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> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
434d53b0 | 70 | #include <linux/bootmem.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
d7876a08 | 99 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 100 | */ |
d6322faf | 101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 102 | |
6aa645ea IM |
103 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
104 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
105 | ||
1da177e4 LT |
106 | /* |
107 | * These are the 'tuning knobs' of the scheduler: | |
108 | * | |
a4ec24b4 | 109 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
110 | * Timeslices get refilled after they expire. |
111 | */ | |
1da177e4 | 112 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 113 | |
d0b27fa7 PZ |
114 | /* |
115 | * single value that denotes runtime == period, ie unlimited time. | |
116 | */ | |
117 | #define RUNTIME_INF ((u64)~0ULL) | |
118 | ||
5517d86b ED |
119 | #ifdef CONFIG_SMP |
120 | /* | |
121 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
122 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
123 | */ | |
124 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
125 | { | |
126 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
127 | } | |
128 | ||
129 | /* | |
130 | * Each time a sched group cpu_power is changed, | |
131 | * we must compute its reciprocal value | |
132 | */ | |
133 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
134 | { | |
135 | sg->__cpu_power += val; | |
136 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
137 | } | |
138 | #endif | |
139 | ||
e05606d3 IM |
140 | static inline int rt_policy(int policy) |
141 | { | |
3f33a7ce | 142 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
143 | return 1; |
144 | return 0; | |
145 | } | |
146 | ||
147 | static inline int task_has_rt_policy(struct task_struct *p) | |
148 | { | |
149 | return rt_policy(p->policy); | |
150 | } | |
151 | ||
1da177e4 | 152 | /* |
6aa645ea | 153 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 154 | */ |
6aa645ea IM |
155 | struct rt_prio_array { |
156 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
157 | struct list_head queue[MAX_RT_PRIO]; | |
158 | }; | |
159 | ||
d0b27fa7 | 160 | struct rt_bandwidth { |
ea736ed5 IM |
161 | /* nests inside the rq lock: */ |
162 | spinlock_t rt_runtime_lock; | |
163 | ktime_t rt_period; | |
164 | u64 rt_runtime; | |
165 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
166 | }; |
167 | ||
168 | static struct rt_bandwidth def_rt_bandwidth; | |
169 | ||
170 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
171 | ||
172 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
173 | { | |
174 | struct rt_bandwidth *rt_b = | |
175 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
176 | ktime_t now; | |
177 | int overrun; | |
178 | int idle = 0; | |
179 | ||
180 | for (;;) { | |
181 | now = hrtimer_cb_get_time(timer); | |
182 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
183 | ||
184 | if (!overrun) | |
185 | break; | |
186 | ||
187 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
188 | } | |
189 | ||
190 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
191 | } | |
192 | ||
193 | static | |
194 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
195 | { | |
196 | rt_b->rt_period = ns_to_ktime(period); | |
197 | rt_b->rt_runtime = runtime; | |
198 | ||
ac086bc2 PZ |
199 | spin_lock_init(&rt_b->rt_runtime_lock); |
200 | ||
d0b27fa7 PZ |
201 | hrtimer_init(&rt_b->rt_period_timer, |
202 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
203 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
204 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
205 | } | |
206 | ||
207 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
208 | { | |
209 | ktime_t now; | |
210 | ||
211 | if (rt_b->rt_runtime == RUNTIME_INF) | |
212 | return; | |
213 | ||
214 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
215 | return; | |
216 | ||
217 | spin_lock(&rt_b->rt_runtime_lock); | |
218 | for (;;) { | |
219 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
220 | break; | |
221 | ||
222 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
223 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
224 | hrtimer_start(&rt_b->rt_period_timer, | |
225 | rt_b->rt_period_timer.expires, | |
226 | HRTIMER_MODE_ABS); | |
227 | } | |
228 | spin_unlock(&rt_b->rt_runtime_lock); | |
229 | } | |
230 | ||
231 | #ifdef CONFIG_RT_GROUP_SCHED | |
232 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
233 | { | |
234 | hrtimer_cancel(&rt_b->rt_period_timer); | |
235 | } | |
236 | #endif | |
237 | ||
712555ee HC |
238 | /* |
239 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
240 | * detach_destroy_domains and partition_sched_domains. | |
241 | */ | |
242 | static DEFINE_MUTEX(sched_domains_mutex); | |
243 | ||
052f1dc7 | 244 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 245 | |
68318b8e SV |
246 | #include <linux/cgroup.h> |
247 | ||
29f59db3 SV |
248 | struct cfs_rq; |
249 | ||
6f505b16 PZ |
250 | static LIST_HEAD(task_groups); |
251 | ||
29f59db3 | 252 | /* task group related information */ |
4cf86d77 | 253 | struct task_group { |
052f1dc7 | 254 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
255 | struct cgroup_subsys_state css; |
256 | #endif | |
052f1dc7 PZ |
257 | |
258 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
259 | /* schedulable entities of this group on each cpu */ |
260 | struct sched_entity **se; | |
261 | /* runqueue "owned" by this group on each cpu */ | |
262 | struct cfs_rq **cfs_rq; | |
263 | unsigned long shares; | |
052f1dc7 PZ |
264 | #endif |
265 | ||
266 | #ifdef CONFIG_RT_GROUP_SCHED | |
267 | struct sched_rt_entity **rt_se; | |
268 | struct rt_rq **rt_rq; | |
269 | ||
d0b27fa7 | 270 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 271 | #endif |
6b2d7700 | 272 | |
ae8393e5 | 273 | struct rcu_head rcu; |
6f505b16 | 274 | struct list_head list; |
f473aa5e PZ |
275 | |
276 | struct task_group *parent; | |
277 | struct list_head siblings; | |
278 | struct list_head children; | |
29f59db3 SV |
279 | }; |
280 | ||
354d60c2 | 281 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
282 | |
283 | /* | |
284 | * Root task group. | |
285 | * Every UID task group (including init_task_group aka UID-0) will | |
286 | * be a child to this group. | |
287 | */ | |
288 | struct task_group root_task_group; | |
289 | ||
052f1dc7 | 290 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
291 | /* Default task group's sched entity on each cpu */ |
292 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
293 | /* Default task group's cfs_rq on each cpu */ | |
294 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 295 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
296 | |
297 | #ifdef CONFIG_RT_GROUP_SCHED | |
298 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
299 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad DG |
300 | #endif /* CONFIG_RT_GROUP_SCHED */ |
301 | #else /* !CONFIG_FAIR_GROUP_SCHED */ | |
eff766a6 | 302 | #define root_task_group init_task_group |
6d6bc0ad | 303 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6f505b16 | 304 | |
8ed36996 | 305 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
306 | * a task group's cpu shares. |
307 | */ | |
8ed36996 | 308 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 309 | |
052f1dc7 | 310 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
311 | #ifdef CONFIG_USER_SCHED |
312 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 313 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 314 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 315 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 316 | |
cb4ad1ff | 317 | /* |
2e084786 LJ |
318 | * A weight of 0 or 1 can cause arithmetics problems. |
319 | * A weight of a cfs_rq is the sum of weights of which entities | |
320 | * are queued on this cfs_rq, so a weight of a entity should not be | |
321 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
322 | * (The default weight is 1024 - so there's no practical |
323 | * limitation from this.) | |
324 | */ | |
18d95a28 | 325 | #define MIN_SHARES 2 |
2e084786 | 326 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 327 | |
052f1dc7 PZ |
328 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
329 | #endif | |
330 | ||
29f59db3 | 331 | /* Default task group. |
3a252015 | 332 | * Every task in system belong to this group at bootup. |
29f59db3 | 333 | */ |
434d53b0 | 334 | struct task_group init_task_group; |
29f59db3 SV |
335 | |
336 | /* return group to which a task belongs */ | |
4cf86d77 | 337 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 338 | { |
4cf86d77 | 339 | struct task_group *tg; |
9b5b7751 | 340 | |
052f1dc7 | 341 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 342 | tg = p->user->tg; |
052f1dc7 | 343 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
344 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
345 | struct task_group, css); | |
24e377a8 | 346 | #else |
41a2d6cf | 347 | tg = &init_task_group; |
24e377a8 | 348 | #endif |
9b5b7751 | 349 | return tg; |
29f59db3 SV |
350 | } |
351 | ||
352 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 353 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 354 | { |
052f1dc7 | 355 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
356 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
357 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 358 | #endif |
6f505b16 | 359 | |
052f1dc7 | 360 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
361 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
362 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 363 | #endif |
29f59db3 SV |
364 | } |
365 | ||
366 | #else | |
367 | ||
6f505b16 | 368 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
369 | static inline struct task_group *task_group(struct task_struct *p) |
370 | { | |
371 | return NULL; | |
372 | } | |
29f59db3 | 373 | |
052f1dc7 | 374 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 375 | |
6aa645ea IM |
376 | /* CFS-related fields in a runqueue */ |
377 | struct cfs_rq { | |
378 | struct load_weight load; | |
379 | unsigned long nr_running; | |
380 | ||
6aa645ea | 381 | u64 exec_clock; |
e9acbff6 | 382 | u64 min_vruntime; |
103638d9 | 383 | u64 pair_start; |
6aa645ea IM |
384 | |
385 | struct rb_root tasks_timeline; | |
386 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
387 | |
388 | struct list_head tasks; | |
389 | struct list_head *balance_iterator; | |
390 | ||
391 | /* | |
392 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
393 | * It is set to NULL otherwise (i.e when none are currently running). |
394 | */ | |
aa2ac252 | 395 | struct sched_entity *curr, *next; |
ddc97297 PZ |
396 | |
397 | unsigned long nr_spread_over; | |
398 | ||
62160e3f | 399 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
400 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
401 | ||
41a2d6cf IM |
402 | /* |
403 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
404 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
405 | * (like users, containers etc.) | |
406 | * | |
407 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
408 | * list is used during load balance. | |
409 | */ | |
41a2d6cf IM |
410 | struct list_head leaf_cfs_rq_list; |
411 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
412 | |
413 | #ifdef CONFIG_SMP | |
c09595f6 | 414 | /* |
c8cba857 | 415 | * the part of load.weight contributed by tasks |
c09595f6 | 416 | */ |
c8cba857 | 417 | unsigned long task_weight; |
c09595f6 | 418 | |
c8cba857 PZ |
419 | /* |
420 | * h_load = weight * f(tg) | |
421 | * | |
422 | * Where f(tg) is the recursive weight fraction assigned to | |
423 | * this group. | |
424 | */ | |
425 | unsigned long h_load; | |
c09595f6 | 426 | |
c8cba857 PZ |
427 | /* |
428 | * this cpu's part of tg->shares | |
429 | */ | |
430 | unsigned long shares; | |
f1d239f7 PZ |
431 | |
432 | /* | |
433 | * load.weight at the time we set shares | |
434 | */ | |
435 | unsigned long rq_weight; | |
c09595f6 | 436 | #endif |
6aa645ea IM |
437 | #endif |
438 | }; | |
1da177e4 | 439 | |
6aa645ea IM |
440 | /* Real-Time classes' related field in a runqueue: */ |
441 | struct rt_rq { | |
442 | struct rt_prio_array active; | |
63489e45 | 443 | unsigned long rt_nr_running; |
052f1dc7 | 444 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
445 | int highest_prio; /* highest queued rt task prio */ |
446 | #endif | |
fa85ae24 | 447 | #ifdef CONFIG_SMP |
73fe6aae | 448 | unsigned long rt_nr_migratory; |
a22d7fc1 | 449 | int overloaded; |
fa85ae24 | 450 | #endif |
6f505b16 | 451 | int rt_throttled; |
fa85ae24 | 452 | u64 rt_time; |
ac086bc2 | 453 | u64 rt_runtime; |
ea736ed5 | 454 | /* Nests inside the rq lock: */ |
ac086bc2 | 455 | spinlock_t rt_runtime_lock; |
6f505b16 | 456 | |
052f1dc7 | 457 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
458 | unsigned long rt_nr_boosted; |
459 | ||
6f505b16 PZ |
460 | struct rq *rq; |
461 | struct list_head leaf_rt_rq_list; | |
462 | struct task_group *tg; | |
463 | struct sched_rt_entity *rt_se; | |
464 | #endif | |
6aa645ea IM |
465 | }; |
466 | ||
57d885fe GH |
467 | #ifdef CONFIG_SMP |
468 | ||
469 | /* | |
470 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
471 | * variables. Each exclusive cpuset essentially defines an island domain by |
472 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
473 | * exclusive cpuset is created, we also create and attach a new root-domain |
474 | * object. | |
475 | * | |
57d885fe GH |
476 | */ |
477 | struct root_domain { | |
478 | atomic_t refcount; | |
479 | cpumask_t span; | |
480 | cpumask_t online; | |
637f5085 | 481 | |
0eab9146 | 482 | /* |
637f5085 GH |
483 | * The "RT overload" flag: it gets set if a CPU has more than |
484 | * one runnable RT task. | |
485 | */ | |
486 | cpumask_t rto_mask; | |
0eab9146 | 487 | atomic_t rto_count; |
6e0534f2 GH |
488 | #ifdef CONFIG_SMP |
489 | struct cpupri cpupri; | |
490 | #endif | |
57d885fe GH |
491 | }; |
492 | ||
dc938520 GH |
493 | /* |
494 | * By default the system creates a single root-domain with all cpus as | |
495 | * members (mimicking the global state we have today). | |
496 | */ | |
57d885fe GH |
497 | static struct root_domain def_root_domain; |
498 | ||
499 | #endif | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * This is the main, per-CPU runqueue data structure. | |
503 | * | |
504 | * Locking rule: those places that want to lock multiple runqueues | |
505 | * (such as the load balancing or the thread migration code), lock | |
506 | * acquire operations must be ordered by ascending &runqueue. | |
507 | */ | |
70b97a7f | 508 | struct rq { |
d8016491 IM |
509 | /* runqueue lock: */ |
510 | spinlock_t lock; | |
1da177e4 LT |
511 | |
512 | /* | |
513 | * nr_running and cpu_load should be in the same cacheline because | |
514 | * remote CPUs use both these fields when doing load calculation. | |
515 | */ | |
516 | unsigned long nr_running; | |
6aa645ea IM |
517 | #define CPU_LOAD_IDX_MAX 5 |
518 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 519 | unsigned char idle_at_tick; |
46cb4b7c | 520 | #ifdef CONFIG_NO_HZ |
15934a37 | 521 | unsigned long last_tick_seen; |
46cb4b7c SS |
522 | unsigned char in_nohz_recently; |
523 | #endif | |
d8016491 IM |
524 | /* capture load from *all* tasks on this cpu: */ |
525 | struct load_weight load; | |
6aa645ea IM |
526 | unsigned long nr_load_updates; |
527 | u64 nr_switches; | |
528 | ||
529 | struct cfs_rq cfs; | |
6f505b16 | 530 | struct rt_rq rt; |
6f505b16 | 531 | |
6aa645ea | 532 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
533 | /* list of leaf cfs_rq on this cpu: */ |
534 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
535 | #endif |
536 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 537 | struct list_head leaf_rt_rq_list; |
1da177e4 | 538 | #endif |
1da177e4 LT |
539 | |
540 | /* | |
541 | * This is part of a global counter where only the total sum | |
542 | * over all CPUs matters. A task can increase this counter on | |
543 | * one CPU and if it got migrated afterwards it may decrease | |
544 | * it on another CPU. Always updated under the runqueue lock: | |
545 | */ | |
546 | unsigned long nr_uninterruptible; | |
547 | ||
36c8b586 | 548 | struct task_struct *curr, *idle; |
c9819f45 | 549 | unsigned long next_balance; |
1da177e4 | 550 | struct mm_struct *prev_mm; |
6aa645ea | 551 | |
3e51f33f | 552 | u64 clock; |
6aa645ea | 553 | |
1da177e4 LT |
554 | atomic_t nr_iowait; |
555 | ||
556 | #ifdef CONFIG_SMP | |
0eab9146 | 557 | struct root_domain *rd; |
1da177e4 LT |
558 | struct sched_domain *sd; |
559 | ||
560 | /* For active balancing */ | |
561 | int active_balance; | |
562 | int push_cpu; | |
d8016491 IM |
563 | /* cpu of this runqueue: */ |
564 | int cpu; | |
1f11eb6a | 565 | int online; |
1da177e4 | 566 | |
a8a51d5e | 567 | unsigned long avg_load_per_task; |
1da177e4 | 568 | |
36c8b586 | 569 | struct task_struct *migration_thread; |
1da177e4 LT |
570 | struct list_head migration_queue; |
571 | #endif | |
572 | ||
8f4d37ec | 573 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
574 | #ifdef CONFIG_SMP |
575 | int hrtick_csd_pending; | |
576 | struct call_single_data hrtick_csd; | |
577 | #endif | |
8f4d37ec PZ |
578 | struct hrtimer hrtick_timer; |
579 | #endif | |
580 | ||
1da177e4 LT |
581 | #ifdef CONFIG_SCHEDSTATS |
582 | /* latency stats */ | |
583 | struct sched_info rq_sched_info; | |
584 | ||
585 | /* sys_sched_yield() stats */ | |
480b9434 KC |
586 | unsigned int yld_exp_empty; |
587 | unsigned int yld_act_empty; | |
588 | unsigned int yld_both_empty; | |
589 | unsigned int yld_count; | |
1da177e4 LT |
590 | |
591 | /* schedule() stats */ | |
480b9434 KC |
592 | unsigned int sched_switch; |
593 | unsigned int sched_count; | |
594 | unsigned int sched_goidle; | |
1da177e4 LT |
595 | |
596 | /* try_to_wake_up() stats */ | |
480b9434 KC |
597 | unsigned int ttwu_count; |
598 | unsigned int ttwu_local; | |
b8efb561 IM |
599 | |
600 | /* BKL stats */ | |
480b9434 | 601 | unsigned int bkl_count; |
1da177e4 | 602 | #endif |
fcb99371 | 603 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
604 | }; |
605 | ||
f34e3b61 | 606 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 607 | |
dd41f596 IM |
608 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
609 | { | |
610 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
611 | } | |
612 | ||
0a2966b4 CL |
613 | static inline int cpu_of(struct rq *rq) |
614 | { | |
615 | #ifdef CONFIG_SMP | |
616 | return rq->cpu; | |
617 | #else | |
618 | return 0; | |
619 | #endif | |
620 | } | |
621 | ||
674311d5 NP |
622 | /* |
623 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 624 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
625 | * |
626 | * The domain tree of any CPU may only be accessed from within | |
627 | * preempt-disabled sections. | |
628 | */ | |
48f24c4d IM |
629 | #define for_each_domain(cpu, __sd) \ |
630 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
631 | |
632 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
633 | #define this_rq() (&__get_cpu_var(runqueues)) | |
634 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
635 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
636 | ||
3e51f33f PZ |
637 | static inline void update_rq_clock(struct rq *rq) |
638 | { | |
639 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
640 | } | |
641 | ||
bf5c91ba IM |
642 | /* |
643 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
644 | */ | |
645 | #ifdef CONFIG_SCHED_DEBUG | |
646 | # define const_debug __read_mostly | |
647 | #else | |
648 | # define const_debug static const | |
649 | #endif | |
650 | ||
017730c1 IM |
651 | /** |
652 | * runqueue_is_locked | |
653 | * | |
654 | * Returns true if the current cpu runqueue is locked. | |
655 | * This interface allows printk to be called with the runqueue lock | |
656 | * held and know whether or not it is OK to wake up the klogd. | |
657 | */ | |
658 | int runqueue_is_locked(void) | |
659 | { | |
660 | int cpu = get_cpu(); | |
661 | struct rq *rq = cpu_rq(cpu); | |
662 | int ret; | |
663 | ||
664 | ret = spin_is_locked(&rq->lock); | |
665 | put_cpu(); | |
666 | return ret; | |
667 | } | |
668 | ||
bf5c91ba IM |
669 | /* |
670 | * Debugging: various feature bits | |
671 | */ | |
f00b45c1 PZ |
672 | |
673 | #define SCHED_FEAT(name, enabled) \ | |
674 | __SCHED_FEAT_##name , | |
675 | ||
bf5c91ba | 676 | enum { |
f00b45c1 | 677 | #include "sched_features.h" |
bf5c91ba IM |
678 | }; |
679 | ||
f00b45c1 PZ |
680 | #undef SCHED_FEAT |
681 | ||
682 | #define SCHED_FEAT(name, enabled) \ | |
683 | (1UL << __SCHED_FEAT_##name) * enabled | | |
684 | ||
bf5c91ba | 685 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
686 | #include "sched_features.h" |
687 | 0; | |
688 | ||
689 | #undef SCHED_FEAT | |
690 | ||
691 | #ifdef CONFIG_SCHED_DEBUG | |
692 | #define SCHED_FEAT(name, enabled) \ | |
693 | #name , | |
694 | ||
983ed7a6 | 695 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
696 | #include "sched_features.h" |
697 | NULL | |
698 | }; | |
699 | ||
700 | #undef SCHED_FEAT | |
701 | ||
983ed7a6 | 702 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
703 | { |
704 | filp->private_data = inode->i_private; | |
705 | return 0; | |
706 | } | |
707 | ||
708 | static ssize_t | |
709 | sched_feat_read(struct file *filp, char __user *ubuf, | |
710 | size_t cnt, loff_t *ppos) | |
711 | { | |
712 | char *buf; | |
713 | int r = 0; | |
714 | int len = 0; | |
715 | int i; | |
716 | ||
717 | for (i = 0; sched_feat_names[i]; i++) { | |
718 | len += strlen(sched_feat_names[i]); | |
719 | len += 4; | |
720 | } | |
721 | ||
722 | buf = kmalloc(len + 2, GFP_KERNEL); | |
723 | if (!buf) | |
724 | return -ENOMEM; | |
725 | ||
726 | for (i = 0; sched_feat_names[i]; i++) { | |
727 | if (sysctl_sched_features & (1UL << i)) | |
728 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
729 | else | |
c24b7c52 | 730 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
731 | } |
732 | ||
733 | r += sprintf(buf + r, "\n"); | |
734 | WARN_ON(r >= len + 2); | |
735 | ||
736 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
737 | ||
738 | kfree(buf); | |
739 | ||
740 | return r; | |
741 | } | |
742 | ||
743 | static ssize_t | |
744 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
745 | size_t cnt, loff_t *ppos) | |
746 | { | |
747 | char buf[64]; | |
748 | char *cmp = buf; | |
749 | int neg = 0; | |
750 | int i; | |
751 | ||
752 | if (cnt > 63) | |
753 | cnt = 63; | |
754 | ||
755 | if (copy_from_user(&buf, ubuf, cnt)) | |
756 | return -EFAULT; | |
757 | ||
758 | buf[cnt] = 0; | |
759 | ||
c24b7c52 | 760 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
761 | neg = 1; |
762 | cmp += 3; | |
763 | } | |
764 | ||
765 | for (i = 0; sched_feat_names[i]; i++) { | |
766 | int len = strlen(sched_feat_names[i]); | |
767 | ||
768 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
769 | if (neg) | |
770 | sysctl_sched_features &= ~(1UL << i); | |
771 | else | |
772 | sysctl_sched_features |= (1UL << i); | |
773 | break; | |
774 | } | |
775 | } | |
776 | ||
777 | if (!sched_feat_names[i]) | |
778 | return -EINVAL; | |
779 | ||
780 | filp->f_pos += cnt; | |
781 | ||
782 | return cnt; | |
783 | } | |
784 | ||
785 | static struct file_operations sched_feat_fops = { | |
786 | .open = sched_feat_open, | |
787 | .read = sched_feat_read, | |
788 | .write = sched_feat_write, | |
789 | }; | |
790 | ||
791 | static __init int sched_init_debug(void) | |
792 | { | |
f00b45c1 PZ |
793 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
794 | &sched_feat_fops); | |
795 | ||
796 | return 0; | |
797 | } | |
798 | late_initcall(sched_init_debug); | |
799 | ||
800 | #endif | |
801 | ||
802 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 803 | |
b82d9fdd PZ |
804 | /* |
805 | * Number of tasks to iterate in a single balance run. | |
806 | * Limited because this is done with IRQs disabled. | |
807 | */ | |
808 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
809 | ||
2398f2c6 PZ |
810 | /* |
811 | * ratelimit for updating the group shares. | |
812 | * default: 0.5ms | |
813 | */ | |
814 | const_debug unsigned int sysctl_sched_shares_ratelimit = 500000; | |
815 | ||
fa85ae24 | 816 | /* |
9f0c1e56 | 817 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
818 | * default: 1s |
819 | */ | |
9f0c1e56 | 820 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 821 | |
6892b75e IM |
822 | static __read_mostly int scheduler_running; |
823 | ||
9f0c1e56 PZ |
824 | /* |
825 | * part of the period that we allow rt tasks to run in us. | |
826 | * default: 0.95s | |
827 | */ | |
828 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 829 | |
d0b27fa7 PZ |
830 | static inline u64 global_rt_period(void) |
831 | { | |
832 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
833 | } | |
834 | ||
835 | static inline u64 global_rt_runtime(void) | |
836 | { | |
837 | if (sysctl_sched_rt_period < 0) | |
838 | return RUNTIME_INF; | |
839 | ||
840 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
841 | } | |
fa85ae24 | 842 | |
1da177e4 | 843 | #ifndef prepare_arch_switch |
4866cde0 NP |
844 | # define prepare_arch_switch(next) do { } while (0) |
845 | #endif | |
846 | #ifndef finish_arch_switch | |
847 | # define finish_arch_switch(prev) do { } while (0) | |
848 | #endif | |
849 | ||
051a1d1a DA |
850 | static inline int task_current(struct rq *rq, struct task_struct *p) |
851 | { | |
852 | return rq->curr == p; | |
853 | } | |
854 | ||
4866cde0 | 855 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 856 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 857 | { |
051a1d1a | 858 | return task_current(rq, p); |
4866cde0 NP |
859 | } |
860 | ||
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
862 | { |
863 | } | |
864 | ||
70b97a7f | 865 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 866 | { |
da04c035 IM |
867 | #ifdef CONFIG_DEBUG_SPINLOCK |
868 | /* this is a valid case when another task releases the spinlock */ | |
869 | rq->lock.owner = current; | |
870 | #endif | |
8a25d5de IM |
871 | /* |
872 | * If we are tracking spinlock dependencies then we have to | |
873 | * fix up the runqueue lock - which gets 'carried over' from | |
874 | * prev into current: | |
875 | */ | |
876 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
877 | ||
4866cde0 NP |
878 | spin_unlock_irq(&rq->lock); |
879 | } | |
880 | ||
881 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 882 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | return p->oncpu; | |
886 | #else | |
051a1d1a | 887 | return task_current(rq, p); |
4866cde0 NP |
888 | #endif |
889 | } | |
890 | ||
70b97a7f | 891 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
892 | { |
893 | #ifdef CONFIG_SMP | |
894 | /* | |
895 | * We can optimise this out completely for !SMP, because the | |
896 | * SMP rebalancing from interrupt is the only thing that cares | |
897 | * here. | |
898 | */ | |
899 | next->oncpu = 1; | |
900 | #endif | |
901 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
902 | spin_unlock_irq(&rq->lock); | |
903 | #else | |
904 | spin_unlock(&rq->lock); | |
905 | #endif | |
906 | } | |
907 | ||
70b97a7f | 908 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
909 | { |
910 | #ifdef CONFIG_SMP | |
911 | /* | |
912 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
913 | * We must ensure this doesn't happen until the switch is completely | |
914 | * finished. | |
915 | */ | |
916 | smp_wmb(); | |
917 | prev->oncpu = 0; | |
918 | #endif | |
919 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
920 | local_irq_enable(); | |
1da177e4 | 921 | #endif |
4866cde0 NP |
922 | } |
923 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 924 | |
b29739f9 IM |
925 | /* |
926 | * __task_rq_lock - lock the runqueue a given task resides on. | |
927 | * Must be called interrupts disabled. | |
928 | */ | |
70b97a7f | 929 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
930 | __acquires(rq->lock) |
931 | { | |
3a5c359a AK |
932 | for (;;) { |
933 | struct rq *rq = task_rq(p); | |
934 | spin_lock(&rq->lock); | |
935 | if (likely(rq == task_rq(p))) | |
936 | return rq; | |
b29739f9 | 937 | spin_unlock(&rq->lock); |
b29739f9 | 938 | } |
b29739f9 IM |
939 | } |
940 | ||
1da177e4 LT |
941 | /* |
942 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 943 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
944 | * explicitly disabling preemption. |
945 | */ | |
70b97a7f | 946 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
947 | __acquires(rq->lock) |
948 | { | |
70b97a7f | 949 | struct rq *rq; |
1da177e4 | 950 | |
3a5c359a AK |
951 | for (;;) { |
952 | local_irq_save(*flags); | |
953 | rq = task_rq(p); | |
954 | spin_lock(&rq->lock); | |
955 | if (likely(rq == task_rq(p))) | |
956 | return rq; | |
1da177e4 | 957 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 958 | } |
1da177e4 LT |
959 | } |
960 | ||
a9957449 | 961 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
962 | __releases(rq->lock) |
963 | { | |
964 | spin_unlock(&rq->lock); | |
965 | } | |
966 | ||
70b97a7f | 967 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
968 | __releases(rq->lock) |
969 | { | |
970 | spin_unlock_irqrestore(&rq->lock, *flags); | |
971 | } | |
972 | ||
1da177e4 | 973 | /* |
cc2a73b5 | 974 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 975 | */ |
a9957449 | 976 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
977 | __acquires(rq->lock) |
978 | { | |
70b97a7f | 979 | struct rq *rq; |
1da177e4 LT |
980 | |
981 | local_irq_disable(); | |
982 | rq = this_rq(); | |
983 | spin_lock(&rq->lock); | |
984 | ||
985 | return rq; | |
986 | } | |
987 | ||
8f4d37ec PZ |
988 | #ifdef CONFIG_SCHED_HRTICK |
989 | /* | |
990 | * Use HR-timers to deliver accurate preemption points. | |
991 | * | |
992 | * Its all a bit involved since we cannot program an hrt while holding the | |
993 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
994 | * reschedule event. | |
995 | * | |
996 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
997 | * rq->lock. | |
998 | */ | |
8f4d37ec PZ |
999 | |
1000 | /* | |
1001 | * Use hrtick when: | |
1002 | * - enabled by features | |
1003 | * - hrtimer is actually high res | |
1004 | */ | |
1005 | static inline int hrtick_enabled(struct rq *rq) | |
1006 | { | |
1007 | if (!sched_feat(HRTICK)) | |
1008 | return 0; | |
ba42059f | 1009 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1010 | return 0; |
8f4d37ec PZ |
1011 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1012 | } | |
1013 | ||
8f4d37ec PZ |
1014 | static void hrtick_clear(struct rq *rq) |
1015 | { | |
1016 | if (hrtimer_active(&rq->hrtick_timer)) | |
1017 | hrtimer_cancel(&rq->hrtick_timer); | |
1018 | } | |
1019 | ||
8f4d37ec PZ |
1020 | /* |
1021 | * High-resolution timer tick. | |
1022 | * Runs from hardirq context with interrupts disabled. | |
1023 | */ | |
1024 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1025 | { | |
1026 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1027 | ||
1028 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1029 | ||
1030 | spin_lock(&rq->lock); | |
3e51f33f | 1031 | update_rq_clock(rq); |
8f4d37ec PZ |
1032 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1033 | spin_unlock(&rq->lock); | |
1034 | ||
1035 | return HRTIMER_NORESTART; | |
1036 | } | |
1037 | ||
95e904c7 | 1038 | #ifdef CONFIG_SMP |
31656519 PZ |
1039 | /* |
1040 | * called from hardirq (IPI) context | |
1041 | */ | |
1042 | static void __hrtick_start(void *arg) | |
b328ca18 | 1043 | { |
31656519 | 1044 | struct rq *rq = arg; |
b328ca18 | 1045 | |
31656519 PZ |
1046 | spin_lock(&rq->lock); |
1047 | hrtimer_restart(&rq->hrtick_timer); | |
1048 | rq->hrtick_csd_pending = 0; | |
1049 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1050 | } |
1051 | ||
31656519 PZ |
1052 | /* |
1053 | * Called to set the hrtick timer state. | |
1054 | * | |
1055 | * called with rq->lock held and irqs disabled | |
1056 | */ | |
1057 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1058 | { |
31656519 PZ |
1059 | struct hrtimer *timer = &rq->hrtick_timer; |
1060 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1061 | |
31656519 PZ |
1062 | timer->expires = time; |
1063 | ||
1064 | if (rq == this_rq()) { | |
1065 | hrtimer_restart(timer); | |
1066 | } else if (!rq->hrtick_csd_pending) { | |
1067 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1068 | rq->hrtick_csd_pending = 1; | |
1069 | } | |
b328ca18 PZ |
1070 | } |
1071 | ||
1072 | static int | |
1073 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1074 | { | |
1075 | int cpu = (int)(long)hcpu; | |
1076 | ||
1077 | switch (action) { | |
1078 | case CPU_UP_CANCELED: | |
1079 | case CPU_UP_CANCELED_FROZEN: | |
1080 | case CPU_DOWN_PREPARE: | |
1081 | case CPU_DOWN_PREPARE_FROZEN: | |
1082 | case CPU_DEAD: | |
1083 | case CPU_DEAD_FROZEN: | |
31656519 | 1084 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1085 | return NOTIFY_OK; |
1086 | } | |
1087 | ||
1088 | return NOTIFY_DONE; | |
1089 | } | |
1090 | ||
1091 | static void init_hrtick(void) | |
1092 | { | |
1093 | hotcpu_notifier(hotplug_hrtick, 0); | |
1094 | } | |
31656519 PZ |
1095 | #else |
1096 | /* | |
1097 | * Called to set the hrtick timer state. | |
1098 | * | |
1099 | * called with rq->lock held and irqs disabled | |
1100 | */ | |
1101 | static void hrtick_start(struct rq *rq, u64 delay) | |
1102 | { | |
1103 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1104 | } | |
b328ca18 | 1105 | |
31656519 | 1106 | static void init_hrtick(void) |
8f4d37ec | 1107 | { |
8f4d37ec | 1108 | } |
31656519 | 1109 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1110 | |
31656519 | 1111 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1112 | { |
31656519 PZ |
1113 | #ifdef CONFIG_SMP |
1114 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1115 | |
31656519 PZ |
1116 | rq->hrtick_csd.flags = 0; |
1117 | rq->hrtick_csd.func = __hrtick_start; | |
1118 | rq->hrtick_csd.info = rq; | |
1119 | #endif | |
8f4d37ec | 1120 | |
31656519 PZ |
1121 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1122 | rq->hrtick_timer.function = hrtick; | |
1123 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
8f4d37ec PZ |
1124 | } |
1125 | #else | |
1126 | static inline void hrtick_clear(struct rq *rq) | |
1127 | { | |
1128 | } | |
1129 | ||
8f4d37ec PZ |
1130 | static inline void init_rq_hrtick(struct rq *rq) |
1131 | { | |
1132 | } | |
1133 | ||
b328ca18 PZ |
1134 | static inline void init_hrtick(void) |
1135 | { | |
1136 | } | |
8f4d37ec PZ |
1137 | #endif |
1138 | ||
c24d20db IM |
1139 | /* |
1140 | * resched_task - mark a task 'to be rescheduled now'. | |
1141 | * | |
1142 | * On UP this means the setting of the need_resched flag, on SMP it | |
1143 | * might also involve a cross-CPU call to trigger the scheduler on | |
1144 | * the target CPU. | |
1145 | */ | |
1146 | #ifdef CONFIG_SMP | |
1147 | ||
1148 | #ifndef tsk_is_polling | |
1149 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1150 | #endif | |
1151 | ||
31656519 | 1152 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1153 | { |
1154 | int cpu; | |
1155 | ||
1156 | assert_spin_locked(&task_rq(p)->lock); | |
1157 | ||
31656519 | 1158 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1159 | return; |
1160 | ||
31656519 | 1161 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1162 | |
1163 | cpu = task_cpu(p); | |
1164 | if (cpu == smp_processor_id()) | |
1165 | return; | |
1166 | ||
1167 | /* NEED_RESCHED must be visible before we test polling */ | |
1168 | smp_mb(); | |
1169 | if (!tsk_is_polling(p)) | |
1170 | smp_send_reschedule(cpu); | |
1171 | } | |
1172 | ||
1173 | static void resched_cpu(int cpu) | |
1174 | { | |
1175 | struct rq *rq = cpu_rq(cpu); | |
1176 | unsigned long flags; | |
1177 | ||
1178 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1179 | return; | |
1180 | resched_task(cpu_curr(cpu)); | |
1181 | spin_unlock_irqrestore(&rq->lock, flags); | |
1182 | } | |
06d8308c TG |
1183 | |
1184 | #ifdef CONFIG_NO_HZ | |
1185 | /* | |
1186 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1187 | * idle CPU then this timer might expire before the next timer event | |
1188 | * which is scheduled to wake up that CPU. In case of a completely | |
1189 | * idle system the next event might even be infinite time into the | |
1190 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1191 | * leaves the inner idle loop so the newly added timer is taken into | |
1192 | * account when the CPU goes back to idle and evaluates the timer | |
1193 | * wheel for the next timer event. | |
1194 | */ | |
1195 | void wake_up_idle_cpu(int cpu) | |
1196 | { | |
1197 | struct rq *rq = cpu_rq(cpu); | |
1198 | ||
1199 | if (cpu == smp_processor_id()) | |
1200 | return; | |
1201 | ||
1202 | /* | |
1203 | * This is safe, as this function is called with the timer | |
1204 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1205 | * to idle and has not yet set rq->curr to idle then it will | |
1206 | * be serialized on the timer wheel base lock and take the new | |
1207 | * timer into account automatically. | |
1208 | */ | |
1209 | if (rq->curr != rq->idle) | |
1210 | return; | |
1211 | ||
1212 | /* | |
1213 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1214 | * lockless. The worst case is that the other CPU runs the | |
1215 | * idle task through an additional NOOP schedule() | |
1216 | */ | |
1217 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1218 | ||
1219 | /* NEED_RESCHED must be visible before we test polling */ | |
1220 | smp_mb(); | |
1221 | if (!tsk_is_polling(rq->idle)) | |
1222 | smp_send_reschedule(cpu); | |
1223 | } | |
6d6bc0ad | 1224 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1225 | |
6d6bc0ad | 1226 | #else /* !CONFIG_SMP */ |
31656519 | 1227 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1228 | { |
1229 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1230 | set_tsk_need_resched(p); |
c24d20db | 1231 | } |
6d6bc0ad | 1232 | #endif /* CONFIG_SMP */ |
c24d20db | 1233 | |
45bf76df IM |
1234 | #if BITS_PER_LONG == 32 |
1235 | # define WMULT_CONST (~0UL) | |
1236 | #else | |
1237 | # define WMULT_CONST (1UL << 32) | |
1238 | #endif | |
1239 | ||
1240 | #define WMULT_SHIFT 32 | |
1241 | ||
194081eb IM |
1242 | /* |
1243 | * Shift right and round: | |
1244 | */ | |
cf2ab469 | 1245 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1246 | |
a7be37ac PZ |
1247 | /* |
1248 | * delta *= weight / lw | |
1249 | */ | |
cb1c4fc9 | 1250 | static unsigned long |
45bf76df IM |
1251 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1252 | struct load_weight *lw) | |
1253 | { | |
1254 | u64 tmp; | |
1255 | ||
7a232e03 LJ |
1256 | if (!lw->inv_weight) { |
1257 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1258 | lw->inv_weight = 1; | |
1259 | else | |
1260 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1261 | / (lw->weight+1); | |
1262 | } | |
45bf76df IM |
1263 | |
1264 | tmp = (u64)delta_exec * weight; | |
1265 | /* | |
1266 | * Check whether we'd overflow the 64-bit multiplication: | |
1267 | */ | |
194081eb | 1268 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1269 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1270 | WMULT_SHIFT/2); |
1271 | else | |
cf2ab469 | 1272 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1273 | |
ecf691da | 1274 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1275 | } |
1276 | ||
1091985b | 1277 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1278 | { |
1279 | lw->weight += inc; | |
e89996ae | 1280 | lw->inv_weight = 0; |
45bf76df IM |
1281 | } |
1282 | ||
1091985b | 1283 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1284 | { |
1285 | lw->weight -= dec; | |
e89996ae | 1286 | lw->inv_weight = 0; |
45bf76df IM |
1287 | } |
1288 | ||
2dd73a4f PW |
1289 | /* |
1290 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1291 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1292 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1293 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1294 | * scaled version of the new time slice allocation that they receive on time |
1295 | * slice expiry etc. | |
1296 | */ | |
1297 | ||
dd41f596 IM |
1298 | #define WEIGHT_IDLEPRIO 2 |
1299 | #define WMULT_IDLEPRIO (1 << 31) | |
1300 | ||
1301 | /* | |
1302 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1303 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1304 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1305 | * that remained on nice 0. | |
1306 | * | |
1307 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1308 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1309 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1310 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1311 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1312 | */ |
1313 | static const int prio_to_weight[40] = { | |
254753dc IM |
1314 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1315 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1316 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1317 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1318 | /* 0 */ 1024, 820, 655, 526, 423, | |
1319 | /* 5 */ 335, 272, 215, 172, 137, | |
1320 | /* 10 */ 110, 87, 70, 56, 45, | |
1321 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1322 | }; |
1323 | ||
5714d2de IM |
1324 | /* |
1325 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1326 | * | |
1327 | * In cases where the weight does not change often, we can use the | |
1328 | * precalculated inverse to speed up arithmetics by turning divisions | |
1329 | * into multiplications: | |
1330 | */ | |
dd41f596 | 1331 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1332 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1333 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1334 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1335 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1336 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1337 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1338 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1339 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1340 | }; |
2dd73a4f | 1341 | |
dd41f596 IM |
1342 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1343 | ||
1344 | /* | |
1345 | * runqueue iterator, to support SMP load-balancing between different | |
1346 | * scheduling classes, without having to expose their internal data | |
1347 | * structures to the load-balancing proper: | |
1348 | */ | |
1349 | struct rq_iterator { | |
1350 | void *arg; | |
1351 | struct task_struct *(*start)(void *); | |
1352 | struct task_struct *(*next)(void *); | |
1353 | }; | |
1354 | ||
e1d1484f PW |
1355 | #ifdef CONFIG_SMP |
1356 | static unsigned long | |
1357 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1358 | unsigned long max_load_move, struct sched_domain *sd, | |
1359 | enum cpu_idle_type idle, int *all_pinned, | |
1360 | int *this_best_prio, struct rq_iterator *iterator); | |
1361 | ||
1362 | static int | |
1363 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1364 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1365 | struct rq_iterator *iterator); | |
e1d1484f | 1366 | #endif |
dd41f596 | 1367 | |
d842de87 SV |
1368 | #ifdef CONFIG_CGROUP_CPUACCT |
1369 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1370 | #else | |
1371 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1372 | #endif | |
1373 | ||
18d95a28 PZ |
1374 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1375 | { | |
1376 | update_load_add(&rq->load, load); | |
1377 | } | |
1378 | ||
1379 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1380 | { | |
1381 | update_load_sub(&rq->load, load); | |
1382 | } | |
1383 | ||
e7693a36 GH |
1384 | #ifdef CONFIG_SMP |
1385 | static unsigned long source_load(int cpu, int type); | |
1386 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1387 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1388 | |
a8a51d5e PZ |
1389 | static unsigned long cpu_avg_load_per_task(int cpu) |
1390 | { | |
1391 | struct rq *rq = cpu_rq(cpu); | |
1392 | ||
1393 | if (rq->nr_running) | |
1394 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1395 | ||
1396 | return rq->avg_load_per_task; | |
1397 | } | |
18d95a28 PZ |
1398 | |
1399 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1400 | |
c8cba857 | 1401 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1402 | |
1403 | /* | |
1404 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1405 | * leaving it for the final time. | |
1406 | */ | |
c8cba857 PZ |
1407 | static void |
1408 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1409 | { |
1410 | struct task_group *parent, *child; | |
1411 | ||
1412 | rcu_read_lock(); | |
1413 | parent = &root_task_group; | |
1414 | down: | |
b6a86c74 | 1415 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1416 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1417 | parent = child; | |
1418 | goto down; | |
1419 | ||
1420 | up: | |
1421 | continue; | |
1422 | } | |
b6a86c74 | 1423 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1424 | |
1425 | child = parent; | |
1426 | parent = parent->parent; | |
1427 | if (parent) | |
1428 | goto up; | |
1429 | rcu_read_unlock(); | |
1430 | } | |
1431 | ||
c09595f6 PZ |
1432 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1433 | ||
1434 | /* | |
1435 | * Calculate and set the cpu's group shares. | |
1436 | */ | |
1437 | static void | |
b6a86c74 | 1438 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1439 | unsigned long sd_shares, unsigned long sd_rq_weight) |
18d95a28 | 1440 | { |
c09595f6 PZ |
1441 | int boost = 0; |
1442 | unsigned long shares; | |
1443 | unsigned long rq_weight; | |
1444 | ||
c8cba857 | 1445 | if (!tg->se[cpu]) |
c09595f6 PZ |
1446 | return; |
1447 | ||
c8cba857 | 1448 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1449 | |
1450 | /* | |
1451 | * If there are currently no tasks on the cpu pretend there is one of | |
1452 | * average load so that when a new task gets to run here it will not | |
1453 | * get delayed by group starvation. | |
1454 | */ | |
1455 | if (!rq_weight) { | |
1456 | boost = 1; | |
1457 | rq_weight = NICE_0_LOAD; | |
1458 | } | |
1459 | ||
c8cba857 PZ |
1460 | if (unlikely(rq_weight > sd_rq_weight)) |
1461 | rq_weight = sd_rq_weight; | |
1462 | ||
c09595f6 PZ |
1463 | /* |
1464 | * \Sum shares * rq_weight | |
1465 | * shares = ----------------------- | |
1466 | * \Sum rq_weight | |
1467 | * | |
1468 | */ | |
c8cba857 | 1469 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1470 | |
1471 | /* | |
1472 | * record the actual number of shares, not the boosted amount. | |
1473 | */ | |
c8cba857 | 1474 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
f1d239f7 | 1475 | tg->cfs_rq[cpu]->rq_weight = rq_weight; |
c09595f6 PZ |
1476 | |
1477 | if (shares < MIN_SHARES) | |
1478 | shares = MIN_SHARES; | |
1479 | else if (shares > MAX_SHARES) | |
1480 | shares = MAX_SHARES; | |
1481 | ||
c8cba857 | 1482 | __set_se_shares(tg->se[cpu], shares); |
18d95a28 | 1483 | } |
c09595f6 PZ |
1484 | |
1485 | /* | |
c8cba857 PZ |
1486 | * Re-compute the task group their per cpu shares over the given domain. |
1487 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1488 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1489 | */ |
1490 | static void | |
c8cba857 | 1491 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1492 | { |
c8cba857 PZ |
1493 | unsigned long rq_weight = 0; |
1494 | unsigned long shares = 0; | |
1495 | int i; | |
c09595f6 | 1496 | |
c8cba857 PZ |
1497 | for_each_cpu_mask(i, sd->span) { |
1498 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1499 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1500 | } |
c09595f6 | 1501 | |
c8cba857 PZ |
1502 | if ((!shares && rq_weight) || shares > tg->shares) |
1503 | shares = tg->shares; | |
1504 | ||
1505 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1506 | shares = tg->shares; | |
c09595f6 | 1507 | |
cd80917e PZ |
1508 | if (!rq_weight) |
1509 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1510 | ||
c09595f6 PZ |
1511 | for_each_cpu_mask(i, sd->span) { |
1512 | struct rq *rq = cpu_rq(i); | |
1513 | unsigned long flags; | |
1514 | ||
1515 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1516 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1517 | spin_unlock_irqrestore(&rq->lock, flags); |
1518 | } | |
c09595f6 PZ |
1519 | } |
1520 | ||
1521 | /* | |
c8cba857 PZ |
1522 | * Compute the cpu's hierarchical load factor for each task group. |
1523 | * This needs to be done in a top-down fashion because the load of a child | |
1524 | * group is a fraction of its parents load. | |
c09595f6 | 1525 | */ |
b6a86c74 | 1526 | static void |
c8cba857 | 1527 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1528 | { |
c8cba857 | 1529 | unsigned long load; |
c09595f6 | 1530 | |
c8cba857 PZ |
1531 | if (!tg->parent) { |
1532 | load = cpu_rq(cpu)->load.weight; | |
1533 | } else { | |
1534 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1535 | load *= tg->cfs_rq[cpu]->shares; | |
1536 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1537 | } | |
c09595f6 | 1538 | |
c8cba857 | 1539 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1540 | } |
1541 | ||
c8cba857 PZ |
1542 | static void |
1543 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1544 | { |
c09595f6 PZ |
1545 | } |
1546 | ||
c8cba857 | 1547 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1548 | { |
2398f2c6 PZ |
1549 | u64 now = cpu_clock(raw_smp_processor_id()); |
1550 | s64 elapsed = now - sd->last_update; | |
1551 | ||
1552 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1553 | sd->last_update = now; | |
1554 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); | |
1555 | } | |
4d8d595d PZ |
1556 | } |
1557 | ||
3e5459b4 PZ |
1558 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1559 | { | |
1560 | spin_unlock(&rq->lock); | |
1561 | update_shares(sd); | |
1562 | spin_lock(&rq->lock); | |
1563 | } | |
1564 | ||
c8cba857 | 1565 | static void update_h_load(int cpu) |
c09595f6 | 1566 | { |
c8cba857 | 1567 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1568 | } |
1569 | ||
c09595f6 PZ |
1570 | #else |
1571 | ||
c8cba857 | 1572 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1573 | { |
1574 | } | |
1575 | ||
3e5459b4 PZ |
1576 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1577 | { | |
1578 | } | |
1579 | ||
18d95a28 PZ |
1580 | #endif |
1581 | ||
18d95a28 PZ |
1582 | #endif |
1583 | ||
30432094 | 1584 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1585 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1586 | { | |
30432094 | 1587 | #ifdef CONFIG_SMP |
34e83e85 IM |
1588 | cfs_rq->shares = shares; |
1589 | #endif | |
1590 | } | |
30432094 | 1591 | #endif |
e7693a36 | 1592 | |
dd41f596 | 1593 | #include "sched_stats.h" |
dd41f596 | 1594 | #include "sched_idletask.c" |
5522d5d5 IM |
1595 | #include "sched_fair.c" |
1596 | #include "sched_rt.c" | |
dd41f596 IM |
1597 | #ifdef CONFIG_SCHED_DEBUG |
1598 | # include "sched_debug.c" | |
1599 | #endif | |
1600 | ||
1601 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1602 | #define for_each_class(class) \ |
1603 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1604 | |
c09595f6 | 1605 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1606 | { |
1607 | rq->nr_running++; | |
9c217245 IM |
1608 | } |
1609 | ||
c09595f6 | 1610 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1611 | { |
1612 | rq->nr_running--; | |
9c217245 IM |
1613 | } |
1614 | ||
45bf76df IM |
1615 | static void set_load_weight(struct task_struct *p) |
1616 | { | |
1617 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1618 | p->se.load.weight = prio_to_weight[0] * 2; |
1619 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1620 | return; | |
1621 | } | |
45bf76df | 1622 | |
dd41f596 IM |
1623 | /* |
1624 | * SCHED_IDLE tasks get minimal weight: | |
1625 | */ | |
1626 | if (p->policy == SCHED_IDLE) { | |
1627 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1628 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1629 | return; | |
1630 | } | |
71f8bd46 | 1631 | |
dd41f596 IM |
1632 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1633 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1634 | } |
1635 | ||
2087a1ad GH |
1636 | static void update_avg(u64 *avg, u64 sample) |
1637 | { | |
1638 | s64 diff = sample - *avg; | |
1639 | *avg += diff >> 3; | |
1640 | } | |
1641 | ||
8159f87e | 1642 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1643 | { |
dd41f596 | 1644 | sched_info_queued(p); |
fd390f6a | 1645 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1646 | p->se.on_rq = 1; |
71f8bd46 IM |
1647 | } |
1648 | ||
69be72c1 | 1649 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1650 | { |
2087a1ad GH |
1651 | if (sleep && p->se.last_wakeup) { |
1652 | update_avg(&p->se.avg_overlap, | |
1653 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1654 | p->se.last_wakeup = 0; | |
1655 | } | |
1656 | ||
46ac22ba | 1657 | sched_info_dequeued(p); |
f02231e5 | 1658 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1659 | p->se.on_rq = 0; |
71f8bd46 IM |
1660 | } |
1661 | ||
14531189 | 1662 | /* |
dd41f596 | 1663 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1664 | */ |
14531189 IM |
1665 | static inline int __normal_prio(struct task_struct *p) |
1666 | { | |
dd41f596 | 1667 | return p->static_prio; |
14531189 IM |
1668 | } |
1669 | ||
b29739f9 IM |
1670 | /* |
1671 | * Calculate the expected normal priority: i.e. priority | |
1672 | * without taking RT-inheritance into account. Might be | |
1673 | * boosted by interactivity modifiers. Changes upon fork, | |
1674 | * setprio syscalls, and whenever the interactivity | |
1675 | * estimator recalculates. | |
1676 | */ | |
36c8b586 | 1677 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1678 | { |
1679 | int prio; | |
1680 | ||
e05606d3 | 1681 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1682 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1683 | else | |
1684 | prio = __normal_prio(p); | |
1685 | return prio; | |
1686 | } | |
1687 | ||
1688 | /* | |
1689 | * Calculate the current priority, i.e. the priority | |
1690 | * taken into account by the scheduler. This value might | |
1691 | * be boosted by RT tasks, or might be boosted by | |
1692 | * interactivity modifiers. Will be RT if the task got | |
1693 | * RT-boosted. If not then it returns p->normal_prio. | |
1694 | */ | |
36c8b586 | 1695 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1696 | { |
1697 | p->normal_prio = normal_prio(p); | |
1698 | /* | |
1699 | * If we are RT tasks or we were boosted to RT priority, | |
1700 | * keep the priority unchanged. Otherwise, update priority | |
1701 | * to the normal priority: | |
1702 | */ | |
1703 | if (!rt_prio(p->prio)) | |
1704 | return p->normal_prio; | |
1705 | return p->prio; | |
1706 | } | |
1707 | ||
1da177e4 | 1708 | /* |
dd41f596 | 1709 | * activate_task - move a task to the runqueue. |
1da177e4 | 1710 | */ |
dd41f596 | 1711 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1712 | { |
d9514f6c | 1713 | if (task_contributes_to_load(p)) |
dd41f596 | 1714 | rq->nr_uninterruptible--; |
1da177e4 | 1715 | |
8159f87e | 1716 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1717 | inc_nr_running(rq); |
1da177e4 LT |
1718 | } |
1719 | ||
1da177e4 LT |
1720 | /* |
1721 | * deactivate_task - remove a task from the runqueue. | |
1722 | */ | |
2e1cb74a | 1723 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1724 | { |
d9514f6c | 1725 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1726 | rq->nr_uninterruptible++; |
1727 | ||
69be72c1 | 1728 | dequeue_task(rq, p, sleep); |
c09595f6 | 1729 | dec_nr_running(rq); |
1da177e4 LT |
1730 | } |
1731 | ||
1da177e4 LT |
1732 | /** |
1733 | * task_curr - is this task currently executing on a CPU? | |
1734 | * @p: the task in question. | |
1735 | */ | |
36c8b586 | 1736 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1737 | { |
1738 | return cpu_curr(task_cpu(p)) == p; | |
1739 | } | |
1740 | ||
dd41f596 IM |
1741 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1742 | { | |
6f505b16 | 1743 | set_task_rq(p, cpu); |
dd41f596 | 1744 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1745 | /* |
1746 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1747 | * successfuly executed on another CPU. We must ensure that updates of | |
1748 | * per-task data have been completed by this moment. | |
1749 | */ | |
1750 | smp_wmb(); | |
dd41f596 | 1751 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1752 | #endif |
2dd73a4f PW |
1753 | } |
1754 | ||
cb469845 SR |
1755 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1756 | const struct sched_class *prev_class, | |
1757 | int oldprio, int running) | |
1758 | { | |
1759 | if (prev_class != p->sched_class) { | |
1760 | if (prev_class->switched_from) | |
1761 | prev_class->switched_from(rq, p, running); | |
1762 | p->sched_class->switched_to(rq, p, running); | |
1763 | } else | |
1764 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1765 | } | |
1766 | ||
1da177e4 | 1767 | #ifdef CONFIG_SMP |
c65cc870 | 1768 | |
e958b360 TG |
1769 | /* Used instead of source_load when we know the type == 0 */ |
1770 | static unsigned long weighted_cpuload(const int cpu) | |
1771 | { | |
1772 | return cpu_rq(cpu)->load.weight; | |
1773 | } | |
1774 | ||
cc367732 IM |
1775 | /* |
1776 | * Is this task likely cache-hot: | |
1777 | */ | |
e7693a36 | 1778 | static int |
cc367732 IM |
1779 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1780 | { | |
1781 | s64 delta; | |
1782 | ||
f540a608 IM |
1783 | /* |
1784 | * Buddy candidates are cache hot: | |
1785 | */ | |
d25ce4cd | 1786 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1787 | return 1; |
1788 | ||
cc367732 IM |
1789 | if (p->sched_class != &fair_sched_class) |
1790 | return 0; | |
1791 | ||
6bc1665b IM |
1792 | if (sysctl_sched_migration_cost == -1) |
1793 | return 1; | |
1794 | if (sysctl_sched_migration_cost == 0) | |
1795 | return 0; | |
1796 | ||
cc367732 IM |
1797 | delta = now - p->se.exec_start; |
1798 | ||
1799 | return delta < (s64)sysctl_sched_migration_cost; | |
1800 | } | |
1801 | ||
1802 | ||
dd41f596 | 1803 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1804 | { |
dd41f596 IM |
1805 | int old_cpu = task_cpu(p); |
1806 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1807 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1808 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1809 | u64 clock_offset; |
dd41f596 IM |
1810 | |
1811 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1812 | |
1813 | #ifdef CONFIG_SCHEDSTATS | |
1814 | if (p->se.wait_start) | |
1815 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1816 | if (p->se.sleep_start) |
1817 | p->se.sleep_start -= clock_offset; | |
1818 | if (p->se.block_start) | |
1819 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1820 | if (old_cpu != new_cpu) { |
1821 | schedstat_inc(p, se.nr_migrations); | |
1822 | if (task_hot(p, old_rq->clock, NULL)) | |
1823 | schedstat_inc(p, se.nr_forced2_migrations); | |
1824 | } | |
6cfb0d5d | 1825 | #endif |
2830cf8c SV |
1826 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1827 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1828 | |
1829 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1830 | } |
1831 | ||
70b97a7f | 1832 | struct migration_req { |
1da177e4 | 1833 | struct list_head list; |
1da177e4 | 1834 | |
36c8b586 | 1835 | struct task_struct *task; |
1da177e4 LT |
1836 | int dest_cpu; |
1837 | ||
1da177e4 | 1838 | struct completion done; |
70b97a7f | 1839 | }; |
1da177e4 LT |
1840 | |
1841 | /* | |
1842 | * The task's runqueue lock must be held. | |
1843 | * Returns true if you have to wait for migration thread. | |
1844 | */ | |
36c8b586 | 1845 | static int |
70b97a7f | 1846 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1847 | { |
70b97a7f | 1848 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1849 | |
1850 | /* | |
1851 | * If the task is not on a runqueue (and not running), then | |
1852 | * it is sufficient to simply update the task's cpu field. | |
1853 | */ | |
dd41f596 | 1854 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1855 | set_task_cpu(p, dest_cpu); |
1856 | return 0; | |
1857 | } | |
1858 | ||
1859 | init_completion(&req->done); | |
1da177e4 LT |
1860 | req->task = p; |
1861 | req->dest_cpu = dest_cpu; | |
1862 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1863 | |
1da177e4 LT |
1864 | return 1; |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * wait_task_inactive - wait for a thread to unschedule. | |
1869 | * | |
1870 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1871 | * else this function might spin for a *long* time. This function can't | |
1872 | * be called with interrupts off, or it may introduce deadlock with | |
1873 | * smp_call_function() if an IPI is sent by the same process we are | |
1874 | * waiting to become inactive. | |
1875 | */ | |
36c8b586 | 1876 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1877 | { |
1878 | unsigned long flags; | |
dd41f596 | 1879 | int running, on_rq; |
70b97a7f | 1880 | struct rq *rq; |
1da177e4 | 1881 | |
3a5c359a AK |
1882 | for (;;) { |
1883 | /* | |
1884 | * We do the initial early heuristics without holding | |
1885 | * any task-queue locks at all. We'll only try to get | |
1886 | * the runqueue lock when things look like they will | |
1887 | * work out! | |
1888 | */ | |
1889 | rq = task_rq(p); | |
fa490cfd | 1890 | |
3a5c359a AK |
1891 | /* |
1892 | * If the task is actively running on another CPU | |
1893 | * still, just relax and busy-wait without holding | |
1894 | * any locks. | |
1895 | * | |
1896 | * NOTE! Since we don't hold any locks, it's not | |
1897 | * even sure that "rq" stays as the right runqueue! | |
1898 | * But we don't care, since "task_running()" will | |
1899 | * return false if the runqueue has changed and p | |
1900 | * is actually now running somewhere else! | |
1901 | */ | |
1902 | while (task_running(rq, p)) | |
1903 | cpu_relax(); | |
fa490cfd | 1904 | |
3a5c359a AK |
1905 | /* |
1906 | * Ok, time to look more closely! We need the rq | |
1907 | * lock now, to be *sure*. If we're wrong, we'll | |
1908 | * just go back and repeat. | |
1909 | */ | |
1910 | rq = task_rq_lock(p, &flags); | |
1911 | running = task_running(rq, p); | |
1912 | on_rq = p->se.on_rq; | |
1913 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1914 | |
3a5c359a AK |
1915 | /* |
1916 | * Was it really running after all now that we | |
1917 | * checked with the proper locks actually held? | |
1918 | * | |
1919 | * Oops. Go back and try again.. | |
1920 | */ | |
1921 | if (unlikely(running)) { | |
1922 | cpu_relax(); | |
1923 | continue; | |
1924 | } | |
fa490cfd | 1925 | |
3a5c359a AK |
1926 | /* |
1927 | * It's not enough that it's not actively running, | |
1928 | * it must be off the runqueue _entirely_, and not | |
1929 | * preempted! | |
1930 | * | |
1931 | * So if it wa still runnable (but just not actively | |
1932 | * running right now), it's preempted, and we should | |
1933 | * yield - it could be a while. | |
1934 | */ | |
1935 | if (unlikely(on_rq)) { | |
1936 | schedule_timeout_uninterruptible(1); | |
1937 | continue; | |
1938 | } | |
fa490cfd | 1939 | |
3a5c359a AK |
1940 | /* |
1941 | * Ahh, all good. It wasn't running, and it wasn't | |
1942 | * runnable, which means that it will never become | |
1943 | * running in the future either. We're all done! | |
1944 | */ | |
1945 | break; | |
1946 | } | |
1da177e4 LT |
1947 | } |
1948 | ||
1949 | /*** | |
1950 | * kick_process - kick a running thread to enter/exit the kernel | |
1951 | * @p: the to-be-kicked thread | |
1952 | * | |
1953 | * Cause a process which is running on another CPU to enter | |
1954 | * kernel-mode, without any delay. (to get signals handled.) | |
1955 | * | |
1956 | * NOTE: this function doesnt have to take the runqueue lock, | |
1957 | * because all it wants to ensure is that the remote task enters | |
1958 | * the kernel. If the IPI races and the task has been migrated | |
1959 | * to another CPU then no harm is done and the purpose has been | |
1960 | * achieved as well. | |
1961 | */ | |
36c8b586 | 1962 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1963 | { |
1964 | int cpu; | |
1965 | ||
1966 | preempt_disable(); | |
1967 | cpu = task_cpu(p); | |
1968 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1969 | smp_send_reschedule(cpu); | |
1970 | preempt_enable(); | |
1971 | } | |
1972 | ||
1973 | /* | |
2dd73a4f PW |
1974 | * Return a low guess at the load of a migration-source cpu weighted |
1975 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1976 | * |
1977 | * We want to under-estimate the load of migration sources, to | |
1978 | * balance conservatively. | |
1979 | */ | |
a9957449 | 1980 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1981 | { |
70b97a7f | 1982 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1983 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1984 | |
93b75217 | 1985 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 1986 | return total; |
b910472d | 1987 | |
dd41f596 | 1988 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1989 | } |
1990 | ||
1991 | /* | |
2dd73a4f PW |
1992 | * Return a high guess at the load of a migration-target cpu weighted |
1993 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1994 | */ |
a9957449 | 1995 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1996 | { |
70b97a7f | 1997 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1998 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1999 | |
93b75217 | 2000 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2001 | return total; |
3b0bd9bc | 2002 | |
dd41f596 | 2003 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2004 | } |
2005 | ||
147cbb4b NP |
2006 | /* |
2007 | * find_idlest_group finds and returns the least busy CPU group within the | |
2008 | * domain. | |
2009 | */ | |
2010 | static struct sched_group * | |
2011 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2012 | { | |
2013 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2014 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2015 | int load_idx = sd->forkexec_idx; | |
2016 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2017 | ||
2018 | do { | |
2019 | unsigned long load, avg_load; | |
2020 | int local_group; | |
2021 | int i; | |
2022 | ||
da5a5522 BD |
2023 | /* Skip over this group if it has no CPUs allowed */ |
2024 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2025 | continue; |
da5a5522 | 2026 | |
147cbb4b | 2027 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2028 | |
2029 | /* Tally up the load of all CPUs in the group */ | |
2030 | avg_load = 0; | |
2031 | ||
363ab6f1 | 2032 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2033 | /* Bias balancing toward cpus of our domain */ |
2034 | if (local_group) | |
2035 | load = source_load(i, load_idx); | |
2036 | else | |
2037 | load = target_load(i, load_idx); | |
2038 | ||
2039 | avg_load += load; | |
2040 | } | |
2041 | ||
2042 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2043 | avg_load = sg_div_cpu_power(group, |
2044 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2045 | |
2046 | if (local_group) { | |
2047 | this_load = avg_load; | |
2048 | this = group; | |
2049 | } else if (avg_load < min_load) { | |
2050 | min_load = avg_load; | |
2051 | idlest = group; | |
2052 | } | |
3a5c359a | 2053 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2054 | |
2055 | if (!idlest || 100*this_load < imbalance*min_load) | |
2056 | return NULL; | |
2057 | return idlest; | |
2058 | } | |
2059 | ||
2060 | /* | |
0feaece9 | 2061 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2062 | */ |
95cdf3b7 | 2063 | static int |
7c16ec58 MT |
2064 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2065 | cpumask_t *tmp) | |
147cbb4b NP |
2066 | { |
2067 | unsigned long load, min_load = ULONG_MAX; | |
2068 | int idlest = -1; | |
2069 | int i; | |
2070 | ||
da5a5522 | 2071 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2072 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2073 | |
363ab6f1 | 2074 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2075 | load = weighted_cpuload(i); |
147cbb4b NP |
2076 | |
2077 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2078 | min_load = load; | |
2079 | idlest = i; | |
2080 | } | |
2081 | } | |
2082 | ||
2083 | return idlest; | |
2084 | } | |
2085 | ||
476d139c NP |
2086 | /* |
2087 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2088 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2089 | * SD_BALANCE_EXEC. | |
2090 | * | |
2091 | * Balance, ie. select the least loaded group. | |
2092 | * | |
2093 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2094 | * | |
2095 | * preempt must be disabled. | |
2096 | */ | |
2097 | static int sched_balance_self(int cpu, int flag) | |
2098 | { | |
2099 | struct task_struct *t = current; | |
2100 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2101 | |
c96d145e | 2102 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2103 | /* |
2104 | * If power savings logic is enabled for a domain, stop there. | |
2105 | */ | |
5c45bf27 SS |
2106 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2107 | break; | |
476d139c NP |
2108 | if (tmp->flags & flag) |
2109 | sd = tmp; | |
c96d145e | 2110 | } |
476d139c | 2111 | |
039a1c41 PZ |
2112 | if (sd) |
2113 | update_shares(sd); | |
2114 | ||
476d139c | 2115 | while (sd) { |
7c16ec58 | 2116 | cpumask_t span, tmpmask; |
476d139c | 2117 | struct sched_group *group; |
1a848870 SS |
2118 | int new_cpu, weight; |
2119 | ||
2120 | if (!(sd->flags & flag)) { | |
2121 | sd = sd->child; | |
2122 | continue; | |
2123 | } | |
476d139c NP |
2124 | |
2125 | span = sd->span; | |
2126 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2127 | if (!group) { |
2128 | sd = sd->child; | |
2129 | continue; | |
2130 | } | |
476d139c | 2131 | |
7c16ec58 | 2132 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2133 | if (new_cpu == -1 || new_cpu == cpu) { |
2134 | /* Now try balancing at a lower domain level of cpu */ | |
2135 | sd = sd->child; | |
2136 | continue; | |
2137 | } | |
476d139c | 2138 | |
1a848870 | 2139 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2140 | cpu = new_cpu; |
476d139c NP |
2141 | sd = NULL; |
2142 | weight = cpus_weight(span); | |
2143 | for_each_domain(cpu, tmp) { | |
2144 | if (weight <= cpus_weight(tmp->span)) | |
2145 | break; | |
2146 | if (tmp->flags & flag) | |
2147 | sd = tmp; | |
2148 | } | |
2149 | /* while loop will break here if sd == NULL */ | |
2150 | } | |
2151 | ||
2152 | return cpu; | |
2153 | } | |
2154 | ||
2155 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2156 | |
1da177e4 LT |
2157 | /*** |
2158 | * try_to_wake_up - wake up a thread | |
2159 | * @p: the to-be-woken-up thread | |
2160 | * @state: the mask of task states that can be woken | |
2161 | * @sync: do a synchronous wakeup? | |
2162 | * | |
2163 | * Put it on the run-queue if it's not already there. The "current" | |
2164 | * thread is always on the run-queue (except when the actual | |
2165 | * re-schedule is in progress), and as such you're allowed to do | |
2166 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2167 | * runnable without the overhead of this. | |
2168 | * | |
2169 | * returns failure only if the task is already active. | |
2170 | */ | |
36c8b586 | 2171 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2172 | { |
cc367732 | 2173 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2174 | unsigned long flags; |
2175 | long old_state; | |
70b97a7f | 2176 | struct rq *rq; |
1da177e4 | 2177 | |
b85d0667 IM |
2178 | if (!sched_feat(SYNC_WAKEUPS)) |
2179 | sync = 0; | |
2180 | ||
2398f2c6 PZ |
2181 | #ifdef CONFIG_SMP |
2182 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2183 | struct sched_domain *sd; | |
2184 | ||
2185 | this_cpu = raw_smp_processor_id(); | |
2186 | cpu = task_cpu(p); | |
2187 | ||
2188 | for_each_domain(this_cpu, sd) { | |
2189 | if (cpu_isset(cpu, sd->span)) { | |
2190 | update_shares(sd); | |
2191 | break; | |
2192 | } | |
2193 | } | |
2194 | } | |
2195 | #endif | |
2196 | ||
04e2f174 | 2197 | smp_wmb(); |
1da177e4 LT |
2198 | rq = task_rq_lock(p, &flags); |
2199 | old_state = p->state; | |
2200 | if (!(old_state & state)) | |
2201 | goto out; | |
2202 | ||
dd41f596 | 2203 | if (p->se.on_rq) |
1da177e4 LT |
2204 | goto out_running; |
2205 | ||
2206 | cpu = task_cpu(p); | |
cc367732 | 2207 | orig_cpu = cpu; |
1da177e4 LT |
2208 | this_cpu = smp_processor_id(); |
2209 | ||
2210 | #ifdef CONFIG_SMP | |
2211 | if (unlikely(task_running(rq, p))) | |
2212 | goto out_activate; | |
2213 | ||
5d2f5a61 DA |
2214 | cpu = p->sched_class->select_task_rq(p, sync); |
2215 | if (cpu != orig_cpu) { | |
2216 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2217 | task_rq_unlock(rq, &flags); |
2218 | /* might preempt at this point */ | |
2219 | rq = task_rq_lock(p, &flags); | |
2220 | old_state = p->state; | |
2221 | if (!(old_state & state)) | |
2222 | goto out; | |
dd41f596 | 2223 | if (p->se.on_rq) |
1da177e4 LT |
2224 | goto out_running; |
2225 | ||
2226 | this_cpu = smp_processor_id(); | |
2227 | cpu = task_cpu(p); | |
2228 | } | |
2229 | ||
e7693a36 GH |
2230 | #ifdef CONFIG_SCHEDSTATS |
2231 | schedstat_inc(rq, ttwu_count); | |
2232 | if (cpu == this_cpu) | |
2233 | schedstat_inc(rq, ttwu_local); | |
2234 | else { | |
2235 | struct sched_domain *sd; | |
2236 | for_each_domain(this_cpu, sd) { | |
2237 | if (cpu_isset(cpu, sd->span)) { | |
2238 | schedstat_inc(sd, ttwu_wake_remote); | |
2239 | break; | |
2240 | } | |
2241 | } | |
2242 | } | |
6d6bc0ad | 2243 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2244 | |
1da177e4 LT |
2245 | out_activate: |
2246 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2247 | schedstat_inc(p, se.nr_wakeups); |
2248 | if (sync) | |
2249 | schedstat_inc(p, se.nr_wakeups_sync); | |
2250 | if (orig_cpu != cpu) | |
2251 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2252 | if (cpu == this_cpu) | |
2253 | schedstat_inc(p, se.nr_wakeups_local); | |
2254 | else | |
2255 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2256 | update_rq_clock(rq); |
dd41f596 | 2257 | activate_task(rq, p, 1); |
1da177e4 LT |
2258 | success = 1; |
2259 | ||
2260 | out_running: | |
5b82a1b0 MD |
2261 | trace_mark(kernel_sched_wakeup, |
2262 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2263 | p->pid, p->state, rq, p, rq->curr); | |
4ae7d5ce IM |
2264 | check_preempt_curr(rq, p); |
2265 | ||
1da177e4 | 2266 | p->state = TASK_RUNNING; |
9a897c5a SR |
2267 | #ifdef CONFIG_SMP |
2268 | if (p->sched_class->task_wake_up) | |
2269 | p->sched_class->task_wake_up(rq, p); | |
2270 | #endif | |
1da177e4 | 2271 | out: |
2087a1ad GH |
2272 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2273 | ||
1da177e4 LT |
2274 | task_rq_unlock(rq, &flags); |
2275 | ||
2276 | return success; | |
2277 | } | |
2278 | ||
7ad5b3a5 | 2279 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2280 | { |
d9514f6c | 2281 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2282 | } |
1da177e4 LT |
2283 | EXPORT_SYMBOL(wake_up_process); |
2284 | ||
7ad5b3a5 | 2285 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2286 | { |
2287 | return try_to_wake_up(p, state, 0); | |
2288 | } | |
2289 | ||
1da177e4 LT |
2290 | /* |
2291 | * Perform scheduler related setup for a newly forked process p. | |
2292 | * p is forked by current. | |
dd41f596 IM |
2293 | * |
2294 | * __sched_fork() is basic setup used by init_idle() too: | |
2295 | */ | |
2296 | static void __sched_fork(struct task_struct *p) | |
2297 | { | |
dd41f596 IM |
2298 | p->se.exec_start = 0; |
2299 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2300 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2301 | p->se.last_wakeup = 0; |
2302 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2303 | |
2304 | #ifdef CONFIG_SCHEDSTATS | |
2305 | p->se.wait_start = 0; | |
dd41f596 IM |
2306 | p->se.sum_sleep_runtime = 0; |
2307 | p->se.sleep_start = 0; | |
dd41f596 IM |
2308 | p->se.block_start = 0; |
2309 | p->se.sleep_max = 0; | |
2310 | p->se.block_max = 0; | |
2311 | p->se.exec_max = 0; | |
eba1ed4b | 2312 | p->se.slice_max = 0; |
dd41f596 | 2313 | p->se.wait_max = 0; |
6cfb0d5d | 2314 | #endif |
476d139c | 2315 | |
fa717060 | 2316 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2317 | p->se.on_rq = 0; |
4a55bd5e | 2318 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2319 | |
e107be36 AK |
2320 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2321 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2322 | #endif | |
2323 | ||
1da177e4 LT |
2324 | /* |
2325 | * We mark the process as running here, but have not actually | |
2326 | * inserted it onto the runqueue yet. This guarantees that | |
2327 | * nobody will actually run it, and a signal or other external | |
2328 | * event cannot wake it up and insert it on the runqueue either. | |
2329 | */ | |
2330 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2331 | } |
2332 | ||
2333 | /* | |
2334 | * fork()/clone()-time setup: | |
2335 | */ | |
2336 | void sched_fork(struct task_struct *p, int clone_flags) | |
2337 | { | |
2338 | int cpu = get_cpu(); | |
2339 | ||
2340 | __sched_fork(p); | |
2341 | ||
2342 | #ifdef CONFIG_SMP | |
2343 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2344 | #endif | |
02e4bac2 | 2345 | set_task_cpu(p, cpu); |
b29739f9 IM |
2346 | |
2347 | /* | |
2348 | * Make sure we do not leak PI boosting priority to the child: | |
2349 | */ | |
2350 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2351 | if (!rt_prio(p->prio)) |
2352 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2353 | |
52f17b6c | 2354 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2355 | if (likely(sched_info_on())) |
52f17b6c | 2356 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2357 | #endif |
d6077cb8 | 2358 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2359 | p->oncpu = 0; |
2360 | #endif | |
1da177e4 | 2361 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2362 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2363 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2364 | #endif |
476d139c | 2365 | put_cpu(); |
1da177e4 LT |
2366 | } |
2367 | ||
2368 | /* | |
2369 | * wake_up_new_task - wake up a newly created task for the first time. | |
2370 | * | |
2371 | * This function will do some initial scheduler statistics housekeeping | |
2372 | * that must be done for every newly created context, then puts the task | |
2373 | * on the runqueue and wakes it. | |
2374 | */ | |
7ad5b3a5 | 2375 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2376 | { |
2377 | unsigned long flags; | |
dd41f596 | 2378 | struct rq *rq; |
1da177e4 LT |
2379 | |
2380 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2381 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2382 | update_rq_clock(rq); |
1da177e4 LT |
2383 | |
2384 | p->prio = effective_prio(p); | |
2385 | ||
b9dca1e0 | 2386 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2387 | activate_task(rq, p, 0); |
1da177e4 | 2388 | } else { |
1da177e4 | 2389 | /* |
dd41f596 IM |
2390 | * Let the scheduling class do new task startup |
2391 | * management (if any): | |
1da177e4 | 2392 | */ |
ee0827d8 | 2393 | p->sched_class->task_new(rq, p); |
c09595f6 | 2394 | inc_nr_running(rq); |
1da177e4 | 2395 | } |
5b82a1b0 MD |
2396 | trace_mark(kernel_sched_wakeup_new, |
2397 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2398 | p->pid, p->state, rq, p, rq->curr); | |
dd41f596 | 2399 | check_preempt_curr(rq, p); |
9a897c5a SR |
2400 | #ifdef CONFIG_SMP |
2401 | if (p->sched_class->task_wake_up) | |
2402 | p->sched_class->task_wake_up(rq, p); | |
2403 | #endif | |
dd41f596 | 2404 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2405 | } |
2406 | ||
e107be36 AK |
2407 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2408 | ||
2409 | /** | |
421cee29 RD |
2410 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2411 | * @notifier: notifier struct to register | |
e107be36 AK |
2412 | */ |
2413 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2414 | { | |
2415 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2416 | } | |
2417 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2418 | ||
2419 | /** | |
2420 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2421 | * @notifier: notifier struct to unregister |
e107be36 AK |
2422 | * |
2423 | * This is safe to call from within a preemption notifier. | |
2424 | */ | |
2425 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2426 | { | |
2427 | hlist_del(¬ifier->link); | |
2428 | } | |
2429 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2430 | ||
2431 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2432 | { | |
2433 | struct preempt_notifier *notifier; | |
2434 | struct hlist_node *node; | |
2435 | ||
2436 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2437 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2438 | } | |
2439 | ||
2440 | static void | |
2441 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2442 | struct task_struct *next) | |
2443 | { | |
2444 | struct preempt_notifier *notifier; | |
2445 | struct hlist_node *node; | |
2446 | ||
2447 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2448 | notifier->ops->sched_out(notifier, next); | |
2449 | } | |
2450 | ||
6d6bc0ad | 2451 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2452 | |
2453 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2454 | { | |
2455 | } | |
2456 | ||
2457 | static void | |
2458 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2459 | struct task_struct *next) | |
2460 | { | |
2461 | } | |
2462 | ||
6d6bc0ad | 2463 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2464 | |
4866cde0 NP |
2465 | /** |
2466 | * prepare_task_switch - prepare to switch tasks | |
2467 | * @rq: the runqueue preparing to switch | |
421cee29 | 2468 | * @prev: the current task that is being switched out |
4866cde0 NP |
2469 | * @next: the task we are going to switch to. |
2470 | * | |
2471 | * This is called with the rq lock held and interrupts off. It must | |
2472 | * be paired with a subsequent finish_task_switch after the context | |
2473 | * switch. | |
2474 | * | |
2475 | * prepare_task_switch sets up locking and calls architecture specific | |
2476 | * hooks. | |
2477 | */ | |
e107be36 AK |
2478 | static inline void |
2479 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2480 | struct task_struct *next) | |
4866cde0 | 2481 | { |
e107be36 | 2482 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2483 | prepare_lock_switch(rq, next); |
2484 | prepare_arch_switch(next); | |
2485 | } | |
2486 | ||
1da177e4 LT |
2487 | /** |
2488 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2489 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2490 | * @prev: the thread we just switched away from. |
2491 | * | |
4866cde0 NP |
2492 | * finish_task_switch must be called after the context switch, paired |
2493 | * with a prepare_task_switch call before the context switch. | |
2494 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2495 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2496 | * |
2497 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2498 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2499 | * with the lock held can cause deadlocks; see schedule() for |
2500 | * details.) | |
2501 | */ | |
a9957449 | 2502 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2503 | __releases(rq->lock) |
2504 | { | |
1da177e4 | 2505 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2506 | long prev_state; |
1da177e4 LT |
2507 | |
2508 | rq->prev_mm = NULL; | |
2509 | ||
2510 | /* | |
2511 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2512 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2513 | * schedule one last time. The schedule call will never return, and |
2514 | * the scheduled task must drop that reference. | |
c394cc9f | 2515 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2516 | * still held, otherwise prev could be scheduled on another cpu, die |
2517 | * there before we look at prev->state, and then the reference would | |
2518 | * be dropped twice. | |
2519 | * Manfred Spraul <manfred@colorfullife.com> | |
2520 | */ | |
55a101f8 | 2521 | prev_state = prev->state; |
4866cde0 NP |
2522 | finish_arch_switch(prev); |
2523 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2524 | #ifdef CONFIG_SMP |
2525 | if (current->sched_class->post_schedule) | |
2526 | current->sched_class->post_schedule(rq); | |
2527 | #endif | |
e8fa1362 | 2528 | |
e107be36 | 2529 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2530 | if (mm) |
2531 | mmdrop(mm); | |
c394cc9f | 2532 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2533 | /* |
2534 | * Remove function-return probe instances associated with this | |
2535 | * task and put them back on the free list. | |
9761eea8 | 2536 | */ |
c6fd91f0 | 2537 | kprobe_flush_task(prev); |
1da177e4 | 2538 | put_task_struct(prev); |
c6fd91f0 | 2539 | } |
1da177e4 LT |
2540 | } |
2541 | ||
2542 | /** | |
2543 | * schedule_tail - first thing a freshly forked thread must call. | |
2544 | * @prev: the thread we just switched away from. | |
2545 | */ | |
36c8b586 | 2546 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2547 | __releases(rq->lock) |
2548 | { | |
70b97a7f IM |
2549 | struct rq *rq = this_rq(); |
2550 | ||
4866cde0 NP |
2551 | finish_task_switch(rq, prev); |
2552 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2553 | /* In this case, finish_task_switch does not reenable preemption */ | |
2554 | preempt_enable(); | |
2555 | #endif | |
1da177e4 | 2556 | if (current->set_child_tid) |
b488893a | 2557 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2558 | } |
2559 | ||
2560 | /* | |
2561 | * context_switch - switch to the new MM and the new | |
2562 | * thread's register state. | |
2563 | */ | |
dd41f596 | 2564 | static inline void |
70b97a7f | 2565 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2566 | struct task_struct *next) |
1da177e4 | 2567 | { |
dd41f596 | 2568 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2569 | |
e107be36 | 2570 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2571 | trace_mark(kernel_sched_schedule, |
2572 | "prev_pid %d next_pid %d prev_state %ld " | |
2573 | "## rq %p prev %p next %p", | |
2574 | prev->pid, next->pid, prev->state, | |
2575 | rq, prev, next); | |
dd41f596 IM |
2576 | mm = next->mm; |
2577 | oldmm = prev->active_mm; | |
9226d125 ZA |
2578 | /* |
2579 | * For paravirt, this is coupled with an exit in switch_to to | |
2580 | * combine the page table reload and the switch backend into | |
2581 | * one hypercall. | |
2582 | */ | |
2583 | arch_enter_lazy_cpu_mode(); | |
2584 | ||
dd41f596 | 2585 | if (unlikely(!mm)) { |
1da177e4 LT |
2586 | next->active_mm = oldmm; |
2587 | atomic_inc(&oldmm->mm_count); | |
2588 | enter_lazy_tlb(oldmm, next); | |
2589 | } else | |
2590 | switch_mm(oldmm, mm, next); | |
2591 | ||
dd41f596 | 2592 | if (unlikely(!prev->mm)) { |
1da177e4 | 2593 | prev->active_mm = NULL; |
1da177e4 LT |
2594 | rq->prev_mm = oldmm; |
2595 | } | |
3a5f5e48 IM |
2596 | /* |
2597 | * Since the runqueue lock will be released by the next | |
2598 | * task (which is an invalid locking op but in the case | |
2599 | * of the scheduler it's an obvious special-case), so we | |
2600 | * do an early lockdep release here: | |
2601 | */ | |
2602 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2603 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2604 | #endif |
1da177e4 LT |
2605 | |
2606 | /* Here we just switch the register state and the stack. */ | |
2607 | switch_to(prev, next, prev); | |
2608 | ||
dd41f596 IM |
2609 | barrier(); |
2610 | /* | |
2611 | * this_rq must be evaluated again because prev may have moved | |
2612 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2613 | * frame will be invalid. | |
2614 | */ | |
2615 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2616 | } |
2617 | ||
2618 | /* | |
2619 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2620 | * | |
2621 | * externally visible scheduler statistics: current number of runnable | |
2622 | * threads, current number of uninterruptible-sleeping threads, total | |
2623 | * number of context switches performed since bootup. | |
2624 | */ | |
2625 | unsigned long nr_running(void) | |
2626 | { | |
2627 | unsigned long i, sum = 0; | |
2628 | ||
2629 | for_each_online_cpu(i) | |
2630 | sum += cpu_rq(i)->nr_running; | |
2631 | ||
2632 | return sum; | |
2633 | } | |
2634 | ||
2635 | unsigned long nr_uninterruptible(void) | |
2636 | { | |
2637 | unsigned long i, sum = 0; | |
2638 | ||
0a945022 | 2639 | for_each_possible_cpu(i) |
1da177e4 LT |
2640 | sum += cpu_rq(i)->nr_uninterruptible; |
2641 | ||
2642 | /* | |
2643 | * Since we read the counters lockless, it might be slightly | |
2644 | * inaccurate. Do not allow it to go below zero though: | |
2645 | */ | |
2646 | if (unlikely((long)sum < 0)) | |
2647 | sum = 0; | |
2648 | ||
2649 | return sum; | |
2650 | } | |
2651 | ||
2652 | unsigned long long nr_context_switches(void) | |
2653 | { | |
cc94abfc SR |
2654 | int i; |
2655 | unsigned long long sum = 0; | |
1da177e4 | 2656 | |
0a945022 | 2657 | for_each_possible_cpu(i) |
1da177e4 LT |
2658 | sum += cpu_rq(i)->nr_switches; |
2659 | ||
2660 | return sum; | |
2661 | } | |
2662 | ||
2663 | unsigned long nr_iowait(void) | |
2664 | { | |
2665 | unsigned long i, sum = 0; | |
2666 | ||
0a945022 | 2667 | for_each_possible_cpu(i) |
1da177e4 LT |
2668 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2669 | ||
2670 | return sum; | |
2671 | } | |
2672 | ||
db1b1fef JS |
2673 | unsigned long nr_active(void) |
2674 | { | |
2675 | unsigned long i, running = 0, uninterruptible = 0; | |
2676 | ||
2677 | for_each_online_cpu(i) { | |
2678 | running += cpu_rq(i)->nr_running; | |
2679 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2680 | } | |
2681 | ||
2682 | if (unlikely((long)uninterruptible < 0)) | |
2683 | uninterruptible = 0; | |
2684 | ||
2685 | return running + uninterruptible; | |
2686 | } | |
2687 | ||
48f24c4d | 2688 | /* |
dd41f596 IM |
2689 | * Update rq->cpu_load[] statistics. This function is usually called every |
2690 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2691 | */ |
dd41f596 | 2692 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2693 | { |
495eca49 | 2694 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2695 | int i, scale; |
2696 | ||
2697 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2698 | |
2699 | /* Update our load: */ | |
2700 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2701 | unsigned long old_load, new_load; | |
2702 | ||
2703 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2704 | ||
2705 | old_load = this_rq->cpu_load[i]; | |
2706 | new_load = this_load; | |
a25707f3 IM |
2707 | /* |
2708 | * Round up the averaging division if load is increasing. This | |
2709 | * prevents us from getting stuck on 9 if the load is 10, for | |
2710 | * example. | |
2711 | */ | |
2712 | if (new_load > old_load) | |
2713 | new_load += scale-1; | |
dd41f596 IM |
2714 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2715 | } | |
48f24c4d IM |
2716 | } |
2717 | ||
dd41f596 IM |
2718 | #ifdef CONFIG_SMP |
2719 | ||
1da177e4 LT |
2720 | /* |
2721 | * double_rq_lock - safely lock two runqueues | |
2722 | * | |
2723 | * Note this does not disable interrupts like task_rq_lock, | |
2724 | * you need to do so manually before calling. | |
2725 | */ | |
70b97a7f | 2726 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2727 | __acquires(rq1->lock) |
2728 | __acquires(rq2->lock) | |
2729 | { | |
054b9108 | 2730 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2731 | if (rq1 == rq2) { |
2732 | spin_lock(&rq1->lock); | |
2733 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2734 | } else { | |
c96d145e | 2735 | if (rq1 < rq2) { |
1da177e4 LT |
2736 | spin_lock(&rq1->lock); |
2737 | spin_lock(&rq2->lock); | |
2738 | } else { | |
2739 | spin_lock(&rq2->lock); | |
2740 | spin_lock(&rq1->lock); | |
2741 | } | |
2742 | } | |
6e82a3be IM |
2743 | update_rq_clock(rq1); |
2744 | update_rq_clock(rq2); | |
1da177e4 LT |
2745 | } |
2746 | ||
2747 | /* | |
2748 | * double_rq_unlock - safely unlock two runqueues | |
2749 | * | |
2750 | * Note this does not restore interrupts like task_rq_unlock, | |
2751 | * you need to do so manually after calling. | |
2752 | */ | |
70b97a7f | 2753 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2754 | __releases(rq1->lock) |
2755 | __releases(rq2->lock) | |
2756 | { | |
2757 | spin_unlock(&rq1->lock); | |
2758 | if (rq1 != rq2) | |
2759 | spin_unlock(&rq2->lock); | |
2760 | else | |
2761 | __release(rq2->lock); | |
2762 | } | |
2763 | ||
2764 | /* | |
2765 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2766 | */ | |
e8fa1362 | 2767 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2768 | __releases(this_rq->lock) |
2769 | __acquires(busiest->lock) | |
2770 | __acquires(this_rq->lock) | |
2771 | { | |
e8fa1362 SR |
2772 | int ret = 0; |
2773 | ||
054b9108 KK |
2774 | if (unlikely(!irqs_disabled())) { |
2775 | /* printk() doesn't work good under rq->lock */ | |
2776 | spin_unlock(&this_rq->lock); | |
2777 | BUG_ON(1); | |
2778 | } | |
1da177e4 | 2779 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2780 | if (busiest < this_rq) { |
1da177e4 LT |
2781 | spin_unlock(&this_rq->lock); |
2782 | spin_lock(&busiest->lock); | |
2783 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2784 | ret = 1; |
1da177e4 LT |
2785 | } else |
2786 | spin_lock(&busiest->lock); | |
2787 | } | |
e8fa1362 | 2788 | return ret; |
1da177e4 LT |
2789 | } |
2790 | ||
1da177e4 LT |
2791 | /* |
2792 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2793 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2794 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2795 | * the cpu_allowed mask is restored. |
2796 | */ | |
36c8b586 | 2797 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2798 | { |
70b97a7f | 2799 | struct migration_req req; |
1da177e4 | 2800 | unsigned long flags; |
70b97a7f | 2801 | struct rq *rq; |
1da177e4 LT |
2802 | |
2803 | rq = task_rq_lock(p, &flags); | |
2804 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2805 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2806 | goto out; |
2807 | ||
2808 | /* force the process onto the specified CPU */ | |
2809 | if (migrate_task(p, dest_cpu, &req)) { | |
2810 | /* Need to wait for migration thread (might exit: take ref). */ | |
2811 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2812 | |
1da177e4 LT |
2813 | get_task_struct(mt); |
2814 | task_rq_unlock(rq, &flags); | |
2815 | wake_up_process(mt); | |
2816 | put_task_struct(mt); | |
2817 | wait_for_completion(&req.done); | |
36c8b586 | 2818 | |
1da177e4 LT |
2819 | return; |
2820 | } | |
2821 | out: | |
2822 | task_rq_unlock(rq, &flags); | |
2823 | } | |
2824 | ||
2825 | /* | |
476d139c NP |
2826 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2827 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2828 | */ |
2829 | void sched_exec(void) | |
2830 | { | |
1da177e4 | 2831 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2832 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2833 | put_cpu(); |
476d139c NP |
2834 | if (new_cpu != this_cpu) |
2835 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2836 | } |
2837 | ||
2838 | /* | |
2839 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2840 | * Both runqueues must be locked. | |
2841 | */ | |
dd41f596 IM |
2842 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2843 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2844 | { |
2e1cb74a | 2845 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2846 | set_task_cpu(p, this_cpu); |
dd41f596 | 2847 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2848 | /* |
2849 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2850 | * to be always true for them. | |
2851 | */ | |
dd41f596 | 2852 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2853 | } |
2854 | ||
2855 | /* | |
2856 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2857 | */ | |
858119e1 | 2858 | static |
70b97a7f | 2859 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2860 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2861 | int *all_pinned) |
1da177e4 LT |
2862 | { |
2863 | /* | |
2864 | * We do not migrate tasks that are: | |
2865 | * 1) running (obviously), or | |
2866 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2867 | * 3) are cache-hot on their current CPU. | |
2868 | */ | |
cc367732 IM |
2869 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2870 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2871 | return 0; |
cc367732 | 2872 | } |
81026794 NP |
2873 | *all_pinned = 0; |
2874 | ||
cc367732 IM |
2875 | if (task_running(rq, p)) { |
2876 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2877 | return 0; |
cc367732 | 2878 | } |
1da177e4 | 2879 | |
da84d961 IM |
2880 | /* |
2881 | * Aggressive migration if: | |
2882 | * 1) task is cache cold, or | |
2883 | * 2) too many balance attempts have failed. | |
2884 | */ | |
2885 | ||
6bc1665b IM |
2886 | if (!task_hot(p, rq->clock, sd) || |
2887 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2888 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2889 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2890 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2891 | schedstat_inc(p, se.nr_forced_migrations); |
2892 | } | |
da84d961 IM |
2893 | #endif |
2894 | return 1; | |
2895 | } | |
2896 | ||
cc367732 IM |
2897 | if (task_hot(p, rq->clock, sd)) { |
2898 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2899 | return 0; |
cc367732 | 2900 | } |
1da177e4 LT |
2901 | return 1; |
2902 | } | |
2903 | ||
e1d1484f PW |
2904 | static unsigned long |
2905 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2906 | unsigned long max_load_move, struct sched_domain *sd, | |
2907 | enum cpu_idle_type idle, int *all_pinned, | |
2908 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2909 | { |
051c6764 | 2910 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2911 | struct task_struct *p; |
2912 | long rem_load_move = max_load_move; | |
1da177e4 | 2913 | |
e1d1484f | 2914 | if (max_load_move == 0) |
1da177e4 LT |
2915 | goto out; |
2916 | ||
81026794 NP |
2917 | pinned = 1; |
2918 | ||
1da177e4 | 2919 | /* |
dd41f596 | 2920 | * Start the load-balancing iterator: |
1da177e4 | 2921 | */ |
dd41f596 IM |
2922 | p = iterator->start(iterator->arg); |
2923 | next: | |
b82d9fdd | 2924 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2925 | goto out; |
051c6764 PZ |
2926 | |
2927 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2928 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2929 | p = iterator->next(iterator->arg); |
2930 | goto next; | |
1da177e4 LT |
2931 | } |
2932 | ||
dd41f596 | 2933 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2934 | pulled++; |
dd41f596 | 2935 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2936 | |
2dd73a4f | 2937 | /* |
b82d9fdd | 2938 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2939 | */ |
e1d1484f | 2940 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2941 | if (p->prio < *this_best_prio) |
2942 | *this_best_prio = p->prio; | |
dd41f596 IM |
2943 | p = iterator->next(iterator->arg); |
2944 | goto next; | |
1da177e4 LT |
2945 | } |
2946 | out: | |
2947 | /* | |
e1d1484f | 2948 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2949 | * so we can safely collect pull_task() stats here rather than |
2950 | * inside pull_task(). | |
2951 | */ | |
2952 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2953 | |
2954 | if (all_pinned) | |
2955 | *all_pinned = pinned; | |
e1d1484f PW |
2956 | |
2957 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2958 | } |
2959 | ||
dd41f596 | 2960 | /* |
43010659 PW |
2961 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2962 | * this_rq, as part of a balancing operation within domain "sd". | |
2963 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2964 | * |
2965 | * Called with both runqueues locked. | |
2966 | */ | |
2967 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2968 | unsigned long max_load_move, |
dd41f596 IM |
2969 | struct sched_domain *sd, enum cpu_idle_type idle, |
2970 | int *all_pinned) | |
2971 | { | |
5522d5d5 | 2972 | const struct sched_class *class = sched_class_highest; |
43010659 | 2973 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2974 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2975 | |
2976 | do { | |
43010659 PW |
2977 | total_load_moved += |
2978 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2979 | max_load_move - total_load_moved, |
a4ac01c3 | 2980 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2981 | class = class->next; |
c4acb2c0 GH |
2982 | |
2983 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
2984 | break; | |
2985 | ||
43010659 | 2986 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2987 | |
43010659 PW |
2988 | return total_load_moved > 0; |
2989 | } | |
2990 | ||
e1d1484f PW |
2991 | static int |
2992 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2993 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2994 | struct rq_iterator *iterator) | |
2995 | { | |
2996 | struct task_struct *p = iterator->start(iterator->arg); | |
2997 | int pinned = 0; | |
2998 | ||
2999 | while (p) { | |
3000 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3001 | pull_task(busiest, p, this_rq, this_cpu); | |
3002 | /* | |
3003 | * Right now, this is only the second place pull_task() | |
3004 | * is called, so we can safely collect pull_task() | |
3005 | * stats here rather than inside pull_task(). | |
3006 | */ | |
3007 | schedstat_inc(sd, lb_gained[idle]); | |
3008 | ||
3009 | return 1; | |
3010 | } | |
3011 | p = iterator->next(iterator->arg); | |
3012 | } | |
3013 | ||
3014 | return 0; | |
3015 | } | |
3016 | ||
43010659 PW |
3017 | /* |
3018 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3019 | * part of active balancing operations within "domain". | |
3020 | * Returns 1 if successful and 0 otherwise. | |
3021 | * | |
3022 | * Called with both runqueues locked. | |
3023 | */ | |
3024 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3025 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3026 | { | |
5522d5d5 | 3027 | const struct sched_class *class; |
43010659 PW |
3028 | |
3029 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3030 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3031 | return 1; |
3032 | ||
3033 | return 0; | |
dd41f596 IM |
3034 | } |
3035 | ||
1da177e4 LT |
3036 | /* |
3037 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3038 | * domain. It calculates and returns the amount of weighted load which |
3039 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3040 | */ |
3041 | static struct sched_group * | |
3042 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3043 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3044 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3045 | { |
3046 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3047 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3048 | unsigned long max_pull; |
2dd73a4f PW |
3049 | unsigned long busiest_load_per_task, busiest_nr_running; |
3050 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3051 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3052 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3053 | int power_savings_balance = 1; | |
3054 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3055 | unsigned long min_nr_running = ULONG_MAX; | |
3056 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3057 | #endif | |
1da177e4 LT |
3058 | |
3059 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3060 | busiest_load_per_task = busiest_nr_running = 0; |
3061 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3062 | |
d15bcfdb | 3063 | if (idle == CPU_NOT_IDLE) |
7897986b | 3064 | load_idx = sd->busy_idx; |
d15bcfdb | 3065 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3066 | load_idx = sd->newidle_idx; |
3067 | else | |
3068 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3069 | |
3070 | do { | |
908a7c1b | 3071 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3072 | int local_group; |
3073 | int i; | |
908a7c1b | 3074 | int __group_imb = 0; |
783609c6 | 3075 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3076 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3077 | unsigned long sum_avg_load_per_task; |
3078 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3079 | |
3080 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3081 | ||
783609c6 SS |
3082 | if (local_group) |
3083 | balance_cpu = first_cpu(group->cpumask); | |
3084 | ||
1da177e4 | 3085 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3086 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3087 | sum_avg_load_per_task = avg_load_per_task = 0; |
3088 | ||
908a7c1b KC |
3089 | max_cpu_load = 0; |
3090 | min_cpu_load = ~0UL; | |
1da177e4 | 3091 | |
363ab6f1 | 3092 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3093 | struct rq *rq; |
3094 | ||
3095 | if (!cpu_isset(i, *cpus)) | |
3096 | continue; | |
3097 | ||
3098 | rq = cpu_rq(i); | |
2dd73a4f | 3099 | |
9439aab8 | 3100 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3101 | *sd_idle = 0; |
3102 | ||
1da177e4 | 3103 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3104 | if (local_group) { |
3105 | if (idle_cpu(i) && !first_idle_cpu) { | |
3106 | first_idle_cpu = 1; | |
3107 | balance_cpu = i; | |
3108 | } | |
3109 | ||
a2000572 | 3110 | load = target_load(i, load_idx); |
908a7c1b | 3111 | } else { |
a2000572 | 3112 | load = source_load(i, load_idx); |
908a7c1b KC |
3113 | if (load > max_cpu_load) |
3114 | max_cpu_load = load; | |
3115 | if (min_cpu_load > load) | |
3116 | min_cpu_load = load; | |
3117 | } | |
1da177e4 LT |
3118 | |
3119 | avg_load += load; | |
2dd73a4f | 3120 | sum_nr_running += rq->nr_running; |
dd41f596 | 3121 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3122 | |
3123 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3124 | } |
3125 | ||
783609c6 SS |
3126 | /* |
3127 | * First idle cpu or the first cpu(busiest) in this sched group | |
3128 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3129 | * domains. In the newly idle case, we will allow all the cpu's |
3130 | * to do the newly idle load balance. | |
783609c6 | 3131 | */ |
9439aab8 SS |
3132 | if (idle != CPU_NEWLY_IDLE && local_group && |
3133 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3134 | *balance = 0; |
3135 | goto ret; | |
3136 | } | |
3137 | ||
1da177e4 | 3138 | total_load += avg_load; |
5517d86b | 3139 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3140 | |
3141 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3142 | avg_load = sg_div_cpu_power(group, |
3143 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3144 | |
408ed066 PZ |
3145 | |
3146 | /* | |
3147 | * Consider the group unbalanced when the imbalance is larger | |
3148 | * than the average weight of two tasks. | |
3149 | * | |
3150 | * APZ: with cgroup the avg task weight can vary wildly and | |
3151 | * might not be a suitable number - should we keep a | |
3152 | * normalized nr_running number somewhere that negates | |
3153 | * the hierarchy? | |
3154 | */ | |
3155 | avg_load_per_task = sg_div_cpu_power(group, | |
3156 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3157 | ||
3158 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3159 | __group_imb = 1; |
3160 | ||
5517d86b | 3161 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3162 | |
1da177e4 LT |
3163 | if (local_group) { |
3164 | this_load = avg_load; | |
3165 | this = group; | |
2dd73a4f PW |
3166 | this_nr_running = sum_nr_running; |
3167 | this_load_per_task = sum_weighted_load; | |
3168 | } else if (avg_load > max_load && | |
908a7c1b | 3169 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3170 | max_load = avg_load; |
3171 | busiest = group; | |
2dd73a4f PW |
3172 | busiest_nr_running = sum_nr_running; |
3173 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3174 | group_imb = __group_imb; |
1da177e4 | 3175 | } |
5c45bf27 SS |
3176 | |
3177 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3178 | /* | |
3179 | * Busy processors will not participate in power savings | |
3180 | * balance. | |
3181 | */ | |
dd41f596 IM |
3182 | if (idle == CPU_NOT_IDLE || |
3183 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3184 | goto group_next; | |
5c45bf27 SS |
3185 | |
3186 | /* | |
3187 | * If the local group is idle or completely loaded | |
3188 | * no need to do power savings balance at this domain | |
3189 | */ | |
3190 | if (local_group && (this_nr_running >= group_capacity || | |
3191 | !this_nr_running)) | |
3192 | power_savings_balance = 0; | |
3193 | ||
dd41f596 | 3194 | /* |
5c45bf27 SS |
3195 | * If a group is already running at full capacity or idle, |
3196 | * don't include that group in power savings calculations | |
dd41f596 IM |
3197 | */ |
3198 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3199 | || !sum_nr_running) |
dd41f596 | 3200 | goto group_next; |
5c45bf27 | 3201 | |
dd41f596 | 3202 | /* |
5c45bf27 | 3203 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3204 | * This is the group from where we need to pick up the load |
3205 | * for saving power | |
3206 | */ | |
3207 | if ((sum_nr_running < min_nr_running) || | |
3208 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3209 | first_cpu(group->cpumask) < |
3210 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3211 | group_min = group; |
3212 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3213 | min_load_per_task = sum_weighted_load / |
3214 | sum_nr_running; | |
dd41f596 | 3215 | } |
5c45bf27 | 3216 | |
dd41f596 | 3217 | /* |
5c45bf27 | 3218 | * Calculate the group which is almost near its |
dd41f596 IM |
3219 | * capacity but still has some space to pick up some load |
3220 | * from other group and save more power | |
3221 | */ | |
3222 | if (sum_nr_running <= group_capacity - 1) { | |
3223 | if (sum_nr_running > leader_nr_running || | |
3224 | (sum_nr_running == leader_nr_running && | |
3225 | first_cpu(group->cpumask) > | |
3226 | first_cpu(group_leader->cpumask))) { | |
3227 | group_leader = group; | |
3228 | leader_nr_running = sum_nr_running; | |
3229 | } | |
48f24c4d | 3230 | } |
5c45bf27 SS |
3231 | group_next: |
3232 | #endif | |
1da177e4 LT |
3233 | group = group->next; |
3234 | } while (group != sd->groups); | |
3235 | ||
2dd73a4f | 3236 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3237 | goto out_balanced; |
3238 | ||
3239 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3240 | ||
3241 | if (this_load >= avg_load || | |
3242 | 100*max_load <= sd->imbalance_pct*this_load) | |
3243 | goto out_balanced; | |
3244 | ||
2dd73a4f | 3245 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3246 | if (group_imb) |
3247 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3248 | ||
1da177e4 LT |
3249 | /* |
3250 | * We're trying to get all the cpus to the average_load, so we don't | |
3251 | * want to push ourselves above the average load, nor do we wish to | |
3252 | * reduce the max loaded cpu below the average load, as either of these | |
3253 | * actions would just result in more rebalancing later, and ping-pong | |
3254 | * tasks around. Thus we look for the minimum possible imbalance. | |
3255 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3256 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3257 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3258 | * appear as very large values with unsigned longs. |
3259 | */ | |
2dd73a4f PW |
3260 | if (max_load <= busiest_load_per_task) |
3261 | goto out_balanced; | |
3262 | ||
3263 | /* | |
3264 | * In the presence of smp nice balancing, certain scenarios can have | |
3265 | * max load less than avg load(as we skip the groups at or below | |
3266 | * its cpu_power, while calculating max_load..) | |
3267 | */ | |
3268 | if (max_load < avg_load) { | |
3269 | *imbalance = 0; | |
3270 | goto small_imbalance; | |
3271 | } | |
0c117f1b SS |
3272 | |
3273 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3274 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3275 | |
1da177e4 | 3276 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3277 | *imbalance = min(max_pull * busiest->__cpu_power, |
3278 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3279 | / SCHED_LOAD_SCALE; |
3280 | ||
2dd73a4f PW |
3281 | /* |
3282 | * if *imbalance is less than the average load per runnable task | |
3283 | * there is no gaurantee that any tasks will be moved so we'll have | |
3284 | * a think about bumping its value to force at least one task to be | |
3285 | * moved | |
3286 | */ | |
7fd0d2dd | 3287 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3288 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3289 | unsigned int imbn; |
3290 | ||
3291 | small_imbalance: | |
3292 | pwr_move = pwr_now = 0; | |
3293 | imbn = 2; | |
3294 | if (this_nr_running) { | |
3295 | this_load_per_task /= this_nr_running; | |
3296 | if (busiest_load_per_task > this_load_per_task) | |
3297 | imbn = 1; | |
3298 | } else | |
408ed066 | 3299 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3300 | |
408ed066 | 3301 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3302 | busiest_load_per_task * imbn) { |
2dd73a4f | 3303 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3304 | return busiest; |
3305 | } | |
3306 | ||
3307 | /* | |
3308 | * OK, we don't have enough imbalance to justify moving tasks, | |
3309 | * however we may be able to increase total CPU power used by | |
3310 | * moving them. | |
3311 | */ | |
3312 | ||
5517d86b ED |
3313 | pwr_now += busiest->__cpu_power * |
3314 | min(busiest_load_per_task, max_load); | |
3315 | pwr_now += this->__cpu_power * | |
3316 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3317 | pwr_now /= SCHED_LOAD_SCALE; |
3318 | ||
3319 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3320 | tmp = sg_div_cpu_power(busiest, |
3321 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3322 | if (max_load > tmp) |
5517d86b | 3323 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3324 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3325 | |
3326 | /* Amount of load we'd add */ | |
5517d86b | 3327 | if (max_load * busiest->__cpu_power < |
33859f7f | 3328 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3329 | tmp = sg_div_cpu_power(this, |
3330 | max_load * busiest->__cpu_power); | |
1da177e4 | 3331 | else |
5517d86b ED |
3332 | tmp = sg_div_cpu_power(this, |
3333 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3334 | pwr_move += this->__cpu_power * | |
3335 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3336 | pwr_move /= SCHED_LOAD_SCALE; |
3337 | ||
3338 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3339 | if (pwr_move > pwr_now) |
3340 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3341 | } |
3342 | ||
1da177e4 LT |
3343 | return busiest; |
3344 | ||
3345 | out_balanced: | |
5c45bf27 | 3346 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3347 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3348 | goto ret; |
1da177e4 | 3349 | |
5c45bf27 SS |
3350 | if (this == group_leader && group_leader != group_min) { |
3351 | *imbalance = min_load_per_task; | |
3352 | return group_min; | |
3353 | } | |
5c45bf27 | 3354 | #endif |
783609c6 | 3355 | ret: |
1da177e4 LT |
3356 | *imbalance = 0; |
3357 | return NULL; | |
3358 | } | |
3359 | ||
3360 | /* | |
3361 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3362 | */ | |
70b97a7f | 3363 | static struct rq * |
d15bcfdb | 3364 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3365 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3366 | { |
70b97a7f | 3367 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3368 | unsigned long max_load = 0; |
1da177e4 LT |
3369 | int i; |
3370 | ||
363ab6f1 | 3371 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3372 | unsigned long wl; |
0a2966b4 CL |
3373 | |
3374 | if (!cpu_isset(i, *cpus)) | |
3375 | continue; | |
3376 | ||
48f24c4d | 3377 | rq = cpu_rq(i); |
dd41f596 | 3378 | wl = weighted_cpuload(i); |
2dd73a4f | 3379 | |
dd41f596 | 3380 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3381 | continue; |
1da177e4 | 3382 | |
dd41f596 IM |
3383 | if (wl > max_load) { |
3384 | max_load = wl; | |
48f24c4d | 3385 | busiest = rq; |
1da177e4 LT |
3386 | } |
3387 | } | |
3388 | ||
3389 | return busiest; | |
3390 | } | |
3391 | ||
77391d71 NP |
3392 | /* |
3393 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3394 | * so long as it is large enough. | |
3395 | */ | |
3396 | #define MAX_PINNED_INTERVAL 512 | |
3397 | ||
1da177e4 LT |
3398 | /* |
3399 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3400 | * tasks if there is an imbalance. | |
1da177e4 | 3401 | */ |
70b97a7f | 3402 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3403 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3404 | int *balance, cpumask_t *cpus) |
1da177e4 | 3405 | { |
43010659 | 3406 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3407 | struct sched_group *group; |
1da177e4 | 3408 | unsigned long imbalance; |
70b97a7f | 3409 | struct rq *busiest; |
fe2eea3f | 3410 | unsigned long flags; |
5969fe06 | 3411 | |
7c16ec58 MT |
3412 | cpus_setall(*cpus); |
3413 | ||
89c4710e SS |
3414 | /* |
3415 | * When power savings policy is enabled for the parent domain, idle | |
3416 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3417 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3418 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3419 | */ |
d15bcfdb | 3420 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3421 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3422 | sd_idle = 1; |
1da177e4 | 3423 | |
2d72376b | 3424 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3425 | |
0a2966b4 | 3426 | redo: |
c8cba857 | 3427 | update_shares(sd); |
0a2966b4 | 3428 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3429 | cpus, balance); |
783609c6 | 3430 | |
06066714 | 3431 | if (*balance == 0) |
783609c6 | 3432 | goto out_balanced; |
783609c6 | 3433 | |
1da177e4 LT |
3434 | if (!group) { |
3435 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3436 | goto out_balanced; | |
3437 | } | |
3438 | ||
7c16ec58 | 3439 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3440 | if (!busiest) { |
3441 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3442 | goto out_balanced; | |
3443 | } | |
3444 | ||
db935dbd | 3445 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3446 | |
3447 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3448 | ||
43010659 | 3449 | ld_moved = 0; |
1da177e4 LT |
3450 | if (busiest->nr_running > 1) { |
3451 | /* | |
3452 | * Attempt to move tasks. If find_busiest_group has found | |
3453 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3454 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3455 | * correctly treated as an imbalance. |
3456 | */ | |
fe2eea3f | 3457 | local_irq_save(flags); |
e17224bf | 3458 | double_rq_lock(this_rq, busiest); |
43010659 | 3459 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3460 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3461 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3462 | local_irq_restore(flags); |
81026794 | 3463 | |
46cb4b7c SS |
3464 | /* |
3465 | * some other cpu did the load balance for us. | |
3466 | */ | |
43010659 | 3467 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3468 | resched_cpu(this_cpu); |
3469 | ||
81026794 | 3470 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3471 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3472 | cpu_clear(cpu_of(busiest), *cpus); |
3473 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3474 | goto redo; |
81026794 | 3475 | goto out_balanced; |
0a2966b4 | 3476 | } |
1da177e4 | 3477 | } |
81026794 | 3478 | |
43010659 | 3479 | if (!ld_moved) { |
1da177e4 LT |
3480 | schedstat_inc(sd, lb_failed[idle]); |
3481 | sd->nr_balance_failed++; | |
3482 | ||
3483 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3484 | |
fe2eea3f | 3485 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3486 | |
3487 | /* don't kick the migration_thread, if the curr | |
3488 | * task on busiest cpu can't be moved to this_cpu | |
3489 | */ | |
3490 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3491 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3492 | all_pinned = 1; |
3493 | goto out_one_pinned; | |
3494 | } | |
3495 | ||
1da177e4 LT |
3496 | if (!busiest->active_balance) { |
3497 | busiest->active_balance = 1; | |
3498 | busiest->push_cpu = this_cpu; | |
81026794 | 3499 | active_balance = 1; |
1da177e4 | 3500 | } |
fe2eea3f | 3501 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3502 | if (active_balance) |
1da177e4 LT |
3503 | wake_up_process(busiest->migration_thread); |
3504 | ||
3505 | /* | |
3506 | * We've kicked active balancing, reset the failure | |
3507 | * counter. | |
3508 | */ | |
39507451 | 3509 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3510 | } |
81026794 | 3511 | } else |
1da177e4 LT |
3512 | sd->nr_balance_failed = 0; |
3513 | ||
81026794 | 3514 | if (likely(!active_balance)) { |
1da177e4 LT |
3515 | /* We were unbalanced, so reset the balancing interval */ |
3516 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3517 | } else { |
3518 | /* | |
3519 | * If we've begun active balancing, start to back off. This | |
3520 | * case may not be covered by the all_pinned logic if there | |
3521 | * is only 1 task on the busy runqueue (because we don't call | |
3522 | * move_tasks). | |
3523 | */ | |
3524 | if (sd->balance_interval < sd->max_interval) | |
3525 | sd->balance_interval *= 2; | |
1da177e4 LT |
3526 | } |
3527 | ||
43010659 | 3528 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3529 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3530 | ld_moved = -1; |
3531 | ||
3532 | goto out; | |
1da177e4 LT |
3533 | |
3534 | out_balanced: | |
1da177e4 LT |
3535 | schedstat_inc(sd, lb_balanced[idle]); |
3536 | ||
16cfb1c0 | 3537 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3538 | |
3539 | out_one_pinned: | |
1da177e4 | 3540 | /* tune up the balancing interval */ |
77391d71 NP |
3541 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3542 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3543 | sd->balance_interval *= 2; |
3544 | ||
48f24c4d | 3545 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3546 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3547 | ld_moved = -1; |
3548 | else | |
3549 | ld_moved = 0; | |
3550 | out: | |
c8cba857 PZ |
3551 | if (ld_moved) |
3552 | update_shares(sd); | |
c09595f6 | 3553 | return ld_moved; |
1da177e4 LT |
3554 | } |
3555 | ||
3556 | /* | |
3557 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3558 | * tasks if there is an imbalance. | |
3559 | * | |
d15bcfdb | 3560 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3561 | * this_rq is locked. |
3562 | */ | |
48f24c4d | 3563 | static int |
7c16ec58 MT |
3564 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3565 | cpumask_t *cpus) | |
1da177e4 LT |
3566 | { |
3567 | struct sched_group *group; | |
70b97a7f | 3568 | struct rq *busiest = NULL; |
1da177e4 | 3569 | unsigned long imbalance; |
43010659 | 3570 | int ld_moved = 0; |
5969fe06 | 3571 | int sd_idle = 0; |
969bb4e4 | 3572 | int all_pinned = 0; |
7c16ec58 MT |
3573 | |
3574 | cpus_setall(*cpus); | |
5969fe06 | 3575 | |
89c4710e SS |
3576 | /* |
3577 | * When power savings policy is enabled for the parent domain, idle | |
3578 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3579 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3580 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3581 | */ |
3582 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3583 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3584 | sd_idle = 1; |
1da177e4 | 3585 | |
2d72376b | 3586 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3587 | redo: |
3e5459b4 | 3588 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3589 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3590 | &sd_idle, cpus, NULL); |
1da177e4 | 3591 | if (!group) { |
d15bcfdb | 3592 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3593 | goto out_balanced; |
1da177e4 LT |
3594 | } |
3595 | ||
7c16ec58 | 3596 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3597 | if (!busiest) { |
d15bcfdb | 3598 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3599 | goto out_balanced; |
1da177e4 LT |
3600 | } |
3601 | ||
db935dbd NP |
3602 | BUG_ON(busiest == this_rq); |
3603 | ||
d15bcfdb | 3604 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3605 | |
43010659 | 3606 | ld_moved = 0; |
d6d5cfaf NP |
3607 | if (busiest->nr_running > 1) { |
3608 | /* Attempt to move tasks */ | |
3609 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3610 | /* this_rq->clock is already updated */ |
3611 | update_rq_clock(busiest); | |
43010659 | 3612 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3613 | imbalance, sd, CPU_NEWLY_IDLE, |
3614 | &all_pinned); | |
d6d5cfaf | 3615 | spin_unlock(&busiest->lock); |
0a2966b4 | 3616 | |
969bb4e4 | 3617 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3618 | cpu_clear(cpu_of(busiest), *cpus); |
3619 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3620 | goto redo; |
3621 | } | |
d6d5cfaf NP |
3622 | } |
3623 | ||
43010659 | 3624 | if (!ld_moved) { |
d15bcfdb | 3625 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3626 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3627 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3628 | return -1; |
3629 | } else | |
16cfb1c0 | 3630 | sd->nr_balance_failed = 0; |
1da177e4 | 3631 | |
3e5459b4 | 3632 | update_shares_locked(this_rq, sd); |
43010659 | 3633 | return ld_moved; |
16cfb1c0 NP |
3634 | |
3635 | out_balanced: | |
d15bcfdb | 3636 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3637 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3638 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3639 | return -1; |
16cfb1c0 | 3640 | sd->nr_balance_failed = 0; |
48f24c4d | 3641 | |
16cfb1c0 | 3642 | return 0; |
1da177e4 LT |
3643 | } |
3644 | ||
3645 | /* | |
3646 | * idle_balance is called by schedule() if this_cpu is about to become | |
3647 | * idle. Attempts to pull tasks from other CPUs. | |
3648 | */ | |
70b97a7f | 3649 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3650 | { |
3651 | struct sched_domain *sd; | |
dd41f596 IM |
3652 | int pulled_task = -1; |
3653 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3654 | cpumask_t tmpmask; |
1da177e4 LT |
3655 | |
3656 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3657 | unsigned long interval; |
3658 | ||
3659 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3660 | continue; | |
3661 | ||
3662 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3663 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3664 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3665 | sd, &tmpmask); | |
92c4ca5c CL |
3666 | |
3667 | interval = msecs_to_jiffies(sd->balance_interval); | |
3668 | if (time_after(next_balance, sd->last_balance + interval)) | |
3669 | next_balance = sd->last_balance + interval; | |
3670 | if (pulled_task) | |
3671 | break; | |
1da177e4 | 3672 | } |
dd41f596 | 3673 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3674 | /* |
3675 | * We are going idle. next_balance may be set based on | |
3676 | * a busy processor. So reset next_balance. | |
3677 | */ | |
3678 | this_rq->next_balance = next_balance; | |
dd41f596 | 3679 | } |
1da177e4 LT |
3680 | } |
3681 | ||
3682 | /* | |
3683 | * active_load_balance is run by migration threads. It pushes running tasks | |
3684 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3685 | * running on each physical CPU where possible, and avoids physical / | |
3686 | * logical imbalances. | |
3687 | * | |
3688 | * Called with busiest_rq locked. | |
3689 | */ | |
70b97a7f | 3690 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3691 | { |
39507451 | 3692 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3693 | struct sched_domain *sd; |
3694 | struct rq *target_rq; | |
39507451 | 3695 | |
48f24c4d | 3696 | /* Is there any task to move? */ |
39507451 | 3697 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3698 | return; |
3699 | ||
3700 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3701 | |
3702 | /* | |
39507451 | 3703 | * This condition is "impossible", if it occurs |
41a2d6cf | 3704 | * we need to fix it. Originally reported by |
39507451 | 3705 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3706 | */ |
39507451 | 3707 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3708 | |
39507451 NP |
3709 | /* move a task from busiest_rq to target_rq */ |
3710 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3711 | update_rq_clock(busiest_rq); |
3712 | update_rq_clock(target_rq); | |
39507451 NP |
3713 | |
3714 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3715 | for_each_domain(target_cpu, sd) { |
39507451 | 3716 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3717 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3718 | break; |
c96d145e | 3719 | } |
39507451 | 3720 | |
48f24c4d | 3721 | if (likely(sd)) { |
2d72376b | 3722 | schedstat_inc(sd, alb_count); |
39507451 | 3723 | |
43010659 PW |
3724 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3725 | sd, CPU_IDLE)) | |
48f24c4d IM |
3726 | schedstat_inc(sd, alb_pushed); |
3727 | else | |
3728 | schedstat_inc(sd, alb_failed); | |
3729 | } | |
39507451 | 3730 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3731 | } |
3732 | ||
46cb4b7c SS |
3733 | #ifdef CONFIG_NO_HZ |
3734 | static struct { | |
3735 | atomic_t load_balancer; | |
41a2d6cf | 3736 | cpumask_t cpu_mask; |
46cb4b7c SS |
3737 | } nohz ____cacheline_aligned = { |
3738 | .load_balancer = ATOMIC_INIT(-1), | |
3739 | .cpu_mask = CPU_MASK_NONE, | |
3740 | }; | |
3741 | ||
7835b98b | 3742 | /* |
46cb4b7c SS |
3743 | * This routine will try to nominate the ilb (idle load balancing) |
3744 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3745 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3746 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3747 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3748 | * arrives... | |
3749 | * | |
3750 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3751 | * for idle load balancing. ilb owner will still be part of | |
3752 | * nohz.cpu_mask.. | |
7835b98b | 3753 | * |
46cb4b7c SS |
3754 | * While stopping the tick, this cpu will become the ilb owner if there |
3755 | * is no other owner. And will be the owner till that cpu becomes busy | |
3756 | * or if all cpus in the system stop their ticks at which point | |
3757 | * there is no need for ilb owner. | |
3758 | * | |
3759 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3760 | * next busy scheduler_tick() | |
3761 | */ | |
3762 | int select_nohz_load_balancer(int stop_tick) | |
3763 | { | |
3764 | int cpu = smp_processor_id(); | |
3765 | ||
3766 | if (stop_tick) { | |
3767 | cpu_set(cpu, nohz.cpu_mask); | |
3768 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3769 | ||
3770 | /* | |
3771 | * If we are going offline and still the leader, give up! | |
3772 | */ | |
e761b772 | 3773 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3774 | atomic_read(&nohz.load_balancer) == cpu) { |
3775 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3776 | BUG(); | |
3777 | return 0; | |
3778 | } | |
3779 | ||
3780 | /* time for ilb owner also to sleep */ | |
3781 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3782 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3783 | atomic_set(&nohz.load_balancer, -1); | |
3784 | return 0; | |
3785 | } | |
3786 | ||
3787 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3788 | /* make me the ilb owner */ | |
3789 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3790 | return 1; | |
3791 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3792 | return 1; | |
3793 | } else { | |
3794 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3795 | return 0; | |
3796 | ||
3797 | cpu_clear(cpu, nohz.cpu_mask); | |
3798 | ||
3799 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3800 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3801 | BUG(); | |
3802 | } | |
3803 | return 0; | |
3804 | } | |
3805 | #endif | |
3806 | ||
3807 | static DEFINE_SPINLOCK(balancing); | |
3808 | ||
3809 | /* | |
7835b98b CL |
3810 | * It checks each scheduling domain to see if it is due to be balanced, |
3811 | * and initiates a balancing operation if so. | |
3812 | * | |
3813 | * Balancing parameters are set up in arch_init_sched_domains. | |
3814 | */ | |
a9957449 | 3815 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3816 | { |
46cb4b7c SS |
3817 | int balance = 1; |
3818 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3819 | unsigned long interval; |
3820 | struct sched_domain *sd; | |
46cb4b7c | 3821 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3822 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3823 | int update_next_balance = 0; |
d07355f5 | 3824 | int need_serialize; |
7c16ec58 | 3825 | cpumask_t tmp; |
1da177e4 | 3826 | |
46cb4b7c | 3827 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3828 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3829 | continue; | |
3830 | ||
3831 | interval = sd->balance_interval; | |
d15bcfdb | 3832 | if (idle != CPU_IDLE) |
1da177e4 LT |
3833 | interval *= sd->busy_factor; |
3834 | ||
3835 | /* scale ms to jiffies */ | |
3836 | interval = msecs_to_jiffies(interval); | |
3837 | if (unlikely(!interval)) | |
3838 | interval = 1; | |
dd41f596 IM |
3839 | if (interval > HZ*NR_CPUS/10) |
3840 | interval = HZ*NR_CPUS/10; | |
3841 | ||
d07355f5 | 3842 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3843 | |
d07355f5 | 3844 | if (need_serialize) { |
08c183f3 CL |
3845 | if (!spin_trylock(&balancing)) |
3846 | goto out; | |
3847 | } | |
3848 | ||
c9819f45 | 3849 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3850 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3851 | /* |
3852 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3853 | * longer idle, or one of our SMT siblings is |
3854 | * not idle. | |
3855 | */ | |
d15bcfdb | 3856 | idle = CPU_NOT_IDLE; |
1da177e4 | 3857 | } |
1bd77f2d | 3858 | sd->last_balance = jiffies; |
1da177e4 | 3859 | } |
d07355f5 | 3860 | if (need_serialize) |
08c183f3 CL |
3861 | spin_unlock(&balancing); |
3862 | out: | |
f549da84 | 3863 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3864 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3865 | update_next_balance = 1; |
3866 | } | |
783609c6 SS |
3867 | |
3868 | /* | |
3869 | * Stop the load balance at this level. There is another | |
3870 | * CPU in our sched group which is doing load balancing more | |
3871 | * actively. | |
3872 | */ | |
3873 | if (!balance) | |
3874 | break; | |
1da177e4 | 3875 | } |
f549da84 SS |
3876 | |
3877 | /* | |
3878 | * next_balance will be updated only when there is a need. | |
3879 | * When the cpu is attached to null domain for ex, it will not be | |
3880 | * updated. | |
3881 | */ | |
3882 | if (likely(update_next_balance)) | |
3883 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3884 | } |
3885 | ||
3886 | /* | |
3887 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3888 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3889 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3890 | */ | |
3891 | static void run_rebalance_domains(struct softirq_action *h) | |
3892 | { | |
dd41f596 IM |
3893 | int this_cpu = smp_processor_id(); |
3894 | struct rq *this_rq = cpu_rq(this_cpu); | |
3895 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3896 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3897 | |
dd41f596 | 3898 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3899 | |
3900 | #ifdef CONFIG_NO_HZ | |
3901 | /* | |
3902 | * If this cpu is the owner for idle load balancing, then do the | |
3903 | * balancing on behalf of the other idle cpus whose ticks are | |
3904 | * stopped. | |
3905 | */ | |
dd41f596 IM |
3906 | if (this_rq->idle_at_tick && |
3907 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3908 | cpumask_t cpus = nohz.cpu_mask; |
3909 | struct rq *rq; | |
3910 | int balance_cpu; | |
3911 | ||
dd41f596 | 3912 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3913 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3914 | /* |
3915 | * If this cpu gets work to do, stop the load balancing | |
3916 | * work being done for other cpus. Next load | |
3917 | * balancing owner will pick it up. | |
3918 | */ | |
3919 | if (need_resched()) | |
3920 | break; | |
3921 | ||
de0cf899 | 3922 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3923 | |
3924 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3925 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3926 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3927 | } |
3928 | } | |
3929 | #endif | |
3930 | } | |
3931 | ||
3932 | /* | |
3933 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3934 | * | |
3935 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3936 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3937 | * if the whole system is idle. | |
3938 | */ | |
dd41f596 | 3939 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3940 | { |
46cb4b7c SS |
3941 | #ifdef CONFIG_NO_HZ |
3942 | /* | |
3943 | * If we were in the nohz mode recently and busy at the current | |
3944 | * scheduler tick, then check if we need to nominate new idle | |
3945 | * load balancer. | |
3946 | */ | |
3947 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3948 | rq->in_nohz_recently = 0; | |
3949 | ||
3950 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3951 | cpu_clear(cpu, nohz.cpu_mask); | |
3952 | atomic_set(&nohz.load_balancer, -1); | |
3953 | } | |
3954 | ||
3955 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3956 | /* | |
3957 | * simple selection for now: Nominate the | |
3958 | * first cpu in the nohz list to be the next | |
3959 | * ilb owner. | |
3960 | * | |
3961 | * TBD: Traverse the sched domains and nominate | |
3962 | * the nearest cpu in the nohz.cpu_mask. | |
3963 | */ | |
3964 | int ilb = first_cpu(nohz.cpu_mask); | |
3965 | ||
434d53b0 | 3966 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3967 | resched_cpu(ilb); |
3968 | } | |
3969 | } | |
3970 | ||
3971 | /* | |
3972 | * If this cpu is idle and doing idle load balancing for all the | |
3973 | * cpus with ticks stopped, is it time for that to stop? | |
3974 | */ | |
3975 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3976 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3977 | resched_cpu(cpu); | |
3978 | return; | |
3979 | } | |
3980 | ||
3981 | /* | |
3982 | * If this cpu is idle and the idle load balancing is done by | |
3983 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3984 | */ | |
3985 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3986 | cpu_isset(cpu, nohz.cpu_mask)) | |
3987 | return; | |
3988 | #endif | |
3989 | if (time_after_eq(jiffies, rq->next_balance)) | |
3990 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3991 | } |
dd41f596 IM |
3992 | |
3993 | #else /* CONFIG_SMP */ | |
3994 | ||
1da177e4 LT |
3995 | /* |
3996 | * on UP we do not need to balance between CPUs: | |
3997 | */ | |
70b97a7f | 3998 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3999 | { |
4000 | } | |
dd41f596 | 4001 | |
1da177e4 LT |
4002 | #endif |
4003 | ||
1da177e4 LT |
4004 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4005 | ||
4006 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4007 | ||
4008 | /* | |
41b86e9c IM |
4009 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4010 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4011 | */ |
41b86e9c | 4012 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4013 | { |
1da177e4 | 4014 | unsigned long flags; |
41b86e9c IM |
4015 | u64 ns, delta_exec; |
4016 | struct rq *rq; | |
48f24c4d | 4017 | |
41b86e9c IM |
4018 | rq = task_rq_lock(p, &flags); |
4019 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4020 | if (task_current(rq, p)) { |
a8e504d2 IM |
4021 | update_rq_clock(rq); |
4022 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4023 | if ((s64)delta_exec > 0) |
4024 | ns += delta_exec; | |
4025 | } | |
4026 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4027 | |
1da177e4 LT |
4028 | return ns; |
4029 | } | |
4030 | ||
1da177e4 LT |
4031 | /* |
4032 | * Account user cpu time to a process. | |
4033 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4034 | * @cputime: the cpu time spent in user space since the last update |
4035 | */ | |
4036 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4037 | { | |
4038 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4039 | cputime64_t tmp; | |
4040 | ||
4041 | p->utime = cputime_add(p->utime, cputime); | |
4042 | ||
4043 | /* Add user time to cpustat. */ | |
4044 | tmp = cputime_to_cputime64(cputime); | |
4045 | if (TASK_NICE(p) > 0) | |
4046 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4047 | else | |
4048 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4049 | /* Account for user time used */ |
4050 | acct_update_integrals(p); | |
1da177e4 LT |
4051 | } |
4052 | ||
94886b84 LV |
4053 | /* |
4054 | * Account guest cpu time to a process. | |
4055 | * @p: the process that the cpu time gets accounted to | |
4056 | * @cputime: the cpu time spent in virtual machine since the last update | |
4057 | */ | |
f7402e03 | 4058 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4059 | { |
4060 | cputime64_t tmp; | |
4061 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4062 | ||
4063 | tmp = cputime_to_cputime64(cputime); | |
4064 | ||
4065 | p->utime = cputime_add(p->utime, cputime); | |
4066 | p->gtime = cputime_add(p->gtime, cputime); | |
4067 | ||
4068 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4069 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4070 | } | |
4071 | ||
c66f08be MN |
4072 | /* |
4073 | * Account scaled user cpu time to a process. | |
4074 | * @p: the process that the cpu time gets accounted to | |
4075 | * @cputime: the cpu time spent in user space since the last update | |
4076 | */ | |
4077 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4078 | { | |
4079 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4080 | } | |
4081 | ||
1da177e4 LT |
4082 | /* |
4083 | * Account system cpu time to a process. | |
4084 | * @p: the process that the cpu time gets accounted to | |
4085 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4086 | * @cputime: the cpu time spent in kernel space since the last update | |
4087 | */ | |
4088 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4089 | cputime_t cputime) | |
4090 | { | |
4091 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4092 | struct rq *rq = this_rq(); |
1da177e4 LT |
4093 | cputime64_t tmp; |
4094 | ||
983ed7a6 HH |
4095 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4096 | account_guest_time(p, cputime); | |
4097 | return; | |
4098 | } | |
94886b84 | 4099 | |
1da177e4 LT |
4100 | p->stime = cputime_add(p->stime, cputime); |
4101 | ||
4102 | /* Add system time to cpustat. */ | |
4103 | tmp = cputime_to_cputime64(cputime); | |
4104 | if (hardirq_count() - hardirq_offset) | |
4105 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4106 | else if (softirq_count()) | |
4107 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4108 | else if (p != rq->idle) |
1da177e4 | 4109 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4110 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4111 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4112 | else | |
4113 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4114 | /* Account for system time used */ | |
4115 | acct_update_integrals(p); | |
1da177e4 LT |
4116 | } |
4117 | ||
c66f08be MN |
4118 | /* |
4119 | * Account scaled system cpu time to a process. | |
4120 | * @p: the process that the cpu time gets accounted to | |
4121 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4122 | * @cputime: the cpu time spent in kernel space since the last update | |
4123 | */ | |
4124 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4125 | { | |
4126 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4127 | } | |
4128 | ||
1da177e4 LT |
4129 | /* |
4130 | * Account for involuntary wait time. | |
4131 | * @p: the process from which the cpu time has been stolen | |
4132 | * @steal: the cpu time spent in involuntary wait | |
4133 | */ | |
4134 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4135 | { | |
4136 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4137 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4138 | struct rq *rq = this_rq(); |
1da177e4 LT |
4139 | |
4140 | if (p == rq->idle) { | |
4141 | p->stime = cputime_add(p->stime, steal); | |
4142 | if (atomic_read(&rq->nr_iowait) > 0) | |
4143 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4144 | else | |
4145 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4146 | } else |
1da177e4 LT |
4147 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4148 | } | |
4149 | ||
7835b98b CL |
4150 | /* |
4151 | * This function gets called by the timer code, with HZ frequency. | |
4152 | * We call it with interrupts disabled. | |
4153 | * | |
4154 | * It also gets called by the fork code, when changing the parent's | |
4155 | * timeslices. | |
4156 | */ | |
4157 | void scheduler_tick(void) | |
4158 | { | |
7835b98b CL |
4159 | int cpu = smp_processor_id(); |
4160 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4161 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4162 | |
4163 | sched_clock_tick(); | |
dd41f596 IM |
4164 | |
4165 | spin_lock(&rq->lock); | |
3e51f33f | 4166 | update_rq_clock(rq); |
f1a438d8 | 4167 | update_cpu_load(rq); |
fa85ae24 | 4168 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4169 | spin_unlock(&rq->lock); |
7835b98b | 4170 | |
e418e1c2 | 4171 | #ifdef CONFIG_SMP |
dd41f596 IM |
4172 | rq->idle_at_tick = idle_cpu(cpu); |
4173 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4174 | #endif |
1da177e4 LT |
4175 | } |
4176 | ||
6cd8a4bb SR |
4177 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4178 | defined(CONFIG_PREEMPT_TRACER)) | |
4179 | ||
4180 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4181 | { | |
4182 | if (in_lock_functions(addr)) { | |
4183 | addr = CALLER_ADDR2; | |
4184 | if (in_lock_functions(addr)) | |
4185 | addr = CALLER_ADDR3; | |
4186 | } | |
4187 | return addr; | |
4188 | } | |
1da177e4 | 4189 | |
43627582 | 4190 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4191 | { |
6cd8a4bb | 4192 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4193 | /* |
4194 | * Underflow? | |
4195 | */ | |
9a11b49a IM |
4196 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4197 | return; | |
6cd8a4bb | 4198 | #endif |
1da177e4 | 4199 | preempt_count() += val; |
6cd8a4bb | 4200 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4201 | /* |
4202 | * Spinlock count overflowing soon? | |
4203 | */ | |
33859f7f MOS |
4204 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4205 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4206 | #endif |
4207 | if (preempt_count() == val) | |
4208 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4209 | } |
4210 | EXPORT_SYMBOL(add_preempt_count); | |
4211 | ||
43627582 | 4212 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4213 | { |
6cd8a4bb | 4214 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4215 | /* |
4216 | * Underflow? | |
4217 | */ | |
9a11b49a IM |
4218 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4219 | return; | |
1da177e4 LT |
4220 | /* |
4221 | * Is the spinlock portion underflowing? | |
4222 | */ | |
9a11b49a IM |
4223 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4224 | !(preempt_count() & PREEMPT_MASK))) | |
4225 | return; | |
6cd8a4bb | 4226 | #endif |
9a11b49a | 4227 | |
6cd8a4bb SR |
4228 | if (preempt_count() == val) |
4229 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4230 | preempt_count() -= val; |
4231 | } | |
4232 | EXPORT_SYMBOL(sub_preempt_count); | |
4233 | ||
4234 | #endif | |
4235 | ||
4236 | /* | |
dd41f596 | 4237 | * Print scheduling while atomic bug: |
1da177e4 | 4238 | */ |
dd41f596 | 4239 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4240 | { |
838225b4 SS |
4241 | struct pt_regs *regs = get_irq_regs(); |
4242 | ||
4243 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4244 | prev->comm, prev->pid, preempt_count()); | |
4245 | ||
dd41f596 | 4246 | debug_show_held_locks(prev); |
e21f5b15 | 4247 | print_modules(); |
dd41f596 IM |
4248 | if (irqs_disabled()) |
4249 | print_irqtrace_events(prev); | |
838225b4 SS |
4250 | |
4251 | if (regs) | |
4252 | show_regs(regs); | |
4253 | else | |
4254 | dump_stack(); | |
dd41f596 | 4255 | } |
1da177e4 | 4256 | |
dd41f596 IM |
4257 | /* |
4258 | * Various schedule()-time debugging checks and statistics: | |
4259 | */ | |
4260 | static inline void schedule_debug(struct task_struct *prev) | |
4261 | { | |
1da177e4 | 4262 | /* |
41a2d6cf | 4263 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4264 | * schedule() atomically, we ignore that path for now. |
4265 | * Otherwise, whine if we are scheduling when we should not be. | |
4266 | */ | |
3f33a7ce | 4267 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4268 | __schedule_bug(prev); |
4269 | ||
1da177e4 LT |
4270 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4271 | ||
2d72376b | 4272 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4273 | #ifdef CONFIG_SCHEDSTATS |
4274 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4275 | schedstat_inc(this_rq(), bkl_count); |
4276 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4277 | } |
4278 | #endif | |
dd41f596 IM |
4279 | } |
4280 | ||
4281 | /* | |
4282 | * Pick up the highest-prio task: | |
4283 | */ | |
4284 | static inline struct task_struct * | |
ff95f3df | 4285 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4286 | { |
5522d5d5 | 4287 | const struct sched_class *class; |
dd41f596 | 4288 | struct task_struct *p; |
1da177e4 LT |
4289 | |
4290 | /* | |
dd41f596 IM |
4291 | * Optimization: we know that if all tasks are in |
4292 | * the fair class we can call that function directly: | |
1da177e4 | 4293 | */ |
dd41f596 | 4294 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4295 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4296 | if (likely(p)) |
4297 | return p; | |
1da177e4 LT |
4298 | } |
4299 | ||
dd41f596 IM |
4300 | class = sched_class_highest; |
4301 | for ( ; ; ) { | |
fb8d4724 | 4302 | p = class->pick_next_task(rq); |
dd41f596 IM |
4303 | if (p) |
4304 | return p; | |
4305 | /* | |
4306 | * Will never be NULL as the idle class always | |
4307 | * returns a non-NULL p: | |
4308 | */ | |
4309 | class = class->next; | |
4310 | } | |
4311 | } | |
1da177e4 | 4312 | |
dd41f596 IM |
4313 | /* |
4314 | * schedule() is the main scheduler function. | |
4315 | */ | |
4316 | asmlinkage void __sched schedule(void) | |
4317 | { | |
4318 | struct task_struct *prev, *next; | |
67ca7bde | 4319 | unsigned long *switch_count; |
dd41f596 | 4320 | struct rq *rq; |
31656519 | 4321 | int cpu; |
dd41f596 IM |
4322 | |
4323 | need_resched: | |
4324 | preempt_disable(); | |
4325 | cpu = smp_processor_id(); | |
4326 | rq = cpu_rq(cpu); | |
4327 | rcu_qsctr_inc(cpu); | |
4328 | prev = rq->curr; | |
4329 | switch_count = &prev->nivcsw; | |
4330 | ||
4331 | release_kernel_lock(prev); | |
4332 | need_resched_nonpreemptible: | |
4333 | ||
4334 | schedule_debug(prev); | |
1da177e4 | 4335 | |
31656519 | 4336 | if (sched_feat(HRTICK)) |
f333fdc9 | 4337 | hrtick_clear(rq); |
8f4d37ec | 4338 | |
1e819950 IM |
4339 | /* |
4340 | * Do the rq-clock update outside the rq lock: | |
4341 | */ | |
4342 | local_irq_disable(); | |
3e51f33f | 4343 | update_rq_clock(rq); |
1e819950 IM |
4344 | spin_lock(&rq->lock); |
4345 | clear_tsk_need_resched(prev); | |
1da177e4 | 4346 | |
1da177e4 | 4347 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4348 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4349 | prev->state = TASK_RUNNING; |
16882c1e | 4350 | else |
2e1cb74a | 4351 | deactivate_task(rq, prev, 1); |
dd41f596 | 4352 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4353 | } |
4354 | ||
9a897c5a SR |
4355 | #ifdef CONFIG_SMP |
4356 | if (prev->sched_class->pre_schedule) | |
4357 | prev->sched_class->pre_schedule(rq, prev); | |
4358 | #endif | |
f65eda4f | 4359 | |
dd41f596 | 4360 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4361 | idle_balance(cpu, rq); |
1da177e4 | 4362 | |
31ee529c | 4363 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4364 | next = pick_next_task(rq, prev); |
1da177e4 | 4365 | |
1da177e4 | 4366 | if (likely(prev != next)) { |
673a90a1 DS |
4367 | sched_info_switch(prev, next); |
4368 | ||
1da177e4 LT |
4369 | rq->nr_switches++; |
4370 | rq->curr = next; | |
4371 | ++*switch_count; | |
4372 | ||
dd41f596 | 4373 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4374 | /* |
4375 | * the context switch might have flipped the stack from under | |
4376 | * us, hence refresh the local variables. | |
4377 | */ | |
4378 | cpu = smp_processor_id(); | |
4379 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4380 | } else |
4381 | spin_unlock_irq(&rq->lock); | |
4382 | ||
8f4d37ec | 4383 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4384 | goto need_resched_nonpreemptible; |
8f4d37ec | 4385 | |
1da177e4 LT |
4386 | preempt_enable_no_resched(); |
4387 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4388 | goto need_resched; | |
4389 | } | |
1da177e4 LT |
4390 | EXPORT_SYMBOL(schedule); |
4391 | ||
4392 | #ifdef CONFIG_PREEMPT | |
4393 | /* | |
2ed6e34f | 4394 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4395 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4396 | * occur there and call schedule directly. |
4397 | */ | |
4398 | asmlinkage void __sched preempt_schedule(void) | |
4399 | { | |
4400 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4401 | |
1da177e4 LT |
4402 | /* |
4403 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4404 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4405 | */ |
beed33a8 | 4406 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4407 | return; |
4408 | ||
3a5c359a AK |
4409 | do { |
4410 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4411 | schedule(); |
3a5c359a | 4412 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4413 | |
3a5c359a AK |
4414 | /* |
4415 | * Check again in case we missed a preemption opportunity | |
4416 | * between schedule and now. | |
4417 | */ | |
4418 | barrier(); | |
4419 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4420 | } |
1da177e4 LT |
4421 | EXPORT_SYMBOL(preempt_schedule); |
4422 | ||
4423 | /* | |
2ed6e34f | 4424 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4425 | * off of irq context. |
4426 | * Note, that this is called and return with irqs disabled. This will | |
4427 | * protect us against recursive calling from irq. | |
4428 | */ | |
4429 | asmlinkage void __sched preempt_schedule_irq(void) | |
4430 | { | |
4431 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4432 | |
2ed6e34f | 4433 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4434 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4435 | ||
3a5c359a AK |
4436 | do { |
4437 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4438 | local_irq_enable(); |
4439 | schedule(); | |
4440 | local_irq_disable(); | |
3a5c359a | 4441 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4442 | |
3a5c359a AK |
4443 | /* |
4444 | * Check again in case we missed a preemption opportunity | |
4445 | * between schedule and now. | |
4446 | */ | |
4447 | barrier(); | |
4448 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4449 | } |
4450 | ||
4451 | #endif /* CONFIG_PREEMPT */ | |
4452 | ||
95cdf3b7 IM |
4453 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4454 | void *key) | |
1da177e4 | 4455 | { |
48f24c4d | 4456 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4457 | } |
1da177e4 LT |
4458 | EXPORT_SYMBOL(default_wake_function); |
4459 | ||
4460 | /* | |
41a2d6cf IM |
4461 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4462 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4463 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4464 | * | |
4465 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4466 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4467 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4468 | */ | |
4469 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4470 | int nr_exclusive, int sync, void *key) | |
4471 | { | |
2e45874c | 4472 | wait_queue_t *curr, *next; |
1da177e4 | 4473 | |
2e45874c | 4474 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4475 | unsigned flags = curr->flags; |
4476 | ||
1da177e4 | 4477 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4478 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4479 | break; |
4480 | } | |
4481 | } | |
4482 | ||
4483 | /** | |
4484 | * __wake_up - wake up threads blocked on a waitqueue. | |
4485 | * @q: the waitqueue | |
4486 | * @mode: which threads | |
4487 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4488 | * @key: is directly passed to the wakeup function |
1da177e4 | 4489 | */ |
7ad5b3a5 | 4490 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4491 | int nr_exclusive, void *key) |
1da177e4 LT |
4492 | { |
4493 | unsigned long flags; | |
4494 | ||
4495 | spin_lock_irqsave(&q->lock, flags); | |
4496 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4497 | spin_unlock_irqrestore(&q->lock, flags); | |
4498 | } | |
1da177e4 LT |
4499 | EXPORT_SYMBOL(__wake_up); |
4500 | ||
4501 | /* | |
4502 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4503 | */ | |
7ad5b3a5 | 4504 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4505 | { |
4506 | __wake_up_common(q, mode, 1, 0, NULL); | |
4507 | } | |
4508 | ||
4509 | /** | |
67be2dd1 | 4510 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4511 | * @q: the waitqueue |
4512 | * @mode: which threads | |
4513 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4514 | * | |
4515 | * The sync wakeup differs that the waker knows that it will schedule | |
4516 | * away soon, so while the target thread will be woken up, it will not | |
4517 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4518 | * with each other. This can prevent needless bouncing between CPUs. | |
4519 | * | |
4520 | * On UP it can prevent extra preemption. | |
4521 | */ | |
7ad5b3a5 | 4522 | void |
95cdf3b7 | 4523 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4524 | { |
4525 | unsigned long flags; | |
4526 | int sync = 1; | |
4527 | ||
4528 | if (unlikely(!q)) | |
4529 | return; | |
4530 | ||
4531 | if (unlikely(!nr_exclusive)) | |
4532 | sync = 0; | |
4533 | ||
4534 | spin_lock_irqsave(&q->lock, flags); | |
4535 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4536 | spin_unlock_irqrestore(&q->lock, flags); | |
4537 | } | |
4538 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4539 | ||
b15136e9 | 4540 | void complete(struct completion *x) |
1da177e4 LT |
4541 | { |
4542 | unsigned long flags; | |
4543 | ||
4544 | spin_lock_irqsave(&x->wait.lock, flags); | |
4545 | x->done++; | |
d9514f6c | 4546 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4547 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4548 | } | |
4549 | EXPORT_SYMBOL(complete); | |
4550 | ||
b15136e9 | 4551 | void complete_all(struct completion *x) |
1da177e4 LT |
4552 | { |
4553 | unsigned long flags; | |
4554 | ||
4555 | spin_lock_irqsave(&x->wait.lock, flags); | |
4556 | x->done += UINT_MAX/2; | |
d9514f6c | 4557 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4558 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4559 | } | |
4560 | EXPORT_SYMBOL(complete_all); | |
4561 | ||
8cbbe86d AK |
4562 | static inline long __sched |
4563 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4564 | { |
1da177e4 LT |
4565 | if (!x->done) { |
4566 | DECLARE_WAITQUEUE(wait, current); | |
4567 | ||
4568 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4569 | __add_wait_queue_tail(&x->wait, &wait); | |
4570 | do { | |
009e577e MW |
4571 | if ((state == TASK_INTERRUPTIBLE && |
4572 | signal_pending(current)) || | |
4573 | (state == TASK_KILLABLE && | |
4574 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4575 | timeout = -ERESTARTSYS; |
4576 | break; | |
8cbbe86d AK |
4577 | } |
4578 | __set_current_state(state); | |
1da177e4 LT |
4579 | spin_unlock_irq(&x->wait.lock); |
4580 | timeout = schedule_timeout(timeout); | |
4581 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4582 | } while (!x->done && timeout); |
1da177e4 | 4583 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4584 | if (!x->done) |
4585 | return timeout; | |
1da177e4 LT |
4586 | } |
4587 | x->done--; | |
ea71a546 | 4588 | return timeout ?: 1; |
1da177e4 | 4589 | } |
1da177e4 | 4590 | |
8cbbe86d AK |
4591 | static long __sched |
4592 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4593 | { |
1da177e4 LT |
4594 | might_sleep(); |
4595 | ||
4596 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4597 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4598 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4599 | return timeout; |
4600 | } | |
1da177e4 | 4601 | |
b15136e9 | 4602 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4603 | { |
4604 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4605 | } |
8cbbe86d | 4606 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4607 | |
b15136e9 | 4608 | unsigned long __sched |
8cbbe86d | 4609 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4610 | { |
8cbbe86d | 4611 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4612 | } |
8cbbe86d | 4613 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4614 | |
8cbbe86d | 4615 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4616 | { |
51e97990 AK |
4617 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4618 | if (t == -ERESTARTSYS) | |
4619 | return t; | |
4620 | return 0; | |
0fec171c | 4621 | } |
8cbbe86d | 4622 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4623 | |
b15136e9 | 4624 | unsigned long __sched |
8cbbe86d AK |
4625 | wait_for_completion_interruptible_timeout(struct completion *x, |
4626 | unsigned long timeout) | |
0fec171c | 4627 | { |
8cbbe86d | 4628 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4629 | } |
8cbbe86d | 4630 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4631 | |
009e577e MW |
4632 | int __sched wait_for_completion_killable(struct completion *x) |
4633 | { | |
4634 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4635 | if (t == -ERESTARTSYS) | |
4636 | return t; | |
4637 | return 0; | |
4638 | } | |
4639 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4640 | ||
8cbbe86d AK |
4641 | static long __sched |
4642 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4643 | { |
0fec171c IM |
4644 | unsigned long flags; |
4645 | wait_queue_t wait; | |
4646 | ||
4647 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4648 | |
8cbbe86d | 4649 | __set_current_state(state); |
1da177e4 | 4650 | |
8cbbe86d AK |
4651 | spin_lock_irqsave(&q->lock, flags); |
4652 | __add_wait_queue(q, &wait); | |
4653 | spin_unlock(&q->lock); | |
4654 | timeout = schedule_timeout(timeout); | |
4655 | spin_lock_irq(&q->lock); | |
4656 | __remove_wait_queue(q, &wait); | |
4657 | spin_unlock_irqrestore(&q->lock, flags); | |
4658 | ||
4659 | return timeout; | |
4660 | } | |
4661 | ||
4662 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4663 | { | |
4664 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4665 | } |
1da177e4 LT |
4666 | EXPORT_SYMBOL(interruptible_sleep_on); |
4667 | ||
0fec171c | 4668 | long __sched |
95cdf3b7 | 4669 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4670 | { |
8cbbe86d | 4671 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4672 | } |
1da177e4 LT |
4673 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4674 | ||
0fec171c | 4675 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4676 | { |
8cbbe86d | 4677 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4678 | } |
1da177e4 LT |
4679 | EXPORT_SYMBOL(sleep_on); |
4680 | ||
0fec171c | 4681 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4682 | { |
8cbbe86d | 4683 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4684 | } |
1da177e4 LT |
4685 | EXPORT_SYMBOL(sleep_on_timeout); |
4686 | ||
b29739f9 IM |
4687 | #ifdef CONFIG_RT_MUTEXES |
4688 | ||
4689 | /* | |
4690 | * rt_mutex_setprio - set the current priority of a task | |
4691 | * @p: task | |
4692 | * @prio: prio value (kernel-internal form) | |
4693 | * | |
4694 | * This function changes the 'effective' priority of a task. It does | |
4695 | * not touch ->normal_prio like __setscheduler(). | |
4696 | * | |
4697 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4698 | */ | |
36c8b586 | 4699 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4700 | { |
4701 | unsigned long flags; | |
83b699ed | 4702 | int oldprio, on_rq, running; |
70b97a7f | 4703 | struct rq *rq; |
cb469845 | 4704 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4705 | |
4706 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4707 | ||
4708 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4709 | update_rq_clock(rq); |
b29739f9 | 4710 | |
d5f9f942 | 4711 | oldprio = p->prio; |
dd41f596 | 4712 | on_rq = p->se.on_rq; |
051a1d1a | 4713 | running = task_current(rq, p); |
0e1f3483 | 4714 | if (on_rq) |
69be72c1 | 4715 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4716 | if (running) |
4717 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4718 | |
4719 | if (rt_prio(prio)) | |
4720 | p->sched_class = &rt_sched_class; | |
4721 | else | |
4722 | p->sched_class = &fair_sched_class; | |
4723 | ||
b29739f9 IM |
4724 | p->prio = prio; |
4725 | ||
0e1f3483 HS |
4726 | if (running) |
4727 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4728 | if (on_rq) { |
8159f87e | 4729 | enqueue_task(rq, p, 0); |
cb469845 SR |
4730 | |
4731 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4732 | } |
4733 | task_rq_unlock(rq, &flags); | |
4734 | } | |
4735 | ||
4736 | #endif | |
4737 | ||
36c8b586 | 4738 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4739 | { |
dd41f596 | 4740 | int old_prio, delta, on_rq; |
1da177e4 | 4741 | unsigned long flags; |
70b97a7f | 4742 | struct rq *rq; |
1da177e4 LT |
4743 | |
4744 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4745 | return; | |
4746 | /* | |
4747 | * We have to be careful, if called from sys_setpriority(), | |
4748 | * the task might be in the middle of scheduling on another CPU. | |
4749 | */ | |
4750 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4751 | update_rq_clock(rq); |
1da177e4 LT |
4752 | /* |
4753 | * The RT priorities are set via sched_setscheduler(), but we still | |
4754 | * allow the 'normal' nice value to be set - but as expected | |
4755 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4756 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4757 | */ |
e05606d3 | 4758 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4759 | p->static_prio = NICE_TO_PRIO(nice); |
4760 | goto out_unlock; | |
4761 | } | |
dd41f596 | 4762 | on_rq = p->se.on_rq; |
c09595f6 | 4763 | if (on_rq) |
69be72c1 | 4764 | dequeue_task(rq, p, 0); |
1da177e4 | 4765 | |
1da177e4 | 4766 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4767 | set_load_weight(p); |
b29739f9 IM |
4768 | old_prio = p->prio; |
4769 | p->prio = effective_prio(p); | |
4770 | delta = p->prio - old_prio; | |
1da177e4 | 4771 | |
dd41f596 | 4772 | if (on_rq) { |
8159f87e | 4773 | enqueue_task(rq, p, 0); |
1da177e4 | 4774 | /* |
d5f9f942 AM |
4775 | * If the task increased its priority or is running and |
4776 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4777 | */ |
d5f9f942 | 4778 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4779 | resched_task(rq->curr); |
4780 | } | |
4781 | out_unlock: | |
4782 | task_rq_unlock(rq, &flags); | |
4783 | } | |
1da177e4 LT |
4784 | EXPORT_SYMBOL(set_user_nice); |
4785 | ||
e43379f1 MM |
4786 | /* |
4787 | * can_nice - check if a task can reduce its nice value | |
4788 | * @p: task | |
4789 | * @nice: nice value | |
4790 | */ | |
36c8b586 | 4791 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4792 | { |
024f4747 MM |
4793 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4794 | int nice_rlim = 20 - nice; | |
48f24c4d | 4795 | |
e43379f1 MM |
4796 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4797 | capable(CAP_SYS_NICE)); | |
4798 | } | |
4799 | ||
1da177e4 LT |
4800 | #ifdef __ARCH_WANT_SYS_NICE |
4801 | ||
4802 | /* | |
4803 | * sys_nice - change the priority of the current process. | |
4804 | * @increment: priority increment | |
4805 | * | |
4806 | * sys_setpriority is a more generic, but much slower function that | |
4807 | * does similar things. | |
4808 | */ | |
4809 | asmlinkage long sys_nice(int increment) | |
4810 | { | |
48f24c4d | 4811 | long nice, retval; |
1da177e4 LT |
4812 | |
4813 | /* | |
4814 | * Setpriority might change our priority at the same moment. | |
4815 | * We don't have to worry. Conceptually one call occurs first | |
4816 | * and we have a single winner. | |
4817 | */ | |
e43379f1 MM |
4818 | if (increment < -40) |
4819 | increment = -40; | |
1da177e4 LT |
4820 | if (increment > 40) |
4821 | increment = 40; | |
4822 | ||
4823 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4824 | if (nice < -20) | |
4825 | nice = -20; | |
4826 | if (nice > 19) | |
4827 | nice = 19; | |
4828 | ||
e43379f1 MM |
4829 | if (increment < 0 && !can_nice(current, nice)) |
4830 | return -EPERM; | |
4831 | ||
1da177e4 LT |
4832 | retval = security_task_setnice(current, nice); |
4833 | if (retval) | |
4834 | return retval; | |
4835 | ||
4836 | set_user_nice(current, nice); | |
4837 | return 0; | |
4838 | } | |
4839 | ||
4840 | #endif | |
4841 | ||
4842 | /** | |
4843 | * task_prio - return the priority value of a given task. | |
4844 | * @p: the task in question. | |
4845 | * | |
4846 | * This is the priority value as seen by users in /proc. | |
4847 | * RT tasks are offset by -200. Normal tasks are centered | |
4848 | * around 0, value goes from -16 to +15. | |
4849 | */ | |
36c8b586 | 4850 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4851 | { |
4852 | return p->prio - MAX_RT_PRIO; | |
4853 | } | |
4854 | ||
4855 | /** | |
4856 | * task_nice - return the nice value of a given task. | |
4857 | * @p: the task in question. | |
4858 | */ | |
36c8b586 | 4859 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4860 | { |
4861 | return TASK_NICE(p); | |
4862 | } | |
150d8bed | 4863 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4864 | |
4865 | /** | |
4866 | * idle_cpu - is a given cpu idle currently? | |
4867 | * @cpu: the processor in question. | |
4868 | */ | |
4869 | int idle_cpu(int cpu) | |
4870 | { | |
4871 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4872 | } | |
4873 | ||
1da177e4 LT |
4874 | /** |
4875 | * idle_task - return the idle task for a given cpu. | |
4876 | * @cpu: the processor in question. | |
4877 | */ | |
36c8b586 | 4878 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4879 | { |
4880 | return cpu_rq(cpu)->idle; | |
4881 | } | |
4882 | ||
4883 | /** | |
4884 | * find_process_by_pid - find a process with a matching PID value. | |
4885 | * @pid: the pid in question. | |
4886 | */ | |
a9957449 | 4887 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4888 | { |
228ebcbe | 4889 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4890 | } |
4891 | ||
4892 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4893 | static void |
4894 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4895 | { |
dd41f596 | 4896 | BUG_ON(p->se.on_rq); |
48f24c4d | 4897 | |
1da177e4 | 4898 | p->policy = policy; |
dd41f596 IM |
4899 | switch (p->policy) { |
4900 | case SCHED_NORMAL: | |
4901 | case SCHED_BATCH: | |
4902 | case SCHED_IDLE: | |
4903 | p->sched_class = &fair_sched_class; | |
4904 | break; | |
4905 | case SCHED_FIFO: | |
4906 | case SCHED_RR: | |
4907 | p->sched_class = &rt_sched_class; | |
4908 | break; | |
4909 | } | |
4910 | ||
1da177e4 | 4911 | p->rt_priority = prio; |
b29739f9 IM |
4912 | p->normal_prio = normal_prio(p); |
4913 | /* we are holding p->pi_lock already */ | |
4914 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4915 | set_load_weight(p); |
1da177e4 LT |
4916 | } |
4917 | ||
961ccddd RR |
4918 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4919 | struct sched_param *param, bool user) | |
1da177e4 | 4920 | { |
83b699ed | 4921 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4922 | unsigned long flags; |
cb469845 | 4923 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4924 | struct rq *rq; |
1da177e4 | 4925 | |
66e5393a SR |
4926 | /* may grab non-irq protected spin_locks */ |
4927 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4928 | recheck: |
4929 | /* double check policy once rq lock held */ | |
4930 | if (policy < 0) | |
4931 | policy = oldpolicy = p->policy; | |
4932 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4933 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4934 | policy != SCHED_IDLE) | |
b0a9499c | 4935 | return -EINVAL; |
1da177e4 LT |
4936 | /* |
4937 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4938 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4939 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4940 | */ |
4941 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4942 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4943 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4944 | return -EINVAL; |
e05606d3 | 4945 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4946 | return -EINVAL; |
4947 | ||
37e4ab3f OC |
4948 | /* |
4949 | * Allow unprivileged RT tasks to decrease priority: | |
4950 | */ | |
961ccddd | 4951 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4952 | if (rt_policy(policy)) { |
8dc3e909 | 4953 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4954 | |
4955 | if (!lock_task_sighand(p, &flags)) | |
4956 | return -ESRCH; | |
4957 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4958 | unlock_task_sighand(p, &flags); | |
4959 | ||
4960 | /* can't set/change the rt policy */ | |
4961 | if (policy != p->policy && !rlim_rtprio) | |
4962 | return -EPERM; | |
4963 | ||
4964 | /* can't increase priority */ | |
4965 | if (param->sched_priority > p->rt_priority && | |
4966 | param->sched_priority > rlim_rtprio) | |
4967 | return -EPERM; | |
4968 | } | |
dd41f596 IM |
4969 | /* |
4970 | * Like positive nice levels, dont allow tasks to | |
4971 | * move out of SCHED_IDLE either: | |
4972 | */ | |
4973 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4974 | return -EPERM; | |
5fe1d75f | 4975 | |
37e4ab3f OC |
4976 | /* can't change other user's priorities */ |
4977 | if ((current->euid != p->euid) && | |
4978 | (current->euid != p->uid)) | |
4979 | return -EPERM; | |
4980 | } | |
1da177e4 | 4981 | |
b68aa230 PZ |
4982 | #ifdef CONFIG_RT_GROUP_SCHED |
4983 | /* | |
4984 | * Do not allow realtime tasks into groups that have no runtime | |
4985 | * assigned. | |
4986 | */ | |
961ccddd RR |
4987 | if (user |
4988 | && rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) | |
b68aa230 PZ |
4989 | return -EPERM; |
4990 | #endif | |
4991 | ||
1da177e4 LT |
4992 | retval = security_task_setscheduler(p, policy, param); |
4993 | if (retval) | |
4994 | return retval; | |
b29739f9 IM |
4995 | /* |
4996 | * make sure no PI-waiters arrive (or leave) while we are | |
4997 | * changing the priority of the task: | |
4998 | */ | |
4999 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5000 | /* |
5001 | * To be able to change p->policy safely, the apropriate | |
5002 | * runqueue lock must be held. | |
5003 | */ | |
b29739f9 | 5004 | rq = __task_rq_lock(p); |
1da177e4 LT |
5005 | /* recheck policy now with rq lock held */ |
5006 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5007 | policy = oldpolicy = -1; | |
b29739f9 IM |
5008 | __task_rq_unlock(rq); |
5009 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5010 | goto recheck; |
5011 | } | |
2daa3577 | 5012 | update_rq_clock(rq); |
dd41f596 | 5013 | on_rq = p->se.on_rq; |
051a1d1a | 5014 | running = task_current(rq, p); |
0e1f3483 | 5015 | if (on_rq) |
2e1cb74a | 5016 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5017 | if (running) |
5018 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5019 | |
1da177e4 | 5020 | oldprio = p->prio; |
dd41f596 | 5021 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5022 | |
0e1f3483 HS |
5023 | if (running) |
5024 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5025 | if (on_rq) { |
5026 | activate_task(rq, p, 0); | |
cb469845 SR |
5027 | |
5028 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5029 | } |
b29739f9 IM |
5030 | __task_rq_unlock(rq); |
5031 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5032 | ||
95e02ca9 TG |
5033 | rt_mutex_adjust_pi(p); |
5034 | ||
1da177e4 LT |
5035 | return 0; |
5036 | } | |
961ccddd RR |
5037 | |
5038 | /** | |
5039 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5040 | * @p: the task in question. | |
5041 | * @policy: new policy. | |
5042 | * @param: structure containing the new RT priority. | |
5043 | * | |
5044 | * NOTE that the task may be already dead. | |
5045 | */ | |
5046 | int sched_setscheduler(struct task_struct *p, int policy, | |
5047 | struct sched_param *param) | |
5048 | { | |
5049 | return __sched_setscheduler(p, policy, param, true); | |
5050 | } | |
1da177e4 LT |
5051 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5052 | ||
961ccddd RR |
5053 | /** |
5054 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5055 | * @p: the task in question. | |
5056 | * @policy: new policy. | |
5057 | * @param: structure containing the new RT priority. | |
5058 | * | |
5059 | * Just like sched_setscheduler, only don't bother checking if the | |
5060 | * current context has permission. For example, this is needed in | |
5061 | * stop_machine(): we create temporary high priority worker threads, | |
5062 | * but our caller might not have that capability. | |
5063 | */ | |
5064 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5065 | struct sched_param *param) | |
5066 | { | |
5067 | return __sched_setscheduler(p, policy, param, false); | |
5068 | } | |
5069 | ||
95cdf3b7 IM |
5070 | static int |
5071 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5072 | { |
1da177e4 LT |
5073 | struct sched_param lparam; |
5074 | struct task_struct *p; | |
36c8b586 | 5075 | int retval; |
1da177e4 LT |
5076 | |
5077 | if (!param || pid < 0) | |
5078 | return -EINVAL; | |
5079 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5080 | return -EFAULT; | |
5fe1d75f ON |
5081 | |
5082 | rcu_read_lock(); | |
5083 | retval = -ESRCH; | |
1da177e4 | 5084 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5085 | if (p != NULL) |
5086 | retval = sched_setscheduler(p, policy, &lparam); | |
5087 | rcu_read_unlock(); | |
36c8b586 | 5088 | |
1da177e4 LT |
5089 | return retval; |
5090 | } | |
5091 | ||
5092 | /** | |
5093 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5094 | * @pid: the pid in question. | |
5095 | * @policy: new policy. | |
5096 | * @param: structure containing the new RT priority. | |
5097 | */ | |
41a2d6cf IM |
5098 | asmlinkage long |
5099 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5100 | { |
c21761f1 JB |
5101 | /* negative values for policy are not valid */ |
5102 | if (policy < 0) | |
5103 | return -EINVAL; | |
5104 | ||
1da177e4 LT |
5105 | return do_sched_setscheduler(pid, policy, param); |
5106 | } | |
5107 | ||
5108 | /** | |
5109 | * sys_sched_setparam - set/change the RT priority of a thread | |
5110 | * @pid: the pid in question. | |
5111 | * @param: structure containing the new RT priority. | |
5112 | */ | |
5113 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5114 | { | |
5115 | return do_sched_setscheduler(pid, -1, param); | |
5116 | } | |
5117 | ||
5118 | /** | |
5119 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5120 | * @pid: the pid in question. | |
5121 | */ | |
5122 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5123 | { | |
36c8b586 | 5124 | struct task_struct *p; |
3a5c359a | 5125 | int retval; |
1da177e4 LT |
5126 | |
5127 | if (pid < 0) | |
3a5c359a | 5128 | return -EINVAL; |
1da177e4 LT |
5129 | |
5130 | retval = -ESRCH; | |
5131 | read_lock(&tasklist_lock); | |
5132 | p = find_process_by_pid(pid); | |
5133 | if (p) { | |
5134 | retval = security_task_getscheduler(p); | |
5135 | if (!retval) | |
5136 | retval = p->policy; | |
5137 | } | |
5138 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5139 | return retval; |
5140 | } | |
5141 | ||
5142 | /** | |
5143 | * sys_sched_getscheduler - get the RT priority of a thread | |
5144 | * @pid: the pid in question. | |
5145 | * @param: structure containing the RT priority. | |
5146 | */ | |
5147 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5148 | { | |
5149 | struct sched_param lp; | |
36c8b586 | 5150 | struct task_struct *p; |
3a5c359a | 5151 | int retval; |
1da177e4 LT |
5152 | |
5153 | if (!param || pid < 0) | |
3a5c359a | 5154 | return -EINVAL; |
1da177e4 LT |
5155 | |
5156 | read_lock(&tasklist_lock); | |
5157 | p = find_process_by_pid(pid); | |
5158 | retval = -ESRCH; | |
5159 | if (!p) | |
5160 | goto out_unlock; | |
5161 | ||
5162 | retval = security_task_getscheduler(p); | |
5163 | if (retval) | |
5164 | goto out_unlock; | |
5165 | ||
5166 | lp.sched_priority = p->rt_priority; | |
5167 | read_unlock(&tasklist_lock); | |
5168 | ||
5169 | /* | |
5170 | * This one might sleep, we cannot do it with a spinlock held ... | |
5171 | */ | |
5172 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5173 | ||
1da177e4 LT |
5174 | return retval; |
5175 | ||
5176 | out_unlock: | |
5177 | read_unlock(&tasklist_lock); | |
5178 | return retval; | |
5179 | } | |
5180 | ||
b53e921b | 5181 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5182 | { |
1da177e4 | 5183 | cpumask_t cpus_allowed; |
b53e921b | 5184 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5185 | struct task_struct *p; |
5186 | int retval; | |
1da177e4 | 5187 | |
95402b38 | 5188 | get_online_cpus(); |
1da177e4 LT |
5189 | read_lock(&tasklist_lock); |
5190 | ||
5191 | p = find_process_by_pid(pid); | |
5192 | if (!p) { | |
5193 | read_unlock(&tasklist_lock); | |
95402b38 | 5194 | put_online_cpus(); |
1da177e4 LT |
5195 | return -ESRCH; |
5196 | } | |
5197 | ||
5198 | /* | |
5199 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5200 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5201 | * usage count and then drop tasklist_lock. |
5202 | */ | |
5203 | get_task_struct(p); | |
5204 | read_unlock(&tasklist_lock); | |
5205 | ||
5206 | retval = -EPERM; | |
5207 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5208 | !capable(CAP_SYS_NICE)) | |
5209 | goto out_unlock; | |
5210 | ||
e7834f8f DQ |
5211 | retval = security_task_setscheduler(p, 0, NULL); |
5212 | if (retval) | |
5213 | goto out_unlock; | |
5214 | ||
f9a86fcb | 5215 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5216 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5217 | again: |
7c16ec58 | 5218 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5219 | |
8707d8b8 | 5220 | if (!retval) { |
f9a86fcb | 5221 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5222 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5223 | /* | |
5224 | * We must have raced with a concurrent cpuset | |
5225 | * update. Just reset the cpus_allowed to the | |
5226 | * cpuset's cpus_allowed | |
5227 | */ | |
5228 | new_mask = cpus_allowed; | |
5229 | goto again; | |
5230 | } | |
5231 | } | |
1da177e4 LT |
5232 | out_unlock: |
5233 | put_task_struct(p); | |
95402b38 | 5234 | put_online_cpus(); |
1da177e4 LT |
5235 | return retval; |
5236 | } | |
5237 | ||
5238 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5239 | cpumask_t *new_mask) | |
5240 | { | |
5241 | if (len < sizeof(cpumask_t)) { | |
5242 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5243 | } else if (len > sizeof(cpumask_t)) { | |
5244 | len = sizeof(cpumask_t); | |
5245 | } | |
5246 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5247 | } | |
5248 | ||
5249 | /** | |
5250 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5251 | * @pid: pid of the process | |
5252 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5253 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5254 | */ | |
5255 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5256 | unsigned long __user *user_mask_ptr) | |
5257 | { | |
5258 | cpumask_t new_mask; | |
5259 | int retval; | |
5260 | ||
5261 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5262 | if (retval) | |
5263 | return retval; | |
5264 | ||
b53e921b | 5265 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5266 | } |
5267 | ||
1da177e4 LT |
5268 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5269 | { | |
36c8b586 | 5270 | struct task_struct *p; |
1da177e4 | 5271 | int retval; |
1da177e4 | 5272 | |
95402b38 | 5273 | get_online_cpus(); |
1da177e4 LT |
5274 | read_lock(&tasklist_lock); |
5275 | ||
5276 | retval = -ESRCH; | |
5277 | p = find_process_by_pid(pid); | |
5278 | if (!p) | |
5279 | goto out_unlock; | |
5280 | ||
e7834f8f DQ |
5281 | retval = security_task_getscheduler(p); |
5282 | if (retval) | |
5283 | goto out_unlock; | |
5284 | ||
2f7016d9 | 5285 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5286 | |
5287 | out_unlock: | |
5288 | read_unlock(&tasklist_lock); | |
95402b38 | 5289 | put_online_cpus(); |
1da177e4 | 5290 | |
9531b62f | 5291 | return retval; |
1da177e4 LT |
5292 | } |
5293 | ||
5294 | /** | |
5295 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5296 | * @pid: pid of the process | |
5297 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5298 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5299 | */ | |
5300 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5301 | unsigned long __user *user_mask_ptr) | |
5302 | { | |
5303 | int ret; | |
5304 | cpumask_t mask; | |
5305 | ||
5306 | if (len < sizeof(cpumask_t)) | |
5307 | return -EINVAL; | |
5308 | ||
5309 | ret = sched_getaffinity(pid, &mask); | |
5310 | if (ret < 0) | |
5311 | return ret; | |
5312 | ||
5313 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5314 | return -EFAULT; | |
5315 | ||
5316 | return sizeof(cpumask_t); | |
5317 | } | |
5318 | ||
5319 | /** | |
5320 | * sys_sched_yield - yield the current processor to other threads. | |
5321 | * | |
dd41f596 IM |
5322 | * This function yields the current CPU to other tasks. If there are no |
5323 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5324 | */ |
5325 | asmlinkage long sys_sched_yield(void) | |
5326 | { | |
70b97a7f | 5327 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5328 | |
2d72376b | 5329 | schedstat_inc(rq, yld_count); |
4530d7ab | 5330 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5331 | |
5332 | /* | |
5333 | * Since we are going to call schedule() anyway, there's | |
5334 | * no need to preempt or enable interrupts: | |
5335 | */ | |
5336 | __release(rq->lock); | |
8a25d5de | 5337 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5338 | _raw_spin_unlock(&rq->lock); |
5339 | preempt_enable_no_resched(); | |
5340 | ||
5341 | schedule(); | |
5342 | ||
5343 | return 0; | |
5344 | } | |
5345 | ||
e7b38404 | 5346 | static void __cond_resched(void) |
1da177e4 | 5347 | { |
8e0a43d8 IM |
5348 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5349 | __might_sleep(__FILE__, __LINE__); | |
5350 | #endif | |
5bbcfd90 IM |
5351 | /* |
5352 | * The BKS might be reacquired before we have dropped | |
5353 | * PREEMPT_ACTIVE, which could trigger a second | |
5354 | * cond_resched() call. | |
5355 | */ | |
1da177e4 LT |
5356 | do { |
5357 | add_preempt_count(PREEMPT_ACTIVE); | |
5358 | schedule(); | |
5359 | sub_preempt_count(PREEMPT_ACTIVE); | |
5360 | } while (need_resched()); | |
5361 | } | |
5362 | ||
02b67cc3 | 5363 | int __sched _cond_resched(void) |
1da177e4 | 5364 | { |
9414232f IM |
5365 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5366 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5367 | __cond_resched(); |
5368 | return 1; | |
5369 | } | |
5370 | return 0; | |
5371 | } | |
02b67cc3 | 5372 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5373 | |
5374 | /* | |
5375 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5376 | * call schedule, and on return reacquire the lock. | |
5377 | * | |
41a2d6cf | 5378 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5379 | * operations here to prevent schedule() from being called twice (once via |
5380 | * spin_unlock(), once by hand). | |
5381 | */ | |
95cdf3b7 | 5382 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5383 | { |
95c354fe | 5384 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5385 | int ret = 0; |
5386 | ||
95c354fe | 5387 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5388 | spin_unlock(lock); |
95c354fe NP |
5389 | if (resched && need_resched()) |
5390 | __cond_resched(); | |
5391 | else | |
5392 | cpu_relax(); | |
6df3cecb | 5393 | ret = 1; |
1da177e4 | 5394 | spin_lock(lock); |
1da177e4 | 5395 | } |
6df3cecb | 5396 | return ret; |
1da177e4 | 5397 | } |
1da177e4 LT |
5398 | EXPORT_SYMBOL(cond_resched_lock); |
5399 | ||
5400 | int __sched cond_resched_softirq(void) | |
5401 | { | |
5402 | BUG_ON(!in_softirq()); | |
5403 | ||
9414232f | 5404 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5405 | local_bh_enable(); |
1da177e4 LT |
5406 | __cond_resched(); |
5407 | local_bh_disable(); | |
5408 | return 1; | |
5409 | } | |
5410 | return 0; | |
5411 | } | |
1da177e4 LT |
5412 | EXPORT_SYMBOL(cond_resched_softirq); |
5413 | ||
1da177e4 LT |
5414 | /** |
5415 | * yield - yield the current processor to other threads. | |
5416 | * | |
72fd4a35 | 5417 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5418 | * thread runnable and calls sys_sched_yield(). |
5419 | */ | |
5420 | void __sched yield(void) | |
5421 | { | |
5422 | set_current_state(TASK_RUNNING); | |
5423 | sys_sched_yield(); | |
5424 | } | |
1da177e4 LT |
5425 | EXPORT_SYMBOL(yield); |
5426 | ||
5427 | /* | |
41a2d6cf | 5428 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5429 | * that process accounting knows that this is a task in IO wait state. |
5430 | * | |
5431 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5432 | * has set its backing_dev_info: the queue against which it should throttle) | |
5433 | */ | |
5434 | void __sched io_schedule(void) | |
5435 | { | |
70b97a7f | 5436 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5437 | |
0ff92245 | 5438 | delayacct_blkio_start(); |
1da177e4 LT |
5439 | atomic_inc(&rq->nr_iowait); |
5440 | schedule(); | |
5441 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5442 | delayacct_blkio_end(); |
1da177e4 | 5443 | } |
1da177e4 LT |
5444 | EXPORT_SYMBOL(io_schedule); |
5445 | ||
5446 | long __sched io_schedule_timeout(long timeout) | |
5447 | { | |
70b97a7f | 5448 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5449 | long ret; |
5450 | ||
0ff92245 | 5451 | delayacct_blkio_start(); |
1da177e4 LT |
5452 | atomic_inc(&rq->nr_iowait); |
5453 | ret = schedule_timeout(timeout); | |
5454 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5455 | delayacct_blkio_end(); |
1da177e4 LT |
5456 | return ret; |
5457 | } | |
5458 | ||
5459 | /** | |
5460 | * sys_sched_get_priority_max - return maximum RT priority. | |
5461 | * @policy: scheduling class. | |
5462 | * | |
5463 | * this syscall returns the maximum rt_priority that can be used | |
5464 | * by a given scheduling class. | |
5465 | */ | |
5466 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5467 | { | |
5468 | int ret = -EINVAL; | |
5469 | ||
5470 | switch (policy) { | |
5471 | case SCHED_FIFO: | |
5472 | case SCHED_RR: | |
5473 | ret = MAX_USER_RT_PRIO-1; | |
5474 | break; | |
5475 | case SCHED_NORMAL: | |
b0a9499c | 5476 | case SCHED_BATCH: |
dd41f596 | 5477 | case SCHED_IDLE: |
1da177e4 LT |
5478 | ret = 0; |
5479 | break; | |
5480 | } | |
5481 | return ret; | |
5482 | } | |
5483 | ||
5484 | /** | |
5485 | * sys_sched_get_priority_min - return minimum RT priority. | |
5486 | * @policy: scheduling class. | |
5487 | * | |
5488 | * this syscall returns the minimum rt_priority that can be used | |
5489 | * by a given scheduling class. | |
5490 | */ | |
5491 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5492 | { | |
5493 | int ret = -EINVAL; | |
5494 | ||
5495 | switch (policy) { | |
5496 | case SCHED_FIFO: | |
5497 | case SCHED_RR: | |
5498 | ret = 1; | |
5499 | break; | |
5500 | case SCHED_NORMAL: | |
b0a9499c | 5501 | case SCHED_BATCH: |
dd41f596 | 5502 | case SCHED_IDLE: |
1da177e4 LT |
5503 | ret = 0; |
5504 | } | |
5505 | return ret; | |
5506 | } | |
5507 | ||
5508 | /** | |
5509 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5510 | * @pid: pid of the process. | |
5511 | * @interval: userspace pointer to the timeslice value. | |
5512 | * | |
5513 | * this syscall writes the default timeslice value of a given process | |
5514 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5515 | */ | |
5516 | asmlinkage | |
5517 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5518 | { | |
36c8b586 | 5519 | struct task_struct *p; |
a4ec24b4 | 5520 | unsigned int time_slice; |
3a5c359a | 5521 | int retval; |
1da177e4 | 5522 | struct timespec t; |
1da177e4 LT |
5523 | |
5524 | if (pid < 0) | |
3a5c359a | 5525 | return -EINVAL; |
1da177e4 LT |
5526 | |
5527 | retval = -ESRCH; | |
5528 | read_lock(&tasklist_lock); | |
5529 | p = find_process_by_pid(pid); | |
5530 | if (!p) | |
5531 | goto out_unlock; | |
5532 | ||
5533 | retval = security_task_getscheduler(p); | |
5534 | if (retval) | |
5535 | goto out_unlock; | |
5536 | ||
77034937 IM |
5537 | /* |
5538 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5539 | * tasks that are on an otherwise idle runqueue: | |
5540 | */ | |
5541 | time_slice = 0; | |
5542 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5543 | time_slice = DEF_TIMESLICE; |
1868f958 | 5544 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5545 | struct sched_entity *se = &p->se; |
5546 | unsigned long flags; | |
5547 | struct rq *rq; | |
5548 | ||
5549 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5550 | if (rq->cfs.load.weight) |
5551 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5552 | task_rq_unlock(rq, &flags); |
5553 | } | |
1da177e4 | 5554 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5555 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5556 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5557 | return retval; |
3a5c359a | 5558 | |
1da177e4 LT |
5559 | out_unlock: |
5560 | read_unlock(&tasklist_lock); | |
5561 | return retval; | |
5562 | } | |
5563 | ||
7c731e0a | 5564 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5565 | |
82a1fcb9 | 5566 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5567 | { |
1da177e4 | 5568 | unsigned long free = 0; |
36c8b586 | 5569 | unsigned state; |
1da177e4 | 5570 | |
1da177e4 | 5571 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5572 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5573 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5574 | #if BITS_PER_LONG == 32 |
1da177e4 | 5575 | if (state == TASK_RUNNING) |
cc4ea795 | 5576 | printk(KERN_CONT " running "); |
1da177e4 | 5577 | else |
cc4ea795 | 5578 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5579 | #else |
5580 | if (state == TASK_RUNNING) | |
cc4ea795 | 5581 | printk(KERN_CONT " running task "); |
1da177e4 | 5582 | else |
cc4ea795 | 5583 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5584 | #endif |
5585 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5586 | { | |
10ebffde | 5587 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5588 | while (!*n) |
5589 | n++; | |
10ebffde | 5590 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5591 | } |
5592 | #endif | |
ba25f9dc | 5593 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5594 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5595 | |
5fb5e6de | 5596 | show_stack(p, NULL); |
1da177e4 LT |
5597 | } |
5598 | ||
e59e2ae2 | 5599 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5600 | { |
36c8b586 | 5601 | struct task_struct *g, *p; |
1da177e4 | 5602 | |
4bd77321 IM |
5603 | #if BITS_PER_LONG == 32 |
5604 | printk(KERN_INFO | |
5605 | " task PC stack pid father\n"); | |
1da177e4 | 5606 | #else |
4bd77321 IM |
5607 | printk(KERN_INFO |
5608 | " task PC stack pid father\n"); | |
1da177e4 LT |
5609 | #endif |
5610 | read_lock(&tasklist_lock); | |
5611 | do_each_thread(g, p) { | |
5612 | /* | |
5613 | * reset the NMI-timeout, listing all files on a slow | |
5614 | * console might take alot of time: | |
5615 | */ | |
5616 | touch_nmi_watchdog(); | |
39bc89fd | 5617 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5618 | sched_show_task(p); |
1da177e4 LT |
5619 | } while_each_thread(g, p); |
5620 | ||
04c9167f JF |
5621 | touch_all_softlockup_watchdogs(); |
5622 | ||
dd41f596 IM |
5623 | #ifdef CONFIG_SCHED_DEBUG |
5624 | sysrq_sched_debug_show(); | |
5625 | #endif | |
1da177e4 | 5626 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5627 | /* |
5628 | * Only show locks if all tasks are dumped: | |
5629 | */ | |
5630 | if (state_filter == -1) | |
5631 | debug_show_all_locks(); | |
1da177e4 LT |
5632 | } |
5633 | ||
1df21055 IM |
5634 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5635 | { | |
dd41f596 | 5636 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5637 | } |
5638 | ||
f340c0d1 IM |
5639 | /** |
5640 | * init_idle - set up an idle thread for a given CPU | |
5641 | * @idle: task in question | |
5642 | * @cpu: cpu the idle task belongs to | |
5643 | * | |
5644 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5645 | * flag, to make booting more robust. | |
5646 | */ | |
5c1e1767 | 5647 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5648 | { |
70b97a7f | 5649 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5650 | unsigned long flags; |
5651 | ||
dd41f596 IM |
5652 | __sched_fork(idle); |
5653 | idle->se.exec_start = sched_clock(); | |
5654 | ||
b29739f9 | 5655 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5656 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5657 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5658 | |
5659 | spin_lock_irqsave(&rq->lock, flags); | |
5660 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5661 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5662 | idle->oncpu = 1; | |
5663 | #endif | |
1da177e4 LT |
5664 | spin_unlock_irqrestore(&rq->lock, flags); |
5665 | ||
5666 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5667 | #if defined(CONFIG_PREEMPT) |
5668 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5669 | #else | |
a1261f54 | 5670 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5671 | #endif |
dd41f596 IM |
5672 | /* |
5673 | * The idle tasks have their own, simple scheduling class: | |
5674 | */ | |
5675 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5676 | } |
5677 | ||
5678 | /* | |
5679 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5680 | * indicates which cpus entered this state. This is used | |
5681 | * in the rcu update to wait only for active cpus. For system | |
5682 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5683 | * always be CPU_MASK_NONE. | |
5684 | */ | |
5685 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5686 | ||
19978ca6 IM |
5687 | /* |
5688 | * Increase the granularity value when there are more CPUs, | |
5689 | * because with more CPUs the 'effective latency' as visible | |
5690 | * to users decreases. But the relationship is not linear, | |
5691 | * so pick a second-best guess by going with the log2 of the | |
5692 | * number of CPUs. | |
5693 | * | |
5694 | * This idea comes from the SD scheduler of Con Kolivas: | |
5695 | */ | |
5696 | static inline void sched_init_granularity(void) | |
5697 | { | |
5698 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5699 | const unsigned long limit = 200000000; | |
5700 | ||
5701 | sysctl_sched_min_granularity *= factor; | |
5702 | if (sysctl_sched_min_granularity > limit) | |
5703 | sysctl_sched_min_granularity = limit; | |
5704 | ||
5705 | sysctl_sched_latency *= factor; | |
5706 | if (sysctl_sched_latency > limit) | |
5707 | sysctl_sched_latency = limit; | |
5708 | ||
5709 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5710 | } |
5711 | ||
1da177e4 LT |
5712 | #ifdef CONFIG_SMP |
5713 | /* | |
5714 | * This is how migration works: | |
5715 | * | |
70b97a7f | 5716 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5717 | * runqueue and wake up that CPU's migration thread. |
5718 | * 2) we down() the locked semaphore => thread blocks. | |
5719 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5720 | * thread off the CPU) | |
5721 | * 4) it gets the migration request and checks whether the migrated | |
5722 | * task is still in the wrong runqueue. | |
5723 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5724 | * it and puts it into the right queue. | |
5725 | * 6) migration thread up()s the semaphore. | |
5726 | * 7) we wake up and the migration is done. | |
5727 | */ | |
5728 | ||
5729 | /* | |
5730 | * Change a given task's CPU affinity. Migrate the thread to a | |
5731 | * proper CPU and schedule it away if the CPU it's executing on | |
5732 | * is removed from the allowed bitmask. | |
5733 | * | |
5734 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5735 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5736 | * call is not atomic; no spinlocks may be held. |
5737 | */ | |
cd8ba7cd | 5738 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5739 | { |
70b97a7f | 5740 | struct migration_req req; |
1da177e4 | 5741 | unsigned long flags; |
70b97a7f | 5742 | struct rq *rq; |
48f24c4d | 5743 | int ret = 0; |
1da177e4 LT |
5744 | |
5745 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5746 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5747 | ret = -EINVAL; |
5748 | goto out; | |
5749 | } | |
5750 | ||
9985b0ba DR |
5751 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5752 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5753 | ret = -EINVAL; | |
5754 | goto out; | |
5755 | } | |
5756 | ||
73fe6aae | 5757 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5758 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5759 | else { |
cd8ba7cd MT |
5760 | p->cpus_allowed = *new_mask; |
5761 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5762 | } |
5763 | ||
1da177e4 | 5764 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5765 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5766 | goto out; |
5767 | ||
cd8ba7cd | 5768 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5769 | /* Need help from migration thread: drop lock and wait. */ |
5770 | task_rq_unlock(rq, &flags); | |
5771 | wake_up_process(rq->migration_thread); | |
5772 | wait_for_completion(&req.done); | |
5773 | tlb_migrate_finish(p->mm); | |
5774 | return 0; | |
5775 | } | |
5776 | out: | |
5777 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5778 | |
1da177e4 LT |
5779 | return ret; |
5780 | } | |
cd8ba7cd | 5781 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5782 | |
5783 | /* | |
41a2d6cf | 5784 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5785 | * this because either it can't run here any more (set_cpus_allowed() |
5786 | * away from this CPU, or CPU going down), or because we're | |
5787 | * attempting to rebalance this task on exec (sched_exec). | |
5788 | * | |
5789 | * So we race with normal scheduler movements, but that's OK, as long | |
5790 | * as the task is no longer on this CPU. | |
efc30814 KK |
5791 | * |
5792 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5793 | */ |
efc30814 | 5794 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5795 | { |
70b97a7f | 5796 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5797 | int ret = 0, on_rq; |
1da177e4 | 5798 | |
e761b772 | 5799 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5800 | return ret; |
1da177e4 LT |
5801 | |
5802 | rq_src = cpu_rq(src_cpu); | |
5803 | rq_dest = cpu_rq(dest_cpu); | |
5804 | ||
5805 | double_rq_lock(rq_src, rq_dest); | |
5806 | /* Already moved. */ | |
5807 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5808 | goto done; |
1da177e4 LT |
5809 | /* Affinity changed (again). */ |
5810 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 5811 | goto fail; |
1da177e4 | 5812 | |
dd41f596 | 5813 | on_rq = p->se.on_rq; |
6e82a3be | 5814 | if (on_rq) |
2e1cb74a | 5815 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5816 | |
1da177e4 | 5817 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5818 | if (on_rq) { |
5819 | activate_task(rq_dest, p, 0); | |
5820 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5821 | } |
b1e38734 | 5822 | done: |
efc30814 | 5823 | ret = 1; |
b1e38734 | 5824 | fail: |
1da177e4 | 5825 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5826 | return ret; |
1da177e4 LT |
5827 | } |
5828 | ||
5829 | /* | |
5830 | * migration_thread - this is a highprio system thread that performs | |
5831 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5832 | * another runqueue. | |
5833 | */ | |
95cdf3b7 | 5834 | static int migration_thread(void *data) |
1da177e4 | 5835 | { |
1da177e4 | 5836 | int cpu = (long)data; |
70b97a7f | 5837 | struct rq *rq; |
1da177e4 LT |
5838 | |
5839 | rq = cpu_rq(cpu); | |
5840 | BUG_ON(rq->migration_thread != current); | |
5841 | ||
5842 | set_current_state(TASK_INTERRUPTIBLE); | |
5843 | while (!kthread_should_stop()) { | |
70b97a7f | 5844 | struct migration_req *req; |
1da177e4 | 5845 | struct list_head *head; |
1da177e4 | 5846 | |
1da177e4 LT |
5847 | spin_lock_irq(&rq->lock); |
5848 | ||
5849 | if (cpu_is_offline(cpu)) { | |
5850 | spin_unlock_irq(&rq->lock); | |
5851 | goto wait_to_die; | |
5852 | } | |
5853 | ||
5854 | if (rq->active_balance) { | |
5855 | active_load_balance(rq, cpu); | |
5856 | rq->active_balance = 0; | |
5857 | } | |
5858 | ||
5859 | head = &rq->migration_queue; | |
5860 | ||
5861 | if (list_empty(head)) { | |
5862 | spin_unlock_irq(&rq->lock); | |
5863 | schedule(); | |
5864 | set_current_state(TASK_INTERRUPTIBLE); | |
5865 | continue; | |
5866 | } | |
70b97a7f | 5867 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5868 | list_del_init(head->next); |
5869 | ||
674311d5 NP |
5870 | spin_unlock(&rq->lock); |
5871 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5872 | local_irq_enable(); | |
1da177e4 LT |
5873 | |
5874 | complete(&req->done); | |
5875 | } | |
5876 | __set_current_state(TASK_RUNNING); | |
5877 | return 0; | |
5878 | ||
5879 | wait_to_die: | |
5880 | /* Wait for kthread_stop */ | |
5881 | set_current_state(TASK_INTERRUPTIBLE); | |
5882 | while (!kthread_should_stop()) { | |
5883 | schedule(); | |
5884 | set_current_state(TASK_INTERRUPTIBLE); | |
5885 | } | |
5886 | __set_current_state(TASK_RUNNING); | |
5887 | return 0; | |
5888 | } | |
5889 | ||
5890 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5891 | |
5892 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5893 | { | |
5894 | int ret; | |
5895 | ||
5896 | local_irq_disable(); | |
5897 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5898 | local_irq_enable(); | |
5899 | return ret; | |
5900 | } | |
5901 | ||
054b9108 | 5902 | /* |
3a4fa0a2 | 5903 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5904 | * NOTE: interrupts should be disabled by the caller |
5905 | */ | |
48f24c4d | 5906 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5907 | { |
efc30814 | 5908 | unsigned long flags; |
1da177e4 | 5909 | cpumask_t mask; |
70b97a7f IM |
5910 | struct rq *rq; |
5911 | int dest_cpu; | |
1da177e4 | 5912 | |
3a5c359a AK |
5913 | do { |
5914 | /* On same node? */ | |
5915 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5916 | cpus_and(mask, mask, p->cpus_allowed); | |
5917 | dest_cpu = any_online_cpu(mask); | |
5918 | ||
5919 | /* On any allowed CPU? */ | |
434d53b0 | 5920 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5921 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5922 | ||
5923 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5924 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5925 | cpumask_t cpus_allowed; |
5926 | ||
5927 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5928 | /* |
5929 | * Try to stay on the same cpuset, where the | |
5930 | * current cpuset may be a subset of all cpus. | |
5931 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5932 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5933 | * called within calls to cpuset_lock/cpuset_unlock. |
5934 | */ | |
3a5c359a | 5935 | rq = task_rq_lock(p, &flags); |
470fd646 | 5936 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5937 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5938 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5939 | |
3a5c359a AK |
5940 | /* |
5941 | * Don't tell them about moving exiting tasks or | |
5942 | * kernel threads (both mm NULL), since they never | |
5943 | * leave kernel. | |
5944 | */ | |
41a2d6cf | 5945 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5946 | printk(KERN_INFO "process %d (%s) no " |
5947 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5948 | task_pid_nr(p), p->comm, dead_cpu); |
5949 | } | |
3a5c359a | 5950 | } |
f7b4cddc | 5951 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5952 | } |
5953 | ||
5954 | /* | |
5955 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5956 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5957 | * for performance reasons the counter is not stricly tracking tasks to | |
5958 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5959 | * to keep the global sum constant after CPU-down: | |
5960 | */ | |
70b97a7f | 5961 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5962 | { |
7c16ec58 | 5963 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5964 | unsigned long flags; |
5965 | ||
5966 | local_irq_save(flags); | |
5967 | double_rq_lock(rq_src, rq_dest); | |
5968 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5969 | rq_src->nr_uninterruptible = 0; | |
5970 | double_rq_unlock(rq_src, rq_dest); | |
5971 | local_irq_restore(flags); | |
5972 | } | |
5973 | ||
5974 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5975 | static void migrate_live_tasks(int src_cpu) | |
5976 | { | |
48f24c4d | 5977 | struct task_struct *p, *t; |
1da177e4 | 5978 | |
f7b4cddc | 5979 | read_lock(&tasklist_lock); |
1da177e4 | 5980 | |
48f24c4d IM |
5981 | do_each_thread(t, p) { |
5982 | if (p == current) | |
1da177e4 LT |
5983 | continue; |
5984 | ||
48f24c4d IM |
5985 | if (task_cpu(p) == src_cpu) |
5986 | move_task_off_dead_cpu(src_cpu, p); | |
5987 | } while_each_thread(t, p); | |
1da177e4 | 5988 | |
f7b4cddc | 5989 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5990 | } |
5991 | ||
dd41f596 IM |
5992 | /* |
5993 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5994 | * It does so by boosting its priority to highest possible. |
5995 | * Used by CPU offline code. | |
1da177e4 LT |
5996 | */ |
5997 | void sched_idle_next(void) | |
5998 | { | |
48f24c4d | 5999 | int this_cpu = smp_processor_id(); |
70b97a7f | 6000 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6001 | struct task_struct *p = rq->idle; |
6002 | unsigned long flags; | |
6003 | ||
6004 | /* cpu has to be offline */ | |
48f24c4d | 6005 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6006 | |
48f24c4d IM |
6007 | /* |
6008 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6009 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6010 | */ |
6011 | spin_lock_irqsave(&rq->lock, flags); | |
6012 | ||
dd41f596 | 6013 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6014 | |
94bc9a7b DA |
6015 | update_rq_clock(rq); |
6016 | activate_task(rq, p, 0); | |
1da177e4 LT |
6017 | |
6018 | spin_unlock_irqrestore(&rq->lock, flags); | |
6019 | } | |
6020 | ||
48f24c4d IM |
6021 | /* |
6022 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6023 | * offline. |
6024 | */ | |
6025 | void idle_task_exit(void) | |
6026 | { | |
6027 | struct mm_struct *mm = current->active_mm; | |
6028 | ||
6029 | BUG_ON(cpu_online(smp_processor_id())); | |
6030 | ||
6031 | if (mm != &init_mm) | |
6032 | switch_mm(mm, &init_mm, current); | |
6033 | mmdrop(mm); | |
6034 | } | |
6035 | ||
054b9108 | 6036 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6037 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6038 | { |
70b97a7f | 6039 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6040 | |
6041 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6042 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6043 | |
6044 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6045 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6046 | |
48f24c4d | 6047 | get_task_struct(p); |
1da177e4 LT |
6048 | |
6049 | /* | |
6050 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6051 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6052 | * fine. |
6053 | */ | |
f7b4cddc | 6054 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6055 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6056 | spin_lock_irq(&rq->lock); |
1da177e4 | 6057 | |
48f24c4d | 6058 | put_task_struct(p); |
1da177e4 LT |
6059 | } |
6060 | ||
6061 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6062 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6063 | { | |
70b97a7f | 6064 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6065 | struct task_struct *next; |
48f24c4d | 6066 | |
dd41f596 IM |
6067 | for ( ; ; ) { |
6068 | if (!rq->nr_running) | |
6069 | break; | |
a8e504d2 | 6070 | update_rq_clock(rq); |
ff95f3df | 6071 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6072 | if (!next) |
6073 | break; | |
79c53799 | 6074 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6075 | migrate_dead(dead_cpu, next); |
e692ab53 | 6076 | |
1da177e4 LT |
6077 | } |
6078 | } | |
6079 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6080 | ||
e692ab53 NP |
6081 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6082 | ||
6083 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6084 | { |
6085 | .procname = "sched_domain", | |
c57baf1e | 6086 | .mode = 0555, |
e0361851 | 6087 | }, |
38605cae | 6088 | {0, }, |
e692ab53 NP |
6089 | }; |
6090 | ||
6091 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6092 | { |
c57baf1e | 6093 | .ctl_name = CTL_KERN, |
e0361851 | 6094 | .procname = "kernel", |
c57baf1e | 6095 | .mode = 0555, |
e0361851 AD |
6096 | .child = sd_ctl_dir, |
6097 | }, | |
38605cae | 6098 | {0, }, |
e692ab53 NP |
6099 | }; |
6100 | ||
6101 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6102 | { | |
6103 | struct ctl_table *entry = | |
5cf9f062 | 6104 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6105 | |
e692ab53 NP |
6106 | return entry; |
6107 | } | |
6108 | ||
6382bc90 MM |
6109 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6110 | { | |
cd790076 | 6111 | struct ctl_table *entry; |
6382bc90 | 6112 | |
cd790076 MM |
6113 | /* |
6114 | * In the intermediate directories, both the child directory and | |
6115 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6116 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6117 | * static strings and all have proc handlers. |
6118 | */ | |
6119 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6120 | if (entry->child) |
6121 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6122 | if (entry->proc_handler == NULL) |
6123 | kfree(entry->procname); | |
6124 | } | |
6382bc90 MM |
6125 | |
6126 | kfree(*tablep); | |
6127 | *tablep = NULL; | |
6128 | } | |
6129 | ||
e692ab53 | 6130 | static void |
e0361851 | 6131 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6132 | const char *procname, void *data, int maxlen, |
6133 | mode_t mode, proc_handler *proc_handler) | |
6134 | { | |
e692ab53 NP |
6135 | entry->procname = procname; |
6136 | entry->data = data; | |
6137 | entry->maxlen = maxlen; | |
6138 | entry->mode = mode; | |
6139 | entry->proc_handler = proc_handler; | |
6140 | } | |
6141 | ||
6142 | static struct ctl_table * | |
6143 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6144 | { | |
ace8b3d6 | 6145 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6146 | |
ad1cdc1d MM |
6147 | if (table == NULL) |
6148 | return NULL; | |
6149 | ||
e0361851 | 6150 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6151 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6152 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6153 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6154 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6155 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6156 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6157 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6158 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6159 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6160 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6161 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6162 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6163 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6164 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6165 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6166 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6167 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6168 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6169 | &sd->cache_nice_tries, |
6170 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6171 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6172 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6173 | /* &table[11] is terminator */ |
e692ab53 NP |
6174 | |
6175 | return table; | |
6176 | } | |
6177 | ||
9a4e7159 | 6178 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6179 | { |
6180 | struct ctl_table *entry, *table; | |
6181 | struct sched_domain *sd; | |
6182 | int domain_num = 0, i; | |
6183 | char buf[32]; | |
6184 | ||
6185 | for_each_domain(cpu, sd) | |
6186 | domain_num++; | |
6187 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6188 | if (table == NULL) |
6189 | return NULL; | |
e692ab53 NP |
6190 | |
6191 | i = 0; | |
6192 | for_each_domain(cpu, sd) { | |
6193 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6194 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6195 | entry->mode = 0555; |
e692ab53 NP |
6196 | entry->child = sd_alloc_ctl_domain_table(sd); |
6197 | entry++; | |
6198 | i++; | |
6199 | } | |
6200 | return table; | |
6201 | } | |
6202 | ||
6203 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6204 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6205 | { |
6206 | int i, cpu_num = num_online_cpus(); | |
6207 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6208 | char buf[32]; | |
6209 | ||
7378547f MM |
6210 | WARN_ON(sd_ctl_dir[0].child); |
6211 | sd_ctl_dir[0].child = entry; | |
6212 | ||
ad1cdc1d MM |
6213 | if (entry == NULL) |
6214 | return; | |
6215 | ||
97b6ea7b | 6216 | for_each_online_cpu(i) { |
e692ab53 | 6217 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6218 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6219 | entry->mode = 0555; |
e692ab53 | 6220 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6221 | entry++; |
e692ab53 | 6222 | } |
7378547f MM |
6223 | |
6224 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6225 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6226 | } | |
6382bc90 | 6227 | |
7378547f | 6228 | /* may be called multiple times per register */ |
6382bc90 MM |
6229 | static void unregister_sched_domain_sysctl(void) |
6230 | { | |
7378547f MM |
6231 | if (sd_sysctl_header) |
6232 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6233 | sd_sysctl_header = NULL; |
7378547f MM |
6234 | if (sd_ctl_dir[0].child) |
6235 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6236 | } |
e692ab53 | 6237 | #else |
6382bc90 MM |
6238 | static void register_sched_domain_sysctl(void) |
6239 | { | |
6240 | } | |
6241 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6242 | { |
6243 | } | |
6244 | #endif | |
6245 | ||
1f11eb6a GH |
6246 | static void set_rq_online(struct rq *rq) |
6247 | { | |
6248 | if (!rq->online) { | |
6249 | const struct sched_class *class; | |
6250 | ||
6251 | cpu_set(rq->cpu, rq->rd->online); | |
6252 | rq->online = 1; | |
6253 | ||
6254 | for_each_class(class) { | |
6255 | if (class->rq_online) | |
6256 | class->rq_online(rq); | |
6257 | } | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | static void set_rq_offline(struct rq *rq) | |
6262 | { | |
6263 | if (rq->online) { | |
6264 | const struct sched_class *class; | |
6265 | ||
6266 | for_each_class(class) { | |
6267 | if (class->rq_offline) | |
6268 | class->rq_offline(rq); | |
6269 | } | |
6270 | ||
6271 | cpu_clear(rq->cpu, rq->rd->online); | |
6272 | rq->online = 0; | |
6273 | } | |
6274 | } | |
6275 | ||
1da177e4 LT |
6276 | /* |
6277 | * migration_call - callback that gets triggered when a CPU is added. | |
6278 | * Here we can start up the necessary migration thread for the new CPU. | |
6279 | */ | |
48f24c4d IM |
6280 | static int __cpuinit |
6281 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6282 | { |
1da177e4 | 6283 | struct task_struct *p; |
48f24c4d | 6284 | int cpu = (long)hcpu; |
1da177e4 | 6285 | unsigned long flags; |
70b97a7f | 6286 | struct rq *rq; |
1da177e4 LT |
6287 | |
6288 | switch (action) { | |
5be9361c | 6289 | |
1da177e4 | 6290 | case CPU_UP_PREPARE: |
8bb78442 | 6291 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6292 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6293 | if (IS_ERR(p)) |
6294 | return NOTIFY_BAD; | |
1da177e4 LT |
6295 | kthread_bind(p, cpu); |
6296 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6297 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6298 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6299 | task_rq_unlock(rq, &flags); |
6300 | cpu_rq(cpu)->migration_thread = p; | |
6301 | break; | |
48f24c4d | 6302 | |
1da177e4 | 6303 | case CPU_ONLINE: |
8bb78442 | 6304 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6305 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6306 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6307 | |
6308 | /* Update our root-domain */ | |
6309 | rq = cpu_rq(cpu); | |
6310 | spin_lock_irqsave(&rq->lock, flags); | |
6311 | if (rq->rd) { | |
6312 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6313 | |
6314 | set_rq_online(rq); | |
1f94ef59 GH |
6315 | } |
6316 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6317 | break; |
48f24c4d | 6318 | |
1da177e4 LT |
6319 | #ifdef CONFIG_HOTPLUG_CPU |
6320 | case CPU_UP_CANCELED: | |
8bb78442 | 6321 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6322 | if (!cpu_rq(cpu)->migration_thread) |
6323 | break; | |
41a2d6cf | 6324 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6325 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6326 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6327 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6328 | cpu_rq(cpu)->migration_thread = NULL; | |
6329 | break; | |
48f24c4d | 6330 | |
1da177e4 | 6331 | case CPU_DEAD: |
8bb78442 | 6332 | case CPU_DEAD_FROZEN: |
470fd646 | 6333 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6334 | migrate_live_tasks(cpu); |
6335 | rq = cpu_rq(cpu); | |
6336 | kthread_stop(rq->migration_thread); | |
6337 | rq->migration_thread = NULL; | |
6338 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6339 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6340 | update_rq_clock(rq); |
2e1cb74a | 6341 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6342 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6343 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6344 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6345 | migrate_dead_tasks(cpu); |
d2da272a | 6346 | spin_unlock_irq(&rq->lock); |
470fd646 | 6347 | cpuset_unlock(); |
1da177e4 LT |
6348 | migrate_nr_uninterruptible(rq); |
6349 | BUG_ON(rq->nr_running != 0); | |
6350 | ||
41a2d6cf IM |
6351 | /* |
6352 | * No need to migrate the tasks: it was best-effort if | |
6353 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6354 | * the requestors. | |
6355 | */ | |
1da177e4 LT |
6356 | spin_lock_irq(&rq->lock); |
6357 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6358 | struct migration_req *req; |
6359 | ||
1da177e4 | 6360 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6361 | struct migration_req, list); |
1da177e4 LT |
6362 | list_del_init(&req->list); |
6363 | complete(&req->done); | |
6364 | } | |
6365 | spin_unlock_irq(&rq->lock); | |
6366 | break; | |
57d885fe | 6367 | |
08f503b0 GH |
6368 | case CPU_DYING: |
6369 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6370 | /* Update our root-domain */ |
6371 | rq = cpu_rq(cpu); | |
6372 | spin_lock_irqsave(&rq->lock, flags); | |
6373 | if (rq->rd) { | |
6374 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6375 | set_rq_offline(rq); |
57d885fe GH |
6376 | } |
6377 | spin_unlock_irqrestore(&rq->lock, flags); | |
6378 | break; | |
1da177e4 LT |
6379 | #endif |
6380 | } | |
6381 | return NOTIFY_OK; | |
6382 | } | |
6383 | ||
6384 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6385 | * happens before everything else. | |
6386 | */ | |
26c2143b | 6387 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6388 | .notifier_call = migration_call, |
6389 | .priority = 10 | |
6390 | }; | |
6391 | ||
e6fe6649 | 6392 | void __init migration_init(void) |
1da177e4 LT |
6393 | { |
6394 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6395 | int err; |
48f24c4d IM |
6396 | |
6397 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6398 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6399 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6400 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6401 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6402 | } |
6403 | #endif | |
6404 | ||
6405 | #ifdef CONFIG_SMP | |
476f3534 | 6406 | |
3e9830dc | 6407 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6408 | |
099f98c8 GS |
6409 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6410 | { | |
6411 | switch (lvl) { | |
6412 | case SD_LV_NONE: | |
6413 | return "NONE"; | |
6414 | case SD_LV_SIBLING: | |
6415 | return "SIBLING"; | |
6416 | case SD_LV_MC: | |
6417 | return "MC"; | |
6418 | case SD_LV_CPU: | |
6419 | return "CPU"; | |
6420 | case SD_LV_NODE: | |
6421 | return "NODE"; | |
6422 | case SD_LV_ALLNODES: | |
6423 | return "ALLNODES"; | |
6424 | case SD_LV_MAX: | |
6425 | return "MAX"; | |
6426 | ||
6427 | } | |
6428 | return "MAX"; | |
6429 | } | |
6430 | ||
7c16ec58 MT |
6431 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6432 | cpumask_t *groupmask) | |
1da177e4 | 6433 | { |
4dcf6aff | 6434 | struct sched_group *group = sd->groups; |
434d53b0 | 6435 | char str[256]; |
1da177e4 | 6436 | |
434d53b0 | 6437 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6438 | cpus_clear(*groupmask); |
4dcf6aff IM |
6439 | |
6440 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6441 | ||
6442 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6443 | printk("does not load-balance\n"); | |
6444 | if (sd->parent) | |
6445 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6446 | " has parent"); | |
6447 | return -1; | |
41c7ce9a NP |
6448 | } |
6449 | ||
099f98c8 GS |
6450 | printk(KERN_CONT "span %s level %s\n", |
6451 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6452 | |
6453 | if (!cpu_isset(cpu, sd->span)) { | |
6454 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6455 | "CPU%d\n", cpu); | |
6456 | } | |
6457 | if (!cpu_isset(cpu, group->cpumask)) { | |
6458 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6459 | " CPU%d\n", cpu); | |
6460 | } | |
1da177e4 | 6461 | |
4dcf6aff | 6462 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6463 | do { |
4dcf6aff IM |
6464 | if (!group) { |
6465 | printk("\n"); | |
6466 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6467 | break; |
6468 | } | |
6469 | ||
4dcf6aff IM |
6470 | if (!group->__cpu_power) { |
6471 | printk(KERN_CONT "\n"); | |
6472 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6473 | "set\n"); | |
6474 | break; | |
6475 | } | |
1da177e4 | 6476 | |
4dcf6aff IM |
6477 | if (!cpus_weight(group->cpumask)) { |
6478 | printk(KERN_CONT "\n"); | |
6479 | printk(KERN_ERR "ERROR: empty group\n"); | |
6480 | break; | |
6481 | } | |
1da177e4 | 6482 | |
7c16ec58 | 6483 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6484 | printk(KERN_CONT "\n"); |
6485 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6486 | break; | |
6487 | } | |
1da177e4 | 6488 | |
7c16ec58 | 6489 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6490 | |
434d53b0 | 6491 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6492 | printk(KERN_CONT " %s", str); |
1da177e4 | 6493 | |
4dcf6aff IM |
6494 | group = group->next; |
6495 | } while (group != sd->groups); | |
6496 | printk(KERN_CONT "\n"); | |
1da177e4 | 6497 | |
7c16ec58 | 6498 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6499 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6500 | |
7c16ec58 | 6501 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6502 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6503 | "of domain->span\n"); | |
6504 | return 0; | |
6505 | } | |
1da177e4 | 6506 | |
4dcf6aff IM |
6507 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6508 | { | |
7c16ec58 | 6509 | cpumask_t *groupmask; |
4dcf6aff | 6510 | int level = 0; |
1da177e4 | 6511 | |
4dcf6aff IM |
6512 | if (!sd) { |
6513 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6514 | return; | |
6515 | } | |
1da177e4 | 6516 | |
4dcf6aff IM |
6517 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6518 | ||
7c16ec58 MT |
6519 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6520 | if (!groupmask) { | |
6521 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6522 | return; | |
6523 | } | |
6524 | ||
4dcf6aff | 6525 | for (;;) { |
7c16ec58 | 6526 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6527 | break; |
1da177e4 LT |
6528 | level++; |
6529 | sd = sd->parent; | |
33859f7f | 6530 | if (!sd) |
4dcf6aff IM |
6531 | break; |
6532 | } | |
7c16ec58 | 6533 | kfree(groupmask); |
1da177e4 | 6534 | } |
6d6bc0ad | 6535 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6536 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6537 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6538 | |
1a20ff27 | 6539 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6540 | { |
6541 | if (cpus_weight(sd->span) == 1) | |
6542 | return 1; | |
6543 | ||
6544 | /* Following flags need at least 2 groups */ | |
6545 | if (sd->flags & (SD_LOAD_BALANCE | | |
6546 | SD_BALANCE_NEWIDLE | | |
6547 | SD_BALANCE_FORK | | |
89c4710e SS |
6548 | SD_BALANCE_EXEC | |
6549 | SD_SHARE_CPUPOWER | | |
6550 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6551 | if (sd->groups != sd->groups->next) |
6552 | return 0; | |
6553 | } | |
6554 | ||
6555 | /* Following flags don't use groups */ | |
6556 | if (sd->flags & (SD_WAKE_IDLE | | |
6557 | SD_WAKE_AFFINE | | |
6558 | SD_WAKE_BALANCE)) | |
6559 | return 0; | |
6560 | ||
6561 | return 1; | |
6562 | } | |
6563 | ||
48f24c4d IM |
6564 | static int |
6565 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6566 | { |
6567 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6568 | ||
6569 | if (sd_degenerate(parent)) | |
6570 | return 1; | |
6571 | ||
6572 | if (!cpus_equal(sd->span, parent->span)) | |
6573 | return 0; | |
6574 | ||
6575 | /* Does parent contain flags not in child? */ | |
6576 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6577 | if (cflags & SD_WAKE_AFFINE) | |
6578 | pflags &= ~SD_WAKE_BALANCE; | |
6579 | /* Flags needing groups don't count if only 1 group in parent */ | |
6580 | if (parent->groups == parent->groups->next) { | |
6581 | pflags &= ~(SD_LOAD_BALANCE | | |
6582 | SD_BALANCE_NEWIDLE | | |
6583 | SD_BALANCE_FORK | | |
89c4710e SS |
6584 | SD_BALANCE_EXEC | |
6585 | SD_SHARE_CPUPOWER | | |
6586 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6587 | } |
6588 | if (~cflags & pflags) | |
6589 | return 0; | |
6590 | ||
6591 | return 1; | |
6592 | } | |
6593 | ||
57d885fe GH |
6594 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6595 | { | |
6596 | unsigned long flags; | |
57d885fe GH |
6597 | |
6598 | spin_lock_irqsave(&rq->lock, flags); | |
6599 | ||
6600 | if (rq->rd) { | |
6601 | struct root_domain *old_rd = rq->rd; | |
6602 | ||
1f11eb6a GH |
6603 | if (cpu_isset(rq->cpu, old_rd->online)) |
6604 | set_rq_offline(rq); | |
57d885fe | 6605 | |
dc938520 | 6606 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6607 | |
57d885fe GH |
6608 | if (atomic_dec_and_test(&old_rd->refcount)) |
6609 | kfree(old_rd); | |
6610 | } | |
6611 | ||
6612 | atomic_inc(&rd->refcount); | |
6613 | rq->rd = rd; | |
6614 | ||
dc938520 | 6615 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6616 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6617 | set_rq_online(rq); |
57d885fe GH |
6618 | |
6619 | spin_unlock_irqrestore(&rq->lock, flags); | |
6620 | } | |
6621 | ||
dc938520 | 6622 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6623 | { |
6624 | memset(rd, 0, sizeof(*rd)); | |
6625 | ||
dc938520 GH |
6626 | cpus_clear(rd->span); |
6627 | cpus_clear(rd->online); | |
6e0534f2 GH |
6628 | |
6629 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6630 | } |
6631 | ||
6632 | static void init_defrootdomain(void) | |
6633 | { | |
dc938520 | 6634 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6635 | atomic_set(&def_root_domain.refcount, 1); |
6636 | } | |
6637 | ||
dc938520 | 6638 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6639 | { |
6640 | struct root_domain *rd; | |
6641 | ||
6642 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6643 | if (!rd) | |
6644 | return NULL; | |
6645 | ||
dc938520 | 6646 | init_rootdomain(rd); |
57d885fe GH |
6647 | |
6648 | return rd; | |
6649 | } | |
6650 | ||
1da177e4 | 6651 | /* |
0eab9146 | 6652 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6653 | * hold the hotplug lock. |
6654 | */ | |
0eab9146 IM |
6655 | static void |
6656 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6657 | { |
70b97a7f | 6658 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6659 | struct sched_domain *tmp; |
6660 | ||
6661 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6662 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6663 | struct sched_domain *parent = tmp->parent; | |
6664 | if (!parent) | |
6665 | break; | |
1a848870 | 6666 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6667 | tmp->parent = parent->parent; |
1a848870 SS |
6668 | if (parent->parent) |
6669 | parent->parent->child = tmp; | |
6670 | } | |
245af2c7 SS |
6671 | } |
6672 | ||
1a848870 | 6673 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6674 | sd = sd->parent; |
1a848870 SS |
6675 | if (sd) |
6676 | sd->child = NULL; | |
6677 | } | |
1da177e4 LT |
6678 | |
6679 | sched_domain_debug(sd, cpu); | |
6680 | ||
57d885fe | 6681 | rq_attach_root(rq, rd); |
674311d5 | 6682 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6683 | } |
6684 | ||
6685 | /* cpus with isolated domains */ | |
67af63a6 | 6686 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6687 | |
6688 | /* Setup the mask of cpus configured for isolated domains */ | |
6689 | static int __init isolated_cpu_setup(char *str) | |
6690 | { | |
13b40c1e MT |
6691 | static int __initdata ints[NR_CPUS]; |
6692 | int i; | |
1da177e4 LT |
6693 | |
6694 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6695 | cpus_clear(cpu_isolated_map); | |
6696 | for (i = 1; i <= ints[0]; i++) | |
6697 | if (ints[i] < NR_CPUS) | |
6698 | cpu_set(ints[i], cpu_isolated_map); | |
6699 | return 1; | |
6700 | } | |
6701 | ||
8927f494 | 6702 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6703 | |
6704 | /* | |
6711cab4 SS |
6705 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6706 | * to a function which identifies what group(along with sched group) a CPU | |
6707 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6708 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6709 | * |
6710 | * init_sched_build_groups will build a circular linked list of the groups | |
6711 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6712 | * and ->cpu_power to 0. | |
6713 | */ | |
a616058b | 6714 | static void |
7c16ec58 | 6715 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6716 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6717 | struct sched_group **sg, |
6718 | cpumask_t *tmpmask), | |
6719 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6720 | { |
6721 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6722 | int i; |
6723 | ||
7c16ec58 MT |
6724 | cpus_clear(*covered); |
6725 | ||
363ab6f1 | 6726 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6727 | struct sched_group *sg; |
7c16ec58 | 6728 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6729 | int j; |
6730 | ||
7c16ec58 | 6731 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6732 | continue; |
6733 | ||
7c16ec58 | 6734 | cpus_clear(sg->cpumask); |
5517d86b | 6735 | sg->__cpu_power = 0; |
1da177e4 | 6736 | |
363ab6f1 | 6737 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6738 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6739 | continue; |
6740 | ||
7c16ec58 | 6741 | cpu_set(j, *covered); |
1da177e4 LT |
6742 | cpu_set(j, sg->cpumask); |
6743 | } | |
6744 | if (!first) | |
6745 | first = sg; | |
6746 | if (last) | |
6747 | last->next = sg; | |
6748 | last = sg; | |
6749 | } | |
6750 | last->next = first; | |
6751 | } | |
6752 | ||
9c1cfda2 | 6753 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6754 | |
9c1cfda2 | 6755 | #ifdef CONFIG_NUMA |
198e2f18 | 6756 | |
9c1cfda2 JH |
6757 | /** |
6758 | * find_next_best_node - find the next node to include in a sched_domain | |
6759 | * @node: node whose sched_domain we're building | |
6760 | * @used_nodes: nodes already in the sched_domain | |
6761 | * | |
41a2d6cf | 6762 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6763 | * finds the closest node not already in the @used_nodes map. |
6764 | * | |
6765 | * Should use nodemask_t. | |
6766 | */ | |
c5f59f08 | 6767 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6768 | { |
6769 | int i, n, val, min_val, best_node = 0; | |
6770 | ||
6771 | min_val = INT_MAX; | |
6772 | ||
076ac2af | 6773 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6774 | /* Start at @node */ |
076ac2af | 6775 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6776 | |
6777 | if (!nr_cpus_node(n)) | |
6778 | continue; | |
6779 | ||
6780 | /* Skip already used nodes */ | |
c5f59f08 | 6781 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6782 | continue; |
6783 | ||
6784 | /* Simple min distance search */ | |
6785 | val = node_distance(node, n); | |
6786 | ||
6787 | if (val < min_val) { | |
6788 | min_val = val; | |
6789 | best_node = n; | |
6790 | } | |
6791 | } | |
6792 | ||
c5f59f08 | 6793 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6794 | return best_node; |
6795 | } | |
6796 | ||
6797 | /** | |
6798 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6799 | * @node: node whose cpumask we're constructing | |
73486722 | 6800 | * @span: resulting cpumask |
9c1cfda2 | 6801 | * |
41a2d6cf | 6802 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6803 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6804 | * out optimally. | |
6805 | */ | |
4bdbaad3 | 6806 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6807 | { |
c5f59f08 | 6808 | nodemask_t used_nodes; |
c5f59f08 | 6809 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6810 | int i; |
9c1cfda2 | 6811 | |
4bdbaad3 | 6812 | cpus_clear(*span); |
c5f59f08 | 6813 | nodes_clear(used_nodes); |
9c1cfda2 | 6814 | |
4bdbaad3 | 6815 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6816 | node_set(node, used_nodes); |
9c1cfda2 JH |
6817 | |
6818 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6819 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6820 | |
c5f59f08 | 6821 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6822 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6823 | } |
9c1cfda2 | 6824 | } |
6d6bc0ad | 6825 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6826 | |
5c45bf27 | 6827 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6828 | |
9c1cfda2 | 6829 | /* |
48f24c4d | 6830 | * SMT sched-domains: |
9c1cfda2 | 6831 | */ |
1da177e4 LT |
6832 | #ifdef CONFIG_SCHED_SMT |
6833 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6834 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6835 | |
41a2d6cf | 6836 | static int |
7c16ec58 MT |
6837 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6838 | cpumask_t *unused) | |
1da177e4 | 6839 | { |
6711cab4 SS |
6840 | if (sg) |
6841 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6842 | return cpu; |
6843 | } | |
6d6bc0ad | 6844 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6845 | |
48f24c4d IM |
6846 | /* |
6847 | * multi-core sched-domains: | |
6848 | */ | |
1e9f28fa SS |
6849 | #ifdef CONFIG_SCHED_MC |
6850 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6851 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6852 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6853 | |
6854 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6855 | static int |
7c16ec58 MT |
6856 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6857 | cpumask_t *mask) | |
1e9f28fa | 6858 | { |
6711cab4 | 6859 | int group; |
7c16ec58 MT |
6860 | |
6861 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6862 | cpus_and(*mask, *mask, *cpu_map); | |
6863 | group = first_cpu(*mask); | |
6711cab4 SS |
6864 | if (sg) |
6865 | *sg = &per_cpu(sched_group_core, group); | |
6866 | return group; | |
1e9f28fa SS |
6867 | } |
6868 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6869 | static int |
7c16ec58 MT |
6870 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6871 | cpumask_t *unused) | |
1e9f28fa | 6872 | { |
6711cab4 SS |
6873 | if (sg) |
6874 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6875 | return cpu; |
6876 | } | |
6877 | #endif | |
6878 | ||
1da177e4 | 6879 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6880 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6881 | |
41a2d6cf | 6882 | static int |
7c16ec58 MT |
6883 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6884 | cpumask_t *mask) | |
1da177e4 | 6885 | { |
6711cab4 | 6886 | int group; |
48f24c4d | 6887 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6888 | *mask = cpu_coregroup_map(cpu); |
6889 | cpus_and(*mask, *mask, *cpu_map); | |
6890 | group = first_cpu(*mask); | |
1e9f28fa | 6891 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6892 | *mask = per_cpu(cpu_sibling_map, cpu); |
6893 | cpus_and(*mask, *mask, *cpu_map); | |
6894 | group = first_cpu(*mask); | |
1da177e4 | 6895 | #else |
6711cab4 | 6896 | group = cpu; |
1da177e4 | 6897 | #endif |
6711cab4 SS |
6898 | if (sg) |
6899 | *sg = &per_cpu(sched_group_phys, group); | |
6900 | return group; | |
1da177e4 LT |
6901 | } |
6902 | ||
6903 | #ifdef CONFIG_NUMA | |
1da177e4 | 6904 | /* |
9c1cfda2 JH |
6905 | * The init_sched_build_groups can't handle what we want to do with node |
6906 | * groups, so roll our own. Now each node has its own list of groups which | |
6907 | * gets dynamically allocated. | |
1da177e4 | 6908 | */ |
9c1cfda2 | 6909 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6910 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6911 | |
9c1cfda2 | 6912 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6913 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6914 | |
6711cab4 | 6915 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6916 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6917 | { |
6711cab4 SS |
6918 | int group; |
6919 | ||
7c16ec58 MT |
6920 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6921 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6922 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6923 | |
6924 | if (sg) | |
6925 | *sg = &per_cpu(sched_group_allnodes, group); | |
6926 | return group; | |
1da177e4 | 6927 | } |
6711cab4 | 6928 | |
08069033 SS |
6929 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6930 | { | |
6931 | struct sched_group *sg = group_head; | |
6932 | int j; | |
6933 | ||
6934 | if (!sg) | |
6935 | return; | |
3a5c359a | 6936 | do { |
363ab6f1 | 6937 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 6938 | struct sched_domain *sd; |
08069033 | 6939 | |
3a5c359a AK |
6940 | sd = &per_cpu(phys_domains, j); |
6941 | if (j != first_cpu(sd->groups->cpumask)) { | |
6942 | /* | |
6943 | * Only add "power" once for each | |
6944 | * physical package. | |
6945 | */ | |
6946 | continue; | |
6947 | } | |
08069033 | 6948 | |
3a5c359a AK |
6949 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6950 | } | |
6951 | sg = sg->next; | |
6952 | } while (sg != group_head); | |
08069033 | 6953 | } |
6d6bc0ad | 6954 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6955 | |
a616058b | 6956 | #ifdef CONFIG_NUMA |
51888ca2 | 6957 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6958 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6959 | { |
a616058b | 6960 | int cpu, i; |
51888ca2 | 6961 | |
363ab6f1 | 6962 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
6963 | struct sched_group **sched_group_nodes |
6964 | = sched_group_nodes_bycpu[cpu]; | |
6965 | ||
51888ca2 SV |
6966 | if (!sched_group_nodes) |
6967 | continue; | |
6968 | ||
076ac2af | 6969 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6970 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6971 | ||
7c16ec58 MT |
6972 | *nodemask = node_to_cpumask(i); |
6973 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6974 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6975 | continue; |
6976 | ||
6977 | if (sg == NULL) | |
6978 | continue; | |
6979 | sg = sg->next; | |
6980 | next_sg: | |
6981 | oldsg = sg; | |
6982 | sg = sg->next; | |
6983 | kfree(oldsg); | |
6984 | if (oldsg != sched_group_nodes[i]) | |
6985 | goto next_sg; | |
6986 | } | |
6987 | kfree(sched_group_nodes); | |
6988 | sched_group_nodes_bycpu[cpu] = NULL; | |
6989 | } | |
51888ca2 | 6990 | } |
6d6bc0ad | 6991 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 6992 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6993 | { |
6994 | } | |
6d6bc0ad | 6995 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6996 | |
89c4710e SS |
6997 | /* |
6998 | * Initialize sched groups cpu_power. | |
6999 | * | |
7000 | * cpu_power indicates the capacity of sched group, which is used while | |
7001 | * distributing the load between different sched groups in a sched domain. | |
7002 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7003 | * there are asymmetries in the topology. If there are asymmetries, group | |
7004 | * having more cpu_power will pickup more load compared to the group having | |
7005 | * less cpu_power. | |
7006 | * | |
7007 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7008 | * the maximum number of tasks a group can handle in the presence of other idle | |
7009 | * or lightly loaded groups in the same sched domain. | |
7010 | */ | |
7011 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7012 | { | |
7013 | struct sched_domain *child; | |
7014 | struct sched_group *group; | |
7015 | ||
7016 | WARN_ON(!sd || !sd->groups); | |
7017 | ||
7018 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7019 | return; | |
7020 | ||
7021 | child = sd->child; | |
7022 | ||
5517d86b ED |
7023 | sd->groups->__cpu_power = 0; |
7024 | ||
89c4710e SS |
7025 | /* |
7026 | * For perf policy, if the groups in child domain share resources | |
7027 | * (for example cores sharing some portions of the cache hierarchy | |
7028 | * or SMT), then set this domain groups cpu_power such that each group | |
7029 | * can handle only one task, when there are other idle groups in the | |
7030 | * same sched domain. | |
7031 | */ | |
7032 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7033 | (child->flags & | |
7034 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7035 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7036 | return; |
7037 | } | |
7038 | ||
89c4710e SS |
7039 | /* |
7040 | * add cpu_power of each child group to this groups cpu_power | |
7041 | */ | |
7042 | group = child->groups; | |
7043 | do { | |
5517d86b | 7044 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7045 | group = group->next; |
7046 | } while (group != child->groups); | |
7047 | } | |
7048 | ||
7c16ec58 MT |
7049 | /* |
7050 | * Initializers for schedule domains | |
7051 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7052 | */ | |
7053 | ||
7054 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7055 | #define SD_INIT_FUNC(type) \ | |
7056 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7057 | { \ | |
7058 | memset(sd, 0, sizeof(*sd)); \ | |
7059 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7060 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7061 | } |
7062 | ||
7063 | SD_INIT_FUNC(CPU) | |
7064 | #ifdef CONFIG_NUMA | |
7065 | SD_INIT_FUNC(ALLNODES) | |
7066 | SD_INIT_FUNC(NODE) | |
7067 | #endif | |
7068 | #ifdef CONFIG_SCHED_SMT | |
7069 | SD_INIT_FUNC(SIBLING) | |
7070 | #endif | |
7071 | #ifdef CONFIG_SCHED_MC | |
7072 | SD_INIT_FUNC(MC) | |
7073 | #endif | |
7074 | ||
7075 | /* | |
7076 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7077 | * space as the usage in build_sched_domains() dictates. Used only | |
7078 | * if the amount of space is significant. | |
7079 | */ | |
7080 | struct allmasks { | |
7081 | cpumask_t tmpmask; /* make this one first */ | |
7082 | union { | |
7083 | cpumask_t nodemask; | |
7084 | cpumask_t this_sibling_map; | |
7085 | cpumask_t this_core_map; | |
7086 | }; | |
7087 | cpumask_t send_covered; | |
7088 | ||
7089 | #ifdef CONFIG_NUMA | |
7090 | cpumask_t domainspan; | |
7091 | cpumask_t covered; | |
7092 | cpumask_t notcovered; | |
7093 | #endif | |
7094 | }; | |
7095 | ||
7096 | #if NR_CPUS > 128 | |
7097 | #define SCHED_CPUMASK_ALLOC 1 | |
7098 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7099 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7100 | #else | |
7101 | #define SCHED_CPUMASK_ALLOC 0 | |
7102 | #define SCHED_CPUMASK_FREE(v) | |
7103 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7104 | #endif | |
7105 | ||
7106 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7107 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7108 | ||
1d3504fc HS |
7109 | static int default_relax_domain_level = -1; |
7110 | ||
7111 | static int __init setup_relax_domain_level(char *str) | |
7112 | { | |
30e0e178 LZ |
7113 | unsigned long val; |
7114 | ||
7115 | val = simple_strtoul(str, NULL, 0); | |
7116 | if (val < SD_LV_MAX) | |
7117 | default_relax_domain_level = val; | |
7118 | ||
1d3504fc HS |
7119 | return 1; |
7120 | } | |
7121 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7122 | ||
7123 | static void set_domain_attribute(struct sched_domain *sd, | |
7124 | struct sched_domain_attr *attr) | |
7125 | { | |
7126 | int request; | |
7127 | ||
7128 | if (!attr || attr->relax_domain_level < 0) { | |
7129 | if (default_relax_domain_level < 0) | |
7130 | return; | |
7131 | else | |
7132 | request = default_relax_domain_level; | |
7133 | } else | |
7134 | request = attr->relax_domain_level; | |
7135 | if (request < sd->level) { | |
7136 | /* turn off idle balance on this domain */ | |
7137 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7138 | } else { | |
7139 | /* turn on idle balance on this domain */ | |
7140 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7141 | } | |
7142 | } | |
7143 | ||
1da177e4 | 7144 | /* |
1a20ff27 DG |
7145 | * Build sched domains for a given set of cpus and attach the sched domains |
7146 | * to the individual cpus | |
1da177e4 | 7147 | */ |
1d3504fc HS |
7148 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7149 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7150 | { |
7151 | int i; | |
57d885fe | 7152 | struct root_domain *rd; |
7c16ec58 MT |
7153 | SCHED_CPUMASK_DECLARE(allmasks); |
7154 | cpumask_t *tmpmask; | |
d1b55138 JH |
7155 | #ifdef CONFIG_NUMA |
7156 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7157 | int sd_allnodes = 0; |
d1b55138 JH |
7158 | |
7159 | /* | |
7160 | * Allocate the per-node list of sched groups | |
7161 | */ | |
076ac2af | 7162 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7163 | GFP_KERNEL); |
d1b55138 JH |
7164 | if (!sched_group_nodes) { |
7165 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7166 | return -ENOMEM; |
d1b55138 | 7167 | } |
d1b55138 | 7168 | #endif |
1da177e4 | 7169 | |
dc938520 | 7170 | rd = alloc_rootdomain(); |
57d885fe GH |
7171 | if (!rd) { |
7172 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7173 | #ifdef CONFIG_NUMA |
7174 | kfree(sched_group_nodes); | |
7175 | #endif | |
57d885fe GH |
7176 | return -ENOMEM; |
7177 | } | |
7178 | ||
7c16ec58 MT |
7179 | #if SCHED_CPUMASK_ALLOC |
7180 | /* get space for all scratch cpumask variables */ | |
7181 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7182 | if (!allmasks) { | |
7183 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7184 | kfree(rd); | |
7185 | #ifdef CONFIG_NUMA | |
7186 | kfree(sched_group_nodes); | |
7187 | #endif | |
7188 | return -ENOMEM; | |
7189 | } | |
7190 | #endif | |
7191 | tmpmask = (cpumask_t *)allmasks; | |
7192 | ||
7193 | ||
7194 | #ifdef CONFIG_NUMA | |
7195 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7196 | #endif | |
7197 | ||
1da177e4 | 7198 | /* |
1a20ff27 | 7199 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7200 | */ |
363ab6f1 | 7201 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7202 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7203 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7204 | |
7c16ec58 MT |
7205 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7206 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7207 | |
7208 | #ifdef CONFIG_NUMA | |
dd41f596 | 7209 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7210 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7211 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7212 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7213 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7214 | sd->span = *cpu_map; |
7c16ec58 | 7215 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7216 | p = sd; |
6711cab4 | 7217 | sd_allnodes = 1; |
9c1cfda2 JH |
7218 | } else |
7219 | p = NULL; | |
7220 | ||
1da177e4 | 7221 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7222 | SD_INIT(sd, NODE); |
1d3504fc | 7223 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7224 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7225 | sd->parent = p; |
1a848870 SS |
7226 | if (p) |
7227 | p->child = sd; | |
9c1cfda2 | 7228 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7229 | #endif |
7230 | ||
7231 | p = sd; | |
7232 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7233 | SD_INIT(sd, CPU); |
1d3504fc | 7234 | set_domain_attribute(sd, attr); |
7c16ec58 | 7235 | sd->span = *nodemask; |
1da177e4 | 7236 | sd->parent = p; |
1a848870 SS |
7237 | if (p) |
7238 | p->child = sd; | |
7c16ec58 | 7239 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7240 | |
1e9f28fa SS |
7241 | #ifdef CONFIG_SCHED_MC |
7242 | p = sd; | |
7243 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7244 | SD_INIT(sd, MC); |
1d3504fc | 7245 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7246 | sd->span = cpu_coregroup_map(i); |
7247 | cpus_and(sd->span, sd->span, *cpu_map); | |
7248 | sd->parent = p; | |
1a848870 | 7249 | p->child = sd; |
7c16ec58 | 7250 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7251 | #endif |
7252 | ||
1da177e4 LT |
7253 | #ifdef CONFIG_SCHED_SMT |
7254 | p = sd; | |
7255 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7256 | SD_INIT(sd, SIBLING); |
1d3504fc | 7257 | set_domain_attribute(sd, attr); |
d5a7430d | 7258 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7259 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7260 | sd->parent = p; |
1a848870 | 7261 | p->child = sd; |
7c16ec58 | 7262 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7263 | #endif |
7264 | } | |
7265 | ||
7266 | #ifdef CONFIG_SCHED_SMT | |
7267 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7268 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7269 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7270 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7271 | ||
7272 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7273 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7274 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7275 | continue; |
7276 | ||
dd41f596 | 7277 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7278 | &cpu_to_cpu_group, |
7279 | send_covered, tmpmask); | |
1da177e4 LT |
7280 | } |
7281 | #endif | |
7282 | ||
1e9f28fa SS |
7283 | #ifdef CONFIG_SCHED_MC |
7284 | /* Set up multi-core groups */ | |
363ab6f1 | 7285 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7286 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7287 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7288 | ||
7289 | *this_core_map = cpu_coregroup_map(i); | |
7290 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7291 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7292 | continue; |
7c16ec58 | 7293 | |
dd41f596 | 7294 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7295 | &cpu_to_core_group, |
7296 | send_covered, tmpmask); | |
1e9f28fa SS |
7297 | } |
7298 | #endif | |
7299 | ||
1da177e4 | 7300 | /* Set up physical groups */ |
076ac2af | 7301 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7302 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7303 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7304 | |
7c16ec58 MT |
7305 | *nodemask = node_to_cpumask(i); |
7306 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7307 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7308 | continue; |
7309 | ||
7c16ec58 MT |
7310 | init_sched_build_groups(nodemask, cpu_map, |
7311 | &cpu_to_phys_group, | |
7312 | send_covered, tmpmask); | |
1da177e4 LT |
7313 | } |
7314 | ||
7315 | #ifdef CONFIG_NUMA | |
7316 | /* Set up node groups */ | |
7c16ec58 MT |
7317 | if (sd_allnodes) { |
7318 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7319 | ||
7320 | init_sched_build_groups(cpu_map, cpu_map, | |
7321 | &cpu_to_allnodes_group, | |
7322 | send_covered, tmpmask); | |
7323 | } | |
9c1cfda2 | 7324 | |
076ac2af | 7325 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7326 | /* Set up node groups */ |
7327 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7328 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7329 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7330 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7331 | int j; |
7332 | ||
7c16ec58 MT |
7333 | *nodemask = node_to_cpumask(i); |
7334 | cpus_clear(*covered); | |
7335 | ||
7336 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7337 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7338 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7339 | continue; |
d1b55138 | 7340 | } |
9c1cfda2 | 7341 | |
4bdbaad3 | 7342 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7343 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7344 | |
15f0b676 | 7345 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7346 | if (!sg) { |
7347 | printk(KERN_WARNING "Can not alloc domain group for " | |
7348 | "node %d\n", i); | |
7349 | goto error; | |
7350 | } | |
9c1cfda2 | 7351 | sched_group_nodes[i] = sg; |
363ab6f1 | 7352 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7353 | struct sched_domain *sd; |
9761eea8 | 7354 | |
9c1cfda2 JH |
7355 | sd = &per_cpu(node_domains, j); |
7356 | sd->groups = sg; | |
9c1cfda2 | 7357 | } |
5517d86b | 7358 | sg->__cpu_power = 0; |
7c16ec58 | 7359 | sg->cpumask = *nodemask; |
51888ca2 | 7360 | sg->next = sg; |
7c16ec58 | 7361 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7362 | prev = sg; |
7363 | ||
076ac2af | 7364 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7365 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7366 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7367 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7368 | |
7c16ec58 MT |
7369 | cpus_complement(*notcovered, *covered); |
7370 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7371 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7372 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7373 | break; |
7374 | ||
7c16ec58 MT |
7375 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7376 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7377 | continue; |
7378 | ||
15f0b676 SV |
7379 | sg = kmalloc_node(sizeof(struct sched_group), |
7380 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7381 | if (!sg) { |
7382 | printk(KERN_WARNING | |
7383 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7384 | goto error; |
9c1cfda2 | 7385 | } |
5517d86b | 7386 | sg->__cpu_power = 0; |
7c16ec58 | 7387 | sg->cpumask = *tmpmask; |
51888ca2 | 7388 | sg->next = prev->next; |
7c16ec58 | 7389 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7390 | prev->next = sg; |
7391 | prev = sg; | |
7392 | } | |
9c1cfda2 | 7393 | } |
1da177e4 LT |
7394 | #endif |
7395 | ||
7396 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7397 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7398 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7399 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7400 | ||
89c4710e | 7401 | init_sched_groups_power(i, sd); |
5c45bf27 | 7402 | } |
1da177e4 | 7403 | #endif |
1e9f28fa | 7404 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7405 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7406 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7407 | ||
89c4710e | 7408 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7409 | } |
7410 | #endif | |
1e9f28fa | 7411 | |
363ab6f1 | 7412 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7413 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7414 | ||
89c4710e | 7415 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7416 | } |
7417 | ||
9c1cfda2 | 7418 | #ifdef CONFIG_NUMA |
076ac2af | 7419 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7420 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7421 | |
6711cab4 SS |
7422 | if (sd_allnodes) { |
7423 | struct sched_group *sg; | |
f712c0c7 | 7424 | |
7c16ec58 MT |
7425 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7426 | tmpmask); | |
f712c0c7 SS |
7427 | init_numa_sched_groups_power(sg); |
7428 | } | |
9c1cfda2 JH |
7429 | #endif |
7430 | ||
1da177e4 | 7431 | /* Attach the domains */ |
363ab6f1 | 7432 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7433 | struct sched_domain *sd; |
7434 | #ifdef CONFIG_SCHED_SMT | |
7435 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7436 | #elif defined(CONFIG_SCHED_MC) |
7437 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7438 | #else |
7439 | sd = &per_cpu(phys_domains, i); | |
7440 | #endif | |
57d885fe | 7441 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7442 | } |
51888ca2 | 7443 | |
7c16ec58 | 7444 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7445 | return 0; |
7446 | ||
a616058b | 7447 | #ifdef CONFIG_NUMA |
51888ca2 | 7448 | error: |
7c16ec58 MT |
7449 | free_sched_groups(cpu_map, tmpmask); |
7450 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7451 | return -ENOMEM; |
a616058b | 7452 | #endif |
1da177e4 | 7453 | } |
029190c5 | 7454 | |
1d3504fc HS |
7455 | static int build_sched_domains(const cpumask_t *cpu_map) |
7456 | { | |
7457 | return __build_sched_domains(cpu_map, NULL); | |
7458 | } | |
7459 | ||
029190c5 PJ |
7460 | static cpumask_t *doms_cur; /* current sched domains */ |
7461 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7462 | static struct sched_domain_attr *dattr_cur; |
7463 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7464 | |
7465 | /* | |
7466 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7467 | * cpumask_t) fails, then fallback to a single sched domain, | |
7468 | * as determined by the single cpumask_t fallback_doms. | |
7469 | */ | |
7470 | static cpumask_t fallback_doms; | |
7471 | ||
22e52b07 HC |
7472 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7473 | { | |
7474 | } | |
7475 | ||
1a20ff27 | 7476 | /* |
41a2d6cf | 7477 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7478 | * For now this just excludes isolated cpus, but could be used to |
7479 | * exclude other special cases in the future. | |
1a20ff27 | 7480 | */ |
51888ca2 | 7481 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7482 | { |
7378547f MM |
7483 | int err; |
7484 | ||
22e52b07 | 7485 | arch_update_cpu_topology(); |
029190c5 PJ |
7486 | ndoms_cur = 1; |
7487 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7488 | if (!doms_cur) | |
7489 | doms_cur = &fallback_doms; | |
7490 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7491 | dattr_cur = NULL; |
7378547f | 7492 | err = build_sched_domains(doms_cur); |
6382bc90 | 7493 | register_sched_domain_sysctl(); |
7378547f MM |
7494 | |
7495 | return err; | |
1a20ff27 DG |
7496 | } |
7497 | ||
7c16ec58 MT |
7498 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7499 | cpumask_t *tmpmask) | |
1da177e4 | 7500 | { |
7c16ec58 | 7501 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7502 | } |
1da177e4 | 7503 | |
1a20ff27 DG |
7504 | /* |
7505 | * Detach sched domains from a group of cpus specified in cpu_map | |
7506 | * These cpus will now be attached to the NULL domain | |
7507 | */ | |
858119e1 | 7508 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7509 | { |
7c16ec58 | 7510 | cpumask_t tmpmask; |
1a20ff27 DG |
7511 | int i; |
7512 | ||
6382bc90 MM |
7513 | unregister_sched_domain_sysctl(); |
7514 | ||
363ab6f1 | 7515 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7516 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7517 | synchronize_sched(); |
7c16ec58 | 7518 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7519 | } |
7520 | ||
1d3504fc HS |
7521 | /* handle null as "default" */ |
7522 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7523 | struct sched_domain_attr *new, int idx_new) | |
7524 | { | |
7525 | struct sched_domain_attr tmp; | |
7526 | ||
7527 | /* fast path */ | |
7528 | if (!new && !cur) | |
7529 | return 1; | |
7530 | ||
7531 | tmp = SD_ATTR_INIT; | |
7532 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7533 | new ? (new + idx_new) : &tmp, | |
7534 | sizeof(struct sched_domain_attr)); | |
7535 | } | |
7536 | ||
029190c5 PJ |
7537 | /* |
7538 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7539 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7540 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7541 | * It destroys each deleted domain and builds each new domain. | |
7542 | * | |
7543 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7544 | * The masks don't intersect (don't overlap.) We should setup one |
7545 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7546 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7547 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7548 | * it as it is. | |
7549 | * | |
41a2d6cf IM |
7550 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7551 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7552 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7553 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7554 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 PJ |
7555 | * |
7556 | * Call with hotplug lock held | |
7557 | */ | |
1d3504fc HS |
7558 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7559 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7560 | { |
7561 | int i, j; | |
7562 | ||
712555ee | 7563 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7564 | |
7378547f MM |
7565 | /* always unregister in case we don't destroy any domains */ |
7566 | unregister_sched_domain_sysctl(); | |
7567 | ||
e761b772 MK |
7568 | if (doms_new == NULL) |
7569 | ndoms_new = 0; | |
029190c5 PJ |
7570 | |
7571 | /* Destroy deleted domains */ | |
7572 | for (i = 0; i < ndoms_cur; i++) { | |
7573 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7574 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7575 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7576 | goto match1; |
7577 | } | |
7578 | /* no match - a current sched domain not in new doms_new[] */ | |
7579 | detach_destroy_domains(doms_cur + i); | |
7580 | match1: | |
7581 | ; | |
7582 | } | |
7583 | ||
e761b772 MK |
7584 | if (doms_new == NULL) { |
7585 | ndoms_cur = 0; | |
7586 | ndoms_new = 1; | |
7587 | doms_new = &fallback_doms; | |
7588 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7589 | dattr_new = NULL; | |
7590 | } | |
7591 | ||
029190c5 PJ |
7592 | /* Build new domains */ |
7593 | for (i = 0; i < ndoms_new; i++) { | |
7594 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7595 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7596 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7597 | goto match2; |
7598 | } | |
7599 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7600 | __build_sched_domains(doms_new + i, |
7601 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7602 | match2: |
7603 | ; | |
7604 | } | |
7605 | ||
7606 | /* Remember the new sched domains */ | |
7607 | if (doms_cur != &fallback_doms) | |
7608 | kfree(doms_cur); | |
1d3504fc | 7609 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7610 | doms_cur = doms_new; |
1d3504fc | 7611 | dattr_cur = dattr_new; |
029190c5 | 7612 | ndoms_cur = ndoms_new; |
7378547f MM |
7613 | |
7614 | register_sched_domain_sysctl(); | |
a1835615 | 7615 | |
712555ee | 7616 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7617 | } |
7618 | ||
5c45bf27 | 7619 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7620 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7621 | { |
95402b38 | 7622 | get_online_cpus(); |
e761b772 | 7623 | rebuild_sched_domains(); |
95402b38 | 7624 | put_online_cpus(); |
e761b772 | 7625 | return 0; |
5c45bf27 SS |
7626 | } |
7627 | ||
7628 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7629 | { | |
7630 | int ret; | |
7631 | ||
7632 | if (buf[0] != '0' && buf[0] != '1') | |
7633 | return -EINVAL; | |
7634 | ||
7635 | if (smt) | |
7636 | sched_smt_power_savings = (buf[0] == '1'); | |
7637 | else | |
7638 | sched_mc_power_savings = (buf[0] == '1'); | |
7639 | ||
7640 | ret = arch_reinit_sched_domains(); | |
7641 | ||
7642 | return ret ? ret : count; | |
7643 | } | |
7644 | ||
5c45bf27 | 7645 | #ifdef CONFIG_SCHED_MC |
4a0b2b4d AK |
7646 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, |
7647 | struct sysdev_attribute *attr, char *page) | |
5c45bf27 SS |
7648 | { |
7649 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7650 | } | |
48f24c4d | 7651 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
4a0b2b4d | 7652 | struct sysdev_attribute *attr, |
48f24c4d | 7653 | const char *buf, size_t count) |
5c45bf27 SS |
7654 | { |
7655 | return sched_power_savings_store(buf, count, 0); | |
7656 | } | |
6707de00 AB |
7657 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7658 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7659 | #endif |
7660 | ||
7661 | #ifdef CONFIG_SCHED_SMT | |
4a0b2b4d AK |
7662 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, |
7663 | struct sysdev_attribute *attr, char *page) | |
5c45bf27 SS |
7664 | { |
7665 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7666 | } | |
48f24c4d | 7667 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
4a0b2b4d | 7668 | struct sysdev_attribute *attr, |
48f24c4d | 7669 | const char *buf, size_t count) |
5c45bf27 SS |
7670 | { |
7671 | return sched_power_savings_store(buf, count, 1); | |
7672 | } | |
6707de00 AB |
7673 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7674 | sched_smt_power_savings_store); | |
7675 | #endif | |
7676 | ||
7677 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7678 | { | |
7679 | int err = 0; | |
7680 | ||
7681 | #ifdef CONFIG_SCHED_SMT | |
7682 | if (smt_capable()) | |
7683 | err = sysfs_create_file(&cls->kset.kobj, | |
7684 | &attr_sched_smt_power_savings.attr); | |
7685 | #endif | |
7686 | #ifdef CONFIG_SCHED_MC | |
7687 | if (!err && mc_capable()) | |
7688 | err = sysfs_create_file(&cls->kset.kobj, | |
7689 | &attr_sched_mc_power_savings.attr); | |
7690 | #endif | |
7691 | return err; | |
7692 | } | |
6d6bc0ad | 7693 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7694 | |
e761b772 | 7695 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7696 | /* |
e761b772 MK |
7697 | * Add online and remove offline CPUs from the scheduler domains. |
7698 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7699 | */ |
7700 | static int update_sched_domains(struct notifier_block *nfb, | |
7701 | unsigned long action, void *hcpu) | |
e761b772 MK |
7702 | { |
7703 | switch (action) { | |
7704 | case CPU_ONLINE: | |
7705 | case CPU_ONLINE_FROZEN: | |
7706 | case CPU_DEAD: | |
7707 | case CPU_DEAD_FROZEN: | |
7708 | partition_sched_domains(0, NULL, NULL); | |
7709 | return NOTIFY_OK; | |
7710 | ||
7711 | default: | |
7712 | return NOTIFY_DONE; | |
7713 | } | |
7714 | } | |
7715 | #endif | |
7716 | ||
7717 | static int update_runtime(struct notifier_block *nfb, | |
7718 | unsigned long action, void *hcpu) | |
1da177e4 | 7719 | { |
7def2be1 PZ |
7720 | int cpu = (int)(long)hcpu; |
7721 | ||
1da177e4 | 7722 | switch (action) { |
1da177e4 | 7723 | case CPU_DOWN_PREPARE: |
8bb78442 | 7724 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7725 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7726 | return NOTIFY_OK; |
7727 | ||
1da177e4 | 7728 | case CPU_DOWN_FAILED: |
8bb78442 | 7729 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7730 | case CPU_ONLINE: |
8bb78442 | 7731 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7732 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7733 | return NOTIFY_OK; |
7734 | ||
1da177e4 LT |
7735 | default: |
7736 | return NOTIFY_DONE; | |
7737 | } | |
1da177e4 | 7738 | } |
1da177e4 LT |
7739 | |
7740 | void __init sched_init_smp(void) | |
7741 | { | |
5c1e1767 NP |
7742 | cpumask_t non_isolated_cpus; |
7743 | ||
434d53b0 MT |
7744 | #if defined(CONFIG_NUMA) |
7745 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7746 | GFP_KERNEL); | |
7747 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7748 | #endif | |
95402b38 | 7749 | get_online_cpus(); |
712555ee | 7750 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7751 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7752 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7753 | if (cpus_empty(non_isolated_cpus)) |
7754 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7755 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7756 | put_online_cpus(); |
e761b772 MK |
7757 | |
7758 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7759 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7760 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7761 | #endif |
7762 | ||
7763 | /* RT runtime code needs to handle some hotplug events */ | |
7764 | hotcpu_notifier(update_runtime, 0); | |
7765 | ||
b328ca18 | 7766 | init_hrtick(); |
5c1e1767 NP |
7767 | |
7768 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7769 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7770 | BUG(); |
19978ca6 | 7771 | sched_init_granularity(); |
1da177e4 LT |
7772 | } |
7773 | #else | |
7774 | void __init sched_init_smp(void) | |
7775 | { | |
19978ca6 | 7776 | sched_init_granularity(); |
1da177e4 LT |
7777 | } |
7778 | #endif /* CONFIG_SMP */ | |
7779 | ||
7780 | int in_sched_functions(unsigned long addr) | |
7781 | { | |
1da177e4 LT |
7782 | return in_lock_functions(addr) || |
7783 | (addr >= (unsigned long)__sched_text_start | |
7784 | && addr < (unsigned long)__sched_text_end); | |
7785 | } | |
7786 | ||
a9957449 | 7787 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7788 | { |
7789 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7790 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7791 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7792 | cfs_rq->rq = rq; | |
7793 | #endif | |
67e9fb2a | 7794 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7795 | } |
7796 | ||
fa85ae24 PZ |
7797 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7798 | { | |
7799 | struct rt_prio_array *array; | |
7800 | int i; | |
7801 | ||
7802 | array = &rt_rq->active; | |
7803 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7804 | INIT_LIST_HEAD(array->queue + i); | |
7805 | __clear_bit(i, array->bitmap); | |
7806 | } | |
7807 | /* delimiter for bitsearch: */ | |
7808 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7809 | ||
052f1dc7 | 7810 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7811 | rt_rq->highest_prio = MAX_RT_PRIO; |
7812 | #endif | |
fa85ae24 PZ |
7813 | #ifdef CONFIG_SMP |
7814 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7815 | rt_rq->overloaded = 0; |
7816 | #endif | |
7817 | ||
7818 | rt_rq->rt_time = 0; | |
7819 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7820 | rt_rq->rt_runtime = 0; |
7821 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7822 | |
052f1dc7 | 7823 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7824 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7825 | rt_rq->rq = rq; |
7826 | #endif | |
fa85ae24 PZ |
7827 | } |
7828 | ||
6f505b16 | 7829 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7830 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7831 | struct sched_entity *se, int cpu, int add, | |
7832 | struct sched_entity *parent) | |
6f505b16 | 7833 | { |
ec7dc8ac | 7834 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7835 | tg->cfs_rq[cpu] = cfs_rq; |
7836 | init_cfs_rq(cfs_rq, rq); | |
7837 | cfs_rq->tg = tg; | |
7838 | if (add) | |
7839 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7840 | ||
7841 | tg->se[cpu] = se; | |
354d60c2 DG |
7842 | /* se could be NULL for init_task_group */ |
7843 | if (!se) | |
7844 | return; | |
7845 | ||
ec7dc8ac DG |
7846 | if (!parent) |
7847 | se->cfs_rq = &rq->cfs; | |
7848 | else | |
7849 | se->cfs_rq = parent->my_q; | |
7850 | ||
6f505b16 PZ |
7851 | se->my_q = cfs_rq; |
7852 | se->load.weight = tg->shares; | |
e05510d0 | 7853 | se->load.inv_weight = 0; |
ec7dc8ac | 7854 | se->parent = parent; |
6f505b16 | 7855 | } |
052f1dc7 | 7856 | #endif |
6f505b16 | 7857 | |
052f1dc7 | 7858 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7859 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7860 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7861 | struct sched_rt_entity *parent) | |
6f505b16 | 7862 | { |
ec7dc8ac DG |
7863 | struct rq *rq = cpu_rq(cpu); |
7864 | ||
6f505b16 PZ |
7865 | tg->rt_rq[cpu] = rt_rq; |
7866 | init_rt_rq(rt_rq, rq); | |
7867 | rt_rq->tg = tg; | |
7868 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7869 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7870 | if (add) |
7871 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7872 | ||
7873 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7874 | if (!rt_se) |
7875 | return; | |
7876 | ||
ec7dc8ac DG |
7877 | if (!parent) |
7878 | rt_se->rt_rq = &rq->rt; | |
7879 | else | |
7880 | rt_se->rt_rq = parent->my_q; | |
7881 | ||
6f505b16 | 7882 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7883 | rt_se->parent = parent; |
6f505b16 PZ |
7884 | INIT_LIST_HEAD(&rt_se->run_list); |
7885 | } | |
7886 | #endif | |
7887 | ||
1da177e4 LT |
7888 | void __init sched_init(void) |
7889 | { | |
dd41f596 | 7890 | int i, j; |
434d53b0 MT |
7891 | unsigned long alloc_size = 0, ptr; |
7892 | ||
7893 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7894 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7895 | #endif | |
7896 | #ifdef CONFIG_RT_GROUP_SCHED | |
7897 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7898 | #endif |
7899 | #ifdef CONFIG_USER_SCHED | |
7900 | alloc_size *= 2; | |
434d53b0 MT |
7901 | #endif |
7902 | /* | |
7903 | * As sched_init() is called before page_alloc is setup, | |
7904 | * we use alloc_bootmem(). | |
7905 | */ | |
7906 | if (alloc_size) { | |
5a9d3225 | 7907 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7908 | |
7909 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7910 | init_task_group.se = (struct sched_entity **)ptr; | |
7911 | ptr += nr_cpu_ids * sizeof(void **); | |
7912 | ||
7913 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7914 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7915 | |
7916 | #ifdef CONFIG_USER_SCHED | |
7917 | root_task_group.se = (struct sched_entity **)ptr; | |
7918 | ptr += nr_cpu_ids * sizeof(void **); | |
7919 | ||
7920 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7921 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7922 | #endif /* CONFIG_USER_SCHED */ |
7923 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7924 | #ifdef CONFIG_RT_GROUP_SCHED |
7925 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7926 | ptr += nr_cpu_ids * sizeof(void **); | |
7927 | ||
7928 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7929 | ptr += nr_cpu_ids * sizeof(void **); |
7930 | ||
7931 | #ifdef CONFIG_USER_SCHED | |
7932 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7933 | ptr += nr_cpu_ids * sizeof(void **); | |
7934 | ||
7935 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7936 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7937 | #endif /* CONFIG_USER_SCHED */ |
7938 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7939 | } |
dd41f596 | 7940 | |
57d885fe GH |
7941 | #ifdef CONFIG_SMP |
7942 | init_defrootdomain(); | |
7943 | #endif | |
7944 | ||
d0b27fa7 PZ |
7945 | init_rt_bandwidth(&def_rt_bandwidth, |
7946 | global_rt_period(), global_rt_runtime()); | |
7947 | ||
7948 | #ifdef CONFIG_RT_GROUP_SCHED | |
7949 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7950 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7951 | #ifdef CONFIG_USER_SCHED |
7952 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7953 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7954 | #endif /* CONFIG_USER_SCHED */ |
7955 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7956 | |
052f1dc7 | 7957 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7958 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7959 | INIT_LIST_HEAD(&init_task_group.children); |
7960 | ||
7961 | #ifdef CONFIG_USER_SCHED | |
7962 | INIT_LIST_HEAD(&root_task_group.children); | |
7963 | init_task_group.parent = &root_task_group; | |
7964 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7965 | #endif /* CONFIG_USER_SCHED */ |
7966 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7967 | |
0a945022 | 7968 | for_each_possible_cpu(i) { |
70b97a7f | 7969 | struct rq *rq; |
1da177e4 LT |
7970 | |
7971 | rq = cpu_rq(i); | |
7972 | spin_lock_init(&rq->lock); | |
fcb99371 | 7973 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7974 | rq->nr_running = 0; |
dd41f596 | 7975 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7976 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7977 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7978 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7979 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7980 | #ifdef CONFIG_CGROUP_SCHED |
7981 | /* | |
7982 | * How much cpu bandwidth does init_task_group get? | |
7983 | * | |
7984 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7985 | * gets 100% of the cpu resources in the system. This overall | |
7986 | * system cpu resource is divided among the tasks of | |
7987 | * init_task_group and its child task-groups in a fair manner, | |
7988 | * based on each entity's (task or task-group's) weight | |
7989 | * (se->load.weight). | |
7990 | * | |
7991 | * In other words, if init_task_group has 10 tasks of weight | |
7992 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7993 | * then A0's share of the cpu resource is: | |
7994 | * | |
7995 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7996 | * | |
7997 | * We achieve this by letting init_task_group's tasks sit | |
7998 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7999 | */ | |
ec7dc8ac | 8000 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8001 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8002 | root_task_group.shares = NICE_0_LOAD; |
8003 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8004 | /* |
8005 | * In case of task-groups formed thr' the user id of tasks, | |
8006 | * init_task_group represents tasks belonging to root user. | |
8007 | * Hence it forms a sibling of all subsequent groups formed. | |
8008 | * In this case, init_task_group gets only a fraction of overall | |
8009 | * system cpu resource, based on the weight assigned to root | |
8010 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8011 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8012 | * (init_cfs_rq) and having one entity represent this group of | |
8013 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8014 | */ | |
ec7dc8ac | 8015 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8016 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8017 | &per_cpu(init_sched_entity, i), i, 1, |
8018 | root_task_group.se[i]); | |
6f505b16 | 8019 | |
052f1dc7 | 8020 | #endif |
354d60c2 DG |
8021 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8022 | ||
8023 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8024 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8025 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8026 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8027 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8028 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8029 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8030 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8031 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8032 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8033 | root_task_group.rt_se[i]); | |
354d60c2 | 8034 | #endif |
dd41f596 | 8035 | #endif |
1da177e4 | 8036 | |
dd41f596 IM |
8037 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8038 | rq->cpu_load[j] = 0; | |
1da177e4 | 8039 | #ifdef CONFIG_SMP |
41c7ce9a | 8040 | rq->sd = NULL; |
57d885fe | 8041 | rq->rd = NULL; |
1da177e4 | 8042 | rq->active_balance = 0; |
dd41f596 | 8043 | rq->next_balance = jiffies; |
1da177e4 | 8044 | rq->push_cpu = 0; |
0a2966b4 | 8045 | rq->cpu = i; |
1f11eb6a | 8046 | rq->online = 0; |
1da177e4 LT |
8047 | rq->migration_thread = NULL; |
8048 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8049 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8050 | #endif |
8f4d37ec | 8051 | init_rq_hrtick(rq); |
1da177e4 | 8052 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8053 | } |
8054 | ||
2dd73a4f | 8055 | set_load_weight(&init_task); |
b50f60ce | 8056 | |
e107be36 AK |
8057 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8058 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8059 | #endif | |
8060 | ||
c9819f45 | 8061 | #ifdef CONFIG_SMP |
962cf36c | 8062 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8063 | #endif |
8064 | ||
b50f60ce HC |
8065 | #ifdef CONFIG_RT_MUTEXES |
8066 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8067 | #endif | |
8068 | ||
1da177e4 LT |
8069 | /* |
8070 | * The boot idle thread does lazy MMU switching as well: | |
8071 | */ | |
8072 | atomic_inc(&init_mm.mm_count); | |
8073 | enter_lazy_tlb(&init_mm, current); | |
8074 | ||
8075 | /* | |
8076 | * Make us the idle thread. Technically, schedule() should not be | |
8077 | * called from this thread, however somewhere below it might be, | |
8078 | * but because we are the idle thread, we just pick up running again | |
8079 | * when this runqueue becomes "idle". | |
8080 | */ | |
8081 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8082 | /* |
8083 | * During early bootup we pretend to be a normal task: | |
8084 | */ | |
8085 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8086 | |
8087 | scheduler_running = 1; | |
1da177e4 LT |
8088 | } |
8089 | ||
8090 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8091 | void __might_sleep(char *file, int line) | |
8092 | { | |
48f24c4d | 8093 | #ifdef in_atomic |
1da177e4 LT |
8094 | static unsigned long prev_jiffy; /* ratelimiting */ |
8095 | ||
8096 | if ((in_atomic() || irqs_disabled()) && | |
8097 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8098 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8099 | return; | |
8100 | prev_jiffy = jiffies; | |
91368d73 | 8101 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8102 | " context at %s:%d\n", file, line); |
8103 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8104 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8105 | debug_show_held_locks(current); |
3117df04 IM |
8106 | if (irqs_disabled()) |
8107 | print_irqtrace_events(current); | |
1da177e4 LT |
8108 | dump_stack(); |
8109 | } | |
8110 | #endif | |
8111 | } | |
8112 | EXPORT_SYMBOL(__might_sleep); | |
8113 | #endif | |
8114 | ||
8115 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8116 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8117 | { | |
8118 | int on_rq; | |
3e51f33f | 8119 | |
3a5e4dc1 AK |
8120 | update_rq_clock(rq); |
8121 | on_rq = p->se.on_rq; | |
8122 | if (on_rq) | |
8123 | deactivate_task(rq, p, 0); | |
8124 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8125 | if (on_rq) { | |
8126 | activate_task(rq, p, 0); | |
8127 | resched_task(rq->curr); | |
8128 | } | |
8129 | } | |
8130 | ||
1da177e4 LT |
8131 | void normalize_rt_tasks(void) |
8132 | { | |
a0f98a1c | 8133 | struct task_struct *g, *p; |
1da177e4 | 8134 | unsigned long flags; |
70b97a7f | 8135 | struct rq *rq; |
1da177e4 | 8136 | |
4cf5d77a | 8137 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8138 | do_each_thread(g, p) { |
178be793 IM |
8139 | /* |
8140 | * Only normalize user tasks: | |
8141 | */ | |
8142 | if (!p->mm) | |
8143 | continue; | |
8144 | ||
6cfb0d5d | 8145 | p->se.exec_start = 0; |
6cfb0d5d | 8146 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8147 | p->se.wait_start = 0; |
dd41f596 | 8148 | p->se.sleep_start = 0; |
dd41f596 | 8149 | p->se.block_start = 0; |
6cfb0d5d | 8150 | #endif |
dd41f596 IM |
8151 | |
8152 | if (!rt_task(p)) { | |
8153 | /* | |
8154 | * Renice negative nice level userspace | |
8155 | * tasks back to 0: | |
8156 | */ | |
8157 | if (TASK_NICE(p) < 0 && p->mm) | |
8158 | set_user_nice(p, 0); | |
1da177e4 | 8159 | continue; |
dd41f596 | 8160 | } |
1da177e4 | 8161 | |
4cf5d77a | 8162 | spin_lock(&p->pi_lock); |
b29739f9 | 8163 | rq = __task_rq_lock(p); |
1da177e4 | 8164 | |
178be793 | 8165 | normalize_task(rq, p); |
3a5e4dc1 | 8166 | |
b29739f9 | 8167 | __task_rq_unlock(rq); |
4cf5d77a | 8168 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8169 | } while_each_thread(g, p); |
8170 | ||
4cf5d77a | 8171 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8172 | } |
8173 | ||
8174 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8175 | |
8176 | #ifdef CONFIG_IA64 | |
8177 | /* | |
8178 | * These functions are only useful for the IA64 MCA handling. | |
8179 | * | |
8180 | * They can only be called when the whole system has been | |
8181 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8182 | * activity can take place. Using them for anything else would | |
8183 | * be a serious bug, and as a result, they aren't even visible | |
8184 | * under any other configuration. | |
8185 | */ | |
8186 | ||
8187 | /** | |
8188 | * curr_task - return the current task for a given cpu. | |
8189 | * @cpu: the processor in question. | |
8190 | * | |
8191 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8192 | */ | |
36c8b586 | 8193 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8194 | { |
8195 | return cpu_curr(cpu); | |
8196 | } | |
8197 | ||
8198 | /** | |
8199 | * set_curr_task - set the current task for a given cpu. | |
8200 | * @cpu: the processor in question. | |
8201 | * @p: the task pointer to set. | |
8202 | * | |
8203 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8204 | * are serviced on a separate stack. It allows the architecture to switch the |
8205 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8206 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8207 | * and caller must save the original value of the current task (see | |
8208 | * curr_task() above) and restore that value before reenabling interrupts and | |
8209 | * re-starting the system. | |
8210 | * | |
8211 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8212 | */ | |
36c8b586 | 8213 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8214 | { |
8215 | cpu_curr(cpu) = p; | |
8216 | } | |
8217 | ||
8218 | #endif | |
29f59db3 | 8219 | |
bccbe08a PZ |
8220 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8221 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8222 | { |
8223 | int i; | |
8224 | ||
8225 | for_each_possible_cpu(i) { | |
8226 | if (tg->cfs_rq) | |
8227 | kfree(tg->cfs_rq[i]); | |
8228 | if (tg->se) | |
8229 | kfree(tg->se[i]); | |
6f505b16 PZ |
8230 | } |
8231 | ||
8232 | kfree(tg->cfs_rq); | |
8233 | kfree(tg->se); | |
6f505b16 PZ |
8234 | } |
8235 | ||
ec7dc8ac DG |
8236 | static |
8237 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8238 | { |
29f59db3 | 8239 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8240 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8241 | struct rq *rq; |
29f59db3 SV |
8242 | int i; |
8243 | ||
434d53b0 | 8244 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8245 | if (!tg->cfs_rq) |
8246 | goto err; | |
434d53b0 | 8247 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8248 | if (!tg->se) |
8249 | goto err; | |
052f1dc7 PZ |
8250 | |
8251 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8252 | |
8253 | for_each_possible_cpu(i) { | |
9b5b7751 | 8254 | rq = cpu_rq(i); |
29f59db3 | 8255 | |
6f505b16 PZ |
8256 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8257 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8258 | if (!cfs_rq) |
8259 | goto err; | |
8260 | ||
6f505b16 PZ |
8261 | se = kmalloc_node(sizeof(struct sched_entity), |
8262 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8263 | if (!se) |
8264 | goto err; | |
8265 | ||
ec7dc8ac DG |
8266 | parent_se = parent ? parent->se[i] : NULL; |
8267 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8268 | } |
8269 | ||
8270 | return 1; | |
8271 | ||
8272 | err: | |
8273 | return 0; | |
8274 | } | |
8275 | ||
8276 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8277 | { | |
8278 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8279 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8280 | } | |
8281 | ||
8282 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8283 | { | |
8284 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8285 | } | |
6d6bc0ad | 8286 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8287 | static inline void free_fair_sched_group(struct task_group *tg) |
8288 | { | |
8289 | } | |
8290 | ||
ec7dc8ac DG |
8291 | static inline |
8292 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8293 | { |
8294 | return 1; | |
8295 | } | |
8296 | ||
8297 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8298 | { | |
8299 | } | |
8300 | ||
8301 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8302 | { | |
8303 | } | |
6d6bc0ad | 8304 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8305 | |
8306 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8307 | static void free_rt_sched_group(struct task_group *tg) |
8308 | { | |
8309 | int i; | |
8310 | ||
d0b27fa7 PZ |
8311 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8312 | ||
bccbe08a PZ |
8313 | for_each_possible_cpu(i) { |
8314 | if (tg->rt_rq) | |
8315 | kfree(tg->rt_rq[i]); | |
8316 | if (tg->rt_se) | |
8317 | kfree(tg->rt_se[i]); | |
8318 | } | |
8319 | ||
8320 | kfree(tg->rt_rq); | |
8321 | kfree(tg->rt_se); | |
8322 | } | |
8323 | ||
ec7dc8ac DG |
8324 | static |
8325 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8326 | { |
8327 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8328 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8329 | struct rq *rq; |
8330 | int i; | |
8331 | ||
434d53b0 | 8332 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8333 | if (!tg->rt_rq) |
8334 | goto err; | |
434d53b0 | 8335 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8336 | if (!tg->rt_se) |
8337 | goto err; | |
8338 | ||
d0b27fa7 PZ |
8339 | init_rt_bandwidth(&tg->rt_bandwidth, |
8340 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8341 | |
8342 | for_each_possible_cpu(i) { | |
8343 | rq = cpu_rq(i); | |
8344 | ||
6f505b16 PZ |
8345 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8346 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8347 | if (!rt_rq) | |
8348 | goto err; | |
29f59db3 | 8349 | |
6f505b16 PZ |
8350 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8351 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8352 | if (!rt_se) | |
8353 | goto err; | |
29f59db3 | 8354 | |
ec7dc8ac DG |
8355 | parent_se = parent ? parent->rt_se[i] : NULL; |
8356 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8357 | } |
8358 | ||
bccbe08a PZ |
8359 | return 1; |
8360 | ||
8361 | err: | |
8362 | return 0; | |
8363 | } | |
8364 | ||
8365 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8366 | { | |
8367 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8368 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8369 | } | |
8370 | ||
8371 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8372 | { | |
8373 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8374 | } | |
6d6bc0ad | 8375 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8376 | static inline void free_rt_sched_group(struct task_group *tg) |
8377 | { | |
8378 | } | |
8379 | ||
ec7dc8ac DG |
8380 | static inline |
8381 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8382 | { |
8383 | return 1; | |
8384 | } | |
8385 | ||
8386 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8387 | { | |
8388 | } | |
8389 | ||
8390 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8391 | { | |
8392 | } | |
6d6bc0ad | 8393 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8394 | |
d0b27fa7 | 8395 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8396 | static void free_sched_group(struct task_group *tg) |
8397 | { | |
8398 | free_fair_sched_group(tg); | |
8399 | free_rt_sched_group(tg); | |
8400 | kfree(tg); | |
8401 | } | |
8402 | ||
8403 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8404 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8405 | { |
8406 | struct task_group *tg; | |
8407 | unsigned long flags; | |
8408 | int i; | |
8409 | ||
8410 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8411 | if (!tg) | |
8412 | return ERR_PTR(-ENOMEM); | |
8413 | ||
ec7dc8ac | 8414 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8415 | goto err; |
8416 | ||
ec7dc8ac | 8417 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8418 | goto err; |
8419 | ||
8ed36996 | 8420 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8421 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8422 | register_fair_sched_group(tg, i); |
8423 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8424 | } |
6f505b16 | 8425 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8426 | |
8427 | WARN_ON(!parent); /* root should already exist */ | |
8428 | ||
8429 | tg->parent = parent; | |
8430 | list_add_rcu(&tg->siblings, &parent->children); | |
8431 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8432 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8433 | |
9b5b7751 | 8434 | return tg; |
29f59db3 SV |
8435 | |
8436 | err: | |
6f505b16 | 8437 | free_sched_group(tg); |
29f59db3 SV |
8438 | return ERR_PTR(-ENOMEM); |
8439 | } | |
8440 | ||
9b5b7751 | 8441 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8442 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8443 | { |
29f59db3 | 8444 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8445 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8446 | } |
8447 | ||
9b5b7751 | 8448 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8449 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8450 | { |
8ed36996 | 8451 | unsigned long flags; |
9b5b7751 | 8452 | int i; |
29f59db3 | 8453 | |
8ed36996 | 8454 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8455 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8456 | unregister_fair_sched_group(tg, i); |
8457 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8458 | } |
6f505b16 | 8459 | list_del_rcu(&tg->list); |
f473aa5e | 8460 | list_del_rcu(&tg->siblings); |
8ed36996 | 8461 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8462 | |
9b5b7751 | 8463 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8464 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8465 | } |
8466 | ||
9b5b7751 | 8467 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8468 | * The caller of this function should have put the task in its new group |
8469 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8470 | * reflect its new group. | |
9b5b7751 SV |
8471 | */ |
8472 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8473 | { |
8474 | int on_rq, running; | |
8475 | unsigned long flags; | |
8476 | struct rq *rq; | |
8477 | ||
8478 | rq = task_rq_lock(tsk, &flags); | |
8479 | ||
29f59db3 SV |
8480 | update_rq_clock(rq); |
8481 | ||
051a1d1a | 8482 | running = task_current(rq, tsk); |
29f59db3 SV |
8483 | on_rq = tsk->se.on_rq; |
8484 | ||
0e1f3483 | 8485 | if (on_rq) |
29f59db3 | 8486 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8487 | if (unlikely(running)) |
8488 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8489 | |
6f505b16 | 8490 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8491 | |
810b3817 PZ |
8492 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8493 | if (tsk->sched_class->moved_group) | |
8494 | tsk->sched_class->moved_group(tsk); | |
8495 | #endif | |
8496 | ||
0e1f3483 HS |
8497 | if (unlikely(running)) |
8498 | tsk->sched_class->set_curr_task(rq); | |
8499 | if (on_rq) | |
7074badb | 8500 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8501 | |
29f59db3 SV |
8502 | task_rq_unlock(rq, &flags); |
8503 | } | |
6d6bc0ad | 8504 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8505 | |
052f1dc7 | 8506 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8507 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8508 | { |
8509 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8510 | int on_rq; |
8511 | ||
29f59db3 | 8512 | on_rq = se->on_rq; |
62fb1851 | 8513 | if (on_rq) |
29f59db3 SV |
8514 | dequeue_entity(cfs_rq, se, 0); |
8515 | ||
8516 | se->load.weight = shares; | |
e05510d0 | 8517 | se->load.inv_weight = 0; |
29f59db3 | 8518 | |
62fb1851 | 8519 | if (on_rq) |
29f59db3 | 8520 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8521 | } |
62fb1851 | 8522 | |
c09595f6 PZ |
8523 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8524 | { | |
8525 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8526 | struct rq *rq = cfs_rq->rq; | |
8527 | unsigned long flags; | |
8528 | ||
8529 | spin_lock_irqsave(&rq->lock, flags); | |
8530 | __set_se_shares(se, shares); | |
8531 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8532 | } |
8533 | ||
8ed36996 PZ |
8534 | static DEFINE_MUTEX(shares_mutex); |
8535 | ||
4cf86d77 | 8536 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8537 | { |
8538 | int i; | |
8ed36996 | 8539 | unsigned long flags; |
c61935fd | 8540 | |
ec7dc8ac DG |
8541 | /* |
8542 | * We can't change the weight of the root cgroup. | |
8543 | */ | |
8544 | if (!tg->se[0]) | |
8545 | return -EINVAL; | |
8546 | ||
18d95a28 PZ |
8547 | if (shares < MIN_SHARES) |
8548 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8549 | else if (shares > MAX_SHARES) |
8550 | shares = MAX_SHARES; | |
62fb1851 | 8551 | |
8ed36996 | 8552 | mutex_lock(&shares_mutex); |
9b5b7751 | 8553 | if (tg->shares == shares) |
5cb350ba | 8554 | goto done; |
29f59db3 | 8555 | |
8ed36996 | 8556 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8557 | for_each_possible_cpu(i) |
8558 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8559 | list_del_rcu(&tg->siblings); |
8ed36996 | 8560 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8561 | |
8562 | /* wait for any ongoing reference to this group to finish */ | |
8563 | synchronize_sched(); | |
8564 | ||
8565 | /* | |
8566 | * Now we are free to modify the group's share on each cpu | |
8567 | * w/o tripping rebalance_share or load_balance_fair. | |
8568 | */ | |
9b5b7751 | 8569 | tg->shares = shares; |
c09595f6 PZ |
8570 | for_each_possible_cpu(i) { |
8571 | /* | |
8572 | * force a rebalance | |
8573 | */ | |
8574 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8575 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8576 | } |
29f59db3 | 8577 | |
6b2d7700 SV |
8578 | /* |
8579 | * Enable load balance activity on this group, by inserting it back on | |
8580 | * each cpu's rq->leaf_cfs_rq_list. | |
8581 | */ | |
8ed36996 | 8582 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8583 | for_each_possible_cpu(i) |
8584 | register_fair_sched_group(tg, i); | |
f473aa5e | 8585 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8586 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8587 | done: |
8ed36996 | 8588 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8589 | return 0; |
29f59db3 SV |
8590 | } |
8591 | ||
5cb350ba DG |
8592 | unsigned long sched_group_shares(struct task_group *tg) |
8593 | { | |
8594 | return tg->shares; | |
8595 | } | |
052f1dc7 | 8596 | #endif |
5cb350ba | 8597 | |
052f1dc7 | 8598 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8599 | /* |
9f0c1e56 | 8600 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8601 | */ |
9f0c1e56 PZ |
8602 | static DEFINE_MUTEX(rt_constraints_mutex); |
8603 | ||
8604 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8605 | { | |
8606 | if (runtime == RUNTIME_INF) | |
8607 | return 1ULL << 16; | |
8608 | ||
6f6d6a1a | 8609 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8610 | } |
8611 | ||
b40b2e8e PZ |
8612 | #ifdef CONFIG_CGROUP_SCHED |
8613 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8614 | { | |
10b612f4 | 8615 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8616 | unsigned long total = 0; |
8617 | ||
8618 | if (!parent) { | |
8619 | if (global_rt_period() < period) | |
8620 | return 0; | |
8621 | ||
8622 | return to_ratio(period, runtime) < | |
8623 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8624 | } | |
8625 | ||
8626 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8627 | return 0; | |
8628 | ||
8629 | rcu_read_lock(); | |
8630 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8631 | if (tgi == tg) | |
8632 | continue; | |
8633 | ||
8634 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8635 | tgi->rt_bandwidth.rt_runtime); | |
8636 | } | |
8637 | rcu_read_unlock(); | |
8638 | ||
10b612f4 | 8639 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8640 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8641 | parent->rt_bandwidth.rt_runtime); | |
8642 | } | |
8643 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8644 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8645 | { |
8646 | struct task_group *tgi; | |
8647 | unsigned long total = 0; | |
9f0c1e56 | 8648 | unsigned long global_ratio = |
d0b27fa7 | 8649 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8650 | |
8651 | rcu_read_lock(); | |
9f0c1e56 PZ |
8652 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8653 | if (tgi == tg) | |
8654 | continue; | |
6f505b16 | 8655 | |
d0b27fa7 PZ |
8656 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8657 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8658 | } |
8659 | rcu_read_unlock(); | |
6f505b16 | 8660 | |
9f0c1e56 | 8661 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8662 | } |
b40b2e8e | 8663 | #endif |
6f505b16 | 8664 | |
521f1a24 DG |
8665 | /* Must be called with tasklist_lock held */ |
8666 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8667 | { | |
8668 | struct task_struct *g, *p; | |
8669 | do_each_thread(g, p) { | |
8670 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8671 | return 1; | |
8672 | } while_each_thread(g, p); | |
8673 | return 0; | |
8674 | } | |
8675 | ||
d0b27fa7 PZ |
8676 | static int tg_set_bandwidth(struct task_group *tg, |
8677 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8678 | { |
ac086bc2 | 8679 | int i, err = 0; |
9f0c1e56 | 8680 | |
9f0c1e56 | 8681 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8682 | read_lock(&tasklist_lock); |
ac086bc2 | 8683 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8684 | err = -EBUSY; |
8685 | goto unlock; | |
8686 | } | |
9f0c1e56 PZ |
8687 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8688 | err = -EINVAL; | |
8689 | goto unlock; | |
8690 | } | |
ac086bc2 PZ |
8691 | |
8692 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8693 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8694 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8695 | |
8696 | for_each_possible_cpu(i) { | |
8697 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8698 | ||
8699 | spin_lock(&rt_rq->rt_runtime_lock); | |
8700 | rt_rq->rt_runtime = rt_runtime; | |
8701 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8702 | } | |
8703 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8704 | unlock: |
521f1a24 | 8705 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8706 | mutex_unlock(&rt_constraints_mutex); |
8707 | ||
8708 | return err; | |
6f505b16 PZ |
8709 | } |
8710 | ||
d0b27fa7 PZ |
8711 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8712 | { | |
8713 | u64 rt_runtime, rt_period; | |
8714 | ||
8715 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8716 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8717 | if (rt_runtime_us < 0) | |
8718 | rt_runtime = RUNTIME_INF; | |
8719 | ||
8720 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8721 | } | |
8722 | ||
9f0c1e56 PZ |
8723 | long sched_group_rt_runtime(struct task_group *tg) |
8724 | { | |
8725 | u64 rt_runtime_us; | |
8726 | ||
d0b27fa7 | 8727 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8728 | return -1; |
8729 | ||
d0b27fa7 | 8730 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8731 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8732 | return rt_runtime_us; | |
8733 | } | |
d0b27fa7 PZ |
8734 | |
8735 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8736 | { | |
8737 | u64 rt_runtime, rt_period; | |
8738 | ||
8739 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8740 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8741 | ||
619b0488 R |
8742 | if (rt_period == 0) |
8743 | return -EINVAL; | |
8744 | ||
d0b27fa7 PZ |
8745 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8746 | } | |
8747 | ||
8748 | long sched_group_rt_period(struct task_group *tg) | |
8749 | { | |
8750 | u64 rt_period_us; | |
8751 | ||
8752 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8753 | do_div(rt_period_us, NSEC_PER_USEC); | |
8754 | return rt_period_us; | |
8755 | } | |
8756 | ||
8757 | static int sched_rt_global_constraints(void) | |
8758 | { | |
10b612f4 PZ |
8759 | struct task_group *tg = &root_task_group; |
8760 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8761 | int ret = 0; |
8762 | ||
10b612f4 PZ |
8763 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8764 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8765 | ||
d0b27fa7 | 8766 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8767 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8768 | ret = -EINVAL; |
8769 | mutex_unlock(&rt_constraints_mutex); | |
8770 | ||
8771 | return ret; | |
8772 | } | |
6d6bc0ad | 8773 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8774 | static int sched_rt_global_constraints(void) |
8775 | { | |
ac086bc2 PZ |
8776 | unsigned long flags; |
8777 | int i; | |
8778 | ||
8779 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8780 | for_each_possible_cpu(i) { | |
8781 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8782 | ||
8783 | spin_lock(&rt_rq->rt_runtime_lock); | |
8784 | rt_rq->rt_runtime = global_rt_runtime(); | |
8785 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8786 | } | |
8787 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8788 | ||
d0b27fa7 PZ |
8789 | return 0; |
8790 | } | |
6d6bc0ad | 8791 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8792 | |
8793 | int sched_rt_handler(struct ctl_table *table, int write, | |
8794 | struct file *filp, void __user *buffer, size_t *lenp, | |
8795 | loff_t *ppos) | |
8796 | { | |
8797 | int ret; | |
8798 | int old_period, old_runtime; | |
8799 | static DEFINE_MUTEX(mutex); | |
8800 | ||
8801 | mutex_lock(&mutex); | |
8802 | old_period = sysctl_sched_rt_period; | |
8803 | old_runtime = sysctl_sched_rt_runtime; | |
8804 | ||
8805 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8806 | ||
8807 | if (!ret && write) { | |
8808 | ret = sched_rt_global_constraints(); | |
8809 | if (ret) { | |
8810 | sysctl_sched_rt_period = old_period; | |
8811 | sysctl_sched_rt_runtime = old_runtime; | |
8812 | } else { | |
8813 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8814 | def_rt_bandwidth.rt_period = | |
8815 | ns_to_ktime(global_rt_period()); | |
8816 | } | |
8817 | } | |
8818 | mutex_unlock(&mutex); | |
8819 | ||
8820 | return ret; | |
8821 | } | |
68318b8e | 8822 | |
052f1dc7 | 8823 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8824 | |
8825 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8826 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8827 | { |
2b01dfe3 PM |
8828 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8829 | struct task_group, css); | |
68318b8e SV |
8830 | } |
8831 | ||
8832 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8833 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8834 | { |
ec7dc8ac | 8835 | struct task_group *tg, *parent; |
68318b8e | 8836 | |
2b01dfe3 | 8837 | if (!cgrp->parent) { |
68318b8e | 8838 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8839 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8840 | return &init_task_group.css; |
8841 | } | |
8842 | ||
ec7dc8ac DG |
8843 | parent = cgroup_tg(cgrp->parent); |
8844 | tg = sched_create_group(parent); | |
68318b8e SV |
8845 | if (IS_ERR(tg)) |
8846 | return ERR_PTR(-ENOMEM); | |
8847 | ||
8848 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8849 | tg->css.cgroup = cgrp; |
68318b8e SV |
8850 | |
8851 | return &tg->css; | |
8852 | } | |
8853 | ||
41a2d6cf IM |
8854 | static void |
8855 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8856 | { |
2b01dfe3 | 8857 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8858 | |
8859 | sched_destroy_group(tg); | |
8860 | } | |
8861 | ||
41a2d6cf IM |
8862 | static int |
8863 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8864 | struct task_struct *tsk) | |
68318b8e | 8865 | { |
b68aa230 PZ |
8866 | #ifdef CONFIG_RT_GROUP_SCHED |
8867 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8868 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8869 | return -EINVAL; |
8870 | #else | |
68318b8e SV |
8871 | /* We don't support RT-tasks being in separate groups */ |
8872 | if (tsk->sched_class != &fair_sched_class) | |
8873 | return -EINVAL; | |
b68aa230 | 8874 | #endif |
68318b8e SV |
8875 | |
8876 | return 0; | |
8877 | } | |
8878 | ||
8879 | static void | |
2b01dfe3 | 8880 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8881 | struct cgroup *old_cont, struct task_struct *tsk) |
8882 | { | |
8883 | sched_move_task(tsk); | |
8884 | } | |
8885 | ||
052f1dc7 | 8886 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8887 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8888 | u64 shareval) |
68318b8e | 8889 | { |
2b01dfe3 | 8890 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8891 | } |
8892 | ||
f4c753b7 | 8893 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8894 | { |
2b01dfe3 | 8895 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8896 | |
8897 | return (u64) tg->shares; | |
8898 | } | |
6d6bc0ad | 8899 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8900 | |
052f1dc7 | 8901 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8902 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8903 | s64 val) |
6f505b16 | 8904 | { |
06ecb27c | 8905 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8906 | } |
8907 | ||
06ecb27c | 8908 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8909 | { |
06ecb27c | 8910 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8911 | } |
d0b27fa7 PZ |
8912 | |
8913 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8914 | u64 rt_period_us) | |
8915 | { | |
8916 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8917 | } | |
8918 | ||
8919 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8920 | { | |
8921 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8922 | } | |
6d6bc0ad | 8923 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8924 | |
fe5c7cc2 | 8925 | static struct cftype cpu_files[] = { |
052f1dc7 | 8926 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8927 | { |
8928 | .name = "shares", | |
f4c753b7 PM |
8929 | .read_u64 = cpu_shares_read_u64, |
8930 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8931 | }, |
052f1dc7 PZ |
8932 | #endif |
8933 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8934 | { |
9f0c1e56 | 8935 | .name = "rt_runtime_us", |
06ecb27c PM |
8936 | .read_s64 = cpu_rt_runtime_read, |
8937 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8938 | }, |
d0b27fa7 PZ |
8939 | { |
8940 | .name = "rt_period_us", | |
f4c753b7 PM |
8941 | .read_u64 = cpu_rt_period_read_uint, |
8942 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8943 | }, |
052f1dc7 | 8944 | #endif |
68318b8e SV |
8945 | }; |
8946 | ||
8947 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8948 | { | |
fe5c7cc2 | 8949 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8950 | } |
8951 | ||
8952 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8953 | .name = "cpu", |
8954 | .create = cpu_cgroup_create, | |
8955 | .destroy = cpu_cgroup_destroy, | |
8956 | .can_attach = cpu_cgroup_can_attach, | |
8957 | .attach = cpu_cgroup_attach, | |
8958 | .populate = cpu_cgroup_populate, | |
8959 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8960 | .early_init = 1, |
8961 | }; | |
8962 | ||
052f1dc7 | 8963 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8964 | |
8965 | #ifdef CONFIG_CGROUP_CPUACCT | |
8966 | ||
8967 | /* | |
8968 | * CPU accounting code for task groups. | |
8969 | * | |
8970 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8971 | * (balbir@in.ibm.com). | |
8972 | */ | |
8973 | ||
8974 | /* track cpu usage of a group of tasks */ | |
8975 | struct cpuacct { | |
8976 | struct cgroup_subsys_state css; | |
8977 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8978 | u64 *cpuusage; | |
8979 | }; | |
8980 | ||
8981 | struct cgroup_subsys cpuacct_subsys; | |
8982 | ||
8983 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8984 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8985 | { |
32cd756a | 8986 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8987 | struct cpuacct, css); |
8988 | } | |
8989 | ||
8990 | /* return cpu accounting group to which this task belongs */ | |
8991 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8992 | { | |
8993 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8994 | struct cpuacct, css); | |
8995 | } | |
8996 | ||
8997 | /* create a new cpu accounting group */ | |
8998 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8999 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9000 | { |
9001 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9002 | ||
9003 | if (!ca) | |
9004 | return ERR_PTR(-ENOMEM); | |
9005 | ||
9006 | ca->cpuusage = alloc_percpu(u64); | |
9007 | if (!ca->cpuusage) { | |
9008 | kfree(ca); | |
9009 | return ERR_PTR(-ENOMEM); | |
9010 | } | |
9011 | ||
9012 | return &ca->css; | |
9013 | } | |
9014 | ||
9015 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9016 | static void |
32cd756a | 9017 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9018 | { |
32cd756a | 9019 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9020 | |
9021 | free_percpu(ca->cpuusage); | |
9022 | kfree(ca); | |
9023 | } | |
9024 | ||
9025 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9026 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9027 | { |
32cd756a | 9028 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9029 | u64 totalcpuusage = 0; |
9030 | int i; | |
9031 | ||
9032 | for_each_possible_cpu(i) { | |
9033 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9034 | ||
9035 | /* | |
9036 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9037 | * platforms. | |
9038 | */ | |
9039 | spin_lock_irq(&cpu_rq(i)->lock); | |
9040 | totalcpuusage += *cpuusage; | |
9041 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9042 | } | |
9043 | ||
9044 | return totalcpuusage; | |
9045 | } | |
9046 | ||
0297b803 DG |
9047 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9048 | u64 reset) | |
9049 | { | |
9050 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9051 | int err = 0; | |
9052 | int i; | |
9053 | ||
9054 | if (reset) { | |
9055 | err = -EINVAL; | |
9056 | goto out; | |
9057 | } | |
9058 | ||
9059 | for_each_possible_cpu(i) { | |
9060 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9061 | ||
9062 | spin_lock_irq(&cpu_rq(i)->lock); | |
9063 | *cpuusage = 0; | |
9064 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9065 | } | |
9066 | out: | |
9067 | return err; | |
9068 | } | |
9069 | ||
d842de87 SV |
9070 | static struct cftype files[] = { |
9071 | { | |
9072 | .name = "usage", | |
f4c753b7 PM |
9073 | .read_u64 = cpuusage_read, |
9074 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9075 | }, |
9076 | }; | |
9077 | ||
32cd756a | 9078 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9079 | { |
32cd756a | 9080 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9081 | } |
9082 | ||
9083 | /* | |
9084 | * charge this task's execution time to its accounting group. | |
9085 | * | |
9086 | * called with rq->lock held. | |
9087 | */ | |
9088 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9089 | { | |
9090 | struct cpuacct *ca; | |
9091 | ||
9092 | if (!cpuacct_subsys.active) | |
9093 | return; | |
9094 | ||
9095 | ca = task_ca(tsk); | |
9096 | if (ca) { | |
9097 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9098 | ||
9099 | *cpuusage += cputime; | |
9100 | } | |
9101 | } | |
9102 | ||
9103 | struct cgroup_subsys cpuacct_subsys = { | |
9104 | .name = "cpuacct", | |
9105 | .create = cpuacct_create, | |
9106 | .destroy = cpuacct_destroy, | |
9107 | .populate = cpuacct_populate, | |
9108 | .subsys_id = cpuacct_subsys_id, | |
9109 | }; | |
9110 | #endif /* CONFIG_CGROUP_CPUACCT */ |