]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
5517d86b | 66 | #include <linux/reciprocal_div.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
434d53b0 | 71 | #include <linux/bootmem.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
0a16b607 | 75 | #include <trace/sched.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
d0b27fa7 PZ |
116 | /* |
117 | * single value that denotes runtime == period, ie unlimited time. | |
118 | */ | |
119 | #define RUNTIME_INF ((u64)~0ULL) | |
120 | ||
7e066fb8 MD |
121 | DEFINE_TRACE(sched_wait_task); |
122 | DEFINE_TRACE(sched_wakeup); | |
123 | DEFINE_TRACE(sched_wakeup_new); | |
124 | DEFINE_TRACE(sched_switch); | |
125 | DEFINE_TRACE(sched_migrate_task); | |
126 | ||
5517d86b | 127 | #ifdef CONFIG_SMP |
fd2ab30b SN |
128 | |
129 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
130 | ||
5517d86b ED |
131 | /* |
132 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
133 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
134 | */ | |
135 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
136 | { | |
137 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Each time a sched group cpu_power is changed, | |
142 | * we must compute its reciprocal value | |
143 | */ | |
144 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
145 | { | |
146 | sg->__cpu_power += val; | |
147 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
148 | } | |
149 | #endif | |
150 | ||
e05606d3 IM |
151 | static inline int rt_policy(int policy) |
152 | { | |
3f33a7ce | 153 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
154 | return 1; |
155 | return 0; | |
156 | } | |
157 | ||
158 | static inline int task_has_rt_policy(struct task_struct *p) | |
159 | { | |
160 | return rt_policy(p->policy); | |
161 | } | |
162 | ||
1da177e4 | 163 | /* |
6aa645ea | 164 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 165 | */ |
6aa645ea IM |
166 | struct rt_prio_array { |
167 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
168 | struct list_head queue[MAX_RT_PRIO]; | |
169 | }; | |
170 | ||
d0b27fa7 | 171 | struct rt_bandwidth { |
ea736ed5 IM |
172 | /* nests inside the rq lock: */ |
173 | spinlock_t rt_runtime_lock; | |
174 | ktime_t rt_period; | |
175 | u64 rt_runtime; | |
176 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
177 | }; |
178 | ||
179 | static struct rt_bandwidth def_rt_bandwidth; | |
180 | ||
181 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
182 | ||
183 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
184 | { | |
185 | struct rt_bandwidth *rt_b = | |
186 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
187 | ktime_t now; | |
188 | int overrun; | |
189 | int idle = 0; | |
190 | ||
191 | for (;;) { | |
192 | now = hrtimer_cb_get_time(timer); | |
193 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
194 | ||
195 | if (!overrun) | |
196 | break; | |
197 | ||
198 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
199 | } | |
200 | ||
201 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
202 | } | |
203 | ||
204 | static | |
205 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
206 | { | |
207 | rt_b->rt_period = ns_to_ktime(period); | |
208 | rt_b->rt_runtime = runtime; | |
209 | ||
ac086bc2 PZ |
210 | spin_lock_init(&rt_b->rt_runtime_lock); |
211 | ||
d0b27fa7 PZ |
212 | hrtimer_init(&rt_b->rt_period_timer, |
213 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
214 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
215 | } |
216 | ||
c8bfff6d KH |
217 | static inline int rt_bandwidth_enabled(void) |
218 | { | |
219 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
220 | } |
221 | ||
222 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
223 | { | |
224 | ktime_t now; | |
225 | ||
0b148fa0 | 226 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
227 | return; |
228 | ||
229 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
230 | return; | |
231 | ||
232 | spin_lock(&rt_b->rt_runtime_lock); | |
233 | for (;;) { | |
234 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
235 | break; | |
236 | ||
237 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
238 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
239 | hrtimer_start_expires(&rt_b->rt_period_timer, |
240 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
241 | } |
242 | spin_unlock(&rt_b->rt_runtime_lock); | |
243 | } | |
244 | ||
245 | #ifdef CONFIG_RT_GROUP_SCHED | |
246 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
247 | { | |
248 | hrtimer_cancel(&rt_b->rt_period_timer); | |
249 | } | |
250 | #endif | |
251 | ||
712555ee HC |
252 | /* |
253 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
254 | * detach_destroy_domains and partition_sched_domains. | |
255 | */ | |
256 | static DEFINE_MUTEX(sched_domains_mutex); | |
257 | ||
052f1dc7 | 258 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 259 | |
68318b8e SV |
260 | #include <linux/cgroup.h> |
261 | ||
29f59db3 SV |
262 | struct cfs_rq; |
263 | ||
6f505b16 PZ |
264 | static LIST_HEAD(task_groups); |
265 | ||
29f59db3 | 266 | /* task group related information */ |
4cf86d77 | 267 | struct task_group { |
052f1dc7 | 268 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
269 | struct cgroup_subsys_state css; |
270 | #endif | |
052f1dc7 | 271 | |
6c415b92 AB |
272 | #ifdef CONFIG_USER_SCHED |
273 | uid_t uid; | |
274 | #endif | |
275 | ||
052f1dc7 | 276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
277 | /* schedulable entities of this group on each cpu */ |
278 | struct sched_entity **se; | |
279 | /* runqueue "owned" by this group on each cpu */ | |
280 | struct cfs_rq **cfs_rq; | |
281 | unsigned long shares; | |
052f1dc7 PZ |
282 | #endif |
283 | ||
284 | #ifdef CONFIG_RT_GROUP_SCHED | |
285 | struct sched_rt_entity **rt_se; | |
286 | struct rt_rq **rt_rq; | |
287 | ||
d0b27fa7 | 288 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 289 | #endif |
6b2d7700 | 290 | |
ae8393e5 | 291 | struct rcu_head rcu; |
6f505b16 | 292 | struct list_head list; |
f473aa5e PZ |
293 | |
294 | struct task_group *parent; | |
295 | struct list_head siblings; | |
296 | struct list_head children; | |
29f59db3 SV |
297 | }; |
298 | ||
354d60c2 | 299 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 300 | |
6c415b92 AB |
301 | /* Helper function to pass uid information to create_sched_user() */ |
302 | void set_tg_uid(struct user_struct *user) | |
303 | { | |
304 | user->tg->uid = user->uid; | |
305 | } | |
306 | ||
eff766a6 PZ |
307 | /* |
308 | * Root task group. | |
309 | * Every UID task group (including init_task_group aka UID-0) will | |
310 | * be a child to this group. | |
311 | */ | |
312 | struct task_group root_task_group; | |
313 | ||
052f1dc7 | 314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
315 | /* Default task group's sched entity on each cpu */ |
316 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
317 | /* Default task group's cfs_rq on each cpu */ | |
318 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 319 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
320 | |
321 | #ifdef CONFIG_RT_GROUP_SCHED | |
322 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
323 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 325 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 326 | #define root_task_group init_task_group |
9a7e0b18 | 327 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 328 | |
8ed36996 | 329 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
330 | * a task group's cpu shares. |
331 | */ | |
8ed36996 | 332 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 333 | |
052f1dc7 | 334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
335 | #ifdef CONFIG_USER_SCHED |
336 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 337 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 338 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 339 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 340 | |
cb4ad1ff | 341 | /* |
2e084786 LJ |
342 | * A weight of 0 or 1 can cause arithmetics problems. |
343 | * A weight of a cfs_rq is the sum of weights of which entities | |
344 | * are queued on this cfs_rq, so a weight of a entity should not be | |
345 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
346 | * (The default weight is 1024 - so there's no practical |
347 | * limitation from this.) | |
348 | */ | |
18d95a28 | 349 | #define MIN_SHARES 2 |
2e084786 | 350 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 351 | |
052f1dc7 PZ |
352 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
353 | #endif | |
354 | ||
29f59db3 | 355 | /* Default task group. |
3a252015 | 356 | * Every task in system belong to this group at bootup. |
29f59db3 | 357 | */ |
434d53b0 | 358 | struct task_group init_task_group; |
29f59db3 SV |
359 | |
360 | /* return group to which a task belongs */ | |
4cf86d77 | 361 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 362 | { |
4cf86d77 | 363 | struct task_group *tg; |
9b5b7751 | 364 | |
052f1dc7 | 365 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
366 | rcu_read_lock(); |
367 | tg = __task_cred(p)->user->tg; | |
368 | rcu_read_unlock(); | |
052f1dc7 | 369 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
370 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
371 | struct task_group, css); | |
24e377a8 | 372 | #else |
41a2d6cf | 373 | tg = &init_task_group; |
24e377a8 | 374 | #endif |
9b5b7751 | 375 | return tg; |
29f59db3 SV |
376 | } |
377 | ||
378 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 379 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 380 | { |
052f1dc7 | 381 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
382 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
383 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 384 | #endif |
6f505b16 | 385 | |
052f1dc7 | 386 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
387 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
388 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 389 | #endif |
29f59db3 SV |
390 | } |
391 | ||
392 | #else | |
393 | ||
6f505b16 | 394 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
395 | static inline struct task_group *task_group(struct task_struct *p) |
396 | { | |
397 | return NULL; | |
398 | } | |
29f59db3 | 399 | |
052f1dc7 | 400 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 401 | |
6aa645ea IM |
402 | /* CFS-related fields in a runqueue */ |
403 | struct cfs_rq { | |
404 | struct load_weight load; | |
405 | unsigned long nr_running; | |
406 | ||
6aa645ea | 407 | u64 exec_clock; |
e9acbff6 | 408 | u64 min_vruntime; |
6aa645ea IM |
409 | |
410 | struct rb_root tasks_timeline; | |
411 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
412 | |
413 | struct list_head tasks; | |
414 | struct list_head *balance_iterator; | |
415 | ||
416 | /* | |
417 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
418 | * It is set to NULL otherwise (i.e when none are currently running). |
419 | */ | |
4793241b | 420 | struct sched_entity *curr, *next, *last; |
ddc97297 | 421 | |
5ac5c4d6 | 422 | unsigned int nr_spread_over; |
ddc97297 | 423 | |
62160e3f | 424 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
425 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
426 | ||
41a2d6cf IM |
427 | /* |
428 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
429 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
430 | * (like users, containers etc.) | |
431 | * | |
432 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
433 | * list is used during load balance. | |
434 | */ | |
41a2d6cf IM |
435 | struct list_head leaf_cfs_rq_list; |
436 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
437 | |
438 | #ifdef CONFIG_SMP | |
c09595f6 | 439 | /* |
c8cba857 | 440 | * the part of load.weight contributed by tasks |
c09595f6 | 441 | */ |
c8cba857 | 442 | unsigned long task_weight; |
c09595f6 | 443 | |
c8cba857 PZ |
444 | /* |
445 | * h_load = weight * f(tg) | |
446 | * | |
447 | * Where f(tg) is the recursive weight fraction assigned to | |
448 | * this group. | |
449 | */ | |
450 | unsigned long h_load; | |
c09595f6 | 451 | |
c8cba857 PZ |
452 | /* |
453 | * this cpu's part of tg->shares | |
454 | */ | |
455 | unsigned long shares; | |
f1d239f7 PZ |
456 | |
457 | /* | |
458 | * load.weight at the time we set shares | |
459 | */ | |
460 | unsigned long rq_weight; | |
c09595f6 | 461 | #endif |
6aa645ea IM |
462 | #endif |
463 | }; | |
1da177e4 | 464 | |
6aa645ea IM |
465 | /* Real-Time classes' related field in a runqueue: */ |
466 | struct rt_rq { | |
467 | struct rt_prio_array active; | |
63489e45 | 468 | unsigned long rt_nr_running; |
052f1dc7 | 469 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
470 | struct { |
471 | int curr; /* highest queued rt task prio */ | |
398a153b | 472 | #ifdef CONFIG_SMP |
e864c499 | 473 | int next; /* next highest */ |
398a153b | 474 | #endif |
e864c499 | 475 | } highest_prio; |
6f505b16 | 476 | #endif |
fa85ae24 | 477 | #ifdef CONFIG_SMP |
73fe6aae | 478 | unsigned long rt_nr_migratory; |
a22d7fc1 | 479 | int overloaded; |
917b627d | 480 | struct plist_head pushable_tasks; |
fa85ae24 | 481 | #endif |
6f505b16 | 482 | int rt_throttled; |
fa85ae24 | 483 | u64 rt_time; |
ac086bc2 | 484 | u64 rt_runtime; |
ea736ed5 | 485 | /* Nests inside the rq lock: */ |
ac086bc2 | 486 | spinlock_t rt_runtime_lock; |
6f505b16 | 487 | |
052f1dc7 | 488 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
489 | unsigned long rt_nr_boosted; |
490 | ||
6f505b16 PZ |
491 | struct rq *rq; |
492 | struct list_head leaf_rt_rq_list; | |
493 | struct task_group *tg; | |
494 | struct sched_rt_entity *rt_se; | |
495 | #endif | |
6aa645ea IM |
496 | }; |
497 | ||
57d885fe GH |
498 | #ifdef CONFIG_SMP |
499 | ||
500 | /* | |
501 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
502 | * variables. Each exclusive cpuset essentially defines an island domain by |
503 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
504 | * exclusive cpuset is created, we also create and attach a new root-domain |
505 | * object. | |
506 | * | |
57d885fe GH |
507 | */ |
508 | struct root_domain { | |
509 | atomic_t refcount; | |
c6c4927b RR |
510 | cpumask_var_t span; |
511 | cpumask_var_t online; | |
637f5085 | 512 | |
0eab9146 | 513 | /* |
637f5085 GH |
514 | * The "RT overload" flag: it gets set if a CPU has more than |
515 | * one runnable RT task. | |
516 | */ | |
c6c4927b | 517 | cpumask_var_t rto_mask; |
0eab9146 | 518 | atomic_t rto_count; |
6e0534f2 GH |
519 | #ifdef CONFIG_SMP |
520 | struct cpupri cpupri; | |
521 | #endif | |
7a09b1a2 VS |
522 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
523 | /* | |
524 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
525 | * used when most cpus are idle in the system indicating overall very | |
526 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
527 | */ | |
528 | unsigned int sched_mc_preferred_wakeup_cpu; | |
529 | #endif | |
57d885fe GH |
530 | }; |
531 | ||
dc938520 GH |
532 | /* |
533 | * By default the system creates a single root-domain with all cpus as | |
534 | * members (mimicking the global state we have today). | |
535 | */ | |
57d885fe GH |
536 | static struct root_domain def_root_domain; |
537 | ||
538 | #endif | |
539 | ||
1da177e4 LT |
540 | /* |
541 | * This is the main, per-CPU runqueue data structure. | |
542 | * | |
543 | * Locking rule: those places that want to lock multiple runqueues | |
544 | * (such as the load balancing or the thread migration code), lock | |
545 | * acquire operations must be ordered by ascending &runqueue. | |
546 | */ | |
70b97a7f | 547 | struct rq { |
d8016491 IM |
548 | /* runqueue lock: */ |
549 | spinlock_t lock; | |
1da177e4 LT |
550 | |
551 | /* | |
552 | * nr_running and cpu_load should be in the same cacheline because | |
553 | * remote CPUs use both these fields when doing load calculation. | |
554 | */ | |
555 | unsigned long nr_running; | |
6aa645ea IM |
556 | #define CPU_LOAD_IDX_MAX 5 |
557 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 558 | #ifdef CONFIG_NO_HZ |
15934a37 | 559 | unsigned long last_tick_seen; |
46cb4b7c SS |
560 | unsigned char in_nohz_recently; |
561 | #endif | |
d8016491 IM |
562 | /* capture load from *all* tasks on this cpu: */ |
563 | struct load_weight load; | |
6aa645ea IM |
564 | unsigned long nr_load_updates; |
565 | u64 nr_switches; | |
566 | ||
567 | struct cfs_rq cfs; | |
6f505b16 | 568 | struct rt_rq rt; |
6f505b16 | 569 | |
6aa645ea | 570 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
571 | /* list of leaf cfs_rq on this cpu: */ |
572 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
573 | #endif |
574 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 575 | struct list_head leaf_rt_rq_list; |
1da177e4 | 576 | #endif |
1da177e4 LT |
577 | |
578 | /* | |
579 | * This is part of a global counter where only the total sum | |
580 | * over all CPUs matters. A task can increase this counter on | |
581 | * one CPU and if it got migrated afterwards it may decrease | |
582 | * it on another CPU. Always updated under the runqueue lock: | |
583 | */ | |
584 | unsigned long nr_uninterruptible; | |
585 | ||
36c8b586 | 586 | struct task_struct *curr, *idle; |
c9819f45 | 587 | unsigned long next_balance; |
1da177e4 | 588 | struct mm_struct *prev_mm; |
6aa645ea | 589 | |
3e51f33f | 590 | u64 clock; |
6aa645ea | 591 | |
1da177e4 LT |
592 | atomic_t nr_iowait; |
593 | ||
594 | #ifdef CONFIG_SMP | |
0eab9146 | 595 | struct root_domain *rd; |
1da177e4 LT |
596 | struct sched_domain *sd; |
597 | ||
a0a522ce | 598 | unsigned char idle_at_tick; |
1da177e4 LT |
599 | /* For active balancing */ |
600 | int active_balance; | |
601 | int push_cpu; | |
d8016491 IM |
602 | /* cpu of this runqueue: */ |
603 | int cpu; | |
1f11eb6a | 604 | int online; |
1da177e4 | 605 | |
a8a51d5e | 606 | unsigned long avg_load_per_task; |
1da177e4 | 607 | |
36c8b586 | 608 | struct task_struct *migration_thread; |
1da177e4 LT |
609 | struct list_head migration_queue; |
610 | #endif | |
611 | ||
8f4d37ec | 612 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
613 | #ifdef CONFIG_SMP |
614 | int hrtick_csd_pending; | |
615 | struct call_single_data hrtick_csd; | |
616 | #endif | |
8f4d37ec PZ |
617 | struct hrtimer hrtick_timer; |
618 | #endif | |
619 | ||
1da177e4 LT |
620 | #ifdef CONFIG_SCHEDSTATS |
621 | /* latency stats */ | |
622 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
623 | unsigned long long rq_cpu_time; |
624 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
625 | |
626 | /* sys_sched_yield() stats */ | |
480b9434 KC |
627 | unsigned int yld_exp_empty; |
628 | unsigned int yld_act_empty; | |
629 | unsigned int yld_both_empty; | |
630 | unsigned int yld_count; | |
1da177e4 LT |
631 | |
632 | /* schedule() stats */ | |
480b9434 KC |
633 | unsigned int sched_switch; |
634 | unsigned int sched_count; | |
635 | unsigned int sched_goidle; | |
1da177e4 LT |
636 | |
637 | /* try_to_wake_up() stats */ | |
480b9434 KC |
638 | unsigned int ttwu_count; |
639 | unsigned int ttwu_local; | |
b8efb561 IM |
640 | |
641 | /* BKL stats */ | |
480b9434 | 642 | unsigned int bkl_count; |
1da177e4 LT |
643 | #endif |
644 | }; | |
645 | ||
f34e3b61 | 646 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 647 | |
15afe09b | 648 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 649 | { |
15afe09b | 650 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
651 | } |
652 | ||
0a2966b4 CL |
653 | static inline int cpu_of(struct rq *rq) |
654 | { | |
655 | #ifdef CONFIG_SMP | |
656 | return rq->cpu; | |
657 | #else | |
658 | return 0; | |
659 | #endif | |
660 | } | |
661 | ||
674311d5 NP |
662 | /* |
663 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 664 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
665 | * |
666 | * The domain tree of any CPU may only be accessed from within | |
667 | * preempt-disabled sections. | |
668 | */ | |
48f24c4d IM |
669 | #define for_each_domain(cpu, __sd) \ |
670 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
671 | |
672 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
673 | #define this_rq() (&__get_cpu_var(runqueues)) | |
674 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
675 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
676 | ||
3e51f33f PZ |
677 | static inline void update_rq_clock(struct rq *rq) |
678 | { | |
679 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
680 | } | |
681 | ||
bf5c91ba IM |
682 | /* |
683 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
684 | */ | |
685 | #ifdef CONFIG_SCHED_DEBUG | |
686 | # define const_debug __read_mostly | |
687 | #else | |
688 | # define const_debug static const | |
689 | #endif | |
690 | ||
017730c1 IM |
691 | /** |
692 | * runqueue_is_locked | |
693 | * | |
694 | * Returns true if the current cpu runqueue is locked. | |
695 | * This interface allows printk to be called with the runqueue lock | |
696 | * held and know whether or not it is OK to wake up the klogd. | |
697 | */ | |
698 | int runqueue_is_locked(void) | |
699 | { | |
700 | int cpu = get_cpu(); | |
701 | struct rq *rq = cpu_rq(cpu); | |
702 | int ret; | |
703 | ||
704 | ret = spin_is_locked(&rq->lock); | |
705 | put_cpu(); | |
706 | return ret; | |
707 | } | |
708 | ||
bf5c91ba IM |
709 | /* |
710 | * Debugging: various feature bits | |
711 | */ | |
f00b45c1 PZ |
712 | |
713 | #define SCHED_FEAT(name, enabled) \ | |
714 | __SCHED_FEAT_##name , | |
715 | ||
bf5c91ba | 716 | enum { |
f00b45c1 | 717 | #include "sched_features.h" |
bf5c91ba IM |
718 | }; |
719 | ||
f00b45c1 PZ |
720 | #undef SCHED_FEAT |
721 | ||
722 | #define SCHED_FEAT(name, enabled) \ | |
723 | (1UL << __SCHED_FEAT_##name) * enabled | | |
724 | ||
bf5c91ba | 725 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
726 | #include "sched_features.h" |
727 | 0; | |
728 | ||
729 | #undef SCHED_FEAT | |
730 | ||
731 | #ifdef CONFIG_SCHED_DEBUG | |
732 | #define SCHED_FEAT(name, enabled) \ | |
733 | #name , | |
734 | ||
983ed7a6 | 735 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
736 | #include "sched_features.h" |
737 | NULL | |
738 | }; | |
739 | ||
740 | #undef SCHED_FEAT | |
741 | ||
34f3a814 | 742 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 743 | { |
f00b45c1 PZ |
744 | int i; |
745 | ||
746 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
747 | if (!(sysctl_sched_features & (1UL << i))) |
748 | seq_puts(m, "NO_"); | |
749 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 750 | } |
34f3a814 | 751 | seq_puts(m, "\n"); |
f00b45c1 | 752 | |
34f3a814 | 753 | return 0; |
f00b45c1 PZ |
754 | } |
755 | ||
756 | static ssize_t | |
757 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
758 | size_t cnt, loff_t *ppos) | |
759 | { | |
760 | char buf[64]; | |
761 | char *cmp = buf; | |
762 | int neg = 0; | |
763 | int i; | |
764 | ||
765 | if (cnt > 63) | |
766 | cnt = 63; | |
767 | ||
768 | if (copy_from_user(&buf, ubuf, cnt)) | |
769 | return -EFAULT; | |
770 | ||
771 | buf[cnt] = 0; | |
772 | ||
c24b7c52 | 773 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
774 | neg = 1; |
775 | cmp += 3; | |
776 | } | |
777 | ||
778 | for (i = 0; sched_feat_names[i]; i++) { | |
779 | int len = strlen(sched_feat_names[i]); | |
780 | ||
781 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
782 | if (neg) | |
783 | sysctl_sched_features &= ~(1UL << i); | |
784 | else | |
785 | sysctl_sched_features |= (1UL << i); | |
786 | break; | |
787 | } | |
788 | } | |
789 | ||
790 | if (!sched_feat_names[i]) | |
791 | return -EINVAL; | |
792 | ||
793 | filp->f_pos += cnt; | |
794 | ||
795 | return cnt; | |
796 | } | |
797 | ||
34f3a814 LZ |
798 | static int sched_feat_open(struct inode *inode, struct file *filp) |
799 | { | |
800 | return single_open(filp, sched_feat_show, NULL); | |
801 | } | |
802 | ||
f00b45c1 | 803 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
804 | .open = sched_feat_open, |
805 | .write = sched_feat_write, | |
806 | .read = seq_read, | |
807 | .llseek = seq_lseek, | |
808 | .release = single_release, | |
f00b45c1 PZ |
809 | }; |
810 | ||
811 | static __init int sched_init_debug(void) | |
812 | { | |
f00b45c1 PZ |
813 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
814 | &sched_feat_fops); | |
815 | ||
816 | return 0; | |
817 | } | |
818 | late_initcall(sched_init_debug); | |
819 | ||
820 | #endif | |
821 | ||
822 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 823 | |
b82d9fdd PZ |
824 | /* |
825 | * Number of tasks to iterate in a single balance run. | |
826 | * Limited because this is done with IRQs disabled. | |
827 | */ | |
828 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
829 | ||
2398f2c6 PZ |
830 | /* |
831 | * ratelimit for updating the group shares. | |
55cd5340 | 832 | * default: 0.25ms |
2398f2c6 | 833 | */ |
55cd5340 | 834 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 835 | |
ffda12a1 PZ |
836 | /* |
837 | * Inject some fuzzyness into changing the per-cpu group shares | |
838 | * this avoids remote rq-locks at the expense of fairness. | |
839 | * default: 4 | |
840 | */ | |
841 | unsigned int sysctl_sched_shares_thresh = 4; | |
842 | ||
fa85ae24 | 843 | /* |
9f0c1e56 | 844 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
845 | * default: 1s |
846 | */ | |
9f0c1e56 | 847 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 848 | |
6892b75e IM |
849 | static __read_mostly int scheduler_running; |
850 | ||
9f0c1e56 PZ |
851 | /* |
852 | * part of the period that we allow rt tasks to run in us. | |
853 | * default: 0.95s | |
854 | */ | |
855 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 856 | |
d0b27fa7 PZ |
857 | static inline u64 global_rt_period(void) |
858 | { | |
859 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
860 | } | |
861 | ||
862 | static inline u64 global_rt_runtime(void) | |
863 | { | |
e26873bb | 864 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
865 | return RUNTIME_INF; |
866 | ||
867 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
868 | } | |
fa85ae24 | 869 | |
1da177e4 | 870 | #ifndef prepare_arch_switch |
4866cde0 NP |
871 | # define prepare_arch_switch(next) do { } while (0) |
872 | #endif | |
873 | #ifndef finish_arch_switch | |
874 | # define finish_arch_switch(prev) do { } while (0) | |
875 | #endif | |
876 | ||
051a1d1a DA |
877 | static inline int task_current(struct rq *rq, struct task_struct *p) |
878 | { | |
879 | return rq->curr == p; | |
880 | } | |
881 | ||
4866cde0 | 882 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 883 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 884 | { |
051a1d1a | 885 | return task_current(rq, p); |
4866cde0 NP |
886 | } |
887 | ||
70b97a7f | 888 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
889 | { |
890 | } | |
891 | ||
70b97a7f | 892 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 893 | { |
da04c035 IM |
894 | #ifdef CONFIG_DEBUG_SPINLOCK |
895 | /* this is a valid case when another task releases the spinlock */ | |
896 | rq->lock.owner = current; | |
897 | #endif | |
8a25d5de IM |
898 | /* |
899 | * If we are tracking spinlock dependencies then we have to | |
900 | * fix up the runqueue lock - which gets 'carried over' from | |
901 | * prev into current: | |
902 | */ | |
903 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
904 | ||
4866cde0 NP |
905 | spin_unlock_irq(&rq->lock); |
906 | } | |
907 | ||
908 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 909 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
910 | { |
911 | #ifdef CONFIG_SMP | |
912 | return p->oncpu; | |
913 | #else | |
051a1d1a | 914 | return task_current(rq, p); |
4866cde0 NP |
915 | #endif |
916 | } | |
917 | ||
70b97a7f | 918 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
919 | { |
920 | #ifdef CONFIG_SMP | |
921 | /* | |
922 | * We can optimise this out completely for !SMP, because the | |
923 | * SMP rebalancing from interrupt is the only thing that cares | |
924 | * here. | |
925 | */ | |
926 | next->oncpu = 1; | |
927 | #endif | |
928 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
929 | spin_unlock_irq(&rq->lock); | |
930 | #else | |
931 | spin_unlock(&rq->lock); | |
932 | #endif | |
933 | } | |
934 | ||
70b97a7f | 935 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
936 | { |
937 | #ifdef CONFIG_SMP | |
938 | /* | |
939 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
940 | * We must ensure this doesn't happen until the switch is completely | |
941 | * finished. | |
942 | */ | |
943 | smp_wmb(); | |
944 | prev->oncpu = 0; | |
945 | #endif | |
946 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
947 | local_irq_enable(); | |
1da177e4 | 948 | #endif |
4866cde0 NP |
949 | } |
950 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 951 | |
b29739f9 IM |
952 | /* |
953 | * __task_rq_lock - lock the runqueue a given task resides on. | |
954 | * Must be called interrupts disabled. | |
955 | */ | |
70b97a7f | 956 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
957 | __acquires(rq->lock) |
958 | { | |
3a5c359a AK |
959 | for (;;) { |
960 | struct rq *rq = task_rq(p); | |
961 | spin_lock(&rq->lock); | |
962 | if (likely(rq == task_rq(p))) | |
963 | return rq; | |
b29739f9 | 964 | spin_unlock(&rq->lock); |
b29739f9 | 965 | } |
b29739f9 IM |
966 | } |
967 | ||
1da177e4 LT |
968 | /* |
969 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 970 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
971 | * explicitly disabling preemption. |
972 | */ | |
70b97a7f | 973 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
974 | __acquires(rq->lock) |
975 | { | |
70b97a7f | 976 | struct rq *rq; |
1da177e4 | 977 | |
3a5c359a AK |
978 | for (;;) { |
979 | local_irq_save(*flags); | |
980 | rq = task_rq(p); | |
981 | spin_lock(&rq->lock); | |
982 | if (likely(rq == task_rq(p))) | |
983 | return rq; | |
1da177e4 | 984 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 985 | } |
1da177e4 LT |
986 | } |
987 | ||
ad474cac ON |
988 | void task_rq_unlock_wait(struct task_struct *p) |
989 | { | |
990 | struct rq *rq = task_rq(p); | |
991 | ||
992 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
993 | spin_unlock_wait(&rq->lock); | |
994 | } | |
995 | ||
a9957449 | 996 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
997 | __releases(rq->lock) |
998 | { | |
999 | spin_unlock(&rq->lock); | |
1000 | } | |
1001 | ||
70b97a7f | 1002 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1003 | __releases(rq->lock) |
1004 | { | |
1005 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1006 | } | |
1007 | ||
1da177e4 | 1008 | /* |
cc2a73b5 | 1009 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1010 | */ |
a9957449 | 1011 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1012 | __acquires(rq->lock) |
1013 | { | |
70b97a7f | 1014 | struct rq *rq; |
1da177e4 LT |
1015 | |
1016 | local_irq_disable(); | |
1017 | rq = this_rq(); | |
1018 | spin_lock(&rq->lock); | |
1019 | ||
1020 | return rq; | |
1021 | } | |
1022 | ||
8f4d37ec PZ |
1023 | #ifdef CONFIG_SCHED_HRTICK |
1024 | /* | |
1025 | * Use HR-timers to deliver accurate preemption points. | |
1026 | * | |
1027 | * Its all a bit involved since we cannot program an hrt while holding the | |
1028 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1029 | * reschedule event. | |
1030 | * | |
1031 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1032 | * rq->lock. | |
1033 | */ | |
8f4d37ec PZ |
1034 | |
1035 | /* | |
1036 | * Use hrtick when: | |
1037 | * - enabled by features | |
1038 | * - hrtimer is actually high res | |
1039 | */ | |
1040 | static inline int hrtick_enabled(struct rq *rq) | |
1041 | { | |
1042 | if (!sched_feat(HRTICK)) | |
1043 | return 0; | |
ba42059f | 1044 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1045 | return 0; |
8f4d37ec PZ |
1046 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1047 | } | |
1048 | ||
8f4d37ec PZ |
1049 | static void hrtick_clear(struct rq *rq) |
1050 | { | |
1051 | if (hrtimer_active(&rq->hrtick_timer)) | |
1052 | hrtimer_cancel(&rq->hrtick_timer); | |
1053 | } | |
1054 | ||
8f4d37ec PZ |
1055 | /* |
1056 | * High-resolution timer tick. | |
1057 | * Runs from hardirq context with interrupts disabled. | |
1058 | */ | |
1059 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1060 | { | |
1061 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1062 | ||
1063 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1064 | ||
1065 | spin_lock(&rq->lock); | |
3e51f33f | 1066 | update_rq_clock(rq); |
8f4d37ec PZ |
1067 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1068 | spin_unlock(&rq->lock); | |
1069 | ||
1070 | return HRTIMER_NORESTART; | |
1071 | } | |
1072 | ||
95e904c7 | 1073 | #ifdef CONFIG_SMP |
31656519 PZ |
1074 | /* |
1075 | * called from hardirq (IPI) context | |
1076 | */ | |
1077 | static void __hrtick_start(void *arg) | |
b328ca18 | 1078 | { |
31656519 | 1079 | struct rq *rq = arg; |
b328ca18 | 1080 | |
31656519 PZ |
1081 | spin_lock(&rq->lock); |
1082 | hrtimer_restart(&rq->hrtick_timer); | |
1083 | rq->hrtick_csd_pending = 0; | |
1084 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1085 | } |
1086 | ||
31656519 PZ |
1087 | /* |
1088 | * Called to set the hrtick timer state. | |
1089 | * | |
1090 | * called with rq->lock held and irqs disabled | |
1091 | */ | |
1092 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1093 | { |
31656519 PZ |
1094 | struct hrtimer *timer = &rq->hrtick_timer; |
1095 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1096 | |
cc584b21 | 1097 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1098 | |
1099 | if (rq == this_rq()) { | |
1100 | hrtimer_restart(timer); | |
1101 | } else if (!rq->hrtick_csd_pending) { | |
1102 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1103 | rq->hrtick_csd_pending = 1; | |
1104 | } | |
b328ca18 PZ |
1105 | } |
1106 | ||
1107 | static int | |
1108 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1109 | { | |
1110 | int cpu = (int)(long)hcpu; | |
1111 | ||
1112 | switch (action) { | |
1113 | case CPU_UP_CANCELED: | |
1114 | case CPU_UP_CANCELED_FROZEN: | |
1115 | case CPU_DOWN_PREPARE: | |
1116 | case CPU_DOWN_PREPARE_FROZEN: | |
1117 | case CPU_DEAD: | |
1118 | case CPU_DEAD_FROZEN: | |
31656519 | 1119 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1120 | return NOTIFY_OK; |
1121 | } | |
1122 | ||
1123 | return NOTIFY_DONE; | |
1124 | } | |
1125 | ||
fa748203 | 1126 | static __init void init_hrtick(void) |
b328ca18 PZ |
1127 | { |
1128 | hotcpu_notifier(hotplug_hrtick, 0); | |
1129 | } | |
31656519 PZ |
1130 | #else |
1131 | /* | |
1132 | * Called to set the hrtick timer state. | |
1133 | * | |
1134 | * called with rq->lock held and irqs disabled | |
1135 | */ | |
1136 | static void hrtick_start(struct rq *rq, u64 delay) | |
1137 | { | |
1138 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1139 | } | |
b328ca18 | 1140 | |
006c75f1 | 1141 | static inline void init_hrtick(void) |
8f4d37ec | 1142 | { |
8f4d37ec | 1143 | } |
31656519 | 1144 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1145 | |
31656519 | 1146 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1147 | { |
31656519 PZ |
1148 | #ifdef CONFIG_SMP |
1149 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1150 | |
31656519 PZ |
1151 | rq->hrtick_csd.flags = 0; |
1152 | rq->hrtick_csd.func = __hrtick_start; | |
1153 | rq->hrtick_csd.info = rq; | |
1154 | #endif | |
8f4d37ec | 1155 | |
31656519 PZ |
1156 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1157 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1158 | } |
006c75f1 | 1159 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1160 | static inline void hrtick_clear(struct rq *rq) |
1161 | { | |
1162 | } | |
1163 | ||
8f4d37ec PZ |
1164 | static inline void init_rq_hrtick(struct rq *rq) |
1165 | { | |
1166 | } | |
1167 | ||
b328ca18 PZ |
1168 | static inline void init_hrtick(void) |
1169 | { | |
1170 | } | |
006c75f1 | 1171 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1172 | |
c24d20db IM |
1173 | /* |
1174 | * resched_task - mark a task 'to be rescheduled now'. | |
1175 | * | |
1176 | * On UP this means the setting of the need_resched flag, on SMP it | |
1177 | * might also involve a cross-CPU call to trigger the scheduler on | |
1178 | * the target CPU. | |
1179 | */ | |
1180 | #ifdef CONFIG_SMP | |
1181 | ||
1182 | #ifndef tsk_is_polling | |
1183 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1184 | #endif | |
1185 | ||
31656519 | 1186 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1187 | { |
1188 | int cpu; | |
1189 | ||
1190 | assert_spin_locked(&task_rq(p)->lock); | |
1191 | ||
31656519 | 1192 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1193 | return; |
1194 | ||
31656519 | 1195 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1196 | |
1197 | cpu = task_cpu(p); | |
1198 | if (cpu == smp_processor_id()) | |
1199 | return; | |
1200 | ||
1201 | /* NEED_RESCHED must be visible before we test polling */ | |
1202 | smp_mb(); | |
1203 | if (!tsk_is_polling(p)) | |
1204 | smp_send_reschedule(cpu); | |
1205 | } | |
1206 | ||
1207 | static void resched_cpu(int cpu) | |
1208 | { | |
1209 | struct rq *rq = cpu_rq(cpu); | |
1210 | unsigned long flags; | |
1211 | ||
1212 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1213 | return; | |
1214 | resched_task(cpu_curr(cpu)); | |
1215 | spin_unlock_irqrestore(&rq->lock, flags); | |
1216 | } | |
06d8308c TG |
1217 | |
1218 | #ifdef CONFIG_NO_HZ | |
1219 | /* | |
1220 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1221 | * idle CPU then this timer might expire before the next timer event | |
1222 | * which is scheduled to wake up that CPU. In case of a completely | |
1223 | * idle system the next event might even be infinite time into the | |
1224 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1225 | * leaves the inner idle loop so the newly added timer is taken into | |
1226 | * account when the CPU goes back to idle and evaluates the timer | |
1227 | * wheel for the next timer event. | |
1228 | */ | |
1229 | void wake_up_idle_cpu(int cpu) | |
1230 | { | |
1231 | struct rq *rq = cpu_rq(cpu); | |
1232 | ||
1233 | if (cpu == smp_processor_id()) | |
1234 | return; | |
1235 | ||
1236 | /* | |
1237 | * This is safe, as this function is called with the timer | |
1238 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1239 | * to idle and has not yet set rq->curr to idle then it will | |
1240 | * be serialized on the timer wheel base lock and take the new | |
1241 | * timer into account automatically. | |
1242 | */ | |
1243 | if (rq->curr != rq->idle) | |
1244 | return; | |
1245 | ||
1246 | /* | |
1247 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1248 | * lockless. The worst case is that the other CPU runs the | |
1249 | * idle task through an additional NOOP schedule() | |
1250 | */ | |
1251 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1252 | ||
1253 | /* NEED_RESCHED must be visible before we test polling */ | |
1254 | smp_mb(); | |
1255 | if (!tsk_is_polling(rq->idle)) | |
1256 | smp_send_reschedule(cpu); | |
1257 | } | |
6d6bc0ad | 1258 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1259 | |
6d6bc0ad | 1260 | #else /* !CONFIG_SMP */ |
31656519 | 1261 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1262 | { |
1263 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1264 | set_tsk_need_resched(p); |
c24d20db | 1265 | } |
6d6bc0ad | 1266 | #endif /* CONFIG_SMP */ |
c24d20db | 1267 | |
45bf76df IM |
1268 | #if BITS_PER_LONG == 32 |
1269 | # define WMULT_CONST (~0UL) | |
1270 | #else | |
1271 | # define WMULT_CONST (1UL << 32) | |
1272 | #endif | |
1273 | ||
1274 | #define WMULT_SHIFT 32 | |
1275 | ||
194081eb IM |
1276 | /* |
1277 | * Shift right and round: | |
1278 | */ | |
cf2ab469 | 1279 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1280 | |
a7be37ac PZ |
1281 | /* |
1282 | * delta *= weight / lw | |
1283 | */ | |
cb1c4fc9 | 1284 | static unsigned long |
45bf76df IM |
1285 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1286 | struct load_weight *lw) | |
1287 | { | |
1288 | u64 tmp; | |
1289 | ||
7a232e03 LJ |
1290 | if (!lw->inv_weight) { |
1291 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1292 | lw->inv_weight = 1; | |
1293 | else | |
1294 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1295 | / (lw->weight+1); | |
1296 | } | |
45bf76df IM |
1297 | |
1298 | tmp = (u64)delta_exec * weight; | |
1299 | /* | |
1300 | * Check whether we'd overflow the 64-bit multiplication: | |
1301 | */ | |
194081eb | 1302 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1303 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1304 | WMULT_SHIFT/2); |
1305 | else | |
cf2ab469 | 1306 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1307 | |
ecf691da | 1308 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1309 | } |
1310 | ||
1091985b | 1311 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1312 | { |
1313 | lw->weight += inc; | |
e89996ae | 1314 | lw->inv_weight = 0; |
45bf76df IM |
1315 | } |
1316 | ||
1091985b | 1317 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1318 | { |
1319 | lw->weight -= dec; | |
e89996ae | 1320 | lw->inv_weight = 0; |
45bf76df IM |
1321 | } |
1322 | ||
2dd73a4f PW |
1323 | /* |
1324 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1325 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1326 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1327 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1328 | * scaled version of the new time slice allocation that they receive on time |
1329 | * slice expiry etc. | |
1330 | */ | |
1331 | ||
cce7ade8 PZ |
1332 | #define WEIGHT_IDLEPRIO 3 |
1333 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1334 | |
1335 | /* | |
1336 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1337 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1338 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1339 | * that remained on nice 0. | |
1340 | * | |
1341 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1342 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1343 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1344 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1345 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1346 | */ |
1347 | static const int prio_to_weight[40] = { | |
254753dc IM |
1348 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1349 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1350 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1351 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1352 | /* 0 */ 1024, 820, 655, 526, 423, | |
1353 | /* 5 */ 335, 272, 215, 172, 137, | |
1354 | /* 10 */ 110, 87, 70, 56, 45, | |
1355 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1356 | }; |
1357 | ||
5714d2de IM |
1358 | /* |
1359 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1360 | * | |
1361 | * In cases where the weight does not change often, we can use the | |
1362 | * precalculated inverse to speed up arithmetics by turning divisions | |
1363 | * into multiplications: | |
1364 | */ | |
dd41f596 | 1365 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1366 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1367 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1368 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1369 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1370 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1371 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1372 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1373 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1374 | }; |
2dd73a4f | 1375 | |
dd41f596 IM |
1376 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1377 | ||
1378 | /* | |
1379 | * runqueue iterator, to support SMP load-balancing between different | |
1380 | * scheduling classes, without having to expose their internal data | |
1381 | * structures to the load-balancing proper: | |
1382 | */ | |
1383 | struct rq_iterator { | |
1384 | void *arg; | |
1385 | struct task_struct *(*start)(void *); | |
1386 | struct task_struct *(*next)(void *); | |
1387 | }; | |
1388 | ||
e1d1484f PW |
1389 | #ifdef CONFIG_SMP |
1390 | static unsigned long | |
1391 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1392 | unsigned long max_load_move, struct sched_domain *sd, | |
1393 | enum cpu_idle_type idle, int *all_pinned, | |
1394 | int *this_best_prio, struct rq_iterator *iterator); | |
1395 | ||
1396 | static int | |
1397 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1398 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1399 | struct rq_iterator *iterator); | |
e1d1484f | 1400 | #endif |
dd41f596 | 1401 | |
d842de87 SV |
1402 | #ifdef CONFIG_CGROUP_CPUACCT |
1403 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1404 | #else | |
1405 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1406 | #endif | |
1407 | ||
18d95a28 PZ |
1408 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1409 | { | |
1410 | update_load_add(&rq->load, load); | |
1411 | } | |
1412 | ||
1413 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1414 | { | |
1415 | update_load_sub(&rq->load, load); | |
1416 | } | |
1417 | ||
7940ca36 | 1418 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1419 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1420 | |
1421 | /* | |
1422 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1423 | * leaving it for the final time. | |
1424 | */ | |
eb755805 | 1425 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1426 | { |
1427 | struct task_group *parent, *child; | |
eb755805 | 1428 | int ret; |
c09595f6 PZ |
1429 | |
1430 | rcu_read_lock(); | |
1431 | parent = &root_task_group; | |
1432 | down: | |
eb755805 PZ |
1433 | ret = (*down)(parent, data); |
1434 | if (ret) | |
1435 | goto out_unlock; | |
c09595f6 PZ |
1436 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1437 | parent = child; | |
1438 | goto down; | |
1439 | ||
1440 | up: | |
1441 | continue; | |
1442 | } | |
eb755805 PZ |
1443 | ret = (*up)(parent, data); |
1444 | if (ret) | |
1445 | goto out_unlock; | |
c09595f6 PZ |
1446 | |
1447 | child = parent; | |
1448 | parent = parent->parent; | |
1449 | if (parent) | |
1450 | goto up; | |
eb755805 | 1451 | out_unlock: |
c09595f6 | 1452 | rcu_read_unlock(); |
eb755805 PZ |
1453 | |
1454 | return ret; | |
c09595f6 PZ |
1455 | } |
1456 | ||
eb755805 PZ |
1457 | static int tg_nop(struct task_group *tg, void *data) |
1458 | { | |
1459 | return 0; | |
c09595f6 | 1460 | } |
eb755805 PZ |
1461 | #endif |
1462 | ||
1463 | #ifdef CONFIG_SMP | |
1464 | static unsigned long source_load(int cpu, int type); | |
1465 | static unsigned long target_load(int cpu, int type); | |
1466 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1467 | ||
1468 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1469 | { | |
1470 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1471 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1472 | |
4cd42620 SR |
1473 | if (nr_running) |
1474 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1475 | else |
1476 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1477 | |
1478 | return rq->avg_load_per_task; | |
1479 | } | |
1480 | ||
1481 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1482 | |
c09595f6 PZ |
1483 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1484 | ||
1485 | /* | |
1486 | * Calculate and set the cpu's group shares. | |
1487 | */ | |
1488 | static void | |
ffda12a1 PZ |
1489 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1490 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1491 | { |
c09595f6 PZ |
1492 | unsigned long shares; |
1493 | unsigned long rq_weight; | |
1494 | ||
c8cba857 | 1495 | if (!tg->se[cpu]) |
c09595f6 PZ |
1496 | return; |
1497 | ||
ec4e0e2f | 1498 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1499 | |
c09595f6 PZ |
1500 | /* |
1501 | * \Sum shares * rq_weight | |
1502 | * shares = ----------------------- | |
1503 | * \Sum rq_weight | |
1504 | * | |
1505 | */ | |
ec4e0e2f | 1506 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1507 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1508 | |
ffda12a1 PZ |
1509 | if (abs(shares - tg->se[cpu]->load.weight) > |
1510 | sysctl_sched_shares_thresh) { | |
1511 | struct rq *rq = cpu_rq(cpu); | |
1512 | unsigned long flags; | |
c09595f6 | 1513 | |
ffda12a1 | 1514 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1515 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1516 | |
ffda12a1 PZ |
1517 | __set_se_shares(tg->se[cpu], shares); |
1518 | spin_unlock_irqrestore(&rq->lock, flags); | |
1519 | } | |
18d95a28 | 1520 | } |
c09595f6 PZ |
1521 | |
1522 | /* | |
c8cba857 PZ |
1523 | * Re-compute the task group their per cpu shares over the given domain. |
1524 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1525 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1526 | */ |
eb755805 | 1527 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1528 | { |
ec4e0e2f | 1529 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1530 | unsigned long shares = 0; |
eb755805 | 1531 | struct sched_domain *sd = data; |
c8cba857 | 1532 | int i; |
c09595f6 | 1533 | |
758b2cdc | 1534 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1535 | /* |
1536 | * If there are currently no tasks on the cpu pretend there | |
1537 | * is one of average load so that when a new task gets to | |
1538 | * run here it will not get delayed by group starvation. | |
1539 | */ | |
1540 | weight = tg->cfs_rq[i]->load.weight; | |
1541 | if (!weight) | |
1542 | weight = NICE_0_LOAD; | |
1543 | ||
1544 | tg->cfs_rq[i]->rq_weight = weight; | |
1545 | rq_weight += weight; | |
c8cba857 | 1546 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1547 | } |
c09595f6 | 1548 | |
c8cba857 PZ |
1549 | if ((!shares && rq_weight) || shares > tg->shares) |
1550 | shares = tg->shares; | |
1551 | ||
1552 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1553 | shares = tg->shares; | |
c09595f6 | 1554 | |
758b2cdc | 1555 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1556 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1557 | |
1558 | return 0; | |
c09595f6 PZ |
1559 | } |
1560 | ||
1561 | /* | |
c8cba857 PZ |
1562 | * Compute the cpu's hierarchical load factor for each task group. |
1563 | * This needs to be done in a top-down fashion because the load of a child | |
1564 | * group is a fraction of its parents load. | |
c09595f6 | 1565 | */ |
eb755805 | 1566 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1567 | { |
c8cba857 | 1568 | unsigned long load; |
eb755805 | 1569 | long cpu = (long)data; |
c09595f6 | 1570 | |
c8cba857 PZ |
1571 | if (!tg->parent) { |
1572 | load = cpu_rq(cpu)->load.weight; | |
1573 | } else { | |
1574 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1575 | load *= tg->cfs_rq[cpu]->shares; | |
1576 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1577 | } | |
c09595f6 | 1578 | |
c8cba857 | 1579 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1580 | |
eb755805 | 1581 | return 0; |
c09595f6 PZ |
1582 | } |
1583 | ||
c8cba857 | 1584 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1585 | { |
2398f2c6 PZ |
1586 | u64 now = cpu_clock(raw_smp_processor_id()); |
1587 | s64 elapsed = now - sd->last_update; | |
1588 | ||
1589 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1590 | sd->last_update = now; | |
eb755805 | 1591 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1592 | } |
4d8d595d PZ |
1593 | } |
1594 | ||
3e5459b4 PZ |
1595 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1596 | { | |
1597 | spin_unlock(&rq->lock); | |
1598 | update_shares(sd); | |
1599 | spin_lock(&rq->lock); | |
1600 | } | |
1601 | ||
eb755805 | 1602 | static void update_h_load(long cpu) |
c09595f6 | 1603 | { |
eb755805 | 1604 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1605 | } |
1606 | ||
c09595f6 PZ |
1607 | #else |
1608 | ||
c8cba857 | 1609 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1610 | { |
1611 | } | |
1612 | ||
3e5459b4 PZ |
1613 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1614 | { | |
1615 | } | |
1616 | ||
18d95a28 PZ |
1617 | #endif |
1618 | ||
8f45e2b5 GH |
1619 | #ifdef CONFIG_PREEMPT |
1620 | ||
70574a99 | 1621 | /* |
8f45e2b5 GH |
1622 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1623 | * way at the expense of forcing extra atomic operations in all | |
1624 | * invocations. This assures that the double_lock is acquired using the | |
1625 | * same underlying policy as the spinlock_t on this architecture, which | |
1626 | * reduces latency compared to the unfair variant below. However, it | |
1627 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1628 | */ |
8f45e2b5 GH |
1629 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1630 | __releases(this_rq->lock) | |
1631 | __acquires(busiest->lock) | |
1632 | __acquires(this_rq->lock) | |
1633 | { | |
1634 | spin_unlock(&this_rq->lock); | |
1635 | double_rq_lock(this_rq, busiest); | |
1636 | ||
1637 | return 1; | |
1638 | } | |
1639 | ||
1640 | #else | |
1641 | /* | |
1642 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1643 | * latency by eliminating extra atomic operations when the locks are | |
1644 | * already in proper order on entry. This favors lower cpu-ids and will | |
1645 | * grant the double lock to lower cpus over higher ids under contention, | |
1646 | * regardless of entry order into the function. | |
1647 | */ | |
1648 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1649 | __releases(this_rq->lock) |
1650 | __acquires(busiest->lock) | |
1651 | __acquires(this_rq->lock) | |
1652 | { | |
1653 | int ret = 0; | |
1654 | ||
70574a99 AD |
1655 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1656 | if (busiest < this_rq) { | |
1657 | spin_unlock(&this_rq->lock); | |
1658 | spin_lock(&busiest->lock); | |
1659 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1660 | ret = 1; | |
1661 | } else | |
1662 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1663 | } | |
1664 | return ret; | |
1665 | } | |
1666 | ||
8f45e2b5 GH |
1667 | #endif /* CONFIG_PREEMPT */ |
1668 | ||
1669 | /* | |
1670 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1671 | */ | |
1672 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1673 | { | |
1674 | if (unlikely(!irqs_disabled())) { | |
1675 | /* printk() doesn't work good under rq->lock */ | |
1676 | spin_unlock(&this_rq->lock); | |
1677 | BUG_ON(1); | |
1678 | } | |
1679 | ||
1680 | return _double_lock_balance(this_rq, busiest); | |
1681 | } | |
1682 | ||
70574a99 AD |
1683 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1684 | __releases(busiest->lock) | |
1685 | { | |
1686 | spin_unlock(&busiest->lock); | |
1687 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1688 | } | |
18d95a28 PZ |
1689 | #endif |
1690 | ||
30432094 | 1691 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1692 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1693 | { | |
30432094 | 1694 | #ifdef CONFIG_SMP |
34e83e85 IM |
1695 | cfs_rq->shares = shares; |
1696 | #endif | |
1697 | } | |
30432094 | 1698 | #endif |
e7693a36 | 1699 | |
dd41f596 | 1700 | #include "sched_stats.h" |
dd41f596 | 1701 | #include "sched_idletask.c" |
5522d5d5 IM |
1702 | #include "sched_fair.c" |
1703 | #include "sched_rt.c" | |
dd41f596 IM |
1704 | #ifdef CONFIG_SCHED_DEBUG |
1705 | # include "sched_debug.c" | |
1706 | #endif | |
1707 | ||
1708 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1709 | #define for_each_class(class) \ |
1710 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1711 | |
c09595f6 | 1712 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1713 | { |
1714 | rq->nr_running++; | |
9c217245 IM |
1715 | } |
1716 | ||
c09595f6 | 1717 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1718 | { |
1719 | rq->nr_running--; | |
9c217245 IM |
1720 | } |
1721 | ||
45bf76df IM |
1722 | static void set_load_weight(struct task_struct *p) |
1723 | { | |
1724 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1725 | p->se.load.weight = prio_to_weight[0] * 2; |
1726 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1727 | return; | |
1728 | } | |
45bf76df | 1729 | |
dd41f596 IM |
1730 | /* |
1731 | * SCHED_IDLE tasks get minimal weight: | |
1732 | */ | |
1733 | if (p->policy == SCHED_IDLE) { | |
1734 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1735 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1736 | return; | |
1737 | } | |
71f8bd46 | 1738 | |
dd41f596 IM |
1739 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1740 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1741 | } |
1742 | ||
2087a1ad GH |
1743 | static void update_avg(u64 *avg, u64 sample) |
1744 | { | |
1745 | s64 diff = sample - *avg; | |
1746 | *avg += diff >> 3; | |
1747 | } | |
1748 | ||
8159f87e | 1749 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1750 | { |
831451ac PZ |
1751 | if (wakeup) |
1752 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1753 | ||
dd41f596 | 1754 | sched_info_queued(p); |
fd390f6a | 1755 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1756 | p->se.on_rq = 1; |
71f8bd46 IM |
1757 | } |
1758 | ||
69be72c1 | 1759 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1760 | { |
831451ac PZ |
1761 | if (sleep) { |
1762 | if (p->se.last_wakeup) { | |
1763 | update_avg(&p->se.avg_overlap, | |
1764 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1765 | p->se.last_wakeup = 0; | |
1766 | } else { | |
1767 | update_avg(&p->se.avg_wakeup, | |
1768 | sysctl_sched_wakeup_granularity); | |
1769 | } | |
2087a1ad GH |
1770 | } |
1771 | ||
46ac22ba | 1772 | sched_info_dequeued(p); |
f02231e5 | 1773 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1774 | p->se.on_rq = 0; |
71f8bd46 IM |
1775 | } |
1776 | ||
14531189 | 1777 | /* |
dd41f596 | 1778 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1779 | */ |
14531189 IM |
1780 | static inline int __normal_prio(struct task_struct *p) |
1781 | { | |
dd41f596 | 1782 | return p->static_prio; |
14531189 IM |
1783 | } |
1784 | ||
b29739f9 IM |
1785 | /* |
1786 | * Calculate the expected normal priority: i.e. priority | |
1787 | * without taking RT-inheritance into account. Might be | |
1788 | * boosted by interactivity modifiers. Changes upon fork, | |
1789 | * setprio syscalls, and whenever the interactivity | |
1790 | * estimator recalculates. | |
1791 | */ | |
36c8b586 | 1792 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1793 | { |
1794 | int prio; | |
1795 | ||
e05606d3 | 1796 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1797 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1798 | else | |
1799 | prio = __normal_prio(p); | |
1800 | return prio; | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * Calculate the current priority, i.e. the priority | |
1805 | * taken into account by the scheduler. This value might | |
1806 | * be boosted by RT tasks, or might be boosted by | |
1807 | * interactivity modifiers. Will be RT if the task got | |
1808 | * RT-boosted. If not then it returns p->normal_prio. | |
1809 | */ | |
36c8b586 | 1810 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1811 | { |
1812 | p->normal_prio = normal_prio(p); | |
1813 | /* | |
1814 | * If we are RT tasks or we were boosted to RT priority, | |
1815 | * keep the priority unchanged. Otherwise, update priority | |
1816 | * to the normal priority: | |
1817 | */ | |
1818 | if (!rt_prio(p->prio)) | |
1819 | return p->normal_prio; | |
1820 | return p->prio; | |
1821 | } | |
1822 | ||
1da177e4 | 1823 | /* |
dd41f596 | 1824 | * activate_task - move a task to the runqueue. |
1da177e4 | 1825 | */ |
dd41f596 | 1826 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1827 | { |
d9514f6c | 1828 | if (task_contributes_to_load(p)) |
dd41f596 | 1829 | rq->nr_uninterruptible--; |
1da177e4 | 1830 | |
8159f87e | 1831 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1832 | inc_nr_running(rq); |
1da177e4 LT |
1833 | } |
1834 | ||
1da177e4 LT |
1835 | /* |
1836 | * deactivate_task - remove a task from the runqueue. | |
1837 | */ | |
2e1cb74a | 1838 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1839 | { |
d9514f6c | 1840 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1841 | rq->nr_uninterruptible++; |
1842 | ||
69be72c1 | 1843 | dequeue_task(rq, p, sleep); |
c09595f6 | 1844 | dec_nr_running(rq); |
1da177e4 LT |
1845 | } |
1846 | ||
1da177e4 LT |
1847 | /** |
1848 | * task_curr - is this task currently executing on a CPU? | |
1849 | * @p: the task in question. | |
1850 | */ | |
36c8b586 | 1851 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1852 | { |
1853 | return cpu_curr(task_cpu(p)) == p; | |
1854 | } | |
1855 | ||
dd41f596 IM |
1856 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1857 | { | |
6f505b16 | 1858 | set_task_rq(p, cpu); |
dd41f596 | 1859 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1860 | /* |
1861 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1862 | * successfuly executed on another CPU. We must ensure that updates of | |
1863 | * per-task data have been completed by this moment. | |
1864 | */ | |
1865 | smp_wmb(); | |
dd41f596 | 1866 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1867 | #endif |
2dd73a4f PW |
1868 | } |
1869 | ||
cb469845 SR |
1870 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1871 | const struct sched_class *prev_class, | |
1872 | int oldprio, int running) | |
1873 | { | |
1874 | if (prev_class != p->sched_class) { | |
1875 | if (prev_class->switched_from) | |
1876 | prev_class->switched_from(rq, p, running); | |
1877 | p->sched_class->switched_to(rq, p, running); | |
1878 | } else | |
1879 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1880 | } | |
1881 | ||
1da177e4 | 1882 | #ifdef CONFIG_SMP |
c65cc870 | 1883 | |
e958b360 TG |
1884 | /* Used instead of source_load when we know the type == 0 */ |
1885 | static unsigned long weighted_cpuload(const int cpu) | |
1886 | { | |
1887 | return cpu_rq(cpu)->load.weight; | |
1888 | } | |
1889 | ||
cc367732 IM |
1890 | /* |
1891 | * Is this task likely cache-hot: | |
1892 | */ | |
e7693a36 | 1893 | static int |
cc367732 IM |
1894 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1895 | { | |
1896 | s64 delta; | |
1897 | ||
f540a608 IM |
1898 | /* |
1899 | * Buddy candidates are cache hot: | |
1900 | */ | |
4793241b PZ |
1901 | if (sched_feat(CACHE_HOT_BUDDY) && |
1902 | (&p->se == cfs_rq_of(&p->se)->next || | |
1903 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1904 | return 1; |
1905 | ||
cc367732 IM |
1906 | if (p->sched_class != &fair_sched_class) |
1907 | return 0; | |
1908 | ||
6bc1665b IM |
1909 | if (sysctl_sched_migration_cost == -1) |
1910 | return 1; | |
1911 | if (sysctl_sched_migration_cost == 0) | |
1912 | return 0; | |
1913 | ||
cc367732 IM |
1914 | delta = now - p->se.exec_start; |
1915 | ||
1916 | return delta < (s64)sysctl_sched_migration_cost; | |
1917 | } | |
1918 | ||
1919 | ||
dd41f596 | 1920 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1921 | { |
dd41f596 IM |
1922 | int old_cpu = task_cpu(p); |
1923 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1924 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1925 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1926 | u64 clock_offset; |
dd41f596 IM |
1927 | |
1928 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1929 | |
cbc34ed1 PZ |
1930 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1931 | ||
6cfb0d5d IM |
1932 | #ifdef CONFIG_SCHEDSTATS |
1933 | if (p->se.wait_start) | |
1934 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1935 | if (p->se.sleep_start) |
1936 | p->se.sleep_start -= clock_offset; | |
1937 | if (p->se.block_start) | |
1938 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1939 | if (old_cpu != new_cpu) { |
1940 | schedstat_inc(p, se.nr_migrations); | |
1941 | if (task_hot(p, old_rq->clock, NULL)) | |
1942 | schedstat_inc(p, se.nr_forced2_migrations); | |
1943 | } | |
6cfb0d5d | 1944 | #endif |
2830cf8c SV |
1945 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1946 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1947 | |
1948 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1949 | } |
1950 | ||
70b97a7f | 1951 | struct migration_req { |
1da177e4 | 1952 | struct list_head list; |
1da177e4 | 1953 | |
36c8b586 | 1954 | struct task_struct *task; |
1da177e4 LT |
1955 | int dest_cpu; |
1956 | ||
1da177e4 | 1957 | struct completion done; |
70b97a7f | 1958 | }; |
1da177e4 LT |
1959 | |
1960 | /* | |
1961 | * The task's runqueue lock must be held. | |
1962 | * Returns true if you have to wait for migration thread. | |
1963 | */ | |
36c8b586 | 1964 | static int |
70b97a7f | 1965 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1966 | { |
70b97a7f | 1967 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1968 | |
1969 | /* | |
1970 | * If the task is not on a runqueue (and not running), then | |
1971 | * it is sufficient to simply update the task's cpu field. | |
1972 | */ | |
dd41f596 | 1973 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1974 | set_task_cpu(p, dest_cpu); |
1975 | return 0; | |
1976 | } | |
1977 | ||
1978 | init_completion(&req->done); | |
1da177e4 LT |
1979 | req->task = p; |
1980 | req->dest_cpu = dest_cpu; | |
1981 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1982 | |
1da177e4 LT |
1983 | return 1; |
1984 | } | |
1985 | ||
1986 | /* | |
1987 | * wait_task_inactive - wait for a thread to unschedule. | |
1988 | * | |
85ba2d86 RM |
1989 | * If @match_state is nonzero, it's the @p->state value just checked and |
1990 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1991 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1992 | * we return a positive number (its total switch count). If a second call | |
1993 | * a short while later returns the same number, the caller can be sure that | |
1994 | * @p has remained unscheduled the whole time. | |
1995 | * | |
1da177e4 LT |
1996 | * The caller must ensure that the task *will* unschedule sometime soon, |
1997 | * else this function might spin for a *long* time. This function can't | |
1998 | * be called with interrupts off, or it may introduce deadlock with | |
1999 | * smp_call_function() if an IPI is sent by the same process we are | |
2000 | * waiting to become inactive. | |
2001 | */ | |
85ba2d86 | 2002 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2003 | { |
2004 | unsigned long flags; | |
dd41f596 | 2005 | int running, on_rq; |
85ba2d86 | 2006 | unsigned long ncsw; |
70b97a7f | 2007 | struct rq *rq; |
1da177e4 | 2008 | |
3a5c359a AK |
2009 | for (;;) { |
2010 | /* | |
2011 | * We do the initial early heuristics without holding | |
2012 | * any task-queue locks at all. We'll only try to get | |
2013 | * the runqueue lock when things look like they will | |
2014 | * work out! | |
2015 | */ | |
2016 | rq = task_rq(p); | |
fa490cfd | 2017 | |
3a5c359a AK |
2018 | /* |
2019 | * If the task is actively running on another CPU | |
2020 | * still, just relax and busy-wait without holding | |
2021 | * any locks. | |
2022 | * | |
2023 | * NOTE! Since we don't hold any locks, it's not | |
2024 | * even sure that "rq" stays as the right runqueue! | |
2025 | * But we don't care, since "task_running()" will | |
2026 | * return false if the runqueue has changed and p | |
2027 | * is actually now running somewhere else! | |
2028 | */ | |
85ba2d86 RM |
2029 | while (task_running(rq, p)) { |
2030 | if (match_state && unlikely(p->state != match_state)) | |
2031 | return 0; | |
3a5c359a | 2032 | cpu_relax(); |
85ba2d86 | 2033 | } |
fa490cfd | 2034 | |
3a5c359a AK |
2035 | /* |
2036 | * Ok, time to look more closely! We need the rq | |
2037 | * lock now, to be *sure*. If we're wrong, we'll | |
2038 | * just go back and repeat. | |
2039 | */ | |
2040 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2041 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2042 | running = task_running(rq, p); |
2043 | on_rq = p->se.on_rq; | |
85ba2d86 | 2044 | ncsw = 0; |
f31e11d8 | 2045 | if (!match_state || p->state == match_state) |
93dcf55f | 2046 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2047 | task_rq_unlock(rq, &flags); |
fa490cfd | 2048 | |
85ba2d86 RM |
2049 | /* |
2050 | * If it changed from the expected state, bail out now. | |
2051 | */ | |
2052 | if (unlikely(!ncsw)) | |
2053 | break; | |
2054 | ||
3a5c359a AK |
2055 | /* |
2056 | * Was it really running after all now that we | |
2057 | * checked with the proper locks actually held? | |
2058 | * | |
2059 | * Oops. Go back and try again.. | |
2060 | */ | |
2061 | if (unlikely(running)) { | |
2062 | cpu_relax(); | |
2063 | continue; | |
2064 | } | |
fa490cfd | 2065 | |
3a5c359a AK |
2066 | /* |
2067 | * It's not enough that it's not actively running, | |
2068 | * it must be off the runqueue _entirely_, and not | |
2069 | * preempted! | |
2070 | * | |
2071 | * So if it wa still runnable (but just not actively | |
2072 | * running right now), it's preempted, and we should | |
2073 | * yield - it could be a while. | |
2074 | */ | |
2075 | if (unlikely(on_rq)) { | |
2076 | schedule_timeout_uninterruptible(1); | |
2077 | continue; | |
2078 | } | |
fa490cfd | 2079 | |
3a5c359a AK |
2080 | /* |
2081 | * Ahh, all good. It wasn't running, and it wasn't | |
2082 | * runnable, which means that it will never become | |
2083 | * running in the future either. We're all done! | |
2084 | */ | |
2085 | break; | |
2086 | } | |
85ba2d86 RM |
2087 | |
2088 | return ncsw; | |
1da177e4 LT |
2089 | } |
2090 | ||
2091 | /*** | |
2092 | * kick_process - kick a running thread to enter/exit the kernel | |
2093 | * @p: the to-be-kicked thread | |
2094 | * | |
2095 | * Cause a process which is running on another CPU to enter | |
2096 | * kernel-mode, without any delay. (to get signals handled.) | |
2097 | * | |
2098 | * NOTE: this function doesnt have to take the runqueue lock, | |
2099 | * because all it wants to ensure is that the remote task enters | |
2100 | * the kernel. If the IPI races and the task has been migrated | |
2101 | * to another CPU then no harm is done and the purpose has been | |
2102 | * achieved as well. | |
2103 | */ | |
36c8b586 | 2104 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2105 | { |
2106 | int cpu; | |
2107 | ||
2108 | preempt_disable(); | |
2109 | cpu = task_cpu(p); | |
2110 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2111 | smp_send_reschedule(cpu); | |
2112 | preempt_enable(); | |
2113 | } | |
2114 | ||
2115 | /* | |
2dd73a4f PW |
2116 | * Return a low guess at the load of a migration-source cpu weighted |
2117 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2118 | * |
2119 | * We want to under-estimate the load of migration sources, to | |
2120 | * balance conservatively. | |
2121 | */ | |
a9957449 | 2122 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2123 | { |
70b97a7f | 2124 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2125 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2126 | |
93b75217 | 2127 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2128 | return total; |
b910472d | 2129 | |
dd41f596 | 2130 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2131 | } |
2132 | ||
2133 | /* | |
2dd73a4f PW |
2134 | * Return a high guess at the load of a migration-target cpu weighted |
2135 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2136 | */ |
a9957449 | 2137 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2138 | { |
70b97a7f | 2139 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2140 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2141 | |
93b75217 | 2142 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2143 | return total; |
3b0bd9bc | 2144 | |
dd41f596 | 2145 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2146 | } |
2147 | ||
147cbb4b NP |
2148 | /* |
2149 | * find_idlest_group finds and returns the least busy CPU group within the | |
2150 | * domain. | |
2151 | */ | |
2152 | static struct sched_group * | |
2153 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2154 | { | |
2155 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2156 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2157 | int load_idx = sd->forkexec_idx; | |
2158 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2159 | ||
2160 | do { | |
2161 | unsigned long load, avg_load; | |
2162 | int local_group; | |
2163 | int i; | |
2164 | ||
da5a5522 | 2165 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2166 | if (!cpumask_intersects(sched_group_cpus(group), |
2167 | &p->cpus_allowed)) | |
3a5c359a | 2168 | continue; |
da5a5522 | 2169 | |
758b2cdc RR |
2170 | local_group = cpumask_test_cpu(this_cpu, |
2171 | sched_group_cpus(group)); | |
147cbb4b NP |
2172 | |
2173 | /* Tally up the load of all CPUs in the group */ | |
2174 | avg_load = 0; | |
2175 | ||
758b2cdc | 2176 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2177 | /* Bias balancing toward cpus of our domain */ |
2178 | if (local_group) | |
2179 | load = source_load(i, load_idx); | |
2180 | else | |
2181 | load = target_load(i, load_idx); | |
2182 | ||
2183 | avg_load += load; | |
2184 | } | |
2185 | ||
2186 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2187 | avg_load = sg_div_cpu_power(group, |
2188 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2189 | |
2190 | if (local_group) { | |
2191 | this_load = avg_load; | |
2192 | this = group; | |
2193 | } else if (avg_load < min_load) { | |
2194 | min_load = avg_load; | |
2195 | idlest = group; | |
2196 | } | |
3a5c359a | 2197 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2198 | |
2199 | if (!idlest || 100*this_load < imbalance*min_load) | |
2200 | return NULL; | |
2201 | return idlest; | |
2202 | } | |
2203 | ||
2204 | /* | |
0feaece9 | 2205 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2206 | */ |
95cdf3b7 | 2207 | static int |
758b2cdc | 2208 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2209 | { |
2210 | unsigned long load, min_load = ULONG_MAX; | |
2211 | int idlest = -1; | |
2212 | int i; | |
2213 | ||
da5a5522 | 2214 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2215 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2216 | load = weighted_cpuload(i); |
147cbb4b NP |
2217 | |
2218 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2219 | min_load = load; | |
2220 | idlest = i; | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | return idlest; | |
2225 | } | |
2226 | ||
476d139c NP |
2227 | /* |
2228 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2229 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2230 | * SD_BALANCE_EXEC. | |
2231 | * | |
2232 | * Balance, ie. select the least loaded group. | |
2233 | * | |
2234 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2235 | * | |
2236 | * preempt must be disabled. | |
2237 | */ | |
2238 | static int sched_balance_self(int cpu, int flag) | |
2239 | { | |
2240 | struct task_struct *t = current; | |
2241 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2242 | |
c96d145e | 2243 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2244 | /* |
2245 | * If power savings logic is enabled for a domain, stop there. | |
2246 | */ | |
5c45bf27 SS |
2247 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2248 | break; | |
476d139c NP |
2249 | if (tmp->flags & flag) |
2250 | sd = tmp; | |
c96d145e | 2251 | } |
476d139c | 2252 | |
039a1c41 PZ |
2253 | if (sd) |
2254 | update_shares(sd); | |
2255 | ||
476d139c | 2256 | while (sd) { |
476d139c | 2257 | struct sched_group *group; |
1a848870 SS |
2258 | int new_cpu, weight; |
2259 | ||
2260 | if (!(sd->flags & flag)) { | |
2261 | sd = sd->child; | |
2262 | continue; | |
2263 | } | |
476d139c | 2264 | |
476d139c | 2265 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2266 | if (!group) { |
2267 | sd = sd->child; | |
2268 | continue; | |
2269 | } | |
476d139c | 2270 | |
758b2cdc | 2271 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2272 | if (new_cpu == -1 || new_cpu == cpu) { |
2273 | /* Now try balancing at a lower domain level of cpu */ | |
2274 | sd = sd->child; | |
2275 | continue; | |
2276 | } | |
476d139c | 2277 | |
1a848870 | 2278 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2279 | cpu = new_cpu; |
758b2cdc | 2280 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2281 | sd = NULL; |
476d139c | 2282 | for_each_domain(cpu, tmp) { |
758b2cdc | 2283 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2284 | break; |
2285 | if (tmp->flags & flag) | |
2286 | sd = tmp; | |
2287 | } | |
2288 | /* while loop will break here if sd == NULL */ | |
2289 | } | |
2290 | ||
2291 | return cpu; | |
2292 | } | |
2293 | ||
2294 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2295 | |
1da177e4 LT |
2296 | /*** |
2297 | * try_to_wake_up - wake up a thread | |
2298 | * @p: the to-be-woken-up thread | |
2299 | * @state: the mask of task states that can be woken | |
2300 | * @sync: do a synchronous wakeup? | |
2301 | * | |
2302 | * Put it on the run-queue if it's not already there. The "current" | |
2303 | * thread is always on the run-queue (except when the actual | |
2304 | * re-schedule is in progress), and as such you're allowed to do | |
2305 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2306 | * runnable without the overhead of this. | |
2307 | * | |
2308 | * returns failure only if the task is already active. | |
2309 | */ | |
36c8b586 | 2310 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2311 | { |
cc367732 | 2312 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2313 | unsigned long flags; |
2314 | long old_state; | |
70b97a7f | 2315 | struct rq *rq; |
1da177e4 | 2316 | |
b85d0667 IM |
2317 | if (!sched_feat(SYNC_WAKEUPS)) |
2318 | sync = 0; | |
2319 | ||
2398f2c6 PZ |
2320 | #ifdef CONFIG_SMP |
2321 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2322 | struct sched_domain *sd; | |
2323 | ||
2324 | this_cpu = raw_smp_processor_id(); | |
2325 | cpu = task_cpu(p); | |
2326 | ||
2327 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2328 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2329 | update_shares(sd); |
2330 | break; | |
2331 | } | |
2332 | } | |
2333 | } | |
2334 | #endif | |
2335 | ||
04e2f174 | 2336 | smp_wmb(); |
1da177e4 | 2337 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2338 | update_rq_clock(rq); |
1da177e4 LT |
2339 | old_state = p->state; |
2340 | if (!(old_state & state)) | |
2341 | goto out; | |
2342 | ||
dd41f596 | 2343 | if (p->se.on_rq) |
1da177e4 LT |
2344 | goto out_running; |
2345 | ||
2346 | cpu = task_cpu(p); | |
cc367732 | 2347 | orig_cpu = cpu; |
1da177e4 LT |
2348 | this_cpu = smp_processor_id(); |
2349 | ||
2350 | #ifdef CONFIG_SMP | |
2351 | if (unlikely(task_running(rq, p))) | |
2352 | goto out_activate; | |
2353 | ||
5d2f5a61 DA |
2354 | cpu = p->sched_class->select_task_rq(p, sync); |
2355 | if (cpu != orig_cpu) { | |
2356 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2357 | task_rq_unlock(rq, &flags); |
2358 | /* might preempt at this point */ | |
2359 | rq = task_rq_lock(p, &flags); | |
2360 | old_state = p->state; | |
2361 | if (!(old_state & state)) | |
2362 | goto out; | |
dd41f596 | 2363 | if (p->se.on_rq) |
1da177e4 LT |
2364 | goto out_running; |
2365 | ||
2366 | this_cpu = smp_processor_id(); | |
2367 | cpu = task_cpu(p); | |
2368 | } | |
2369 | ||
e7693a36 GH |
2370 | #ifdef CONFIG_SCHEDSTATS |
2371 | schedstat_inc(rq, ttwu_count); | |
2372 | if (cpu == this_cpu) | |
2373 | schedstat_inc(rq, ttwu_local); | |
2374 | else { | |
2375 | struct sched_domain *sd; | |
2376 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2377 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2378 | schedstat_inc(sd, ttwu_wake_remote); |
2379 | break; | |
2380 | } | |
2381 | } | |
2382 | } | |
6d6bc0ad | 2383 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2384 | |
1da177e4 LT |
2385 | out_activate: |
2386 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2387 | schedstat_inc(p, se.nr_wakeups); |
2388 | if (sync) | |
2389 | schedstat_inc(p, se.nr_wakeups_sync); | |
2390 | if (orig_cpu != cpu) | |
2391 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2392 | if (cpu == this_cpu) | |
2393 | schedstat_inc(p, se.nr_wakeups_local); | |
2394 | else | |
2395 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2396 | activate_task(rq, p, 1); |
1da177e4 LT |
2397 | success = 1; |
2398 | ||
831451ac PZ |
2399 | /* |
2400 | * Only attribute actual wakeups done by this task. | |
2401 | */ | |
2402 | if (!in_interrupt()) { | |
2403 | struct sched_entity *se = ¤t->se; | |
2404 | u64 sample = se->sum_exec_runtime; | |
2405 | ||
2406 | if (se->last_wakeup) | |
2407 | sample -= se->last_wakeup; | |
2408 | else | |
2409 | sample -= se->start_runtime; | |
2410 | update_avg(&se->avg_wakeup, sample); | |
2411 | ||
2412 | se->last_wakeup = se->sum_exec_runtime; | |
2413 | } | |
2414 | ||
1da177e4 | 2415 | out_running: |
468a15bb | 2416 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2417 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2418 | |
1da177e4 | 2419 | p->state = TASK_RUNNING; |
9a897c5a SR |
2420 | #ifdef CONFIG_SMP |
2421 | if (p->sched_class->task_wake_up) | |
2422 | p->sched_class->task_wake_up(rq, p); | |
2423 | #endif | |
1da177e4 LT |
2424 | out: |
2425 | task_rq_unlock(rq, &flags); | |
2426 | ||
2427 | return success; | |
2428 | } | |
2429 | ||
7ad5b3a5 | 2430 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2431 | { |
d9514f6c | 2432 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2433 | } |
1da177e4 LT |
2434 | EXPORT_SYMBOL(wake_up_process); |
2435 | ||
7ad5b3a5 | 2436 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2437 | { |
2438 | return try_to_wake_up(p, state, 0); | |
2439 | } | |
2440 | ||
1da177e4 LT |
2441 | /* |
2442 | * Perform scheduler related setup for a newly forked process p. | |
2443 | * p is forked by current. | |
dd41f596 IM |
2444 | * |
2445 | * __sched_fork() is basic setup used by init_idle() too: | |
2446 | */ | |
2447 | static void __sched_fork(struct task_struct *p) | |
2448 | { | |
dd41f596 IM |
2449 | p->se.exec_start = 0; |
2450 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2451 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2452 | p->se.last_wakeup = 0; |
2453 | p->se.avg_overlap = 0; | |
831451ac PZ |
2454 | p->se.start_runtime = 0; |
2455 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2456 | |
2457 | #ifdef CONFIG_SCHEDSTATS | |
2458 | p->se.wait_start = 0; | |
dd41f596 IM |
2459 | p->se.sum_sleep_runtime = 0; |
2460 | p->se.sleep_start = 0; | |
dd41f596 IM |
2461 | p->se.block_start = 0; |
2462 | p->se.sleep_max = 0; | |
2463 | p->se.block_max = 0; | |
2464 | p->se.exec_max = 0; | |
eba1ed4b | 2465 | p->se.slice_max = 0; |
dd41f596 | 2466 | p->se.wait_max = 0; |
6cfb0d5d | 2467 | #endif |
476d139c | 2468 | |
fa717060 | 2469 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2470 | p->se.on_rq = 0; |
4a55bd5e | 2471 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2472 | |
e107be36 AK |
2473 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2474 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2475 | #endif | |
2476 | ||
1da177e4 LT |
2477 | /* |
2478 | * We mark the process as running here, but have not actually | |
2479 | * inserted it onto the runqueue yet. This guarantees that | |
2480 | * nobody will actually run it, and a signal or other external | |
2481 | * event cannot wake it up and insert it on the runqueue either. | |
2482 | */ | |
2483 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2484 | } |
2485 | ||
2486 | /* | |
2487 | * fork()/clone()-time setup: | |
2488 | */ | |
2489 | void sched_fork(struct task_struct *p, int clone_flags) | |
2490 | { | |
2491 | int cpu = get_cpu(); | |
2492 | ||
2493 | __sched_fork(p); | |
2494 | ||
2495 | #ifdef CONFIG_SMP | |
2496 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2497 | #endif | |
02e4bac2 | 2498 | set_task_cpu(p, cpu); |
b29739f9 IM |
2499 | |
2500 | /* | |
2501 | * Make sure we do not leak PI boosting priority to the child: | |
2502 | */ | |
2503 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2504 | if (!rt_prio(p->prio)) |
2505 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2506 | |
52f17b6c | 2507 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2508 | if (likely(sched_info_on())) |
52f17b6c | 2509 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2510 | #endif |
d6077cb8 | 2511 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2512 | p->oncpu = 0; |
2513 | #endif | |
1da177e4 | 2514 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2515 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2516 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2517 | #endif |
917b627d GH |
2518 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2519 | ||
476d139c | 2520 | put_cpu(); |
1da177e4 LT |
2521 | } |
2522 | ||
2523 | /* | |
2524 | * wake_up_new_task - wake up a newly created task for the first time. | |
2525 | * | |
2526 | * This function will do some initial scheduler statistics housekeeping | |
2527 | * that must be done for every newly created context, then puts the task | |
2528 | * on the runqueue and wakes it. | |
2529 | */ | |
7ad5b3a5 | 2530 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2531 | { |
2532 | unsigned long flags; | |
dd41f596 | 2533 | struct rq *rq; |
1da177e4 LT |
2534 | |
2535 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2536 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2537 | update_rq_clock(rq); |
1da177e4 LT |
2538 | |
2539 | p->prio = effective_prio(p); | |
2540 | ||
b9dca1e0 | 2541 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2542 | activate_task(rq, p, 0); |
1da177e4 | 2543 | } else { |
1da177e4 | 2544 | /* |
dd41f596 IM |
2545 | * Let the scheduling class do new task startup |
2546 | * management (if any): | |
1da177e4 | 2547 | */ |
ee0827d8 | 2548 | p->sched_class->task_new(rq, p); |
c09595f6 | 2549 | inc_nr_running(rq); |
1da177e4 | 2550 | } |
c71dd42d | 2551 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2552 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2553 | #ifdef CONFIG_SMP |
2554 | if (p->sched_class->task_wake_up) | |
2555 | p->sched_class->task_wake_up(rq, p); | |
2556 | #endif | |
dd41f596 | 2557 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2558 | } |
2559 | ||
e107be36 AK |
2560 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2561 | ||
2562 | /** | |
421cee29 RD |
2563 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2564 | * @notifier: notifier struct to register | |
e107be36 AK |
2565 | */ |
2566 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2567 | { | |
2568 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2569 | } | |
2570 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2571 | ||
2572 | /** | |
2573 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2574 | * @notifier: notifier struct to unregister |
e107be36 AK |
2575 | * |
2576 | * This is safe to call from within a preemption notifier. | |
2577 | */ | |
2578 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2579 | { | |
2580 | hlist_del(¬ifier->link); | |
2581 | } | |
2582 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2583 | ||
2584 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2585 | { | |
2586 | struct preempt_notifier *notifier; | |
2587 | struct hlist_node *node; | |
2588 | ||
2589 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2590 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2591 | } | |
2592 | ||
2593 | static void | |
2594 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2595 | struct task_struct *next) | |
2596 | { | |
2597 | struct preempt_notifier *notifier; | |
2598 | struct hlist_node *node; | |
2599 | ||
2600 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2601 | notifier->ops->sched_out(notifier, next); | |
2602 | } | |
2603 | ||
6d6bc0ad | 2604 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2605 | |
2606 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2607 | { | |
2608 | } | |
2609 | ||
2610 | static void | |
2611 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2612 | struct task_struct *next) | |
2613 | { | |
2614 | } | |
2615 | ||
6d6bc0ad | 2616 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2617 | |
4866cde0 NP |
2618 | /** |
2619 | * prepare_task_switch - prepare to switch tasks | |
2620 | * @rq: the runqueue preparing to switch | |
421cee29 | 2621 | * @prev: the current task that is being switched out |
4866cde0 NP |
2622 | * @next: the task we are going to switch to. |
2623 | * | |
2624 | * This is called with the rq lock held and interrupts off. It must | |
2625 | * be paired with a subsequent finish_task_switch after the context | |
2626 | * switch. | |
2627 | * | |
2628 | * prepare_task_switch sets up locking and calls architecture specific | |
2629 | * hooks. | |
2630 | */ | |
e107be36 AK |
2631 | static inline void |
2632 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2633 | struct task_struct *next) | |
4866cde0 | 2634 | { |
e107be36 | 2635 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2636 | prepare_lock_switch(rq, next); |
2637 | prepare_arch_switch(next); | |
2638 | } | |
2639 | ||
1da177e4 LT |
2640 | /** |
2641 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2642 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2643 | * @prev: the thread we just switched away from. |
2644 | * | |
4866cde0 NP |
2645 | * finish_task_switch must be called after the context switch, paired |
2646 | * with a prepare_task_switch call before the context switch. | |
2647 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2648 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2649 | * |
2650 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2651 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2652 | * with the lock held can cause deadlocks; see schedule() for |
2653 | * details.) | |
2654 | */ | |
a9957449 | 2655 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2656 | __releases(rq->lock) |
2657 | { | |
1da177e4 | 2658 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2659 | long prev_state; |
967fc046 GH |
2660 | #ifdef CONFIG_SMP |
2661 | int post_schedule = 0; | |
2662 | ||
2663 | if (current->sched_class->needs_post_schedule) | |
2664 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2665 | #endif | |
1da177e4 LT |
2666 | |
2667 | rq->prev_mm = NULL; | |
2668 | ||
2669 | /* | |
2670 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2671 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2672 | * schedule one last time. The schedule call will never return, and |
2673 | * the scheduled task must drop that reference. | |
c394cc9f | 2674 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2675 | * still held, otherwise prev could be scheduled on another cpu, die |
2676 | * there before we look at prev->state, and then the reference would | |
2677 | * be dropped twice. | |
2678 | * Manfred Spraul <manfred@colorfullife.com> | |
2679 | */ | |
55a101f8 | 2680 | prev_state = prev->state; |
4866cde0 NP |
2681 | finish_arch_switch(prev); |
2682 | finish_lock_switch(rq, prev); | |
9a897c5a | 2683 | #ifdef CONFIG_SMP |
967fc046 | 2684 | if (post_schedule) |
9a897c5a SR |
2685 | current->sched_class->post_schedule(rq); |
2686 | #endif | |
e8fa1362 | 2687 | |
e107be36 | 2688 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2689 | if (mm) |
2690 | mmdrop(mm); | |
c394cc9f | 2691 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2692 | /* |
2693 | * Remove function-return probe instances associated with this | |
2694 | * task and put them back on the free list. | |
9761eea8 | 2695 | */ |
c6fd91f0 | 2696 | kprobe_flush_task(prev); |
1da177e4 | 2697 | put_task_struct(prev); |
c6fd91f0 | 2698 | } |
1da177e4 LT |
2699 | } |
2700 | ||
2701 | /** | |
2702 | * schedule_tail - first thing a freshly forked thread must call. | |
2703 | * @prev: the thread we just switched away from. | |
2704 | */ | |
36c8b586 | 2705 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2706 | __releases(rq->lock) |
2707 | { | |
70b97a7f IM |
2708 | struct rq *rq = this_rq(); |
2709 | ||
4866cde0 NP |
2710 | finish_task_switch(rq, prev); |
2711 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2712 | /* In this case, finish_task_switch does not reenable preemption */ | |
2713 | preempt_enable(); | |
2714 | #endif | |
1da177e4 | 2715 | if (current->set_child_tid) |
b488893a | 2716 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2717 | } |
2718 | ||
2719 | /* | |
2720 | * context_switch - switch to the new MM and the new | |
2721 | * thread's register state. | |
2722 | */ | |
dd41f596 | 2723 | static inline void |
70b97a7f | 2724 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2725 | struct task_struct *next) |
1da177e4 | 2726 | { |
dd41f596 | 2727 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2728 | |
e107be36 | 2729 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2730 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2731 | mm = next->mm; |
2732 | oldmm = prev->active_mm; | |
9226d125 ZA |
2733 | /* |
2734 | * For paravirt, this is coupled with an exit in switch_to to | |
2735 | * combine the page table reload and the switch backend into | |
2736 | * one hypercall. | |
2737 | */ | |
2738 | arch_enter_lazy_cpu_mode(); | |
2739 | ||
dd41f596 | 2740 | if (unlikely(!mm)) { |
1da177e4 LT |
2741 | next->active_mm = oldmm; |
2742 | atomic_inc(&oldmm->mm_count); | |
2743 | enter_lazy_tlb(oldmm, next); | |
2744 | } else | |
2745 | switch_mm(oldmm, mm, next); | |
2746 | ||
dd41f596 | 2747 | if (unlikely(!prev->mm)) { |
1da177e4 | 2748 | prev->active_mm = NULL; |
1da177e4 LT |
2749 | rq->prev_mm = oldmm; |
2750 | } | |
3a5f5e48 IM |
2751 | /* |
2752 | * Since the runqueue lock will be released by the next | |
2753 | * task (which is an invalid locking op but in the case | |
2754 | * of the scheduler it's an obvious special-case), so we | |
2755 | * do an early lockdep release here: | |
2756 | */ | |
2757 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2758 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2759 | #endif |
1da177e4 LT |
2760 | |
2761 | /* Here we just switch the register state and the stack. */ | |
2762 | switch_to(prev, next, prev); | |
2763 | ||
dd41f596 IM |
2764 | barrier(); |
2765 | /* | |
2766 | * this_rq must be evaluated again because prev may have moved | |
2767 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2768 | * frame will be invalid. | |
2769 | */ | |
2770 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2771 | } |
2772 | ||
2773 | /* | |
2774 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2775 | * | |
2776 | * externally visible scheduler statistics: current number of runnable | |
2777 | * threads, current number of uninterruptible-sleeping threads, total | |
2778 | * number of context switches performed since bootup. | |
2779 | */ | |
2780 | unsigned long nr_running(void) | |
2781 | { | |
2782 | unsigned long i, sum = 0; | |
2783 | ||
2784 | for_each_online_cpu(i) | |
2785 | sum += cpu_rq(i)->nr_running; | |
2786 | ||
2787 | return sum; | |
2788 | } | |
2789 | ||
2790 | unsigned long nr_uninterruptible(void) | |
2791 | { | |
2792 | unsigned long i, sum = 0; | |
2793 | ||
0a945022 | 2794 | for_each_possible_cpu(i) |
1da177e4 LT |
2795 | sum += cpu_rq(i)->nr_uninterruptible; |
2796 | ||
2797 | /* | |
2798 | * Since we read the counters lockless, it might be slightly | |
2799 | * inaccurate. Do not allow it to go below zero though: | |
2800 | */ | |
2801 | if (unlikely((long)sum < 0)) | |
2802 | sum = 0; | |
2803 | ||
2804 | return sum; | |
2805 | } | |
2806 | ||
2807 | unsigned long long nr_context_switches(void) | |
2808 | { | |
cc94abfc SR |
2809 | int i; |
2810 | unsigned long long sum = 0; | |
1da177e4 | 2811 | |
0a945022 | 2812 | for_each_possible_cpu(i) |
1da177e4 LT |
2813 | sum += cpu_rq(i)->nr_switches; |
2814 | ||
2815 | return sum; | |
2816 | } | |
2817 | ||
2818 | unsigned long nr_iowait(void) | |
2819 | { | |
2820 | unsigned long i, sum = 0; | |
2821 | ||
0a945022 | 2822 | for_each_possible_cpu(i) |
1da177e4 LT |
2823 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2824 | ||
2825 | return sum; | |
2826 | } | |
2827 | ||
db1b1fef JS |
2828 | unsigned long nr_active(void) |
2829 | { | |
2830 | unsigned long i, running = 0, uninterruptible = 0; | |
2831 | ||
2832 | for_each_online_cpu(i) { | |
2833 | running += cpu_rq(i)->nr_running; | |
2834 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2835 | } | |
2836 | ||
2837 | if (unlikely((long)uninterruptible < 0)) | |
2838 | uninterruptible = 0; | |
2839 | ||
2840 | return running + uninterruptible; | |
2841 | } | |
2842 | ||
48f24c4d | 2843 | /* |
dd41f596 IM |
2844 | * Update rq->cpu_load[] statistics. This function is usually called every |
2845 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2846 | */ |
dd41f596 | 2847 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2848 | { |
495eca49 | 2849 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2850 | int i, scale; |
2851 | ||
2852 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2853 | |
2854 | /* Update our load: */ | |
2855 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2856 | unsigned long old_load, new_load; | |
2857 | ||
2858 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2859 | ||
2860 | old_load = this_rq->cpu_load[i]; | |
2861 | new_load = this_load; | |
a25707f3 IM |
2862 | /* |
2863 | * Round up the averaging division if load is increasing. This | |
2864 | * prevents us from getting stuck on 9 if the load is 10, for | |
2865 | * example. | |
2866 | */ | |
2867 | if (new_load > old_load) | |
2868 | new_load += scale-1; | |
dd41f596 IM |
2869 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2870 | } | |
48f24c4d IM |
2871 | } |
2872 | ||
dd41f596 IM |
2873 | #ifdef CONFIG_SMP |
2874 | ||
1da177e4 LT |
2875 | /* |
2876 | * double_rq_lock - safely lock two runqueues | |
2877 | * | |
2878 | * Note this does not disable interrupts like task_rq_lock, | |
2879 | * you need to do so manually before calling. | |
2880 | */ | |
70b97a7f | 2881 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2882 | __acquires(rq1->lock) |
2883 | __acquires(rq2->lock) | |
2884 | { | |
054b9108 | 2885 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2886 | if (rq1 == rq2) { |
2887 | spin_lock(&rq1->lock); | |
2888 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2889 | } else { | |
c96d145e | 2890 | if (rq1 < rq2) { |
1da177e4 | 2891 | spin_lock(&rq1->lock); |
5e710e37 | 2892 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2893 | } else { |
2894 | spin_lock(&rq2->lock); | |
5e710e37 | 2895 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2896 | } |
2897 | } | |
6e82a3be IM |
2898 | update_rq_clock(rq1); |
2899 | update_rq_clock(rq2); | |
1da177e4 LT |
2900 | } |
2901 | ||
2902 | /* | |
2903 | * double_rq_unlock - safely unlock two runqueues | |
2904 | * | |
2905 | * Note this does not restore interrupts like task_rq_unlock, | |
2906 | * you need to do so manually after calling. | |
2907 | */ | |
70b97a7f | 2908 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2909 | __releases(rq1->lock) |
2910 | __releases(rq2->lock) | |
2911 | { | |
2912 | spin_unlock(&rq1->lock); | |
2913 | if (rq1 != rq2) | |
2914 | spin_unlock(&rq2->lock); | |
2915 | else | |
2916 | __release(rq2->lock); | |
2917 | } | |
2918 | ||
1da177e4 LT |
2919 | /* |
2920 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2921 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2922 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2923 | * the cpu_allowed mask is restored. |
2924 | */ | |
36c8b586 | 2925 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2926 | { |
70b97a7f | 2927 | struct migration_req req; |
1da177e4 | 2928 | unsigned long flags; |
70b97a7f | 2929 | struct rq *rq; |
1da177e4 LT |
2930 | |
2931 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2932 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2933 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2934 | goto out; |
2935 | ||
2936 | /* force the process onto the specified CPU */ | |
2937 | if (migrate_task(p, dest_cpu, &req)) { | |
2938 | /* Need to wait for migration thread (might exit: take ref). */ | |
2939 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2940 | |
1da177e4 LT |
2941 | get_task_struct(mt); |
2942 | task_rq_unlock(rq, &flags); | |
2943 | wake_up_process(mt); | |
2944 | put_task_struct(mt); | |
2945 | wait_for_completion(&req.done); | |
36c8b586 | 2946 | |
1da177e4 LT |
2947 | return; |
2948 | } | |
2949 | out: | |
2950 | task_rq_unlock(rq, &flags); | |
2951 | } | |
2952 | ||
2953 | /* | |
476d139c NP |
2954 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2955 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2956 | */ |
2957 | void sched_exec(void) | |
2958 | { | |
1da177e4 | 2959 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2960 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2961 | put_cpu(); |
476d139c NP |
2962 | if (new_cpu != this_cpu) |
2963 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2964 | } |
2965 | ||
2966 | /* | |
2967 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2968 | * Both runqueues must be locked. | |
2969 | */ | |
dd41f596 IM |
2970 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2971 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2972 | { |
2e1cb74a | 2973 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2974 | set_task_cpu(p, this_cpu); |
dd41f596 | 2975 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2976 | /* |
2977 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2978 | * to be always true for them. | |
2979 | */ | |
15afe09b | 2980 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2981 | } |
2982 | ||
2983 | /* | |
2984 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2985 | */ | |
858119e1 | 2986 | static |
70b97a7f | 2987 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2988 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2989 | int *all_pinned) |
1da177e4 LT |
2990 | { |
2991 | /* | |
2992 | * We do not migrate tasks that are: | |
2993 | * 1) running (obviously), or | |
2994 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2995 | * 3) are cache-hot on their current CPU. | |
2996 | */ | |
96f874e2 | 2997 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 2998 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 2999 | return 0; |
cc367732 | 3000 | } |
81026794 NP |
3001 | *all_pinned = 0; |
3002 | ||
cc367732 IM |
3003 | if (task_running(rq, p)) { |
3004 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3005 | return 0; |
cc367732 | 3006 | } |
1da177e4 | 3007 | |
da84d961 IM |
3008 | /* |
3009 | * Aggressive migration if: | |
3010 | * 1) task is cache cold, or | |
3011 | * 2) too many balance attempts have failed. | |
3012 | */ | |
3013 | ||
6bc1665b IM |
3014 | if (!task_hot(p, rq->clock, sd) || |
3015 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3016 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 3017 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 3018 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3019 | schedstat_inc(p, se.nr_forced_migrations); |
3020 | } | |
da84d961 IM |
3021 | #endif |
3022 | return 1; | |
3023 | } | |
3024 | ||
cc367732 IM |
3025 | if (task_hot(p, rq->clock, sd)) { |
3026 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 3027 | return 0; |
cc367732 | 3028 | } |
1da177e4 LT |
3029 | return 1; |
3030 | } | |
3031 | ||
e1d1484f PW |
3032 | static unsigned long |
3033 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3034 | unsigned long max_load_move, struct sched_domain *sd, | |
3035 | enum cpu_idle_type idle, int *all_pinned, | |
3036 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3037 | { |
051c6764 | 3038 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3039 | struct task_struct *p; |
3040 | long rem_load_move = max_load_move; | |
1da177e4 | 3041 | |
e1d1484f | 3042 | if (max_load_move == 0) |
1da177e4 LT |
3043 | goto out; |
3044 | ||
81026794 NP |
3045 | pinned = 1; |
3046 | ||
1da177e4 | 3047 | /* |
dd41f596 | 3048 | * Start the load-balancing iterator: |
1da177e4 | 3049 | */ |
dd41f596 IM |
3050 | p = iterator->start(iterator->arg); |
3051 | next: | |
b82d9fdd | 3052 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3053 | goto out; |
051c6764 PZ |
3054 | |
3055 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3056 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3057 | p = iterator->next(iterator->arg); |
3058 | goto next; | |
1da177e4 LT |
3059 | } |
3060 | ||
dd41f596 | 3061 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3062 | pulled++; |
dd41f596 | 3063 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3064 | |
7e96fa58 GH |
3065 | #ifdef CONFIG_PREEMPT |
3066 | /* | |
3067 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3068 | * will stop after the first task is pulled to minimize the critical | |
3069 | * section. | |
3070 | */ | |
3071 | if (idle == CPU_NEWLY_IDLE) | |
3072 | goto out; | |
3073 | #endif | |
3074 | ||
2dd73a4f | 3075 | /* |
b82d9fdd | 3076 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3077 | */ |
e1d1484f | 3078 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3079 | if (p->prio < *this_best_prio) |
3080 | *this_best_prio = p->prio; | |
dd41f596 IM |
3081 | p = iterator->next(iterator->arg); |
3082 | goto next; | |
1da177e4 LT |
3083 | } |
3084 | out: | |
3085 | /* | |
e1d1484f | 3086 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3087 | * so we can safely collect pull_task() stats here rather than |
3088 | * inside pull_task(). | |
3089 | */ | |
3090 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3091 | |
3092 | if (all_pinned) | |
3093 | *all_pinned = pinned; | |
e1d1484f PW |
3094 | |
3095 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3096 | } |
3097 | ||
dd41f596 | 3098 | /* |
43010659 PW |
3099 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3100 | * this_rq, as part of a balancing operation within domain "sd". | |
3101 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3102 | * |
3103 | * Called with both runqueues locked. | |
3104 | */ | |
3105 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3106 | unsigned long max_load_move, |
dd41f596 IM |
3107 | struct sched_domain *sd, enum cpu_idle_type idle, |
3108 | int *all_pinned) | |
3109 | { | |
5522d5d5 | 3110 | const struct sched_class *class = sched_class_highest; |
43010659 | 3111 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3112 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3113 | |
3114 | do { | |
43010659 PW |
3115 | total_load_moved += |
3116 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3117 | max_load_move - total_load_moved, |
a4ac01c3 | 3118 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3119 | class = class->next; |
c4acb2c0 | 3120 | |
7e96fa58 GH |
3121 | #ifdef CONFIG_PREEMPT |
3122 | /* | |
3123 | * NEWIDLE balancing is a source of latency, so preemptible | |
3124 | * kernels will stop after the first task is pulled to minimize | |
3125 | * the critical section. | |
3126 | */ | |
c4acb2c0 GH |
3127 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3128 | break; | |
7e96fa58 | 3129 | #endif |
43010659 | 3130 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3131 | |
43010659 PW |
3132 | return total_load_moved > 0; |
3133 | } | |
3134 | ||
e1d1484f PW |
3135 | static int |
3136 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3137 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3138 | struct rq_iterator *iterator) | |
3139 | { | |
3140 | struct task_struct *p = iterator->start(iterator->arg); | |
3141 | int pinned = 0; | |
3142 | ||
3143 | while (p) { | |
3144 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3145 | pull_task(busiest, p, this_rq, this_cpu); | |
3146 | /* | |
3147 | * Right now, this is only the second place pull_task() | |
3148 | * is called, so we can safely collect pull_task() | |
3149 | * stats here rather than inside pull_task(). | |
3150 | */ | |
3151 | schedstat_inc(sd, lb_gained[idle]); | |
3152 | ||
3153 | return 1; | |
3154 | } | |
3155 | p = iterator->next(iterator->arg); | |
3156 | } | |
3157 | ||
3158 | return 0; | |
3159 | } | |
3160 | ||
43010659 PW |
3161 | /* |
3162 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3163 | * part of active balancing operations within "domain". | |
3164 | * Returns 1 if successful and 0 otherwise. | |
3165 | * | |
3166 | * Called with both runqueues locked. | |
3167 | */ | |
3168 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3169 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3170 | { | |
5522d5d5 | 3171 | const struct sched_class *class; |
43010659 PW |
3172 | |
3173 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3174 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3175 | return 1; |
3176 | ||
3177 | return 0; | |
dd41f596 IM |
3178 | } |
3179 | ||
1da177e4 LT |
3180 | /* |
3181 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3182 | * domain. It calculates and returns the amount of weighted load which |
3183 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3184 | */ |
3185 | static struct sched_group * | |
3186 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3187 | unsigned long *imbalance, enum cpu_idle_type idle, |
96f874e2 | 3188 | int *sd_idle, const struct cpumask *cpus, int *balance) |
1da177e4 LT |
3189 | { |
3190 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3191 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3192 | unsigned long max_pull; |
2dd73a4f PW |
3193 | unsigned long busiest_load_per_task, busiest_nr_running; |
3194 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3195 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3196 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3197 | int power_savings_balance = 1; | |
3198 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3199 | unsigned long min_nr_running = ULONG_MAX; | |
3200 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3201 | #endif | |
1da177e4 LT |
3202 | |
3203 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3204 | busiest_load_per_task = busiest_nr_running = 0; |
3205 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3206 | |
d15bcfdb | 3207 | if (idle == CPU_NOT_IDLE) |
7897986b | 3208 | load_idx = sd->busy_idx; |
d15bcfdb | 3209 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3210 | load_idx = sd->newidle_idx; |
3211 | else | |
3212 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3213 | |
3214 | do { | |
908a7c1b | 3215 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3216 | int local_group; |
3217 | int i; | |
908a7c1b | 3218 | int __group_imb = 0; |
783609c6 | 3219 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3220 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3221 | unsigned long sum_avg_load_per_task; |
3222 | unsigned long avg_load_per_task; | |
1da177e4 | 3223 | |
758b2cdc RR |
3224 | local_group = cpumask_test_cpu(this_cpu, |
3225 | sched_group_cpus(group)); | |
1da177e4 | 3226 | |
783609c6 | 3227 | if (local_group) |
758b2cdc | 3228 | balance_cpu = cpumask_first(sched_group_cpus(group)); |
783609c6 | 3229 | |
1da177e4 | 3230 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3231 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3232 | sum_avg_load_per_task = avg_load_per_task = 0; |
3233 | ||
908a7c1b KC |
3234 | max_cpu_load = 0; |
3235 | min_cpu_load = ~0UL; | |
1da177e4 | 3236 | |
758b2cdc RR |
3237 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3238 | struct rq *rq = cpu_rq(i); | |
2dd73a4f | 3239 | |
9439aab8 | 3240 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3241 | *sd_idle = 0; |
3242 | ||
1da177e4 | 3243 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3244 | if (local_group) { |
3245 | if (idle_cpu(i) && !first_idle_cpu) { | |
3246 | first_idle_cpu = 1; | |
3247 | balance_cpu = i; | |
3248 | } | |
3249 | ||
a2000572 | 3250 | load = target_load(i, load_idx); |
908a7c1b | 3251 | } else { |
a2000572 | 3252 | load = source_load(i, load_idx); |
908a7c1b KC |
3253 | if (load > max_cpu_load) |
3254 | max_cpu_load = load; | |
3255 | if (min_cpu_load > load) | |
3256 | min_cpu_load = load; | |
3257 | } | |
1da177e4 LT |
3258 | |
3259 | avg_load += load; | |
2dd73a4f | 3260 | sum_nr_running += rq->nr_running; |
dd41f596 | 3261 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3262 | |
3263 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3264 | } |
3265 | ||
783609c6 SS |
3266 | /* |
3267 | * First idle cpu or the first cpu(busiest) in this sched group | |
3268 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3269 | * domains. In the newly idle case, we will allow all the cpu's |
3270 | * to do the newly idle load balance. | |
783609c6 | 3271 | */ |
9439aab8 SS |
3272 | if (idle != CPU_NEWLY_IDLE && local_group && |
3273 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3274 | *balance = 0; |
3275 | goto ret; | |
3276 | } | |
3277 | ||
1da177e4 | 3278 | total_load += avg_load; |
5517d86b | 3279 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3280 | |
3281 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3282 | avg_load = sg_div_cpu_power(group, |
3283 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3284 | |
408ed066 PZ |
3285 | |
3286 | /* | |
3287 | * Consider the group unbalanced when the imbalance is larger | |
3288 | * than the average weight of two tasks. | |
3289 | * | |
3290 | * APZ: with cgroup the avg task weight can vary wildly and | |
3291 | * might not be a suitable number - should we keep a | |
3292 | * normalized nr_running number somewhere that negates | |
3293 | * the hierarchy? | |
3294 | */ | |
3295 | avg_load_per_task = sg_div_cpu_power(group, | |
3296 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3297 | ||
3298 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3299 | __group_imb = 1; |
3300 | ||
5517d86b | 3301 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3302 | |
1da177e4 LT |
3303 | if (local_group) { |
3304 | this_load = avg_load; | |
3305 | this = group; | |
2dd73a4f PW |
3306 | this_nr_running = sum_nr_running; |
3307 | this_load_per_task = sum_weighted_load; | |
3308 | } else if (avg_load > max_load && | |
908a7c1b | 3309 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3310 | max_load = avg_load; |
3311 | busiest = group; | |
2dd73a4f PW |
3312 | busiest_nr_running = sum_nr_running; |
3313 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3314 | group_imb = __group_imb; |
1da177e4 | 3315 | } |
5c45bf27 SS |
3316 | |
3317 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3318 | /* | |
3319 | * Busy processors will not participate in power savings | |
3320 | * balance. | |
3321 | */ | |
dd41f596 IM |
3322 | if (idle == CPU_NOT_IDLE || |
3323 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3324 | goto group_next; | |
5c45bf27 SS |
3325 | |
3326 | /* | |
3327 | * If the local group is idle or completely loaded | |
3328 | * no need to do power savings balance at this domain | |
3329 | */ | |
3330 | if (local_group && (this_nr_running >= group_capacity || | |
3331 | !this_nr_running)) | |
3332 | power_savings_balance = 0; | |
3333 | ||
dd41f596 | 3334 | /* |
5c45bf27 SS |
3335 | * If a group is already running at full capacity or idle, |
3336 | * don't include that group in power savings calculations | |
dd41f596 IM |
3337 | */ |
3338 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3339 | || !sum_nr_running) |
dd41f596 | 3340 | goto group_next; |
5c45bf27 | 3341 | |
dd41f596 | 3342 | /* |
5c45bf27 | 3343 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3344 | * This is the group from where we need to pick up the load |
3345 | * for saving power | |
3346 | */ | |
3347 | if ((sum_nr_running < min_nr_running) || | |
3348 | (sum_nr_running == min_nr_running && | |
d5679bd1 | 3349 | cpumask_first(sched_group_cpus(group)) > |
758b2cdc | 3350 | cpumask_first(sched_group_cpus(group_min)))) { |
dd41f596 IM |
3351 | group_min = group; |
3352 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3353 | min_load_per_task = sum_weighted_load / |
3354 | sum_nr_running; | |
dd41f596 | 3355 | } |
5c45bf27 | 3356 | |
dd41f596 | 3357 | /* |
5c45bf27 | 3358 | * Calculate the group which is almost near its |
dd41f596 IM |
3359 | * capacity but still has some space to pick up some load |
3360 | * from other group and save more power | |
3361 | */ | |
3362 | if (sum_nr_running <= group_capacity - 1) { | |
3363 | if (sum_nr_running > leader_nr_running || | |
3364 | (sum_nr_running == leader_nr_running && | |
d5679bd1 | 3365 | cpumask_first(sched_group_cpus(group)) < |
758b2cdc | 3366 | cpumask_first(sched_group_cpus(group_leader)))) { |
dd41f596 IM |
3367 | group_leader = group; |
3368 | leader_nr_running = sum_nr_running; | |
3369 | } | |
48f24c4d | 3370 | } |
5c45bf27 SS |
3371 | group_next: |
3372 | #endif | |
1da177e4 LT |
3373 | group = group->next; |
3374 | } while (group != sd->groups); | |
3375 | ||
2dd73a4f | 3376 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3377 | goto out_balanced; |
3378 | ||
3379 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3380 | ||
3381 | if (this_load >= avg_load || | |
3382 | 100*max_load <= sd->imbalance_pct*this_load) | |
3383 | goto out_balanced; | |
3384 | ||
2dd73a4f | 3385 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3386 | if (group_imb) |
3387 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3388 | ||
1da177e4 LT |
3389 | /* |
3390 | * We're trying to get all the cpus to the average_load, so we don't | |
3391 | * want to push ourselves above the average load, nor do we wish to | |
3392 | * reduce the max loaded cpu below the average load, as either of these | |
3393 | * actions would just result in more rebalancing later, and ping-pong | |
3394 | * tasks around. Thus we look for the minimum possible imbalance. | |
3395 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3396 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3397 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3398 | * appear as very large values with unsigned longs. |
3399 | */ | |
2dd73a4f PW |
3400 | if (max_load <= busiest_load_per_task) |
3401 | goto out_balanced; | |
3402 | ||
3403 | /* | |
3404 | * In the presence of smp nice balancing, certain scenarios can have | |
3405 | * max load less than avg load(as we skip the groups at or below | |
3406 | * its cpu_power, while calculating max_load..) | |
3407 | */ | |
3408 | if (max_load < avg_load) { | |
3409 | *imbalance = 0; | |
3410 | goto small_imbalance; | |
3411 | } | |
0c117f1b SS |
3412 | |
3413 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3414 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3415 | |
1da177e4 | 3416 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3417 | *imbalance = min(max_pull * busiest->__cpu_power, |
3418 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3419 | / SCHED_LOAD_SCALE; |
3420 | ||
2dd73a4f PW |
3421 | /* |
3422 | * if *imbalance is less than the average load per runnable task | |
3423 | * there is no gaurantee that any tasks will be moved so we'll have | |
3424 | * a think about bumping its value to force at least one task to be | |
3425 | * moved | |
3426 | */ | |
7fd0d2dd | 3427 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3428 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3429 | unsigned int imbn; |
3430 | ||
3431 | small_imbalance: | |
3432 | pwr_move = pwr_now = 0; | |
3433 | imbn = 2; | |
3434 | if (this_nr_running) { | |
3435 | this_load_per_task /= this_nr_running; | |
3436 | if (busiest_load_per_task > this_load_per_task) | |
3437 | imbn = 1; | |
3438 | } else | |
408ed066 | 3439 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3440 | |
01c8c57d | 3441 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3442 | busiest_load_per_task * imbn) { |
2dd73a4f | 3443 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3444 | return busiest; |
3445 | } | |
3446 | ||
3447 | /* | |
3448 | * OK, we don't have enough imbalance to justify moving tasks, | |
3449 | * however we may be able to increase total CPU power used by | |
3450 | * moving them. | |
3451 | */ | |
3452 | ||
5517d86b ED |
3453 | pwr_now += busiest->__cpu_power * |
3454 | min(busiest_load_per_task, max_load); | |
3455 | pwr_now += this->__cpu_power * | |
3456 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3457 | pwr_now /= SCHED_LOAD_SCALE; |
3458 | ||
3459 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3460 | tmp = sg_div_cpu_power(busiest, |
3461 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3462 | if (max_load > tmp) |
5517d86b | 3463 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3464 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3465 | |
3466 | /* Amount of load we'd add */ | |
5517d86b | 3467 | if (max_load * busiest->__cpu_power < |
33859f7f | 3468 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3469 | tmp = sg_div_cpu_power(this, |
3470 | max_load * busiest->__cpu_power); | |
1da177e4 | 3471 | else |
5517d86b ED |
3472 | tmp = sg_div_cpu_power(this, |
3473 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3474 | pwr_move += this->__cpu_power * | |
3475 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3476 | pwr_move /= SCHED_LOAD_SCALE; |
3477 | ||
3478 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3479 | if (pwr_move > pwr_now) |
3480 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3481 | } |
3482 | ||
1da177e4 LT |
3483 | return busiest; |
3484 | ||
3485 | out_balanced: | |
5c45bf27 | 3486 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3487 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3488 | goto ret; |
1da177e4 | 3489 | |
5c45bf27 SS |
3490 | if (this == group_leader && group_leader != group_min) { |
3491 | *imbalance = min_load_per_task; | |
7a09b1a2 VS |
3492 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3493 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
9924da43 | 3494 | cpumask_first(sched_group_cpus(group_leader)); |
7a09b1a2 | 3495 | } |
5c45bf27 SS |
3496 | return group_min; |
3497 | } | |
5c45bf27 | 3498 | #endif |
783609c6 | 3499 | ret: |
1da177e4 LT |
3500 | *imbalance = 0; |
3501 | return NULL; | |
3502 | } | |
3503 | ||
3504 | /* | |
3505 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3506 | */ | |
70b97a7f | 3507 | static struct rq * |
d15bcfdb | 3508 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3509 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3510 | { |
70b97a7f | 3511 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3512 | unsigned long max_load = 0; |
1da177e4 LT |
3513 | int i; |
3514 | ||
758b2cdc | 3515 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3516 | unsigned long wl; |
0a2966b4 | 3517 | |
96f874e2 | 3518 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3519 | continue; |
3520 | ||
48f24c4d | 3521 | rq = cpu_rq(i); |
dd41f596 | 3522 | wl = weighted_cpuload(i); |
2dd73a4f | 3523 | |
dd41f596 | 3524 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3525 | continue; |
1da177e4 | 3526 | |
dd41f596 IM |
3527 | if (wl > max_load) { |
3528 | max_load = wl; | |
48f24c4d | 3529 | busiest = rq; |
1da177e4 LT |
3530 | } |
3531 | } | |
3532 | ||
3533 | return busiest; | |
3534 | } | |
3535 | ||
77391d71 NP |
3536 | /* |
3537 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3538 | * so long as it is large enough. | |
3539 | */ | |
3540 | #define MAX_PINNED_INTERVAL 512 | |
3541 | ||
1da177e4 LT |
3542 | /* |
3543 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3544 | * tasks if there is an imbalance. | |
1da177e4 | 3545 | */ |
70b97a7f | 3546 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3547 | struct sched_domain *sd, enum cpu_idle_type idle, |
96f874e2 | 3548 | int *balance, struct cpumask *cpus) |
1da177e4 | 3549 | { |
43010659 | 3550 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3551 | struct sched_group *group; |
1da177e4 | 3552 | unsigned long imbalance; |
70b97a7f | 3553 | struct rq *busiest; |
fe2eea3f | 3554 | unsigned long flags; |
5969fe06 | 3555 | |
96f874e2 | 3556 | cpumask_setall(cpus); |
7c16ec58 | 3557 | |
89c4710e SS |
3558 | /* |
3559 | * When power savings policy is enabled for the parent domain, idle | |
3560 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3561 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3562 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3563 | */ |
d15bcfdb | 3564 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3565 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3566 | sd_idle = 1; |
1da177e4 | 3567 | |
2d72376b | 3568 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3569 | |
0a2966b4 | 3570 | redo: |
c8cba857 | 3571 | update_shares(sd); |
0a2966b4 | 3572 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3573 | cpus, balance); |
783609c6 | 3574 | |
06066714 | 3575 | if (*balance == 0) |
783609c6 | 3576 | goto out_balanced; |
783609c6 | 3577 | |
1da177e4 LT |
3578 | if (!group) { |
3579 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3580 | goto out_balanced; | |
3581 | } | |
3582 | ||
7c16ec58 | 3583 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3584 | if (!busiest) { |
3585 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3586 | goto out_balanced; | |
3587 | } | |
3588 | ||
db935dbd | 3589 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3590 | |
3591 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3592 | ||
43010659 | 3593 | ld_moved = 0; |
1da177e4 LT |
3594 | if (busiest->nr_running > 1) { |
3595 | /* | |
3596 | * Attempt to move tasks. If find_busiest_group has found | |
3597 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3598 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3599 | * correctly treated as an imbalance. |
3600 | */ | |
fe2eea3f | 3601 | local_irq_save(flags); |
e17224bf | 3602 | double_rq_lock(this_rq, busiest); |
43010659 | 3603 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3604 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3605 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3606 | local_irq_restore(flags); |
81026794 | 3607 | |
46cb4b7c SS |
3608 | /* |
3609 | * some other cpu did the load balance for us. | |
3610 | */ | |
43010659 | 3611 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3612 | resched_cpu(this_cpu); |
3613 | ||
81026794 | 3614 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3615 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3616 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3617 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3618 | goto redo; |
81026794 | 3619 | goto out_balanced; |
0a2966b4 | 3620 | } |
1da177e4 | 3621 | } |
81026794 | 3622 | |
43010659 | 3623 | if (!ld_moved) { |
1da177e4 LT |
3624 | schedstat_inc(sd, lb_failed[idle]); |
3625 | sd->nr_balance_failed++; | |
3626 | ||
3627 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3628 | |
fe2eea3f | 3629 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3630 | |
3631 | /* don't kick the migration_thread, if the curr | |
3632 | * task on busiest cpu can't be moved to this_cpu | |
3633 | */ | |
96f874e2 RR |
3634 | if (!cpumask_test_cpu(this_cpu, |
3635 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3636 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3637 | all_pinned = 1; |
3638 | goto out_one_pinned; | |
3639 | } | |
3640 | ||
1da177e4 LT |
3641 | if (!busiest->active_balance) { |
3642 | busiest->active_balance = 1; | |
3643 | busiest->push_cpu = this_cpu; | |
81026794 | 3644 | active_balance = 1; |
1da177e4 | 3645 | } |
fe2eea3f | 3646 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3647 | if (active_balance) |
1da177e4 LT |
3648 | wake_up_process(busiest->migration_thread); |
3649 | ||
3650 | /* | |
3651 | * We've kicked active balancing, reset the failure | |
3652 | * counter. | |
3653 | */ | |
39507451 | 3654 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3655 | } |
81026794 | 3656 | } else |
1da177e4 LT |
3657 | sd->nr_balance_failed = 0; |
3658 | ||
81026794 | 3659 | if (likely(!active_balance)) { |
1da177e4 LT |
3660 | /* We were unbalanced, so reset the balancing interval */ |
3661 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3662 | } else { |
3663 | /* | |
3664 | * If we've begun active balancing, start to back off. This | |
3665 | * case may not be covered by the all_pinned logic if there | |
3666 | * is only 1 task on the busy runqueue (because we don't call | |
3667 | * move_tasks). | |
3668 | */ | |
3669 | if (sd->balance_interval < sd->max_interval) | |
3670 | sd->balance_interval *= 2; | |
1da177e4 LT |
3671 | } |
3672 | ||
43010659 | 3673 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3674 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3675 | ld_moved = -1; |
3676 | ||
3677 | goto out; | |
1da177e4 LT |
3678 | |
3679 | out_balanced: | |
1da177e4 LT |
3680 | schedstat_inc(sd, lb_balanced[idle]); |
3681 | ||
16cfb1c0 | 3682 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3683 | |
3684 | out_one_pinned: | |
1da177e4 | 3685 | /* tune up the balancing interval */ |
77391d71 NP |
3686 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3687 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3688 | sd->balance_interval *= 2; |
3689 | ||
48f24c4d | 3690 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3691 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3692 | ld_moved = -1; |
3693 | else | |
3694 | ld_moved = 0; | |
3695 | out: | |
c8cba857 PZ |
3696 | if (ld_moved) |
3697 | update_shares(sd); | |
c09595f6 | 3698 | return ld_moved; |
1da177e4 LT |
3699 | } |
3700 | ||
3701 | /* | |
3702 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3703 | * tasks if there is an imbalance. | |
3704 | * | |
d15bcfdb | 3705 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3706 | * this_rq is locked. |
3707 | */ | |
48f24c4d | 3708 | static int |
7c16ec58 | 3709 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
96f874e2 | 3710 | struct cpumask *cpus) |
1da177e4 LT |
3711 | { |
3712 | struct sched_group *group; | |
70b97a7f | 3713 | struct rq *busiest = NULL; |
1da177e4 | 3714 | unsigned long imbalance; |
43010659 | 3715 | int ld_moved = 0; |
5969fe06 | 3716 | int sd_idle = 0; |
969bb4e4 | 3717 | int all_pinned = 0; |
7c16ec58 | 3718 | |
96f874e2 | 3719 | cpumask_setall(cpus); |
5969fe06 | 3720 | |
89c4710e SS |
3721 | /* |
3722 | * When power savings policy is enabled for the parent domain, idle | |
3723 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3724 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3725 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3726 | */ |
3727 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3728 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3729 | sd_idle = 1; |
1da177e4 | 3730 | |
2d72376b | 3731 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3732 | redo: |
3e5459b4 | 3733 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3734 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3735 | &sd_idle, cpus, NULL); |
1da177e4 | 3736 | if (!group) { |
d15bcfdb | 3737 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3738 | goto out_balanced; |
1da177e4 LT |
3739 | } |
3740 | ||
7c16ec58 | 3741 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3742 | if (!busiest) { |
d15bcfdb | 3743 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3744 | goto out_balanced; |
1da177e4 LT |
3745 | } |
3746 | ||
db935dbd NP |
3747 | BUG_ON(busiest == this_rq); |
3748 | ||
d15bcfdb | 3749 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3750 | |
43010659 | 3751 | ld_moved = 0; |
d6d5cfaf NP |
3752 | if (busiest->nr_running > 1) { |
3753 | /* Attempt to move tasks */ | |
3754 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3755 | /* this_rq->clock is already updated */ |
3756 | update_rq_clock(busiest); | |
43010659 | 3757 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3758 | imbalance, sd, CPU_NEWLY_IDLE, |
3759 | &all_pinned); | |
1b12bbc7 | 3760 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3761 | |
969bb4e4 | 3762 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3763 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3764 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
3765 | goto redo; |
3766 | } | |
d6d5cfaf NP |
3767 | } |
3768 | ||
43010659 | 3769 | if (!ld_moved) { |
36dffab6 | 3770 | int active_balance = 0; |
ad273b32 | 3771 | |
d15bcfdb | 3772 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3773 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3774 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3775 | return -1; |
ad273b32 VS |
3776 | |
3777 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
3778 | return -1; | |
3779 | ||
3780 | if (sd->nr_balance_failed++ < 2) | |
3781 | return -1; | |
3782 | ||
3783 | /* | |
3784 | * The only task running in a non-idle cpu can be moved to this | |
3785 | * cpu in an attempt to completely freeup the other CPU | |
3786 | * package. The same method used to move task in load_balance() | |
3787 | * have been extended for load_balance_newidle() to speedup | |
3788 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
3789 | * | |
3790 | * The package power saving logic comes from | |
3791 | * find_busiest_group(). If there are no imbalance, then | |
3792 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
3793 | * f_b_g() will select a group from which a running task may be | |
3794 | * pulled to this cpu in order to make the other package idle. | |
3795 | * If there is no opportunity to make a package idle and if | |
3796 | * there are no imbalance, then f_b_g() will return NULL and no | |
3797 | * action will be taken in load_balance_newidle(). | |
3798 | * | |
3799 | * Under normal task pull operation due to imbalance, there | |
3800 | * will be more than one task in the source run queue and | |
3801 | * move_tasks() will succeed. ld_moved will be true and this | |
3802 | * active balance code will not be triggered. | |
3803 | */ | |
3804 | ||
3805 | /* Lock busiest in correct order while this_rq is held */ | |
3806 | double_lock_balance(this_rq, busiest); | |
3807 | ||
3808 | /* | |
3809 | * don't kick the migration_thread, if the curr | |
3810 | * task on busiest cpu can't be moved to this_cpu | |
3811 | */ | |
6ca09dfc | 3812 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
3813 | double_unlock_balance(this_rq, busiest); |
3814 | all_pinned = 1; | |
3815 | return ld_moved; | |
3816 | } | |
3817 | ||
3818 | if (!busiest->active_balance) { | |
3819 | busiest->active_balance = 1; | |
3820 | busiest->push_cpu = this_cpu; | |
3821 | active_balance = 1; | |
3822 | } | |
3823 | ||
3824 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
3825 | /* |
3826 | * Should not call ttwu while holding a rq->lock | |
3827 | */ | |
3828 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
3829 | if (active_balance) |
3830 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 3831 | spin_lock(&this_rq->lock); |
ad273b32 | 3832 | |
5969fe06 | 3833 | } else |
16cfb1c0 | 3834 | sd->nr_balance_failed = 0; |
1da177e4 | 3835 | |
3e5459b4 | 3836 | update_shares_locked(this_rq, sd); |
43010659 | 3837 | return ld_moved; |
16cfb1c0 NP |
3838 | |
3839 | out_balanced: | |
d15bcfdb | 3840 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3841 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3842 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3843 | return -1; |
16cfb1c0 | 3844 | sd->nr_balance_failed = 0; |
48f24c4d | 3845 | |
16cfb1c0 | 3846 | return 0; |
1da177e4 LT |
3847 | } |
3848 | ||
3849 | /* | |
3850 | * idle_balance is called by schedule() if this_cpu is about to become | |
3851 | * idle. Attempts to pull tasks from other CPUs. | |
3852 | */ | |
70b97a7f | 3853 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3854 | { |
3855 | struct sched_domain *sd; | |
efbe027e | 3856 | int pulled_task = 0; |
dd41f596 | 3857 | unsigned long next_balance = jiffies + HZ; |
4d2732c6 RR |
3858 | cpumask_var_t tmpmask; |
3859 | ||
3860 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | |
3861 | return; | |
1da177e4 LT |
3862 | |
3863 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3864 | unsigned long interval; |
3865 | ||
3866 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3867 | continue; | |
3868 | ||
3869 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3870 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 3871 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4d2732c6 | 3872 | sd, tmpmask); |
92c4ca5c CL |
3873 | |
3874 | interval = msecs_to_jiffies(sd->balance_interval); | |
3875 | if (time_after(next_balance, sd->last_balance + interval)) | |
3876 | next_balance = sd->last_balance + interval; | |
3877 | if (pulled_task) | |
3878 | break; | |
1da177e4 | 3879 | } |
dd41f596 | 3880 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3881 | /* |
3882 | * We are going idle. next_balance may be set based on | |
3883 | * a busy processor. So reset next_balance. | |
3884 | */ | |
3885 | this_rq->next_balance = next_balance; | |
dd41f596 | 3886 | } |
4d2732c6 | 3887 | free_cpumask_var(tmpmask); |
1da177e4 LT |
3888 | } |
3889 | ||
3890 | /* | |
3891 | * active_load_balance is run by migration threads. It pushes running tasks | |
3892 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3893 | * running on each physical CPU where possible, and avoids physical / | |
3894 | * logical imbalances. | |
3895 | * | |
3896 | * Called with busiest_rq locked. | |
3897 | */ | |
70b97a7f | 3898 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3899 | { |
39507451 | 3900 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3901 | struct sched_domain *sd; |
3902 | struct rq *target_rq; | |
39507451 | 3903 | |
48f24c4d | 3904 | /* Is there any task to move? */ |
39507451 | 3905 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3906 | return; |
3907 | ||
3908 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3909 | |
3910 | /* | |
39507451 | 3911 | * This condition is "impossible", if it occurs |
41a2d6cf | 3912 | * we need to fix it. Originally reported by |
39507451 | 3913 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3914 | */ |
39507451 | 3915 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3916 | |
39507451 NP |
3917 | /* move a task from busiest_rq to target_rq */ |
3918 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3919 | update_rq_clock(busiest_rq); |
3920 | update_rq_clock(target_rq); | |
39507451 NP |
3921 | |
3922 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3923 | for_each_domain(target_cpu, sd) { |
39507451 | 3924 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 3925 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 3926 | break; |
c96d145e | 3927 | } |
39507451 | 3928 | |
48f24c4d | 3929 | if (likely(sd)) { |
2d72376b | 3930 | schedstat_inc(sd, alb_count); |
39507451 | 3931 | |
43010659 PW |
3932 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3933 | sd, CPU_IDLE)) | |
48f24c4d IM |
3934 | schedstat_inc(sd, alb_pushed); |
3935 | else | |
3936 | schedstat_inc(sd, alb_failed); | |
3937 | } | |
1b12bbc7 | 3938 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3939 | } |
3940 | ||
46cb4b7c SS |
3941 | #ifdef CONFIG_NO_HZ |
3942 | static struct { | |
3943 | atomic_t load_balancer; | |
7d1e6a9b | 3944 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
3945 | } nohz ____cacheline_aligned = { |
3946 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
3947 | }; |
3948 | ||
7835b98b | 3949 | /* |
46cb4b7c SS |
3950 | * This routine will try to nominate the ilb (idle load balancing) |
3951 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3952 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3953 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3954 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3955 | * arrives... | |
3956 | * | |
3957 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3958 | * for idle load balancing. ilb owner will still be part of | |
3959 | * nohz.cpu_mask.. | |
7835b98b | 3960 | * |
46cb4b7c SS |
3961 | * While stopping the tick, this cpu will become the ilb owner if there |
3962 | * is no other owner. And will be the owner till that cpu becomes busy | |
3963 | * or if all cpus in the system stop their ticks at which point | |
3964 | * there is no need for ilb owner. | |
3965 | * | |
3966 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3967 | * next busy scheduler_tick() | |
3968 | */ | |
3969 | int select_nohz_load_balancer(int stop_tick) | |
3970 | { | |
3971 | int cpu = smp_processor_id(); | |
3972 | ||
3973 | if (stop_tick) { | |
46cb4b7c SS |
3974 | cpu_rq(cpu)->in_nohz_recently = 1; |
3975 | ||
483b4ee6 SS |
3976 | if (!cpu_active(cpu)) { |
3977 | if (atomic_read(&nohz.load_balancer) != cpu) | |
3978 | return 0; | |
3979 | ||
3980 | /* | |
3981 | * If we are going offline and still the leader, | |
3982 | * give up! | |
3983 | */ | |
46cb4b7c SS |
3984 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
3985 | BUG(); | |
483b4ee6 | 3986 | |
46cb4b7c SS |
3987 | return 0; |
3988 | } | |
3989 | ||
483b4ee6 SS |
3990 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
3991 | ||
46cb4b7c | 3992 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 3993 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
3994 | if (atomic_read(&nohz.load_balancer) == cpu) |
3995 | atomic_set(&nohz.load_balancer, -1); | |
3996 | return 0; | |
3997 | } | |
3998 | ||
3999 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4000 | /* make me the ilb owner */ | |
4001 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4002 | return 1; | |
4003 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
4004 | return 1; | |
4005 | } else { | |
7d1e6a9b | 4006 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4007 | return 0; |
4008 | ||
7d1e6a9b | 4009 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4010 | |
4011 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4012 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4013 | BUG(); | |
4014 | } | |
4015 | return 0; | |
4016 | } | |
4017 | #endif | |
4018 | ||
4019 | static DEFINE_SPINLOCK(balancing); | |
4020 | ||
4021 | /* | |
7835b98b CL |
4022 | * It checks each scheduling domain to see if it is due to be balanced, |
4023 | * and initiates a balancing operation if so. | |
4024 | * | |
4025 | * Balancing parameters are set up in arch_init_sched_domains. | |
4026 | */ | |
a9957449 | 4027 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4028 | { |
46cb4b7c SS |
4029 | int balance = 1; |
4030 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4031 | unsigned long interval; |
4032 | struct sched_domain *sd; | |
46cb4b7c | 4033 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4034 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4035 | int update_next_balance = 0; |
d07355f5 | 4036 | int need_serialize; |
a0e90245 RR |
4037 | cpumask_var_t tmp; |
4038 | ||
4039 | /* Fails alloc? Rebalancing probably not a priority right now. */ | |
4040 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | |
4041 | return; | |
1da177e4 | 4042 | |
46cb4b7c | 4043 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4044 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4045 | continue; | |
4046 | ||
4047 | interval = sd->balance_interval; | |
d15bcfdb | 4048 | if (idle != CPU_IDLE) |
1da177e4 LT |
4049 | interval *= sd->busy_factor; |
4050 | ||
4051 | /* scale ms to jiffies */ | |
4052 | interval = msecs_to_jiffies(interval); | |
4053 | if (unlikely(!interval)) | |
4054 | interval = 1; | |
dd41f596 IM |
4055 | if (interval > HZ*NR_CPUS/10) |
4056 | interval = HZ*NR_CPUS/10; | |
4057 | ||
d07355f5 | 4058 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4059 | |
d07355f5 | 4060 | if (need_serialize) { |
08c183f3 CL |
4061 | if (!spin_trylock(&balancing)) |
4062 | goto out; | |
4063 | } | |
4064 | ||
c9819f45 | 4065 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
a0e90245 | 4066 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { |
fa3b6ddc SS |
4067 | /* |
4068 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4069 | * longer idle, or one of our SMT siblings is |
4070 | * not idle. | |
4071 | */ | |
d15bcfdb | 4072 | idle = CPU_NOT_IDLE; |
1da177e4 | 4073 | } |
1bd77f2d | 4074 | sd->last_balance = jiffies; |
1da177e4 | 4075 | } |
d07355f5 | 4076 | if (need_serialize) |
08c183f3 CL |
4077 | spin_unlock(&balancing); |
4078 | out: | |
f549da84 | 4079 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4080 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4081 | update_next_balance = 1; |
4082 | } | |
783609c6 SS |
4083 | |
4084 | /* | |
4085 | * Stop the load balance at this level. There is another | |
4086 | * CPU in our sched group which is doing load balancing more | |
4087 | * actively. | |
4088 | */ | |
4089 | if (!balance) | |
4090 | break; | |
1da177e4 | 4091 | } |
f549da84 SS |
4092 | |
4093 | /* | |
4094 | * next_balance will be updated only when there is a need. | |
4095 | * When the cpu is attached to null domain for ex, it will not be | |
4096 | * updated. | |
4097 | */ | |
4098 | if (likely(update_next_balance)) | |
4099 | rq->next_balance = next_balance; | |
a0e90245 RR |
4100 | |
4101 | free_cpumask_var(tmp); | |
46cb4b7c SS |
4102 | } |
4103 | ||
4104 | /* | |
4105 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4106 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4107 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4108 | */ | |
4109 | static void run_rebalance_domains(struct softirq_action *h) | |
4110 | { | |
dd41f596 IM |
4111 | int this_cpu = smp_processor_id(); |
4112 | struct rq *this_rq = cpu_rq(this_cpu); | |
4113 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4114 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4115 | |
dd41f596 | 4116 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4117 | |
4118 | #ifdef CONFIG_NO_HZ | |
4119 | /* | |
4120 | * If this cpu is the owner for idle load balancing, then do the | |
4121 | * balancing on behalf of the other idle cpus whose ticks are | |
4122 | * stopped. | |
4123 | */ | |
dd41f596 IM |
4124 | if (this_rq->idle_at_tick && |
4125 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4126 | struct rq *rq; |
4127 | int balance_cpu; | |
4128 | ||
7d1e6a9b RR |
4129 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4130 | if (balance_cpu == this_cpu) | |
4131 | continue; | |
4132 | ||
46cb4b7c SS |
4133 | /* |
4134 | * If this cpu gets work to do, stop the load balancing | |
4135 | * work being done for other cpus. Next load | |
4136 | * balancing owner will pick it up. | |
4137 | */ | |
4138 | if (need_resched()) | |
4139 | break; | |
4140 | ||
de0cf899 | 4141 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4142 | |
4143 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4144 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4145 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4146 | } |
4147 | } | |
4148 | #endif | |
4149 | } | |
4150 | ||
4151 | /* | |
4152 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4153 | * | |
4154 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4155 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4156 | * if the whole system is idle. | |
4157 | */ | |
dd41f596 | 4158 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4159 | { |
46cb4b7c SS |
4160 | #ifdef CONFIG_NO_HZ |
4161 | /* | |
4162 | * If we were in the nohz mode recently and busy at the current | |
4163 | * scheduler tick, then check if we need to nominate new idle | |
4164 | * load balancer. | |
4165 | */ | |
4166 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4167 | rq->in_nohz_recently = 0; | |
4168 | ||
4169 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4170 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4171 | atomic_set(&nohz.load_balancer, -1); |
4172 | } | |
4173 | ||
4174 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4175 | /* | |
4176 | * simple selection for now: Nominate the | |
4177 | * first cpu in the nohz list to be the next | |
4178 | * ilb owner. | |
4179 | * | |
4180 | * TBD: Traverse the sched domains and nominate | |
4181 | * the nearest cpu in the nohz.cpu_mask. | |
4182 | */ | |
7d1e6a9b | 4183 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4184 | |
434d53b0 | 4185 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4186 | resched_cpu(ilb); |
4187 | } | |
4188 | } | |
4189 | ||
4190 | /* | |
4191 | * If this cpu is idle and doing idle load balancing for all the | |
4192 | * cpus with ticks stopped, is it time for that to stop? | |
4193 | */ | |
4194 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4195 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4196 | resched_cpu(cpu); |
4197 | return; | |
4198 | } | |
4199 | ||
4200 | /* | |
4201 | * If this cpu is idle and the idle load balancing is done by | |
4202 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4203 | */ | |
4204 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4205 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4206 | return; |
4207 | #endif | |
4208 | if (time_after_eq(jiffies, rq->next_balance)) | |
4209 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4210 | } |
dd41f596 IM |
4211 | |
4212 | #else /* CONFIG_SMP */ | |
4213 | ||
1da177e4 LT |
4214 | /* |
4215 | * on UP we do not need to balance between CPUs: | |
4216 | */ | |
70b97a7f | 4217 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4218 | { |
4219 | } | |
dd41f596 | 4220 | |
1da177e4 LT |
4221 | #endif |
4222 | ||
1da177e4 LT |
4223 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4224 | ||
4225 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4226 | ||
4227 | /* | |
f06febc9 FM |
4228 | * Return any ns on the sched_clock that have not yet been banked in |
4229 | * @p in case that task is currently running. | |
1da177e4 | 4230 | */ |
bb34d92f | 4231 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4232 | { |
1da177e4 | 4233 | unsigned long flags; |
41b86e9c | 4234 | struct rq *rq; |
bb34d92f | 4235 | u64 ns = 0; |
48f24c4d | 4236 | |
41b86e9c | 4237 | rq = task_rq_lock(p, &flags); |
1508487e | 4238 | |
051a1d1a | 4239 | if (task_current(rq, p)) { |
f06febc9 FM |
4240 | u64 delta_exec; |
4241 | ||
a8e504d2 IM |
4242 | update_rq_clock(rq); |
4243 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4244 | if ((s64)delta_exec > 0) |
bb34d92f | 4245 | ns = delta_exec; |
41b86e9c | 4246 | } |
48f24c4d | 4247 | |
41b86e9c | 4248 | task_rq_unlock(rq, &flags); |
48f24c4d | 4249 | |
1da177e4 LT |
4250 | return ns; |
4251 | } | |
4252 | ||
1da177e4 LT |
4253 | /* |
4254 | * Account user cpu time to a process. | |
4255 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4256 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4257 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4258 | */ |
457533a7 MS |
4259 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4260 | cputime_t cputime_scaled) | |
1da177e4 LT |
4261 | { |
4262 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4263 | cputime64_t tmp; | |
4264 | ||
457533a7 | 4265 | /* Add user time to process. */ |
1da177e4 | 4266 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4267 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4268 | account_group_user_time(p, cputime); |
1da177e4 LT |
4269 | |
4270 | /* Add user time to cpustat. */ | |
4271 | tmp = cputime_to_cputime64(cputime); | |
4272 | if (TASK_NICE(p) > 0) | |
4273 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4274 | else | |
4275 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4276 | /* Account for user time used */ |
4277 | acct_update_integrals(p); | |
1da177e4 LT |
4278 | } |
4279 | ||
94886b84 LV |
4280 | /* |
4281 | * Account guest cpu time to a process. | |
4282 | * @p: the process that the cpu time gets accounted to | |
4283 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4284 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4285 | */ |
457533a7 MS |
4286 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4287 | cputime_t cputime_scaled) | |
94886b84 LV |
4288 | { |
4289 | cputime64_t tmp; | |
4290 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4291 | ||
4292 | tmp = cputime_to_cputime64(cputime); | |
4293 | ||
457533a7 | 4294 | /* Add guest time to process. */ |
94886b84 | 4295 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4296 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4297 | account_group_user_time(p, cputime); |
94886b84 LV |
4298 | p->gtime = cputime_add(p->gtime, cputime); |
4299 | ||
457533a7 | 4300 | /* Add guest time to cpustat. */ |
94886b84 LV |
4301 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4302 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4303 | } | |
4304 | ||
1da177e4 LT |
4305 | /* |
4306 | * Account system cpu time to a process. | |
4307 | * @p: the process that the cpu time gets accounted to | |
4308 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4309 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4310 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4311 | */ |
4312 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4313 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4314 | { |
4315 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4316 | cputime64_t tmp; |
4317 | ||
983ed7a6 | 4318 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4319 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4320 | return; |
4321 | } | |
94886b84 | 4322 | |
457533a7 | 4323 | /* Add system time to process. */ |
1da177e4 | 4324 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4325 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4326 | account_group_system_time(p, cputime); |
1da177e4 LT |
4327 | |
4328 | /* Add system time to cpustat. */ | |
4329 | tmp = cputime_to_cputime64(cputime); | |
4330 | if (hardirq_count() - hardirq_offset) | |
4331 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4332 | else if (softirq_count()) | |
4333 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4334 | else |
79741dd3 MS |
4335 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4336 | ||
1da177e4 LT |
4337 | /* Account for system time used */ |
4338 | acct_update_integrals(p); | |
1da177e4 LT |
4339 | } |
4340 | ||
c66f08be | 4341 | /* |
1da177e4 | 4342 | * Account for involuntary wait time. |
1da177e4 | 4343 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4344 | */ |
79741dd3 | 4345 | void account_steal_time(cputime_t cputime) |
c66f08be | 4346 | { |
79741dd3 MS |
4347 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4348 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4349 | ||
4350 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4351 | } |
4352 | ||
1da177e4 | 4353 | /* |
79741dd3 MS |
4354 | * Account for idle time. |
4355 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4356 | */ |
79741dd3 | 4357 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4358 | { |
4359 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4360 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4361 | struct rq *rq = this_rq(); |
1da177e4 | 4362 | |
79741dd3 MS |
4363 | if (atomic_read(&rq->nr_iowait) > 0) |
4364 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4365 | else | |
4366 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4367 | } |
4368 | ||
79741dd3 MS |
4369 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4370 | ||
4371 | /* | |
4372 | * Account a single tick of cpu time. | |
4373 | * @p: the process that the cpu time gets accounted to | |
4374 | * @user_tick: indicates if the tick is a user or a system tick | |
4375 | */ | |
4376 | void account_process_tick(struct task_struct *p, int user_tick) | |
4377 | { | |
4378 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4379 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4380 | struct rq *rq = this_rq(); | |
4381 | ||
4382 | if (user_tick) | |
4383 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
4384 | else if (p != rq->idle) | |
4385 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, | |
4386 | one_jiffy_scaled); | |
4387 | else | |
4388 | account_idle_time(one_jiffy); | |
4389 | } | |
4390 | ||
4391 | /* | |
4392 | * Account multiple ticks of steal time. | |
4393 | * @p: the process from which the cpu time has been stolen | |
4394 | * @ticks: number of stolen ticks | |
4395 | */ | |
4396 | void account_steal_ticks(unsigned long ticks) | |
4397 | { | |
4398 | account_steal_time(jiffies_to_cputime(ticks)); | |
4399 | } | |
4400 | ||
4401 | /* | |
4402 | * Account multiple ticks of idle time. | |
4403 | * @ticks: number of stolen ticks | |
4404 | */ | |
4405 | void account_idle_ticks(unsigned long ticks) | |
4406 | { | |
4407 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4408 | } |
4409 | ||
79741dd3 MS |
4410 | #endif |
4411 | ||
49048622 BS |
4412 | /* |
4413 | * Use precise platform statistics if available: | |
4414 | */ | |
4415 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4416 | cputime_t task_utime(struct task_struct *p) | |
4417 | { | |
4418 | return p->utime; | |
4419 | } | |
4420 | ||
4421 | cputime_t task_stime(struct task_struct *p) | |
4422 | { | |
4423 | return p->stime; | |
4424 | } | |
4425 | #else | |
4426 | cputime_t task_utime(struct task_struct *p) | |
4427 | { | |
4428 | clock_t utime = cputime_to_clock_t(p->utime), | |
4429 | total = utime + cputime_to_clock_t(p->stime); | |
4430 | u64 temp; | |
4431 | ||
4432 | /* | |
4433 | * Use CFS's precise accounting: | |
4434 | */ | |
4435 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4436 | ||
4437 | if (total) { | |
4438 | temp *= utime; | |
4439 | do_div(temp, total); | |
4440 | } | |
4441 | utime = (clock_t)temp; | |
4442 | ||
4443 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4444 | return p->prev_utime; | |
4445 | } | |
4446 | ||
4447 | cputime_t task_stime(struct task_struct *p) | |
4448 | { | |
4449 | clock_t stime; | |
4450 | ||
4451 | /* | |
4452 | * Use CFS's precise accounting. (we subtract utime from | |
4453 | * the total, to make sure the total observed by userspace | |
4454 | * grows monotonically - apps rely on that): | |
4455 | */ | |
4456 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4457 | cputime_to_clock_t(task_utime(p)); | |
4458 | ||
4459 | if (stime >= 0) | |
4460 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4461 | ||
4462 | return p->prev_stime; | |
4463 | } | |
4464 | #endif | |
4465 | ||
4466 | inline cputime_t task_gtime(struct task_struct *p) | |
4467 | { | |
4468 | return p->gtime; | |
4469 | } | |
4470 | ||
7835b98b CL |
4471 | /* |
4472 | * This function gets called by the timer code, with HZ frequency. | |
4473 | * We call it with interrupts disabled. | |
4474 | * | |
4475 | * It also gets called by the fork code, when changing the parent's | |
4476 | * timeslices. | |
4477 | */ | |
4478 | void scheduler_tick(void) | |
4479 | { | |
7835b98b CL |
4480 | int cpu = smp_processor_id(); |
4481 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4482 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4483 | |
4484 | sched_clock_tick(); | |
dd41f596 IM |
4485 | |
4486 | spin_lock(&rq->lock); | |
3e51f33f | 4487 | update_rq_clock(rq); |
f1a438d8 | 4488 | update_cpu_load(rq); |
fa85ae24 | 4489 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4490 | spin_unlock(&rq->lock); |
7835b98b | 4491 | |
e418e1c2 | 4492 | #ifdef CONFIG_SMP |
dd41f596 IM |
4493 | rq->idle_at_tick = idle_cpu(cpu); |
4494 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4495 | #endif |
1da177e4 LT |
4496 | } |
4497 | ||
6cd8a4bb SR |
4498 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4499 | defined(CONFIG_PREEMPT_TRACER)) | |
4500 | ||
4501 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4502 | { | |
4503 | if (in_lock_functions(addr)) { | |
4504 | addr = CALLER_ADDR2; | |
4505 | if (in_lock_functions(addr)) | |
4506 | addr = CALLER_ADDR3; | |
4507 | } | |
4508 | return addr; | |
4509 | } | |
1da177e4 | 4510 | |
43627582 | 4511 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4512 | { |
6cd8a4bb | 4513 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4514 | /* |
4515 | * Underflow? | |
4516 | */ | |
9a11b49a IM |
4517 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4518 | return; | |
6cd8a4bb | 4519 | #endif |
1da177e4 | 4520 | preempt_count() += val; |
6cd8a4bb | 4521 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4522 | /* |
4523 | * Spinlock count overflowing soon? | |
4524 | */ | |
33859f7f MOS |
4525 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4526 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4527 | #endif |
4528 | if (preempt_count() == val) | |
4529 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4530 | } |
4531 | EXPORT_SYMBOL(add_preempt_count); | |
4532 | ||
43627582 | 4533 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4534 | { |
6cd8a4bb | 4535 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4536 | /* |
4537 | * Underflow? | |
4538 | */ | |
01e3eb82 | 4539 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4540 | return; |
1da177e4 LT |
4541 | /* |
4542 | * Is the spinlock portion underflowing? | |
4543 | */ | |
9a11b49a IM |
4544 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4545 | !(preempt_count() & PREEMPT_MASK))) | |
4546 | return; | |
6cd8a4bb | 4547 | #endif |
9a11b49a | 4548 | |
6cd8a4bb SR |
4549 | if (preempt_count() == val) |
4550 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4551 | preempt_count() -= val; |
4552 | } | |
4553 | EXPORT_SYMBOL(sub_preempt_count); | |
4554 | ||
4555 | #endif | |
4556 | ||
4557 | /* | |
dd41f596 | 4558 | * Print scheduling while atomic bug: |
1da177e4 | 4559 | */ |
dd41f596 | 4560 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4561 | { |
838225b4 SS |
4562 | struct pt_regs *regs = get_irq_regs(); |
4563 | ||
4564 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4565 | prev->comm, prev->pid, preempt_count()); | |
4566 | ||
dd41f596 | 4567 | debug_show_held_locks(prev); |
e21f5b15 | 4568 | print_modules(); |
dd41f596 IM |
4569 | if (irqs_disabled()) |
4570 | print_irqtrace_events(prev); | |
838225b4 SS |
4571 | |
4572 | if (regs) | |
4573 | show_regs(regs); | |
4574 | else | |
4575 | dump_stack(); | |
dd41f596 | 4576 | } |
1da177e4 | 4577 | |
dd41f596 IM |
4578 | /* |
4579 | * Various schedule()-time debugging checks and statistics: | |
4580 | */ | |
4581 | static inline void schedule_debug(struct task_struct *prev) | |
4582 | { | |
1da177e4 | 4583 | /* |
41a2d6cf | 4584 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4585 | * schedule() atomically, we ignore that path for now. |
4586 | * Otherwise, whine if we are scheduling when we should not be. | |
4587 | */ | |
3f33a7ce | 4588 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4589 | __schedule_bug(prev); |
4590 | ||
1da177e4 LT |
4591 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4592 | ||
2d72376b | 4593 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4594 | #ifdef CONFIG_SCHEDSTATS |
4595 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4596 | schedstat_inc(this_rq(), bkl_count); |
4597 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4598 | } |
4599 | #endif | |
dd41f596 IM |
4600 | } |
4601 | ||
4602 | /* | |
4603 | * Pick up the highest-prio task: | |
4604 | */ | |
4605 | static inline struct task_struct * | |
ff95f3df | 4606 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4607 | { |
5522d5d5 | 4608 | const struct sched_class *class; |
dd41f596 | 4609 | struct task_struct *p; |
1da177e4 LT |
4610 | |
4611 | /* | |
dd41f596 IM |
4612 | * Optimization: we know that if all tasks are in |
4613 | * the fair class we can call that function directly: | |
1da177e4 | 4614 | */ |
dd41f596 | 4615 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4616 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4617 | if (likely(p)) |
4618 | return p; | |
1da177e4 LT |
4619 | } |
4620 | ||
dd41f596 IM |
4621 | class = sched_class_highest; |
4622 | for ( ; ; ) { | |
fb8d4724 | 4623 | p = class->pick_next_task(rq); |
dd41f596 IM |
4624 | if (p) |
4625 | return p; | |
4626 | /* | |
4627 | * Will never be NULL as the idle class always | |
4628 | * returns a non-NULL p: | |
4629 | */ | |
4630 | class = class->next; | |
4631 | } | |
4632 | } | |
1da177e4 | 4633 | |
dd41f596 IM |
4634 | /* |
4635 | * schedule() is the main scheduler function. | |
4636 | */ | |
4637 | asmlinkage void __sched schedule(void) | |
4638 | { | |
4639 | struct task_struct *prev, *next; | |
67ca7bde | 4640 | unsigned long *switch_count; |
dd41f596 | 4641 | struct rq *rq; |
31656519 | 4642 | int cpu; |
dd41f596 IM |
4643 | |
4644 | need_resched: | |
4645 | preempt_disable(); | |
4646 | cpu = smp_processor_id(); | |
4647 | rq = cpu_rq(cpu); | |
4648 | rcu_qsctr_inc(cpu); | |
4649 | prev = rq->curr; | |
4650 | switch_count = &prev->nivcsw; | |
4651 | ||
4652 | release_kernel_lock(prev); | |
4653 | need_resched_nonpreemptible: | |
4654 | ||
4655 | schedule_debug(prev); | |
1da177e4 | 4656 | |
31656519 | 4657 | if (sched_feat(HRTICK)) |
f333fdc9 | 4658 | hrtick_clear(rq); |
8f4d37ec | 4659 | |
8cd162ce | 4660 | spin_lock_irq(&rq->lock); |
3e51f33f | 4661 | update_rq_clock(rq); |
1e819950 | 4662 | clear_tsk_need_resched(prev); |
1da177e4 | 4663 | |
1da177e4 | 4664 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4665 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4666 | prev->state = TASK_RUNNING; |
16882c1e | 4667 | else |
2e1cb74a | 4668 | deactivate_task(rq, prev, 1); |
dd41f596 | 4669 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4670 | } |
4671 | ||
9a897c5a SR |
4672 | #ifdef CONFIG_SMP |
4673 | if (prev->sched_class->pre_schedule) | |
4674 | prev->sched_class->pre_schedule(rq, prev); | |
4675 | #endif | |
f65eda4f | 4676 | |
dd41f596 | 4677 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4678 | idle_balance(cpu, rq); |
1da177e4 | 4679 | |
31ee529c | 4680 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4681 | next = pick_next_task(rq, prev); |
1da177e4 | 4682 | |
1da177e4 | 4683 | if (likely(prev != next)) { |
673a90a1 DS |
4684 | sched_info_switch(prev, next); |
4685 | ||
1da177e4 LT |
4686 | rq->nr_switches++; |
4687 | rq->curr = next; | |
4688 | ++*switch_count; | |
4689 | ||
dd41f596 | 4690 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4691 | /* |
4692 | * the context switch might have flipped the stack from under | |
4693 | * us, hence refresh the local variables. | |
4694 | */ | |
4695 | cpu = smp_processor_id(); | |
4696 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4697 | } else |
4698 | spin_unlock_irq(&rq->lock); | |
4699 | ||
8f4d37ec | 4700 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4701 | goto need_resched_nonpreemptible; |
8f4d37ec | 4702 | |
1da177e4 LT |
4703 | preempt_enable_no_resched(); |
4704 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4705 | goto need_resched; | |
4706 | } | |
1da177e4 LT |
4707 | EXPORT_SYMBOL(schedule); |
4708 | ||
4709 | #ifdef CONFIG_PREEMPT | |
4710 | /* | |
2ed6e34f | 4711 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4712 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4713 | * occur there and call schedule directly. |
4714 | */ | |
4715 | asmlinkage void __sched preempt_schedule(void) | |
4716 | { | |
4717 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4718 | |
1da177e4 LT |
4719 | /* |
4720 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4721 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4722 | */ |
beed33a8 | 4723 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4724 | return; |
4725 | ||
3a5c359a AK |
4726 | do { |
4727 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4728 | schedule(); |
3a5c359a | 4729 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4730 | |
3a5c359a AK |
4731 | /* |
4732 | * Check again in case we missed a preemption opportunity | |
4733 | * between schedule and now. | |
4734 | */ | |
4735 | barrier(); | |
4736 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4737 | } |
1da177e4 LT |
4738 | EXPORT_SYMBOL(preempt_schedule); |
4739 | ||
4740 | /* | |
2ed6e34f | 4741 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4742 | * off of irq context. |
4743 | * Note, that this is called and return with irqs disabled. This will | |
4744 | * protect us against recursive calling from irq. | |
4745 | */ | |
4746 | asmlinkage void __sched preempt_schedule_irq(void) | |
4747 | { | |
4748 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4749 | |
2ed6e34f | 4750 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4751 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4752 | ||
3a5c359a AK |
4753 | do { |
4754 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4755 | local_irq_enable(); |
4756 | schedule(); | |
4757 | local_irq_disable(); | |
3a5c359a | 4758 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4759 | |
3a5c359a AK |
4760 | /* |
4761 | * Check again in case we missed a preemption opportunity | |
4762 | * between schedule and now. | |
4763 | */ | |
4764 | barrier(); | |
4765 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4766 | } |
4767 | ||
4768 | #endif /* CONFIG_PREEMPT */ | |
4769 | ||
95cdf3b7 IM |
4770 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4771 | void *key) | |
1da177e4 | 4772 | { |
48f24c4d | 4773 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4774 | } |
1da177e4 LT |
4775 | EXPORT_SYMBOL(default_wake_function); |
4776 | ||
4777 | /* | |
41a2d6cf IM |
4778 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4779 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4780 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4781 | * | |
4782 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4783 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4784 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4785 | */ | |
777c6c5f JW |
4786 | void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
4787 | int nr_exclusive, int sync, void *key) | |
1da177e4 | 4788 | { |
2e45874c | 4789 | wait_queue_t *curr, *next; |
1da177e4 | 4790 | |
2e45874c | 4791 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4792 | unsigned flags = curr->flags; |
4793 | ||
1da177e4 | 4794 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4795 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4796 | break; |
4797 | } | |
4798 | } | |
4799 | ||
4800 | /** | |
4801 | * __wake_up - wake up threads blocked on a waitqueue. | |
4802 | * @q: the waitqueue | |
4803 | * @mode: which threads | |
4804 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4805 | * @key: is directly passed to the wakeup function |
1da177e4 | 4806 | */ |
7ad5b3a5 | 4807 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4808 | int nr_exclusive, void *key) |
1da177e4 LT |
4809 | { |
4810 | unsigned long flags; | |
4811 | ||
4812 | spin_lock_irqsave(&q->lock, flags); | |
4813 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4814 | spin_unlock_irqrestore(&q->lock, flags); | |
4815 | } | |
1da177e4 LT |
4816 | EXPORT_SYMBOL(__wake_up); |
4817 | ||
4818 | /* | |
4819 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4820 | */ | |
7ad5b3a5 | 4821 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4822 | { |
4823 | __wake_up_common(q, mode, 1, 0, NULL); | |
4824 | } | |
4825 | ||
4826 | /** | |
67be2dd1 | 4827 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4828 | * @q: the waitqueue |
4829 | * @mode: which threads | |
4830 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4831 | * | |
4832 | * The sync wakeup differs that the waker knows that it will schedule | |
4833 | * away soon, so while the target thread will be woken up, it will not | |
4834 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4835 | * with each other. This can prevent needless bouncing between CPUs. | |
4836 | * | |
4837 | * On UP it can prevent extra preemption. | |
4838 | */ | |
7ad5b3a5 | 4839 | void |
95cdf3b7 | 4840 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4841 | { |
4842 | unsigned long flags; | |
4843 | int sync = 1; | |
4844 | ||
4845 | if (unlikely(!q)) | |
4846 | return; | |
4847 | ||
4848 | if (unlikely(!nr_exclusive)) | |
4849 | sync = 0; | |
4850 | ||
4851 | spin_lock_irqsave(&q->lock, flags); | |
4852 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4853 | spin_unlock_irqrestore(&q->lock, flags); | |
4854 | } | |
4855 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4856 | ||
65eb3dc6 KD |
4857 | /** |
4858 | * complete: - signals a single thread waiting on this completion | |
4859 | * @x: holds the state of this particular completion | |
4860 | * | |
4861 | * This will wake up a single thread waiting on this completion. Threads will be | |
4862 | * awakened in the same order in which they were queued. | |
4863 | * | |
4864 | * See also complete_all(), wait_for_completion() and related routines. | |
4865 | */ | |
b15136e9 | 4866 | void complete(struct completion *x) |
1da177e4 LT |
4867 | { |
4868 | unsigned long flags; | |
4869 | ||
4870 | spin_lock_irqsave(&x->wait.lock, flags); | |
4871 | x->done++; | |
d9514f6c | 4872 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4873 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4874 | } | |
4875 | EXPORT_SYMBOL(complete); | |
4876 | ||
65eb3dc6 KD |
4877 | /** |
4878 | * complete_all: - signals all threads waiting on this completion | |
4879 | * @x: holds the state of this particular completion | |
4880 | * | |
4881 | * This will wake up all threads waiting on this particular completion event. | |
4882 | */ | |
b15136e9 | 4883 | void complete_all(struct completion *x) |
1da177e4 LT |
4884 | { |
4885 | unsigned long flags; | |
4886 | ||
4887 | spin_lock_irqsave(&x->wait.lock, flags); | |
4888 | x->done += UINT_MAX/2; | |
d9514f6c | 4889 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4890 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4891 | } | |
4892 | EXPORT_SYMBOL(complete_all); | |
4893 | ||
8cbbe86d AK |
4894 | static inline long __sched |
4895 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4896 | { |
1da177e4 LT |
4897 | if (!x->done) { |
4898 | DECLARE_WAITQUEUE(wait, current); | |
4899 | ||
4900 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4901 | __add_wait_queue_tail(&x->wait, &wait); | |
4902 | do { | |
94d3d824 | 4903 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4904 | timeout = -ERESTARTSYS; |
4905 | break; | |
8cbbe86d AK |
4906 | } |
4907 | __set_current_state(state); | |
1da177e4 LT |
4908 | spin_unlock_irq(&x->wait.lock); |
4909 | timeout = schedule_timeout(timeout); | |
4910 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4911 | } while (!x->done && timeout); |
1da177e4 | 4912 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4913 | if (!x->done) |
4914 | return timeout; | |
1da177e4 LT |
4915 | } |
4916 | x->done--; | |
ea71a546 | 4917 | return timeout ?: 1; |
1da177e4 | 4918 | } |
1da177e4 | 4919 | |
8cbbe86d AK |
4920 | static long __sched |
4921 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4922 | { |
1da177e4 LT |
4923 | might_sleep(); |
4924 | ||
4925 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4926 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4927 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4928 | return timeout; |
4929 | } | |
1da177e4 | 4930 | |
65eb3dc6 KD |
4931 | /** |
4932 | * wait_for_completion: - waits for completion of a task | |
4933 | * @x: holds the state of this particular completion | |
4934 | * | |
4935 | * This waits to be signaled for completion of a specific task. It is NOT | |
4936 | * interruptible and there is no timeout. | |
4937 | * | |
4938 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4939 | * and interrupt capability. Also see complete(). | |
4940 | */ | |
b15136e9 | 4941 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4942 | { |
4943 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4944 | } |
8cbbe86d | 4945 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4946 | |
65eb3dc6 KD |
4947 | /** |
4948 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4949 | * @x: holds the state of this particular completion | |
4950 | * @timeout: timeout value in jiffies | |
4951 | * | |
4952 | * This waits for either a completion of a specific task to be signaled or for a | |
4953 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4954 | * interruptible. | |
4955 | */ | |
b15136e9 | 4956 | unsigned long __sched |
8cbbe86d | 4957 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4958 | { |
8cbbe86d | 4959 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4960 | } |
8cbbe86d | 4961 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4962 | |
65eb3dc6 KD |
4963 | /** |
4964 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4965 | * @x: holds the state of this particular completion | |
4966 | * | |
4967 | * This waits for completion of a specific task to be signaled. It is | |
4968 | * interruptible. | |
4969 | */ | |
8cbbe86d | 4970 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4971 | { |
51e97990 AK |
4972 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4973 | if (t == -ERESTARTSYS) | |
4974 | return t; | |
4975 | return 0; | |
0fec171c | 4976 | } |
8cbbe86d | 4977 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4978 | |
65eb3dc6 KD |
4979 | /** |
4980 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4981 | * @x: holds the state of this particular completion | |
4982 | * @timeout: timeout value in jiffies | |
4983 | * | |
4984 | * This waits for either a completion of a specific task to be signaled or for a | |
4985 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4986 | */ | |
b15136e9 | 4987 | unsigned long __sched |
8cbbe86d AK |
4988 | wait_for_completion_interruptible_timeout(struct completion *x, |
4989 | unsigned long timeout) | |
0fec171c | 4990 | { |
8cbbe86d | 4991 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4992 | } |
8cbbe86d | 4993 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4994 | |
65eb3dc6 KD |
4995 | /** |
4996 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4997 | * @x: holds the state of this particular completion | |
4998 | * | |
4999 | * This waits to be signaled for completion of a specific task. It can be | |
5000 | * interrupted by a kill signal. | |
5001 | */ | |
009e577e MW |
5002 | int __sched wait_for_completion_killable(struct completion *x) |
5003 | { | |
5004 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5005 | if (t == -ERESTARTSYS) | |
5006 | return t; | |
5007 | return 0; | |
5008 | } | |
5009 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5010 | ||
be4de352 DC |
5011 | /** |
5012 | * try_wait_for_completion - try to decrement a completion without blocking | |
5013 | * @x: completion structure | |
5014 | * | |
5015 | * Returns: 0 if a decrement cannot be done without blocking | |
5016 | * 1 if a decrement succeeded. | |
5017 | * | |
5018 | * If a completion is being used as a counting completion, | |
5019 | * attempt to decrement the counter without blocking. This | |
5020 | * enables us to avoid waiting if the resource the completion | |
5021 | * is protecting is not available. | |
5022 | */ | |
5023 | bool try_wait_for_completion(struct completion *x) | |
5024 | { | |
5025 | int ret = 1; | |
5026 | ||
5027 | spin_lock_irq(&x->wait.lock); | |
5028 | if (!x->done) | |
5029 | ret = 0; | |
5030 | else | |
5031 | x->done--; | |
5032 | spin_unlock_irq(&x->wait.lock); | |
5033 | return ret; | |
5034 | } | |
5035 | EXPORT_SYMBOL(try_wait_for_completion); | |
5036 | ||
5037 | /** | |
5038 | * completion_done - Test to see if a completion has any waiters | |
5039 | * @x: completion structure | |
5040 | * | |
5041 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5042 | * 1 if there are no waiters. | |
5043 | * | |
5044 | */ | |
5045 | bool completion_done(struct completion *x) | |
5046 | { | |
5047 | int ret = 1; | |
5048 | ||
5049 | spin_lock_irq(&x->wait.lock); | |
5050 | if (!x->done) | |
5051 | ret = 0; | |
5052 | spin_unlock_irq(&x->wait.lock); | |
5053 | return ret; | |
5054 | } | |
5055 | EXPORT_SYMBOL(completion_done); | |
5056 | ||
8cbbe86d AK |
5057 | static long __sched |
5058 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5059 | { |
0fec171c IM |
5060 | unsigned long flags; |
5061 | wait_queue_t wait; | |
5062 | ||
5063 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5064 | |
8cbbe86d | 5065 | __set_current_state(state); |
1da177e4 | 5066 | |
8cbbe86d AK |
5067 | spin_lock_irqsave(&q->lock, flags); |
5068 | __add_wait_queue(q, &wait); | |
5069 | spin_unlock(&q->lock); | |
5070 | timeout = schedule_timeout(timeout); | |
5071 | spin_lock_irq(&q->lock); | |
5072 | __remove_wait_queue(q, &wait); | |
5073 | spin_unlock_irqrestore(&q->lock, flags); | |
5074 | ||
5075 | return timeout; | |
5076 | } | |
5077 | ||
5078 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5079 | { | |
5080 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5081 | } |
1da177e4 LT |
5082 | EXPORT_SYMBOL(interruptible_sleep_on); |
5083 | ||
0fec171c | 5084 | long __sched |
95cdf3b7 | 5085 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5086 | { |
8cbbe86d | 5087 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5088 | } |
1da177e4 LT |
5089 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5090 | ||
0fec171c | 5091 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5092 | { |
8cbbe86d | 5093 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5094 | } |
1da177e4 LT |
5095 | EXPORT_SYMBOL(sleep_on); |
5096 | ||
0fec171c | 5097 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5098 | { |
8cbbe86d | 5099 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5100 | } |
1da177e4 LT |
5101 | EXPORT_SYMBOL(sleep_on_timeout); |
5102 | ||
b29739f9 IM |
5103 | #ifdef CONFIG_RT_MUTEXES |
5104 | ||
5105 | /* | |
5106 | * rt_mutex_setprio - set the current priority of a task | |
5107 | * @p: task | |
5108 | * @prio: prio value (kernel-internal form) | |
5109 | * | |
5110 | * This function changes the 'effective' priority of a task. It does | |
5111 | * not touch ->normal_prio like __setscheduler(). | |
5112 | * | |
5113 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5114 | */ | |
36c8b586 | 5115 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5116 | { |
5117 | unsigned long flags; | |
83b699ed | 5118 | int oldprio, on_rq, running; |
70b97a7f | 5119 | struct rq *rq; |
cb469845 | 5120 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5121 | |
5122 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5123 | ||
5124 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5125 | update_rq_clock(rq); |
b29739f9 | 5126 | |
d5f9f942 | 5127 | oldprio = p->prio; |
dd41f596 | 5128 | on_rq = p->se.on_rq; |
051a1d1a | 5129 | running = task_current(rq, p); |
0e1f3483 | 5130 | if (on_rq) |
69be72c1 | 5131 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5132 | if (running) |
5133 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5134 | |
5135 | if (rt_prio(prio)) | |
5136 | p->sched_class = &rt_sched_class; | |
5137 | else | |
5138 | p->sched_class = &fair_sched_class; | |
5139 | ||
b29739f9 IM |
5140 | p->prio = prio; |
5141 | ||
0e1f3483 HS |
5142 | if (running) |
5143 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5144 | if (on_rq) { |
8159f87e | 5145 | enqueue_task(rq, p, 0); |
cb469845 SR |
5146 | |
5147 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5148 | } |
5149 | task_rq_unlock(rq, &flags); | |
5150 | } | |
5151 | ||
5152 | #endif | |
5153 | ||
36c8b586 | 5154 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5155 | { |
dd41f596 | 5156 | int old_prio, delta, on_rq; |
1da177e4 | 5157 | unsigned long flags; |
70b97a7f | 5158 | struct rq *rq; |
1da177e4 LT |
5159 | |
5160 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5161 | return; | |
5162 | /* | |
5163 | * We have to be careful, if called from sys_setpriority(), | |
5164 | * the task might be in the middle of scheduling on another CPU. | |
5165 | */ | |
5166 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5167 | update_rq_clock(rq); |
1da177e4 LT |
5168 | /* |
5169 | * The RT priorities are set via sched_setscheduler(), but we still | |
5170 | * allow the 'normal' nice value to be set - but as expected | |
5171 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5172 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5173 | */ |
e05606d3 | 5174 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5175 | p->static_prio = NICE_TO_PRIO(nice); |
5176 | goto out_unlock; | |
5177 | } | |
dd41f596 | 5178 | on_rq = p->se.on_rq; |
c09595f6 | 5179 | if (on_rq) |
69be72c1 | 5180 | dequeue_task(rq, p, 0); |
1da177e4 | 5181 | |
1da177e4 | 5182 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5183 | set_load_weight(p); |
b29739f9 IM |
5184 | old_prio = p->prio; |
5185 | p->prio = effective_prio(p); | |
5186 | delta = p->prio - old_prio; | |
1da177e4 | 5187 | |
dd41f596 | 5188 | if (on_rq) { |
8159f87e | 5189 | enqueue_task(rq, p, 0); |
1da177e4 | 5190 | /* |
d5f9f942 AM |
5191 | * If the task increased its priority or is running and |
5192 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5193 | */ |
d5f9f942 | 5194 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5195 | resched_task(rq->curr); |
5196 | } | |
5197 | out_unlock: | |
5198 | task_rq_unlock(rq, &flags); | |
5199 | } | |
1da177e4 LT |
5200 | EXPORT_SYMBOL(set_user_nice); |
5201 | ||
e43379f1 MM |
5202 | /* |
5203 | * can_nice - check if a task can reduce its nice value | |
5204 | * @p: task | |
5205 | * @nice: nice value | |
5206 | */ | |
36c8b586 | 5207 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5208 | { |
024f4747 MM |
5209 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5210 | int nice_rlim = 20 - nice; | |
48f24c4d | 5211 | |
e43379f1 MM |
5212 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5213 | capable(CAP_SYS_NICE)); | |
5214 | } | |
5215 | ||
1da177e4 LT |
5216 | #ifdef __ARCH_WANT_SYS_NICE |
5217 | ||
5218 | /* | |
5219 | * sys_nice - change the priority of the current process. | |
5220 | * @increment: priority increment | |
5221 | * | |
5222 | * sys_setpriority is a more generic, but much slower function that | |
5223 | * does similar things. | |
5224 | */ | |
5add95d4 | 5225 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5226 | { |
48f24c4d | 5227 | long nice, retval; |
1da177e4 LT |
5228 | |
5229 | /* | |
5230 | * Setpriority might change our priority at the same moment. | |
5231 | * We don't have to worry. Conceptually one call occurs first | |
5232 | * and we have a single winner. | |
5233 | */ | |
e43379f1 MM |
5234 | if (increment < -40) |
5235 | increment = -40; | |
1da177e4 LT |
5236 | if (increment > 40) |
5237 | increment = 40; | |
5238 | ||
2b8f836f | 5239 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5240 | if (nice < -20) |
5241 | nice = -20; | |
5242 | if (nice > 19) | |
5243 | nice = 19; | |
5244 | ||
e43379f1 MM |
5245 | if (increment < 0 && !can_nice(current, nice)) |
5246 | return -EPERM; | |
5247 | ||
1da177e4 LT |
5248 | retval = security_task_setnice(current, nice); |
5249 | if (retval) | |
5250 | return retval; | |
5251 | ||
5252 | set_user_nice(current, nice); | |
5253 | return 0; | |
5254 | } | |
5255 | ||
5256 | #endif | |
5257 | ||
5258 | /** | |
5259 | * task_prio - return the priority value of a given task. | |
5260 | * @p: the task in question. | |
5261 | * | |
5262 | * This is the priority value as seen by users in /proc. | |
5263 | * RT tasks are offset by -200. Normal tasks are centered | |
5264 | * around 0, value goes from -16 to +15. | |
5265 | */ | |
36c8b586 | 5266 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5267 | { |
5268 | return p->prio - MAX_RT_PRIO; | |
5269 | } | |
5270 | ||
5271 | /** | |
5272 | * task_nice - return the nice value of a given task. | |
5273 | * @p: the task in question. | |
5274 | */ | |
36c8b586 | 5275 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5276 | { |
5277 | return TASK_NICE(p); | |
5278 | } | |
150d8bed | 5279 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5280 | |
5281 | /** | |
5282 | * idle_cpu - is a given cpu idle currently? | |
5283 | * @cpu: the processor in question. | |
5284 | */ | |
5285 | int idle_cpu(int cpu) | |
5286 | { | |
5287 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5288 | } | |
5289 | ||
1da177e4 LT |
5290 | /** |
5291 | * idle_task - return the idle task for a given cpu. | |
5292 | * @cpu: the processor in question. | |
5293 | */ | |
36c8b586 | 5294 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5295 | { |
5296 | return cpu_rq(cpu)->idle; | |
5297 | } | |
5298 | ||
5299 | /** | |
5300 | * find_process_by_pid - find a process with a matching PID value. | |
5301 | * @pid: the pid in question. | |
5302 | */ | |
a9957449 | 5303 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5304 | { |
228ebcbe | 5305 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5306 | } |
5307 | ||
5308 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5309 | static void |
5310 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5311 | { |
dd41f596 | 5312 | BUG_ON(p->se.on_rq); |
48f24c4d | 5313 | |
1da177e4 | 5314 | p->policy = policy; |
dd41f596 IM |
5315 | switch (p->policy) { |
5316 | case SCHED_NORMAL: | |
5317 | case SCHED_BATCH: | |
5318 | case SCHED_IDLE: | |
5319 | p->sched_class = &fair_sched_class; | |
5320 | break; | |
5321 | case SCHED_FIFO: | |
5322 | case SCHED_RR: | |
5323 | p->sched_class = &rt_sched_class; | |
5324 | break; | |
5325 | } | |
5326 | ||
1da177e4 | 5327 | p->rt_priority = prio; |
b29739f9 IM |
5328 | p->normal_prio = normal_prio(p); |
5329 | /* we are holding p->pi_lock already */ | |
5330 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5331 | set_load_weight(p); |
1da177e4 LT |
5332 | } |
5333 | ||
c69e8d9c DH |
5334 | /* |
5335 | * check the target process has a UID that matches the current process's | |
5336 | */ | |
5337 | static bool check_same_owner(struct task_struct *p) | |
5338 | { | |
5339 | const struct cred *cred = current_cred(), *pcred; | |
5340 | bool match; | |
5341 | ||
5342 | rcu_read_lock(); | |
5343 | pcred = __task_cred(p); | |
5344 | match = (cred->euid == pcred->euid || | |
5345 | cred->euid == pcred->uid); | |
5346 | rcu_read_unlock(); | |
5347 | return match; | |
5348 | } | |
5349 | ||
961ccddd RR |
5350 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5351 | struct sched_param *param, bool user) | |
1da177e4 | 5352 | { |
83b699ed | 5353 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5354 | unsigned long flags; |
cb469845 | 5355 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5356 | struct rq *rq; |
1da177e4 | 5357 | |
66e5393a SR |
5358 | /* may grab non-irq protected spin_locks */ |
5359 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5360 | recheck: |
5361 | /* double check policy once rq lock held */ | |
5362 | if (policy < 0) | |
5363 | policy = oldpolicy = p->policy; | |
5364 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5365 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5366 | policy != SCHED_IDLE) | |
b0a9499c | 5367 | return -EINVAL; |
1da177e4 LT |
5368 | /* |
5369 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5370 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5371 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5372 | */ |
5373 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5374 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5375 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5376 | return -EINVAL; |
e05606d3 | 5377 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5378 | return -EINVAL; |
5379 | ||
37e4ab3f OC |
5380 | /* |
5381 | * Allow unprivileged RT tasks to decrease priority: | |
5382 | */ | |
961ccddd | 5383 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5384 | if (rt_policy(policy)) { |
8dc3e909 | 5385 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5386 | |
5387 | if (!lock_task_sighand(p, &flags)) | |
5388 | return -ESRCH; | |
5389 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5390 | unlock_task_sighand(p, &flags); | |
5391 | ||
5392 | /* can't set/change the rt policy */ | |
5393 | if (policy != p->policy && !rlim_rtprio) | |
5394 | return -EPERM; | |
5395 | ||
5396 | /* can't increase priority */ | |
5397 | if (param->sched_priority > p->rt_priority && | |
5398 | param->sched_priority > rlim_rtprio) | |
5399 | return -EPERM; | |
5400 | } | |
dd41f596 IM |
5401 | /* |
5402 | * Like positive nice levels, dont allow tasks to | |
5403 | * move out of SCHED_IDLE either: | |
5404 | */ | |
5405 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5406 | return -EPERM; | |
5fe1d75f | 5407 | |
37e4ab3f | 5408 | /* can't change other user's priorities */ |
c69e8d9c | 5409 | if (!check_same_owner(p)) |
37e4ab3f OC |
5410 | return -EPERM; |
5411 | } | |
1da177e4 | 5412 | |
725aad24 | 5413 | if (user) { |
b68aa230 | 5414 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5415 | /* |
5416 | * Do not allow realtime tasks into groups that have no runtime | |
5417 | * assigned. | |
5418 | */ | |
9a7e0b18 PZ |
5419 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5420 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5421 | return -EPERM; |
b68aa230 PZ |
5422 | #endif |
5423 | ||
725aad24 JF |
5424 | retval = security_task_setscheduler(p, policy, param); |
5425 | if (retval) | |
5426 | return retval; | |
5427 | } | |
5428 | ||
b29739f9 IM |
5429 | /* |
5430 | * make sure no PI-waiters arrive (or leave) while we are | |
5431 | * changing the priority of the task: | |
5432 | */ | |
5433 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5434 | /* |
5435 | * To be able to change p->policy safely, the apropriate | |
5436 | * runqueue lock must be held. | |
5437 | */ | |
b29739f9 | 5438 | rq = __task_rq_lock(p); |
1da177e4 LT |
5439 | /* recheck policy now with rq lock held */ |
5440 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5441 | policy = oldpolicy = -1; | |
b29739f9 IM |
5442 | __task_rq_unlock(rq); |
5443 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5444 | goto recheck; |
5445 | } | |
2daa3577 | 5446 | update_rq_clock(rq); |
dd41f596 | 5447 | on_rq = p->se.on_rq; |
051a1d1a | 5448 | running = task_current(rq, p); |
0e1f3483 | 5449 | if (on_rq) |
2e1cb74a | 5450 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5451 | if (running) |
5452 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5453 | |
1da177e4 | 5454 | oldprio = p->prio; |
dd41f596 | 5455 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5456 | |
0e1f3483 HS |
5457 | if (running) |
5458 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5459 | if (on_rq) { |
5460 | activate_task(rq, p, 0); | |
cb469845 SR |
5461 | |
5462 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5463 | } |
b29739f9 IM |
5464 | __task_rq_unlock(rq); |
5465 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5466 | ||
95e02ca9 TG |
5467 | rt_mutex_adjust_pi(p); |
5468 | ||
1da177e4 LT |
5469 | return 0; |
5470 | } | |
961ccddd RR |
5471 | |
5472 | /** | |
5473 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5474 | * @p: the task in question. | |
5475 | * @policy: new policy. | |
5476 | * @param: structure containing the new RT priority. | |
5477 | * | |
5478 | * NOTE that the task may be already dead. | |
5479 | */ | |
5480 | int sched_setscheduler(struct task_struct *p, int policy, | |
5481 | struct sched_param *param) | |
5482 | { | |
5483 | return __sched_setscheduler(p, policy, param, true); | |
5484 | } | |
1da177e4 LT |
5485 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5486 | ||
961ccddd RR |
5487 | /** |
5488 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5489 | * @p: the task in question. | |
5490 | * @policy: new policy. | |
5491 | * @param: structure containing the new RT priority. | |
5492 | * | |
5493 | * Just like sched_setscheduler, only don't bother checking if the | |
5494 | * current context has permission. For example, this is needed in | |
5495 | * stop_machine(): we create temporary high priority worker threads, | |
5496 | * but our caller might not have that capability. | |
5497 | */ | |
5498 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5499 | struct sched_param *param) | |
5500 | { | |
5501 | return __sched_setscheduler(p, policy, param, false); | |
5502 | } | |
5503 | ||
95cdf3b7 IM |
5504 | static int |
5505 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5506 | { |
1da177e4 LT |
5507 | struct sched_param lparam; |
5508 | struct task_struct *p; | |
36c8b586 | 5509 | int retval; |
1da177e4 LT |
5510 | |
5511 | if (!param || pid < 0) | |
5512 | return -EINVAL; | |
5513 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5514 | return -EFAULT; | |
5fe1d75f ON |
5515 | |
5516 | rcu_read_lock(); | |
5517 | retval = -ESRCH; | |
1da177e4 | 5518 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5519 | if (p != NULL) |
5520 | retval = sched_setscheduler(p, policy, &lparam); | |
5521 | rcu_read_unlock(); | |
36c8b586 | 5522 | |
1da177e4 LT |
5523 | return retval; |
5524 | } | |
5525 | ||
5526 | /** | |
5527 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5528 | * @pid: the pid in question. | |
5529 | * @policy: new policy. | |
5530 | * @param: structure containing the new RT priority. | |
5531 | */ | |
5add95d4 HC |
5532 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5533 | struct sched_param __user *, param) | |
1da177e4 | 5534 | { |
c21761f1 JB |
5535 | /* negative values for policy are not valid */ |
5536 | if (policy < 0) | |
5537 | return -EINVAL; | |
5538 | ||
1da177e4 LT |
5539 | return do_sched_setscheduler(pid, policy, param); |
5540 | } | |
5541 | ||
5542 | /** | |
5543 | * sys_sched_setparam - set/change the RT priority of a thread | |
5544 | * @pid: the pid in question. | |
5545 | * @param: structure containing the new RT priority. | |
5546 | */ | |
5add95d4 | 5547 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5548 | { |
5549 | return do_sched_setscheduler(pid, -1, param); | |
5550 | } | |
5551 | ||
5552 | /** | |
5553 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5554 | * @pid: the pid in question. | |
5555 | */ | |
5add95d4 | 5556 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5557 | { |
36c8b586 | 5558 | struct task_struct *p; |
3a5c359a | 5559 | int retval; |
1da177e4 LT |
5560 | |
5561 | if (pid < 0) | |
3a5c359a | 5562 | return -EINVAL; |
1da177e4 LT |
5563 | |
5564 | retval = -ESRCH; | |
5565 | read_lock(&tasklist_lock); | |
5566 | p = find_process_by_pid(pid); | |
5567 | if (p) { | |
5568 | retval = security_task_getscheduler(p); | |
5569 | if (!retval) | |
5570 | retval = p->policy; | |
5571 | } | |
5572 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5573 | return retval; |
5574 | } | |
5575 | ||
5576 | /** | |
5577 | * sys_sched_getscheduler - get the RT priority of a thread | |
5578 | * @pid: the pid in question. | |
5579 | * @param: structure containing the RT priority. | |
5580 | */ | |
5add95d4 | 5581 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5582 | { |
5583 | struct sched_param lp; | |
36c8b586 | 5584 | struct task_struct *p; |
3a5c359a | 5585 | int retval; |
1da177e4 LT |
5586 | |
5587 | if (!param || pid < 0) | |
3a5c359a | 5588 | return -EINVAL; |
1da177e4 LT |
5589 | |
5590 | read_lock(&tasklist_lock); | |
5591 | p = find_process_by_pid(pid); | |
5592 | retval = -ESRCH; | |
5593 | if (!p) | |
5594 | goto out_unlock; | |
5595 | ||
5596 | retval = security_task_getscheduler(p); | |
5597 | if (retval) | |
5598 | goto out_unlock; | |
5599 | ||
5600 | lp.sched_priority = p->rt_priority; | |
5601 | read_unlock(&tasklist_lock); | |
5602 | ||
5603 | /* | |
5604 | * This one might sleep, we cannot do it with a spinlock held ... | |
5605 | */ | |
5606 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5607 | ||
1da177e4 LT |
5608 | return retval; |
5609 | ||
5610 | out_unlock: | |
5611 | read_unlock(&tasklist_lock); | |
5612 | return retval; | |
5613 | } | |
5614 | ||
96f874e2 | 5615 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5616 | { |
5a16f3d3 | 5617 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5618 | struct task_struct *p; |
5619 | int retval; | |
1da177e4 | 5620 | |
95402b38 | 5621 | get_online_cpus(); |
1da177e4 LT |
5622 | read_lock(&tasklist_lock); |
5623 | ||
5624 | p = find_process_by_pid(pid); | |
5625 | if (!p) { | |
5626 | read_unlock(&tasklist_lock); | |
95402b38 | 5627 | put_online_cpus(); |
1da177e4 LT |
5628 | return -ESRCH; |
5629 | } | |
5630 | ||
5631 | /* | |
5632 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5633 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5634 | * usage count and then drop tasklist_lock. |
5635 | */ | |
5636 | get_task_struct(p); | |
5637 | read_unlock(&tasklist_lock); | |
5638 | ||
5a16f3d3 RR |
5639 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5640 | retval = -ENOMEM; | |
5641 | goto out_put_task; | |
5642 | } | |
5643 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5644 | retval = -ENOMEM; | |
5645 | goto out_free_cpus_allowed; | |
5646 | } | |
1da177e4 | 5647 | retval = -EPERM; |
c69e8d9c | 5648 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5649 | goto out_unlock; |
5650 | ||
e7834f8f DQ |
5651 | retval = security_task_setscheduler(p, 0, NULL); |
5652 | if (retval) | |
5653 | goto out_unlock; | |
5654 | ||
5a16f3d3 RR |
5655 | cpuset_cpus_allowed(p, cpus_allowed); |
5656 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 5657 | again: |
5a16f3d3 | 5658 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5659 | |
8707d8b8 | 5660 | if (!retval) { |
5a16f3d3 RR |
5661 | cpuset_cpus_allowed(p, cpus_allowed); |
5662 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5663 | /* |
5664 | * We must have raced with a concurrent cpuset | |
5665 | * update. Just reset the cpus_allowed to the | |
5666 | * cpuset's cpus_allowed | |
5667 | */ | |
5a16f3d3 | 5668 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5669 | goto again; |
5670 | } | |
5671 | } | |
1da177e4 | 5672 | out_unlock: |
5a16f3d3 RR |
5673 | free_cpumask_var(new_mask); |
5674 | out_free_cpus_allowed: | |
5675 | free_cpumask_var(cpus_allowed); | |
5676 | out_put_task: | |
1da177e4 | 5677 | put_task_struct(p); |
95402b38 | 5678 | put_online_cpus(); |
1da177e4 LT |
5679 | return retval; |
5680 | } | |
5681 | ||
5682 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5683 | struct cpumask *new_mask) |
1da177e4 | 5684 | { |
96f874e2 RR |
5685 | if (len < cpumask_size()) |
5686 | cpumask_clear(new_mask); | |
5687 | else if (len > cpumask_size()) | |
5688 | len = cpumask_size(); | |
5689 | ||
1da177e4 LT |
5690 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5691 | } | |
5692 | ||
5693 | /** | |
5694 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5695 | * @pid: pid of the process | |
5696 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5697 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5698 | */ | |
5add95d4 HC |
5699 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5700 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5701 | { |
5a16f3d3 | 5702 | cpumask_var_t new_mask; |
1da177e4 LT |
5703 | int retval; |
5704 | ||
5a16f3d3 RR |
5705 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5706 | return -ENOMEM; | |
1da177e4 | 5707 | |
5a16f3d3 RR |
5708 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5709 | if (retval == 0) | |
5710 | retval = sched_setaffinity(pid, new_mask); | |
5711 | free_cpumask_var(new_mask); | |
5712 | return retval; | |
1da177e4 LT |
5713 | } |
5714 | ||
96f874e2 | 5715 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5716 | { |
36c8b586 | 5717 | struct task_struct *p; |
1da177e4 | 5718 | int retval; |
1da177e4 | 5719 | |
95402b38 | 5720 | get_online_cpus(); |
1da177e4 LT |
5721 | read_lock(&tasklist_lock); |
5722 | ||
5723 | retval = -ESRCH; | |
5724 | p = find_process_by_pid(pid); | |
5725 | if (!p) | |
5726 | goto out_unlock; | |
5727 | ||
e7834f8f DQ |
5728 | retval = security_task_getscheduler(p); |
5729 | if (retval) | |
5730 | goto out_unlock; | |
5731 | ||
96f874e2 | 5732 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
5733 | |
5734 | out_unlock: | |
5735 | read_unlock(&tasklist_lock); | |
95402b38 | 5736 | put_online_cpus(); |
1da177e4 | 5737 | |
9531b62f | 5738 | return retval; |
1da177e4 LT |
5739 | } |
5740 | ||
5741 | /** | |
5742 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5743 | * @pid: pid of the process | |
5744 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5745 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5746 | */ | |
5add95d4 HC |
5747 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5748 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5749 | { |
5750 | int ret; | |
f17c8607 | 5751 | cpumask_var_t mask; |
1da177e4 | 5752 | |
f17c8607 | 5753 | if (len < cpumask_size()) |
1da177e4 LT |
5754 | return -EINVAL; |
5755 | ||
f17c8607 RR |
5756 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5757 | return -ENOMEM; | |
1da177e4 | 5758 | |
f17c8607 RR |
5759 | ret = sched_getaffinity(pid, mask); |
5760 | if (ret == 0) { | |
5761 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
5762 | ret = -EFAULT; | |
5763 | else | |
5764 | ret = cpumask_size(); | |
5765 | } | |
5766 | free_cpumask_var(mask); | |
1da177e4 | 5767 | |
f17c8607 | 5768 | return ret; |
1da177e4 LT |
5769 | } |
5770 | ||
5771 | /** | |
5772 | * sys_sched_yield - yield the current processor to other threads. | |
5773 | * | |
dd41f596 IM |
5774 | * This function yields the current CPU to other tasks. If there are no |
5775 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5776 | */ |
5add95d4 | 5777 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5778 | { |
70b97a7f | 5779 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5780 | |
2d72376b | 5781 | schedstat_inc(rq, yld_count); |
4530d7ab | 5782 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5783 | |
5784 | /* | |
5785 | * Since we are going to call schedule() anyway, there's | |
5786 | * no need to preempt or enable interrupts: | |
5787 | */ | |
5788 | __release(rq->lock); | |
8a25d5de | 5789 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5790 | _raw_spin_unlock(&rq->lock); |
5791 | preempt_enable_no_resched(); | |
5792 | ||
5793 | schedule(); | |
5794 | ||
5795 | return 0; | |
5796 | } | |
5797 | ||
e7b38404 | 5798 | static void __cond_resched(void) |
1da177e4 | 5799 | { |
8e0a43d8 IM |
5800 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5801 | __might_sleep(__FILE__, __LINE__); | |
5802 | #endif | |
5bbcfd90 IM |
5803 | /* |
5804 | * The BKS might be reacquired before we have dropped | |
5805 | * PREEMPT_ACTIVE, which could trigger a second | |
5806 | * cond_resched() call. | |
5807 | */ | |
1da177e4 LT |
5808 | do { |
5809 | add_preempt_count(PREEMPT_ACTIVE); | |
5810 | schedule(); | |
5811 | sub_preempt_count(PREEMPT_ACTIVE); | |
5812 | } while (need_resched()); | |
5813 | } | |
5814 | ||
02b67cc3 | 5815 | int __sched _cond_resched(void) |
1da177e4 | 5816 | { |
9414232f IM |
5817 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5818 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5819 | __cond_resched(); |
5820 | return 1; | |
5821 | } | |
5822 | return 0; | |
5823 | } | |
02b67cc3 | 5824 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5825 | |
5826 | /* | |
5827 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5828 | * call schedule, and on return reacquire the lock. | |
5829 | * | |
41a2d6cf | 5830 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5831 | * operations here to prevent schedule() from being called twice (once via |
5832 | * spin_unlock(), once by hand). | |
5833 | */ | |
95cdf3b7 | 5834 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5835 | { |
95c354fe | 5836 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5837 | int ret = 0; |
5838 | ||
95c354fe | 5839 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5840 | spin_unlock(lock); |
95c354fe NP |
5841 | if (resched && need_resched()) |
5842 | __cond_resched(); | |
5843 | else | |
5844 | cpu_relax(); | |
6df3cecb | 5845 | ret = 1; |
1da177e4 | 5846 | spin_lock(lock); |
1da177e4 | 5847 | } |
6df3cecb | 5848 | return ret; |
1da177e4 | 5849 | } |
1da177e4 LT |
5850 | EXPORT_SYMBOL(cond_resched_lock); |
5851 | ||
5852 | int __sched cond_resched_softirq(void) | |
5853 | { | |
5854 | BUG_ON(!in_softirq()); | |
5855 | ||
9414232f | 5856 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5857 | local_bh_enable(); |
1da177e4 LT |
5858 | __cond_resched(); |
5859 | local_bh_disable(); | |
5860 | return 1; | |
5861 | } | |
5862 | return 0; | |
5863 | } | |
1da177e4 LT |
5864 | EXPORT_SYMBOL(cond_resched_softirq); |
5865 | ||
1da177e4 LT |
5866 | /** |
5867 | * yield - yield the current processor to other threads. | |
5868 | * | |
72fd4a35 | 5869 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5870 | * thread runnable and calls sys_sched_yield(). |
5871 | */ | |
5872 | void __sched yield(void) | |
5873 | { | |
5874 | set_current_state(TASK_RUNNING); | |
5875 | sys_sched_yield(); | |
5876 | } | |
1da177e4 LT |
5877 | EXPORT_SYMBOL(yield); |
5878 | ||
5879 | /* | |
41a2d6cf | 5880 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5881 | * that process accounting knows that this is a task in IO wait state. |
5882 | * | |
5883 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5884 | * has set its backing_dev_info: the queue against which it should throttle) | |
5885 | */ | |
5886 | void __sched io_schedule(void) | |
5887 | { | |
70b97a7f | 5888 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5889 | |
0ff92245 | 5890 | delayacct_blkio_start(); |
1da177e4 LT |
5891 | atomic_inc(&rq->nr_iowait); |
5892 | schedule(); | |
5893 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5894 | delayacct_blkio_end(); |
1da177e4 | 5895 | } |
1da177e4 LT |
5896 | EXPORT_SYMBOL(io_schedule); |
5897 | ||
5898 | long __sched io_schedule_timeout(long timeout) | |
5899 | { | |
70b97a7f | 5900 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5901 | long ret; |
5902 | ||
0ff92245 | 5903 | delayacct_blkio_start(); |
1da177e4 LT |
5904 | atomic_inc(&rq->nr_iowait); |
5905 | ret = schedule_timeout(timeout); | |
5906 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5907 | delayacct_blkio_end(); |
1da177e4 LT |
5908 | return ret; |
5909 | } | |
5910 | ||
5911 | /** | |
5912 | * sys_sched_get_priority_max - return maximum RT priority. | |
5913 | * @policy: scheduling class. | |
5914 | * | |
5915 | * this syscall returns the maximum rt_priority that can be used | |
5916 | * by a given scheduling class. | |
5917 | */ | |
5add95d4 | 5918 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5919 | { |
5920 | int ret = -EINVAL; | |
5921 | ||
5922 | switch (policy) { | |
5923 | case SCHED_FIFO: | |
5924 | case SCHED_RR: | |
5925 | ret = MAX_USER_RT_PRIO-1; | |
5926 | break; | |
5927 | case SCHED_NORMAL: | |
b0a9499c | 5928 | case SCHED_BATCH: |
dd41f596 | 5929 | case SCHED_IDLE: |
1da177e4 LT |
5930 | ret = 0; |
5931 | break; | |
5932 | } | |
5933 | return ret; | |
5934 | } | |
5935 | ||
5936 | /** | |
5937 | * sys_sched_get_priority_min - return minimum RT priority. | |
5938 | * @policy: scheduling class. | |
5939 | * | |
5940 | * this syscall returns the minimum rt_priority that can be used | |
5941 | * by a given scheduling class. | |
5942 | */ | |
5add95d4 | 5943 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5944 | { |
5945 | int ret = -EINVAL; | |
5946 | ||
5947 | switch (policy) { | |
5948 | case SCHED_FIFO: | |
5949 | case SCHED_RR: | |
5950 | ret = 1; | |
5951 | break; | |
5952 | case SCHED_NORMAL: | |
b0a9499c | 5953 | case SCHED_BATCH: |
dd41f596 | 5954 | case SCHED_IDLE: |
1da177e4 LT |
5955 | ret = 0; |
5956 | } | |
5957 | return ret; | |
5958 | } | |
5959 | ||
5960 | /** | |
5961 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5962 | * @pid: pid of the process. | |
5963 | * @interval: userspace pointer to the timeslice value. | |
5964 | * | |
5965 | * this syscall writes the default timeslice value of a given process | |
5966 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5967 | */ | |
17da2bd9 | 5968 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5969 | struct timespec __user *, interval) |
1da177e4 | 5970 | { |
36c8b586 | 5971 | struct task_struct *p; |
a4ec24b4 | 5972 | unsigned int time_slice; |
3a5c359a | 5973 | int retval; |
1da177e4 | 5974 | struct timespec t; |
1da177e4 LT |
5975 | |
5976 | if (pid < 0) | |
3a5c359a | 5977 | return -EINVAL; |
1da177e4 LT |
5978 | |
5979 | retval = -ESRCH; | |
5980 | read_lock(&tasklist_lock); | |
5981 | p = find_process_by_pid(pid); | |
5982 | if (!p) | |
5983 | goto out_unlock; | |
5984 | ||
5985 | retval = security_task_getscheduler(p); | |
5986 | if (retval) | |
5987 | goto out_unlock; | |
5988 | ||
77034937 IM |
5989 | /* |
5990 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5991 | * tasks that are on an otherwise idle runqueue: | |
5992 | */ | |
5993 | time_slice = 0; | |
5994 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5995 | time_slice = DEF_TIMESLICE; |
1868f958 | 5996 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5997 | struct sched_entity *se = &p->se; |
5998 | unsigned long flags; | |
5999 | struct rq *rq; | |
6000 | ||
6001 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6002 | if (rq->cfs.load.weight) |
6003 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6004 | task_rq_unlock(rq, &flags); |
6005 | } | |
1da177e4 | 6006 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6007 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6008 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6009 | return retval; |
3a5c359a | 6010 | |
1da177e4 LT |
6011 | out_unlock: |
6012 | read_unlock(&tasklist_lock); | |
6013 | return retval; | |
6014 | } | |
6015 | ||
7c731e0a | 6016 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6017 | |
82a1fcb9 | 6018 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6019 | { |
1da177e4 | 6020 | unsigned long free = 0; |
36c8b586 | 6021 | unsigned state; |
1da177e4 | 6022 | |
1da177e4 | 6023 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6024 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6025 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6026 | #if BITS_PER_LONG == 32 |
1da177e4 | 6027 | if (state == TASK_RUNNING) |
cc4ea795 | 6028 | printk(KERN_CONT " running "); |
1da177e4 | 6029 | else |
cc4ea795 | 6030 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6031 | #else |
6032 | if (state == TASK_RUNNING) | |
cc4ea795 | 6033 | printk(KERN_CONT " running task "); |
1da177e4 | 6034 | else |
cc4ea795 | 6035 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6036 | #endif |
6037 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
6038 | { | |
10ebffde | 6039 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
6040 | while (!*n) |
6041 | n++; | |
10ebffde | 6042 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
6043 | } |
6044 | #endif | |
ba25f9dc | 6045 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 6046 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 6047 | |
5fb5e6de | 6048 | show_stack(p, NULL); |
1da177e4 LT |
6049 | } |
6050 | ||
e59e2ae2 | 6051 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6052 | { |
36c8b586 | 6053 | struct task_struct *g, *p; |
1da177e4 | 6054 | |
4bd77321 IM |
6055 | #if BITS_PER_LONG == 32 |
6056 | printk(KERN_INFO | |
6057 | " task PC stack pid father\n"); | |
1da177e4 | 6058 | #else |
4bd77321 IM |
6059 | printk(KERN_INFO |
6060 | " task PC stack pid father\n"); | |
1da177e4 LT |
6061 | #endif |
6062 | read_lock(&tasklist_lock); | |
6063 | do_each_thread(g, p) { | |
6064 | /* | |
6065 | * reset the NMI-timeout, listing all files on a slow | |
6066 | * console might take alot of time: | |
6067 | */ | |
6068 | touch_nmi_watchdog(); | |
39bc89fd | 6069 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6070 | sched_show_task(p); |
1da177e4 LT |
6071 | } while_each_thread(g, p); |
6072 | ||
04c9167f JF |
6073 | touch_all_softlockup_watchdogs(); |
6074 | ||
dd41f596 IM |
6075 | #ifdef CONFIG_SCHED_DEBUG |
6076 | sysrq_sched_debug_show(); | |
6077 | #endif | |
1da177e4 | 6078 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6079 | /* |
6080 | * Only show locks if all tasks are dumped: | |
6081 | */ | |
6082 | if (state_filter == -1) | |
6083 | debug_show_all_locks(); | |
1da177e4 LT |
6084 | } |
6085 | ||
1df21055 IM |
6086 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6087 | { | |
dd41f596 | 6088 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6089 | } |
6090 | ||
f340c0d1 IM |
6091 | /** |
6092 | * init_idle - set up an idle thread for a given CPU | |
6093 | * @idle: task in question | |
6094 | * @cpu: cpu the idle task belongs to | |
6095 | * | |
6096 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6097 | * flag, to make booting more robust. | |
6098 | */ | |
5c1e1767 | 6099 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6100 | { |
70b97a7f | 6101 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6102 | unsigned long flags; |
6103 | ||
5cbd54ef IM |
6104 | spin_lock_irqsave(&rq->lock, flags); |
6105 | ||
dd41f596 IM |
6106 | __sched_fork(idle); |
6107 | idle->se.exec_start = sched_clock(); | |
6108 | ||
b29739f9 | 6109 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6110 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6111 | __set_task_cpu(idle, cpu); |
1da177e4 | 6112 | |
1da177e4 | 6113 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6114 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6115 | idle->oncpu = 1; | |
6116 | #endif | |
1da177e4 LT |
6117 | spin_unlock_irqrestore(&rq->lock, flags); |
6118 | ||
6119 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6120 | #if defined(CONFIG_PREEMPT) |
6121 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6122 | #else | |
a1261f54 | 6123 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6124 | #endif |
dd41f596 IM |
6125 | /* |
6126 | * The idle tasks have their own, simple scheduling class: | |
6127 | */ | |
6128 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6129 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6130 | } |
6131 | ||
6132 | /* | |
6133 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6134 | * indicates which cpus entered this state. This is used | |
6135 | * in the rcu update to wait only for active cpus. For system | |
6136 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6137 | * always be CPU_BITS_NONE. |
1da177e4 | 6138 | */ |
6a7b3dc3 | 6139 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6140 | |
19978ca6 IM |
6141 | /* |
6142 | * Increase the granularity value when there are more CPUs, | |
6143 | * because with more CPUs the 'effective latency' as visible | |
6144 | * to users decreases. But the relationship is not linear, | |
6145 | * so pick a second-best guess by going with the log2 of the | |
6146 | * number of CPUs. | |
6147 | * | |
6148 | * This idea comes from the SD scheduler of Con Kolivas: | |
6149 | */ | |
6150 | static inline void sched_init_granularity(void) | |
6151 | { | |
6152 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6153 | const unsigned long limit = 200000000; | |
6154 | ||
6155 | sysctl_sched_min_granularity *= factor; | |
6156 | if (sysctl_sched_min_granularity > limit) | |
6157 | sysctl_sched_min_granularity = limit; | |
6158 | ||
6159 | sysctl_sched_latency *= factor; | |
6160 | if (sysctl_sched_latency > limit) | |
6161 | sysctl_sched_latency = limit; | |
6162 | ||
6163 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6164 | |
6165 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6166 | } |
6167 | ||
1da177e4 LT |
6168 | #ifdef CONFIG_SMP |
6169 | /* | |
6170 | * This is how migration works: | |
6171 | * | |
70b97a7f | 6172 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6173 | * runqueue and wake up that CPU's migration thread. |
6174 | * 2) we down() the locked semaphore => thread blocks. | |
6175 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6176 | * thread off the CPU) | |
6177 | * 4) it gets the migration request and checks whether the migrated | |
6178 | * task is still in the wrong runqueue. | |
6179 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6180 | * it and puts it into the right queue. | |
6181 | * 6) migration thread up()s the semaphore. | |
6182 | * 7) we wake up and the migration is done. | |
6183 | */ | |
6184 | ||
6185 | /* | |
6186 | * Change a given task's CPU affinity. Migrate the thread to a | |
6187 | * proper CPU and schedule it away if the CPU it's executing on | |
6188 | * is removed from the allowed bitmask. | |
6189 | * | |
6190 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6191 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6192 | * call is not atomic; no spinlocks may be held. |
6193 | */ | |
96f874e2 | 6194 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6195 | { |
70b97a7f | 6196 | struct migration_req req; |
1da177e4 | 6197 | unsigned long flags; |
70b97a7f | 6198 | struct rq *rq; |
48f24c4d | 6199 | int ret = 0; |
1da177e4 LT |
6200 | |
6201 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6202 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6203 | ret = -EINVAL; |
6204 | goto out; | |
6205 | } | |
6206 | ||
9985b0ba | 6207 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6208 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6209 | ret = -EINVAL; |
6210 | goto out; | |
6211 | } | |
6212 | ||
73fe6aae | 6213 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6214 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6215 | else { |
96f874e2 RR |
6216 | cpumask_copy(&p->cpus_allowed, new_mask); |
6217 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6218 | } |
6219 | ||
1da177e4 | 6220 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6221 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6222 | goto out; |
6223 | ||
1e5ce4f4 | 6224 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6225 | /* Need help from migration thread: drop lock and wait. */ |
6226 | task_rq_unlock(rq, &flags); | |
6227 | wake_up_process(rq->migration_thread); | |
6228 | wait_for_completion(&req.done); | |
6229 | tlb_migrate_finish(p->mm); | |
6230 | return 0; | |
6231 | } | |
6232 | out: | |
6233 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6234 | |
1da177e4 LT |
6235 | return ret; |
6236 | } | |
cd8ba7cd | 6237 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6238 | |
6239 | /* | |
41a2d6cf | 6240 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6241 | * this because either it can't run here any more (set_cpus_allowed() |
6242 | * away from this CPU, or CPU going down), or because we're | |
6243 | * attempting to rebalance this task on exec (sched_exec). | |
6244 | * | |
6245 | * So we race with normal scheduler movements, but that's OK, as long | |
6246 | * as the task is no longer on this CPU. | |
efc30814 KK |
6247 | * |
6248 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6249 | */ |
efc30814 | 6250 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6251 | { |
70b97a7f | 6252 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6253 | int ret = 0, on_rq; |
1da177e4 | 6254 | |
e761b772 | 6255 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6256 | return ret; |
1da177e4 LT |
6257 | |
6258 | rq_src = cpu_rq(src_cpu); | |
6259 | rq_dest = cpu_rq(dest_cpu); | |
6260 | ||
6261 | double_rq_lock(rq_src, rq_dest); | |
6262 | /* Already moved. */ | |
6263 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6264 | goto done; |
1da177e4 | 6265 | /* Affinity changed (again). */ |
96f874e2 | 6266 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6267 | goto fail; |
1da177e4 | 6268 | |
dd41f596 | 6269 | on_rq = p->se.on_rq; |
6e82a3be | 6270 | if (on_rq) |
2e1cb74a | 6271 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6272 | |
1da177e4 | 6273 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6274 | if (on_rq) { |
6275 | activate_task(rq_dest, p, 0); | |
15afe09b | 6276 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6277 | } |
b1e38734 | 6278 | done: |
efc30814 | 6279 | ret = 1; |
b1e38734 | 6280 | fail: |
1da177e4 | 6281 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6282 | return ret; |
1da177e4 LT |
6283 | } |
6284 | ||
6285 | /* | |
6286 | * migration_thread - this is a highprio system thread that performs | |
6287 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6288 | * another runqueue. | |
6289 | */ | |
95cdf3b7 | 6290 | static int migration_thread(void *data) |
1da177e4 | 6291 | { |
1da177e4 | 6292 | int cpu = (long)data; |
70b97a7f | 6293 | struct rq *rq; |
1da177e4 LT |
6294 | |
6295 | rq = cpu_rq(cpu); | |
6296 | BUG_ON(rq->migration_thread != current); | |
6297 | ||
6298 | set_current_state(TASK_INTERRUPTIBLE); | |
6299 | while (!kthread_should_stop()) { | |
70b97a7f | 6300 | struct migration_req *req; |
1da177e4 | 6301 | struct list_head *head; |
1da177e4 | 6302 | |
1da177e4 LT |
6303 | spin_lock_irq(&rq->lock); |
6304 | ||
6305 | if (cpu_is_offline(cpu)) { | |
6306 | spin_unlock_irq(&rq->lock); | |
6307 | goto wait_to_die; | |
6308 | } | |
6309 | ||
6310 | if (rq->active_balance) { | |
6311 | active_load_balance(rq, cpu); | |
6312 | rq->active_balance = 0; | |
6313 | } | |
6314 | ||
6315 | head = &rq->migration_queue; | |
6316 | ||
6317 | if (list_empty(head)) { | |
6318 | spin_unlock_irq(&rq->lock); | |
6319 | schedule(); | |
6320 | set_current_state(TASK_INTERRUPTIBLE); | |
6321 | continue; | |
6322 | } | |
70b97a7f | 6323 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6324 | list_del_init(head->next); |
6325 | ||
674311d5 NP |
6326 | spin_unlock(&rq->lock); |
6327 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6328 | local_irq_enable(); | |
1da177e4 LT |
6329 | |
6330 | complete(&req->done); | |
6331 | } | |
6332 | __set_current_state(TASK_RUNNING); | |
6333 | return 0; | |
6334 | ||
6335 | wait_to_die: | |
6336 | /* Wait for kthread_stop */ | |
6337 | set_current_state(TASK_INTERRUPTIBLE); | |
6338 | while (!kthread_should_stop()) { | |
6339 | schedule(); | |
6340 | set_current_state(TASK_INTERRUPTIBLE); | |
6341 | } | |
6342 | __set_current_state(TASK_RUNNING); | |
6343 | return 0; | |
6344 | } | |
6345 | ||
6346 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6347 | |
6348 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6349 | { | |
6350 | int ret; | |
6351 | ||
6352 | local_irq_disable(); | |
6353 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6354 | local_irq_enable(); | |
6355 | return ret; | |
6356 | } | |
6357 | ||
054b9108 | 6358 | /* |
3a4fa0a2 | 6359 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6360 | */ |
48f24c4d | 6361 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6362 | { |
70b97a7f | 6363 | int dest_cpu; |
6ca09dfc | 6364 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
6365 | |
6366 | again: | |
6367 | /* Look for allowed, online CPU in same node. */ | |
6368 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6369 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6370 | goto move; | |
6371 | ||
6372 | /* Any allowed, online CPU? */ | |
6373 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6374 | if (dest_cpu < nr_cpu_ids) | |
6375 | goto move; | |
6376 | ||
6377 | /* No more Mr. Nice Guy. */ | |
6378 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6379 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6380 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6381 | |
e76bd8d9 RR |
6382 | /* |
6383 | * Don't tell them about moving exiting tasks or | |
6384 | * kernel threads (both mm NULL), since they never | |
6385 | * leave kernel. | |
6386 | */ | |
6387 | if (p->mm && printk_ratelimit()) { | |
6388 | printk(KERN_INFO "process %d (%s) no " | |
6389 | "longer affine to cpu%d\n", | |
6390 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6391 | } |
e76bd8d9 RR |
6392 | } |
6393 | ||
6394 | move: | |
6395 | /* It can have affinity changed while we were choosing. */ | |
6396 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6397 | goto again; | |
1da177e4 LT |
6398 | } |
6399 | ||
6400 | /* | |
6401 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6402 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6403 | * for performance reasons the counter is not stricly tracking tasks to | |
6404 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6405 | * to keep the global sum constant after CPU-down: | |
6406 | */ | |
70b97a7f | 6407 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6408 | { |
1e5ce4f4 | 6409 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6410 | unsigned long flags; |
6411 | ||
6412 | local_irq_save(flags); | |
6413 | double_rq_lock(rq_src, rq_dest); | |
6414 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6415 | rq_src->nr_uninterruptible = 0; | |
6416 | double_rq_unlock(rq_src, rq_dest); | |
6417 | local_irq_restore(flags); | |
6418 | } | |
6419 | ||
6420 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6421 | static void migrate_live_tasks(int src_cpu) | |
6422 | { | |
48f24c4d | 6423 | struct task_struct *p, *t; |
1da177e4 | 6424 | |
f7b4cddc | 6425 | read_lock(&tasklist_lock); |
1da177e4 | 6426 | |
48f24c4d IM |
6427 | do_each_thread(t, p) { |
6428 | if (p == current) | |
1da177e4 LT |
6429 | continue; |
6430 | ||
48f24c4d IM |
6431 | if (task_cpu(p) == src_cpu) |
6432 | move_task_off_dead_cpu(src_cpu, p); | |
6433 | } while_each_thread(t, p); | |
1da177e4 | 6434 | |
f7b4cddc | 6435 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6436 | } |
6437 | ||
dd41f596 IM |
6438 | /* |
6439 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6440 | * It does so by boosting its priority to highest possible. |
6441 | * Used by CPU offline code. | |
1da177e4 LT |
6442 | */ |
6443 | void sched_idle_next(void) | |
6444 | { | |
48f24c4d | 6445 | int this_cpu = smp_processor_id(); |
70b97a7f | 6446 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6447 | struct task_struct *p = rq->idle; |
6448 | unsigned long flags; | |
6449 | ||
6450 | /* cpu has to be offline */ | |
48f24c4d | 6451 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6452 | |
48f24c4d IM |
6453 | /* |
6454 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6455 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6456 | */ |
6457 | spin_lock_irqsave(&rq->lock, flags); | |
6458 | ||
dd41f596 | 6459 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6460 | |
94bc9a7b DA |
6461 | update_rq_clock(rq); |
6462 | activate_task(rq, p, 0); | |
1da177e4 LT |
6463 | |
6464 | spin_unlock_irqrestore(&rq->lock, flags); | |
6465 | } | |
6466 | ||
48f24c4d IM |
6467 | /* |
6468 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6469 | * offline. |
6470 | */ | |
6471 | void idle_task_exit(void) | |
6472 | { | |
6473 | struct mm_struct *mm = current->active_mm; | |
6474 | ||
6475 | BUG_ON(cpu_online(smp_processor_id())); | |
6476 | ||
6477 | if (mm != &init_mm) | |
6478 | switch_mm(mm, &init_mm, current); | |
6479 | mmdrop(mm); | |
6480 | } | |
6481 | ||
054b9108 | 6482 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6483 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6484 | { |
70b97a7f | 6485 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6486 | |
6487 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6488 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6489 | |
6490 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6491 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6492 | |
48f24c4d | 6493 | get_task_struct(p); |
1da177e4 LT |
6494 | |
6495 | /* | |
6496 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6497 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6498 | * fine. |
6499 | */ | |
f7b4cddc | 6500 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6501 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6502 | spin_lock_irq(&rq->lock); |
1da177e4 | 6503 | |
48f24c4d | 6504 | put_task_struct(p); |
1da177e4 LT |
6505 | } |
6506 | ||
6507 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6508 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6509 | { | |
70b97a7f | 6510 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6511 | struct task_struct *next; |
48f24c4d | 6512 | |
dd41f596 IM |
6513 | for ( ; ; ) { |
6514 | if (!rq->nr_running) | |
6515 | break; | |
a8e504d2 | 6516 | update_rq_clock(rq); |
ff95f3df | 6517 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6518 | if (!next) |
6519 | break; | |
79c53799 | 6520 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6521 | migrate_dead(dead_cpu, next); |
e692ab53 | 6522 | |
1da177e4 LT |
6523 | } |
6524 | } | |
6525 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6526 | ||
e692ab53 NP |
6527 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6528 | ||
6529 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6530 | { |
6531 | .procname = "sched_domain", | |
c57baf1e | 6532 | .mode = 0555, |
e0361851 | 6533 | }, |
38605cae | 6534 | {0, }, |
e692ab53 NP |
6535 | }; |
6536 | ||
6537 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6538 | { |
c57baf1e | 6539 | .ctl_name = CTL_KERN, |
e0361851 | 6540 | .procname = "kernel", |
c57baf1e | 6541 | .mode = 0555, |
e0361851 AD |
6542 | .child = sd_ctl_dir, |
6543 | }, | |
38605cae | 6544 | {0, }, |
e692ab53 NP |
6545 | }; |
6546 | ||
6547 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6548 | { | |
6549 | struct ctl_table *entry = | |
5cf9f062 | 6550 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6551 | |
e692ab53 NP |
6552 | return entry; |
6553 | } | |
6554 | ||
6382bc90 MM |
6555 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6556 | { | |
cd790076 | 6557 | struct ctl_table *entry; |
6382bc90 | 6558 | |
cd790076 MM |
6559 | /* |
6560 | * In the intermediate directories, both the child directory and | |
6561 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6562 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6563 | * static strings and all have proc handlers. |
6564 | */ | |
6565 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6566 | if (entry->child) |
6567 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6568 | if (entry->proc_handler == NULL) |
6569 | kfree(entry->procname); | |
6570 | } | |
6382bc90 MM |
6571 | |
6572 | kfree(*tablep); | |
6573 | *tablep = NULL; | |
6574 | } | |
6575 | ||
e692ab53 | 6576 | static void |
e0361851 | 6577 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6578 | const char *procname, void *data, int maxlen, |
6579 | mode_t mode, proc_handler *proc_handler) | |
6580 | { | |
e692ab53 NP |
6581 | entry->procname = procname; |
6582 | entry->data = data; | |
6583 | entry->maxlen = maxlen; | |
6584 | entry->mode = mode; | |
6585 | entry->proc_handler = proc_handler; | |
6586 | } | |
6587 | ||
6588 | static struct ctl_table * | |
6589 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6590 | { | |
a5d8c348 | 6591 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6592 | |
ad1cdc1d MM |
6593 | if (table == NULL) |
6594 | return NULL; | |
6595 | ||
e0361851 | 6596 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6597 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6598 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6599 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6600 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6601 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6602 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6603 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6604 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6605 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6606 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6607 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6608 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6609 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6610 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6611 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6612 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6613 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6614 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6615 | &sd->cache_nice_tries, |
6616 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6617 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6618 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6619 | set_table_entry(&table[11], "name", sd->name, |
6620 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6621 | /* &table[12] is terminator */ | |
e692ab53 NP |
6622 | |
6623 | return table; | |
6624 | } | |
6625 | ||
9a4e7159 | 6626 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6627 | { |
6628 | struct ctl_table *entry, *table; | |
6629 | struct sched_domain *sd; | |
6630 | int domain_num = 0, i; | |
6631 | char buf[32]; | |
6632 | ||
6633 | for_each_domain(cpu, sd) | |
6634 | domain_num++; | |
6635 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6636 | if (table == NULL) |
6637 | return NULL; | |
e692ab53 NP |
6638 | |
6639 | i = 0; | |
6640 | for_each_domain(cpu, sd) { | |
6641 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6642 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6643 | entry->mode = 0555; |
e692ab53 NP |
6644 | entry->child = sd_alloc_ctl_domain_table(sd); |
6645 | entry++; | |
6646 | i++; | |
6647 | } | |
6648 | return table; | |
6649 | } | |
6650 | ||
6651 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6652 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6653 | { |
6654 | int i, cpu_num = num_online_cpus(); | |
6655 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6656 | char buf[32]; | |
6657 | ||
7378547f MM |
6658 | WARN_ON(sd_ctl_dir[0].child); |
6659 | sd_ctl_dir[0].child = entry; | |
6660 | ||
ad1cdc1d MM |
6661 | if (entry == NULL) |
6662 | return; | |
6663 | ||
97b6ea7b | 6664 | for_each_online_cpu(i) { |
e692ab53 | 6665 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6666 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6667 | entry->mode = 0555; |
e692ab53 | 6668 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6669 | entry++; |
e692ab53 | 6670 | } |
7378547f MM |
6671 | |
6672 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6673 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6674 | } | |
6382bc90 | 6675 | |
7378547f | 6676 | /* may be called multiple times per register */ |
6382bc90 MM |
6677 | static void unregister_sched_domain_sysctl(void) |
6678 | { | |
7378547f MM |
6679 | if (sd_sysctl_header) |
6680 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6681 | sd_sysctl_header = NULL; |
7378547f MM |
6682 | if (sd_ctl_dir[0].child) |
6683 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6684 | } |
e692ab53 | 6685 | #else |
6382bc90 MM |
6686 | static void register_sched_domain_sysctl(void) |
6687 | { | |
6688 | } | |
6689 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6690 | { |
6691 | } | |
6692 | #endif | |
6693 | ||
1f11eb6a GH |
6694 | static void set_rq_online(struct rq *rq) |
6695 | { | |
6696 | if (!rq->online) { | |
6697 | const struct sched_class *class; | |
6698 | ||
c6c4927b | 6699 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6700 | rq->online = 1; |
6701 | ||
6702 | for_each_class(class) { | |
6703 | if (class->rq_online) | |
6704 | class->rq_online(rq); | |
6705 | } | |
6706 | } | |
6707 | } | |
6708 | ||
6709 | static void set_rq_offline(struct rq *rq) | |
6710 | { | |
6711 | if (rq->online) { | |
6712 | const struct sched_class *class; | |
6713 | ||
6714 | for_each_class(class) { | |
6715 | if (class->rq_offline) | |
6716 | class->rq_offline(rq); | |
6717 | } | |
6718 | ||
c6c4927b | 6719 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6720 | rq->online = 0; |
6721 | } | |
6722 | } | |
6723 | ||
1da177e4 LT |
6724 | /* |
6725 | * migration_call - callback that gets triggered when a CPU is added. | |
6726 | * Here we can start up the necessary migration thread for the new CPU. | |
6727 | */ | |
48f24c4d IM |
6728 | static int __cpuinit |
6729 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6730 | { |
1da177e4 | 6731 | struct task_struct *p; |
48f24c4d | 6732 | int cpu = (long)hcpu; |
1da177e4 | 6733 | unsigned long flags; |
70b97a7f | 6734 | struct rq *rq; |
1da177e4 LT |
6735 | |
6736 | switch (action) { | |
5be9361c | 6737 | |
1da177e4 | 6738 | case CPU_UP_PREPARE: |
8bb78442 | 6739 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6740 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6741 | if (IS_ERR(p)) |
6742 | return NOTIFY_BAD; | |
1da177e4 LT |
6743 | kthread_bind(p, cpu); |
6744 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6745 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6746 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6747 | task_rq_unlock(rq, &flags); |
6748 | cpu_rq(cpu)->migration_thread = p; | |
6749 | break; | |
48f24c4d | 6750 | |
1da177e4 | 6751 | case CPU_ONLINE: |
8bb78442 | 6752 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6753 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6754 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6755 | |
6756 | /* Update our root-domain */ | |
6757 | rq = cpu_rq(cpu); | |
6758 | spin_lock_irqsave(&rq->lock, flags); | |
6759 | if (rq->rd) { | |
c6c4927b | 6760 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6761 | |
6762 | set_rq_online(rq); | |
1f94ef59 GH |
6763 | } |
6764 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6765 | break; |
48f24c4d | 6766 | |
1da177e4 LT |
6767 | #ifdef CONFIG_HOTPLUG_CPU |
6768 | case CPU_UP_CANCELED: | |
8bb78442 | 6769 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6770 | if (!cpu_rq(cpu)->migration_thread) |
6771 | break; | |
41a2d6cf | 6772 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 6773 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 6774 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6775 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6776 | cpu_rq(cpu)->migration_thread = NULL; | |
6777 | break; | |
48f24c4d | 6778 | |
1da177e4 | 6779 | case CPU_DEAD: |
8bb78442 | 6780 | case CPU_DEAD_FROZEN: |
470fd646 | 6781 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6782 | migrate_live_tasks(cpu); |
6783 | rq = cpu_rq(cpu); | |
6784 | kthread_stop(rq->migration_thread); | |
6785 | rq->migration_thread = NULL; | |
6786 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6787 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6788 | update_rq_clock(rq); |
2e1cb74a | 6789 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6790 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6791 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6792 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6793 | migrate_dead_tasks(cpu); |
d2da272a | 6794 | spin_unlock_irq(&rq->lock); |
470fd646 | 6795 | cpuset_unlock(); |
1da177e4 LT |
6796 | migrate_nr_uninterruptible(rq); |
6797 | BUG_ON(rq->nr_running != 0); | |
6798 | ||
41a2d6cf IM |
6799 | /* |
6800 | * No need to migrate the tasks: it was best-effort if | |
6801 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6802 | * the requestors. | |
6803 | */ | |
1da177e4 LT |
6804 | spin_lock_irq(&rq->lock); |
6805 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6806 | struct migration_req *req; |
6807 | ||
1da177e4 | 6808 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6809 | struct migration_req, list); |
1da177e4 | 6810 | list_del_init(&req->list); |
9a2bd244 | 6811 | spin_unlock_irq(&rq->lock); |
1da177e4 | 6812 | complete(&req->done); |
9a2bd244 | 6813 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
6814 | } |
6815 | spin_unlock_irq(&rq->lock); | |
6816 | break; | |
57d885fe | 6817 | |
08f503b0 GH |
6818 | case CPU_DYING: |
6819 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6820 | /* Update our root-domain */ |
6821 | rq = cpu_rq(cpu); | |
6822 | spin_lock_irqsave(&rq->lock, flags); | |
6823 | if (rq->rd) { | |
c6c4927b | 6824 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6825 | set_rq_offline(rq); |
57d885fe GH |
6826 | } |
6827 | spin_unlock_irqrestore(&rq->lock, flags); | |
6828 | break; | |
1da177e4 LT |
6829 | #endif |
6830 | } | |
6831 | return NOTIFY_OK; | |
6832 | } | |
6833 | ||
6834 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6835 | * happens before everything else. | |
6836 | */ | |
26c2143b | 6837 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6838 | .notifier_call = migration_call, |
6839 | .priority = 10 | |
6840 | }; | |
6841 | ||
7babe8db | 6842 | static int __init migration_init(void) |
1da177e4 LT |
6843 | { |
6844 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6845 | int err; |
48f24c4d IM |
6846 | |
6847 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6848 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6849 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6850 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6851 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6852 | |
6853 | return err; | |
1da177e4 | 6854 | } |
7babe8db | 6855 | early_initcall(migration_init); |
1da177e4 LT |
6856 | #endif |
6857 | ||
6858 | #ifdef CONFIG_SMP | |
476f3534 | 6859 | |
3e9830dc | 6860 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6861 | |
7c16ec58 | 6862 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6863 | struct cpumask *groupmask) |
1da177e4 | 6864 | { |
4dcf6aff | 6865 | struct sched_group *group = sd->groups; |
434d53b0 | 6866 | char str[256]; |
1da177e4 | 6867 | |
968ea6d8 | 6868 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6869 | cpumask_clear(groupmask); |
4dcf6aff IM |
6870 | |
6871 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6872 | ||
6873 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6874 | printk("does not load-balance\n"); | |
6875 | if (sd->parent) | |
6876 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6877 | " has parent"); | |
6878 | return -1; | |
41c7ce9a NP |
6879 | } |
6880 | ||
eefd796a | 6881 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6882 | |
758b2cdc | 6883 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
6884 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6885 | "CPU%d\n", cpu); | |
6886 | } | |
758b2cdc | 6887 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
6888 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6889 | " CPU%d\n", cpu); | |
6890 | } | |
1da177e4 | 6891 | |
4dcf6aff | 6892 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6893 | do { |
4dcf6aff IM |
6894 | if (!group) { |
6895 | printk("\n"); | |
6896 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6897 | break; |
6898 | } | |
6899 | ||
4dcf6aff IM |
6900 | if (!group->__cpu_power) { |
6901 | printk(KERN_CONT "\n"); | |
6902 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6903 | "set\n"); | |
6904 | break; | |
6905 | } | |
1da177e4 | 6906 | |
758b2cdc | 6907 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
6908 | printk(KERN_CONT "\n"); |
6909 | printk(KERN_ERR "ERROR: empty group\n"); | |
6910 | break; | |
6911 | } | |
1da177e4 | 6912 | |
758b2cdc | 6913 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
6914 | printk(KERN_CONT "\n"); |
6915 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6916 | break; | |
6917 | } | |
1da177e4 | 6918 | |
758b2cdc | 6919 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6920 | |
968ea6d8 | 6921 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
4dcf6aff | 6922 | printk(KERN_CONT " %s", str); |
1da177e4 | 6923 | |
4dcf6aff IM |
6924 | group = group->next; |
6925 | } while (group != sd->groups); | |
6926 | printk(KERN_CONT "\n"); | |
1da177e4 | 6927 | |
758b2cdc | 6928 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 6929 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6930 | |
758b2cdc RR |
6931 | if (sd->parent && |
6932 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
6933 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6934 | "of domain->span\n"); | |
6935 | return 0; | |
6936 | } | |
1da177e4 | 6937 | |
4dcf6aff IM |
6938 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6939 | { | |
d5dd3db1 | 6940 | cpumask_var_t groupmask; |
4dcf6aff | 6941 | int level = 0; |
1da177e4 | 6942 | |
4dcf6aff IM |
6943 | if (!sd) { |
6944 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6945 | return; | |
6946 | } | |
1da177e4 | 6947 | |
4dcf6aff IM |
6948 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6949 | ||
d5dd3db1 | 6950 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6951 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6952 | return; | |
6953 | } | |
6954 | ||
4dcf6aff | 6955 | for (;;) { |
7c16ec58 | 6956 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6957 | break; |
1da177e4 LT |
6958 | level++; |
6959 | sd = sd->parent; | |
33859f7f | 6960 | if (!sd) |
4dcf6aff IM |
6961 | break; |
6962 | } | |
d5dd3db1 | 6963 | free_cpumask_var(groupmask); |
1da177e4 | 6964 | } |
6d6bc0ad | 6965 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6966 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6967 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6968 | |
1a20ff27 | 6969 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6970 | { |
758b2cdc | 6971 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6972 | return 1; |
6973 | ||
6974 | /* Following flags need at least 2 groups */ | |
6975 | if (sd->flags & (SD_LOAD_BALANCE | | |
6976 | SD_BALANCE_NEWIDLE | | |
6977 | SD_BALANCE_FORK | | |
89c4710e SS |
6978 | SD_BALANCE_EXEC | |
6979 | SD_SHARE_CPUPOWER | | |
6980 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6981 | if (sd->groups != sd->groups->next) |
6982 | return 0; | |
6983 | } | |
6984 | ||
6985 | /* Following flags don't use groups */ | |
6986 | if (sd->flags & (SD_WAKE_IDLE | | |
6987 | SD_WAKE_AFFINE | | |
6988 | SD_WAKE_BALANCE)) | |
6989 | return 0; | |
6990 | ||
6991 | return 1; | |
6992 | } | |
6993 | ||
48f24c4d IM |
6994 | static int |
6995 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6996 | { |
6997 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6998 | ||
6999 | if (sd_degenerate(parent)) | |
7000 | return 1; | |
7001 | ||
758b2cdc | 7002 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7003 | return 0; |
7004 | ||
7005 | /* Does parent contain flags not in child? */ | |
7006 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7007 | if (cflags & SD_WAKE_AFFINE) | |
7008 | pflags &= ~SD_WAKE_BALANCE; | |
7009 | /* Flags needing groups don't count if only 1 group in parent */ | |
7010 | if (parent->groups == parent->groups->next) { | |
7011 | pflags &= ~(SD_LOAD_BALANCE | | |
7012 | SD_BALANCE_NEWIDLE | | |
7013 | SD_BALANCE_FORK | | |
89c4710e SS |
7014 | SD_BALANCE_EXEC | |
7015 | SD_SHARE_CPUPOWER | | |
7016 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7017 | if (nr_node_ids == 1) |
7018 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7019 | } |
7020 | if (~cflags & pflags) | |
7021 | return 0; | |
7022 | ||
7023 | return 1; | |
7024 | } | |
7025 | ||
c6c4927b RR |
7026 | static void free_rootdomain(struct root_domain *rd) |
7027 | { | |
68e74568 RR |
7028 | cpupri_cleanup(&rd->cpupri); |
7029 | ||
c6c4927b RR |
7030 | free_cpumask_var(rd->rto_mask); |
7031 | free_cpumask_var(rd->online); | |
7032 | free_cpumask_var(rd->span); | |
7033 | kfree(rd); | |
7034 | } | |
7035 | ||
57d885fe GH |
7036 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7037 | { | |
a0490fa3 | 7038 | struct root_domain *old_rd = NULL; |
57d885fe | 7039 | unsigned long flags; |
57d885fe GH |
7040 | |
7041 | spin_lock_irqsave(&rq->lock, flags); | |
7042 | ||
7043 | if (rq->rd) { | |
a0490fa3 | 7044 | old_rd = rq->rd; |
57d885fe | 7045 | |
c6c4927b | 7046 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7047 | set_rq_offline(rq); |
57d885fe | 7048 | |
c6c4927b | 7049 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7050 | |
a0490fa3 IM |
7051 | /* |
7052 | * If we dont want to free the old_rt yet then | |
7053 | * set old_rd to NULL to skip the freeing later | |
7054 | * in this function: | |
7055 | */ | |
7056 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7057 | old_rd = NULL; | |
57d885fe GH |
7058 | } |
7059 | ||
7060 | atomic_inc(&rd->refcount); | |
7061 | rq->rd = rd; | |
7062 | ||
c6c4927b RR |
7063 | cpumask_set_cpu(rq->cpu, rd->span); |
7064 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7065 | set_rq_online(rq); |
57d885fe GH |
7066 | |
7067 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7068 | |
7069 | if (old_rd) | |
7070 | free_rootdomain(old_rd); | |
57d885fe GH |
7071 | } |
7072 | ||
db2f59c8 | 7073 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7074 | { |
7075 | memset(rd, 0, sizeof(*rd)); | |
7076 | ||
c6c4927b RR |
7077 | if (bootmem) { |
7078 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7079 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7080 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7081 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7082 | return 0; |
7083 | } | |
7084 | ||
7085 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7086 | goto out; |
c6c4927b RR |
7087 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7088 | goto free_span; | |
7089 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7090 | goto free_online; | |
6e0534f2 | 7091 | |
68e74568 RR |
7092 | if (cpupri_init(&rd->cpupri, false) != 0) |
7093 | goto free_rto_mask; | |
c6c4927b | 7094 | return 0; |
6e0534f2 | 7095 | |
68e74568 RR |
7096 | free_rto_mask: |
7097 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7098 | free_online: |
7099 | free_cpumask_var(rd->online); | |
7100 | free_span: | |
7101 | free_cpumask_var(rd->span); | |
0c910d28 | 7102 | out: |
c6c4927b | 7103 | return -ENOMEM; |
57d885fe GH |
7104 | } |
7105 | ||
7106 | static void init_defrootdomain(void) | |
7107 | { | |
c6c4927b RR |
7108 | init_rootdomain(&def_root_domain, true); |
7109 | ||
57d885fe GH |
7110 | atomic_set(&def_root_domain.refcount, 1); |
7111 | } | |
7112 | ||
dc938520 | 7113 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7114 | { |
7115 | struct root_domain *rd; | |
7116 | ||
7117 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7118 | if (!rd) | |
7119 | return NULL; | |
7120 | ||
c6c4927b RR |
7121 | if (init_rootdomain(rd, false) != 0) { |
7122 | kfree(rd); | |
7123 | return NULL; | |
7124 | } | |
57d885fe GH |
7125 | |
7126 | return rd; | |
7127 | } | |
7128 | ||
1da177e4 | 7129 | /* |
0eab9146 | 7130 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7131 | * hold the hotplug lock. |
7132 | */ | |
0eab9146 IM |
7133 | static void |
7134 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7135 | { |
70b97a7f | 7136 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7137 | struct sched_domain *tmp; |
7138 | ||
7139 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7140 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7141 | struct sched_domain *parent = tmp->parent; |
7142 | if (!parent) | |
7143 | break; | |
f29c9b1c | 7144 | |
1a848870 | 7145 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7146 | tmp->parent = parent->parent; |
1a848870 SS |
7147 | if (parent->parent) |
7148 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7149 | } else |
7150 | tmp = tmp->parent; | |
245af2c7 SS |
7151 | } |
7152 | ||
1a848870 | 7153 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7154 | sd = sd->parent; |
1a848870 SS |
7155 | if (sd) |
7156 | sd->child = NULL; | |
7157 | } | |
1da177e4 LT |
7158 | |
7159 | sched_domain_debug(sd, cpu); | |
7160 | ||
57d885fe | 7161 | rq_attach_root(rq, rd); |
674311d5 | 7162 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7163 | } |
7164 | ||
7165 | /* cpus with isolated domains */ | |
dcc30a35 | 7166 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7167 | |
7168 | /* Setup the mask of cpus configured for isolated domains */ | |
7169 | static int __init isolated_cpu_setup(char *str) | |
7170 | { | |
968ea6d8 | 7171 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7172 | return 1; |
7173 | } | |
7174 | ||
8927f494 | 7175 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7176 | |
7177 | /* | |
6711cab4 SS |
7178 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7179 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7180 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7181 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7182 | * |
7183 | * init_sched_build_groups will build a circular linked list of the groups | |
7184 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7185 | * and ->cpu_power to 0. | |
7186 | */ | |
a616058b | 7187 | static void |
96f874e2 RR |
7188 | init_sched_build_groups(const struct cpumask *span, |
7189 | const struct cpumask *cpu_map, | |
7190 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7191 | struct sched_group **sg, |
96f874e2 RR |
7192 | struct cpumask *tmpmask), |
7193 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7194 | { |
7195 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7196 | int i; |
7197 | ||
96f874e2 | 7198 | cpumask_clear(covered); |
7c16ec58 | 7199 | |
abcd083a | 7200 | for_each_cpu(i, span) { |
6711cab4 | 7201 | struct sched_group *sg; |
7c16ec58 | 7202 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7203 | int j; |
7204 | ||
758b2cdc | 7205 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7206 | continue; |
7207 | ||
758b2cdc | 7208 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7209 | sg->__cpu_power = 0; |
1da177e4 | 7210 | |
abcd083a | 7211 | for_each_cpu(j, span) { |
7c16ec58 | 7212 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7213 | continue; |
7214 | ||
96f874e2 | 7215 | cpumask_set_cpu(j, covered); |
758b2cdc | 7216 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7217 | } |
7218 | if (!first) | |
7219 | first = sg; | |
7220 | if (last) | |
7221 | last->next = sg; | |
7222 | last = sg; | |
7223 | } | |
7224 | last->next = first; | |
7225 | } | |
7226 | ||
9c1cfda2 | 7227 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7228 | |
9c1cfda2 | 7229 | #ifdef CONFIG_NUMA |
198e2f18 | 7230 | |
9c1cfda2 JH |
7231 | /** |
7232 | * find_next_best_node - find the next node to include in a sched_domain | |
7233 | * @node: node whose sched_domain we're building | |
7234 | * @used_nodes: nodes already in the sched_domain | |
7235 | * | |
41a2d6cf | 7236 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7237 | * finds the closest node not already in the @used_nodes map. |
7238 | * | |
7239 | * Should use nodemask_t. | |
7240 | */ | |
c5f59f08 | 7241 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7242 | { |
7243 | int i, n, val, min_val, best_node = 0; | |
7244 | ||
7245 | min_val = INT_MAX; | |
7246 | ||
076ac2af | 7247 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7248 | /* Start at @node */ |
076ac2af | 7249 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7250 | |
7251 | if (!nr_cpus_node(n)) | |
7252 | continue; | |
7253 | ||
7254 | /* Skip already used nodes */ | |
c5f59f08 | 7255 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7256 | continue; |
7257 | ||
7258 | /* Simple min distance search */ | |
7259 | val = node_distance(node, n); | |
7260 | ||
7261 | if (val < min_val) { | |
7262 | min_val = val; | |
7263 | best_node = n; | |
7264 | } | |
7265 | } | |
7266 | ||
c5f59f08 | 7267 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7268 | return best_node; |
7269 | } | |
7270 | ||
7271 | /** | |
7272 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7273 | * @node: node whose cpumask we're constructing | |
73486722 | 7274 | * @span: resulting cpumask |
9c1cfda2 | 7275 | * |
41a2d6cf | 7276 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7277 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7278 | * out optimally. | |
7279 | */ | |
96f874e2 | 7280 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7281 | { |
c5f59f08 | 7282 | nodemask_t used_nodes; |
48f24c4d | 7283 | int i; |
9c1cfda2 | 7284 | |
6ca09dfc | 7285 | cpumask_clear(span); |
c5f59f08 | 7286 | nodes_clear(used_nodes); |
9c1cfda2 | 7287 | |
6ca09dfc | 7288 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7289 | node_set(node, used_nodes); |
9c1cfda2 JH |
7290 | |
7291 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7292 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7293 | |
6ca09dfc | 7294 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7295 | } |
9c1cfda2 | 7296 | } |
6d6bc0ad | 7297 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7298 | |
5c45bf27 | 7299 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7300 | |
6c99e9ad RR |
7301 | /* |
7302 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7303 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7304 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7305 | */ | |
7306 | struct static_sched_group { | |
7307 | struct sched_group sg; | |
7308 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7309 | }; | |
7310 | ||
7311 | struct static_sched_domain { | |
7312 | struct sched_domain sd; | |
7313 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7314 | }; | |
7315 | ||
9c1cfda2 | 7316 | /* |
48f24c4d | 7317 | * SMT sched-domains: |
9c1cfda2 | 7318 | */ |
1da177e4 | 7319 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7320 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7321 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7322 | |
41a2d6cf | 7323 | static int |
96f874e2 RR |
7324 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7325 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7326 | { |
6711cab4 | 7327 | if (sg) |
6c99e9ad | 7328 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7329 | return cpu; |
7330 | } | |
6d6bc0ad | 7331 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7332 | |
48f24c4d IM |
7333 | /* |
7334 | * multi-core sched-domains: | |
7335 | */ | |
1e9f28fa | 7336 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7337 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7338 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7339 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7340 | |
7341 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7342 | static int |
96f874e2 RR |
7343 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7344 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7345 | { |
6711cab4 | 7346 | int group; |
7c16ec58 | 7347 | |
96f874e2 RR |
7348 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7349 | group = cpumask_first(mask); | |
6711cab4 | 7350 | if (sg) |
6c99e9ad | 7351 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7352 | return group; |
1e9f28fa SS |
7353 | } |
7354 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7355 | static int |
96f874e2 RR |
7356 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7357 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7358 | { |
6711cab4 | 7359 | if (sg) |
6c99e9ad | 7360 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7361 | return cpu; |
7362 | } | |
7363 | #endif | |
7364 | ||
6c99e9ad RR |
7365 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7366 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7367 | |
41a2d6cf | 7368 | static int |
96f874e2 RR |
7369 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7370 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7371 | { |
6711cab4 | 7372 | int group; |
48f24c4d | 7373 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7374 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7375 | group = cpumask_first(mask); |
1e9f28fa | 7376 | #elif defined(CONFIG_SCHED_SMT) |
96f874e2 RR |
7377 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7378 | group = cpumask_first(mask); | |
1da177e4 | 7379 | #else |
6711cab4 | 7380 | group = cpu; |
1da177e4 | 7381 | #endif |
6711cab4 | 7382 | if (sg) |
6c99e9ad | 7383 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7384 | return group; |
1da177e4 LT |
7385 | } |
7386 | ||
7387 | #ifdef CONFIG_NUMA | |
1da177e4 | 7388 | /* |
9c1cfda2 JH |
7389 | * The init_sched_build_groups can't handle what we want to do with node |
7390 | * groups, so roll our own. Now each node has its own list of groups which | |
7391 | * gets dynamically allocated. | |
1da177e4 | 7392 | */ |
62ea9ceb | 7393 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7394 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7395 | |
62ea9ceb | 7396 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7397 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7398 | |
96f874e2 RR |
7399 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7400 | struct sched_group **sg, | |
7401 | struct cpumask *nodemask) | |
9c1cfda2 | 7402 | { |
6711cab4 SS |
7403 | int group; |
7404 | ||
6ca09dfc | 7405 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7406 | group = cpumask_first(nodemask); |
6711cab4 SS |
7407 | |
7408 | if (sg) | |
6c99e9ad | 7409 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7410 | return group; |
1da177e4 | 7411 | } |
6711cab4 | 7412 | |
08069033 SS |
7413 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7414 | { | |
7415 | struct sched_group *sg = group_head; | |
7416 | int j; | |
7417 | ||
7418 | if (!sg) | |
7419 | return; | |
3a5c359a | 7420 | do { |
758b2cdc | 7421 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7422 | struct sched_domain *sd; |
08069033 | 7423 | |
6c99e9ad | 7424 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7425 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7426 | /* |
7427 | * Only add "power" once for each | |
7428 | * physical package. | |
7429 | */ | |
7430 | continue; | |
7431 | } | |
08069033 | 7432 | |
3a5c359a AK |
7433 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7434 | } | |
7435 | sg = sg->next; | |
7436 | } while (sg != group_head); | |
08069033 | 7437 | } |
6d6bc0ad | 7438 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7439 | |
a616058b | 7440 | #ifdef CONFIG_NUMA |
51888ca2 | 7441 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7442 | static void free_sched_groups(const struct cpumask *cpu_map, |
7443 | struct cpumask *nodemask) | |
51888ca2 | 7444 | { |
a616058b | 7445 | int cpu, i; |
51888ca2 | 7446 | |
abcd083a | 7447 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7448 | struct sched_group **sched_group_nodes |
7449 | = sched_group_nodes_bycpu[cpu]; | |
7450 | ||
51888ca2 SV |
7451 | if (!sched_group_nodes) |
7452 | continue; | |
7453 | ||
076ac2af | 7454 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7455 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7456 | ||
6ca09dfc | 7457 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7458 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7459 | continue; |
7460 | ||
7461 | if (sg == NULL) | |
7462 | continue; | |
7463 | sg = sg->next; | |
7464 | next_sg: | |
7465 | oldsg = sg; | |
7466 | sg = sg->next; | |
7467 | kfree(oldsg); | |
7468 | if (oldsg != sched_group_nodes[i]) | |
7469 | goto next_sg; | |
7470 | } | |
7471 | kfree(sched_group_nodes); | |
7472 | sched_group_nodes_bycpu[cpu] = NULL; | |
7473 | } | |
51888ca2 | 7474 | } |
6d6bc0ad | 7475 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7476 | static void free_sched_groups(const struct cpumask *cpu_map, |
7477 | struct cpumask *nodemask) | |
a616058b SS |
7478 | { |
7479 | } | |
6d6bc0ad | 7480 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7481 | |
89c4710e SS |
7482 | /* |
7483 | * Initialize sched groups cpu_power. | |
7484 | * | |
7485 | * cpu_power indicates the capacity of sched group, which is used while | |
7486 | * distributing the load between different sched groups in a sched domain. | |
7487 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7488 | * there are asymmetries in the topology. If there are asymmetries, group | |
7489 | * having more cpu_power will pickup more load compared to the group having | |
7490 | * less cpu_power. | |
7491 | * | |
7492 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7493 | * the maximum number of tasks a group can handle in the presence of other idle | |
7494 | * or lightly loaded groups in the same sched domain. | |
7495 | */ | |
7496 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7497 | { | |
7498 | struct sched_domain *child; | |
7499 | struct sched_group *group; | |
7500 | ||
7501 | WARN_ON(!sd || !sd->groups); | |
7502 | ||
758b2cdc | 7503 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
7504 | return; |
7505 | ||
7506 | child = sd->child; | |
7507 | ||
5517d86b ED |
7508 | sd->groups->__cpu_power = 0; |
7509 | ||
89c4710e SS |
7510 | /* |
7511 | * For perf policy, if the groups in child domain share resources | |
7512 | * (for example cores sharing some portions of the cache hierarchy | |
7513 | * or SMT), then set this domain groups cpu_power such that each group | |
7514 | * can handle only one task, when there are other idle groups in the | |
7515 | * same sched domain. | |
7516 | */ | |
7517 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7518 | (child->flags & | |
7519 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7520 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7521 | return; |
7522 | } | |
7523 | ||
89c4710e SS |
7524 | /* |
7525 | * add cpu_power of each child group to this groups cpu_power | |
7526 | */ | |
7527 | group = child->groups; | |
7528 | do { | |
5517d86b | 7529 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7530 | group = group->next; |
7531 | } while (group != child->groups); | |
7532 | } | |
7533 | ||
7c16ec58 MT |
7534 | /* |
7535 | * Initializers for schedule domains | |
7536 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7537 | */ | |
7538 | ||
a5d8c348 IM |
7539 | #ifdef CONFIG_SCHED_DEBUG |
7540 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7541 | #else | |
7542 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7543 | #endif | |
7544 | ||
7c16ec58 | 7545 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7546 | |
7c16ec58 MT |
7547 | #define SD_INIT_FUNC(type) \ |
7548 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7549 | { \ | |
7550 | memset(sd, 0, sizeof(*sd)); \ | |
7551 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7552 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7553 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7554 | } |
7555 | ||
7556 | SD_INIT_FUNC(CPU) | |
7557 | #ifdef CONFIG_NUMA | |
7558 | SD_INIT_FUNC(ALLNODES) | |
7559 | SD_INIT_FUNC(NODE) | |
7560 | #endif | |
7561 | #ifdef CONFIG_SCHED_SMT | |
7562 | SD_INIT_FUNC(SIBLING) | |
7563 | #endif | |
7564 | #ifdef CONFIG_SCHED_MC | |
7565 | SD_INIT_FUNC(MC) | |
7566 | #endif | |
7567 | ||
1d3504fc HS |
7568 | static int default_relax_domain_level = -1; |
7569 | ||
7570 | static int __init setup_relax_domain_level(char *str) | |
7571 | { | |
30e0e178 LZ |
7572 | unsigned long val; |
7573 | ||
7574 | val = simple_strtoul(str, NULL, 0); | |
7575 | if (val < SD_LV_MAX) | |
7576 | default_relax_domain_level = val; | |
7577 | ||
1d3504fc HS |
7578 | return 1; |
7579 | } | |
7580 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7581 | ||
7582 | static void set_domain_attribute(struct sched_domain *sd, | |
7583 | struct sched_domain_attr *attr) | |
7584 | { | |
7585 | int request; | |
7586 | ||
7587 | if (!attr || attr->relax_domain_level < 0) { | |
7588 | if (default_relax_domain_level < 0) | |
7589 | return; | |
7590 | else | |
7591 | request = default_relax_domain_level; | |
7592 | } else | |
7593 | request = attr->relax_domain_level; | |
7594 | if (request < sd->level) { | |
7595 | /* turn off idle balance on this domain */ | |
7596 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7597 | } else { | |
7598 | /* turn on idle balance on this domain */ | |
7599 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7600 | } | |
7601 | } | |
7602 | ||
1da177e4 | 7603 | /* |
1a20ff27 DG |
7604 | * Build sched domains for a given set of cpus and attach the sched domains |
7605 | * to the individual cpus | |
1da177e4 | 7606 | */ |
96f874e2 | 7607 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 7608 | struct sched_domain_attr *attr) |
1da177e4 | 7609 | { |
3404c8d9 | 7610 | int i, err = -ENOMEM; |
57d885fe | 7611 | struct root_domain *rd; |
3404c8d9 RR |
7612 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
7613 | tmpmask; | |
d1b55138 | 7614 | #ifdef CONFIG_NUMA |
3404c8d9 | 7615 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 7616 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 7617 | int sd_allnodes = 0; |
d1b55138 | 7618 | |
3404c8d9 RR |
7619 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
7620 | goto out; | |
7621 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
7622 | goto free_domainspan; | |
7623 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
7624 | goto free_covered; | |
7625 | #endif | |
7626 | ||
7627 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
7628 | goto free_notcovered; | |
7629 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
7630 | goto free_nodemask; | |
7631 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
7632 | goto free_this_sibling_map; | |
7633 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
7634 | goto free_this_core_map; | |
7635 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
7636 | goto free_send_covered; | |
7637 | ||
7638 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
7639 | /* |
7640 | * Allocate the per-node list of sched groups | |
7641 | */ | |
076ac2af | 7642 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7643 | GFP_KERNEL); |
d1b55138 JH |
7644 | if (!sched_group_nodes) { |
7645 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 7646 | goto free_tmpmask; |
d1b55138 | 7647 | } |
d1b55138 | 7648 | #endif |
1da177e4 | 7649 | |
dc938520 | 7650 | rd = alloc_rootdomain(); |
57d885fe GH |
7651 | if (!rd) { |
7652 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 7653 | goto free_sched_groups; |
57d885fe GH |
7654 | } |
7655 | ||
7c16ec58 | 7656 | #ifdef CONFIG_NUMA |
96f874e2 | 7657 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
7658 | #endif |
7659 | ||
1da177e4 | 7660 | /* |
1a20ff27 | 7661 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7662 | */ |
abcd083a | 7663 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7664 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 7665 | |
6ca09dfc | 7666 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
7667 | |
7668 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
7669 | if (cpumask_weight(cpu_map) > |
7670 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 7671 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 7672 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7673 | set_domain_attribute(sd, attr); |
758b2cdc | 7674 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 7675 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7676 | p = sd; |
6711cab4 | 7677 | sd_allnodes = 1; |
9c1cfda2 JH |
7678 | } else |
7679 | p = NULL; | |
7680 | ||
62ea9ceb | 7681 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 7682 | SD_INIT(sd, NODE); |
1d3504fc | 7683 | set_domain_attribute(sd, attr); |
758b2cdc | 7684 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 7685 | sd->parent = p; |
1a848870 SS |
7686 | if (p) |
7687 | p->child = sd; | |
758b2cdc RR |
7688 | cpumask_and(sched_domain_span(sd), |
7689 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
7690 | #endif |
7691 | ||
7692 | p = sd; | |
6c99e9ad | 7693 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 7694 | SD_INIT(sd, CPU); |
1d3504fc | 7695 | set_domain_attribute(sd, attr); |
758b2cdc | 7696 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 7697 | sd->parent = p; |
1a848870 SS |
7698 | if (p) |
7699 | p->child = sd; | |
7c16ec58 | 7700 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7701 | |
1e9f28fa SS |
7702 | #ifdef CONFIG_SCHED_MC |
7703 | p = sd; | |
6c99e9ad | 7704 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 7705 | SD_INIT(sd, MC); |
1d3504fc | 7706 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
7707 | cpumask_and(sched_domain_span(sd), cpu_map, |
7708 | cpu_coregroup_mask(i)); | |
1e9f28fa | 7709 | sd->parent = p; |
1a848870 | 7710 | p->child = sd; |
7c16ec58 | 7711 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7712 | #endif |
7713 | ||
1da177e4 LT |
7714 | #ifdef CONFIG_SCHED_SMT |
7715 | p = sd; | |
6c99e9ad | 7716 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 7717 | SD_INIT(sd, SIBLING); |
1d3504fc | 7718 | set_domain_attribute(sd, attr); |
758b2cdc RR |
7719 | cpumask_and(sched_domain_span(sd), |
7720 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
1da177e4 | 7721 | sd->parent = p; |
1a848870 | 7722 | p->child = sd; |
7c16ec58 | 7723 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7724 | #endif |
7725 | } | |
7726 | ||
7727 | #ifdef CONFIG_SCHED_SMT | |
7728 | /* Set up CPU (sibling) groups */ | |
abcd083a | 7729 | for_each_cpu(i, cpu_map) { |
96f874e2 RR |
7730 | cpumask_and(this_sibling_map, |
7731 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
7732 | if (i != cpumask_first(this_sibling_map)) | |
1da177e4 LT |
7733 | continue; |
7734 | ||
dd41f596 | 7735 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7736 | &cpu_to_cpu_group, |
7737 | send_covered, tmpmask); | |
1da177e4 LT |
7738 | } |
7739 | #endif | |
7740 | ||
1e9f28fa SS |
7741 | #ifdef CONFIG_SCHED_MC |
7742 | /* Set up multi-core groups */ | |
abcd083a | 7743 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 7744 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 7745 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 7746 | continue; |
7c16ec58 | 7747 | |
dd41f596 | 7748 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7749 | &cpu_to_core_group, |
7750 | send_covered, tmpmask); | |
1e9f28fa SS |
7751 | } |
7752 | #endif | |
7753 | ||
1da177e4 | 7754 | /* Set up physical groups */ |
076ac2af | 7755 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 7756 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7757 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
7758 | continue; |
7759 | ||
7c16ec58 MT |
7760 | init_sched_build_groups(nodemask, cpu_map, |
7761 | &cpu_to_phys_group, | |
7762 | send_covered, tmpmask); | |
1da177e4 LT |
7763 | } |
7764 | ||
7765 | #ifdef CONFIG_NUMA | |
7766 | /* Set up node groups */ | |
7c16ec58 | 7767 | if (sd_allnodes) { |
7c16ec58 MT |
7768 | init_sched_build_groups(cpu_map, cpu_map, |
7769 | &cpu_to_allnodes_group, | |
7770 | send_covered, tmpmask); | |
7771 | } | |
9c1cfda2 | 7772 | |
076ac2af | 7773 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7774 | /* Set up node groups */ |
7775 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
7776 | int j; |
7777 | ||
96f874e2 | 7778 | cpumask_clear(covered); |
6ca09dfc | 7779 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7780 | if (cpumask_empty(nodemask)) { |
d1b55138 | 7781 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7782 | continue; |
d1b55138 | 7783 | } |
9c1cfda2 | 7784 | |
4bdbaad3 | 7785 | sched_domain_node_span(i, domainspan); |
96f874e2 | 7786 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 7787 | |
6c99e9ad RR |
7788 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
7789 | GFP_KERNEL, i); | |
51888ca2 SV |
7790 | if (!sg) { |
7791 | printk(KERN_WARNING "Can not alloc domain group for " | |
7792 | "node %d\n", i); | |
7793 | goto error; | |
7794 | } | |
9c1cfda2 | 7795 | sched_group_nodes[i] = sg; |
abcd083a | 7796 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 7797 | struct sched_domain *sd; |
9761eea8 | 7798 | |
62ea9ceb | 7799 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 7800 | sd->groups = sg; |
9c1cfda2 | 7801 | } |
5517d86b | 7802 | sg->__cpu_power = 0; |
758b2cdc | 7803 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 7804 | sg->next = sg; |
96f874e2 | 7805 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
7806 | prev = sg; |
7807 | ||
076ac2af | 7808 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 7809 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 7810 | |
96f874e2 RR |
7811 | cpumask_complement(notcovered, covered); |
7812 | cpumask_and(tmpmask, notcovered, cpu_map); | |
7813 | cpumask_and(tmpmask, tmpmask, domainspan); | |
7814 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
7815 | break; |
7816 | ||
6ca09dfc | 7817 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 7818 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
7819 | continue; |
7820 | ||
6c99e9ad RR |
7821 | sg = kmalloc_node(sizeof(struct sched_group) + |
7822 | cpumask_size(), | |
15f0b676 | 7823 | GFP_KERNEL, i); |
9c1cfda2 JH |
7824 | if (!sg) { |
7825 | printk(KERN_WARNING | |
7826 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7827 | goto error; |
9c1cfda2 | 7828 | } |
5517d86b | 7829 | sg->__cpu_power = 0; |
758b2cdc | 7830 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 7831 | sg->next = prev->next; |
96f874e2 | 7832 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
7833 | prev->next = sg; |
7834 | prev = sg; | |
7835 | } | |
9c1cfda2 | 7836 | } |
1da177e4 LT |
7837 | #endif |
7838 | ||
7839 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7840 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7841 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7842 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 7843 | |
89c4710e | 7844 | init_sched_groups_power(i, sd); |
5c45bf27 | 7845 | } |
1da177e4 | 7846 | #endif |
1e9f28fa | 7847 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7848 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7849 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 7850 | |
89c4710e | 7851 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7852 | } |
7853 | #endif | |
1e9f28fa | 7854 | |
abcd083a | 7855 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7856 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 7857 | |
89c4710e | 7858 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7859 | } |
7860 | ||
9c1cfda2 | 7861 | #ifdef CONFIG_NUMA |
076ac2af | 7862 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7863 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7864 | |
6711cab4 SS |
7865 | if (sd_allnodes) { |
7866 | struct sched_group *sg; | |
f712c0c7 | 7867 | |
96f874e2 | 7868 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 7869 | tmpmask); |
f712c0c7 SS |
7870 | init_numa_sched_groups_power(sg); |
7871 | } | |
9c1cfda2 JH |
7872 | #endif |
7873 | ||
1da177e4 | 7874 | /* Attach the domains */ |
abcd083a | 7875 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
7876 | struct sched_domain *sd; |
7877 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 7878 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7879 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7880 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7881 | #else |
6c99e9ad | 7882 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7883 | #endif |
57d885fe | 7884 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7885 | } |
51888ca2 | 7886 | |
3404c8d9 RR |
7887 | err = 0; |
7888 | ||
7889 | free_tmpmask: | |
7890 | free_cpumask_var(tmpmask); | |
7891 | free_send_covered: | |
7892 | free_cpumask_var(send_covered); | |
7893 | free_this_core_map: | |
7894 | free_cpumask_var(this_core_map); | |
7895 | free_this_sibling_map: | |
7896 | free_cpumask_var(this_sibling_map); | |
7897 | free_nodemask: | |
7898 | free_cpumask_var(nodemask); | |
7899 | free_notcovered: | |
7900 | #ifdef CONFIG_NUMA | |
7901 | free_cpumask_var(notcovered); | |
7902 | free_covered: | |
7903 | free_cpumask_var(covered); | |
7904 | free_domainspan: | |
7905 | free_cpumask_var(domainspan); | |
7906 | out: | |
7907 | #endif | |
7908 | return err; | |
7909 | ||
7910 | free_sched_groups: | |
7911 | #ifdef CONFIG_NUMA | |
7912 | kfree(sched_group_nodes); | |
7913 | #endif | |
7914 | goto free_tmpmask; | |
51888ca2 | 7915 | |
a616058b | 7916 | #ifdef CONFIG_NUMA |
51888ca2 | 7917 | error: |
7c16ec58 | 7918 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 7919 | free_rootdomain(rd); |
3404c8d9 | 7920 | goto free_tmpmask; |
a616058b | 7921 | #endif |
1da177e4 | 7922 | } |
029190c5 | 7923 | |
96f874e2 | 7924 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7925 | { |
7926 | return __build_sched_domains(cpu_map, NULL); | |
7927 | } | |
7928 | ||
96f874e2 | 7929 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 7930 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7931 | static struct sched_domain_attr *dattr_cur; |
7932 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7933 | |
7934 | /* | |
7935 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7936 | * cpumask) fails, then fallback to a single sched domain, |
7937 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7938 | */ |
4212823f | 7939 | static cpumask_var_t fallback_doms; |
029190c5 | 7940 | |
ee79d1bd HC |
7941 | /* |
7942 | * arch_update_cpu_topology lets virtualized architectures update the | |
7943 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7944 | * or 0 if it stayed the same. | |
7945 | */ | |
7946 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7947 | { |
ee79d1bd | 7948 | return 0; |
22e52b07 HC |
7949 | } |
7950 | ||
1a20ff27 | 7951 | /* |
41a2d6cf | 7952 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7953 | * For now this just excludes isolated cpus, but could be used to |
7954 | * exclude other special cases in the future. | |
1a20ff27 | 7955 | */ |
96f874e2 | 7956 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7957 | { |
7378547f MM |
7958 | int err; |
7959 | ||
22e52b07 | 7960 | arch_update_cpu_topology(); |
029190c5 | 7961 | ndoms_cur = 1; |
96f874e2 | 7962 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 7963 | if (!doms_cur) |
4212823f | 7964 | doms_cur = fallback_doms; |
dcc30a35 | 7965 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 7966 | dattr_cur = NULL; |
7378547f | 7967 | err = build_sched_domains(doms_cur); |
6382bc90 | 7968 | register_sched_domain_sysctl(); |
7378547f MM |
7969 | |
7970 | return err; | |
1a20ff27 DG |
7971 | } |
7972 | ||
96f874e2 RR |
7973 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7974 | struct cpumask *tmpmask) | |
1da177e4 | 7975 | { |
7c16ec58 | 7976 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7977 | } |
1da177e4 | 7978 | |
1a20ff27 DG |
7979 | /* |
7980 | * Detach sched domains from a group of cpus specified in cpu_map | |
7981 | * These cpus will now be attached to the NULL domain | |
7982 | */ | |
96f874e2 | 7983 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7984 | { |
96f874e2 RR |
7985 | /* Save because hotplug lock held. */ |
7986 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7987 | int i; |
7988 | ||
abcd083a | 7989 | for_each_cpu(i, cpu_map) |
57d885fe | 7990 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7991 | synchronize_sched(); |
96f874e2 | 7992 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7993 | } |
7994 | ||
1d3504fc HS |
7995 | /* handle null as "default" */ |
7996 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7997 | struct sched_domain_attr *new, int idx_new) | |
7998 | { | |
7999 | struct sched_domain_attr tmp; | |
8000 | ||
8001 | /* fast path */ | |
8002 | if (!new && !cur) | |
8003 | return 1; | |
8004 | ||
8005 | tmp = SD_ATTR_INIT; | |
8006 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8007 | new ? (new + idx_new) : &tmp, | |
8008 | sizeof(struct sched_domain_attr)); | |
8009 | } | |
8010 | ||
029190c5 PJ |
8011 | /* |
8012 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8013 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8014 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8015 | * It destroys each deleted domain and builds each new domain. | |
8016 | * | |
96f874e2 | 8017 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8018 | * The masks don't intersect (don't overlap.) We should setup one |
8019 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8020 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8021 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8022 | * it as it is. | |
8023 | * | |
41a2d6cf IM |
8024 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8025 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8026 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8027 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8028 | * the single partition 'fallback_doms', it also forces the domains | |
8029 | * to be rebuilt. | |
029190c5 | 8030 | * |
96f874e2 | 8031 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8032 | * ndoms_new == 0 is a special case for destroying existing domains, |
8033 | * and it will not create the default domain. | |
dfb512ec | 8034 | * |
029190c5 PJ |
8035 | * Call with hotplug lock held |
8036 | */ | |
96f874e2 RR |
8037 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8038 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8039 | struct sched_domain_attr *dattr_new) |
029190c5 | 8040 | { |
dfb512ec | 8041 | int i, j, n; |
d65bd5ec | 8042 | int new_topology; |
029190c5 | 8043 | |
712555ee | 8044 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8045 | |
7378547f MM |
8046 | /* always unregister in case we don't destroy any domains */ |
8047 | unregister_sched_domain_sysctl(); | |
8048 | ||
d65bd5ec HC |
8049 | /* Let architecture update cpu core mappings. */ |
8050 | new_topology = arch_update_cpu_topology(); | |
8051 | ||
dfb512ec | 8052 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8053 | |
8054 | /* Destroy deleted domains */ | |
8055 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8056 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8057 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8058 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8059 | goto match1; |
8060 | } | |
8061 | /* no match - a current sched domain not in new doms_new[] */ | |
8062 | detach_destroy_domains(doms_cur + i); | |
8063 | match1: | |
8064 | ; | |
8065 | } | |
8066 | ||
e761b772 MK |
8067 | if (doms_new == NULL) { |
8068 | ndoms_cur = 0; | |
4212823f | 8069 | doms_new = fallback_doms; |
dcc30a35 | 8070 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8071 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8072 | } |
8073 | ||
029190c5 PJ |
8074 | /* Build new domains */ |
8075 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8076 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8077 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8078 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8079 | goto match2; |
8080 | } | |
8081 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8082 | __build_sched_domains(doms_new + i, |
8083 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8084 | match2: |
8085 | ; | |
8086 | } | |
8087 | ||
8088 | /* Remember the new sched domains */ | |
4212823f | 8089 | if (doms_cur != fallback_doms) |
029190c5 | 8090 | kfree(doms_cur); |
1d3504fc | 8091 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8092 | doms_cur = doms_new; |
1d3504fc | 8093 | dattr_cur = dattr_new; |
029190c5 | 8094 | ndoms_cur = ndoms_new; |
7378547f MM |
8095 | |
8096 | register_sched_domain_sysctl(); | |
a1835615 | 8097 | |
712555ee | 8098 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8099 | } |
8100 | ||
5c45bf27 | 8101 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8102 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8103 | { |
95402b38 | 8104 | get_online_cpus(); |
dfb512ec MK |
8105 | |
8106 | /* Destroy domains first to force the rebuild */ | |
8107 | partition_sched_domains(0, NULL, NULL); | |
8108 | ||
e761b772 | 8109 | rebuild_sched_domains(); |
95402b38 | 8110 | put_online_cpus(); |
5c45bf27 SS |
8111 | } |
8112 | ||
8113 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8114 | { | |
afb8a9b7 | 8115 | unsigned int level = 0; |
5c45bf27 | 8116 | |
afb8a9b7 GS |
8117 | if (sscanf(buf, "%u", &level) != 1) |
8118 | return -EINVAL; | |
8119 | ||
8120 | /* | |
8121 | * level is always be positive so don't check for | |
8122 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8123 | * What happens on 0 or 1 byte write, | |
8124 | * need to check for count as well? | |
8125 | */ | |
8126 | ||
8127 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8128 | return -EINVAL; |
8129 | ||
8130 | if (smt) | |
afb8a9b7 | 8131 | sched_smt_power_savings = level; |
5c45bf27 | 8132 | else |
afb8a9b7 | 8133 | sched_mc_power_savings = level; |
5c45bf27 | 8134 | |
c70f22d2 | 8135 | arch_reinit_sched_domains(); |
5c45bf27 | 8136 | |
c70f22d2 | 8137 | return count; |
5c45bf27 SS |
8138 | } |
8139 | ||
5c45bf27 | 8140 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8141 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8142 | char *page) | |
5c45bf27 SS |
8143 | { |
8144 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8145 | } | |
f718cd4a | 8146 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8147 | const char *buf, size_t count) |
5c45bf27 SS |
8148 | { |
8149 | return sched_power_savings_store(buf, count, 0); | |
8150 | } | |
f718cd4a AK |
8151 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8152 | sched_mc_power_savings_show, | |
8153 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8154 | #endif |
8155 | ||
8156 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8157 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8158 | char *page) | |
5c45bf27 SS |
8159 | { |
8160 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8161 | } | |
f718cd4a | 8162 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8163 | const char *buf, size_t count) |
5c45bf27 SS |
8164 | { |
8165 | return sched_power_savings_store(buf, count, 1); | |
8166 | } | |
f718cd4a AK |
8167 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8168 | sched_smt_power_savings_show, | |
6707de00 AB |
8169 | sched_smt_power_savings_store); |
8170 | #endif | |
8171 | ||
39aac648 | 8172 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8173 | { |
8174 | int err = 0; | |
8175 | ||
8176 | #ifdef CONFIG_SCHED_SMT | |
8177 | if (smt_capable()) | |
8178 | err = sysfs_create_file(&cls->kset.kobj, | |
8179 | &attr_sched_smt_power_savings.attr); | |
8180 | #endif | |
8181 | #ifdef CONFIG_SCHED_MC | |
8182 | if (!err && mc_capable()) | |
8183 | err = sysfs_create_file(&cls->kset.kobj, | |
8184 | &attr_sched_mc_power_savings.attr); | |
8185 | #endif | |
8186 | return err; | |
8187 | } | |
6d6bc0ad | 8188 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8189 | |
e761b772 | 8190 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8191 | /* |
e761b772 MK |
8192 | * Add online and remove offline CPUs from the scheduler domains. |
8193 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8194 | */ |
8195 | static int update_sched_domains(struct notifier_block *nfb, | |
8196 | unsigned long action, void *hcpu) | |
e761b772 MK |
8197 | { |
8198 | switch (action) { | |
8199 | case CPU_ONLINE: | |
8200 | case CPU_ONLINE_FROZEN: | |
8201 | case CPU_DEAD: | |
8202 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8203 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8204 | return NOTIFY_OK; |
8205 | ||
8206 | default: | |
8207 | return NOTIFY_DONE; | |
8208 | } | |
8209 | } | |
8210 | #endif | |
8211 | ||
8212 | static int update_runtime(struct notifier_block *nfb, | |
8213 | unsigned long action, void *hcpu) | |
1da177e4 | 8214 | { |
7def2be1 PZ |
8215 | int cpu = (int)(long)hcpu; |
8216 | ||
1da177e4 | 8217 | switch (action) { |
1da177e4 | 8218 | case CPU_DOWN_PREPARE: |
8bb78442 | 8219 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8220 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8221 | return NOTIFY_OK; |
8222 | ||
1da177e4 | 8223 | case CPU_DOWN_FAILED: |
8bb78442 | 8224 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8225 | case CPU_ONLINE: |
8bb78442 | 8226 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8227 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8228 | return NOTIFY_OK; |
8229 | ||
1da177e4 LT |
8230 | default: |
8231 | return NOTIFY_DONE; | |
8232 | } | |
1da177e4 | 8233 | } |
1da177e4 LT |
8234 | |
8235 | void __init sched_init_smp(void) | |
8236 | { | |
dcc30a35 RR |
8237 | cpumask_var_t non_isolated_cpus; |
8238 | ||
8239 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8240 | |
434d53b0 MT |
8241 | #if defined(CONFIG_NUMA) |
8242 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8243 | GFP_KERNEL); | |
8244 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8245 | #endif | |
95402b38 | 8246 | get_online_cpus(); |
712555ee | 8247 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8248 | arch_init_sched_domains(cpu_online_mask); |
8249 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8250 | if (cpumask_empty(non_isolated_cpus)) | |
8251 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8252 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8253 | put_online_cpus(); |
e761b772 MK |
8254 | |
8255 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8256 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8257 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8258 | #endif |
8259 | ||
8260 | /* RT runtime code needs to handle some hotplug events */ | |
8261 | hotcpu_notifier(update_runtime, 0); | |
8262 | ||
b328ca18 | 8263 | init_hrtick(); |
5c1e1767 NP |
8264 | |
8265 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8266 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8267 | BUG(); |
19978ca6 | 8268 | sched_init_granularity(); |
dcc30a35 | 8269 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8270 | |
8271 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8272 | init_sched_rt_class(); |
1da177e4 LT |
8273 | } |
8274 | #else | |
8275 | void __init sched_init_smp(void) | |
8276 | { | |
19978ca6 | 8277 | sched_init_granularity(); |
1da177e4 LT |
8278 | } |
8279 | #endif /* CONFIG_SMP */ | |
8280 | ||
8281 | int in_sched_functions(unsigned long addr) | |
8282 | { | |
1da177e4 LT |
8283 | return in_lock_functions(addr) || |
8284 | (addr >= (unsigned long)__sched_text_start | |
8285 | && addr < (unsigned long)__sched_text_end); | |
8286 | } | |
8287 | ||
a9957449 | 8288 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8289 | { |
8290 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8291 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8292 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8293 | cfs_rq->rq = rq; | |
8294 | #endif | |
67e9fb2a | 8295 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8296 | } |
8297 | ||
fa85ae24 PZ |
8298 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8299 | { | |
8300 | struct rt_prio_array *array; | |
8301 | int i; | |
8302 | ||
8303 | array = &rt_rq->active; | |
8304 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8305 | INIT_LIST_HEAD(array->queue + i); | |
8306 | __clear_bit(i, array->bitmap); | |
8307 | } | |
8308 | /* delimiter for bitsearch: */ | |
8309 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8310 | ||
052f1dc7 | 8311 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8312 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8313 | #ifdef CONFIG_SMP |
e864c499 | 8314 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8315 | #endif |
48d5e258 | 8316 | #endif |
fa85ae24 PZ |
8317 | #ifdef CONFIG_SMP |
8318 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8319 | rt_rq->overloaded = 0; |
917b627d | 8320 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8321 | #endif |
8322 | ||
8323 | rt_rq->rt_time = 0; | |
8324 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8325 | rt_rq->rt_runtime = 0; |
8326 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8327 | |
052f1dc7 | 8328 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8329 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8330 | rt_rq->rq = rq; |
8331 | #endif | |
fa85ae24 PZ |
8332 | } |
8333 | ||
6f505b16 | 8334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8335 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8336 | struct sched_entity *se, int cpu, int add, | |
8337 | struct sched_entity *parent) | |
6f505b16 | 8338 | { |
ec7dc8ac | 8339 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8340 | tg->cfs_rq[cpu] = cfs_rq; |
8341 | init_cfs_rq(cfs_rq, rq); | |
8342 | cfs_rq->tg = tg; | |
8343 | if (add) | |
8344 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8345 | ||
8346 | tg->se[cpu] = se; | |
354d60c2 DG |
8347 | /* se could be NULL for init_task_group */ |
8348 | if (!se) | |
8349 | return; | |
8350 | ||
ec7dc8ac DG |
8351 | if (!parent) |
8352 | se->cfs_rq = &rq->cfs; | |
8353 | else | |
8354 | se->cfs_rq = parent->my_q; | |
8355 | ||
6f505b16 PZ |
8356 | se->my_q = cfs_rq; |
8357 | se->load.weight = tg->shares; | |
e05510d0 | 8358 | se->load.inv_weight = 0; |
ec7dc8ac | 8359 | se->parent = parent; |
6f505b16 | 8360 | } |
052f1dc7 | 8361 | #endif |
6f505b16 | 8362 | |
052f1dc7 | 8363 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8364 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8365 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8366 | struct sched_rt_entity *parent) | |
6f505b16 | 8367 | { |
ec7dc8ac DG |
8368 | struct rq *rq = cpu_rq(cpu); |
8369 | ||
6f505b16 PZ |
8370 | tg->rt_rq[cpu] = rt_rq; |
8371 | init_rt_rq(rt_rq, rq); | |
8372 | rt_rq->tg = tg; | |
8373 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8374 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8375 | if (add) |
8376 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8377 | ||
8378 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8379 | if (!rt_se) |
8380 | return; | |
8381 | ||
ec7dc8ac DG |
8382 | if (!parent) |
8383 | rt_se->rt_rq = &rq->rt; | |
8384 | else | |
8385 | rt_se->rt_rq = parent->my_q; | |
8386 | ||
6f505b16 | 8387 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8388 | rt_se->parent = parent; |
6f505b16 PZ |
8389 | INIT_LIST_HEAD(&rt_se->run_list); |
8390 | } | |
8391 | #endif | |
8392 | ||
1da177e4 LT |
8393 | void __init sched_init(void) |
8394 | { | |
dd41f596 | 8395 | int i, j; |
434d53b0 MT |
8396 | unsigned long alloc_size = 0, ptr; |
8397 | ||
8398 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8399 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8400 | #endif | |
8401 | #ifdef CONFIG_RT_GROUP_SCHED | |
8402 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8403 | #endif |
8404 | #ifdef CONFIG_USER_SCHED | |
8405 | alloc_size *= 2; | |
434d53b0 MT |
8406 | #endif |
8407 | /* | |
8408 | * As sched_init() is called before page_alloc is setup, | |
8409 | * we use alloc_bootmem(). | |
8410 | */ | |
8411 | if (alloc_size) { | |
5a9d3225 | 8412 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8413 | |
8414 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8415 | init_task_group.se = (struct sched_entity **)ptr; | |
8416 | ptr += nr_cpu_ids * sizeof(void **); | |
8417 | ||
8418 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8419 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8420 | |
8421 | #ifdef CONFIG_USER_SCHED | |
8422 | root_task_group.se = (struct sched_entity **)ptr; | |
8423 | ptr += nr_cpu_ids * sizeof(void **); | |
8424 | ||
8425 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8426 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8427 | #endif /* CONFIG_USER_SCHED */ |
8428 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8429 | #ifdef CONFIG_RT_GROUP_SCHED |
8430 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8431 | ptr += nr_cpu_ids * sizeof(void **); | |
8432 | ||
8433 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8434 | ptr += nr_cpu_ids * sizeof(void **); |
8435 | ||
8436 | #ifdef CONFIG_USER_SCHED | |
8437 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8438 | ptr += nr_cpu_ids * sizeof(void **); | |
8439 | ||
8440 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8441 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8442 | #endif /* CONFIG_USER_SCHED */ |
8443 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8444 | } |
dd41f596 | 8445 | |
57d885fe GH |
8446 | #ifdef CONFIG_SMP |
8447 | init_defrootdomain(); | |
8448 | #endif | |
8449 | ||
d0b27fa7 PZ |
8450 | init_rt_bandwidth(&def_rt_bandwidth, |
8451 | global_rt_period(), global_rt_runtime()); | |
8452 | ||
8453 | #ifdef CONFIG_RT_GROUP_SCHED | |
8454 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8455 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8456 | #ifdef CONFIG_USER_SCHED |
8457 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8458 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8459 | #endif /* CONFIG_USER_SCHED */ |
8460 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8461 | |
052f1dc7 | 8462 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8463 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8464 | INIT_LIST_HEAD(&init_task_group.children); |
8465 | ||
8466 | #ifdef CONFIG_USER_SCHED | |
8467 | INIT_LIST_HEAD(&root_task_group.children); | |
8468 | init_task_group.parent = &root_task_group; | |
8469 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8470 | #endif /* CONFIG_USER_SCHED */ |
8471 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8472 | |
0a945022 | 8473 | for_each_possible_cpu(i) { |
70b97a7f | 8474 | struct rq *rq; |
1da177e4 LT |
8475 | |
8476 | rq = cpu_rq(i); | |
8477 | spin_lock_init(&rq->lock); | |
7897986b | 8478 | rq->nr_running = 0; |
dd41f596 | 8479 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8480 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8481 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8482 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8483 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8484 | #ifdef CONFIG_CGROUP_SCHED |
8485 | /* | |
8486 | * How much cpu bandwidth does init_task_group get? | |
8487 | * | |
8488 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8489 | * gets 100% of the cpu resources in the system. This overall | |
8490 | * system cpu resource is divided among the tasks of | |
8491 | * init_task_group and its child task-groups in a fair manner, | |
8492 | * based on each entity's (task or task-group's) weight | |
8493 | * (se->load.weight). | |
8494 | * | |
8495 | * In other words, if init_task_group has 10 tasks of weight | |
8496 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8497 | * then A0's share of the cpu resource is: | |
8498 | * | |
8499 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8500 | * | |
8501 | * We achieve this by letting init_task_group's tasks sit | |
8502 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8503 | */ | |
ec7dc8ac | 8504 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8505 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8506 | root_task_group.shares = NICE_0_LOAD; |
8507 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8508 | /* |
8509 | * In case of task-groups formed thr' the user id of tasks, | |
8510 | * init_task_group represents tasks belonging to root user. | |
8511 | * Hence it forms a sibling of all subsequent groups formed. | |
8512 | * In this case, init_task_group gets only a fraction of overall | |
8513 | * system cpu resource, based on the weight assigned to root | |
8514 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8515 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8516 | * (init_cfs_rq) and having one entity represent this group of | |
8517 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8518 | */ | |
ec7dc8ac | 8519 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8520 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8521 | &per_cpu(init_sched_entity, i), i, 1, |
8522 | root_task_group.se[i]); | |
6f505b16 | 8523 | |
052f1dc7 | 8524 | #endif |
354d60c2 DG |
8525 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8526 | ||
8527 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8528 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8529 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8530 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8531 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8532 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8533 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8534 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8535 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8536 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8537 | root_task_group.rt_se[i]); | |
354d60c2 | 8538 | #endif |
dd41f596 | 8539 | #endif |
1da177e4 | 8540 | |
dd41f596 IM |
8541 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8542 | rq->cpu_load[j] = 0; | |
1da177e4 | 8543 | #ifdef CONFIG_SMP |
41c7ce9a | 8544 | rq->sd = NULL; |
57d885fe | 8545 | rq->rd = NULL; |
1da177e4 | 8546 | rq->active_balance = 0; |
dd41f596 | 8547 | rq->next_balance = jiffies; |
1da177e4 | 8548 | rq->push_cpu = 0; |
0a2966b4 | 8549 | rq->cpu = i; |
1f11eb6a | 8550 | rq->online = 0; |
1da177e4 LT |
8551 | rq->migration_thread = NULL; |
8552 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8553 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8554 | #endif |
8f4d37ec | 8555 | init_rq_hrtick(rq); |
1da177e4 | 8556 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8557 | } |
8558 | ||
2dd73a4f | 8559 | set_load_weight(&init_task); |
b50f60ce | 8560 | |
e107be36 AK |
8561 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8562 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8563 | #endif | |
8564 | ||
c9819f45 | 8565 | #ifdef CONFIG_SMP |
962cf36c | 8566 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8567 | #endif |
8568 | ||
b50f60ce HC |
8569 | #ifdef CONFIG_RT_MUTEXES |
8570 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8571 | #endif | |
8572 | ||
1da177e4 LT |
8573 | /* |
8574 | * The boot idle thread does lazy MMU switching as well: | |
8575 | */ | |
8576 | atomic_inc(&init_mm.mm_count); | |
8577 | enter_lazy_tlb(&init_mm, current); | |
8578 | ||
8579 | /* | |
8580 | * Make us the idle thread. Technically, schedule() should not be | |
8581 | * called from this thread, however somewhere below it might be, | |
8582 | * but because we are the idle thread, we just pick up running again | |
8583 | * when this runqueue becomes "idle". | |
8584 | */ | |
8585 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8586 | /* |
8587 | * During early bootup we pretend to be a normal task: | |
8588 | */ | |
8589 | current->sched_class = &fair_sched_class; | |
6892b75e | 8590 | |
6a7b3dc3 RR |
8591 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
8592 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 8593 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
8594 | #ifdef CONFIG_NO_HZ |
8595 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
8596 | #endif | |
dcc30a35 | 8597 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 8598 | #endif /* SMP */ |
6a7b3dc3 | 8599 | |
6892b75e | 8600 | scheduler_running = 1; |
1da177e4 LT |
8601 | } |
8602 | ||
8603 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8604 | void __might_sleep(char *file, int line) | |
8605 | { | |
48f24c4d | 8606 | #ifdef in_atomic |
1da177e4 LT |
8607 | static unsigned long prev_jiffy; /* ratelimiting */ |
8608 | ||
aef745fc IM |
8609 | if ((!in_atomic() && !irqs_disabled()) || |
8610 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8611 | return; | |
8612 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8613 | return; | |
8614 | prev_jiffy = jiffies; | |
8615 | ||
8616 | printk(KERN_ERR | |
8617 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8618 | file, line); | |
8619 | printk(KERN_ERR | |
8620 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8621 | in_atomic(), irqs_disabled(), | |
8622 | current->pid, current->comm); | |
8623 | ||
8624 | debug_show_held_locks(current); | |
8625 | if (irqs_disabled()) | |
8626 | print_irqtrace_events(current); | |
8627 | dump_stack(); | |
1da177e4 LT |
8628 | #endif |
8629 | } | |
8630 | EXPORT_SYMBOL(__might_sleep); | |
8631 | #endif | |
8632 | ||
8633 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8634 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8635 | { | |
8636 | int on_rq; | |
3e51f33f | 8637 | |
3a5e4dc1 AK |
8638 | update_rq_clock(rq); |
8639 | on_rq = p->se.on_rq; | |
8640 | if (on_rq) | |
8641 | deactivate_task(rq, p, 0); | |
8642 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8643 | if (on_rq) { | |
8644 | activate_task(rq, p, 0); | |
8645 | resched_task(rq->curr); | |
8646 | } | |
8647 | } | |
8648 | ||
1da177e4 LT |
8649 | void normalize_rt_tasks(void) |
8650 | { | |
a0f98a1c | 8651 | struct task_struct *g, *p; |
1da177e4 | 8652 | unsigned long flags; |
70b97a7f | 8653 | struct rq *rq; |
1da177e4 | 8654 | |
4cf5d77a | 8655 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8656 | do_each_thread(g, p) { |
178be793 IM |
8657 | /* |
8658 | * Only normalize user tasks: | |
8659 | */ | |
8660 | if (!p->mm) | |
8661 | continue; | |
8662 | ||
6cfb0d5d | 8663 | p->se.exec_start = 0; |
6cfb0d5d | 8664 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8665 | p->se.wait_start = 0; |
dd41f596 | 8666 | p->se.sleep_start = 0; |
dd41f596 | 8667 | p->se.block_start = 0; |
6cfb0d5d | 8668 | #endif |
dd41f596 IM |
8669 | |
8670 | if (!rt_task(p)) { | |
8671 | /* | |
8672 | * Renice negative nice level userspace | |
8673 | * tasks back to 0: | |
8674 | */ | |
8675 | if (TASK_NICE(p) < 0 && p->mm) | |
8676 | set_user_nice(p, 0); | |
1da177e4 | 8677 | continue; |
dd41f596 | 8678 | } |
1da177e4 | 8679 | |
4cf5d77a | 8680 | spin_lock(&p->pi_lock); |
b29739f9 | 8681 | rq = __task_rq_lock(p); |
1da177e4 | 8682 | |
178be793 | 8683 | normalize_task(rq, p); |
3a5e4dc1 | 8684 | |
b29739f9 | 8685 | __task_rq_unlock(rq); |
4cf5d77a | 8686 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8687 | } while_each_thread(g, p); |
8688 | ||
4cf5d77a | 8689 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8690 | } |
8691 | ||
8692 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8693 | |
8694 | #ifdef CONFIG_IA64 | |
8695 | /* | |
8696 | * These functions are only useful for the IA64 MCA handling. | |
8697 | * | |
8698 | * They can only be called when the whole system has been | |
8699 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8700 | * activity can take place. Using them for anything else would | |
8701 | * be a serious bug, and as a result, they aren't even visible | |
8702 | * under any other configuration. | |
8703 | */ | |
8704 | ||
8705 | /** | |
8706 | * curr_task - return the current task for a given cpu. | |
8707 | * @cpu: the processor in question. | |
8708 | * | |
8709 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8710 | */ | |
36c8b586 | 8711 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8712 | { |
8713 | return cpu_curr(cpu); | |
8714 | } | |
8715 | ||
8716 | /** | |
8717 | * set_curr_task - set the current task for a given cpu. | |
8718 | * @cpu: the processor in question. | |
8719 | * @p: the task pointer to set. | |
8720 | * | |
8721 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8722 | * are serviced on a separate stack. It allows the architecture to switch the |
8723 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8724 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8725 | * and caller must save the original value of the current task (see | |
8726 | * curr_task() above) and restore that value before reenabling interrupts and | |
8727 | * re-starting the system. | |
8728 | * | |
8729 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8730 | */ | |
36c8b586 | 8731 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8732 | { |
8733 | cpu_curr(cpu) = p; | |
8734 | } | |
8735 | ||
8736 | #endif | |
29f59db3 | 8737 | |
bccbe08a PZ |
8738 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8739 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8740 | { |
8741 | int i; | |
8742 | ||
8743 | for_each_possible_cpu(i) { | |
8744 | if (tg->cfs_rq) | |
8745 | kfree(tg->cfs_rq[i]); | |
8746 | if (tg->se) | |
8747 | kfree(tg->se[i]); | |
6f505b16 PZ |
8748 | } |
8749 | ||
8750 | kfree(tg->cfs_rq); | |
8751 | kfree(tg->se); | |
6f505b16 PZ |
8752 | } |
8753 | ||
ec7dc8ac DG |
8754 | static |
8755 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8756 | { |
29f59db3 | 8757 | struct cfs_rq *cfs_rq; |
eab17229 | 8758 | struct sched_entity *se; |
9b5b7751 | 8759 | struct rq *rq; |
29f59db3 SV |
8760 | int i; |
8761 | ||
434d53b0 | 8762 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8763 | if (!tg->cfs_rq) |
8764 | goto err; | |
434d53b0 | 8765 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8766 | if (!tg->se) |
8767 | goto err; | |
052f1dc7 PZ |
8768 | |
8769 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8770 | |
8771 | for_each_possible_cpu(i) { | |
9b5b7751 | 8772 | rq = cpu_rq(i); |
29f59db3 | 8773 | |
eab17229 LZ |
8774 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8775 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8776 | if (!cfs_rq) |
8777 | goto err; | |
8778 | ||
eab17229 LZ |
8779 | se = kzalloc_node(sizeof(struct sched_entity), |
8780 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8781 | if (!se) |
8782 | goto err; | |
8783 | ||
eab17229 | 8784 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8785 | } |
8786 | ||
8787 | return 1; | |
8788 | ||
8789 | err: | |
8790 | return 0; | |
8791 | } | |
8792 | ||
8793 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8794 | { | |
8795 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8796 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8797 | } | |
8798 | ||
8799 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8800 | { | |
8801 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8802 | } | |
6d6bc0ad | 8803 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8804 | static inline void free_fair_sched_group(struct task_group *tg) |
8805 | { | |
8806 | } | |
8807 | ||
ec7dc8ac DG |
8808 | static inline |
8809 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8810 | { |
8811 | return 1; | |
8812 | } | |
8813 | ||
8814 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8815 | { | |
8816 | } | |
8817 | ||
8818 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8819 | { | |
8820 | } | |
6d6bc0ad | 8821 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8822 | |
8823 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8824 | static void free_rt_sched_group(struct task_group *tg) |
8825 | { | |
8826 | int i; | |
8827 | ||
d0b27fa7 PZ |
8828 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8829 | ||
bccbe08a PZ |
8830 | for_each_possible_cpu(i) { |
8831 | if (tg->rt_rq) | |
8832 | kfree(tg->rt_rq[i]); | |
8833 | if (tg->rt_se) | |
8834 | kfree(tg->rt_se[i]); | |
8835 | } | |
8836 | ||
8837 | kfree(tg->rt_rq); | |
8838 | kfree(tg->rt_se); | |
8839 | } | |
8840 | ||
ec7dc8ac DG |
8841 | static |
8842 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8843 | { |
8844 | struct rt_rq *rt_rq; | |
eab17229 | 8845 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8846 | struct rq *rq; |
8847 | int i; | |
8848 | ||
434d53b0 | 8849 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8850 | if (!tg->rt_rq) |
8851 | goto err; | |
434d53b0 | 8852 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8853 | if (!tg->rt_se) |
8854 | goto err; | |
8855 | ||
d0b27fa7 PZ |
8856 | init_rt_bandwidth(&tg->rt_bandwidth, |
8857 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8858 | |
8859 | for_each_possible_cpu(i) { | |
8860 | rq = cpu_rq(i); | |
8861 | ||
eab17229 LZ |
8862 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8863 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8864 | if (!rt_rq) |
8865 | goto err; | |
29f59db3 | 8866 | |
eab17229 LZ |
8867 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8868 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8869 | if (!rt_se) |
8870 | goto err; | |
29f59db3 | 8871 | |
eab17229 | 8872 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8873 | } |
8874 | ||
bccbe08a PZ |
8875 | return 1; |
8876 | ||
8877 | err: | |
8878 | return 0; | |
8879 | } | |
8880 | ||
8881 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8882 | { | |
8883 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8884 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8885 | } | |
8886 | ||
8887 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8888 | { | |
8889 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8890 | } | |
6d6bc0ad | 8891 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8892 | static inline void free_rt_sched_group(struct task_group *tg) |
8893 | { | |
8894 | } | |
8895 | ||
ec7dc8ac DG |
8896 | static inline |
8897 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8898 | { |
8899 | return 1; | |
8900 | } | |
8901 | ||
8902 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8903 | { | |
8904 | } | |
8905 | ||
8906 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8907 | { | |
8908 | } | |
6d6bc0ad | 8909 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8910 | |
d0b27fa7 | 8911 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8912 | static void free_sched_group(struct task_group *tg) |
8913 | { | |
8914 | free_fair_sched_group(tg); | |
8915 | free_rt_sched_group(tg); | |
8916 | kfree(tg); | |
8917 | } | |
8918 | ||
8919 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8920 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8921 | { |
8922 | struct task_group *tg; | |
8923 | unsigned long flags; | |
8924 | int i; | |
8925 | ||
8926 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8927 | if (!tg) | |
8928 | return ERR_PTR(-ENOMEM); | |
8929 | ||
ec7dc8ac | 8930 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8931 | goto err; |
8932 | ||
ec7dc8ac | 8933 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8934 | goto err; |
8935 | ||
8ed36996 | 8936 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8937 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8938 | register_fair_sched_group(tg, i); |
8939 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8940 | } |
6f505b16 | 8941 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8942 | |
8943 | WARN_ON(!parent); /* root should already exist */ | |
8944 | ||
8945 | tg->parent = parent; | |
f473aa5e | 8946 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8947 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8948 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8949 | |
9b5b7751 | 8950 | return tg; |
29f59db3 SV |
8951 | |
8952 | err: | |
6f505b16 | 8953 | free_sched_group(tg); |
29f59db3 SV |
8954 | return ERR_PTR(-ENOMEM); |
8955 | } | |
8956 | ||
9b5b7751 | 8957 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8958 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8959 | { |
29f59db3 | 8960 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8961 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8962 | } |
8963 | ||
9b5b7751 | 8964 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8965 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8966 | { |
8ed36996 | 8967 | unsigned long flags; |
9b5b7751 | 8968 | int i; |
29f59db3 | 8969 | |
8ed36996 | 8970 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8971 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8972 | unregister_fair_sched_group(tg, i); |
8973 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8974 | } |
6f505b16 | 8975 | list_del_rcu(&tg->list); |
f473aa5e | 8976 | list_del_rcu(&tg->siblings); |
8ed36996 | 8977 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8978 | |
9b5b7751 | 8979 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8980 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8981 | } |
8982 | ||
9b5b7751 | 8983 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8984 | * The caller of this function should have put the task in its new group |
8985 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8986 | * reflect its new group. | |
9b5b7751 SV |
8987 | */ |
8988 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8989 | { |
8990 | int on_rq, running; | |
8991 | unsigned long flags; | |
8992 | struct rq *rq; | |
8993 | ||
8994 | rq = task_rq_lock(tsk, &flags); | |
8995 | ||
29f59db3 SV |
8996 | update_rq_clock(rq); |
8997 | ||
051a1d1a | 8998 | running = task_current(rq, tsk); |
29f59db3 SV |
8999 | on_rq = tsk->se.on_rq; |
9000 | ||
0e1f3483 | 9001 | if (on_rq) |
29f59db3 | 9002 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9003 | if (unlikely(running)) |
9004 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9005 | |
6f505b16 | 9006 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9007 | |
810b3817 PZ |
9008 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9009 | if (tsk->sched_class->moved_group) | |
9010 | tsk->sched_class->moved_group(tsk); | |
9011 | #endif | |
9012 | ||
0e1f3483 HS |
9013 | if (unlikely(running)) |
9014 | tsk->sched_class->set_curr_task(rq); | |
9015 | if (on_rq) | |
7074badb | 9016 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9017 | |
29f59db3 SV |
9018 | task_rq_unlock(rq, &flags); |
9019 | } | |
6d6bc0ad | 9020 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9021 | |
052f1dc7 | 9022 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9023 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9024 | { |
9025 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9026 | int on_rq; |
9027 | ||
29f59db3 | 9028 | on_rq = se->on_rq; |
62fb1851 | 9029 | if (on_rq) |
29f59db3 SV |
9030 | dequeue_entity(cfs_rq, se, 0); |
9031 | ||
9032 | se->load.weight = shares; | |
e05510d0 | 9033 | se->load.inv_weight = 0; |
29f59db3 | 9034 | |
62fb1851 | 9035 | if (on_rq) |
29f59db3 | 9036 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9037 | } |
62fb1851 | 9038 | |
c09595f6 PZ |
9039 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9040 | { | |
9041 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9042 | struct rq *rq = cfs_rq->rq; | |
9043 | unsigned long flags; | |
9044 | ||
9045 | spin_lock_irqsave(&rq->lock, flags); | |
9046 | __set_se_shares(se, shares); | |
9047 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9048 | } |
9049 | ||
8ed36996 PZ |
9050 | static DEFINE_MUTEX(shares_mutex); |
9051 | ||
4cf86d77 | 9052 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9053 | { |
9054 | int i; | |
8ed36996 | 9055 | unsigned long flags; |
c61935fd | 9056 | |
ec7dc8ac DG |
9057 | /* |
9058 | * We can't change the weight of the root cgroup. | |
9059 | */ | |
9060 | if (!tg->se[0]) | |
9061 | return -EINVAL; | |
9062 | ||
18d95a28 PZ |
9063 | if (shares < MIN_SHARES) |
9064 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9065 | else if (shares > MAX_SHARES) |
9066 | shares = MAX_SHARES; | |
62fb1851 | 9067 | |
8ed36996 | 9068 | mutex_lock(&shares_mutex); |
9b5b7751 | 9069 | if (tg->shares == shares) |
5cb350ba | 9070 | goto done; |
29f59db3 | 9071 | |
8ed36996 | 9072 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9073 | for_each_possible_cpu(i) |
9074 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9075 | list_del_rcu(&tg->siblings); |
8ed36996 | 9076 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9077 | |
9078 | /* wait for any ongoing reference to this group to finish */ | |
9079 | synchronize_sched(); | |
9080 | ||
9081 | /* | |
9082 | * Now we are free to modify the group's share on each cpu | |
9083 | * w/o tripping rebalance_share or load_balance_fair. | |
9084 | */ | |
9b5b7751 | 9085 | tg->shares = shares; |
c09595f6 PZ |
9086 | for_each_possible_cpu(i) { |
9087 | /* | |
9088 | * force a rebalance | |
9089 | */ | |
9090 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9091 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9092 | } |
29f59db3 | 9093 | |
6b2d7700 SV |
9094 | /* |
9095 | * Enable load balance activity on this group, by inserting it back on | |
9096 | * each cpu's rq->leaf_cfs_rq_list. | |
9097 | */ | |
8ed36996 | 9098 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9099 | for_each_possible_cpu(i) |
9100 | register_fair_sched_group(tg, i); | |
f473aa5e | 9101 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9102 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9103 | done: |
8ed36996 | 9104 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9105 | return 0; |
29f59db3 SV |
9106 | } |
9107 | ||
5cb350ba DG |
9108 | unsigned long sched_group_shares(struct task_group *tg) |
9109 | { | |
9110 | return tg->shares; | |
9111 | } | |
052f1dc7 | 9112 | #endif |
5cb350ba | 9113 | |
052f1dc7 | 9114 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9115 | /* |
9f0c1e56 | 9116 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9117 | */ |
9f0c1e56 PZ |
9118 | static DEFINE_MUTEX(rt_constraints_mutex); |
9119 | ||
9120 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9121 | { | |
9122 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9123 | return 1ULL << 20; |
9f0c1e56 | 9124 | |
9a7e0b18 | 9125 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9126 | } |
9127 | ||
9a7e0b18 PZ |
9128 | /* Must be called with tasklist_lock held */ |
9129 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9130 | { |
9a7e0b18 | 9131 | struct task_struct *g, *p; |
b40b2e8e | 9132 | |
9a7e0b18 PZ |
9133 | do_each_thread(g, p) { |
9134 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9135 | return 1; | |
9136 | } while_each_thread(g, p); | |
b40b2e8e | 9137 | |
9a7e0b18 PZ |
9138 | return 0; |
9139 | } | |
b40b2e8e | 9140 | |
9a7e0b18 PZ |
9141 | struct rt_schedulable_data { |
9142 | struct task_group *tg; | |
9143 | u64 rt_period; | |
9144 | u64 rt_runtime; | |
9145 | }; | |
b40b2e8e | 9146 | |
9a7e0b18 PZ |
9147 | static int tg_schedulable(struct task_group *tg, void *data) |
9148 | { | |
9149 | struct rt_schedulable_data *d = data; | |
9150 | struct task_group *child; | |
9151 | unsigned long total, sum = 0; | |
9152 | u64 period, runtime; | |
b40b2e8e | 9153 | |
9a7e0b18 PZ |
9154 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9155 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9156 | |
9a7e0b18 PZ |
9157 | if (tg == d->tg) { |
9158 | period = d->rt_period; | |
9159 | runtime = d->rt_runtime; | |
b40b2e8e | 9160 | } |
b40b2e8e | 9161 | |
98a4826b PZ |
9162 | #ifdef CONFIG_USER_SCHED |
9163 | if (tg == &root_task_group) { | |
9164 | period = global_rt_period(); | |
9165 | runtime = global_rt_runtime(); | |
9166 | } | |
9167 | #endif | |
9168 | ||
4653f803 PZ |
9169 | /* |
9170 | * Cannot have more runtime than the period. | |
9171 | */ | |
9172 | if (runtime > period && runtime != RUNTIME_INF) | |
9173 | return -EINVAL; | |
6f505b16 | 9174 | |
4653f803 PZ |
9175 | /* |
9176 | * Ensure we don't starve existing RT tasks. | |
9177 | */ | |
9a7e0b18 PZ |
9178 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9179 | return -EBUSY; | |
6f505b16 | 9180 | |
9a7e0b18 | 9181 | total = to_ratio(period, runtime); |
6f505b16 | 9182 | |
4653f803 PZ |
9183 | /* |
9184 | * Nobody can have more than the global setting allows. | |
9185 | */ | |
9186 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9187 | return -EINVAL; | |
6f505b16 | 9188 | |
4653f803 PZ |
9189 | /* |
9190 | * The sum of our children's runtime should not exceed our own. | |
9191 | */ | |
9a7e0b18 PZ |
9192 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9193 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9194 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9195 | |
9a7e0b18 PZ |
9196 | if (child == d->tg) { |
9197 | period = d->rt_period; | |
9198 | runtime = d->rt_runtime; | |
9199 | } | |
6f505b16 | 9200 | |
9a7e0b18 | 9201 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9202 | } |
6f505b16 | 9203 | |
9a7e0b18 PZ |
9204 | if (sum > total) |
9205 | return -EINVAL; | |
9206 | ||
9207 | return 0; | |
6f505b16 PZ |
9208 | } |
9209 | ||
9a7e0b18 | 9210 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9211 | { |
9a7e0b18 PZ |
9212 | struct rt_schedulable_data data = { |
9213 | .tg = tg, | |
9214 | .rt_period = period, | |
9215 | .rt_runtime = runtime, | |
9216 | }; | |
9217 | ||
9218 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9219 | } |
9220 | ||
d0b27fa7 PZ |
9221 | static int tg_set_bandwidth(struct task_group *tg, |
9222 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9223 | { |
ac086bc2 | 9224 | int i, err = 0; |
9f0c1e56 | 9225 | |
9f0c1e56 | 9226 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9227 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9228 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9229 | if (err) | |
9f0c1e56 | 9230 | goto unlock; |
ac086bc2 PZ |
9231 | |
9232 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9233 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9234 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9235 | |
9236 | for_each_possible_cpu(i) { | |
9237 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9238 | ||
9239 | spin_lock(&rt_rq->rt_runtime_lock); | |
9240 | rt_rq->rt_runtime = rt_runtime; | |
9241 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9242 | } | |
9243 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9244 | unlock: |
521f1a24 | 9245 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9246 | mutex_unlock(&rt_constraints_mutex); |
9247 | ||
9248 | return err; | |
6f505b16 PZ |
9249 | } |
9250 | ||
d0b27fa7 PZ |
9251 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9252 | { | |
9253 | u64 rt_runtime, rt_period; | |
9254 | ||
9255 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9256 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9257 | if (rt_runtime_us < 0) | |
9258 | rt_runtime = RUNTIME_INF; | |
9259 | ||
9260 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9261 | } | |
9262 | ||
9f0c1e56 PZ |
9263 | long sched_group_rt_runtime(struct task_group *tg) |
9264 | { | |
9265 | u64 rt_runtime_us; | |
9266 | ||
d0b27fa7 | 9267 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9268 | return -1; |
9269 | ||
d0b27fa7 | 9270 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9271 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9272 | return rt_runtime_us; | |
9273 | } | |
d0b27fa7 PZ |
9274 | |
9275 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9276 | { | |
9277 | u64 rt_runtime, rt_period; | |
9278 | ||
9279 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9280 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9281 | ||
619b0488 R |
9282 | if (rt_period == 0) |
9283 | return -EINVAL; | |
9284 | ||
d0b27fa7 PZ |
9285 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9286 | } | |
9287 | ||
9288 | long sched_group_rt_period(struct task_group *tg) | |
9289 | { | |
9290 | u64 rt_period_us; | |
9291 | ||
9292 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9293 | do_div(rt_period_us, NSEC_PER_USEC); | |
9294 | return rt_period_us; | |
9295 | } | |
9296 | ||
9297 | static int sched_rt_global_constraints(void) | |
9298 | { | |
4653f803 | 9299 | u64 runtime, period; |
d0b27fa7 PZ |
9300 | int ret = 0; |
9301 | ||
ec5d4989 HS |
9302 | if (sysctl_sched_rt_period <= 0) |
9303 | return -EINVAL; | |
9304 | ||
4653f803 PZ |
9305 | runtime = global_rt_runtime(); |
9306 | period = global_rt_period(); | |
9307 | ||
9308 | /* | |
9309 | * Sanity check on the sysctl variables. | |
9310 | */ | |
9311 | if (runtime > period && runtime != RUNTIME_INF) | |
9312 | return -EINVAL; | |
10b612f4 | 9313 | |
d0b27fa7 | 9314 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9315 | read_lock(&tasklist_lock); |
4653f803 | 9316 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9317 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9318 | mutex_unlock(&rt_constraints_mutex); |
9319 | ||
9320 | return ret; | |
9321 | } | |
6d6bc0ad | 9322 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9323 | static int sched_rt_global_constraints(void) |
9324 | { | |
ac086bc2 PZ |
9325 | unsigned long flags; |
9326 | int i; | |
9327 | ||
ec5d4989 HS |
9328 | if (sysctl_sched_rt_period <= 0) |
9329 | return -EINVAL; | |
9330 | ||
ac086bc2 PZ |
9331 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9332 | for_each_possible_cpu(i) { | |
9333 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9334 | ||
9335 | spin_lock(&rt_rq->rt_runtime_lock); | |
9336 | rt_rq->rt_runtime = global_rt_runtime(); | |
9337 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9338 | } | |
9339 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9340 | ||
d0b27fa7 PZ |
9341 | return 0; |
9342 | } | |
6d6bc0ad | 9343 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9344 | |
9345 | int sched_rt_handler(struct ctl_table *table, int write, | |
9346 | struct file *filp, void __user *buffer, size_t *lenp, | |
9347 | loff_t *ppos) | |
9348 | { | |
9349 | int ret; | |
9350 | int old_period, old_runtime; | |
9351 | static DEFINE_MUTEX(mutex); | |
9352 | ||
9353 | mutex_lock(&mutex); | |
9354 | old_period = sysctl_sched_rt_period; | |
9355 | old_runtime = sysctl_sched_rt_runtime; | |
9356 | ||
9357 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9358 | ||
9359 | if (!ret && write) { | |
9360 | ret = sched_rt_global_constraints(); | |
9361 | if (ret) { | |
9362 | sysctl_sched_rt_period = old_period; | |
9363 | sysctl_sched_rt_runtime = old_runtime; | |
9364 | } else { | |
9365 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9366 | def_rt_bandwidth.rt_period = | |
9367 | ns_to_ktime(global_rt_period()); | |
9368 | } | |
9369 | } | |
9370 | mutex_unlock(&mutex); | |
9371 | ||
9372 | return ret; | |
9373 | } | |
68318b8e | 9374 | |
052f1dc7 | 9375 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9376 | |
9377 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9378 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9379 | { |
2b01dfe3 PM |
9380 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9381 | struct task_group, css); | |
68318b8e SV |
9382 | } |
9383 | ||
9384 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9385 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9386 | { |
ec7dc8ac | 9387 | struct task_group *tg, *parent; |
68318b8e | 9388 | |
2b01dfe3 | 9389 | if (!cgrp->parent) { |
68318b8e | 9390 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9391 | return &init_task_group.css; |
9392 | } | |
9393 | ||
ec7dc8ac DG |
9394 | parent = cgroup_tg(cgrp->parent); |
9395 | tg = sched_create_group(parent); | |
68318b8e SV |
9396 | if (IS_ERR(tg)) |
9397 | return ERR_PTR(-ENOMEM); | |
9398 | ||
68318b8e SV |
9399 | return &tg->css; |
9400 | } | |
9401 | ||
41a2d6cf IM |
9402 | static void |
9403 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9404 | { |
2b01dfe3 | 9405 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9406 | |
9407 | sched_destroy_group(tg); | |
9408 | } | |
9409 | ||
41a2d6cf IM |
9410 | static int |
9411 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9412 | struct task_struct *tsk) | |
68318b8e | 9413 | { |
b68aa230 PZ |
9414 | #ifdef CONFIG_RT_GROUP_SCHED |
9415 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9416 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9417 | return -EINVAL; |
9418 | #else | |
68318b8e SV |
9419 | /* We don't support RT-tasks being in separate groups */ |
9420 | if (tsk->sched_class != &fair_sched_class) | |
9421 | return -EINVAL; | |
b68aa230 | 9422 | #endif |
68318b8e SV |
9423 | |
9424 | return 0; | |
9425 | } | |
9426 | ||
9427 | static void | |
2b01dfe3 | 9428 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9429 | struct cgroup *old_cont, struct task_struct *tsk) |
9430 | { | |
9431 | sched_move_task(tsk); | |
9432 | } | |
9433 | ||
052f1dc7 | 9434 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9435 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9436 | u64 shareval) |
68318b8e | 9437 | { |
2b01dfe3 | 9438 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9439 | } |
9440 | ||
f4c753b7 | 9441 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9442 | { |
2b01dfe3 | 9443 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9444 | |
9445 | return (u64) tg->shares; | |
9446 | } | |
6d6bc0ad | 9447 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9448 | |
052f1dc7 | 9449 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9450 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9451 | s64 val) |
6f505b16 | 9452 | { |
06ecb27c | 9453 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9454 | } |
9455 | ||
06ecb27c | 9456 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9457 | { |
06ecb27c | 9458 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9459 | } |
d0b27fa7 PZ |
9460 | |
9461 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9462 | u64 rt_period_us) | |
9463 | { | |
9464 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9465 | } | |
9466 | ||
9467 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9468 | { | |
9469 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9470 | } | |
6d6bc0ad | 9471 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9472 | |
fe5c7cc2 | 9473 | static struct cftype cpu_files[] = { |
052f1dc7 | 9474 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9475 | { |
9476 | .name = "shares", | |
f4c753b7 PM |
9477 | .read_u64 = cpu_shares_read_u64, |
9478 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9479 | }, |
052f1dc7 PZ |
9480 | #endif |
9481 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9482 | { |
9f0c1e56 | 9483 | .name = "rt_runtime_us", |
06ecb27c PM |
9484 | .read_s64 = cpu_rt_runtime_read, |
9485 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9486 | }, |
d0b27fa7 PZ |
9487 | { |
9488 | .name = "rt_period_us", | |
f4c753b7 PM |
9489 | .read_u64 = cpu_rt_period_read_uint, |
9490 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9491 | }, |
052f1dc7 | 9492 | #endif |
68318b8e SV |
9493 | }; |
9494 | ||
9495 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9496 | { | |
fe5c7cc2 | 9497 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9498 | } |
9499 | ||
9500 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9501 | .name = "cpu", |
9502 | .create = cpu_cgroup_create, | |
9503 | .destroy = cpu_cgroup_destroy, | |
9504 | .can_attach = cpu_cgroup_can_attach, | |
9505 | .attach = cpu_cgroup_attach, | |
9506 | .populate = cpu_cgroup_populate, | |
9507 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9508 | .early_init = 1, |
9509 | }; | |
9510 | ||
052f1dc7 | 9511 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9512 | |
9513 | #ifdef CONFIG_CGROUP_CPUACCT | |
9514 | ||
9515 | /* | |
9516 | * CPU accounting code for task groups. | |
9517 | * | |
9518 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9519 | * (balbir@in.ibm.com). | |
9520 | */ | |
9521 | ||
934352f2 | 9522 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9523 | struct cpuacct { |
9524 | struct cgroup_subsys_state css; | |
9525 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9526 | u64 *cpuusage; | |
934352f2 | 9527 | struct cpuacct *parent; |
d842de87 SV |
9528 | }; |
9529 | ||
9530 | struct cgroup_subsys cpuacct_subsys; | |
9531 | ||
9532 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9533 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9534 | { |
32cd756a | 9535 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9536 | struct cpuacct, css); |
9537 | } | |
9538 | ||
9539 | /* return cpu accounting group to which this task belongs */ | |
9540 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9541 | { | |
9542 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9543 | struct cpuacct, css); | |
9544 | } | |
9545 | ||
9546 | /* create a new cpu accounting group */ | |
9547 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9548 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9549 | { |
9550 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9551 | ||
9552 | if (!ca) | |
9553 | return ERR_PTR(-ENOMEM); | |
9554 | ||
9555 | ca->cpuusage = alloc_percpu(u64); | |
9556 | if (!ca->cpuusage) { | |
9557 | kfree(ca); | |
9558 | return ERR_PTR(-ENOMEM); | |
9559 | } | |
9560 | ||
934352f2 BR |
9561 | if (cgrp->parent) |
9562 | ca->parent = cgroup_ca(cgrp->parent); | |
9563 | ||
d842de87 SV |
9564 | return &ca->css; |
9565 | } | |
9566 | ||
9567 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9568 | static void |
32cd756a | 9569 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9570 | { |
32cd756a | 9571 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9572 | |
9573 | free_percpu(ca->cpuusage); | |
9574 | kfree(ca); | |
9575 | } | |
9576 | ||
720f5498 KC |
9577 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9578 | { | |
9579 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9580 | u64 data; | |
9581 | ||
9582 | #ifndef CONFIG_64BIT | |
9583 | /* | |
9584 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9585 | */ | |
9586 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9587 | data = *cpuusage; | |
9588 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9589 | #else | |
9590 | data = *cpuusage; | |
9591 | #endif | |
9592 | ||
9593 | return data; | |
9594 | } | |
9595 | ||
9596 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9597 | { | |
9598 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9599 | ||
9600 | #ifndef CONFIG_64BIT | |
9601 | /* | |
9602 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9603 | */ | |
9604 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9605 | *cpuusage = val; | |
9606 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9607 | #else | |
9608 | *cpuusage = val; | |
9609 | #endif | |
9610 | } | |
9611 | ||
d842de87 | 9612 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9613 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9614 | { |
32cd756a | 9615 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9616 | u64 totalcpuusage = 0; |
9617 | int i; | |
9618 | ||
720f5498 KC |
9619 | for_each_present_cpu(i) |
9620 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9621 | |
9622 | return totalcpuusage; | |
9623 | } | |
9624 | ||
0297b803 DG |
9625 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9626 | u64 reset) | |
9627 | { | |
9628 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9629 | int err = 0; | |
9630 | int i; | |
9631 | ||
9632 | if (reset) { | |
9633 | err = -EINVAL; | |
9634 | goto out; | |
9635 | } | |
9636 | ||
720f5498 KC |
9637 | for_each_present_cpu(i) |
9638 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9639 | |
0297b803 DG |
9640 | out: |
9641 | return err; | |
9642 | } | |
9643 | ||
e9515c3c KC |
9644 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9645 | struct seq_file *m) | |
9646 | { | |
9647 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9648 | u64 percpu; | |
9649 | int i; | |
9650 | ||
9651 | for_each_present_cpu(i) { | |
9652 | percpu = cpuacct_cpuusage_read(ca, i); | |
9653 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9654 | } | |
9655 | seq_printf(m, "\n"); | |
9656 | return 0; | |
9657 | } | |
9658 | ||
d842de87 SV |
9659 | static struct cftype files[] = { |
9660 | { | |
9661 | .name = "usage", | |
f4c753b7 PM |
9662 | .read_u64 = cpuusage_read, |
9663 | .write_u64 = cpuusage_write, | |
d842de87 | 9664 | }, |
e9515c3c KC |
9665 | { |
9666 | .name = "usage_percpu", | |
9667 | .read_seq_string = cpuacct_percpu_seq_read, | |
9668 | }, | |
9669 | ||
d842de87 SV |
9670 | }; |
9671 | ||
32cd756a | 9672 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9673 | { |
32cd756a | 9674 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9675 | } |
9676 | ||
9677 | /* | |
9678 | * charge this task's execution time to its accounting group. | |
9679 | * | |
9680 | * called with rq->lock held. | |
9681 | */ | |
9682 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9683 | { | |
9684 | struct cpuacct *ca; | |
934352f2 | 9685 | int cpu; |
d842de87 SV |
9686 | |
9687 | if (!cpuacct_subsys.active) | |
9688 | return; | |
9689 | ||
934352f2 | 9690 | cpu = task_cpu(tsk); |
d842de87 | 9691 | ca = task_ca(tsk); |
d842de87 | 9692 | |
934352f2 BR |
9693 | for (; ca; ca = ca->parent) { |
9694 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
d842de87 SV |
9695 | *cpuusage += cputime; |
9696 | } | |
9697 | } | |
9698 | ||
9699 | struct cgroup_subsys cpuacct_subsys = { | |
9700 | .name = "cpuacct", | |
9701 | .create = cpuacct_create, | |
9702 | .destroy = cpuacct_destroy, | |
9703 | .populate = cpuacct_populate, | |
9704 | .subsys_id = cpuacct_subsys_id, | |
9705 | }; | |
9706 | #endif /* CONFIG_CGROUP_CPUACCT */ |