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 | ||
5517d86b ED |
121 | #ifdef CONFIG_SMP |
122 | /* | |
123 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
124 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
125 | */ | |
126 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
127 | { | |
128 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
129 | } | |
130 | ||
131 | /* | |
132 | * Each time a sched group cpu_power is changed, | |
133 | * we must compute its reciprocal value | |
134 | */ | |
135 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
136 | { | |
137 | sg->__cpu_power += val; | |
138 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
139 | } | |
140 | #endif | |
141 | ||
e05606d3 IM |
142 | static inline int rt_policy(int policy) |
143 | { | |
3f33a7ce | 144 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
145 | return 1; |
146 | return 0; | |
147 | } | |
148 | ||
149 | static inline int task_has_rt_policy(struct task_struct *p) | |
150 | { | |
151 | return rt_policy(p->policy); | |
152 | } | |
153 | ||
1da177e4 | 154 | /* |
6aa645ea | 155 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 156 | */ |
6aa645ea IM |
157 | struct rt_prio_array { |
158 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
159 | struct list_head queue[MAX_RT_PRIO]; | |
160 | }; | |
161 | ||
d0b27fa7 | 162 | struct rt_bandwidth { |
ea736ed5 IM |
163 | /* nests inside the rq lock: */ |
164 | spinlock_t rt_runtime_lock; | |
165 | ktime_t rt_period; | |
166 | u64 rt_runtime; | |
167 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
168 | }; |
169 | ||
170 | static struct rt_bandwidth def_rt_bandwidth; | |
171 | ||
172 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
173 | ||
174 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
175 | { | |
176 | struct rt_bandwidth *rt_b = | |
177 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
178 | ktime_t now; | |
179 | int overrun; | |
180 | int idle = 0; | |
181 | ||
182 | for (;;) { | |
183 | now = hrtimer_cb_get_time(timer); | |
184 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
185 | ||
186 | if (!overrun) | |
187 | break; | |
188 | ||
189 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
190 | } | |
191 | ||
192 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
193 | } | |
194 | ||
195 | static | |
196 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
197 | { | |
198 | rt_b->rt_period = ns_to_ktime(period); | |
199 | rt_b->rt_runtime = runtime; | |
200 | ||
ac086bc2 PZ |
201 | spin_lock_init(&rt_b->rt_runtime_lock); |
202 | ||
d0b27fa7 PZ |
203 | hrtimer_init(&rt_b->rt_period_timer, |
204 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
205 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
ccc7dadf | 206 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; |
d0b27fa7 PZ |
207 | } |
208 | ||
c8bfff6d KH |
209 | static inline int rt_bandwidth_enabled(void) |
210 | { | |
211 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
212 | } |
213 | ||
214 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
215 | { | |
216 | ktime_t now; | |
217 | ||
0b148fa0 | 218 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
219 | return; |
220 | ||
221 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
222 | return; | |
223 | ||
224 | spin_lock(&rt_b->rt_runtime_lock); | |
225 | for (;;) { | |
226 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
227 | break; | |
228 | ||
229 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
230 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
231 | hrtimer_start_expires(&rt_b->rt_period_timer, |
232 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
233 | } |
234 | spin_unlock(&rt_b->rt_runtime_lock); | |
235 | } | |
236 | ||
237 | #ifdef CONFIG_RT_GROUP_SCHED | |
238 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
239 | { | |
240 | hrtimer_cancel(&rt_b->rt_period_timer); | |
241 | } | |
242 | #endif | |
243 | ||
712555ee HC |
244 | /* |
245 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
246 | * detach_destroy_domains and partition_sched_domains. | |
247 | */ | |
248 | static DEFINE_MUTEX(sched_domains_mutex); | |
249 | ||
052f1dc7 | 250 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 251 | |
68318b8e SV |
252 | #include <linux/cgroup.h> |
253 | ||
29f59db3 SV |
254 | struct cfs_rq; |
255 | ||
6f505b16 PZ |
256 | static LIST_HEAD(task_groups); |
257 | ||
29f59db3 | 258 | /* task group related information */ |
4cf86d77 | 259 | struct task_group { |
052f1dc7 | 260 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
261 | struct cgroup_subsys_state css; |
262 | #endif | |
052f1dc7 PZ |
263 | |
264 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
265 | /* schedulable entities of this group on each cpu */ |
266 | struct sched_entity **se; | |
267 | /* runqueue "owned" by this group on each cpu */ | |
268 | struct cfs_rq **cfs_rq; | |
269 | unsigned long shares; | |
052f1dc7 PZ |
270 | #endif |
271 | ||
272 | #ifdef CONFIG_RT_GROUP_SCHED | |
273 | struct sched_rt_entity **rt_se; | |
274 | struct rt_rq **rt_rq; | |
275 | ||
d0b27fa7 | 276 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 277 | #endif |
6b2d7700 | 278 | |
ae8393e5 | 279 | struct rcu_head rcu; |
6f505b16 | 280 | struct list_head list; |
f473aa5e PZ |
281 | |
282 | struct task_group *parent; | |
283 | struct list_head siblings; | |
284 | struct list_head children; | |
29f59db3 SV |
285 | }; |
286 | ||
354d60c2 | 287 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
288 | |
289 | /* | |
290 | * Root task group. | |
291 | * Every UID task group (including init_task_group aka UID-0) will | |
292 | * be a child to this group. | |
293 | */ | |
294 | struct task_group root_task_group; | |
295 | ||
052f1dc7 | 296 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
297 | /* Default task group's sched entity on each cpu */ |
298 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
299 | /* Default task group's cfs_rq on each cpu */ | |
300 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 301 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
302 | |
303 | #ifdef CONFIG_RT_GROUP_SCHED | |
304 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
305 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 306 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 307 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 308 | #define root_task_group init_task_group |
9a7e0b18 | 309 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 310 | |
8ed36996 | 311 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
312 | * a task group's cpu shares. |
313 | */ | |
8ed36996 | 314 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 315 | |
052f1dc7 | 316 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
317 | #ifdef CONFIG_USER_SCHED |
318 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 319 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 320 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 321 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 322 | |
cb4ad1ff | 323 | /* |
2e084786 LJ |
324 | * A weight of 0 or 1 can cause arithmetics problems. |
325 | * A weight of a cfs_rq is the sum of weights of which entities | |
326 | * are queued on this cfs_rq, so a weight of a entity should not be | |
327 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
328 | * (The default weight is 1024 - so there's no practical |
329 | * limitation from this.) | |
330 | */ | |
18d95a28 | 331 | #define MIN_SHARES 2 |
2e084786 | 332 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 333 | |
052f1dc7 PZ |
334 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
335 | #endif | |
336 | ||
29f59db3 | 337 | /* Default task group. |
3a252015 | 338 | * Every task in system belong to this group at bootup. |
29f59db3 | 339 | */ |
434d53b0 | 340 | struct task_group init_task_group; |
29f59db3 SV |
341 | |
342 | /* return group to which a task belongs */ | |
4cf86d77 | 343 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 344 | { |
4cf86d77 | 345 | struct task_group *tg; |
9b5b7751 | 346 | |
052f1dc7 | 347 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 348 | tg = p->user->tg; |
052f1dc7 | 349 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
350 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
351 | struct task_group, css); | |
24e377a8 | 352 | #else |
41a2d6cf | 353 | tg = &init_task_group; |
24e377a8 | 354 | #endif |
9b5b7751 | 355 | return tg; |
29f59db3 SV |
356 | } |
357 | ||
358 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 359 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 360 | { |
052f1dc7 | 361 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
362 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
363 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 364 | #endif |
6f505b16 | 365 | |
052f1dc7 | 366 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
367 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
368 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 369 | #endif |
29f59db3 SV |
370 | } |
371 | ||
372 | #else | |
373 | ||
6f505b16 | 374 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
375 | static inline struct task_group *task_group(struct task_struct *p) |
376 | { | |
377 | return NULL; | |
378 | } | |
29f59db3 | 379 | |
052f1dc7 | 380 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 381 | |
6aa645ea IM |
382 | /* CFS-related fields in a runqueue */ |
383 | struct cfs_rq { | |
384 | struct load_weight load; | |
385 | unsigned long nr_running; | |
386 | ||
6aa645ea | 387 | u64 exec_clock; |
e9acbff6 | 388 | u64 min_vruntime; |
6aa645ea IM |
389 | |
390 | struct rb_root tasks_timeline; | |
391 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
392 | |
393 | struct list_head tasks; | |
394 | struct list_head *balance_iterator; | |
395 | ||
396 | /* | |
397 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
398 | * It is set to NULL otherwise (i.e when none are currently running). |
399 | */ | |
aa2ac252 | 400 | struct sched_entity *curr, *next; |
ddc97297 PZ |
401 | |
402 | unsigned long nr_spread_over; | |
403 | ||
62160e3f | 404 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
405 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
406 | ||
41a2d6cf IM |
407 | /* |
408 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
409 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
410 | * (like users, containers etc.) | |
411 | * | |
412 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
413 | * list is used during load balance. | |
414 | */ | |
41a2d6cf IM |
415 | struct list_head leaf_cfs_rq_list; |
416 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
417 | |
418 | #ifdef CONFIG_SMP | |
c09595f6 | 419 | /* |
c8cba857 | 420 | * the part of load.weight contributed by tasks |
c09595f6 | 421 | */ |
c8cba857 | 422 | unsigned long task_weight; |
c09595f6 | 423 | |
c8cba857 PZ |
424 | /* |
425 | * h_load = weight * f(tg) | |
426 | * | |
427 | * Where f(tg) is the recursive weight fraction assigned to | |
428 | * this group. | |
429 | */ | |
430 | unsigned long h_load; | |
c09595f6 | 431 | |
c8cba857 PZ |
432 | /* |
433 | * this cpu's part of tg->shares | |
434 | */ | |
435 | unsigned long shares; | |
f1d239f7 PZ |
436 | |
437 | /* | |
438 | * load.weight at the time we set shares | |
439 | */ | |
440 | unsigned long rq_weight; | |
c09595f6 | 441 | #endif |
6aa645ea IM |
442 | #endif |
443 | }; | |
1da177e4 | 444 | |
6aa645ea IM |
445 | /* Real-Time classes' related field in a runqueue: */ |
446 | struct rt_rq { | |
447 | struct rt_prio_array active; | |
63489e45 | 448 | unsigned long rt_nr_running; |
052f1dc7 | 449 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
450 | int highest_prio; /* highest queued rt task prio */ |
451 | #endif | |
fa85ae24 | 452 | #ifdef CONFIG_SMP |
73fe6aae | 453 | unsigned long rt_nr_migratory; |
a22d7fc1 | 454 | int overloaded; |
fa85ae24 | 455 | #endif |
6f505b16 | 456 | int rt_throttled; |
fa85ae24 | 457 | u64 rt_time; |
ac086bc2 | 458 | u64 rt_runtime; |
ea736ed5 | 459 | /* Nests inside the rq lock: */ |
ac086bc2 | 460 | spinlock_t rt_runtime_lock; |
6f505b16 | 461 | |
052f1dc7 | 462 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
463 | unsigned long rt_nr_boosted; |
464 | ||
6f505b16 PZ |
465 | struct rq *rq; |
466 | struct list_head leaf_rt_rq_list; | |
467 | struct task_group *tg; | |
468 | struct sched_rt_entity *rt_se; | |
469 | #endif | |
6aa645ea IM |
470 | }; |
471 | ||
57d885fe GH |
472 | #ifdef CONFIG_SMP |
473 | ||
474 | /* | |
475 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
476 | * variables. Each exclusive cpuset essentially defines an island domain by |
477 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
478 | * exclusive cpuset is created, we also create and attach a new root-domain |
479 | * object. | |
480 | * | |
57d885fe GH |
481 | */ |
482 | struct root_domain { | |
483 | atomic_t refcount; | |
484 | cpumask_t span; | |
485 | cpumask_t online; | |
637f5085 | 486 | |
0eab9146 | 487 | /* |
637f5085 GH |
488 | * The "RT overload" flag: it gets set if a CPU has more than |
489 | * one runnable RT task. | |
490 | */ | |
491 | cpumask_t rto_mask; | |
0eab9146 | 492 | atomic_t rto_count; |
6e0534f2 GH |
493 | #ifdef CONFIG_SMP |
494 | struct cpupri cpupri; | |
495 | #endif | |
57d885fe GH |
496 | }; |
497 | ||
dc938520 GH |
498 | /* |
499 | * By default the system creates a single root-domain with all cpus as | |
500 | * members (mimicking the global state we have today). | |
501 | */ | |
57d885fe GH |
502 | static struct root_domain def_root_domain; |
503 | ||
504 | #endif | |
505 | ||
1da177e4 LT |
506 | /* |
507 | * This is the main, per-CPU runqueue data structure. | |
508 | * | |
509 | * Locking rule: those places that want to lock multiple runqueues | |
510 | * (such as the load balancing or the thread migration code), lock | |
511 | * acquire operations must be ordered by ascending &runqueue. | |
512 | */ | |
70b97a7f | 513 | struct rq { |
d8016491 IM |
514 | /* runqueue lock: */ |
515 | spinlock_t lock; | |
1da177e4 LT |
516 | |
517 | /* | |
518 | * nr_running and cpu_load should be in the same cacheline because | |
519 | * remote CPUs use both these fields when doing load calculation. | |
520 | */ | |
521 | unsigned long nr_running; | |
6aa645ea IM |
522 | #define CPU_LOAD_IDX_MAX 5 |
523 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 524 | unsigned char idle_at_tick; |
46cb4b7c | 525 | #ifdef CONFIG_NO_HZ |
15934a37 | 526 | unsigned long last_tick_seen; |
46cb4b7c SS |
527 | unsigned char in_nohz_recently; |
528 | #endif | |
d8016491 IM |
529 | /* capture load from *all* tasks on this cpu: */ |
530 | struct load_weight load; | |
6aa645ea IM |
531 | unsigned long nr_load_updates; |
532 | u64 nr_switches; | |
533 | ||
534 | struct cfs_rq cfs; | |
6f505b16 | 535 | struct rt_rq rt; |
6f505b16 | 536 | |
6aa645ea | 537 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
538 | /* list of leaf cfs_rq on this cpu: */ |
539 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
540 | #endif |
541 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 542 | struct list_head leaf_rt_rq_list; |
1da177e4 | 543 | #endif |
1da177e4 LT |
544 | |
545 | /* | |
546 | * This is part of a global counter where only the total sum | |
547 | * over all CPUs matters. A task can increase this counter on | |
548 | * one CPU and if it got migrated afterwards it may decrease | |
549 | * it on another CPU. Always updated under the runqueue lock: | |
550 | */ | |
551 | unsigned long nr_uninterruptible; | |
552 | ||
36c8b586 | 553 | struct task_struct *curr, *idle; |
c9819f45 | 554 | unsigned long next_balance; |
1da177e4 | 555 | struct mm_struct *prev_mm; |
6aa645ea | 556 | |
3e51f33f | 557 | u64 clock; |
6aa645ea | 558 | |
1da177e4 LT |
559 | atomic_t nr_iowait; |
560 | ||
561 | #ifdef CONFIG_SMP | |
0eab9146 | 562 | struct root_domain *rd; |
1da177e4 LT |
563 | struct sched_domain *sd; |
564 | ||
565 | /* For active balancing */ | |
566 | int active_balance; | |
567 | int push_cpu; | |
d8016491 IM |
568 | /* cpu of this runqueue: */ |
569 | int cpu; | |
1f11eb6a | 570 | int online; |
1da177e4 | 571 | |
a8a51d5e | 572 | unsigned long avg_load_per_task; |
1da177e4 | 573 | |
36c8b586 | 574 | struct task_struct *migration_thread; |
1da177e4 LT |
575 | struct list_head migration_queue; |
576 | #endif | |
577 | ||
8f4d37ec | 578 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
579 | #ifdef CONFIG_SMP |
580 | int hrtick_csd_pending; | |
581 | struct call_single_data hrtick_csd; | |
582 | #endif | |
8f4d37ec PZ |
583 | struct hrtimer hrtick_timer; |
584 | #endif | |
585 | ||
1da177e4 LT |
586 | #ifdef CONFIG_SCHEDSTATS |
587 | /* latency stats */ | |
588 | struct sched_info rq_sched_info; | |
589 | ||
590 | /* sys_sched_yield() stats */ | |
480b9434 KC |
591 | unsigned int yld_exp_empty; |
592 | unsigned int yld_act_empty; | |
593 | unsigned int yld_both_empty; | |
594 | unsigned int yld_count; | |
1da177e4 LT |
595 | |
596 | /* schedule() stats */ | |
480b9434 KC |
597 | unsigned int sched_switch; |
598 | unsigned int sched_count; | |
599 | unsigned int sched_goidle; | |
1da177e4 LT |
600 | |
601 | /* try_to_wake_up() stats */ | |
480b9434 KC |
602 | unsigned int ttwu_count; |
603 | unsigned int ttwu_local; | |
b8efb561 IM |
604 | |
605 | /* BKL stats */ | |
480b9434 | 606 | unsigned int bkl_count; |
1da177e4 LT |
607 | #endif |
608 | }; | |
609 | ||
f34e3b61 | 610 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 611 | |
15afe09b | 612 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 613 | { |
15afe09b | 614 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
615 | } |
616 | ||
0a2966b4 CL |
617 | static inline int cpu_of(struct rq *rq) |
618 | { | |
619 | #ifdef CONFIG_SMP | |
620 | return rq->cpu; | |
621 | #else | |
622 | return 0; | |
623 | #endif | |
624 | } | |
625 | ||
674311d5 NP |
626 | /* |
627 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 628 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
629 | * |
630 | * The domain tree of any CPU may only be accessed from within | |
631 | * preempt-disabled sections. | |
632 | */ | |
48f24c4d IM |
633 | #define for_each_domain(cpu, __sd) \ |
634 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
635 | |
636 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
637 | #define this_rq() (&__get_cpu_var(runqueues)) | |
638 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
639 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
640 | ||
3e51f33f PZ |
641 | static inline void update_rq_clock(struct rq *rq) |
642 | { | |
643 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
644 | } | |
645 | ||
bf5c91ba IM |
646 | /* |
647 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
648 | */ | |
649 | #ifdef CONFIG_SCHED_DEBUG | |
650 | # define const_debug __read_mostly | |
651 | #else | |
652 | # define const_debug static const | |
653 | #endif | |
654 | ||
017730c1 IM |
655 | /** |
656 | * runqueue_is_locked | |
657 | * | |
658 | * Returns true if the current cpu runqueue is locked. | |
659 | * This interface allows printk to be called with the runqueue lock | |
660 | * held and know whether or not it is OK to wake up the klogd. | |
661 | */ | |
662 | int runqueue_is_locked(void) | |
663 | { | |
664 | int cpu = get_cpu(); | |
665 | struct rq *rq = cpu_rq(cpu); | |
666 | int ret; | |
667 | ||
668 | ret = spin_is_locked(&rq->lock); | |
669 | put_cpu(); | |
670 | return ret; | |
671 | } | |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Debugging: various feature bits | |
675 | */ | |
f00b45c1 PZ |
676 | |
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | __SCHED_FEAT_##name , | |
679 | ||
bf5c91ba | 680 | enum { |
f00b45c1 | 681 | #include "sched_features.h" |
bf5c91ba IM |
682 | }; |
683 | ||
f00b45c1 PZ |
684 | #undef SCHED_FEAT |
685 | ||
686 | #define SCHED_FEAT(name, enabled) \ | |
687 | (1UL << __SCHED_FEAT_##name) * enabled | | |
688 | ||
bf5c91ba | 689 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
690 | #include "sched_features.h" |
691 | 0; | |
692 | ||
693 | #undef SCHED_FEAT | |
694 | ||
695 | #ifdef CONFIG_SCHED_DEBUG | |
696 | #define SCHED_FEAT(name, enabled) \ | |
697 | #name , | |
698 | ||
983ed7a6 | 699 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
700 | #include "sched_features.h" |
701 | NULL | |
702 | }; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
34f3a814 | 706 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 707 | { |
f00b45c1 PZ |
708 | int i; |
709 | ||
710 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
711 | if (!(sysctl_sched_features & (1UL << i))) |
712 | seq_puts(m, "NO_"); | |
713 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 714 | } |
34f3a814 | 715 | seq_puts(m, "\n"); |
f00b45c1 | 716 | |
34f3a814 | 717 | return 0; |
f00b45c1 PZ |
718 | } |
719 | ||
720 | static ssize_t | |
721 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
722 | size_t cnt, loff_t *ppos) | |
723 | { | |
724 | char buf[64]; | |
725 | char *cmp = buf; | |
726 | int neg = 0; | |
727 | int i; | |
728 | ||
729 | if (cnt > 63) | |
730 | cnt = 63; | |
731 | ||
732 | if (copy_from_user(&buf, ubuf, cnt)) | |
733 | return -EFAULT; | |
734 | ||
735 | buf[cnt] = 0; | |
736 | ||
c24b7c52 | 737 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
738 | neg = 1; |
739 | cmp += 3; | |
740 | } | |
741 | ||
742 | for (i = 0; sched_feat_names[i]; i++) { | |
743 | int len = strlen(sched_feat_names[i]); | |
744 | ||
745 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
746 | if (neg) | |
747 | sysctl_sched_features &= ~(1UL << i); | |
748 | else | |
749 | sysctl_sched_features |= (1UL << i); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (!sched_feat_names[i]) | |
755 | return -EINVAL; | |
756 | ||
757 | filp->f_pos += cnt; | |
758 | ||
759 | return cnt; | |
760 | } | |
761 | ||
34f3a814 LZ |
762 | static int sched_feat_open(struct inode *inode, struct file *filp) |
763 | { | |
764 | return single_open(filp, sched_feat_show, NULL); | |
765 | } | |
766 | ||
f00b45c1 | 767 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
768 | .open = sched_feat_open, |
769 | .write = sched_feat_write, | |
770 | .read = seq_read, | |
771 | .llseek = seq_lseek, | |
772 | .release = single_release, | |
f00b45c1 PZ |
773 | }; |
774 | ||
775 | static __init int sched_init_debug(void) | |
776 | { | |
f00b45c1 PZ |
777 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
778 | &sched_feat_fops); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | late_initcall(sched_init_debug); | |
783 | ||
784 | #endif | |
785 | ||
786 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 787 | |
b82d9fdd PZ |
788 | /* |
789 | * Number of tasks to iterate in a single balance run. | |
790 | * Limited because this is done with IRQs disabled. | |
791 | */ | |
792 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
793 | ||
2398f2c6 PZ |
794 | /* |
795 | * ratelimit for updating the group shares. | |
55cd5340 | 796 | * default: 0.25ms |
2398f2c6 | 797 | */ |
55cd5340 | 798 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 799 | |
ffda12a1 PZ |
800 | /* |
801 | * Inject some fuzzyness into changing the per-cpu group shares | |
802 | * this avoids remote rq-locks at the expense of fairness. | |
803 | * default: 4 | |
804 | */ | |
805 | unsigned int sysctl_sched_shares_thresh = 4; | |
806 | ||
fa85ae24 | 807 | /* |
9f0c1e56 | 808 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
809 | * default: 1s |
810 | */ | |
9f0c1e56 | 811 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 812 | |
6892b75e IM |
813 | static __read_mostly int scheduler_running; |
814 | ||
9f0c1e56 PZ |
815 | /* |
816 | * part of the period that we allow rt tasks to run in us. | |
817 | * default: 0.95s | |
818 | */ | |
819 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 820 | |
d0b27fa7 PZ |
821 | static inline u64 global_rt_period(void) |
822 | { | |
823 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
824 | } | |
825 | ||
826 | static inline u64 global_rt_runtime(void) | |
827 | { | |
e26873bb | 828 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
829 | return RUNTIME_INF; |
830 | ||
831 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
832 | } | |
fa85ae24 | 833 | |
1da177e4 | 834 | #ifndef prepare_arch_switch |
4866cde0 NP |
835 | # define prepare_arch_switch(next) do { } while (0) |
836 | #endif | |
837 | #ifndef finish_arch_switch | |
838 | # define finish_arch_switch(prev) do { } while (0) | |
839 | #endif | |
840 | ||
051a1d1a DA |
841 | static inline int task_current(struct rq *rq, struct task_struct *p) |
842 | { | |
843 | return rq->curr == p; | |
844 | } | |
845 | ||
4866cde0 | 846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 847 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 848 | { |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | } |
851 | ||
70b97a7f | 852 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
853 | { |
854 | } | |
855 | ||
70b97a7f | 856 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 857 | { |
da04c035 IM |
858 | #ifdef CONFIG_DEBUG_SPINLOCK |
859 | /* this is a valid case when another task releases the spinlock */ | |
860 | rq->lock.owner = current; | |
861 | #endif | |
8a25d5de IM |
862 | /* |
863 | * If we are tracking spinlock dependencies then we have to | |
864 | * fix up the runqueue lock - which gets 'carried over' from | |
865 | * prev into current: | |
866 | */ | |
867 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
868 | ||
4866cde0 NP |
869 | spin_unlock_irq(&rq->lock); |
870 | } | |
871 | ||
872 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
874 | { |
875 | #ifdef CONFIG_SMP | |
876 | return p->oncpu; | |
877 | #else | |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | #endif |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * We can optimise this out completely for !SMP, because the | |
887 | * SMP rebalancing from interrupt is the only thing that cares | |
888 | * here. | |
889 | */ | |
890 | next->oncpu = 1; | |
891 | #endif | |
892 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
893 | spin_unlock_irq(&rq->lock); | |
894 | #else | |
895 | spin_unlock(&rq->lock); | |
896 | #endif | |
897 | } | |
898 | ||
70b97a7f | 899 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
904 | * We must ensure this doesn't happen until the switch is completely | |
905 | * finished. | |
906 | */ | |
907 | smp_wmb(); | |
908 | prev->oncpu = 0; | |
909 | #endif | |
910 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
911 | local_irq_enable(); | |
1da177e4 | 912 | #endif |
4866cde0 NP |
913 | } |
914 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 915 | |
b29739f9 IM |
916 | /* |
917 | * __task_rq_lock - lock the runqueue a given task resides on. | |
918 | * Must be called interrupts disabled. | |
919 | */ | |
70b97a7f | 920 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
921 | __acquires(rq->lock) |
922 | { | |
3a5c359a AK |
923 | for (;;) { |
924 | struct rq *rq = task_rq(p); | |
925 | spin_lock(&rq->lock); | |
926 | if (likely(rq == task_rq(p))) | |
927 | return rq; | |
b29739f9 | 928 | spin_unlock(&rq->lock); |
b29739f9 | 929 | } |
b29739f9 IM |
930 | } |
931 | ||
1da177e4 LT |
932 | /* |
933 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 934 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
935 | * explicitly disabling preemption. |
936 | */ | |
70b97a7f | 937 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
938 | __acquires(rq->lock) |
939 | { | |
70b97a7f | 940 | struct rq *rq; |
1da177e4 | 941 | |
3a5c359a AK |
942 | for (;;) { |
943 | local_irq_save(*flags); | |
944 | rq = task_rq(p); | |
945 | spin_lock(&rq->lock); | |
946 | if (likely(rq == task_rq(p))) | |
947 | return rq; | |
1da177e4 | 948 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 949 | } |
1da177e4 LT |
950 | } |
951 | ||
a9957449 | 952 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
953 | __releases(rq->lock) |
954 | { | |
955 | spin_unlock(&rq->lock); | |
956 | } | |
957 | ||
70b97a7f | 958 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
959 | __releases(rq->lock) |
960 | { | |
961 | spin_unlock_irqrestore(&rq->lock, *flags); | |
962 | } | |
963 | ||
1da177e4 | 964 | /* |
cc2a73b5 | 965 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 966 | */ |
a9957449 | 967 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
968 | __acquires(rq->lock) |
969 | { | |
70b97a7f | 970 | struct rq *rq; |
1da177e4 LT |
971 | |
972 | local_irq_disable(); | |
973 | rq = this_rq(); | |
974 | spin_lock(&rq->lock); | |
975 | ||
976 | return rq; | |
977 | } | |
978 | ||
8f4d37ec PZ |
979 | #ifdef CONFIG_SCHED_HRTICK |
980 | /* | |
981 | * Use HR-timers to deliver accurate preemption points. | |
982 | * | |
983 | * Its all a bit involved since we cannot program an hrt while holding the | |
984 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
985 | * reschedule event. | |
986 | * | |
987 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
988 | * rq->lock. | |
989 | */ | |
8f4d37ec PZ |
990 | |
991 | /* | |
992 | * Use hrtick when: | |
993 | * - enabled by features | |
994 | * - hrtimer is actually high res | |
995 | */ | |
996 | static inline int hrtick_enabled(struct rq *rq) | |
997 | { | |
998 | if (!sched_feat(HRTICK)) | |
999 | return 0; | |
ba42059f | 1000 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1001 | return 0; |
8f4d37ec PZ |
1002 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1003 | } | |
1004 | ||
8f4d37ec PZ |
1005 | static void hrtick_clear(struct rq *rq) |
1006 | { | |
1007 | if (hrtimer_active(&rq->hrtick_timer)) | |
1008 | hrtimer_cancel(&rq->hrtick_timer); | |
1009 | } | |
1010 | ||
8f4d37ec PZ |
1011 | /* |
1012 | * High-resolution timer tick. | |
1013 | * Runs from hardirq context with interrupts disabled. | |
1014 | */ | |
1015 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1016 | { | |
1017 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1018 | ||
1019 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1020 | ||
1021 | spin_lock(&rq->lock); | |
3e51f33f | 1022 | update_rq_clock(rq); |
8f4d37ec PZ |
1023 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1024 | spin_unlock(&rq->lock); | |
1025 | ||
1026 | return HRTIMER_NORESTART; | |
1027 | } | |
1028 | ||
95e904c7 | 1029 | #ifdef CONFIG_SMP |
31656519 PZ |
1030 | /* |
1031 | * called from hardirq (IPI) context | |
1032 | */ | |
1033 | static void __hrtick_start(void *arg) | |
b328ca18 | 1034 | { |
31656519 | 1035 | struct rq *rq = arg; |
b328ca18 | 1036 | |
31656519 PZ |
1037 | spin_lock(&rq->lock); |
1038 | hrtimer_restart(&rq->hrtick_timer); | |
1039 | rq->hrtick_csd_pending = 0; | |
1040 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1041 | } |
1042 | ||
31656519 PZ |
1043 | /* |
1044 | * Called to set the hrtick timer state. | |
1045 | * | |
1046 | * called with rq->lock held and irqs disabled | |
1047 | */ | |
1048 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1049 | { |
31656519 PZ |
1050 | struct hrtimer *timer = &rq->hrtick_timer; |
1051 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1052 | |
cc584b21 | 1053 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1054 | |
1055 | if (rq == this_rq()) { | |
1056 | hrtimer_restart(timer); | |
1057 | } else if (!rq->hrtick_csd_pending) { | |
1058 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1059 | rq->hrtick_csd_pending = 1; | |
1060 | } | |
b328ca18 PZ |
1061 | } |
1062 | ||
1063 | static int | |
1064 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1065 | { | |
1066 | int cpu = (int)(long)hcpu; | |
1067 | ||
1068 | switch (action) { | |
1069 | case CPU_UP_CANCELED: | |
1070 | case CPU_UP_CANCELED_FROZEN: | |
1071 | case CPU_DOWN_PREPARE: | |
1072 | case CPU_DOWN_PREPARE_FROZEN: | |
1073 | case CPU_DEAD: | |
1074 | case CPU_DEAD_FROZEN: | |
31656519 | 1075 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1076 | return NOTIFY_OK; |
1077 | } | |
1078 | ||
1079 | return NOTIFY_DONE; | |
1080 | } | |
1081 | ||
fa748203 | 1082 | static __init void init_hrtick(void) |
b328ca18 PZ |
1083 | { |
1084 | hotcpu_notifier(hotplug_hrtick, 0); | |
1085 | } | |
31656519 PZ |
1086 | #else |
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) | |
1093 | { | |
1094 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1095 | } | |
b328ca18 | 1096 | |
006c75f1 | 1097 | static inline void init_hrtick(void) |
8f4d37ec | 1098 | { |
8f4d37ec | 1099 | } |
31656519 | 1100 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1101 | |
31656519 | 1102 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1103 | { |
31656519 PZ |
1104 | #ifdef CONFIG_SMP |
1105 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1106 | |
31656519 PZ |
1107 | rq->hrtick_csd.flags = 0; |
1108 | rq->hrtick_csd.func = __hrtick_start; | |
1109 | rq->hrtick_csd.info = rq; | |
1110 | #endif | |
8f4d37ec | 1111 | |
31656519 PZ |
1112 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1113 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1114 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1115 | } |
006c75f1 | 1116 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1117 | static inline void hrtick_clear(struct rq *rq) |
1118 | { | |
1119 | } | |
1120 | ||
8f4d37ec PZ |
1121 | static inline void init_rq_hrtick(struct rq *rq) |
1122 | { | |
1123 | } | |
1124 | ||
b328ca18 PZ |
1125 | static inline void init_hrtick(void) |
1126 | { | |
1127 | } | |
006c75f1 | 1128 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1129 | |
c24d20db IM |
1130 | /* |
1131 | * resched_task - mark a task 'to be rescheduled now'. | |
1132 | * | |
1133 | * On UP this means the setting of the need_resched flag, on SMP it | |
1134 | * might also involve a cross-CPU call to trigger the scheduler on | |
1135 | * the target CPU. | |
1136 | */ | |
1137 | #ifdef CONFIG_SMP | |
1138 | ||
1139 | #ifndef tsk_is_polling | |
1140 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1141 | #endif | |
1142 | ||
31656519 | 1143 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1144 | { |
1145 | int cpu; | |
1146 | ||
1147 | assert_spin_locked(&task_rq(p)->lock); | |
1148 | ||
31656519 | 1149 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1150 | return; |
1151 | ||
31656519 | 1152 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1153 | |
1154 | cpu = task_cpu(p); | |
1155 | if (cpu == smp_processor_id()) | |
1156 | return; | |
1157 | ||
1158 | /* NEED_RESCHED must be visible before we test polling */ | |
1159 | smp_mb(); | |
1160 | if (!tsk_is_polling(p)) | |
1161 | smp_send_reschedule(cpu); | |
1162 | } | |
1163 | ||
1164 | static void resched_cpu(int cpu) | |
1165 | { | |
1166 | struct rq *rq = cpu_rq(cpu); | |
1167 | unsigned long flags; | |
1168 | ||
1169 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1170 | return; | |
1171 | resched_task(cpu_curr(cpu)); | |
1172 | spin_unlock_irqrestore(&rq->lock, flags); | |
1173 | } | |
06d8308c TG |
1174 | |
1175 | #ifdef CONFIG_NO_HZ | |
1176 | /* | |
1177 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1178 | * idle CPU then this timer might expire before the next timer event | |
1179 | * which is scheduled to wake up that CPU. In case of a completely | |
1180 | * idle system the next event might even be infinite time into the | |
1181 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1182 | * leaves the inner idle loop so the newly added timer is taken into | |
1183 | * account when the CPU goes back to idle and evaluates the timer | |
1184 | * wheel for the next timer event. | |
1185 | */ | |
1186 | void wake_up_idle_cpu(int cpu) | |
1187 | { | |
1188 | struct rq *rq = cpu_rq(cpu); | |
1189 | ||
1190 | if (cpu == smp_processor_id()) | |
1191 | return; | |
1192 | ||
1193 | /* | |
1194 | * This is safe, as this function is called with the timer | |
1195 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1196 | * to idle and has not yet set rq->curr to idle then it will | |
1197 | * be serialized on the timer wheel base lock and take the new | |
1198 | * timer into account automatically. | |
1199 | */ | |
1200 | if (rq->curr != rq->idle) | |
1201 | return; | |
1202 | ||
1203 | /* | |
1204 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1205 | * lockless. The worst case is that the other CPU runs the | |
1206 | * idle task through an additional NOOP schedule() | |
1207 | */ | |
1208 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1209 | ||
1210 | /* NEED_RESCHED must be visible before we test polling */ | |
1211 | smp_mb(); | |
1212 | if (!tsk_is_polling(rq->idle)) | |
1213 | smp_send_reschedule(cpu); | |
1214 | } | |
6d6bc0ad | 1215 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1216 | |
6d6bc0ad | 1217 | #else /* !CONFIG_SMP */ |
31656519 | 1218 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1219 | { |
1220 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1221 | set_tsk_need_resched(p); |
c24d20db | 1222 | } |
6d6bc0ad | 1223 | #endif /* CONFIG_SMP */ |
c24d20db | 1224 | |
45bf76df IM |
1225 | #if BITS_PER_LONG == 32 |
1226 | # define WMULT_CONST (~0UL) | |
1227 | #else | |
1228 | # define WMULT_CONST (1UL << 32) | |
1229 | #endif | |
1230 | ||
1231 | #define WMULT_SHIFT 32 | |
1232 | ||
194081eb IM |
1233 | /* |
1234 | * Shift right and round: | |
1235 | */ | |
cf2ab469 | 1236 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1237 | |
a7be37ac PZ |
1238 | /* |
1239 | * delta *= weight / lw | |
1240 | */ | |
cb1c4fc9 | 1241 | static unsigned long |
45bf76df IM |
1242 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1243 | struct load_weight *lw) | |
1244 | { | |
1245 | u64 tmp; | |
1246 | ||
7a232e03 LJ |
1247 | if (!lw->inv_weight) { |
1248 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1249 | lw->inv_weight = 1; | |
1250 | else | |
1251 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1252 | / (lw->weight+1); | |
1253 | } | |
45bf76df IM |
1254 | |
1255 | tmp = (u64)delta_exec * weight; | |
1256 | /* | |
1257 | * Check whether we'd overflow the 64-bit multiplication: | |
1258 | */ | |
194081eb | 1259 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1260 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1261 | WMULT_SHIFT/2); |
1262 | else | |
cf2ab469 | 1263 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1264 | |
ecf691da | 1265 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1266 | } |
1267 | ||
1091985b | 1268 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1269 | { |
1270 | lw->weight += inc; | |
e89996ae | 1271 | lw->inv_weight = 0; |
45bf76df IM |
1272 | } |
1273 | ||
1091985b | 1274 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1275 | { |
1276 | lw->weight -= dec; | |
e89996ae | 1277 | lw->inv_weight = 0; |
45bf76df IM |
1278 | } |
1279 | ||
2dd73a4f PW |
1280 | /* |
1281 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1282 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1283 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1284 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1285 | * scaled version of the new time slice allocation that they receive on time |
1286 | * slice expiry etc. | |
1287 | */ | |
1288 | ||
dd41f596 IM |
1289 | #define WEIGHT_IDLEPRIO 2 |
1290 | #define WMULT_IDLEPRIO (1 << 31) | |
1291 | ||
1292 | /* | |
1293 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1294 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1295 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1296 | * that remained on nice 0. | |
1297 | * | |
1298 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1299 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1300 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1301 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1302 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1303 | */ |
1304 | static const int prio_to_weight[40] = { | |
254753dc IM |
1305 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1306 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1307 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1308 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1309 | /* 0 */ 1024, 820, 655, 526, 423, | |
1310 | /* 5 */ 335, 272, 215, 172, 137, | |
1311 | /* 10 */ 110, 87, 70, 56, 45, | |
1312 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1313 | }; |
1314 | ||
5714d2de IM |
1315 | /* |
1316 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1317 | * | |
1318 | * In cases where the weight does not change often, we can use the | |
1319 | * precalculated inverse to speed up arithmetics by turning divisions | |
1320 | * into multiplications: | |
1321 | */ | |
dd41f596 | 1322 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1323 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1324 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1325 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1326 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1327 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1328 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1329 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1330 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1331 | }; |
2dd73a4f | 1332 | |
dd41f596 IM |
1333 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1334 | ||
1335 | /* | |
1336 | * runqueue iterator, to support SMP load-balancing between different | |
1337 | * scheduling classes, without having to expose their internal data | |
1338 | * structures to the load-balancing proper: | |
1339 | */ | |
1340 | struct rq_iterator { | |
1341 | void *arg; | |
1342 | struct task_struct *(*start)(void *); | |
1343 | struct task_struct *(*next)(void *); | |
1344 | }; | |
1345 | ||
e1d1484f PW |
1346 | #ifdef CONFIG_SMP |
1347 | static unsigned long | |
1348 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1349 | unsigned long max_load_move, struct sched_domain *sd, | |
1350 | enum cpu_idle_type idle, int *all_pinned, | |
1351 | int *this_best_prio, struct rq_iterator *iterator); | |
1352 | ||
1353 | static int | |
1354 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1355 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1356 | struct rq_iterator *iterator); | |
e1d1484f | 1357 | #endif |
dd41f596 | 1358 | |
d842de87 SV |
1359 | #ifdef CONFIG_CGROUP_CPUACCT |
1360 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1361 | #else | |
1362 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1363 | #endif | |
1364 | ||
18d95a28 PZ |
1365 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1366 | { | |
1367 | update_load_add(&rq->load, load); | |
1368 | } | |
1369 | ||
1370 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1371 | { | |
1372 | update_load_sub(&rq->load, load); | |
1373 | } | |
1374 | ||
7940ca36 | 1375 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1376 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1377 | |
1378 | /* | |
1379 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1380 | * leaving it for the final time. | |
1381 | */ | |
eb755805 | 1382 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1383 | { |
1384 | struct task_group *parent, *child; | |
eb755805 | 1385 | int ret; |
c09595f6 PZ |
1386 | |
1387 | rcu_read_lock(); | |
1388 | parent = &root_task_group; | |
1389 | down: | |
eb755805 PZ |
1390 | ret = (*down)(parent, data); |
1391 | if (ret) | |
1392 | goto out_unlock; | |
c09595f6 PZ |
1393 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1394 | parent = child; | |
1395 | goto down; | |
1396 | ||
1397 | up: | |
1398 | continue; | |
1399 | } | |
eb755805 PZ |
1400 | ret = (*up)(parent, data); |
1401 | if (ret) | |
1402 | goto out_unlock; | |
c09595f6 PZ |
1403 | |
1404 | child = parent; | |
1405 | parent = parent->parent; | |
1406 | if (parent) | |
1407 | goto up; | |
eb755805 | 1408 | out_unlock: |
c09595f6 | 1409 | rcu_read_unlock(); |
eb755805 PZ |
1410 | |
1411 | return ret; | |
c09595f6 PZ |
1412 | } |
1413 | ||
eb755805 PZ |
1414 | static int tg_nop(struct task_group *tg, void *data) |
1415 | { | |
1416 | return 0; | |
c09595f6 | 1417 | } |
eb755805 PZ |
1418 | #endif |
1419 | ||
1420 | #ifdef CONFIG_SMP | |
1421 | static unsigned long source_load(int cpu, int type); | |
1422 | static unsigned long target_load(int cpu, int type); | |
1423 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1424 | ||
1425 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1426 | { | |
1427 | struct rq *rq = cpu_rq(cpu); | |
1428 | ||
1429 | if (rq->nr_running) | |
1430 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1431 | ||
1432 | return rq->avg_load_per_task; | |
1433 | } | |
1434 | ||
1435 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1436 | |
c09595f6 PZ |
1437 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1438 | ||
1439 | /* | |
1440 | * Calculate and set the cpu's group shares. | |
1441 | */ | |
1442 | static void | |
ffda12a1 PZ |
1443 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1444 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1445 | { |
c09595f6 PZ |
1446 | int boost = 0; |
1447 | unsigned long shares; | |
1448 | unsigned long rq_weight; | |
1449 | ||
c8cba857 | 1450 | if (!tg->se[cpu]) |
c09595f6 PZ |
1451 | return; |
1452 | ||
c8cba857 | 1453 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1454 | |
1455 | /* | |
1456 | * If there are currently no tasks on the cpu pretend there is one of | |
1457 | * average load so that when a new task gets to run here it will not | |
1458 | * get delayed by group starvation. | |
1459 | */ | |
1460 | if (!rq_weight) { | |
1461 | boost = 1; | |
1462 | rq_weight = NICE_0_LOAD; | |
1463 | } | |
1464 | ||
c8cba857 PZ |
1465 | if (unlikely(rq_weight > sd_rq_weight)) |
1466 | rq_weight = sd_rq_weight; | |
1467 | ||
c09595f6 PZ |
1468 | /* |
1469 | * \Sum shares * rq_weight | |
1470 | * shares = ----------------------- | |
1471 | * \Sum rq_weight | |
1472 | * | |
1473 | */ | |
c8cba857 | 1474 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
ffda12a1 | 1475 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1476 | |
ffda12a1 PZ |
1477 | if (abs(shares - tg->se[cpu]->load.weight) > |
1478 | sysctl_sched_shares_thresh) { | |
1479 | struct rq *rq = cpu_rq(cpu); | |
1480 | unsigned long flags; | |
c09595f6 | 1481 | |
ffda12a1 PZ |
1482 | spin_lock_irqsave(&rq->lock, flags); |
1483 | /* | |
1484 | * record the actual number of shares, not the boosted amount. | |
1485 | */ | |
1486 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; | |
1487 | tg->cfs_rq[cpu]->rq_weight = rq_weight; | |
c09595f6 | 1488 | |
ffda12a1 PZ |
1489 | __set_se_shares(tg->se[cpu], shares); |
1490 | spin_unlock_irqrestore(&rq->lock, flags); | |
1491 | } | |
18d95a28 | 1492 | } |
c09595f6 PZ |
1493 | |
1494 | /* | |
c8cba857 PZ |
1495 | * Re-compute the task group their per cpu shares over the given domain. |
1496 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1497 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1498 | */ |
eb755805 | 1499 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1500 | { |
c8cba857 PZ |
1501 | unsigned long rq_weight = 0; |
1502 | unsigned long shares = 0; | |
eb755805 | 1503 | struct sched_domain *sd = data; |
c8cba857 | 1504 | int i; |
c09595f6 | 1505 | |
c8cba857 PZ |
1506 | for_each_cpu_mask(i, sd->span) { |
1507 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1508 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1509 | } |
c09595f6 | 1510 | |
c8cba857 PZ |
1511 | if ((!shares && rq_weight) || shares > tg->shares) |
1512 | shares = tg->shares; | |
1513 | ||
1514 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1515 | shares = tg->shares; | |
c09595f6 | 1516 | |
cd80917e PZ |
1517 | if (!rq_weight) |
1518 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1519 | ||
ffda12a1 PZ |
1520 | for_each_cpu_mask(i, sd->span) |
1521 | update_group_shares_cpu(tg, i, shares, rq_weight); | |
eb755805 PZ |
1522 | |
1523 | return 0; | |
c09595f6 PZ |
1524 | } |
1525 | ||
1526 | /* | |
c8cba857 PZ |
1527 | * Compute the cpu's hierarchical load factor for each task group. |
1528 | * This needs to be done in a top-down fashion because the load of a child | |
1529 | * group is a fraction of its parents load. | |
c09595f6 | 1530 | */ |
eb755805 | 1531 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1532 | { |
c8cba857 | 1533 | unsigned long load; |
eb755805 | 1534 | long cpu = (long)data; |
c09595f6 | 1535 | |
c8cba857 PZ |
1536 | if (!tg->parent) { |
1537 | load = cpu_rq(cpu)->load.weight; | |
1538 | } else { | |
1539 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1540 | load *= tg->cfs_rq[cpu]->shares; | |
1541 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1542 | } | |
c09595f6 | 1543 | |
c8cba857 | 1544 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1545 | |
eb755805 | 1546 | return 0; |
c09595f6 PZ |
1547 | } |
1548 | ||
c8cba857 | 1549 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1550 | { |
2398f2c6 PZ |
1551 | u64 now = cpu_clock(raw_smp_processor_id()); |
1552 | s64 elapsed = now - sd->last_update; | |
1553 | ||
1554 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1555 | sd->last_update = now; | |
eb755805 | 1556 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1557 | } |
4d8d595d PZ |
1558 | } |
1559 | ||
3e5459b4 PZ |
1560 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1561 | { | |
1562 | spin_unlock(&rq->lock); | |
1563 | update_shares(sd); | |
1564 | spin_lock(&rq->lock); | |
1565 | } | |
1566 | ||
eb755805 | 1567 | static void update_h_load(long cpu) |
c09595f6 | 1568 | { |
eb755805 | 1569 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1570 | } |
1571 | ||
c09595f6 PZ |
1572 | #else |
1573 | ||
c8cba857 | 1574 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1575 | { |
1576 | } | |
1577 | ||
3e5459b4 PZ |
1578 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1579 | { | |
1580 | } | |
1581 | ||
18d95a28 PZ |
1582 | #endif |
1583 | ||
18d95a28 PZ |
1584 | #endif |
1585 | ||
30432094 | 1586 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1587 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1588 | { | |
30432094 | 1589 | #ifdef CONFIG_SMP |
34e83e85 IM |
1590 | cfs_rq->shares = shares; |
1591 | #endif | |
1592 | } | |
30432094 | 1593 | #endif |
e7693a36 | 1594 | |
dd41f596 | 1595 | #include "sched_stats.h" |
dd41f596 | 1596 | #include "sched_idletask.c" |
5522d5d5 IM |
1597 | #include "sched_fair.c" |
1598 | #include "sched_rt.c" | |
dd41f596 IM |
1599 | #ifdef CONFIG_SCHED_DEBUG |
1600 | # include "sched_debug.c" | |
1601 | #endif | |
1602 | ||
1603 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1604 | #define for_each_class(class) \ |
1605 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1606 | |
c09595f6 | 1607 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1608 | { |
1609 | rq->nr_running++; | |
9c217245 IM |
1610 | } |
1611 | ||
c09595f6 | 1612 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1613 | { |
1614 | rq->nr_running--; | |
9c217245 IM |
1615 | } |
1616 | ||
45bf76df IM |
1617 | static void set_load_weight(struct task_struct *p) |
1618 | { | |
1619 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1620 | p->se.load.weight = prio_to_weight[0] * 2; |
1621 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1622 | return; | |
1623 | } | |
45bf76df | 1624 | |
dd41f596 IM |
1625 | /* |
1626 | * SCHED_IDLE tasks get minimal weight: | |
1627 | */ | |
1628 | if (p->policy == SCHED_IDLE) { | |
1629 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1630 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1631 | return; | |
1632 | } | |
71f8bd46 | 1633 | |
dd41f596 IM |
1634 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1635 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1636 | } |
1637 | ||
2087a1ad GH |
1638 | static void update_avg(u64 *avg, u64 sample) |
1639 | { | |
1640 | s64 diff = sample - *avg; | |
1641 | *avg += diff >> 3; | |
1642 | } | |
1643 | ||
8159f87e | 1644 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1645 | { |
dd41f596 | 1646 | sched_info_queued(p); |
fd390f6a | 1647 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1648 | p->se.on_rq = 1; |
71f8bd46 IM |
1649 | } |
1650 | ||
69be72c1 | 1651 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1652 | { |
2087a1ad GH |
1653 | if (sleep && p->se.last_wakeup) { |
1654 | update_avg(&p->se.avg_overlap, | |
1655 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1656 | p->se.last_wakeup = 0; | |
1657 | } | |
1658 | ||
46ac22ba | 1659 | sched_info_dequeued(p); |
f02231e5 | 1660 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1661 | p->se.on_rq = 0; |
71f8bd46 IM |
1662 | } |
1663 | ||
14531189 | 1664 | /* |
dd41f596 | 1665 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1666 | */ |
14531189 IM |
1667 | static inline int __normal_prio(struct task_struct *p) |
1668 | { | |
dd41f596 | 1669 | return p->static_prio; |
14531189 IM |
1670 | } |
1671 | ||
b29739f9 IM |
1672 | /* |
1673 | * Calculate the expected normal priority: i.e. priority | |
1674 | * without taking RT-inheritance into account. Might be | |
1675 | * boosted by interactivity modifiers. Changes upon fork, | |
1676 | * setprio syscalls, and whenever the interactivity | |
1677 | * estimator recalculates. | |
1678 | */ | |
36c8b586 | 1679 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1680 | { |
1681 | int prio; | |
1682 | ||
e05606d3 | 1683 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1684 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1685 | else | |
1686 | prio = __normal_prio(p); | |
1687 | return prio; | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Calculate the current priority, i.e. the priority | |
1692 | * taken into account by the scheduler. This value might | |
1693 | * be boosted by RT tasks, or might be boosted by | |
1694 | * interactivity modifiers. Will be RT if the task got | |
1695 | * RT-boosted. If not then it returns p->normal_prio. | |
1696 | */ | |
36c8b586 | 1697 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1698 | { |
1699 | p->normal_prio = normal_prio(p); | |
1700 | /* | |
1701 | * If we are RT tasks or we were boosted to RT priority, | |
1702 | * keep the priority unchanged. Otherwise, update priority | |
1703 | * to the normal priority: | |
1704 | */ | |
1705 | if (!rt_prio(p->prio)) | |
1706 | return p->normal_prio; | |
1707 | return p->prio; | |
1708 | } | |
1709 | ||
1da177e4 | 1710 | /* |
dd41f596 | 1711 | * activate_task - move a task to the runqueue. |
1da177e4 | 1712 | */ |
dd41f596 | 1713 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1714 | { |
d9514f6c | 1715 | if (task_contributes_to_load(p)) |
dd41f596 | 1716 | rq->nr_uninterruptible--; |
1da177e4 | 1717 | |
8159f87e | 1718 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1719 | inc_nr_running(rq); |
1da177e4 LT |
1720 | } |
1721 | ||
1da177e4 LT |
1722 | /* |
1723 | * deactivate_task - remove a task from the runqueue. | |
1724 | */ | |
2e1cb74a | 1725 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1726 | { |
d9514f6c | 1727 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1728 | rq->nr_uninterruptible++; |
1729 | ||
69be72c1 | 1730 | dequeue_task(rq, p, sleep); |
c09595f6 | 1731 | dec_nr_running(rq); |
1da177e4 LT |
1732 | } |
1733 | ||
1da177e4 LT |
1734 | /** |
1735 | * task_curr - is this task currently executing on a CPU? | |
1736 | * @p: the task in question. | |
1737 | */ | |
36c8b586 | 1738 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1739 | { |
1740 | return cpu_curr(task_cpu(p)) == p; | |
1741 | } | |
1742 | ||
dd41f596 IM |
1743 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1744 | { | |
6f505b16 | 1745 | set_task_rq(p, cpu); |
dd41f596 | 1746 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1747 | /* |
1748 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1749 | * successfuly executed on another CPU. We must ensure that updates of | |
1750 | * per-task data have been completed by this moment. | |
1751 | */ | |
1752 | smp_wmb(); | |
dd41f596 | 1753 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1754 | #endif |
2dd73a4f PW |
1755 | } |
1756 | ||
cb469845 SR |
1757 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1758 | const struct sched_class *prev_class, | |
1759 | int oldprio, int running) | |
1760 | { | |
1761 | if (prev_class != p->sched_class) { | |
1762 | if (prev_class->switched_from) | |
1763 | prev_class->switched_from(rq, p, running); | |
1764 | p->sched_class->switched_to(rq, p, running); | |
1765 | } else | |
1766 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1767 | } | |
1768 | ||
1da177e4 | 1769 | #ifdef CONFIG_SMP |
c65cc870 | 1770 | |
e958b360 TG |
1771 | /* Used instead of source_load when we know the type == 0 */ |
1772 | static unsigned long weighted_cpuload(const int cpu) | |
1773 | { | |
1774 | return cpu_rq(cpu)->load.weight; | |
1775 | } | |
1776 | ||
cc367732 IM |
1777 | /* |
1778 | * Is this task likely cache-hot: | |
1779 | */ | |
e7693a36 | 1780 | static int |
cc367732 IM |
1781 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1782 | { | |
1783 | s64 delta; | |
1784 | ||
f540a608 IM |
1785 | /* |
1786 | * Buddy candidates are cache hot: | |
1787 | */ | |
d25ce4cd | 1788 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1789 | return 1; |
1790 | ||
cc367732 IM |
1791 | if (p->sched_class != &fair_sched_class) |
1792 | return 0; | |
1793 | ||
6bc1665b IM |
1794 | if (sysctl_sched_migration_cost == -1) |
1795 | return 1; | |
1796 | if (sysctl_sched_migration_cost == 0) | |
1797 | return 0; | |
1798 | ||
cc367732 IM |
1799 | delta = now - p->se.exec_start; |
1800 | ||
1801 | return delta < (s64)sysctl_sched_migration_cost; | |
1802 | } | |
1803 | ||
1804 | ||
dd41f596 | 1805 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1806 | { |
dd41f596 IM |
1807 | int old_cpu = task_cpu(p); |
1808 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1809 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1810 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1811 | u64 clock_offset; |
dd41f596 IM |
1812 | |
1813 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1814 | |
1815 | #ifdef CONFIG_SCHEDSTATS | |
1816 | if (p->se.wait_start) | |
1817 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1818 | if (p->se.sleep_start) |
1819 | p->se.sleep_start -= clock_offset; | |
1820 | if (p->se.block_start) | |
1821 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1822 | if (old_cpu != new_cpu) { |
1823 | schedstat_inc(p, se.nr_migrations); | |
1824 | if (task_hot(p, old_rq->clock, NULL)) | |
1825 | schedstat_inc(p, se.nr_forced2_migrations); | |
1826 | } | |
6cfb0d5d | 1827 | #endif |
2830cf8c SV |
1828 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1829 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1830 | |
1831 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1832 | } |
1833 | ||
70b97a7f | 1834 | struct migration_req { |
1da177e4 | 1835 | struct list_head list; |
1da177e4 | 1836 | |
36c8b586 | 1837 | struct task_struct *task; |
1da177e4 LT |
1838 | int dest_cpu; |
1839 | ||
1da177e4 | 1840 | struct completion done; |
70b97a7f | 1841 | }; |
1da177e4 LT |
1842 | |
1843 | /* | |
1844 | * The task's runqueue lock must be held. | |
1845 | * Returns true if you have to wait for migration thread. | |
1846 | */ | |
36c8b586 | 1847 | static int |
70b97a7f | 1848 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1849 | { |
70b97a7f | 1850 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1851 | |
1852 | /* | |
1853 | * If the task is not on a runqueue (and not running), then | |
1854 | * it is sufficient to simply update the task's cpu field. | |
1855 | */ | |
dd41f596 | 1856 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1857 | set_task_cpu(p, dest_cpu); |
1858 | return 0; | |
1859 | } | |
1860 | ||
1861 | init_completion(&req->done); | |
1da177e4 LT |
1862 | req->task = p; |
1863 | req->dest_cpu = dest_cpu; | |
1864 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1865 | |
1da177e4 LT |
1866 | return 1; |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * wait_task_inactive - wait for a thread to unschedule. | |
1871 | * | |
85ba2d86 RM |
1872 | * If @match_state is nonzero, it's the @p->state value just checked and |
1873 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1874 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1875 | * we return a positive number (its total switch count). If a second call | |
1876 | * a short while later returns the same number, the caller can be sure that | |
1877 | * @p has remained unscheduled the whole time. | |
1878 | * | |
1da177e4 LT |
1879 | * The caller must ensure that the task *will* unschedule sometime soon, |
1880 | * else this function might spin for a *long* time. This function can't | |
1881 | * be called with interrupts off, or it may introduce deadlock with | |
1882 | * smp_call_function() if an IPI is sent by the same process we are | |
1883 | * waiting to become inactive. | |
1884 | */ | |
85ba2d86 | 1885 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1886 | { |
1887 | unsigned long flags; | |
dd41f596 | 1888 | int running, on_rq; |
85ba2d86 | 1889 | unsigned long ncsw; |
70b97a7f | 1890 | struct rq *rq; |
1da177e4 | 1891 | |
3a5c359a AK |
1892 | for (;;) { |
1893 | /* | |
1894 | * We do the initial early heuristics without holding | |
1895 | * any task-queue locks at all. We'll only try to get | |
1896 | * the runqueue lock when things look like they will | |
1897 | * work out! | |
1898 | */ | |
1899 | rq = task_rq(p); | |
fa490cfd | 1900 | |
3a5c359a AK |
1901 | /* |
1902 | * If the task is actively running on another CPU | |
1903 | * still, just relax and busy-wait without holding | |
1904 | * any locks. | |
1905 | * | |
1906 | * NOTE! Since we don't hold any locks, it's not | |
1907 | * even sure that "rq" stays as the right runqueue! | |
1908 | * But we don't care, since "task_running()" will | |
1909 | * return false if the runqueue has changed and p | |
1910 | * is actually now running somewhere else! | |
1911 | */ | |
85ba2d86 RM |
1912 | while (task_running(rq, p)) { |
1913 | if (match_state && unlikely(p->state != match_state)) | |
1914 | return 0; | |
3a5c359a | 1915 | cpu_relax(); |
85ba2d86 | 1916 | } |
fa490cfd | 1917 | |
3a5c359a AK |
1918 | /* |
1919 | * Ok, time to look more closely! We need the rq | |
1920 | * lock now, to be *sure*. If we're wrong, we'll | |
1921 | * just go back and repeat. | |
1922 | */ | |
1923 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 1924 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
1925 | running = task_running(rq, p); |
1926 | on_rq = p->se.on_rq; | |
85ba2d86 | 1927 | ncsw = 0; |
f31e11d8 | 1928 | if (!match_state || p->state == match_state) |
93dcf55f | 1929 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1930 | task_rq_unlock(rq, &flags); |
fa490cfd | 1931 | |
85ba2d86 RM |
1932 | /* |
1933 | * If it changed from the expected state, bail out now. | |
1934 | */ | |
1935 | if (unlikely(!ncsw)) | |
1936 | break; | |
1937 | ||
3a5c359a AK |
1938 | /* |
1939 | * Was it really running after all now that we | |
1940 | * checked with the proper locks actually held? | |
1941 | * | |
1942 | * Oops. Go back and try again.. | |
1943 | */ | |
1944 | if (unlikely(running)) { | |
1945 | cpu_relax(); | |
1946 | continue; | |
1947 | } | |
fa490cfd | 1948 | |
3a5c359a AK |
1949 | /* |
1950 | * It's not enough that it's not actively running, | |
1951 | * it must be off the runqueue _entirely_, and not | |
1952 | * preempted! | |
1953 | * | |
1954 | * So if it wa still runnable (but just not actively | |
1955 | * running right now), it's preempted, and we should | |
1956 | * yield - it could be a while. | |
1957 | */ | |
1958 | if (unlikely(on_rq)) { | |
1959 | schedule_timeout_uninterruptible(1); | |
1960 | continue; | |
1961 | } | |
fa490cfd | 1962 | |
3a5c359a AK |
1963 | /* |
1964 | * Ahh, all good. It wasn't running, and it wasn't | |
1965 | * runnable, which means that it will never become | |
1966 | * running in the future either. We're all done! | |
1967 | */ | |
1968 | break; | |
1969 | } | |
85ba2d86 RM |
1970 | |
1971 | return ncsw; | |
1da177e4 LT |
1972 | } |
1973 | ||
1974 | /*** | |
1975 | * kick_process - kick a running thread to enter/exit the kernel | |
1976 | * @p: the to-be-kicked thread | |
1977 | * | |
1978 | * Cause a process which is running on another CPU to enter | |
1979 | * kernel-mode, without any delay. (to get signals handled.) | |
1980 | * | |
1981 | * NOTE: this function doesnt have to take the runqueue lock, | |
1982 | * because all it wants to ensure is that the remote task enters | |
1983 | * the kernel. If the IPI races and the task has been migrated | |
1984 | * to another CPU then no harm is done and the purpose has been | |
1985 | * achieved as well. | |
1986 | */ | |
36c8b586 | 1987 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1988 | { |
1989 | int cpu; | |
1990 | ||
1991 | preempt_disable(); | |
1992 | cpu = task_cpu(p); | |
1993 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1994 | smp_send_reschedule(cpu); | |
1995 | preempt_enable(); | |
1996 | } | |
1997 | ||
1998 | /* | |
2dd73a4f PW |
1999 | * Return a low guess at the load of a migration-source cpu weighted |
2000 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2001 | * |
2002 | * We want to under-estimate the load of migration sources, to | |
2003 | * balance conservatively. | |
2004 | */ | |
a9957449 | 2005 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2006 | { |
70b97a7f | 2007 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2008 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2009 | |
93b75217 | 2010 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2011 | return total; |
b910472d | 2012 | |
dd41f596 | 2013 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2014 | } |
2015 | ||
2016 | /* | |
2dd73a4f PW |
2017 | * Return a high guess at the load of a migration-target cpu weighted |
2018 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2019 | */ |
a9957449 | 2020 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2021 | { |
70b97a7f | 2022 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2023 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2024 | |
93b75217 | 2025 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2026 | return total; |
3b0bd9bc | 2027 | |
dd41f596 | 2028 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2029 | } |
2030 | ||
147cbb4b NP |
2031 | /* |
2032 | * find_idlest_group finds and returns the least busy CPU group within the | |
2033 | * domain. | |
2034 | */ | |
2035 | static struct sched_group * | |
2036 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2037 | { | |
2038 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2039 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2040 | int load_idx = sd->forkexec_idx; | |
2041 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2042 | ||
2043 | do { | |
2044 | unsigned long load, avg_load; | |
2045 | int local_group; | |
2046 | int i; | |
2047 | ||
da5a5522 BD |
2048 | /* Skip over this group if it has no CPUs allowed */ |
2049 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2050 | continue; |
da5a5522 | 2051 | |
147cbb4b | 2052 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2053 | |
2054 | /* Tally up the load of all CPUs in the group */ | |
2055 | avg_load = 0; | |
2056 | ||
363ab6f1 | 2057 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2058 | /* Bias balancing toward cpus of our domain */ |
2059 | if (local_group) | |
2060 | load = source_load(i, load_idx); | |
2061 | else | |
2062 | load = target_load(i, load_idx); | |
2063 | ||
2064 | avg_load += load; | |
2065 | } | |
2066 | ||
2067 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2068 | avg_load = sg_div_cpu_power(group, |
2069 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2070 | |
2071 | if (local_group) { | |
2072 | this_load = avg_load; | |
2073 | this = group; | |
2074 | } else if (avg_load < min_load) { | |
2075 | min_load = avg_load; | |
2076 | idlest = group; | |
2077 | } | |
3a5c359a | 2078 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2079 | |
2080 | if (!idlest || 100*this_load < imbalance*min_load) | |
2081 | return NULL; | |
2082 | return idlest; | |
2083 | } | |
2084 | ||
2085 | /* | |
0feaece9 | 2086 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2087 | */ |
95cdf3b7 | 2088 | static int |
7c16ec58 MT |
2089 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2090 | cpumask_t *tmp) | |
147cbb4b NP |
2091 | { |
2092 | unsigned long load, min_load = ULONG_MAX; | |
2093 | int idlest = -1; | |
2094 | int i; | |
2095 | ||
da5a5522 | 2096 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2097 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2098 | |
363ab6f1 | 2099 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2100 | load = weighted_cpuload(i); |
147cbb4b NP |
2101 | |
2102 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2103 | min_load = load; | |
2104 | idlest = i; | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | return idlest; | |
2109 | } | |
2110 | ||
476d139c NP |
2111 | /* |
2112 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2113 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2114 | * SD_BALANCE_EXEC. | |
2115 | * | |
2116 | * Balance, ie. select the least loaded group. | |
2117 | * | |
2118 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2119 | * | |
2120 | * preempt must be disabled. | |
2121 | */ | |
2122 | static int sched_balance_self(int cpu, int flag) | |
2123 | { | |
2124 | struct task_struct *t = current; | |
2125 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2126 | |
c96d145e | 2127 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2128 | /* |
2129 | * If power savings logic is enabled for a domain, stop there. | |
2130 | */ | |
5c45bf27 SS |
2131 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2132 | break; | |
476d139c NP |
2133 | if (tmp->flags & flag) |
2134 | sd = tmp; | |
c96d145e | 2135 | } |
476d139c | 2136 | |
039a1c41 PZ |
2137 | if (sd) |
2138 | update_shares(sd); | |
2139 | ||
476d139c | 2140 | while (sd) { |
7c16ec58 | 2141 | cpumask_t span, tmpmask; |
476d139c | 2142 | struct sched_group *group; |
1a848870 SS |
2143 | int new_cpu, weight; |
2144 | ||
2145 | if (!(sd->flags & flag)) { | |
2146 | sd = sd->child; | |
2147 | continue; | |
2148 | } | |
476d139c NP |
2149 | |
2150 | span = sd->span; | |
2151 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2152 | if (!group) { |
2153 | sd = sd->child; | |
2154 | continue; | |
2155 | } | |
476d139c | 2156 | |
7c16ec58 | 2157 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2158 | if (new_cpu == -1 || new_cpu == cpu) { |
2159 | /* Now try balancing at a lower domain level of cpu */ | |
2160 | sd = sd->child; | |
2161 | continue; | |
2162 | } | |
476d139c | 2163 | |
1a848870 | 2164 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2165 | cpu = new_cpu; |
476d139c NP |
2166 | sd = NULL; |
2167 | weight = cpus_weight(span); | |
2168 | for_each_domain(cpu, tmp) { | |
2169 | if (weight <= cpus_weight(tmp->span)) | |
2170 | break; | |
2171 | if (tmp->flags & flag) | |
2172 | sd = tmp; | |
2173 | } | |
2174 | /* while loop will break here if sd == NULL */ | |
2175 | } | |
2176 | ||
2177 | return cpu; | |
2178 | } | |
2179 | ||
2180 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2181 | |
1da177e4 LT |
2182 | /*** |
2183 | * try_to_wake_up - wake up a thread | |
2184 | * @p: the to-be-woken-up thread | |
2185 | * @state: the mask of task states that can be woken | |
2186 | * @sync: do a synchronous wakeup? | |
2187 | * | |
2188 | * Put it on the run-queue if it's not already there. The "current" | |
2189 | * thread is always on the run-queue (except when the actual | |
2190 | * re-schedule is in progress), and as such you're allowed to do | |
2191 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2192 | * runnable without the overhead of this. | |
2193 | * | |
2194 | * returns failure only if the task is already active. | |
2195 | */ | |
36c8b586 | 2196 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2197 | { |
cc367732 | 2198 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2199 | unsigned long flags; |
2200 | long old_state; | |
70b97a7f | 2201 | struct rq *rq; |
1da177e4 | 2202 | |
b85d0667 IM |
2203 | if (!sched_feat(SYNC_WAKEUPS)) |
2204 | sync = 0; | |
2205 | ||
2398f2c6 PZ |
2206 | #ifdef CONFIG_SMP |
2207 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2208 | struct sched_domain *sd; | |
2209 | ||
2210 | this_cpu = raw_smp_processor_id(); | |
2211 | cpu = task_cpu(p); | |
2212 | ||
2213 | for_each_domain(this_cpu, sd) { | |
2214 | if (cpu_isset(cpu, sd->span)) { | |
2215 | update_shares(sd); | |
2216 | break; | |
2217 | } | |
2218 | } | |
2219 | } | |
2220 | #endif | |
2221 | ||
04e2f174 | 2222 | smp_wmb(); |
1da177e4 LT |
2223 | rq = task_rq_lock(p, &flags); |
2224 | old_state = p->state; | |
2225 | if (!(old_state & state)) | |
2226 | goto out; | |
2227 | ||
dd41f596 | 2228 | if (p->se.on_rq) |
1da177e4 LT |
2229 | goto out_running; |
2230 | ||
2231 | cpu = task_cpu(p); | |
cc367732 | 2232 | orig_cpu = cpu; |
1da177e4 LT |
2233 | this_cpu = smp_processor_id(); |
2234 | ||
2235 | #ifdef CONFIG_SMP | |
2236 | if (unlikely(task_running(rq, p))) | |
2237 | goto out_activate; | |
2238 | ||
5d2f5a61 DA |
2239 | cpu = p->sched_class->select_task_rq(p, sync); |
2240 | if (cpu != orig_cpu) { | |
2241 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2242 | task_rq_unlock(rq, &flags); |
2243 | /* might preempt at this point */ | |
2244 | rq = task_rq_lock(p, &flags); | |
2245 | old_state = p->state; | |
2246 | if (!(old_state & state)) | |
2247 | goto out; | |
dd41f596 | 2248 | if (p->se.on_rq) |
1da177e4 LT |
2249 | goto out_running; |
2250 | ||
2251 | this_cpu = smp_processor_id(); | |
2252 | cpu = task_cpu(p); | |
2253 | } | |
2254 | ||
e7693a36 GH |
2255 | #ifdef CONFIG_SCHEDSTATS |
2256 | schedstat_inc(rq, ttwu_count); | |
2257 | if (cpu == this_cpu) | |
2258 | schedstat_inc(rq, ttwu_local); | |
2259 | else { | |
2260 | struct sched_domain *sd; | |
2261 | for_each_domain(this_cpu, sd) { | |
2262 | if (cpu_isset(cpu, sd->span)) { | |
2263 | schedstat_inc(sd, ttwu_wake_remote); | |
2264 | break; | |
2265 | } | |
2266 | } | |
2267 | } | |
6d6bc0ad | 2268 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2269 | |
1da177e4 LT |
2270 | out_activate: |
2271 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2272 | schedstat_inc(p, se.nr_wakeups); |
2273 | if (sync) | |
2274 | schedstat_inc(p, se.nr_wakeups_sync); | |
2275 | if (orig_cpu != cpu) | |
2276 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2277 | if (cpu == this_cpu) | |
2278 | schedstat_inc(p, se.nr_wakeups_local); | |
2279 | else | |
2280 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2281 | update_rq_clock(rq); |
dd41f596 | 2282 | activate_task(rq, p, 1); |
1da177e4 LT |
2283 | success = 1; |
2284 | ||
2285 | out_running: | |
0a16b607 | 2286 | trace_sched_wakeup(rq, p); |
15afe09b | 2287 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2288 | |
1da177e4 | 2289 | p->state = TASK_RUNNING; |
9a897c5a SR |
2290 | #ifdef CONFIG_SMP |
2291 | if (p->sched_class->task_wake_up) | |
2292 | p->sched_class->task_wake_up(rq, p); | |
2293 | #endif | |
1da177e4 | 2294 | out: |
2087a1ad GH |
2295 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2296 | ||
1da177e4 LT |
2297 | task_rq_unlock(rq, &flags); |
2298 | ||
2299 | return success; | |
2300 | } | |
2301 | ||
7ad5b3a5 | 2302 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2303 | { |
d9514f6c | 2304 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2305 | } |
1da177e4 LT |
2306 | EXPORT_SYMBOL(wake_up_process); |
2307 | ||
7ad5b3a5 | 2308 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2309 | { |
2310 | return try_to_wake_up(p, state, 0); | |
2311 | } | |
2312 | ||
1da177e4 LT |
2313 | /* |
2314 | * Perform scheduler related setup for a newly forked process p. | |
2315 | * p is forked by current. | |
dd41f596 IM |
2316 | * |
2317 | * __sched_fork() is basic setup used by init_idle() too: | |
2318 | */ | |
2319 | static void __sched_fork(struct task_struct *p) | |
2320 | { | |
dd41f596 IM |
2321 | p->se.exec_start = 0; |
2322 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2323 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2324 | p->se.last_wakeup = 0; |
2325 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2326 | |
2327 | #ifdef CONFIG_SCHEDSTATS | |
2328 | p->se.wait_start = 0; | |
dd41f596 IM |
2329 | p->se.sum_sleep_runtime = 0; |
2330 | p->se.sleep_start = 0; | |
dd41f596 IM |
2331 | p->se.block_start = 0; |
2332 | p->se.sleep_max = 0; | |
2333 | p->se.block_max = 0; | |
2334 | p->se.exec_max = 0; | |
eba1ed4b | 2335 | p->se.slice_max = 0; |
dd41f596 | 2336 | p->se.wait_max = 0; |
6cfb0d5d | 2337 | #endif |
476d139c | 2338 | |
fa717060 | 2339 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2340 | p->se.on_rq = 0; |
4a55bd5e | 2341 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2342 | |
e107be36 AK |
2343 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2344 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2345 | #endif | |
2346 | ||
1da177e4 LT |
2347 | /* |
2348 | * We mark the process as running here, but have not actually | |
2349 | * inserted it onto the runqueue yet. This guarantees that | |
2350 | * nobody will actually run it, and a signal or other external | |
2351 | * event cannot wake it up and insert it on the runqueue either. | |
2352 | */ | |
2353 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2354 | } |
2355 | ||
2356 | /* | |
2357 | * fork()/clone()-time setup: | |
2358 | */ | |
2359 | void sched_fork(struct task_struct *p, int clone_flags) | |
2360 | { | |
2361 | int cpu = get_cpu(); | |
2362 | ||
2363 | __sched_fork(p); | |
2364 | ||
2365 | #ifdef CONFIG_SMP | |
2366 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2367 | #endif | |
02e4bac2 | 2368 | set_task_cpu(p, cpu); |
b29739f9 IM |
2369 | |
2370 | /* | |
2371 | * Make sure we do not leak PI boosting priority to the child: | |
2372 | */ | |
2373 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2374 | if (!rt_prio(p->prio)) |
2375 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2376 | |
52f17b6c | 2377 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2378 | if (likely(sched_info_on())) |
52f17b6c | 2379 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2380 | #endif |
d6077cb8 | 2381 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2382 | p->oncpu = 0; |
2383 | #endif | |
1da177e4 | 2384 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2385 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2386 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2387 | #endif |
476d139c | 2388 | put_cpu(); |
1da177e4 LT |
2389 | } |
2390 | ||
2391 | /* | |
2392 | * wake_up_new_task - wake up a newly created task for the first time. | |
2393 | * | |
2394 | * This function will do some initial scheduler statistics housekeeping | |
2395 | * that must be done for every newly created context, then puts the task | |
2396 | * on the runqueue and wakes it. | |
2397 | */ | |
7ad5b3a5 | 2398 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2399 | { |
2400 | unsigned long flags; | |
dd41f596 | 2401 | struct rq *rq; |
1da177e4 LT |
2402 | |
2403 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2404 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2405 | update_rq_clock(rq); |
1da177e4 LT |
2406 | |
2407 | p->prio = effective_prio(p); | |
2408 | ||
b9dca1e0 | 2409 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2410 | activate_task(rq, p, 0); |
1da177e4 | 2411 | } else { |
1da177e4 | 2412 | /* |
dd41f596 IM |
2413 | * Let the scheduling class do new task startup |
2414 | * management (if any): | |
1da177e4 | 2415 | */ |
ee0827d8 | 2416 | p->sched_class->task_new(rq, p); |
c09595f6 | 2417 | inc_nr_running(rq); |
1da177e4 | 2418 | } |
0a16b607 | 2419 | trace_sched_wakeup_new(rq, p); |
15afe09b | 2420 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2421 | #ifdef CONFIG_SMP |
2422 | if (p->sched_class->task_wake_up) | |
2423 | p->sched_class->task_wake_up(rq, p); | |
2424 | #endif | |
dd41f596 | 2425 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2426 | } |
2427 | ||
e107be36 AK |
2428 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2429 | ||
2430 | /** | |
421cee29 RD |
2431 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2432 | * @notifier: notifier struct to register | |
e107be36 AK |
2433 | */ |
2434 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2435 | { | |
2436 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2437 | } | |
2438 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2439 | ||
2440 | /** | |
2441 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2442 | * @notifier: notifier struct to unregister |
e107be36 AK |
2443 | * |
2444 | * This is safe to call from within a preemption notifier. | |
2445 | */ | |
2446 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2447 | { | |
2448 | hlist_del(¬ifier->link); | |
2449 | } | |
2450 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2451 | ||
2452 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2453 | { | |
2454 | struct preempt_notifier *notifier; | |
2455 | struct hlist_node *node; | |
2456 | ||
2457 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2458 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2459 | } | |
2460 | ||
2461 | static void | |
2462 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2463 | struct task_struct *next) | |
2464 | { | |
2465 | struct preempt_notifier *notifier; | |
2466 | struct hlist_node *node; | |
2467 | ||
2468 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2469 | notifier->ops->sched_out(notifier, next); | |
2470 | } | |
2471 | ||
6d6bc0ad | 2472 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2473 | |
2474 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2475 | { | |
2476 | } | |
2477 | ||
2478 | static void | |
2479 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2480 | struct task_struct *next) | |
2481 | { | |
2482 | } | |
2483 | ||
6d6bc0ad | 2484 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2485 | |
4866cde0 NP |
2486 | /** |
2487 | * prepare_task_switch - prepare to switch tasks | |
2488 | * @rq: the runqueue preparing to switch | |
421cee29 | 2489 | * @prev: the current task that is being switched out |
4866cde0 NP |
2490 | * @next: the task we are going to switch to. |
2491 | * | |
2492 | * This is called with the rq lock held and interrupts off. It must | |
2493 | * be paired with a subsequent finish_task_switch after the context | |
2494 | * switch. | |
2495 | * | |
2496 | * prepare_task_switch sets up locking and calls architecture specific | |
2497 | * hooks. | |
2498 | */ | |
e107be36 AK |
2499 | static inline void |
2500 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2501 | struct task_struct *next) | |
4866cde0 | 2502 | { |
e107be36 | 2503 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2504 | prepare_lock_switch(rq, next); |
2505 | prepare_arch_switch(next); | |
2506 | } | |
2507 | ||
1da177e4 LT |
2508 | /** |
2509 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2510 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2511 | * @prev: the thread we just switched away from. |
2512 | * | |
4866cde0 NP |
2513 | * finish_task_switch must be called after the context switch, paired |
2514 | * with a prepare_task_switch call before the context switch. | |
2515 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2516 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2517 | * |
2518 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2519 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2520 | * with the lock held can cause deadlocks; see schedule() for |
2521 | * details.) | |
2522 | */ | |
a9957449 | 2523 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2524 | __releases(rq->lock) |
2525 | { | |
1da177e4 | 2526 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2527 | long prev_state; |
1da177e4 LT |
2528 | |
2529 | rq->prev_mm = NULL; | |
2530 | ||
2531 | /* | |
2532 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2533 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2534 | * schedule one last time. The schedule call will never return, and |
2535 | * the scheduled task must drop that reference. | |
c394cc9f | 2536 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2537 | * still held, otherwise prev could be scheduled on another cpu, die |
2538 | * there before we look at prev->state, and then the reference would | |
2539 | * be dropped twice. | |
2540 | * Manfred Spraul <manfred@colorfullife.com> | |
2541 | */ | |
55a101f8 | 2542 | prev_state = prev->state; |
4866cde0 NP |
2543 | finish_arch_switch(prev); |
2544 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2545 | #ifdef CONFIG_SMP |
2546 | if (current->sched_class->post_schedule) | |
2547 | current->sched_class->post_schedule(rq); | |
2548 | #endif | |
e8fa1362 | 2549 | |
e107be36 | 2550 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2551 | if (mm) |
2552 | mmdrop(mm); | |
c394cc9f | 2553 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2554 | /* |
2555 | * Remove function-return probe instances associated with this | |
2556 | * task and put them back on the free list. | |
9761eea8 | 2557 | */ |
c6fd91f0 | 2558 | kprobe_flush_task(prev); |
1da177e4 | 2559 | put_task_struct(prev); |
c6fd91f0 | 2560 | } |
1da177e4 LT |
2561 | } |
2562 | ||
2563 | /** | |
2564 | * schedule_tail - first thing a freshly forked thread must call. | |
2565 | * @prev: the thread we just switched away from. | |
2566 | */ | |
36c8b586 | 2567 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2568 | __releases(rq->lock) |
2569 | { | |
70b97a7f IM |
2570 | struct rq *rq = this_rq(); |
2571 | ||
4866cde0 NP |
2572 | finish_task_switch(rq, prev); |
2573 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2574 | /* In this case, finish_task_switch does not reenable preemption */ | |
2575 | preempt_enable(); | |
2576 | #endif | |
1da177e4 | 2577 | if (current->set_child_tid) |
b488893a | 2578 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2579 | } |
2580 | ||
2581 | /* | |
2582 | * context_switch - switch to the new MM and the new | |
2583 | * thread's register state. | |
2584 | */ | |
dd41f596 | 2585 | static inline void |
70b97a7f | 2586 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2587 | struct task_struct *next) |
1da177e4 | 2588 | { |
dd41f596 | 2589 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2590 | |
e107be36 | 2591 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2592 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2593 | mm = next->mm; |
2594 | oldmm = prev->active_mm; | |
9226d125 ZA |
2595 | /* |
2596 | * For paravirt, this is coupled with an exit in switch_to to | |
2597 | * combine the page table reload and the switch backend into | |
2598 | * one hypercall. | |
2599 | */ | |
2600 | arch_enter_lazy_cpu_mode(); | |
2601 | ||
dd41f596 | 2602 | if (unlikely(!mm)) { |
1da177e4 LT |
2603 | next->active_mm = oldmm; |
2604 | atomic_inc(&oldmm->mm_count); | |
2605 | enter_lazy_tlb(oldmm, next); | |
2606 | } else | |
2607 | switch_mm(oldmm, mm, next); | |
2608 | ||
dd41f596 | 2609 | if (unlikely(!prev->mm)) { |
1da177e4 | 2610 | prev->active_mm = NULL; |
1da177e4 LT |
2611 | rq->prev_mm = oldmm; |
2612 | } | |
3a5f5e48 IM |
2613 | /* |
2614 | * Since the runqueue lock will be released by the next | |
2615 | * task (which is an invalid locking op but in the case | |
2616 | * of the scheduler it's an obvious special-case), so we | |
2617 | * do an early lockdep release here: | |
2618 | */ | |
2619 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2620 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2621 | #endif |
1da177e4 LT |
2622 | |
2623 | /* Here we just switch the register state and the stack. */ | |
2624 | switch_to(prev, next, prev); | |
2625 | ||
dd41f596 IM |
2626 | barrier(); |
2627 | /* | |
2628 | * this_rq must be evaluated again because prev may have moved | |
2629 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2630 | * frame will be invalid. | |
2631 | */ | |
2632 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2633 | } |
2634 | ||
2635 | /* | |
2636 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2637 | * | |
2638 | * externally visible scheduler statistics: current number of runnable | |
2639 | * threads, current number of uninterruptible-sleeping threads, total | |
2640 | * number of context switches performed since bootup. | |
2641 | */ | |
2642 | unsigned long nr_running(void) | |
2643 | { | |
2644 | unsigned long i, sum = 0; | |
2645 | ||
2646 | for_each_online_cpu(i) | |
2647 | sum += cpu_rq(i)->nr_running; | |
2648 | ||
2649 | return sum; | |
2650 | } | |
2651 | ||
2652 | unsigned long nr_uninterruptible(void) | |
2653 | { | |
2654 | unsigned long i, sum = 0; | |
2655 | ||
0a945022 | 2656 | for_each_possible_cpu(i) |
1da177e4 LT |
2657 | sum += cpu_rq(i)->nr_uninterruptible; |
2658 | ||
2659 | /* | |
2660 | * Since we read the counters lockless, it might be slightly | |
2661 | * inaccurate. Do not allow it to go below zero though: | |
2662 | */ | |
2663 | if (unlikely((long)sum < 0)) | |
2664 | sum = 0; | |
2665 | ||
2666 | return sum; | |
2667 | } | |
2668 | ||
2669 | unsigned long long nr_context_switches(void) | |
2670 | { | |
cc94abfc SR |
2671 | int i; |
2672 | unsigned long long sum = 0; | |
1da177e4 | 2673 | |
0a945022 | 2674 | for_each_possible_cpu(i) |
1da177e4 LT |
2675 | sum += cpu_rq(i)->nr_switches; |
2676 | ||
2677 | return sum; | |
2678 | } | |
2679 | ||
2680 | unsigned long nr_iowait(void) | |
2681 | { | |
2682 | unsigned long i, sum = 0; | |
2683 | ||
0a945022 | 2684 | for_each_possible_cpu(i) |
1da177e4 LT |
2685 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2686 | ||
2687 | return sum; | |
2688 | } | |
2689 | ||
db1b1fef JS |
2690 | unsigned long nr_active(void) |
2691 | { | |
2692 | unsigned long i, running = 0, uninterruptible = 0; | |
2693 | ||
2694 | for_each_online_cpu(i) { | |
2695 | running += cpu_rq(i)->nr_running; | |
2696 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2697 | } | |
2698 | ||
2699 | if (unlikely((long)uninterruptible < 0)) | |
2700 | uninterruptible = 0; | |
2701 | ||
2702 | return running + uninterruptible; | |
2703 | } | |
2704 | ||
48f24c4d | 2705 | /* |
dd41f596 IM |
2706 | * Update rq->cpu_load[] statistics. This function is usually called every |
2707 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2708 | */ |
dd41f596 | 2709 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2710 | { |
495eca49 | 2711 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2712 | int i, scale; |
2713 | ||
2714 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2715 | |
2716 | /* Update our load: */ | |
2717 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2718 | unsigned long old_load, new_load; | |
2719 | ||
2720 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2721 | ||
2722 | old_load = this_rq->cpu_load[i]; | |
2723 | new_load = this_load; | |
a25707f3 IM |
2724 | /* |
2725 | * Round up the averaging division if load is increasing. This | |
2726 | * prevents us from getting stuck on 9 if the load is 10, for | |
2727 | * example. | |
2728 | */ | |
2729 | if (new_load > old_load) | |
2730 | new_load += scale-1; | |
dd41f596 IM |
2731 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2732 | } | |
48f24c4d IM |
2733 | } |
2734 | ||
dd41f596 IM |
2735 | #ifdef CONFIG_SMP |
2736 | ||
1da177e4 LT |
2737 | /* |
2738 | * double_rq_lock - safely lock two runqueues | |
2739 | * | |
2740 | * Note this does not disable interrupts like task_rq_lock, | |
2741 | * you need to do so manually before calling. | |
2742 | */ | |
70b97a7f | 2743 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2744 | __acquires(rq1->lock) |
2745 | __acquires(rq2->lock) | |
2746 | { | |
054b9108 | 2747 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2748 | if (rq1 == rq2) { |
2749 | spin_lock(&rq1->lock); | |
2750 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2751 | } else { | |
c96d145e | 2752 | if (rq1 < rq2) { |
1da177e4 | 2753 | spin_lock(&rq1->lock); |
5e710e37 | 2754 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2755 | } else { |
2756 | spin_lock(&rq2->lock); | |
5e710e37 | 2757 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2758 | } |
2759 | } | |
6e82a3be IM |
2760 | update_rq_clock(rq1); |
2761 | update_rq_clock(rq2); | |
1da177e4 LT |
2762 | } |
2763 | ||
2764 | /* | |
2765 | * double_rq_unlock - safely unlock two runqueues | |
2766 | * | |
2767 | * Note this does not restore interrupts like task_rq_unlock, | |
2768 | * you need to do so manually after calling. | |
2769 | */ | |
70b97a7f | 2770 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2771 | __releases(rq1->lock) |
2772 | __releases(rq2->lock) | |
2773 | { | |
2774 | spin_unlock(&rq1->lock); | |
2775 | if (rq1 != rq2) | |
2776 | spin_unlock(&rq2->lock); | |
2777 | else | |
2778 | __release(rq2->lock); | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2783 | */ | |
e8fa1362 | 2784 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2785 | __releases(this_rq->lock) |
2786 | __acquires(busiest->lock) | |
2787 | __acquires(this_rq->lock) | |
2788 | { | |
e8fa1362 SR |
2789 | int ret = 0; |
2790 | ||
054b9108 KK |
2791 | if (unlikely(!irqs_disabled())) { |
2792 | /* printk() doesn't work good under rq->lock */ | |
2793 | spin_unlock(&this_rq->lock); | |
2794 | BUG_ON(1); | |
2795 | } | |
1da177e4 | 2796 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2797 | if (busiest < this_rq) { |
1da177e4 LT |
2798 | spin_unlock(&this_rq->lock); |
2799 | spin_lock(&busiest->lock); | |
5e710e37 | 2800 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2801 | ret = 1; |
1da177e4 | 2802 | } else |
5e710e37 | 2803 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2804 | } |
e8fa1362 | 2805 | return ret; |
1da177e4 LT |
2806 | } |
2807 | ||
1b12bbc7 PZ |
2808 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2809 | __releases(busiest->lock) | |
2810 | { | |
2811 | spin_unlock(&busiest->lock); | |
2812 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2813 | } | |
2814 | ||
1da177e4 LT |
2815 | /* |
2816 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2817 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2818 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2819 | * the cpu_allowed mask is restored. |
2820 | */ | |
36c8b586 | 2821 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2822 | { |
70b97a7f | 2823 | struct migration_req req; |
1da177e4 | 2824 | unsigned long flags; |
70b97a7f | 2825 | struct rq *rq; |
1da177e4 LT |
2826 | |
2827 | rq = task_rq_lock(p, &flags); | |
2828 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2829 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2830 | goto out; |
2831 | ||
0a16b607 | 2832 | trace_sched_migrate_task(rq, p, dest_cpu); |
1da177e4 LT |
2833 | /* force the process onto the specified CPU */ |
2834 | if (migrate_task(p, dest_cpu, &req)) { | |
2835 | /* Need to wait for migration thread (might exit: take ref). */ | |
2836 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2837 | |
1da177e4 LT |
2838 | get_task_struct(mt); |
2839 | task_rq_unlock(rq, &flags); | |
2840 | wake_up_process(mt); | |
2841 | put_task_struct(mt); | |
2842 | wait_for_completion(&req.done); | |
36c8b586 | 2843 | |
1da177e4 LT |
2844 | return; |
2845 | } | |
2846 | out: | |
2847 | task_rq_unlock(rq, &flags); | |
2848 | } | |
2849 | ||
2850 | /* | |
476d139c NP |
2851 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2852 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2853 | */ |
2854 | void sched_exec(void) | |
2855 | { | |
1da177e4 | 2856 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2857 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2858 | put_cpu(); |
476d139c NP |
2859 | if (new_cpu != this_cpu) |
2860 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2861 | } |
2862 | ||
2863 | /* | |
2864 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2865 | * Both runqueues must be locked. | |
2866 | */ | |
dd41f596 IM |
2867 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2868 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2869 | { |
2e1cb74a | 2870 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2871 | set_task_cpu(p, this_cpu); |
dd41f596 | 2872 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2873 | /* |
2874 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2875 | * to be always true for them. | |
2876 | */ | |
15afe09b | 2877 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2878 | } |
2879 | ||
2880 | /* | |
2881 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2882 | */ | |
858119e1 | 2883 | static |
70b97a7f | 2884 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2885 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2886 | int *all_pinned) |
1da177e4 LT |
2887 | { |
2888 | /* | |
2889 | * We do not migrate tasks that are: | |
2890 | * 1) running (obviously), or | |
2891 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2892 | * 3) are cache-hot on their current CPU. | |
2893 | */ | |
cc367732 IM |
2894 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2895 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2896 | return 0; |
cc367732 | 2897 | } |
81026794 NP |
2898 | *all_pinned = 0; |
2899 | ||
cc367732 IM |
2900 | if (task_running(rq, p)) { |
2901 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2902 | return 0; |
cc367732 | 2903 | } |
1da177e4 | 2904 | |
da84d961 IM |
2905 | /* |
2906 | * Aggressive migration if: | |
2907 | * 1) task is cache cold, or | |
2908 | * 2) too many balance attempts have failed. | |
2909 | */ | |
2910 | ||
6bc1665b IM |
2911 | if (!task_hot(p, rq->clock, sd) || |
2912 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2913 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2914 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2915 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2916 | schedstat_inc(p, se.nr_forced_migrations); |
2917 | } | |
da84d961 IM |
2918 | #endif |
2919 | return 1; | |
2920 | } | |
2921 | ||
cc367732 IM |
2922 | if (task_hot(p, rq->clock, sd)) { |
2923 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2924 | return 0; |
cc367732 | 2925 | } |
1da177e4 LT |
2926 | return 1; |
2927 | } | |
2928 | ||
e1d1484f PW |
2929 | static unsigned long |
2930 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2931 | unsigned long max_load_move, struct sched_domain *sd, | |
2932 | enum cpu_idle_type idle, int *all_pinned, | |
2933 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2934 | { |
051c6764 | 2935 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2936 | struct task_struct *p; |
2937 | long rem_load_move = max_load_move; | |
1da177e4 | 2938 | |
e1d1484f | 2939 | if (max_load_move == 0) |
1da177e4 LT |
2940 | goto out; |
2941 | ||
81026794 NP |
2942 | pinned = 1; |
2943 | ||
1da177e4 | 2944 | /* |
dd41f596 | 2945 | * Start the load-balancing iterator: |
1da177e4 | 2946 | */ |
dd41f596 IM |
2947 | p = iterator->start(iterator->arg); |
2948 | next: | |
b82d9fdd | 2949 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2950 | goto out; |
051c6764 PZ |
2951 | |
2952 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2953 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2954 | p = iterator->next(iterator->arg); |
2955 | goto next; | |
1da177e4 LT |
2956 | } |
2957 | ||
dd41f596 | 2958 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2959 | pulled++; |
dd41f596 | 2960 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2961 | |
2dd73a4f | 2962 | /* |
b82d9fdd | 2963 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2964 | */ |
e1d1484f | 2965 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2966 | if (p->prio < *this_best_prio) |
2967 | *this_best_prio = p->prio; | |
dd41f596 IM |
2968 | p = iterator->next(iterator->arg); |
2969 | goto next; | |
1da177e4 LT |
2970 | } |
2971 | out: | |
2972 | /* | |
e1d1484f | 2973 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2974 | * so we can safely collect pull_task() stats here rather than |
2975 | * inside pull_task(). | |
2976 | */ | |
2977 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2978 | |
2979 | if (all_pinned) | |
2980 | *all_pinned = pinned; | |
e1d1484f PW |
2981 | |
2982 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2983 | } |
2984 | ||
dd41f596 | 2985 | /* |
43010659 PW |
2986 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2987 | * this_rq, as part of a balancing operation within domain "sd". | |
2988 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2989 | * |
2990 | * Called with both runqueues locked. | |
2991 | */ | |
2992 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2993 | unsigned long max_load_move, |
dd41f596 IM |
2994 | struct sched_domain *sd, enum cpu_idle_type idle, |
2995 | int *all_pinned) | |
2996 | { | |
5522d5d5 | 2997 | const struct sched_class *class = sched_class_highest; |
43010659 | 2998 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2999 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3000 | |
3001 | do { | |
43010659 PW |
3002 | total_load_moved += |
3003 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3004 | max_load_move - total_load_moved, |
a4ac01c3 | 3005 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3006 | class = class->next; |
c4acb2c0 GH |
3007 | |
3008 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3009 | break; | |
3010 | ||
43010659 | 3011 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3012 | |
43010659 PW |
3013 | return total_load_moved > 0; |
3014 | } | |
3015 | ||
e1d1484f PW |
3016 | static int |
3017 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3018 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3019 | struct rq_iterator *iterator) | |
3020 | { | |
3021 | struct task_struct *p = iterator->start(iterator->arg); | |
3022 | int pinned = 0; | |
3023 | ||
3024 | while (p) { | |
3025 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3026 | pull_task(busiest, p, this_rq, this_cpu); | |
3027 | /* | |
3028 | * Right now, this is only the second place pull_task() | |
3029 | * is called, so we can safely collect pull_task() | |
3030 | * stats here rather than inside pull_task(). | |
3031 | */ | |
3032 | schedstat_inc(sd, lb_gained[idle]); | |
3033 | ||
3034 | return 1; | |
3035 | } | |
3036 | p = iterator->next(iterator->arg); | |
3037 | } | |
3038 | ||
3039 | return 0; | |
3040 | } | |
3041 | ||
43010659 PW |
3042 | /* |
3043 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3044 | * part of active balancing operations within "domain". | |
3045 | * Returns 1 if successful and 0 otherwise. | |
3046 | * | |
3047 | * Called with both runqueues locked. | |
3048 | */ | |
3049 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3050 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3051 | { | |
5522d5d5 | 3052 | const struct sched_class *class; |
43010659 PW |
3053 | |
3054 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3055 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3056 | return 1; |
3057 | ||
3058 | return 0; | |
dd41f596 IM |
3059 | } |
3060 | ||
1da177e4 LT |
3061 | /* |
3062 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3063 | * domain. It calculates and returns the amount of weighted load which |
3064 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3065 | */ |
3066 | static struct sched_group * | |
3067 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3068 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3069 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3070 | { |
3071 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3072 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3073 | unsigned long max_pull; |
2dd73a4f PW |
3074 | unsigned long busiest_load_per_task, busiest_nr_running; |
3075 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3076 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3077 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3078 | int power_savings_balance = 1; | |
3079 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3080 | unsigned long min_nr_running = ULONG_MAX; | |
3081 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3082 | #endif | |
1da177e4 LT |
3083 | |
3084 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3085 | busiest_load_per_task = busiest_nr_running = 0; |
3086 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3087 | |
d15bcfdb | 3088 | if (idle == CPU_NOT_IDLE) |
7897986b | 3089 | load_idx = sd->busy_idx; |
d15bcfdb | 3090 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3091 | load_idx = sd->newidle_idx; |
3092 | else | |
3093 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3094 | |
3095 | do { | |
908a7c1b | 3096 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3097 | int local_group; |
3098 | int i; | |
908a7c1b | 3099 | int __group_imb = 0; |
783609c6 | 3100 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3101 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3102 | unsigned long sum_avg_load_per_task; |
3103 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3104 | |
3105 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3106 | ||
783609c6 SS |
3107 | if (local_group) |
3108 | balance_cpu = first_cpu(group->cpumask); | |
3109 | ||
1da177e4 | 3110 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3111 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3112 | sum_avg_load_per_task = avg_load_per_task = 0; |
3113 | ||
908a7c1b KC |
3114 | max_cpu_load = 0; |
3115 | min_cpu_load = ~0UL; | |
1da177e4 | 3116 | |
363ab6f1 | 3117 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3118 | struct rq *rq; |
3119 | ||
3120 | if (!cpu_isset(i, *cpus)) | |
3121 | continue; | |
3122 | ||
3123 | rq = cpu_rq(i); | |
2dd73a4f | 3124 | |
9439aab8 | 3125 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3126 | *sd_idle = 0; |
3127 | ||
1da177e4 | 3128 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3129 | if (local_group) { |
3130 | if (idle_cpu(i) && !first_idle_cpu) { | |
3131 | first_idle_cpu = 1; | |
3132 | balance_cpu = i; | |
3133 | } | |
3134 | ||
a2000572 | 3135 | load = target_load(i, load_idx); |
908a7c1b | 3136 | } else { |
a2000572 | 3137 | load = source_load(i, load_idx); |
908a7c1b KC |
3138 | if (load > max_cpu_load) |
3139 | max_cpu_load = load; | |
3140 | if (min_cpu_load > load) | |
3141 | min_cpu_load = load; | |
3142 | } | |
1da177e4 LT |
3143 | |
3144 | avg_load += load; | |
2dd73a4f | 3145 | sum_nr_running += rq->nr_running; |
dd41f596 | 3146 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3147 | |
3148 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3149 | } |
3150 | ||
783609c6 SS |
3151 | /* |
3152 | * First idle cpu or the first cpu(busiest) in this sched group | |
3153 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3154 | * domains. In the newly idle case, we will allow all the cpu's |
3155 | * to do the newly idle load balance. | |
783609c6 | 3156 | */ |
9439aab8 SS |
3157 | if (idle != CPU_NEWLY_IDLE && local_group && |
3158 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3159 | *balance = 0; |
3160 | goto ret; | |
3161 | } | |
3162 | ||
1da177e4 | 3163 | total_load += avg_load; |
5517d86b | 3164 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3165 | |
3166 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3167 | avg_load = sg_div_cpu_power(group, |
3168 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3169 | |
408ed066 PZ |
3170 | |
3171 | /* | |
3172 | * Consider the group unbalanced when the imbalance is larger | |
3173 | * than the average weight of two tasks. | |
3174 | * | |
3175 | * APZ: with cgroup the avg task weight can vary wildly and | |
3176 | * might not be a suitable number - should we keep a | |
3177 | * normalized nr_running number somewhere that negates | |
3178 | * the hierarchy? | |
3179 | */ | |
3180 | avg_load_per_task = sg_div_cpu_power(group, | |
3181 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3182 | ||
3183 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3184 | __group_imb = 1; |
3185 | ||
5517d86b | 3186 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3187 | |
1da177e4 LT |
3188 | if (local_group) { |
3189 | this_load = avg_load; | |
3190 | this = group; | |
2dd73a4f PW |
3191 | this_nr_running = sum_nr_running; |
3192 | this_load_per_task = sum_weighted_load; | |
3193 | } else if (avg_load > max_load && | |
908a7c1b | 3194 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3195 | max_load = avg_load; |
3196 | busiest = group; | |
2dd73a4f PW |
3197 | busiest_nr_running = sum_nr_running; |
3198 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3199 | group_imb = __group_imb; |
1da177e4 | 3200 | } |
5c45bf27 SS |
3201 | |
3202 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3203 | /* | |
3204 | * Busy processors will not participate in power savings | |
3205 | * balance. | |
3206 | */ | |
dd41f596 IM |
3207 | if (idle == CPU_NOT_IDLE || |
3208 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3209 | goto group_next; | |
5c45bf27 SS |
3210 | |
3211 | /* | |
3212 | * If the local group is idle or completely loaded | |
3213 | * no need to do power savings balance at this domain | |
3214 | */ | |
3215 | if (local_group && (this_nr_running >= group_capacity || | |
3216 | !this_nr_running)) | |
3217 | power_savings_balance = 0; | |
3218 | ||
dd41f596 | 3219 | /* |
5c45bf27 SS |
3220 | * If a group is already running at full capacity or idle, |
3221 | * don't include that group in power savings calculations | |
dd41f596 IM |
3222 | */ |
3223 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3224 | || !sum_nr_running) |
dd41f596 | 3225 | goto group_next; |
5c45bf27 | 3226 | |
dd41f596 | 3227 | /* |
5c45bf27 | 3228 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3229 | * This is the group from where we need to pick up the load |
3230 | * for saving power | |
3231 | */ | |
3232 | if ((sum_nr_running < min_nr_running) || | |
3233 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3234 | first_cpu(group->cpumask) < |
3235 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3236 | group_min = group; |
3237 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3238 | min_load_per_task = sum_weighted_load / |
3239 | sum_nr_running; | |
dd41f596 | 3240 | } |
5c45bf27 | 3241 | |
dd41f596 | 3242 | /* |
5c45bf27 | 3243 | * Calculate the group which is almost near its |
dd41f596 IM |
3244 | * capacity but still has some space to pick up some load |
3245 | * from other group and save more power | |
3246 | */ | |
3247 | if (sum_nr_running <= group_capacity - 1) { | |
3248 | if (sum_nr_running > leader_nr_running || | |
3249 | (sum_nr_running == leader_nr_running && | |
3250 | first_cpu(group->cpumask) > | |
3251 | first_cpu(group_leader->cpumask))) { | |
3252 | group_leader = group; | |
3253 | leader_nr_running = sum_nr_running; | |
3254 | } | |
48f24c4d | 3255 | } |
5c45bf27 SS |
3256 | group_next: |
3257 | #endif | |
1da177e4 LT |
3258 | group = group->next; |
3259 | } while (group != sd->groups); | |
3260 | ||
2dd73a4f | 3261 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3262 | goto out_balanced; |
3263 | ||
3264 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3265 | ||
3266 | if (this_load >= avg_load || | |
3267 | 100*max_load <= sd->imbalance_pct*this_load) | |
3268 | goto out_balanced; | |
3269 | ||
2dd73a4f | 3270 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3271 | if (group_imb) |
3272 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3273 | ||
1da177e4 LT |
3274 | /* |
3275 | * We're trying to get all the cpus to the average_load, so we don't | |
3276 | * want to push ourselves above the average load, nor do we wish to | |
3277 | * reduce the max loaded cpu below the average load, as either of these | |
3278 | * actions would just result in more rebalancing later, and ping-pong | |
3279 | * tasks around. Thus we look for the minimum possible imbalance. | |
3280 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3281 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3282 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3283 | * appear as very large values with unsigned longs. |
3284 | */ | |
2dd73a4f PW |
3285 | if (max_load <= busiest_load_per_task) |
3286 | goto out_balanced; | |
3287 | ||
3288 | /* | |
3289 | * In the presence of smp nice balancing, certain scenarios can have | |
3290 | * max load less than avg load(as we skip the groups at or below | |
3291 | * its cpu_power, while calculating max_load..) | |
3292 | */ | |
3293 | if (max_load < avg_load) { | |
3294 | *imbalance = 0; | |
3295 | goto small_imbalance; | |
3296 | } | |
0c117f1b SS |
3297 | |
3298 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3299 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3300 | |
1da177e4 | 3301 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3302 | *imbalance = min(max_pull * busiest->__cpu_power, |
3303 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3304 | / SCHED_LOAD_SCALE; |
3305 | ||
2dd73a4f PW |
3306 | /* |
3307 | * if *imbalance is less than the average load per runnable task | |
3308 | * there is no gaurantee that any tasks will be moved so we'll have | |
3309 | * a think about bumping its value to force at least one task to be | |
3310 | * moved | |
3311 | */ | |
7fd0d2dd | 3312 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3313 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3314 | unsigned int imbn; |
3315 | ||
3316 | small_imbalance: | |
3317 | pwr_move = pwr_now = 0; | |
3318 | imbn = 2; | |
3319 | if (this_nr_running) { | |
3320 | this_load_per_task /= this_nr_running; | |
3321 | if (busiest_load_per_task > this_load_per_task) | |
3322 | imbn = 1; | |
3323 | } else | |
408ed066 | 3324 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3325 | |
01c8c57d | 3326 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3327 | busiest_load_per_task * imbn) { |
2dd73a4f | 3328 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3329 | return busiest; |
3330 | } | |
3331 | ||
3332 | /* | |
3333 | * OK, we don't have enough imbalance to justify moving tasks, | |
3334 | * however we may be able to increase total CPU power used by | |
3335 | * moving them. | |
3336 | */ | |
3337 | ||
5517d86b ED |
3338 | pwr_now += busiest->__cpu_power * |
3339 | min(busiest_load_per_task, max_load); | |
3340 | pwr_now += this->__cpu_power * | |
3341 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3342 | pwr_now /= SCHED_LOAD_SCALE; |
3343 | ||
3344 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3345 | tmp = sg_div_cpu_power(busiest, |
3346 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3347 | if (max_load > tmp) |
5517d86b | 3348 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3349 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3350 | |
3351 | /* Amount of load we'd add */ | |
5517d86b | 3352 | if (max_load * busiest->__cpu_power < |
33859f7f | 3353 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3354 | tmp = sg_div_cpu_power(this, |
3355 | max_load * busiest->__cpu_power); | |
1da177e4 | 3356 | else |
5517d86b ED |
3357 | tmp = sg_div_cpu_power(this, |
3358 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3359 | pwr_move += this->__cpu_power * | |
3360 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3361 | pwr_move /= SCHED_LOAD_SCALE; |
3362 | ||
3363 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3364 | if (pwr_move > pwr_now) |
3365 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3366 | } |
3367 | ||
1da177e4 LT |
3368 | return busiest; |
3369 | ||
3370 | out_balanced: | |
5c45bf27 | 3371 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3372 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3373 | goto ret; |
1da177e4 | 3374 | |
5c45bf27 SS |
3375 | if (this == group_leader && group_leader != group_min) { |
3376 | *imbalance = min_load_per_task; | |
3377 | return group_min; | |
3378 | } | |
5c45bf27 | 3379 | #endif |
783609c6 | 3380 | ret: |
1da177e4 LT |
3381 | *imbalance = 0; |
3382 | return NULL; | |
3383 | } | |
3384 | ||
3385 | /* | |
3386 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3387 | */ | |
70b97a7f | 3388 | static struct rq * |
d15bcfdb | 3389 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3390 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3391 | { |
70b97a7f | 3392 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3393 | unsigned long max_load = 0; |
1da177e4 LT |
3394 | int i; |
3395 | ||
363ab6f1 | 3396 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3397 | unsigned long wl; |
0a2966b4 CL |
3398 | |
3399 | if (!cpu_isset(i, *cpus)) | |
3400 | continue; | |
3401 | ||
48f24c4d | 3402 | rq = cpu_rq(i); |
dd41f596 | 3403 | wl = weighted_cpuload(i); |
2dd73a4f | 3404 | |
dd41f596 | 3405 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3406 | continue; |
1da177e4 | 3407 | |
dd41f596 IM |
3408 | if (wl > max_load) { |
3409 | max_load = wl; | |
48f24c4d | 3410 | busiest = rq; |
1da177e4 LT |
3411 | } |
3412 | } | |
3413 | ||
3414 | return busiest; | |
3415 | } | |
3416 | ||
77391d71 NP |
3417 | /* |
3418 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3419 | * so long as it is large enough. | |
3420 | */ | |
3421 | #define MAX_PINNED_INTERVAL 512 | |
3422 | ||
1da177e4 LT |
3423 | /* |
3424 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3425 | * tasks if there is an imbalance. | |
1da177e4 | 3426 | */ |
70b97a7f | 3427 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3428 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3429 | int *balance, cpumask_t *cpus) |
1da177e4 | 3430 | { |
43010659 | 3431 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3432 | struct sched_group *group; |
1da177e4 | 3433 | unsigned long imbalance; |
70b97a7f | 3434 | struct rq *busiest; |
fe2eea3f | 3435 | unsigned long flags; |
5969fe06 | 3436 | |
7c16ec58 MT |
3437 | cpus_setall(*cpus); |
3438 | ||
89c4710e SS |
3439 | /* |
3440 | * When power savings policy is enabled for the parent domain, idle | |
3441 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3442 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3443 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3444 | */ |
d15bcfdb | 3445 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3446 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3447 | sd_idle = 1; |
1da177e4 | 3448 | |
2d72376b | 3449 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3450 | |
0a2966b4 | 3451 | redo: |
c8cba857 | 3452 | update_shares(sd); |
0a2966b4 | 3453 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3454 | cpus, balance); |
783609c6 | 3455 | |
06066714 | 3456 | if (*balance == 0) |
783609c6 | 3457 | goto out_balanced; |
783609c6 | 3458 | |
1da177e4 LT |
3459 | if (!group) { |
3460 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3461 | goto out_balanced; | |
3462 | } | |
3463 | ||
7c16ec58 | 3464 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3465 | if (!busiest) { |
3466 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3467 | goto out_balanced; | |
3468 | } | |
3469 | ||
db935dbd | 3470 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3471 | |
3472 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3473 | ||
43010659 | 3474 | ld_moved = 0; |
1da177e4 LT |
3475 | if (busiest->nr_running > 1) { |
3476 | /* | |
3477 | * Attempt to move tasks. If find_busiest_group has found | |
3478 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3479 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3480 | * correctly treated as an imbalance. |
3481 | */ | |
fe2eea3f | 3482 | local_irq_save(flags); |
e17224bf | 3483 | double_rq_lock(this_rq, busiest); |
43010659 | 3484 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3485 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3486 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3487 | local_irq_restore(flags); |
81026794 | 3488 | |
46cb4b7c SS |
3489 | /* |
3490 | * some other cpu did the load balance for us. | |
3491 | */ | |
43010659 | 3492 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3493 | resched_cpu(this_cpu); |
3494 | ||
81026794 | 3495 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3496 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3497 | cpu_clear(cpu_of(busiest), *cpus); |
3498 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3499 | goto redo; |
81026794 | 3500 | goto out_balanced; |
0a2966b4 | 3501 | } |
1da177e4 | 3502 | } |
81026794 | 3503 | |
43010659 | 3504 | if (!ld_moved) { |
1da177e4 LT |
3505 | schedstat_inc(sd, lb_failed[idle]); |
3506 | sd->nr_balance_failed++; | |
3507 | ||
3508 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3509 | |
fe2eea3f | 3510 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3511 | |
3512 | /* don't kick the migration_thread, if the curr | |
3513 | * task on busiest cpu can't be moved to this_cpu | |
3514 | */ | |
3515 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3516 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3517 | all_pinned = 1; |
3518 | goto out_one_pinned; | |
3519 | } | |
3520 | ||
1da177e4 LT |
3521 | if (!busiest->active_balance) { |
3522 | busiest->active_balance = 1; | |
3523 | busiest->push_cpu = this_cpu; | |
81026794 | 3524 | active_balance = 1; |
1da177e4 | 3525 | } |
fe2eea3f | 3526 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3527 | if (active_balance) |
1da177e4 LT |
3528 | wake_up_process(busiest->migration_thread); |
3529 | ||
3530 | /* | |
3531 | * We've kicked active balancing, reset the failure | |
3532 | * counter. | |
3533 | */ | |
39507451 | 3534 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3535 | } |
81026794 | 3536 | } else |
1da177e4 LT |
3537 | sd->nr_balance_failed = 0; |
3538 | ||
81026794 | 3539 | if (likely(!active_balance)) { |
1da177e4 LT |
3540 | /* We were unbalanced, so reset the balancing interval */ |
3541 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3542 | } else { |
3543 | /* | |
3544 | * If we've begun active balancing, start to back off. This | |
3545 | * case may not be covered by the all_pinned logic if there | |
3546 | * is only 1 task on the busy runqueue (because we don't call | |
3547 | * move_tasks). | |
3548 | */ | |
3549 | if (sd->balance_interval < sd->max_interval) | |
3550 | sd->balance_interval *= 2; | |
1da177e4 LT |
3551 | } |
3552 | ||
43010659 | 3553 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3554 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3555 | ld_moved = -1; |
3556 | ||
3557 | goto out; | |
1da177e4 LT |
3558 | |
3559 | out_balanced: | |
1da177e4 LT |
3560 | schedstat_inc(sd, lb_balanced[idle]); |
3561 | ||
16cfb1c0 | 3562 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3563 | |
3564 | out_one_pinned: | |
1da177e4 | 3565 | /* tune up the balancing interval */ |
77391d71 NP |
3566 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3567 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3568 | sd->balance_interval *= 2; |
3569 | ||
48f24c4d | 3570 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3571 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3572 | ld_moved = -1; |
3573 | else | |
3574 | ld_moved = 0; | |
3575 | out: | |
c8cba857 PZ |
3576 | if (ld_moved) |
3577 | update_shares(sd); | |
c09595f6 | 3578 | return ld_moved; |
1da177e4 LT |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3583 | * tasks if there is an imbalance. | |
3584 | * | |
d15bcfdb | 3585 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3586 | * this_rq is locked. |
3587 | */ | |
48f24c4d | 3588 | static int |
7c16ec58 MT |
3589 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3590 | cpumask_t *cpus) | |
1da177e4 LT |
3591 | { |
3592 | struct sched_group *group; | |
70b97a7f | 3593 | struct rq *busiest = NULL; |
1da177e4 | 3594 | unsigned long imbalance; |
43010659 | 3595 | int ld_moved = 0; |
5969fe06 | 3596 | int sd_idle = 0; |
969bb4e4 | 3597 | int all_pinned = 0; |
7c16ec58 MT |
3598 | |
3599 | cpus_setall(*cpus); | |
5969fe06 | 3600 | |
89c4710e SS |
3601 | /* |
3602 | * When power savings policy is enabled for the parent domain, idle | |
3603 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3604 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3605 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3606 | */ |
3607 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3608 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3609 | sd_idle = 1; |
1da177e4 | 3610 | |
2d72376b | 3611 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3612 | redo: |
3e5459b4 | 3613 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3614 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3615 | &sd_idle, cpus, NULL); |
1da177e4 | 3616 | if (!group) { |
d15bcfdb | 3617 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3618 | goto out_balanced; |
1da177e4 LT |
3619 | } |
3620 | ||
7c16ec58 | 3621 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3622 | if (!busiest) { |
d15bcfdb | 3623 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3624 | goto out_balanced; |
1da177e4 LT |
3625 | } |
3626 | ||
db935dbd NP |
3627 | BUG_ON(busiest == this_rq); |
3628 | ||
d15bcfdb | 3629 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3630 | |
43010659 | 3631 | ld_moved = 0; |
d6d5cfaf NP |
3632 | if (busiest->nr_running > 1) { |
3633 | /* Attempt to move tasks */ | |
3634 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3635 | /* this_rq->clock is already updated */ |
3636 | update_rq_clock(busiest); | |
43010659 | 3637 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3638 | imbalance, sd, CPU_NEWLY_IDLE, |
3639 | &all_pinned); | |
1b12bbc7 | 3640 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3641 | |
969bb4e4 | 3642 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3643 | cpu_clear(cpu_of(busiest), *cpus); |
3644 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3645 | goto redo; |
3646 | } | |
d6d5cfaf NP |
3647 | } |
3648 | ||
43010659 | 3649 | if (!ld_moved) { |
d15bcfdb | 3650 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3651 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3652 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3653 | return -1; |
3654 | } else | |
16cfb1c0 | 3655 | sd->nr_balance_failed = 0; |
1da177e4 | 3656 | |
3e5459b4 | 3657 | update_shares_locked(this_rq, sd); |
43010659 | 3658 | return ld_moved; |
16cfb1c0 NP |
3659 | |
3660 | out_balanced: | |
d15bcfdb | 3661 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3662 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3663 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3664 | return -1; |
16cfb1c0 | 3665 | sd->nr_balance_failed = 0; |
48f24c4d | 3666 | |
16cfb1c0 | 3667 | return 0; |
1da177e4 LT |
3668 | } |
3669 | ||
3670 | /* | |
3671 | * idle_balance is called by schedule() if this_cpu is about to become | |
3672 | * idle. Attempts to pull tasks from other CPUs. | |
3673 | */ | |
70b97a7f | 3674 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3675 | { |
3676 | struct sched_domain *sd; | |
dd41f596 IM |
3677 | int pulled_task = -1; |
3678 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3679 | cpumask_t tmpmask; |
1da177e4 LT |
3680 | |
3681 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3682 | unsigned long interval; |
3683 | ||
3684 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3685 | continue; | |
3686 | ||
3687 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3688 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3689 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3690 | sd, &tmpmask); | |
92c4ca5c CL |
3691 | |
3692 | interval = msecs_to_jiffies(sd->balance_interval); | |
3693 | if (time_after(next_balance, sd->last_balance + interval)) | |
3694 | next_balance = sd->last_balance + interval; | |
3695 | if (pulled_task) | |
3696 | break; | |
1da177e4 | 3697 | } |
dd41f596 | 3698 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3699 | /* |
3700 | * We are going idle. next_balance may be set based on | |
3701 | * a busy processor. So reset next_balance. | |
3702 | */ | |
3703 | this_rq->next_balance = next_balance; | |
dd41f596 | 3704 | } |
1da177e4 LT |
3705 | } |
3706 | ||
3707 | /* | |
3708 | * active_load_balance is run by migration threads. It pushes running tasks | |
3709 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3710 | * running on each physical CPU where possible, and avoids physical / | |
3711 | * logical imbalances. | |
3712 | * | |
3713 | * Called with busiest_rq locked. | |
3714 | */ | |
70b97a7f | 3715 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3716 | { |
39507451 | 3717 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3718 | struct sched_domain *sd; |
3719 | struct rq *target_rq; | |
39507451 | 3720 | |
48f24c4d | 3721 | /* Is there any task to move? */ |
39507451 | 3722 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3723 | return; |
3724 | ||
3725 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3726 | |
3727 | /* | |
39507451 | 3728 | * This condition is "impossible", if it occurs |
41a2d6cf | 3729 | * we need to fix it. Originally reported by |
39507451 | 3730 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3731 | */ |
39507451 | 3732 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3733 | |
39507451 NP |
3734 | /* move a task from busiest_rq to target_rq */ |
3735 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3736 | update_rq_clock(busiest_rq); |
3737 | update_rq_clock(target_rq); | |
39507451 NP |
3738 | |
3739 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3740 | for_each_domain(target_cpu, sd) { |
39507451 | 3741 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3742 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3743 | break; |
c96d145e | 3744 | } |
39507451 | 3745 | |
48f24c4d | 3746 | if (likely(sd)) { |
2d72376b | 3747 | schedstat_inc(sd, alb_count); |
39507451 | 3748 | |
43010659 PW |
3749 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3750 | sd, CPU_IDLE)) | |
48f24c4d IM |
3751 | schedstat_inc(sd, alb_pushed); |
3752 | else | |
3753 | schedstat_inc(sd, alb_failed); | |
3754 | } | |
1b12bbc7 | 3755 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3756 | } |
3757 | ||
46cb4b7c SS |
3758 | #ifdef CONFIG_NO_HZ |
3759 | static struct { | |
3760 | atomic_t load_balancer; | |
41a2d6cf | 3761 | cpumask_t cpu_mask; |
46cb4b7c SS |
3762 | } nohz ____cacheline_aligned = { |
3763 | .load_balancer = ATOMIC_INIT(-1), | |
3764 | .cpu_mask = CPU_MASK_NONE, | |
3765 | }; | |
3766 | ||
7835b98b | 3767 | /* |
46cb4b7c SS |
3768 | * This routine will try to nominate the ilb (idle load balancing) |
3769 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3770 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3771 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3772 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3773 | * arrives... | |
3774 | * | |
3775 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3776 | * for idle load balancing. ilb owner will still be part of | |
3777 | * nohz.cpu_mask.. | |
7835b98b | 3778 | * |
46cb4b7c SS |
3779 | * While stopping the tick, this cpu will become the ilb owner if there |
3780 | * is no other owner. And will be the owner till that cpu becomes busy | |
3781 | * or if all cpus in the system stop their ticks at which point | |
3782 | * there is no need for ilb owner. | |
3783 | * | |
3784 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3785 | * next busy scheduler_tick() | |
3786 | */ | |
3787 | int select_nohz_load_balancer(int stop_tick) | |
3788 | { | |
3789 | int cpu = smp_processor_id(); | |
3790 | ||
3791 | if (stop_tick) { | |
3792 | cpu_set(cpu, nohz.cpu_mask); | |
3793 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3794 | ||
3795 | /* | |
3796 | * If we are going offline and still the leader, give up! | |
3797 | */ | |
e761b772 | 3798 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3799 | atomic_read(&nohz.load_balancer) == cpu) { |
3800 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3801 | BUG(); | |
3802 | return 0; | |
3803 | } | |
3804 | ||
3805 | /* time for ilb owner also to sleep */ | |
3806 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3807 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3808 | atomic_set(&nohz.load_balancer, -1); | |
3809 | return 0; | |
3810 | } | |
3811 | ||
3812 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3813 | /* make me the ilb owner */ | |
3814 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3815 | return 1; | |
3816 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3817 | return 1; | |
3818 | } else { | |
3819 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3820 | return 0; | |
3821 | ||
3822 | cpu_clear(cpu, nohz.cpu_mask); | |
3823 | ||
3824 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3825 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3826 | BUG(); | |
3827 | } | |
3828 | return 0; | |
3829 | } | |
3830 | #endif | |
3831 | ||
3832 | static DEFINE_SPINLOCK(balancing); | |
3833 | ||
3834 | /* | |
7835b98b CL |
3835 | * It checks each scheduling domain to see if it is due to be balanced, |
3836 | * and initiates a balancing operation if so. | |
3837 | * | |
3838 | * Balancing parameters are set up in arch_init_sched_domains. | |
3839 | */ | |
a9957449 | 3840 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3841 | { |
46cb4b7c SS |
3842 | int balance = 1; |
3843 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3844 | unsigned long interval; |
3845 | struct sched_domain *sd; | |
46cb4b7c | 3846 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3847 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3848 | int update_next_balance = 0; |
d07355f5 | 3849 | int need_serialize; |
7c16ec58 | 3850 | cpumask_t tmp; |
1da177e4 | 3851 | |
46cb4b7c | 3852 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3853 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3854 | continue; | |
3855 | ||
3856 | interval = sd->balance_interval; | |
d15bcfdb | 3857 | if (idle != CPU_IDLE) |
1da177e4 LT |
3858 | interval *= sd->busy_factor; |
3859 | ||
3860 | /* scale ms to jiffies */ | |
3861 | interval = msecs_to_jiffies(interval); | |
3862 | if (unlikely(!interval)) | |
3863 | interval = 1; | |
dd41f596 IM |
3864 | if (interval > HZ*NR_CPUS/10) |
3865 | interval = HZ*NR_CPUS/10; | |
3866 | ||
d07355f5 | 3867 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3868 | |
d07355f5 | 3869 | if (need_serialize) { |
08c183f3 CL |
3870 | if (!spin_trylock(&balancing)) |
3871 | goto out; | |
3872 | } | |
3873 | ||
c9819f45 | 3874 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3875 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3876 | /* |
3877 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3878 | * longer idle, or one of our SMT siblings is |
3879 | * not idle. | |
3880 | */ | |
d15bcfdb | 3881 | idle = CPU_NOT_IDLE; |
1da177e4 | 3882 | } |
1bd77f2d | 3883 | sd->last_balance = jiffies; |
1da177e4 | 3884 | } |
d07355f5 | 3885 | if (need_serialize) |
08c183f3 CL |
3886 | spin_unlock(&balancing); |
3887 | out: | |
f549da84 | 3888 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3889 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3890 | update_next_balance = 1; |
3891 | } | |
783609c6 SS |
3892 | |
3893 | /* | |
3894 | * Stop the load balance at this level. There is another | |
3895 | * CPU in our sched group which is doing load balancing more | |
3896 | * actively. | |
3897 | */ | |
3898 | if (!balance) | |
3899 | break; | |
1da177e4 | 3900 | } |
f549da84 SS |
3901 | |
3902 | /* | |
3903 | * next_balance will be updated only when there is a need. | |
3904 | * When the cpu is attached to null domain for ex, it will not be | |
3905 | * updated. | |
3906 | */ | |
3907 | if (likely(update_next_balance)) | |
3908 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3909 | } |
3910 | ||
3911 | /* | |
3912 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3913 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3914 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3915 | */ | |
3916 | static void run_rebalance_domains(struct softirq_action *h) | |
3917 | { | |
dd41f596 IM |
3918 | int this_cpu = smp_processor_id(); |
3919 | struct rq *this_rq = cpu_rq(this_cpu); | |
3920 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3921 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3922 | |
dd41f596 | 3923 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3924 | |
3925 | #ifdef CONFIG_NO_HZ | |
3926 | /* | |
3927 | * If this cpu is the owner for idle load balancing, then do the | |
3928 | * balancing on behalf of the other idle cpus whose ticks are | |
3929 | * stopped. | |
3930 | */ | |
dd41f596 IM |
3931 | if (this_rq->idle_at_tick && |
3932 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3933 | cpumask_t cpus = nohz.cpu_mask; |
3934 | struct rq *rq; | |
3935 | int balance_cpu; | |
3936 | ||
dd41f596 | 3937 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3938 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3939 | /* |
3940 | * If this cpu gets work to do, stop the load balancing | |
3941 | * work being done for other cpus. Next load | |
3942 | * balancing owner will pick it up. | |
3943 | */ | |
3944 | if (need_resched()) | |
3945 | break; | |
3946 | ||
de0cf899 | 3947 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3948 | |
3949 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3950 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3951 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3952 | } |
3953 | } | |
3954 | #endif | |
3955 | } | |
3956 | ||
3957 | /* | |
3958 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3959 | * | |
3960 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3961 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3962 | * if the whole system is idle. | |
3963 | */ | |
dd41f596 | 3964 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3965 | { |
46cb4b7c SS |
3966 | #ifdef CONFIG_NO_HZ |
3967 | /* | |
3968 | * If we were in the nohz mode recently and busy at the current | |
3969 | * scheduler tick, then check if we need to nominate new idle | |
3970 | * load balancer. | |
3971 | */ | |
3972 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3973 | rq->in_nohz_recently = 0; | |
3974 | ||
3975 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3976 | cpu_clear(cpu, nohz.cpu_mask); | |
3977 | atomic_set(&nohz.load_balancer, -1); | |
3978 | } | |
3979 | ||
3980 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3981 | /* | |
3982 | * simple selection for now: Nominate the | |
3983 | * first cpu in the nohz list to be the next | |
3984 | * ilb owner. | |
3985 | * | |
3986 | * TBD: Traverse the sched domains and nominate | |
3987 | * the nearest cpu in the nohz.cpu_mask. | |
3988 | */ | |
3989 | int ilb = first_cpu(nohz.cpu_mask); | |
3990 | ||
434d53b0 | 3991 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3992 | resched_cpu(ilb); |
3993 | } | |
3994 | } | |
3995 | ||
3996 | /* | |
3997 | * If this cpu is idle and doing idle load balancing for all the | |
3998 | * cpus with ticks stopped, is it time for that to stop? | |
3999 | */ | |
4000 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4001 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4002 | resched_cpu(cpu); | |
4003 | return; | |
4004 | } | |
4005 | ||
4006 | /* | |
4007 | * If this cpu is idle and the idle load balancing is done by | |
4008 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4009 | */ | |
4010 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4011 | cpu_isset(cpu, nohz.cpu_mask)) | |
4012 | return; | |
4013 | #endif | |
4014 | if (time_after_eq(jiffies, rq->next_balance)) | |
4015 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4016 | } |
dd41f596 IM |
4017 | |
4018 | #else /* CONFIG_SMP */ | |
4019 | ||
1da177e4 LT |
4020 | /* |
4021 | * on UP we do not need to balance between CPUs: | |
4022 | */ | |
70b97a7f | 4023 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4024 | { |
4025 | } | |
dd41f596 | 4026 | |
1da177e4 LT |
4027 | #endif |
4028 | ||
1da177e4 LT |
4029 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4030 | ||
4031 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4032 | ||
4033 | /* | |
f06febc9 FM |
4034 | * Return any ns on the sched_clock that have not yet been banked in |
4035 | * @p in case that task is currently running. | |
1da177e4 | 4036 | */ |
bb34d92f | 4037 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4038 | { |
1da177e4 | 4039 | unsigned long flags; |
41b86e9c | 4040 | struct rq *rq; |
bb34d92f | 4041 | u64 ns = 0; |
48f24c4d | 4042 | |
41b86e9c | 4043 | rq = task_rq_lock(p, &flags); |
1508487e | 4044 | |
051a1d1a | 4045 | if (task_current(rq, p)) { |
f06febc9 FM |
4046 | u64 delta_exec; |
4047 | ||
a8e504d2 IM |
4048 | update_rq_clock(rq); |
4049 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4050 | if ((s64)delta_exec > 0) |
bb34d92f | 4051 | ns = delta_exec; |
41b86e9c | 4052 | } |
48f24c4d | 4053 | |
41b86e9c | 4054 | task_rq_unlock(rq, &flags); |
48f24c4d | 4055 | |
1da177e4 LT |
4056 | return ns; |
4057 | } | |
4058 | ||
1da177e4 LT |
4059 | /* |
4060 | * Account user cpu time to a process. | |
4061 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4062 | * @cputime: the cpu time spent in user space since the last update |
4063 | */ | |
4064 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4065 | { | |
4066 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4067 | cputime64_t tmp; | |
4068 | ||
4069 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4070 | account_group_user_time(p, cputime); |
1da177e4 LT |
4071 | |
4072 | /* Add user time to cpustat. */ | |
4073 | tmp = cputime_to_cputime64(cputime); | |
4074 | if (TASK_NICE(p) > 0) | |
4075 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4076 | else | |
4077 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4078 | /* Account for user time used */ |
4079 | acct_update_integrals(p); | |
1da177e4 LT |
4080 | } |
4081 | ||
94886b84 LV |
4082 | /* |
4083 | * Account guest cpu time to a process. | |
4084 | * @p: the process that the cpu time gets accounted to | |
4085 | * @cputime: the cpu time spent in virtual machine since the last update | |
4086 | */ | |
f7402e03 | 4087 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4088 | { |
4089 | cputime64_t tmp; | |
4090 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4091 | ||
4092 | tmp = cputime_to_cputime64(cputime); | |
4093 | ||
4094 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4095 | account_group_user_time(p, cputime); |
94886b84 LV |
4096 | p->gtime = cputime_add(p->gtime, cputime); |
4097 | ||
4098 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4099 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4100 | } | |
4101 | ||
c66f08be MN |
4102 | /* |
4103 | * Account scaled user cpu time to a process. | |
4104 | * @p: the process that the cpu time gets accounted to | |
4105 | * @cputime: the cpu time spent in user space since the last update | |
4106 | */ | |
4107 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4108 | { | |
4109 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4110 | } | |
4111 | ||
1da177e4 LT |
4112 | /* |
4113 | * Account system cpu time to a process. | |
4114 | * @p: the process that the cpu time gets accounted to | |
4115 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4116 | * @cputime: the cpu time spent in kernel space since the last update | |
4117 | */ | |
4118 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4119 | cputime_t cputime) | |
4120 | { | |
4121 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4122 | struct rq *rq = this_rq(); |
1da177e4 LT |
4123 | cputime64_t tmp; |
4124 | ||
983ed7a6 HH |
4125 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4126 | account_guest_time(p, cputime); | |
4127 | return; | |
4128 | } | |
94886b84 | 4129 | |
1da177e4 | 4130 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4131 | account_group_system_time(p, cputime); |
1da177e4 LT |
4132 | |
4133 | /* Add system time to cpustat. */ | |
4134 | tmp = cputime_to_cputime64(cputime); | |
4135 | if (hardirq_count() - hardirq_offset) | |
4136 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4137 | else if (softirq_count()) | |
4138 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4139 | else if (p != rq->idle) |
1da177e4 | 4140 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4141 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4142 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4143 | else | |
4144 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4145 | /* Account for system time used */ | |
4146 | acct_update_integrals(p); | |
1da177e4 LT |
4147 | } |
4148 | ||
c66f08be MN |
4149 | /* |
4150 | * Account scaled system cpu time to a process. | |
4151 | * @p: the process that the cpu time gets accounted to | |
4152 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4153 | * @cputime: the cpu time spent in kernel space since the last update | |
4154 | */ | |
4155 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4156 | { | |
4157 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4158 | } | |
4159 | ||
1da177e4 LT |
4160 | /* |
4161 | * Account for involuntary wait time. | |
4162 | * @p: the process from which the cpu time has been stolen | |
4163 | * @steal: the cpu time spent in involuntary wait | |
4164 | */ | |
4165 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4166 | { | |
4167 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4168 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4169 | struct rq *rq = this_rq(); |
1da177e4 LT |
4170 | |
4171 | if (p == rq->idle) { | |
4172 | p->stime = cputime_add(p->stime, steal); | |
f06febc9 | 4173 | account_group_system_time(p, steal); |
1da177e4 LT |
4174 | if (atomic_read(&rq->nr_iowait) > 0) |
4175 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4176 | else | |
4177 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4178 | } else |
1da177e4 LT |
4179 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4180 | } | |
4181 | ||
49048622 BS |
4182 | /* |
4183 | * Use precise platform statistics if available: | |
4184 | */ | |
4185 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4186 | cputime_t task_utime(struct task_struct *p) | |
4187 | { | |
4188 | return p->utime; | |
4189 | } | |
4190 | ||
4191 | cputime_t task_stime(struct task_struct *p) | |
4192 | { | |
4193 | return p->stime; | |
4194 | } | |
4195 | #else | |
4196 | cputime_t task_utime(struct task_struct *p) | |
4197 | { | |
4198 | clock_t utime = cputime_to_clock_t(p->utime), | |
4199 | total = utime + cputime_to_clock_t(p->stime); | |
4200 | u64 temp; | |
4201 | ||
4202 | /* | |
4203 | * Use CFS's precise accounting: | |
4204 | */ | |
4205 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4206 | ||
4207 | if (total) { | |
4208 | temp *= utime; | |
4209 | do_div(temp, total); | |
4210 | } | |
4211 | utime = (clock_t)temp; | |
4212 | ||
4213 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4214 | return p->prev_utime; | |
4215 | } | |
4216 | ||
4217 | cputime_t task_stime(struct task_struct *p) | |
4218 | { | |
4219 | clock_t stime; | |
4220 | ||
4221 | /* | |
4222 | * Use CFS's precise accounting. (we subtract utime from | |
4223 | * the total, to make sure the total observed by userspace | |
4224 | * grows monotonically - apps rely on that): | |
4225 | */ | |
4226 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4227 | cputime_to_clock_t(task_utime(p)); | |
4228 | ||
4229 | if (stime >= 0) | |
4230 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4231 | ||
4232 | return p->prev_stime; | |
4233 | } | |
4234 | #endif | |
4235 | ||
4236 | inline cputime_t task_gtime(struct task_struct *p) | |
4237 | { | |
4238 | return p->gtime; | |
4239 | } | |
4240 | ||
7835b98b CL |
4241 | /* |
4242 | * This function gets called by the timer code, with HZ frequency. | |
4243 | * We call it with interrupts disabled. | |
4244 | * | |
4245 | * It also gets called by the fork code, when changing the parent's | |
4246 | * timeslices. | |
4247 | */ | |
4248 | void scheduler_tick(void) | |
4249 | { | |
7835b98b CL |
4250 | int cpu = smp_processor_id(); |
4251 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4252 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4253 | |
4254 | sched_clock_tick(); | |
dd41f596 IM |
4255 | |
4256 | spin_lock(&rq->lock); | |
3e51f33f | 4257 | update_rq_clock(rq); |
f1a438d8 | 4258 | update_cpu_load(rq); |
fa85ae24 | 4259 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4260 | spin_unlock(&rq->lock); |
7835b98b | 4261 | |
e418e1c2 | 4262 | #ifdef CONFIG_SMP |
dd41f596 IM |
4263 | rq->idle_at_tick = idle_cpu(cpu); |
4264 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4265 | #endif |
1da177e4 LT |
4266 | } |
4267 | ||
6cd8a4bb SR |
4268 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4269 | defined(CONFIG_PREEMPT_TRACER)) | |
4270 | ||
4271 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4272 | { | |
4273 | if (in_lock_functions(addr)) { | |
4274 | addr = CALLER_ADDR2; | |
4275 | if (in_lock_functions(addr)) | |
4276 | addr = CALLER_ADDR3; | |
4277 | } | |
4278 | return addr; | |
4279 | } | |
1da177e4 | 4280 | |
43627582 | 4281 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4282 | { |
6cd8a4bb | 4283 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4284 | /* |
4285 | * Underflow? | |
4286 | */ | |
9a11b49a IM |
4287 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4288 | return; | |
6cd8a4bb | 4289 | #endif |
1da177e4 | 4290 | preempt_count() += val; |
6cd8a4bb | 4291 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4292 | /* |
4293 | * Spinlock count overflowing soon? | |
4294 | */ | |
33859f7f MOS |
4295 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4296 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4297 | #endif |
4298 | if (preempt_count() == val) | |
4299 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4300 | } |
4301 | EXPORT_SYMBOL(add_preempt_count); | |
4302 | ||
43627582 | 4303 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4304 | { |
6cd8a4bb | 4305 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4306 | /* |
4307 | * Underflow? | |
4308 | */ | |
9a11b49a IM |
4309 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4310 | return; | |
1da177e4 LT |
4311 | /* |
4312 | * Is the spinlock portion underflowing? | |
4313 | */ | |
9a11b49a IM |
4314 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4315 | !(preempt_count() & PREEMPT_MASK))) | |
4316 | return; | |
6cd8a4bb | 4317 | #endif |
9a11b49a | 4318 | |
6cd8a4bb SR |
4319 | if (preempt_count() == val) |
4320 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4321 | preempt_count() -= val; |
4322 | } | |
4323 | EXPORT_SYMBOL(sub_preempt_count); | |
4324 | ||
4325 | #endif | |
4326 | ||
4327 | /* | |
dd41f596 | 4328 | * Print scheduling while atomic bug: |
1da177e4 | 4329 | */ |
dd41f596 | 4330 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4331 | { |
838225b4 SS |
4332 | struct pt_regs *regs = get_irq_regs(); |
4333 | ||
4334 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4335 | prev->comm, prev->pid, preempt_count()); | |
4336 | ||
dd41f596 | 4337 | debug_show_held_locks(prev); |
e21f5b15 | 4338 | print_modules(); |
dd41f596 IM |
4339 | if (irqs_disabled()) |
4340 | print_irqtrace_events(prev); | |
838225b4 SS |
4341 | |
4342 | if (regs) | |
4343 | show_regs(regs); | |
4344 | else | |
4345 | dump_stack(); | |
dd41f596 | 4346 | } |
1da177e4 | 4347 | |
dd41f596 IM |
4348 | /* |
4349 | * Various schedule()-time debugging checks and statistics: | |
4350 | */ | |
4351 | static inline void schedule_debug(struct task_struct *prev) | |
4352 | { | |
1da177e4 | 4353 | /* |
41a2d6cf | 4354 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4355 | * schedule() atomically, we ignore that path for now. |
4356 | * Otherwise, whine if we are scheduling when we should not be. | |
4357 | */ | |
3f33a7ce | 4358 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4359 | __schedule_bug(prev); |
4360 | ||
1da177e4 LT |
4361 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4362 | ||
2d72376b | 4363 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4364 | #ifdef CONFIG_SCHEDSTATS |
4365 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4366 | schedstat_inc(this_rq(), bkl_count); |
4367 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4368 | } |
4369 | #endif | |
dd41f596 IM |
4370 | } |
4371 | ||
4372 | /* | |
4373 | * Pick up the highest-prio task: | |
4374 | */ | |
4375 | static inline struct task_struct * | |
ff95f3df | 4376 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4377 | { |
5522d5d5 | 4378 | const struct sched_class *class; |
dd41f596 | 4379 | struct task_struct *p; |
1da177e4 LT |
4380 | |
4381 | /* | |
dd41f596 IM |
4382 | * Optimization: we know that if all tasks are in |
4383 | * the fair class we can call that function directly: | |
1da177e4 | 4384 | */ |
dd41f596 | 4385 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4386 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4387 | if (likely(p)) |
4388 | return p; | |
1da177e4 LT |
4389 | } |
4390 | ||
dd41f596 IM |
4391 | class = sched_class_highest; |
4392 | for ( ; ; ) { | |
fb8d4724 | 4393 | p = class->pick_next_task(rq); |
dd41f596 IM |
4394 | if (p) |
4395 | return p; | |
4396 | /* | |
4397 | * Will never be NULL as the idle class always | |
4398 | * returns a non-NULL p: | |
4399 | */ | |
4400 | class = class->next; | |
4401 | } | |
4402 | } | |
1da177e4 | 4403 | |
dd41f596 IM |
4404 | /* |
4405 | * schedule() is the main scheduler function. | |
4406 | */ | |
4407 | asmlinkage void __sched schedule(void) | |
4408 | { | |
4409 | struct task_struct *prev, *next; | |
67ca7bde | 4410 | unsigned long *switch_count; |
dd41f596 | 4411 | struct rq *rq; |
31656519 | 4412 | int cpu; |
dd41f596 IM |
4413 | |
4414 | need_resched: | |
4415 | preempt_disable(); | |
4416 | cpu = smp_processor_id(); | |
4417 | rq = cpu_rq(cpu); | |
4418 | rcu_qsctr_inc(cpu); | |
4419 | prev = rq->curr; | |
4420 | switch_count = &prev->nivcsw; | |
4421 | ||
4422 | release_kernel_lock(prev); | |
4423 | need_resched_nonpreemptible: | |
4424 | ||
4425 | schedule_debug(prev); | |
1da177e4 | 4426 | |
31656519 | 4427 | if (sched_feat(HRTICK)) |
f333fdc9 | 4428 | hrtick_clear(rq); |
8f4d37ec | 4429 | |
8cd162ce | 4430 | spin_lock_irq(&rq->lock); |
3e51f33f | 4431 | update_rq_clock(rq); |
1e819950 | 4432 | clear_tsk_need_resched(prev); |
1da177e4 | 4433 | |
1da177e4 | 4434 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4435 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4436 | prev->state = TASK_RUNNING; |
16882c1e | 4437 | else |
2e1cb74a | 4438 | deactivate_task(rq, prev, 1); |
dd41f596 | 4439 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4440 | } |
4441 | ||
9a897c5a SR |
4442 | #ifdef CONFIG_SMP |
4443 | if (prev->sched_class->pre_schedule) | |
4444 | prev->sched_class->pre_schedule(rq, prev); | |
4445 | #endif | |
f65eda4f | 4446 | |
dd41f596 | 4447 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4448 | idle_balance(cpu, rq); |
1da177e4 | 4449 | |
31ee529c | 4450 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4451 | next = pick_next_task(rq, prev); |
1da177e4 | 4452 | |
1da177e4 | 4453 | if (likely(prev != next)) { |
673a90a1 DS |
4454 | sched_info_switch(prev, next); |
4455 | ||
1da177e4 LT |
4456 | rq->nr_switches++; |
4457 | rq->curr = next; | |
4458 | ++*switch_count; | |
4459 | ||
dd41f596 | 4460 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4461 | /* |
4462 | * the context switch might have flipped the stack from under | |
4463 | * us, hence refresh the local variables. | |
4464 | */ | |
4465 | cpu = smp_processor_id(); | |
4466 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4467 | } else |
4468 | spin_unlock_irq(&rq->lock); | |
4469 | ||
8f4d37ec | 4470 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4471 | goto need_resched_nonpreemptible; |
8f4d37ec | 4472 | |
1da177e4 LT |
4473 | preempt_enable_no_resched(); |
4474 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4475 | goto need_resched; | |
4476 | } | |
1da177e4 LT |
4477 | EXPORT_SYMBOL(schedule); |
4478 | ||
4479 | #ifdef CONFIG_PREEMPT | |
4480 | /* | |
2ed6e34f | 4481 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4482 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4483 | * occur there and call schedule directly. |
4484 | */ | |
4485 | asmlinkage void __sched preempt_schedule(void) | |
4486 | { | |
4487 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4488 | |
1da177e4 LT |
4489 | /* |
4490 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4491 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4492 | */ |
beed33a8 | 4493 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4494 | return; |
4495 | ||
3a5c359a AK |
4496 | do { |
4497 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4498 | schedule(); |
3a5c359a | 4499 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4500 | |
3a5c359a AK |
4501 | /* |
4502 | * Check again in case we missed a preemption opportunity | |
4503 | * between schedule and now. | |
4504 | */ | |
4505 | barrier(); | |
4506 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4507 | } |
1da177e4 LT |
4508 | EXPORT_SYMBOL(preempt_schedule); |
4509 | ||
4510 | /* | |
2ed6e34f | 4511 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4512 | * off of irq context. |
4513 | * Note, that this is called and return with irqs disabled. This will | |
4514 | * protect us against recursive calling from irq. | |
4515 | */ | |
4516 | asmlinkage void __sched preempt_schedule_irq(void) | |
4517 | { | |
4518 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4519 | |
2ed6e34f | 4520 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4521 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4522 | ||
3a5c359a AK |
4523 | do { |
4524 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4525 | local_irq_enable(); |
4526 | schedule(); | |
4527 | local_irq_disable(); | |
3a5c359a | 4528 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4529 | |
3a5c359a AK |
4530 | /* |
4531 | * Check again in case we missed a preemption opportunity | |
4532 | * between schedule and now. | |
4533 | */ | |
4534 | barrier(); | |
4535 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4536 | } |
4537 | ||
4538 | #endif /* CONFIG_PREEMPT */ | |
4539 | ||
95cdf3b7 IM |
4540 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4541 | void *key) | |
1da177e4 | 4542 | { |
48f24c4d | 4543 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4544 | } |
1da177e4 LT |
4545 | EXPORT_SYMBOL(default_wake_function); |
4546 | ||
4547 | /* | |
41a2d6cf IM |
4548 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4549 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4550 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4551 | * | |
4552 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4553 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4554 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4555 | */ | |
4556 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4557 | int nr_exclusive, int sync, void *key) | |
4558 | { | |
2e45874c | 4559 | wait_queue_t *curr, *next; |
1da177e4 | 4560 | |
2e45874c | 4561 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4562 | unsigned flags = curr->flags; |
4563 | ||
1da177e4 | 4564 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4565 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4566 | break; |
4567 | } | |
4568 | } | |
4569 | ||
4570 | /** | |
4571 | * __wake_up - wake up threads blocked on a waitqueue. | |
4572 | * @q: the waitqueue | |
4573 | * @mode: which threads | |
4574 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4575 | * @key: is directly passed to the wakeup function |
1da177e4 | 4576 | */ |
7ad5b3a5 | 4577 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4578 | int nr_exclusive, void *key) |
1da177e4 LT |
4579 | { |
4580 | unsigned long flags; | |
4581 | ||
4582 | spin_lock_irqsave(&q->lock, flags); | |
4583 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4584 | spin_unlock_irqrestore(&q->lock, flags); | |
4585 | } | |
1da177e4 LT |
4586 | EXPORT_SYMBOL(__wake_up); |
4587 | ||
4588 | /* | |
4589 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4590 | */ | |
7ad5b3a5 | 4591 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4592 | { |
4593 | __wake_up_common(q, mode, 1, 0, NULL); | |
4594 | } | |
4595 | ||
4596 | /** | |
67be2dd1 | 4597 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4598 | * @q: the waitqueue |
4599 | * @mode: which threads | |
4600 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4601 | * | |
4602 | * The sync wakeup differs that the waker knows that it will schedule | |
4603 | * away soon, so while the target thread will be woken up, it will not | |
4604 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4605 | * with each other. This can prevent needless bouncing between CPUs. | |
4606 | * | |
4607 | * On UP it can prevent extra preemption. | |
4608 | */ | |
7ad5b3a5 | 4609 | void |
95cdf3b7 | 4610 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4611 | { |
4612 | unsigned long flags; | |
4613 | int sync = 1; | |
4614 | ||
4615 | if (unlikely(!q)) | |
4616 | return; | |
4617 | ||
4618 | if (unlikely(!nr_exclusive)) | |
4619 | sync = 0; | |
4620 | ||
4621 | spin_lock_irqsave(&q->lock, flags); | |
4622 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4623 | spin_unlock_irqrestore(&q->lock, flags); | |
4624 | } | |
4625 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4626 | ||
65eb3dc6 KD |
4627 | /** |
4628 | * complete: - signals a single thread waiting on this completion | |
4629 | * @x: holds the state of this particular completion | |
4630 | * | |
4631 | * This will wake up a single thread waiting on this completion. Threads will be | |
4632 | * awakened in the same order in which they were queued. | |
4633 | * | |
4634 | * See also complete_all(), wait_for_completion() and related routines. | |
4635 | */ | |
b15136e9 | 4636 | void complete(struct completion *x) |
1da177e4 LT |
4637 | { |
4638 | unsigned long flags; | |
4639 | ||
4640 | spin_lock_irqsave(&x->wait.lock, flags); | |
4641 | x->done++; | |
d9514f6c | 4642 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4643 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4644 | } | |
4645 | EXPORT_SYMBOL(complete); | |
4646 | ||
65eb3dc6 KD |
4647 | /** |
4648 | * complete_all: - signals all threads waiting on this completion | |
4649 | * @x: holds the state of this particular completion | |
4650 | * | |
4651 | * This will wake up all threads waiting on this particular completion event. | |
4652 | */ | |
b15136e9 | 4653 | void complete_all(struct completion *x) |
1da177e4 LT |
4654 | { |
4655 | unsigned long flags; | |
4656 | ||
4657 | spin_lock_irqsave(&x->wait.lock, flags); | |
4658 | x->done += UINT_MAX/2; | |
d9514f6c | 4659 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4660 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4661 | } | |
4662 | EXPORT_SYMBOL(complete_all); | |
4663 | ||
8cbbe86d AK |
4664 | static inline long __sched |
4665 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4666 | { |
1da177e4 LT |
4667 | if (!x->done) { |
4668 | DECLARE_WAITQUEUE(wait, current); | |
4669 | ||
4670 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4671 | __add_wait_queue_tail(&x->wait, &wait); | |
4672 | do { | |
94d3d824 | 4673 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4674 | timeout = -ERESTARTSYS; |
4675 | break; | |
8cbbe86d AK |
4676 | } |
4677 | __set_current_state(state); | |
1da177e4 LT |
4678 | spin_unlock_irq(&x->wait.lock); |
4679 | timeout = schedule_timeout(timeout); | |
4680 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4681 | } while (!x->done && timeout); |
1da177e4 | 4682 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4683 | if (!x->done) |
4684 | return timeout; | |
1da177e4 LT |
4685 | } |
4686 | x->done--; | |
ea71a546 | 4687 | return timeout ?: 1; |
1da177e4 | 4688 | } |
1da177e4 | 4689 | |
8cbbe86d AK |
4690 | static long __sched |
4691 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4692 | { |
1da177e4 LT |
4693 | might_sleep(); |
4694 | ||
4695 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4696 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4697 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4698 | return timeout; |
4699 | } | |
1da177e4 | 4700 | |
65eb3dc6 KD |
4701 | /** |
4702 | * wait_for_completion: - waits for completion of a task | |
4703 | * @x: holds the state of this particular completion | |
4704 | * | |
4705 | * This waits to be signaled for completion of a specific task. It is NOT | |
4706 | * interruptible and there is no timeout. | |
4707 | * | |
4708 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4709 | * and interrupt capability. Also see complete(). | |
4710 | */ | |
b15136e9 | 4711 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4712 | { |
4713 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4714 | } |
8cbbe86d | 4715 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4716 | |
65eb3dc6 KD |
4717 | /** |
4718 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4719 | * @x: holds the state of this particular completion | |
4720 | * @timeout: timeout value in jiffies | |
4721 | * | |
4722 | * This waits for either a completion of a specific task to be signaled or for a | |
4723 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4724 | * interruptible. | |
4725 | */ | |
b15136e9 | 4726 | unsigned long __sched |
8cbbe86d | 4727 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4728 | { |
8cbbe86d | 4729 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4730 | } |
8cbbe86d | 4731 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4732 | |
65eb3dc6 KD |
4733 | /** |
4734 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4735 | * @x: holds the state of this particular completion | |
4736 | * | |
4737 | * This waits for completion of a specific task to be signaled. It is | |
4738 | * interruptible. | |
4739 | */ | |
8cbbe86d | 4740 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4741 | { |
51e97990 AK |
4742 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4743 | if (t == -ERESTARTSYS) | |
4744 | return t; | |
4745 | return 0; | |
0fec171c | 4746 | } |
8cbbe86d | 4747 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4748 | |
65eb3dc6 KD |
4749 | /** |
4750 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4751 | * @x: holds the state of this particular completion | |
4752 | * @timeout: timeout value in jiffies | |
4753 | * | |
4754 | * This waits for either a completion of a specific task to be signaled or for a | |
4755 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4756 | */ | |
b15136e9 | 4757 | unsigned long __sched |
8cbbe86d AK |
4758 | wait_for_completion_interruptible_timeout(struct completion *x, |
4759 | unsigned long timeout) | |
0fec171c | 4760 | { |
8cbbe86d | 4761 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4762 | } |
8cbbe86d | 4763 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4764 | |
65eb3dc6 KD |
4765 | /** |
4766 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4767 | * @x: holds the state of this particular completion | |
4768 | * | |
4769 | * This waits to be signaled for completion of a specific task. It can be | |
4770 | * interrupted by a kill signal. | |
4771 | */ | |
009e577e MW |
4772 | int __sched wait_for_completion_killable(struct completion *x) |
4773 | { | |
4774 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4775 | if (t == -ERESTARTSYS) | |
4776 | return t; | |
4777 | return 0; | |
4778 | } | |
4779 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4780 | ||
be4de352 DC |
4781 | /** |
4782 | * try_wait_for_completion - try to decrement a completion without blocking | |
4783 | * @x: completion structure | |
4784 | * | |
4785 | * Returns: 0 if a decrement cannot be done without blocking | |
4786 | * 1 if a decrement succeeded. | |
4787 | * | |
4788 | * If a completion is being used as a counting completion, | |
4789 | * attempt to decrement the counter without blocking. This | |
4790 | * enables us to avoid waiting if the resource the completion | |
4791 | * is protecting is not available. | |
4792 | */ | |
4793 | bool try_wait_for_completion(struct completion *x) | |
4794 | { | |
4795 | int ret = 1; | |
4796 | ||
4797 | spin_lock_irq(&x->wait.lock); | |
4798 | if (!x->done) | |
4799 | ret = 0; | |
4800 | else | |
4801 | x->done--; | |
4802 | spin_unlock_irq(&x->wait.lock); | |
4803 | return ret; | |
4804 | } | |
4805 | EXPORT_SYMBOL(try_wait_for_completion); | |
4806 | ||
4807 | /** | |
4808 | * completion_done - Test to see if a completion has any waiters | |
4809 | * @x: completion structure | |
4810 | * | |
4811 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4812 | * 1 if there are no waiters. | |
4813 | * | |
4814 | */ | |
4815 | bool completion_done(struct completion *x) | |
4816 | { | |
4817 | int ret = 1; | |
4818 | ||
4819 | spin_lock_irq(&x->wait.lock); | |
4820 | if (!x->done) | |
4821 | ret = 0; | |
4822 | spin_unlock_irq(&x->wait.lock); | |
4823 | return ret; | |
4824 | } | |
4825 | EXPORT_SYMBOL(completion_done); | |
4826 | ||
8cbbe86d AK |
4827 | static long __sched |
4828 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4829 | { |
0fec171c IM |
4830 | unsigned long flags; |
4831 | wait_queue_t wait; | |
4832 | ||
4833 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4834 | |
8cbbe86d | 4835 | __set_current_state(state); |
1da177e4 | 4836 | |
8cbbe86d AK |
4837 | spin_lock_irqsave(&q->lock, flags); |
4838 | __add_wait_queue(q, &wait); | |
4839 | spin_unlock(&q->lock); | |
4840 | timeout = schedule_timeout(timeout); | |
4841 | spin_lock_irq(&q->lock); | |
4842 | __remove_wait_queue(q, &wait); | |
4843 | spin_unlock_irqrestore(&q->lock, flags); | |
4844 | ||
4845 | return timeout; | |
4846 | } | |
4847 | ||
4848 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4849 | { | |
4850 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4851 | } |
1da177e4 LT |
4852 | EXPORT_SYMBOL(interruptible_sleep_on); |
4853 | ||
0fec171c | 4854 | long __sched |
95cdf3b7 | 4855 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4856 | { |
8cbbe86d | 4857 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4858 | } |
1da177e4 LT |
4859 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4860 | ||
0fec171c | 4861 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4862 | { |
8cbbe86d | 4863 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4864 | } |
1da177e4 LT |
4865 | EXPORT_SYMBOL(sleep_on); |
4866 | ||
0fec171c | 4867 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4868 | { |
8cbbe86d | 4869 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4870 | } |
1da177e4 LT |
4871 | EXPORT_SYMBOL(sleep_on_timeout); |
4872 | ||
b29739f9 IM |
4873 | #ifdef CONFIG_RT_MUTEXES |
4874 | ||
4875 | /* | |
4876 | * rt_mutex_setprio - set the current priority of a task | |
4877 | * @p: task | |
4878 | * @prio: prio value (kernel-internal form) | |
4879 | * | |
4880 | * This function changes the 'effective' priority of a task. It does | |
4881 | * not touch ->normal_prio like __setscheduler(). | |
4882 | * | |
4883 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4884 | */ | |
36c8b586 | 4885 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4886 | { |
4887 | unsigned long flags; | |
83b699ed | 4888 | int oldprio, on_rq, running; |
70b97a7f | 4889 | struct rq *rq; |
cb469845 | 4890 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4891 | |
4892 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4893 | ||
4894 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4895 | update_rq_clock(rq); |
b29739f9 | 4896 | |
d5f9f942 | 4897 | oldprio = p->prio; |
dd41f596 | 4898 | on_rq = p->se.on_rq; |
051a1d1a | 4899 | running = task_current(rq, p); |
0e1f3483 | 4900 | if (on_rq) |
69be72c1 | 4901 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4902 | if (running) |
4903 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4904 | |
4905 | if (rt_prio(prio)) | |
4906 | p->sched_class = &rt_sched_class; | |
4907 | else | |
4908 | p->sched_class = &fair_sched_class; | |
4909 | ||
b29739f9 IM |
4910 | p->prio = prio; |
4911 | ||
0e1f3483 HS |
4912 | if (running) |
4913 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4914 | if (on_rq) { |
8159f87e | 4915 | enqueue_task(rq, p, 0); |
cb469845 SR |
4916 | |
4917 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4918 | } |
4919 | task_rq_unlock(rq, &flags); | |
4920 | } | |
4921 | ||
4922 | #endif | |
4923 | ||
36c8b586 | 4924 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4925 | { |
dd41f596 | 4926 | int old_prio, delta, on_rq; |
1da177e4 | 4927 | unsigned long flags; |
70b97a7f | 4928 | struct rq *rq; |
1da177e4 LT |
4929 | |
4930 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4931 | return; | |
4932 | /* | |
4933 | * We have to be careful, if called from sys_setpriority(), | |
4934 | * the task might be in the middle of scheduling on another CPU. | |
4935 | */ | |
4936 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4937 | update_rq_clock(rq); |
1da177e4 LT |
4938 | /* |
4939 | * The RT priorities are set via sched_setscheduler(), but we still | |
4940 | * allow the 'normal' nice value to be set - but as expected | |
4941 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4942 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4943 | */ |
e05606d3 | 4944 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4945 | p->static_prio = NICE_TO_PRIO(nice); |
4946 | goto out_unlock; | |
4947 | } | |
dd41f596 | 4948 | on_rq = p->se.on_rq; |
c09595f6 | 4949 | if (on_rq) |
69be72c1 | 4950 | dequeue_task(rq, p, 0); |
1da177e4 | 4951 | |
1da177e4 | 4952 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4953 | set_load_weight(p); |
b29739f9 IM |
4954 | old_prio = p->prio; |
4955 | p->prio = effective_prio(p); | |
4956 | delta = p->prio - old_prio; | |
1da177e4 | 4957 | |
dd41f596 | 4958 | if (on_rq) { |
8159f87e | 4959 | enqueue_task(rq, p, 0); |
1da177e4 | 4960 | /* |
d5f9f942 AM |
4961 | * If the task increased its priority or is running and |
4962 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4963 | */ |
d5f9f942 | 4964 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4965 | resched_task(rq->curr); |
4966 | } | |
4967 | out_unlock: | |
4968 | task_rq_unlock(rq, &flags); | |
4969 | } | |
1da177e4 LT |
4970 | EXPORT_SYMBOL(set_user_nice); |
4971 | ||
e43379f1 MM |
4972 | /* |
4973 | * can_nice - check if a task can reduce its nice value | |
4974 | * @p: task | |
4975 | * @nice: nice value | |
4976 | */ | |
36c8b586 | 4977 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4978 | { |
024f4747 MM |
4979 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4980 | int nice_rlim = 20 - nice; | |
48f24c4d | 4981 | |
e43379f1 MM |
4982 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4983 | capable(CAP_SYS_NICE)); | |
4984 | } | |
4985 | ||
1da177e4 LT |
4986 | #ifdef __ARCH_WANT_SYS_NICE |
4987 | ||
4988 | /* | |
4989 | * sys_nice - change the priority of the current process. | |
4990 | * @increment: priority increment | |
4991 | * | |
4992 | * sys_setpriority is a more generic, but much slower function that | |
4993 | * does similar things. | |
4994 | */ | |
4995 | asmlinkage long sys_nice(int increment) | |
4996 | { | |
48f24c4d | 4997 | long nice, retval; |
1da177e4 LT |
4998 | |
4999 | /* | |
5000 | * Setpriority might change our priority at the same moment. | |
5001 | * We don't have to worry. Conceptually one call occurs first | |
5002 | * and we have a single winner. | |
5003 | */ | |
e43379f1 MM |
5004 | if (increment < -40) |
5005 | increment = -40; | |
1da177e4 LT |
5006 | if (increment > 40) |
5007 | increment = 40; | |
5008 | ||
5009 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5010 | if (nice < -20) | |
5011 | nice = -20; | |
5012 | if (nice > 19) | |
5013 | nice = 19; | |
5014 | ||
e43379f1 MM |
5015 | if (increment < 0 && !can_nice(current, nice)) |
5016 | return -EPERM; | |
5017 | ||
1da177e4 LT |
5018 | retval = security_task_setnice(current, nice); |
5019 | if (retval) | |
5020 | return retval; | |
5021 | ||
5022 | set_user_nice(current, nice); | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | #endif | |
5027 | ||
5028 | /** | |
5029 | * task_prio - return the priority value of a given task. | |
5030 | * @p: the task in question. | |
5031 | * | |
5032 | * This is the priority value as seen by users in /proc. | |
5033 | * RT tasks are offset by -200. Normal tasks are centered | |
5034 | * around 0, value goes from -16 to +15. | |
5035 | */ | |
36c8b586 | 5036 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5037 | { |
5038 | return p->prio - MAX_RT_PRIO; | |
5039 | } | |
5040 | ||
5041 | /** | |
5042 | * task_nice - return the nice value of a given task. | |
5043 | * @p: the task in question. | |
5044 | */ | |
36c8b586 | 5045 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5046 | { |
5047 | return TASK_NICE(p); | |
5048 | } | |
150d8bed | 5049 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5050 | |
5051 | /** | |
5052 | * idle_cpu - is a given cpu idle currently? | |
5053 | * @cpu: the processor in question. | |
5054 | */ | |
5055 | int idle_cpu(int cpu) | |
5056 | { | |
5057 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5058 | } | |
5059 | ||
1da177e4 LT |
5060 | /** |
5061 | * idle_task - return the idle task for a given cpu. | |
5062 | * @cpu: the processor in question. | |
5063 | */ | |
36c8b586 | 5064 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5065 | { |
5066 | return cpu_rq(cpu)->idle; | |
5067 | } | |
5068 | ||
5069 | /** | |
5070 | * find_process_by_pid - find a process with a matching PID value. | |
5071 | * @pid: the pid in question. | |
5072 | */ | |
a9957449 | 5073 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5074 | { |
228ebcbe | 5075 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5076 | } |
5077 | ||
5078 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5079 | static void |
5080 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5081 | { |
dd41f596 | 5082 | BUG_ON(p->se.on_rq); |
48f24c4d | 5083 | |
1da177e4 | 5084 | p->policy = policy; |
dd41f596 IM |
5085 | switch (p->policy) { |
5086 | case SCHED_NORMAL: | |
5087 | case SCHED_BATCH: | |
5088 | case SCHED_IDLE: | |
5089 | p->sched_class = &fair_sched_class; | |
5090 | break; | |
5091 | case SCHED_FIFO: | |
5092 | case SCHED_RR: | |
5093 | p->sched_class = &rt_sched_class; | |
5094 | break; | |
5095 | } | |
5096 | ||
1da177e4 | 5097 | p->rt_priority = prio; |
b29739f9 IM |
5098 | p->normal_prio = normal_prio(p); |
5099 | /* we are holding p->pi_lock already */ | |
5100 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5101 | set_load_weight(p); |
1da177e4 LT |
5102 | } |
5103 | ||
961ccddd RR |
5104 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5105 | struct sched_param *param, bool user) | |
1da177e4 | 5106 | { |
83b699ed | 5107 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5108 | unsigned long flags; |
cb469845 | 5109 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5110 | struct rq *rq; |
1da177e4 | 5111 | |
66e5393a SR |
5112 | /* may grab non-irq protected spin_locks */ |
5113 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5114 | recheck: |
5115 | /* double check policy once rq lock held */ | |
5116 | if (policy < 0) | |
5117 | policy = oldpolicy = p->policy; | |
5118 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5119 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5120 | policy != SCHED_IDLE) | |
b0a9499c | 5121 | return -EINVAL; |
1da177e4 LT |
5122 | /* |
5123 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5124 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5125 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5126 | */ |
5127 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5128 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5129 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5130 | return -EINVAL; |
e05606d3 | 5131 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5132 | return -EINVAL; |
5133 | ||
37e4ab3f OC |
5134 | /* |
5135 | * Allow unprivileged RT tasks to decrease priority: | |
5136 | */ | |
961ccddd | 5137 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5138 | if (rt_policy(policy)) { |
8dc3e909 | 5139 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5140 | |
5141 | if (!lock_task_sighand(p, &flags)) | |
5142 | return -ESRCH; | |
5143 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5144 | unlock_task_sighand(p, &flags); | |
5145 | ||
5146 | /* can't set/change the rt policy */ | |
5147 | if (policy != p->policy && !rlim_rtprio) | |
5148 | return -EPERM; | |
5149 | ||
5150 | /* can't increase priority */ | |
5151 | if (param->sched_priority > p->rt_priority && | |
5152 | param->sched_priority > rlim_rtprio) | |
5153 | return -EPERM; | |
5154 | } | |
dd41f596 IM |
5155 | /* |
5156 | * Like positive nice levels, dont allow tasks to | |
5157 | * move out of SCHED_IDLE either: | |
5158 | */ | |
5159 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5160 | return -EPERM; | |
5fe1d75f | 5161 | |
37e4ab3f OC |
5162 | /* can't change other user's priorities */ |
5163 | if ((current->euid != p->euid) && | |
5164 | (current->euid != p->uid)) | |
5165 | return -EPERM; | |
5166 | } | |
1da177e4 | 5167 | |
725aad24 | 5168 | if (user) { |
b68aa230 | 5169 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5170 | /* |
5171 | * Do not allow realtime tasks into groups that have no runtime | |
5172 | * assigned. | |
5173 | */ | |
9a7e0b18 PZ |
5174 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5175 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5176 | return -EPERM; |
b68aa230 PZ |
5177 | #endif |
5178 | ||
725aad24 JF |
5179 | retval = security_task_setscheduler(p, policy, param); |
5180 | if (retval) | |
5181 | return retval; | |
5182 | } | |
5183 | ||
b29739f9 IM |
5184 | /* |
5185 | * make sure no PI-waiters arrive (or leave) while we are | |
5186 | * changing the priority of the task: | |
5187 | */ | |
5188 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5189 | /* |
5190 | * To be able to change p->policy safely, the apropriate | |
5191 | * runqueue lock must be held. | |
5192 | */ | |
b29739f9 | 5193 | rq = __task_rq_lock(p); |
1da177e4 LT |
5194 | /* recheck policy now with rq lock held */ |
5195 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5196 | policy = oldpolicy = -1; | |
b29739f9 IM |
5197 | __task_rq_unlock(rq); |
5198 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5199 | goto recheck; |
5200 | } | |
2daa3577 | 5201 | update_rq_clock(rq); |
dd41f596 | 5202 | on_rq = p->se.on_rq; |
051a1d1a | 5203 | running = task_current(rq, p); |
0e1f3483 | 5204 | if (on_rq) |
2e1cb74a | 5205 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5206 | if (running) |
5207 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5208 | |
1da177e4 | 5209 | oldprio = p->prio; |
dd41f596 | 5210 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5211 | |
0e1f3483 HS |
5212 | if (running) |
5213 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5214 | if (on_rq) { |
5215 | activate_task(rq, p, 0); | |
cb469845 SR |
5216 | |
5217 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5218 | } |
b29739f9 IM |
5219 | __task_rq_unlock(rq); |
5220 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5221 | ||
95e02ca9 TG |
5222 | rt_mutex_adjust_pi(p); |
5223 | ||
1da177e4 LT |
5224 | return 0; |
5225 | } | |
961ccddd RR |
5226 | |
5227 | /** | |
5228 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5229 | * @p: the task in question. | |
5230 | * @policy: new policy. | |
5231 | * @param: structure containing the new RT priority. | |
5232 | * | |
5233 | * NOTE that the task may be already dead. | |
5234 | */ | |
5235 | int sched_setscheduler(struct task_struct *p, int policy, | |
5236 | struct sched_param *param) | |
5237 | { | |
5238 | return __sched_setscheduler(p, policy, param, true); | |
5239 | } | |
1da177e4 LT |
5240 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5241 | ||
961ccddd RR |
5242 | /** |
5243 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5244 | * @p: the task in question. | |
5245 | * @policy: new policy. | |
5246 | * @param: structure containing the new RT priority. | |
5247 | * | |
5248 | * Just like sched_setscheduler, only don't bother checking if the | |
5249 | * current context has permission. For example, this is needed in | |
5250 | * stop_machine(): we create temporary high priority worker threads, | |
5251 | * but our caller might not have that capability. | |
5252 | */ | |
5253 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5254 | struct sched_param *param) | |
5255 | { | |
5256 | return __sched_setscheduler(p, policy, param, false); | |
5257 | } | |
5258 | ||
95cdf3b7 IM |
5259 | static int |
5260 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5261 | { |
1da177e4 LT |
5262 | struct sched_param lparam; |
5263 | struct task_struct *p; | |
36c8b586 | 5264 | int retval; |
1da177e4 LT |
5265 | |
5266 | if (!param || pid < 0) | |
5267 | return -EINVAL; | |
5268 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5269 | return -EFAULT; | |
5fe1d75f ON |
5270 | |
5271 | rcu_read_lock(); | |
5272 | retval = -ESRCH; | |
1da177e4 | 5273 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5274 | if (p != NULL) |
5275 | retval = sched_setscheduler(p, policy, &lparam); | |
5276 | rcu_read_unlock(); | |
36c8b586 | 5277 | |
1da177e4 LT |
5278 | return retval; |
5279 | } | |
5280 | ||
5281 | /** | |
5282 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5283 | * @pid: the pid in question. | |
5284 | * @policy: new policy. | |
5285 | * @param: structure containing the new RT priority. | |
5286 | */ | |
41a2d6cf IM |
5287 | asmlinkage long |
5288 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5289 | { |
c21761f1 JB |
5290 | /* negative values for policy are not valid */ |
5291 | if (policy < 0) | |
5292 | return -EINVAL; | |
5293 | ||
1da177e4 LT |
5294 | return do_sched_setscheduler(pid, policy, param); |
5295 | } | |
5296 | ||
5297 | /** | |
5298 | * sys_sched_setparam - set/change the RT priority of a thread | |
5299 | * @pid: the pid in question. | |
5300 | * @param: structure containing the new RT priority. | |
5301 | */ | |
5302 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5303 | { | |
5304 | return do_sched_setscheduler(pid, -1, param); | |
5305 | } | |
5306 | ||
5307 | /** | |
5308 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5309 | * @pid: the pid in question. | |
5310 | */ | |
5311 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5312 | { | |
36c8b586 | 5313 | struct task_struct *p; |
3a5c359a | 5314 | int retval; |
1da177e4 LT |
5315 | |
5316 | if (pid < 0) | |
3a5c359a | 5317 | return -EINVAL; |
1da177e4 LT |
5318 | |
5319 | retval = -ESRCH; | |
5320 | read_lock(&tasklist_lock); | |
5321 | p = find_process_by_pid(pid); | |
5322 | if (p) { | |
5323 | retval = security_task_getscheduler(p); | |
5324 | if (!retval) | |
5325 | retval = p->policy; | |
5326 | } | |
5327 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5328 | return retval; |
5329 | } | |
5330 | ||
5331 | /** | |
5332 | * sys_sched_getscheduler - get the RT priority of a thread | |
5333 | * @pid: the pid in question. | |
5334 | * @param: structure containing the RT priority. | |
5335 | */ | |
5336 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5337 | { | |
5338 | struct sched_param lp; | |
36c8b586 | 5339 | struct task_struct *p; |
3a5c359a | 5340 | int retval; |
1da177e4 LT |
5341 | |
5342 | if (!param || pid < 0) | |
3a5c359a | 5343 | return -EINVAL; |
1da177e4 LT |
5344 | |
5345 | read_lock(&tasklist_lock); | |
5346 | p = find_process_by_pid(pid); | |
5347 | retval = -ESRCH; | |
5348 | if (!p) | |
5349 | goto out_unlock; | |
5350 | ||
5351 | retval = security_task_getscheduler(p); | |
5352 | if (retval) | |
5353 | goto out_unlock; | |
5354 | ||
5355 | lp.sched_priority = p->rt_priority; | |
5356 | read_unlock(&tasklist_lock); | |
5357 | ||
5358 | /* | |
5359 | * This one might sleep, we cannot do it with a spinlock held ... | |
5360 | */ | |
5361 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5362 | ||
1da177e4 LT |
5363 | return retval; |
5364 | ||
5365 | out_unlock: | |
5366 | read_unlock(&tasklist_lock); | |
5367 | return retval; | |
5368 | } | |
5369 | ||
b53e921b | 5370 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5371 | { |
1da177e4 | 5372 | cpumask_t cpus_allowed; |
b53e921b | 5373 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5374 | struct task_struct *p; |
5375 | int retval; | |
1da177e4 | 5376 | |
95402b38 | 5377 | get_online_cpus(); |
1da177e4 LT |
5378 | read_lock(&tasklist_lock); |
5379 | ||
5380 | p = find_process_by_pid(pid); | |
5381 | if (!p) { | |
5382 | read_unlock(&tasklist_lock); | |
95402b38 | 5383 | put_online_cpus(); |
1da177e4 LT |
5384 | return -ESRCH; |
5385 | } | |
5386 | ||
5387 | /* | |
5388 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5389 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5390 | * usage count and then drop tasklist_lock. |
5391 | */ | |
5392 | get_task_struct(p); | |
5393 | read_unlock(&tasklist_lock); | |
5394 | ||
5395 | retval = -EPERM; | |
5396 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5397 | !capable(CAP_SYS_NICE)) | |
5398 | goto out_unlock; | |
5399 | ||
e7834f8f DQ |
5400 | retval = security_task_setscheduler(p, 0, NULL); |
5401 | if (retval) | |
5402 | goto out_unlock; | |
5403 | ||
f9a86fcb | 5404 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5405 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5406 | again: |
7c16ec58 | 5407 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5408 | |
8707d8b8 | 5409 | if (!retval) { |
f9a86fcb | 5410 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5411 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5412 | /* | |
5413 | * We must have raced with a concurrent cpuset | |
5414 | * update. Just reset the cpus_allowed to the | |
5415 | * cpuset's cpus_allowed | |
5416 | */ | |
5417 | new_mask = cpus_allowed; | |
5418 | goto again; | |
5419 | } | |
5420 | } | |
1da177e4 LT |
5421 | out_unlock: |
5422 | put_task_struct(p); | |
95402b38 | 5423 | put_online_cpus(); |
1da177e4 LT |
5424 | return retval; |
5425 | } | |
5426 | ||
5427 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5428 | cpumask_t *new_mask) | |
5429 | { | |
5430 | if (len < sizeof(cpumask_t)) { | |
5431 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5432 | } else if (len > sizeof(cpumask_t)) { | |
5433 | len = sizeof(cpumask_t); | |
5434 | } | |
5435 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5436 | } | |
5437 | ||
5438 | /** | |
5439 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5440 | * @pid: pid of the process | |
5441 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5442 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5443 | */ | |
5444 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5445 | unsigned long __user *user_mask_ptr) | |
5446 | { | |
5447 | cpumask_t new_mask; | |
5448 | int retval; | |
5449 | ||
5450 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5451 | if (retval) | |
5452 | return retval; | |
5453 | ||
b53e921b | 5454 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5455 | } |
5456 | ||
1da177e4 LT |
5457 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5458 | { | |
36c8b586 | 5459 | struct task_struct *p; |
1da177e4 | 5460 | int retval; |
1da177e4 | 5461 | |
95402b38 | 5462 | get_online_cpus(); |
1da177e4 LT |
5463 | read_lock(&tasklist_lock); |
5464 | ||
5465 | retval = -ESRCH; | |
5466 | p = find_process_by_pid(pid); | |
5467 | if (!p) | |
5468 | goto out_unlock; | |
5469 | ||
e7834f8f DQ |
5470 | retval = security_task_getscheduler(p); |
5471 | if (retval) | |
5472 | goto out_unlock; | |
5473 | ||
2f7016d9 | 5474 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5475 | |
5476 | out_unlock: | |
5477 | read_unlock(&tasklist_lock); | |
95402b38 | 5478 | put_online_cpus(); |
1da177e4 | 5479 | |
9531b62f | 5480 | return retval; |
1da177e4 LT |
5481 | } |
5482 | ||
5483 | /** | |
5484 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5485 | * @pid: pid of the process | |
5486 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5487 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5488 | */ | |
5489 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5490 | unsigned long __user *user_mask_ptr) | |
5491 | { | |
5492 | int ret; | |
5493 | cpumask_t mask; | |
5494 | ||
5495 | if (len < sizeof(cpumask_t)) | |
5496 | return -EINVAL; | |
5497 | ||
5498 | ret = sched_getaffinity(pid, &mask); | |
5499 | if (ret < 0) | |
5500 | return ret; | |
5501 | ||
5502 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5503 | return -EFAULT; | |
5504 | ||
5505 | return sizeof(cpumask_t); | |
5506 | } | |
5507 | ||
5508 | /** | |
5509 | * sys_sched_yield - yield the current processor to other threads. | |
5510 | * | |
dd41f596 IM |
5511 | * This function yields the current CPU to other tasks. If there are no |
5512 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5513 | */ |
5514 | asmlinkage long sys_sched_yield(void) | |
5515 | { | |
70b97a7f | 5516 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5517 | |
2d72376b | 5518 | schedstat_inc(rq, yld_count); |
4530d7ab | 5519 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5520 | |
5521 | /* | |
5522 | * Since we are going to call schedule() anyway, there's | |
5523 | * no need to preempt or enable interrupts: | |
5524 | */ | |
5525 | __release(rq->lock); | |
8a25d5de | 5526 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5527 | _raw_spin_unlock(&rq->lock); |
5528 | preempt_enable_no_resched(); | |
5529 | ||
5530 | schedule(); | |
5531 | ||
5532 | return 0; | |
5533 | } | |
5534 | ||
e7b38404 | 5535 | static void __cond_resched(void) |
1da177e4 | 5536 | { |
8e0a43d8 IM |
5537 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5538 | __might_sleep(__FILE__, __LINE__); | |
5539 | #endif | |
5bbcfd90 IM |
5540 | /* |
5541 | * The BKS might be reacquired before we have dropped | |
5542 | * PREEMPT_ACTIVE, which could trigger a second | |
5543 | * cond_resched() call. | |
5544 | */ | |
1da177e4 LT |
5545 | do { |
5546 | add_preempt_count(PREEMPT_ACTIVE); | |
5547 | schedule(); | |
5548 | sub_preempt_count(PREEMPT_ACTIVE); | |
5549 | } while (need_resched()); | |
5550 | } | |
5551 | ||
02b67cc3 | 5552 | int __sched _cond_resched(void) |
1da177e4 | 5553 | { |
9414232f IM |
5554 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5555 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5556 | __cond_resched(); |
5557 | return 1; | |
5558 | } | |
5559 | return 0; | |
5560 | } | |
02b67cc3 | 5561 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5562 | |
5563 | /* | |
5564 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5565 | * call schedule, and on return reacquire the lock. | |
5566 | * | |
41a2d6cf | 5567 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5568 | * operations here to prevent schedule() from being called twice (once via |
5569 | * spin_unlock(), once by hand). | |
5570 | */ | |
95cdf3b7 | 5571 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5572 | { |
95c354fe | 5573 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5574 | int ret = 0; |
5575 | ||
95c354fe | 5576 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5577 | spin_unlock(lock); |
95c354fe NP |
5578 | if (resched && need_resched()) |
5579 | __cond_resched(); | |
5580 | else | |
5581 | cpu_relax(); | |
6df3cecb | 5582 | ret = 1; |
1da177e4 | 5583 | spin_lock(lock); |
1da177e4 | 5584 | } |
6df3cecb | 5585 | return ret; |
1da177e4 | 5586 | } |
1da177e4 LT |
5587 | EXPORT_SYMBOL(cond_resched_lock); |
5588 | ||
5589 | int __sched cond_resched_softirq(void) | |
5590 | { | |
5591 | BUG_ON(!in_softirq()); | |
5592 | ||
9414232f | 5593 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5594 | local_bh_enable(); |
1da177e4 LT |
5595 | __cond_resched(); |
5596 | local_bh_disable(); | |
5597 | return 1; | |
5598 | } | |
5599 | return 0; | |
5600 | } | |
1da177e4 LT |
5601 | EXPORT_SYMBOL(cond_resched_softirq); |
5602 | ||
1da177e4 LT |
5603 | /** |
5604 | * yield - yield the current processor to other threads. | |
5605 | * | |
72fd4a35 | 5606 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5607 | * thread runnable and calls sys_sched_yield(). |
5608 | */ | |
5609 | void __sched yield(void) | |
5610 | { | |
5611 | set_current_state(TASK_RUNNING); | |
5612 | sys_sched_yield(); | |
5613 | } | |
1da177e4 LT |
5614 | EXPORT_SYMBOL(yield); |
5615 | ||
5616 | /* | |
41a2d6cf | 5617 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5618 | * that process accounting knows that this is a task in IO wait state. |
5619 | * | |
5620 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5621 | * has set its backing_dev_info: the queue against which it should throttle) | |
5622 | */ | |
5623 | void __sched io_schedule(void) | |
5624 | { | |
70b97a7f | 5625 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5626 | |
0ff92245 | 5627 | delayacct_blkio_start(); |
1da177e4 LT |
5628 | atomic_inc(&rq->nr_iowait); |
5629 | schedule(); | |
5630 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5631 | delayacct_blkio_end(); |
1da177e4 | 5632 | } |
1da177e4 LT |
5633 | EXPORT_SYMBOL(io_schedule); |
5634 | ||
5635 | long __sched io_schedule_timeout(long timeout) | |
5636 | { | |
70b97a7f | 5637 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5638 | long ret; |
5639 | ||
0ff92245 | 5640 | delayacct_blkio_start(); |
1da177e4 LT |
5641 | atomic_inc(&rq->nr_iowait); |
5642 | ret = schedule_timeout(timeout); | |
5643 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5644 | delayacct_blkio_end(); |
1da177e4 LT |
5645 | return ret; |
5646 | } | |
5647 | ||
5648 | /** | |
5649 | * sys_sched_get_priority_max - return maximum RT priority. | |
5650 | * @policy: scheduling class. | |
5651 | * | |
5652 | * this syscall returns the maximum rt_priority that can be used | |
5653 | * by a given scheduling class. | |
5654 | */ | |
5655 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5656 | { | |
5657 | int ret = -EINVAL; | |
5658 | ||
5659 | switch (policy) { | |
5660 | case SCHED_FIFO: | |
5661 | case SCHED_RR: | |
5662 | ret = MAX_USER_RT_PRIO-1; | |
5663 | break; | |
5664 | case SCHED_NORMAL: | |
b0a9499c | 5665 | case SCHED_BATCH: |
dd41f596 | 5666 | case SCHED_IDLE: |
1da177e4 LT |
5667 | ret = 0; |
5668 | break; | |
5669 | } | |
5670 | return ret; | |
5671 | } | |
5672 | ||
5673 | /** | |
5674 | * sys_sched_get_priority_min - return minimum RT priority. | |
5675 | * @policy: scheduling class. | |
5676 | * | |
5677 | * this syscall returns the minimum rt_priority that can be used | |
5678 | * by a given scheduling class. | |
5679 | */ | |
5680 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5681 | { | |
5682 | int ret = -EINVAL; | |
5683 | ||
5684 | switch (policy) { | |
5685 | case SCHED_FIFO: | |
5686 | case SCHED_RR: | |
5687 | ret = 1; | |
5688 | break; | |
5689 | case SCHED_NORMAL: | |
b0a9499c | 5690 | case SCHED_BATCH: |
dd41f596 | 5691 | case SCHED_IDLE: |
1da177e4 LT |
5692 | ret = 0; |
5693 | } | |
5694 | return ret; | |
5695 | } | |
5696 | ||
5697 | /** | |
5698 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5699 | * @pid: pid of the process. | |
5700 | * @interval: userspace pointer to the timeslice value. | |
5701 | * | |
5702 | * this syscall writes the default timeslice value of a given process | |
5703 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5704 | */ | |
5705 | asmlinkage | |
5706 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5707 | { | |
36c8b586 | 5708 | struct task_struct *p; |
a4ec24b4 | 5709 | unsigned int time_slice; |
3a5c359a | 5710 | int retval; |
1da177e4 | 5711 | struct timespec t; |
1da177e4 LT |
5712 | |
5713 | if (pid < 0) | |
3a5c359a | 5714 | return -EINVAL; |
1da177e4 LT |
5715 | |
5716 | retval = -ESRCH; | |
5717 | read_lock(&tasklist_lock); | |
5718 | p = find_process_by_pid(pid); | |
5719 | if (!p) | |
5720 | goto out_unlock; | |
5721 | ||
5722 | retval = security_task_getscheduler(p); | |
5723 | if (retval) | |
5724 | goto out_unlock; | |
5725 | ||
77034937 IM |
5726 | /* |
5727 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5728 | * tasks that are on an otherwise idle runqueue: | |
5729 | */ | |
5730 | time_slice = 0; | |
5731 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5732 | time_slice = DEF_TIMESLICE; |
1868f958 | 5733 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5734 | struct sched_entity *se = &p->se; |
5735 | unsigned long flags; | |
5736 | struct rq *rq; | |
5737 | ||
5738 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5739 | if (rq->cfs.load.weight) |
5740 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5741 | task_rq_unlock(rq, &flags); |
5742 | } | |
1da177e4 | 5743 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5744 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5745 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5746 | return retval; |
3a5c359a | 5747 | |
1da177e4 LT |
5748 | out_unlock: |
5749 | read_unlock(&tasklist_lock); | |
5750 | return retval; | |
5751 | } | |
5752 | ||
7c731e0a | 5753 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5754 | |
82a1fcb9 | 5755 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5756 | { |
1da177e4 | 5757 | unsigned long free = 0; |
36c8b586 | 5758 | unsigned state; |
1da177e4 | 5759 | |
1da177e4 | 5760 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5761 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5762 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5763 | #if BITS_PER_LONG == 32 |
1da177e4 | 5764 | if (state == TASK_RUNNING) |
cc4ea795 | 5765 | printk(KERN_CONT " running "); |
1da177e4 | 5766 | else |
cc4ea795 | 5767 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5768 | #else |
5769 | if (state == TASK_RUNNING) | |
cc4ea795 | 5770 | printk(KERN_CONT " running task "); |
1da177e4 | 5771 | else |
cc4ea795 | 5772 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5773 | #endif |
5774 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5775 | { | |
10ebffde | 5776 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5777 | while (!*n) |
5778 | n++; | |
10ebffde | 5779 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5780 | } |
5781 | #endif | |
ba25f9dc | 5782 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5783 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5784 | |
5fb5e6de | 5785 | show_stack(p, NULL); |
1da177e4 LT |
5786 | } |
5787 | ||
e59e2ae2 | 5788 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5789 | { |
36c8b586 | 5790 | struct task_struct *g, *p; |
1da177e4 | 5791 | |
4bd77321 IM |
5792 | #if BITS_PER_LONG == 32 |
5793 | printk(KERN_INFO | |
5794 | " task PC stack pid father\n"); | |
1da177e4 | 5795 | #else |
4bd77321 IM |
5796 | printk(KERN_INFO |
5797 | " task PC stack pid father\n"); | |
1da177e4 LT |
5798 | #endif |
5799 | read_lock(&tasklist_lock); | |
5800 | do_each_thread(g, p) { | |
5801 | /* | |
5802 | * reset the NMI-timeout, listing all files on a slow | |
5803 | * console might take alot of time: | |
5804 | */ | |
5805 | touch_nmi_watchdog(); | |
39bc89fd | 5806 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5807 | sched_show_task(p); |
1da177e4 LT |
5808 | } while_each_thread(g, p); |
5809 | ||
04c9167f JF |
5810 | touch_all_softlockup_watchdogs(); |
5811 | ||
dd41f596 IM |
5812 | #ifdef CONFIG_SCHED_DEBUG |
5813 | sysrq_sched_debug_show(); | |
5814 | #endif | |
1da177e4 | 5815 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5816 | /* |
5817 | * Only show locks if all tasks are dumped: | |
5818 | */ | |
5819 | if (state_filter == -1) | |
5820 | debug_show_all_locks(); | |
1da177e4 LT |
5821 | } |
5822 | ||
1df21055 IM |
5823 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5824 | { | |
dd41f596 | 5825 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5826 | } |
5827 | ||
f340c0d1 IM |
5828 | /** |
5829 | * init_idle - set up an idle thread for a given CPU | |
5830 | * @idle: task in question | |
5831 | * @cpu: cpu the idle task belongs to | |
5832 | * | |
5833 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5834 | * flag, to make booting more robust. | |
5835 | */ | |
5c1e1767 | 5836 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5837 | { |
70b97a7f | 5838 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5839 | unsigned long flags; |
5840 | ||
dd41f596 IM |
5841 | __sched_fork(idle); |
5842 | idle->se.exec_start = sched_clock(); | |
5843 | ||
b29739f9 | 5844 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5845 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5846 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5847 | |
5848 | spin_lock_irqsave(&rq->lock, flags); | |
5849 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5850 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5851 | idle->oncpu = 1; | |
5852 | #endif | |
1da177e4 LT |
5853 | spin_unlock_irqrestore(&rq->lock, flags); |
5854 | ||
5855 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5856 | #if defined(CONFIG_PREEMPT) |
5857 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5858 | #else | |
a1261f54 | 5859 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5860 | #endif |
dd41f596 IM |
5861 | /* |
5862 | * The idle tasks have their own, simple scheduling class: | |
5863 | */ | |
5864 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5865 | } |
5866 | ||
5867 | /* | |
5868 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5869 | * indicates which cpus entered this state. This is used | |
5870 | * in the rcu update to wait only for active cpus. For system | |
5871 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5872 | * always be CPU_MASK_NONE. | |
5873 | */ | |
5874 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5875 | ||
19978ca6 IM |
5876 | /* |
5877 | * Increase the granularity value when there are more CPUs, | |
5878 | * because with more CPUs the 'effective latency' as visible | |
5879 | * to users decreases. But the relationship is not linear, | |
5880 | * so pick a second-best guess by going with the log2 of the | |
5881 | * number of CPUs. | |
5882 | * | |
5883 | * This idea comes from the SD scheduler of Con Kolivas: | |
5884 | */ | |
5885 | static inline void sched_init_granularity(void) | |
5886 | { | |
5887 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5888 | const unsigned long limit = 200000000; | |
5889 | ||
5890 | sysctl_sched_min_granularity *= factor; | |
5891 | if (sysctl_sched_min_granularity > limit) | |
5892 | sysctl_sched_min_granularity = limit; | |
5893 | ||
5894 | sysctl_sched_latency *= factor; | |
5895 | if (sysctl_sched_latency > limit) | |
5896 | sysctl_sched_latency = limit; | |
5897 | ||
5898 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5899 | |
5900 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5901 | } |
5902 | ||
1da177e4 LT |
5903 | #ifdef CONFIG_SMP |
5904 | /* | |
5905 | * This is how migration works: | |
5906 | * | |
70b97a7f | 5907 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5908 | * runqueue and wake up that CPU's migration thread. |
5909 | * 2) we down() the locked semaphore => thread blocks. | |
5910 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5911 | * thread off the CPU) | |
5912 | * 4) it gets the migration request and checks whether the migrated | |
5913 | * task is still in the wrong runqueue. | |
5914 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5915 | * it and puts it into the right queue. | |
5916 | * 6) migration thread up()s the semaphore. | |
5917 | * 7) we wake up and the migration is done. | |
5918 | */ | |
5919 | ||
5920 | /* | |
5921 | * Change a given task's CPU affinity. Migrate the thread to a | |
5922 | * proper CPU and schedule it away if the CPU it's executing on | |
5923 | * is removed from the allowed bitmask. | |
5924 | * | |
5925 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5926 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5927 | * call is not atomic; no spinlocks may be held. |
5928 | */ | |
cd8ba7cd | 5929 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5930 | { |
70b97a7f | 5931 | struct migration_req req; |
1da177e4 | 5932 | unsigned long flags; |
70b97a7f | 5933 | struct rq *rq; |
48f24c4d | 5934 | int ret = 0; |
1da177e4 LT |
5935 | |
5936 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5937 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5938 | ret = -EINVAL; |
5939 | goto out; | |
5940 | } | |
5941 | ||
9985b0ba DR |
5942 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5943 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5944 | ret = -EINVAL; | |
5945 | goto out; | |
5946 | } | |
5947 | ||
73fe6aae | 5948 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5949 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5950 | else { |
cd8ba7cd MT |
5951 | p->cpus_allowed = *new_mask; |
5952 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5953 | } |
5954 | ||
1da177e4 | 5955 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5956 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5957 | goto out; |
5958 | ||
cd8ba7cd | 5959 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5960 | /* Need help from migration thread: drop lock and wait. */ |
5961 | task_rq_unlock(rq, &flags); | |
5962 | wake_up_process(rq->migration_thread); | |
5963 | wait_for_completion(&req.done); | |
5964 | tlb_migrate_finish(p->mm); | |
5965 | return 0; | |
5966 | } | |
5967 | out: | |
5968 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5969 | |
1da177e4 LT |
5970 | return ret; |
5971 | } | |
cd8ba7cd | 5972 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5973 | |
5974 | /* | |
41a2d6cf | 5975 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5976 | * this because either it can't run here any more (set_cpus_allowed() |
5977 | * away from this CPU, or CPU going down), or because we're | |
5978 | * attempting to rebalance this task on exec (sched_exec). | |
5979 | * | |
5980 | * So we race with normal scheduler movements, but that's OK, as long | |
5981 | * as the task is no longer on this CPU. | |
efc30814 KK |
5982 | * |
5983 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5984 | */ |
efc30814 | 5985 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5986 | { |
70b97a7f | 5987 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5988 | int ret = 0, on_rq; |
1da177e4 | 5989 | |
e761b772 | 5990 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5991 | return ret; |
1da177e4 LT |
5992 | |
5993 | rq_src = cpu_rq(src_cpu); | |
5994 | rq_dest = cpu_rq(dest_cpu); | |
5995 | ||
5996 | double_rq_lock(rq_src, rq_dest); | |
5997 | /* Already moved. */ | |
5998 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5999 | goto done; |
1da177e4 LT |
6000 | /* Affinity changed (again). */ |
6001 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 6002 | goto fail; |
1da177e4 | 6003 | |
dd41f596 | 6004 | on_rq = p->se.on_rq; |
6e82a3be | 6005 | if (on_rq) |
2e1cb74a | 6006 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6007 | |
1da177e4 | 6008 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6009 | if (on_rq) { |
6010 | activate_task(rq_dest, p, 0); | |
15afe09b | 6011 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6012 | } |
b1e38734 | 6013 | done: |
efc30814 | 6014 | ret = 1; |
b1e38734 | 6015 | fail: |
1da177e4 | 6016 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6017 | return ret; |
1da177e4 LT |
6018 | } |
6019 | ||
6020 | /* | |
6021 | * migration_thread - this is a highprio system thread that performs | |
6022 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6023 | * another runqueue. | |
6024 | */ | |
95cdf3b7 | 6025 | static int migration_thread(void *data) |
1da177e4 | 6026 | { |
1da177e4 | 6027 | int cpu = (long)data; |
70b97a7f | 6028 | struct rq *rq; |
1da177e4 LT |
6029 | |
6030 | rq = cpu_rq(cpu); | |
6031 | BUG_ON(rq->migration_thread != current); | |
6032 | ||
6033 | set_current_state(TASK_INTERRUPTIBLE); | |
6034 | while (!kthread_should_stop()) { | |
70b97a7f | 6035 | struct migration_req *req; |
1da177e4 | 6036 | struct list_head *head; |
1da177e4 | 6037 | |
1da177e4 LT |
6038 | spin_lock_irq(&rq->lock); |
6039 | ||
6040 | if (cpu_is_offline(cpu)) { | |
6041 | spin_unlock_irq(&rq->lock); | |
6042 | goto wait_to_die; | |
6043 | } | |
6044 | ||
6045 | if (rq->active_balance) { | |
6046 | active_load_balance(rq, cpu); | |
6047 | rq->active_balance = 0; | |
6048 | } | |
6049 | ||
6050 | head = &rq->migration_queue; | |
6051 | ||
6052 | if (list_empty(head)) { | |
6053 | spin_unlock_irq(&rq->lock); | |
6054 | schedule(); | |
6055 | set_current_state(TASK_INTERRUPTIBLE); | |
6056 | continue; | |
6057 | } | |
70b97a7f | 6058 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6059 | list_del_init(head->next); |
6060 | ||
674311d5 NP |
6061 | spin_unlock(&rq->lock); |
6062 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6063 | local_irq_enable(); | |
1da177e4 LT |
6064 | |
6065 | complete(&req->done); | |
6066 | } | |
6067 | __set_current_state(TASK_RUNNING); | |
6068 | return 0; | |
6069 | ||
6070 | wait_to_die: | |
6071 | /* Wait for kthread_stop */ | |
6072 | set_current_state(TASK_INTERRUPTIBLE); | |
6073 | while (!kthread_should_stop()) { | |
6074 | schedule(); | |
6075 | set_current_state(TASK_INTERRUPTIBLE); | |
6076 | } | |
6077 | __set_current_state(TASK_RUNNING); | |
6078 | return 0; | |
6079 | } | |
6080 | ||
6081 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6082 | |
6083 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6084 | { | |
6085 | int ret; | |
6086 | ||
6087 | local_irq_disable(); | |
6088 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6089 | local_irq_enable(); | |
6090 | return ret; | |
6091 | } | |
6092 | ||
054b9108 | 6093 | /* |
3a4fa0a2 | 6094 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
6095 | * NOTE: interrupts should be disabled by the caller |
6096 | */ | |
48f24c4d | 6097 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6098 | { |
efc30814 | 6099 | unsigned long flags; |
1da177e4 | 6100 | cpumask_t mask; |
70b97a7f IM |
6101 | struct rq *rq; |
6102 | int dest_cpu; | |
1da177e4 | 6103 | |
3a5c359a AK |
6104 | do { |
6105 | /* On same node? */ | |
6106 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6107 | cpus_and(mask, mask, p->cpus_allowed); | |
6108 | dest_cpu = any_online_cpu(mask); | |
6109 | ||
6110 | /* On any allowed CPU? */ | |
434d53b0 | 6111 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6112 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6113 | ||
6114 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6115 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6116 | cpumask_t cpus_allowed; |
6117 | ||
6118 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6119 | /* |
6120 | * Try to stay on the same cpuset, where the | |
6121 | * current cpuset may be a subset of all cpus. | |
6122 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6123 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6124 | * called within calls to cpuset_lock/cpuset_unlock. |
6125 | */ | |
3a5c359a | 6126 | rq = task_rq_lock(p, &flags); |
470fd646 | 6127 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6128 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6129 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6130 | |
3a5c359a AK |
6131 | /* |
6132 | * Don't tell them about moving exiting tasks or | |
6133 | * kernel threads (both mm NULL), since they never | |
6134 | * leave kernel. | |
6135 | */ | |
41a2d6cf | 6136 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6137 | printk(KERN_INFO "process %d (%s) no " |
6138 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6139 | task_pid_nr(p), p->comm, dead_cpu); |
6140 | } | |
3a5c359a | 6141 | } |
f7b4cddc | 6142 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6143 | } |
6144 | ||
6145 | /* | |
6146 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6147 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6148 | * for performance reasons the counter is not stricly tracking tasks to | |
6149 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6150 | * to keep the global sum constant after CPU-down: | |
6151 | */ | |
70b97a7f | 6152 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6153 | { |
7c16ec58 | 6154 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6155 | unsigned long flags; |
6156 | ||
6157 | local_irq_save(flags); | |
6158 | double_rq_lock(rq_src, rq_dest); | |
6159 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6160 | rq_src->nr_uninterruptible = 0; | |
6161 | double_rq_unlock(rq_src, rq_dest); | |
6162 | local_irq_restore(flags); | |
6163 | } | |
6164 | ||
6165 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6166 | static void migrate_live_tasks(int src_cpu) | |
6167 | { | |
48f24c4d | 6168 | struct task_struct *p, *t; |
1da177e4 | 6169 | |
f7b4cddc | 6170 | read_lock(&tasklist_lock); |
1da177e4 | 6171 | |
48f24c4d IM |
6172 | do_each_thread(t, p) { |
6173 | if (p == current) | |
1da177e4 LT |
6174 | continue; |
6175 | ||
48f24c4d IM |
6176 | if (task_cpu(p) == src_cpu) |
6177 | move_task_off_dead_cpu(src_cpu, p); | |
6178 | } while_each_thread(t, p); | |
1da177e4 | 6179 | |
f7b4cddc | 6180 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6181 | } |
6182 | ||
dd41f596 IM |
6183 | /* |
6184 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6185 | * It does so by boosting its priority to highest possible. |
6186 | * Used by CPU offline code. | |
1da177e4 LT |
6187 | */ |
6188 | void sched_idle_next(void) | |
6189 | { | |
48f24c4d | 6190 | int this_cpu = smp_processor_id(); |
70b97a7f | 6191 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6192 | struct task_struct *p = rq->idle; |
6193 | unsigned long flags; | |
6194 | ||
6195 | /* cpu has to be offline */ | |
48f24c4d | 6196 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6197 | |
48f24c4d IM |
6198 | /* |
6199 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6200 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6201 | */ |
6202 | spin_lock_irqsave(&rq->lock, flags); | |
6203 | ||
dd41f596 | 6204 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6205 | |
94bc9a7b DA |
6206 | update_rq_clock(rq); |
6207 | activate_task(rq, p, 0); | |
1da177e4 LT |
6208 | |
6209 | spin_unlock_irqrestore(&rq->lock, flags); | |
6210 | } | |
6211 | ||
48f24c4d IM |
6212 | /* |
6213 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6214 | * offline. |
6215 | */ | |
6216 | void idle_task_exit(void) | |
6217 | { | |
6218 | struct mm_struct *mm = current->active_mm; | |
6219 | ||
6220 | BUG_ON(cpu_online(smp_processor_id())); | |
6221 | ||
6222 | if (mm != &init_mm) | |
6223 | switch_mm(mm, &init_mm, current); | |
6224 | mmdrop(mm); | |
6225 | } | |
6226 | ||
054b9108 | 6227 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6228 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6229 | { |
70b97a7f | 6230 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6231 | |
6232 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6233 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6234 | |
6235 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6236 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6237 | |
48f24c4d | 6238 | get_task_struct(p); |
1da177e4 LT |
6239 | |
6240 | /* | |
6241 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6242 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6243 | * fine. |
6244 | */ | |
f7b4cddc | 6245 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6246 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6247 | spin_lock_irq(&rq->lock); |
1da177e4 | 6248 | |
48f24c4d | 6249 | put_task_struct(p); |
1da177e4 LT |
6250 | } |
6251 | ||
6252 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6253 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6254 | { | |
70b97a7f | 6255 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6256 | struct task_struct *next; |
48f24c4d | 6257 | |
dd41f596 IM |
6258 | for ( ; ; ) { |
6259 | if (!rq->nr_running) | |
6260 | break; | |
a8e504d2 | 6261 | update_rq_clock(rq); |
ff95f3df | 6262 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6263 | if (!next) |
6264 | break; | |
79c53799 | 6265 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6266 | migrate_dead(dead_cpu, next); |
e692ab53 | 6267 | |
1da177e4 LT |
6268 | } |
6269 | } | |
6270 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6271 | ||
e692ab53 NP |
6272 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6273 | ||
6274 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6275 | { |
6276 | .procname = "sched_domain", | |
c57baf1e | 6277 | .mode = 0555, |
e0361851 | 6278 | }, |
38605cae | 6279 | {0, }, |
e692ab53 NP |
6280 | }; |
6281 | ||
6282 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6283 | { |
c57baf1e | 6284 | .ctl_name = CTL_KERN, |
e0361851 | 6285 | .procname = "kernel", |
c57baf1e | 6286 | .mode = 0555, |
e0361851 AD |
6287 | .child = sd_ctl_dir, |
6288 | }, | |
38605cae | 6289 | {0, }, |
e692ab53 NP |
6290 | }; |
6291 | ||
6292 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6293 | { | |
6294 | struct ctl_table *entry = | |
5cf9f062 | 6295 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6296 | |
e692ab53 NP |
6297 | return entry; |
6298 | } | |
6299 | ||
6382bc90 MM |
6300 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6301 | { | |
cd790076 | 6302 | struct ctl_table *entry; |
6382bc90 | 6303 | |
cd790076 MM |
6304 | /* |
6305 | * In the intermediate directories, both the child directory and | |
6306 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6307 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6308 | * static strings and all have proc handlers. |
6309 | */ | |
6310 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6311 | if (entry->child) |
6312 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6313 | if (entry->proc_handler == NULL) |
6314 | kfree(entry->procname); | |
6315 | } | |
6382bc90 MM |
6316 | |
6317 | kfree(*tablep); | |
6318 | *tablep = NULL; | |
6319 | } | |
6320 | ||
e692ab53 | 6321 | static void |
e0361851 | 6322 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6323 | const char *procname, void *data, int maxlen, |
6324 | mode_t mode, proc_handler *proc_handler) | |
6325 | { | |
e692ab53 NP |
6326 | entry->procname = procname; |
6327 | entry->data = data; | |
6328 | entry->maxlen = maxlen; | |
6329 | entry->mode = mode; | |
6330 | entry->proc_handler = proc_handler; | |
6331 | } | |
6332 | ||
6333 | static struct ctl_table * | |
6334 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6335 | { | |
a5d8c348 | 6336 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6337 | |
ad1cdc1d MM |
6338 | if (table == NULL) |
6339 | return NULL; | |
6340 | ||
e0361851 | 6341 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6342 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6343 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6344 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6345 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6346 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6347 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6348 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6349 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6350 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6351 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6352 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6353 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6354 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6355 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6356 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6357 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6358 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6359 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6360 | &sd->cache_nice_tries, |
6361 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6362 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6363 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6364 | set_table_entry(&table[11], "name", sd->name, |
6365 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6366 | /* &table[12] is terminator */ | |
e692ab53 NP |
6367 | |
6368 | return table; | |
6369 | } | |
6370 | ||
9a4e7159 | 6371 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6372 | { |
6373 | struct ctl_table *entry, *table; | |
6374 | struct sched_domain *sd; | |
6375 | int domain_num = 0, i; | |
6376 | char buf[32]; | |
6377 | ||
6378 | for_each_domain(cpu, sd) | |
6379 | domain_num++; | |
6380 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6381 | if (table == NULL) |
6382 | return NULL; | |
e692ab53 NP |
6383 | |
6384 | i = 0; | |
6385 | for_each_domain(cpu, sd) { | |
6386 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6387 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6388 | entry->mode = 0555; |
e692ab53 NP |
6389 | entry->child = sd_alloc_ctl_domain_table(sd); |
6390 | entry++; | |
6391 | i++; | |
6392 | } | |
6393 | return table; | |
6394 | } | |
6395 | ||
6396 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6397 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6398 | { |
6399 | int i, cpu_num = num_online_cpus(); | |
6400 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6401 | char buf[32]; | |
6402 | ||
7378547f MM |
6403 | WARN_ON(sd_ctl_dir[0].child); |
6404 | sd_ctl_dir[0].child = entry; | |
6405 | ||
ad1cdc1d MM |
6406 | if (entry == NULL) |
6407 | return; | |
6408 | ||
97b6ea7b | 6409 | for_each_online_cpu(i) { |
e692ab53 | 6410 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6411 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6412 | entry->mode = 0555; |
e692ab53 | 6413 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6414 | entry++; |
e692ab53 | 6415 | } |
7378547f MM |
6416 | |
6417 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6418 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6419 | } | |
6382bc90 | 6420 | |
7378547f | 6421 | /* may be called multiple times per register */ |
6382bc90 MM |
6422 | static void unregister_sched_domain_sysctl(void) |
6423 | { | |
7378547f MM |
6424 | if (sd_sysctl_header) |
6425 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6426 | sd_sysctl_header = NULL; |
7378547f MM |
6427 | if (sd_ctl_dir[0].child) |
6428 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6429 | } |
e692ab53 | 6430 | #else |
6382bc90 MM |
6431 | static void register_sched_domain_sysctl(void) |
6432 | { | |
6433 | } | |
6434 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6435 | { |
6436 | } | |
6437 | #endif | |
6438 | ||
1f11eb6a GH |
6439 | static void set_rq_online(struct rq *rq) |
6440 | { | |
6441 | if (!rq->online) { | |
6442 | const struct sched_class *class; | |
6443 | ||
6444 | cpu_set(rq->cpu, rq->rd->online); | |
6445 | rq->online = 1; | |
6446 | ||
6447 | for_each_class(class) { | |
6448 | if (class->rq_online) | |
6449 | class->rq_online(rq); | |
6450 | } | |
6451 | } | |
6452 | } | |
6453 | ||
6454 | static void set_rq_offline(struct rq *rq) | |
6455 | { | |
6456 | if (rq->online) { | |
6457 | const struct sched_class *class; | |
6458 | ||
6459 | for_each_class(class) { | |
6460 | if (class->rq_offline) | |
6461 | class->rq_offline(rq); | |
6462 | } | |
6463 | ||
6464 | cpu_clear(rq->cpu, rq->rd->online); | |
6465 | rq->online = 0; | |
6466 | } | |
6467 | } | |
6468 | ||
1da177e4 LT |
6469 | /* |
6470 | * migration_call - callback that gets triggered when a CPU is added. | |
6471 | * Here we can start up the necessary migration thread for the new CPU. | |
6472 | */ | |
48f24c4d IM |
6473 | static int __cpuinit |
6474 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6475 | { |
1da177e4 | 6476 | struct task_struct *p; |
48f24c4d | 6477 | int cpu = (long)hcpu; |
1da177e4 | 6478 | unsigned long flags; |
70b97a7f | 6479 | struct rq *rq; |
1da177e4 LT |
6480 | |
6481 | switch (action) { | |
5be9361c | 6482 | |
1da177e4 | 6483 | case CPU_UP_PREPARE: |
8bb78442 | 6484 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6485 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6486 | if (IS_ERR(p)) |
6487 | return NOTIFY_BAD; | |
1da177e4 LT |
6488 | kthread_bind(p, cpu); |
6489 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6490 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6491 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6492 | task_rq_unlock(rq, &flags); |
6493 | cpu_rq(cpu)->migration_thread = p; | |
6494 | break; | |
48f24c4d | 6495 | |
1da177e4 | 6496 | case CPU_ONLINE: |
8bb78442 | 6497 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6498 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6499 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6500 | |
6501 | /* Update our root-domain */ | |
6502 | rq = cpu_rq(cpu); | |
6503 | spin_lock_irqsave(&rq->lock, flags); | |
6504 | if (rq->rd) { | |
6505 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6506 | |
6507 | set_rq_online(rq); | |
1f94ef59 GH |
6508 | } |
6509 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6510 | break; |
48f24c4d | 6511 | |
1da177e4 LT |
6512 | #ifdef CONFIG_HOTPLUG_CPU |
6513 | case CPU_UP_CANCELED: | |
8bb78442 | 6514 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6515 | if (!cpu_rq(cpu)->migration_thread) |
6516 | break; | |
41a2d6cf | 6517 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6518 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6519 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6520 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6521 | cpu_rq(cpu)->migration_thread = NULL; | |
6522 | break; | |
48f24c4d | 6523 | |
1da177e4 | 6524 | case CPU_DEAD: |
8bb78442 | 6525 | case CPU_DEAD_FROZEN: |
470fd646 | 6526 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6527 | migrate_live_tasks(cpu); |
6528 | rq = cpu_rq(cpu); | |
6529 | kthread_stop(rq->migration_thread); | |
6530 | rq->migration_thread = NULL; | |
6531 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6532 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6533 | update_rq_clock(rq); |
2e1cb74a | 6534 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6535 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6536 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6537 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6538 | migrate_dead_tasks(cpu); |
d2da272a | 6539 | spin_unlock_irq(&rq->lock); |
470fd646 | 6540 | cpuset_unlock(); |
1da177e4 LT |
6541 | migrate_nr_uninterruptible(rq); |
6542 | BUG_ON(rq->nr_running != 0); | |
6543 | ||
41a2d6cf IM |
6544 | /* |
6545 | * No need to migrate the tasks: it was best-effort if | |
6546 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6547 | * the requestors. | |
6548 | */ | |
1da177e4 LT |
6549 | spin_lock_irq(&rq->lock); |
6550 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6551 | struct migration_req *req; |
6552 | ||
1da177e4 | 6553 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6554 | struct migration_req, list); |
1da177e4 LT |
6555 | list_del_init(&req->list); |
6556 | complete(&req->done); | |
6557 | } | |
6558 | spin_unlock_irq(&rq->lock); | |
6559 | break; | |
57d885fe | 6560 | |
08f503b0 GH |
6561 | case CPU_DYING: |
6562 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6563 | /* Update our root-domain */ |
6564 | rq = cpu_rq(cpu); | |
6565 | spin_lock_irqsave(&rq->lock, flags); | |
6566 | if (rq->rd) { | |
6567 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6568 | set_rq_offline(rq); |
57d885fe GH |
6569 | } |
6570 | spin_unlock_irqrestore(&rq->lock, flags); | |
6571 | break; | |
1da177e4 LT |
6572 | #endif |
6573 | } | |
6574 | return NOTIFY_OK; | |
6575 | } | |
6576 | ||
6577 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6578 | * happens before everything else. | |
6579 | */ | |
26c2143b | 6580 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6581 | .notifier_call = migration_call, |
6582 | .priority = 10 | |
6583 | }; | |
6584 | ||
7babe8db | 6585 | static int __init migration_init(void) |
1da177e4 LT |
6586 | { |
6587 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6588 | int err; |
48f24c4d IM |
6589 | |
6590 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6591 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6592 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6593 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6594 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6595 | |
6596 | return err; | |
1da177e4 | 6597 | } |
7babe8db | 6598 | early_initcall(migration_init); |
1da177e4 LT |
6599 | #endif |
6600 | ||
6601 | #ifdef CONFIG_SMP | |
476f3534 | 6602 | |
3e9830dc | 6603 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6604 | |
099f98c8 GS |
6605 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6606 | { | |
6607 | switch (lvl) { | |
6608 | case SD_LV_NONE: | |
6609 | return "NONE"; | |
6610 | case SD_LV_SIBLING: | |
6611 | return "SIBLING"; | |
6612 | case SD_LV_MC: | |
6613 | return "MC"; | |
6614 | case SD_LV_CPU: | |
6615 | return "CPU"; | |
6616 | case SD_LV_NODE: | |
6617 | return "NODE"; | |
6618 | case SD_LV_ALLNODES: | |
6619 | return "ALLNODES"; | |
6620 | case SD_LV_MAX: | |
6621 | return "MAX"; | |
6622 | ||
6623 | } | |
6624 | return "MAX"; | |
6625 | } | |
6626 | ||
7c16ec58 MT |
6627 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6628 | cpumask_t *groupmask) | |
1da177e4 | 6629 | { |
4dcf6aff | 6630 | struct sched_group *group = sd->groups; |
434d53b0 | 6631 | char str[256]; |
1da177e4 | 6632 | |
434d53b0 | 6633 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6634 | cpus_clear(*groupmask); |
4dcf6aff IM |
6635 | |
6636 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6637 | ||
6638 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6639 | printk("does not load-balance\n"); | |
6640 | if (sd->parent) | |
6641 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6642 | " has parent"); | |
6643 | return -1; | |
41c7ce9a NP |
6644 | } |
6645 | ||
099f98c8 GS |
6646 | printk(KERN_CONT "span %s level %s\n", |
6647 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6648 | |
6649 | if (!cpu_isset(cpu, sd->span)) { | |
6650 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6651 | "CPU%d\n", cpu); | |
6652 | } | |
6653 | if (!cpu_isset(cpu, group->cpumask)) { | |
6654 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6655 | " CPU%d\n", cpu); | |
6656 | } | |
1da177e4 | 6657 | |
4dcf6aff | 6658 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6659 | do { |
4dcf6aff IM |
6660 | if (!group) { |
6661 | printk("\n"); | |
6662 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6663 | break; |
6664 | } | |
6665 | ||
4dcf6aff IM |
6666 | if (!group->__cpu_power) { |
6667 | printk(KERN_CONT "\n"); | |
6668 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6669 | "set\n"); | |
6670 | break; | |
6671 | } | |
1da177e4 | 6672 | |
4dcf6aff IM |
6673 | if (!cpus_weight(group->cpumask)) { |
6674 | printk(KERN_CONT "\n"); | |
6675 | printk(KERN_ERR "ERROR: empty group\n"); | |
6676 | break; | |
6677 | } | |
1da177e4 | 6678 | |
7c16ec58 | 6679 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6680 | printk(KERN_CONT "\n"); |
6681 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6682 | break; | |
6683 | } | |
1da177e4 | 6684 | |
7c16ec58 | 6685 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6686 | |
434d53b0 | 6687 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6688 | printk(KERN_CONT " %s", str); |
1da177e4 | 6689 | |
4dcf6aff IM |
6690 | group = group->next; |
6691 | } while (group != sd->groups); | |
6692 | printk(KERN_CONT "\n"); | |
1da177e4 | 6693 | |
7c16ec58 | 6694 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6695 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6696 | |
7c16ec58 | 6697 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6698 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6699 | "of domain->span\n"); | |
6700 | return 0; | |
6701 | } | |
1da177e4 | 6702 | |
4dcf6aff IM |
6703 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6704 | { | |
7c16ec58 | 6705 | cpumask_t *groupmask; |
4dcf6aff | 6706 | int level = 0; |
1da177e4 | 6707 | |
4dcf6aff IM |
6708 | if (!sd) { |
6709 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6710 | return; | |
6711 | } | |
1da177e4 | 6712 | |
4dcf6aff IM |
6713 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6714 | ||
7c16ec58 MT |
6715 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6716 | if (!groupmask) { | |
6717 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6718 | return; | |
6719 | } | |
6720 | ||
4dcf6aff | 6721 | for (;;) { |
7c16ec58 | 6722 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6723 | break; |
1da177e4 LT |
6724 | level++; |
6725 | sd = sd->parent; | |
33859f7f | 6726 | if (!sd) |
4dcf6aff IM |
6727 | break; |
6728 | } | |
7c16ec58 | 6729 | kfree(groupmask); |
1da177e4 | 6730 | } |
6d6bc0ad | 6731 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6732 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6733 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6734 | |
1a20ff27 | 6735 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6736 | { |
6737 | if (cpus_weight(sd->span) == 1) | |
6738 | return 1; | |
6739 | ||
6740 | /* Following flags need at least 2 groups */ | |
6741 | if (sd->flags & (SD_LOAD_BALANCE | | |
6742 | SD_BALANCE_NEWIDLE | | |
6743 | SD_BALANCE_FORK | | |
89c4710e SS |
6744 | SD_BALANCE_EXEC | |
6745 | SD_SHARE_CPUPOWER | | |
6746 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6747 | if (sd->groups != sd->groups->next) |
6748 | return 0; | |
6749 | } | |
6750 | ||
6751 | /* Following flags don't use groups */ | |
6752 | if (sd->flags & (SD_WAKE_IDLE | | |
6753 | SD_WAKE_AFFINE | | |
6754 | SD_WAKE_BALANCE)) | |
6755 | return 0; | |
6756 | ||
6757 | return 1; | |
6758 | } | |
6759 | ||
48f24c4d IM |
6760 | static int |
6761 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6762 | { |
6763 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6764 | ||
6765 | if (sd_degenerate(parent)) | |
6766 | return 1; | |
6767 | ||
6768 | if (!cpus_equal(sd->span, parent->span)) | |
6769 | return 0; | |
6770 | ||
6771 | /* Does parent contain flags not in child? */ | |
6772 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6773 | if (cflags & SD_WAKE_AFFINE) | |
6774 | pflags &= ~SD_WAKE_BALANCE; | |
6775 | /* Flags needing groups don't count if only 1 group in parent */ | |
6776 | if (parent->groups == parent->groups->next) { | |
6777 | pflags &= ~(SD_LOAD_BALANCE | | |
6778 | SD_BALANCE_NEWIDLE | | |
6779 | SD_BALANCE_FORK | | |
89c4710e SS |
6780 | SD_BALANCE_EXEC | |
6781 | SD_SHARE_CPUPOWER | | |
6782 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6783 | } |
6784 | if (~cflags & pflags) | |
6785 | return 0; | |
6786 | ||
6787 | return 1; | |
6788 | } | |
6789 | ||
57d885fe GH |
6790 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6791 | { | |
6792 | unsigned long flags; | |
57d885fe GH |
6793 | |
6794 | spin_lock_irqsave(&rq->lock, flags); | |
6795 | ||
6796 | if (rq->rd) { | |
6797 | struct root_domain *old_rd = rq->rd; | |
6798 | ||
1f11eb6a GH |
6799 | if (cpu_isset(rq->cpu, old_rd->online)) |
6800 | set_rq_offline(rq); | |
57d885fe | 6801 | |
dc938520 | 6802 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6803 | |
57d885fe GH |
6804 | if (atomic_dec_and_test(&old_rd->refcount)) |
6805 | kfree(old_rd); | |
6806 | } | |
6807 | ||
6808 | atomic_inc(&rd->refcount); | |
6809 | rq->rd = rd; | |
6810 | ||
dc938520 | 6811 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6812 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6813 | set_rq_online(rq); |
57d885fe GH |
6814 | |
6815 | spin_unlock_irqrestore(&rq->lock, flags); | |
6816 | } | |
6817 | ||
dc938520 | 6818 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6819 | { |
6820 | memset(rd, 0, sizeof(*rd)); | |
6821 | ||
dc938520 GH |
6822 | cpus_clear(rd->span); |
6823 | cpus_clear(rd->online); | |
6e0534f2 GH |
6824 | |
6825 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6826 | } |
6827 | ||
6828 | static void init_defrootdomain(void) | |
6829 | { | |
dc938520 | 6830 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6831 | atomic_set(&def_root_domain.refcount, 1); |
6832 | } | |
6833 | ||
dc938520 | 6834 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6835 | { |
6836 | struct root_domain *rd; | |
6837 | ||
6838 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6839 | if (!rd) | |
6840 | return NULL; | |
6841 | ||
dc938520 | 6842 | init_rootdomain(rd); |
57d885fe GH |
6843 | |
6844 | return rd; | |
6845 | } | |
6846 | ||
1da177e4 | 6847 | /* |
0eab9146 | 6848 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6849 | * hold the hotplug lock. |
6850 | */ | |
0eab9146 IM |
6851 | static void |
6852 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6853 | { |
70b97a7f | 6854 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6855 | struct sched_domain *tmp; |
6856 | ||
6857 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6858 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6859 | struct sched_domain *parent = tmp->parent; | |
6860 | if (!parent) | |
6861 | break; | |
1a848870 | 6862 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6863 | tmp->parent = parent->parent; |
1a848870 SS |
6864 | if (parent->parent) |
6865 | parent->parent->child = tmp; | |
6866 | } | |
245af2c7 SS |
6867 | } |
6868 | ||
1a848870 | 6869 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6870 | sd = sd->parent; |
1a848870 SS |
6871 | if (sd) |
6872 | sd->child = NULL; | |
6873 | } | |
1da177e4 LT |
6874 | |
6875 | sched_domain_debug(sd, cpu); | |
6876 | ||
57d885fe | 6877 | rq_attach_root(rq, rd); |
674311d5 | 6878 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6879 | } |
6880 | ||
6881 | /* cpus with isolated domains */ | |
67af63a6 | 6882 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6883 | |
6884 | /* Setup the mask of cpus configured for isolated domains */ | |
6885 | static int __init isolated_cpu_setup(char *str) | |
6886 | { | |
13b40c1e MT |
6887 | static int __initdata ints[NR_CPUS]; |
6888 | int i; | |
1da177e4 LT |
6889 | |
6890 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6891 | cpus_clear(cpu_isolated_map); | |
6892 | for (i = 1; i <= ints[0]; i++) | |
6893 | if (ints[i] < NR_CPUS) | |
6894 | cpu_set(ints[i], cpu_isolated_map); | |
6895 | return 1; | |
6896 | } | |
6897 | ||
8927f494 | 6898 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6899 | |
6900 | /* | |
6711cab4 SS |
6901 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6902 | * to a function which identifies what group(along with sched group) a CPU | |
6903 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6904 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6905 | * |
6906 | * init_sched_build_groups will build a circular linked list of the groups | |
6907 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6908 | * and ->cpu_power to 0. | |
6909 | */ | |
a616058b | 6910 | static void |
7c16ec58 | 6911 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6912 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6913 | struct sched_group **sg, |
6914 | cpumask_t *tmpmask), | |
6915 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6916 | { |
6917 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6918 | int i; |
6919 | ||
7c16ec58 MT |
6920 | cpus_clear(*covered); |
6921 | ||
363ab6f1 | 6922 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6923 | struct sched_group *sg; |
7c16ec58 | 6924 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6925 | int j; |
6926 | ||
7c16ec58 | 6927 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6928 | continue; |
6929 | ||
7c16ec58 | 6930 | cpus_clear(sg->cpumask); |
5517d86b | 6931 | sg->__cpu_power = 0; |
1da177e4 | 6932 | |
363ab6f1 | 6933 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6934 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6935 | continue; |
6936 | ||
7c16ec58 | 6937 | cpu_set(j, *covered); |
1da177e4 LT |
6938 | cpu_set(j, sg->cpumask); |
6939 | } | |
6940 | if (!first) | |
6941 | first = sg; | |
6942 | if (last) | |
6943 | last->next = sg; | |
6944 | last = sg; | |
6945 | } | |
6946 | last->next = first; | |
6947 | } | |
6948 | ||
9c1cfda2 | 6949 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6950 | |
9c1cfda2 | 6951 | #ifdef CONFIG_NUMA |
198e2f18 | 6952 | |
9c1cfda2 JH |
6953 | /** |
6954 | * find_next_best_node - find the next node to include in a sched_domain | |
6955 | * @node: node whose sched_domain we're building | |
6956 | * @used_nodes: nodes already in the sched_domain | |
6957 | * | |
41a2d6cf | 6958 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6959 | * finds the closest node not already in the @used_nodes map. |
6960 | * | |
6961 | * Should use nodemask_t. | |
6962 | */ | |
c5f59f08 | 6963 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6964 | { |
6965 | int i, n, val, min_val, best_node = 0; | |
6966 | ||
6967 | min_val = INT_MAX; | |
6968 | ||
076ac2af | 6969 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6970 | /* Start at @node */ |
076ac2af | 6971 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6972 | |
6973 | if (!nr_cpus_node(n)) | |
6974 | continue; | |
6975 | ||
6976 | /* Skip already used nodes */ | |
c5f59f08 | 6977 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6978 | continue; |
6979 | ||
6980 | /* Simple min distance search */ | |
6981 | val = node_distance(node, n); | |
6982 | ||
6983 | if (val < min_val) { | |
6984 | min_val = val; | |
6985 | best_node = n; | |
6986 | } | |
6987 | } | |
6988 | ||
c5f59f08 | 6989 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6990 | return best_node; |
6991 | } | |
6992 | ||
6993 | /** | |
6994 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6995 | * @node: node whose cpumask we're constructing | |
73486722 | 6996 | * @span: resulting cpumask |
9c1cfda2 | 6997 | * |
41a2d6cf | 6998 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6999 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7000 | * out optimally. | |
7001 | */ | |
4bdbaad3 | 7002 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 7003 | { |
c5f59f08 | 7004 | nodemask_t used_nodes; |
c5f59f08 | 7005 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7006 | int i; |
9c1cfda2 | 7007 | |
4bdbaad3 | 7008 | cpus_clear(*span); |
c5f59f08 | 7009 | nodes_clear(used_nodes); |
9c1cfda2 | 7010 | |
4bdbaad3 | 7011 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7012 | node_set(node, used_nodes); |
9c1cfda2 JH |
7013 | |
7014 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7015 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7016 | |
c5f59f08 | 7017 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7018 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7019 | } |
9c1cfda2 | 7020 | } |
6d6bc0ad | 7021 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7022 | |
5c45bf27 | 7023 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7024 | |
9c1cfda2 | 7025 | /* |
48f24c4d | 7026 | * SMT sched-domains: |
9c1cfda2 | 7027 | */ |
1da177e4 LT |
7028 | #ifdef CONFIG_SCHED_SMT |
7029 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 7030 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 7031 | |
41a2d6cf | 7032 | static int |
7c16ec58 MT |
7033 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7034 | cpumask_t *unused) | |
1da177e4 | 7035 | { |
6711cab4 SS |
7036 | if (sg) |
7037 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
7038 | return cpu; |
7039 | } | |
6d6bc0ad | 7040 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7041 | |
48f24c4d IM |
7042 | /* |
7043 | * multi-core sched-domains: | |
7044 | */ | |
1e9f28fa SS |
7045 | #ifdef CONFIG_SCHED_MC |
7046 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 7047 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 7048 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7049 | |
7050 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7051 | static int |
7c16ec58 MT |
7052 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7053 | cpumask_t *mask) | |
1e9f28fa | 7054 | { |
6711cab4 | 7055 | int group; |
7c16ec58 MT |
7056 | |
7057 | *mask = per_cpu(cpu_sibling_map, cpu); | |
7058 | cpus_and(*mask, *mask, *cpu_map); | |
7059 | group = first_cpu(*mask); | |
6711cab4 SS |
7060 | if (sg) |
7061 | *sg = &per_cpu(sched_group_core, group); | |
7062 | return group; | |
1e9f28fa SS |
7063 | } |
7064 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7065 | static int |
7c16ec58 MT |
7066 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7067 | cpumask_t *unused) | |
1e9f28fa | 7068 | { |
6711cab4 SS |
7069 | if (sg) |
7070 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
7071 | return cpu; |
7072 | } | |
7073 | #endif | |
7074 | ||
1da177e4 | 7075 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 7076 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 7077 | |
41a2d6cf | 7078 | static int |
7c16ec58 MT |
7079 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
7080 | cpumask_t *mask) | |
1da177e4 | 7081 | { |
6711cab4 | 7082 | int group; |
48f24c4d | 7083 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
7084 | *mask = cpu_coregroup_map(cpu); |
7085 | cpus_and(*mask, *mask, *cpu_map); | |
7086 | group = first_cpu(*mask); | |
1e9f28fa | 7087 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
7088 | *mask = per_cpu(cpu_sibling_map, cpu); |
7089 | cpus_and(*mask, *mask, *cpu_map); | |
7090 | group = first_cpu(*mask); | |
1da177e4 | 7091 | #else |
6711cab4 | 7092 | group = cpu; |
1da177e4 | 7093 | #endif |
6711cab4 SS |
7094 | if (sg) |
7095 | *sg = &per_cpu(sched_group_phys, group); | |
7096 | return group; | |
1da177e4 LT |
7097 | } |
7098 | ||
7099 | #ifdef CONFIG_NUMA | |
1da177e4 | 7100 | /* |
9c1cfda2 JH |
7101 | * The init_sched_build_groups can't handle what we want to do with node |
7102 | * groups, so roll our own. Now each node has its own list of groups which | |
7103 | * gets dynamically allocated. | |
1da177e4 | 7104 | */ |
9c1cfda2 | 7105 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7106 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7107 | |
9c1cfda2 | 7108 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7109 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7110 | |
6711cab4 | 7111 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7112 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7113 | { |
6711cab4 SS |
7114 | int group; |
7115 | ||
7c16ec58 MT |
7116 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7117 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7118 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7119 | |
7120 | if (sg) | |
7121 | *sg = &per_cpu(sched_group_allnodes, group); | |
7122 | return group; | |
1da177e4 | 7123 | } |
6711cab4 | 7124 | |
08069033 SS |
7125 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7126 | { | |
7127 | struct sched_group *sg = group_head; | |
7128 | int j; | |
7129 | ||
7130 | if (!sg) | |
7131 | return; | |
3a5c359a | 7132 | do { |
363ab6f1 | 7133 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7134 | struct sched_domain *sd; |
08069033 | 7135 | |
3a5c359a AK |
7136 | sd = &per_cpu(phys_domains, j); |
7137 | if (j != first_cpu(sd->groups->cpumask)) { | |
7138 | /* | |
7139 | * Only add "power" once for each | |
7140 | * physical package. | |
7141 | */ | |
7142 | continue; | |
7143 | } | |
08069033 | 7144 | |
3a5c359a AK |
7145 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7146 | } | |
7147 | sg = sg->next; | |
7148 | } while (sg != group_head); | |
08069033 | 7149 | } |
6d6bc0ad | 7150 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7151 | |
a616058b | 7152 | #ifdef CONFIG_NUMA |
51888ca2 | 7153 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7154 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7155 | { |
a616058b | 7156 | int cpu, i; |
51888ca2 | 7157 | |
363ab6f1 | 7158 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7159 | struct sched_group **sched_group_nodes |
7160 | = sched_group_nodes_bycpu[cpu]; | |
7161 | ||
51888ca2 SV |
7162 | if (!sched_group_nodes) |
7163 | continue; | |
7164 | ||
076ac2af | 7165 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7166 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7167 | ||
7c16ec58 MT |
7168 | *nodemask = node_to_cpumask(i); |
7169 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7170 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7171 | continue; |
7172 | ||
7173 | if (sg == NULL) | |
7174 | continue; | |
7175 | sg = sg->next; | |
7176 | next_sg: | |
7177 | oldsg = sg; | |
7178 | sg = sg->next; | |
7179 | kfree(oldsg); | |
7180 | if (oldsg != sched_group_nodes[i]) | |
7181 | goto next_sg; | |
7182 | } | |
7183 | kfree(sched_group_nodes); | |
7184 | sched_group_nodes_bycpu[cpu] = NULL; | |
7185 | } | |
51888ca2 | 7186 | } |
6d6bc0ad | 7187 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7188 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7189 | { |
7190 | } | |
6d6bc0ad | 7191 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7192 | |
89c4710e SS |
7193 | /* |
7194 | * Initialize sched groups cpu_power. | |
7195 | * | |
7196 | * cpu_power indicates the capacity of sched group, which is used while | |
7197 | * distributing the load between different sched groups in a sched domain. | |
7198 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7199 | * there are asymmetries in the topology. If there are asymmetries, group | |
7200 | * having more cpu_power will pickup more load compared to the group having | |
7201 | * less cpu_power. | |
7202 | * | |
7203 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7204 | * the maximum number of tasks a group can handle in the presence of other idle | |
7205 | * or lightly loaded groups in the same sched domain. | |
7206 | */ | |
7207 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7208 | { | |
7209 | struct sched_domain *child; | |
7210 | struct sched_group *group; | |
7211 | ||
7212 | WARN_ON(!sd || !sd->groups); | |
7213 | ||
7214 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7215 | return; | |
7216 | ||
7217 | child = sd->child; | |
7218 | ||
5517d86b ED |
7219 | sd->groups->__cpu_power = 0; |
7220 | ||
89c4710e SS |
7221 | /* |
7222 | * For perf policy, if the groups in child domain share resources | |
7223 | * (for example cores sharing some portions of the cache hierarchy | |
7224 | * or SMT), then set this domain groups cpu_power such that each group | |
7225 | * can handle only one task, when there are other idle groups in the | |
7226 | * same sched domain. | |
7227 | */ | |
7228 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7229 | (child->flags & | |
7230 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7231 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7232 | return; |
7233 | } | |
7234 | ||
89c4710e SS |
7235 | /* |
7236 | * add cpu_power of each child group to this groups cpu_power | |
7237 | */ | |
7238 | group = child->groups; | |
7239 | do { | |
5517d86b | 7240 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7241 | group = group->next; |
7242 | } while (group != child->groups); | |
7243 | } | |
7244 | ||
7c16ec58 MT |
7245 | /* |
7246 | * Initializers for schedule domains | |
7247 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7248 | */ | |
7249 | ||
a5d8c348 IM |
7250 | #ifdef CONFIG_SCHED_DEBUG |
7251 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7252 | #else | |
7253 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7254 | #endif | |
7255 | ||
7c16ec58 | 7256 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7257 | |
7c16ec58 MT |
7258 | #define SD_INIT_FUNC(type) \ |
7259 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7260 | { \ | |
7261 | memset(sd, 0, sizeof(*sd)); \ | |
7262 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7263 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7264 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7265 | } |
7266 | ||
7267 | SD_INIT_FUNC(CPU) | |
7268 | #ifdef CONFIG_NUMA | |
7269 | SD_INIT_FUNC(ALLNODES) | |
7270 | SD_INIT_FUNC(NODE) | |
7271 | #endif | |
7272 | #ifdef CONFIG_SCHED_SMT | |
7273 | SD_INIT_FUNC(SIBLING) | |
7274 | #endif | |
7275 | #ifdef CONFIG_SCHED_MC | |
7276 | SD_INIT_FUNC(MC) | |
7277 | #endif | |
7278 | ||
7279 | /* | |
7280 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7281 | * space as the usage in build_sched_domains() dictates. Used only | |
7282 | * if the amount of space is significant. | |
7283 | */ | |
7284 | struct allmasks { | |
7285 | cpumask_t tmpmask; /* make this one first */ | |
7286 | union { | |
7287 | cpumask_t nodemask; | |
7288 | cpumask_t this_sibling_map; | |
7289 | cpumask_t this_core_map; | |
7290 | }; | |
7291 | cpumask_t send_covered; | |
7292 | ||
7293 | #ifdef CONFIG_NUMA | |
7294 | cpumask_t domainspan; | |
7295 | cpumask_t covered; | |
7296 | cpumask_t notcovered; | |
7297 | #endif | |
7298 | }; | |
7299 | ||
7300 | #if NR_CPUS > 128 | |
7301 | #define SCHED_CPUMASK_ALLOC 1 | |
7302 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7303 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7304 | #else | |
7305 | #define SCHED_CPUMASK_ALLOC 0 | |
7306 | #define SCHED_CPUMASK_FREE(v) | |
7307 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7308 | #endif | |
7309 | ||
7310 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7311 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7312 | ||
1d3504fc HS |
7313 | static int default_relax_domain_level = -1; |
7314 | ||
7315 | static int __init setup_relax_domain_level(char *str) | |
7316 | { | |
30e0e178 LZ |
7317 | unsigned long val; |
7318 | ||
7319 | val = simple_strtoul(str, NULL, 0); | |
7320 | if (val < SD_LV_MAX) | |
7321 | default_relax_domain_level = val; | |
7322 | ||
1d3504fc HS |
7323 | return 1; |
7324 | } | |
7325 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7326 | ||
7327 | static void set_domain_attribute(struct sched_domain *sd, | |
7328 | struct sched_domain_attr *attr) | |
7329 | { | |
7330 | int request; | |
7331 | ||
7332 | if (!attr || attr->relax_domain_level < 0) { | |
7333 | if (default_relax_domain_level < 0) | |
7334 | return; | |
7335 | else | |
7336 | request = default_relax_domain_level; | |
7337 | } else | |
7338 | request = attr->relax_domain_level; | |
7339 | if (request < sd->level) { | |
7340 | /* turn off idle balance on this domain */ | |
7341 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7342 | } else { | |
7343 | /* turn on idle balance on this domain */ | |
7344 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7345 | } | |
7346 | } | |
7347 | ||
1da177e4 | 7348 | /* |
1a20ff27 DG |
7349 | * Build sched domains for a given set of cpus and attach the sched domains |
7350 | * to the individual cpus | |
1da177e4 | 7351 | */ |
1d3504fc HS |
7352 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7353 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7354 | { |
7355 | int i; | |
57d885fe | 7356 | struct root_domain *rd; |
7c16ec58 MT |
7357 | SCHED_CPUMASK_DECLARE(allmasks); |
7358 | cpumask_t *tmpmask; | |
d1b55138 JH |
7359 | #ifdef CONFIG_NUMA |
7360 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7361 | int sd_allnodes = 0; |
d1b55138 JH |
7362 | |
7363 | /* | |
7364 | * Allocate the per-node list of sched groups | |
7365 | */ | |
076ac2af | 7366 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7367 | GFP_KERNEL); |
d1b55138 JH |
7368 | if (!sched_group_nodes) { |
7369 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7370 | return -ENOMEM; |
d1b55138 | 7371 | } |
d1b55138 | 7372 | #endif |
1da177e4 | 7373 | |
dc938520 | 7374 | rd = alloc_rootdomain(); |
57d885fe GH |
7375 | if (!rd) { |
7376 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7377 | #ifdef CONFIG_NUMA |
7378 | kfree(sched_group_nodes); | |
7379 | #endif | |
57d885fe GH |
7380 | return -ENOMEM; |
7381 | } | |
7382 | ||
7c16ec58 MT |
7383 | #if SCHED_CPUMASK_ALLOC |
7384 | /* get space for all scratch cpumask variables */ | |
7385 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7386 | if (!allmasks) { | |
7387 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7388 | kfree(rd); | |
7389 | #ifdef CONFIG_NUMA | |
7390 | kfree(sched_group_nodes); | |
7391 | #endif | |
7392 | return -ENOMEM; | |
7393 | } | |
7394 | #endif | |
7395 | tmpmask = (cpumask_t *)allmasks; | |
7396 | ||
7397 | ||
7398 | #ifdef CONFIG_NUMA | |
7399 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7400 | #endif | |
7401 | ||
1da177e4 | 7402 | /* |
1a20ff27 | 7403 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7404 | */ |
363ab6f1 | 7405 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7406 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7407 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7408 | |
7c16ec58 MT |
7409 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7410 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7411 | |
7412 | #ifdef CONFIG_NUMA | |
dd41f596 | 7413 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7414 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7415 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7416 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7417 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7418 | sd->span = *cpu_map; |
7c16ec58 | 7419 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7420 | p = sd; |
6711cab4 | 7421 | sd_allnodes = 1; |
9c1cfda2 JH |
7422 | } else |
7423 | p = NULL; | |
7424 | ||
1da177e4 | 7425 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7426 | SD_INIT(sd, NODE); |
1d3504fc | 7427 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7428 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7429 | sd->parent = p; |
1a848870 SS |
7430 | if (p) |
7431 | p->child = sd; | |
9c1cfda2 | 7432 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7433 | #endif |
7434 | ||
7435 | p = sd; | |
7436 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7437 | SD_INIT(sd, CPU); |
1d3504fc | 7438 | set_domain_attribute(sd, attr); |
7c16ec58 | 7439 | sd->span = *nodemask; |
1da177e4 | 7440 | sd->parent = p; |
1a848870 SS |
7441 | if (p) |
7442 | p->child = sd; | |
7c16ec58 | 7443 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7444 | |
1e9f28fa SS |
7445 | #ifdef CONFIG_SCHED_MC |
7446 | p = sd; | |
7447 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7448 | SD_INIT(sd, MC); |
1d3504fc | 7449 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7450 | sd->span = cpu_coregroup_map(i); |
7451 | cpus_and(sd->span, sd->span, *cpu_map); | |
7452 | sd->parent = p; | |
1a848870 | 7453 | p->child = sd; |
7c16ec58 | 7454 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7455 | #endif |
7456 | ||
1da177e4 LT |
7457 | #ifdef CONFIG_SCHED_SMT |
7458 | p = sd; | |
7459 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7460 | SD_INIT(sd, SIBLING); |
1d3504fc | 7461 | set_domain_attribute(sd, attr); |
d5a7430d | 7462 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7463 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7464 | sd->parent = p; |
1a848870 | 7465 | p->child = sd; |
7c16ec58 | 7466 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7467 | #endif |
7468 | } | |
7469 | ||
7470 | #ifdef CONFIG_SCHED_SMT | |
7471 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7472 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7473 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7474 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7475 | ||
7476 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7477 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7478 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7479 | continue; |
7480 | ||
dd41f596 | 7481 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7482 | &cpu_to_cpu_group, |
7483 | send_covered, tmpmask); | |
1da177e4 LT |
7484 | } |
7485 | #endif | |
7486 | ||
1e9f28fa SS |
7487 | #ifdef CONFIG_SCHED_MC |
7488 | /* Set up multi-core groups */ | |
363ab6f1 | 7489 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7490 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7491 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7492 | ||
7493 | *this_core_map = cpu_coregroup_map(i); | |
7494 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7495 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7496 | continue; |
7c16ec58 | 7497 | |
dd41f596 | 7498 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7499 | &cpu_to_core_group, |
7500 | send_covered, tmpmask); | |
1e9f28fa SS |
7501 | } |
7502 | #endif | |
7503 | ||
1da177e4 | 7504 | /* Set up physical groups */ |
076ac2af | 7505 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7506 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7507 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7508 | |
7c16ec58 MT |
7509 | *nodemask = node_to_cpumask(i); |
7510 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7511 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7512 | continue; |
7513 | ||
7c16ec58 MT |
7514 | init_sched_build_groups(nodemask, cpu_map, |
7515 | &cpu_to_phys_group, | |
7516 | send_covered, tmpmask); | |
1da177e4 LT |
7517 | } |
7518 | ||
7519 | #ifdef CONFIG_NUMA | |
7520 | /* Set up node groups */ | |
7c16ec58 MT |
7521 | if (sd_allnodes) { |
7522 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7523 | ||
7524 | init_sched_build_groups(cpu_map, cpu_map, | |
7525 | &cpu_to_allnodes_group, | |
7526 | send_covered, tmpmask); | |
7527 | } | |
9c1cfda2 | 7528 | |
076ac2af | 7529 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7530 | /* Set up node groups */ |
7531 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7532 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7533 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7534 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7535 | int j; |
7536 | ||
7c16ec58 MT |
7537 | *nodemask = node_to_cpumask(i); |
7538 | cpus_clear(*covered); | |
7539 | ||
7540 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7541 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7542 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7543 | continue; |
d1b55138 | 7544 | } |
9c1cfda2 | 7545 | |
4bdbaad3 | 7546 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7547 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7548 | |
15f0b676 | 7549 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7550 | if (!sg) { |
7551 | printk(KERN_WARNING "Can not alloc domain group for " | |
7552 | "node %d\n", i); | |
7553 | goto error; | |
7554 | } | |
9c1cfda2 | 7555 | sched_group_nodes[i] = sg; |
363ab6f1 | 7556 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7557 | struct sched_domain *sd; |
9761eea8 | 7558 | |
9c1cfda2 JH |
7559 | sd = &per_cpu(node_domains, j); |
7560 | sd->groups = sg; | |
9c1cfda2 | 7561 | } |
5517d86b | 7562 | sg->__cpu_power = 0; |
7c16ec58 | 7563 | sg->cpumask = *nodemask; |
51888ca2 | 7564 | sg->next = sg; |
7c16ec58 | 7565 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7566 | prev = sg; |
7567 | ||
076ac2af | 7568 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7569 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7570 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7571 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7572 | |
7c16ec58 MT |
7573 | cpus_complement(*notcovered, *covered); |
7574 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7575 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7576 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7577 | break; |
7578 | ||
7c16ec58 MT |
7579 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7580 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7581 | continue; |
7582 | ||
15f0b676 SV |
7583 | sg = kmalloc_node(sizeof(struct sched_group), |
7584 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7585 | if (!sg) { |
7586 | printk(KERN_WARNING | |
7587 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7588 | goto error; |
9c1cfda2 | 7589 | } |
5517d86b | 7590 | sg->__cpu_power = 0; |
7c16ec58 | 7591 | sg->cpumask = *tmpmask; |
51888ca2 | 7592 | sg->next = prev->next; |
7c16ec58 | 7593 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7594 | prev->next = sg; |
7595 | prev = sg; | |
7596 | } | |
9c1cfda2 | 7597 | } |
1da177e4 LT |
7598 | #endif |
7599 | ||
7600 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7601 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7602 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7603 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7604 | ||
89c4710e | 7605 | init_sched_groups_power(i, sd); |
5c45bf27 | 7606 | } |
1da177e4 | 7607 | #endif |
1e9f28fa | 7608 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7609 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7610 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7611 | ||
89c4710e | 7612 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7613 | } |
7614 | #endif | |
1e9f28fa | 7615 | |
363ab6f1 | 7616 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7617 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7618 | ||
89c4710e | 7619 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7620 | } |
7621 | ||
9c1cfda2 | 7622 | #ifdef CONFIG_NUMA |
076ac2af | 7623 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7624 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7625 | |
6711cab4 SS |
7626 | if (sd_allnodes) { |
7627 | struct sched_group *sg; | |
f712c0c7 | 7628 | |
7c16ec58 MT |
7629 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7630 | tmpmask); | |
f712c0c7 SS |
7631 | init_numa_sched_groups_power(sg); |
7632 | } | |
9c1cfda2 JH |
7633 | #endif |
7634 | ||
1da177e4 | 7635 | /* Attach the domains */ |
363ab6f1 | 7636 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7637 | struct sched_domain *sd; |
7638 | #ifdef CONFIG_SCHED_SMT | |
7639 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7640 | #elif defined(CONFIG_SCHED_MC) |
7641 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7642 | #else |
7643 | sd = &per_cpu(phys_domains, i); | |
7644 | #endif | |
57d885fe | 7645 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7646 | } |
51888ca2 | 7647 | |
7c16ec58 | 7648 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7649 | return 0; |
7650 | ||
a616058b | 7651 | #ifdef CONFIG_NUMA |
51888ca2 | 7652 | error: |
7c16ec58 MT |
7653 | free_sched_groups(cpu_map, tmpmask); |
7654 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7655 | return -ENOMEM; |
a616058b | 7656 | #endif |
1da177e4 | 7657 | } |
029190c5 | 7658 | |
1d3504fc HS |
7659 | static int build_sched_domains(const cpumask_t *cpu_map) |
7660 | { | |
7661 | return __build_sched_domains(cpu_map, NULL); | |
7662 | } | |
7663 | ||
029190c5 PJ |
7664 | static cpumask_t *doms_cur; /* current sched domains */ |
7665 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7666 | static struct sched_domain_attr *dattr_cur; |
7667 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7668 | |
7669 | /* | |
7670 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7671 | * cpumask_t) fails, then fallback to a single sched domain, | |
7672 | * as determined by the single cpumask_t fallback_doms. | |
7673 | */ | |
7674 | static cpumask_t fallback_doms; | |
7675 | ||
22e52b07 HC |
7676 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7677 | { | |
7678 | } | |
7679 | ||
1a20ff27 | 7680 | /* |
41a2d6cf | 7681 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7682 | * For now this just excludes isolated cpus, but could be used to |
7683 | * exclude other special cases in the future. | |
1a20ff27 | 7684 | */ |
51888ca2 | 7685 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7686 | { |
7378547f MM |
7687 | int err; |
7688 | ||
22e52b07 | 7689 | arch_update_cpu_topology(); |
029190c5 PJ |
7690 | ndoms_cur = 1; |
7691 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7692 | if (!doms_cur) | |
7693 | doms_cur = &fallback_doms; | |
7694 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7695 | dattr_cur = NULL; |
7378547f | 7696 | err = build_sched_domains(doms_cur); |
6382bc90 | 7697 | register_sched_domain_sysctl(); |
7378547f MM |
7698 | |
7699 | return err; | |
1a20ff27 DG |
7700 | } |
7701 | ||
7c16ec58 MT |
7702 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7703 | cpumask_t *tmpmask) | |
1da177e4 | 7704 | { |
7c16ec58 | 7705 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7706 | } |
1da177e4 | 7707 | |
1a20ff27 DG |
7708 | /* |
7709 | * Detach sched domains from a group of cpus specified in cpu_map | |
7710 | * These cpus will now be attached to the NULL domain | |
7711 | */ | |
858119e1 | 7712 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7713 | { |
7c16ec58 | 7714 | cpumask_t tmpmask; |
1a20ff27 DG |
7715 | int i; |
7716 | ||
6382bc90 MM |
7717 | unregister_sched_domain_sysctl(); |
7718 | ||
363ab6f1 | 7719 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7720 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7721 | synchronize_sched(); |
7c16ec58 | 7722 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7723 | } |
7724 | ||
1d3504fc HS |
7725 | /* handle null as "default" */ |
7726 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7727 | struct sched_domain_attr *new, int idx_new) | |
7728 | { | |
7729 | struct sched_domain_attr tmp; | |
7730 | ||
7731 | /* fast path */ | |
7732 | if (!new && !cur) | |
7733 | return 1; | |
7734 | ||
7735 | tmp = SD_ATTR_INIT; | |
7736 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7737 | new ? (new + idx_new) : &tmp, | |
7738 | sizeof(struct sched_domain_attr)); | |
7739 | } | |
7740 | ||
029190c5 PJ |
7741 | /* |
7742 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7743 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7744 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7745 | * It destroys each deleted domain and builds each new domain. | |
7746 | * | |
7747 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7748 | * The masks don't intersect (don't overlap.) We should setup one |
7749 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7750 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7751 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7752 | * it as it is. | |
7753 | * | |
41a2d6cf IM |
7754 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7755 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7756 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7757 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7758 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 | 7759 | * |
dfb512ec MK |
7760 | * If doms_new==NULL it will be replaced with cpu_online_map. |
7761 | * ndoms_new==0 is a special case for destroying existing domains. | |
7762 | * It will not create the default domain. | |
7763 | * | |
029190c5 PJ |
7764 | * Call with hotplug lock held |
7765 | */ | |
1d3504fc HS |
7766 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7767 | struct sched_domain_attr *dattr_new) | |
029190c5 | 7768 | { |
dfb512ec | 7769 | int i, j, n; |
029190c5 | 7770 | |
712555ee | 7771 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7772 | |
7378547f MM |
7773 | /* always unregister in case we don't destroy any domains */ |
7774 | unregister_sched_domain_sysctl(); | |
7775 | ||
dfb512ec | 7776 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7777 | |
7778 | /* Destroy deleted domains */ | |
7779 | for (i = 0; i < ndoms_cur; i++) { | |
dfb512ec | 7780 | for (j = 0; j < n; j++) { |
1d3504fc HS |
7781 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7782 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7783 | goto match1; |
7784 | } | |
7785 | /* no match - a current sched domain not in new doms_new[] */ | |
7786 | detach_destroy_domains(doms_cur + i); | |
7787 | match1: | |
7788 | ; | |
7789 | } | |
7790 | ||
e761b772 MK |
7791 | if (doms_new == NULL) { |
7792 | ndoms_cur = 0; | |
e761b772 MK |
7793 | doms_new = &fallback_doms; |
7794 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7795 | dattr_new = NULL; | |
7796 | } | |
7797 | ||
029190c5 PJ |
7798 | /* Build new domains */ |
7799 | for (i = 0; i < ndoms_new; i++) { | |
7800 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7801 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7802 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7803 | goto match2; |
7804 | } | |
7805 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7806 | __build_sched_domains(doms_new + i, |
7807 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7808 | match2: |
7809 | ; | |
7810 | } | |
7811 | ||
7812 | /* Remember the new sched domains */ | |
7813 | if (doms_cur != &fallback_doms) | |
7814 | kfree(doms_cur); | |
1d3504fc | 7815 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7816 | doms_cur = doms_new; |
1d3504fc | 7817 | dattr_cur = dattr_new; |
029190c5 | 7818 | ndoms_cur = ndoms_new; |
7378547f MM |
7819 | |
7820 | register_sched_domain_sysctl(); | |
a1835615 | 7821 | |
712555ee | 7822 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7823 | } |
7824 | ||
5c45bf27 | 7825 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7826 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7827 | { |
95402b38 | 7828 | get_online_cpus(); |
dfb512ec MK |
7829 | |
7830 | /* Destroy domains first to force the rebuild */ | |
7831 | partition_sched_domains(0, NULL, NULL); | |
7832 | ||
e761b772 | 7833 | rebuild_sched_domains(); |
95402b38 | 7834 | put_online_cpus(); |
dfb512ec | 7835 | |
e761b772 | 7836 | return 0; |
5c45bf27 SS |
7837 | } |
7838 | ||
7839 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7840 | { | |
7841 | int ret; | |
7842 | ||
7843 | if (buf[0] != '0' && buf[0] != '1') | |
7844 | return -EINVAL; | |
7845 | ||
7846 | if (smt) | |
7847 | sched_smt_power_savings = (buf[0] == '1'); | |
7848 | else | |
7849 | sched_mc_power_savings = (buf[0] == '1'); | |
7850 | ||
7851 | ret = arch_reinit_sched_domains(); | |
7852 | ||
7853 | return ret ? ret : count; | |
7854 | } | |
7855 | ||
5c45bf27 | 7856 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7857 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7858 | char *page) | |
5c45bf27 SS |
7859 | { |
7860 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7861 | } | |
f718cd4a | 7862 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7863 | const char *buf, size_t count) |
5c45bf27 SS |
7864 | { |
7865 | return sched_power_savings_store(buf, count, 0); | |
7866 | } | |
f718cd4a AK |
7867 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7868 | sched_mc_power_savings_show, | |
7869 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7870 | #endif |
7871 | ||
7872 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7873 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7874 | char *page) | |
5c45bf27 SS |
7875 | { |
7876 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7877 | } | |
f718cd4a | 7878 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7879 | const char *buf, size_t count) |
5c45bf27 SS |
7880 | { |
7881 | return sched_power_savings_store(buf, count, 1); | |
7882 | } | |
f718cd4a AK |
7883 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7884 | sched_smt_power_savings_show, | |
6707de00 AB |
7885 | sched_smt_power_savings_store); |
7886 | #endif | |
7887 | ||
7888 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7889 | { | |
7890 | int err = 0; | |
7891 | ||
7892 | #ifdef CONFIG_SCHED_SMT | |
7893 | if (smt_capable()) | |
7894 | err = sysfs_create_file(&cls->kset.kobj, | |
7895 | &attr_sched_smt_power_savings.attr); | |
7896 | #endif | |
7897 | #ifdef CONFIG_SCHED_MC | |
7898 | if (!err && mc_capable()) | |
7899 | err = sysfs_create_file(&cls->kset.kobj, | |
7900 | &attr_sched_mc_power_savings.attr); | |
7901 | #endif | |
7902 | return err; | |
7903 | } | |
6d6bc0ad | 7904 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7905 | |
e761b772 | 7906 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7907 | /* |
e761b772 MK |
7908 | * Add online and remove offline CPUs from the scheduler domains. |
7909 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7910 | */ |
7911 | static int update_sched_domains(struct notifier_block *nfb, | |
7912 | unsigned long action, void *hcpu) | |
e761b772 MK |
7913 | { |
7914 | switch (action) { | |
7915 | case CPU_ONLINE: | |
7916 | case CPU_ONLINE_FROZEN: | |
7917 | case CPU_DEAD: | |
7918 | case CPU_DEAD_FROZEN: | |
dfb512ec | 7919 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7920 | return NOTIFY_OK; |
7921 | ||
7922 | default: | |
7923 | return NOTIFY_DONE; | |
7924 | } | |
7925 | } | |
7926 | #endif | |
7927 | ||
7928 | static int update_runtime(struct notifier_block *nfb, | |
7929 | unsigned long action, void *hcpu) | |
1da177e4 | 7930 | { |
7def2be1 PZ |
7931 | int cpu = (int)(long)hcpu; |
7932 | ||
1da177e4 | 7933 | switch (action) { |
1da177e4 | 7934 | case CPU_DOWN_PREPARE: |
8bb78442 | 7935 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7936 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7937 | return NOTIFY_OK; |
7938 | ||
1da177e4 | 7939 | case CPU_DOWN_FAILED: |
8bb78442 | 7940 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7941 | case CPU_ONLINE: |
8bb78442 | 7942 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7943 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7944 | return NOTIFY_OK; |
7945 | ||
1da177e4 LT |
7946 | default: |
7947 | return NOTIFY_DONE; | |
7948 | } | |
1da177e4 | 7949 | } |
1da177e4 LT |
7950 | |
7951 | void __init sched_init_smp(void) | |
7952 | { | |
5c1e1767 NP |
7953 | cpumask_t non_isolated_cpus; |
7954 | ||
434d53b0 MT |
7955 | #if defined(CONFIG_NUMA) |
7956 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7957 | GFP_KERNEL); | |
7958 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7959 | #endif | |
95402b38 | 7960 | get_online_cpus(); |
712555ee | 7961 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7962 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7963 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7964 | if (cpus_empty(non_isolated_cpus)) |
7965 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7966 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7967 | put_online_cpus(); |
e761b772 MK |
7968 | |
7969 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7970 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7971 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7972 | #endif |
7973 | ||
7974 | /* RT runtime code needs to handle some hotplug events */ | |
7975 | hotcpu_notifier(update_runtime, 0); | |
7976 | ||
b328ca18 | 7977 | init_hrtick(); |
5c1e1767 NP |
7978 | |
7979 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7980 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7981 | BUG(); |
19978ca6 | 7982 | sched_init_granularity(); |
1da177e4 LT |
7983 | } |
7984 | #else | |
7985 | void __init sched_init_smp(void) | |
7986 | { | |
19978ca6 | 7987 | sched_init_granularity(); |
1da177e4 LT |
7988 | } |
7989 | #endif /* CONFIG_SMP */ | |
7990 | ||
7991 | int in_sched_functions(unsigned long addr) | |
7992 | { | |
1da177e4 LT |
7993 | return in_lock_functions(addr) || |
7994 | (addr >= (unsigned long)__sched_text_start | |
7995 | && addr < (unsigned long)__sched_text_end); | |
7996 | } | |
7997 | ||
a9957449 | 7998 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7999 | { |
8000 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8001 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8002 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8003 | cfs_rq->rq = rq; | |
8004 | #endif | |
67e9fb2a | 8005 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8006 | } |
8007 | ||
fa85ae24 PZ |
8008 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8009 | { | |
8010 | struct rt_prio_array *array; | |
8011 | int i; | |
8012 | ||
8013 | array = &rt_rq->active; | |
8014 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8015 | INIT_LIST_HEAD(array->queue + i); | |
8016 | __clear_bit(i, array->bitmap); | |
8017 | } | |
8018 | /* delimiter for bitsearch: */ | |
8019 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8020 | ||
052f1dc7 | 8021 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8022 | rt_rq->highest_prio = MAX_RT_PRIO; |
8023 | #endif | |
fa85ae24 PZ |
8024 | #ifdef CONFIG_SMP |
8025 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8026 | rt_rq->overloaded = 0; |
8027 | #endif | |
8028 | ||
8029 | rt_rq->rt_time = 0; | |
8030 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8031 | rt_rq->rt_runtime = 0; |
8032 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8033 | |
052f1dc7 | 8034 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8035 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8036 | rt_rq->rq = rq; |
8037 | #endif | |
fa85ae24 PZ |
8038 | } |
8039 | ||
6f505b16 | 8040 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8041 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8042 | struct sched_entity *se, int cpu, int add, | |
8043 | struct sched_entity *parent) | |
6f505b16 | 8044 | { |
ec7dc8ac | 8045 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8046 | tg->cfs_rq[cpu] = cfs_rq; |
8047 | init_cfs_rq(cfs_rq, rq); | |
8048 | cfs_rq->tg = tg; | |
8049 | if (add) | |
8050 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8051 | ||
8052 | tg->se[cpu] = se; | |
354d60c2 DG |
8053 | /* se could be NULL for init_task_group */ |
8054 | if (!se) | |
8055 | return; | |
8056 | ||
ec7dc8ac DG |
8057 | if (!parent) |
8058 | se->cfs_rq = &rq->cfs; | |
8059 | else | |
8060 | se->cfs_rq = parent->my_q; | |
8061 | ||
6f505b16 PZ |
8062 | se->my_q = cfs_rq; |
8063 | se->load.weight = tg->shares; | |
e05510d0 | 8064 | se->load.inv_weight = 0; |
ec7dc8ac | 8065 | se->parent = parent; |
6f505b16 | 8066 | } |
052f1dc7 | 8067 | #endif |
6f505b16 | 8068 | |
052f1dc7 | 8069 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8070 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8071 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8072 | struct sched_rt_entity *parent) | |
6f505b16 | 8073 | { |
ec7dc8ac DG |
8074 | struct rq *rq = cpu_rq(cpu); |
8075 | ||
6f505b16 PZ |
8076 | tg->rt_rq[cpu] = rt_rq; |
8077 | init_rt_rq(rt_rq, rq); | |
8078 | rt_rq->tg = tg; | |
8079 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8080 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8081 | if (add) |
8082 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8083 | ||
8084 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8085 | if (!rt_se) |
8086 | return; | |
8087 | ||
ec7dc8ac DG |
8088 | if (!parent) |
8089 | rt_se->rt_rq = &rq->rt; | |
8090 | else | |
8091 | rt_se->rt_rq = parent->my_q; | |
8092 | ||
6f505b16 | 8093 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8094 | rt_se->parent = parent; |
6f505b16 PZ |
8095 | INIT_LIST_HEAD(&rt_se->run_list); |
8096 | } | |
8097 | #endif | |
8098 | ||
1da177e4 LT |
8099 | void __init sched_init(void) |
8100 | { | |
dd41f596 | 8101 | int i, j; |
434d53b0 MT |
8102 | unsigned long alloc_size = 0, ptr; |
8103 | ||
8104 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8105 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8106 | #endif | |
8107 | #ifdef CONFIG_RT_GROUP_SCHED | |
8108 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8109 | #endif |
8110 | #ifdef CONFIG_USER_SCHED | |
8111 | alloc_size *= 2; | |
434d53b0 MT |
8112 | #endif |
8113 | /* | |
8114 | * As sched_init() is called before page_alloc is setup, | |
8115 | * we use alloc_bootmem(). | |
8116 | */ | |
8117 | if (alloc_size) { | |
5a9d3225 | 8118 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8119 | |
8120 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8121 | init_task_group.se = (struct sched_entity **)ptr; | |
8122 | ptr += nr_cpu_ids * sizeof(void **); | |
8123 | ||
8124 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8125 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8126 | |
8127 | #ifdef CONFIG_USER_SCHED | |
8128 | root_task_group.se = (struct sched_entity **)ptr; | |
8129 | ptr += nr_cpu_ids * sizeof(void **); | |
8130 | ||
8131 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8132 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8133 | #endif /* CONFIG_USER_SCHED */ |
8134 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8135 | #ifdef CONFIG_RT_GROUP_SCHED |
8136 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8137 | ptr += nr_cpu_ids * sizeof(void **); | |
8138 | ||
8139 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8140 | ptr += nr_cpu_ids * sizeof(void **); |
8141 | ||
8142 | #ifdef CONFIG_USER_SCHED | |
8143 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8144 | ptr += nr_cpu_ids * sizeof(void **); | |
8145 | ||
8146 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8147 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8148 | #endif /* CONFIG_USER_SCHED */ |
8149 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8150 | } |
dd41f596 | 8151 | |
57d885fe GH |
8152 | #ifdef CONFIG_SMP |
8153 | init_defrootdomain(); | |
8154 | #endif | |
8155 | ||
d0b27fa7 PZ |
8156 | init_rt_bandwidth(&def_rt_bandwidth, |
8157 | global_rt_period(), global_rt_runtime()); | |
8158 | ||
8159 | #ifdef CONFIG_RT_GROUP_SCHED | |
8160 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8161 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8162 | #ifdef CONFIG_USER_SCHED |
8163 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8164 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8165 | #endif /* CONFIG_USER_SCHED */ |
8166 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8167 | |
052f1dc7 | 8168 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8169 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8170 | INIT_LIST_HEAD(&init_task_group.children); |
8171 | ||
8172 | #ifdef CONFIG_USER_SCHED | |
8173 | INIT_LIST_HEAD(&root_task_group.children); | |
8174 | init_task_group.parent = &root_task_group; | |
8175 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8176 | #endif /* CONFIG_USER_SCHED */ |
8177 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8178 | |
0a945022 | 8179 | for_each_possible_cpu(i) { |
70b97a7f | 8180 | struct rq *rq; |
1da177e4 LT |
8181 | |
8182 | rq = cpu_rq(i); | |
8183 | spin_lock_init(&rq->lock); | |
7897986b | 8184 | rq->nr_running = 0; |
dd41f596 | 8185 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8186 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8187 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8188 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8189 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8190 | #ifdef CONFIG_CGROUP_SCHED |
8191 | /* | |
8192 | * How much cpu bandwidth does init_task_group get? | |
8193 | * | |
8194 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8195 | * gets 100% of the cpu resources in the system. This overall | |
8196 | * system cpu resource is divided among the tasks of | |
8197 | * init_task_group and its child task-groups in a fair manner, | |
8198 | * based on each entity's (task or task-group's) weight | |
8199 | * (se->load.weight). | |
8200 | * | |
8201 | * In other words, if init_task_group has 10 tasks of weight | |
8202 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8203 | * then A0's share of the cpu resource is: | |
8204 | * | |
8205 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8206 | * | |
8207 | * We achieve this by letting init_task_group's tasks sit | |
8208 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8209 | */ | |
ec7dc8ac | 8210 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8211 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8212 | root_task_group.shares = NICE_0_LOAD; |
8213 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8214 | /* |
8215 | * In case of task-groups formed thr' the user id of tasks, | |
8216 | * init_task_group represents tasks belonging to root user. | |
8217 | * Hence it forms a sibling of all subsequent groups formed. | |
8218 | * In this case, init_task_group gets only a fraction of overall | |
8219 | * system cpu resource, based on the weight assigned to root | |
8220 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8221 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8222 | * (init_cfs_rq) and having one entity represent this group of | |
8223 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8224 | */ | |
ec7dc8ac | 8225 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8226 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8227 | &per_cpu(init_sched_entity, i), i, 1, |
8228 | root_task_group.se[i]); | |
6f505b16 | 8229 | |
052f1dc7 | 8230 | #endif |
354d60c2 DG |
8231 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8232 | ||
8233 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8234 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8235 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8236 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8237 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8238 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8239 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8240 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8241 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8242 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8243 | root_task_group.rt_se[i]); | |
354d60c2 | 8244 | #endif |
dd41f596 | 8245 | #endif |
1da177e4 | 8246 | |
dd41f596 IM |
8247 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8248 | rq->cpu_load[j] = 0; | |
1da177e4 | 8249 | #ifdef CONFIG_SMP |
41c7ce9a | 8250 | rq->sd = NULL; |
57d885fe | 8251 | rq->rd = NULL; |
1da177e4 | 8252 | rq->active_balance = 0; |
dd41f596 | 8253 | rq->next_balance = jiffies; |
1da177e4 | 8254 | rq->push_cpu = 0; |
0a2966b4 | 8255 | rq->cpu = i; |
1f11eb6a | 8256 | rq->online = 0; |
1da177e4 LT |
8257 | rq->migration_thread = NULL; |
8258 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8259 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8260 | #endif |
8f4d37ec | 8261 | init_rq_hrtick(rq); |
1da177e4 | 8262 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8263 | } |
8264 | ||
2dd73a4f | 8265 | set_load_weight(&init_task); |
b50f60ce | 8266 | |
e107be36 AK |
8267 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8268 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8269 | #endif | |
8270 | ||
c9819f45 | 8271 | #ifdef CONFIG_SMP |
962cf36c | 8272 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8273 | #endif |
8274 | ||
b50f60ce HC |
8275 | #ifdef CONFIG_RT_MUTEXES |
8276 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8277 | #endif | |
8278 | ||
1da177e4 LT |
8279 | /* |
8280 | * The boot idle thread does lazy MMU switching as well: | |
8281 | */ | |
8282 | atomic_inc(&init_mm.mm_count); | |
8283 | enter_lazy_tlb(&init_mm, current); | |
8284 | ||
8285 | /* | |
8286 | * Make us the idle thread. Technically, schedule() should not be | |
8287 | * called from this thread, however somewhere below it might be, | |
8288 | * but because we are the idle thread, we just pick up running again | |
8289 | * when this runqueue becomes "idle". | |
8290 | */ | |
8291 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8292 | /* |
8293 | * During early bootup we pretend to be a normal task: | |
8294 | */ | |
8295 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8296 | |
8297 | scheduler_running = 1; | |
1da177e4 LT |
8298 | } |
8299 | ||
8300 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8301 | void __might_sleep(char *file, int line) | |
8302 | { | |
48f24c4d | 8303 | #ifdef in_atomic |
1da177e4 LT |
8304 | static unsigned long prev_jiffy; /* ratelimiting */ |
8305 | ||
aef745fc IM |
8306 | if ((!in_atomic() && !irqs_disabled()) || |
8307 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8308 | return; | |
8309 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8310 | return; | |
8311 | prev_jiffy = jiffies; | |
8312 | ||
8313 | printk(KERN_ERR | |
8314 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8315 | file, line); | |
8316 | printk(KERN_ERR | |
8317 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8318 | in_atomic(), irqs_disabled(), | |
8319 | current->pid, current->comm); | |
8320 | ||
8321 | debug_show_held_locks(current); | |
8322 | if (irqs_disabled()) | |
8323 | print_irqtrace_events(current); | |
8324 | dump_stack(); | |
1da177e4 LT |
8325 | #endif |
8326 | } | |
8327 | EXPORT_SYMBOL(__might_sleep); | |
8328 | #endif | |
8329 | ||
8330 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8331 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8332 | { | |
8333 | int on_rq; | |
3e51f33f | 8334 | |
3a5e4dc1 AK |
8335 | update_rq_clock(rq); |
8336 | on_rq = p->se.on_rq; | |
8337 | if (on_rq) | |
8338 | deactivate_task(rq, p, 0); | |
8339 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8340 | if (on_rq) { | |
8341 | activate_task(rq, p, 0); | |
8342 | resched_task(rq->curr); | |
8343 | } | |
8344 | } | |
8345 | ||
1da177e4 LT |
8346 | void normalize_rt_tasks(void) |
8347 | { | |
a0f98a1c | 8348 | struct task_struct *g, *p; |
1da177e4 | 8349 | unsigned long flags; |
70b97a7f | 8350 | struct rq *rq; |
1da177e4 | 8351 | |
4cf5d77a | 8352 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8353 | do_each_thread(g, p) { |
178be793 IM |
8354 | /* |
8355 | * Only normalize user tasks: | |
8356 | */ | |
8357 | if (!p->mm) | |
8358 | continue; | |
8359 | ||
6cfb0d5d | 8360 | p->se.exec_start = 0; |
6cfb0d5d | 8361 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8362 | p->se.wait_start = 0; |
dd41f596 | 8363 | p->se.sleep_start = 0; |
dd41f596 | 8364 | p->se.block_start = 0; |
6cfb0d5d | 8365 | #endif |
dd41f596 IM |
8366 | |
8367 | if (!rt_task(p)) { | |
8368 | /* | |
8369 | * Renice negative nice level userspace | |
8370 | * tasks back to 0: | |
8371 | */ | |
8372 | if (TASK_NICE(p) < 0 && p->mm) | |
8373 | set_user_nice(p, 0); | |
1da177e4 | 8374 | continue; |
dd41f596 | 8375 | } |
1da177e4 | 8376 | |
4cf5d77a | 8377 | spin_lock(&p->pi_lock); |
b29739f9 | 8378 | rq = __task_rq_lock(p); |
1da177e4 | 8379 | |
178be793 | 8380 | normalize_task(rq, p); |
3a5e4dc1 | 8381 | |
b29739f9 | 8382 | __task_rq_unlock(rq); |
4cf5d77a | 8383 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8384 | } while_each_thread(g, p); |
8385 | ||
4cf5d77a | 8386 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8387 | } |
8388 | ||
8389 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8390 | |
8391 | #ifdef CONFIG_IA64 | |
8392 | /* | |
8393 | * These functions are only useful for the IA64 MCA handling. | |
8394 | * | |
8395 | * They can only be called when the whole system has been | |
8396 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8397 | * activity can take place. Using them for anything else would | |
8398 | * be a serious bug, and as a result, they aren't even visible | |
8399 | * under any other configuration. | |
8400 | */ | |
8401 | ||
8402 | /** | |
8403 | * curr_task - return the current task for a given cpu. | |
8404 | * @cpu: the processor in question. | |
8405 | * | |
8406 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8407 | */ | |
36c8b586 | 8408 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8409 | { |
8410 | return cpu_curr(cpu); | |
8411 | } | |
8412 | ||
8413 | /** | |
8414 | * set_curr_task - set the current task for a given cpu. | |
8415 | * @cpu: the processor in question. | |
8416 | * @p: the task pointer to set. | |
8417 | * | |
8418 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8419 | * are serviced on a separate stack. It allows the architecture to switch the |
8420 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8421 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8422 | * and caller must save the original value of the current task (see | |
8423 | * curr_task() above) and restore that value before reenabling interrupts and | |
8424 | * re-starting the system. | |
8425 | * | |
8426 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8427 | */ | |
36c8b586 | 8428 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8429 | { |
8430 | cpu_curr(cpu) = p; | |
8431 | } | |
8432 | ||
8433 | #endif | |
29f59db3 | 8434 | |
bccbe08a PZ |
8435 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8436 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8437 | { |
8438 | int i; | |
8439 | ||
8440 | for_each_possible_cpu(i) { | |
8441 | if (tg->cfs_rq) | |
8442 | kfree(tg->cfs_rq[i]); | |
8443 | if (tg->se) | |
8444 | kfree(tg->se[i]); | |
6f505b16 PZ |
8445 | } |
8446 | ||
8447 | kfree(tg->cfs_rq); | |
8448 | kfree(tg->se); | |
6f505b16 PZ |
8449 | } |
8450 | ||
ec7dc8ac DG |
8451 | static |
8452 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8453 | { |
29f59db3 | 8454 | struct cfs_rq *cfs_rq; |
eab17229 | 8455 | struct sched_entity *se; |
9b5b7751 | 8456 | struct rq *rq; |
29f59db3 SV |
8457 | int i; |
8458 | ||
434d53b0 | 8459 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8460 | if (!tg->cfs_rq) |
8461 | goto err; | |
434d53b0 | 8462 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8463 | if (!tg->se) |
8464 | goto err; | |
052f1dc7 PZ |
8465 | |
8466 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8467 | |
8468 | for_each_possible_cpu(i) { | |
9b5b7751 | 8469 | rq = cpu_rq(i); |
29f59db3 | 8470 | |
eab17229 LZ |
8471 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8472 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8473 | if (!cfs_rq) |
8474 | goto err; | |
8475 | ||
eab17229 LZ |
8476 | se = kzalloc_node(sizeof(struct sched_entity), |
8477 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8478 | if (!se) |
8479 | goto err; | |
8480 | ||
eab17229 | 8481 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8482 | } |
8483 | ||
8484 | return 1; | |
8485 | ||
8486 | err: | |
8487 | return 0; | |
8488 | } | |
8489 | ||
8490 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8491 | { | |
8492 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8493 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8494 | } | |
8495 | ||
8496 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8497 | { | |
8498 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8499 | } | |
6d6bc0ad | 8500 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8501 | static inline void free_fair_sched_group(struct task_group *tg) |
8502 | { | |
8503 | } | |
8504 | ||
ec7dc8ac DG |
8505 | static inline |
8506 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8507 | { |
8508 | return 1; | |
8509 | } | |
8510 | ||
8511 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8512 | { | |
8513 | } | |
8514 | ||
8515 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8516 | { | |
8517 | } | |
6d6bc0ad | 8518 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8519 | |
8520 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8521 | static void free_rt_sched_group(struct task_group *tg) |
8522 | { | |
8523 | int i; | |
8524 | ||
d0b27fa7 PZ |
8525 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8526 | ||
bccbe08a PZ |
8527 | for_each_possible_cpu(i) { |
8528 | if (tg->rt_rq) | |
8529 | kfree(tg->rt_rq[i]); | |
8530 | if (tg->rt_se) | |
8531 | kfree(tg->rt_se[i]); | |
8532 | } | |
8533 | ||
8534 | kfree(tg->rt_rq); | |
8535 | kfree(tg->rt_se); | |
8536 | } | |
8537 | ||
ec7dc8ac DG |
8538 | static |
8539 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8540 | { |
8541 | struct rt_rq *rt_rq; | |
eab17229 | 8542 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8543 | struct rq *rq; |
8544 | int i; | |
8545 | ||
434d53b0 | 8546 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8547 | if (!tg->rt_rq) |
8548 | goto err; | |
434d53b0 | 8549 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8550 | if (!tg->rt_se) |
8551 | goto err; | |
8552 | ||
d0b27fa7 PZ |
8553 | init_rt_bandwidth(&tg->rt_bandwidth, |
8554 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8555 | |
8556 | for_each_possible_cpu(i) { | |
8557 | rq = cpu_rq(i); | |
8558 | ||
eab17229 LZ |
8559 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8560 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8561 | if (!rt_rq) |
8562 | goto err; | |
29f59db3 | 8563 | |
eab17229 LZ |
8564 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8565 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8566 | if (!rt_se) |
8567 | goto err; | |
29f59db3 | 8568 | |
eab17229 | 8569 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8570 | } |
8571 | ||
bccbe08a PZ |
8572 | return 1; |
8573 | ||
8574 | err: | |
8575 | return 0; | |
8576 | } | |
8577 | ||
8578 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8579 | { | |
8580 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8581 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8582 | } | |
8583 | ||
8584 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8585 | { | |
8586 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8587 | } | |
6d6bc0ad | 8588 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8589 | static inline void free_rt_sched_group(struct task_group *tg) |
8590 | { | |
8591 | } | |
8592 | ||
ec7dc8ac DG |
8593 | static inline |
8594 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8595 | { |
8596 | return 1; | |
8597 | } | |
8598 | ||
8599 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8600 | { | |
8601 | } | |
8602 | ||
8603 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8604 | { | |
8605 | } | |
6d6bc0ad | 8606 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8607 | |
d0b27fa7 | 8608 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8609 | static void free_sched_group(struct task_group *tg) |
8610 | { | |
8611 | free_fair_sched_group(tg); | |
8612 | free_rt_sched_group(tg); | |
8613 | kfree(tg); | |
8614 | } | |
8615 | ||
8616 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8617 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8618 | { |
8619 | struct task_group *tg; | |
8620 | unsigned long flags; | |
8621 | int i; | |
8622 | ||
8623 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8624 | if (!tg) | |
8625 | return ERR_PTR(-ENOMEM); | |
8626 | ||
ec7dc8ac | 8627 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8628 | goto err; |
8629 | ||
ec7dc8ac | 8630 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8631 | goto err; |
8632 | ||
8ed36996 | 8633 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8634 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8635 | register_fair_sched_group(tg, i); |
8636 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8637 | } |
6f505b16 | 8638 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8639 | |
8640 | WARN_ON(!parent); /* root should already exist */ | |
8641 | ||
8642 | tg->parent = parent; | |
f473aa5e | 8643 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8644 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8645 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8646 | |
9b5b7751 | 8647 | return tg; |
29f59db3 SV |
8648 | |
8649 | err: | |
6f505b16 | 8650 | free_sched_group(tg); |
29f59db3 SV |
8651 | return ERR_PTR(-ENOMEM); |
8652 | } | |
8653 | ||
9b5b7751 | 8654 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8655 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8656 | { |
29f59db3 | 8657 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8658 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8659 | } |
8660 | ||
9b5b7751 | 8661 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8662 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8663 | { |
8ed36996 | 8664 | unsigned long flags; |
9b5b7751 | 8665 | int i; |
29f59db3 | 8666 | |
8ed36996 | 8667 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8668 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8669 | unregister_fair_sched_group(tg, i); |
8670 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8671 | } |
6f505b16 | 8672 | list_del_rcu(&tg->list); |
f473aa5e | 8673 | list_del_rcu(&tg->siblings); |
8ed36996 | 8674 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8675 | |
9b5b7751 | 8676 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8677 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8678 | } |
8679 | ||
9b5b7751 | 8680 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8681 | * The caller of this function should have put the task in its new group |
8682 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8683 | * reflect its new group. | |
9b5b7751 SV |
8684 | */ |
8685 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8686 | { |
8687 | int on_rq, running; | |
8688 | unsigned long flags; | |
8689 | struct rq *rq; | |
8690 | ||
8691 | rq = task_rq_lock(tsk, &flags); | |
8692 | ||
29f59db3 SV |
8693 | update_rq_clock(rq); |
8694 | ||
051a1d1a | 8695 | running = task_current(rq, tsk); |
29f59db3 SV |
8696 | on_rq = tsk->se.on_rq; |
8697 | ||
0e1f3483 | 8698 | if (on_rq) |
29f59db3 | 8699 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8700 | if (unlikely(running)) |
8701 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8702 | |
6f505b16 | 8703 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8704 | |
810b3817 PZ |
8705 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8706 | if (tsk->sched_class->moved_group) | |
8707 | tsk->sched_class->moved_group(tsk); | |
8708 | #endif | |
8709 | ||
0e1f3483 HS |
8710 | if (unlikely(running)) |
8711 | tsk->sched_class->set_curr_task(rq); | |
8712 | if (on_rq) | |
7074badb | 8713 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8714 | |
29f59db3 SV |
8715 | task_rq_unlock(rq, &flags); |
8716 | } | |
6d6bc0ad | 8717 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8718 | |
052f1dc7 | 8719 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8720 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8721 | { |
8722 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8723 | int on_rq; |
8724 | ||
29f59db3 | 8725 | on_rq = se->on_rq; |
62fb1851 | 8726 | if (on_rq) |
29f59db3 SV |
8727 | dequeue_entity(cfs_rq, se, 0); |
8728 | ||
8729 | se->load.weight = shares; | |
e05510d0 | 8730 | se->load.inv_weight = 0; |
29f59db3 | 8731 | |
62fb1851 | 8732 | if (on_rq) |
29f59db3 | 8733 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8734 | } |
62fb1851 | 8735 | |
c09595f6 PZ |
8736 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8737 | { | |
8738 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8739 | struct rq *rq = cfs_rq->rq; | |
8740 | unsigned long flags; | |
8741 | ||
8742 | spin_lock_irqsave(&rq->lock, flags); | |
8743 | __set_se_shares(se, shares); | |
8744 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8745 | } |
8746 | ||
8ed36996 PZ |
8747 | static DEFINE_MUTEX(shares_mutex); |
8748 | ||
4cf86d77 | 8749 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8750 | { |
8751 | int i; | |
8ed36996 | 8752 | unsigned long flags; |
c61935fd | 8753 | |
ec7dc8ac DG |
8754 | /* |
8755 | * We can't change the weight of the root cgroup. | |
8756 | */ | |
8757 | if (!tg->se[0]) | |
8758 | return -EINVAL; | |
8759 | ||
18d95a28 PZ |
8760 | if (shares < MIN_SHARES) |
8761 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8762 | else if (shares > MAX_SHARES) |
8763 | shares = MAX_SHARES; | |
62fb1851 | 8764 | |
8ed36996 | 8765 | mutex_lock(&shares_mutex); |
9b5b7751 | 8766 | if (tg->shares == shares) |
5cb350ba | 8767 | goto done; |
29f59db3 | 8768 | |
8ed36996 | 8769 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8770 | for_each_possible_cpu(i) |
8771 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8772 | list_del_rcu(&tg->siblings); |
8ed36996 | 8773 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8774 | |
8775 | /* wait for any ongoing reference to this group to finish */ | |
8776 | synchronize_sched(); | |
8777 | ||
8778 | /* | |
8779 | * Now we are free to modify the group's share on each cpu | |
8780 | * w/o tripping rebalance_share or load_balance_fair. | |
8781 | */ | |
9b5b7751 | 8782 | tg->shares = shares; |
c09595f6 PZ |
8783 | for_each_possible_cpu(i) { |
8784 | /* | |
8785 | * force a rebalance | |
8786 | */ | |
8787 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8788 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8789 | } |
29f59db3 | 8790 | |
6b2d7700 SV |
8791 | /* |
8792 | * Enable load balance activity on this group, by inserting it back on | |
8793 | * each cpu's rq->leaf_cfs_rq_list. | |
8794 | */ | |
8ed36996 | 8795 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8796 | for_each_possible_cpu(i) |
8797 | register_fair_sched_group(tg, i); | |
f473aa5e | 8798 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8799 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8800 | done: |
8ed36996 | 8801 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8802 | return 0; |
29f59db3 SV |
8803 | } |
8804 | ||
5cb350ba DG |
8805 | unsigned long sched_group_shares(struct task_group *tg) |
8806 | { | |
8807 | return tg->shares; | |
8808 | } | |
052f1dc7 | 8809 | #endif |
5cb350ba | 8810 | |
052f1dc7 | 8811 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8812 | /* |
9f0c1e56 | 8813 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8814 | */ |
9f0c1e56 PZ |
8815 | static DEFINE_MUTEX(rt_constraints_mutex); |
8816 | ||
8817 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8818 | { | |
8819 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8820 | return 1ULL << 20; |
9f0c1e56 | 8821 | |
9a7e0b18 | 8822 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8823 | } |
8824 | ||
9a7e0b18 PZ |
8825 | /* Must be called with tasklist_lock held */ |
8826 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8827 | { |
9a7e0b18 | 8828 | struct task_struct *g, *p; |
b40b2e8e | 8829 | |
9a7e0b18 PZ |
8830 | do_each_thread(g, p) { |
8831 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8832 | return 1; | |
8833 | } while_each_thread(g, p); | |
b40b2e8e | 8834 | |
9a7e0b18 PZ |
8835 | return 0; |
8836 | } | |
b40b2e8e | 8837 | |
9a7e0b18 PZ |
8838 | struct rt_schedulable_data { |
8839 | struct task_group *tg; | |
8840 | u64 rt_period; | |
8841 | u64 rt_runtime; | |
8842 | }; | |
b40b2e8e | 8843 | |
9a7e0b18 PZ |
8844 | static int tg_schedulable(struct task_group *tg, void *data) |
8845 | { | |
8846 | struct rt_schedulable_data *d = data; | |
8847 | struct task_group *child; | |
8848 | unsigned long total, sum = 0; | |
8849 | u64 period, runtime; | |
b40b2e8e | 8850 | |
9a7e0b18 PZ |
8851 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8852 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8853 | |
9a7e0b18 PZ |
8854 | if (tg == d->tg) { |
8855 | period = d->rt_period; | |
8856 | runtime = d->rt_runtime; | |
b40b2e8e | 8857 | } |
b40b2e8e | 8858 | |
4653f803 PZ |
8859 | /* |
8860 | * Cannot have more runtime than the period. | |
8861 | */ | |
8862 | if (runtime > period && runtime != RUNTIME_INF) | |
8863 | return -EINVAL; | |
6f505b16 | 8864 | |
4653f803 PZ |
8865 | /* |
8866 | * Ensure we don't starve existing RT tasks. | |
8867 | */ | |
9a7e0b18 PZ |
8868 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8869 | return -EBUSY; | |
6f505b16 | 8870 | |
9a7e0b18 | 8871 | total = to_ratio(period, runtime); |
6f505b16 | 8872 | |
4653f803 PZ |
8873 | /* |
8874 | * Nobody can have more than the global setting allows. | |
8875 | */ | |
8876 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8877 | return -EINVAL; | |
6f505b16 | 8878 | |
4653f803 PZ |
8879 | /* |
8880 | * The sum of our children's runtime should not exceed our own. | |
8881 | */ | |
9a7e0b18 PZ |
8882 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8883 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8884 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8885 | |
9a7e0b18 PZ |
8886 | if (child == d->tg) { |
8887 | period = d->rt_period; | |
8888 | runtime = d->rt_runtime; | |
8889 | } | |
6f505b16 | 8890 | |
9a7e0b18 | 8891 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8892 | } |
6f505b16 | 8893 | |
9a7e0b18 PZ |
8894 | if (sum > total) |
8895 | return -EINVAL; | |
8896 | ||
8897 | return 0; | |
6f505b16 PZ |
8898 | } |
8899 | ||
9a7e0b18 | 8900 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8901 | { |
9a7e0b18 PZ |
8902 | struct rt_schedulable_data data = { |
8903 | .tg = tg, | |
8904 | .rt_period = period, | |
8905 | .rt_runtime = runtime, | |
8906 | }; | |
8907 | ||
8908 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8909 | } |
8910 | ||
d0b27fa7 PZ |
8911 | static int tg_set_bandwidth(struct task_group *tg, |
8912 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8913 | { |
ac086bc2 | 8914 | int i, err = 0; |
9f0c1e56 | 8915 | |
9f0c1e56 | 8916 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8917 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8918 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8919 | if (err) | |
9f0c1e56 | 8920 | goto unlock; |
ac086bc2 PZ |
8921 | |
8922 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8923 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8924 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8925 | |
8926 | for_each_possible_cpu(i) { | |
8927 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8928 | ||
8929 | spin_lock(&rt_rq->rt_runtime_lock); | |
8930 | rt_rq->rt_runtime = rt_runtime; | |
8931 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8932 | } | |
8933 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8934 | unlock: |
521f1a24 | 8935 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8936 | mutex_unlock(&rt_constraints_mutex); |
8937 | ||
8938 | return err; | |
6f505b16 PZ |
8939 | } |
8940 | ||
d0b27fa7 PZ |
8941 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8942 | { | |
8943 | u64 rt_runtime, rt_period; | |
8944 | ||
8945 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8946 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8947 | if (rt_runtime_us < 0) | |
8948 | rt_runtime = RUNTIME_INF; | |
8949 | ||
8950 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8951 | } | |
8952 | ||
9f0c1e56 PZ |
8953 | long sched_group_rt_runtime(struct task_group *tg) |
8954 | { | |
8955 | u64 rt_runtime_us; | |
8956 | ||
d0b27fa7 | 8957 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8958 | return -1; |
8959 | ||
d0b27fa7 | 8960 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8961 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8962 | return rt_runtime_us; | |
8963 | } | |
d0b27fa7 PZ |
8964 | |
8965 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8966 | { | |
8967 | u64 rt_runtime, rt_period; | |
8968 | ||
8969 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8970 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8971 | ||
619b0488 R |
8972 | if (rt_period == 0) |
8973 | return -EINVAL; | |
8974 | ||
d0b27fa7 PZ |
8975 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8976 | } | |
8977 | ||
8978 | long sched_group_rt_period(struct task_group *tg) | |
8979 | { | |
8980 | u64 rt_period_us; | |
8981 | ||
8982 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8983 | do_div(rt_period_us, NSEC_PER_USEC); | |
8984 | return rt_period_us; | |
8985 | } | |
8986 | ||
8987 | static int sched_rt_global_constraints(void) | |
8988 | { | |
4653f803 | 8989 | u64 runtime, period; |
d0b27fa7 PZ |
8990 | int ret = 0; |
8991 | ||
ec5d4989 HS |
8992 | if (sysctl_sched_rt_period <= 0) |
8993 | return -EINVAL; | |
8994 | ||
4653f803 PZ |
8995 | runtime = global_rt_runtime(); |
8996 | period = global_rt_period(); | |
8997 | ||
8998 | /* | |
8999 | * Sanity check on the sysctl variables. | |
9000 | */ | |
9001 | if (runtime > period && runtime != RUNTIME_INF) | |
9002 | return -EINVAL; | |
10b612f4 | 9003 | |
d0b27fa7 | 9004 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9005 | read_lock(&tasklist_lock); |
4653f803 | 9006 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9007 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9008 | mutex_unlock(&rt_constraints_mutex); |
9009 | ||
9010 | return ret; | |
9011 | } | |
6d6bc0ad | 9012 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9013 | static int sched_rt_global_constraints(void) |
9014 | { | |
ac086bc2 PZ |
9015 | unsigned long flags; |
9016 | int i; | |
9017 | ||
ec5d4989 HS |
9018 | if (sysctl_sched_rt_period <= 0) |
9019 | return -EINVAL; | |
9020 | ||
ac086bc2 PZ |
9021 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9022 | for_each_possible_cpu(i) { | |
9023 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9024 | ||
9025 | spin_lock(&rt_rq->rt_runtime_lock); | |
9026 | rt_rq->rt_runtime = global_rt_runtime(); | |
9027 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9028 | } | |
9029 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9030 | ||
d0b27fa7 PZ |
9031 | return 0; |
9032 | } | |
6d6bc0ad | 9033 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9034 | |
9035 | int sched_rt_handler(struct ctl_table *table, int write, | |
9036 | struct file *filp, void __user *buffer, size_t *lenp, | |
9037 | loff_t *ppos) | |
9038 | { | |
9039 | int ret; | |
9040 | int old_period, old_runtime; | |
9041 | static DEFINE_MUTEX(mutex); | |
9042 | ||
9043 | mutex_lock(&mutex); | |
9044 | old_period = sysctl_sched_rt_period; | |
9045 | old_runtime = sysctl_sched_rt_runtime; | |
9046 | ||
9047 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9048 | ||
9049 | if (!ret && write) { | |
9050 | ret = sched_rt_global_constraints(); | |
9051 | if (ret) { | |
9052 | sysctl_sched_rt_period = old_period; | |
9053 | sysctl_sched_rt_runtime = old_runtime; | |
9054 | } else { | |
9055 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9056 | def_rt_bandwidth.rt_period = | |
9057 | ns_to_ktime(global_rt_period()); | |
9058 | } | |
9059 | } | |
9060 | mutex_unlock(&mutex); | |
9061 | ||
9062 | return ret; | |
9063 | } | |
68318b8e | 9064 | |
052f1dc7 | 9065 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9066 | |
9067 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9068 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9069 | { |
2b01dfe3 PM |
9070 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9071 | struct task_group, css); | |
68318b8e SV |
9072 | } |
9073 | ||
9074 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9075 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9076 | { |
ec7dc8ac | 9077 | struct task_group *tg, *parent; |
68318b8e | 9078 | |
2b01dfe3 | 9079 | if (!cgrp->parent) { |
68318b8e | 9080 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9081 | return &init_task_group.css; |
9082 | } | |
9083 | ||
ec7dc8ac DG |
9084 | parent = cgroup_tg(cgrp->parent); |
9085 | tg = sched_create_group(parent); | |
68318b8e SV |
9086 | if (IS_ERR(tg)) |
9087 | return ERR_PTR(-ENOMEM); | |
9088 | ||
68318b8e SV |
9089 | return &tg->css; |
9090 | } | |
9091 | ||
41a2d6cf IM |
9092 | static void |
9093 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9094 | { |
2b01dfe3 | 9095 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9096 | |
9097 | sched_destroy_group(tg); | |
9098 | } | |
9099 | ||
41a2d6cf IM |
9100 | static int |
9101 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9102 | struct task_struct *tsk) | |
68318b8e | 9103 | { |
b68aa230 PZ |
9104 | #ifdef CONFIG_RT_GROUP_SCHED |
9105 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9106 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9107 | return -EINVAL; |
9108 | #else | |
68318b8e SV |
9109 | /* We don't support RT-tasks being in separate groups */ |
9110 | if (tsk->sched_class != &fair_sched_class) | |
9111 | return -EINVAL; | |
b68aa230 | 9112 | #endif |
68318b8e SV |
9113 | |
9114 | return 0; | |
9115 | } | |
9116 | ||
9117 | static void | |
2b01dfe3 | 9118 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9119 | struct cgroup *old_cont, struct task_struct *tsk) |
9120 | { | |
9121 | sched_move_task(tsk); | |
9122 | } | |
9123 | ||
052f1dc7 | 9124 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9125 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9126 | u64 shareval) |
68318b8e | 9127 | { |
2b01dfe3 | 9128 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9129 | } |
9130 | ||
f4c753b7 | 9131 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9132 | { |
2b01dfe3 | 9133 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9134 | |
9135 | return (u64) tg->shares; | |
9136 | } | |
6d6bc0ad | 9137 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9138 | |
052f1dc7 | 9139 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9140 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9141 | s64 val) |
6f505b16 | 9142 | { |
06ecb27c | 9143 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9144 | } |
9145 | ||
06ecb27c | 9146 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9147 | { |
06ecb27c | 9148 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9149 | } |
d0b27fa7 PZ |
9150 | |
9151 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9152 | u64 rt_period_us) | |
9153 | { | |
9154 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9155 | } | |
9156 | ||
9157 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9158 | { | |
9159 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9160 | } | |
6d6bc0ad | 9161 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9162 | |
fe5c7cc2 | 9163 | static struct cftype cpu_files[] = { |
052f1dc7 | 9164 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9165 | { |
9166 | .name = "shares", | |
f4c753b7 PM |
9167 | .read_u64 = cpu_shares_read_u64, |
9168 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9169 | }, |
052f1dc7 PZ |
9170 | #endif |
9171 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9172 | { |
9f0c1e56 | 9173 | .name = "rt_runtime_us", |
06ecb27c PM |
9174 | .read_s64 = cpu_rt_runtime_read, |
9175 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9176 | }, |
d0b27fa7 PZ |
9177 | { |
9178 | .name = "rt_period_us", | |
f4c753b7 PM |
9179 | .read_u64 = cpu_rt_period_read_uint, |
9180 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9181 | }, |
052f1dc7 | 9182 | #endif |
68318b8e SV |
9183 | }; |
9184 | ||
9185 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9186 | { | |
fe5c7cc2 | 9187 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9188 | } |
9189 | ||
9190 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9191 | .name = "cpu", |
9192 | .create = cpu_cgroup_create, | |
9193 | .destroy = cpu_cgroup_destroy, | |
9194 | .can_attach = cpu_cgroup_can_attach, | |
9195 | .attach = cpu_cgroup_attach, | |
9196 | .populate = cpu_cgroup_populate, | |
9197 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9198 | .early_init = 1, |
9199 | }; | |
9200 | ||
052f1dc7 | 9201 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9202 | |
9203 | #ifdef CONFIG_CGROUP_CPUACCT | |
9204 | ||
9205 | /* | |
9206 | * CPU accounting code for task groups. | |
9207 | * | |
9208 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9209 | * (balbir@in.ibm.com). | |
9210 | */ | |
9211 | ||
9212 | /* track cpu usage of a group of tasks */ | |
9213 | struct cpuacct { | |
9214 | struct cgroup_subsys_state css; | |
9215 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9216 | u64 *cpuusage; | |
9217 | }; | |
9218 | ||
9219 | struct cgroup_subsys cpuacct_subsys; | |
9220 | ||
9221 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9222 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9223 | { |
32cd756a | 9224 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9225 | struct cpuacct, css); |
9226 | } | |
9227 | ||
9228 | /* return cpu accounting group to which this task belongs */ | |
9229 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9230 | { | |
9231 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9232 | struct cpuacct, css); | |
9233 | } | |
9234 | ||
9235 | /* create a new cpu accounting group */ | |
9236 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9237 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9238 | { |
9239 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9240 | ||
9241 | if (!ca) | |
9242 | return ERR_PTR(-ENOMEM); | |
9243 | ||
9244 | ca->cpuusage = alloc_percpu(u64); | |
9245 | if (!ca->cpuusage) { | |
9246 | kfree(ca); | |
9247 | return ERR_PTR(-ENOMEM); | |
9248 | } | |
9249 | ||
9250 | return &ca->css; | |
9251 | } | |
9252 | ||
9253 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9254 | static void |
32cd756a | 9255 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9256 | { |
32cd756a | 9257 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9258 | |
9259 | free_percpu(ca->cpuusage); | |
9260 | kfree(ca); | |
9261 | } | |
9262 | ||
9263 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9264 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9265 | { |
32cd756a | 9266 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9267 | u64 totalcpuusage = 0; |
9268 | int i; | |
9269 | ||
9270 | for_each_possible_cpu(i) { | |
9271 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9272 | ||
9273 | /* | |
9274 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9275 | * platforms. | |
9276 | */ | |
9277 | spin_lock_irq(&cpu_rq(i)->lock); | |
9278 | totalcpuusage += *cpuusage; | |
9279 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9280 | } | |
9281 | ||
9282 | return totalcpuusage; | |
9283 | } | |
9284 | ||
0297b803 DG |
9285 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9286 | u64 reset) | |
9287 | { | |
9288 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9289 | int err = 0; | |
9290 | int i; | |
9291 | ||
9292 | if (reset) { | |
9293 | err = -EINVAL; | |
9294 | goto out; | |
9295 | } | |
9296 | ||
9297 | for_each_possible_cpu(i) { | |
9298 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9299 | ||
9300 | spin_lock_irq(&cpu_rq(i)->lock); | |
9301 | *cpuusage = 0; | |
9302 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9303 | } | |
9304 | out: | |
9305 | return err; | |
9306 | } | |
9307 | ||
d842de87 SV |
9308 | static struct cftype files[] = { |
9309 | { | |
9310 | .name = "usage", | |
f4c753b7 PM |
9311 | .read_u64 = cpuusage_read, |
9312 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9313 | }, |
9314 | }; | |
9315 | ||
32cd756a | 9316 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9317 | { |
32cd756a | 9318 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9319 | } |
9320 | ||
9321 | /* | |
9322 | * charge this task's execution time to its accounting group. | |
9323 | * | |
9324 | * called with rq->lock held. | |
9325 | */ | |
9326 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9327 | { | |
9328 | struct cpuacct *ca; | |
9329 | ||
9330 | if (!cpuacct_subsys.active) | |
9331 | return; | |
9332 | ||
9333 | ca = task_ca(tsk); | |
9334 | if (ca) { | |
9335 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9336 | ||
9337 | *cpuusage += cputime; | |
9338 | } | |
9339 | } | |
9340 | ||
9341 | struct cgroup_subsys cpuacct_subsys = { | |
9342 | .name = "cpuacct", | |
9343 | .create = cpuacct_create, | |
9344 | .destroy = cpuacct_destroy, | |
9345 | .populate = cpuacct_populate, | |
9346 | .subsys_id = cpuacct_subsys_id, | |
9347 | }; | |
9348 | #endif /* CONFIG_CGROUP_CPUACCT */ |