580 | } |
5517d86b | 8581 | sg->__cpu_power = 0; |
758b2cdc | 8582 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8583 | sg->next = prev->next; |
96f874e2 | 8584 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8585 | prev->next = sg; |
8586 | prev = sg; | |
8587 | } | |
9c1cfda2 | 8588 | } |
1da177e4 LT |
8589 | #endif |
8590 | ||
8591 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8592 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8593 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8594 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8595 | |
89c4710e | 8596 | init_sched_groups_power(i, sd); |
5c45bf27 | 8597 | } |
1da177e4 | 8598 | #endif |
1e9f28fa | 8599 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8600 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8601 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8602 | |
89c4710e | 8603 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8604 | } |
8605 | #endif | |
1e9f28fa | 8606 | |
abcd083a | 8607 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8608 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8609 | |
89c4710e | 8610 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8611 | } |
8612 | ||
9c1cfda2 | 8613 | #ifdef CONFIG_NUMA |
076ac2af | 8614 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8615 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8616 | |
6711cab4 SS |
8617 | if (sd_allnodes) { |
8618 | struct sched_group *sg; | |
f712c0c7 | 8619 | |
96f874e2 | 8620 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8621 | tmpmask); |
f712c0c7 SS |
8622 | init_numa_sched_groups_power(sg); |
8623 | } | |
9c1cfda2 JH |
8624 | #endif |
8625 | ||
1da177e4 | 8626 | /* Attach the domains */ |
abcd083a | 8627 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8628 | struct sched_domain *sd; |
8629 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8630 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8631 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8632 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8633 | #else |
6c99e9ad | 8634 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8635 | #endif |
57d885fe | 8636 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8637 | } |
51888ca2 | 8638 | |
3404c8d9 RR |
8639 | err = 0; |
8640 | ||
8641 | free_tmpmask: | |
8642 | free_cpumask_var(tmpmask); | |
8643 | free_send_covered: | |
8644 | free_cpumask_var(send_covered); | |
8645 | free_this_core_map: | |
8646 | free_cpumask_var(this_core_map); | |
8647 | free_this_sibling_map: | |
8648 | free_cpumask_var(this_sibling_map); | |
8649 | free_nodemask: | |
8650 | free_cpumask_var(nodemask); | |
8651 | free_notcovered: | |
8652 | #ifdef CONFIG_NUMA | |
8653 | free_cpumask_var(notcovered); | |
8654 | free_covered: | |
8655 | free_cpumask_var(covered); | |
8656 | free_domainspan: | |
8657 | free_cpumask_var(domainspan); | |
8658 | out: | |
8659 | #endif | |
8660 | return err; | |
8661 | ||
8662 | free_sched_groups: | |
8663 | #ifdef CONFIG_NUMA | |
8664 | kfree(sched_group_nodes); | |
8665 | #endif | |
8666 | goto free_tmpmask; | |
51888ca2 | 8667 | |
a616058b | 8668 | #ifdef CONFIG_NUMA |
51888ca2 | 8669 | error: |
7c16ec58 | 8670 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8671 | free_rootdomain(rd); |
3404c8d9 | 8672 | goto free_tmpmask; |
a616058b | 8673 | #endif |
1da177e4 | 8674 | } |
029190c5 | 8675 | |
96f874e2 | 8676 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8677 | { |
8678 | return __build_sched_domains(cpu_map, NULL); | |
8679 | } | |
8680 | ||
96f874e2 | 8681 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8682 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8683 | static struct sched_domain_attr *dattr_cur; |
8684 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8685 | |
8686 | /* | |
8687 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8688 | * cpumask) fails, then fallback to a single sched domain, |
8689 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8690 | */ |
4212823f | 8691 | static cpumask_var_t fallback_doms; |
029190c5 | 8692 | |
ee79d1bd HC |
8693 | /* |
8694 | * arch_update_cpu_topology lets virtualized architectures update the | |
8695 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8696 | * or 0 if it stayed the same. | |
8697 | */ | |
8698 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8699 | { |
ee79d1bd | 8700 | return 0; |
22e52b07 HC |
8701 | } |
8702 | ||
1a20ff27 | 8703 | /* |
41a2d6cf | 8704 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8705 | * For now this just excludes isolated cpus, but could be used to |
8706 | * exclude other special cases in the future. | |
1a20ff27 | 8707 | */ |
96f874e2 | 8708 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8709 | { |
7378547f MM |
8710 | int err; |
8711 | ||
22e52b07 | 8712 | arch_update_cpu_topology(); |
029190c5 | 8713 | ndoms_cur = 1; |
96f874e2 | 8714 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8715 | if (!doms_cur) |
4212823f | 8716 | doms_cur = fallback_doms; |
dcc30a35 | 8717 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8718 | dattr_cur = NULL; |
7378547f | 8719 | err = build_sched_domains(doms_cur); |
6382bc90 | 8720 | register_sched_domain_sysctl(); |
7378547f MM |
8721 | |
8722 | return err; | |
1a20ff27 DG |
8723 | } |
8724 | ||
96f874e2 RR |
8725 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8726 | struct cpumask *tmpmask) | |
1da177e4 | 8727 | { |
7c16ec58 | 8728 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8729 | } |
1da177e4 | 8730 | |
1a20ff27 DG |
8731 | /* |
8732 | * Detach sched domains from a group of cpus specified in cpu_map | |
8733 | * These cpus will now be attached to the NULL domain | |
8734 | */ | |
96f874e2 | 8735 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8736 | { |
96f874e2 RR |
8737 | /* Save because hotplug lock held. */ |
8738 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8739 | int i; |
8740 | ||
abcd083a | 8741 | for_each_cpu(i, cpu_map) |
57d885fe | 8742 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8743 | synchronize_sched(); |
96f874e2 | 8744 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8745 | } |
8746 | ||
1d3504fc HS |
8747 | /* handle null as "default" */ |
8748 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8749 | struct sched_domain_attr *new, int idx_new) | |
8750 | { | |
8751 | struct sched_domain_attr tmp; | |
8752 | ||
8753 | /* fast path */ | |
8754 | if (!new && !cur) | |
8755 | return 1; | |
8756 | ||
8757 | tmp = SD_ATTR_INIT; | |
8758 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8759 | new ? (new + idx_new) : &tmp, | |
8760 | sizeof(struct sched_domain_attr)); | |
8761 | } | |
8762 | ||
029190c5 PJ |
8763 | /* |
8764 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8765 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8766 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8767 | * It destroys each deleted domain and builds each new domain. | |
8768 | * | |
96f874e2 | 8769 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8770 | * The masks don't intersect (don't overlap.) We should setup one |
8771 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8772 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8773 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8774 | * it as it is. | |
8775 | * | |
41a2d6cf IM |
8776 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8777 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8778 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8779 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8780 | * the single partition 'fallback_doms', it also forces the domains | |
8781 | * to be rebuilt. | |
029190c5 | 8782 | * |
96f874e2 | 8783 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8784 | * ndoms_new == 0 is a special case for destroying existing domains, |
8785 | * and it will not create the default domain. | |
dfb512ec | 8786 | * |
029190c5 PJ |
8787 | * Call with hotplug lock held |
8788 | */ | |
96f874e2 RR |
8789 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8790 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8791 | struct sched_domain_attr *dattr_new) |
029190c5 | 8792 | { |
dfb512ec | 8793 | int i, j, n; |
d65bd5ec | 8794 | int new_topology; |
029190c5 | 8795 | |
712555ee | 8796 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8797 | |
7378547f MM |
8798 | /* always unregister in case we don't destroy any domains */ |
8799 | unregister_sched_domain_sysctl(); | |
8800 | ||
d65bd5ec HC |
8801 | /* Let architecture update cpu core mappings. */ |
8802 | new_topology = arch_update_cpu_topology(); | |
8803 | ||
dfb512ec | 8804 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8805 | |
8806 | /* Destroy deleted domains */ | |
8807 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8808 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8809 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8810 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8811 | goto match1; |
8812 | } | |
8813 | /* no match - a current sched domain not in new doms_new[] */ | |
8814 | detach_destroy_domains(doms_cur + i); | |
8815 | match1: | |
8816 | ; | |
8817 | } | |
8818 | ||
e761b772 MK |
8819 | if (doms_new == NULL) { |
8820 | ndoms_cur = 0; | |
4212823f | 8821 | doms_new = fallback_doms; |
dcc30a35 | 8822 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8823 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8824 | } |
8825 | ||
029190c5 PJ |
8826 | /* Build new domains */ |
8827 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8828 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8829 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8830 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8831 | goto match2; |
8832 | } | |
8833 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8834 | __build_sched_domains(doms_new + i, |
8835 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8836 | match2: |
8837 | ; | |
8838 | } | |
8839 | ||
8840 | /* Remember the new sched domains */ | |
4212823f | 8841 | if (doms_cur != fallback_doms) |
029190c5 | 8842 | kfree(doms_cur); |
1d3504fc | 8843 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8844 | doms_cur = doms_new; |
1d3504fc | 8845 | dattr_cur = dattr_new; |
029190c5 | 8846 | ndoms_cur = ndoms_new; |
7378547f MM |
8847 | |
8848 | register_sched_domain_sysctl(); | |
a1835615 | 8849 | |
712555ee | 8850 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8851 | } |
8852 | ||
5c45bf27 | 8853 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8854 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8855 | { |
95402b38 | 8856 | get_online_cpus(); |
dfb512ec MK |
8857 | |
8858 | /* Destroy domains first to force the rebuild */ | |
8859 | partition_sched_domains(0, NULL, NULL); | |
8860 | ||
e761b772 | 8861 | rebuild_sched_domains(); |
95402b38 | 8862 | put_online_cpus(); |
5c45bf27 SS |
8863 | } |
8864 | ||
8865 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8866 | { | |
afb8a9b7 | 8867 | unsigned int level = 0; |
5c45bf27 | 8868 | |
afb8a9b7 GS |
8869 | if (sscanf(buf, "%u", &level) != 1) |
8870 | return -EINVAL; | |
8871 | ||
8872 | /* | |
8873 | * level is always be positive so don't check for | |
8874 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8875 | * What happens on 0 or 1 byte write, | |
8876 | * need to check for count as well? | |
8877 | */ | |
8878 | ||
8879 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8880 | return -EINVAL; |
8881 | ||
8882 | if (smt) | |
afb8a9b7 | 8883 | sched_smt_power_savings = level; |
5c45bf27 | 8884 | else |
afb8a9b7 | 8885 | sched_mc_power_savings = level; |
5c45bf27 | 8886 | |
c70f22d2 | 8887 | arch_reinit_sched_domains(); |
5c45bf27 | 8888 | |
c70f22d2 | 8889 | return count; |
5c45bf27 SS |
8890 | } |
8891 | ||
5c45bf27 | 8892 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8893 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8894 | char *page) | |
5c45bf27 SS |
8895 | { |
8896 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8897 | } | |
f718cd4a | 8898 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8899 | const char *buf, size_t count) |
5c45bf27 SS |
8900 | { |
8901 | return sched_power_savings_store(buf, count, 0); | |
8902 | } | |
f718cd4a AK |
8903 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8904 | sched_mc_power_savings_show, | |
8905 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8906 | #endif |
8907 | ||
8908 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8909 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8910 | char *page) | |
5c45bf27 SS |
8911 | { |
8912 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8913 | } | |
f718cd4a | 8914 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8915 | const char *buf, size_t count) |
5c45bf27 SS |
8916 | { |
8917 | return sched_power_savings_store(buf, count, 1); | |
8918 | } | |
f718cd4a AK |
8919 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8920 | sched_smt_power_savings_show, | |
6707de00 AB |
8921 | sched_smt_power_savings_store); |
8922 | #endif | |
8923 | ||
39aac648 | 8924 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8925 | { |
8926 | int err = 0; | |
8927 | ||
8928 | #ifdef CONFIG_SCHED_SMT | |
8929 | if (smt_capable()) | |
8930 | err = sysfs_create_file(&cls->kset.kobj, | |
8931 | &attr_sched_smt_power_savings.attr); | |
8932 | #endif | |
8933 | #ifdef CONFIG_SCHED_MC | |
8934 | if (!err && mc_capable()) | |
8935 | err = sysfs_create_file(&cls->kset.kobj, | |
8936 | &attr_sched_mc_power_savings.attr); | |
8937 | #endif | |
8938 | return err; | |
8939 | } | |
6d6bc0ad | 8940 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8941 | |
e761b772 | 8942 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8943 | /* |
e761b772 MK |
8944 | * Add online and remove offline CPUs from the scheduler domains. |
8945 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8946 | */ |
8947 | static int update_sched_domains(struct notifier_block *nfb, | |
8948 | unsigned long action, void *hcpu) | |
e761b772 MK |
8949 | { |
8950 | switch (action) { | |
8951 | case CPU_ONLINE: | |
8952 | case CPU_ONLINE_FROZEN: | |
8953 | case CPU_DEAD: | |
8954 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8955 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8956 | return NOTIFY_OK; |
8957 | ||
8958 | default: | |
8959 | return NOTIFY_DONE; | |
8960 | } | |
8961 | } | |
8962 | #endif | |
8963 | ||
8964 | static int update_runtime(struct notifier_block *nfb, | |
8965 | unsigned long action, void *hcpu) | |
1da177e4 | 8966 | { |
7def2be1 PZ |
8967 | int cpu = (int)(long)hcpu; |
8968 | ||
1da177e4 | 8969 | switch (action) { |
1da177e4 | 8970 | case CPU_DOWN_PREPARE: |
8bb78442 | 8971 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8972 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8973 | return NOTIFY_OK; |
8974 | ||
1da177e4 | 8975 | case CPU_DOWN_FAILED: |
8bb78442 | 8976 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8977 | case CPU_ONLINE: |
8bb78442 | 8978 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8979 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8980 | return NOTIFY_OK; |
8981 | ||
1da177e4 LT |
8982 | default: |
8983 | return NOTIFY_DONE; | |
8984 | } | |
1da177e4 | 8985 | } |
1da177e4 LT |
8986 | |
8987 | void __init sched_init_smp(void) | |
8988 | { | |
dcc30a35 RR |
8989 | cpumask_var_t non_isolated_cpus; |
8990 | ||
8991 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8992 | |
434d53b0 MT |
8993 | #if defined(CONFIG_NUMA) |
8994 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8995 | GFP_KERNEL); | |
8996 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8997 | #endif | |
95402b38 | 8998 | get_online_cpus(); |
712555ee | 8999 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9000 | arch_init_sched_domains(cpu_online_mask); |
9001 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9002 | if (cpumask_empty(non_isolated_cpus)) | |
9003 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9004 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9005 | put_online_cpus(); |
e761b772 MK |
9006 | |
9007 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9008 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9009 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9010 | #endif |
9011 | ||
9012 | /* RT runtime code needs to handle some hotplug events */ | |
9013 | hotcpu_notifier(update_runtime, 0); | |
9014 | ||
b328ca18 | 9015 | init_hrtick(); |
5c1e1767 NP |
9016 | |
9017 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9018 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9019 | BUG(); |
19978ca6 | 9020 | sched_init_granularity(); |
dcc30a35 | 9021 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9022 | |
9023 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9024 | init_sched_rt_class(); |
1da177e4 LT |
9025 | } |
9026 | #else | |
9027 | void __init sched_init_smp(void) | |
9028 | { | |
19978ca6 | 9029 | sched_init_granularity(); |
1da177e4 LT |
9030 | } |
9031 | #endif /* CONFIG_SMP */ | |
9032 | ||
cd1bb94b AB |
9033 | const_debug unsigned int sysctl_timer_migration = 1; |
9034 | ||
1da177e4 LT |
9035 | int in_sched_functions(unsigned long addr) |
9036 | { | |
1da177e4 LT |
9037 | return in_lock_functions(addr) || |
9038 | (addr >= (unsigned long)__sched_text_start | |
9039 | && addr < (unsigned long)__sched_text_end); | |
9040 | } | |
9041 | ||
a9957449 | 9042 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9043 | { |
9044 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9045 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9046 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9047 | cfs_rq->rq = rq; | |
9048 | #endif | |
67e9fb2a | 9049 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9050 | } |
9051 | ||
fa85ae24 PZ |
9052 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9053 | { | |
9054 | struct rt_prio_array *array; | |
9055 | int i; | |
9056 | ||
9057 | array = &rt_rq->active; | |
9058 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9059 | INIT_LIST_HEAD(array->queue + i); | |
9060 | __clear_bit(i, array->bitmap); | |
9061 | } | |
9062 | /* delimiter for bitsearch: */ | |
9063 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9064 | ||
052f1dc7 | 9065 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9066 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9067 | #ifdef CONFIG_SMP |
e864c499 | 9068 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9069 | #endif |
48d5e258 | 9070 | #endif |
fa85ae24 PZ |
9071 | #ifdef CONFIG_SMP |
9072 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9073 | rt_rq->overloaded = 0; |
917b627d | 9074 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9075 | #endif |
9076 | ||
9077 | rt_rq->rt_time = 0; | |
9078 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9079 | rt_rq->rt_runtime = 0; |
9080 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9081 | |
052f1dc7 | 9082 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9083 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9084 | rt_rq->rq = rq; |
9085 | #endif | |
fa85ae24 PZ |
9086 | } |
9087 | ||
6f505b16 | 9088 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9089 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9090 | struct sched_entity *se, int cpu, int add, | |
9091 | struct sched_entity *parent) | |
6f505b16 | 9092 | { |
ec7dc8ac | 9093 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9094 | tg->cfs_rq[cpu] = cfs_rq; |
9095 | init_cfs_rq(cfs_rq, rq); | |
9096 | cfs_rq->tg = tg; | |
9097 | if (add) | |
9098 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9099 | ||
9100 | tg->se[cpu] = se; | |
354d60c2 DG |
9101 | /* se could be NULL for init_task_group */ |
9102 | if (!se) | |
9103 | return; | |
9104 | ||
ec7dc8ac DG |
9105 | if (!parent) |
9106 | se->cfs_rq = &rq->cfs; | |
9107 | else | |
9108 | se->cfs_rq = parent->my_q; | |
9109 | ||
6f505b16 PZ |
9110 | se->my_q = cfs_rq; |
9111 | se->load.weight = tg->shares; | |
e05510d0 | 9112 | se->load.inv_weight = 0; |
ec7dc8ac | 9113 | se->parent = parent; |
6f505b16 | 9114 | } |
052f1dc7 | 9115 | #endif |
6f505b16 | 9116 | |
052f1dc7 | 9117 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9118 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9119 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9120 | struct sched_rt_entity *parent) | |
6f505b16 | 9121 | { |
ec7dc8ac DG |
9122 | struct rq *rq = cpu_rq(cpu); |
9123 | ||
6f505b16 PZ |
9124 | tg->rt_rq[cpu] = rt_rq; |
9125 | init_rt_rq(rt_rq, rq); | |
9126 | rt_rq->tg = tg; | |
9127 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9128 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9129 | if (add) |
9130 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9131 | ||
9132 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9133 | if (!rt_se) |
9134 | return; | |
9135 | ||
ec7dc8ac DG |
9136 | if (!parent) |
9137 | rt_se->rt_rq = &rq->rt; | |
9138 | else | |
9139 | rt_se->rt_rq = parent->my_q; | |
9140 | ||
6f505b16 | 9141 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9142 | rt_se->parent = parent; |
6f505b16 PZ |
9143 | INIT_LIST_HEAD(&rt_se->run_list); |
9144 | } | |
9145 | #endif | |
9146 | ||
1da177e4 LT |
9147 | void __init sched_init(void) |
9148 | { | |
dd41f596 | 9149 | int i, j; |
434d53b0 MT |
9150 | unsigned long alloc_size = 0, ptr; |
9151 | ||
9152 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9153 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9154 | #endif | |
9155 | #ifdef CONFIG_RT_GROUP_SCHED | |
9156 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9157 | #endif |
9158 | #ifdef CONFIG_USER_SCHED | |
9159 | alloc_size *= 2; | |
df7c8e84 RR |
9160 | #endif |
9161 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9162 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9163 | #endif |
9164 | /* | |
9165 | * As sched_init() is called before page_alloc is setup, | |
9166 | * we use alloc_bootmem(). | |
9167 | */ | |
9168 | if (alloc_size) { | |
36b7b6d4 | 9169 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9170 | |
9171 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9172 | init_task_group.se = (struct sched_entity **)ptr; | |
9173 | ptr += nr_cpu_ids * sizeof(void **); | |
9174 | ||
9175 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9176 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9177 | |
9178 | #ifdef CONFIG_USER_SCHED | |
9179 | root_task_group.se = (struct sched_entity **)ptr; | |
9180 | ptr += nr_cpu_ids * sizeof(void **); | |
9181 | ||
9182 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9183 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9184 | #endif /* CONFIG_USER_SCHED */ |
9185 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9186 | #ifdef CONFIG_RT_GROUP_SCHED |
9187 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9188 | ptr += nr_cpu_ids * sizeof(void **); | |
9189 | ||
9190 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9191 | ptr += nr_cpu_ids * sizeof(void **); |
9192 | ||
9193 | #ifdef CONFIG_USER_SCHED | |
9194 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9195 | ptr += nr_cpu_ids * sizeof(void **); | |
9196 | ||
9197 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9198 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9199 | #endif /* CONFIG_USER_SCHED */ |
9200 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9201 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9202 | for_each_possible_cpu(i) { | |
9203 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9204 | ptr += cpumask_size(); | |
9205 | } | |
9206 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9207 | } |
dd41f596 | 9208 | |
57d885fe GH |
9209 | #ifdef CONFIG_SMP |
9210 | init_defrootdomain(); | |
9211 | #endif | |
9212 | ||
d0b27fa7 PZ |
9213 | init_rt_bandwidth(&def_rt_bandwidth, |
9214 | global_rt_period(), global_rt_runtime()); | |
9215 | ||
9216 | #ifdef CONFIG_RT_GROUP_SCHED | |
9217 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9218 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9219 | #ifdef CONFIG_USER_SCHED |
9220 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9221 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9222 | #endif /* CONFIG_USER_SCHED */ |
9223 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9224 | |
052f1dc7 | 9225 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9226 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9227 | INIT_LIST_HEAD(&init_task_group.children); |
9228 | ||
9229 | #ifdef CONFIG_USER_SCHED | |
9230 | INIT_LIST_HEAD(&root_task_group.children); | |
9231 | init_task_group.parent = &root_task_group; | |
9232 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9233 | #endif /* CONFIG_USER_SCHED */ |
9234 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9235 | |
0a945022 | 9236 | for_each_possible_cpu(i) { |
70b97a7f | 9237 | struct rq *rq; |
1da177e4 LT |
9238 | |
9239 | rq = cpu_rq(i); | |
9240 | spin_lock_init(&rq->lock); | |
7897986b | 9241 | rq->nr_running = 0; |
dce48a84 TG |
9242 | rq->calc_load_active = 0; |
9243 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9244 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9245 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9246 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9247 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9248 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9249 | #ifdef CONFIG_CGROUP_SCHED |
9250 | /* | |
9251 | * How much cpu bandwidth does init_task_group get? | |
9252 | * | |
9253 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9254 | * gets 100% of the cpu resources in the system. This overall | |
9255 | * system cpu resource is divided among the tasks of | |
9256 | * init_task_group and its child task-groups in a fair manner, | |
9257 | * based on each entity's (task or task-group's) weight | |
9258 | * (se->load.weight). | |
9259 | * | |
9260 | * In other words, if init_task_group has 10 tasks of weight | |
9261 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9262 | * then A0's share of the cpu resource is: | |
9263 | * | |
0d905bca | 9264 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9265 | * |
9266 | * We achieve this by letting init_task_group's tasks sit | |
9267 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9268 | */ | |
ec7dc8ac | 9269 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9270 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9271 | root_task_group.shares = NICE_0_LOAD; |
9272 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9273 | /* |
9274 | * In case of task-groups formed thr' the user id of tasks, | |
9275 | * init_task_group represents tasks belonging to root user. | |
9276 | * Hence it forms a sibling of all subsequent groups formed. | |
9277 | * In this case, init_task_group gets only a fraction of overall | |
9278 | * system cpu resource, based on the weight assigned to root | |
9279 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9280 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9281 | * (init_cfs_rq) and having one entity represent this group of | |
9282 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9283 | */ | |
ec7dc8ac | 9284 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9285 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9286 | &per_cpu(init_sched_entity, i), i, 1, |
9287 | root_task_group.se[i]); | |
6f505b16 | 9288 | |
052f1dc7 | 9289 | #endif |
354d60c2 DG |
9290 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9291 | ||
9292 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9293 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9294 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9295 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9296 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9297 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9298 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9299 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9300 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9301 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9302 | root_task_group.rt_se[i]); | |
354d60c2 | 9303 | #endif |
dd41f596 | 9304 | #endif |
1da177e4 | 9305 | |
dd41f596 IM |
9306 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9307 | rq->cpu_load[j] = 0; | |
1da177e4 | 9308 | #ifdef CONFIG_SMP |
41c7ce9a | 9309 | rq->sd = NULL; |
57d885fe | 9310 | rq->rd = NULL; |
1da177e4 | 9311 | rq->active_balance = 0; |
dd41f596 | 9312 | rq->next_balance = jiffies; |
1da177e4 | 9313 | rq->push_cpu = 0; |
0a2966b4 | 9314 | rq->cpu = i; |
1f11eb6a | 9315 | rq->online = 0; |
1da177e4 LT |
9316 | rq->migration_thread = NULL; |
9317 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9318 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9319 | #endif |
8f4d37ec | 9320 | init_rq_hrtick(rq); |
1da177e4 | 9321 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9322 | } |
9323 | ||
2dd73a4f | 9324 | set_load_weight(&init_task); |
b50f60ce | 9325 | |
e107be36 AK |
9326 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9327 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9328 | #endif | |
9329 | ||
c9819f45 | 9330 | #ifdef CONFIG_SMP |
962cf36c | 9331 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9332 | #endif |
9333 | ||
b50f60ce HC |
9334 | #ifdef CONFIG_RT_MUTEXES |
9335 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9336 | #endif | |
9337 | ||
1da177e4 LT |
9338 | /* |
9339 | * The boot idle thread does lazy MMU switching as well: | |
9340 | */ | |
9341 | atomic_inc(&init_mm.mm_count); | |
9342 | enter_lazy_tlb(&init_mm, current); | |
9343 | ||
9344 | /* | |
9345 | * Make us the idle thread. Technically, schedule() should not be | |
9346 | * called from this thread, however somewhere below it might be, | |
9347 | * but because we are the idle thread, we just pick up running again | |
9348 | * when this runqueue becomes "idle". | |
9349 | */ | |
9350 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9351 | |
9352 | calc_load_update = jiffies + LOAD_FREQ; | |
9353 | ||
dd41f596 IM |
9354 | /* |
9355 | * During early bootup we pretend to be a normal task: | |
9356 | */ | |
9357 | current->sched_class = &fair_sched_class; | |
6892b75e | 9358 | |
6a7b3dc3 | 9359 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9360 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9361 | #ifdef CONFIG_SMP |
7d1e6a9b | 9362 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9363 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9364 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9365 | #endif |
4bdddf8f | 9366 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9367 | #endif /* SMP */ |
6a7b3dc3 | 9368 | |
0d905bca IM |
9369 | perf_counter_init(); |
9370 | ||
6892b75e | 9371 | scheduler_running = 1; |
1da177e4 LT |
9372 | } |
9373 | ||
9374 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9375 | void __might_sleep(char *file, int line) | |
9376 | { | |
48f24c4d | 9377 | #ifdef in_atomic |
1da177e4 LT |
9378 | static unsigned long prev_jiffy; /* ratelimiting */ |
9379 | ||
aef745fc IM |
9380 | if ((!in_atomic() && !irqs_disabled()) || |
9381 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9382 | return; | |
9383 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9384 | return; | |
9385 | prev_jiffy = jiffies; | |
9386 | ||
9387 | printk(KERN_ERR | |
9388 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9389 | file, line); | |
9390 | printk(KERN_ERR | |
9391 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9392 | in_atomic(), irqs_disabled(), | |
9393 | current->pid, current->comm); | |
9394 | ||
9395 | debug_show_held_locks(current); | |
9396 | if (irqs_disabled()) | |
9397 | print_irqtrace_events(current); | |
9398 | dump_stack(); | |
1da177e4 LT |
9399 | #endif |
9400 | } | |
9401 | EXPORT_SYMBOL(__might_sleep); | |
9402 | #endif | |
9403 | ||
9404 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9405 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9406 | { | |
9407 | int on_rq; | |
3e51f33f | 9408 | |
3a5e4dc1 AK |
9409 | update_rq_clock(rq); |
9410 | on_rq = p->se.on_rq; | |
9411 | if (on_rq) | |
9412 | deactivate_task(rq, p, 0); | |
9413 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9414 | if (on_rq) { | |
9415 | activate_task(rq, p, 0); | |
9416 | resched_task(rq->curr); | |
9417 | } | |
9418 | } | |
9419 | ||
1da177e4 LT |
9420 | void normalize_rt_tasks(void) |
9421 | { | |
a0f98a1c | 9422 | struct task_struct *g, *p; |
1da177e4 | 9423 | unsigned long flags; |
70b97a7f | 9424 | struct rq *rq; |
1da177e4 | 9425 | |
4cf5d77a | 9426 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9427 | do_each_thread(g, p) { |
178be793 IM |
9428 | /* |
9429 | * Only normalize user tasks: | |
9430 | */ | |
9431 | if (!p->mm) | |
9432 | continue; | |
9433 | ||
6cfb0d5d | 9434 | p->se.exec_start = 0; |
6cfb0d5d | 9435 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9436 | p->se.wait_start = 0; |
dd41f596 | 9437 | p->se.sleep_start = 0; |
dd41f596 | 9438 | p->se.block_start = 0; |
6cfb0d5d | 9439 | #endif |
dd41f596 IM |
9440 | |
9441 | if (!rt_task(p)) { | |
9442 | /* | |
9443 | * Renice negative nice level userspace | |
9444 | * tasks back to 0: | |
9445 | */ | |
9446 | if (TASK_NICE(p) < 0 && p->mm) | |
9447 | set_user_nice(p, 0); | |
1da177e4 | 9448 | continue; |
dd41f596 | 9449 | } |
1da177e4 | 9450 | |
4cf5d77a | 9451 | spin_lock(&p->pi_lock); |
b29739f9 | 9452 | rq = __task_rq_lock(p); |
1da177e4 | 9453 | |
178be793 | 9454 | normalize_task(rq, p); |
3a5e4dc1 | 9455 | |
b29739f9 | 9456 | __task_rq_unlock(rq); |
4cf5d77a | 9457 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9458 | } while_each_thread(g, p); |
9459 | ||
4cf5d77a | 9460 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9461 | } |
9462 | ||
9463 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9464 | |
9465 | #ifdef CONFIG_IA64 | |
9466 | /* | |
9467 | * These functions are only useful for the IA64 MCA handling. | |
9468 | * | |
9469 | * They can only be called when the whole system has been | |
9470 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9471 | * activity can take place. Using them for anything else would | |
9472 | * be a serious bug, and as a result, they aren't even visible | |
9473 | * under any other configuration. | |
9474 | */ | |
9475 | ||
9476 | /** | |
9477 | * curr_task - return the current task for a given cpu. | |
9478 | * @cpu: the processor in question. | |
9479 | * | |
9480 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9481 | */ | |
36c8b586 | 9482 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9483 | { |
9484 | return cpu_curr(cpu); | |
9485 | } | |
9486 | ||
9487 | /** | |
9488 | * set_curr_task - set the current task for a given cpu. | |
9489 | * @cpu: the processor in question. | |
9490 | * @p: the task pointer to set. | |
9491 | * | |
9492 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9493 | * are serviced on a separate stack. It allows the architecture to switch the |
9494 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9495 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9496 | * and caller must save the original value of the current task (see | |
9497 | * curr_task() above) and restore that value before reenabling interrupts and | |
9498 | * re-starting the system. | |
9499 | * | |
9500 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9501 | */ | |
36c8b586 | 9502 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9503 | { |
9504 | cpu_curr(cpu) = p; | |
9505 | } | |
9506 | ||
9507 | #endif | |
29f59db3 | 9508 | |
bccbe08a PZ |
9509 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9510 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9511 | { |
9512 | int i; | |
9513 | ||
9514 | for_each_possible_cpu(i) { | |
9515 | if (tg->cfs_rq) | |
9516 | kfree(tg->cfs_rq[i]); | |
9517 | if (tg->se) | |
9518 | kfree(tg->se[i]); | |
6f505b16 PZ |
9519 | } |
9520 | ||
9521 | kfree(tg->cfs_rq); | |
9522 | kfree(tg->se); | |
6f505b16 PZ |
9523 | } |
9524 | ||
ec7dc8ac DG |
9525 | static |
9526 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9527 | { |
29f59db3 | 9528 | struct cfs_rq *cfs_rq; |
eab17229 | 9529 | struct sched_entity *se; |
9b5b7751 | 9530 | struct rq *rq; |
29f59db3 SV |
9531 | int i; |
9532 | ||
434d53b0 | 9533 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9534 | if (!tg->cfs_rq) |
9535 | goto err; | |
434d53b0 | 9536 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9537 | if (!tg->se) |
9538 | goto err; | |
052f1dc7 PZ |
9539 | |
9540 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9541 | |
9542 | for_each_possible_cpu(i) { | |
9b5b7751 | 9543 | rq = cpu_rq(i); |
29f59db3 | 9544 | |
eab17229 LZ |
9545 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9546 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9547 | if (!cfs_rq) |
9548 | goto err; | |
9549 | ||
eab17229 LZ |
9550 | se = kzalloc_node(sizeof(struct sched_entity), |
9551 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9552 | if (!se) |
9553 | goto err; | |
9554 | ||
eab17229 | 9555 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9556 | } |
9557 | ||
9558 | return 1; | |
9559 | ||
9560 | err: | |
9561 | return 0; | |
9562 | } | |
9563 | ||
9564 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9565 | { | |
9566 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9567 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9568 | } | |
9569 | ||
9570 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9571 | { | |
9572 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9573 | } | |
6d6bc0ad | 9574 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9575 | static inline void free_fair_sched_group(struct task_group *tg) |
9576 | { | |
9577 | } | |
9578 | ||
ec7dc8ac DG |
9579 | static inline |
9580 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9581 | { |
9582 | return 1; | |
9583 | } | |
9584 | ||
9585 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9586 | { | |
9587 | } | |
9588 | ||
9589 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9590 | { | |
9591 | } | |
6d6bc0ad | 9592 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9593 | |
9594 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9595 | static void free_rt_sched_group(struct task_group *tg) |
9596 | { | |
9597 | int i; | |
9598 | ||
d0b27fa7 PZ |
9599 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9600 | ||
bccbe08a PZ |
9601 | for_each_possible_cpu(i) { |
9602 | if (tg->rt_rq) | |
9603 | kfree(tg->rt_rq[i]); | |
9604 | if (tg->rt_se) | |
9605 | kfree(tg->rt_se[i]); | |
9606 | } | |
9607 | ||
9608 | kfree(tg->rt_rq); | |
9609 | kfree(tg->rt_se); | |
9610 | } | |
9611 | ||
ec7dc8ac DG |
9612 | static |
9613 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9614 | { |
9615 | struct rt_rq *rt_rq; | |
eab17229 | 9616 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9617 | struct rq *rq; |
9618 | int i; | |
9619 | ||
434d53b0 | 9620 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9621 | if (!tg->rt_rq) |
9622 | goto err; | |
434d53b0 | 9623 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9624 | if (!tg->rt_se) |
9625 | goto err; | |
9626 | ||
d0b27fa7 PZ |
9627 | init_rt_bandwidth(&tg->rt_bandwidth, |
9628 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9629 | |
9630 | for_each_possible_cpu(i) { | |
9631 | rq = cpu_rq(i); | |
9632 | ||
eab17229 LZ |
9633 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9634 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9635 | if (!rt_rq) |
9636 | goto err; | |
29f59db3 | 9637 | |
eab17229 LZ |
9638 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9639 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9640 | if (!rt_se) |
9641 | goto err; | |
29f59db3 | 9642 | |
eab17229 | 9643 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9644 | } |
9645 | ||
bccbe08a PZ |
9646 | return 1; |
9647 | ||
9648 | err: | |
9649 | return 0; | |
9650 | } | |
9651 | ||
9652 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9653 | { | |
9654 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9655 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9656 | } | |
9657 | ||
9658 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9659 | { | |
9660 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9661 | } | |
6d6bc0ad | 9662 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9663 | static inline void free_rt_sched_group(struct task_group *tg) |
9664 | { | |
9665 | } | |
9666 | ||
ec7dc8ac DG |
9667 | static inline |
9668 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9669 | { |
9670 | return 1; | |
9671 | } | |
9672 | ||
9673 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9674 | { | |
9675 | } | |
9676 | ||
9677 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9678 | { | |
9679 | } | |
6d6bc0ad | 9680 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9681 | |
d0b27fa7 | 9682 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9683 | static void free_sched_group(struct task_group *tg) |
9684 | { | |
9685 | free_fair_sched_group(tg); | |
9686 | free_rt_sched_group(tg); | |
9687 | kfree(tg); | |
9688 | } | |
9689 | ||
9690 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9691 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9692 | { |
9693 | struct task_group *tg; | |
9694 | unsigned long flags; | |
9695 | int i; | |
9696 | ||
9697 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9698 | if (!tg) | |
9699 | return ERR_PTR(-ENOMEM); | |
9700 | ||
ec7dc8ac | 9701 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9702 | goto err; |
9703 | ||
ec7dc8ac | 9704 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9705 | goto err; |
9706 | ||
8ed36996 | 9707 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9708 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9709 | register_fair_sched_group(tg, i); |
9710 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9711 | } |
6f505b16 | 9712 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9713 | |
9714 | WARN_ON(!parent); /* root should already exist */ | |
9715 | ||
9716 | tg->parent = parent; | |
f473aa5e | 9717 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9718 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9719 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9720 | |
9b5b7751 | 9721 | return tg; |
29f59db3 SV |
9722 | |
9723 | err: | |
6f505b16 | 9724 | free_sched_group(tg); |
29f59db3 SV |
9725 | return ERR_PTR(-ENOMEM); |
9726 | } | |
9727 | ||
9b5b7751 | 9728 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9729 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9730 | { |
29f59db3 | 9731 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9732 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9733 | } |
9734 | ||
9b5b7751 | 9735 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9736 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9737 | { |
8ed36996 | 9738 | unsigned long flags; |
9b5b7751 | 9739 | int i; |
29f59db3 | 9740 | |
8ed36996 | 9741 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9742 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9743 | unregister_fair_sched_group(tg, i); |
9744 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9745 | } |
6f505b16 | 9746 | list_del_rcu(&tg->list); |
f473aa5e | 9747 | list_del_rcu(&tg->siblings); |
8ed36996 | 9748 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9749 | |
9b5b7751 | 9750 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9751 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9752 | } |
9753 | ||
9b5b7751 | 9754 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9755 | * The caller of this function should have put the task in its new group |
9756 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9757 | * reflect its new group. | |
9b5b7751 SV |
9758 | */ |
9759 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9760 | { |
9761 | int on_rq, running; | |
9762 | unsigned long flags; | |
9763 | struct rq *rq; | |
9764 | ||
9765 | rq = task_rq_lock(tsk, &flags); | |
9766 | ||
29f59db3 SV |
9767 | update_rq_clock(rq); |
9768 | ||
051a1d1a | 9769 | running = task_current(rq, tsk); |
29f59db3 SV |
9770 | on_rq = tsk->se.on_rq; |
9771 | ||
0e1f3483 | 9772 | if (on_rq) |
29f59db3 | 9773 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9774 | if (unlikely(running)) |
9775 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9776 | |
6f505b16 | 9777 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9778 | |
810b3817 PZ |
9779 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9780 | if (tsk->sched_class->moved_group) | |
9781 | tsk->sched_class->moved_group(tsk); | |
9782 | #endif | |
9783 | ||
0e1f3483 HS |
9784 | if (unlikely(running)) |
9785 | tsk->sched_class->set_curr_task(rq); | |
9786 | if (on_rq) | |
7074badb | 9787 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9788 | |
29f59db3 SV |
9789 | task_rq_unlock(rq, &flags); |
9790 | } | |
6d6bc0ad | 9791 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9792 | |
052f1dc7 | 9793 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9794 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9795 | { |
9796 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9797 | int on_rq; |
9798 | ||
29f59db3 | 9799 | on_rq = se->on_rq; |
62fb1851 | 9800 | if (on_rq) |
29f59db3 SV |
9801 | dequeue_entity(cfs_rq, se, 0); |
9802 | ||
9803 | se->load.weight = shares; | |
e05510d0 | 9804 | se->load.inv_weight = 0; |
29f59db3 | 9805 | |
62fb1851 | 9806 | if (on_rq) |
29f59db3 | 9807 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9808 | } |
62fb1851 | 9809 | |
c09595f6 PZ |
9810 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9811 | { | |
9812 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9813 | struct rq *rq = cfs_rq->rq; | |
9814 | unsigned long flags; | |
9815 | ||
9816 | spin_lock_irqsave(&rq->lock, flags); | |
9817 | __set_se_shares(se, shares); | |
9818 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9819 | } |
9820 | ||
8ed36996 PZ |
9821 | static DEFINE_MUTEX(shares_mutex); |
9822 | ||
4cf86d77 | 9823 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9824 | { |
9825 | int i; | |
8ed36996 | 9826 | unsigned long flags; |
c61935fd | 9827 | |
ec7dc8ac DG |
9828 | /* |
9829 | * We can't change the weight of the root cgroup. | |
9830 | */ | |
9831 | if (!tg->se[0]) | |
9832 | return -EINVAL; | |
9833 | ||
18d95a28 PZ |
9834 | if (shares < MIN_SHARES) |
9835 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9836 | else if (shares > MAX_SHARES) |
9837 | shares = MAX_SHARES; | |
62fb1851 | 9838 | |
8ed36996 | 9839 | mutex_lock(&shares_mutex); |
9b5b7751 | 9840 | if (tg->shares == shares) |
5cb350ba | 9841 | goto done; |
29f59db3 | 9842 | |
8ed36996 | 9843 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9844 | for_each_possible_cpu(i) |
9845 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9846 | list_del_rcu(&tg->siblings); |
8ed36996 | 9847 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9848 | |
9849 | /* wait for any ongoing reference to this group to finish */ | |
9850 | synchronize_sched(); | |
9851 | ||
9852 | /* | |
9853 | * Now we are free to modify the group's share on each cpu | |
9854 | * w/o tripping rebalance_share or load_balance_fair. | |
9855 | */ | |
9b5b7751 | 9856 | tg->shares = shares; |
c09595f6 PZ |
9857 | for_each_possible_cpu(i) { |
9858 | /* | |
9859 | * force a rebalance | |
9860 | */ | |
9861 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9862 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9863 | } |
29f59db3 | 9864 | |
6b2d7700 SV |
9865 | /* |
9866 | * Enable load balance activity on this group, by inserting it back on | |
9867 | * each cpu's rq->leaf_cfs_rq_list. | |
9868 | */ | |
8ed36996 | 9869 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9870 | for_each_possible_cpu(i) |
9871 | register_fair_sched_group(tg, i); | |
f473aa5e | 9872 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9873 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9874 | done: |
8ed36996 | 9875 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9876 | return 0; |
29f59db3 SV |
9877 | } |
9878 | ||
5cb350ba DG |
9879 | unsigned long sched_group_shares(struct task_group *tg) |
9880 | { | |
9881 | return tg->shares; | |
9882 | } | |
052f1dc7 | 9883 | #endif |
5cb350ba | 9884 | |
052f1dc7 | 9885 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9886 | /* |
9f0c1e56 | 9887 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9888 | */ |
9f0c1e56 PZ |
9889 | static DEFINE_MUTEX(rt_constraints_mutex); |
9890 | ||
9891 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9892 | { | |
9893 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9894 | return 1ULL << 20; |
9f0c1e56 | 9895 | |
9a7e0b18 | 9896 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9897 | } |
9898 | ||
9a7e0b18 PZ |
9899 | /* Must be called with tasklist_lock held */ |
9900 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9901 | { |
9a7e0b18 | 9902 | struct task_struct *g, *p; |
b40b2e8e | 9903 | |
9a7e0b18 PZ |
9904 | do_each_thread(g, p) { |
9905 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9906 | return 1; | |
9907 | } while_each_thread(g, p); | |
b40b2e8e | 9908 | |
9a7e0b18 PZ |
9909 | return 0; |
9910 | } | |
b40b2e8e | 9911 | |
9a7e0b18 PZ |
9912 | struct rt_schedulable_data { |
9913 | struct task_group *tg; | |
9914 | u64 rt_period; | |
9915 | u64 rt_runtime; | |
9916 | }; | |
b40b2e8e | 9917 | |
9a7e0b18 PZ |
9918 | static int tg_schedulable(struct task_group *tg, void *data) |
9919 | { | |
9920 | struct rt_schedulable_data *d = data; | |
9921 | struct task_group *child; | |
9922 | unsigned long total, sum = 0; | |
9923 | u64 period, runtime; | |
b40b2e8e | 9924 | |
9a7e0b18 PZ |
9925 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9926 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9927 | |
9a7e0b18 PZ |
9928 | if (tg == d->tg) { |
9929 | period = d->rt_period; | |
9930 | runtime = d->rt_runtime; | |
b40b2e8e | 9931 | } |
b40b2e8e | 9932 | |
98a4826b PZ |
9933 | #ifdef CONFIG_USER_SCHED |
9934 | if (tg == &root_task_group) { | |
9935 | period = global_rt_period(); | |
9936 | runtime = global_rt_runtime(); | |
9937 | } | |
9938 | #endif | |
9939 | ||
4653f803 PZ |
9940 | /* |
9941 | * Cannot have more runtime than the period. | |
9942 | */ | |
9943 | if (runtime > period && runtime != RUNTIME_INF) | |
9944 | return -EINVAL; | |
6f505b16 | 9945 | |
4653f803 PZ |
9946 | /* |
9947 | * Ensure we don't starve existing RT tasks. | |
9948 | */ | |
9a7e0b18 PZ |
9949 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9950 | return -EBUSY; | |
6f505b16 | 9951 | |
9a7e0b18 | 9952 | total = to_ratio(period, runtime); |
6f505b16 | 9953 | |
4653f803 PZ |
9954 | /* |
9955 | * Nobody can have more than the global setting allows. | |
9956 | */ | |
9957 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9958 | return -EINVAL; | |
6f505b16 | 9959 | |
4653f803 PZ |
9960 | /* |
9961 | * The sum of our children's runtime should not exceed our own. | |
9962 | */ | |
9a7e0b18 PZ |
9963 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9964 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9965 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9966 | |
9a7e0b18 PZ |
9967 | if (child == d->tg) { |
9968 | period = d->rt_period; | |
9969 | runtime = d->rt_runtime; | |
9970 | } | |
6f505b16 | 9971 | |
9a7e0b18 | 9972 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9973 | } |
6f505b16 | 9974 | |
9a7e0b18 PZ |
9975 | if (sum > total) |
9976 | return -EINVAL; | |
9977 | ||
9978 | return 0; | |
6f505b16 PZ |
9979 | } |
9980 | ||
9a7e0b18 | 9981 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9982 | { |
9a7e0b18 PZ |
9983 | struct rt_schedulable_data data = { |
9984 | .tg = tg, | |
9985 | .rt_period = period, | |
9986 | .rt_runtime = runtime, | |
9987 | }; | |
9988 | ||
9989 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9990 | } |
9991 | ||
d0b27fa7 PZ |
9992 | static int tg_set_bandwidth(struct task_group *tg, |
9993 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9994 | { |
ac086bc2 | 9995 | int i, err = 0; |
9f0c1e56 | 9996 | |
9f0c1e56 | 9997 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9998 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9999 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10000 | if (err) | |
9f0c1e56 | 10001 | goto unlock; |
ac086bc2 PZ |
10002 | |
10003 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10004 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10005 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10006 | |
10007 | for_each_possible_cpu(i) { | |
10008 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10009 | ||
10010 | spin_lock(&rt_rq->rt_runtime_lock); | |
10011 | rt_rq->rt_runtime = rt_runtime; | |
10012 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10013 | } | |
10014 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10015 | unlock: |
521f1a24 | 10016 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10017 | mutex_unlock(&rt_constraints_mutex); |
10018 | ||
10019 | return err; | |
6f505b16 PZ |
10020 | } |
10021 | ||
d0b27fa7 PZ |
10022 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10023 | { | |
10024 | u64 rt_runtime, rt_period; | |
10025 | ||
10026 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10027 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10028 | if (rt_runtime_us < 0) | |
10029 | rt_runtime = RUNTIME_INF; | |
10030 | ||
10031 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10032 | } | |
10033 | ||
9f0c1e56 PZ |
10034 | long sched_group_rt_runtime(struct task_group *tg) |
10035 | { | |
10036 | u64 rt_runtime_us; | |
10037 | ||
d0b27fa7 | 10038 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10039 | return -1; |
10040 | ||
d0b27fa7 | 10041 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10042 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10043 | return rt_runtime_us; | |
10044 | } | |
d0b27fa7 PZ |
10045 | |
10046 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10047 | { | |
10048 | u64 rt_runtime, rt_period; | |
10049 | ||
10050 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10051 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10052 | ||
619b0488 R |
10053 | if (rt_period == 0) |
10054 | return -EINVAL; | |
10055 | ||
d0b27fa7 PZ |
10056 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10057 | } | |
10058 | ||
10059 | long sched_group_rt_period(struct task_group *tg) | |
10060 | { | |
10061 | u64 rt_period_us; | |
10062 | ||
10063 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10064 | do_div(rt_period_us, NSEC_PER_USEC); | |
10065 | return rt_period_us; | |
10066 | } | |
10067 | ||
10068 | static int sched_rt_global_constraints(void) | |
10069 | { | |
4653f803 | 10070 | u64 runtime, period; |
d0b27fa7 PZ |
10071 | int ret = 0; |
10072 | ||
ec5d4989 HS |
10073 | if (sysctl_sched_rt_period <= 0) |
10074 | return -EINVAL; | |
10075 | ||
4653f803 PZ |
10076 | runtime = global_rt_runtime(); |
10077 | period = global_rt_period(); | |
10078 | ||
10079 | /* | |
10080 | * Sanity check on the sysctl variables. | |
10081 | */ | |
10082 | if (runtime > period && runtime != RUNTIME_INF) | |
10083 | return -EINVAL; | |
10b612f4 | 10084 | |
d0b27fa7 | 10085 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10086 | read_lock(&tasklist_lock); |
4653f803 | 10087 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10088 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10089 | mutex_unlock(&rt_constraints_mutex); |
10090 | ||
10091 | return ret; | |
10092 | } | |
54e99124 DG |
10093 | |
10094 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10095 | { | |
10096 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10097 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10098 | return 0; | |
10099 | ||
10100 | return 1; | |
10101 | } | |
10102 | ||
6d6bc0ad | 10103 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10104 | static int sched_rt_global_constraints(void) |
10105 | { | |
ac086bc2 PZ |
10106 | unsigned long flags; |
10107 | int i; | |
10108 | ||
ec5d4989 HS |
10109 | if (sysctl_sched_rt_period <= 0) |
10110 | return -EINVAL; | |
10111 | ||
60aa605d PZ |
10112 | /* |
10113 | * There's always some RT tasks in the root group | |
10114 | * -- migration, kstopmachine etc.. | |
10115 | */ | |
10116 | if (sysctl_sched_rt_runtime == 0) | |
10117 | return -EBUSY; | |
10118 | ||
ac086bc2 PZ |
10119 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10120 | for_each_possible_cpu(i) { | |
10121 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10122 | ||
10123 | spin_lock(&rt_rq->rt_runtime_lock); | |
10124 | rt_rq->rt_runtime = global_rt_runtime(); | |
10125 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10126 | } | |
10127 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10128 | ||
d0b27fa7 PZ |
10129 | return 0; |
10130 | } | |
6d6bc0ad | 10131 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10132 | |
10133 | int sched_rt_handler(struct ctl_table *table, int write, | |
10134 | struct file *filp, void __user *buffer, size_t *lenp, | |
10135 | loff_t *ppos) | |
10136 | { | |
10137 | int ret; | |
10138 | int old_period, old_runtime; | |
10139 | static DEFINE_MUTEX(mutex); | |
10140 | ||
10141 | mutex_lock(&mutex); | |
10142 | old_period = sysctl_sched_rt_period; | |
10143 | old_runtime = sysctl_sched_rt_runtime; | |
10144 | ||
10145 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10146 | ||
10147 | if (!ret && write) { | |
10148 | ret = sched_rt_global_constraints(); | |
10149 | if (ret) { | |
10150 | sysctl_sched_rt_period = old_period; | |
10151 | sysctl_sched_rt_runtime = old_runtime; | |
10152 | } else { | |
10153 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10154 | def_rt_bandwidth.rt_period = | |
10155 | ns_to_ktime(global_rt_period()); | |
10156 | } | |
10157 | } | |
10158 | mutex_unlock(&mutex); | |
10159 | ||
10160 | return ret; | |
10161 | } | |
68318b8e | 10162 | |
052f1dc7 | 10163 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10164 | |
10165 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10166 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10167 | { |
2b01dfe3 PM |
10168 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10169 | struct task_group, css); | |
68318b8e SV |
10170 | } |
10171 | ||
10172 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10173 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10174 | { |
ec7dc8ac | 10175 | struct task_group *tg, *parent; |
68318b8e | 10176 | |
2b01dfe3 | 10177 | if (!cgrp->parent) { |
68318b8e | 10178 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10179 | return &init_task_group.css; |
10180 | } | |
10181 | ||
ec7dc8ac DG |
10182 | parent = cgroup_tg(cgrp->parent); |
10183 | tg = sched_create_group(parent); | |
68318b8e SV |
10184 | if (IS_ERR(tg)) |
10185 | return ERR_PTR(-ENOMEM); | |
10186 | ||
68318b8e SV |
10187 | return &tg->css; |
10188 | } | |
10189 | ||
41a2d6cf IM |
10190 | static void |
10191 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10192 | { |
2b01dfe3 | 10193 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10194 | |
10195 | sched_destroy_group(tg); | |
10196 | } | |
10197 | ||
41a2d6cf IM |
10198 | static int |
10199 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10200 | struct task_struct *tsk) | |
68318b8e | 10201 | { |
b68aa230 | 10202 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10203 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10204 | return -EINVAL; |
10205 | #else | |
68318b8e SV |
10206 | /* We don't support RT-tasks being in separate groups */ |
10207 | if (tsk->sched_class != &fair_sched_class) | |
10208 | return -EINVAL; | |
b68aa230 | 10209 | #endif |
68318b8e SV |
10210 | |
10211 | return 0; | |
10212 | } | |
10213 | ||
10214 | static void | |
2b01dfe3 | 10215 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10216 | struct cgroup *old_cont, struct task_struct *tsk) |
10217 | { | |
10218 | sched_move_task(tsk); | |
10219 | } | |
10220 | ||
052f1dc7 | 10221 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10222 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10223 | u64 shareval) |
68318b8e | 10224 | { |
2b01dfe3 | 10225 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10226 | } |
10227 | ||
f4c753b7 | 10228 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10229 | { |
2b01dfe3 | 10230 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10231 | |
10232 | return (u64) tg->shares; | |
10233 | } | |
6d6bc0ad | 10234 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10235 | |
052f1dc7 | 10236 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10237 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10238 | s64 val) |
6f505b16 | 10239 | { |
06ecb27c | 10240 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10241 | } |
10242 | ||
06ecb27c | 10243 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10244 | { |
06ecb27c | 10245 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10246 | } |
d0b27fa7 PZ |
10247 | |
10248 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10249 | u64 rt_period_us) | |
10250 | { | |
10251 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10252 | } | |
10253 | ||
10254 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10255 | { | |
10256 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10257 | } | |
6d6bc0ad | 10258 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10259 | |
fe5c7cc2 | 10260 | static struct cftype cpu_files[] = { |
052f1dc7 | 10261 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10262 | { |
10263 | .name = "shares", | |
f4c753b7 PM |
10264 | .read_u64 = cpu_shares_read_u64, |
10265 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10266 | }, |
052f1dc7 PZ |
10267 | #endif |
10268 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10269 | { |
9f0c1e56 | 10270 | .name = "rt_runtime_us", |
06ecb27c PM |
10271 | .read_s64 = cpu_rt_runtime_read, |
10272 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10273 | }, |
d0b27fa7 PZ |
10274 | { |
10275 | .name = "rt_period_us", | |
f4c753b7 PM |
10276 | .read_u64 = cpu_rt_period_read_uint, |
10277 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10278 | }, |
052f1dc7 | 10279 | #endif |
68318b8e SV |
10280 | }; |
10281 | ||
10282 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10283 | { | |
fe5c7cc2 | 10284 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10285 | } |
10286 | ||
10287 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10288 | .name = "cpu", |
10289 | .create = cpu_cgroup_create, | |
10290 | .destroy = cpu_cgroup_destroy, | |
10291 | .can_attach = cpu_cgroup_can_attach, | |
10292 | .attach = cpu_cgroup_attach, | |
10293 | .populate = cpu_cgroup_populate, | |
10294 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10295 | .early_init = 1, |
10296 | }; | |
10297 | ||
052f1dc7 | 10298 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10299 | |
10300 | #ifdef CONFIG_CGROUP_CPUACCT | |
10301 | ||
10302 | /* | |
10303 | * CPU accounting code for task groups. | |
10304 | * | |
10305 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10306 | * (balbir@in.ibm.com). | |
10307 | */ | |
10308 | ||
934352f2 | 10309 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10310 | struct cpuacct { |
10311 | struct cgroup_subsys_state css; | |
10312 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10313 | u64 *cpuusage; | |
ef12fefa | 10314 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10315 | struct cpuacct *parent; |
d842de87 SV |
10316 | }; |
10317 | ||
10318 | struct cgroup_subsys cpuacct_subsys; | |
10319 | ||
10320 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10321 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10322 | { |
32cd756a | 10323 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10324 | struct cpuacct, css); |
10325 | } | |
10326 | ||
10327 | /* return cpu accounting group to which this task belongs */ | |
10328 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10329 | { | |
10330 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10331 | struct cpuacct, css); | |
10332 | } | |
10333 | ||
10334 | /* create a new cpu accounting group */ | |
10335 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10336 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10337 | { |
10338 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10339 | int i; |
d842de87 SV |
10340 | |
10341 | if (!ca) | |
ef12fefa | 10342 | goto out; |
d842de87 SV |
10343 | |
10344 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10345 | if (!ca->cpuusage) |
10346 | goto out_free_ca; | |
10347 | ||
10348 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10349 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10350 | goto out_free_counters; | |
d842de87 | 10351 | |
934352f2 BR |
10352 | if (cgrp->parent) |
10353 | ca->parent = cgroup_ca(cgrp->parent); | |
10354 | ||
d842de87 | 10355 | return &ca->css; |
ef12fefa BR |
10356 | |
10357 | out_free_counters: | |
10358 | while (--i >= 0) | |
10359 | percpu_counter_destroy(&ca->cpustat[i]); | |
10360 | free_percpu(ca->cpuusage); | |
10361 | out_free_ca: | |
10362 | kfree(ca); | |
10363 | out: | |
10364 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10365 | } |
10366 | ||
10367 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10368 | static void |
32cd756a | 10369 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10370 | { |
32cd756a | 10371 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10372 | int i; |
d842de87 | 10373 | |
ef12fefa BR |
10374 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10375 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10376 | free_percpu(ca->cpuusage); |
10377 | kfree(ca); | |
10378 | } | |
10379 | ||
720f5498 KC |
10380 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10381 | { | |
b36128c8 | 10382 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10383 | u64 data; |
10384 | ||
10385 | #ifndef CONFIG_64BIT | |
10386 | /* | |
10387 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10388 | */ | |
10389 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10390 | data = *cpuusage; | |
10391 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10392 | #else | |
10393 | data = *cpuusage; | |
10394 | #endif | |
10395 | ||
10396 | return data; | |
10397 | } | |
10398 | ||
10399 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10400 | { | |
b36128c8 | 10401 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10402 | |
10403 | #ifndef CONFIG_64BIT | |
10404 | /* | |
10405 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10406 | */ | |
10407 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10408 | *cpuusage = val; | |
10409 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10410 | #else | |
10411 | *cpuusage = val; | |
10412 | #endif | |
10413 | } | |
10414 | ||
d842de87 | 10415 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10416 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10417 | { |
32cd756a | 10418 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10419 | u64 totalcpuusage = 0; |
10420 | int i; | |
10421 | ||
720f5498 KC |
10422 | for_each_present_cpu(i) |
10423 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10424 | |
10425 | return totalcpuusage; | |
10426 | } | |
10427 | ||
0297b803 DG |
10428 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10429 | u64 reset) | |
10430 | { | |
10431 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10432 | int err = 0; | |
10433 | int i; | |
10434 | ||
10435 | if (reset) { | |
10436 | err = -EINVAL; | |
10437 | goto out; | |
10438 | } | |
10439 | ||
720f5498 KC |
10440 | for_each_present_cpu(i) |
10441 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10442 | |
0297b803 DG |
10443 | out: |
10444 | return err; | |
10445 | } | |
10446 | ||
e9515c3c KC |
10447 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10448 | struct seq_file *m) | |
10449 | { | |
10450 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10451 | u64 percpu; | |
10452 | int i; | |
10453 | ||
10454 | for_each_present_cpu(i) { | |
10455 | percpu = cpuacct_cpuusage_read(ca, i); | |
10456 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10457 | } | |
10458 | seq_printf(m, "\n"); | |
10459 | return 0; | |
10460 | } | |
10461 | ||
ef12fefa BR |
10462 | static const char *cpuacct_stat_desc[] = { |
10463 | [CPUACCT_STAT_USER] = "user", | |
10464 | [CPUACCT_STAT_SYSTEM] = "system", | |
10465 | }; | |
10466 | ||
10467 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10468 | struct cgroup_map_cb *cb) | |
10469 | { | |
10470 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10471 | int i; | |
10472 | ||
10473 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10474 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10475 | val = cputime64_to_clock_t(val); | |
10476 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10477 | } | |
10478 | return 0; | |
10479 | } | |
10480 | ||
d842de87 SV |
10481 | static struct cftype files[] = { |
10482 | { | |
10483 | .name = "usage", | |
f4c753b7 PM |
10484 | .read_u64 = cpuusage_read, |
10485 | .write_u64 = cpuusage_write, | |
d842de87 | 10486 | }, |
e9515c3c KC |
10487 | { |
10488 | .name = "usage_percpu", | |
10489 | .read_seq_string = cpuacct_percpu_seq_read, | |
10490 | }, | |
ef12fefa BR |
10491 | { |
10492 | .name = "stat", | |
10493 | .read_map = cpuacct_stats_show, | |
10494 | }, | |
d842de87 SV |
10495 | }; |
10496 | ||
32cd756a | 10497 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10498 | { |
32cd756a | 10499 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10500 | } |
10501 | ||
10502 | /* | |
10503 | * charge this task's execution time to its accounting group. | |
10504 | * | |
10505 | * called with rq->lock held. | |
10506 | */ | |
10507 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10508 | { | |
10509 | struct cpuacct *ca; | |
934352f2 | 10510 | int cpu; |
d842de87 | 10511 | |
c40c6f85 | 10512 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10513 | return; |
10514 | ||
934352f2 | 10515 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10516 | |
10517 | rcu_read_lock(); | |
10518 | ||
d842de87 | 10519 | ca = task_ca(tsk); |
d842de87 | 10520 | |
934352f2 | 10521 | for (; ca; ca = ca->parent) { |
b36128c8 | 10522 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10523 | *cpuusage += cputime; |
10524 | } | |
a18b83b7 BR |
10525 | |
10526 | rcu_read_unlock(); | |
d842de87 SV |
10527 | } |
10528 | ||
ef12fefa BR |
10529 | /* |
10530 | * Charge the system/user time to the task's accounting group. | |
10531 | */ | |
10532 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10533 | enum cpuacct_stat_index idx, cputime_t val) | |
10534 | { | |
10535 | struct cpuacct *ca; | |
10536 | ||
10537 | if (unlikely(!cpuacct_subsys.active)) | |
10538 | return; | |
10539 | ||
10540 | rcu_read_lock(); | |
10541 | ca = task_ca(tsk); | |
10542 | ||
10543 | do { | |
10544 | percpu_counter_add(&ca->cpustat[idx], val); | |
10545 | ca = ca->parent; | |
10546 | } while (ca); | |
10547 | rcu_read_unlock(); | |
10548 | } | |
10549 | ||
d842de87 SV |
10550 | struct cgroup_subsys cpuacct_subsys = { |
10551 | .name = "cpuacct", | |
10552 | .create = cpuacct_create, | |
10553 | .destroy = cpuacct_destroy, | |
10554 | .populate = cpuacct_populate, | |
10555 | .subsys_id = cpuacct_subsys_id, | |
10556 | }; | |
10557 | #endif /* CONFIG_CGROUP_CPUACCT */ |