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 | * | |
85ba2d86 RM |
1870 | * If @match_state is nonzero, it's the @p->state value just checked and |
1871 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1872 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1873 | * we return a positive number (its total switch count). If a second call | |
1874 | * a short while later returns the same number, the caller can be sure that | |
1875 | * @p has remained unscheduled the whole time. | |
1876 | * | |
1da177e4 LT |
1877 | * The caller must ensure that the task *will* unschedule sometime soon, |
1878 | * else this function might spin for a *long* time. This function can't | |
1879 | * be called with interrupts off, or it may introduce deadlock with | |
1880 | * smp_call_function() if an IPI is sent by the same process we are | |
1881 | * waiting to become inactive. | |
1882 | */ | |
85ba2d86 | 1883 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1884 | { |
1885 | unsigned long flags; | |
dd41f596 | 1886 | int running, on_rq; |
85ba2d86 | 1887 | unsigned long ncsw; |
70b97a7f | 1888 | struct rq *rq; |
1da177e4 | 1889 | |
3a5c359a AK |
1890 | for (;;) { |
1891 | /* | |
1892 | * We do the initial early heuristics without holding | |
1893 | * any task-queue locks at all. We'll only try to get | |
1894 | * the runqueue lock when things look like they will | |
1895 | * work out! | |
1896 | */ | |
1897 | rq = task_rq(p); | |
fa490cfd | 1898 | |
3a5c359a AK |
1899 | /* |
1900 | * If the task is actively running on another CPU | |
1901 | * still, just relax and busy-wait without holding | |
1902 | * any locks. | |
1903 | * | |
1904 | * NOTE! Since we don't hold any locks, it's not | |
1905 | * even sure that "rq" stays as the right runqueue! | |
1906 | * But we don't care, since "task_running()" will | |
1907 | * return false if the runqueue has changed and p | |
1908 | * is actually now running somewhere else! | |
1909 | */ | |
85ba2d86 RM |
1910 | while (task_running(rq, p)) { |
1911 | if (match_state && unlikely(p->state != match_state)) | |
1912 | return 0; | |
3a5c359a | 1913 | cpu_relax(); |
85ba2d86 | 1914 | } |
fa490cfd | 1915 | |
3a5c359a AK |
1916 | /* |
1917 | * Ok, time to look more closely! We need the rq | |
1918 | * lock now, to be *sure*. If we're wrong, we'll | |
1919 | * just go back and repeat. | |
1920 | */ | |
1921 | rq = task_rq_lock(p, &flags); | |
1922 | running = task_running(rq, p); | |
1923 | on_rq = p->se.on_rq; | |
85ba2d86 RM |
1924 | ncsw = 0; |
1925 | if (!match_state || p->state == match_state) { | |
1926 | ncsw = p->nivcsw + p->nvcsw; | |
1927 | if (unlikely(!ncsw)) | |
1928 | ncsw = 1; | |
1929 | } | |
3a5c359a | 1930 | task_rq_unlock(rq, &flags); |
fa490cfd | 1931 | |
85ba2d86 RM |
1932 | /* |
1933 | * If it changed from the expected state, bail out now. | |
1934 | */ | |
1935 | if (unlikely(!ncsw)) | |
1936 | break; | |
1937 | ||
3a5c359a AK |
1938 | /* |
1939 | * Was it really running after all now that we | |
1940 | * checked with the proper locks actually held? | |
1941 | * | |
1942 | * Oops. Go back and try again.. | |
1943 | */ | |
1944 | if (unlikely(running)) { | |
1945 | cpu_relax(); | |
1946 | continue; | |
1947 | } | |
fa490cfd | 1948 | |
3a5c359a AK |
1949 | /* |
1950 | * It's not enough that it's not actively running, | |
1951 | * it must be off the runqueue _entirely_, and not | |
1952 | * preempted! | |
1953 | * | |
1954 | * So if it wa still runnable (but just not actively | |
1955 | * running right now), it's preempted, and we should | |
1956 | * yield - it could be a while. | |
1957 | */ | |
1958 | if (unlikely(on_rq)) { | |
1959 | schedule_timeout_uninterruptible(1); | |
1960 | continue; | |
1961 | } | |
fa490cfd | 1962 | |
3a5c359a AK |
1963 | /* |
1964 | * Ahh, all good. It wasn't running, and it wasn't | |
1965 | * runnable, which means that it will never become | |
1966 | * running in the future either. We're all done! | |
1967 | */ | |
1968 | break; | |
1969 | } | |
85ba2d86 RM |
1970 | |
1971 | return ncsw; | |
1da177e4 LT |
1972 | } |
1973 | ||
1974 | /*** | |
1975 | * kick_process - kick a running thread to enter/exit the kernel | |
1976 | * @p: the to-be-kicked thread | |
1977 | * | |
1978 | * Cause a process which is running on another CPU to enter | |
1979 | * kernel-mode, without any delay. (to get signals handled.) | |
1980 | * | |
1981 | * NOTE: this function doesnt have to take the runqueue lock, | |
1982 | * because all it wants to ensure is that the remote task enters | |
1983 | * the kernel. If the IPI races and the task has been migrated | |
1984 | * to another CPU then no harm is done and the purpose has been | |
1985 | * achieved as well. | |
1986 | */ | |
36c8b586 | 1987 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1988 | { |
1989 | int cpu; | |
1990 | ||
1991 | preempt_disable(); | |
1992 | cpu = task_cpu(p); | |
1993 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1994 | smp_send_reschedule(cpu); | |
1995 | preempt_enable(); | |
1996 | } | |
1997 | ||
1998 | /* | |
2dd73a4f PW |
1999 | * Return a low guess at the load of a migration-source cpu weighted |
2000 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2001 | * |
2002 | * We want to under-estimate the load of migration sources, to | |
2003 | * balance conservatively. | |
2004 | */ | |
a9957449 | 2005 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2006 | { |
70b97a7f | 2007 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2008 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2009 | |
93b75217 | 2010 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2011 | return total; |
b910472d | 2012 | |
dd41f596 | 2013 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2014 | } |
2015 | ||
2016 | /* | |
2dd73a4f PW |
2017 | * Return a high guess at the load of a migration-target cpu weighted |
2018 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2019 | */ |
a9957449 | 2020 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2021 | { |
70b97a7f | 2022 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2023 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2024 | |
93b75217 | 2025 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2026 | return total; |
3b0bd9bc | 2027 | |
dd41f596 | 2028 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2029 | } |
2030 | ||
147cbb4b NP |
2031 | /* |
2032 | * find_idlest_group finds and returns the least busy CPU group within the | |
2033 | * domain. | |
2034 | */ | |
2035 | static struct sched_group * | |
2036 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2037 | { | |
2038 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2039 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2040 | int load_idx = sd->forkexec_idx; | |
2041 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2042 | ||
2043 | do { | |
2044 | unsigned long load, avg_load; | |
2045 | int local_group; | |
2046 | int i; | |
2047 | ||
da5a5522 BD |
2048 | /* Skip over this group if it has no CPUs allowed */ |
2049 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2050 | continue; |
da5a5522 | 2051 | |
147cbb4b | 2052 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2053 | |
2054 | /* Tally up the load of all CPUs in the group */ | |
2055 | avg_load = 0; | |
2056 | ||
363ab6f1 | 2057 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2058 | /* Bias balancing toward cpus of our domain */ |
2059 | if (local_group) | |
2060 | load = source_load(i, load_idx); | |
2061 | else | |
2062 | load = target_load(i, load_idx); | |
2063 | ||
2064 | avg_load += load; | |
2065 | } | |
2066 | ||
2067 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2068 | avg_load = sg_div_cpu_power(group, |
2069 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2070 | |
2071 | if (local_group) { | |
2072 | this_load = avg_load; | |
2073 | this = group; | |
2074 | } else if (avg_load < min_load) { | |
2075 | min_load = avg_load; | |
2076 | idlest = group; | |
2077 | } | |
3a5c359a | 2078 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2079 | |
2080 | if (!idlest || 100*this_load < imbalance*min_load) | |
2081 | return NULL; | |
2082 | return idlest; | |
2083 | } | |
2084 | ||
2085 | /* | |
0feaece9 | 2086 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2087 | */ |
95cdf3b7 | 2088 | static int |
7c16ec58 MT |
2089 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2090 | cpumask_t *tmp) | |
147cbb4b NP |
2091 | { |
2092 | unsigned long load, min_load = ULONG_MAX; | |
2093 | int idlest = -1; | |
2094 | int i; | |
2095 | ||
da5a5522 | 2096 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2097 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2098 | |
363ab6f1 | 2099 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2100 | load = weighted_cpuload(i); |
147cbb4b NP |
2101 | |
2102 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2103 | min_load = load; | |
2104 | idlest = i; | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | return idlest; | |
2109 | } | |
2110 | ||
476d139c NP |
2111 | /* |
2112 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2113 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2114 | * SD_BALANCE_EXEC. | |
2115 | * | |
2116 | * Balance, ie. select the least loaded group. | |
2117 | * | |
2118 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2119 | * | |
2120 | * preempt must be disabled. | |
2121 | */ | |
2122 | static int sched_balance_self(int cpu, int flag) | |
2123 | { | |
2124 | struct task_struct *t = current; | |
2125 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2126 | |
c96d145e | 2127 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2128 | /* |
2129 | * If power savings logic is enabled for a domain, stop there. | |
2130 | */ | |
5c45bf27 SS |
2131 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2132 | break; | |
476d139c NP |
2133 | if (tmp->flags & flag) |
2134 | sd = tmp; | |
c96d145e | 2135 | } |
476d139c | 2136 | |
039a1c41 PZ |
2137 | if (sd) |
2138 | update_shares(sd); | |
2139 | ||
476d139c | 2140 | while (sd) { |
7c16ec58 | 2141 | cpumask_t span, tmpmask; |
476d139c | 2142 | struct sched_group *group; |
1a848870 SS |
2143 | int new_cpu, weight; |
2144 | ||
2145 | if (!(sd->flags & flag)) { | |
2146 | sd = sd->child; | |
2147 | continue; | |
2148 | } | |
476d139c NP |
2149 | |
2150 | span = sd->span; | |
2151 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2152 | if (!group) { |
2153 | sd = sd->child; | |
2154 | continue; | |
2155 | } | |
476d139c | 2156 | |
7c16ec58 | 2157 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2158 | if (new_cpu == -1 || new_cpu == cpu) { |
2159 | /* Now try balancing at a lower domain level of cpu */ | |
2160 | sd = sd->child; | |
2161 | continue; | |
2162 | } | |
476d139c | 2163 | |
1a848870 | 2164 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2165 | cpu = new_cpu; |
476d139c NP |
2166 | sd = NULL; |
2167 | weight = cpus_weight(span); | |
2168 | for_each_domain(cpu, tmp) { | |
2169 | if (weight <= cpus_weight(tmp->span)) | |
2170 | break; | |
2171 | if (tmp->flags & flag) | |
2172 | sd = tmp; | |
2173 | } | |
2174 | /* while loop will break here if sd == NULL */ | |
2175 | } | |
2176 | ||
2177 | return cpu; | |
2178 | } | |
2179 | ||
2180 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2181 | |
1da177e4 LT |
2182 | /*** |
2183 | * try_to_wake_up - wake up a thread | |
2184 | * @p: the to-be-woken-up thread | |
2185 | * @state: the mask of task states that can be woken | |
2186 | * @sync: do a synchronous wakeup? | |
2187 | * | |
2188 | * Put it on the run-queue if it's not already there. The "current" | |
2189 | * thread is always on the run-queue (except when the actual | |
2190 | * re-schedule is in progress), and as such you're allowed to do | |
2191 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2192 | * runnable without the overhead of this. | |
2193 | * | |
2194 | * returns failure only if the task is already active. | |
2195 | */ | |
36c8b586 | 2196 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2197 | { |
cc367732 | 2198 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2199 | unsigned long flags; |
2200 | long old_state; | |
70b97a7f | 2201 | struct rq *rq; |
1da177e4 | 2202 | |
b85d0667 IM |
2203 | if (!sched_feat(SYNC_WAKEUPS)) |
2204 | sync = 0; | |
2205 | ||
2398f2c6 PZ |
2206 | #ifdef CONFIG_SMP |
2207 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2208 | struct sched_domain *sd; | |
2209 | ||
2210 | this_cpu = raw_smp_processor_id(); | |
2211 | cpu = task_cpu(p); | |
2212 | ||
2213 | for_each_domain(this_cpu, sd) { | |
2214 | if (cpu_isset(cpu, sd->span)) { | |
2215 | update_shares(sd); | |
2216 | break; | |
2217 | } | |
2218 | } | |
2219 | } | |
2220 | #endif | |
2221 | ||
04e2f174 | 2222 | smp_wmb(); |
1da177e4 LT |
2223 | rq = task_rq_lock(p, &flags); |
2224 | old_state = p->state; | |
2225 | if (!(old_state & state)) | |
2226 | goto out; | |
2227 | ||
dd41f596 | 2228 | if (p->se.on_rq) |
1da177e4 LT |
2229 | goto out_running; |
2230 | ||
2231 | cpu = task_cpu(p); | |
cc367732 | 2232 | orig_cpu = cpu; |
1da177e4 LT |
2233 | this_cpu = smp_processor_id(); |
2234 | ||
2235 | #ifdef CONFIG_SMP | |
2236 | if (unlikely(task_running(rq, p))) | |
2237 | goto out_activate; | |
2238 | ||
5d2f5a61 DA |
2239 | cpu = p->sched_class->select_task_rq(p, sync); |
2240 | if (cpu != orig_cpu) { | |
2241 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2242 | task_rq_unlock(rq, &flags); |
2243 | /* might preempt at this point */ | |
2244 | rq = task_rq_lock(p, &flags); | |
2245 | old_state = p->state; | |
2246 | if (!(old_state & state)) | |
2247 | goto out; | |
dd41f596 | 2248 | if (p->se.on_rq) |
1da177e4 LT |
2249 | goto out_running; |
2250 | ||
2251 | this_cpu = smp_processor_id(); | |
2252 | cpu = task_cpu(p); | |
2253 | } | |
2254 | ||
e7693a36 GH |
2255 | #ifdef CONFIG_SCHEDSTATS |
2256 | schedstat_inc(rq, ttwu_count); | |
2257 | if (cpu == this_cpu) | |
2258 | schedstat_inc(rq, ttwu_local); | |
2259 | else { | |
2260 | struct sched_domain *sd; | |
2261 | for_each_domain(this_cpu, sd) { | |
2262 | if (cpu_isset(cpu, sd->span)) { | |
2263 | schedstat_inc(sd, ttwu_wake_remote); | |
2264 | break; | |
2265 | } | |
2266 | } | |
2267 | } | |
6d6bc0ad | 2268 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2269 | |
1da177e4 LT |
2270 | out_activate: |
2271 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2272 | schedstat_inc(p, se.nr_wakeups); |
2273 | if (sync) | |
2274 | schedstat_inc(p, se.nr_wakeups_sync); | |
2275 | if (orig_cpu != cpu) | |
2276 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2277 | if (cpu == this_cpu) | |
2278 | schedstat_inc(p, se.nr_wakeups_local); | |
2279 | else | |
2280 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2281 | update_rq_clock(rq); |
dd41f596 | 2282 | activate_task(rq, p, 1); |
1da177e4 LT |
2283 | success = 1; |
2284 | ||
2285 | out_running: | |
5b82a1b0 MD |
2286 | trace_mark(kernel_sched_wakeup, |
2287 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2288 | p->pid, p->state, rq, p, rq->curr); | |
4ae7d5ce IM |
2289 | check_preempt_curr(rq, p); |
2290 | ||
1da177e4 | 2291 | p->state = TASK_RUNNING; |
9a897c5a SR |
2292 | #ifdef CONFIG_SMP |
2293 | if (p->sched_class->task_wake_up) | |
2294 | p->sched_class->task_wake_up(rq, p); | |
2295 | #endif | |
1da177e4 | 2296 | out: |
2087a1ad GH |
2297 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2298 | ||
1da177e4 LT |
2299 | task_rq_unlock(rq, &flags); |
2300 | ||
2301 | return success; | |
2302 | } | |
2303 | ||
7ad5b3a5 | 2304 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2305 | { |
d9514f6c | 2306 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2307 | } |
1da177e4 LT |
2308 | EXPORT_SYMBOL(wake_up_process); |
2309 | ||
7ad5b3a5 | 2310 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2311 | { |
2312 | return try_to_wake_up(p, state, 0); | |
2313 | } | |
2314 | ||
1da177e4 LT |
2315 | /* |
2316 | * Perform scheduler related setup for a newly forked process p. | |
2317 | * p is forked by current. | |
dd41f596 IM |
2318 | * |
2319 | * __sched_fork() is basic setup used by init_idle() too: | |
2320 | */ | |
2321 | static void __sched_fork(struct task_struct *p) | |
2322 | { | |
dd41f596 IM |
2323 | p->se.exec_start = 0; |
2324 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2325 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2326 | p->se.last_wakeup = 0; |
2327 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2328 | |
2329 | #ifdef CONFIG_SCHEDSTATS | |
2330 | p->se.wait_start = 0; | |
dd41f596 IM |
2331 | p->se.sum_sleep_runtime = 0; |
2332 | p->se.sleep_start = 0; | |
dd41f596 IM |
2333 | p->se.block_start = 0; |
2334 | p->se.sleep_max = 0; | |
2335 | p->se.block_max = 0; | |
2336 | p->se.exec_max = 0; | |
eba1ed4b | 2337 | p->se.slice_max = 0; |
dd41f596 | 2338 | p->se.wait_max = 0; |
6cfb0d5d | 2339 | #endif |
476d139c | 2340 | |
fa717060 | 2341 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2342 | p->se.on_rq = 0; |
4a55bd5e | 2343 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2344 | |
e107be36 AK |
2345 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2346 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2347 | #endif | |
2348 | ||
1da177e4 LT |
2349 | /* |
2350 | * We mark the process as running here, but have not actually | |
2351 | * inserted it onto the runqueue yet. This guarantees that | |
2352 | * nobody will actually run it, and a signal or other external | |
2353 | * event cannot wake it up and insert it on the runqueue either. | |
2354 | */ | |
2355 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2356 | } |
2357 | ||
2358 | /* | |
2359 | * fork()/clone()-time setup: | |
2360 | */ | |
2361 | void sched_fork(struct task_struct *p, int clone_flags) | |
2362 | { | |
2363 | int cpu = get_cpu(); | |
2364 | ||
2365 | __sched_fork(p); | |
2366 | ||
2367 | #ifdef CONFIG_SMP | |
2368 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2369 | #endif | |
02e4bac2 | 2370 | set_task_cpu(p, cpu); |
b29739f9 IM |
2371 | |
2372 | /* | |
2373 | * Make sure we do not leak PI boosting priority to the child: | |
2374 | */ | |
2375 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2376 | if (!rt_prio(p->prio)) |
2377 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2378 | |
52f17b6c | 2379 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2380 | if (likely(sched_info_on())) |
52f17b6c | 2381 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2382 | #endif |
d6077cb8 | 2383 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2384 | p->oncpu = 0; |
2385 | #endif | |
1da177e4 | 2386 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2387 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2388 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2389 | #endif |
476d139c | 2390 | put_cpu(); |
1da177e4 LT |
2391 | } |
2392 | ||
2393 | /* | |
2394 | * wake_up_new_task - wake up a newly created task for the first time. | |
2395 | * | |
2396 | * This function will do some initial scheduler statistics housekeeping | |
2397 | * that must be done for every newly created context, then puts the task | |
2398 | * on the runqueue and wakes it. | |
2399 | */ | |
7ad5b3a5 | 2400 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2401 | { |
2402 | unsigned long flags; | |
dd41f596 | 2403 | struct rq *rq; |
1da177e4 LT |
2404 | |
2405 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2406 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2407 | update_rq_clock(rq); |
1da177e4 LT |
2408 | |
2409 | p->prio = effective_prio(p); | |
2410 | ||
b9dca1e0 | 2411 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2412 | activate_task(rq, p, 0); |
1da177e4 | 2413 | } else { |
1da177e4 | 2414 | /* |
dd41f596 IM |
2415 | * Let the scheduling class do new task startup |
2416 | * management (if any): | |
1da177e4 | 2417 | */ |
ee0827d8 | 2418 | p->sched_class->task_new(rq, p); |
c09595f6 | 2419 | inc_nr_running(rq); |
1da177e4 | 2420 | } |
5b82a1b0 MD |
2421 | trace_mark(kernel_sched_wakeup_new, |
2422 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2423 | p->pid, p->state, rq, p, rq->curr); | |
dd41f596 | 2424 | check_preempt_curr(rq, p); |
9a897c5a SR |
2425 | #ifdef CONFIG_SMP |
2426 | if (p->sched_class->task_wake_up) | |
2427 | p->sched_class->task_wake_up(rq, p); | |
2428 | #endif | |
dd41f596 | 2429 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2430 | } |
2431 | ||
e107be36 AK |
2432 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2433 | ||
2434 | /** | |
421cee29 RD |
2435 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2436 | * @notifier: notifier struct to register | |
e107be36 AK |
2437 | */ |
2438 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2439 | { | |
2440 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2441 | } | |
2442 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2443 | ||
2444 | /** | |
2445 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2446 | * @notifier: notifier struct to unregister |
e107be36 AK |
2447 | * |
2448 | * This is safe to call from within a preemption notifier. | |
2449 | */ | |
2450 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2451 | { | |
2452 | hlist_del(¬ifier->link); | |
2453 | } | |
2454 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2455 | ||
2456 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2457 | { | |
2458 | struct preempt_notifier *notifier; | |
2459 | struct hlist_node *node; | |
2460 | ||
2461 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2462 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2463 | } | |
2464 | ||
2465 | static void | |
2466 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2467 | struct task_struct *next) | |
2468 | { | |
2469 | struct preempt_notifier *notifier; | |
2470 | struct hlist_node *node; | |
2471 | ||
2472 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2473 | notifier->ops->sched_out(notifier, next); | |
2474 | } | |
2475 | ||
6d6bc0ad | 2476 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2477 | |
2478 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2479 | { | |
2480 | } | |
2481 | ||
2482 | static void | |
2483 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2484 | struct task_struct *next) | |
2485 | { | |
2486 | } | |
2487 | ||
6d6bc0ad | 2488 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2489 | |
4866cde0 NP |
2490 | /** |
2491 | * prepare_task_switch - prepare to switch tasks | |
2492 | * @rq: the runqueue preparing to switch | |
421cee29 | 2493 | * @prev: the current task that is being switched out |
4866cde0 NP |
2494 | * @next: the task we are going to switch to. |
2495 | * | |
2496 | * This is called with the rq lock held and interrupts off. It must | |
2497 | * be paired with a subsequent finish_task_switch after the context | |
2498 | * switch. | |
2499 | * | |
2500 | * prepare_task_switch sets up locking and calls architecture specific | |
2501 | * hooks. | |
2502 | */ | |
e107be36 AK |
2503 | static inline void |
2504 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2505 | struct task_struct *next) | |
4866cde0 | 2506 | { |
e107be36 | 2507 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2508 | prepare_lock_switch(rq, next); |
2509 | prepare_arch_switch(next); | |
2510 | } | |
2511 | ||
1da177e4 LT |
2512 | /** |
2513 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2514 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2515 | * @prev: the thread we just switched away from. |
2516 | * | |
4866cde0 NP |
2517 | * finish_task_switch must be called after the context switch, paired |
2518 | * with a prepare_task_switch call before the context switch. | |
2519 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2520 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2521 | * |
2522 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2523 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2524 | * with the lock held can cause deadlocks; see schedule() for |
2525 | * details.) | |
2526 | */ | |
a9957449 | 2527 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2528 | __releases(rq->lock) |
2529 | { | |
1da177e4 | 2530 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2531 | long prev_state; |
1da177e4 LT |
2532 | |
2533 | rq->prev_mm = NULL; | |
2534 | ||
2535 | /* | |
2536 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2537 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2538 | * schedule one last time. The schedule call will never return, and |
2539 | * the scheduled task must drop that reference. | |
c394cc9f | 2540 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2541 | * still held, otherwise prev could be scheduled on another cpu, die |
2542 | * there before we look at prev->state, and then the reference would | |
2543 | * be dropped twice. | |
2544 | * Manfred Spraul <manfred@colorfullife.com> | |
2545 | */ | |
55a101f8 | 2546 | prev_state = prev->state; |
4866cde0 NP |
2547 | finish_arch_switch(prev); |
2548 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2549 | #ifdef CONFIG_SMP |
2550 | if (current->sched_class->post_schedule) | |
2551 | current->sched_class->post_schedule(rq); | |
2552 | #endif | |
e8fa1362 | 2553 | |
e107be36 | 2554 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2555 | if (mm) |
2556 | mmdrop(mm); | |
c394cc9f | 2557 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2558 | /* |
2559 | * Remove function-return probe instances associated with this | |
2560 | * task and put them back on the free list. | |
9761eea8 | 2561 | */ |
c6fd91f0 | 2562 | kprobe_flush_task(prev); |
1da177e4 | 2563 | put_task_struct(prev); |
c6fd91f0 | 2564 | } |
1da177e4 LT |
2565 | } |
2566 | ||
2567 | /** | |
2568 | * schedule_tail - first thing a freshly forked thread must call. | |
2569 | * @prev: the thread we just switched away from. | |
2570 | */ | |
36c8b586 | 2571 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2572 | __releases(rq->lock) |
2573 | { | |
70b97a7f IM |
2574 | struct rq *rq = this_rq(); |
2575 | ||
4866cde0 NP |
2576 | finish_task_switch(rq, prev); |
2577 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2578 | /* In this case, finish_task_switch does not reenable preemption */ | |
2579 | preempt_enable(); | |
2580 | #endif | |
1da177e4 | 2581 | if (current->set_child_tid) |
b488893a | 2582 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2583 | } |
2584 | ||
2585 | /* | |
2586 | * context_switch - switch to the new MM and the new | |
2587 | * thread's register state. | |
2588 | */ | |
dd41f596 | 2589 | static inline void |
70b97a7f | 2590 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2591 | struct task_struct *next) |
1da177e4 | 2592 | { |
dd41f596 | 2593 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2594 | |
e107be36 | 2595 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2596 | trace_mark(kernel_sched_schedule, |
2597 | "prev_pid %d next_pid %d prev_state %ld " | |
2598 | "## rq %p prev %p next %p", | |
2599 | prev->pid, next->pid, prev->state, | |
2600 | rq, prev, next); | |
dd41f596 IM |
2601 | mm = next->mm; |
2602 | oldmm = prev->active_mm; | |
9226d125 ZA |
2603 | /* |
2604 | * For paravirt, this is coupled with an exit in switch_to to | |
2605 | * combine the page table reload and the switch backend into | |
2606 | * one hypercall. | |
2607 | */ | |
2608 | arch_enter_lazy_cpu_mode(); | |
2609 | ||
dd41f596 | 2610 | if (unlikely(!mm)) { |
1da177e4 LT |
2611 | next->active_mm = oldmm; |
2612 | atomic_inc(&oldmm->mm_count); | |
2613 | enter_lazy_tlb(oldmm, next); | |
2614 | } else | |
2615 | switch_mm(oldmm, mm, next); | |
2616 | ||
dd41f596 | 2617 | if (unlikely(!prev->mm)) { |
1da177e4 | 2618 | prev->active_mm = NULL; |
1da177e4 LT |
2619 | rq->prev_mm = oldmm; |
2620 | } | |
3a5f5e48 IM |
2621 | /* |
2622 | * Since the runqueue lock will be released by the next | |
2623 | * task (which is an invalid locking op but in the case | |
2624 | * of the scheduler it's an obvious special-case), so we | |
2625 | * do an early lockdep release here: | |
2626 | */ | |
2627 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2628 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2629 | #endif |
1da177e4 LT |
2630 | |
2631 | /* Here we just switch the register state and the stack. */ | |
2632 | switch_to(prev, next, prev); | |
2633 | ||
dd41f596 IM |
2634 | barrier(); |
2635 | /* | |
2636 | * this_rq must be evaluated again because prev may have moved | |
2637 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2638 | * frame will be invalid. | |
2639 | */ | |
2640 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2641 | } |
2642 | ||
2643 | /* | |
2644 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2645 | * | |
2646 | * externally visible scheduler statistics: current number of runnable | |
2647 | * threads, current number of uninterruptible-sleeping threads, total | |
2648 | * number of context switches performed since bootup. | |
2649 | */ | |
2650 | unsigned long nr_running(void) | |
2651 | { | |
2652 | unsigned long i, sum = 0; | |
2653 | ||
2654 | for_each_online_cpu(i) | |
2655 | sum += cpu_rq(i)->nr_running; | |
2656 | ||
2657 | return sum; | |
2658 | } | |
2659 | ||
2660 | unsigned long nr_uninterruptible(void) | |
2661 | { | |
2662 | unsigned long i, sum = 0; | |
2663 | ||
0a945022 | 2664 | for_each_possible_cpu(i) |
1da177e4 LT |
2665 | sum += cpu_rq(i)->nr_uninterruptible; |
2666 | ||
2667 | /* | |
2668 | * Since we read the counters lockless, it might be slightly | |
2669 | * inaccurate. Do not allow it to go below zero though: | |
2670 | */ | |
2671 | if (unlikely((long)sum < 0)) | |
2672 | sum = 0; | |
2673 | ||
2674 | return sum; | |
2675 | } | |
2676 | ||
2677 | unsigned long long nr_context_switches(void) | |
2678 | { | |
cc94abfc SR |
2679 | int i; |
2680 | unsigned long long sum = 0; | |
1da177e4 | 2681 | |
0a945022 | 2682 | for_each_possible_cpu(i) |
1da177e4 LT |
2683 | sum += cpu_rq(i)->nr_switches; |
2684 | ||
2685 | return sum; | |
2686 | } | |
2687 | ||
2688 | unsigned long nr_iowait(void) | |
2689 | { | |
2690 | unsigned long i, sum = 0; | |
2691 | ||
0a945022 | 2692 | for_each_possible_cpu(i) |
1da177e4 LT |
2693 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2694 | ||
2695 | return sum; | |
2696 | } | |
2697 | ||
db1b1fef JS |
2698 | unsigned long nr_active(void) |
2699 | { | |
2700 | unsigned long i, running = 0, uninterruptible = 0; | |
2701 | ||
2702 | for_each_online_cpu(i) { | |
2703 | running += cpu_rq(i)->nr_running; | |
2704 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2705 | } | |
2706 | ||
2707 | if (unlikely((long)uninterruptible < 0)) | |
2708 | uninterruptible = 0; | |
2709 | ||
2710 | return running + uninterruptible; | |
2711 | } | |
2712 | ||
48f24c4d | 2713 | /* |
dd41f596 IM |
2714 | * Update rq->cpu_load[] statistics. This function is usually called every |
2715 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2716 | */ |
dd41f596 | 2717 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2718 | { |
495eca49 | 2719 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2720 | int i, scale; |
2721 | ||
2722 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2723 | |
2724 | /* Update our load: */ | |
2725 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2726 | unsigned long old_load, new_load; | |
2727 | ||
2728 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2729 | ||
2730 | old_load = this_rq->cpu_load[i]; | |
2731 | new_load = this_load; | |
a25707f3 IM |
2732 | /* |
2733 | * Round up the averaging division if load is increasing. This | |
2734 | * prevents us from getting stuck on 9 if the load is 10, for | |
2735 | * example. | |
2736 | */ | |
2737 | if (new_load > old_load) | |
2738 | new_load += scale-1; | |
dd41f596 IM |
2739 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2740 | } | |
48f24c4d IM |
2741 | } |
2742 | ||
dd41f596 IM |
2743 | #ifdef CONFIG_SMP |
2744 | ||
1da177e4 LT |
2745 | /* |
2746 | * double_rq_lock - safely lock two runqueues | |
2747 | * | |
2748 | * Note this does not disable interrupts like task_rq_lock, | |
2749 | * you need to do so manually before calling. | |
2750 | */ | |
70b97a7f | 2751 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2752 | __acquires(rq1->lock) |
2753 | __acquires(rq2->lock) | |
2754 | { | |
054b9108 | 2755 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2756 | if (rq1 == rq2) { |
2757 | spin_lock(&rq1->lock); | |
2758 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2759 | } else { | |
c96d145e | 2760 | if (rq1 < rq2) { |
1da177e4 LT |
2761 | spin_lock(&rq1->lock); |
2762 | spin_lock(&rq2->lock); | |
2763 | } else { | |
2764 | spin_lock(&rq2->lock); | |
2765 | spin_lock(&rq1->lock); | |
2766 | } | |
2767 | } | |
6e82a3be IM |
2768 | update_rq_clock(rq1); |
2769 | update_rq_clock(rq2); | |
1da177e4 LT |
2770 | } |
2771 | ||
2772 | /* | |
2773 | * double_rq_unlock - safely unlock two runqueues | |
2774 | * | |
2775 | * Note this does not restore interrupts like task_rq_unlock, | |
2776 | * you need to do so manually after calling. | |
2777 | */ | |
70b97a7f | 2778 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2779 | __releases(rq1->lock) |
2780 | __releases(rq2->lock) | |
2781 | { | |
2782 | spin_unlock(&rq1->lock); | |
2783 | if (rq1 != rq2) | |
2784 | spin_unlock(&rq2->lock); | |
2785 | else | |
2786 | __release(rq2->lock); | |
2787 | } | |
2788 | ||
2789 | /* | |
2790 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2791 | */ | |
e8fa1362 | 2792 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2793 | __releases(this_rq->lock) |
2794 | __acquires(busiest->lock) | |
2795 | __acquires(this_rq->lock) | |
2796 | { | |
e8fa1362 SR |
2797 | int ret = 0; |
2798 | ||
054b9108 KK |
2799 | if (unlikely(!irqs_disabled())) { |
2800 | /* printk() doesn't work good under rq->lock */ | |
2801 | spin_unlock(&this_rq->lock); | |
2802 | BUG_ON(1); | |
2803 | } | |
1da177e4 | 2804 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2805 | if (busiest < this_rq) { |
1da177e4 LT |
2806 | spin_unlock(&this_rq->lock); |
2807 | spin_lock(&busiest->lock); | |
2808 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2809 | ret = 1; |
1da177e4 LT |
2810 | } else |
2811 | spin_lock(&busiest->lock); | |
2812 | } | |
e8fa1362 | 2813 | return ret; |
1da177e4 LT |
2814 | } |
2815 | ||
1da177e4 LT |
2816 | /* |
2817 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2818 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2819 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2820 | * the cpu_allowed mask is restored. |
2821 | */ | |
36c8b586 | 2822 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2823 | { |
70b97a7f | 2824 | struct migration_req req; |
1da177e4 | 2825 | unsigned long flags; |
70b97a7f | 2826 | struct rq *rq; |
1da177e4 LT |
2827 | |
2828 | rq = task_rq_lock(p, &flags); | |
2829 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2830 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2831 | goto out; |
2832 | ||
2833 | /* force the process onto the specified CPU */ | |
2834 | if (migrate_task(p, dest_cpu, &req)) { | |
2835 | /* Need to wait for migration thread (might exit: take ref). */ | |
2836 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2837 | |
1da177e4 LT |
2838 | get_task_struct(mt); |
2839 | task_rq_unlock(rq, &flags); | |
2840 | wake_up_process(mt); | |
2841 | put_task_struct(mt); | |
2842 | wait_for_completion(&req.done); | |
36c8b586 | 2843 | |
1da177e4 LT |
2844 | return; |
2845 | } | |
2846 | out: | |
2847 | task_rq_unlock(rq, &flags); | |
2848 | } | |
2849 | ||
2850 | /* | |
476d139c NP |
2851 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2852 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2853 | */ |
2854 | void sched_exec(void) | |
2855 | { | |
1da177e4 | 2856 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2857 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2858 | put_cpu(); |
476d139c NP |
2859 | if (new_cpu != this_cpu) |
2860 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2861 | } |
2862 | ||
2863 | /* | |
2864 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2865 | * Both runqueues must be locked. | |
2866 | */ | |
dd41f596 IM |
2867 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2868 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2869 | { |
2e1cb74a | 2870 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2871 | set_task_cpu(p, this_cpu); |
dd41f596 | 2872 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2873 | /* |
2874 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2875 | * to be always true for them. | |
2876 | */ | |
dd41f596 | 2877 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2878 | } |
2879 | ||
2880 | /* | |
2881 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2882 | */ | |
858119e1 | 2883 | static |
70b97a7f | 2884 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2885 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2886 | int *all_pinned) |
1da177e4 LT |
2887 | { |
2888 | /* | |
2889 | * We do not migrate tasks that are: | |
2890 | * 1) running (obviously), or | |
2891 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2892 | * 3) are cache-hot on their current CPU. | |
2893 | */ | |
cc367732 IM |
2894 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2895 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2896 | return 0; |
cc367732 | 2897 | } |
81026794 NP |
2898 | *all_pinned = 0; |
2899 | ||
cc367732 IM |
2900 | if (task_running(rq, p)) { |
2901 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2902 | return 0; |
cc367732 | 2903 | } |
1da177e4 | 2904 | |
da84d961 IM |
2905 | /* |
2906 | * Aggressive migration if: | |
2907 | * 1) task is cache cold, or | |
2908 | * 2) too many balance attempts have failed. | |
2909 | */ | |
2910 | ||
6bc1665b IM |
2911 | if (!task_hot(p, rq->clock, sd) || |
2912 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2913 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2914 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2915 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2916 | schedstat_inc(p, se.nr_forced_migrations); |
2917 | } | |
da84d961 IM |
2918 | #endif |
2919 | return 1; | |
2920 | } | |
2921 | ||
cc367732 IM |
2922 | if (task_hot(p, rq->clock, sd)) { |
2923 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2924 | return 0; |
cc367732 | 2925 | } |
1da177e4 LT |
2926 | return 1; |
2927 | } | |
2928 | ||
e1d1484f PW |
2929 | static unsigned long |
2930 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2931 | unsigned long max_load_move, struct sched_domain *sd, | |
2932 | enum cpu_idle_type idle, int *all_pinned, | |
2933 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2934 | { |
051c6764 | 2935 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2936 | struct task_struct *p; |
2937 | long rem_load_move = max_load_move; | |
1da177e4 | 2938 | |
e1d1484f | 2939 | if (max_load_move == 0) |
1da177e4 LT |
2940 | goto out; |
2941 | ||
81026794 NP |
2942 | pinned = 1; |
2943 | ||
1da177e4 | 2944 | /* |
dd41f596 | 2945 | * Start the load-balancing iterator: |
1da177e4 | 2946 | */ |
dd41f596 IM |
2947 | p = iterator->start(iterator->arg); |
2948 | next: | |
b82d9fdd | 2949 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2950 | goto out; |
051c6764 PZ |
2951 | |
2952 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2953 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2954 | p = iterator->next(iterator->arg); |
2955 | goto next; | |
1da177e4 LT |
2956 | } |
2957 | ||
dd41f596 | 2958 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2959 | pulled++; |
dd41f596 | 2960 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2961 | |
2dd73a4f | 2962 | /* |
b82d9fdd | 2963 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2964 | */ |
e1d1484f | 2965 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2966 | if (p->prio < *this_best_prio) |
2967 | *this_best_prio = p->prio; | |
dd41f596 IM |
2968 | p = iterator->next(iterator->arg); |
2969 | goto next; | |
1da177e4 LT |
2970 | } |
2971 | out: | |
2972 | /* | |
e1d1484f | 2973 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2974 | * so we can safely collect pull_task() stats here rather than |
2975 | * inside pull_task(). | |
2976 | */ | |
2977 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2978 | |
2979 | if (all_pinned) | |
2980 | *all_pinned = pinned; | |
e1d1484f PW |
2981 | |
2982 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2983 | } |
2984 | ||
dd41f596 | 2985 | /* |
43010659 PW |
2986 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2987 | * this_rq, as part of a balancing operation within domain "sd". | |
2988 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2989 | * |
2990 | * Called with both runqueues locked. | |
2991 | */ | |
2992 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2993 | unsigned long max_load_move, |
dd41f596 IM |
2994 | struct sched_domain *sd, enum cpu_idle_type idle, |
2995 | int *all_pinned) | |
2996 | { | |
5522d5d5 | 2997 | const struct sched_class *class = sched_class_highest; |
43010659 | 2998 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2999 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3000 | |
3001 | do { | |
43010659 PW |
3002 | total_load_moved += |
3003 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3004 | max_load_move - total_load_moved, |
a4ac01c3 | 3005 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3006 | class = class->next; |
c4acb2c0 GH |
3007 | |
3008 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3009 | break; | |
3010 | ||
43010659 | 3011 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3012 | |
43010659 PW |
3013 | return total_load_moved > 0; |
3014 | } | |
3015 | ||
e1d1484f PW |
3016 | static int |
3017 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3018 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3019 | struct rq_iterator *iterator) | |
3020 | { | |
3021 | struct task_struct *p = iterator->start(iterator->arg); | |
3022 | int pinned = 0; | |
3023 | ||
3024 | while (p) { | |
3025 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3026 | pull_task(busiest, p, this_rq, this_cpu); | |
3027 | /* | |
3028 | * Right now, this is only the second place pull_task() | |
3029 | * is called, so we can safely collect pull_task() | |
3030 | * stats here rather than inside pull_task(). | |
3031 | */ | |
3032 | schedstat_inc(sd, lb_gained[idle]); | |
3033 | ||
3034 | return 1; | |
3035 | } | |
3036 | p = iterator->next(iterator->arg); | |
3037 | } | |
3038 | ||
3039 | return 0; | |
3040 | } | |
3041 | ||
43010659 PW |
3042 | /* |
3043 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3044 | * part of active balancing operations within "domain". | |
3045 | * Returns 1 if successful and 0 otherwise. | |
3046 | * | |
3047 | * Called with both runqueues locked. | |
3048 | */ | |
3049 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3050 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3051 | { | |
5522d5d5 | 3052 | const struct sched_class *class; |
43010659 PW |
3053 | |
3054 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3055 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3056 | return 1; |
3057 | ||
3058 | return 0; | |
dd41f596 IM |
3059 | } |
3060 | ||
1da177e4 LT |
3061 | /* |
3062 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3063 | * domain. It calculates and returns the amount of weighted load which |
3064 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3065 | */ |
3066 | static struct sched_group * | |
3067 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3068 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3069 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3070 | { |
3071 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3072 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3073 | unsigned long max_pull; |
2dd73a4f PW |
3074 | unsigned long busiest_load_per_task, busiest_nr_running; |
3075 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3076 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3077 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3078 | int power_savings_balance = 1; | |
3079 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3080 | unsigned long min_nr_running = ULONG_MAX; | |
3081 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3082 | #endif | |
1da177e4 LT |
3083 | |
3084 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3085 | busiest_load_per_task = busiest_nr_running = 0; |
3086 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3087 | |
d15bcfdb | 3088 | if (idle == CPU_NOT_IDLE) |
7897986b | 3089 | load_idx = sd->busy_idx; |
d15bcfdb | 3090 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3091 | load_idx = sd->newidle_idx; |
3092 | else | |
3093 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3094 | |
3095 | do { | |
908a7c1b | 3096 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3097 | int local_group; |
3098 | int i; | |
908a7c1b | 3099 | int __group_imb = 0; |
783609c6 | 3100 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3101 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3102 | unsigned long sum_avg_load_per_task; |
3103 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3104 | |
3105 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3106 | ||
783609c6 SS |
3107 | if (local_group) |
3108 | balance_cpu = first_cpu(group->cpumask); | |
3109 | ||
1da177e4 | 3110 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3111 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3112 | sum_avg_load_per_task = avg_load_per_task = 0; |
3113 | ||
908a7c1b KC |
3114 | max_cpu_load = 0; |
3115 | min_cpu_load = ~0UL; | |
1da177e4 | 3116 | |
363ab6f1 | 3117 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3118 | struct rq *rq; |
3119 | ||
3120 | if (!cpu_isset(i, *cpus)) | |
3121 | continue; | |
3122 | ||
3123 | rq = cpu_rq(i); | |
2dd73a4f | 3124 | |
9439aab8 | 3125 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3126 | *sd_idle = 0; |
3127 | ||
1da177e4 | 3128 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3129 | if (local_group) { |
3130 | if (idle_cpu(i) && !first_idle_cpu) { | |
3131 | first_idle_cpu = 1; | |
3132 | balance_cpu = i; | |
3133 | } | |
3134 | ||
a2000572 | 3135 | load = target_load(i, load_idx); |
908a7c1b | 3136 | } else { |
a2000572 | 3137 | load = source_load(i, load_idx); |
908a7c1b KC |
3138 | if (load > max_cpu_load) |
3139 | max_cpu_load = load; | |
3140 | if (min_cpu_load > load) | |
3141 | min_cpu_load = load; | |
3142 | } | |
1da177e4 LT |
3143 | |
3144 | avg_load += load; | |
2dd73a4f | 3145 | sum_nr_running += rq->nr_running; |
dd41f596 | 3146 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3147 | |
3148 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3149 | } |
3150 | ||
783609c6 SS |
3151 | /* |
3152 | * First idle cpu or the first cpu(busiest) in this sched group | |
3153 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3154 | * domains. In the newly idle case, we will allow all the cpu's |
3155 | * to do the newly idle load balance. | |
783609c6 | 3156 | */ |
9439aab8 SS |
3157 | if (idle != CPU_NEWLY_IDLE && local_group && |
3158 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3159 | *balance = 0; |
3160 | goto ret; | |
3161 | } | |
3162 | ||
1da177e4 | 3163 | total_load += avg_load; |
5517d86b | 3164 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3165 | |
3166 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3167 | avg_load = sg_div_cpu_power(group, |
3168 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3169 | |
408ed066 PZ |
3170 | |
3171 | /* | |
3172 | * Consider the group unbalanced when the imbalance is larger | |
3173 | * than the average weight of two tasks. | |
3174 | * | |
3175 | * APZ: with cgroup the avg task weight can vary wildly and | |
3176 | * might not be a suitable number - should we keep a | |
3177 | * normalized nr_running number somewhere that negates | |
3178 | * the hierarchy? | |
3179 | */ | |
3180 | avg_load_per_task = sg_div_cpu_power(group, | |
3181 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3182 | ||
3183 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3184 | __group_imb = 1; |
3185 | ||
5517d86b | 3186 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3187 | |
1da177e4 LT |
3188 | if (local_group) { |
3189 | this_load = avg_load; | |
3190 | this = group; | |
2dd73a4f PW |
3191 | this_nr_running = sum_nr_running; |
3192 | this_load_per_task = sum_weighted_load; | |
3193 | } else if (avg_load > max_load && | |
908a7c1b | 3194 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3195 | max_load = avg_load; |
3196 | busiest = group; | |
2dd73a4f PW |
3197 | busiest_nr_running = sum_nr_running; |
3198 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3199 | group_imb = __group_imb; |
1da177e4 | 3200 | } |
5c45bf27 SS |
3201 | |
3202 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3203 | /* | |
3204 | * Busy processors will not participate in power savings | |
3205 | * balance. | |
3206 | */ | |
dd41f596 IM |
3207 | if (idle == CPU_NOT_IDLE || |
3208 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3209 | goto group_next; | |
5c45bf27 SS |
3210 | |
3211 | /* | |
3212 | * If the local group is idle or completely loaded | |
3213 | * no need to do power savings balance at this domain | |
3214 | */ | |
3215 | if (local_group && (this_nr_running >= group_capacity || | |
3216 | !this_nr_running)) | |
3217 | power_savings_balance = 0; | |
3218 | ||
dd41f596 | 3219 | /* |
5c45bf27 SS |
3220 | * If a group is already running at full capacity or idle, |
3221 | * don't include that group in power savings calculations | |
dd41f596 IM |
3222 | */ |
3223 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3224 | || !sum_nr_running) |
dd41f596 | 3225 | goto group_next; |
5c45bf27 | 3226 | |
dd41f596 | 3227 | /* |
5c45bf27 | 3228 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3229 | * This is the group from where we need to pick up the load |
3230 | * for saving power | |
3231 | */ | |
3232 | if ((sum_nr_running < min_nr_running) || | |
3233 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3234 | first_cpu(group->cpumask) < |
3235 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3236 | group_min = group; |
3237 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3238 | min_load_per_task = sum_weighted_load / |
3239 | sum_nr_running; | |
dd41f596 | 3240 | } |
5c45bf27 | 3241 | |
dd41f596 | 3242 | /* |
5c45bf27 | 3243 | * Calculate the group which is almost near its |
dd41f596 IM |
3244 | * capacity but still has some space to pick up some load |
3245 | * from other group and save more power | |
3246 | */ | |
3247 | if (sum_nr_running <= group_capacity - 1) { | |
3248 | if (sum_nr_running > leader_nr_running || | |
3249 | (sum_nr_running == leader_nr_running && | |
3250 | first_cpu(group->cpumask) > | |
3251 | first_cpu(group_leader->cpumask))) { | |
3252 | group_leader = group; | |
3253 | leader_nr_running = sum_nr_running; | |
3254 | } | |
48f24c4d | 3255 | } |
5c45bf27 SS |
3256 | group_next: |
3257 | #endif | |
1da177e4 LT |
3258 | group = group->next; |
3259 | } while (group != sd->groups); | |
3260 | ||
2dd73a4f | 3261 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3262 | goto out_balanced; |
3263 | ||
3264 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3265 | ||
3266 | if (this_load >= avg_load || | |
3267 | 100*max_load <= sd->imbalance_pct*this_load) | |
3268 | goto out_balanced; | |
3269 | ||
2dd73a4f | 3270 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3271 | if (group_imb) |
3272 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3273 | ||
1da177e4 LT |
3274 | /* |
3275 | * We're trying to get all the cpus to the average_load, so we don't | |
3276 | * want to push ourselves above the average load, nor do we wish to | |
3277 | * reduce the max loaded cpu below the average load, as either of these | |
3278 | * actions would just result in more rebalancing later, and ping-pong | |
3279 | * tasks around. Thus we look for the minimum possible imbalance. | |
3280 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3281 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3282 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3283 | * appear as very large values with unsigned longs. |
3284 | */ | |
2dd73a4f PW |
3285 | if (max_load <= busiest_load_per_task) |
3286 | goto out_balanced; | |
3287 | ||
3288 | /* | |
3289 | * In the presence of smp nice balancing, certain scenarios can have | |
3290 | * max load less than avg load(as we skip the groups at or below | |
3291 | * its cpu_power, while calculating max_load..) | |
3292 | */ | |
3293 | if (max_load < avg_load) { | |
3294 | *imbalance = 0; | |
3295 | goto small_imbalance; | |
3296 | } | |
0c117f1b SS |
3297 | |
3298 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3299 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3300 | |
1da177e4 | 3301 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3302 | *imbalance = min(max_pull * busiest->__cpu_power, |
3303 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3304 | / SCHED_LOAD_SCALE; |
3305 | ||
2dd73a4f PW |
3306 | /* |
3307 | * if *imbalance is less than the average load per runnable task | |
3308 | * there is no gaurantee that any tasks will be moved so we'll have | |
3309 | * a think about bumping its value to force at least one task to be | |
3310 | * moved | |
3311 | */ | |
7fd0d2dd | 3312 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3313 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3314 | unsigned int imbn; |
3315 | ||
3316 | small_imbalance: | |
3317 | pwr_move = pwr_now = 0; | |
3318 | imbn = 2; | |
3319 | if (this_nr_running) { | |
3320 | this_load_per_task /= this_nr_running; | |
3321 | if (busiest_load_per_task > this_load_per_task) | |
3322 | imbn = 1; | |
3323 | } else | |
408ed066 | 3324 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3325 | |
408ed066 | 3326 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3327 | busiest_load_per_task * imbn) { |
2dd73a4f | 3328 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3329 | return busiest; |
3330 | } | |
3331 | ||
3332 | /* | |
3333 | * OK, we don't have enough imbalance to justify moving tasks, | |
3334 | * however we may be able to increase total CPU power used by | |
3335 | * moving them. | |
3336 | */ | |
3337 | ||
5517d86b ED |
3338 | pwr_now += busiest->__cpu_power * |
3339 | min(busiest_load_per_task, max_load); | |
3340 | pwr_now += this->__cpu_power * | |
3341 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3342 | pwr_now /= SCHED_LOAD_SCALE; |
3343 | ||
3344 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3345 | tmp = sg_div_cpu_power(busiest, |
3346 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3347 | if (max_load > tmp) |
5517d86b | 3348 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3349 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3350 | |
3351 | /* Amount of load we'd add */ | |
5517d86b | 3352 | if (max_load * busiest->__cpu_power < |
33859f7f | 3353 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3354 | tmp = sg_div_cpu_power(this, |
3355 | max_load * busiest->__cpu_power); | |
1da177e4 | 3356 | else |
5517d86b ED |
3357 | tmp = sg_div_cpu_power(this, |
3358 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3359 | pwr_move += this->__cpu_power * | |
3360 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3361 | pwr_move /= SCHED_LOAD_SCALE; |
3362 | ||
3363 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3364 | if (pwr_move > pwr_now) |
3365 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3366 | } |
3367 | ||
1da177e4 LT |
3368 | return busiest; |
3369 | ||
3370 | out_balanced: | |
5c45bf27 | 3371 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3372 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3373 | goto ret; |
1da177e4 | 3374 | |
5c45bf27 SS |
3375 | if (this == group_leader && group_leader != group_min) { |
3376 | *imbalance = min_load_per_task; | |
3377 | return group_min; | |
3378 | } | |
5c45bf27 | 3379 | #endif |
783609c6 | 3380 | ret: |
1da177e4 LT |
3381 | *imbalance = 0; |
3382 | return NULL; | |
3383 | } | |
3384 | ||
3385 | /* | |
3386 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3387 | */ | |
70b97a7f | 3388 | static struct rq * |
d15bcfdb | 3389 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3390 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3391 | { |
70b97a7f | 3392 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3393 | unsigned long max_load = 0; |
1da177e4 LT |
3394 | int i; |
3395 | ||
363ab6f1 | 3396 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3397 | unsigned long wl; |
0a2966b4 CL |
3398 | |
3399 | if (!cpu_isset(i, *cpus)) | |
3400 | continue; | |
3401 | ||
48f24c4d | 3402 | rq = cpu_rq(i); |
dd41f596 | 3403 | wl = weighted_cpuload(i); |
2dd73a4f | 3404 | |
dd41f596 | 3405 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3406 | continue; |
1da177e4 | 3407 | |
dd41f596 IM |
3408 | if (wl > max_load) { |
3409 | max_load = wl; | |
48f24c4d | 3410 | busiest = rq; |
1da177e4 LT |
3411 | } |
3412 | } | |
3413 | ||
3414 | return busiest; | |
3415 | } | |
3416 | ||
77391d71 NP |
3417 | /* |
3418 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3419 | * so long as it is large enough. | |
3420 | */ | |
3421 | #define MAX_PINNED_INTERVAL 512 | |
3422 | ||
1da177e4 LT |
3423 | /* |
3424 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3425 | * tasks if there is an imbalance. | |
1da177e4 | 3426 | */ |
70b97a7f | 3427 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3428 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3429 | int *balance, cpumask_t *cpus) |
1da177e4 | 3430 | { |
43010659 | 3431 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3432 | struct sched_group *group; |
1da177e4 | 3433 | unsigned long imbalance; |
70b97a7f | 3434 | struct rq *busiest; |
fe2eea3f | 3435 | unsigned long flags; |
5969fe06 | 3436 | |
7c16ec58 MT |
3437 | cpus_setall(*cpus); |
3438 | ||
89c4710e SS |
3439 | /* |
3440 | * When power savings policy is enabled for the parent domain, idle | |
3441 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3442 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3443 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3444 | */ |
d15bcfdb | 3445 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3446 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3447 | sd_idle = 1; |
1da177e4 | 3448 | |
2d72376b | 3449 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3450 | |
0a2966b4 | 3451 | redo: |
c8cba857 | 3452 | update_shares(sd); |
0a2966b4 | 3453 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3454 | cpus, balance); |
783609c6 | 3455 | |
06066714 | 3456 | if (*balance == 0) |
783609c6 | 3457 | goto out_balanced; |
783609c6 | 3458 | |
1da177e4 LT |
3459 | if (!group) { |
3460 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3461 | goto out_balanced; | |
3462 | } | |
3463 | ||
7c16ec58 | 3464 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3465 | if (!busiest) { |
3466 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3467 | goto out_balanced; | |
3468 | } | |
3469 | ||
db935dbd | 3470 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3471 | |
3472 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3473 | ||
43010659 | 3474 | ld_moved = 0; |
1da177e4 LT |
3475 | if (busiest->nr_running > 1) { |
3476 | /* | |
3477 | * Attempt to move tasks. If find_busiest_group has found | |
3478 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3479 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3480 | * correctly treated as an imbalance. |
3481 | */ | |
fe2eea3f | 3482 | local_irq_save(flags); |
e17224bf | 3483 | double_rq_lock(this_rq, busiest); |
43010659 | 3484 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3485 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3486 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3487 | local_irq_restore(flags); |
81026794 | 3488 | |
46cb4b7c SS |
3489 | /* |
3490 | * some other cpu did the load balance for us. | |
3491 | */ | |
43010659 | 3492 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3493 | resched_cpu(this_cpu); |
3494 | ||
81026794 | 3495 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3496 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3497 | cpu_clear(cpu_of(busiest), *cpus); |
3498 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3499 | goto redo; |
81026794 | 3500 | goto out_balanced; |
0a2966b4 | 3501 | } |
1da177e4 | 3502 | } |
81026794 | 3503 | |
43010659 | 3504 | if (!ld_moved) { |
1da177e4 LT |
3505 | schedstat_inc(sd, lb_failed[idle]); |
3506 | sd->nr_balance_failed++; | |
3507 | ||
3508 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3509 | |
fe2eea3f | 3510 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3511 | |
3512 | /* don't kick the migration_thread, if the curr | |
3513 | * task on busiest cpu can't be moved to this_cpu | |
3514 | */ | |
3515 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3516 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3517 | all_pinned = 1; |
3518 | goto out_one_pinned; | |
3519 | } | |
3520 | ||
1da177e4 LT |
3521 | if (!busiest->active_balance) { |
3522 | busiest->active_balance = 1; | |
3523 | busiest->push_cpu = this_cpu; | |
81026794 | 3524 | active_balance = 1; |
1da177e4 | 3525 | } |
fe2eea3f | 3526 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3527 | if (active_balance) |
1da177e4 LT |
3528 | wake_up_process(busiest->migration_thread); |
3529 | ||
3530 | /* | |
3531 | * We've kicked active balancing, reset the failure | |
3532 | * counter. | |
3533 | */ | |
39507451 | 3534 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3535 | } |
81026794 | 3536 | } else |
1da177e4 LT |
3537 | sd->nr_balance_failed = 0; |
3538 | ||
81026794 | 3539 | if (likely(!active_balance)) { |
1da177e4 LT |
3540 | /* We were unbalanced, so reset the balancing interval */ |
3541 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3542 | } else { |
3543 | /* | |
3544 | * If we've begun active balancing, start to back off. This | |
3545 | * case may not be covered by the all_pinned logic if there | |
3546 | * is only 1 task on the busy runqueue (because we don't call | |
3547 | * move_tasks). | |
3548 | */ | |
3549 | if (sd->balance_interval < sd->max_interval) | |
3550 | sd->balance_interval *= 2; | |
1da177e4 LT |
3551 | } |
3552 | ||
43010659 | 3553 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3554 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3555 | ld_moved = -1; |
3556 | ||
3557 | goto out; | |
1da177e4 LT |
3558 | |
3559 | out_balanced: | |
1da177e4 LT |
3560 | schedstat_inc(sd, lb_balanced[idle]); |
3561 | ||
16cfb1c0 | 3562 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3563 | |
3564 | out_one_pinned: | |
1da177e4 | 3565 | /* tune up the balancing interval */ |
77391d71 NP |
3566 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3567 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3568 | sd->balance_interval *= 2; |
3569 | ||
48f24c4d | 3570 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3571 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3572 | ld_moved = -1; |
3573 | else | |
3574 | ld_moved = 0; | |
3575 | out: | |
c8cba857 PZ |
3576 | if (ld_moved) |
3577 | update_shares(sd); | |
c09595f6 | 3578 | return ld_moved; |
1da177e4 LT |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3583 | * tasks if there is an imbalance. | |
3584 | * | |
d15bcfdb | 3585 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3586 | * this_rq is locked. |
3587 | */ | |
48f24c4d | 3588 | static int |
7c16ec58 MT |
3589 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3590 | cpumask_t *cpus) | |
1da177e4 LT |
3591 | { |
3592 | struct sched_group *group; | |
70b97a7f | 3593 | struct rq *busiest = NULL; |
1da177e4 | 3594 | unsigned long imbalance; |
43010659 | 3595 | int ld_moved = 0; |
5969fe06 | 3596 | int sd_idle = 0; |
969bb4e4 | 3597 | int all_pinned = 0; |
7c16ec58 MT |
3598 | |
3599 | cpus_setall(*cpus); | |
5969fe06 | 3600 | |
89c4710e SS |
3601 | /* |
3602 | * When power savings policy is enabled for the parent domain, idle | |
3603 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3604 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3605 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3606 | */ |
3607 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3608 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3609 | sd_idle = 1; |
1da177e4 | 3610 | |
2d72376b | 3611 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3612 | redo: |
3e5459b4 | 3613 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3614 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3615 | &sd_idle, cpus, NULL); |
1da177e4 | 3616 | if (!group) { |
d15bcfdb | 3617 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3618 | goto out_balanced; |
1da177e4 LT |
3619 | } |
3620 | ||
7c16ec58 | 3621 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3622 | if (!busiest) { |
d15bcfdb | 3623 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3624 | goto out_balanced; |
1da177e4 LT |
3625 | } |
3626 | ||
db935dbd NP |
3627 | BUG_ON(busiest == this_rq); |
3628 | ||
d15bcfdb | 3629 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3630 | |
43010659 | 3631 | ld_moved = 0; |
d6d5cfaf NP |
3632 | if (busiest->nr_running > 1) { |
3633 | /* Attempt to move tasks */ | |
3634 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3635 | /* this_rq->clock is already updated */ |
3636 | update_rq_clock(busiest); | |
43010659 | 3637 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3638 | imbalance, sd, CPU_NEWLY_IDLE, |
3639 | &all_pinned); | |
d6d5cfaf | 3640 | spin_unlock(&busiest->lock); |
0a2966b4 | 3641 | |
969bb4e4 | 3642 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3643 | cpu_clear(cpu_of(busiest), *cpus); |
3644 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3645 | goto redo; |
3646 | } | |
d6d5cfaf NP |
3647 | } |
3648 | ||
43010659 | 3649 | if (!ld_moved) { |
d15bcfdb | 3650 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3651 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3652 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3653 | return -1; |
3654 | } else | |
16cfb1c0 | 3655 | sd->nr_balance_failed = 0; |
1da177e4 | 3656 | |
3e5459b4 | 3657 | update_shares_locked(this_rq, sd); |
43010659 | 3658 | return ld_moved; |
16cfb1c0 NP |
3659 | |
3660 | out_balanced: | |
d15bcfdb | 3661 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3662 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3663 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3664 | return -1; |
16cfb1c0 | 3665 | sd->nr_balance_failed = 0; |
48f24c4d | 3666 | |
16cfb1c0 | 3667 | return 0; |
1da177e4 LT |
3668 | } |
3669 | ||
3670 | /* | |
3671 | * idle_balance is called by schedule() if this_cpu is about to become | |
3672 | * idle. Attempts to pull tasks from other CPUs. | |
3673 | */ | |
70b97a7f | 3674 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3675 | { |
3676 | struct sched_domain *sd; | |
dd41f596 IM |
3677 | int pulled_task = -1; |
3678 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3679 | cpumask_t tmpmask; |
1da177e4 LT |
3680 | |
3681 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3682 | unsigned long interval; |
3683 | ||
3684 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3685 | continue; | |
3686 | ||
3687 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3688 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3689 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3690 | sd, &tmpmask); | |
92c4ca5c CL |
3691 | |
3692 | interval = msecs_to_jiffies(sd->balance_interval); | |
3693 | if (time_after(next_balance, sd->last_balance + interval)) | |
3694 | next_balance = sd->last_balance + interval; | |
3695 | if (pulled_task) | |
3696 | break; | |
1da177e4 | 3697 | } |
dd41f596 | 3698 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3699 | /* |
3700 | * We are going idle. next_balance may be set based on | |
3701 | * a busy processor. So reset next_balance. | |
3702 | */ | |
3703 | this_rq->next_balance = next_balance; | |
dd41f596 | 3704 | } |
1da177e4 LT |
3705 | } |
3706 | ||
3707 | /* | |
3708 | * active_load_balance is run by migration threads. It pushes running tasks | |
3709 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3710 | * running on each physical CPU where possible, and avoids physical / | |
3711 | * logical imbalances. | |
3712 | * | |
3713 | * Called with busiest_rq locked. | |
3714 | */ | |
70b97a7f | 3715 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3716 | { |
39507451 | 3717 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3718 | struct sched_domain *sd; |
3719 | struct rq *target_rq; | |
39507451 | 3720 | |
48f24c4d | 3721 | /* Is there any task to move? */ |
39507451 | 3722 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3723 | return; |
3724 | ||
3725 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3726 | |
3727 | /* | |
39507451 | 3728 | * This condition is "impossible", if it occurs |
41a2d6cf | 3729 | * we need to fix it. Originally reported by |
39507451 | 3730 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3731 | */ |
39507451 | 3732 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3733 | |
39507451 NP |
3734 | /* move a task from busiest_rq to target_rq */ |
3735 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3736 | update_rq_clock(busiest_rq); |
3737 | update_rq_clock(target_rq); | |
39507451 NP |
3738 | |
3739 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3740 | for_each_domain(target_cpu, sd) { |
39507451 | 3741 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3742 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3743 | break; |
c96d145e | 3744 | } |
39507451 | 3745 | |
48f24c4d | 3746 | if (likely(sd)) { |
2d72376b | 3747 | schedstat_inc(sd, alb_count); |
39507451 | 3748 | |
43010659 PW |
3749 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3750 | sd, CPU_IDLE)) | |
48f24c4d IM |
3751 | schedstat_inc(sd, alb_pushed); |
3752 | else | |
3753 | schedstat_inc(sd, alb_failed); | |
3754 | } | |
39507451 | 3755 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3756 | } |
3757 | ||
46cb4b7c SS |
3758 | #ifdef CONFIG_NO_HZ |
3759 | static struct { | |
3760 | atomic_t load_balancer; | |
41a2d6cf | 3761 | cpumask_t cpu_mask; |
46cb4b7c SS |
3762 | } nohz ____cacheline_aligned = { |
3763 | .load_balancer = ATOMIC_INIT(-1), | |
3764 | .cpu_mask = CPU_MASK_NONE, | |
3765 | }; | |
3766 | ||
7835b98b | 3767 | /* |
46cb4b7c SS |
3768 | * This routine will try to nominate the ilb (idle load balancing) |
3769 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3770 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3771 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3772 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3773 | * arrives... | |
3774 | * | |
3775 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3776 | * for idle load balancing. ilb owner will still be part of | |
3777 | * nohz.cpu_mask.. | |
7835b98b | 3778 | * |
46cb4b7c SS |
3779 | * While stopping the tick, this cpu will become the ilb owner if there |
3780 | * is no other owner. And will be the owner till that cpu becomes busy | |
3781 | * or if all cpus in the system stop their ticks at which point | |
3782 | * there is no need for ilb owner. | |
3783 | * | |
3784 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3785 | * next busy scheduler_tick() | |
3786 | */ | |
3787 | int select_nohz_load_balancer(int stop_tick) | |
3788 | { | |
3789 | int cpu = smp_processor_id(); | |
3790 | ||
3791 | if (stop_tick) { | |
3792 | cpu_set(cpu, nohz.cpu_mask); | |
3793 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3794 | ||
3795 | /* | |
3796 | * If we are going offline and still the leader, give up! | |
3797 | */ | |
e761b772 | 3798 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3799 | atomic_read(&nohz.load_balancer) == cpu) { |
3800 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3801 | BUG(); | |
3802 | return 0; | |
3803 | } | |
3804 | ||
3805 | /* time for ilb owner also to sleep */ | |
3806 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3807 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3808 | atomic_set(&nohz.load_balancer, -1); | |
3809 | return 0; | |
3810 | } | |
3811 | ||
3812 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3813 | /* make me the ilb owner */ | |
3814 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3815 | return 1; | |
3816 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3817 | return 1; | |
3818 | } else { | |
3819 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3820 | return 0; | |
3821 | ||
3822 | cpu_clear(cpu, nohz.cpu_mask); | |
3823 | ||
3824 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3825 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3826 | BUG(); | |
3827 | } | |
3828 | return 0; | |
3829 | } | |
3830 | #endif | |
3831 | ||
3832 | static DEFINE_SPINLOCK(balancing); | |
3833 | ||
3834 | /* | |
7835b98b CL |
3835 | * It checks each scheduling domain to see if it is due to be balanced, |
3836 | * and initiates a balancing operation if so. | |
3837 | * | |
3838 | * Balancing parameters are set up in arch_init_sched_domains. | |
3839 | */ | |
a9957449 | 3840 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3841 | { |
46cb4b7c SS |
3842 | int balance = 1; |
3843 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3844 | unsigned long interval; |
3845 | struct sched_domain *sd; | |
46cb4b7c | 3846 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3847 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3848 | int update_next_balance = 0; |
d07355f5 | 3849 | int need_serialize; |
7c16ec58 | 3850 | cpumask_t tmp; |
1da177e4 | 3851 | |
46cb4b7c | 3852 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3853 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3854 | continue; | |
3855 | ||
3856 | interval = sd->balance_interval; | |
d15bcfdb | 3857 | if (idle != CPU_IDLE) |
1da177e4 LT |
3858 | interval *= sd->busy_factor; |
3859 | ||
3860 | /* scale ms to jiffies */ | |
3861 | interval = msecs_to_jiffies(interval); | |
3862 | if (unlikely(!interval)) | |
3863 | interval = 1; | |
dd41f596 IM |
3864 | if (interval > HZ*NR_CPUS/10) |
3865 | interval = HZ*NR_CPUS/10; | |
3866 | ||
d07355f5 | 3867 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3868 | |
d07355f5 | 3869 | if (need_serialize) { |
08c183f3 CL |
3870 | if (!spin_trylock(&balancing)) |
3871 | goto out; | |
3872 | } | |
3873 | ||
c9819f45 | 3874 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3875 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3876 | /* |
3877 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3878 | * longer idle, or one of our SMT siblings is |
3879 | * not idle. | |
3880 | */ | |
d15bcfdb | 3881 | idle = CPU_NOT_IDLE; |
1da177e4 | 3882 | } |
1bd77f2d | 3883 | sd->last_balance = jiffies; |
1da177e4 | 3884 | } |
d07355f5 | 3885 | if (need_serialize) |
08c183f3 CL |
3886 | spin_unlock(&balancing); |
3887 | out: | |
f549da84 | 3888 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3889 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3890 | update_next_balance = 1; |
3891 | } | |
783609c6 SS |
3892 | |
3893 | /* | |
3894 | * Stop the load balance at this level. There is another | |
3895 | * CPU in our sched group which is doing load balancing more | |
3896 | * actively. | |
3897 | */ | |
3898 | if (!balance) | |
3899 | break; | |
1da177e4 | 3900 | } |
f549da84 SS |
3901 | |
3902 | /* | |
3903 | * next_balance will be updated only when there is a need. | |
3904 | * When the cpu is attached to null domain for ex, it will not be | |
3905 | * updated. | |
3906 | */ | |
3907 | if (likely(update_next_balance)) | |
3908 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3909 | } |
3910 | ||
3911 | /* | |
3912 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3913 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3914 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3915 | */ | |
3916 | static void run_rebalance_domains(struct softirq_action *h) | |
3917 | { | |
dd41f596 IM |
3918 | int this_cpu = smp_processor_id(); |
3919 | struct rq *this_rq = cpu_rq(this_cpu); | |
3920 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3921 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3922 | |
dd41f596 | 3923 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3924 | |
3925 | #ifdef CONFIG_NO_HZ | |
3926 | /* | |
3927 | * If this cpu is the owner for idle load balancing, then do the | |
3928 | * balancing on behalf of the other idle cpus whose ticks are | |
3929 | * stopped. | |
3930 | */ | |
dd41f596 IM |
3931 | if (this_rq->idle_at_tick && |
3932 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3933 | cpumask_t cpus = nohz.cpu_mask; |
3934 | struct rq *rq; | |
3935 | int balance_cpu; | |
3936 | ||
dd41f596 | 3937 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3938 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3939 | /* |
3940 | * If this cpu gets work to do, stop the load balancing | |
3941 | * work being done for other cpus. Next load | |
3942 | * balancing owner will pick it up. | |
3943 | */ | |
3944 | if (need_resched()) | |
3945 | break; | |
3946 | ||
de0cf899 | 3947 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3948 | |
3949 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3950 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3951 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3952 | } |
3953 | } | |
3954 | #endif | |
3955 | } | |
3956 | ||
3957 | /* | |
3958 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3959 | * | |
3960 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3961 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3962 | * if the whole system is idle. | |
3963 | */ | |
dd41f596 | 3964 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3965 | { |
46cb4b7c SS |
3966 | #ifdef CONFIG_NO_HZ |
3967 | /* | |
3968 | * If we were in the nohz mode recently and busy at the current | |
3969 | * scheduler tick, then check if we need to nominate new idle | |
3970 | * load balancer. | |
3971 | */ | |
3972 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3973 | rq->in_nohz_recently = 0; | |
3974 | ||
3975 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3976 | cpu_clear(cpu, nohz.cpu_mask); | |
3977 | atomic_set(&nohz.load_balancer, -1); | |
3978 | } | |
3979 | ||
3980 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3981 | /* | |
3982 | * simple selection for now: Nominate the | |
3983 | * first cpu in the nohz list to be the next | |
3984 | * ilb owner. | |
3985 | * | |
3986 | * TBD: Traverse the sched domains and nominate | |
3987 | * the nearest cpu in the nohz.cpu_mask. | |
3988 | */ | |
3989 | int ilb = first_cpu(nohz.cpu_mask); | |
3990 | ||
434d53b0 | 3991 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3992 | resched_cpu(ilb); |
3993 | } | |
3994 | } | |
3995 | ||
3996 | /* | |
3997 | * If this cpu is idle and doing idle load balancing for all the | |
3998 | * cpus with ticks stopped, is it time for that to stop? | |
3999 | */ | |
4000 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4001 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4002 | resched_cpu(cpu); | |
4003 | return; | |
4004 | } | |
4005 | ||
4006 | /* | |
4007 | * If this cpu is idle and the idle load balancing is done by | |
4008 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4009 | */ | |
4010 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4011 | cpu_isset(cpu, nohz.cpu_mask)) | |
4012 | return; | |
4013 | #endif | |
4014 | if (time_after_eq(jiffies, rq->next_balance)) | |
4015 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4016 | } |
dd41f596 IM |
4017 | |
4018 | #else /* CONFIG_SMP */ | |
4019 | ||
1da177e4 LT |
4020 | /* |
4021 | * on UP we do not need to balance between CPUs: | |
4022 | */ | |
70b97a7f | 4023 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4024 | { |
4025 | } | |
dd41f596 | 4026 | |
1da177e4 LT |
4027 | #endif |
4028 | ||
1da177e4 LT |
4029 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4030 | ||
4031 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4032 | ||
4033 | /* | |
41b86e9c IM |
4034 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4035 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4036 | */ |
41b86e9c | 4037 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4038 | { |
1da177e4 | 4039 | unsigned long flags; |
41b86e9c IM |
4040 | u64 ns, delta_exec; |
4041 | struct rq *rq; | |
48f24c4d | 4042 | |
41b86e9c IM |
4043 | rq = task_rq_lock(p, &flags); |
4044 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4045 | if (task_current(rq, p)) { |
a8e504d2 IM |
4046 | update_rq_clock(rq); |
4047 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4048 | if ((s64)delta_exec > 0) |
4049 | ns += delta_exec; | |
4050 | } | |
4051 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4052 | |
1da177e4 LT |
4053 | return ns; |
4054 | } | |
4055 | ||
1da177e4 LT |
4056 | /* |
4057 | * Account user cpu time to a process. | |
4058 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4059 | * @cputime: the cpu time spent in user space since the last update |
4060 | */ | |
4061 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4062 | { | |
4063 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4064 | cputime64_t tmp; | |
4065 | ||
4066 | p->utime = cputime_add(p->utime, cputime); | |
4067 | ||
4068 | /* Add user time to cpustat. */ | |
4069 | tmp = cputime_to_cputime64(cputime); | |
4070 | if (TASK_NICE(p) > 0) | |
4071 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4072 | else | |
4073 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4074 | /* Account for user time used */ |
4075 | acct_update_integrals(p); | |
1da177e4 LT |
4076 | } |
4077 | ||
94886b84 LV |
4078 | /* |
4079 | * Account guest cpu time to a process. | |
4080 | * @p: the process that the cpu time gets accounted to | |
4081 | * @cputime: the cpu time spent in virtual machine since the last update | |
4082 | */ | |
f7402e03 | 4083 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4084 | { |
4085 | cputime64_t tmp; | |
4086 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4087 | ||
4088 | tmp = cputime_to_cputime64(cputime); | |
4089 | ||
4090 | p->utime = cputime_add(p->utime, cputime); | |
4091 | p->gtime = cputime_add(p->gtime, cputime); | |
4092 | ||
4093 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4094 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4095 | } | |
4096 | ||
c66f08be MN |
4097 | /* |
4098 | * Account scaled user cpu time to a process. | |
4099 | * @p: the process that the cpu time gets accounted to | |
4100 | * @cputime: the cpu time spent in user space since the last update | |
4101 | */ | |
4102 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4103 | { | |
4104 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4105 | } | |
4106 | ||
1da177e4 LT |
4107 | /* |
4108 | * Account system cpu time to a process. | |
4109 | * @p: the process that the cpu time gets accounted to | |
4110 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4111 | * @cputime: the cpu time spent in kernel space since the last update | |
4112 | */ | |
4113 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4114 | cputime_t cputime) | |
4115 | { | |
4116 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4117 | struct rq *rq = this_rq(); |
1da177e4 LT |
4118 | cputime64_t tmp; |
4119 | ||
983ed7a6 HH |
4120 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4121 | account_guest_time(p, cputime); | |
4122 | return; | |
4123 | } | |
94886b84 | 4124 | |
1da177e4 LT |
4125 | p->stime = cputime_add(p->stime, cputime); |
4126 | ||
4127 | /* Add system time to cpustat. */ | |
4128 | tmp = cputime_to_cputime64(cputime); | |
4129 | if (hardirq_count() - hardirq_offset) | |
4130 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4131 | else if (softirq_count()) | |
4132 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4133 | else if (p != rq->idle) |
1da177e4 | 4134 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4135 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4136 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4137 | else | |
4138 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4139 | /* Account for system time used */ | |
4140 | acct_update_integrals(p); | |
1da177e4 LT |
4141 | } |
4142 | ||
c66f08be MN |
4143 | /* |
4144 | * Account scaled system cpu time to a process. | |
4145 | * @p: the process that the cpu time gets accounted to | |
4146 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4147 | * @cputime: the cpu time spent in kernel space since the last update | |
4148 | */ | |
4149 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4150 | { | |
4151 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4152 | } | |
4153 | ||
1da177e4 LT |
4154 | /* |
4155 | * Account for involuntary wait time. | |
4156 | * @p: the process from which the cpu time has been stolen | |
4157 | * @steal: the cpu time spent in involuntary wait | |
4158 | */ | |
4159 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4160 | { | |
4161 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4162 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4163 | struct rq *rq = this_rq(); |
1da177e4 LT |
4164 | |
4165 | if (p == rq->idle) { | |
4166 | p->stime = cputime_add(p->stime, steal); | |
4167 | if (atomic_read(&rq->nr_iowait) > 0) | |
4168 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4169 | else | |
4170 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4171 | } else |
1da177e4 LT |
4172 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4173 | } | |
4174 | ||
7835b98b CL |
4175 | /* |
4176 | * This function gets called by the timer code, with HZ frequency. | |
4177 | * We call it with interrupts disabled. | |
4178 | * | |
4179 | * It also gets called by the fork code, when changing the parent's | |
4180 | * timeslices. | |
4181 | */ | |
4182 | void scheduler_tick(void) | |
4183 | { | |
7835b98b CL |
4184 | int cpu = smp_processor_id(); |
4185 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4186 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4187 | |
4188 | sched_clock_tick(); | |
dd41f596 IM |
4189 | |
4190 | spin_lock(&rq->lock); | |
3e51f33f | 4191 | update_rq_clock(rq); |
f1a438d8 | 4192 | update_cpu_load(rq); |
fa85ae24 | 4193 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4194 | spin_unlock(&rq->lock); |
7835b98b | 4195 | |
e418e1c2 | 4196 | #ifdef CONFIG_SMP |
dd41f596 IM |
4197 | rq->idle_at_tick = idle_cpu(cpu); |
4198 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4199 | #endif |
1da177e4 LT |
4200 | } |
4201 | ||
6cd8a4bb SR |
4202 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4203 | defined(CONFIG_PREEMPT_TRACER)) | |
4204 | ||
4205 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4206 | { | |
4207 | if (in_lock_functions(addr)) { | |
4208 | addr = CALLER_ADDR2; | |
4209 | if (in_lock_functions(addr)) | |
4210 | addr = CALLER_ADDR3; | |
4211 | } | |
4212 | return addr; | |
4213 | } | |
1da177e4 | 4214 | |
43627582 | 4215 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4216 | { |
6cd8a4bb | 4217 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4218 | /* |
4219 | * Underflow? | |
4220 | */ | |
9a11b49a IM |
4221 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4222 | return; | |
6cd8a4bb | 4223 | #endif |
1da177e4 | 4224 | preempt_count() += val; |
6cd8a4bb | 4225 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4226 | /* |
4227 | * Spinlock count overflowing soon? | |
4228 | */ | |
33859f7f MOS |
4229 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4230 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4231 | #endif |
4232 | if (preempt_count() == val) | |
4233 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4234 | } |
4235 | EXPORT_SYMBOL(add_preempt_count); | |
4236 | ||
43627582 | 4237 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4238 | { |
6cd8a4bb | 4239 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4240 | /* |
4241 | * Underflow? | |
4242 | */ | |
9a11b49a IM |
4243 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4244 | return; | |
1da177e4 LT |
4245 | /* |
4246 | * Is the spinlock portion underflowing? | |
4247 | */ | |
9a11b49a IM |
4248 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4249 | !(preempt_count() & PREEMPT_MASK))) | |
4250 | return; | |
6cd8a4bb | 4251 | #endif |
9a11b49a | 4252 | |
6cd8a4bb SR |
4253 | if (preempt_count() == val) |
4254 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4255 | preempt_count() -= val; |
4256 | } | |
4257 | EXPORT_SYMBOL(sub_preempt_count); | |
4258 | ||
4259 | #endif | |
4260 | ||
4261 | /* | |
dd41f596 | 4262 | * Print scheduling while atomic bug: |
1da177e4 | 4263 | */ |
dd41f596 | 4264 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4265 | { |
838225b4 SS |
4266 | struct pt_regs *regs = get_irq_regs(); |
4267 | ||
4268 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4269 | prev->comm, prev->pid, preempt_count()); | |
4270 | ||
dd41f596 | 4271 | debug_show_held_locks(prev); |
e21f5b15 | 4272 | print_modules(); |
dd41f596 IM |
4273 | if (irqs_disabled()) |
4274 | print_irqtrace_events(prev); | |
838225b4 SS |
4275 | |
4276 | if (regs) | |
4277 | show_regs(regs); | |
4278 | else | |
4279 | dump_stack(); | |
dd41f596 | 4280 | } |
1da177e4 | 4281 | |
dd41f596 IM |
4282 | /* |
4283 | * Various schedule()-time debugging checks and statistics: | |
4284 | */ | |
4285 | static inline void schedule_debug(struct task_struct *prev) | |
4286 | { | |
1da177e4 | 4287 | /* |
41a2d6cf | 4288 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4289 | * schedule() atomically, we ignore that path for now. |
4290 | * Otherwise, whine if we are scheduling when we should not be. | |
4291 | */ | |
3f33a7ce | 4292 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4293 | __schedule_bug(prev); |
4294 | ||
1da177e4 LT |
4295 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4296 | ||
2d72376b | 4297 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4298 | #ifdef CONFIG_SCHEDSTATS |
4299 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4300 | schedstat_inc(this_rq(), bkl_count); |
4301 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4302 | } |
4303 | #endif | |
dd41f596 IM |
4304 | } |
4305 | ||
4306 | /* | |
4307 | * Pick up the highest-prio task: | |
4308 | */ | |
4309 | static inline struct task_struct * | |
ff95f3df | 4310 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4311 | { |
5522d5d5 | 4312 | const struct sched_class *class; |
dd41f596 | 4313 | struct task_struct *p; |
1da177e4 LT |
4314 | |
4315 | /* | |
dd41f596 IM |
4316 | * Optimization: we know that if all tasks are in |
4317 | * the fair class we can call that function directly: | |
1da177e4 | 4318 | */ |
dd41f596 | 4319 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4320 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4321 | if (likely(p)) |
4322 | return p; | |
1da177e4 LT |
4323 | } |
4324 | ||
dd41f596 IM |
4325 | class = sched_class_highest; |
4326 | for ( ; ; ) { | |
fb8d4724 | 4327 | p = class->pick_next_task(rq); |
dd41f596 IM |
4328 | if (p) |
4329 | return p; | |
4330 | /* | |
4331 | * Will never be NULL as the idle class always | |
4332 | * returns a non-NULL p: | |
4333 | */ | |
4334 | class = class->next; | |
4335 | } | |
4336 | } | |
1da177e4 | 4337 | |
dd41f596 IM |
4338 | /* |
4339 | * schedule() is the main scheduler function. | |
4340 | */ | |
4341 | asmlinkage void __sched schedule(void) | |
4342 | { | |
4343 | struct task_struct *prev, *next; | |
67ca7bde | 4344 | unsigned long *switch_count; |
dd41f596 | 4345 | struct rq *rq; |
31656519 | 4346 | int cpu; |
dd41f596 IM |
4347 | |
4348 | need_resched: | |
4349 | preempt_disable(); | |
4350 | cpu = smp_processor_id(); | |
4351 | rq = cpu_rq(cpu); | |
4352 | rcu_qsctr_inc(cpu); | |
4353 | prev = rq->curr; | |
4354 | switch_count = &prev->nivcsw; | |
4355 | ||
4356 | release_kernel_lock(prev); | |
4357 | need_resched_nonpreemptible: | |
4358 | ||
4359 | schedule_debug(prev); | |
1da177e4 | 4360 | |
31656519 | 4361 | if (sched_feat(HRTICK)) |
f333fdc9 | 4362 | hrtick_clear(rq); |
8f4d37ec | 4363 | |
1e819950 IM |
4364 | /* |
4365 | * Do the rq-clock update outside the rq lock: | |
4366 | */ | |
4367 | local_irq_disable(); | |
3e51f33f | 4368 | update_rq_clock(rq); |
1e819950 IM |
4369 | spin_lock(&rq->lock); |
4370 | clear_tsk_need_resched(prev); | |
1da177e4 | 4371 | |
1da177e4 | 4372 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4373 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4374 | prev->state = TASK_RUNNING; |
16882c1e | 4375 | else |
2e1cb74a | 4376 | deactivate_task(rq, prev, 1); |
dd41f596 | 4377 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4378 | } |
4379 | ||
9a897c5a SR |
4380 | #ifdef CONFIG_SMP |
4381 | if (prev->sched_class->pre_schedule) | |
4382 | prev->sched_class->pre_schedule(rq, prev); | |
4383 | #endif | |
f65eda4f | 4384 | |
dd41f596 | 4385 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4386 | idle_balance(cpu, rq); |
1da177e4 | 4387 | |
31ee529c | 4388 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4389 | next = pick_next_task(rq, prev); |
1da177e4 | 4390 | |
1da177e4 | 4391 | if (likely(prev != next)) { |
673a90a1 DS |
4392 | sched_info_switch(prev, next); |
4393 | ||
1da177e4 LT |
4394 | rq->nr_switches++; |
4395 | rq->curr = next; | |
4396 | ++*switch_count; | |
4397 | ||
dd41f596 | 4398 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4399 | /* |
4400 | * the context switch might have flipped the stack from under | |
4401 | * us, hence refresh the local variables. | |
4402 | */ | |
4403 | cpu = smp_processor_id(); | |
4404 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4405 | } else |
4406 | spin_unlock_irq(&rq->lock); | |
4407 | ||
8f4d37ec | 4408 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4409 | goto need_resched_nonpreemptible; |
8f4d37ec | 4410 | |
1da177e4 LT |
4411 | preempt_enable_no_resched(); |
4412 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4413 | goto need_resched; | |
4414 | } | |
1da177e4 LT |
4415 | EXPORT_SYMBOL(schedule); |
4416 | ||
4417 | #ifdef CONFIG_PREEMPT | |
4418 | /* | |
2ed6e34f | 4419 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4420 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4421 | * occur there and call schedule directly. |
4422 | */ | |
4423 | asmlinkage void __sched preempt_schedule(void) | |
4424 | { | |
4425 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4426 | |
1da177e4 LT |
4427 | /* |
4428 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4429 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4430 | */ |
beed33a8 | 4431 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4432 | return; |
4433 | ||
3a5c359a AK |
4434 | do { |
4435 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4436 | schedule(); |
3a5c359a | 4437 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4438 | |
3a5c359a AK |
4439 | /* |
4440 | * Check again in case we missed a preemption opportunity | |
4441 | * between schedule and now. | |
4442 | */ | |
4443 | barrier(); | |
4444 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4445 | } |
1da177e4 LT |
4446 | EXPORT_SYMBOL(preempt_schedule); |
4447 | ||
4448 | /* | |
2ed6e34f | 4449 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4450 | * off of irq context. |
4451 | * Note, that this is called and return with irqs disabled. This will | |
4452 | * protect us against recursive calling from irq. | |
4453 | */ | |
4454 | asmlinkage void __sched preempt_schedule_irq(void) | |
4455 | { | |
4456 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4457 | |
2ed6e34f | 4458 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4459 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4460 | ||
3a5c359a AK |
4461 | do { |
4462 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4463 | local_irq_enable(); |
4464 | schedule(); | |
4465 | local_irq_disable(); | |
3a5c359a | 4466 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4467 | |
3a5c359a AK |
4468 | /* |
4469 | * Check again in case we missed a preemption opportunity | |
4470 | * between schedule and now. | |
4471 | */ | |
4472 | barrier(); | |
4473 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4474 | } |
4475 | ||
4476 | #endif /* CONFIG_PREEMPT */ | |
4477 | ||
95cdf3b7 IM |
4478 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4479 | void *key) | |
1da177e4 | 4480 | { |
48f24c4d | 4481 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4482 | } |
1da177e4 LT |
4483 | EXPORT_SYMBOL(default_wake_function); |
4484 | ||
4485 | /* | |
41a2d6cf IM |
4486 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4487 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4488 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4489 | * | |
4490 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4491 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4492 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4493 | */ | |
4494 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4495 | int nr_exclusive, int sync, void *key) | |
4496 | { | |
2e45874c | 4497 | wait_queue_t *curr, *next; |
1da177e4 | 4498 | |
2e45874c | 4499 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4500 | unsigned flags = curr->flags; |
4501 | ||
1da177e4 | 4502 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4503 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4504 | break; |
4505 | } | |
4506 | } | |
4507 | ||
4508 | /** | |
4509 | * __wake_up - wake up threads blocked on a waitqueue. | |
4510 | * @q: the waitqueue | |
4511 | * @mode: which threads | |
4512 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4513 | * @key: is directly passed to the wakeup function |
1da177e4 | 4514 | */ |
7ad5b3a5 | 4515 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4516 | int nr_exclusive, void *key) |
1da177e4 LT |
4517 | { |
4518 | unsigned long flags; | |
4519 | ||
4520 | spin_lock_irqsave(&q->lock, flags); | |
4521 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4522 | spin_unlock_irqrestore(&q->lock, flags); | |
4523 | } | |
1da177e4 LT |
4524 | EXPORT_SYMBOL(__wake_up); |
4525 | ||
4526 | /* | |
4527 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4528 | */ | |
7ad5b3a5 | 4529 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4530 | { |
4531 | __wake_up_common(q, mode, 1, 0, NULL); | |
4532 | } | |
4533 | ||
4534 | /** | |
67be2dd1 | 4535 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4536 | * @q: the waitqueue |
4537 | * @mode: which threads | |
4538 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4539 | * | |
4540 | * The sync wakeup differs that the waker knows that it will schedule | |
4541 | * away soon, so while the target thread will be woken up, it will not | |
4542 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4543 | * with each other. This can prevent needless bouncing between CPUs. | |
4544 | * | |
4545 | * On UP it can prevent extra preemption. | |
4546 | */ | |
7ad5b3a5 | 4547 | void |
95cdf3b7 | 4548 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4549 | { |
4550 | unsigned long flags; | |
4551 | int sync = 1; | |
4552 | ||
4553 | if (unlikely(!q)) | |
4554 | return; | |
4555 | ||
4556 | if (unlikely(!nr_exclusive)) | |
4557 | sync = 0; | |
4558 | ||
4559 | spin_lock_irqsave(&q->lock, flags); | |
4560 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4561 | spin_unlock_irqrestore(&q->lock, flags); | |
4562 | } | |
4563 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4564 | ||
b15136e9 | 4565 | void complete(struct completion *x) |
1da177e4 LT |
4566 | { |
4567 | unsigned long flags; | |
4568 | ||
4569 | spin_lock_irqsave(&x->wait.lock, flags); | |
4570 | x->done++; | |
d9514f6c | 4571 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4572 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4573 | } | |
4574 | EXPORT_SYMBOL(complete); | |
4575 | ||
b15136e9 | 4576 | void complete_all(struct completion *x) |
1da177e4 LT |
4577 | { |
4578 | unsigned long flags; | |
4579 | ||
4580 | spin_lock_irqsave(&x->wait.lock, flags); | |
4581 | x->done += UINT_MAX/2; | |
d9514f6c | 4582 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4583 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4584 | } | |
4585 | EXPORT_SYMBOL(complete_all); | |
4586 | ||
8cbbe86d AK |
4587 | static inline long __sched |
4588 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4589 | { |
1da177e4 LT |
4590 | if (!x->done) { |
4591 | DECLARE_WAITQUEUE(wait, current); | |
4592 | ||
4593 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4594 | __add_wait_queue_tail(&x->wait, &wait); | |
4595 | do { | |
009e577e MW |
4596 | if ((state == TASK_INTERRUPTIBLE && |
4597 | signal_pending(current)) || | |
4598 | (state == TASK_KILLABLE && | |
4599 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4600 | timeout = -ERESTARTSYS; |
4601 | break; | |
8cbbe86d AK |
4602 | } |
4603 | __set_current_state(state); | |
1da177e4 LT |
4604 | spin_unlock_irq(&x->wait.lock); |
4605 | timeout = schedule_timeout(timeout); | |
4606 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4607 | } while (!x->done && timeout); |
1da177e4 | 4608 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4609 | if (!x->done) |
4610 | return timeout; | |
1da177e4 LT |
4611 | } |
4612 | x->done--; | |
ea71a546 | 4613 | return timeout ?: 1; |
1da177e4 | 4614 | } |
1da177e4 | 4615 | |
8cbbe86d AK |
4616 | static long __sched |
4617 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4618 | { |
1da177e4 LT |
4619 | might_sleep(); |
4620 | ||
4621 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4622 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4623 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4624 | return timeout; |
4625 | } | |
1da177e4 | 4626 | |
b15136e9 | 4627 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4628 | { |
4629 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4630 | } |
8cbbe86d | 4631 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4632 | |
b15136e9 | 4633 | unsigned long __sched |
8cbbe86d | 4634 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4635 | { |
8cbbe86d | 4636 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4637 | } |
8cbbe86d | 4638 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4639 | |
8cbbe86d | 4640 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4641 | { |
51e97990 AK |
4642 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4643 | if (t == -ERESTARTSYS) | |
4644 | return t; | |
4645 | return 0; | |
0fec171c | 4646 | } |
8cbbe86d | 4647 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4648 | |
b15136e9 | 4649 | unsigned long __sched |
8cbbe86d AK |
4650 | wait_for_completion_interruptible_timeout(struct completion *x, |
4651 | unsigned long timeout) | |
0fec171c | 4652 | { |
8cbbe86d | 4653 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4654 | } |
8cbbe86d | 4655 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4656 | |
009e577e MW |
4657 | int __sched wait_for_completion_killable(struct completion *x) |
4658 | { | |
4659 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4660 | if (t == -ERESTARTSYS) | |
4661 | return t; | |
4662 | return 0; | |
4663 | } | |
4664 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4665 | ||
8cbbe86d AK |
4666 | static long __sched |
4667 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4668 | { |
0fec171c IM |
4669 | unsigned long flags; |
4670 | wait_queue_t wait; | |
4671 | ||
4672 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4673 | |
8cbbe86d | 4674 | __set_current_state(state); |
1da177e4 | 4675 | |
8cbbe86d AK |
4676 | spin_lock_irqsave(&q->lock, flags); |
4677 | __add_wait_queue(q, &wait); | |
4678 | spin_unlock(&q->lock); | |
4679 | timeout = schedule_timeout(timeout); | |
4680 | spin_lock_irq(&q->lock); | |
4681 | __remove_wait_queue(q, &wait); | |
4682 | spin_unlock_irqrestore(&q->lock, flags); | |
4683 | ||
4684 | return timeout; | |
4685 | } | |
4686 | ||
4687 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4688 | { | |
4689 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4690 | } |
1da177e4 LT |
4691 | EXPORT_SYMBOL(interruptible_sleep_on); |
4692 | ||
0fec171c | 4693 | long __sched |
95cdf3b7 | 4694 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4695 | { |
8cbbe86d | 4696 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4697 | } |
1da177e4 LT |
4698 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4699 | ||
0fec171c | 4700 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4701 | { |
8cbbe86d | 4702 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4703 | } |
1da177e4 LT |
4704 | EXPORT_SYMBOL(sleep_on); |
4705 | ||
0fec171c | 4706 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4707 | { |
8cbbe86d | 4708 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4709 | } |
1da177e4 LT |
4710 | EXPORT_SYMBOL(sleep_on_timeout); |
4711 | ||
b29739f9 IM |
4712 | #ifdef CONFIG_RT_MUTEXES |
4713 | ||
4714 | /* | |
4715 | * rt_mutex_setprio - set the current priority of a task | |
4716 | * @p: task | |
4717 | * @prio: prio value (kernel-internal form) | |
4718 | * | |
4719 | * This function changes the 'effective' priority of a task. It does | |
4720 | * not touch ->normal_prio like __setscheduler(). | |
4721 | * | |
4722 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4723 | */ | |
36c8b586 | 4724 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4725 | { |
4726 | unsigned long flags; | |
83b699ed | 4727 | int oldprio, on_rq, running; |
70b97a7f | 4728 | struct rq *rq; |
cb469845 | 4729 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4730 | |
4731 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4732 | ||
4733 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4734 | update_rq_clock(rq); |
b29739f9 | 4735 | |
d5f9f942 | 4736 | oldprio = p->prio; |
dd41f596 | 4737 | on_rq = p->se.on_rq; |
051a1d1a | 4738 | running = task_current(rq, p); |
0e1f3483 | 4739 | if (on_rq) |
69be72c1 | 4740 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4741 | if (running) |
4742 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4743 | |
4744 | if (rt_prio(prio)) | |
4745 | p->sched_class = &rt_sched_class; | |
4746 | else | |
4747 | p->sched_class = &fair_sched_class; | |
4748 | ||
b29739f9 IM |
4749 | p->prio = prio; |
4750 | ||
0e1f3483 HS |
4751 | if (running) |
4752 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4753 | if (on_rq) { |
8159f87e | 4754 | enqueue_task(rq, p, 0); |
cb469845 SR |
4755 | |
4756 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4757 | } |
4758 | task_rq_unlock(rq, &flags); | |
4759 | } | |
4760 | ||
4761 | #endif | |
4762 | ||
36c8b586 | 4763 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4764 | { |
dd41f596 | 4765 | int old_prio, delta, on_rq; |
1da177e4 | 4766 | unsigned long flags; |
70b97a7f | 4767 | struct rq *rq; |
1da177e4 LT |
4768 | |
4769 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4770 | return; | |
4771 | /* | |
4772 | * We have to be careful, if called from sys_setpriority(), | |
4773 | * the task might be in the middle of scheduling on another CPU. | |
4774 | */ | |
4775 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4776 | update_rq_clock(rq); |
1da177e4 LT |
4777 | /* |
4778 | * The RT priorities are set via sched_setscheduler(), but we still | |
4779 | * allow the 'normal' nice value to be set - but as expected | |
4780 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4781 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4782 | */ |
e05606d3 | 4783 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4784 | p->static_prio = NICE_TO_PRIO(nice); |
4785 | goto out_unlock; | |
4786 | } | |
dd41f596 | 4787 | on_rq = p->se.on_rq; |
c09595f6 | 4788 | if (on_rq) |
69be72c1 | 4789 | dequeue_task(rq, p, 0); |
1da177e4 | 4790 | |
1da177e4 | 4791 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4792 | set_load_weight(p); |
b29739f9 IM |
4793 | old_prio = p->prio; |
4794 | p->prio = effective_prio(p); | |
4795 | delta = p->prio - old_prio; | |
1da177e4 | 4796 | |
dd41f596 | 4797 | if (on_rq) { |
8159f87e | 4798 | enqueue_task(rq, p, 0); |
1da177e4 | 4799 | /* |
d5f9f942 AM |
4800 | * If the task increased its priority or is running and |
4801 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4802 | */ |
d5f9f942 | 4803 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4804 | resched_task(rq->curr); |
4805 | } | |
4806 | out_unlock: | |
4807 | task_rq_unlock(rq, &flags); | |
4808 | } | |
1da177e4 LT |
4809 | EXPORT_SYMBOL(set_user_nice); |
4810 | ||
e43379f1 MM |
4811 | /* |
4812 | * can_nice - check if a task can reduce its nice value | |
4813 | * @p: task | |
4814 | * @nice: nice value | |
4815 | */ | |
36c8b586 | 4816 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4817 | { |
024f4747 MM |
4818 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4819 | int nice_rlim = 20 - nice; | |
48f24c4d | 4820 | |
e43379f1 MM |
4821 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4822 | capable(CAP_SYS_NICE)); | |
4823 | } | |
4824 | ||
1da177e4 LT |
4825 | #ifdef __ARCH_WANT_SYS_NICE |
4826 | ||
4827 | /* | |
4828 | * sys_nice - change the priority of the current process. | |
4829 | * @increment: priority increment | |
4830 | * | |
4831 | * sys_setpriority is a more generic, but much slower function that | |
4832 | * does similar things. | |
4833 | */ | |
4834 | asmlinkage long sys_nice(int increment) | |
4835 | { | |
48f24c4d | 4836 | long nice, retval; |
1da177e4 LT |
4837 | |
4838 | /* | |
4839 | * Setpriority might change our priority at the same moment. | |
4840 | * We don't have to worry. Conceptually one call occurs first | |
4841 | * and we have a single winner. | |
4842 | */ | |
e43379f1 MM |
4843 | if (increment < -40) |
4844 | increment = -40; | |
1da177e4 LT |
4845 | if (increment > 40) |
4846 | increment = 40; | |
4847 | ||
4848 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4849 | if (nice < -20) | |
4850 | nice = -20; | |
4851 | if (nice > 19) | |
4852 | nice = 19; | |
4853 | ||
e43379f1 MM |
4854 | if (increment < 0 && !can_nice(current, nice)) |
4855 | return -EPERM; | |
4856 | ||
1da177e4 LT |
4857 | retval = security_task_setnice(current, nice); |
4858 | if (retval) | |
4859 | return retval; | |
4860 | ||
4861 | set_user_nice(current, nice); | |
4862 | return 0; | |
4863 | } | |
4864 | ||
4865 | #endif | |
4866 | ||
4867 | /** | |
4868 | * task_prio - return the priority value of a given task. | |
4869 | * @p: the task in question. | |
4870 | * | |
4871 | * This is the priority value as seen by users in /proc. | |
4872 | * RT tasks are offset by -200. Normal tasks are centered | |
4873 | * around 0, value goes from -16 to +15. | |
4874 | */ | |
36c8b586 | 4875 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4876 | { |
4877 | return p->prio - MAX_RT_PRIO; | |
4878 | } | |
4879 | ||
4880 | /** | |
4881 | * task_nice - return the nice value of a given task. | |
4882 | * @p: the task in question. | |
4883 | */ | |
36c8b586 | 4884 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4885 | { |
4886 | return TASK_NICE(p); | |
4887 | } | |
150d8bed | 4888 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4889 | |
4890 | /** | |
4891 | * idle_cpu - is a given cpu idle currently? | |
4892 | * @cpu: the processor in question. | |
4893 | */ | |
4894 | int idle_cpu(int cpu) | |
4895 | { | |
4896 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4897 | } | |
4898 | ||
1da177e4 LT |
4899 | /** |
4900 | * idle_task - return the idle task for a given cpu. | |
4901 | * @cpu: the processor in question. | |
4902 | */ | |
36c8b586 | 4903 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4904 | { |
4905 | return cpu_rq(cpu)->idle; | |
4906 | } | |
4907 | ||
4908 | /** | |
4909 | * find_process_by_pid - find a process with a matching PID value. | |
4910 | * @pid: the pid in question. | |
4911 | */ | |
a9957449 | 4912 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4913 | { |
228ebcbe | 4914 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4915 | } |
4916 | ||
4917 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4918 | static void |
4919 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4920 | { |
dd41f596 | 4921 | BUG_ON(p->se.on_rq); |
48f24c4d | 4922 | |
1da177e4 | 4923 | p->policy = policy; |
dd41f596 IM |
4924 | switch (p->policy) { |
4925 | case SCHED_NORMAL: | |
4926 | case SCHED_BATCH: | |
4927 | case SCHED_IDLE: | |
4928 | p->sched_class = &fair_sched_class; | |
4929 | break; | |
4930 | case SCHED_FIFO: | |
4931 | case SCHED_RR: | |
4932 | p->sched_class = &rt_sched_class; | |
4933 | break; | |
4934 | } | |
4935 | ||
1da177e4 | 4936 | p->rt_priority = prio; |
b29739f9 IM |
4937 | p->normal_prio = normal_prio(p); |
4938 | /* we are holding p->pi_lock already */ | |
4939 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4940 | set_load_weight(p); |
1da177e4 LT |
4941 | } |
4942 | ||
961ccddd RR |
4943 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4944 | struct sched_param *param, bool user) | |
1da177e4 | 4945 | { |
83b699ed | 4946 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4947 | unsigned long flags; |
cb469845 | 4948 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4949 | struct rq *rq; |
1da177e4 | 4950 | |
66e5393a SR |
4951 | /* may grab non-irq protected spin_locks */ |
4952 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4953 | recheck: |
4954 | /* double check policy once rq lock held */ | |
4955 | if (policy < 0) | |
4956 | policy = oldpolicy = p->policy; | |
4957 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4958 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4959 | policy != SCHED_IDLE) | |
b0a9499c | 4960 | return -EINVAL; |
1da177e4 LT |
4961 | /* |
4962 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4963 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4964 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4965 | */ |
4966 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4967 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4968 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4969 | return -EINVAL; |
e05606d3 | 4970 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4971 | return -EINVAL; |
4972 | ||
37e4ab3f OC |
4973 | /* |
4974 | * Allow unprivileged RT tasks to decrease priority: | |
4975 | */ | |
961ccddd | 4976 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4977 | if (rt_policy(policy)) { |
8dc3e909 | 4978 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4979 | |
4980 | if (!lock_task_sighand(p, &flags)) | |
4981 | return -ESRCH; | |
4982 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4983 | unlock_task_sighand(p, &flags); | |
4984 | ||
4985 | /* can't set/change the rt policy */ | |
4986 | if (policy != p->policy && !rlim_rtprio) | |
4987 | return -EPERM; | |
4988 | ||
4989 | /* can't increase priority */ | |
4990 | if (param->sched_priority > p->rt_priority && | |
4991 | param->sched_priority > rlim_rtprio) | |
4992 | return -EPERM; | |
4993 | } | |
dd41f596 IM |
4994 | /* |
4995 | * Like positive nice levels, dont allow tasks to | |
4996 | * move out of SCHED_IDLE either: | |
4997 | */ | |
4998 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4999 | return -EPERM; | |
5fe1d75f | 5000 | |
37e4ab3f OC |
5001 | /* can't change other user's priorities */ |
5002 | if ((current->euid != p->euid) && | |
5003 | (current->euid != p->uid)) | |
5004 | return -EPERM; | |
5005 | } | |
1da177e4 | 5006 | |
725aad24 | 5007 | if (user) { |
b68aa230 | 5008 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5009 | /* |
5010 | * Do not allow realtime tasks into groups that have no runtime | |
5011 | * assigned. | |
5012 | */ | |
5013 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) | |
5014 | return -EPERM; | |
b68aa230 PZ |
5015 | #endif |
5016 | ||
725aad24 JF |
5017 | retval = security_task_setscheduler(p, policy, param); |
5018 | if (retval) | |
5019 | return retval; | |
5020 | } | |
5021 | ||
b29739f9 IM |
5022 | /* |
5023 | * make sure no PI-waiters arrive (or leave) while we are | |
5024 | * changing the priority of the task: | |
5025 | */ | |
5026 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5027 | /* |
5028 | * To be able to change p->policy safely, the apropriate | |
5029 | * runqueue lock must be held. | |
5030 | */ | |
b29739f9 | 5031 | rq = __task_rq_lock(p); |
1da177e4 LT |
5032 | /* recheck policy now with rq lock held */ |
5033 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5034 | policy = oldpolicy = -1; | |
b29739f9 IM |
5035 | __task_rq_unlock(rq); |
5036 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5037 | goto recheck; |
5038 | } | |
2daa3577 | 5039 | update_rq_clock(rq); |
dd41f596 | 5040 | on_rq = p->se.on_rq; |
051a1d1a | 5041 | running = task_current(rq, p); |
0e1f3483 | 5042 | if (on_rq) |
2e1cb74a | 5043 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5044 | if (running) |
5045 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5046 | |
1da177e4 | 5047 | oldprio = p->prio; |
dd41f596 | 5048 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5049 | |
0e1f3483 HS |
5050 | if (running) |
5051 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5052 | if (on_rq) { |
5053 | activate_task(rq, p, 0); | |
cb469845 SR |
5054 | |
5055 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5056 | } |
b29739f9 IM |
5057 | __task_rq_unlock(rq); |
5058 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5059 | ||
95e02ca9 TG |
5060 | rt_mutex_adjust_pi(p); |
5061 | ||
1da177e4 LT |
5062 | return 0; |
5063 | } | |
961ccddd RR |
5064 | |
5065 | /** | |
5066 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5067 | * @p: the task in question. | |
5068 | * @policy: new policy. | |
5069 | * @param: structure containing the new RT priority. | |
5070 | * | |
5071 | * NOTE that the task may be already dead. | |
5072 | */ | |
5073 | int sched_setscheduler(struct task_struct *p, int policy, | |
5074 | struct sched_param *param) | |
5075 | { | |
5076 | return __sched_setscheduler(p, policy, param, true); | |
5077 | } | |
1da177e4 LT |
5078 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5079 | ||
961ccddd RR |
5080 | /** |
5081 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5082 | * @p: the task in question. | |
5083 | * @policy: new policy. | |
5084 | * @param: structure containing the new RT priority. | |
5085 | * | |
5086 | * Just like sched_setscheduler, only don't bother checking if the | |
5087 | * current context has permission. For example, this is needed in | |
5088 | * stop_machine(): we create temporary high priority worker threads, | |
5089 | * but our caller might not have that capability. | |
5090 | */ | |
5091 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5092 | struct sched_param *param) | |
5093 | { | |
5094 | return __sched_setscheduler(p, policy, param, false); | |
5095 | } | |
5096 | ||
95cdf3b7 IM |
5097 | static int |
5098 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5099 | { |
1da177e4 LT |
5100 | struct sched_param lparam; |
5101 | struct task_struct *p; | |
36c8b586 | 5102 | int retval; |
1da177e4 LT |
5103 | |
5104 | if (!param || pid < 0) | |
5105 | return -EINVAL; | |
5106 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5107 | return -EFAULT; | |
5fe1d75f ON |
5108 | |
5109 | rcu_read_lock(); | |
5110 | retval = -ESRCH; | |
1da177e4 | 5111 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5112 | if (p != NULL) |
5113 | retval = sched_setscheduler(p, policy, &lparam); | |
5114 | rcu_read_unlock(); | |
36c8b586 | 5115 | |
1da177e4 LT |
5116 | return retval; |
5117 | } | |
5118 | ||
5119 | /** | |
5120 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5121 | * @pid: the pid in question. | |
5122 | * @policy: new policy. | |
5123 | * @param: structure containing the new RT priority. | |
5124 | */ | |
41a2d6cf IM |
5125 | asmlinkage long |
5126 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5127 | { |
c21761f1 JB |
5128 | /* negative values for policy are not valid */ |
5129 | if (policy < 0) | |
5130 | return -EINVAL; | |
5131 | ||
1da177e4 LT |
5132 | return do_sched_setscheduler(pid, policy, param); |
5133 | } | |
5134 | ||
5135 | /** | |
5136 | * sys_sched_setparam - set/change the RT priority of a thread | |
5137 | * @pid: the pid in question. | |
5138 | * @param: structure containing the new RT priority. | |
5139 | */ | |
5140 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5141 | { | |
5142 | return do_sched_setscheduler(pid, -1, param); | |
5143 | } | |
5144 | ||
5145 | /** | |
5146 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5147 | * @pid: the pid in question. | |
5148 | */ | |
5149 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5150 | { | |
36c8b586 | 5151 | struct task_struct *p; |
3a5c359a | 5152 | int retval; |
1da177e4 LT |
5153 | |
5154 | if (pid < 0) | |
3a5c359a | 5155 | return -EINVAL; |
1da177e4 LT |
5156 | |
5157 | retval = -ESRCH; | |
5158 | read_lock(&tasklist_lock); | |
5159 | p = find_process_by_pid(pid); | |
5160 | if (p) { | |
5161 | retval = security_task_getscheduler(p); | |
5162 | if (!retval) | |
5163 | retval = p->policy; | |
5164 | } | |
5165 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5166 | return retval; |
5167 | } | |
5168 | ||
5169 | /** | |
5170 | * sys_sched_getscheduler - get the RT priority of a thread | |
5171 | * @pid: the pid in question. | |
5172 | * @param: structure containing the RT priority. | |
5173 | */ | |
5174 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5175 | { | |
5176 | struct sched_param lp; | |
36c8b586 | 5177 | struct task_struct *p; |
3a5c359a | 5178 | int retval; |
1da177e4 LT |
5179 | |
5180 | if (!param || pid < 0) | |
3a5c359a | 5181 | return -EINVAL; |
1da177e4 LT |
5182 | |
5183 | read_lock(&tasklist_lock); | |
5184 | p = find_process_by_pid(pid); | |
5185 | retval = -ESRCH; | |
5186 | if (!p) | |
5187 | goto out_unlock; | |
5188 | ||
5189 | retval = security_task_getscheduler(p); | |
5190 | if (retval) | |
5191 | goto out_unlock; | |
5192 | ||
5193 | lp.sched_priority = p->rt_priority; | |
5194 | read_unlock(&tasklist_lock); | |
5195 | ||
5196 | /* | |
5197 | * This one might sleep, we cannot do it with a spinlock held ... | |
5198 | */ | |
5199 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5200 | ||
1da177e4 LT |
5201 | return retval; |
5202 | ||
5203 | out_unlock: | |
5204 | read_unlock(&tasklist_lock); | |
5205 | return retval; | |
5206 | } | |
5207 | ||
b53e921b | 5208 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5209 | { |
1da177e4 | 5210 | cpumask_t cpus_allowed; |
b53e921b | 5211 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5212 | struct task_struct *p; |
5213 | int retval; | |
1da177e4 | 5214 | |
95402b38 | 5215 | get_online_cpus(); |
1da177e4 LT |
5216 | read_lock(&tasklist_lock); |
5217 | ||
5218 | p = find_process_by_pid(pid); | |
5219 | if (!p) { | |
5220 | read_unlock(&tasklist_lock); | |
95402b38 | 5221 | put_online_cpus(); |
1da177e4 LT |
5222 | return -ESRCH; |
5223 | } | |
5224 | ||
5225 | /* | |
5226 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5227 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5228 | * usage count and then drop tasklist_lock. |
5229 | */ | |
5230 | get_task_struct(p); | |
5231 | read_unlock(&tasklist_lock); | |
5232 | ||
5233 | retval = -EPERM; | |
5234 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5235 | !capable(CAP_SYS_NICE)) | |
5236 | goto out_unlock; | |
5237 | ||
e7834f8f DQ |
5238 | retval = security_task_setscheduler(p, 0, NULL); |
5239 | if (retval) | |
5240 | goto out_unlock; | |
5241 | ||
f9a86fcb | 5242 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5243 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5244 | again: |
7c16ec58 | 5245 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5246 | |
8707d8b8 | 5247 | if (!retval) { |
f9a86fcb | 5248 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5249 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5250 | /* | |
5251 | * We must have raced with a concurrent cpuset | |
5252 | * update. Just reset the cpus_allowed to the | |
5253 | * cpuset's cpus_allowed | |
5254 | */ | |
5255 | new_mask = cpus_allowed; | |
5256 | goto again; | |
5257 | } | |
5258 | } | |
1da177e4 LT |
5259 | out_unlock: |
5260 | put_task_struct(p); | |
95402b38 | 5261 | put_online_cpus(); |
1da177e4 LT |
5262 | return retval; |
5263 | } | |
5264 | ||
5265 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5266 | cpumask_t *new_mask) | |
5267 | { | |
5268 | if (len < sizeof(cpumask_t)) { | |
5269 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5270 | } else if (len > sizeof(cpumask_t)) { | |
5271 | len = sizeof(cpumask_t); | |
5272 | } | |
5273 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5274 | } | |
5275 | ||
5276 | /** | |
5277 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5278 | * @pid: pid of the process | |
5279 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5280 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5281 | */ | |
5282 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5283 | unsigned long __user *user_mask_ptr) | |
5284 | { | |
5285 | cpumask_t new_mask; | |
5286 | int retval; | |
5287 | ||
5288 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5289 | if (retval) | |
5290 | return retval; | |
5291 | ||
b53e921b | 5292 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5293 | } |
5294 | ||
1da177e4 LT |
5295 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5296 | { | |
36c8b586 | 5297 | struct task_struct *p; |
1da177e4 | 5298 | int retval; |
1da177e4 | 5299 | |
95402b38 | 5300 | get_online_cpus(); |
1da177e4 LT |
5301 | read_lock(&tasklist_lock); |
5302 | ||
5303 | retval = -ESRCH; | |
5304 | p = find_process_by_pid(pid); | |
5305 | if (!p) | |
5306 | goto out_unlock; | |
5307 | ||
e7834f8f DQ |
5308 | retval = security_task_getscheduler(p); |
5309 | if (retval) | |
5310 | goto out_unlock; | |
5311 | ||
2f7016d9 | 5312 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5313 | |
5314 | out_unlock: | |
5315 | read_unlock(&tasklist_lock); | |
95402b38 | 5316 | put_online_cpus(); |
1da177e4 | 5317 | |
9531b62f | 5318 | return retval; |
1da177e4 LT |
5319 | } |
5320 | ||
5321 | /** | |
5322 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5323 | * @pid: pid of the process | |
5324 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5325 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5326 | */ | |
5327 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5328 | unsigned long __user *user_mask_ptr) | |
5329 | { | |
5330 | int ret; | |
5331 | cpumask_t mask; | |
5332 | ||
5333 | if (len < sizeof(cpumask_t)) | |
5334 | return -EINVAL; | |
5335 | ||
5336 | ret = sched_getaffinity(pid, &mask); | |
5337 | if (ret < 0) | |
5338 | return ret; | |
5339 | ||
5340 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5341 | return -EFAULT; | |
5342 | ||
5343 | return sizeof(cpumask_t); | |
5344 | } | |
5345 | ||
5346 | /** | |
5347 | * sys_sched_yield - yield the current processor to other threads. | |
5348 | * | |
dd41f596 IM |
5349 | * This function yields the current CPU to other tasks. If there are no |
5350 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5351 | */ |
5352 | asmlinkage long sys_sched_yield(void) | |
5353 | { | |
70b97a7f | 5354 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5355 | |
2d72376b | 5356 | schedstat_inc(rq, yld_count); |
4530d7ab | 5357 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5358 | |
5359 | /* | |
5360 | * Since we are going to call schedule() anyway, there's | |
5361 | * no need to preempt or enable interrupts: | |
5362 | */ | |
5363 | __release(rq->lock); | |
8a25d5de | 5364 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5365 | _raw_spin_unlock(&rq->lock); |
5366 | preempt_enable_no_resched(); | |
5367 | ||
5368 | schedule(); | |
5369 | ||
5370 | return 0; | |
5371 | } | |
5372 | ||
e7b38404 | 5373 | static void __cond_resched(void) |
1da177e4 | 5374 | { |
8e0a43d8 IM |
5375 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5376 | __might_sleep(__FILE__, __LINE__); | |
5377 | #endif | |
5bbcfd90 IM |
5378 | /* |
5379 | * The BKS might be reacquired before we have dropped | |
5380 | * PREEMPT_ACTIVE, which could trigger a second | |
5381 | * cond_resched() call. | |
5382 | */ | |
1da177e4 LT |
5383 | do { |
5384 | add_preempt_count(PREEMPT_ACTIVE); | |
5385 | schedule(); | |
5386 | sub_preempt_count(PREEMPT_ACTIVE); | |
5387 | } while (need_resched()); | |
5388 | } | |
5389 | ||
02b67cc3 | 5390 | int __sched _cond_resched(void) |
1da177e4 | 5391 | { |
9414232f IM |
5392 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5393 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5394 | __cond_resched(); |
5395 | return 1; | |
5396 | } | |
5397 | return 0; | |
5398 | } | |
02b67cc3 | 5399 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5400 | |
5401 | /* | |
5402 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5403 | * call schedule, and on return reacquire the lock. | |
5404 | * | |
41a2d6cf | 5405 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5406 | * operations here to prevent schedule() from being called twice (once via |
5407 | * spin_unlock(), once by hand). | |
5408 | */ | |
95cdf3b7 | 5409 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5410 | { |
95c354fe | 5411 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5412 | int ret = 0; |
5413 | ||
95c354fe | 5414 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5415 | spin_unlock(lock); |
95c354fe NP |
5416 | if (resched && need_resched()) |
5417 | __cond_resched(); | |
5418 | else | |
5419 | cpu_relax(); | |
6df3cecb | 5420 | ret = 1; |
1da177e4 | 5421 | spin_lock(lock); |
1da177e4 | 5422 | } |
6df3cecb | 5423 | return ret; |
1da177e4 | 5424 | } |
1da177e4 LT |
5425 | EXPORT_SYMBOL(cond_resched_lock); |
5426 | ||
5427 | int __sched cond_resched_softirq(void) | |
5428 | { | |
5429 | BUG_ON(!in_softirq()); | |
5430 | ||
9414232f | 5431 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5432 | local_bh_enable(); |
1da177e4 LT |
5433 | __cond_resched(); |
5434 | local_bh_disable(); | |
5435 | return 1; | |
5436 | } | |
5437 | return 0; | |
5438 | } | |
1da177e4 LT |
5439 | EXPORT_SYMBOL(cond_resched_softirq); |
5440 | ||
1da177e4 LT |
5441 | /** |
5442 | * yield - yield the current processor to other threads. | |
5443 | * | |
72fd4a35 | 5444 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5445 | * thread runnable and calls sys_sched_yield(). |
5446 | */ | |
5447 | void __sched yield(void) | |
5448 | { | |
5449 | set_current_state(TASK_RUNNING); | |
5450 | sys_sched_yield(); | |
5451 | } | |
1da177e4 LT |
5452 | EXPORT_SYMBOL(yield); |
5453 | ||
5454 | /* | |
41a2d6cf | 5455 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5456 | * that process accounting knows that this is a task in IO wait state. |
5457 | * | |
5458 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5459 | * has set its backing_dev_info: the queue against which it should throttle) | |
5460 | */ | |
5461 | void __sched io_schedule(void) | |
5462 | { | |
70b97a7f | 5463 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5464 | |
0ff92245 | 5465 | delayacct_blkio_start(); |
1da177e4 LT |
5466 | atomic_inc(&rq->nr_iowait); |
5467 | schedule(); | |
5468 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5469 | delayacct_blkio_end(); |
1da177e4 | 5470 | } |
1da177e4 LT |
5471 | EXPORT_SYMBOL(io_schedule); |
5472 | ||
5473 | long __sched io_schedule_timeout(long timeout) | |
5474 | { | |
70b97a7f | 5475 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5476 | long ret; |
5477 | ||
0ff92245 | 5478 | delayacct_blkio_start(); |
1da177e4 LT |
5479 | atomic_inc(&rq->nr_iowait); |
5480 | ret = schedule_timeout(timeout); | |
5481 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5482 | delayacct_blkio_end(); |
1da177e4 LT |
5483 | return ret; |
5484 | } | |
5485 | ||
5486 | /** | |
5487 | * sys_sched_get_priority_max - return maximum RT priority. | |
5488 | * @policy: scheduling class. | |
5489 | * | |
5490 | * this syscall returns the maximum rt_priority that can be used | |
5491 | * by a given scheduling class. | |
5492 | */ | |
5493 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5494 | { | |
5495 | int ret = -EINVAL; | |
5496 | ||
5497 | switch (policy) { | |
5498 | case SCHED_FIFO: | |
5499 | case SCHED_RR: | |
5500 | ret = MAX_USER_RT_PRIO-1; | |
5501 | break; | |
5502 | case SCHED_NORMAL: | |
b0a9499c | 5503 | case SCHED_BATCH: |
dd41f596 | 5504 | case SCHED_IDLE: |
1da177e4 LT |
5505 | ret = 0; |
5506 | break; | |
5507 | } | |
5508 | return ret; | |
5509 | } | |
5510 | ||
5511 | /** | |
5512 | * sys_sched_get_priority_min - return minimum RT priority. | |
5513 | * @policy: scheduling class. | |
5514 | * | |
5515 | * this syscall returns the minimum rt_priority that can be used | |
5516 | * by a given scheduling class. | |
5517 | */ | |
5518 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5519 | { | |
5520 | int ret = -EINVAL; | |
5521 | ||
5522 | switch (policy) { | |
5523 | case SCHED_FIFO: | |
5524 | case SCHED_RR: | |
5525 | ret = 1; | |
5526 | break; | |
5527 | case SCHED_NORMAL: | |
b0a9499c | 5528 | case SCHED_BATCH: |
dd41f596 | 5529 | case SCHED_IDLE: |
1da177e4 LT |
5530 | ret = 0; |
5531 | } | |
5532 | return ret; | |
5533 | } | |
5534 | ||
5535 | /** | |
5536 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5537 | * @pid: pid of the process. | |
5538 | * @interval: userspace pointer to the timeslice value. | |
5539 | * | |
5540 | * this syscall writes the default timeslice value of a given process | |
5541 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5542 | */ | |
5543 | asmlinkage | |
5544 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5545 | { | |
36c8b586 | 5546 | struct task_struct *p; |
a4ec24b4 | 5547 | unsigned int time_slice; |
3a5c359a | 5548 | int retval; |
1da177e4 | 5549 | struct timespec t; |
1da177e4 LT |
5550 | |
5551 | if (pid < 0) | |
3a5c359a | 5552 | return -EINVAL; |
1da177e4 LT |
5553 | |
5554 | retval = -ESRCH; | |
5555 | read_lock(&tasklist_lock); | |
5556 | p = find_process_by_pid(pid); | |
5557 | if (!p) | |
5558 | goto out_unlock; | |
5559 | ||
5560 | retval = security_task_getscheduler(p); | |
5561 | if (retval) | |
5562 | goto out_unlock; | |
5563 | ||
77034937 IM |
5564 | /* |
5565 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5566 | * tasks that are on an otherwise idle runqueue: | |
5567 | */ | |
5568 | time_slice = 0; | |
5569 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5570 | time_slice = DEF_TIMESLICE; |
1868f958 | 5571 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5572 | struct sched_entity *se = &p->se; |
5573 | unsigned long flags; | |
5574 | struct rq *rq; | |
5575 | ||
5576 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5577 | if (rq->cfs.load.weight) |
5578 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5579 | task_rq_unlock(rq, &flags); |
5580 | } | |
1da177e4 | 5581 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5582 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5583 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5584 | return retval; |
3a5c359a | 5585 | |
1da177e4 LT |
5586 | out_unlock: |
5587 | read_unlock(&tasklist_lock); | |
5588 | return retval; | |
5589 | } | |
5590 | ||
7c731e0a | 5591 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5592 | |
82a1fcb9 | 5593 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5594 | { |
1da177e4 | 5595 | unsigned long free = 0; |
36c8b586 | 5596 | unsigned state; |
1da177e4 | 5597 | |
1da177e4 | 5598 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5599 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5600 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5601 | #if BITS_PER_LONG == 32 |
1da177e4 | 5602 | if (state == TASK_RUNNING) |
cc4ea795 | 5603 | printk(KERN_CONT " running "); |
1da177e4 | 5604 | else |
cc4ea795 | 5605 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5606 | #else |
5607 | if (state == TASK_RUNNING) | |
cc4ea795 | 5608 | printk(KERN_CONT " running task "); |
1da177e4 | 5609 | else |
cc4ea795 | 5610 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5611 | #endif |
5612 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5613 | { | |
10ebffde | 5614 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5615 | while (!*n) |
5616 | n++; | |
10ebffde | 5617 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5618 | } |
5619 | #endif | |
ba25f9dc | 5620 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5621 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5622 | |
5fb5e6de | 5623 | show_stack(p, NULL); |
1da177e4 LT |
5624 | } |
5625 | ||
e59e2ae2 | 5626 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5627 | { |
36c8b586 | 5628 | struct task_struct *g, *p; |
1da177e4 | 5629 | |
4bd77321 IM |
5630 | #if BITS_PER_LONG == 32 |
5631 | printk(KERN_INFO | |
5632 | " task PC stack pid father\n"); | |
1da177e4 | 5633 | #else |
4bd77321 IM |
5634 | printk(KERN_INFO |
5635 | " task PC stack pid father\n"); | |
1da177e4 LT |
5636 | #endif |
5637 | read_lock(&tasklist_lock); | |
5638 | do_each_thread(g, p) { | |
5639 | /* | |
5640 | * reset the NMI-timeout, listing all files on a slow | |
5641 | * console might take alot of time: | |
5642 | */ | |
5643 | touch_nmi_watchdog(); | |
39bc89fd | 5644 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5645 | sched_show_task(p); |
1da177e4 LT |
5646 | } while_each_thread(g, p); |
5647 | ||
04c9167f JF |
5648 | touch_all_softlockup_watchdogs(); |
5649 | ||
dd41f596 IM |
5650 | #ifdef CONFIG_SCHED_DEBUG |
5651 | sysrq_sched_debug_show(); | |
5652 | #endif | |
1da177e4 | 5653 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5654 | /* |
5655 | * Only show locks if all tasks are dumped: | |
5656 | */ | |
5657 | if (state_filter == -1) | |
5658 | debug_show_all_locks(); | |
1da177e4 LT |
5659 | } |
5660 | ||
1df21055 IM |
5661 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5662 | { | |
dd41f596 | 5663 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5664 | } |
5665 | ||
f340c0d1 IM |
5666 | /** |
5667 | * init_idle - set up an idle thread for a given CPU | |
5668 | * @idle: task in question | |
5669 | * @cpu: cpu the idle task belongs to | |
5670 | * | |
5671 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5672 | * flag, to make booting more robust. | |
5673 | */ | |
5c1e1767 | 5674 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5675 | { |
70b97a7f | 5676 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5677 | unsigned long flags; |
5678 | ||
dd41f596 IM |
5679 | __sched_fork(idle); |
5680 | idle->se.exec_start = sched_clock(); | |
5681 | ||
b29739f9 | 5682 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5683 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5684 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5685 | |
5686 | spin_lock_irqsave(&rq->lock, flags); | |
5687 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5688 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5689 | idle->oncpu = 1; | |
5690 | #endif | |
1da177e4 LT |
5691 | spin_unlock_irqrestore(&rq->lock, flags); |
5692 | ||
5693 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5694 | #if defined(CONFIG_PREEMPT) |
5695 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5696 | #else | |
a1261f54 | 5697 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5698 | #endif |
dd41f596 IM |
5699 | /* |
5700 | * The idle tasks have their own, simple scheduling class: | |
5701 | */ | |
5702 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5703 | } |
5704 | ||
5705 | /* | |
5706 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5707 | * indicates which cpus entered this state. This is used | |
5708 | * in the rcu update to wait only for active cpus. For system | |
5709 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5710 | * always be CPU_MASK_NONE. | |
5711 | */ | |
5712 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5713 | ||
19978ca6 IM |
5714 | /* |
5715 | * Increase the granularity value when there are more CPUs, | |
5716 | * because with more CPUs the 'effective latency' as visible | |
5717 | * to users decreases. But the relationship is not linear, | |
5718 | * so pick a second-best guess by going with the log2 of the | |
5719 | * number of CPUs. | |
5720 | * | |
5721 | * This idea comes from the SD scheduler of Con Kolivas: | |
5722 | */ | |
5723 | static inline void sched_init_granularity(void) | |
5724 | { | |
5725 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5726 | const unsigned long limit = 200000000; | |
5727 | ||
5728 | sysctl_sched_min_granularity *= factor; | |
5729 | if (sysctl_sched_min_granularity > limit) | |
5730 | sysctl_sched_min_granularity = limit; | |
5731 | ||
5732 | sysctl_sched_latency *= factor; | |
5733 | if (sysctl_sched_latency > limit) | |
5734 | sysctl_sched_latency = limit; | |
5735 | ||
5736 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5737 | } |
5738 | ||
1da177e4 LT |
5739 | #ifdef CONFIG_SMP |
5740 | /* | |
5741 | * This is how migration works: | |
5742 | * | |
70b97a7f | 5743 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5744 | * runqueue and wake up that CPU's migration thread. |
5745 | * 2) we down() the locked semaphore => thread blocks. | |
5746 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5747 | * thread off the CPU) | |
5748 | * 4) it gets the migration request and checks whether the migrated | |
5749 | * task is still in the wrong runqueue. | |
5750 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5751 | * it and puts it into the right queue. | |
5752 | * 6) migration thread up()s the semaphore. | |
5753 | * 7) we wake up and the migration is done. | |
5754 | */ | |
5755 | ||
5756 | /* | |
5757 | * Change a given task's CPU affinity. Migrate the thread to a | |
5758 | * proper CPU and schedule it away if the CPU it's executing on | |
5759 | * is removed from the allowed bitmask. | |
5760 | * | |
5761 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5762 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5763 | * call is not atomic; no spinlocks may be held. |
5764 | */ | |
cd8ba7cd | 5765 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5766 | { |
70b97a7f | 5767 | struct migration_req req; |
1da177e4 | 5768 | unsigned long flags; |
70b97a7f | 5769 | struct rq *rq; |
48f24c4d | 5770 | int ret = 0; |
1da177e4 LT |
5771 | |
5772 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5773 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5774 | ret = -EINVAL; |
5775 | goto out; | |
5776 | } | |
5777 | ||
9985b0ba DR |
5778 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5779 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5780 | ret = -EINVAL; | |
5781 | goto out; | |
5782 | } | |
5783 | ||
73fe6aae | 5784 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5785 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5786 | else { |
cd8ba7cd MT |
5787 | p->cpus_allowed = *new_mask; |
5788 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5789 | } |
5790 | ||
1da177e4 | 5791 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5792 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5793 | goto out; |
5794 | ||
cd8ba7cd | 5795 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5796 | /* Need help from migration thread: drop lock and wait. */ |
5797 | task_rq_unlock(rq, &flags); | |
5798 | wake_up_process(rq->migration_thread); | |
5799 | wait_for_completion(&req.done); | |
5800 | tlb_migrate_finish(p->mm); | |
5801 | return 0; | |
5802 | } | |
5803 | out: | |
5804 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5805 | |
1da177e4 LT |
5806 | return ret; |
5807 | } | |
cd8ba7cd | 5808 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5809 | |
5810 | /* | |
41a2d6cf | 5811 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5812 | * this because either it can't run here any more (set_cpus_allowed() |
5813 | * away from this CPU, or CPU going down), or because we're | |
5814 | * attempting to rebalance this task on exec (sched_exec). | |
5815 | * | |
5816 | * So we race with normal scheduler movements, but that's OK, as long | |
5817 | * as the task is no longer on this CPU. | |
efc30814 KK |
5818 | * |
5819 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5820 | */ |
efc30814 | 5821 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5822 | { |
70b97a7f | 5823 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5824 | int ret = 0, on_rq; |
1da177e4 | 5825 | |
e761b772 | 5826 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5827 | return ret; |
1da177e4 LT |
5828 | |
5829 | rq_src = cpu_rq(src_cpu); | |
5830 | rq_dest = cpu_rq(dest_cpu); | |
5831 | ||
5832 | double_rq_lock(rq_src, rq_dest); | |
5833 | /* Already moved. */ | |
5834 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5835 | goto done; |
1da177e4 LT |
5836 | /* Affinity changed (again). */ |
5837 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 5838 | goto fail; |
1da177e4 | 5839 | |
dd41f596 | 5840 | on_rq = p->se.on_rq; |
6e82a3be | 5841 | if (on_rq) |
2e1cb74a | 5842 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5843 | |
1da177e4 | 5844 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5845 | if (on_rq) { |
5846 | activate_task(rq_dest, p, 0); | |
5847 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5848 | } |
b1e38734 | 5849 | done: |
efc30814 | 5850 | ret = 1; |
b1e38734 | 5851 | fail: |
1da177e4 | 5852 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5853 | return ret; |
1da177e4 LT |
5854 | } |
5855 | ||
5856 | /* | |
5857 | * migration_thread - this is a highprio system thread that performs | |
5858 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5859 | * another runqueue. | |
5860 | */ | |
95cdf3b7 | 5861 | static int migration_thread(void *data) |
1da177e4 | 5862 | { |
1da177e4 | 5863 | int cpu = (long)data; |
70b97a7f | 5864 | struct rq *rq; |
1da177e4 LT |
5865 | |
5866 | rq = cpu_rq(cpu); | |
5867 | BUG_ON(rq->migration_thread != current); | |
5868 | ||
5869 | set_current_state(TASK_INTERRUPTIBLE); | |
5870 | while (!kthread_should_stop()) { | |
70b97a7f | 5871 | struct migration_req *req; |
1da177e4 | 5872 | struct list_head *head; |
1da177e4 | 5873 | |
1da177e4 LT |
5874 | spin_lock_irq(&rq->lock); |
5875 | ||
5876 | if (cpu_is_offline(cpu)) { | |
5877 | spin_unlock_irq(&rq->lock); | |
5878 | goto wait_to_die; | |
5879 | } | |
5880 | ||
5881 | if (rq->active_balance) { | |
5882 | active_load_balance(rq, cpu); | |
5883 | rq->active_balance = 0; | |
5884 | } | |
5885 | ||
5886 | head = &rq->migration_queue; | |
5887 | ||
5888 | if (list_empty(head)) { | |
5889 | spin_unlock_irq(&rq->lock); | |
5890 | schedule(); | |
5891 | set_current_state(TASK_INTERRUPTIBLE); | |
5892 | continue; | |
5893 | } | |
70b97a7f | 5894 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5895 | list_del_init(head->next); |
5896 | ||
674311d5 NP |
5897 | spin_unlock(&rq->lock); |
5898 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5899 | local_irq_enable(); | |
1da177e4 LT |
5900 | |
5901 | complete(&req->done); | |
5902 | } | |
5903 | __set_current_state(TASK_RUNNING); | |
5904 | return 0; | |
5905 | ||
5906 | wait_to_die: | |
5907 | /* Wait for kthread_stop */ | |
5908 | set_current_state(TASK_INTERRUPTIBLE); | |
5909 | while (!kthread_should_stop()) { | |
5910 | schedule(); | |
5911 | set_current_state(TASK_INTERRUPTIBLE); | |
5912 | } | |
5913 | __set_current_state(TASK_RUNNING); | |
5914 | return 0; | |
5915 | } | |
5916 | ||
5917 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5918 | |
5919 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5920 | { | |
5921 | int ret; | |
5922 | ||
5923 | local_irq_disable(); | |
5924 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5925 | local_irq_enable(); | |
5926 | return ret; | |
5927 | } | |
5928 | ||
054b9108 | 5929 | /* |
3a4fa0a2 | 5930 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5931 | * NOTE: interrupts should be disabled by the caller |
5932 | */ | |
48f24c4d | 5933 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5934 | { |
efc30814 | 5935 | unsigned long flags; |
1da177e4 | 5936 | cpumask_t mask; |
70b97a7f IM |
5937 | struct rq *rq; |
5938 | int dest_cpu; | |
1da177e4 | 5939 | |
3a5c359a AK |
5940 | do { |
5941 | /* On same node? */ | |
5942 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5943 | cpus_and(mask, mask, p->cpus_allowed); | |
5944 | dest_cpu = any_online_cpu(mask); | |
5945 | ||
5946 | /* On any allowed CPU? */ | |
434d53b0 | 5947 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5948 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5949 | ||
5950 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5951 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5952 | cpumask_t cpus_allowed; |
5953 | ||
5954 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5955 | /* |
5956 | * Try to stay on the same cpuset, where the | |
5957 | * current cpuset may be a subset of all cpus. | |
5958 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5959 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5960 | * called within calls to cpuset_lock/cpuset_unlock. |
5961 | */ | |
3a5c359a | 5962 | rq = task_rq_lock(p, &flags); |
470fd646 | 5963 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5964 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5965 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5966 | |
3a5c359a AK |
5967 | /* |
5968 | * Don't tell them about moving exiting tasks or | |
5969 | * kernel threads (both mm NULL), since they never | |
5970 | * leave kernel. | |
5971 | */ | |
41a2d6cf | 5972 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5973 | printk(KERN_INFO "process %d (%s) no " |
5974 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5975 | task_pid_nr(p), p->comm, dead_cpu); |
5976 | } | |
3a5c359a | 5977 | } |
f7b4cddc | 5978 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5979 | } |
5980 | ||
5981 | /* | |
5982 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5983 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5984 | * for performance reasons the counter is not stricly tracking tasks to | |
5985 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5986 | * to keep the global sum constant after CPU-down: | |
5987 | */ | |
70b97a7f | 5988 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5989 | { |
7c16ec58 | 5990 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5991 | unsigned long flags; |
5992 | ||
5993 | local_irq_save(flags); | |
5994 | double_rq_lock(rq_src, rq_dest); | |
5995 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5996 | rq_src->nr_uninterruptible = 0; | |
5997 | double_rq_unlock(rq_src, rq_dest); | |
5998 | local_irq_restore(flags); | |
5999 | } | |
6000 | ||
6001 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6002 | static void migrate_live_tasks(int src_cpu) | |
6003 | { | |
48f24c4d | 6004 | struct task_struct *p, *t; |
1da177e4 | 6005 | |
f7b4cddc | 6006 | read_lock(&tasklist_lock); |
1da177e4 | 6007 | |
48f24c4d IM |
6008 | do_each_thread(t, p) { |
6009 | if (p == current) | |
1da177e4 LT |
6010 | continue; |
6011 | ||
48f24c4d IM |
6012 | if (task_cpu(p) == src_cpu) |
6013 | move_task_off_dead_cpu(src_cpu, p); | |
6014 | } while_each_thread(t, p); | |
1da177e4 | 6015 | |
f7b4cddc | 6016 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6017 | } |
6018 | ||
dd41f596 IM |
6019 | /* |
6020 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6021 | * It does so by boosting its priority to highest possible. |
6022 | * Used by CPU offline code. | |
1da177e4 LT |
6023 | */ |
6024 | void sched_idle_next(void) | |
6025 | { | |
48f24c4d | 6026 | int this_cpu = smp_processor_id(); |
70b97a7f | 6027 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6028 | struct task_struct *p = rq->idle; |
6029 | unsigned long flags; | |
6030 | ||
6031 | /* cpu has to be offline */ | |
48f24c4d | 6032 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6033 | |
48f24c4d IM |
6034 | /* |
6035 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6036 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6037 | */ |
6038 | spin_lock_irqsave(&rq->lock, flags); | |
6039 | ||
dd41f596 | 6040 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6041 | |
94bc9a7b DA |
6042 | update_rq_clock(rq); |
6043 | activate_task(rq, p, 0); | |
1da177e4 LT |
6044 | |
6045 | spin_unlock_irqrestore(&rq->lock, flags); | |
6046 | } | |
6047 | ||
48f24c4d IM |
6048 | /* |
6049 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6050 | * offline. |
6051 | */ | |
6052 | void idle_task_exit(void) | |
6053 | { | |
6054 | struct mm_struct *mm = current->active_mm; | |
6055 | ||
6056 | BUG_ON(cpu_online(smp_processor_id())); | |
6057 | ||
6058 | if (mm != &init_mm) | |
6059 | switch_mm(mm, &init_mm, current); | |
6060 | mmdrop(mm); | |
6061 | } | |
6062 | ||
054b9108 | 6063 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6064 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6065 | { |
70b97a7f | 6066 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6067 | |
6068 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6069 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6070 | |
6071 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6072 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6073 | |
48f24c4d | 6074 | get_task_struct(p); |
1da177e4 LT |
6075 | |
6076 | /* | |
6077 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6078 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6079 | * fine. |
6080 | */ | |
f7b4cddc | 6081 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6082 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6083 | spin_lock_irq(&rq->lock); |
1da177e4 | 6084 | |
48f24c4d | 6085 | put_task_struct(p); |
1da177e4 LT |
6086 | } |
6087 | ||
6088 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6089 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6090 | { | |
70b97a7f | 6091 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6092 | struct task_struct *next; |
48f24c4d | 6093 | |
dd41f596 IM |
6094 | for ( ; ; ) { |
6095 | if (!rq->nr_running) | |
6096 | break; | |
a8e504d2 | 6097 | update_rq_clock(rq); |
ff95f3df | 6098 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6099 | if (!next) |
6100 | break; | |
79c53799 | 6101 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6102 | migrate_dead(dead_cpu, next); |
e692ab53 | 6103 | |
1da177e4 LT |
6104 | } |
6105 | } | |
6106 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6107 | ||
e692ab53 NP |
6108 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6109 | ||
6110 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6111 | { |
6112 | .procname = "sched_domain", | |
c57baf1e | 6113 | .mode = 0555, |
e0361851 | 6114 | }, |
38605cae | 6115 | {0, }, |
e692ab53 NP |
6116 | }; |
6117 | ||
6118 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6119 | { |
c57baf1e | 6120 | .ctl_name = CTL_KERN, |
e0361851 | 6121 | .procname = "kernel", |
c57baf1e | 6122 | .mode = 0555, |
e0361851 AD |
6123 | .child = sd_ctl_dir, |
6124 | }, | |
38605cae | 6125 | {0, }, |
e692ab53 NP |
6126 | }; |
6127 | ||
6128 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6129 | { | |
6130 | struct ctl_table *entry = | |
5cf9f062 | 6131 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6132 | |
e692ab53 NP |
6133 | return entry; |
6134 | } | |
6135 | ||
6382bc90 MM |
6136 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6137 | { | |
cd790076 | 6138 | struct ctl_table *entry; |
6382bc90 | 6139 | |
cd790076 MM |
6140 | /* |
6141 | * In the intermediate directories, both the child directory and | |
6142 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6143 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6144 | * static strings and all have proc handlers. |
6145 | */ | |
6146 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6147 | if (entry->child) |
6148 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6149 | if (entry->proc_handler == NULL) |
6150 | kfree(entry->procname); | |
6151 | } | |
6382bc90 MM |
6152 | |
6153 | kfree(*tablep); | |
6154 | *tablep = NULL; | |
6155 | } | |
6156 | ||
e692ab53 | 6157 | static void |
e0361851 | 6158 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6159 | const char *procname, void *data, int maxlen, |
6160 | mode_t mode, proc_handler *proc_handler) | |
6161 | { | |
e692ab53 NP |
6162 | entry->procname = procname; |
6163 | entry->data = data; | |
6164 | entry->maxlen = maxlen; | |
6165 | entry->mode = mode; | |
6166 | entry->proc_handler = proc_handler; | |
6167 | } | |
6168 | ||
6169 | static struct ctl_table * | |
6170 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6171 | { | |
ace8b3d6 | 6172 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6173 | |
ad1cdc1d MM |
6174 | if (table == NULL) |
6175 | return NULL; | |
6176 | ||
e0361851 | 6177 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6178 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6179 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6180 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6181 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6182 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6183 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6184 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6185 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6186 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6187 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6188 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6189 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6190 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6191 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6192 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6193 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6194 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6195 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6196 | &sd->cache_nice_tries, |
6197 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6198 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6199 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6200 | /* &table[11] is terminator */ |
e692ab53 NP |
6201 | |
6202 | return table; | |
6203 | } | |
6204 | ||
9a4e7159 | 6205 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6206 | { |
6207 | struct ctl_table *entry, *table; | |
6208 | struct sched_domain *sd; | |
6209 | int domain_num = 0, i; | |
6210 | char buf[32]; | |
6211 | ||
6212 | for_each_domain(cpu, sd) | |
6213 | domain_num++; | |
6214 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6215 | if (table == NULL) |
6216 | return NULL; | |
e692ab53 NP |
6217 | |
6218 | i = 0; | |
6219 | for_each_domain(cpu, sd) { | |
6220 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6221 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6222 | entry->mode = 0555; |
e692ab53 NP |
6223 | entry->child = sd_alloc_ctl_domain_table(sd); |
6224 | entry++; | |
6225 | i++; | |
6226 | } | |
6227 | return table; | |
6228 | } | |
6229 | ||
6230 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6231 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6232 | { |
6233 | int i, cpu_num = num_online_cpus(); | |
6234 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6235 | char buf[32]; | |
6236 | ||
7378547f MM |
6237 | WARN_ON(sd_ctl_dir[0].child); |
6238 | sd_ctl_dir[0].child = entry; | |
6239 | ||
ad1cdc1d MM |
6240 | if (entry == NULL) |
6241 | return; | |
6242 | ||
97b6ea7b | 6243 | for_each_online_cpu(i) { |
e692ab53 | 6244 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6245 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6246 | entry->mode = 0555; |
e692ab53 | 6247 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6248 | entry++; |
e692ab53 | 6249 | } |
7378547f MM |
6250 | |
6251 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6252 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6253 | } | |
6382bc90 | 6254 | |
7378547f | 6255 | /* may be called multiple times per register */ |
6382bc90 MM |
6256 | static void unregister_sched_domain_sysctl(void) |
6257 | { | |
7378547f MM |
6258 | if (sd_sysctl_header) |
6259 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6260 | sd_sysctl_header = NULL; |
7378547f MM |
6261 | if (sd_ctl_dir[0].child) |
6262 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6263 | } |
e692ab53 | 6264 | #else |
6382bc90 MM |
6265 | static void register_sched_domain_sysctl(void) |
6266 | { | |
6267 | } | |
6268 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6269 | { |
6270 | } | |
6271 | #endif | |
6272 | ||
1f11eb6a GH |
6273 | static void set_rq_online(struct rq *rq) |
6274 | { | |
6275 | if (!rq->online) { | |
6276 | const struct sched_class *class; | |
6277 | ||
6278 | cpu_set(rq->cpu, rq->rd->online); | |
6279 | rq->online = 1; | |
6280 | ||
6281 | for_each_class(class) { | |
6282 | if (class->rq_online) | |
6283 | class->rq_online(rq); | |
6284 | } | |
6285 | } | |
6286 | } | |
6287 | ||
6288 | static void set_rq_offline(struct rq *rq) | |
6289 | { | |
6290 | if (rq->online) { | |
6291 | const struct sched_class *class; | |
6292 | ||
6293 | for_each_class(class) { | |
6294 | if (class->rq_offline) | |
6295 | class->rq_offline(rq); | |
6296 | } | |
6297 | ||
6298 | cpu_clear(rq->cpu, rq->rd->online); | |
6299 | rq->online = 0; | |
6300 | } | |
6301 | } | |
6302 | ||
1da177e4 LT |
6303 | /* |
6304 | * migration_call - callback that gets triggered when a CPU is added. | |
6305 | * Here we can start up the necessary migration thread for the new CPU. | |
6306 | */ | |
48f24c4d IM |
6307 | static int __cpuinit |
6308 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6309 | { |
1da177e4 | 6310 | struct task_struct *p; |
48f24c4d | 6311 | int cpu = (long)hcpu; |
1da177e4 | 6312 | unsigned long flags; |
70b97a7f | 6313 | struct rq *rq; |
1da177e4 LT |
6314 | |
6315 | switch (action) { | |
5be9361c | 6316 | |
1da177e4 | 6317 | case CPU_UP_PREPARE: |
8bb78442 | 6318 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6319 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6320 | if (IS_ERR(p)) |
6321 | return NOTIFY_BAD; | |
1da177e4 LT |
6322 | kthread_bind(p, cpu); |
6323 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6324 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6325 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6326 | task_rq_unlock(rq, &flags); |
6327 | cpu_rq(cpu)->migration_thread = p; | |
6328 | break; | |
48f24c4d | 6329 | |
1da177e4 | 6330 | case CPU_ONLINE: |
8bb78442 | 6331 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6332 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6333 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6334 | |
6335 | /* Update our root-domain */ | |
6336 | rq = cpu_rq(cpu); | |
6337 | spin_lock_irqsave(&rq->lock, flags); | |
6338 | if (rq->rd) { | |
6339 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6340 | |
6341 | set_rq_online(rq); | |
1f94ef59 GH |
6342 | } |
6343 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6344 | break; |
48f24c4d | 6345 | |
1da177e4 LT |
6346 | #ifdef CONFIG_HOTPLUG_CPU |
6347 | case CPU_UP_CANCELED: | |
8bb78442 | 6348 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6349 | if (!cpu_rq(cpu)->migration_thread) |
6350 | break; | |
41a2d6cf | 6351 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6352 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6353 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6354 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6355 | cpu_rq(cpu)->migration_thread = NULL; | |
6356 | break; | |
48f24c4d | 6357 | |
1da177e4 | 6358 | case CPU_DEAD: |
8bb78442 | 6359 | case CPU_DEAD_FROZEN: |
470fd646 | 6360 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6361 | migrate_live_tasks(cpu); |
6362 | rq = cpu_rq(cpu); | |
6363 | kthread_stop(rq->migration_thread); | |
6364 | rq->migration_thread = NULL; | |
6365 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6366 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6367 | update_rq_clock(rq); |
2e1cb74a | 6368 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6369 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6370 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6371 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6372 | migrate_dead_tasks(cpu); |
d2da272a | 6373 | spin_unlock_irq(&rq->lock); |
470fd646 | 6374 | cpuset_unlock(); |
1da177e4 LT |
6375 | migrate_nr_uninterruptible(rq); |
6376 | BUG_ON(rq->nr_running != 0); | |
6377 | ||
41a2d6cf IM |
6378 | /* |
6379 | * No need to migrate the tasks: it was best-effort if | |
6380 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6381 | * the requestors. | |
6382 | */ | |
1da177e4 LT |
6383 | spin_lock_irq(&rq->lock); |
6384 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6385 | struct migration_req *req; |
6386 | ||
1da177e4 | 6387 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6388 | struct migration_req, list); |
1da177e4 LT |
6389 | list_del_init(&req->list); |
6390 | complete(&req->done); | |
6391 | } | |
6392 | spin_unlock_irq(&rq->lock); | |
6393 | break; | |
57d885fe | 6394 | |
08f503b0 GH |
6395 | case CPU_DYING: |
6396 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6397 | /* Update our root-domain */ |
6398 | rq = cpu_rq(cpu); | |
6399 | spin_lock_irqsave(&rq->lock, flags); | |
6400 | if (rq->rd) { | |
6401 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6402 | set_rq_offline(rq); |
57d885fe GH |
6403 | } |
6404 | spin_unlock_irqrestore(&rq->lock, flags); | |
6405 | break; | |
1da177e4 LT |
6406 | #endif |
6407 | } | |
6408 | return NOTIFY_OK; | |
6409 | } | |
6410 | ||
6411 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6412 | * happens before everything else. | |
6413 | */ | |
26c2143b | 6414 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6415 | .notifier_call = migration_call, |
6416 | .priority = 10 | |
6417 | }; | |
6418 | ||
7babe8db | 6419 | static int __init migration_init(void) |
1da177e4 LT |
6420 | { |
6421 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6422 | int err; |
48f24c4d IM |
6423 | |
6424 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6425 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6426 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6427 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6428 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6429 | |
6430 | return err; | |
1da177e4 | 6431 | } |
7babe8db | 6432 | early_initcall(migration_init); |
1da177e4 LT |
6433 | #endif |
6434 | ||
6435 | #ifdef CONFIG_SMP | |
476f3534 | 6436 | |
3e9830dc | 6437 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6438 | |
099f98c8 GS |
6439 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6440 | { | |
6441 | switch (lvl) { | |
6442 | case SD_LV_NONE: | |
6443 | return "NONE"; | |
6444 | case SD_LV_SIBLING: | |
6445 | return "SIBLING"; | |
6446 | case SD_LV_MC: | |
6447 | return "MC"; | |
6448 | case SD_LV_CPU: | |
6449 | return "CPU"; | |
6450 | case SD_LV_NODE: | |
6451 | return "NODE"; | |
6452 | case SD_LV_ALLNODES: | |
6453 | return "ALLNODES"; | |
6454 | case SD_LV_MAX: | |
6455 | return "MAX"; | |
6456 | ||
6457 | } | |
6458 | return "MAX"; | |
6459 | } | |
6460 | ||
7c16ec58 MT |
6461 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6462 | cpumask_t *groupmask) | |
1da177e4 | 6463 | { |
4dcf6aff | 6464 | struct sched_group *group = sd->groups; |
434d53b0 | 6465 | char str[256]; |
1da177e4 | 6466 | |
434d53b0 | 6467 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6468 | cpus_clear(*groupmask); |
4dcf6aff IM |
6469 | |
6470 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6471 | ||
6472 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6473 | printk("does not load-balance\n"); | |
6474 | if (sd->parent) | |
6475 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6476 | " has parent"); | |
6477 | return -1; | |
41c7ce9a NP |
6478 | } |
6479 | ||
099f98c8 GS |
6480 | printk(KERN_CONT "span %s level %s\n", |
6481 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6482 | |
6483 | if (!cpu_isset(cpu, sd->span)) { | |
6484 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6485 | "CPU%d\n", cpu); | |
6486 | } | |
6487 | if (!cpu_isset(cpu, group->cpumask)) { | |
6488 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6489 | " CPU%d\n", cpu); | |
6490 | } | |
1da177e4 | 6491 | |
4dcf6aff | 6492 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6493 | do { |
4dcf6aff IM |
6494 | if (!group) { |
6495 | printk("\n"); | |
6496 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6497 | break; |
6498 | } | |
6499 | ||
4dcf6aff IM |
6500 | if (!group->__cpu_power) { |
6501 | printk(KERN_CONT "\n"); | |
6502 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6503 | "set\n"); | |
6504 | break; | |
6505 | } | |
1da177e4 | 6506 | |
4dcf6aff IM |
6507 | if (!cpus_weight(group->cpumask)) { |
6508 | printk(KERN_CONT "\n"); | |
6509 | printk(KERN_ERR "ERROR: empty group\n"); | |
6510 | break; | |
6511 | } | |
1da177e4 | 6512 | |
7c16ec58 | 6513 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6514 | printk(KERN_CONT "\n"); |
6515 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6516 | break; | |
6517 | } | |
1da177e4 | 6518 | |
7c16ec58 | 6519 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6520 | |
434d53b0 | 6521 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6522 | printk(KERN_CONT " %s", str); |
1da177e4 | 6523 | |
4dcf6aff IM |
6524 | group = group->next; |
6525 | } while (group != sd->groups); | |
6526 | printk(KERN_CONT "\n"); | |
1da177e4 | 6527 | |
7c16ec58 | 6528 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6529 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6530 | |
7c16ec58 | 6531 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6532 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6533 | "of domain->span\n"); | |
6534 | return 0; | |
6535 | } | |
1da177e4 | 6536 | |
4dcf6aff IM |
6537 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6538 | { | |
7c16ec58 | 6539 | cpumask_t *groupmask; |
4dcf6aff | 6540 | int level = 0; |
1da177e4 | 6541 | |
4dcf6aff IM |
6542 | if (!sd) { |
6543 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6544 | return; | |
6545 | } | |
1da177e4 | 6546 | |
4dcf6aff IM |
6547 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6548 | ||
7c16ec58 MT |
6549 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6550 | if (!groupmask) { | |
6551 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6552 | return; | |
6553 | } | |
6554 | ||
4dcf6aff | 6555 | for (;;) { |
7c16ec58 | 6556 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6557 | break; |
1da177e4 LT |
6558 | level++; |
6559 | sd = sd->parent; | |
33859f7f | 6560 | if (!sd) |
4dcf6aff IM |
6561 | break; |
6562 | } | |
7c16ec58 | 6563 | kfree(groupmask); |
1da177e4 | 6564 | } |
6d6bc0ad | 6565 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6566 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6567 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6568 | |
1a20ff27 | 6569 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6570 | { |
6571 | if (cpus_weight(sd->span) == 1) | |
6572 | return 1; | |
6573 | ||
6574 | /* Following flags need at least 2 groups */ | |
6575 | if (sd->flags & (SD_LOAD_BALANCE | | |
6576 | SD_BALANCE_NEWIDLE | | |
6577 | SD_BALANCE_FORK | | |
89c4710e SS |
6578 | SD_BALANCE_EXEC | |
6579 | SD_SHARE_CPUPOWER | | |
6580 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6581 | if (sd->groups != sd->groups->next) |
6582 | return 0; | |
6583 | } | |
6584 | ||
6585 | /* Following flags don't use groups */ | |
6586 | if (sd->flags & (SD_WAKE_IDLE | | |
6587 | SD_WAKE_AFFINE | | |
6588 | SD_WAKE_BALANCE)) | |
6589 | return 0; | |
6590 | ||
6591 | return 1; | |
6592 | } | |
6593 | ||
48f24c4d IM |
6594 | static int |
6595 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6596 | { |
6597 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6598 | ||
6599 | if (sd_degenerate(parent)) | |
6600 | return 1; | |
6601 | ||
6602 | if (!cpus_equal(sd->span, parent->span)) | |
6603 | return 0; | |
6604 | ||
6605 | /* Does parent contain flags not in child? */ | |
6606 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6607 | if (cflags & SD_WAKE_AFFINE) | |
6608 | pflags &= ~SD_WAKE_BALANCE; | |
6609 | /* Flags needing groups don't count if only 1 group in parent */ | |
6610 | if (parent->groups == parent->groups->next) { | |
6611 | pflags &= ~(SD_LOAD_BALANCE | | |
6612 | SD_BALANCE_NEWIDLE | | |
6613 | SD_BALANCE_FORK | | |
89c4710e SS |
6614 | SD_BALANCE_EXEC | |
6615 | SD_SHARE_CPUPOWER | | |
6616 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6617 | } |
6618 | if (~cflags & pflags) | |
6619 | return 0; | |
6620 | ||
6621 | return 1; | |
6622 | } | |
6623 | ||
57d885fe GH |
6624 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6625 | { | |
6626 | unsigned long flags; | |
57d885fe GH |
6627 | |
6628 | spin_lock_irqsave(&rq->lock, flags); | |
6629 | ||
6630 | if (rq->rd) { | |
6631 | struct root_domain *old_rd = rq->rd; | |
6632 | ||
1f11eb6a GH |
6633 | if (cpu_isset(rq->cpu, old_rd->online)) |
6634 | set_rq_offline(rq); | |
57d885fe | 6635 | |
dc938520 | 6636 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6637 | |
57d885fe GH |
6638 | if (atomic_dec_and_test(&old_rd->refcount)) |
6639 | kfree(old_rd); | |
6640 | } | |
6641 | ||
6642 | atomic_inc(&rd->refcount); | |
6643 | rq->rd = rd; | |
6644 | ||
dc938520 | 6645 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6646 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6647 | set_rq_online(rq); |
57d885fe GH |
6648 | |
6649 | spin_unlock_irqrestore(&rq->lock, flags); | |
6650 | } | |
6651 | ||
dc938520 | 6652 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6653 | { |
6654 | memset(rd, 0, sizeof(*rd)); | |
6655 | ||
dc938520 GH |
6656 | cpus_clear(rd->span); |
6657 | cpus_clear(rd->online); | |
6e0534f2 GH |
6658 | |
6659 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6660 | } |
6661 | ||
6662 | static void init_defrootdomain(void) | |
6663 | { | |
dc938520 | 6664 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6665 | atomic_set(&def_root_domain.refcount, 1); |
6666 | } | |
6667 | ||
dc938520 | 6668 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6669 | { |
6670 | struct root_domain *rd; | |
6671 | ||
6672 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6673 | if (!rd) | |
6674 | return NULL; | |
6675 | ||
dc938520 | 6676 | init_rootdomain(rd); |
57d885fe GH |
6677 | |
6678 | return rd; | |
6679 | } | |
6680 | ||
1da177e4 | 6681 | /* |
0eab9146 | 6682 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6683 | * hold the hotplug lock. |
6684 | */ | |
0eab9146 IM |
6685 | static void |
6686 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6687 | { |
70b97a7f | 6688 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6689 | struct sched_domain *tmp; |
6690 | ||
6691 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6692 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6693 | struct sched_domain *parent = tmp->parent; | |
6694 | if (!parent) | |
6695 | break; | |
1a848870 | 6696 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6697 | tmp->parent = parent->parent; |
1a848870 SS |
6698 | if (parent->parent) |
6699 | parent->parent->child = tmp; | |
6700 | } | |
245af2c7 SS |
6701 | } |
6702 | ||
1a848870 | 6703 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6704 | sd = sd->parent; |
1a848870 SS |
6705 | if (sd) |
6706 | sd->child = NULL; | |
6707 | } | |
1da177e4 LT |
6708 | |
6709 | sched_domain_debug(sd, cpu); | |
6710 | ||
57d885fe | 6711 | rq_attach_root(rq, rd); |
674311d5 | 6712 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6713 | } |
6714 | ||
6715 | /* cpus with isolated domains */ | |
67af63a6 | 6716 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6717 | |
6718 | /* Setup the mask of cpus configured for isolated domains */ | |
6719 | static int __init isolated_cpu_setup(char *str) | |
6720 | { | |
13b40c1e MT |
6721 | static int __initdata ints[NR_CPUS]; |
6722 | int i; | |
1da177e4 LT |
6723 | |
6724 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6725 | cpus_clear(cpu_isolated_map); | |
6726 | for (i = 1; i <= ints[0]; i++) | |
6727 | if (ints[i] < NR_CPUS) | |
6728 | cpu_set(ints[i], cpu_isolated_map); | |
6729 | return 1; | |
6730 | } | |
6731 | ||
8927f494 | 6732 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6733 | |
6734 | /* | |
6711cab4 SS |
6735 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6736 | * to a function which identifies what group(along with sched group) a CPU | |
6737 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6738 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6739 | * |
6740 | * init_sched_build_groups will build a circular linked list of the groups | |
6741 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6742 | * and ->cpu_power to 0. | |
6743 | */ | |
a616058b | 6744 | static void |
7c16ec58 | 6745 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6746 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6747 | struct sched_group **sg, |
6748 | cpumask_t *tmpmask), | |
6749 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6750 | { |
6751 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6752 | int i; |
6753 | ||
7c16ec58 MT |
6754 | cpus_clear(*covered); |
6755 | ||
363ab6f1 | 6756 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6757 | struct sched_group *sg; |
7c16ec58 | 6758 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6759 | int j; |
6760 | ||
7c16ec58 | 6761 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6762 | continue; |
6763 | ||
7c16ec58 | 6764 | cpus_clear(sg->cpumask); |
5517d86b | 6765 | sg->__cpu_power = 0; |
1da177e4 | 6766 | |
363ab6f1 | 6767 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6768 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6769 | continue; |
6770 | ||
7c16ec58 | 6771 | cpu_set(j, *covered); |
1da177e4 LT |
6772 | cpu_set(j, sg->cpumask); |
6773 | } | |
6774 | if (!first) | |
6775 | first = sg; | |
6776 | if (last) | |
6777 | last->next = sg; | |
6778 | last = sg; | |
6779 | } | |
6780 | last->next = first; | |
6781 | } | |
6782 | ||
9c1cfda2 | 6783 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6784 | |
9c1cfda2 | 6785 | #ifdef CONFIG_NUMA |
198e2f18 | 6786 | |
9c1cfda2 JH |
6787 | /** |
6788 | * find_next_best_node - find the next node to include in a sched_domain | |
6789 | * @node: node whose sched_domain we're building | |
6790 | * @used_nodes: nodes already in the sched_domain | |
6791 | * | |
41a2d6cf | 6792 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6793 | * finds the closest node not already in the @used_nodes map. |
6794 | * | |
6795 | * Should use nodemask_t. | |
6796 | */ | |
c5f59f08 | 6797 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6798 | { |
6799 | int i, n, val, min_val, best_node = 0; | |
6800 | ||
6801 | min_val = INT_MAX; | |
6802 | ||
076ac2af | 6803 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6804 | /* Start at @node */ |
076ac2af | 6805 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6806 | |
6807 | if (!nr_cpus_node(n)) | |
6808 | continue; | |
6809 | ||
6810 | /* Skip already used nodes */ | |
c5f59f08 | 6811 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6812 | continue; |
6813 | ||
6814 | /* Simple min distance search */ | |
6815 | val = node_distance(node, n); | |
6816 | ||
6817 | if (val < min_val) { | |
6818 | min_val = val; | |
6819 | best_node = n; | |
6820 | } | |
6821 | } | |
6822 | ||
c5f59f08 | 6823 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6824 | return best_node; |
6825 | } | |
6826 | ||
6827 | /** | |
6828 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6829 | * @node: node whose cpumask we're constructing | |
73486722 | 6830 | * @span: resulting cpumask |
9c1cfda2 | 6831 | * |
41a2d6cf | 6832 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6833 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6834 | * out optimally. | |
6835 | */ | |
4bdbaad3 | 6836 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6837 | { |
c5f59f08 | 6838 | nodemask_t used_nodes; |
c5f59f08 | 6839 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6840 | int i; |
9c1cfda2 | 6841 | |
4bdbaad3 | 6842 | cpus_clear(*span); |
c5f59f08 | 6843 | nodes_clear(used_nodes); |
9c1cfda2 | 6844 | |
4bdbaad3 | 6845 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6846 | node_set(node, used_nodes); |
9c1cfda2 JH |
6847 | |
6848 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6849 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6850 | |
c5f59f08 | 6851 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6852 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6853 | } |
9c1cfda2 | 6854 | } |
6d6bc0ad | 6855 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6856 | |
5c45bf27 | 6857 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6858 | |
9c1cfda2 | 6859 | /* |
48f24c4d | 6860 | * SMT sched-domains: |
9c1cfda2 | 6861 | */ |
1da177e4 LT |
6862 | #ifdef CONFIG_SCHED_SMT |
6863 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6864 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6865 | |
41a2d6cf | 6866 | static int |
7c16ec58 MT |
6867 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6868 | cpumask_t *unused) | |
1da177e4 | 6869 | { |
6711cab4 SS |
6870 | if (sg) |
6871 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6872 | return cpu; |
6873 | } | |
6d6bc0ad | 6874 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6875 | |
48f24c4d IM |
6876 | /* |
6877 | * multi-core sched-domains: | |
6878 | */ | |
1e9f28fa SS |
6879 | #ifdef CONFIG_SCHED_MC |
6880 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6881 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6882 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6883 | |
6884 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6885 | static int |
7c16ec58 MT |
6886 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6887 | cpumask_t *mask) | |
1e9f28fa | 6888 | { |
6711cab4 | 6889 | int group; |
7c16ec58 MT |
6890 | |
6891 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6892 | cpus_and(*mask, *mask, *cpu_map); | |
6893 | group = first_cpu(*mask); | |
6711cab4 SS |
6894 | if (sg) |
6895 | *sg = &per_cpu(sched_group_core, group); | |
6896 | return group; | |
1e9f28fa SS |
6897 | } |
6898 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6899 | static int |
7c16ec58 MT |
6900 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6901 | cpumask_t *unused) | |
1e9f28fa | 6902 | { |
6711cab4 SS |
6903 | if (sg) |
6904 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6905 | return cpu; |
6906 | } | |
6907 | #endif | |
6908 | ||
1da177e4 | 6909 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6910 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6911 | |
41a2d6cf | 6912 | static int |
7c16ec58 MT |
6913 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6914 | cpumask_t *mask) | |
1da177e4 | 6915 | { |
6711cab4 | 6916 | int group; |
48f24c4d | 6917 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6918 | *mask = cpu_coregroup_map(cpu); |
6919 | cpus_and(*mask, *mask, *cpu_map); | |
6920 | group = first_cpu(*mask); | |
1e9f28fa | 6921 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6922 | *mask = per_cpu(cpu_sibling_map, cpu); |
6923 | cpus_and(*mask, *mask, *cpu_map); | |
6924 | group = first_cpu(*mask); | |
1da177e4 | 6925 | #else |
6711cab4 | 6926 | group = cpu; |
1da177e4 | 6927 | #endif |
6711cab4 SS |
6928 | if (sg) |
6929 | *sg = &per_cpu(sched_group_phys, group); | |
6930 | return group; | |
1da177e4 LT |
6931 | } |
6932 | ||
6933 | #ifdef CONFIG_NUMA | |
1da177e4 | 6934 | /* |
9c1cfda2 JH |
6935 | * The init_sched_build_groups can't handle what we want to do with node |
6936 | * groups, so roll our own. Now each node has its own list of groups which | |
6937 | * gets dynamically allocated. | |
1da177e4 | 6938 | */ |
9c1cfda2 | 6939 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6940 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6941 | |
9c1cfda2 | 6942 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6943 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6944 | |
6711cab4 | 6945 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6946 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6947 | { |
6711cab4 SS |
6948 | int group; |
6949 | ||
7c16ec58 MT |
6950 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6951 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6952 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6953 | |
6954 | if (sg) | |
6955 | *sg = &per_cpu(sched_group_allnodes, group); | |
6956 | return group; | |
1da177e4 | 6957 | } |
6711cab4 | 6958 | |
08069033 SS |
6959 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6960 | { | |
6961 | struct sched_group *sg = group_head; | |
6962 | int j; | |
6963 | ||
6964 | if (!sg) | |
6965 | return; | |
3a5c359a | 6966 | do { |
363ab6f1 | 6967 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 6968 | struct sched_domain *sd; |
08069033 | 6969 | |
3a5c359a AK |
6970 | sd = &per_cpu(phys_domains, j); |
6971 | if (j != first_cpu(sd->groups->cpumask)) { | |
6972 | /* | |
6973 | * Only add "power" once for each | |
6974 | * physical package. | |
6975 | */ | |
6976 | continue; | |
6977 | } | |
08069033 | 6978 | |
3a5c359a AK |
6979 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6980 | } | |
6981 | sg = sg->next; | |
6982 | } while (sg != group_head); | |
08069033 | 6983 | } |
6d6bc0ad | 6984 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6985 | |
a616058b | 6986 | #ifdef CONFIG_NUMA |
51888ca2 | 6987 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6988 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6989 | { |
a616058b | 6990 | int cpu, i; |
51888ca2 | 6991 | |
363ab6f1 | 6992 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
6993 | struct sched_group **sched_group_nodes |
6994 | = sched_group_nodes_bycpu[cpu]; | |
6995 | ||
51888ca2 SV |
6996 | if (!sched_group_nodes) |
6997 | continue; | |
6998 | ||
076ac2af | 6999 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7000 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7001 | ||
7c16ec58 MT |
7002 | *nodemask = node_to_cpumask(i); |
7003 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7004 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7005 | continue; |
7006 | ||
7007 | if (sg == NULL) | |
7008 | continue; | |
7009 | sg = sg->next; | |
7010 | next_sg: | |
7011 | oldsg = sg; | |
7012 | sg = sg->next; | |
7013 | kfree(oldsg); | |
7014 | if (oldsg != sched_group_nodes[i]) | |
7015 | goto next_sg; | |
7016 | } | |
7017 | kfree(sched_group_nodes); | |
7018 | sched_group_nodes_bycpu[cpu] = NULL; | |
7019 | } | |
51888ca2 | 7020 | } |
6d6bc0ad | 7021 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7022 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7023 | { |
7024 | } | |
6d6bc0ad | 7025 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7026 | |
89c4710e SS |
7027 | /* |
7028 | * Initialize sched groups cpu_power. | |
7029 | * | |
7030 | * cpu_power indicates the capacity of sched group, which is used while | |
7031 | * distributing the load between different sched groups in a sched domain. | |
7032 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7033 | * there are asymmetries in the topology. If there are asymmetries, group | |
7034 | * having more cpu_power will pickup more load compared to the group having | |
7035 | * less cpu_power. | |
7036 | * | |
7037 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7038 | * the maximum number of tasks a group can handle in the presence of other idle | |
7039 | * or lightly loaded groups in the same sched domain. | |
7040 | */ | |
7041 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7042 | { | |
7043 | struct sched_domain *child; | |
7044 | struct sched_group *group; | |
7045 | ||
7046 | WARN_ON(!sd || !sd->groups); | |
7047 | ||
7048 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7049 | return; | |
7050 | ||
7051 | child = sd->child; | |
7052 | ||
5517d86b ED |
7053 | sd->groups->__cpu_power = 0; |
7054 | ||
89c4710e SS |
7055 | /* |
7056 | * For perf policy, if the groups in child domain share resources | |
7057 | * (for example cores sharing some portions of the cache hierarchy | |
7058 | * or SMT), then set this domain groups cpu_power such that each group | |
7059 | * can handle only one task, when there are other idle groups in the | |
7060 | * same sched domain. | |
7061 | */ | |
7062 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7063 | (child->flags & | |
7064 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7065 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7066 | return; |
7067 | } | |
7068 | ||
89c4710e SS |
7069 | /* |
7070 | * add cpu_power of each child group to this groups cpu_power | |
7071 | */ | |
7072 | group = child->groups; | |
7073 | do { | |
5517d86b | 7074 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7075 | group = group->next; |
7076 | } while (group != child->groups); | |
7077 | } | |
7078 | ||
7c16ec58 MT |
7079 | /* |
7080 | * Initializers for schedule domains | |
7081 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7082 | */ | |
7083 | ||
7084 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7085 | #define SD_INIT_FUNC(type) \ | |
7086 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7087 | { \ | |
7088 | memset(sd, 0, sizeof(*sd)); \ | |
7089 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7090 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7091 | } |
7092 | ||
7093 | SD_INIT_FUNC(CPU) | |
7094 | #ifdef CONFIG_NUMA | |
7095 | SD_INIT_FUNC(ALLNODES) | |
7096 | SD_INIT_FUNC(NODE) | |
7097 | #endif | |
7098 | #ifdef CONFIG_SCHED_SMT | |
7099 | SD_INIT_FUNC(SIBLING) | |
7100 | #endif | |
7101 | #ifdef CONFIG_SCHED_MC | |
7102 | SD_INIT_FUNC(MC) | |
7103 | #endif | |
7104 | ||
7105 | /* | |
7106 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7107 | * space as the usage in build_sched_domains() dictates. Used only | |
7108 | * if the amount of space is significant. | |
7109 | */ | |
7110 | struct allmasks { | |
7111 | cpumask_t tmpmask; /* make this one first */ | |
7112 | union { | |
7113 | cpumask_t nodemask; | |
7114 | cpumask_t this_sibling_map; | |
7115 | cpumask_t this_core_map; | |
7116 | }; | |
7117 | cpumask_t send_covered; | |
7118 | ||
7119 | #ifdef CONFIG_NUMA | |
7120 | cpumask_t domainspan; | |
7121 | cpumask_t covered; | |
7122 | cpumask_t notcovered; | |
7123 | #endif | |
7124 | }; | |
7125 | ||
7126 | #if NR_CPUS > 128 | |
7127 | #define SCHED_CPUMASK_ALLOC 1 | |
7128 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7129 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7130 | #else | |
7131 | #define SCHED_CPUMASK_ALLOC 0 | |
7132 | #define SCHED_CPUMASK_FREE(v) | |
7133 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7134 | #endif | |
7135 | ||
7136 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7137 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7138 | ||
1d3504fc HS |
7139 | static int default_relax_domain_level = -1; |
7140 | ||
7141 | static int __init setup_relax_domain_level(char *str) | |
7142 | { | |
30e0e178 LZ |
7143 | unsigned long val; |
7144 | ||
7145 | val = simple_strtoul(str, NULL, 0); | |
7146 | if (val < SD_LV_MAX) | |
7147 | default_relax_domain_level = val; | |
7148 | ||
1d3504fc HS |
7149 | return 1; |
7150 | } | |
7151 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7152 | ||
7153 | static void set_domain_attribute(struct sched_domain *sd, | |
7154 | struct sched_domain_attr *attr) | |
7155 | { | |
7156 | int request; | |
7157 | ||
7158 | if (!attr || attr->relax_domain_level < 0) { | |
7159 | if (default_relax_domain_level < 0) | |
7160 | return; | |
7161 | else | |
7162 | request = default_relax_domain_level; | |
7163 | } else | |
7164 | request = attr->relax_domain_level; | |
7165 | if (request < sd->level) { | |
7166 | /* turn off idle balance on this domain */ | |
7167 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7168 | } else { | |
7169 | /* turn on idle balance on this domain */ | |
7170 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7171 | } | |
7172 | } | |
7173 | ||
1da177e4 | 7174 | /* |
1a20ff27 DG |
7175 | * Build sched domains for a given set of cpus and attach the sched domains |
7176 | * to the individual cpus | |
1da177e4 | 7177 | */ |
1d3504fc HS |
7178 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7179 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7180 | { |
7181 | int i; | |
57d885fe | 7182 | struct root_domain *rd; |
7c16ec58 MT |
7183 | SCHED_CPUMASK_DECLARE(allmasks); |
7184 | cpumask_t *tmpmask; | |
d1b55138 JH |
7185 | #ifdef CONFIG_NUMA |
7186 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7187 | int sd_allnodes = 0; |
d1b55138 JH |
7188 | |
7189 | /* | |
7190 | * Allocate the per-node list of sched groups | |
7191 | */ | |
076ac2af | 7192 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7193 | GFP_KERNEL); |
d1b55138 JH |
7194 | if (!sched_group_nodes) { |
7195 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7196 | return -ENOMEM; |
d1b55138 | 7197 | } |
d1b55138 | 7198 | #endif |
1da177e4 | 7199 | |
dc938520 | 7200 | rd = alloc_rootdomain(); |
57d885fe GH |
7201 | if (!rd) { |
7202 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7203 | #ifdef CONFIG_NUMA |
7204 | kfree(sched_group_nodes); | |
7205 | #endif | |
57d885fe GH |
7206 | return -ENOMEM; |
7207 | } | |
7208 | ||
7c16ec58 MT |
7209 | #if SCHED_CPUMASK_ALLOC |
7210 | /* get space for all scratch cpumask variables */ | |
7211 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7212 | if (!allmasks) { | |
7213 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7214 | kfree(rd); | |
7215 | #ifdef CONFIG_NUMA | |
7216 | kfree(sched_group_nodes); | |
7217 | #endif | |
7218 | return -ENOMEM; | |
7219 | } | |
7220 | #endif | |
7221 | tmpmask = (cpumask_t *)allmasks; | |
7222 | ||
7223 | ||
7224 | #ifdef CONFIG_NUMA | |
7225 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7226 | #endif | |
7227 | ||
1da177e4 | 7228 | /* |
1a20ff27 | 7229 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7230 | */ |
363ab6f1 | 7231 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7232 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7233 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7234 | |
7c16ec58 MT |
7235 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7236 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7237 | |
7238 | #ifdef CONFIG_NUMA | |
dd41f596 | 7239 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7240 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7241 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7242 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7243 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7244 | sd->span = *cpu_map; |
7c16ec58 | 7245 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7246 | p = sd; |
6711cab4 | 7247 | sd_allnodes = 1; |
9c1cfda2 JH |
7248 | } else |
7249 | p = NULL; | |
7250 | ||
1da177e4 | 7251 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7252 | SD_INIT(sd, NODE); |
1d3504fc | 7253 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7254 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7255 | sd->parent = p; |
1a848870 SS |
7256 | if (p) |
7257 | p->child = sd; | |
9c1cfda2 | 7258 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7259 | #endif |
7260 | ||
7261 | p = sd; | |
7262 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7263 | SD_INIT(sd, CPU); |
1d3504fc | 7264 | set_domain_attribute(sd, attr); |
7c16ec58 | 7265 | sd->span = *nodemask; |
1da177e4 | 7266 | sd->parent = p; |
1a848870 SS |
7267 | if (p) |
7268 | p->child = sd; | |
7c16ec58 | 7269 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7270 | |
1e9f28fa SS |
7271 | #ifdef CONFIG_SCHED_MC |
7272 | p = sd; | |
7273 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7274 | SD_INIT(sd, MC); |
1d3504fc | 7275 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7276 | sd->span = cpu_coregroup_map(i); |
7277 | cpus_and(sd->span, sd->span, *cpu_map); | |
7278 | sd->parent = p; | |
1a848870 | 7279 | p->child = sd; |
7c16ec58 | 7280 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7281 | #endif |
7282 | ||
1da177e4 LT |
7283 | #ifdef CONFIG_SCHED_SMT |
7284 | p = sd; | |
7285 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7286 | SD_INIT(sd, SIBLING); |
1d3504fc | 7287 | set_domain_attribute(sd, attr); |
d5a7430d | 7288 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7289 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7290 | sd->parent = p; |
1a848870 | 7291 | p->child = sd; |
7c16ec58 | 7292 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7293 | #endif |
7294 | } | |
7295 | ||
7296 | #ifdef CONFIG_SCHED_SMT | |
7297 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7298 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7299 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7300 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7301 | ||
7302 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7303 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7304 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7305 | continue; |
7306 | ||
dd41f596 | 7307 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7308 | &cpu_to_cpu_group, |
7309 | send_covered, tmpmask); | |
1da177e4 LT |
7310 | } |
7311 | #endif | |
7312 | ||
1e9f28fa SS |
7313 | #ifdef CONFIG_SCHED_MC |
7314 | /* Set up multi-core groups */ | |
363ab6f1 | 7315 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7316 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7317 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7318 | ||
7319 | *this_core_map = cpu_coregroup_map(i); | |
7320 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7321 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7322 | continue; |
7c16ec58 | 7323 | |
dd41f596 | 7324 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7325 | &cpu_to_core_group, |
7326 | send_covered, tmpmask); | |
1e9f28fa SS |
7327 | } |
7328 | #endif | |
7329 | ||
1da177e4 | 7330 | /* Set up physical groups */ |
076ac2af | 7331 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7332 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7333 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7334 | |
7c16ec58 MT |
7335 | *nodemask = node_to_cpumask(i); |
7336 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7337 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7338 | continue; |
7339 | ||
7c16ec58 MT |
7340 | init_sched_build_groups(nodemask, cpu_map, |
7341 | &cpu_to_phys_group, | |
7342 | send_covered, tmpmask); | |
1da177e4 LT |
7343 | } |
7344 | ||
7345 | #ifdef CONFIG_NUMA | |
7346 | /* Set up node groups */ | |
7c16ec58 MT |
7347 | if (sd_allnodes) { |
7348 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7349 | ||
7350 | init_sched_build_groups(cpu_map, cpu_map, | |
7351 | &cpu_to_allnodes_group, | |
7352 | send_covered, tmpmask); | |
7353 | } | |
9c1cfda2 | 7354 | |
076ac2af | 7355 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7356 | /* Set up node groups */ |
7357 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7358 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7359 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7360 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7361 | int j; |
7362 | ||
7c16ec58 MT |
7363 | *nodemask = node_to_cpumask(i); |
7364 | cpus_clear(*covered); | |
7365 | ||
7366 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7367 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7368 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7369 | continue; |
d1b55138 | 7370 | } |
9c1cfda2 | 7371 | |
4bdbaad3 | 7372 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7373 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7374 | |
15f0b676 | 7375 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7376 | if (!sg) { |
7377 | printk(KERN_WARNING "Can not alloc domain group for " | |
7378 | "node %d\n", i); | |
7379 | goto error; | |
7380 | } | |
9c1cfda2 | 7381 | sched_group_nodes[i] = sg; |
363ab6f1 | 7382 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7383 | struct sched_domain *sd; |
9761eea8 | 7384 | |
9c1cfda2 JH |
7385 | sd = &per_cpu(node_domains, j); |
7386 | sd->groups = sg; | |
9c1cfda2 | 7387 | } |
5517d86b | 7388 | sg->__cpu_power = 0; |
7c16ec58 | 7389 | sg->cpumask = *nodemask; |
51888ca2 | 7390 | sg->next = sg; |
7c16ec58 | 7391 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7392 | prev = sg; |
7393 | ||
076ac2af | 7394 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7395 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7396 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7397 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7398 | |
7c16ec58 MT |
7399 | cpus_complement(*notcovered, *covered); |
7400 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7401 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7402 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7403 | break; |
7404 | ||
7c16ec58 MT |
7405 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7406 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7407 | continue; |
7408 | ||
15f0b676 SV |
7409 | sg = kmalloc_node(sizeof(struct sched_group), |
7410 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7411 | if (!sg) { |
7412 | printk(KERN_WARNING | |
7413 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7414 | goto error; |
9c1cfda2 | 7415 | } |
5517d86b | 7416 | sg->__cpu_power = 0; |
7c16ec58 | 7417 | sg->cpumask = *tmpmask; |
51888ca2 | 7418 | sg->next = prev->next; |
7c16ec58 | 7419 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7420 | prev->next = sg; |
7421 | prev = sg; | |
7422 | } | |
9c1cfda2 | 7423 | } |
1da177e4 LT |
7424 | #endif |
7425 | ||
7426 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7427 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7428 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7429 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7430 | ||
89c4710e | 7431 | init_sched_groups_power(i, sd); |
5c45bf27 | 7432 | } |
1da177e4 | 7433 | #endif |
1e9f28fa | 7434 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7435 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7436 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7437 | ||
89c4710e | 7438 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7439 | } |
7440 | #endif | |
1e9f28fa | 7441 | |
363ab6f1 | 7442 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7443 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7444 | ||
89c4710e | 7445 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7446 | } |
7447 | ||
9c1cfda2 | 7448 | #ifdef CONFIG_NUMA |
076ac2af | 7449 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7450 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7451 | |
6711cab4 SS |
7452 | if (sd_allnodes) { |
7453 | struct sched_group *sg; | |
f712c0c7 | 7454 | |
7c16ec58 MT |
7455 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7456 | tmpmask); | |
f712c0c7 SS |
7457 | init_numa_sched_groups_power(sg); |
7458 | } | |
9c1cfda2 JH |
7459 | #endif |
7460 | ||
1da177e4 | 7461 | /* Attach the domains */ |
363ab6f1 | 7462 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7463 | struct sched_domain *sd; |
7464 | #ifdef CONFIG_SCHED_SMT | |
7465 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7466 | #elif defined(CONFIG_SCHED_MC) |
7467 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7468 | #else |
7469 | sd = &per_cpu(phys_domains, i); | |
7470 | #endif | |
57d885fe | 7471 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7472 | } |
51888ca2 | 7473 | |
7c16ec58 | 7474 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7475 | return 0; |
7476 | ||
a616058b | 7477 | #ifdef CONFIG_NUMA |
51888ca2 | 7478 | error: |
7c16ec58 MT |
7479 | free_sched_groups(cpu_map, tmpmask); |
7480 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7481 | return -ENOMEM; |
a616058b | 7482 | #endif |
1da177e4 | 7483 | } |
029190c5 | 7484 | |
1d3504fc HS |
7485 | static int build_sched_domains(const cpumask_t *cpu_map) |
7486 | { | |
7487 | return __build_sched_domains(cpu_map, NULL); | |
7488 | } | |
7489 | ||
029190c5 PJ |
7490 | static cpumask_t *doms_cur; /* current sched domains */ |
7491 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7492 | static struct sched_domain_attr *dattr_cur; |
7493 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7494 | |
7495 | /* | |
7496 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7497 | * cpumask_t) fails, then fallback to a single sched domain, | |
7498 | * as determined by the single cpumask_t fallback_doms. | |
7499 | */ | |
7500 | static cpumask_t fallback_doms; | |
7501 | ||
22e52b07 HC |
7502 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7503 | { | |
7504 | } | |
7505 | ||
1a20ff27 | 7506 | /* |
41a2d6cf | 7507 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7508 | * For now this just excludes isolated cpus, but could be used to |
7509 | * exclude other special cases in the future. | |
1a20ff27 | 7510 | */ |
51888ca2 | 7511 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7512 | { |
7378547f MM |
7513 | int err; |
7514 | ||
22e52b07 | 7515 | arch_update_cpu_topology(); |
029190c5 PJ |
7516 | ndoms_cur = 1; |
7517 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7518 | if (!doms_cur) | |
7519 | doms_cur = &fallback_doms; | |
7520 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7521 | dattr_cur = NULL; |
7378547f | 7522 | err = build_sched_domains(doms_cur); |
6382bc90 | 7523 | register_sched_domain_sysctl(); |
7378547f MM |
7524 | |
7525 | return err; | |
1a20ff27 DG |
7526 | } |
7527 | ||
7c16ec58 MT |
7528 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7529 | cpumask_t *tmpmask) | |
1da177e4 | 7530 | { |
7c16ec58 | 7531 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7532 | } |
1da177e4 | 7533 | |
1a20ff27 DG |
7534 | /* |
7535 | * Detach sched domains from a group of cpus specified in cpu_map | |
7536 | * These cpus will now be attached to the NULL domain | |
7537 | */ | |
858119e1 | 7538 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7539 | { |
7c16ec58 | 7540 | cpumask_t tmpmask; |
1a20ff27 DG |
7541 | int i; |
7542 | ||
6382bc90 MM |
7543 | unregister_sched_domain_sysctl(); |
7544 | ||
363ab6f1 | 7545 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7546 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7547 | synchronize_sched(); |
7c16ec58 | 7548 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7549 | } |
7550 | ||
1d3504fc HS |
7551 | /* handle null as "default" */ |
7552 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7553 | struct sched_domain_attr *new, int idx_new) | |
7554 | { | |
7555 | struct sched_domain_attr tmp; | |
7556 | ||
7557 | /* fast path */ | |
7558 | if (!new && !cur) | |
7559 | return 1; | |
7560 | ||
7561 | tmp = SD_ATTR_INIT; | |
7562 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7563 | new ? (new + idx_new) : &tmp, | |
7564 | sizeof(struct sched_domain_attr)); | |
7565 | } | |
7566 | ||
029190c5 PJ |
7567 | /* |
7568 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7569 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7570 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7571 | * It destroys each deleted domain and builds each new domain. | |
7572 | * | |
7573 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7574 | * The masks don't intersect (don't overlap.) We should setup one |
7575 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7576 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7577 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7578 | * it as it is. | |
7579 | * | |
41a2d6cf IM |
7580 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7581 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7582 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7583 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7584 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 PJ |
7585 | * |
7586 | * Call with hotplug lock held | |
7587 | */ | |
1d3504fc HS |
7588 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7589 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7590 | { |
7591 | int i, j; | |
7592 | ||
712555ee | 7593 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7594 | |
7378547f MM |
7595 | /* always unregister in case we don't destroy any domains */ |
7596 | unregister_sched_domain_sysctl(); | |
7597 | ||
e761b772 MK |
7598 | if (doms_new == NULL) |
7599 | ndoms_new = 0; | |
029190c5 PJ |
7600 | |
7601 | /* Destroy deleted domains */ | |
7602 | for (i = 0; i < ndoms_cur; i++) { | |
7603 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7604 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7605 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7606 | goto match1; |
7607 | } | |
7608 | /* no match - a current sched domain not in new doms_new[] */ | |
7609 | detach_destroy_domains(doms_cur + i); | |
7610 | match1: | |
7611 | ; | |
7612 | } | |
7613 | ||
e761b772 MK |
7614 | if (doms_new == NULL) { |
7615 | ndoms_cur = 0; | |
7616 | ndoms_new = 1; | |
7617 | doms_new = &fallback_doms; | |
7618 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7619 | dattr_new = NULL; | |
7620 | } | |
7621 | ||
029190c5 PJ |
7622 | /* Build new domains */ |
7623 | for (i = 0; i < ndoms_new; i++) { | |
7624 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7625 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7626 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7627 | goto match2; |
7628 | } | |
7629 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7630 | __build_sched_domains(doms_new + i, |
7631 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7632 | match2: |
7633 | ; | |
7634 | } | |
7635 | ||
7636 | /* Remember the new sched domains */ | |
7637 | if (doms_cur != &fallback_doms) | |
7638 | kfree(doms_cur); | |
1d3504fc | 7639 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7640 | doms_cur = doms_new; |
1d3504fc | 7641 | dattr_cur = dattr_new; |
029190c5 | 7642 | ndoms_cur = ndoms_new; |
7378547f MM |
7643 | |
7644 | register_sched_domain_sysctl(); | |
a1835615 | 7645 | |
712555ee | 7646 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7647 | } |
7648 | ||
5c45bf27 | 7649 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7650 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7651 | { |
95402b38 | 7652 | get_online_cpus(); |
e761b772 | 7653 | rebuild_sched_domains(); |
95402b38 | 7654 | put_online_cpus(); |
e761b772 | 7655 | return 0; |
5c45bf27 SS |
7656 | } |
7657 | ||
7658 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7659 | { | |
7660 | int ret; | |
7661 | ||
7662 | if (buf[0] != '0' && buf[0] != '1') | |
7663 | return -EINVAL; | |
7664 | ||
7665 | if (smt) | |
7666 | sched_smt_power_savings = (buf[0] == '1'); | |
7667 | else | |
7668 | sched_mc_power_savings = (buf[0] == '1'); | |
7669 | ||
7670 | ret = arch_reinit_sched_domains(); | |
7671 | ||
7672 | return ret ? ret : count; | |
7673 | } | |
7674 | ||
5c45bf27 | 7675 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7676 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7677 | char *page) | |
5c45bf27 SS |
7678 | { |
7679 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7680 | } | |
f718cd4a | 7681 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7682 | const char *buf, size_t count) |
5c45bf27 SS |
7683 | { |
7684 | return sched_power_savings_store(buf, count, 0); | |
7685 | } | |
f718cd4a AK |
7686 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7687 | sched_mc_power_savings_show, | |
7688 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7689 | #endif |
7690 | ||
7691 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7692 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7693 | char *page) | |
5c45bf27 SS |
7694 | { |
7695 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7696 | } | |
f718cd4a | 7697 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7698 | const char *buf, size_t count) |
5c45bf27 SS |
7699 | { |
7700 | return sched_power_savings_store(buf, count, 1); | |
7701 | } | |
f718cd4a AK |
7702 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7703 | sched_smt_power_savings_show, | |
6707de00 AB |
7704 | sched_smt_power_savings_store); |
7705 | #endif | |
7706 | ||
7707 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7708 | { | |
7709 | int err = 0; | |
7710 | ||
7711 | #ifdef CONFIG_SCHED_SMT | |
7712 | if (smt_capable()) | |
7713 | err = sysfs_create_file(&cls->kset.kobj, | |
7714 | &attr_sched_smt_power_savings.attr); | |
7715 | #endif | |
7716 | #ifdef CONFIG_SCHED_MC | |
7717 | if (!err && mc_capable()) | |
7718 | err = sysfs_create_file(&cls->kset.kobj, | |
7719 | &attr_sched_mc_power_savings.attr); | |
7720 | #endif | |
7721 | return err; | |
7722 | } | |
6d6bc0ad | 7723 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7724 | |
e761b772 | 7725 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7726 | /* |
e761b772 MK |
7727 | * Add online and remove offline CPUs from the scheduler domains. |
7728 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7729 | */ |
7730 | static int update_sched_domains(struct notifier_block *nfb, | |
7731 | unsigned long action, void *hcpu) | |
e761b772 MK |
7732 | { |
7733 | switch (action) { | |
7734 | case CPU_ONLINE: | |
7735 | case CPU_ONLINE_FROZEN: | |
7736 | case CPU_DEAD: | |
7737 | case CPU_DEAD_FROZEN: | |
7738 | partition_sched_domains(0, NULL, NULL); | |
7739 | return NOTIFY_OK; | |
7740 | ||
7741 | default: | |
7742 | return NOTIFY_DONE; | |
7743 | } | |
7744 | } | |
7745 | #endif | |
7746 | ||
7747 | static int update_runtime(struct notifier_block *nfb, | |
7748 | unsigned long action, void *hcpu) | |
1da177e4 | 7749 | { |
7def2be1 PZ |
7750 | int cpu = (int)(long)hcpu; |
7751 | ||
1da177e4 | 7752 | switch (action) { |
1da177e4 | 7753 | case CPU_DOWN_PREPARE: |
8bb78442 | 7754 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7755 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7756 | return NOTIFY_OK; |
7757 | ||
1da177e4 | 7758 | case CPU_DOWN_FAILED: |
8bb78442 | 7759 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7760 | case CPU_ONLINE: |
8bb78442 | 7761 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7762 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7763 | return NOTIFY_OK; |
7764 | ||
1da177e4 LT |
7765 | default: |
7766 | return NOTIFY_DONE; | |
7767 | } | |
1da177e4 | 7768 | } |
1da177e4 LT |
7769 | |
7770 | void __init sched_init_smp(void) | |
7771 | { | |
5c1e1767 NP |
7772 | cpumask_t non_isolated_cpus; |
7773 | ||
434d53b0 MT |
7774 | #if defined(CONFIG_NUMA) |
7775 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7776 | GFP_KERNEL); | |
7777 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7778 | #endif | |
95402b38 | 7779 | get_online_cpus(); |
712555ee | 7780 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7781 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7782 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7783 | if (cpus_empty(non_isolated_cpus)) |
7784 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7785 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7786 | put_online_cpus(); |
e761b772 MK |
7787 | |
7788 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7789 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7790 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7791 | #endif |
7792 | ||
7793 | /* RT runtime code needs to handle some hotplug events */ | |
7794 | hotcpu_notifier(update_runtime, 0); | |
7795 | ||
b328ca18 | 7796 | init_hrtick(); |
5c1e1767 NP |
7797 | |
7798 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7799 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7800 | BUG(); |
19978ca6 | 7801 | sched_init_granularity(); |
1da177e4 LT |
7802 | } |
7803 | #else | |
7804 | void __init sched_init_smp(void) | |
7805 | { | |
19978ca6 | 7806 | sched_init_granularity(); |
1da177e4 LT |
7807 | } |
7808 | #endif /* CONFIG_SMP */ | |
7809 | ||
7810 | int in_sched_functions(unsigned long addr) | |
7811 | { | |
1da177e4 LT |
7812 | return in_lock_functions(addr) || |
7813 | (addr >= (unsigned long)__sched_text_start | |
7814 | && addr < (unsigned long)__sched_text_end); | |
7815 | } | |
7816 | ||
a9957449 | 7817 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7818 | { |
7819 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7820 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7821 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7822 | cfs_rq->rq = rq; | |
7823 | #endif | |
67e9fb2a | 7824 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7825 | } |
7826 | ||
fa85ae24 PZ |
7827 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7828 | { | |
7829 | struct rt_prio_array *array; | |
7830 | int i; | |
7831 | ||
7832 | array = &rt_rq->active; | |
7833 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7834 | INIT_LIST_HEAD(array->queue + i); | |
7835 | __clear_bit(i, array->bitmap); | |
7836 | } | |
7837 | /* delimiter for bitsearch: */ | |
7838 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7839 | ||
052f1dc7 | 7840 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7841 | rt_rq->highest_prio = MAX_RT_PRIO; |
7842 | #endif | |
fa85ae24 PZ |
7843 | #ifdef CONFIG_SMP |
7844 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7845 | rt_rq->overloaded = 0; |
7846 | #endif | |
7847 | ||
7848 | rt_rq->rt_time = 0; | |
7849 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7850 | rt_rq->rt_runtime = 0; |
7851 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7852 | |
052f1dc7 | 7853 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7854 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7855 | rt_rq->rq = rq; |
7856 | #endif | |
fa85ae24 PZ |
7857 | } |
7858 | ||
6f505b16 | 7859 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7860 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7861 | struct sched_entity *se, int cpu, int add, | |
7862 | struct sched_entity *parent) | |
6f505b16 | 7863 | { |
ec7dc8ac | 7864 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7865 | tg->cfs_rq[cpu] = cfs_rq; |
7866 | init_cfs_rq(cfs_rq, rq); | |
7867 | cfs_rq->tg = tg; | |
7868 | if (add) | |
7869 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7870 | ||
7871 | tg->se[cpu] = se; | |
354d60c2 DG |
7872 | /* se could be NULL for init_task_group */ |
7873 | if (!se) | |
7874 | return; | |
7875 | ||
ec7dc8ac DG |
7876 | if (!parent) |
7877 | se->cfs_rq = &rq->cfs; | |
7878 | else | |
7879 | se->cfs_rq = parent->my_q; | |
7880 | ||
6f505b16 PZ |
7881 | se->my_q = cfs_rq; |
7882 | se->load.weight = tg->shares; | |
e05510d0 | 7883 | se->load.inv_weight = 0; |
ec7dc8ac | 7884 | se->parent = parent; |
6f505b16 | 7885 | } |
052f1dc7 | 7886 | #endif |
6f505b16 | 7887 | |
052f1dc7 | 7888 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7889 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7890 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7891 | struct sched_rt_entity *parent) | |
6f505b16 | 7892 | { |
ec7dc8ac DG |
7893 | struct rq *rq = cpu_rq(cpu); |
7894 | ||
6f505b16 PZ |
7895 | tg->rt_rq[cpu] = rt_rq; |
7896 | init_rt_rq(rt_rq, rq); | |
7897 | rt_rq->tg = tg; | |
7898 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7899 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7900 | if (add) |
7901 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7902 | ||
7903 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7904 | if (!rt_se) |
7905 | return; | |
7906 | ||
ec7dc8ac DG |
7907 | if (!parent) |
7908 | rt_se->rt_rq = &rq->rt; | |
7909 | else | |
7910 | rt_se->rt_rq = parent->my_q; | |
7911 | ||
6f505b16 | 7912 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7913 | rt_se->parent = parent; |
6f505b16 PZ |
7914 | INIT_LIST_HEAD(&rt_se->run_list); |
7915 | } | |
7916 | #endif | |
7917 | ||
1da177e4 LT |
7918 | void __init sched_init(void) |
7919 | { | |
dd41f596 | 7920 | int i, j; |
434d53b0 MT |
7921 | unsigned long alloc_size = 0, ptr; |
7922 | ||
7923 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7924 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7925 | #endif | |
7926 | #ifdef CONFIG_RT_GROUP_SCHED | |
7927 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7928 | #endif |
7929 | #ifdef CONFIG_USER_SCHED | |
7930 | alloc_size *= 2; | |
434d53b0 MT |
7931 | #endif |
7932 | /* | |
7933 | * As sched_init() is called before page_alloc is setup, | |
7934 | * we use alloc_bootmem(). | |
7935 | */ | |
7936 | if (alloc_size) { | |
5a9d3225 | 7937 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7938 | |
7939 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7940 | init_task_group.se = (struct sched_entity **)ptr; | |
7941 | ptr += nr_cpu_ids * sizeof(void **); | |
7942 | ||
7943 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7944 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7945 | |
7946 | #ifdef CONFIG_USER_SCHED | |
7947 | root_task_group.se = (struct sched_entity **)ptr; | |
7948 | ptr += nr_cpu_ids * sizeof(void **); | |
7949 | ||
7950 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7951 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7952 | #endif /* CONFIG_USER_SCHED */ |
7953 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7954 | #ifdef CONFIG_RT_GROUP_SCHED |
7955 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7956 | ptr += nr_cpu_ids * sizeof(void **); | |
7957 | ||
7958 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7959 | ptr += nr_cpu_ids * sizeof(void **); |
7960 | ||
7961 | #ifdef CONFIG_USER_SCHED | |
7962 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7963 | ptr += nr_cpu_ids * sizeof(void **); | |
7964 | ||
7965 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7966 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7967 | #endif /* CONFIG_USER_SCHED */ |
7968 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7969 | } |
dd41f596 | 7970 | |
57d885fe GH |
7971 | #ifdef CONFIG_SMP |
7972 | init_defrootdomain(); | |
7973 | #endif | |
7974 | ||
d0b27fa7 PZ |
7975 | init_rt_bandwidth(&def_rt_bandwidth, |
7976 | global_rt_period(), global_rt_runtime()); | |
7977 | ||
7978 | #ifdef CONFIG_RT_GROUP_SCHED | |
7979 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7980 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7981 | #ifdef CONFIG_USER_SCHED |
7982 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7983 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7984 | #endif /* CONFIG_USER_SCHED */ |
7985 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7986 | |
052f1dc7 | 7987 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7988 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7989 | INIT_LIST_HEAD(&init_task_group.children); |
7990 | ||
7991 | #ifdef CONFIG_USER_SCHED | |
7992 | INIT_LIST_HEAD(&root_task_group.children); | |
7993 | init_task_group.parent = &root_task_group; | |
7994 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7995 | #endif /* CONFIG_USER_SCHED */ |
7996 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7997 | |
0a945022 | 7998 | for_each_possible_cpu(i) { |
70b97a7f | 7999 | struct rq *rq; |
1da177e4 LT |
8000 | |
8001 | rq = cpu_rq(i); | |
8002 | spin_lock_init(&rq->lock); | |
fcb99371 | 8003 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 8004 | rq->nr_running = 0; |
dd41f596 | 8005 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8006 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8007 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8008 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8009 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8010 | #ifdef CONFIG_CGROUP_SCHED |
8011 | /* | |
8012 | * How much cpu bandwidth does init_task_group get? | |
8013 | * | |
8014 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8015 | * gets 100% of the cpu resources in the system. This overall | |
8016 | * system cpu resource is divided among the tasks of | |
8017 | * init_task_group and its child task-groups in a fair manner, | |
8018 | * based on each entity's (task or task-group's) weight | |
8019 | * (se->load.weight). | |
8020 | * | |
8021 | * In other words, if init_task_group has 10 tasks of weight | |
8022 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8023 | * then A0's share of the cpu resource is: | |
8024 | * | |
8025 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8026 | * | |
8027 | * We achieve this by letting init_task_group's tasks sit | |
8028 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8029 | */ | |
ec7dc8ac | 8030 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8031 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8032 | root_task_group.shares = NICE_0_LOAD; |
8033 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8034 | /* |
8035 | * In case of task-groups formed thr' the user id of tasks, | |
8036 | * init_task_group represents tasks belonging to root user. | |
8037 | * Hence it forms a sibling of all subsequent groups formed. | |
8038 | * In this case, init_task_group gets only a fraction of overall | |
8039 | * system cpu resource, based on the weight assigned to root | |
8040 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8041 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8042 | * (init_cfs_rq) and having one entity represent this group of | |
8043 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8044 | */ | |
ec7dc8ac | 8045 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8046 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8047 | &per_cpu(init_sched_entity, i), i, 1, |
8048 | root_task_group.se[i]); | |
6f505b16 | 8049 | |
052f1dc7 | 8050 | #endif |
354d60c2 DG |
8051 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8052 | ||
8053 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8054 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8055 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8056 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8057 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8058 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8059 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8060 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8061 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8062 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8063 | root_task_group.rt_se[i]); | |
354d60c2 | 8064 | #endif |
dd41f596 | 8065 | #endif |
1da177e4 | 8066 | |
dd41f596 IM |
8067 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8068 | rq->cpu_load[j] = 0; | |
1da177e4 | 8069 | #ifdef CONFIG_SMP |
41c7ce9a | 8070 | rq->sd = NULL; |
57d885fe | 8071 | rq->rd = NULL; |
1da177e4 | 8072 | rq->active_balance = 0; |
dd41f596 | 8073 | rq->next_balance = jiffies; |
1da177e4 | 8074 | rq->push_cpu = 0; |
0a2966b4 | 8075 | rq->cpu = i; |
1f11eb6a | 8076 | rq->online = 0; |
1da177e4 LT |
8077 | rq->migration_thread = NULL; |
8078 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8079 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8080 | #endif |
8f4d37ec | 8081 | init_rq_hrtick(rq); |
1da177e4 | 8082 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8083 | } |
8084 | ||
2dd73a4f | 8085 | set_load_weight(&init_task); |
b50f60ce | 8086 | |
e107be36 AK |
8087 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8088 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8089 | #endif | |
8090 | ||
c9819f45 | 8091 | #ifdef CONFIG_SMP |
962cf36c | 8092 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8093 | #endif |
8094 | ||
b50f60ce HC |
8095 | #ifdef CONFIG_RT_MUTEXES |
8096 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8097 | #endif | |
8098 | ||
1da177e4 LT |
8099 | /* |
8100 | * The boot idle thread does lazy MMU switching as well: | |
8101 | */ | |
8102 | atomic_inc(&init_mm.mm_count); | |
8103 | enter_lazy_tlb(&init_mm, current); | |
8104 | ||
8105 | /* | |
8106 | * Make us the idle thread. Technically, schedule() should not be | |
8107 | * called from this thread, however somewhere below it might be, | |
8108 | * but because we are the idle thread, we just pick up running again | |
8109 | * when this runqueue becomes "idle". | |
8110 | */ | |
8111 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8112 | /* |
8113 | * During early bootup we pretend to be a normal task: | |
8114 | */ | |
8115 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8116 | |
8117 | scheduler_running = 1; | |
1da177e4 LT |
8118 | } |
8119 | ||
8120 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8121 | void __might_sleep(char *file, int line) | |
8122 | { | |
48f24c4d | 8123 | #ifdef in_atomic |
1da177e4 LT |
8124 | static unsigned long prev_jiffy; /* ratelimiting */ |
8125 | ||
8126 | if ((in_atomic() || irqs_disabled()) && | |
8127 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8128 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8129 | return; | |
8130 | prev_jiffy = jiffies; | |
91368d73 | 8131 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8132 | " context at %s:%d\n", file, line); |
8133 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8134 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8135 | debug_show_held_locks(current); |
3117df04 IM |
8136 | if (irqs_disabled()) |
8137 | print_irqtrace_events(current); | |
1da177e4 LT |
8138 | dump_stack(); |
8139 | } | |
8140 | #endif | |
8141 | } | |
8142 | EXPORT_SYMBOL(__might_sleep); | |
8143 | #endif | |
8144 | ||
8145 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8146 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8147 | { | |
8148 | int on_rq; | |
3e51f33f | 8149 | |
3a5e4dc1 AK |
8150 | update_rq_clock(rq); |
8151 | on_rq = p->se.on_rq; | |
8152 | if (on_rq) | |
8153 | deactivate_task(rq, p, 0); | |
8154 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8155 | if (on_rq) { | |
8156 | activate_task(rq, p, 0); | |
8157 | resched_task(rq->curr); | |
8158 | } | |
8159 | } | |
8160 | ||
1da177e4 LT |
8161 | void normalize_rt_tasks(void) |
8162 | { | |
a0f98a1c | 8163 | struct task_struct *g, *p; |
1da177e4 | 8164 | unsigned long flags; |
70b97a7f | 8165 | struct rq *rq; |
1da177e4 | 8166 | |
4cf5d77a | 8167 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8168 | do_each_thread(g, p) { |
178be793 IM |
8169 | /* |
8170 | * Only normalize user tasks: | |
8171 | */ | |
8172 | if (!p->mm) | |
8173 | continue; | |
8174 | ||
6cfb0d5d | 8175 | p->se.exec_start = 0; |
6cfb0d5d | 8176 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8177 | p->se.wait_start = 0; |
dd41f596 | 8178 | p->se.sleep_start = 0; |
dd41f596 | 8179 | p->se.block_start = 0; |
6cfb0d5d | 8180 | #endif |
dd41f596 IM |
8181 | |
8182 | if (!rt_task(p)) { | |
8183 | /* | |
8184 | * Renice negative nice level userspace | |
8185 | * tasks back to 0: | |
8186 | */ | |
8187 | if (TASK_NICE(p) < 0 && p->mm) | |
8188 | set_user_nice(p, 0); | |
1da177e4 | 8189 | continue; |
dd41f596 | 8190 | } |
1da177e4 | 8191 | |
4cf5d77a | 8192 | spin_lock(&p->pi_lock); |
b29739f9 | 8193 | rq = __task_rq_lock(p); |
1da177e4 | 8194 | |
178be793 | 8195 | normalize_task(rq, p); |
3a5e4dc1 | 8196 | |
b29739f9 | 8197 | __task_rq_unlock(rq); |
4cf5d77a | 8198 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8199 | } while_each_thread(g, p); |
8200 | ||
4cf5d77a | 8201 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8202 | } |
8203 | ||
8204 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8205 | |
8206 | #ifdef CONFIG_IA64 | |
8207 | /* | |
8208 | * These functions are only useful for the IA64 MCA handling. | |
8209 | * | |
8210 | * They can only be called when the whole system has been | |
8211 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8212 | * activity can take place. Using them for anything else would | |
8213 | * be a serious bug, and as a result, they aren't even visible | |
8214 | * under any other configuration. | |
8215 | */ | |
8216 | ||
8217 | /** | |
8218 | * curr_task - return the current task for a given cpu. | |
8219 | * @cpu: the processor in question. | |
8220 | * | |
8221 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8222 | */ | |
36c8b586 | 8223 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8224 | { |
8225 | return cpu_curr(cpu); | |
8226 | } | |
8227 | ||
8228 | /** | |
8229 | * set_curr_task - set the current task for a given cpu. | |
8230 | * @cpu: the processor in question. | |
8231 | * @p: the task pointer to set. | |
8232 | * | |
8233 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8234 | * are serviced on a separate stack. It allows the architecture to switch the |
8235 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8236 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8237 | * and caller must save the original value of the current task (see | |
8238 | * curr_task() above) and restore that value before reenabling interrupts and | |
8239 | * re-starting the system. | |
8240 | * | |
8241 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8242 | */ | |
36c8b586 | 8243 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8244 | { |
8245 | cpu_curr(cpu) = p; | |
8246 | } | |
8247 | ||
8248 | #endif | |
29f59db3 | 8249 | |
bccbe08a PZ |
8250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8251 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8252 | { |
8253 | int i; | |
8254 | ||
8255 | for_each_possible_cpu(i) { | |
8256 | if (tg->cfs_rq) | |
8257 | kfree(tg->cfs_rq[i]); | |
8258 | if (tg->se) | |
8259 | kfree(tg->se[i]); | |
6f505b16 PZ |
8260 | } |
8261 | ||
8262 | kfree(tg->cfs_rq); | |
8263 | kfree(tg->se); | |
6f505b16 PZ |
8264 | } |
8265 | ||
ec7dc8ac DG |
8266 | static |
8267 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8268 | { |
29f59db3 | 8269 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8270 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8271 | struct rq *rq; |
29f59db3 SV |
8272 | int i; |
8273 | ||
434d53b0 | 8274 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8275 | if (!tg->cfs_rq) |
8276 | goto err; | |
434d53b0 | 8277 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8278 | if (!tg->se) |
8279 | goto err; | |
052f1dc7 PZ |
8280 | |
8281 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8282 | |
8283 | for_each_possible_cpu(i) { | |
9b5b7751 | 8284 | rq = cpu_rq(i); |
29f59db3 | 8285 | |
6f505b16 PZ |
8286 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8287 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8288 | if (!cfs_rq) |
8289 | goto err; | |
8290 | ||
6f505b16 PZ |
8291 | se = kmalloc_node(sizeof(struct sched_entity), |
8292 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8293 | if (!se) |
8294 | goto err; | |
8295 | ||
ec7dc8ac DG |
8296 | parent_se = parent ? parent->se[i] : NULL; |
8297 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8298 | } |
8299 | ||
8300 | return 1; | |
8301 | ||
8302 | err: | |
8303 | return 0; | |
8304 | } | |
8305 | ||
8306 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8307 | { | |
8308 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8309 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8310 | } | |
8311 | ||
8312 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8313 | { | |
8314 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8315 | } | |
6d6bc0ad | 8316 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8317 | static inline void free_fair_sched_group(struct task_group *tg) |
8318 | { | |
8319 | } | |
8320 | ||
ec7dc8ac DG |
8321 | static inline |
8322 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8323 | { |
8324 | return 1; | |
8325 | } | |
8326 | ||
8327 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8328 | { | |
8329 | } | |
8330 | ||
8331 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8332 | { | |
8333 | } | |
6d6bc0ad | 8334 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8335 | |
8336 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8337 | static void free_rt_sched_group(struct task_group *tg) |
8338 | { | |
8339 | int i; | |
8340 | ||
d0b27fa7 PZ |
8341 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8342 | ||
bccbe08a PZ |
8343 | for_each_possible_cpu(i) { |
8344 | if (tg->rt_rq) | |
8345 | kfree(tg->rt_rq[i]); | |
8346 | if (tg->rt_se) | |
8347 | kfree(tg->rt_se[i]); | |
8348 | } | |
8349 | ||
8350 | kfree(tg->rt_rq); | |
8351 | kfree(tg->rt_se); | |
8352 | } | |
8353 | ||
ec7dc8ac DG |
8354 | static |
8355 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8356 | { |
8357 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8358 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8359 | struct rq *rq; |
8360 | int i; | |
8361 | ||
434d53b0 | 8362 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8363 | if (!tg->rt_rq) |
8364 | goto err; | |
434d53b0 | 8365 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8366 | if (!tg->rt_se) |
8367 | goto err; | |
8368 | ||
d0b27fa7 PZ |
8369 | init_rt_bandwidth(&tg->rt_bandwidth, |
8370 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8371 | |
8372 | for_each_possible_cpu(i) { | |
8373 | rq = cpu_rq(i); | |
8374 | ||
6f505b16 PZ |
8375 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8376 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8377 | if (!rt_rq) | |
8378 | goto err; | |
29f59db3 | 8379 | |
6f505b16 PZ |
8380 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8381 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8382 | if (!rt_se) | |
8383 | goto err; | |
29f59db3 | 8384 | |
ec7dc8ac DG |
8385 | parent_se = parent ? parent->rt_se[i] : NULL; |
8386 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8387 | } |
8388 | ||
bccbe08a PZ |
8389 | return 1; |
8390 | ||
8391 | err: | |
8392 | return 0; | |
8393 | } | |
8394 | ||
8395 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8396 | { | |
8397 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8398 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8399 | } | |
8400 | ||
8401 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8402 | { | |
8403 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8404 | } | |
6d6bc0ad | 8405 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8406 | static inline void free_rt_sched_group(struct task_group *tg) |
8407 | { | |
8408 | } | |
8409 | ||
ec7dc8ac DG |
8410 | static inline |
8411 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8412 | { |
8413 | return 1; | |
8414 | } | |
8415 | ||
8416 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8417 | { | |
8418 | } | |
8419 | ||
8420 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8421 | { | |
8422 | } | |
6d6bc0ad | 8423 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8424 | |
d0b27fa7 | 8425 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8426 | static void free_sched_group(struct task_group *tg) |
8427 | { | |
8428 | free_fair_sched_group(tg); | |
8429 | free_rt_sched_group(tg); | |
8430 | kfree(tg); | |
8431 | } | |
8432 | ||
8433 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8434 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8435 | { |
8436 | struct task_group *tg; | |
8437 | unsigned long flags; | |
8438 | int i; | |
8439 | ||
8440 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8441 | if (!tg) | |
8442 | return ERR_PTR(-ENOMEM); | |
8443 | ||
ec7dc8ac | 8444 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8445 | goto err; |
8446 | ||
ec7dc8ac | 8447 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8448 | goto err; |
8449 | ||
8ed36996 | 8450 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8451 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8452 | register_fair_sched_group(tg, i); |
8453 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8454 | } |
6f505b16 | 8455 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8456 | |
8457 | WARN_ON(!parent); /* root should already exist */ | |
8458 | ||
8459 | tg->parent = parent; | |
8460 | list_add_rcu(&tg->siblings, &parent->children); | |
8461 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8462 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8463 | |
9b5b7751 | 8464 | return tg; |
29f59db3 SV |
8465 | |
8466 | err: | |
6f505b16 | 8467 | free_sched_group(tg); |
29f59db3 SV |
8468 | return ERR_PTR(-ENOMEM); |
8469 | } | |
8470 | ||
9b5b7751 | 8471 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8472 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8473 | { |
29f59db3 | 8474 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8475 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8476 | } |
8477 | ||
9b5b7751 | 8478 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8479 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8480 | { |
8ed36996 | 8481 | unsigned long flags; |
9b5b7751 | 8482 | int i; |
29f59db3 | 8483 | |
8ed36996 | 8484 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8485 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8486 | unregister_fair_sched_group(tg, i); |
8487 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8488 | } |
6f505b16 | 8489 | list_del_rcu(&tg->list); |
f473aa5e | 8490 | list_del_rcu(&tg->siblings); |
8ed36996 | 8491 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8492 | |
9b5b7751 | 8493 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8494 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8495 | } |
8496 | ||
9b5b7751 | 8497 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8498 | * The caller of this function should have put the task in its new group |
8499 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8500 | * reflect its new group. | |
9b5b7751 SV |
8501 | */ |
8502 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8503 | { |
8504 | int on_rq, running; | |
8505 | unsigned long flags; | |
8506 | struct rq *rq; | |
8507 | ||
8508 | rq = task_rq_lock(tsk, &flags); | |
8509 | ||
29f59db3 SV |
8510 | update_rq_clock(rq); |
8511 | ||
051a1d1a | 8512 | running = task_current(rq, tsk); |
29f59db3 SV |
8513 | on_rq = tsk->se.on_rq; |
8514 | ||
0e1f3483 | 8515 | if (on_rq) |
29f59db3 | 8516 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8517 | if (unlikely(running)) |
8518 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8519 | |
6f505b16 | 8520 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8521 | |
810b3817 PZ |
8522 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8523 | if (tsk->sched_class->moved_group) | |
8524 | tsk->sched_class->moved_group(tsk); | |
8525 | #endif | |
8526 | ||
0e1f3483 HS |
8527 | if (unlikely(running)) |
8528 | tsk->sched_class->set_curr_task(rq); | |
8529 | if (on_rq) | |
7074badb | 8530 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8531 | |
29f59db3 SV |
8532 | task_rq_unlock(rq, &flags); |
8533 | } | |
6d6bc0ad | 8534 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8535 | |
052f1dc7 | 8536 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8537 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8538 | { |
8539 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8540 | int on_rq; |
8541 | ||
29f59db3 | 8542 | on_rq = se->on_rq; |
62fb1851 | 8543 | if (on_rq) |
29f59db3 SV |
8544 | dequeue_entity(cfs_rq, se, 0); |
8545 | ||
8546 | se->load.weight = shares; | |
e05510d0 | 8547 | se->load.inv_weight = 0; |
29f59db3 | 8548 | |
62fb1851 | 8549 | if (on_rq) |
29f59db3 | 8550 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8551 | } |
62fb1851 | 8552 | |
c09595f6 PZ |
8553 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8554 | { | |
8555 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8556 | struct rq *rq = cfs_rq->rq; | |
8557 | unsigned long flags; | |
8558 | ||
8559 | spin_lock_irqsave(&rq->lock, flags); | |
8560 | __set_se_shares(se, shares); | |
8561 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8562 | } |
8563 | ||
8ed36996 PZ |
8564 | static DEFINE_MUTEX(shares_mutex); |
8565 | ||
4cf86d77 | 8566 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8567 | { |
8568 | int i; | |
8ed36996 | 8569 | unsigned long flags; |
c61935fd | 8570 | |
ec7dc8ac DG |
8571 | /* |
8572 | * We can't change the weight of the root cgroup. | |
8573 | */ | |
8574 | if (!tg->se[0]) | |
8575 | return -EINVAL; | |
8576 | ||
18d95a28 PZ |
8577 | if (shares < MIN_SHARES) |
8578 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8579 | else if (shares > MAX_SHARES) |
8580 | shares = MAX_SHARES; | |
62fb1851 | 8581 | |
8ed36996 | 8582 | mutex_lock(&shares_mutex); |
9b5b7751 | 8583 | if (tg->shares == shares) |
5cb350ba | 8584 | goto done; |
29f59db3 | 8585 | |
8ed36996 | 8586 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8587 | for_each_possible_cpu(i) |
8588 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8589 | list_del_rcu(&tg->siblings); |
8ed36996 | 8590 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8591 | |
8592 | /* wait for any ongoing reference to this group to finish */ | |
8593 | synchronize_sched(); | |
8594 | ||
8595 | /* | |
8596 | * Now we are free to modify the group's share on each cpu | |
8597 | * w/o tripping rebalance_share or load_balance_fair. | |
8598 | */ | |
9b5b7751 | 8599 | tg->shares = shares; |
c09595f6 PZ |
8600 | for_each_possible_cpu(i) { |
8601 | /* | |
8602 | * force a rebalance | |
8603 | */ | |
8604 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8605 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8606 | } |
29f59db3 | 8607 | |
6b2d7700 SV |
8608 | /* |
8609 | * Enable load balance activity on this group, by inserting it back on | |
8610 | * each cpu's rq->leaf_cfs_rq_list. | |
8611 | */ | |
8ed36996 | 8612 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8613 | for_each_possible_cpu(i) |
8614 | register_fair_sched_group(tg, i); | |
f473aa5e | 8615 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8616 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8617 | done: |
8ed36996 | 8618 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8619 | return 0; |
29f59db3 SV |
8620 | } |
8621 | ||
5cb350ba DG |
8622 | unsigned long sched_group_shares(struct task_group *tg) |
8623 | { | |
8624 | return tg->shares; | |
8625 | } | |
052f1dc7 | 8626 | #endif |
5cb350ba | 8627 | |
052f1dc7 | 8628 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8629 | /* |
9f0c1e56 | 8630 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8631 | */ |
9f0c1e56 PZ |
8632 | static DEFINE_MUTEX(rt_constraints_mutex); |
8633 | ||
8634 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8635 | { | |
8636 | if (runtime == RUNTIME_INF) | |
8637 | return 1ULL << 16; | |
8638 | ||
6f6d6a1a | 8639 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8640 | } |
8641 | ||
b40b2e8e PZ |
8642 | #ifdef CONFIG_CGROUP_SCHED |
8643 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8644 | { | |
10b612f4 | 8645 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8646 | unsigned long total = 0; |
8647 | ||
8648 | if (!parent) { | |
8649 | if (global_rt_period() < period) | |
8650 | return 0; | |
8651 | ||
8652 | return to_ratio(period, runtime) < | |
8653 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8654 | } | |
8655 | ||
8656 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8657 | return 0; | |
8658 | ||
8659 | rcu_read_lock(); | |
8660 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8661 | if (tgi == tg) | |
8662 | continue; | |
8663 | ||
8664 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8665 | tgi->rt_bandwidth.rt_runtime); | |
8666 | } | |
8667 | rcu_read_unlock(); | |
8668 | ||
10b612f4 | 8669 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8670 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8671 | parent->rt_bandwidth.rt_runtime); | |
8672 | } | |
8673 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8674 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8675 | { |
8676 | struct task_group *tgi; | |
8677 | unsigned long total = 0; | |
9f0c1e56 | 8678 | unsigned long global_ratio = |
d0b27fa7 | 8679 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8680 | |
8681 | rcu_read_lock(); | |
9f0c1e56 PZ |
8682 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8683 | if (tgi == tg) | |
8684 | continue; | |
6f505b16 | 8685 | |
d0b27fa7 PZ |
8686 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8687 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8688 | } |
8689 | rcu_read_unlock(); | |
6f505b16 | 8690 | |
9f0c1e56 | 8691 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8692 | } |
b40b2e8e | 8693 | #endif |
6f505b16 | 8694 | |
521f1a24 DG |
8695 | /* Must be called with tasklist_lock held */ |
8696 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8697 | { | |
8698 | struct task_struct *g, *p; | |
8699 | do_each_thread(g, p) { | |
8700 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8701 | return 1; | |
8702 | } while_each_thread(g, p); | |
8703 | return 0; | |
8704 | } | |
8705 | ||
d0b27fa7 PZ |
8706 | static int tg_set_bandwidth(struct task_group *tg, |
8707 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8708 | { |
ac086bc2 | 8709 | int i, err = 0; |
9f0c1e56 | 8710 | |
9f0c1e56 | 8711 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8712 | read_lock(&tasklist_lock); |
ac086bc2 | 8713 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8714 | err = -EBUSY; |
8715 | goto unlock; | |
8716 | } | |
9f0c1e56 PZ |
8717 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8718 | err = -EINVAL; | |
8719 | goto unlock; | |
8720 | } | |
ac086bc2 PZ |
8721 | |
8722 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8723 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8724 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8725 | |
8726 | for_each_possible_cpu(i) { | |
8727 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8728 | ||
8729 | spin_lock(&rt_rq->rt_runtime_lock); | |
8730 | rt_rq->rt_runtime = rt_runtime; | |
8731 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8732 | } | |
8733 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8734 | unlock: |
521f1a24 | 8735 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8736 | mutex_unlock(&rt_constraints_mutex); |
8737 | ||
8738 | return err; | |
6f505b16 PZ |
8739 | } |
8740 | ||
d0b27fa7 PZ |
8741 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8742 | { | |
8743 | u64 rt_runtime, rt_period; | |
8744 | ||
8745 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8746 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8747 | if (rt_runtime_us < 0) | |
8748 | rt_runtime = RUNTIME_INF; | |
8749 | ||
8750 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8751 | } | |
8752 | ||
9f0c1e56 PZ |
8753 | long sched_group_rt_runtime(struct task_group *tg) |
8754 | { | |
8755 | u64 rt_runtime_us; | |
8756 | ||
d0b27fa7 | 8757 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8758 | return -1; |
8759 | ||
d0b27fa7 | 8760 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8761 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8762 | return rt_runtime_us; | |
8763 | } | |
d0b27fa7 PZ |
8764 | |
8765 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8766 | { | |
8767 | u64 rt_runtime, rt_period; | |
8768 | ||
8769 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8770 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8771 | ||
619b0488 R |
8772 | if (rt_period == 0) |
8773 | return -EINVAL; | |
8774 | ||
d0b27fa7 PZ |
8775 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8776 | } | |
8777 | ||
8778 | long sched_group_rt_period(struct task_group *tg) | |
8779 | { | |
8780 | u64 rt_period_us; | |
8781 | ||
8782 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8783 | do_div(rt_period_us, NSEC_PER_USEC); | |
8784 | return rt_period_us; | |
8785 | } | |
8786 | ||
8787 | static int sched_rt_global_constraints(void) | |
8788 | { | |
10b612f4 PZ |
8789 | struct task_group *tg = &root_task_group; |
8790 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8791 | int ret = 0; |
8792 | ||
10b612f4 PZ |
8793 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8794 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8795 | ||
d0b27fa7 | 8796 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8797 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8798 | ret = -EINVAL; |
8799 | mutex_unlock(&rt_constraints_mutex); | |
8800 | ||
8801 | return ret; | |
8802 | } | |
6d6bc0ad | 8803 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8804 | static int sched_rt_global_constraints(void) |
8805 | { | |
ac086bc2 PZ |
8806 | unsigned long flags; |
8807 | int i; | |
8808 | ||
8809 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8810 | for_each_possible_cpu(i) { | |
8811 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8812 | ||
8813 | spin_lock(&rt_rq->rt_runtime_lock); | |
8814 | rt_rq->rt_runtime = global_rt_runtime(); | |
8815 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8816 | } | |
8817 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8818 | ||
d0b27fa7 PZ |
8819 | return 0; |
8820 | } | |
6d6bc0ad | 8821 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8822 | |
8823 | int sched_rt_handler(struct ctl_table *table, int write, | |
8824 | struct file *filp, void __user *buffer, size_t *lenp, | |
8825 | loff_t *ppos) | |
8826 | { | |
8827 | int ret; | |
8828 | int old_period, old_runtime; | |
8829 | static DEFINE_MUTEX(mutex); | |
8830 | ||
8831 | mutex_lock(&mutex); | |
8832 | old_period = sysctl_sched_rt_period; | |
8833 | old_runtime = sysctl_sched_rt_runtime; | |
8834 | ||
8835 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8836 | ||
8837 | if (!ret && write) { | |
8838 | ret = sched_rt_global_constraints(); | |
8839 | if (ret) { | |
8840 | sysctl_sched_rt_period = old_period; | |
8841 | sysctl_sched_rt_runtime = old_runtime; | |
8842 | } else { | |
8843 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8844 | def_rt_bandwidth.rt_period = | |
8845 | ns_to_ktime(global_rt_period()); | |
8846 | } | |
8847 | } | |
8848 | mutex_unlock(&mutex); | |
8849 | ||
8850 | return ret; | |
8851 | } | |
68318b8e | 8852 | |
052f1dc7 | 8853 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8854 | |
8855 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8856 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8857 | { |
2b01dfe3 PM |
8858 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8859 | struct task_group, css); | |
68318b8e SV |
8860 | } |
8861 | ||
8862 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8863 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8864 | { |
ec7dc8ac | 8865 | struct task_group *tg, *parent; |
68318b8e | 8866 | |
2b01dfe3 | 8867 | if (!cgrp->parent) { |
68318b8e | 8868 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8869 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8870 | return &init_task_group.css; |
8871 | } | |
8872 | ||
ec7dc8ac DG |
8873 | parent = cgroup_tg(cgrp->parent); |
8874 | tg = sched_create_group(parent); | |
68318b8e SV |
8875 | if (IS_ERR(tg)) |
8876 | return ERR_PTR(-ENOMEM); | |
8877 | ||
8878 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8879 | tg->css.cgroup = cgrp; |
68318b8e SV |
8880 | |
8881 | return &tg->css; | |
8882 | } | |
8883 | ||
41a2d6cf IM |
8884 | static void |
8885 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8886 | { |
2b01dfe3 | 8887 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8888 | |
8889 | sched_destroy_group(tg); | |
8890 | } | |
8891 | ||
41a2d6cf IM |
8892 | static int |
8893 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8894 | struct task_struct *tsk) | |
68318b8e | 8895 | { |
b68aa230 PZ |
8896 | #ifdef CONFIG_RT_GROUP_SCHED |
8897 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8898 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8899 | return -EINVAL; |
8900 | #else | |
68318b8e SV |
8901 | /* We don't support RT-tasks being in separate groups */ |
8902 | if (tsk->sched_class != &fair_sched_class) | |
8903 | return -EINVAL; | |
b68aa230 | 8904 | #endif |
68318b8e SV |
8905 | |
8906 | return 0; | |
8907 | } | |
8908 | ||
8909 | static void | |
2b01dfe3 | 8910 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8911 | struct cgroup *old_cont, struct task_struct *tsk) |
8912 | { | |
8913 | sched_move_task(tsk); | |
8914 | } | |
8915 | ||
052f1dc7 | 8916 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8917 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8918 | u64 shareval) |
68318b8e | 8919 | { |
2b01dfe3 | 8920 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8921 | } |
8922 | ||
f4c753b7 | 8923 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8924 | { |
2b01dfe3 | 8925 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8926 | |
8927 | return (u64) tg->shares; | |
8928 | } | |
6d6bc0ad | 8929 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8930 | |
052f1dc7 | 8931 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8932 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8933 | s64 val) |
6f505b16 | 8934 | { |
06ecb27c | 8935 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8936 | } |
8937 | ||
06ecb27c | 8938 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8939 | { |
06ecb27c | 8940 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8941 | } |
d0b27fa7 PZ |
8942 | |
8943 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8944 | u64 rt_period_us) | |
8945 | { | |
8946 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8947 | } | |
8948 | ||
8949 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8950 | { | |
8951 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8952 | } | |
6d6bc0ad | 8953 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8954 | |
fe5c7cc2 | 8955 | static struct cftype cpu_files[] = { |
052f1dc7 | 8956 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8957 | { |
8958 | .name = "shares", | |
f4c753b7 PM |
8959 | .read_u64 = cpu_shares_read_u64, |
8960 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8961 | }, |
052f1dc7 PZ |
8962 | #endif |
8963 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8964 | { |
9f0c1e56 | 8965 | .name = "rt_runtime_us", |
06ecb27c PM |
8966 | .read_s64 = cpu_rt_runtime_read, |
8967 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8968 | }, |
d0b27fa7 PZ |
8969 | { |
8970 | .name = "rt_period_us", | |
f4c753b7 PM |
8971 | .read_u64 = cpu_rt_period_read_uint, |
8972 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8973 | }, |
052f1dc7 | 8974 | #endif |
68318b8e SV |
8975 | }; |
8976 | ||
8977 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8978 | { | |
fe5c7cc2 | 8979 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8980 | } |
8981 | ||
8982 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8983 | .name = "cpu", |
8984 | .create = cpu_cgroup_create, | |
8985 | .destroy = cpu_cgroup_destroy, | |
8986 | .can_attach = cpu_cgroup_can_attach, | |
8987 | .attach = cpu_cgroup_attach, | |
8988 | .populate = cpu_cgroup_populate, | |
8989 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8990 | .early_init = 1, |
8991 | }; | |
8992 | ||
052f1dc7 | 8993 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8994 | |
8995 | #ifdef CONFIG_CGROUP_CPUACCT | |
8996 | ||
8997 | /* | |
8998 | * CPU accounting code for task groups. | |
8999 | * | |
9000 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9001 | * (balbir@in.ibm.com). | |
9002 | */ | |
9003 | ||
9004 | /* track cpu usage of a group of tasks */ | |
9005 | struct cpuacct { | |
9006 | struct cgroup_subsys_state css; | |
9007 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9008 | u64 *cpuusage; | |
9009 | }; | |
9010 | ||
9011 | struct cgroup_subsys cpuacct_subsys; | |
9012 | ||
9013 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9014 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9015 | { |
32cd756a | 9016 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9017 | struct cpuacct, css); |
9018 | } | |
9019 | ||
9020 | /* return cpu accounting group to which this task belongs */ | |
9021 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9022 | { | |
9023 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9024 | struct cpuacct, css); | |
9025 | } | |
9026 | ||
9027 | /* create a new cpu accounting group */ | |
9028 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9029 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9030 | { |
9031 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9032 | ||
9033 | if (!ca) | |
9034 | return ERR_PTR(-ENOMEM); | |
9035 | ||
9036 | ca->cpuusage = alloc_percpu(u64); | |
9037 | if (!ca->cpuusage) { | |
9038 | kfree(ca); | |
9039 | return ERR_PTR(-ENOMEM); | |
9040 | } | |
9041 | ||
9042 | return &ca->css; | |
9043 | } | |
9044 | ||
9045 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9046 | static void |
32cd756a | 9047 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9048 | { |
32cd756a | 9049 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9050 | |
9051 | free_percpu(ca->cpuusage); | |
9052 | kfree(ca); | |
9053 | } | |
9054 | ||
9055 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9056 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9057 | { |
32cd756a | 9058 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9059 | u64 totalcpuusage = 0; |
9060 | int i; | |
9061 | ||
9062 | for_each_possible_cpu(i) { | |
9063 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9064 | ||
9065 | /* | |
9066 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9067 | * platforms. | |
9068 | */ | |
9069 | spin_lock_irq(&cpu_rq(i)->lock); | |
9070 | totalcpuusage += *cpuusage; | |
9071 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9072 | } | |
9073 | ||
9074 | return totalcpuusage; | |
9075 | } | |
9076 | ||
0297b803 DG |
9077 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9078 | u64 reset) | |
9079 | { | |
9080 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9081 | int err = 0; | |
9082 | int i; | |
9083 | ||
9084 | if (reset) { | |
9085 | err = -EINVAL; | |
9086 | goto out; | |
9087 | } | |
9088 | ||
9089 | for_each_possible_cpu(i) { | |
9090 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9091 | ||
9092 | spin_lock_irq(&cpu_rq(i)->lock); | |
9093 | *cpuusage = 0; | |
9094 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9095 | } | |
9096 | out: | |
9097 | return err; | |
9098 | } | |
9099 | ||
d842de87 SV |
9100 | static struct cftype files[] = { |
9101 | { | |
9102 | .name = "usage", | |
f4c753b7 PM |
9103 | .read_u64 = cpuusage_read, |
9104 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9105 | }, |
9106 | }; | |
9107 | ||
32cd756a | 9108 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9109 | { |
32cd756a | 9110 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9111 | } |
9112 | ||
9113 | /* | |
9114 | * charge this task's execution time to its accounting group. | |
9115 | * | |
9116 | * called with rq->lock held. | |
9117 | */ | |
9118 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9119 | { | |
9120 | struct cpuacct *ca; | |
9121 | ||
9122 | if (!cpuacct_subsys.active) | |
9123 | return; | |
9124 | ||
9125 | ca = task_ca(tsk); | |
9126 | if (ca) { | |
9127 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9128 | ||
9129 | *cpuusage += cputime; | |
9130 | } | |
9131 | } | |
9132 | ||
9133 | struct cgroup_subsys cpuacct_subsys = { | |
9134 | .name = "cpuacct", | |
9135 | .create = cpuacct_create, | |
9136 | .destroy = cpuacct_destroy, | |
9137 | .populate = cpuacct_populate, | |
9138 | .subsys_id = cpuacct_subsys_id, | |
9139 | }; | |
9140 | #endif /* CONFIG_CGROUP_CPUACCT */ |