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); | |
4049 | } | |
4050 | ||
94886b84 LV |
4051 | /* |
4052 | * Account guest cpu time to a process. | |
4053 | * @p: the process that the cpu time gets accounted to | |
4054 | * @cputime: the cpu time spent in virtual machine since the last update | |
4055 | */ | |
f7402e03 | 4056 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4057 | { |
4058 | cputime64_t tmp; | |
4059 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4060 | ||
4061 | tmp = cputime_to_cputime64(cputime); | |
4062 | ||
4063 | p->utime = cputime_add(p->utime, cputime); | |
4064 | p->gtime = cputime_add(p->gtime, cputime); | |
4065 | ||
4066 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4067 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4068 | } | |
4069 | ||
c66f08be MN |
4070 | /* |
4071 | * Account scaled user cpu time to a process. | |
4072 | * @p: the process that the cpu time gets accounted to | |
4073 | * @cputime: the cpu time spent in user space since the last update | |
4074 | */ | |
4075 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4076 | { | |
4077 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4078 | } | |
4079 | ||
1da177e4 LT |
4080 | /* |
4081 | * Account system cpu time to a process. | |
4082 | * @p: the process that the cpu time gets accounted to | |
4083 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4084 | * @cputime: the cpu time spent in kernel space since the last update | |
4085 | */ | |
4086 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4087 | cputime_t cputime) | |
4088 | { | |
4089 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4090 | struct rq *rq = this_rq(); |
1da177e4 LT |
4091 | cputime64_t tmp; |
4092 | ||
983ed7a6 HH |
4093 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4094 | account_guest_time(p, cputime); | |
4095 | return; | |
4096 | } | |
94886b84 | 4097 | |
1da177e4 LT |
4098 | p->stime = cputime_add(p->stime, cputime); |
4099 | ||
4100 | /* Add system time to cpustat. */ | |
4101 | tmp = cputime_to_cputime64(cputime); | |
4102 | if (hardirq_count() - hardirq_offset) | |
4103 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4104 | else if (softirq_count()) | |
4105 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4106 | else if (p != rq->idle) |
1da177e4 | 4107 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4108 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4109 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4110 | else | |
4111 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4112 | /* Account for system time used */ | |
4113 | acct_update_integrals(p); | |
1da177e4 LT |
4114 | } |
4115 | ||
c66f08be MN |
4116 | /* |
4117 | * Account scaled system cpu time to a process. | |
4118 | * @p: the process that the cpu time gets accounted to | |
4119 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4120 | * @cputime: the cpu time spent in kernel space since the last update | |
4121 | */ | |
4122 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4123 | { | |
4124 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4125 | } | |
4126 | ||
1da177e4 LT |
4127 | /* |
4128 | * Account for involuntary wait time. | |
4129 | * @p: the process from which the cpu time has been stolen | |
4130 | * @steal: the cpu time spent in involuntary wait | |
4131 | */ | |
4132 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4133 | { | |
4134 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4135 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4136 | struct rq *rq = this_rq(); |
1da177e4 LT |
4137 | |
4138 | if (p == rq->idle) { | |
4139 | p->stime = cputime_add(p->stime, steal); | |
4140 | if (atomic_read(&rq->nr_iowait) > 0) | |
4141 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4142 | else | |
4143 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4144 | } else |
1da177e4 LT |
4145 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4146 | } | |
4147 | ||
7835b98b CL |
4148 | /* |
4149 | * This function gets called by the timer code, with HZ frequency. | |
4150 | * We call it with interrupts disabled. | |
4151 | * | |
4152 | * It also gets called by the fork code, when changing the parent's | |
4153 | * timeslices. | |
4154 | */ | |
4155 | void scheduler_tick(void) | |
4156 | { | |
7835b98b CL |
4157 | int cpu = smp_processor_id(); |
4158 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4159 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4160 | |
4161 | sched_clock_tick(); | |
dd41f596 IM |
4162 | |
4163 | spin_lock(&rq->lock); | |
3e51f33f | 4164 | update_rq_clock(rq); |
f1a438d8 | 4165 | update_cpu_load(rq); |
fa85ae24 | 4166 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4167 | spin_unlock(&rq->lock); |
7835b98b | 4168 | |
e418e1c2 | 4169 | #ifdef CONFIG_SMP |
dd41f596 IM |
4170 | rq->idle_at_tick = idle_cpu(cpu); |
4171 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4172 | #endif |
1da177e4 LT |
4173 | } |
4174 | ||
6cd8a4bb SR |
4175 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4176 | defined(CONFIG_PREEMPT_TRACER)) | |
4177 | ||
4178 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4179 | { | |
4180 | if (in_lock_functions(addr)) { | |
4181 | addr = CALLER_ADDR2; | |
4182 | if (in_lock_functions(addr)) | |
4183 | addr = CALLER_ADDR3; | |
4184 | } | |
4185 | return addr; | |
4186 | } | |
1da177e4 | 4187 | |
43627582 | 4188 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4189 | { |
6cd8a4bb | 4190 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4191 | /* |
4192 | * Underflow? | |
4193 | */ | |
9a11b49a IM |
4194 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4195 | return; | |
6cd8a4bb | 4196 | #endif |
1da177e4 | 4197 | preempt_count() += val; |
6cd8a4bb | 4198 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4199 | /* |
4200 | * Spinlock count overflowing soon? | |
4201 | */ | |
33859f7f MOS |
4202 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4203 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4204 | #endif |
4205 | if (preempt_count() == val) | |
4206 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4207 | } |
4208 | EXPORT_SYMBOL(add_preempt_count); | |
4209 | ||
43627582 | 4210 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4211 | { |
6cd8a4bb | 4212 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4213 | /* |
4214 | * Underflow? | |
4215 | */ | |
9a11b49a IM |
4216 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4217 | return; | |
1da177e4 LT |
4218 | /* |
4219 | * Is the spinlock portion underflowing? | |
4220 | */ | |
9a11b49a IM |
4221 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4222 | !(preempt_count() & PREEMPT_MASK))) | |
4223 | return; | |
6cd8a4bb | 4224 | #endif |
9a11b49a | 4225 | |
6cd8a4bb SR |
4226 | if (preempt_count() == val) |
4227 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4228 | preempt_count() -= val; |
4229 | } | |
4230 | EXPORT_SYMBOL(sub_preempt_count); | |
4231 | ||
4232 | #endif | |
4233 | ||
4234 | /* | |
dd41f596 | 4235 | * Print scheduling while atomic bug: |
1da177e4 | 4236 | */ |
dd41f596 | 4237 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4238 | { |
838225b4 SS |
4239 | struct pt_regs *regs = get_irq_regs(); |
4240 | ||
4241 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4242 | prev->comm, prev->pid, preempt_count()); | |
4243 | ||
dd41f596 | 4244 | debug_show_held_locks(prev); |
e21f5b15 | 4245 | print_modules(); |
dd41f596 IM |
4246 | if (irqs_disabled()) |
4247 | print_irqtrace_events(prev); | |
838225b4 SS |
4248 | |
4249 | if (regs) | |
4250 | show_regs(regs); | |
4251 | else | |
4252 | dump_stack(); | |
dd41f596 | 4253 | } |
1da177e4 | 4254 | |
dd41f596 IM |
4255 | /* |
4256 | * Various schedule()-time debugging checks and statistics: | |
4257 | */ | |
4258 | static inline void schedule_debug(struct task_struct *prev) | |
4259 | { | |
1da177e4 | 4260 | /* |
41a2d6cf | 4261 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4262 | * schedule() atomically, we ignore that path for now. |
4263 | * Otherwise, whine if we are scheduling when we should not be. | |
4264 | */ | |
3f33a7ce | 4265 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4266 | __schedule_bug(prev); |
4267 | ||
1da177e4 LT |
4268 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4269 | ||
2d72376b | 4270 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4271 | #ifdef CONFIG_SCHEDSTATS |
4272 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4273 | schedstat_inc(this_rq(), bkl_count); |
4274 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4275 | } |
4276 | #endif | |
dd41f596 IM |
4277 | } |
4278 | ||
4279 | /* | |
4280 | * Pick up the highest-prio task: | |
4281 | */ | |
4282 | static inline struct task_struct * | |
ff95f3df | 4283 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4284 | { |
5522d5d5 | 4285 | const struct sched_class *class; |
dd41f596 | 4286 | struct task_struct *p; |
1da177e4 LT |
4287 | |
4288 | /* | |
dd41f596 IM |
4289 | * Optimization: we know that if all tasks are in |
4290 | * the fair class we can call that function directly: | |
1da177e4 | 4291 | */ |
dd41f596 | 4292 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4293 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4294 | if (likely(p)) |
4295 | return p; | |
1da177e4 LT |
4296 | } |
4297 | ||
dd41f596 IM |
4298 | class = sched_class_highest; |
4299 | for ( ; ; ) { | |
fb8d4724 | 4300 | p = class->pick_next_task(rq); |
dd41f596 IM |
4301 | if (p) |
4302 | return p; | |
4303 | /* | |
4304 | * Will never be NULL as the idle class always | |
4305 | * returns a non-NULL p: | |
4306 | */ | |
4307 | class = class->next; | |
4308 | } | |
4309 | } | |
1da177e4 | 4310 | |
dd41f596 IM |
4311 | /* |
4312 | * schedule() is the main scheduler function. | |
4313 | */ | |
4314 | asmlinkage void __sched schedule(void) | |
4315 | { | |
4316 | struct task_struct *prev, *next; | |
67ca7bde | 4317 | unsigned long *switch_count; |
dd41f596 | 4318 | struct rq *rq; |
31656519 | 4319 | int cpu; |
dd41f596 IM |
4320 | |
4321 | need_resched: | |
4322 | preempt_disable(); | |
4323 | cpu = smp_processor_id(); | |
4324 | rq = cpu_rq(cpu); | |
4325 | rcu_qsctr_inc(cpu); | |
4326 | prev = rq->curr; | |
4327 | switch_count = &prev->nivcsw; | |
4328 | ||
4329 | release_kernel_lock(prev); | |
4330 | need_resched_nonpreemptible: | |
4331 | ||
4332 | schedule_debug(prev); | |
1da177e4 | 4333 | |
31656519 | 4334 | if (sched_feat(HRTICK)) |
f333fdc9 | 4335 | hrtick_clear(rq); |
8f4d37ec | 4336 | |
1e819950 IM |
4337 | /* |
4338 | * Do the rq-clock update outside the rq lock: | |
4339 | */ | |
4340 | local_irq_disable(); | |
3e51f33f | 4341 | update_rq_clock(rq); |
1e819950 IM |
4342 | spin_lock(&rq->lock); |
4343 | clear_tsk_need_resched(prev); | |
1da177e4 | 4344 | |
1da177e4 | 4345 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4346 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4347 | prev->state = TASK_RUNNING; |
16882c1e | 4348 | else |
2e1cb74a | 4349 | deactivate_task(rq, prev, 1); |
dd41f596 | 4350 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4351 | } |
4352 | ||
9a897c5a SR |
4353 | #ifdef CONFIG_SMP |
4354 | if (prev->sched_class->pre_schedule) | |
4355 | prev->sched_class->pre_schedule(rq, prev); | |
4356 | #endif | |
f65eda4f | 4357 | |
dd41f596 | 4358 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4359 | idle_balance(cpu, rq); |
1da177e4 | 4360 | |
31ee529c | 4361 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4362 | next = pick_next_task(rq, prev); |
1da177e4 | 4363 | |
1da177e4 | 4364 | if (likely(prev != next)) { |
673a90a1 DS |
4365 | sched_info_switch(prev, next); |
4366 | ||
1da177e4 LT |
4367 | rq->nr_switches++; |
4368 | rq->curr = next; | |
4369 | ++*switch_count; | |
4370 | ||
dd41f596 | 4371 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4372 | /* |
4373 | * the context switch might have flipped the stack from under | |
4374 | * us, hence refresh the local variables. | |
4375 | */ | |
4376 | cpu = smp_processor_id(); | |
4377 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4378 | } else |
4379 | spin_unlock_irq(&rq->lock); | |
4380 | ||
8f4d37ec | 4381 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4382 | goto need_resched_nonpreemptible; |
8f4d37ec | 4383 | |
1da177e4 LT |
4384 | preempt_enable_no_resched(); |
4385 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4386 | goto need_resched; | |
4387 | } | |
1da177e4 LT |
4388 | EXPORT_SYMBOL(schedule); |
4389 | ||
4390 | #ifdef CONFIG_PREEMPT | |
4391 | /* | |
2ed6e34f | 4392 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4393 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4394 | * occur there and call schedule directly. |
4395 | */ | |
4396 | asmlinkage void __sched preempt_schedule(void) | |
4397 | { | |
4398 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4399 | |
1da177e4 LT |
4400 | /* |
4401 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4402 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4403 | */ |
beed33a8 | 4404 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4405 | return; |
4406 | ||
3a5c359a AK |
4407 | do { |
4408 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4409 | schedule(); |
3a5c359a | 4410 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4411 | |
3a5c359a AK |
4412 | /* |
4413 | * Check again in case we missed a preemption opportunity | |
4414 | * between schedule and now. | |
4415 | */ | |
4416 | barrier(); | |
4417 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4418 | } |
1da177e4 LT |
4419 | EXPORT_SYMBOL(preempt_schedule); |
4420 | ||
4421 | /* | |
2ed6e34f | 4422 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4423 | * off of irq context. |
4424 | * Note, that this is called and return with irqs disabled. This will | |
4425 | * protect us against recursive calling from irq. | |
4426 | */ | |
4427 | asmlinkage void __sched preempt_schedule_irq(void) | |
4428 | { | |
4429 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4430 | |
2ed6e34f | 4431 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4432 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4433 | ||
3a5c359a AK |
4434 | do { |
4435 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4436 | local_irq_enable(); |
4437 | schedule(); | |
4438 | local_irq_disable(); | |
3a5c359a | 4439 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4440 | |
3a5c359a AK |
4441 | /* |
4442 | * Check again in case we missed a preemption opportunity | |
4443 | * between schedule and now. | |
4444 | */ | |
4445 | barrier(); | |
4446 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4447 | } |
4448 | ||
4449 | #endif /* CONFIG_PREEMPT */ | |
4450 | ||
95cdf3b7 IM |
4451 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4452 | void *key) | |
1da177e4 | 4453 | { |
48f24c4d | 4454 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4455 | } |
1da177e4 LT |
4456 | EXPORT_SYMBOL(default_wake_function); |
4457 | ||
4458 | /* | |
41a2d6cf IM |
4459 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4460 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4461 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4462 | * | |
4463 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4464 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4465 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4466 | */ | |
4467 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4468 | int nr_exclusive, int sync, void *key) | |
4469 | { | |
2e45874c | 4470 | wait_queue_t *curr, *next; |
1da177e4 | 4471 | |
2e45874c | 4472 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4473 | unsigned flags = curr->flags; |
4474 | ||
1da177e4 | 4475 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4476 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4477 | break; |
4478 | } | |
4479 | } | |
4480 | ||
4481 | /** | |
4482 | * __wake_up - wake up threads blocked on a waitqueue. | |
4483 | * @q: the waitqueue | |
4484 | * @mode: which threads | |
4485 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4486 | * @key: is directly passed to the wakeup function |
1da177e4 | 4487 | */ |
7ad5b3a5 | 4488 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4489 | int nr_exclusive, void *key) |
1da177e4 LT |
4490 | { |
4491 | unsigned long flags; | |
4492 | ||
4493 | spin_lock_irqsave(&q->lock, flags); | |
4494 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4495 | spin_unlock_irqrestore(&q->lock, flags); | |
4496 | } | |
1da177e4 LT |
4497 | EXPORT_SYMBOL(__wake_up); |
4498 | ||
4499 | /* | |
4500 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4501 | */ | |
7ad5b3a5 | 4502 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4503 | { |
4504 | __wake_up_common(q, mode, 1, 0, NULL); | |
4505 | } | |
4506 | ||
4507 | /** | |
67be2dd1 | 4508 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4509 | * @q: the waitqueue |
4510 | * @mode: which threads | |
4511 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4512 | * | |
4513 | * The sync wakeup differs that the waker knows that it will schedule | |
4514 | * away soon, so while the target thread will be woken up, it will not | |
4515 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4516 | * with each other. This can prevent needless bouncing between CPUs. | |
4517 | * | |
4518 | * On UP it can prevent extra preemption. | |
4519 | */ | |
7ad5b3a5 | 4520 | void |
95cdf3b7 | 4521 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4522 | { |
4523 | unsigned long flags; | |
4524 | int sync = 1; | |
4525 | ||
4526 | if (unlikely(!q)) | |
4527 | return; | |
4528 | ||
4529 | if (unlikely(!nr_exclusive)) | |
4530 | sync = 0; | |
4531 | ||
4532 | spin_lock_irqsave(&q->lock, flags); | |
4533 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4534 | spin_unlock_irqrestore(&q->lock, flags); | |
4535 | } | |
4536 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4537 | ||
b15136e9 | 4538 | void complete(struct completion *x) |
1da177e4 LT |
4539 | { |
4540 | unsigned long flags; | |
4541 | ||
4542 | spin_lock_irqsave(&x->wait.lock, flags); | |
4543 | x->done++; | |
d9514f6c | 4544 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4545 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4546 | } | |
4547 | EXPORT_SYMBOL(complete); | |
4548 | ||
b15136e9 | 4549 | void complete_all(struct completion *x) |
1da177e4 LT |
4550 | { |
4551 | unsigned long flags; | |
4552 | ||
4553 | spin_lock_irqsave(&x->wait.lock, flags); | |
4554 | x->done += UINT_MAX/2; | |
d9514f6c | 4555 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4556 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4557 | } | |
4558 | EXPORT_SYMBOL(complete_all); | |
4559 | ||
8cbbe86d AK |
4560 | static inline long __sched |
4561 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4562 | { |
1da177e4 LT |
4563 | if (!x->done) { |
4564 | DECLARE_WAITQUEUE(wait, current); | |
4565 | ||
4566 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4567 | __add_wait_queue_tail(&x->wait, &wait); | |
4568 | do { | |
009e577e MW |
4569 | if ((state == TASK_INTERRUPTIBLE && |
4570 | signal_pending(current)) || | |
4571 | (state == TASK_KILLABLE && | |
4572 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4573 | timeout = -ERESTARTSYS; |
4574 | break; | |
8cbbe86d AK |
4575 | } |
4576 | __set_current_state(state); | |
1da177e4 LT |
4577 | spin_unlock_irq(&x->wait.lock); |
4578 | timeout = schedule_timeout(timeout); | |
4579 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4580 | } while (!x->done && timeout); |
1da177e4 | 4581 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4582 | if (!x->done) |
4583 | return timeout; | |
1da177e4 LT |
4584 | } |
4585 | x->done--; | |
ea71a546 | 4586 | return timeout ?: 1; |
1da177e4 | 4587 | } |
1da177e4 | 4588 | |
8cbbe86d AK |
4589 | static long __sched |
4590 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4591 | { |
1da177e4 LT |
4592 | might_sleep(); |
4593 | ||
4594 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4595 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4596 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4597 | return timeout; |
4598 | } | |
1da177e4 | 4599 | |
b15136e9 | 4600 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4601 | { |
4602 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4603 | } |
8cbbe86d | 4604 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4605 | |
b15136e9 | 4606 | unsigned long __sched |
8cbbe86d | 4607 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4608 | { |
8cbbe86d | 4609 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4610 | } |
8cbbe86d | 4611 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4612 | |
8cbbe86d | 4613 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4614 | { |
51e97990 AK |
4615 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4616 | if (t == -ERESTARTSYS) | |
4617 | return t; | |
4618 | return 0; | |
0fec171c | 4619 | } |
8cbbe86d | 4620 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4621 | |
b15136e9 | 4622 | unsigned long __sched |
8cbbe86d AK |
4623 | wait_for_completion_interruptible_timeout(struct completion *x, |
4624 | unsigned long timeout) | |
0fec171c | 4625 | { |
8cbbe86d | 4626 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4627 | } |
8cbbe86d | 4628 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4629 | |
009e577e MW |
4630 | int __sched wait_for_completion_killable(struct completion *x) |
4631 | { | |
4632 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4633 | if (t == -ERESTARTSYS) | |
4634 | return t; | |
4635 | return 0; | |
4636 | } | |
4637 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4638 | ||
8cbbe86d AK |
4639 | static long __sched |
4640 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4641 | { |
0fec171c IM |
4642 | unsigned long flags; |
4643 | wait_queue_t wait; | |
4644 | ||
4645 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4646 | |
8cbbe86d | 4647 | __set_current_state(state); |
1da177e4 | 4648 | |
8cbbe86d AK |
4649 | spin_lock_irqsave(&q->lock, flags); |
4650 | __add_wait_queue(q, &wait); | |
4651 | spin_unlock(&q->lock); | |
4652 | timeout = schedule_timeout(timeout); | |
4653 | spin_lock_irq(&q->lock); | |
4654 | __remove_wait_queue(q, &wait); | |
4655 | spin_unlock_irqrestore(&q->lock, flags); | |
4656 | ||
4657 | return timeout; | |
4658 | } | |
4659 | ||
4660 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4661 | { | |
4662 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4663 | } |
1da177e4 LT |
4664 | EXPORT_SYMBOL(interruptible_sleep_on); |
4665 | ||
0fec171c | 4666 | long __sched |
95cdf3b7 | 4667 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4668 | { |
8cbbe86d | 4669 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4670 | } |
1da177e4 LT |
4671 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4672 | ||
0fec171c | 4673 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4674 | { |
8cbbe86d | 4675 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4676 | } |
1da177e4 LT |
4677 | EXPORT_SYMBOL(sleep_on); |
4678 | ||
0fec171c | 4679 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4680 | { |
8cbbe86d | 4681 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4682 | } |
1da177e4 LT |
4683 | EXPORT_SYMBOL(sleep_on_timeout); |
4684 | ||
b29739f9 IM |
4685 | #ifdef CONFIG_RT_MUTEXES |
4686 | ||
4687 | /* | |
4688 | * rt_mutex_setprio - set the current priority of a task | |
4689 | * @p: task | |
4690 | * @prio: prio value (kernel-internal form) | |
4691 | * | |
4692 | * This function changes the 'effective' priority of a task. It does | |
4693 | * not touch ->normal_prio like __setscheduler(). | |
4694 | * | |
4695 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4696 | */ | |
36c8b586 | 4697 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4698 | { |
4699 | unsigned long flags; | |
83b699ed | 4700 | int oldprio, on_rq, running; |
70b97a7f | 4701 | struct rq *rq; |
cb469845 | 4702 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4703 | |
4704 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4705 | ||
4706 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4707 | update_rq_clock(rq); |
b29739f9 | 4708 | |
d5f9f942 | 4709 | oldprio = p->prio; |
dd41f596 | 4710 | on_rq = p->se.on_rq; |
051a1d1a | 4711 | running = task_current(rq, p); |
0e1f3483 | 4712 | if (on_rq) |
69be72c1 | 4713 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4714 | if (running) |
4715 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4716 | |
4717 | if (rt_prio(prio)) | |
4718 | p->sched_class = &rt_sched_class; | |
4719 | else | |
4720 | p->sched_class = &fair_sched_class; | |
4721 | ||
b29739f9 IM |
4722 | p->prio = prio; |
4723 | ||
0e1f3483 HS |
4724 | if (running) |
4725 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4726 | if (on_rq) { |
8159f87e | 4727 | enqueue_task(rq, p, 0); |
cb469845 SR |
4728 | |
4729 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4730 | } |
4731 | task_rq_unlock(rq, &flags); | |
4732 | } | |
4733 | ||
4734 | #endif | |
4735 | ||
36c8b586 | 4736 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4737 | { |
dd41f596 | 4738 | int old_prio, delta, on_rq; |
1da177e4 | 4739 | unsigned long flags; |
70b97a7f | 4740 | struct rq *rq; |
1da177e4 LT |
4741 | |
4742 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4743 | return; | |
4744 | /* | |
4745 | * We have to be careful, if called from sys_setpriority(), | |
4746 | * the task might be in the middle of scheduling on another CPU. | |
4747 | */ | |
4748 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4749 | update_rq_clock(rq); |
1da177e4 LT |
4750 | /* |
4751 | * The RT priorities are set via sched_setscheduler(), but we still | |
4752 | * allow the 'normal' nice value to be set - but as expected | |
4753 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4754 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4755 | */ |
e05606d3 | 4756 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4757 | p->static_prio = NICE_TO_PRIO(nice); |
4758 | goto out_unlock; | |
4759 | } | |
dd41f596 | 4760 | on_rq = p->se.on_rq; |
c09595f6 | 4761 | if (on_rq) |
69be72c1 | 4762 | dequeue_task(rq, p, 0); |
1da177e4 | 4763 | |
1da177e4 | 4764 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4765 | set_load_weight(p); |
b29739f9 IM |
4766 | old_prio = p->prio; |
4767 | p->prio = effective_prio(p); | |
4768 | delta = p->prio - old_prio; | |
1da177e4 | 4769 | |
dd41f596 | 4770 | if (on_rq) { |
8159f87e | 4771 | enqueue_task(rq, p, 0); |
1da177e4 | 4772 | /* |
d5f9f942 AM |
4773 | * If the task increased its priority or is running and |
4774 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4775 | */ |
d5f9f942 | 4776 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4777 | resched_task(rq->curr); |
4778 | } | |
4779 | out_unlock: | |
4780 | task_rq_unlock(rq, &flags); | |
4781 | } | |
1da177e4 LT |
4782 | EXPORT_SYMBOL(set_user_nice); |
4783 | ||
e43379f1 MM |
4784 | /* |
4785 | * can_nice - check if a task can reduce its nice value | |
4786 | * @p: task | |
4787 | * @nice: nice value | |
4788 | */ | |
36c8b586 | 4789 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4790 | { |
024f4747 MM |
4791 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4792 | int nice_rlim = 20 - nice; | |
48f24c4d | 4793 | |
e43379f1 MM |
4794 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4795 | capable(CAP_SYS_NICE)); | |
4796 | } | |
4797 | ||
1da177e4 LT |
4798 | #ifdef __ARCH_WANT_SYS_NICE |
4799 | ||
4800 | /* | |
4801 | * sys_nice - change the priority of the current process. | |
4802 | * @increment: priority increment | |
4803 | * | |
4804 | * sys_setpriority is a more generic, but much slower function that | |
4805 | * does similar things. | |
4806 | */ | |
4807 | asmlinkage long sys_nice(int increment) | |
4808 | { | |
48f24c4d | 4809 | long nice, retval; |
1da177e4 LT |
4810 | |
4811 | /* | |
4812 | * Setpriority might change our priority at the same moment. | |
4813 | * We don't have to worry. Conceptually one call occurs first | |
4814 | * and we have a single winner. | |
4815 | */ | |
e43379f1 MM |
4816 | if (increment < -40) |
4817 | increment = -40; | |
1da177e4 LT |
4818 | if (increment > 40) |
4819 | increment = 40; | |
4820 | ||
4821 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4822 | if (nice < -20) | |
4823 | nice = -20; | |
4824 | if (nice > 19) | |
4825 | nice = 19; | |
4826 | ||
e43379f1 MM |
4827 | if (increment < 0 && !can_nice(current, nice)) |
4828 | return -EPERM; | |
4829 | ||
1da177e4 LT |
4830 | retval = security_task_setnice(current, nice); |
4831 | if (retval) | |
4832 | return retval; | |
4833 | ||
4834 | set_user_nice(current, nice); | |
4835 | return 0; | |
4836 | } | |
4837 | ||
4838 | #endif | |
4839 | ||
4840 | /** | |
4841 | * task_prio - return the priority value of a given task. | |
4842 | * @p: the task in question. | |
4843 | * | |
4844 | * This is the priority value as seen by users in /proc. | |
4845 | * RT tasks are offset by -200. Normal tasks are centered | |
4846 | * around 0, value goes from -16 to +15. | |
4847 | */ | |
36c8b586 | 4848 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4849 | { |
4850 | return p->prio - MAX_RT_PRIO; | |
4851 | } | |
4852 | ||
4853 | /** | |
4854 | * task_nice - return the nice value of a given task. | |
4855 | * @p: the task in question. | |
4856 | */ | |
36c8b586 | 4857 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4858 | { |
4859 | return TASK_NICE(p); | |
4860 | } | |
150d8bed | 4861 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4862 | |
4863 | /** | |
4864 | * idle_cpu - is a given cpu idle currently? | |
4865 | * @cpu: the processor in question. | |
4866 | */ | |
4867 | int idle_cpu(int cpu) | |
4868 | { | |
4869 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4870 | } | |
4871 | ||
1da177e4 LT |
4872 | /** |
4873 | * idle_task - return the idle task for a given cpu. | |
4874 | * @cpu: the processor in question. | |
4875 | */ | |
36c8b586 | 4876 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4877 | { |
4878 | return cpu_rq(cpu)->idle; | |
4879 | } | |
4880 | ||
4881 | /** | |
4882 | * find_process_by_pid - find a process with a matching PID value. | |
4883 | * @pid: the pid in question. | |
4884 | */ | |
a9957449 | 4885 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4886 | { |
228ebcbe | 4887 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4888 | } |
4889 | ||
4890 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4891 | static void |
4892 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4893 | { |
dd41f596 | 4894 | BUG_ON(p->se.on_rq); |
48f24c4d | 4895 | |
1da177e4 | 4896 | p->policy = policy; |
dd41f596 IM |
4897 | switch (p->policy) { |
4898 | case SCHED_NORMAL: | |
4899 | case SCHED_BATCH: | |
4900 | case SCHED_IDLE: | |
4901 | p->sched_class = &fair_sched_class; | |
4902 | break; | |
4903 | case SCHED_FIFO: | |
4904 | case SCHED_RR: | |
4905 | p->sched_class = &rt_sched_class; | |
4906 | break; | |
4907 | } | |
4908 | ||
1da177e4 | 4909 | p->rt_priority = prio; |
b29739f9 IM |
4910 | p->normal_prio = normal_prio(p); |
4911 | /* we are holding p->pi_lock already */ | |
4912 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4913 | set_load_weight(p); |
1da177e4 LT |
4914 | } |
4915 | ||
961ccddd RR |
4916 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4917 | struct sched_param *param, bool user) | |
1da177e4 | 4918 | { |
83b699ed | 4919 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4920 | unsigned long flags; |
cb469845 | 4921 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4922 | struct rq *rq; |
1da177e4 | 4923 | |
66e5393a SR |
4924 | /* may grab non-irq protected spin_locks */ |
4925 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4926 | recheck: |
4927 | /* double check policy once rq lock held */ | |
4928 | if (policy < 0) | |
4929 | policy = oldpolicy = p->policy; | |
4930 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4931 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4932 | policy != SCHED_IDLE) | |
b0a9499c | 4933 | return -EINVAL; |
1da177e4 LT |
4934 | /* |
4935 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4936 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4937 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4938 | */ |
4939 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4940 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4941 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4942 | return -EINVAL; |
e05606d3 | 4943 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4944 | return -EINVAL; |
4945 | ||
37e4ab3f OC |
4946 | /* |
4947 | * Allow unprivileged RT tasks to decrease priority: | |
4948 | */ | |
961ccddd | 4949 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4950 | if (rt_policy(policy)) { |
8dc3e909 | 4951 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4952 | |
4953 | if (!lock_task_sighand(p, &flags)) | |
4954 | return -ESRCH; | |
4955 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4956 | unlock_task_sighand(p, &flags); | |
4957 | ||
4958 | /* can't set/change the rt policy */ | |
4959 | if (policy != p->policy && !rlim_rtprio) | |
4960 | return -EPERM; | |
4961 | ||
4962 | /* can't increase priority */ | |
4963 | if (param->sched_priority > p->rt_priority && | |
4964 | param->sched_priority > rlim_rtprio) | |
4965 | return -EPERM; | |
4966 | } | |
dd41f596 IM |
4967 | /* |
4968 | * Like positive nice levels, dont allow tasks to | |
4969 | * move out of SCHED_IDLE either: | |
4970 | */ | |
4971 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4972 | return -EPERM; | |
5fe1d75f | 4973 | |
37e4ab3f OC |
4974 | /* can't change other user's priorities */ |
4975 | if ((current->euid != p->euid) && | |
4976 | (current->euid != p->uid)) | |
4977 | return -EPERM; | |
4978 | } | |
1da177e4 | 4979 | |
b68aa230 PZ |
4980 | #ifdef CONFIG_RT_GROUP_SCHED |
4981 | /* | |
4982 | * Do not allow realtime tasks into groups that have no runtime | |
4983 | * assigned. | |
4984 | */ | |
961ccddd RR |
4985 | if (user |
4986 | && rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) | |
b68aa230 PZ |
4987 | return -EPERM; |
4988 | #endif | |
4989 | ||
1da177e4 LT |
4990 | retval = security_task_setscheduler(p, policy, param); |
4991 | if (retval) | |
4992 | return retval; | |
b29739f9 IM |
4993 | /* |
4994 | * make sure no PI-waiters arrive (or leave) while we are | |
4995 | * changing the priority of the task: | |
4996 | */ | |
4997 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4998 | /* |
4999 | * To be able to change p->policy safely, the apropriate | |
5000 | * runqueue lock must be held. | |
5001 | */ | |
b29739f9 | 5002 | rq = __task_rq_lock(p); |
1da177e4 LT |
5003 | /* recheck policy now with rq lock held */ |
5004 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5005 | policy = oldpolicy = -1; | |
b29739f9 IM |
5006 | __task_rq_unlock(rq); |
5007 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5008 | goto recheck; |
5009 | } | |
2daa3577 | 5010 | update_rq_clock(rq); |
dd41f596 | 5011 | on_rq = p->se.on_rq; |
051a1d1a | 5012 | running = task_current(rq, p); |
0e1f3483 | 5013 | if (on_rq) |
2e1cb74a | 5014 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5015 | if (running) |
5016 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5017 | |
1da177e4 | 5018 | oldprio = p->prio; |
dd41f596 | 5019 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5020 | |
0e1f3483 HS |
5021 | if (running) |
5022 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5023 | if (on_rq) { |
5024 | activate_task(rq, p, 0); | |
cb469845 SR |
5025 | |
5026 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5027 | } |
b29739f9 IM |
5028 | __task_rq_unlock(rq); |
5029 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5030 | ||
95e02ca9 TG |
5031 | rt_mutex_adjust_pi(p); |
5032 | ||
1da177e4 LT |
5033 | return 0; |
5034 | } | |
961ccddd RR |
5035 | |
5036 | /** | |
5037 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5038 | * @p: the task in question. | |
5039 | * @policy: new policy. | |
5040 | * @param: structure containing the new RT priority. | |
5041 | * | |
5042 | * NOTE that the task may be already dead. | |
5043 | */ | |
5044 | int sched_setscheduler(struct task_struct *p, int policy, | |
5045 | struct sched_param *param) | |
5046 | { | |
5047 | return __sched_setscheduler(p, policy, param, true); | |
5048 | } | |
1da177e4 LT |
5049 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5050 | ||
961ccddd RR |
5051 | /** |
5052 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5053 | * @p: the task in question. | |
5054 | * @policy: new policy. | |
5055 | * @param: structure containing the new RT priority. | |
5056 | * | |
5057 | * Just like sched_setscheduler, only don't bother checking if the | |
5058 | * current context has permission. For example, this is needed in | |
5059 | * stop_machine(): we create temporary high priority worker threads, | |
5060 | * but our caller might not have that capability. | |
5061 | */ | |
5062 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5063 | struct sched_param *param) | |
5064 | { | |
5065 | return __sched_setscheduler(p, policy, param, false); | |
5066 | } | |
5067 | ||
95cdf3b7 IM |
5068 | static int |
5069 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5070 | { |
1da177e4 LT |
5071 | struct sched_param lparam; |
5072 | struct task_struct *p; | |
36c8b586 | 5073 | int retval; |
1da177e4 LT |
5074 | |
5075 | if (!param || pid < 0) | |
5076 | return -EINVAL; | |
5077 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5078 | return -EFAULT; | |
5fe1d75f ON |
5079 | |
5080 | rcu_read_lock(); | |
5081 | retval = -ESRCH; | |
1da177e4 | 5082 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5083 | if (p != NULL) |
5084 | retval = sched_setscheduler(p, policy, &lparam); | |
5085 | rcu_read_unlock(); | |
36c8b586 | 5086 | |
1da177e4 LT |
5087 | return retval; |
5088 | } | |
5089 | ||
5090 | /** | |
5091 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5092 | * @pid: the pid in question. | |
5093 | * @policy: new policy. | |
5094 | * @param: structure containing the new RT priority. | |
5095 | */ | |
41a2d6cf IM |
5096 | asmlinkage long |
5097 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5098 | { |
c21761f1 JB |
5099 | /* negative values for policy are not valid */ |
5100 | if (policy < 0) | |
5101 | return -EINVAL; | |
5102 | ||
1da177e4 LT |
5103 | return do_sched_setscheduler(pid, policy, param); |
5104 | } | |
5105 | ||
5106 | /** | |
5107 | * sys_sched_setparam - set/change the RT priority of a thread | |
5108 | * @pid: the pid in question. | |
5109 | * @param: structure containing the new RT priority. | |
5110 | */ | |
5111 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5112 | { | |
5113 | return do_sched_setscheduler(pid, -1, param); | |
5114 | } | |
5115 | ||
5116 | /** | |
5117 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5118 | * @pid: the pid in question. | |
5119 | */ | |
5120 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5121 | { | |
36c8b586 | 5122 | struct task_struct *p; |
3a5c359a | 5123 | int retval; |
1da177e4 LT |
5124 | |
5125 | if (pid < 0) | |
3a5c359a | 5126 | return -EINVAL; |
1da177e4 LT |
5127 | |
5128 | retval = -ESRCH; | |
5129 | read_lock(&tasklist_lock); | |
5130 | p = find_process_by_pid(pid); | |
5131 | if (p) { | |
5132 | retval = security_task_getscheduler(p); | |
5133 | if (!retval) | |
5134 | retval = p->policy; | |
5135 | } | |
5136 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5137 | return retval; |
5138 | } | |
5139 | ||
5140 | /** | |
5141 | * sys_sched_getscheduler - get the RT priority of a thread | |
5142 | * @pid: the pid in question. | |
5143 | * @param: structure containing the RT priority. | |
5144 | */ | |
5145 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5146 | { | |
5147 | struct sched_param lp; | |
36c8b586 | 5148 | struct task_struct *p; |
3a5c359a | 5149 | int retval; |
1da177e4 LT |
5150 | |
5151 | if (!param || pid < 0) | |
3a5c359a | 5152 | return -EINVAL; |
1da177e4 LT |
5153 | |
5154 | read_lock(&tasklist_lock); | |
5155 | p = find_process_by_pid(pid); | |
5156 | retval = -ESRCH; | |
5157 | if (!p) | |
5158 | goto out_unlock; | |
5159 | ||
5160 | retval = security_task_getscheduler(p); | |
5161 | if (retval) | |
5162 | goto out_unlock; | |
5163 | ||
5164 | lp.sched_priority = p->rt_priority; | |
5165 | read_unlock(&tasklist_lock); | |
5166 | ||
5167 | /* | |
5168 | * This one might sleep, we cannot do it with a spinlock held ... | |
5169 | */ | |
5170 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5171 | ||
1da177e4 LT |
5172 | return retval; |
5173 | ||
5174 | out_unlock: | |
5175 | read_unlock(&tasklist_lock); | |
5176 | return retval; | |
5177 | } | |
5178 | ||
b53e921b | 5179 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5180 | { |
1da177e4 | 5181 | cpumask_t cpus_allowed; |
b53e921b | 5182 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5183 | struct task_struct *p; |
5184 | int retval; | |
1da177e4 | 5185 | |
95402b38 | 5186 | get_online_cpus(); |
1da177e4 LT |
5187 | read_lock(&tasklist_lock); |
5188 | ||
5189 | p = find_process_by_pid(pid); | |
5190 | if (!p) { | |
5191 | read_unlock(&tasklist_lock); | |
95402b38 | 5192 | put_online_cpus(); |
1da177e4 LT |
5193 | return -ESRCH; |
5194 | } | |
5195 | ||
5196 | /* | |
5197 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5198 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5199 | * usage count and then drop tasklist_lock. |
5200 | */ | |
5201 | get_task_struct(p); | |
5202 | read_unlock(&tasklist_lock); | |
5203 | ||
5204 | retval = -EPERM; | |
5205 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5206 | !capable(CAP_SYS_NICE)) | |
5207 | goto out_unlock; | |
5208 | ||
e7834f8f DQ |
5209 | retval = security_task_setscheduler(p, 0, NULL); |
5210 | if (retval) | |
5211 | goto out_unlock; | |
5212 | ||
f9a86fcb | 5213 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5214 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5215 | again: |
7c16ec58 | 5216 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5217 | |
8707d8b8 | 5218 | if (!retval) { |
f9a86fcb | 5219 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5220 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5221 | /* | |
5222 | * We must have raced with a concurrent cpuset | |
5223 | * update. Just reset the cpus_allowed to the | |
5224 | * cpuset's cpus_allowed | |
5225 | */ | |
5226 | new_mask = cpus_allowed; | |
5227 | goto again; | |
5228 | } | |
5229 | } | |
1da177e4 LT |
5230 | out_unlock: |
5231 | put_task_struct(p); | |
95402b38 | 5232 | put_online_cpus(); |
1da177e4 LT |
5233 | return retval; |
5234 | } | |
5235 | ||
5236 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5237 | cpumask_t *new_mask) | |
5238 | { | |
5239 | if (len < sizeof(cpumask_t)) { | |
5240 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5241 | } else if (len > sizeof(cpumask_t)) { | |
5242 | len = sizeof(cpumask_t); | |
5243 | } | |
5244 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5245 | } | |
5246 | ||
5247 | /** | |
5248 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5249 | * @pid: pid of the process | |
5250 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5251 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5252 | */ | |
5253 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5254 | unsigned long __user *user_mask_ptr) | |
5255 | { | |
5256 | cpumask_t new_mask; | |
5257 | int retval; | |
5258 | ||
5259 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5260 | if (retval) | |
5261 | return retval; | |
5262 | ||
b53e921b | 5263 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5264 | } |
5265 | ||
1da177e4 LT |
5266 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5267 | { | |
36c8b586 | 5268 | struct task_struct *p; |
1da177e4 | 5269 | int retval; |
1da177e4 | 5270 | |
95402b38 | 5271 | get_online_cpus(); |
1da177e4 LT |
5272 | read_lock(&tasklist_lock); |
5273 | ||
5274 | retval = -ESRCH; | |
5275 | p = find_process_by_pid(pid); | |
5276 | if (!p) | |
5277 | goto out_unlock; | |
5278 | ||
e7834f8f DQ |
5279 | retval = security_task_getscheduler(p); |
5280 | if (retval) | |
5281 | goto out_unlock; | |
5282 | ||
2f7016d9 | 5283 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5284 | |
5285 | out_unlock: | |
5286 | read_unlock(&tasklist_lock); | |
95402b38 | 5287 | put_online_cpus(); |
1da177e4 | 5288 | |
9531b62f | 5289 | return retval; |
1da177e4 LT |
5290 | } |
5291 | ||
5292 | /** | |
5293 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5294 | * @pid: pid of the process | |
5295 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5296 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5297 | */ | |
5298 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5299 | unsigned long __user *user_mask_ptr) | |
5300 | { | |
5301 | int ret; | |
5302 | cpumask_t mask; | |
5303 | ||
5304 | if (len < sizeof(cpumask_t)) | |
5305 | return -EINVAL; | |
5306 | ||
5307 | ret = sched_getaffinity(pid, &mask); | |
5308 | if (ret < 0) | |
5309 | return ret; | |
5310 | ||
5311 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5312 | return -EFAULT; | |
5313 | ||
5314 | return sizeof(cpumask_t); | |
5315 | } | |
5316 | ||
5317 | /** | |
5318 | * sys_sched_yield - yield the current processor to other threads. | |
5319 | * | |
dd41f596 IM |
5320 | * This function yields the current CPU to other tasks. If there are no |
5321 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5322 | */ |
5323 | asmlinkage long sys_sched_yield(void) | |
5324 | { | |
70b97a7f | 5325 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5326 | |
2d72376b | 5327 | schedstat_inc(rq, yld_count); |
4530d7ab | 5328 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5329 | |
5330 | /* | |
5331 | * Since we are going to call schedule() anyway, there's | |
5332 | * no need to preempt or enable interrupts: | |
5333 | */ | |
5334 | __release(rq->lock); | |
8a25d5de | 5335 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5336 | _raw_spin_unlock(&rq->lock); |
5337 | preempt_enable_no_resched(); | |
5338 | ||
5339 | schedule(); | |
5340 | ||
5341 | return 0; | |
5342 | } | |
5343 | ||
e7b38404 | 5344 | static void __cond_resched(void) |
1da177e4 | 5345 | { |
8e0a43d8 IM |
5346 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5347 | __might_sleep(__FILE__, __LINE__); | |
5348 | #endif | |
5bbcfd90 IM |
5349 | /* |
5350 | * The BKS might be reacquired before we have dropped | |
5351 | * PREEMPT_ACTIVE, which could trigger a second | |
5352 | * cond_resched() call. | |
5353 | */ | |
1da177e4 LT |
5354 | do { |
5355 | add_preempt_count(PREEMPT_ACTIVE); | |
5356 | schedule(); | |
5357 | sub_preempt_count(PREEMPT_ACTIVE); | |
5358 | } while (need_resched()); | |
5359 | } | |
5360 | ||
02b67cc3 | 5361 | int __sched _cond_resched(void) |
1da177e4 | 5362 | { |
9414232f IM |
5363 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5364 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5365 | __cond_resched(); |
5366 | return 1; | |
5367 | } | |
5368 | return 0; | |
5369 | } | |
02b67cc3 | 5370 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5371 | |
5372 | /* | |
5373 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5374 | * call schedule, and on return reacquire the lock. | |
5375 | * | |
41a2d6cf | 5376 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5377 | * operations here to prevent schedule() from being called twice (once via |
5378 | * spin_unlock(), once by hand). | |
5379 | */ | |
95cdf3b7 | 5380 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5381 | { |
95c354fe | 5382 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5383 | int ret = 0; |
5384 | ||
95c354fe | 5385 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5386 | spin_unlock(lock); |
95c354fe NP |
5387 | if (resched && need_resched()) |
5388 | __cond_resched(); | |
5389 | else | |
5390 | cpu_relax(); | |
6df3cecb | 5391 | ret = 1; |
1da177e4 | 5392 | spin_lock(lock); |
1da177e4 | 5393 | } |
6df3cecb | 5394 | return ret; |
1da177e4 | 5395 | } |
1da177e4 LT |
5396 | EXPORT_SYMBOL(cond_resched_lock); |
5397 | ||
5398 | int __sched cond_resched_softirq(void) | |
5399 | { | |
5400 | BUG_ON(!in_softirq()); | |
5401 | ||
9414232f | 5402 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5403 | local_bh_enable(); |
1da177e4 LT |
5404 | __cond_resched(); |
5405 | local_bh_disable(); | |
5406 | return 1; | |
5407 | } | |
5408 | return 0; | |
5409 | } | |
1da177e4 LT |
5410 | EXPORT_SYMBOL(cond_resched_softirq); |
5411 | ||
1da177e4 LT |
5412 | /** |
5413 | * yield - yield the current processor to other threads. | |
5414 | * | |
72fd4a35 | 5415 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5416 | * thread runnable and calls sys_sched_yield(). |
5417 | */ | |
5418 | void __sched yield(void) | |
5419 | { | |
5420 | set_current_state(TASK_RUNNING); | |
5421 | sys_sched_yield(); | |
5422 | } | |
1da177e4 LT |
5423 | EXPORT_SYMBOL(yield); |
5424 | ||
5425 | /* | |
41a2d6cf | 5426 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5427 | * that process accounting knows that this is a task in IO wait state. |
5428 | * | |
5429 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5430 | * has set its backing_dev_info: the queue against which it should throttle) | |
5431 | */ | |
5432 | void __sched io_schedule(void) | |
5433 | { | |
70b97a7f | 5434 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5435 | |
0ff92245 | 5436 | delayacct_blkio_start(); |
1da177e4 LT |
5437 | atomic_inc(&rq->nr_iowait); |
5438 | schedule(); | |
5439 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5440 | delayacct_blkio_end(); |
1da177e4 | 5441 | } |
1da177e4 LT |
5442 | EXPORT_SYMBOL(io_schedule); |
5443 | ||
5444 | long __sched io_schedule_timeout(long timeout) | |
5445 | { | |
70b97a7f | 5446 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5447 | long ret; |
5448 | ||
0ff92245 | 5449 | delayacct_blkio_start(); |
1da177e4 LT |
5450 | atomic_inc(&rq->nr_iowait); |
5451 | ret = schedule_timeout(timeout); | |
5452 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5453 | delayacct_blkio_end(); |
1da177e4 LT |
5454 | return ret; |
5455 | } | |
5456 | ||
5457 | /** | |
5458 | * sys_sched_get_priority_max - return maximum RT priority. | |
5459 | * @policy: scheduling class. | |
5460 | * | |
5461 | * this syscall returns the maximum rt_priority that can be used | |
5462 | * by a given scheduling class. | |
5463 | */ | |
5464 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5465 | { | |
5466 | int ret = -EINVAL; | |
5467 | ||
5468 | switch (policy) { | |
5469 | case SCHED_FIFO: | |
5470 | case SCHED_RR: | |
5471 | ret = MAX_USER_RT_PRIO-1; | |
5472 | break; | |
5473 | case SCHED_NORMAL: | |
b0a9499c | 5474 | case SCHED_BATCH: |
dd41f596 | 5475 | case SCHED_IDLE: |
1da177e4 LT |
5476 | ret = 0; |
5477 | break; | |
5478 | } | |
5479 | return ret; | |
5480 | } | |
5481 | ||
5482 | /** | |
5483 | * sys_sched_get_priority_min - return minimum RT priority. | |
5484 | * @policy: scheduling class. | |
5485 | * | |
5486 | * this syscall returns the minimum rt_priority that can be used | |
5487 | * by a given scheduling class. | |
5488 | */ | |
5489 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5490 | { | |
5491 | int ret = -EINVAL; | |
5492 | ||
5493 | switch (policy) { | |
5494 | case SCHED_FIFO: | |
5495 | case SCHED_RR: | |
5496 | ret = 1; | |
5497 | break; | |
5498 | case SCHED_NORMAL: | |
b0a9499c | 5499 | case SCHED_BATCH: |
dd41f596 | 5500 | case SCHED_IDLE: |
1da177e4 LT |
5501 | ret = 0; |
5502 | } | |
5503 | return ret; | |
5504 | } | |
5505 | ||
5506 | /** | |
5507 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5508 | * @pid: pid of the process. | |
5509 | * @interval: userspace pointer to the timeslice value. | |
5510 | * | |
5511 | * this syscall writes the default timeslice value of a given process | |
5512 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5513 | */ | |
5514 | asmlinkage | |
5515 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5516 | { | |
36c8b586 | 5517 | struct task_struct *p; |
a4ec24b4 | 5518 | unsigned int time_slice; |
3a5c359a | 5519 | int retval; |
1da177e4 | 5520 | struct timespec t; |
1da177e4 LT |
5521 | |
5522 | if (pid < 0) | |
3a5c359a | 5523 | return -EINVAL; |
1da177e4 LT |
5524 | |
5525 | retval = -ESRCH; | |
5526 | read_lock(&tasklist_lock); | |
5527 | p = find_process_by_pid(pid); | |
5528 | if (!p) | |
5529 | goto out_unlock; | |
5530 | ||
5531 | retval = security_task_getscheduler(p); | |
5532 | if (retval) | |
5533 | goto out_unlock; | |
5534 | ||
77034937 IM |
5535 | /* |
5536 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5537 | * tasks that are on an otherwise idle runqueue: | |
5538 | */ | |
5539 | time_slice = 0; | |
5540 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5541 | time_slice = DEF_TIMESLICE; |
1868f958 | 5542 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5543 | struct sched_entity *se = &p->se; |
5544 | unsigned long flags; | |
5545 | struct rq *rq; | |
5546 | ||
5547 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5548 | if (rq->cfs.load.weight) |
5549 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5550 | task_rq_unlock(rq, &flags); |
5551 | } | |
1da177e4 | 5552 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5553 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5554 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5555 | return retval; |
3a5c359a | 5556 | |
1da177e4 LT |
5557 | out_unlock: |
5558 | read_unlock(&tasklist_lock); | |
5559 | return retval; | |
5560 | } | |
5561 | ||
7c731e0a | 5562 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5563 | |
82a1fcb9 | 5564 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5565 | { |
1da177e4 | 5566 | unsigned long free = 0; |
36c8b586 | 5567 | unsigned state; |
1da177e4 | 5568 | |
1da177e4 | 5569 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5570 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5571 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5572 | #if BITS_PER_LONG == 32 |
1da177e4 | 5573 | if (state == TASK_RUNNING) |
cc4ea795 | 5574 | printk(KERN_CONT " running "); |
1da177e4 | 5575 | else |
cc4ea795 | 5576 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5577 | #else |
5578 | if (state == TASK_RUNNING) | |
cc4ea795 | 5579 | printk(KERN_CONT " running task "); |
1da177e4 | 5580 | else |
cc4ea795 | 5581 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5582 | #endif |
5583 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5584 | { | |
10ebffde | 5585 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5586 | while (!*n) |
5587 | n++; | |
10ebffde | 5588 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5589 | } |
5590 | #endif | |
ba25f9dc | 5591 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5592 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5593 | |
5fb5e6de | 5594 | show_stack(p, NULL); |
1da177e4 LT |
5595 | } |
5596 | ||
e59e2ae2 | 5597 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5598 | { |
36c8b586 | 5599 | struct task_struct *g, *p; |
1da177e4 | 5600 | |
4bd77321 IM |
5601 | #if BITS_PER_LONG == 32 |
5602 | printk(KERN_INFO | |
5603 | " task PC stack pid father\n"); | |
1da177e4 | 5604 | #else |
4bd77321 IM |
5605 | printk(KERN_INFO |
5606 | " task PC stack pid father\n"); | |
1da177e4 LT |
5607 | #endif |
5608 | read_lock(&tasklist_lock); | |
5609 | do_each_thread(g, p) { | |
5610 | /* | |
5611 | * reset the NMI-timeout, listing all files on a slow | |
5612 | * console might take alot of time: | |
5613 | */ | |
5614 | touch_nmi_watchdog(); | |
39bc89fd | 5615 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5616 | sched_show_task(p); |
1da177e4 LT |
5617 | } while_each_thread(g, p); |
5618 | ||
04c9167f JF |
5619 | touch_all_softlockup_watchdogs(); |
5620 | ||
dd41f596 IM |
5621 | #ifdef CONFIG_SCHED_DEBUG |
5622 | sysrq_sched_debug_show(); | |
5623 | #endif | |
1da177e4 | 5624 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5625 | /* |
5626 | * Only show locks if all tasks are dumped: | |
5627 | */ | |
5628 | if (state_filter == -1) | |
5629 | debug_show_all_locks(); | |
1da177e4 LT |
5630 | } |
5631 | ||
1df21055 IM |
5632 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5633 | { | |
dd41f596 | 5634 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5635 | } |
5636 | ||
f340c0d1 IM |
5637 | /** |
5638 | * init_idle - set up an idle thread for a given CPU | |
5639 | * @idle: task in question | |
5640 | * @cpu: cpu the idle task belongs to | |
5641 | * | |
5642 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5643 | * flag, to make booting more robust. | |
5644 | */ | |
5c1e1767 | 5645 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5646 | { |
70b97a7f | 5647 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5648 | unsigned long flags; |
5649 | ||
dd41f596 IM |
5650 | __sched_fork(idle); |
5651 | idle->se.exec_start = sched_clock(); | |
5652 | ||
b29739f9 | 5653 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5654 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5655 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5656 | |
5657 | spin_lock_irqsave(&rq->lock, flags); | |
5658 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5659 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5660 | idle->oncpu = 1; | |
5661 | #endif | |
1da177e4 LT |
5662 | spin_unlock_irqrestore(&rq->lock, flags); |
5663 | ||
5664 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5665 | #if defined(CONFIG_PREEMPT) |
5666 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5667 | #else | |
a1261f54 | 5668 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5669 | #endif |
dd41f596 IM |
5670 | /* |
5671 | * The idle tasks have their own, simple scheduling class: | |
5672 | */ | |
5673 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5674 | } |
5675 | ||
5676 | /* | |
5677 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5678 | * indicates which cpus entered this state. This is used | |
5679 | * in the rcu update to wait only for active cpus. For system | |
5680 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5681 | * always be CPU_MASK_NONE. | |
5682 | */ | |
5683 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5684 | ||
19978ca6 IM |
5685 | /* |
5686 | * Increase the granularity value when there are more CPUs, | |
5687 | * because with more CPUs the 'effective latency' as visible | |
5688 | * to users decreases. But the relationship is not linear, | |
5689 | * so pick a second-best guess by going with the log2 of the | |
5690 | * number of CPUs. | |
5691 | * | |
5692 | * This idea comes from the SD scheduler of Con Kolivas: | |
5693 | */ | |
5694 | static inline void sched_init_granularity(void) | |
5695 | { | |
5696 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5697 | const unsigned long limit = 200000000; | |
5698 | ||
5699 | sysctl_sched_min_granularity *= factor; | |
5700 | if (sysctl_sched_min_granularity > limit) | |
5701 | sysctl_sched_min_granularity = limit; | |
5702 | ||
5703 | sysctl_sched_latency *= factor; | |
5704 | if (sysctl_sched_latency > limit) | |
5705 | sysctl_sched_latency = limit; | |
5706 | ||
5707 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5708 | } |
5709 | ||
1da177e4 LT |
5710 | #ifdef CONFIG_SMP |
5711 | /* | |
5712 | * This is how migration works: | |
5713 | * | |
70b97a7f | 5714 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5715 | * runqueue and wake up that CPU's migration thread. |
5716 | * 2) we down() the locked semaphore => thread blocks. | |
5717 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5718 | * thread off the CPU) | |
5719 | * 4) it gets the migration request and checks whether the migrated | |
5720 | * task is still in the wrong runqueue. | |
5721 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5722 | * it and puts it into the right queue. | |
5723 | * 6) migration thread up()s the semaphore. | |
5724 | * 7) we wake up and the migration is done. | |
5725 | */ | |
5726 | ||
5727 | /* | |
5728 | * Change a given task's CPU affinity. Migrate the thread to a | |
5729 | * proper CPU and schedule it away if the CPU it's executing on | |
5730 | * is removed from the allowed bitmask. | |
5731 | * | |
5732 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5733 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5734 | * call is not atomic; no spinlocks may be held. |
5735 | */ | |
cd8ba7cd | 5736 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5737 | { |
70b97a7f | 5738 | struct migration_req req; |
1da177e4 | 5739 | unsigned long flags; |
70b97a7f | 5740 | struct rq *rq; |
48f24c4d | 5741 | int ret = 0; |
1da177e4 LT |
5742 | |
5743 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5744 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5745 | ret = -EINVAL; |
5746 | goto out; | |
5747 | } | |
5748 | ||
9985b0ba DR |
5749 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5750 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5751 | ret = -EINVAL; | |
5752 | goto out; | |
5753 | } | |
5754 | ||
73fe6aae | 5755 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5756 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5757 | else { |
cd8ba7cd MT |
5758 | p->cpus_allowed = *new_mask; |
5759 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5760 | } |
5761 | ||
1da177e4 | 5762 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5763 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5764 | goto out; |
5765 | ||
cd8ba7cd | 5766 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5767 | /* Need help from migration thread: drop lock and wait. */ |
5768 | task_rq_unlock(rq, &flags); | |
5769 | wake_up_process(rq->migration_thread); | |
5770 | wait_for_completion(&req.done); | |
5771 | tlb_migrate_finish(p->mm); | |
5772 | return 0; | |
5773 | } | |
5774 | out: | |
5775 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5776 | |
1da177e4 LT |
5777 | return ret; |
5778 | } | |
cd8ba7cd | 5779 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5780 | |
5781 | /* | |
41a2d6cf | 5782 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5783 | * this because either it can't run here any more (set_cpus_allowed() |
5784 | * away from this CPU, or CPU going down), or because we're | |
5785 | * attempting to rebalance this task on exec (sched_exec). | |
5786 | * | |
5787 | * So we race with normal scheduler movements, but that's OK, as long | |
5788 | * as the task is no longer on this CPU. | |
efc30814 KK |
5789 | * |
5790 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5791 | */ |
efc30814 | 5792 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5793 | { |
70b97a7f | 5794 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5795 | int ret = 0, on_rq; |
1da177e4 | 5796 | |
e761b772 | 5797 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5798 | return ret; |
1da177e4 LT |
5799 | |
5800 | rq_src = cpu_rq(src_cpu); | |
5801 | rq_dest = cpu_rq(dest_cpu); | |
5802 | ||
5803 | double_rq_lock(rq_src, rq_dest); | |
5804 | /* Already moved. */ | |
5805 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5806 | goto done; |
1da177e4 LT |
5807 | /* Affinity changed (again). */ |
5808 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 5809 | goto fail; |
1da177e4 | 5810 | |
dd41f596 | 5811 | on_rq = p->se.on_rq; |
6e82a3be | 5812 | if (on_rq) |
2e1cb74a | 5813 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5814 | |
1da177e4 | 5815 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5816 | if (on_rq) { |
5817 | activate_task(rq_dest, p, 0); | |
5818 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5819 | } |
b1e38734 | 5820 | done: |
efc30814 | 5821 | ret = 1; |
b1e38734 | 5822 | fail: |
1da177e4 | 5823 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5824 | return ret; |
1da177e4 LT |
5825 | } |
5826 | ||
5827 | /* | |
5828 | * migration_thread - this is a highprio system thread that performs | |
5829 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5830 | * another runqueue. | |
5831 | */ | |
95cdf3b7 | 5832 | static int migration_thread(void *data) |
1da177e4 | 5833 | { |
1da177e4 | 5834 | int cpu = (long)data; |
70b97a7f | 5835 | struct rq *rq; |
1da177e4 LT |
5836 | |
5837 | rq = cpu_rq(cpu); | |
5838 | BUG_ON(rq->migration_thread != current); | |
5839 | ||
5840 | set_current_state(TASK_INTERRUPTIBLE); | |
5841 | while (!kthread_should_stop()) { | |
70b97a7f | 5842 | struct migration_req *req; |
1da177e4 | 5843 | struct list_head *head; |
1da177e4 | 5844 | |
1da177e4 LT |
5845 | spin_lock_irq(&rq->lock); |
5846 | ||
5847 | if (cpu_is_offline(cpu)) { | |
5848 | spin_unlock_irq(&rq->lock); | |
5849 | goto wait_to_die; | |
5850 | } | |
5851 | ||
5852 | if (rq->active_balance) { | |
5853 | active_load_balance(rq, cpu); | |
5854 | rq->active_balance = 0; | |
5855 | } | |
5856 | ||
5857 | head = &rq->migration_queue; | |
5858 | ||
5859 | if (list_empty(head)) { | |
5860 | spin_unlock_irq(&rq->lock); | |
5861 | schedule(); | |
5862 | set_current_state(TASK_INTERRUPTIBLE); | |
5863 | continue; | |
5864 | } | |
70b97a7f | 5865 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5866 | list_del_init(head->next); |
5867 | ||
674311d5 NP |
5868 | spin_unlock(&rq->lock); |
5869 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5870 | local_irq_enable(); | |
1da177e4 LT |
5871 | |
5872 | complete(&req->done); | |
5873 | } | |
5874 | __set_current_state(TASK_RUNNING); | |
5875 | return 0; | |
5876 | ||
5877 | wait_to_die: | |
5878 | /* Wait for kthread_stop */ | |
5879 | set_current_state(TASK_INTERRUPTIBLE); | |
5880 | while (!kthread_should_stop()) { | |
5881 | schedule(); | |
5882 | set_current_state(TASK_INTERRUPTIBLE); | |
5883 | } | |
5884 | __set_current_state(TASK_RUNNING); | |
5885 | return 0; | |
5886 | } | |
5887 | ||
5888 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5889 | |
5890 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5891 | { | |
5892 | int ret; | |
5893 | ||
5894 | local_irq_disable(); | |
5895 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5896 | local_irq_enable(); | |
5897 | return ret; | |
5898 | } | |
5899 | ||
054b9108 | 5900 | /* |
3a4fa0a2 | 5901 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5902 | * NOTE: interrupts should be disabled by the caller |
5903 | */ | |
48f24c4d | 5904 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5905 | { |
efc30814 | 5906 | unsigned long flags; |
1da177e4 | 5907 | cpumask_t mask; |
70b97a7f IM |
5908 | struct rq *rq; |
5909 | int dest_cpu; | |
1da177e4 | 5910 | |
3a5c359a AK |
5911 | do { |
5912 | /* On same node? */ | |
5913 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5914 | cpus_and(mask, mask, p->cpus_allowed); | |
5915 | dest_cpu = any_online_cpu(mask); | |
5916 | ||
5917 | /* On any allowed CPU? */ | |
434d53b0 | 5918 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5919 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5920 | ||
5921 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5922 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5923 | cpumask_t cpus_allowed; |
5924 | ||
5925 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5926 | /* |
5927 | * Try to stay on the same cpuset, where the | |
5928 | * current cpuset may be a subset of all cpus. | |
5929 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5930 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5931 | * called within calls to cpuset_lock/cpuset_unlock. |
5932 | */ | |
3a5c359a | 5933 | rq = task_rq_lock(p, &flags); |
470fd646 | 5934 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5935 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5936 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5937 | |
3a5c359a AK |
5938 | /* |
5939 | * Don't tell them about moving exiting tasks or | |
5940 | * kernel threads (both mm NULL), since they never | |
5941 | * leave kernel. | |
5942 | */ | |
41a2d6cf | 5943 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5944 | printk(KERN_INFO "process %d (%s) no " |
5945 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5946 | task_pid_nr(p), p->comm, dead_cpu); |
5947 | } | |
3a5c359a | 5948 | } |
f7b4cddc | 5949 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5950 | } |
5951 | ||
5952 | /* | |
5953 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5954 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5955 | * for performance reasons the counter is not stricly tracking tasks to | |
5956 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5957 | * to keep the global sum constant after CPU-down: | |
5958 | */ | |
70b97a7f | 5959 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5960 | { |
7c16ec58 | 5961 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5962 | unsigned long flags; |
5963 | ||
5964 | local_irq_save(flags); | |
5965 | double_rq_lock(rq_src, rq_dest); | |
5966 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5967 | rq_src->nr_uninterruptible = 0; | |
5968 | double_rq_unlock(rq_src, rq_dest); | |
5969 | local_irq_restore(flags); | |
5970 | } | |
5971 | ||
5972 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5973 | static void migrate_live_tasks(int src_cpu) | |
5974 | { | |
48f24c4d | 5975 | struct task_struct *p, *t; |
1da177e4 | 5976 | |
f7b4cddc | 5977 | read_lock(&tasklist_lock); |
1da177e4 | 5978 | |
48f24c4d IM |
5979 | do_each_thread(t, p) { |
5980 | if (p == current) | |
1da177e4 LT |
5981 | continue; |
5982 | ||
48f24c4d IM |
5983 | if (task_cpu(p) == src_cpu) |
5984 | move_task_off_dead_cpu(src_cpu, p); | |
5985 | } while_each_thread(t, p); | |
1da177e4 | 5986 | |
f7b4cddc | 5987 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5988 | } |
5989 | ||
dd41f596 IM |
5990 | /* |
5991 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5992 | * It does so by boosting its priority to highest possible. |
5993 | * Used by CPU offline code. | |
1da177e4 LT |
5994 | */ |
5995 | void sched_idle_next(void) | |
5996 | { | |
48f24c4d | 5997 | int this_cpu = smp_processor_id(); |
70b97a7f | 5998 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5999 | struct task_struct *p = rq->idle; |
6000 | unsigned long flags; | |
6001 | ||
6002 | /* cpu has to be offline */ | |
48f24c4d | 6003 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6004 | |
48f24c4d IM |
6005 | /* |
6006 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6007 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6008 | */ |
6009 | spin_lock_irqsave(&rq->lock, flags); | |
6010 | ||
dd41f596 | 6011 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6012 | |
94bc9a7b DA |
6013 | update_rq_clock(rq); |
6014 | activate_task(rq, p, 0); | |
1da177e4 LT |
6015 | |
6016 | spin_unlock_irqrestore(&rq->lock, flags); | |
6017 | } | |
6018 | ||
48f24c4d IM |
6019 | /* |
6020 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6021 | * offline. |
6022 | */ | |
6023 | void idle_task_exit(void) | |
6024 | { | |
6025 | struct mm_struct *mm = current->active_mm; | |
6026 | ||
6027 | BUG_ON(cpu_online(smp_processor_id())); | |
6028 | ||
6029 | if (mm != &init_mm) | |
6030 | switch_mm(mm, &init_mm, current); | |
6031 | mmdrop(mm); | |
6032 | } | |
6033 | ||
054b9108 | 6034 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6035 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6036 | { |
70b97a7f | 6037 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6038 | |
6039 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6040 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6041 | |
6042 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6043 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6044 | |
48f24c4d | 6045 | get_task_struct(p); |
1da177e4 LT |
6046 | |
6047 | /* | |
6048 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6049 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6050 | * fine. |
6051 | */ | |
f7b4cddc | 6052 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6053 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6054 | spin_lock_irq(&rq->lock); |
1da177e4 | 6055 | |
48f24c4d | 6056 | put_task_struct(p); |
1da177e4 LT |
6057 | } |
6058 | ||
6059 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6060 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6061 | { | |
70b97a7f | 6062 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6063 | struct task_struct *next; |
48f24c4d | 6064 | |
dd41f596 IM |
6065 | for ( ; ; ) { |
6066 | if (!rq->nr_running) | |
6067 | break; | |
a8e504d2 | 6068 | update_rq_clock(rq); |
ff95f3df | 6069 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6070 | if (!next) |
6071 | break; | |
79c53799 | 6072 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6073 | migrate_dead(dead_cpu, next); |
e692ab53 | 6074 | |
1da177e4 LT |
6075 | } |
6076 | } | |
6077 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6078 | ||
e692ab53 NP |
6079 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6080 | ||
6081 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6082 | { |
6083 | .procname = "sched_domain", | |
c57baf1e | 6084 | .mode = 0555, |
e0361851 | 6085 | }, |
38605cae | 6086 | {0, }, |
e692ab53 NP |
6087 | }; |
6088 | ||
6089 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6090 | { |
c57baf1e | 6091 | .ctl_name = CTL_KERN, |
e0361851 | 6092 | .procname = "kernel", |
c57baf1e | 6093 | .mode = 0555, |
e0361851 AD |
6094 | .child = sd_ctl_dir, |
6095 | }, | |
38605cae | 6096 | {0, }, |
e692ab53 NP |
6097 | }; |
6098 | ||
6099 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6100 | { | |
6101 | struct ctl_table *entry = | |
5cf9f062 | 6102 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6103 | |
e692ab53 NP |
6104 | return entry; |
6105 | } | |
6106 | ||
6382bc90 MM |
6107 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6108 | { | |
cd790076 | 6109 | struct ctl_table *entry; |
6382bc90 | 6110 | |
cd790076 MM |
6111 | /* |
6112 | * In the intermediate directories, both the child directory and | |
6113 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6114 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6115 | * static strings and all have proc handlers. |
6116 | */ | |
6117 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6118 | if (entry->child) |
6119 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6120 | if (entry->proc_handler == NULL) |
6121 | kfree(entry->procname); | |
6122 | } | |
6382bc90 MM |
6123 | |
6124 | kfree(*tablep); | |
6125 | *tablep = NULL; | |
6126 | } | |
6127 | ||
e692ab53 | 6128 | static void |
e0361851 | 6129 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6130 | const char *procname, void *data, int maxlen, |
6131 | mode_t mode, proc_handler *proc_handler) | |
6132 | { | |
e692ab53 NP |
6133 | entry->procname = procname; |
6134 | entry->data = data; | |
6135 | entry->maxlen = maxlen; | |
6136 | entry->mode = mode; | |
6137 | entry->proc_handler = proc_handler; | |
6138 | } | |
6139 | ||
6140 | static struct ctl_table * | |
6141 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6142 | { | |
ace8b3d6 | 6143 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6144 | |
ad1cdc1d MM |
6145 | if (table == NULL) |
6146 | return NULL; | |
6147 | ||
e0361851 | 6148 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6149 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6150 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6151 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6152 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6153 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6154 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6155 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6156 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6157 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6158 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6159 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6160 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6161 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6162 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6163 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6164 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6165 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6166 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6167 | &sd->cache_nice_tries, |
6168 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6169 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6170 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6171 | /* &table[11] is terminator */ |
e692ab53 NP |
6172 | |
6173 | return table; | |
6174 | } | |
6175 | ||
9a4e7159 | 6176 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6177 | { |
6178 | struct ctl_table *entry, *table; | |
6179 | struct sched_domain *sd; | |
6180 | int domain_num = 0, i; | |
6181 | char buf[32]; | |
6182 | ||
6183 | for_each_domain(cpu, sd) | |
6184 | domain_num++; | |
6185 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6186 | if (table == NULL) |
6187 | return NULL; | |
e692ab53 NP |
6188 | |
6189 | i = 0; | |
6190 | for_each_domain(cpu, sd) { | |
6191 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6192 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6193 | entry->mode = 0555; |
e692ab53 NP |
6194 | entry->child = sd_alloc_ctl_domain_table(sd); |
6195 | entry++; | |
6196 | i++; | |
6197 | } | |
6198 | return table; | |
6199 | } | |
6200 | ||
6201 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6202 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6203 | { |
6204 | int i, cpu_num = num_online_cpus(); | |
6205 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6206 | char buf[32]; | |
6207 | ||
7378547f MM |
6208 | WARN_ON(sd_ctl_dir[0].child); |
6209 | sd_ctl_dir[0].child = entry; | |
6210 | ||
ad1cdc1d MM |
6211 | if (entry == NULL) |
6212 | return; | |
6213 | ||
97b6ea7b | 6214 | for_each_online_cpu(i) { |
e692ab53 | 6215 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6216 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6217 | entry->mode = 0555; |
e692ab53 | 6218 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6219 | entry++; |
e692ab53 | 6220 | } |
7378547f MM |
6221 | |
6222 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6223 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6224 | } | |
6382bc90 | 6225 | |
7378547f | 6226 | /* may be called multiple times per register */ |
6382bc90 MM |
6227 | static void unregister_sched_domain_sysctl(void) |
6228 | { | |
7378547f MM |
6229 | if (sd_sysctl_header) |
6230 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6231 | sd_sysctl_header = NULL; |
7378547f MM |
6232 | if (sd_ctl_dir[0].child) |
6233 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6234 | } |
e692ab53 | 6235 | #else |
6382bc90 MM |
6236 | static void register_sched_domain_sysctl(void) |
6237 | { | |
6238 | } | |
6239 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6240 | { |
6241 | } | |
6242 | #endif | |
6243 | ||
1f11eb6a GH |
6244 | static void set_rq_online(struct rq *rq) |
6245 | { | |
6246 | if (!rq->online) { | |
6247 | const struct sched_class *class; | |
6248 | ||
6249 | cpu_set(rq->cpu, rq->rd->online); | |
6250 | rq->online = 1; | |
6251 | ||
6252 | for_each_class(class) { | |
6253 | if (class->rq_online) | |
6254 | class->rq_online(rq); | |
6255 | } | |
6256 | } | |
6257 | } | |
6258 | ||
6259 | static void set_rq_offline(struct rq *rq) | |
6260 | { | |
6261 | if (rq->online) { | |
6262 | const struct sched_class *class; | |
6263 | ||
6264 | for_each_class(class) { | |
6265 | if (class->rq_offline) | |
6266 | class->rq_offline(rq); | |
6267 | } | |
6268 | ||
6269 | cpu_clear(rq->cpu, rq->rd->online); | |
6270 | rq->online = 0; | |
6271 | } | |
6272 | } | |
6273 | ||
1da177e4 LT |
6274 | /* |
6275 | * migration_call - callback that gets triggered when a CPU is added. | |
6276 | * Here we can start up the necessary migration thread for the new CPU. | |
6277 | */ | |
48f24c4d IM |
6278 | static int __cpuinit |
6279 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6280 | { |
1da177e4 | 6281 | struct task_struct *p; |
48f24c4d | 6282 | int cpu = (long)hcpu; |
1da177e4 | 6283 | unsigned long flags; |
70b97a7f | 6284 | struct rq *rq; |
1da177e4 LT |
6285 | |
6286 | switch (action) { | |
5be9361c | 6287 | |
1da177e4 | 6288 | case CPU_UP_PREPARE: |
8bb78442 | 6289 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6290 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6291 | if (IS_ERR(p)) |
6292 | return NOTIFY_BAD; | |
1da177e4 LT |
6293 | kthread_bind(p, cpu); |
6294 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6295 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6296 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6297 | task_rq_unlock(rq, &flags); |
6298 | cpu_rq(cpu)->migration_thread = p; | |
6299 | break; | |
48f24c4d | 6300 | |
1da177e4 | 6301 | case CPU_ONLINE: |
8bb78442 | 6302 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6303 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6304 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6305 | |
6306 | /* Update our root-domain */ | |
6307 | rq = cpu_rq(cpu); | |
6308 | spin_lock_irqsave(&rq->lock, flags); | |
6309 | if (rq->rd) { | |
6310 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6311 | |
6312 | set_rq_online(rq); | |
1f94ef59 GH |
6313 | } |
6314 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6315 | break; |
48f24c4d | 6316 | |
1da177e4 LT |
6317 | #ifdef CONFIG_HOTPLUG_CPU |
6318 | case CPU_UP_CANCELED: | |
8bb78442 | 6319 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6320 | if (!cpu_rq(cpu)->migration_thread) |
6321 | break; | |
41a2d6cf | 6322 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6323 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6324 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6325 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6326 | cpu_rq(cpu)->migration_thread = NULL; | |
6327 | break; | |
48f24c4d | 6328 | |
1da177e4 | 6329 | case CPU_DEAD: |
8bb78442 | 6330 | case CPU_DEAD_FROZEN: |
470fd646 | 6331 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6332 | migrate_live_tasks(cpu); |
6333 | rq = cpu_rq(cpu); | |
6334 | kthread_stop(rq->migration_thread); | |
6335 | rq->migration_thread = NULL; | |
6336 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6337 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6338 | update_rq_clock(rq); |
2e1cb74a | 6339 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6340 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6341 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6342 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6343 | migrate_dead_tasks(cpu); |
d2da272a | 6344 | spin_unlock_irq(&rq->lock); |
470fd646 | 6345 | cpuset_unlock(); |
1da177e4 LT |
6346 | migrate_nr_uninterruptible(rq); |
6347 | BUG_ON(rq->nr_running != 0); | |
6348 | ||
41a2d6cf IM |
6349 | /* |
6350 | * No need to migrate the tasks: it was best-effort if | |
6351 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6352 | * the requestors. | |
6353 | */ | |
1da177e4 LT |
6354 | spin_lock_irq(&rq->lock); |
6355 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6356 | struct migration_req *req; |
6357 | ||
1da177e4 | 6358 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6359 | struct migration_req, list); |
1da177e4 LT |
6360 | list_del_init(&req->list); |
6361 | complete(&req->done); | |
6362 | } | |
6363 | spin_unlock_irq(&rq->lock); | |
6364 | break; | |
57d885fe | 6365 | |
08f503b0 GH |
6366 | case CPU_DYING: |
6367 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6368 | /* Update our root-domain */ |
6369 | rq = cpu_rq(cpu); | |
6370 | spin_lock_irqsave(&rq->lock, flags); | |
6371 | if (rq->rd) { | |
6372 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6373 | set_rq_offline(rq); |
57d885fe GH |
6374 | } |
6375 | spin_unlock_irqrestore(&rq->lock, flags); | |
6376 | break; | |
1da177e4 LT |
6377 | #endif |
6378 | } | |
6379 | return NOTIFY_OK; | |
6380 | } | |
6381 | ||
6382 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6383 | * happens before everything else. | |
6384 | */ | |
26c2143b | 6385 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6386 | .notifier_call = migration_call, |
6387 | .priority = 10 | |
6388 | }; | |
6389 | ||
e6fe6649 | 6390 | void __init migration_init(void) |
1da177e4 LT |
6391 | { |
6392 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6393 | int err; |
48f24c4d IM |
6394 | |
6395 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6396 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6397 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6398 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6399 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6400 | } |
6401 | #endif | |
6402 | ||
6403 | #ifdef CONFIG_SMP | |
476f3534 | 6404 | |
3e9830dc | 6405 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6406 | |
099f98c8 GS |
6407 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6408 | { | |
6409 | switch (lvl) { | |
6410 | case SD_LV_NONE: | |
6411 | return "NONE"; | |
6412 | case SD_LV_SIBLING: | |
6413 | return "SIBLING"; | |
6414 | case SD_LV_MC: | |
6415 | return "MC"; | |
6416 | case SD_LV_CPU: | |
6417 | return "CPU"; | |
6418 | case SD_LV_NODE: | |
6419 | return "NODE"; | |
6420 | case SD_LV_ALLNODES: | |
6421 | return "ALLNODES"; | |
6422 | case SD_LV_MAX: | |
6423 | return "MAX"; | |
6424 | ||
6425 | } | |
6426 | return "MAX"; | |
6427 | } | |
6428 | ||
7c16ec58 MT |
6429 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6430 | cpumask_t *groupmask) | |
1da177e4 | 6431 | { |
4dcf6aff | 6432 | struct sched_group *group = sd->groups; |
434d53b0 | 6433 | char str[256]; |
1da177e4 | 6434 | |
434d53b0 | 6435 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6436 | cpus_clear(*groupmask); |
4dcf6aff IM |
6437 | |
6438 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6439 | ||
6440 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6441 | printk("does not load-balance\n"); | |
6442 | if (sd->parent) | |
6443 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6444 | " has parent"); | |
6445 | return -1; | |
41c7ce9a NP |
6446 | } |
6447 | ||
099f98c8 GS |
6448 | printk(KERN_CONT "span %s level %s\n", |
6449 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6450 | |
6451 | if (!cpu_isset(cpu, sd->span)) { | |
6452 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6453 | "CPU%d\n", cpu); | |
6454 | } | |
6455 | if (!cpu_isset(cpu, group->cpumask)) { | |
6456 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6457 | " CPU%d\n", cpu); | |
6458 | } | |
1da177e4 | 6459 | |
4dcf6aff | 6460 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6461 | do { |
4dcf6aff IM |
6462 | if (!group) { |
6463 | printk("\n"); | |
6464 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6465 | break; |
6466 | } | |
6467 | ||
4dcf6aff IM |
6468 | if (!group->__cpu_power) { |
6469 | printk(KERN_CONT "\n"); | |
6470 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6471 | "set\n"); | |
6472 | break; | |
6473 | } | |
1da177e4 | 6474 | |
4dcf6aff IM |
6475 | if (!cpus_weight(group->cpumask)) { |
6476 | printk(KERN_CONT "\n"); | |
6477 | printk(KERN_ERR "ERROR: empty group\n"); | |
6478 | break; | |
6479 | } | |
1da177e4 | 6480 | |
7c16ec58 | 6481 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6482 | printk(KERN_CONT "\n"); |
6483 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6484 | break; | |
6485 | } | |
1da177e4 | 6486 | |
7c16ec58 | 6487 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6488 | |
434d53b0 | 6489 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6490 | printk(KERN_CONT " %s", str); |
1da177e4 | 6491 | |
4dcf6aff IM |
6492 | group = group->next; |
6493 | } while (group != sd->groups); | |
6494 | printk(KERN_CONT "\n"); | |
1da177e4 | 6495 | |
7c16ec58 | 6496 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6497 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6498 | |
7c16ec58 | 6499 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6500 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6501 | "of domain->span\n"); | |
6502 | return 0; | |
6503 | } | |
1da177e4 | 6504 | |
4dcf6aff IM |
6505 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6506 | { | |
7c16ec58 | 6507 | cpumask_t *groupmask; |
4dcf6aff | 6508 | int level = 0; |
1da177e4 | 6509 | |
4dcf6aff IM |
6510 | if (!sd) { |
6511 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6512 | return; | |
6513 | } | |
1da177e4 | 6514 | |
4dcf6aff IM |
6515 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6516 | ||
7c16ec58 MT |
6517 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6518 | if (!groupmask) { | |
6519 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6520 | return; | |
6521 | } | |
6522 | ||
4dcf6aff | 6523 | for (;;) { |
7c16ec58 | 6524 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6525 | break; |
1da177e4 LT |
6526 | level++; |
6527 | sd = sd->parent; | |
33859f7f | 6528 | if (!sd) |
4dcf6aff IM |
6529 | break; |
6530 | } | |
7c16ec58 | 6531 | kfree(groupmask); |
1da177e4 | 6532 | } |
6d6bc0ad | 6533 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6534 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6535 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6536 | |
1a20ff27 | 6537 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6538 | { |
6539 | if (cpus_weight(sd->span) == 1) | |
6540 | return 1; | |
6541 | ||
6542 | /* Following flags need at least 2 groups */ | |
6543 | if (sd->flags & (SD_LOAD_BALANCE | | |
6544 | SD_BALANCE_NEWIDLE | | |
6545 | SD_BALANCE_FORK | | |
89c4710e SS |
6546 | SD_BALANCE_EXEC | |
6547 | SD_SHARE_CPUPOWER | | |
6548 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6549 | if (sd->groups != sd->groups->next) |
6550 | return 0; | |
6551 | } | |
6552 | ||
6553 | /* Following flags don't use groups */ | |
6554 | if (sd->flags & (SD_WAKE_IDLE | | |
6555 | SD_WAKE_AFFINE | | |
6556 | SD_WAKE_BALANCE)) | |
6557 | return 0; | |
6558 | ||
6559 | return 1; | |
6560 | } | |
6561 | ||
48f24c4d IM |
6562 | static int |
6563 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6564 | { |
6565 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6566 | ||
6567 | if (sd_degenerate(parent)) | |
6568 | return 1; | |
6569 | ||
6570 | if (!cpus_equal(sd->span, parent->span)) | |
6571 | return 0; | |
6572 | ||
6573 | /* Does parent contain flags not in child? */ | |
6574 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6575 | if (cflags & SD_WAKE_AFFINE) | |
6576 | pflags &= ~SD_WAKE_BALANCE; | |
6577 | /* Flags needing groups don't count if only 1 group in parent */ | |
6578 | if (parent->groups == parent->groups->next) { | |
6579 | pflags &= ~(SD_LOAD_BALANCE | | |
6580 | SD_BALANCE_NEWIDLE | | |
6581 | SD_BALANCE_FORK | | |
89c4710e SS |
6582 | SD_BALANCE_EXEC | |
6583 | SD_SHARE_CPUPOWER | | |
6584 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6585 | } |
6586 | if (~cflags & pflags) | |
6587 | return 0; | |
6588 | ||
6589 | return 1; | |
6590 | } | |
6591 | ||
57d885fe GH |
6592 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6593 | { | |
6594 | unsigned long flags; | |
57d885fe GH |
6595 | |
6596 | spin_lock_irqsave(&rq->lock, flags); | |
6597 | ||
6598 | if (rq->rd) { | |
6599 | struct root_domain *old_rd = rq->rd; | |
6600 | ||
1f11eb6a GH |
6601 | if (cpu_isset(rq->cpu, old_rd->online)) |
6602 | set_rq_offline(rq); | |
57d885fe | 6603 | |
dc938520 | 6604 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6605 | |
57d885fe GH |
6606 | if (atomic_dec_and_test(&old_rd->refcount)) |
6607 | kfree(old_rd); | |
6608 | } | |
6609 | ||
6610 | atomic_inc(&rd->refcount); | |
6611 | rq->rd = rd; | |
6612 | ||
dc938520 | 6613 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6614 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6615 | set_rq_online(rq); |
57d885fe GH |
6616 | |
6617 | spin_unlock_irqrestore(&rq->lock, flags); | |
6618 | } | |
6619 | ||
dc938520 | 6620 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6621 | { |
6622 | memset(rd, 0, sizeof(*rd)); | |
6623 | ||
dc938520 GH |
6624 | cpus_clear(rd->span); |
6625 | cpus_clear(rd->online); | |
6e0534f2 GH |
6626 | |
6627 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6628 | } |
6629 | ||
6630 | static void init_defrootdomain(void) | |
6631 | { | |
dc938520 | 6632 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6633 | atomic_set(&def_root_domain.refcount, 1); |
6634 | } | |
6635 | ||
dc938520 | 6636 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6637 | { |
6638 | struct root_domain *rd; | |
6639 | ||
6640 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6641 | if (!rd) | |
6642 | return NULL; | |
6643 | ||
dc938520 | 6644 | init_rootdomain(rd); |
57d885fe GH |
6645 | |
6646 | return rd; | |
6647 | } | |
6648 | ||
1da177e4 | 6649 | /* |
0eab9146 | 6650 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6651 | * hold the hotplug lock. |
6652 | */ | |
0eab9146 IM |
6653 | static void |
6654 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6655 | { |
70b97a7f | 6656 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6657 | struct sched_domain *tmp; |
6658 | ||
6659 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6660 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6661 | struct sched_domain *parent = tmp->parent; | |
6662 | if (!parent) | |
6663 | break; | |
1a848870 | 6664 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6665 | tmp->parent = parent->parent; |
1a848870 SS |
6666 | if (parent->parent) |
6667 | parent->parent->child = tmp; | |
6668 | } | |
245af2c7 SS |
6669 | } |
6670 | ||
1a848870 | 6671 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6672 | sd = sd->parent; |
1a848870 SS |
6673 | if (sd) |
6674 | sd->child = NULL; | |
6675 | } | |
1da177e4 LT |
6676 | |
6677 | sched_domain_debug(sd, cpu); | |
6678 | ||
57d885fe | 6679 | rq_attach_root(rq, rd); |
674311d5 | 6680 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6681 | } |
6682 | ||
6683 | /* cpus with isolated domains */ | |
67af63a6 | 6684 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6685 | |
6686 | /* Setup the mask of cpus configured for isolated domains */ | |
6687 | static int __init isolated_cpu_setup(char *str) | |
6688 | { | |
13b40c1e MT |
6689 | static int __initdata ints[NR_CPUS]; |
6690 | int i; | |
1da177e4 LT |
6691 | |
6692 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6693 | cpus_clear(cpu_isolated_map); | |
6694 | for (i = 1; i <= ints[0]; i++) | |
6695 | if (ints[i] < NR_CPUS) | |
6696 | cpu_set(ints[i], cpu_isolated_map); | |
6697 | return 1; | |
6698 | } | |
6699 | ||
8927f494 | 6700 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6701 | |
6702 | /* | |
6711cab4 SS |
6703 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6704 | * to a function which identifies what group(along with sched group) a CPU | |
6705 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6706 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6707 | * |
6708 | * init_sched_build_groups will build a circular linked list of the groups | |
6709 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6710 | * and ->cpu_power to 0. | |
6711 | */ | |
a616058b | 6712 | static void |
7c16ec58 | 6713 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6714 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6715 | struct sched_group **sg, |
6716 | cpumask_t *tmpmask), | |
6717 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6718 | { |
6719 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6720 | int i; |
6721 | ||
7c16ec58 MT |
6722 | cpus_clear(*covered); |
6723 | ||
363ab6f1 | 6724 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6725 | struct sched_group *sg; |
7c16ec58 | 6726 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6727 | int j; |
6728 | ||
7c16ec58 | 6729 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6730 | continue; |
6731 | ||
7c16ec58 | 6732 | cpus_clear(sg->cpumask); |
5517d86b | 6733 | sg->__cpu_power = 0; |
1da177e4 | 6734 | |
363ab6f1 | 6735 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6736 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6737 | continue; |
6738 | ||
7c16ec58 | 6739 | cpu_set(j, *covered); |
1da177e4 LT |
6740 | cpu_set(j, sg->cpumask); |
6741 | } | |
6742 | if (!first) | |
6743 | first = sg; | |
6744 | if (last) | |
6745 | last->next = sg; | |
6746 | last = sg; | |
6747 | } | |
6748 | last->next = first; | |
6749 | } | |
6750 | ||
9c1cfda2 | 6751 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6752 | |
9c1cfda2 | 6753 | #ifdef CONFIG_NUMA |
198e2f18 | 6754 | |
9c1cfda2 JH |
6755 | /** |
6756 | * find_next_best_node - find the next node to include in a sched_domain | |
6757 | * @node: node whose sched_domain we're building | |
6758 | * @used_nodes: nodes already in the sched_domain | |
6759 | * | |
41a2d6cf | 6760 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6761 | * finds the closest node not already in the @used_nodes map. |
6762 | * | |
6763 | * Should use nodemask_t. | |
6764 | */ | |
c5f59f08 | 6765 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6766 | { |
6767 | int i, n, val, min_val, best_node = 0; | |
6768 | ||
6769 | min_val = INT_MAX; | |
6770 | ||
076ac2af | 6771 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6772 | /* Start at @node */ |
076ac2af | 6773 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6774 | |
6775 | if (!nr_cpus_node(n)) | |
6776 | continue; | |
6777 | ||
6778 | /* Skip already used nodes */ | |
c5f59f08 | 6779 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6780 | continue; |
6781 | ||
6782 | /* Simple min distance search */ | |
6783 | val = node_distance(node, n); | |
6784 | ||
6785 | if (val < min_val) { | |
6786 | min_val = val; | |
6787 | best_node = n; | |
6788 | } | |
6789 | } | |
6790 | ||
c5f59f08 | 6791 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6792 | return best_node; |
6793 | } | |
6794 | ||
6795 | /** | |
6796 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6797 | * @node: node whose cpumask we're constructing | |
73486722 | 6798 | * @span: resulting cpumask |
9c1cfda2 | 6799 | * |
41a2d6cf | 6800 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6801 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6802 | * out optimally. | |
6803 | */ | |
4bdbaad3 | 6804 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6805 | { |
c5f59f08 | 6806 | nodemask_t used_nodes; |
c5f59f08 | 6807 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6808 | int i; |
9c1cfda2 | 6809 | |
4bdbaad3 | 6810 | cpus_clear(*span); |
c5f59f08 | 6811 | nodes_clear(used_nodes); |
9c1cfda2 | 6812 | |
4bdbaad3 | 6813 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6814 | node_set(node, used_nodes); |
9c1cfda2 JH |
6815 | |
6816 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6817 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6818 | |
c5f59f08 | 6819 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6820 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6821 | } |
9c1cfda2 | 6822 | } |
6d6bc0ad | 6823 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6824 | |
5c45bf27 | 6825 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6826 | |
9c1cfda2 | 6827 | /* |
48f24c4d | 6828 | * SMT sched-domains: |
9c1cfda2 | 6829 | */ |
1da177e4 LT |
6830 | #ifdef CONFIG_SCHED_SMT |
6831 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6832 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6833 | |
41a2d6cf | 6834 | static int |
7c16ec58 MT |
6835 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6836 | cpumask_t *unused) | |
1da177e4 | 6837 | { |
6711cab4 SS |
6838 | if (sg) |
6839 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6840 | return cpu; |
6841 | } | |
6d6bc0ad | 6842 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6843 | |
48f24c4d IM |
6844 | /* |
6845 | * multi-core sched-domains: | |
6846 | */ | |
1e9f28fa SS |
6847 | #ifdef CONFIG_SCHED_MC |
6848 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6849 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6850 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6851 | |
6852 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6853 | static int |
7c16ec58 MT |
6854 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6855 | cpumask_t *mask) | |
1e9f28fa | 6856 | { |
6711cab4 | 6857 | int group; |
7c16ec58 MT |
6858 | |
6859 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6860 | cpus_and(*mask, *mask, *cpu_map); | |
6861 | group = first_cpu(*mask); | |
6711cab4 SS |
6862 | if (sg) |
6863 | *sg = &per_cpu(sched_group_core, group); | |
6864 | return group; | |
1e9f28fa SS |
6865 | } |
6866 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6867 | static int |
7c16ec58 MT |
6868 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6869 | cpumask_t *unused) | |
1e9f28fa | 6870 | { |
6711cab4 SS |
6871 | if (sg) |
6872 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6873 | return cpu; |
6874 | } | |
6875 | #endif | |
6876 | ||
1da177e4 | 6877 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6878 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6879 | |
41a2d6cf | 6880 | static int |
7c16ec58 MT |
6881 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6882 | cpumask_t *mask) | |
1da177e4 | 6883 | { |
6711cab4 | 6884 | int group; |
48f24c4d | 6885 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6886 | *mask = cpu_coregroup_map(cpu); |
6887 | cpus_and(*mask, *mask, *cpu_map); | |
6888 | group = first_cpu(*mask); | |
1e9f28fa | 6889 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6890 | *mask = per_cpu(cpu_sibling_map, cpu); |
6891 | cpus_and(*mask, *mask, *cpu_map); | |
6892 | group = first_cpu(*mask); | |
1da177e4 | 6893 | #else |
6711cab4 | 6894 | group = cpu; |
1da177e4 | 6895 | #endif |
6711cab4 SS |
6896 | if (sg) |
6897 | *sg = &per_cpu(sched_group_phys, group); | |
6898 | return group; | |
1da177e4 LT |
6899 | } |
6900 | ||
6901 | #ifdef CONFIG_NUMA | |
1da177e4 | 6902 | /* |
9c1cfda2 JH |
6903 | * The init_sched_build_groups can't handle what we want to do with node |
6904 | * groups, so roll our own. Now each node has its own list of groups which | |
6905 | * gets dynamically allocated. | |
1da177e4 | 6906 | */ |
9c1cfda2 | 6907 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6908 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6909 | |
9c1cfda2 | 6910 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6911 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6912 | |
6711cab4 | 6913 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6914 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6915 | { |
6711cab4 SS |
6916 | int group; |
6917 | ||
7c16ec58 MT |
6918 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6919 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6920 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6921 | |
6922 | if (sg) | |
6923 | *sg = &per_cpu(sched_group_allnodes, group); | |
6924 | return group; | |
1da177e4 | 6925 | } |
6711cab4 | 6926 | |
08069033 SS |
6927 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6928 | { | |
6929 | struct sched_group *sg = group_head; | |
6930 | int j; | |
6931 | ||
6932 | if (!sg) | |
6933 | return; | |
3a5c359a | 6934 | do { |
363ab6f1 | 6935 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 6936 | struct sched_domain *sd; |
08069033 | 6937 | |
3a5c359a AK |
6938 | sd = &per_cpu(phys_domains, j); |
6939 | if (j != first_cpu(sd->groups->cpumask)) { | |
6940 | /* | |
6941 | * Only add "power" once for each | |
6942 | * physical package. | |
6943 | */ | |
6944 | continue; | |
6945 | } | |
08069033 | 6946 | |
3a5c359a AK |
6947 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6948 | } | |
6949 | sg = sg->next; | |
6950 | } while (sg != group_head); | |
08069033 | 6951 | } |
6d6bc0ad | 6952 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6953 | |
a616058b | 6954 | #ifdef CONFIG_NUMA |
51888ca2 | 6955 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6956 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6957 | { |
a616058b | 6958 | int cpu, i; |
51888ca2 | 6959 | |
363ab6f1 | 6960 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
6961 | struct sched_group **sched_group_nodes |
6962 | = sched_group_nodes_bycpu[cpu]; | |
6963 | ||
51888ca2 SV |
6964 | if (!sched_group_nodes) |
6965 | continue; | |
6966 | ||
076ac2af | 6967 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6968 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6969 | ||
7c16ec58 MT |
6970 | *nodemask = node_to_cpumask(i); |
6971 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6972 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6973 | continue; |
6974 | ||
6975 | if (sg == NULL) | |
6976 | continue; | |
6977 | sg = sg->next; | |
6978 | next_sg: | |
6979 | oldsg = sg; | |
6980 | sg = sg->next; | |
6981 | kfree(oldsg); | |
6982 | if (oldsg != sched_group_nodes[i]) | |
6983 | goto next_sg; | |
6984 | } | |
6985 | kfree(sched_group_nodes); | |
6986 | sched_group_nodes_bycpu[cpu] = NULL; | |
6987 | } | |
51888ca2 | 6988 | } |
6d6bc0ad | 6989 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 6990 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6991 | { |
6992 | } | |
6d6bc0ad | 6993 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6994 | |
89c4710e SS |
6995 | /* |
6996 | * Initialize sched groups cpu_power. | |
6997 | * | |
6998 | * cpu_power indicates the capacity of sched group, which is used while | |
6999 | * distributing the load between different sched groups in a sched domain. | |
7000 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7001 | * there are asymmetries in the topology. If there are asymmetries, group | |
7002 | * having more cpu_power will pickup more load compared to the group having | |
7003 | * less cpu_power. | |
7004 | * | |
7005 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7006 | * the maximum number of tasks a group can handle in the presence of other idle | |
7007 | * or lightly loaded groups in the same sched domain. | |
7008 | */ | |
7009 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7010 | { | |
7011 | struct sched_domain *child; | |
7012 | struct sched_group *group; | |
7013 | ||
7014 | WARN_ON(!sd || !sd->groups); | |
7015 | ||
7016 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7017 | return; | |
7018 | ||
7019 | child = sd->child; | |
7020 | ||
5517d86b ED |
7021 | sd->groups->__cpu_power = 0; |
7022 | ||
89c4710e SS |
7023 | /* |
7024 | * For perf policy, if the groups in child domain share resources | |
7025 | * (for example cores sharing some portions of the cache hierarchy | |
7026 | * or SMT), then set this domain groups cpu_power such that each group | |
7027 | * can handle only one task, when there are other idle groups in the | |
7028 | * same sched domain. | |
7029 | */ | |
7030 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7031 | (child->flags & | |
7032 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7033 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7034 | return; |
7035 | } | |
7036 | ||
89c4710e SS |
7037 | /* |
7038 | * add cpu_power of each child group to this groups cpu_power | |
7039 | */ | |
7040 | group = child->groups; | |
7041 | do { | |
5517d86b | 7042 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7043 | group = group->next; |
7044 | } while (group != child->groups); | |
7045 | } | |
7046 | ||
7c16ec58 MT |
7047 | /* |
7048 | * Initializers for schedule domains | |
7049 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7050 | */ | |
7051 | ||
7052 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7053 | #define SD_INIT_FUNC(type) \ | |
7054 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7055 | { \ | |
7056 | memset(sd, 0, sizeof(*sd)); \ | |
7057 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7058 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7059 | } |
7060 | ||
7061 | SD_INIT_FUNC(CPU) | |
7062 | #ifdef CONFIG_NUMA | |
7063 | SD_INIT_FUNC(ALLNODES) | |
7064 | SD_INIT_FUNC(NODE) | |
7065 | #endif | |
7066 | #ifdef CONFIG_SCHED_SMT | |
7067 | SD_INIT_FUNC(SIBLING) | |
7068 | #endif | |
7069 | #ifdef CONFIG_SCHED_MC | |
7070 | SD_INIT_FUNC(MC) | |
7071 | #endif | |
7072 | ||
7073 | /* | |
7074 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7075 | * space as the usage in build_sched_domains() dictates. Used only | |
7076 | * if the amount of space is significant. | |
7077 | */ | |
7078 | struct allmasks { | |
7079 | cpumask_t tmpmask; /* make this one first */ | |
7080 | union { | |
7081 | cpumask_t nodemask; | |
7082 | cpumask_t this_sibling_map; | |
7083 | cpumask_t this_core_map; | |
7084 | }; | |
7085 | cpumask_t send_covered; | |
7086 | ||
7087 | #ifdef CONFIG_NUMA | |
7088 | cpumask_t domainspan; | |
7089 | cpumask_t covered; | |
7090 | cpumask_t notcovered; | |
7091 | #endif | |
7092 | }; | |
7093 | ||
7094 | #if NR_CPUS > 128 | |
7095 | #define SCHED_CPUMASK_ALLOC 1 | |
7096 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7097 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7098 | #else | |
7099 | #define SCHED_CPUMASK_ALLOC 0 | |
7100 | #define SCHED_CPUMASK_FREE(v) | |
7101 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7102 | #endif | |
7103 | ||
7104 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7105 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7106 | ||
1d3504fc HS |
7107 | static int default_relax_domain_level = -1; |
7108 | ||
7109 | static int __init setup_relax_domain_level(char *str) | |
7110 | { | |
30e0e178 LZ |
7111 | unsigned long val; |
7112 | ||
7113 | val = simple_strtoul(str, NULL, 0); | |
7114 | if (val < SD_LV_MAX) | |
7115 | default_relax_domain_level = val; | |
7116 | ||
1d3504fc HS |
7117 | return 1; |
7118 | } | |
7119 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7120 | ||
7121 | static void set_domain_attribute(struct sched_domain *sd, | |
7122 | struct sched_domain_attr *attr) | |
7123 | { | |
7124 | int request; | |
7125 | ||
7126 | if (!attr || attr->relax_domain_level < 0) { | |
7127 | if (default_relax_domain_level < 0) | |
7128 | return; | |
7129 | else | |
7130 | request = default_relax_domain_level; | |
7131 | } else | |
7132 | request = attr->relax_domain_level; | |
7133 | if (request < sd->level) { | |
7134 | /* turn off idle balance on this domain */ | |
7135 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7136 | } else { | |
7137 | /* turn on idle balance on this domain */ | |
7138 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7139 | } | |
7140 | } | |
7141 | ||
1da177e4 | 7142 | /* |
1a20ff27 DG |
7143 | * Build sched domains for a given set of cpus and attach the sched domains |
7144 | * to the individual cpus | |
1da177e4 | 7145 | */ |
1d3504fc HS |
7146 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7147 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7148 | { |
7149 | int i; | |
57d885fe | 7150 | struct root_domain *rd; |
7c16ec58 MT |
7151 | SCHED_CPUMASK_DECLARE(allmasks); |
7152 | cpumask_t *tmpmask; | |
d1b55138 JH |
7153 | #ifdef CONFIG_NUMA |
7154 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7155 | int sd_allnodes = 0; |
d1b55138 JH |
7156 | |
7157 | /* | |
7158 | * Allocate the per-node list of sched groups | |
7159 | */ | |
076ac2af | 7160 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7161 | GFP_KERNEL); |
d1b55138 JH |
7162 | if (!sched_group_nodes) { |
7163 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7164 | return -ENOMEM; |
d1b55138 | 7165 | } |
d1b55138 | 7166 | #endif |
1da177e4 | 7167 | |
dc938520 | 7168 | rd = alloc_rootdomain(); |
57d885fe GH |
7169 | if (!rd) { |
7170 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7171 | #ifdef CONFIG_NUMA |
7172 | kfree(sched_group_nodes); | |
7173 | #endif | |
57d885fe GH |
7174 | return -ENOMEM; |
7175 | } | |
7176 | ||
7c16ec58 MT |
7177 | #if SCHED_CPUMASK_ALLOC |
7178 | /* get space for all scratch cpumask variables */ | |
7179 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7180 | if (!allmasks) { | |
7181 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7182 | kfree(rd); | |
7183 | #ifdef CONFIG_NUMA | |
7184 | kfree(sched_group_nodes); | |
7185 | #endif | |
7186 | return -ENOMEM; | |
7187 | } | |
7188 | #endif | |
7189 | tmpmask = (cpumask_t *)allmasks; | |
7190 | ||
7191 | ||
7192 | #ifdef CONFIG_NUMA | |
7193 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7194 | #endif | |
7195 | ||
1da177e4 | 7196 | /* |
1a20ff27 | 7197 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7198 | */ |
363ab6f1 | 7199 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7200 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7201 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7202 | |
7c16ec58 MT |
7203 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7204 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7205 | |
7206 | #ifdef CONFIG_NUMA | |
dd41f596 | 7207 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7208 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7209 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7210 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7211 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7212 | sd->span = *cpu_map; |
7c16ec58 | 7213 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7214 | p = sd; |
6711cab4 | 7215 | sd_allnodes = 1; |
9c1cfda2 JH |
7216 | } else |
7217 | p = NULL; | |
7218 | ||
1da177e4 | 7219 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7220 | SD_INIT(sd, NODE); |
1d3504fc | 7221 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7222 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7223 | sd->parent = p; |
1a848870 SS |
7224 | if (p) |
7225 | p->child = sd; | |
9c1cfda2 | 7226 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7227 | #endif |
7228 | ||
7229 | p = sd; | |
7230 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7231 | SD_INIT(sd, CPU); |
1d3504fc | 7232 | set_domain_attribute(sd, attr); |
7c16ec58 | 7233 | sd->span = *nodemask; |
1da177e4 | 7234 | sd->parent = p; |
1a848870 SS |
7235 | if (p) |
7236 | p->child = sd; | |
7c16ec58 | 7237 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7238 | |
1e9f28fa SS |
7239 | #ifdef CONFIG_SCHED_MC |
7240 | p = sd; | |
7241 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7242 | SD_INIT(sd, MC); |
1d3504fc | 7243 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7244 | sd->span = cpu_coregroup_map(i); |
7245 | cpus_and(sd->span, sd->span, *cpu_map); | |
7246 | sd->parent = p; | |
1a848870 | 7247 | p->child = sd; |
7c16ec58 | 7248 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7249 | #endif |
7250 | ||
1da177e4 LT |
7251 | #ifdef CONFIG_SCHED_SMT |
7252 | p = sd; | |
7253 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7254 | SD_INIT(sd, SIBLING); |
1d3504fc | 7255 | set_domain_attribute(sd, attr); |
d5a7430d | 7256 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7257 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7258 | sd->parent = p; |
1a848870 | 7259 | p->child = sd; |
7c16ec58 | 7260 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7261 | #endif |
7262 | } | |
7263 | ||
7264 | #ifdef CONFIG_SCHED_SMT | |
7265 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7266 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7267 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7268 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7269 | ||
7270 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7271 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7272 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7273 | continue; |
7274 | ||
dd41f596 | 7275 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7276 | &cpu_to_cpu_group, |
7277 | send_covered, tmpmask); | |
1da177e4 LT |
7278 | } |
7279 | #endif | |
7280 | ||
1e9f28fa SS |
7281 | #ifdef CONFIG_SCHED_MC |
7282 | /* Set up multi-core groups */ | |
363ab6f1 | 7283 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7284 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7285 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7286 | ||
7287 | *this_core_map = cpu_coregroup_map(i); | |
7288 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7289 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7290 | continue; |
7c16ec58 | 7291 | |
dd41f596 | 7292 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7293 | &cpu_to_core_group, |
7294 | send_covered, tmpmask); | |
1e9f28fa SS |
7295 | } |
7296 | #endif | |
7297 | ||
1da177e4 | 7298 | /* Set up physical groups */ |
076ac2af | 7299 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7300 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7301 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7302 | |
7c16ec58 MT |
7303 | *nodemask = node_to_cpumask(i); |
7304 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7305 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7306 | continue; |
7307 | ||
7c16ec58 MT |
7308 | init_sched_build_groups(nodemask, cpu_map, |
7309 | &cpu_to_phys_group, | |
7310 | send_covered, tmpmask); | |
1da177e4 LT |
7311 | } |
7312 | ||
7313 | #ifdef CONFIG_NUMA | |
7314 | /* Set up node groups */ | |
7c16ec58 MT |
7315 | if (sd_allnodes) { |
7316 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7317 | ||
7318 | init_sched_build_groups(cpu_map, cpu_map, | |
7319 | &cpu_to_allnodes_group, | |
7320 | send_covered, tmpmask); | |
7321 | } | |
9c1cfda2 | 7322 | |
076ac2af | 7323 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7324 | /* Set up node groups */ |
7325 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7326 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7327 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7328 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7329 | int j; |
7330 | ||
7c16ec58 MT |
7331 | *nodemask = node_to_cpumask(i); |
7332 | cpus_clear(*covered); | |
7333 | ||
7334 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7335 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7336 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7337 | continue; |
d1b55138 | 7338 | } |
9c1cfda2 | 7339 | |
4bdbaad3 | 7340 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7341 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7342 | |
15f0b676 | 7343 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7344 | if (!sg) { |
7345 | printk(KERN_WARNING "Can not alloc domain group for " | |
7346 | "node %d\n", i); | |
7347 | goto error; | |
7348 | } | |
9c1cfda2 | 7349 | sched_group_nodes[i] = sg; |
363ab6f1 | 7350 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7351 | struct sched_domain *sd; |
9761eea8 | 7352 | |
9c1cfda2 JH |
7353 | sd = &per_cpu(node_domains, j); |
7354 | sd->groups = sg; | |
9c1cfda2 | 7355 | } |
5517d86b | 7356 | sg->__cpu_power = 0; |
7c16ec58 | 7357 | sg->cpumask = *nodemask; |
51888ca2 | 7358 | sg->next = sg; |
7c16ec58 | 7359 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7360 | prev = sg; |
7361 | ||
076ac2af | 7362 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7363 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7364 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7365 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7366 | |
7c16ec58 MT |
7367 | cpus_complement(*notcovered, *covered); |
7368 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7369 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7370 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7371 | break; |
7372 | ||
7c16ec58 MT |
7373 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7374 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7375 | continue; |
7376 | ||
15f0b676 SV |
7377 | sg = kmalloc_node(sizeof(struct sched_group), |
7378 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7379 | if (!sg) { |
7380 | printk(KERN_WARNING | |
7381 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7382 | goto error; |
9c1cfda2 | 7383 | } |
5517d86b | 7384 | sg->__cpu_power = 0; |
7c16ec58 | 7385 | sg->cpumask = *tmpmask; |
51888ca2 | 7386 | sg->next = prev->next; |
7c16ec58 | 7387 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7388 | prev->next = sg; |
7389 | prev = sg; | |
7390 | } | |
9c1cfda2 | 7391 | } |
1da177e4 LT |
7392 | #endif |
7393 | ||
7394 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7395 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7396 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7397 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7398 | ||
89c4710e | 7399 | init_sched_groups_power(i, sd); |
5c45bf27 | 7400 | } |
1da177e4 | 7401 | #endif |
1e9f28fa | 7402 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7403 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7404 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7405 | ||
89c4710e | 7406 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7407 | } |
7408 | #endif | |
1e9f28fa | 7409 | |
363ab6f1 | 7410 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7411 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7412 | ||
89c4710e | 7413 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7414 | } |
7415 | ||
9c1cfda2 | 7416 | #ifdef CONFIG_NUMA |
076ac2af | 7417 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7418 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7419 | |
6711cab4 SS |
7420 | if (sd_allnodes) { |
7421 | struct sched_group *sg; | |
f712c0c7 | 7422 | |
7c16ec58 MT |
7423 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7424 | tmpmask); | |
f712c0c7 SS |
7425 | init_numa_sched_groups_power(sg); |
7426 | } | |
9c1cfda2 JH |
7427 | #endif |
7428 | ||
1da177e4 | 7429 | /* Attach the domains */ |
363ab6f1 | 7430 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7431 | struct sched_domain *sd; |
7432 | #ifdef CONFIG_SCHED_SMT | |
7433 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7434 | #elif defined(CONFIG_SCHED_MC) |
7435 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7436 | #else |
7437 | sd = &per_cpu(phys_domains, i); | |
7438 | #endif | |
57d885fe | 7439 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7440 | } |
51888ca2 | 7441 | |
7c16ec58 | 7442 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7443 | return 0; |
7444 | ||
a616058b | 7445 | #ifdef CONFIG_NUMA |
51888ca2 | 7446 | error: |
7c16ec58 MT |
7447 | free_sched_groups(cpu_map, tmpmask); |
7448 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7449 | return -ENOMEM; |
a616058b | 7450 | #endif |
1da177e4 | 7451 | } |
029190c5 | 7452 | |
1d3504fc HS |
7453 | static int build_sched_domains(const cpumask_t *cpu_map) |
7454 | { | |
7455 | return __build_sched_domains(cpu_map, NULL); | |
7456 | } | |
7457 | ||
029190c5 PJ |
7458 | static cpumask_t *doms_cur; /* current sched domains */ |
7459 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7460 | static struct sched_domain_attr *dattr_cur; |
7461 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7462 | |
7463 | /* | |
7464 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7465 | * cpumask_t) fails, then fallback to a single sched domain, | |
7466 | * as determined by the single cpumask_t fallback_doms. | |
7467 | */ | |
7468 | static cpumask_t fallback_doms; | |
7469 | ||
22e52b07 HC |
7470 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7471 | { | |
7472 | } | |
7473 | ||
1a20ff27 | 7474 | /* |
41a2d6cf | 7475 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7476 | * For now this just excludes isolated cpus, but could be used to |
7477 | * exclude other special cases in the future. | |
1a20ff27 | 7478 | */ |
51888ca2 | 7479 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7480 | { |
7378547f MM |
7481 | int err; |
7482 | ||
22e52b07 | 7483 | arch_update_cpu_topology(); |
029190c5 PJ |
7484 | ndoms_cur = 1; |
7485 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7486 | if (!doms_cur) | |
7487 | doms_cur = &fallback_doms; | |
7488 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7489 | dattr_cur = NULL; |
7378547f | 7490 | err = build_sched_domains(doms_cur); |
6382bc90 | 7491 | register_sched_domain_sysctl(); |
7378547f MM |
7492 | |
7493 | return err; | |
1a20ff27 DG |
7494 | } |
7495 | ||
7c16ec58 MT |
7496 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7497 | cpumask_t *tmpmask) | |
1da177e4 | 7498 | { |
7c16ec58 | 7499 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7500 | } |
1da177e4 | 7501 | |
1a20ff27 DG |
7502 | /* |
7503 | * Detach sched domains from a group of cpus specified in cpu_map | |
7504 | * These cpus will now be attached to the NULL domain | |
7505 | */ | |
858119e1 | 7506 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7507 | { |
7c16ec58 | 7508 | cpumask_t tmpmask; |
1a20ff27 DG |
7509 | int i; |
7510 | ||
6382bc90 MM |
7511 | unregister_sched_domain_sysctl(); |
7512 | ||
363ab6f1 | 7513 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7514 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7515 | synchronize_sched(); |
7c16ec58 | 7516 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7517 | } |
7518 | ||
1d3504fc HS |
7519 | /* handle null as "default" */ |
7520 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7521 | struct sched_domain_attr *new, int idx_new) | |
7522 | { | |
7523 | struct sched_domain_attr tmp; | |
7524 | ||
7525 | /* fast path */ | |
7526 | if (!new && !cur) | |
7527 | return 1; | |
7528 | ||
7529 | tmp = SD_ATTR_INIT; | |
7530 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7531 | new ? (new + idx_new) : &tmp, | |
7532 | sizeof(struct sched_domain_attr)); | |
7533 | } | |
7534 | ||
029190c5 PJ |
7535 | /* |
7536 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7537 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7538 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7539 | * It destroys each deleted domain and builds each new domain. | |
7540 | * | |
7541 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7542 | * The masks don't intersect (don't overlap.) We should setup one |
7543 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7544 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7545 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7546 | * it as it is. | |
7547 | * | |
41a2d6cf IM |
7548 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7549 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7550 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7551 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7552 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 PJ |
7553 | * |
7554 | * Call with hotplug lock held | |
7555 | */ | |
1d3504fc HS |
7556 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7557 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7558 | { |
7559 | int i, j; | |
7560 | ||
712555ee | 7561 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7562 | |
7378547f MM |
7563 | /* always unregister in case we don't destroy any domains */ |
7564 | unregister_sched_domain_sysctl(); | |
7565 | ||
e761b772 MK |
7566 | if (doms_new == NULL) |
7567 | ndoms_new = 0; | |
029190c5 PJ |
7568 | |
7569 | /* Destroy deleted domains */ | |
7570 | for (i = 0; i < ndoms_cur; i++) { | |
7571 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7572 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7573 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7574 | goto match1; |
7575 | } | |
7576 | /* no match - a current sched domain not in new doms_new[] */ | |
7577 | detach_destroy_domains(doms_cur + i); | |
7578 | match1: | |
7579 | ; | |
7580 | } | |
7581 | ||
e761b772 MK |
7582 | if (doms_new == NULL) { |
7583 | ndoms_cur = 0; | |
7584 | ndoms_new = 1; | |
7585 | doms_new = &fallback_doms; | |
7586 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7587 | dattr_new = NULL; | |
7588 | } | |
7589 | ||
029190c5 PJ |
7590 | /* Build new domains */ |
7591 | for (i = 0; i < ndoms_new; i++) { | |
7592 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7593 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7594 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7595 | goto match2; |
7596 | } | |
7597 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7598 | __build_sched_domains(doms_new + i, |
7599 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7600 | match2: |
7601 | ; | |
7602 | } | |
7603 | ||
7604 | /* Remember the new sched domains */ | |
7605 | if (doms_cur != &fallback_doms) | |
7606 | kfree(doms_cur); | |
1d3504fc | 7607 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7608 | doms_cur = doms_new; |
1d3504fc | 7609 | dattr_cur = dattr_new; |
029190c5 | 7610 | ndoms_cur = ndoms_new; |
7378547f MM |
7611 | |
7612 | register_sched_domain_sysctl(); | |
a1835615 | 7613 | |
712555ee | 7614 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7615 | } |
7616 | ||
5c45bf27 | 7617 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7618 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7619 | { |
95402b38 | 7620 | get_online_cpus(); |
e761b772 | 7621 | rebuild_sched_domains(); |
95402b38 | 7622 | put_online_cpus(); |
e761b772 | 7623 | return 0; |
5c45bf27 SS |
7624 | } |
7625 | ||
7626 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7627 | { | |
7628 | int ret; | |
7629 | ||
7630 | if (buf[0] != '0' && buf[0] != '1') | |
7631 | return -EINVAL; | |
7632 | ||
7633 | if (smt) | |
7634 | sched_smt_power_savings = (buf[0] == '1'); | |
7635 | else | |
7636 | sched_mc_power_savings = (buf[0] == '1'); | |
7637 | ||
7638 | ret = arch_reinit_sched_domains(); | |
7639 | ||
7640 | return ret ? ret : count; | |
7641 | } | |
7642 | ||
5c45bf27 | 7643 | #ifdef CONFIG_SCHED_MC |
4a0b2b4d AK |
7644 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, |
7645 | struct sysdev_attribute *attr, char *page) | |
5c45bf27 SS |
7646 | { |
7647 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7648 | } | |
48f24c4d | 7649 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
4a0b2b4d | 7650 | struct sysdev_attribute *attr, |
48f24c4d | 7651 | const char *buf, size_t count) |
5c45bf27 SS |
7652 | { |
7653 | return sched_power_savings_store(buf, count, 0); | |
7654 | } | |
6707de00 AB |
7655 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7656 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7657 | #endif |
7658 | ||
7659 | #ifdef CONFIG_SCHED_SMT | |
4a0b2b4d AK |
7660 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, |
7661 | struct sysdev_attribute *attr, char *page) | |
5c45bf27 SS |
7662 | { |
7663 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7664 | } | |
48f24c4d | 7665 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
4a0b2b4d | 7666 | struct sysdev_attribute *attr, |
48f24c4d | 7667 | const char *buf, size_t count) |
5c45bf27 SS |
7668 | { |
7669 | return sched_power_savings_store(buf, count, 1); | |
7670 | } | |
6707de00 AB |
7671 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7672 | sched_smt_power_savings_store); | |
7673 | #endif | |
7674 | ||
7675 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7676 | { | |
7677 | int err = 0; | |
7678 | ||
7679 | #ifdef CONFIG_SCHED_SMT | |
7680 | if (smt_capable()) | |
7681 | err = sysfs_create_file(&cls->kset.kobj, | |
7682 | &attr_sched_smt_power_savings.attr); | |
7683 | #endif | |
7684 | #ifdef CONFIG_SCHED_MC | |
7685 | if (!err && mc_capable()) | |
7686 | err = sysfs_create_file(&cls->kset.kobj, | |
7687 | &attr_sched_mc_power_savings.attr); | |
7688 | #endif | |
7689 | return err; | |
7690 | } | |
6d6bc0ad | 7691 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7692 | |
e761b772 | 7693 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7694 | /* |
e761b772 MK |
7695 | * Add online and remove offline CPUs from the scheduler domains. |
7696 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7697 | */ |
7698 | static int update_sched_domains(struct notifier_block *nfb, | |
7699 | unsigned long action, void *hcpu) | |
e761b772 MK |
7700 | { |
7701 | switch (action) { | |
7702 | case CPU_ONLINE: | |
7703 | case CPU_ONLINE_FROZEN: | |
7704 | case CPU_DEAD: | |
7705 | case CPU_DEAD_FROZEN: | |
7706 | partition_sched_domains(0, NULL, NULL); | |
7707 | return NOTIFY_OK; | |
7708 | ||
7709 | default: | |
7710 | return NOTIFY_DONE; | |
7711 | } | |
7712 | } | |
7713 | #endif | |
7714 | ||
7715 | static int update_runtime(struct notifier_block *nfb, | |
7716 | unsigned long action, void *hcpu) | |
1da177e4 | 7717 | { |
7def2be1 PZ |
7718 | int cpu = (int)(long)hcpu; |
7719 | ||
1da177e4 | 7720 | switch (action) { |
1da177e4 | 7721 | case CPU_DOWN_PREPARE: |
8bb78442 | 7722 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7723 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7724 | return NOTIFY_OK; |
7725 | ||
1da177e4 | 7726 | case CPU_DOWN_FAILED: |
8bb78442 | 7727 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7728 | case CPU_ONLINE: |
8bb78442 | 7729 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7730 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7731 | return NOTIFY_OK; |
7732 | ||
1da177e4 LT |
7733 | default: |
7734 | return NOTIFY_DONE; | |
7735 | } | |
1da177e4 | 7736 | } |
1da177e4 LT |
7737 | |
7738 | void __init sched_init_smp(void) | |
7739 | { | |
5c1e1767 NP |
7740 | cpumask_t non_isolated_cpus; |
7741 | ||
434d53b0 MT |
7742 | #if defined(CONFIG_NUMA) |
7743 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7744 | GFP_KERNEL); | |
7745 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7746 | #endif | |
95402b38 | 7747 | get_online_cpus(); |
712555ee | 7748 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7749 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7750 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7751 | if (cpus_empty(non_isolated_cpus)) |
7752 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7753 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7754 | put_online_cpus(); |
e761b772 MK |
7755 | |
7756 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7757 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7758 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7759 | #endif |
7760 | ||
7761 | /* RT runtime code needs to handle some hotplug events */ | |
7762 | hotcpu_notifier(update_runtime, 0); | |
7763 | ||
b328ca18 | 7764 | init_hrtick(); |
5c1e1767 NP |
7765 | |
7766 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7767 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7768 | BUG(); |
19978ca6 | 7769 | sched_init_granularity(); |
1da177e4 LT |
7770 | } |
7771 | #else | |
7772 | void __init sched_init_smp(void) | |
7773 | { | |
19978ca6 | 7774 | sched_init_granularity(); |
1da177e4 LT |
7775 | } |
7776 | #endif /* CONFIG_SMP */ | |
7777 | ||
7778 | int in_sched_functions(unsigned long addr) | |
7779 | { | |
1da177e4 LT |
7780 | return in_lock_functions(addr) || |
7781 | (addr >= (unsigned long)__sched_text_start | |
7782 | && addr < (unsigned long)__sched_text_end); | |
7783 | } | |
7784 | ||
a9957449 | 7785 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7786 | { |
7787 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7788 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7789 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7790 | cfs_rq->rq = rq; | |
7791 | #endif | |
67e9fb2a | 7792 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7793 | } |
7794 | ||
fa85ae24 PZ |
7795 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7796 | { | |
7797 | struct rt_prio_array *array; | |
7798 | int i; | |
7799 | ||
7800 | array = &rt_rq->active; | |
7801 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7802 | INIT_LIST_HEAD(array->queue + i); | |
7803 | __clear_bit(i, array->bitmap); | |
7804 | } | |
7805 | /* delimiter for bitsearch: */ | |
7806 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7807 | ||
052f1dc7 | 7808 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7809 | rt_rq->highest_prio = MAX_RT_PRIO; |
7810 | #endif | |
fa85ae24 PZ |
7811 | #ifdef CONFIG_SMP |
7812 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7813 | rt_rq->overloaded = 0; |
7814 | #endif | |
7815 | ||
7816 | rt_rq->rt_time = 0; | |
7817 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7818 | rt_rq->rt_runtime = 0; |
7819 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7820 | |
052f1dc7 | 7821 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7822 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7823 | rt_rq->rq = rq; |
7824 | #endif | |
fa85ae24 PZ |
7825 | } |
7826 | ||
6f505b16 | 7827 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7828 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7829 | struct sched_entity *se, int cpu, int add, | |
7830 | struct sched_entity *parent) | |
6f505b16 | 7831 | { |
ec7dc8ac | 7832 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7833 | tg->cfs_rq[cpu] = cfs_rq; |
7834 | init_cfs_rq(cfs_rq, rq); | |
7835 | cfs_rq->tg = tg; | |
7836 | if (add) | |
7837 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7838 | ||
7839 | tg->se[cpu] = se; | |
354d60c2 DG |
7840 | /* se could be NULL for init_task_group */ |
7841 | if (!se) | |
7842 | return; | |
7843 | ||
ec7dc8ac DG |
7844 | if (!parent) |
7845 | se->cfs_rq = &rq->cfs; | |
7846 | else | |
7847 | se->cfs_rq = parent->my_q; | |
7848 | ||
6f505b16 PZ |
7849 | se->my_q = cfs_rq; |
7850 | se->load.weight = tg->shares; | |
e05510d0 | 7851 | se->load.inv_weight = 0; |
ec7dc8ac | 7852 | se->parent = parent; |
6f505b16 | 7853 | } |
052f1dc7 | 7854 | #endif |
6f505b16 | 7855 | |
052f1dc7 | 7856 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7857 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7858 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7859 | struct sched_rt_entity *parent) | |
6f505b16 | 7860 | { |
ec7dc8ac DG |
7861 | struct rq *rq = cpu_rq(cpu); |
7862 | ||
6f505b16 PZ |
7863 | tg->rt_rq[cpu] = rt_rq; |
7864 | init_rt_rq(rt_rq, rq); | |
7865 | rt_rq->tg = tg; | |
7866 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7867 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7868 | if (add) |
7869 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7870 | ||
7871 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7872 | if (!rt_se) |
7873 | return; | |
7874 | ||
ec7dc8ac DG |
7875 | if (!parent) |
7876 | rt_se->rt_rq = &rq->rt; | |
7877 | else | |
7878 | rt_se->rt_rq = parent->my_q; | |
7879 | ||
6f505b16 | 7880 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7881 | rt_se->parent = parent; |
6f505b16 PZ |
7882 | INIT_LIST_HEAD(&rt_se->run_list); |
7883 | } | |
7884 | #endif | |
7885 | ||
1da177e4 LT |
7886 | void __init sched_init(void) |
7887 | { | |
dd41f596 | 7888 | int i, j; |
434d53b0 MT |
7889 | unsigned long alloc_size = 0, ptr; |
7890 | ||
7891 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7892 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7893 | #endif | |
7894 | #ifdef CONFIG_RT_GROUP_SCHED | |
7895 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7896 | #endif |
7897 | #ifdef CONFIG_USER_SCHED | |
7898 | alloc_size *= 2; | |
434d53b0 MT |
7899 | #endif |
7900 | /* | |
7901 | * As sched_init() is called before page_alloc is setup, | |
7902 | * we use alloc_bootmem(). | |
7903 | */ | |
7904 | if (alloc_size) { | |
5a9d3225 | 7905 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7906 | |
7907 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7908 | init_task_group.se = (struct sched_entity **)ptr; | |
7909 | ptr += nr_cpu_ids * sizeof(void **); | |
7910 | ||
7911 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7912 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7913 | |
7914 | #ifdef CONFIG_USER_SCHED | |
7915 | root_task_group.se = (struct sched_entity **)ptr; | |
7916 | ptr += nr_cpu_ids * sizeof(void **); | |
7917 | ||
7918 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7919 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7920 | #endif /* CONFIG_USER_SCHED */ |
7921 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7922 | #ifdef CONFIG_RT_GROUP_SCHED |
7923 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7924 | ptr += nr_cpu_ids * sizeof(void **); | |
7925 | ||
7926 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7927 | ptr += nr_cpu_ids * sizeof(void **); |
7928 | ||
7929 | #ifdef CONFIG_USER_SCHED | |
7930 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7931 | ptr += nr_cpu_ids * sizeof(void **); | |
7932 | ||
7933 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7934 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7935 | #endif /* CONFIG_USER_SCHED */ |
7936 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7937 | } |
dd41f596 | 7938 | |
57d885fe GH |
7939 | #ifdef CONFIG_SMP |
7940 | init_defrootdomain(); | |
7941 | #endif | |
7942 | ||
d0b27fa7 PZ |
7943 | init_rt_bandwidth(&def_rt_bandwidth, |
7944 | global_rt_period(), global_rt_runtime()); | |
7945 | ||
7946 | #ifdef CONFIG_RT_GROUP_SCHED | |
7947 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7948 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7949 | #ifdef CONFIG_USER_SCHED |
7950 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7951 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7952 | #endif /* CONFIG_USER_SCHED */ |
7953 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7954 | |
052f1dc7 | 7955 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7956 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7957 | INIT_LIST_HEAD(&init_task_group.children); |
7958 | ||
7959 | #ifdef CONFIG_USER_SCHED | |
7960 | INIT_LIST_HEAD(&root_task_group.children); | |
7961 | init_task_group.parent = &root_task_group; | |
7962 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7963 | #endif /* CONFIG_USER_SCHED */ |
7964 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7965 | |
0a945022 | 7966 | for_each_possible_cpu(i) { |
70b97a7f | 7967 | struct rq *rq; |
1da177e4 LT |
7968 | |
7969 | rq = cpu_rq(i); | |
7970 | spin_lock_init(&rq->lock); | |
fcb99371 | 7971 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7972 | rq->nr_running = 0; |
dd41f596 | 7973 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7974 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7975 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7976 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7977 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7978 | #ifdef CONFIG_CGROUP_SCHED |
7979 | /* | |
7980 | * How much cpu bandwidth does init_task_group get? | |
7981 | * | |
7982 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7983 | * gets 100% of the cpu resources in the system. This overall | |
7984 | * system cpu resource is divided among the tasks of | |
7985 | * init_task_group and its child task-groups in a fair manner, | |
7986 | * based on each entity's (task or task-group's) weight | |
7987 | * (se->load.weight). | |
7988 | * | |
7989 | * In other words, if init_task_group has 10 tasks of weight | |
7990 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7991 | * then A0's share of the cpu resource is: | |
7992 | * | |
7993 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7994 | * | |
7995 | * We achieve this by letting init_task_group's tasks sit | |
7996 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7997 | */ | |
ec7dc8ac | 7998 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7999 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8000 | root_task_group.shares = NICE_0_LOAD; |
8001 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8002 | /* |
8003 | * In case of task-groups formed thr' the user id of tasks, | |
8004 | * init_task_group represents tasks belonging to root user. | |
8005 | * Hence it forms a sibling of all subsequent groups formed. | |
8006 | * In this case, init_task_group gets only a fraction of overall | |
8007 | * system cpu resource, based on the weight assigned to root | |
8008 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8009 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8010 | * (init_cfs_rq) and having one entity represent this group of | |
8011 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8012 | */ | |
ec7dc8ac | 8013 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8014 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8015 | &per_cpu(init_sched_entity, i), i, 1, |
8016 | root_task_group.se[i]); | |
6f505b16 | 8017 | |
052f1dc7 | 8018 | #endif |
354d60c2 DG |
8019 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8020 | ||
8021 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8022 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8023 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8024 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8025 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8026 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8027 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8028 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8029 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8030 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8031 | root_task_group.rt_se[i]); | |
354d60c2 | 8032 | #endif |
dd41f596 | 8033 | #endif |
1da177e4 | 8034 | |
dd41f596 IM |
8035 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8036 | rq->cpu_load[j] = 0; | |
1da177e4 | 8037 | #ifdef CONFIG_SMP |
41c7ce9a | 8038 | rq->sd = NULL; |
57d885fe | 8039 | rq->rd = NULL; |
1da177e4 | 8040 | rq->active_balance = 0; |
dd41f596 | 8041 | rq->next_balance = jiffies; |
1da177e4 | 8042 | rq->push_cpu = 0; |
0a2966b4 | 8043 | rq->cpu = i; |
1f11eb6a | 8044 | rq->online = 0; |
1da177e4 LT |
8045 | rq->migration_thread = NULL; |
8046 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8047 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8048 | #endif |
8f4d37ec | 8049 | init_rq_hrtick(rq); |
1da177e4 | 8050 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8051 | } |
8052 | ||
2dd73a4f | 8053 | set_load_weight(&init_task); |
b50f60ce | 8054 | |
e107be36 AK |
8055 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8056 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8057 | #endif | |
8058 | ||
c9819f45 | 8059 | #ifdef CONFIG_SMP |
962cf36c | 8060 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8061 | #endif |
8062 | ||
b50f60ce HC |
8063 | #ifdef CONFIG_RT_MUTEXES |
8064 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8065 | #endif | |
8066 | ||
1da177e4 LT |
8067 | /* |
8068 | * The boot idle thread does lazy MMU switching as well: | |
8069 | */ | |
8070 | atomic_inc(&init_mm.mm_count); | |
8071 | enter_lazy_tlb(&init_mm, current); | |
8072 | ||
8073 | /* | |
8074 | * Make us the idle thread. Technically, schedule() should not be | |
8075 | * called from this thread, however somewhere below it might be, | |
8076 | * but because we are the idle thread, we just pick up running again | |
8077 | * when this runqueue becomes "idle". | |
8078 | */ | |
8079 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8080 | /* |
8081 | * During early bootup we pretend to be a normal task: | |
8082 | */ | |
8083 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8084 | |
8085 | scheduler_running = 1; | |
1da177e4 LT |
8086 | } |
8087 | ||
8088 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8089 | void __might_sleep(char *file, int line) | |
8090 | { | |
48f24c4d | 8091 | #ifdef in_atomic |
1da177e4 LT |
8092 | static unsigned long prev_jiffy; /* ratelimiting */ |
8093 | ||
8094 | if ((in_atomic() || irqs_disabled()) && | |
8095 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8096 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8097 | return; | |
8098 | prev_jiffy = jiffies; | |
91368d73 | 8099 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8100 | " context at %s:%d\n", file, line); |
8101 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8102 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8103 | debug_show_held_locks(current); |
3117df04 IM |
8104 | if (irqs_disabled()) |
8105 | print_irqtrace_events(current); | |
1da177e4 LT |
8106 | dump_stack(); |
8107 | } | |
8108 | #endif | |
8109 | } | |
8110 | EXPORT_SYMBOL(__might_sleep); | |
8111 | #endif | |
8112 | ||
8113 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8114 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8115 | { | |
8116 | int on_rq; | |
3e51f33f | 8117 | |
3a5e4dc1 AK |
8118 | update_rq_clock(rq); |
8119 | on_rq = p->se.on_rq; | |
8120 | if (on_rq) | |
8121 | deactivate_task(rq, p, 0); | |
8122 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8123 | if (on_rq) { | |
8124 | activate_task(rq, p, 0); | |
8125 | resched_task(rq->curr); | |
8126 | } | |
8127 | } | |
8128 | ||
1da177e4 LT |
8129 | void normalize_rt_tasks(void) |
8130 | { | |
a0f98a1c | 8131 | struct task_struct *g, *p; |
1da177e4 | 8132 | unsigned long flags; |
70b97a7f | 8133 | struct rq *rq; |
1da177e4 | 8134 | |
4cf5d77a | 8135 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8136 | do_each_thread(g, p) { |
178be793 IM |
8137 | /* |
8138 | * Only normalize user tasks: | |
8139 | */ | |
8140 | if (!p->mm) | |
8141 | continue; | |
8142 | ||
6cfb0d5d | 8143 | p->se.exec_start = 0; |
6cfb0d5d | 8144 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8145 | p->se.wait_start = 0; |
dd41f596 | 8146 | p->se.sleep_start = 0; |
dd41f596 | 8147 | p->se.block_start = 0; |
6cfb0d5d | 8148 | #endif |
dd41f596 IM |
8149 | |
8150 | if (!rt_task(p)) { | |
8151 | /* | |
8152 | * Renice negative nice level userspace | |
8153 | * tasks back to 0: | |
8154 | */ | |
8155 | if (TASK_NICE(p) < 0 && p->mm) | |
8156 | set_user_nice(p, 0); | |
1da177e4 | 8157 | continue; |
dd41f596 | 8158 | } |
1da177e4 | 8159 | |
4cf5d77a | 8160 | spin_lock(&p->pi_lock); |
b29739f9 | 8161 | rq = __task_rq_lock(p); |
1da177e4 | 8162 | |
178be793 | 8163 | normalize_task(rq, p); |
3a5e4dc1 | 8164 | |
b29739f9 | 8165 | __task_rq_unlock(rq); |
4cf5d77a | 8166 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8167 | } while_each_thread(g, p); |
8168 | ||
4cf5d77a | 8169 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8170 | } |
8171 | ||
8172 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8173 | |
8174 | #ifdef CONFIG_IA64 | |
8175 | /* | |
8176 | * These functions are only useful for the IA64 MCA handling. | |
8177 | * | |
8178 | * They can only be called when the whole system has been | |
8179 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8180 | * activity can take place. Using them for anything else would | |
8181 | * be a serious bug, and as a result, they aren't even visible | |
8182 | * under any other configuration. | |
8183 | */ | |
8184 | ||
8185 | /** | |
8186 | * curr_task - return the current task for a given cpu. | |
8187 | * @cpu: the processor in question. | |
8188 | * | |
8189 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8190 | */ | |
36c8b586 | 8191 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8192 | { |
8193 | return cpu_curr(cpu); | |
8194 | } | |
8195 | ||
8196 | /** | |
8197 | * set_curr_task - set the current task for a given cpu. | |
8198 | * @cpu: the processor in question. | |
8199 | * @p: the task pointer to set. | |
8200 | * | |
8201 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8202 | * are serviced on a separate stack. It allows the architecture to switch the |
8203 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8204 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8205 | * and caller must save the original value of the current task (see | |
8206 | * curr_task() above) and restore that value before reenabling interrupts and | |
8207 | * re-starting the system. | |
8208 | * | |
8209 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8210 | */ | |
36c8b586 | 8211 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8212 | { |
8213 | cpu_curr(cpu) = p; | |
8214 | } | |
8215 | ||
8216 | #endif | |
29f59db3 | 8217 | |
bccbe08a PZ |
8218 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8219 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8220 | { |
8221 | int i; | |
8222 | ||
8223 | for_each_possible_cpu(i) { | |
8224 | if (tg->cfs_rq) | |
8225 | kfree(tg->cfs_rq[i]); | |
8226 | if (tg->se) | |
8227 | kfree(tg->se[i]); | |
6f505b16 PZ |
8228 | } |
8229 | ||
8230 | kfree(tg->cfs_rq); | |
8231 | kfree(tg->se); | |
6f505b16 PZ |
8232 | } |
8233 | ||
ec7dc8ac DG |
8234 | static |
8235 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8236 | { |
29f59db3 | 8237 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8238 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8239 | struct rq *rq; |
29f59db3 SV |
8240 | int i; |
8241 | ||
434d53b0 | 8242 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8243 | if (!tg->cfs_rq) |
8244 | goto err; | |
434d53b0 | 8245 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8246 | if (!tg->se) |
8247 | goto err; | |
052f1dc7 PZ |
8248 | |
8249 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8250 | |
8251 | for_each_possible_cpu(i) { | |
9b5b7751 | 8252 | rq = cpu_rq(i); |
29f59db3 | 8253 | |
6f505b16 PZ |
8254 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8255 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8256 | if (!cfs_rq) |
8257 | goto err; | |
8258 | ||
6f505b16 PZ |
8259 | se = kmalloc_node(sizeof(struct sched_entity), |
8260 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8261 | if (!se) |
8262 | goto err; | |
8263 | ||
ec7dc8ac DG |
8264 | parent_se = parent ? parent->se[i] : NULL; |
8265 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8266 | } |
8267 | ||
8268 | return 1; | |
8269 | ||
8270 | err: | |
8271 | return 0; | |
8272 | } | |
8273 | ||
8274 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8275 | { | |
8276 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8277 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8278 | } | |
8279 | ||
8280 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8281 | { | |
8282 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8283 | } | |
6d6bc0ad | 8284 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8285 | static inline void free_fair_sched_group(struct task_group *tg) |
8286 | { | |
8287 | } | |
8288 | ||
ec7dc8ac DG |
8289 | static inline |
8290 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8291 | { |
8292 | return 1; | |
8293 | } | |
8294 | ||
8295 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8296 | { | |
8297 | } | |
8298 | ||
8299 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8300 | { | |
8301 | } | |
6d6bc0ad | 8302 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8303 | |
8304 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8305 | static void free_rt_sched_group(struct task_group *tg) |
8306 | { | |
8307 | int i; | |
8308 | ||
d0b27fa7 PZ |
8309 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8310 | ||
bccbe08a PZ |
8311 | for_each_possible_cpu(i) { |
8312 | if (tg->rt_rq) | |
8313 | kfree(tg->rt_rq[i]); | |
8314 | if (tg->rt_se) | |
8315 | kfree(tg->rt_se[i]); | |
8316 | } | |
8317 | ||
8318 | kfree(tg->rt_rq); | |
8319 | kfree(tg->rt_se); | |
8320 | } | |
8321 | ||
ec7dc8ac DG |
8322 | static |
8323 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8324 | { |
8325 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8326 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8327 | struct rq *rq; |
8328 | int i; | |
8329 | ||
434d53b0 | 8330 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8331 | if (!tg->rt_rq) |
8332 | goto err; | |
434d53b0 | 8333 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8334 | if (!tg->rt_se) |
8335 | goto err; | |
8336 | ||
d0b27fa7 PZ |
8337 | init_rt_bandwidth(&tg->rt_bandwidth, |
8338 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8339 | |
8340 | for_each_possible_cpu(i) { | |
8341 | rq = cpu_rq(i); | |
8342 | ||
6f505b16 PZ |
8343 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8344 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8345 | if (!rt_rq) | |
8346 | goto err; | |
29f59db3 | 8347 | |
6f505b16 PZ |
8348 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8349 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8350 | if (!rt_se) | |
8351 | goto err; | |
29f59db3 | 8352 | |
ec7dc8ac DG |
8353 | parent_se = parent ? parent->rt_se[i] : NULL; |
8354 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8355 | } |
8356 | ||
bccbe08a PZ |
8357 | return 1; |
8358 | ||
8359 | err: | |
8360 | return 0; | |
8361 | } | |
8362 | ||
8363 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8364 | { | |
8365 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8366 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8367 | } | |
8368 | ||
8369 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8370 | { | |
8371 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8372 | } | |
6d6bc0ad | 8373 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8374 | static inline void free_rt_sched_group(struct task_group *tg) |
8375 | { | |
8376 | } | |
8377 | ||
ec7dc8ac DG |
8378 | static inline |
8379 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8380 | { |
8381 | return 1; | |
8382 | } | |
8383 | ||
8384 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8385 | { | |
8386 | } | |
8387 | ||
8388 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8389 | { | |
8390 | } | |
6d6bc0ad | 8391 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8392 | |
d0b27fa7 | 8393 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8394 | static void free_sched_group(struct task_group *tg) |
8395 | { | |
8396 | free_fair_sched_group(tg); | |
8397 | free_rt_sched_group(tg); | |
8398 | kfree(tg); | |
8399 | } | |
8400 | ||
8401 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8402 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8403 | { |
8404 | struct task_group *tg; | |
8405 | unsigned long flags; | |
8406 | int i; | |
8407 | ||
8408 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8409 | if (!tg) | |
8410 | return ERR_PTR(-ENOMEM); | |
8411 | ||
ec7dc8ac | 8412 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8413 | goto err; |
8414 | ||
ec7dc8ac | 8415 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8416 | goto err; |
8417 | ||
8ed36996 | 8418 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8419 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8420 | register_fair_sched_group(tg, i); |
8421 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8422 | } |
6f505b16 | 8423 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8424 | |
8425 | WARN_ON(!parent); /* root should already exist */ | |
8426 | ||
8427 | tg->parent = parent; | |
8428 | list_add_rcu(&tg->siblings, &parent->children); | |
8429 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8430 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8431 | |
9b5b7751 | 8432 | return tg; |
29f59db3 SV |
8433 | |
8434 | err: | |
6f505b16 | 8435 | free_sched_group(tg); |
29f59db3 SV |
8436 | return ERR_PTR(-ENOMEM); |
8437 | } | |
8438 | ||
9b5b7751 | 8439 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8440 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8441 | { |
29f59db3 | 8442 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8443 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8444 | } |
8445 | ||
9b5b7751 | 8446 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8447 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8448 | { |
8ed36996 | 8449 | unsigned long flags; |
9b5b7751 | 8450 | int i; |
29f59db3 | 8451 | |
8ed36996 | 8452 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8453 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8454 | unregister_fair_sched_group(tg, i); |
8455 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8456 | } |
6f505b16 | 8457 | list_del_rcu(&tg->list); |
f473aa5e | 8458 | list_del_rcu(&tg->siblings); |
8ed36996 | 8459 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8460 | |
9b5b7751 | 8461 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8462 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8463 | } |
8464 | ||
9b5b7751 | 8465 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8466 | * The caller of this function should have put the task in its new group |
8467 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8468 | * reflect its new group. | |
9b5b7751 SV |
8469 | */ |
8470 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8471 | { |
8472 | int on_rq, running; | |
8473 | unsigned long flags; | |
8474 | struct rq *rq; | |
8475 | ||
8476 | rq = task_rq_lock(tsk, &flags); | |
8477 | ||
29f59db3 SV |
8478 | update_rq_clock(rq); |
8479 | ||
051a1d1a | 8480 | running = task_current(rq, tsk); |
29f59db3 SV |
8481 | on_rq = tsk->se.on_rq; |
8482 | ||
0e1f3483 | 8483 | if (on_rq) |
29f59db3 | 8484 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8485 | if (unlikely(running)) |
8486 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8487 | |
6f505b16 | 8488 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8489 | |
810b3817 PZ |
8490 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8491 | if (tsk->sched_class->moved_group) | |
8492 | tsk->sched_class->moved_group(tsk); | |
8493 | #endif | |
8494 | ||
0e1f3483 HS |
8495 | if (unlikely(running)) |
8496 | tsk->sched_class->set_curr_task(rq); | |
8497 | if (on_rq) | |
7074badb | 8498 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8499 | |
29f59db3 SV |
8500 | task_rq_unlock(rq, &flags); |
8501 | } | |
6d6bc0ad | 8502 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8503 | |
052f1dc7 | 8504 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8505 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8506 | { |
8507 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8508 | int on_rq; |
8509 | ||
29f59db3 | 8510 | on_rq = se->on_rq; |
62fb1851 | 8511 | if (on_rq) |
29f59db3 SV |
8512 | dequeue_entity(cfs_rq, se, 0); |
8513 | ||
8514 | se->load.weight = shares; | |
e05510d0 | 8515 | se->load.inv_weight = 0; |
29f59db3 | 8516 | |
62fb1851 | 8517 | if (on_rq) |
29f59db3 | 8518 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8519 | } |
62fb1851 | 8520 | |
c09595f6 PZ |
8521 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8522 | { | |
8523 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8524 | struct rq *rq = cfs_rq->rq; | |
8525 | unsigned long flags; | |
8526 | ||
8527 | spin_lock_irqsave(&rq->lock, flags); | |
8528 | __set_se_shares(se, shares); | |
8529 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8530 | } |
8531 | ||
8ed36996 PZ |
8532 | static DEFINE_MUTEX(shares_mutex); |
8533 | ||
4cf86d77 | 8534 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8535 | { |
8536 | int i; | |
8ed36996 | 8537 | unsigned long flags; |
c61935fd | 8538 | |
ec7dc8ac DG |
8539 | /* |
8540 | * We can't change the weight of the root cgroup. | |
8541 | */ | |
8542 | if (!tg->se[0]) | |
8543 | return -EINVAL; | |
8544 | ||
18d95a28 PZ |
8545 | if (shares < MIN_SHARES) |
8546 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8547 | else if (shares > MAX_SHARES) |
8548 | shares = MAX_SHARES; | |
62fb1851 | 8549 | |
8ed36996 | 8550 | mutex_lock(&shares_mutex); |
9b5b7751 | 8551 | if (tg->shares == shares) |
5cb350ba | 8552 | goto done; |
29f59db3 | 8553 | |
8ed36996 | 8554 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8555 | for_each_possible_cpu(i) |
8556 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8557 | list_del_rcu(&tg->siblings); |
8ed36996 | 8558 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8559 | |
8560 | /* wait for any ongoing reference to this group to finish */ | |
8561 | synchronize_sched(); | |
8562 | ||
8563 | /* | |
8564 | * Now we are free to modify the group's share on each cpu | |
8565 | * w/o tripping rebalance_share or load_balance_fair. | |
8566 | */ | |
9b5b7751 | 8567 | tg->shares = shares; |
c09595f6 PZ |
8568 | for_each_possible_cpu(i) { |
8569 | /* | |
8570 | * force a rebalance | |
8571 | */ | |
8572 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8573 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8574 | } |
29f59db3 | 8575 | |
6b2d7700 SV |
8576 | /* |
8577 | * Enable load balance activity on this group, by inserting it back on | |
8578 | * each cpu's rq->leaf_cfs_rq_list. | |
8579 | */ | |
8ed36996 | 8580 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8581 | for_each_possible_cpu(i) |
8582 | register_fair_sched_group(tg, i); | |
f473aa5e | 8583 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8584 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8585 | done: |
8ed36996 | 8586 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8587 | return 0; |
29f59db3 SV |
8588 | } |
8589 | ||
5cb350ba DG |
8590 | unsigned long sched_group_shares(struct task_group *tg) |
8591 | { | |
8592 | return tg->shares; | |
8593 | } | |
052f1dc7 | 8594 | #endif |
5cb350ba | 8595 | |
052f1dc7 | 8596 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8597 | /* |
9f0c1e56 | 8598 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8599 | */ |
9f0c1e56 PZ |
8600 | static DEFINE_MUTEX(rt_constraints_mutex); |
8601 | ||
8602 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8603 | { | |
8604 | if (runtime == RUNTIME_INF) | |
8605 | return 1ULL << 16; | |
8606 | ||
6f6d6a1a | 8607 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8608 | } |
8609 | ||
b40b2e8e PZ |
8610 | #ifdef CONFIG_CGROUP_SCHED |
8611 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8612 | { | |
10b612f4 | 8613 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8614 | unsigned long total = 0; |
8615 | ||
8616 | if (!parent) { | |
8617 | if (global_rt_period() < period) | |
8618 | return 0; | |
8619 | ||
8620 | return to_ratio(period, runtime) < | |
8621 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8622 | } | |
8623 | ||
8624 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8625 | return 0; | |
8626 | ||
8627 | rcu_read_lock(); | |
8628 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8629 | if (tgi == tg) | |
8630 | continue; | |
8631 | ||
8632 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8633 | tgi->rt_bandwidth.rt_runtime); | |
8634 | } | |
8635 | rcu_read_unlock(); | |
8636 | ||
10b612f4 | 8637 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8638 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8639 | parent->rt_bandwidth.rt_runtime); | |
8640 | } | |
8641 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8642 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8643 | { |
8644 | struct task_group *tgi; | |
8645 | unsigned long total = 0; | |
9f0c1e56 | 8646 | unsigned long global_ratio = |
d0b27fa7 | 8647 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8648 | |
8649 | rcu_read_lock(); | |
9f0c1e56 PZ |
8650 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8651 | if (tgi == tg) | |
8652 | continue; | |
6f505b16 | 8653 | |
d0b27fa7 PZ |
8654 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8655 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8656 | } |
8657 | rcu_read_unlock(); | |
6f505b16 | 8658 | |
9f0c1e56 | 8659 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8660 | } |
b40b2e8e | 8661 | #endif |
6f505b16 | 8662 | |
521f1a24 DG |
8663 | /* Must be called with tasklist_lock held */ |
8664 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8665 | { | |
8666 | struct task_struct *g, *p; | |
8667 | do_each_thread(g, p) { | |
8668 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8669 | return 1; | |
8670 | } while_each_thread(g, p); | |
8671 | return 0; | |
8672 | } | |
8673 | ||
d0b27fa7 PZ |
8674 | static int tg_set_bandwidth(struct task_group *tg, |
8675 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8676 | { |
ac086bc2 | 8677 | int i, err = 0; |
9f0c1e56 | 8678 | |
9f0c1e56 | 8679 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8680 | read_lock(&tasklist_lock); |
ac086bc2 | 8681 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8682 | err = -EBUSY; |
8683 | goto unlock; | |
8684 | } | |
9f0c1e56 PZ |
8685 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8686 | err = -EINVAL; | |
8687 | goto unlock; | |
8688 | } | |
ac086bc2 PZ |
8689 | |
8690 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8691 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8692 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8693 | |
8694 | for_each_possible_cpu(i) { | |
8695 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8696 | ||
8697 | spin_lock(&rt_rq->rt_runtime_lock); | |
8698 | rt_rq->rt_runtime = rt_runtime; | |
8699 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8700 | } | |
8701 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8702 | unlock: |
521f1a24 | 8703 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8704 | mutex_unlock(&rt_constraints_mutex); |
8705 | ||
8706 | return err; | |
6f505b16 PZ |
8707 | } |
8708 | ||
d0b27fa7 PZ |
8709 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8710 | { | |
8711 | u64 rt_runtime, rt_period; | |
8712 | ||
8713 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8714 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8715 | if (rt_runtime_us < 0) | |
8716 | rt_runtime = RUNTIME_INF; | |
8717 | ||
8718 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8719 | } | |
8720 | ||
9f0c1e56 PZ |
8721 | long sched_group_rt_runtime(struct task_group *tg) |
8722 | { | |
8723 | u64 rt_runtime_us; | |
8724 | ||
d0b27fa7 | 8725 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8726 | return -1; |
8727 | ||
d0b27fa7 | 8728 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8729 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8730 | return rt_runtime_us; | |
8731 | } | |
d0b27fa7 PZ |
8732 | |
8733 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8734 | { | |
8735 | u64 rt_runtime, rt_period; | |
8736 | ||
8737 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8738 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8739 | ||
619b0488 R |
8740 | if (rt_period == 0) |
8741 | return -EINVAL; | |
8742 | ||
d0b27fa7 PZ |
8743 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8744 | } | |
8745 | ||
8746 | long sched_group_rt_period(struct task_group *tg) | |
8747 | { | |
8748 | u64 rt_period_us; | |
8749 | ||
8750 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8751 | do_div(rt_period_us, NSEC_PER_USEC); | |
8752 | return rt_period_us; | |
8753 | } | |
8754 | ||
8755 | static int sched_rt_global_constraints(void) | |
8756 | { | |
10b612f4 PZ |
8757 | struct task_group *tg = &root_task_group; |
8758 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8759 | int ret = 0; |
8760 | ||
10b612f4 PZ |
8761 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8762 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8763 | ||
d0b27fa7 | 8764 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8765 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8766 | ret = -EINVAL; |
8767 | mutex_unlock(&rt_constraints_mutex); | |
8768 | ||
8769 | return ret; | |
8770 | } | |
6d6bc0ad | 8771 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8772 | static int sched_rt_global_constraints(void) |
8773 | { | |
ac086bc2 PZ |
8774 | unsigned long flags; |
8775 | int i; | |
8776 | ||
8777 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8778 | for_each_possible_cpu(i) { | |
8779 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8780 | ||
8781 | spin_lock(&rt_rq->rt_runtime_lock); | |
8782 | rt_rq->rt_runtime = global_rt_runtime(); | |
8783 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8784 | } | |
8785 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8786 | ||
d0b27fa7 PZ |
8787 | return 0; |
8788 | } | |
6d6bc0ad | 8789 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8790 | |
8791 | int sched_rt_handler(struct ctl_table *table, int write, | |
8792 | struct file *filp, void __user *buffer, size_t *lenp, | |
8793 | loff_t *ppos) | |
8794 | { | |
8795 | int ret; | |
8796 | int old_period, old_runtime; | |
8797 | static DEFINE_MUTEX(mutex); | |
8798 | ||
8799 | mutex_lock(&mutex); | |
8800 | old_period = sysctl_sched_rt_period; | |
8801 | old_runtime = sysctl_sched_rt_runtime; | |
8802 | ||
8803 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8804 | ||
8805 | if (!ret && write) { | |
8806 | ret = sched_rt_global_constraints(); | |
8807 | if (ret) { | |
8808 | sysctl_sched_rt_period = old_period; | |
8809 | sysctl_sched_rt_runtime = old_runtime; | |
8810 | } else { | |
8811 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8812 | def_rt_bandwidth.rt_period = | |
8813 | ns_to_ktime(global_rt_period()); | |
8814 | } | |
8815 | } | |
8816 | mutex_unlock(&mutex); | |
8817 | ||
8818 | return ret; | |
8819 | } | |
68318b8e | 8820 | |
052f1dc7 | 8821 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8822 | |
8823 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8824 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8825 | { |
2b01dfe3 PM |
8826 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8827 | struct task_group, css); | |
68318b8e SV |
8828 | } |
8829 | ||
8830 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8831 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8832 | { |
ec7dc8ac | 8833 | struct task_group *tg, *parent; |
68318b8e | 8834 | |
2b01dfe3 | 8835 | if (!cgrp->parent) { |
68318b8e | 8836 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8837 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8838 | return &init_task_group.css; |
8839 | } | |
8840 | ||
ec7dc8ac DG |
8841 | parent = cgroup_tg(cgrp->parent); |
8842 | tg = sched_create_group(parent); | |
68318b8e SV |
8843 | if (IS_ERR(tg)) |
8844 | return ERR_PTR(-ENOMEM); | |
8845 | ||
8846 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8847 | tg->css.cgroup = cgrp; |
68318b8e SV |
8848 | |
8849 | return &tg->css; | |
8850 | } | |
8851 | ||
41a2d6cf IM |
8852 | static void |
8853 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8854 | { |
2b01dfe3 | 8855 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8856 | |
8857 | sched_destroy_group(tg); | |
8858 | } | |
8859 | ||
41a2d6cf IM |
8860 | static int |
8861 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8862 | struct task_struct *tsk) | |
68318b8e | 8863 | { |
b68aa230 PZ |
8864 | #ifdef CONFIG_RT_GROUP_SCHED |
8865 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8866 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8867 | return -EINVAL; |
8868 | #else | |
68318b8e SV |
8869 | /* We don't support RT-tasks being in separate groups */ |
8870 | if (tsk->sched_class != &fair_sched_class) | |
8871 | return -EINVAL; | |
b68aa230 | 8872 | #endif |
68318b8e SV |
8873 | |
8874 | return 0; | |
8875 | } | |
8876 | ||
8877 | static void | |
2b01dfe3 | 8878 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8879 | struct cgroup *old_cont, struct task_struct *tsk) |
8880 | { | |
8881 | sched_move_task(tsk); | |
8882 | } | |
8883 | ||
052f1dc7 | 8884 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8885 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8886 | u64 shareval) |
68318b8e | 8887 | { |
2b01dfe3 | 8888 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8889 | } |
8890 | ||
f4c753b7 | 8891 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8892 | { |
2b01dfe3 | 8893 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8894 | |
8895 | return (u64) tg->shares; | |
8896 | } | |
6d6bc0ad | 8897 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8898 | |
052f1dc7 | 8899 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8900 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8901 | s64 val) |
6f505b16 | 8902 | { |
06ecb27c | 8903 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8904 | } |
8905 | ||
06ecb27c | 8906 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8907 | { |
06ecb27c | 8908 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8909 | } |
d0b27fa7 PZ |
8910 | |
8911 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8912 | u64 rt_period_us) | |
8913 | { | |
8914 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8915 | } | |
8916 | ||
8917 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8918 | { | |
8919 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8920 | } | |
6d6bc0ad | 8921 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8922 | |
fe5c7cc2 | 8923 | static struct cftype cpu_files[] = { |
052f1dc7 | 8924 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8925 | { |
8926 | .name = "shares", | |
f4c753b7 PM |
8927 | .read_u64 = cpu_shares_read_u64, |
8928 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8929 | }, |
052f1dc7 PZ |
8930 | #endif |
8931 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8932 | { |
9f0c1e56 | 8933 | .name = "rt_runtime_us", |
06ecb27c PM |
8934 | .read_s64 = cpu_rt_runtime_read, |
8935 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8936 | }, |
d0b27fa7 PZ |
8937 | { |
8938 | .name = "rt_period_us", | |
f4c753b7 PM |
8939 | .read_u64 = cpu_rt_period_read_uint, |
8940 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8941 | }, |
052f1dc7 | 8942 | #endif |
68318b8e SV |
8943 | }; |
8944 | ||
8945 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8946 | { | |
fe5c7cc2 | 8947 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8948 | } |
8949 | ||
8950 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8951 | .name = "cpu", |
8952 | .create = cpu_cgroup_create, | |
8953 | .destroy = cpu_cgroup_destroy, | |
8954 | .can_attach = cpu_cgroup_can_attach, | |
8955 | .attach = cpu_cgroup_attach, | |
8956 | .populate = cpu_cgroup_populate, | |
8957 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8958 | .early_init = 1, |
8959 | }; | |
8960 | ||
052f1dc7 | 8961 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8962 | |
8963 | #ifdef CONFIG_CGROUP_CPUACCT | |
8964 | ||
8965 | /* | |
8966 | * CPU accounting code for task groups. | |
8967 | * | |
8968 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8969 | * (balbir@in.ibm.com). | |
8970 | */ | |
8971 | ||
8972 | /* track cpu usage of a group of tasks */ | |
8973 | struct cpuacct { | |
8974 | struct cgroup_subsys_state css; | |
8975 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8976 | u64 *cpuusage; | |
8977 | }; | |
8978 | ||
8979 | struct cgroup_subsys cpuacct_subsys; | |
8980 | ||
8981 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8982 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8983 | { |
32cd756a | 8984 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8985 | struct cpuacct, css); |
8986 | } | |
8987 | ||
8988 | /* return cpu accounting group to which this task belongs */ | |
8989 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8990 | { | |
8991 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8992 | struct cpuacct, css); | |
8993 | } | |
8994 | ||
8995 | /* create a new cpu accounting group */ | |
8996 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8997 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8998 | { |
8999 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9000 | ||
9001 | if (!ca) | |
9002 | return ERR_PTR(-ENOMEM); | |
9003 | ||
9004 | ca->cpuusage = alloc_percpu(u64); | |
9005 | if (!ca->cpuusage) { | |
9006 | kfree(ca); | |
9007 | return ERR_PTR(-ENOMEM); | |
9008 | } | |
9009 | ||
9010 | return &ca->css; | |
9011 | } | |
9012 | ||
9013 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9014 | static void |
32cd756a | 9015 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9016 | { |
32cd756a | 9017 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9018 | |
9019 | free_percpu(ca->cpuusage); | |
9020 | kfree(ca); | |
9021 | } | |
9022 | ||
9023 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9024 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9025 | { |
32cd756a | 9026 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9027 | u64 totalcpuusage = 0; |
9028 | int i; | |
9029 | ||
9030 | for_each_possible_cpu(i) { | |
9031 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9032 | ||
9033 | /* | |
9034 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9035 | * platforms. | |
9036 | */ | |
9037 | spin_lock_irq(&cpu_rq(i)->lock); | |
9038 | totalcpuusage += *cpuusage; | |
9039 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9040 | } | |
9041 | ||
9042 | return totalcpuusage; | |
9043 | } | |
9044 | ||
0297b803 DG |
9045 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9046 | u64 reset) | |
9047 | { | |
9048 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9049 | int err = 0; | |
9050 | int i; | |
9051 | ||
9052 | if (reset) { | |
9053 | err = -EINVAL; | |
9054 | goto out; | |
9055 | } | |
9056 | ||
9057 | for_each_possible_cpu(i) { | |
9058 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9059 | ||
9060 | spin_lock_irq(&cpu_rq(i)->lock); | |
9061 | *cpuusage = 0; | |
9062 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9063 | } | |
9064 | out: | |
9065 | return err; | |
9066 | } | |
9067 | ||
d842de87 SV |
9068 | static struct cftype files[] = { |
9069 | { | |
9070 | .name = "usage", | |
f4c753b7 PM |
9071 | .read_u64 = cpuusage_read, |
9072 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9073 | }, |
9074 | }; | |
9075 | ||
32cd756a | 9076 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9077 | { |
32cd756a | 9078 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9079 | } |
9080 | ||
9081 | /* | |
9082 | * charge this task's execution time to its accounting group. | |
9083 | * | |
9084 | * called with rq->lock held. | |
9085 | */ | |
9086 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9087 | { | |
9088 | struct cpuacct *ca; | |
9089 | ||
9090 | if (!cpuacct_subsys.active) | |
9091 | return; | |
9092 | ||
9093 | ca = task_ca(tsk); | |
9094 | if (ca) { | |
9095 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9096 | ||
9097 | *cpuusage += cputime; | |
9098 | } | |
9099 | } | |
9100 | ||
9101 | struct cgroup_subsys cpuacct_subsys = { | |
9102 | .name = "cpuacct", | |
9103 | .create = cpuacct_create, | |
9104 | .destroy = cpuacct_destroy, | |
9105 | .populate = cpuacct_populate, | |
9106 | .subsys_id = cpuacct_subsys_id, | |
9107 | }; | |
9108 | #endif /* CONFIG_CGROUP_CPUACCT */ |