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 | ||
0b148fa0 PZ |
207 | static inline int rt_bandwidth_enabled(void); |
208 | ||
d0b27fa7 PZ |
209 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) |
210 | { | |
211 | ktime_t now; | |
212 | ||
0b148fa0 | 213 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
214 | return; |
215 | ||
216 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
217 | return; | |
218 | ||
219 | spin_lock(&rt_b->rt_runtime_lock); | |
220 | for (;;) { | |
221 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
222 | break; | |
223 | ||
224 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
225 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
226 | hrtimer_start(&rt_b->rt_period_timer, | |
227 | rt_b->rt_period_timer.expires, | |
228 | HRTIMER_MODE_ABS); | |
229 | } | |
230 | spin_unlock(&rt_b->rt_runtime_lock); | |
231 | } | |
232 | ||
233 | #ifdef CONFIG_RT_GROUP_SCHED | |
234 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
235 | { | |
236 | hrtimer_cancel(&rt_b->rt_period_timer); | |
237 | } | |
238 | #endif | |
239 | ||
712555ee HC |
240 | /* |
241 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
242 | * detach_destroy_domains and partition_sched_domains. | |
243 | */ | |
244 | static DEFINE_MUTEX(sched_domains_mutex); | |
245 | ||
052f1dc7 | 246 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 247 | |
68318b8e SV |
248 | #include <linux/cgroup.h> |
249 | ||
29f59db3 SV |
250 | struct cfs_rq; |
251 | ||
6f505b16 PZ |
252 | static LIST_HEAD(task_groups); |
253 | ||
29f59db3 | 254 | /* task group related information */ |
4cf86d77 | 255 | struct task_group { |
052f1dc7 | 256 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
257 | struct cgroup_subsys_state css; |
258 | #endif | |
052f1dc7 PZ |
259 | |
260 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
261 | /* schedulable entities of this group on each cpu */ |
262 | struct sched_entity **se; | |
263 | /* runqueue "owned" by this group on each cpu */ | |
264 | struct cfs_rq **cfs_rq; | |
265 | unsigned long shares; | |
052f1dc7 PZ |
266 | #endif |
267 | ||
268 | #ifdef CONFIG_RT_GROUP_SCHED | |
269 | struct sched_rt_entity **rt_se; | |
270 | struct rt_rq **rt_rq; | |
271 | ||
d0b27fa7 | 272 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 273 | #endif |
6b2d7700 | 274 | |
ae8393e5 | 275 | struct rcu_head rcu; |
6f505b16 | 276 | struct list_head list; |
f473aa5e PZ |
277 | |
278 | struct task_group *parent; | |
279 | struct list_head siblings; | |
280 | struct list_head children; | |
29f59db3 SV |
281 | }; |
282 | ||
354d60c2 | 283 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
284 | |
285 | /* | |
286 | * Root task group. | |
287 | * Every UID task group (including init_task_group aka UID-0) will | |
288 | * be a child to this group. | |
289 | */ | |
290 | struct task_group root_task_group; | |
291 | ||
052f1dc7 | 292 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
293 | /* Default task group's sched entity on each cpu */ |
294 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
295 | /* Default task group's cfs_rq on each cpu */ | |
296 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 297 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
298 | |
299 | #ifdef CONFIG_RT_GROUP_SCHED | |
300 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
301 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad DG |
302 | #endif /* CONFIG_RT_GROUP_SCHED */ |
303 | #else /* !CONFIG_FAIR_GROUP_SCHED */ | |
eff766a6 | 304 | #define root_task_group init_task_group |
6d6bc0ad | 305 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6f505b16 | 306 | |
8ed36996 | 307 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
308 | * a task group's cpu shares. |
309 | */ | |
8ed36996 | 310 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 311 | |
052f1dc7 | 312 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
313 | #ifdef CONFIG_USER_SCHED |
314 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 315 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 316 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 317 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 318 | |
cb4ad1ff | 319 | /* |
2e084786 LJ |
320 | * A weight of 0 or 1 can cause arithmetics problems. |
321 | * A weight of a cfs_rq is the sum of weights of which entities | |
322 | * are queued on this cfs_rq, so a weight of a entity should not be | |
323 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
324 | * (The default weight is 1024 - so there's no practical |
325 | * limitation from this.) | |
326 | */ | |
18d95a28 | 327 | #define MIN_SHARES 2 |
2e084786 | 328 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 329 | |
052f1dc7 PZ |
330 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
331 | #endif | |
332 | ||
29f59db3 | 333 | /* Default task group. |
3a252015 | 334 | * Every task in system belong to this group at bootup. |
29f59db3 | 335 | */ |
434d53b0 | 336 | struct task_group init_task_group; |
29f59db3 SV |
337 | |
338 | /* return group to which a task belongs */ | |
4cf86d77 | 339 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 340 | { |
4cf86d77 | 341 | struct task_group *tg; |
9b5b7751 | 342 | |
052f1dc7 | 343 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 344 | tg = p->user->tg; |
052f1dc7 | 345 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
346 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
347 | struct task_group, css); | |
24e377a8 | 348 | #else |
41a2d6cf | 349 | tg = &init_task_group; |
24e377a8 | 350 | #endif |
9b5b7751 | 351 | return tg; |
29f59db3 SV |
352 | } |
353 | ||
354 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 355 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 356 | { |
052f1dc7 | 357 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
358 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
359 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 360 | #endif |
6f505b16 | 361 | |
052f1dc7 | 362 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
363 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
364 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 365 | #endif |
29f59db3 SV |
366 | } |
367 | ||
368 | #else | |
369 | ||
6f505b16 | 370 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
371 | static inline struct task_group *task_group(struct task_struct *p) |
372 | { | |
373 | return NULL; | |
374 | } | |
29f59db3 | 375 | |
052f1dc7 | 376 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 377 | |
6aa645ea IM |
378 | /* CFS-related fields in a runqueue */ |
379 | struct cfs_rq { | |
380 | struct load_weight load; | |
381 | unsigned long nr_running; | |
382 | ||
6aa645ea | 383 | u64 exec_clock; |
e9acbff6 | 384 | u64 min_vruntime; |
103638d9 | 385 | u64 pair_start; |
6aa645ea IM |
386 | |
387 | struct rb_root tasks_timeline; | |
388 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
389 | |
390 | struct list_head tasks; | |
391 | struct list_head *balance_iterator; | |
392 | ||
393 | /* | |
394 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
395 | * It is set to NULL otherwise (i.e when none are currently running). |
396 | */ | |
aa2ac252 | 397 | struct sched_entity *curr, *next; |
ddc97297 PZ |
398 | |
399 | unsigned long nr_spread_over; | |
400 | ||
62160e3f | 401 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
402 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
403 | ||
41a2d6cf IM |
404 | /* |
405 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
406 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
407 | * (like users, containers etc.) | |
408 | * | |
409 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
410 | * list is used during load balance. | |
411 | */ | |
41a2d6cf IM |
412 | struct list_head leaf_cfs_rq_list; |
413 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
414 | |
415 | #ifdef CONFIG_SMP | |
c09595f6 | 416 | /* |
c8cba857 | 417 | * the part of load.weight contributed by tasks |
c09595f6 | 418 | */ |
c8cba857 | 419 | unsigned long task_weight; |
c09595f6 | 420 | |
c8cba857 PZ |
421 | /* |
422 | * h_load = weight * f(tg) | |
423 | * | |
424 | * Where f(tg) is the recursive weight fraction assigned to | |
425 | * this group. | |
426 | */ | |
427 | unsigned long h_load; | |
c09595f6 | 428 | |
c8cba857 PZ |
429 | /* |
430 | * this cpu's part of tg->shares | |
431 | */ | |
432 | unsigned long shares; | |
f1d239f7 PZ |
433 | |
434 | /* | |
435 | * load.weight at the time we set shares | |
436 | */ | |
437 | unsigned long rq_weight; | |
c09595f6 | 438 | #endif |
6aa645ea IM |
439 | #endif |
440 | }; | |
1da177e4 | 441 | |
6aa645ea IM |
442 | /* Real-Time classes' related field in a runqueue: */ |
443 | struct rt_rq { | |
444 | struct rt_prio_array active; | |
63489e45 | 445 | unsigned long rt_nr_running; |
052f1dc7 | 446 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
447 | int highest_prio; /* highest queued rt task prio */ |
448 | #endif | |
fa85ae24 | 449 | #ifdef CONFIG_SMP |
73fe6aae | 450 | unsigned long rt_nr_migratory; |
a22d7fc1 | 451 | int overloaded; |
fa85ae24 | 452 | #endif |
6f505b16 | 453 | int rt_throttled; |
fa85ae24 | 454 | u64 rt_time; |
ac086bc2 | 455 | u64 rt_runtime; |
ea736ed5 | 456 | /* Nests inside the rq lock: */ |
ac086bc2 | 457 | spinlock_t rt_runtime_lock; |
6f505b16 | 458 | |
052f1dc7 | 459 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
460 | unsigned long rt_nr_boosted; |
461 | ||
6f505b16 PZ |
462 | struct rq *rq; |
463 | struct list_head leaf_rt_rq_list; | |
464 | struct task_group *tg; | |
465 | struct sched_rt_entity *rt_se; | |
466 | #endif | |
6aa645ea IM |
467 | }; |
468 | ||
57d885fe GH |
469 | #ifdef CONFIG_SMP |
470 | ||
471 | /* | |
472 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
473 | * variables. Each exclusive cpuset essentially defines an island domain by |
474 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
475 | * exclusive cpuset is created, we also create and attach a new root-domain |
476 | * object. | |
477 | * | |
57d885fe GH |
478 | */ |
479 | struct root_domain { | |
480 | atomic_t refcount; | |
481 | cpumask_t span; | |
482 | cpumask_t online; | |
637f5085 | 483 | |
0eab9146 | 484 | /* |
637f5085 GH |
485 | * The "RT overload" flag: it gets set if a CPU has more than |
486 | * one runnable RT task. | |
487 | */ | |
488 | cpumask_t rto_mask; | |
0eab9146 | 489 | atomic_t rto_count; |
6e0534f2 GH |
490 | #ifdef CONFIG_SMP |
491 | struct cpupri cpupri; | |
492 | #endif | |
57d885fe GH |
493 | }; |
494 | ||
dc938520 GH |
495 | /* |
496 | * By default the system creates a single root-domain with all cpus as | |
497 | * members (mimicking the global state we have today). | |
498 | */ | |
57d885fe GH |
499 | static struct root_domain def_root_domain; |
500 | ||
501 | #endif | |
502 | ||
1da177e4 LT |
503 | /* |
504 | * This is the main, per-CPU runqueue data structure. | |
505 | * | |
506 | * Locking rule: those places that want to lock multiple runqueues | |
507 | * (such as the load balancing or the thread migration code), lock | |
508 | * acquire operations must be ordered by ascending &runqueue. | |
509 | */ | |
70b97a7f | 510 | struct rq { |
d8016491 IM |
511 | /* runqueue lock: */ |
512 | spinlock_t lock; | |
1da177e4 LT |
513 | |
514 | /* | |
515 | * nr_running and cpu_load should be in the same cacheline because | |
516 | * remote CPUs use both these fields when doing load calculation. | |
517 | */ | |
518 | unsigned long nr_running; | |
6aa645ea IM |
519 | #define CPU_LOAD_IDX_MAX 5 |
520 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 521 | unsigned char idle_at_tick; |
46cb4b7c | 522 | #ifdef CONFIG_NO_HZ |
15934a37 | 523 | unsigned long last_tick_seen; |
46cb4b7c SS |
524 | unsigned char in_nohz_recently; |
525 | #endif | |
d8016491 IM |
526 | /* capture load from *all* tasks on this cpu: */ |
527 | struct load_weight load; | |
6aa645ea IM |
528 | unsigned long nr_load_updates; |
529 | u64 nr_switches; | |
530 | ||
531 | struct cfs_rq cfs; | |
6f505b16 | 532 | struct rt_rq rt; |
6f505b16 | 533 | |
6aa645ea | 534 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
535 | /* list of leaf cfs_rq on this cpu: */ |
536 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
537 | #endif |
538 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 539 | struct list_head leaf_rt_rq_list; |
1da177e4 | 540 | #endif |
1da177e4 LT |
541 | |
542 | /* | |
543 | * This is part of a global counter where only the total sum | |
544 | * over all CPUs matters. A task can increase this counter on | |
545 | * one CPU and if it got migrated afterwards it may decrease | |
546 | * it on another CPU. Always updated under the runqueue lock: | |
547 | */ | |
548 | unsigned long nr_uninterruptible; | |
549 | ||
36c8b586 | 550 | struct task_struct *curr, *idle; |
c9819f45 | 551 | unsigned long next_balance; |
1da177e4 | 552 | struct mm_struct *prev_mm; |
6aa645ea | 553 | |
3e51f33f | 554 | u64 clock; |
6aa645ea | 555 | |
1da177e4 LT |
556 | atomic_t nr_iowait; |
557 | ||
558 | #ifdef CONFIG_SMP | |
0eab9146 | 559 | struct root_domain *rd; |
1da177e4 LT |
560 | struct sched_domain *sd; |
561 | ||
562 | /* For active balancing */ | |
563 | int active_balance; | |
564 | int push_cpu; | |
d8016491 IM |
565 | /* cpu of this runqueue: */ |
566 | int cpu; | |
1f11eb6a | 567 | int online; |
1da177e4 | 568 | |
a8a51d5e | 569 | unsigned long avg_load_per_task; |
1da177e4 | 570 | |
36c8b586 | 571 | struct task_struct *migration_thread; |
1da177e4 LT |
572 | struct list_head migration_queue; |
573 | #endif | |
574 | ||
8f4d37ec | 575 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
576 | #ifdef CONFIG_SMP |
577 | int hrtick_csd_pending; | |
578 | struct call_single_data hrtick_csd; | |
579 | #endif | |
8f4d37ec PZ |
580 | struct hrtimer hrtick_timer; |
581 | #endif | |
582 | ||
1da177e4 LT |
583 | #ifdef CONFIG_SCHEDSTATS |
584 | /* latency stats */ | |
585 | struct sched_info rq_sched_info; | |
586 | ||
587 | /* sys_sched_yield() stats */ | |
480b9434 KC |
588 | unsigned int yld_exp_empty; |
589 | unsigned int yld_act_empty; | |
590 | unsigned int yld_both_empty; | |
591 | unsigned int yld_count; | |
1da177e4 LT |
592 | |
593 | /* schedule() stats */ | |
480b9434 KC |
594 | unsigned int sched_switch; |
595 | unsigned int sched_count; | |
596 | unsigned int sched_goidle; | |
1da177e4 LT |
597 | |
598 | /* try_to_wake_up() stats */ | |
480b9434 KC |
599 | unsigned int ttwu_count; |
600 | unsigned int ttwu_local; | |
b8efb561 IM |
601 | |
602 | /* BKL stats */ | |
480b9434 | 603 | unsigned int bkl_count; |
1da177e4 LT |
604 | #endif |
605 | }; | |
606 | ||
f34e3b61 | 607 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 608 | |
dd41f596 IM |
609 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
610 | { | |
611 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
612 | } | |
613 | ||
0a2966b4 CL |
614 | static inline int cpu_of(struct rq *rq) |
615 | { | |
616 | #ifdef CONFIG_SMP | |
617 | return rq->cpu; | |
618 | #else | |
619 | return 0; | |
620 | #endif | |
621 | } | |
622 | ||
674311d5 NP |
623 | /* |
624 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 625 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
626 | * |
627 | * The domain tree of any CPU may only be accessed from within | |
628 | * preempt-disabled sections. | |
629 | */ | |
48f24c4d IM |
630 | #define for_each_domain(cpu, __sd) \ |
631 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
632 | |
633 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
634 | #define this_rq() (&__get_cpu_var(runqueues)) | |
635 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
636 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
637 | ||
3e51f33f PZ |
638 | static inline void update_rq_clock(struct rq *rq) |
639 | { | |
640 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
641 | } | |
642 | ||
bf5c91ba IM |
643 | /* |
644 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
645 | */ | |
646 | #ifdef CONFIG_SCHED_DEBUG | |
647 | # define const_debug __read_mostly | |
648 | #else | |
649 | # define const_debug static const | |
650 | #endif | |
651 | ||
017730c1 IM |
652 | /** |
653 | * runqueue_is_locked | |
654 | * | |
655 | * Returns true if the current cpu runqueue is locked. | |
656 | * This interface allows printk to be called with the runqueue lock | |
657 | * held and know whether or not it is OK to wake up the klogd. | |
658 | */ | |
659 | int runqueue_is_locked(void) | |
660 | { | |
661 | int cpu = get_cpu(); | |
662 | struct rq *rq = cpu_rq(cpu); | |
663 | int ret; | |
664 | ||
665 | ret = spin_is_locked(&rq->lock); | |
666 | put_cpu(); | |
667 | return ret; | |
668 | } | |
669 | ||
bf5c91ba IM |
670 | /* |
671 | * Debugging: various feature bits | |
672 | */ | |
f00b45c1 PZ |
673 | |
674 | #define SCHED_FEAT(name, enabled) \ | |
675 | __SCHED_FEAT_##name , | |
676 | ||
bf5c91ba | 677 | enum { |
f00b45c1 | 678 | #include "sched_features.h" |
bf5c91ba IM |
679 | }; |
680 | ||
f00b45c1 PZ |
681 | #undef SCHED_FEAT |
682 | ||
683 | #define SCHED_FEAT(name, enabled) \ | |
684 | (1UL << __SCHED_FEAT_##name) * enabled | | |
685 | ||
bf5c91ba | 686 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
687 | #include "sched_features.h" |
688 | 0; | |
689 | ||
690 | #undef SCHED_FEAT | |
691 | ||
692 | #ifdef CONFIG_SCHED_DEBUG | |
693 | #define SCHED_FEAT(name, enabled) \ | |
694 | #name , | |
695 | ||
983ed7a6 | 696 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
697 | #include "sched_features.h" |
698 | NULL | |
699 | }; | |
700 | ||
701 | #undef SCHED_FEAT | |
702 | ||
983ed7a6 | 703 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
704 | { |
705 | filp->private_data = inode->i_private; | |
706 | return 0; | |
707 | } | |
708 | ||
709 | static ssize_t | |
710 | sched_feat_read(struct file *filp, char __user *ubuf, | |
711 | size_t cnt, loff_t *ppos) | |
712 | { | |
713 | char *buf; | |
714 | int r = 0; | |
715 | int len = 0; | |
716 | int i; | |
717 | ||
718 | for (i = 0; sched_feat_names[i]; i++) { | |
719 | len += strlen(sched_feat_names[i]); | |
720 | len += 4; | |
721 | } | |
722 | ||
723 | buf = kmalloc(len + 2, GFP_KERNEL); | |
724 | if (!buf) | |
725 | return -ENOMEM; | |
726 | ||
727 | for (i = 0; sched_feat_names[i]; i++) { | |
728 | if (sysctl_sched_features & (1UL << i)) | |
729 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
730 | else | |
c24b7c52 | 731 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
732 | } |
733 | ||
734 | r += sprintf(buf + r, "\n"); | |
735 | WARN_ON(r >= len + 2); | |
736 | ||
737 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
738 | ||
739 | kfree(buf); | |
740 | ||
741 | return r; | |
742 | } | |
743 | ||
744 | static ssize_t | |
745 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
746 | size_t cnt, loff_t *ppos) | |
747 | { | |
748 | char buf[64]; | |
749 | char *cmp = buf; | |
750 | int neg = 0; | |
751 | int i; | |
752 | ||
753 | if (cnt > 63) | |
754 | cnt = 63; | |
755 | ||
756 | if (copy_from_user(&buf, ubuf, cnt)) | |
757 | return -EFAULT; | |
758 | ||
759 | buf[cnt] = 0; | |
760 | ||
c24b7c52 | 761 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
762 | neg = 1; |
763 | cmp += 3; | |
764 | } | |
765 | ||
766 | for (i = 0; sched_feat_names[i]; i++) { | |
767 | int len = strlen(sched_feat_names[i]); | |
768 | ||
769 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
770 | if (neg) | |
771 | sysctl_sched_features &= ~(1UL << i); | |
772 | else | |
773 | sysctl_sched_features |= (1UL << i); | |
774 | break; | |
775 | } | |
776 | } | |
777 | ||
778 | if (!sched_feat_names[i]) | |
779 | return -EINVAL; | |
780 | ||
781 | filp->f_pos += cnt; | |
782 | ||
783 | return cnt; | |
784 | } | |
785 | ||
786 | static struct file_operations sched_feat_fops = { | |
787 | .open = sched_feat_open, | |
788 | .read = sched_feat_read, | |
789 | .write = sched_feat_write, | |
790 | }; | |
791 | ||
792 | static __init int sched_init_debug(void) | |
793 | { | |
f00b45c1 PZ |
794 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
795 | &sched_feat_fops); | |
796 | ||
797 | return 0; | |
798 | } | |
799 | late_initcall(sched_init_debug); | |
800 | ||
801 | #endif | |
802 | ||
803 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 804 | |
b82d9fdd PZ |
805 | /* |
806 | * Number of tasks to iterate in a single balance run. | |
807 | * Limited because this is done with IRQs disabled. | |
808 | */ | |
809 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
810 | ||
2398f2c6 PZ |
811 | /* |
812 | * ratelimit for updating the group shares. | |
55cd5340 | 813 | * default: 0.25ms |
2398f2c6 | 814 | */ |
55cd5340 | 815 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 816 | |
fa85ae24 | 817 | /* |
9f0c1e56 | 818 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
819 | * default: 1s |
820 | */ | |
9f0c1e56 | 821 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 822 | |
6892b75e IM |
823 | static __read_mostly int scheduler_running; |
824 | ||
9f0c1e56 PZ |
825 | /* |
826 | * part of the period that we allow rt tasks to run in us. | |
827 | * default: 0.95s | |
828 | */ | |
829 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 830 | |
d0b27fa7 PZ |
831 | static inline u64 global_rt_period(void) |
832 | { | |
833 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
834 | } | |
835 | ||
836 | static inline u64 global_rt_runtime(void) | |
837 | { | |
e26873bb | 838 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
839 | return RUNTIME_INF; |
840 | ||
841 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
842 | } | |
fa85ae24 | 843 | |
0b148fa0 PZ |
844 | static inline int rt_bandwidth_enabled(void) |
845 | { | |
846 | return sysctl_sched_rt_runtime >= 0; | |
847 | } | |
848 | ||
1da177e4 | 849 | #ifndef prepare_arch_switch |
4866cde0 NP |
850 | # define prepare_arch_switch(next) do { } while (0) |
851 | #endif | |
852 | #ifndef finish_arch_switch | |
853 | # define finish_arch_switch(prev) do { } while (0) | |
854 | #endif | |
855 | ||
051a1d1a DA |
856 | static inline int task_current(struct rq *rq, struct task_struct *p) |
857 | { | |
858 | return rq->curr == p; | |
859 | } | |
860 | ||
4866cde0 | 861 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 862 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 863 | { |
051a1d1a | 864 | return task_current(rq, p); |
4866cde0 NP |
865 | } |
866 | ||
70b97a7f | 867 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
868 | { |
869 | } | |
870 | ||
70b97a7f | 871 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 872 | { |
da04c035 IM |
873 | #ifdef CONFIG_DEBUG_SPINLOCK |
874 | /* this is a valid case when another task releases the spinlock */ | |
875 | rq->lock.owner = current; | |
876 | #endif | |
8a25d5de IM |
877 | /* |
878 | * If we are tracking spinlock dependencies then we have to | |
879 | * fix up the runqueue lock - which gets 'carried over' from | |
880 | * prev into current: | |
881 | */ | |
882 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
883 | ||
4866cde0 NP |
884 | spin_unlock_irq(&rq->lock); |
885 | } | |
886 | ||
887 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 888 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
889 | { |
890 | #ifdef CONFIG_SMP | |
891 | return p->oncpu; | |
892 | #else | |
051a1d1a | 893 | return task_current(rq, p); |
4866cde0 NP |
894 | #endif |
895 | } | |
896 | ||
70b97a7f | 897 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
898 | { |
899 | #ifdef CONFIG_SMP | |
900 | /* | |
901 | * We can optimise this out completely for !SMP, because the | |
902 | * SMP rebalancing from interrupt is the only thing that cares | |
903 | * here. | |
904 | */ | |
905 | next->oncpu = 1; | |
906 | #endif | |
907 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
908 | spin_unlock_irq(&rq->lock); | |
909 | #else | |
910 | spin_unlock(&rq->lock); | |
911 | #endif | |
912 | } | |
913 | ||
70b97a7f | 914 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
915 | { |
916 | #ifdef CONFIG_SMP | |
917 | /* | |
918 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
919 | * We must ensure this doesn't happen until the switch is completely | |
920 | * finished. | |
921 | */ | |
922 | smp_wmb(); | |
923 | prev->oncpu = 0; | |
924 | #endif | |
925 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
926 | local_irq_enable(); | |
1da177e4 | 927 | #endif |
4866cde0 NP |
928 | } |
929 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 930 | |
b29739f9 IM |
931 | /* |
932 | * __task_rq_lock - lock the runqueue a given task resides on. | |
933 | * Must be called interrupts disabled. | |
934 | */ | |
70b97a7f | 935 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
936 | __acquires(rq->lock) |
937 | { | |
3a5c359a AK |
938 | for (;;) { |
939 | struct rq *rq = task_rq(p); | |
940 | spin_lock(&rq->lock); | |
941 | if (likely(rq == task_rq(p))) | |
942 | return rq; | |
b29739f9 | 943 | spin_unlock(&rq->lock); |
b29739f9 | 944 | } |
b29739f9 IM |
945 | } |
946 | ||
1da177e4 LT |
947 | /* |
948 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 949 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
950 | * explicitly disabling preemption. |
951 | */ | |
70b97a7f | 952 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
953 | __acquires(rq->lock) |
954 | { | |
70b97a7f | 955 | struct rq *rq; |
1da177e4 | 956 | |
3a5c359a AK |
957 | for (;;) { |
958 | local_irq_save(*flags); | |
959 | rq = task_rq(p); | |
960 | spin_lock(&rq->lock); | |
961 | if (likely(rq == task_rq(p))) | |
962 | return rq; | |
1da177e4 | 963 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 964 | } |
1da177e4 LT |
965 | } |
966 | ||
a9957449 | 967 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
968 | __releases(rq->lock) |
969 | { | |
970 | spin_unlock(&rq->lock); | |
971 | } | |
972 | ||
70b97a7f | 973 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
974 | __releases(rq->lock) |
975 | { | |
976 | spin_unlock_irqrestore(&rq->lock, *flags); | |
977 | } | |
978 | ||
1da177e4 | 979 | /* |
cc2a73b5 | 980 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 981 | */ |
a9957449 | 982 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
983 | __acquires(rq->lock) |
984 | { | |
70b97a7f | 985 | struct rq *rq; |
1da177e4 LT |
986 | |
987 | local_irq_disable(); | |
988 | rq = this_rq(); | |
989 | spin_lock(&rq->lock); | |
990 | ||
991 | return rq; | |
992 | } | |
993 | ||
8f4d37ec PZ |
994 | #ifdef CONFIG_SCHED_HRTICK |
995 | /* | |
996 | * Use HR-timers to deliver accurate preemption points. | |
997 | * | |
998 | * Its all a bit involved since we cannot program an hrt while holding the | |
999 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1000 | * reschedule event. | |
1001 | * | |
1002 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1003 | * rq->lock. | |
1004 | */ | |
8f4d37ec PZ |
1005 | |
1006 | /* | |
1007 | * Use hrtick when: | |
1008 | * - enabled by features | |
1009 | * - hrtimer is actually high res | |
1010 | */ | |
1011 | static inline int hrtick_enabled(struct rq *rq) | |
1012 | { | |
1013 | if (!sched_feat(HRTICK)) | |
1014 | return 0; | |
ba42059f | 1015 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1016 | return 0; |
8f4d37ec PZ |
1017 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1018 | } | |
1019 | ||
8f4d37ec PZ |
1020 | static void hrtick_clear(struct rq *rq) |
1021 | { | |
1022 | if (hrtimer_active(&rq->hrtick_timer)) | |
1023 | hrtimer_cancel(&rq->hrtick_timer); | |
1024 | } | |
1025 | ||
8f4d37ec PZ |
1026 | /* |
1027 | * High-resolution timer tick. | |
1028 | * Runs from hardirq context with interrupts disabled. | |
1029 | */ | |
1030 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1031 | { | |
1032 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1033 | ||
1034 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1035 | ||
1036 | spin_lock(&rq->lock); | |
3e51f33f | 1037 | update_rq_clock(rq); |
8f4d37ec PZ |
1038 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1039 | spin_unlock(&rq->lock); | |
1040 | ||
1041 | return HRTIMER_NORESTART; | |
1042 | } | |
1043 | ||
95e904c7 | 1044 | #ifdef CONFIG_SMP |
31656519 PZ |
1045 | /* |
1046 | * called from hardirq (IPI) context | |
1047 | */ | |
1048 | static void __hrtick_start(void *arg) | |
b328ca18 | 1049 | { |
31656519 | 1050 | struct rq *rq = arg; |
b328ca18 | 1051 | |
31656519 PZ |
1052 | spin_lock(&rq->lock); |
1053 | hrtimer_restart(&rq->hrtick_timer); | |
1054 | rq->hrtick_csd_pending = 0; | |
1055 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1056 | } |
1057 | ||
31656519 PZ |
1058 | /* |
1059 | * Called to set the hrtick timer state. | |
1060 | * | |
1061 | * called with rq->lock held and irqs disabled | |
1062 | */ | |
1063 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1064 | { |
31656519 PZ |
1065 | struct hrtimer *timer = &rq->hrtick_timer; |
1066 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1067 | |
31656519 PZ |
1068 | timer->expires = time; |
1069 | ||
1070 | if (rq == this_rq()) { | |
1071 | hrtimer_restart(timer); | |
1072 | } else if (!rq->hrtick_csd_pending) { | |
1073 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1074 | rq->hrtick_csd_pending = 1; | |
1075 | } | |
b328ca18 PZ |
1076 | } |
1077 | ||
1078 | static int | |
1079 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1080 | { | |
1081 | int cpu = (int)(long)hcpu; | |
1082 | ||
1083 | switch (action) { | |
1084 | case CPU_UP_CANCELED: | |
1085 | case CPU_UP_CANCELED_FROZEN: | |
1086 | case CPU_DOWN_PREPARE: | |
1087 | case CPU_DOWN_PREPARE_FROZEN: | |
1088 | case CPU_DEAD: | |
1089 | case CPU_DEAD_FROZEN: | |
31656519 | 1090 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1091 | return NOTIFY_OK; |
1092 | } | |
1093 | ||
1094 | return NOTIFY_DONE; | |
1095 | } | |
1096 | ||
1097 | static void init_hrtick(void) | |
1098 | { | |
1099 | hotcpu_notifier(hotplug_hrtick, 0); | |
1100 | } | |
31656519 PZ |
1101 | #else |
1102 | /* | |
1103 | * Called to set the hrtick timer state. | |
1104 | * | |
1105 | * called with rq->lock held and irqs disabled | |
1106 | */ | |
1107 | static void hrtick_start(struct rq *rq, u64 delay) | |
1108 | { | |
1109 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1110 | } | |
b328ca18 | 1111 | |
31656519 | 1112 | static void init_hrtick(void) |
8f4d37ec | 1113 | { |
8f4d37ec | 1114 | } |
31656519 | 1115 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1116 | |
31656519 | 1117 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1118 | { |
31656519 PZ |
1119 | #ifdef CONFIG_SMP |
1120 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1121 | |
31656519 PZ |
1122 | rq->hrtick_csd.flags = 0; |
1123 | rq->hrtick_csd.func = __hrtick_start; | |
1124 | rq->hrtick_csd.info = rq; | |
1125 | #endif | |
8f4d37ec | 1126 | |
31656519 PZ |
1127 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1128 | rq->hrtick_timer.function = hrtick; | |
1129 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
8f4d37ec PZ |
1130 | } |
1131 | #else | |
1132 | static inline void hrtick_clear(struct rq *rq) | |
1133 | { | |
1134 | } | |
1135 | ||
8f4d37ec PZ |
1136 | static inline void init_rq_hrtick(struct rq *rq) |
1137 | { | |
1138 | } | |
1139 | ||
b328ca18 PZ |
1140 | static inline void init_hrtick(void) |
1141 | { | |
1142 | } | |
8f4d37ec PZ |
1143 | #endif |
1144 | ||
c24d20db IM |
1145 | /* |
1146 | * resched_task - mark a task 'to be rescheduled now'. | |
1147 | * | |
1148 | * On UP this means the setting of the need_resched flag, on SMP it | |
1149 | * might also involve a cross-CPU call to trigger the scheduler on | |
1150 | * the target CPU. | |
1151 | */ | |
1152 | #ifdef CONFIG_SMP | |
1153 | ||
1154 | #ifndef tsk_is_polling | |
1155 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1156 | #endif | |
1157 | ||
31656519 | 1158 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1159 | { |
1160 | int cpu; | |
1161 | ||
1162 | assert_spin_locked(&task_rq(p)->lock); | |
1163 | ||
31656519 | 1164 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1165 | return; |
1166 | ||
31656519 | 1167 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1168 | |
1169 | cpu = task_cpu(p); | |
1170 | if (cpu == smp_processor_id()) | |
1171 | return; | |
1172 | ||
1173 | /* NEED_RESCHED must be visible before we test polling */ | |
1174 | smp_mb(); | |
1175 | if (!tsk_is_polling(p)) | |
1176 | smp_send_reschedule(cpu); | |
1177 | } | |
1178 | ||
1179 | static void resched_cpu(int cpu) | |
1180 | { | |
1181 | struct rq *rq = cpu_rq(cpu); | |
1182 | unsigned long flags; | |
1183 | ||
1184 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1185 | return; | |
1186 | resched_task(cpu_curr(cpu)); | |
1187 | spin_unlock_irqrestore(&rq->lock, flags); | |
1188 | } | |
06d8308c TG |
1189 | |
1190 | #ifdef CONFIG_NO_HZ | |
1191 | /* | |
1192 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1193 | * idle CPU then this timer might expire before the next timer event | |
1194 | * which is scheduled to wake up that CPU. In case of a completely | |
1195 | * idle system the next event might even be infinite time into the | |
1196 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1197 | * leaves the inner idle loop so the newly added timer is taken into | |
1198 | * account when the CPU goes back to idle and evaluates the timer | |
1199 | * wheel for the next timer event. | |
1200 | */ | |
1201 | void wake_up_idle_cpu(int cpu) | |
1202 | { | |
1203 | struct rq *rq = cpu_rq(cpu); | |
1204 | ||
1205 | if (cpu == smp_processor_id()) | |
1206 | return; | |
1207 | ||
1208 | /* | |
1209 | * This is safe, as this function is called with the timer | |
1210 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1211 | * to idle and has not yet set rq->curr to idle then it will | |
1212 | * be serialized on the timer wheel base lock and take the new | |
1213 | * timer into account automatically. | |
1214 | */ | |
1215 | if (rq->curr != rq->idle) | |
1216 | return; | |
1217 | ||
1218 | /* | |
1219 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1220 | * lockless. The worst case is that the other CPU runs the | |
1221 | * idle task through an additional NOOP schedule() | |
1222 | */ | |
1223 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1224 | ||
1225 | /* NEED_RESCHED must be visible before we test polling */ | |
1226 | smp_mb(); | |
1227 | if (!tsk_is_polling(rq->idle)) | |
1228 | smp_send_reschedule(cpu); | |
1229 | } | |
6d6bc0ad | 1230 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1231 | |
6d6bc0ad | 1232 | #else /* !CONFIG_SMP */ |
31656519 | 1233 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1234 | { |
1235 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1236 | set_tsk_need_resched(p); |
c24d20db | 1237 | } |
6d6bc0ad | 1238 | #endif /* CONFIG_SMP */ |
c24d20db | 1239 | |
45bf76df IM |
1240 | #if BITS_PER_LONG == 32 |
1241 | # define WMULT_CONST (~0UL) | |
1242 | #else | |
1243 | # define WMULT_CONST (1UL << 32) | |
1244 | #endif | |
1245 | ||
1246 | #define WMULT_SHIFT 32 | |
1247 | ||
194081eb IM |
1248 | /* |
1249 | * Shift right and round: | |
1250 | */ | |
cf2ab469 | 1251 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1252 | |
a7be37ac PZ |
1253 | /* |
1254 | * delta *= weight / lw | |
1255 | */ | |
cb1c4fc9 | 1256 | static unsigned long |
45bf76df IM |
1257 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1258 | struct load_weight *lw) | |
1259 | { | |
1260 | u64 tmp; | |
1261 | ||
7a232e03 LJ |
1262 | if (!lw->inv_weight) { |
1263 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1264 | lw->inv_weight = 1; | |
1265 | else | |
1266 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1267 | / (lw->weight+1); | |
1268 | } | |
45bf76df IM |
1269 | |
1270 | tmp = (u64)delta_exec * weight; | |
1271 | /* | |
1272 | * Check whether we'd overflow the 64-bit multiplication: | |
1273 | */ | |
194081eb | 1274 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1275 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1276 | WMULT_SHIFT/2); |
1277 | else | |
cf2ab469 | 1278 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1279 | |
ecf691da | 1280 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1281 | } |
1282 | ||
1091985b | 1283 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1284 | { |
1285 | lw->weight += inc; | |
e89996ae | 1286 | lw->inv_weight = 0; |
45bf76df IM |
1287 | } |
1288 | ||
1091985b | 1289 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1290 | { |
1291 | lw->weight -= dec; | |
e89996ae | 1292 | lw->inv_weight = 0; |
45bf76df IM |
1293 | } |
1294 | ||
2dd73a4f PW |
1295 | /* |
1296 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1297 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1298 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1299 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1300 | * scaled version of the new time slice allocation that they receive on time |
1301 | * slice expiry etc. | |
1302 | */ | |
1303 | ||
dd41f596 IM |
1304 | #define WEIGHT_IDLEPRIO 2 |
1305 | #define WMULT_IDLEPRIO (1 << 31) | |
1306 | ||
1307 | /* | |
1308 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1309 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1310 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1311 | * that remained on nice 0. | |
1312 | * | |
1313 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1314 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1315 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1316 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1317 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1318 | */ |
1319 | static const int prio_to_weight[40] = { | |
254753dc IM |
1320 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1321 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1322 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1323 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1324 | /* 0 */ 1024, 820, 655, 526, 423, | |
1325 | /* 5 */ 335, 272, 215, 172, 137, | |
1326 | /* 10 */ 110, 87, 70, 56, 45, | |
1327 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1328 | }; |
1329 | ||
5714d2de IM |
1330 | /* |
1331 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1332 | * | |
1333 | * In cases where the weight does not change often, we can use the | |
1334 | * precalculated inverse to speed up arithmetics by turning divisions | |
1335 | * into multiplications: | |
1336 | */ | |
dd41f596 | 1337 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1338 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1339 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1340 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1341 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1342 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1343 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1344 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1345 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1346 | }; |
2dd73a4f | 1347 | |
dd41f596 IM |
1348 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1349 | ||
1350 | /* | |
1351 | * runqueue iterator, to support SMP load-balancing between different | |
1352 | * scheduling classes, without having to expose their internal data | |
1353 | * structures to the load-balancing proper: | |
1354 | */ | |
1355 | struct rq_iterator { | |
1356 | void *arg; | |
1357 | struct task_struct *(*start)(void *); | |
1358 | struct task_struct *(*next)(void *); | |
1359 | }; | |
1360 | ||
e1d1484f PW |
1361 | #ifdef CONFIG_SMP |
1362 | static unsigned long | |
1363 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1364 | unsigned long max_load_move, struct sched_domain *sd, | |
1365 | enum cpu_idle_type idle, int *all_pinned, | |
1366 | int *this_best_prio, struct rq_iterator *iterator); | |
1367 | ||
1368 | static int | |
1369 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1370 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1371 | struct rq_iterator *iterator); | |
e1d1484f | 1372 | #endif |
dd41f596 | 1373 | |
d842de87 SV |
1374 | #ifdef CONFIG_CGROUP_CPUACCT |
1375 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1376 | #else | |
1377 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1378 | #endif | |
1379 | ||
18d95a28 PZ |
1380 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1381 | { | |
1382 | update_load_add(&rq->load, load); | |
1383 | } | |
1384 | ||
1385 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1386 | { | |
1387 | update_load_sub(&rq->load, load); | |
1388 | } | |
1389 | ||
e7693a36 GH |
1390 | #ifdef CONFIG_SMP |
1391 | static unsigned long source_load(int cpu, int type); | |
1392 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1393 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1394 | |
a8a51d5e PZ |
1395 | static unsigned long cpu_avg_load_per_task(int cpu) |
1396 | { | |
1397 | struct rq *rq = cpu_rq(cpu); | |
1398 | ||
1399 | if (rq->nr_running) | |
1400 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1401 | ||
1402 | return rq->avg_load_per_task; | |
1403 | } | |
18d95a28 PZ |
1404 | |
1405 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1406 | |
c8cba857 | 1407 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1408 | |
1409 | /* | |
1410 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1411 | * leaving it for the final time. | |
1412 | */ | |
c8cba857 PZ |
1413 | static void |
1414 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1415 | { |
1416 | struct task_group *parent, *child; | |
1417 | ||
1418 | rcu_read_lock(); | |
1419 | parent = &root_task_group; | |
1420 | down: | |
b6a86c74 | 1421 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1422 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1423 | parent = child; | |
1424 | goto down; | |
1425 | ||
1426 | up: | |
1427 | continue; | |
1428 | } | |
b6a86c74 | 1429 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1430 | |
1431 | child = parent; | |
1432 | parent = parent->parent; | |
1433 | if (parent) | |
1434 | goto up; | |
1435 | rcu_read_unlock(); | |
1436 | } | |
1437 | ||
c09595f6 PZ |
1438 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1439 | ||
1440 | /* | |
1441 | * Calculate and set the cpu's group shares. | |
1442 | */ | |
1443 | static void | |
b6a86c74 | 1444 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1445 | unsigned long sd_shares, unsigned long sd_rq_weight) |
18d95a28 | 1446 | { |
c09595f6 PZ |
1447 | int boost = 0; |
1448 | unsigned long shares; | |
1449 | unsigned long rq_weight; | |
1450 | ||
c8cba857 | 1451 | if (!tg->se[cpu]) |
c09595f6 PZ |
1452 | return; |
1453 | ||
c8cba857 | 1454 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1455 | |
1456 | /* | |
1457 | * If there are currently no tasks on the cpu pretend there is one of | |
1458 | * average load so that when a new task gets to run here it will not | |
1459 | * get delayed by group starvation. | |
1460 | */ | |
1461 | if (!rq_weight) { | |
1462 | boost = 1; | |
1463 | rq_weight = NICE_0_LOAD; | |
1464 | } | |
1465 | ||
c8cba857 PZ |
1466 | if (unlikely(rq_weight > sd_rq_weight)) |
1467 | rq_weight = sd_rq_weight; | |
1468 | ||
c09595f6 PZ |
1469 | /* |
1470 | * \Sum shares * rq_weight | |
1471 | * shares = ----------------------- | |
1472 | * \Sum rq_weight | |
1473 | * | |
1474 | */ | |
c8cba857 | 1475 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1476 | |
1477 | /* | |
1478 | * record the actual number of shares, not the boosted amount. | |
1479 | */ | |
c8cba857 | 1480 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
f1d239f7 | 1481 | tg->cfs_rq[cpu]->rq_weight = rq_weight; |
c09595f6 PZ |
1482 | |
1483 | if (shares < MIN_SHARES) | |
1484 | shares = MIN_SHARES; | |
1485 | else if (shares > MAX_SHARES) | |
1486 | shares = MAX_SHARES; | |
1487 | ||
c8cba857 | 1488 | __set_se_shares(tg->se[cpu], shares); |
18d95a28 | 1489 | } |
c09595f6 PZ |
1490 | |
1491 | /* | |
c8cba857 PZ |
1492 | * Re-compute the task group their per cpu shares over the given domain. |
1493 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1494 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1495 | */ |
1496 | static void | |
c8cba857 | 1497 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1498 | { |
c8cba857 PZ |
1499 | unsigned long rq_weight = 0; |
1500 | unsigned long shares = 0; | |
1501 | int i; | |
c09595f6 | 1502 | |
c8cba857 PZ |
1503 | for_each_cpu_mask(i, sd->span) { |
1504 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1505 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1506 | } |
c09595f6 | 1507 | |
c8cba857 PZ |
1508 | if ((!shares && rq_weight) || shares > tg->shares) |
1509 | shares = tg->shares; | |
1510 | ||
1511 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1512 | shares = tg->shares; | |
c09595f6 | 1513 | |
cd80917e PZ |
1514 | if (!rq_weight) |
1515 | rq_weight = cpus_weight(sd->span) * NICE_0_LOAD; | |
1516 | ||
c09595f6 PZ |
1517 | for_each_cpu_mask(i, sd->span) { |
1518 | struct rq *rq = cpu_rq(i); | |
1519 | unsigned long flags; | |
1520 | ||
1521 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1522 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1523 | spin_unlock_irqrestore(&rq->lock, flags); |
1524 | } | |
c09595f6 PZ |
1525 | } |
1526 | ||
1527 | /* | |
c8cba857 PZ |
1528 | * Compute the cpu's hierarchical load factor for each task group. |
1529 | * This needs to be done in a top-down fashion because the load of a child | |
1530 | * group is a fraction of its parents load. | |
c09595f6 | 1531 | */ |
b6a86c74 | 1532 | static void |
c8cba857 | 1533 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1534 | { |
c8cba857 | 1535 | unsigned long load; |
c09595f6 | 1536 | |
c8cba857 PZ |
1537 | if (!tg->parent) { |
1538 | load = cpu_rq(cpu)->load.weight; | |
1539 | } else { | |
1540 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1541 | load *= tg->cfs_rq[cpu]->shares; | |
1542 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1543 | } | |
c09595f6 | 1544 | |
c8cba857 | 1545 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1546 | } |
1547 | ||
c8cba857 PZ |
1548 | static void |
1549 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1550 | { |
c09595f6 PZ |
1551 | } |
1552 | ||
c8cba857 | 1553 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1554 | { |
2398f2c6 PZ |
1555 | u64 now = cpu_clock(raw_smp_processor_id()); |
1556 | s64 elapsed = now - sd->last_update; | |
1557 | ||
1558 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1559 | sd->last_update = now; | |
1560 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); | |
1561 | } | |
4d8d595d PZ |
1562 | } |
1563 | ||
3e5459b4 PZ |
1564 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1565 | { | |
1566 | spin_unlock(&rq->lock); | |
1567 | update_shares(sd); | |
1568 | spin_lock(&rq->lock); | |
1569 | } | |
1570 | ||
c8cba857 | 1571 | static void update_h_load(int cpu) |
c09595f6 | 1572 | { |
c8cba857 | 1573 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1574 | } |
1575 | ||
c09595f6 PZ |
1576 | #else |
1577 | ||
c8cba857 | 1578 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1579 | { |
1580 | } | |
1581 | ||
3e5459b4 PZ |
1582 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1583 | { | |
1584 | } | |
1585 | ||
18d95a28 PZ |
1586 | #endif |
1587 | ||
18d95a28 PZ |
1588 | #endif |
1589 | ||
30432094 | 1590 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1591 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1592 | { | |
30432094 | 1593 | #ifdef CONFIG_SMP |
34e83e85 IM |
1594 | cfs_rq->shares = shares; |
1595 | #endif | |
1596 | } | |
30432094 | 1597 | #endif |
e7693a36 | 1598 | |
dd41f596 | 1599 | #include "sched_stats.h" |
dd41f596 | 1600 | #include "sched_idletask.c" |
5522d5d5 IM |
1601 | #include "sched_fair.c" |
1602 | #include "sched_rt.c" | |
dd41f596 IM |
1603 | #ifdef CONFIG_SCHED_DEBUG |
1604 | # include "sched_debug.c" | |
1605 | #endif | |
1606 | ||
1607 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1608 | #define for_each_class(class) \ |
1609 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1610 | |
c09595f6 | 1611 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1612 | { |
1613 | rq->nr_running++; | |
9c217245 IM |
1614 | } |
1615 | ||
c09595f6 | 1616 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1617 | { |
1618 | rq->nr_running--; | |
9c217245 IM |
1619 | } |
1620 | ||
45bf76df IM |
1621 | static void set_load_weight(struct task_struct *p) |
1622 | { | |
1623 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1624 | p->se.load.weight = prio_to_weight[0] * 2; |
1625 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1626 | return; | |
1627 | } | |
45bf76df | 1628 | |
dd41f596 IM |
1629 | /* |
1630 | * SCHED_IDLE tasks get minimal weight: | |
1631 | */ | |
1632 | if (p->policy == SCHED_IDLE) { | |
1633 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1634 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1635 | return; | |
1636 | } | |
71f8bd46 | 1637 | |
dd41f596 IM |
1638 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1639 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1640 | } |
1641 | ||
2087a1ad GH |
1642 | static void update_avg(u64 *avg, u64 sample) |
1643 | { | |
1644 | s64 diff = sample - *avg; | |
1645 | *avg += diff >> 3; | |
1646 | } | |
1647 | ||
8159f87e | 1648 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1649 | { |
dd41f596 | 1650 | sched_info_queued(p); |
fd390f6a | 1651 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1652 | p->se.on_rq = 1; |
71f8bd46 IM |
1653 | } |
1654 | ||
69be72c1 | 1655 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1656 | { |
2087a1ad GH |
1657 | if (sleep && p->se.last_wakeup) { |
1658 | update_avg(&p->se.avg_overlap, | |
1659 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1660 | p->se.last_wakeup = 0; | |
1661 | } | |
1662 | ||
46ac22ba | 1663 | sched_info_dequeued(p); |
f02231e5 | 1664 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1665 | p->se.on_rq = 0; |
71f8bd46 IM |
1666 | } |
1667 | ||
14531189 | 1668 | /* |
dd41f596 | 1669 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1670 | */ |
14531189 IM |
1671 | static inline int __normal_prio(struct task_struct *p) |
1672 | { | |
dd41f596 | 1673 | return p->static_prio; |
14531189 IM |
1674 | } |
1675 | ||
b29739f9 IM |
1676 | /* |
1677 | * Calculate the expected normal priority: i.e. priority | |
1678 | * without taking RT-inheritance into account. Might be | |
1679 | * boosted by interactivity modifiers. Changes upon fork, | |
1680 | * setprio syscalls, and whenever the interactivity | |
1681 | * estimator recalculates. | |
1682 | */ | |
36c8b586 | 1683 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1684 | { |
1685 | int prio; | |
1686 | ||
e05606d3 | 1687 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1688 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1689 | else | |
1690 | prio = __normal_prio(p); | |
1691 | return prio; | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * Calculate the current priority, i.e. the priority | |
1696 | * taken into account by the scheduler. This value might | |
1697 | * be boosted by RT tasks, or might be boosted by | |
1698 | * interactivity modifiers. Will be RT if the task got | |
1699 | * RT-boosted. If not then it returns p->normal_prio. | |
1700 | */ | |
36c8b586 | 1701 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1702 | { |
1703 | p->normal_prio = normal_prio(p); | |
1704 | /* | |
1705 | * If we are RT tasks or we were boosted to RT priority, | |
1706 | * keep the priority unchanged. Otherwise, update priority | |
1707 | * to the normal priority: | |
1708 | */ | |
1709 | if (!rt_prio(p->prio)) | |
1710 | return p->normal_prio; | |
1711 | return p->prio; | |
1712 | } | |
1713 | ||
1da177e4 | 1714 | /* |
dd41f596 | 1715 | * activate_task - move a task to the runqueue. |
1da177e4 | 1716 | */ |
dd41f596 | 1717 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1718 | { |
d9514f6c | 1719 | if (task_contributes_to_load(p)) |
dd41f596 | 1720 | rq->nr_uninterruptible--; |
1da177e4 | 1721 | |
8159f87e | 1722 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1723 | inc_nr_running(rq); |
1da177e4 LT |
1724 | } |
1725 | ||
1da177e4 LT |
1726 | /* |
1727 | * deactivate_task - remove a task from the runqueue. | |
1728 | */ | |
2e1cb74a | 1729 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1730 | { |
d9514f6c | 1731 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1732 | rq->nr_uninterruptible++; |
1733 | ||
69be72c1 | 1734 | dequeue_task(rq, p, sleep); |
c09595f6 | 1735 | dec_nr_running(rq); |
1da177e4 LT |
1736 | } |
1737 | ||
1da177e4 LT |
1738 | /** |
1739 | * task_curr - is this task currently executing on a CPU? | |
1740 | * @p: the task in question. | |
1741 | */ | |
36c8b586 | 1742 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1743 | { |
1744 | return cpu_curr(task_cpu(p)) == p; | |
1745 | } | |
1746 | ||
dd41f596 IM |
1747 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1748 | { | |
6f505b16 | 1749 | set_task_rq(p, cpu); |
dd41f596 | 1750 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1751 | /* |
1752 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1753 | * successfuly executed on another CPU. We must ensure that updates of | |
1754 | * per-task data have been completed by this moment. | |
1755 | */ | |
1756 | smp_wmb(); | |
dd41f596 | 1757 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1758 | #endif |
2dd73a4f PW |
1759 | } |
1760 | ||
cb469845 SR |
1761 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1762 | const struct sched_class *prev_class, | |
1763 | int oldprio, int running) | |
1764 | { | |
1765 | if (prev_class != p->sched_class) { | |
1766 | if (prev_class->switched_from) | |
1767 | prev_class->switched_from(rq, p, running); | |
1768 | p->sched_class->switched_to(rq, p, running); | |
1769 | } else | |
1770 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1771 | } | |
1772 | ||
1da177e4 | 1773 | #ifdef CONFIG_SMP |
c65cc870 | 1774 | |
e958b360 TG |
1775 | /* Used instead of source_load when we know the type == 0 */ |
1776 | static unsigned long weighted_cpuload(const int cpu) | |
1777 | { | |
1778 | return cpu_rq(cpu)->load.weight; | |
1779 | } | |
1780 | ||
cc367732 IM |
1781 | /* |
1782 | * Is this task likely cache-hot: | |
1783 | */ | |
e7693a36 | 1784 | static int |
cc367732 IM |
1785 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1786 | { | |
1787 | s64 delta; | |
1788 | ||
f540a608 IM |
1789 | /* |
1790 | * Buddy candidates are cache hot: | |
1791 | */ | |
d25ce4cd | 1792 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1793 | return 1; |
1794 | ||
cc367732 IM |
1795 | if (p->sched_class != &fair_sched_class) |
1796 | return 0; | |
1797 | ||
6bc1665b IM |
1798 | if (sysctl_sched_migration_cost == -1) |
1799 | return 1; | |
1800 | if (sysctl_sched_migration_cost == 0) | |
1801 | return 0; | |
1802 | ||
cc367732 IM |
1803 | delta = now - p->se.exec_start; |
1804 | ||
1805 | return delta < (s64)sysctl_sched_migration_cost; | |
1806 | } | |
1807 | ||
1808 | ||
dd41f596 | 1809 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1810 | { |
dd41f596 IM |
1811 | int old_cpu = task_cpu(p); |
1812 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1813 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1814 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1815 | u64 clock_offset; |
dd41f596 IM |
1816 | |
1817 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1818 | |
1819 | #ifdef CONFIG_SCHEDSTATS | |
1820 | if (p->se.wait_start) | |
1821 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1822 | if (p->se.sleep_start) |
1823 | p->se.sleep_start -= clock_offset; | |
1824 | if (p->se.block_start) | |
1825 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1826 | if (old_cpu != new_cpu) { |
1827 | schedstat_inc(p, se.nr_migrations); | |
1828 | if (task_hot(p, old_rq->clock, NULL)) | |
1829 | schedstat_inc(p, se.nr_forced2_migrations); | |
1830 | } | |
6cfb0d5d | 1831 | #endif |
2830cf8c SV |
1832 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1833 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1834 | |
1835 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1836 | } |
1837 | ||
70b97a7f | 1838 | struct migration_req { |
1da177e4 | 1839 | struct list_head list; |
1da177e4 | 1840 | |
36c8b586 | 1841 | struct task_struct *task; |
1da177e4 LT |
1842 | int dest_cpu; |
1843 | ||
1da177e4 | 1844 | struct completion done; |
70b97a7f | 1845 | }; |
1da177e4 LT |
1846 | |
1847 | /* | |
1848 | * The task's runqueue lock must be held. | |
1849 | * Returns true if you have to wait for migration thread. | |
1850 | */ | |
36c8b586 | 1851 | static int |
70b97a7f | 1852 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1853 | { |
70b97a7f | 1854 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1855 | |
1856 | /* | |
1857 | * If the task is not on a runqueue (and not running), then | |
1858 | * it is sufficient to simply update the task's cpu field. | |
1859 | */ | |
dd41f596 | 1860 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1861 | set_task_cpu(p, dest_cpu); |
1862 | return 0; | |
1863 | } | |
1864 | ||
1865 | init_completion(&req->done); | |
1da177e4 LT |
1866 | req->task = p; |
1867 | req->dest_cpu = dest_cpu; | |
1868 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1869 | |
1da177e4 LT |
1870 | return 1; |
1871 | } | |
1872 | ||
1873 | /* | |
1874 | * wait_task_inactive - wait for a thread to unschedule. | |
1875 | * | |
85ba2d86 RM |
1876 | * If @match_state is nonzero, it's the @p->state value just checked and |
1877 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1878 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1879 | * we return a positive number (its total switch count). If a second call | |
1880 | * a short while later returns the same number, the caller can be sure that | |
1881 | * @p has remained unscheduled the whole time. | |
1882 | * | |
1da177e4 LT |
1883 | * The caller must ensure that the task *will* unschedule sometime soon, |
1884 | * else this function might spin for a *long* time. This function can't | |
1885 | * be called with interrupts off, or it may introduce deadlock with | |
1886 | * smp_call_function() if an IPI is sent by the same process we are | |
1887 | * waiting to become inactive. | |
1888 | */ | |
85ba2d86 | 1889 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1890 | { |
1891 | unsigned long flags; | |
dd41f596 | 1892 | int running, on_rq; |
85ba2d86 | 1893 | unsigned long ncsw; |
70b97a7f | 1894 | struct rq *rq; |
1da177e4 | 1895 | |
3a5c359a AK |
1896 | for (;;) { |
1897 | /* | |
1898 | * We do the initial early heuristics without holding | |
1899 | * any task-queue locks at all. We'll only try to get | |
1900 | * the runqueue lock when things look like they will | |
1901 | * work out! | |
1902 | */ | |
1903 | rq = task_rq(p); | |
fa490cfd | 1904 | |
3a5c359a AK |
1905 | /* |
1906 | * If the task is actively running on another CPU | |
1907 | * still, just relax and busy-wait without holding | |
1908 | * any locks. | |
1909 | * | |
1910 | * NOTE! Since we don't hold any locks, it's not | |
1911 | * even sure that "rq" stays as the right runqueue! | |
1912 | * But we don't care, since "task_running()" will | |
1913 | * return false if the runqueue has changed and p | |
1914 | * is actually now running somewhere else! | |
1915 | */ | |
85ba2d86 RM |
1916 | while (task_running(rq, p)) { |
1917 | if (match_state && unlikely(p->state != match_state)) | |
1918 | return 0; | |
3a5c359a | 1919 | cpu_relax(); |
85ba2d86 | 1920 | } |
fa490cfd | 1921 | |
3a5c359a AK |
1922 | /* |
1923 | * Ok, time to look more closely! We need the rq | |
1924 | * lock now, to be *sure*. If we're wrong, we'll | |
1925 | * just go back and repeat. | |
1926 | */ | |
1927 | rq = task_rq_lock(p, &flags); | |
1928 | running = task_running(rq, p); | |
1929 | on_rq = p->se.on_rq; | |
85ba2d86 RM |
1930 | ncsw = 0; |
1931 | if (!match_state || p->state == match_state) { | |
1932 | ncsw = p->nivcsw + p->nvcsw; | |
1933 | if (unlikely(!ncsw)) | |
1934 | ncsw = 1; | |
1935 | } | |
3a5c359a | 1936 | task_rq_unlock(rq, &flags); |
fa490cfd | 1937 | |
85ba2d86 RM |
1938 | /* |
1939 | * If it changed from the expected state, bail out now. | |
1940 | */ | |
1941 | if (unlikely(!ncsw)) | |
1942 | break; | |
1943 | ||
3a5c359a AK |
1944 | /* |
1945 | * Was it really running after all now that we | |
1946 | * checked with the proper locks actually held? | |
1947 | * | |
1948 | * Oops. Go back and try again.. | |
1949 | */ | |
1950 | if (unlikely(running)) { | |
1951 | cpu_relax(); | |
1952 | continue; | |
1953 | } | |
fa490cfd | 1954 | |
3a5c359a AK |
1955 | /* |
1956 | * It's not enough that it's not actively running, | |
1957 | * it must be off the runqueue _entirely_, and not | |
1958 | * preempted! | |
1959 | * | |
1960 | * So if it wa still runnable (but just not actively | |
1961 | * running right now), it's preempted, and we should | |
1962 | * yield - it could be a while. | |
1963 | */ | |
1964 | if (unlikely(on_rq)) { | |
1965 | schedule_timeout_uninterruptible(1); | |
1966 | continue; | |
1967 | } | |
fa490cfd | 1968 | |
3a5c359a AK |
1969 | /* |
1970 | * Ahh, all good. It wasn't running, and it wasn't | |
1971 | * runnable, which means that it will never become | |
1972 | * running in the future either. We're all done! | |
1973 | */ | |
1974 | break; | |
1975 | } | |
85ba2d86 RM |
1976 | |
1977 | return ncsw; | |
1da177e4 LT |
1978 | } |
1979 | ||
1980 | /*** | |
1981 | * kick_process - kick a running thread to enter/exit the kernel | |
1982 | * @p: the to-be-kicked thread | |
1983 | * | |
1984 | * Cause a process which is running on another CPU to enter | |
1985 | * kernel-mode, without any delay. (to get signals handled.) | |
1986 | * | |
1987 | * NOTE: this function doesnt have to take the runqueue lock, | |
1988 | * because all it wants to ensure is that the remote task enters | |
1989 | * the kernel. If the IPI races and the task has been migrated | |
1990 | * to another CPU then no harm is done and the purpose has been | |
1991 | * achieved as well. | |
1992 | */ | |
36c8b586 | 1993 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1994 | { |
1995 | int cpu; | |
1996 | ||
1997 | preempt_disable(); | |
1998 | cpu = task_cpu(p); | |
1999 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2000 | smp_send_reschedule(cpu); | |
2001 | preempt_enable(); | |
2002 | } | |
2003 | ||
2004 | /* | |
2dd73a4f PW |
2005 | * Return a low guess at the load of a migration-source cpu weighted |
2006 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2007 | * |
2008 | * We want to under-estimate the load of migration sources, to | |
2009 | * balance conservatively. | |
2010 | */ | |
a9957449 | 2011 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2012 | { |
70b97a7f | 2013 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2014 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2015 | |
93b75217 | 2016 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2017 | return total; |
b910472d | 2018 | |
dd41f596 | 2019 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2020 | } |
2021 | ||
2022 | /* | |
2dd73a4f PW |
2023 | * Return a high guess at the load of a migration-target cpu weighted |
2024 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2025 | */ |
a9957449 | 2026 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2027 | { |
70b97a7f | 2028 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2029 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2030 | |
93b75217 | 2031 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2032 | return total; |
3b0bd9bc | 2033 | |
dd41f596 | 2034 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2035 | } |
2036 | ||
147cbb4b NP |
2037 | /* |
2038 | * find_idlest_group finds and returns the least busy CPU group within the | |
2039 | * domain. | |
2040 | */ | |
2041 | static struct sched_group * | |
2042 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2043 | { | |
2044 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2045 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2046 | int load_idx = sd->forkexec_idx; | |
2047 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2048 | ||
2049 | do { | |
2050 | unsigned long load, avg_load; | |
2051 | int local_group; | |
2052 | int i; | |
2053 | ||
da5a5522 BD |
2054 | /* Skip over this group if it has no CPUs allowed */ |
2055 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2056 | continue; |
da5a5522 | 2057 | |
147cbb4b | 2058 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2059 | |
2060 | /* Tally up the load of all CPUs in the group */ | |
2061 | avg_load = 0; | |
2062 | ||
363ab6f1 | 2063 | for_each_cpu_mask_nr(i, group->cpumask) { |
147cbb4b NP |
2064 | /* Bias balancing toward cpus of our domain */ |
2065 | if (local_group) | |
2066 | load = source_load(i, load_idx); | |
2067 | else | |
2068 | load = target_load(i, load_idx); | |
2069 | ||
2070 | avg_load += load; | |
2071 | } | |
2072 | ||
2073 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2074 | avg_load = sg_div_cpu_power(group, |
2075 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2076 | |
2077 | if (local_group) { | |
2078 | this_load = avg_load; | |
2079 | this = group; | |
2080 | } else if (avg_load < min_load) { | |
2081 | min_load = avg_load; | |
2082 | idlest = group; | |
2083 | } | |
3a5c359a | 2084 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2085 | |
2086 | if (!idlest || 100*this_load < imbalance*min_load) | |
2087 | return NULL; | |
2088 | return idlest; | |
2089 | } | |
2090 | ||
2091 | /* | |
0feaece9 | 2092 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2093 | */ |
95cdf3b7 | 2094 | static int |
7c16ec58 MT |
2095 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2096 | cpumask_t *tmp) | |
147cbb4b NP |
2097 | { |
2098 | unsigned long load, min_load = ULONG_MAX; | |
2099 | int idlest = -1; | |
2100 | int i; | |
2101 | ||
da5a5522 | 2102 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2103 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2104 | |
363ab6f1 | 2105 | for_each_cpu_mask_nr(i, *tmp) { |
2dd73a4f | 2106 | load = weighted_cpuload(i); |
147cbb4b NP |
2107 | |
2108 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2109 | min_load = load; | |
2110 | idlest = i; | |
2111 | } | |
2112 | } | |
2113 | ||
2114 | return idlest; | |
2115 | } | |
2116 | ||
476d139c NP |
2117 | /* |
2118 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2119 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2120 | * SD_BALANCE_EXEC. | |
2121 | * | |
2122 | * Balance, ie. select the least loaded group. | |
2123 | * | |
2124 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2125 | * | |
2126 | * preempt must be disabled. | |
2127 | */ | |
2128 | static int sched_balance_self(int cpu, int flag) | |
2129 | { | |
2130 | struct task_struct *t = current; | |
2131 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2132 | |
c96d145e | 2133 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2134 | /* |
2135 | * If power savings logic is enabled for a domain, stop there. | |
2136 | */ | |
5c45bf27 SS |
2137 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2138 | break; | |
476d139c NP |
2139 | if (tmp->flags & flag) |
2140 | sd = tmp; | |
c96d145e | 2141 | } |
476d139c | 2142 | |
039a1c41 PZ |
2143 | if (sd) |
2144 | update_shares(sd); | |
2145 | ||
476d139c | 2146 | while (sd) { |
7c16ec58 | 2147 | cpumask_t span, tmpmask; |
476d139c | 2148 | struct sched_group *group; |
1a848870 SS |
2149 | int new_cpu, weight; |
2150 | ||
2151 | if (!(sd->flags & flag)) { | |
2152 | sd = sd->child; | |
2153 | continue; | |
2154 | } | |
476d139c NP |
2155 | |
2156 | span = sd->span; | |
2157 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2158 | if (!group) { |
2159 | sd = sd->child; | |
2160 | continue; | |
2161 | } | |
476d139c | 2162 | |
7c16ec58 | 2163 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2164 | if (new_cpu == -1 || new_cpu == cpu) { |
2165 | /* Now try balancing at a lower domain level of cpu */ | |
2166 | sd = sd->child; | |
2167 | continue; | |
2168 | } | |
476d139c | 2169 | |
1a848870 | 2170 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2171 | cpu = new_cpu; |
476d139c NP |
2172 | sd = NULL; |
2173 | weight = cpus_weight(span); | |
2174 | for_each_domain(cpu, tmp) { | |
2175 | if (weight <= cpus_weight(tmp->span)) | |
2176 | break; | |
2177 | if (tmp->flags & flag) | |
2178 | sd = tmp; | |
2179 | } | |
2180 | /* while loop will break here if sd == NULL */ | |
2181 | } | |
2182 | ||
2183 | return cpu; | |
2184 | } | |
2185 | ||
2186 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2187 | |
1da177e4 LT |
2188 | /*** |
2189 | * try_to_wake_up - wake up a thread | |
2190 | * @p: the to-be-woken-up thread | |
2191 | * @state: the mask of task states that can be woken | |
2192 | * @sync: do a synchronous wakeup? | |
2193 | * | |
2194 | * Put it on the run-queue if it's not already there. The "current" | |
2195 | * thread is always on the run-queue (except when the actual | |
2196 | * re-schedule is in progress), and as such you're allowed to do | |
2197 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2198 | * runnable without the overhead of this. | |
2199 | * | |
2200 | * returns failure only if the task is already active. | |
2201 | */ | |
36c8b586 | 2202 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2203 | { |
cc367732 | 2204 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2205 | unsigned long flags; |
2206 | long old_state; | |
70b97a7f | 2207 | struct rq *rq; |
1da177e4 | 2208 | |
b85d0667 IM |
2209 | if (!sched_feat(SYNC_WAKEUPS)) |
2210 | sync = 0; | |
2211 | ||
2398f2c6 PZ |
2212 | #ifdef CONFIG_SMP |
2213 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2214 | struct sched_domain *sd; | |
2215 | ||
2216 | this_cpu = raw_smp_processor_id(); | |
2217 | cpu = task_cpu(p); | |
2218 | ||
2219 | for_each_domain(this_cpu, sd) { | |
2220 | if (cpu_isset(cpu, sd->span)) { | |
2221 | update_shares(sd); | |
2222 | break; | |
2223 | } | |
2224 | } | |
2225 | } | |
2226 | #endif | |
2227 | ||
04e2f174 | 2228 | smp_wmb(); |
1da177e4 LT |
2229 | rq = task_rq_lock(p, &flags); |
2230 | old_state = p->state; | |
2231 | if (!(old_state & state)) | |
2232 | goto out; | |
2233 | ||
dd41f596 | 2234 | if (p->se.on_rq) |
1da177e4 LT |
2235 | goto out_running; |
2236 | ||
2237 | cpu = task_cpu(p); | |
cc367732 | 2238 | orig_cpu = cpu; |
1da177e4 LT |
2239 | this_cpu = smp_processor_id(); |
2240 | ||
2241 | #ifdef CONFIG_SMP | |
2242 | if (unlikely(task_running(rq, p))) | |
2243 | goto out_activate; | |
2244 | ||
5d2f5a61 DA |
2245 | cpu = p->sched_class->select_task_rq(p, sync); |
2246 | if (cpu != orig_cpu) { | |
2247 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2248 | task_rq_unlock(rq, &flags); |
2249 | /* might preempt at this point */ | |
2250 | rq = task_rq_lock(p, &flags); | |
2251 | old_state = p->state; | |
2252 | if (!(old_state & state)) | |
2253 | goto out; | |
dd41f596 | 2254 | if (p->se.on_rq) |
1da177e4 LT |
2255 | goto out_running; |
2256 | ||
2257 | this_cpu = smp_processor_id(); | |
2258 | cpu = task_cpu(p); | |
2259 | } | |
2260 | ||
e7693a36 GH |
2261 | #ifdef CONFIG_SCHEDSTATS |
2262 | schedstat_inc(rq, ttwu_count); | |
2263 | if (cpu == this_cpu) | |
2264 | schedstat_inc(rq, ttwu_local); | |
2265 | else { | |
2266 | struct sched_domain *sd; | |
2267 | for_each_domain(this_cpu, sd) { | |
2268 | if (cpu_isset(cpu, sd->span)) { | |
2269 | schedstat_inc(sd, ttwu_wake_remote); | |
2270 | break; | |
2271 | } | |
2272 | } | |
2273 | } | |
6d6bc0ad | 2274 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2275 | |
1da177e4 LT |
2276 | out_activate: |
2277 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2278 | schedstat_inc(p, se.nr_wakeups); |
2279 | if (sync) | |
2280 | schedstat_inc(p, se.nr_wakeups_sync); | |
2281 | if (orig_cpu != cpu) | |
2282 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2283 | if (cpu == this_cpu) | |
2284 | schedstat_inc(p, se.nr_wakeups_local); | |
2285 | else | |
2286 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2287 | update_rq_clock(rq); |
dd41f596 | 2288 | activate_task(rq, p, 1); |
1da177e4 LT |
2289 | success = 1; |
2290 | ||
2291 | out_running: | |
5b82a1b0 MD |
2292 | trace_mark(kernel_sched_wakeup, |
2293 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2294 | p->pid, p->state, rq, p, rq->curr); | |
4ae7d5ce IM |
2295 | check_preempt_curr(rq, p); |
2296 | ||
1da177e4 | 2297 | p->state = TASK_RUNNING; |
9a897c5a SR |
2298 | #ifdef CONFIG_SMP |
2299 | if (p->sched_class->task_wake_up) | |
2300 | p->sched_class->task_wake_up(rq, p); | |
2301 | #endif | |
1da177e4 | 2302 | out: |
2087a1ad GH |
2303 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2304 | ||
1da177e4 LT |
2305 | task_rq_unlock(rq, &flags); |
2306 | ||
2307 | return success; | |
2308 | } | |
2309 | ||
7ad5b3a5 | 2310 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2311 | { |
d9514f6c | 2312 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2313 | } |
1da177e4 LT |
2314 | EXPORT_SYMBOL(wake_up_process); |
2315 | ||
7ad5b3a5 | 2316 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2317 | { |
2318 | return try_to_wake_up(p, state, 0); | |
2319 | } | |
2320 | ||
1da177e4 LT |
2321 | /* |
2322 | * Perform scheduler related setup for a newly forked process p. | |
2323 | * p is forked by current. | |
dd41f596 IM |
2324 | * |
2325 | * __sched_fork() is basic setup used by init_idle() too: | |
2326 | */ | |
2327 | static void __sched_fork(struct task_struct *p) | |
2328 | { | |
dd41f596 IM |
2329 | p->se.exec_start = 0; |
2330 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2331 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2332 | p->se.last_wakeup = 0; |
2333 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2334 | |
2335 | #ifdef CONFIG_SCHEDSTATS | |
2336 | p->se.wait_start = 0; | |
dd41f596 IM |
2337 | p->se.sum_sleep_runtime = 0; |
2338 | p->se.sleep_start = 0; | |
dd41f596 IM |
2339 | p->se.block_start = 0; |
2340 | p->se.sleep_max = 0; | |
2341 | p->se.block_max = 0; | |
2342 | p->se.exec_max = 0; | |
eba1ed4b | 2343 | p->se.slice_max = 0; |
dd41f596 | 2344 | p->se.wait_max = 0; |
6cfb0d5d | 2345 | #endif |
476d139c | 2346 | |
fa717060 | 2347 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2348 | p->se.on_rq = 0; |
4a55bd5e | 2349 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2350 | |
e107be36 AK |
2351 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2352 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2353 | #endif | |
2354 | ||
1da177e4 LT |
2355 | /* |
2356 | * We mark the process as running here, but have not actually | |
2357 | * inserted it onto the runqueue yet. This guarantees that | |
2358 | * nobody will actually run it, and a signal or other external | |
2359 | * event cannot wake it up and insert it on the runqueue either. | |
2360 | */ | |
2361 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2362 | } |
2363 | ||
2364 | /* | |
2365 | * fork()/clone()-time setup: | |
2366 | */ | |
2367 | void sched_fork(struct task_struct *p, int clone_flags) | |
2368 | { | |
2369 | int cpu = get_cpu(); | |
2370 | ||
2371 | __sched_fork(p); | |
2372 | ||
2373 | #ifdef CONFIG_SMP | |
2374 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2375 | #endif | |
02e4bac2 | 2376 | set_task_cpu(p, cpu); |
b29739f9 IM |
2377 | |
2378 | /* | |
2379 | * Make sure we do not leak PI boosting priority to the child: | |
2380 | */ | |
2381 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2382 | if (!rt_prio(p->prio)) |
2383 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2384 | |
52f17b6c | 2385 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2386 | if (likely(sched_info_on())) |
52f17b6c | 2387 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2388 | #endif |
d6077cb8 | 2389 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2390 | p->oncpu = 0; |
2391 | #endif | |
1da177e4 | 2392 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2393 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2394 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2395 | #endif |
476d139c | 2396 | put_cpu(); |
1da177e4 LT |
2397 | } |
2398 | ||
2399 | /* | |
2400 | * wake_up_new_task - wake up a newly created task for the first time. | |
2401 | * | |
2402 | * This function will do some initial scheduler statistics housekeeping | |
2403 | * that must be done for every newly created context, then puts the task | |
2404 | * on the runqueue and wakes it. | |
2405 | */ | |
7ad5b3a5 | 2406 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2407 | { |
2408 | unsigned long flags; | |
dd41f596 | 2409 | struct rq *rq; |
1da177e4 LT |
2410 | |
2411 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2412 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2413 | update_rq_clock(rq); |
1da177e4 LT |
2414 | |
2415 | p->prio = effective_prio(p); | |
2416 | ||
b9dca1e0 | 2417 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2418 | activate_task(rq, p, 0); |
1da177e4 | 2419 | } else { |
1da177e4 | 2420 | /* |
dd41f596 IM |
2421 | * Let the scheduling class do new task startup |
2422 | * management (if any): | |
1da177e4 | 2423 | */ |
ee0827d8 | 2424 | p->sched_class->task_new(rq, p); |
c09595f6 | 2425 | inc_nr_running(rq); |
1da177e4 | 2426 | } |
5b82a1b0 MD |
2427 | trace_mark(kernel_sched_wakeup_new, |
2428 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2429 | p->pid, p->state, rq, p, rq->curr); | |
dd41f596 | 2430 | check_preempt_curr(rq, p); |
9a897c5a SR |
2431 | #ifdef CONFIG_SMP |
2432 | if (p->sched_class->task_wake_up) | |
2433 | p->sched_class->task_wake_up(rq, p); | |
2434 | #endif | |
dd41f596 | 2435 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2436 | } |
2437 | ||
e107be36 AK |
2438 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2439 | ||
2440 | /** | |
421cee29 RD |
2441 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2442 | * @notifier: notifier struct to register | |
e107be36 AK |
2443 | */ |
2444 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2445 | { | |
2446 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2447 | } | |
2448 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2449 | ||
2450 | /** | |
2451 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2452 | * @notifier: notifier struct to unregister |
e107be36 AK |
2453 | * |
2454 | * This is safe to call from within a preemption notifier. | |
2455 | */ | |
2456 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2457 | { | |
2458 | hlist_del(¬ifier->link); | |
2459 | } | |
2460 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2461 | ||
2462 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2463 | { | |
2464 | struct preempt_notifier *notifier; | |
2465 | struct hlist_node *node; | |
2466 | ||
2467 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2468 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2469 | } | |
2470 | ||
2471 | static void | |
2472 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2473 | struct task_struct *next) | |
2474 | { | |
2475 | struct preempt_notifier *notifier; | |
2476 | struct hlist_node *node; | |
2477 | ||
2478 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2479 | notifier->ops->sched_out(notifier, next); | |
2480 | } | |
2481 | ||
6d6bc0ad | 2482 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2483 | |
2484 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2485 | { | |
2486 | } | |
2487 | ||
2488 | static void | |
2489 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2490 | struct task_struct *next) | |
2491 | { | |
2492 | } | |
2493 | ||
6d6bc0ad | 2494 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2495 | |
4866cde0 NP |
2496 | /** |
2497 | * prepare_task_switch - prepare to switch tasks | |
2498 | * @rq: the runqueue preparing to switch | |
421cee29 | 2499 | * @prev: the current task that is being switched out |
4866cde0 NP |
2500 | * @next: the task we are going to switch to. |
2501 | * | |
2502 | * This is called with the rq lock held and interrupts off. It must | |
2503 | * be paired with a subsequent finish_task_switch after the context | |
2504 | * switch. | |
2505 | * | |
2506 | * prepare_task_switch sets up locking and calls architecture specific | |
2507 | * hooks. | |
2508 | */ | |
e107be36 AK |
2509 | static inline void |
2510 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2511 | struct task_struct *next) | |
4866cde0 | 2512 | { |
e107be36 | 2513 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2514 | prepare_lock_switch(rq, next); |
2515 | prepare_arch_switch(next); | |
2516 | } | |
2517 | ||
1da177e4 LT |
2518 | /** |
2519 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2520 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2521 | * @prev: the thread we just switched away from. |
2522 | * | |
4866cde0 NP |
2523 | * finish_task_switch must be called after the context switch, paired |
2524 | * with a prepare_task_switch call before the context switch. | |
2525 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2526 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2527 | * |
2528 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2529 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2530 | * with the lock held can cause deadlocks; see schedule() for |
2531 | * details.) | |
2532 | */ | |
a9957449 | 2533 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2534 | __releases(rq->lock) |
2535 | { | |
1da177e4 | 2536 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2537 | long prev_state; |
1da177e4 LT |
2538 | |
2539 | rq->prev_mm = NULL; | |
2540 | ||
2541 | /* | |
2542 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2543 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2544 | * schedule one last time. The schedule call will never return, and |
2545 | * the scheduled task must drop that reference. | |
c394cc9f | 2546 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2547 | * still held, otherwise prev could be scheduled on another cpu, die |
2548 | * there before we look at prev->state, and then the reference would | |
2549 | * be dropped twice. | |
2550 | * Manfred Spraul <manfred@colorfullife.com> | |
2551 | */ | |
55a101f8 | 2552 | prev_state = prev->state; |
4866cde0 NP |
2553 | finish_arch_switch(prev); |
2554 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2555 | #ifdef CONFIG_SMP |
2556 | if (current->sched_class->post_schedule) | |
2557 | current->sched_class->post_schedule(rq); | |
2558 | #endif | |
e8fa1362 | 2559 | |
e107be36 | 2560 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2561 | if (mm) |
2562 | mmdrop(mm); | |
c394cc9f | 2563 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2564 | /* |
2565 | * Remove function-return probe instances associated with this | |
2566 | * task and put them back on the free list. | |
9761eea8 | 2567 | */ |
c6fd91f0 | 2568 | kprobe_flush_task(prev); |
1da177e4 | 2569 | put_task_struct(prev); |
c6fd91f0 | 2570 | } |
1da177e4 LT |
2571 | } |
2572 | ||
2573 | /** | |
2574 | * schedule_tail - first thing a freshly forked thread must call. | |
2575 | * @prev: the thread we just switched away from. | |
2576 | */ | |
36c8b586 | 2577 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2578 | __releases(rq->lock) |
2579 | { | |
70b97a7f IM |
2580 | struct rq *rq = this_rq(); |
2581 | ||
4866cde0 NP |
2582 | finish_task_switch(rq, prev); |
2583 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2584 | /* In this case, finish_task_switch does not reenable preemption */ | |
2585 | preempt_enable(); | |
2586 | #endif | |
1da177e4 | 2587 | if (current->set_child_tid) |
b488893a | 2588 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2589 | } |
2590 | ||
2591 | /* | |
2592 | * context_switch - switch to the new MM and the new | |
2593 | * thread's register state. | |
2594 | */ | |
dd41f596 | 2595 | static inline void |
70b97a7f | 2596 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2597 | struct task_struct *next) |
1da177e4 | 2598 | { |
dd41f596 | 2599 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2600 | |
e107be36 | 2601 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2602 | trace_mark(kernel_sched_schedule, |
2603 | "prev_pid %d next_pid %d prev_state %ld " | |
2604 | "## rq %p prev %p next %p", | |
2605 | prev->pid, next->pid, prev->state, | |
2606 | rq, prev, next); | |
dd41f596 IM |
2607 | mm = next->mm; |
2608 | oldmm = prev->active_mm; | |
9226d125 ZA |
2609 | /* |
2610 | * For paravirt, this is coupled with an exit in switch_to to | |
2611 | * combine the page table reload and the switch backend into | |
2612 | * one hypercall. | |
2613 | */ | |
2614 | arch_enter_lazy_cpu_mode(); | |
2615 | ||
dd41f596 | 2616 | if (unlikely(!mm)) { |
1da177e4 LT |
2617 | next->active_mm = oldmm; |
2618 | atomic_inc(&oldmm->mm_count); | |
2619 | enter_lazy_tlb(oldmm, next); | |
2620 | } else | |
2621 | switch_mm(oldmm, mm, next); | |
2622 | ||
dd41f596 | 2623 | if (unlikely(!prev->mm)) { |
1da177e4 | 2624 | prev->active_mm = NULL; |
1da177e4 LT |
2625 | rq->prev_mm = oldmm; |
2626 | } | |
3a5f5e48 IM |
2627 | /* |
2628 | * Since the runqueue lock will be released by the next | |
2629 | * task (which is an invalid locking op but in the case | |
2630 | * of the scheduler it's an obvious special-case), so we | |
2631 | * do an early lockdep release here: | |
2632 | */ | |
2633 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2634 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2635 | #endif |
1da177e4 LT |
2636 | |
2637 | /* Here we just switch the register state and the stack. */ | |
2638 | switch_to(prev, next, prev); | |
2639 | ||
dd41f596 IM |
2640 | barrier(); |
2641 | /* | |
2642 | * this_rq must be evaluated again because prev may have moved | |
2643 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2644 | * frame will be invalid. | |
2645 | */ | |
2646 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2647 | } |
2648 | ||
2649 | /* | |
2650 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2651 | * | |
2652 | * externally visible scheduler statistics: current number of runnable | |
2653 | * threads, current number of uninterruptible-sleeping threads, total | |
2654 | * number of context switches performed since bootup. | |
2655 | */ | |
2656 | unsigned long nr_running(void) | |
2657 | { | |
2658 | unsigned long i, sum = 0; | |
2659 | ||
2660 | for_each_online_cpu(i) | |
2661 | sum += cpu_rq(i)->nr_running; | |
2662 | ||
2663 | return sum; | |
2664 | } | |
2665 | ||
2666 | unsigned long nr_uninterruptible(void) | |
2667 | { | |
2668 | unsigned long i, sum = 0; | |
2669 | ||
0a945022 | 2670 | for_each_possible_cpu(i) |
1da177e4 LT |
2671 | sum += cpu_rq(i)->nr_uninterruptible; |
2672 | ||
2673 | /* | |
2674 | * Since we read the counters lockless, it might be slightly | |
2675 | * inaccurate. Do not allow it to go below zero though: | |
2676 | */ | |
2677 | if (unlikely((long)sum < 0)) | |
2678 | sum = 0; | |
2679 | ||
2680 | return sum; | |
2681 | } | |
2682 | ||
2683 | unsigned long long nr_context_switches(void) | |
2684 | { | |
cc94abfc SR |
2685 | int i; |
2686 | unsigned long long sum = 0; | |
1da177e4 | 2687 | |
0a945022 | 2688 | for_each_possible_cpu(i) |
1da177e4 LT |
2689 | sum += cpu_rq(i)->nr_switches; |
2690 | ||
2691 | return sum; | |
2692 | } | |
2693 | ||
2694 | unsigned long nr_iowait(void) | |
2695 | { | |
2696 | unsigned long i, sum = 0; | |
2697 | ||
0a945022 | 2698 | for_each_possible_cpu(i) |
1da177e4 LT |
2699 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2700 | ||
2701 | return sum; | |
2702 | } | |
2703 | ||
db1b1fef JS |
2704 | unsigned long nr_active(void) |
2705 | { | |
2706 | unsigned long i, running = 0, uninterruptible = 0; | |
2707 | ||
2708 | for_each_online_cpu(i) { | |
2709 | running += cpu_rq(i)->nr_running; | |
2710 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2711 | } | |
2712 | ||
2713 | if (unlikely((long)uninterruptible < 0)) | |
2714 | uninterruptible = 0; | |
2715 | ||
2716 | return running + uninterruptible; | |
2717 | } | |
2718 | ||
48f24c4d | 2719 | /* |
dd41f596 IM |
2720 | * Update rq->cpu_load[] statistics. This function is usually called every |
2721 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2722 | */ |
dd41f596 | 2723 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2724 | { |
495eca49 | 2725 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2726 | int i, scale; |
2727 | ||
2728 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2729 | |
2730 | /* Update our load: */ | |
2731 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2732 | unsigned long old_load, new_load; | |
2733 | ||
2734 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2735 | ||
2736 | old_load = this_rq->cpu_load[i]; | |
2737 | new_load = this_load; | |
a25707f3 IM |
2738 | /* |
2739 | * Round up the averaging division if load is increasing. This | |
2740 | * prevents us from getting stuck on 9 if the load is 10, for | |
2741 | * example. | |
2742 | */ | |
2743 | if (new_load > old_load) | |
2744 | new_load += scale-1; | |
dd41f596 IM |
2745 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2746 | } | |
48f24c4d IM |
2747 | } |
2748 | ||
dd41f596 IM |
2749 | #ifdef CONFIG_SMP |
2750 | ||
1da177e4 LT |
2751 | /* |
2752 | * double_rq_lock - safely lock two runqueues | |
2753 | * | |
2754 | * Note this does not disable interrupts like task_rq_lock, | |
2755 | * you need to do so manually before calling. | |
2756 | */ | |
70b97a7f | 2757 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2758 | __acquires(rq1->lock) |
2759 | __acquires(rq2->lock) | |
2760 | { | |
054b9108 | 2761 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2762 | if (rq1 == rq2) { |
2763 | spin_lock(&rq1->lock); | |
2764 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2765 | } else { | |
c96d145e | 2766 | if (rq1 < rq2) { |
1da177e4 | 2767 | spin_lock(&rq1->lock); |
5e710e37 | 2768 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2769 | } else { |
2770 | spin_lock(&rq2->lock); | |
5e710e37 | 2771 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2772 | } |
2773 | } | |
6e82a3be IM |
2774 | update_rq_clock(rq1); |
2775 | update_rq_clock(rq2); | |
1da177e4 LT |
2776 | } |
2777 | ||
2778 | /* | |
2779 | * double_rq_unlock - safely unlock two runqueues | |
2780 | * | |
2781 | * Note this does not restore interrupts like task_rq_unlock, | |
2782 | * you need to do so manually after calling. | |
2783 | */ | |
70b97a7f | 2784 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2785 | __releases(rq1->lock) |
2786 | __releases(rq2->lock) | |
2787 | { | |
2788 | spin_unlock(&rq1->lock); | |
2789 | if (rq1 != rq2) | |
2790 | spin_unlock(&rq2->lock); | |
2791 | else | |
2792 | __release(rq2->lock); | |
2793 | } | |
2794 | ||
2795 | /* | |
2796 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2797 | */ | |
e8fa1362 | 2798 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2799 | __releases(this_rq->lock) |
2800 | __acquires(busiest->lock) | |
2801 | __acquires(this_rq->lock) | |
2802 | { | |
e8fa1362 SR |
2803 | int ret = 0; |
2804 | ||
054b9108 KK |
2805 | if (unlikely(!irqs_disabled())) { |
2806 | /* printk() doesn't work good under rq->lock */ | |
2807 | spin_unlock(&this_rq->lock); | |
2808 | BUG_ON(1); | |
2809 | } | |
1da177e4 | 2810 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2811 | if (busiest < this_rq) { |
1da177e4 LT |
2812 | spin_unlock(&this_rq->lock); |
2813 | spin_lock(&busiest->lock); | |
5e710e37 | 2814 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); |
e8fa1362 | 2815 | ret = 1; |
1da177e4 | 2816 | } else |
5e710e37 | 2817 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); |
1da177e4 | 2818 | } |
e8fa1362 | 2819 | return ret; |
1da177e4 LT |
2820 | } |
2821 | ||
1b12bbc7 PZ |
2822 | static void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
2823 | __releases(busiest->lock) | |
2824 | { | |
2825 | spin_unlock(&busiest->lock); | |
2826 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2827 | } | |
2828 | ||
1da177e4 LT |
2829 | /* |
2830 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2831 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2832 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2833 | * the cpu_allowed mask is restored. |
2834 | */ | |
36c8b586 | 2835 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2836 | { |
70b97a7f | 2837 | struct migration_req req; |
1da177e4 | 2838 | unsigned long flags; |
70b97a7f | 2839 | struct rq *rq; |
1da177e4 LT |
2840 | |
2841 | rq = task_rq_lock(p, &flags); | |
2842 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
e761b772 | 2843 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2844 | goto out; |
2845 | ||
2846 | /* force the process onto the specified CPU */ | |
2847 | if (migrate_task(p, dest_cpu, &req)) { | |
2848 | /* Need to wait for migration thread (might exit: take ref). */ | |
2849 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2850 | |
1da177e4 LT |
2851 | get_task_struct(mt); |
2852 | task_rq_unlock(rq, &flags); | |
2853 | wake_up_process(mt); | |
2854 | put_task_struct(mt); | |
2855 | wait_for_completion(&req.done); | |
36c8b586 | 2856 | |
1da177e4 LT |
2857 | return; |
2858 | } | |
2859 | out: | |
2860 | task_rq_unlock(rq, &flags); | |
2861 | } | |
2862 | ||
2863 | /* | |
476d139c NP |
2864 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2865 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2866 | */ |
2867 | void sched_exec(void) | |
2868 | { | |
1da177e4 | 2869 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2870 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2871 | put_cpu(); |
476d139c NP |
2872 | if (new_cpu != this_cpu) |
2873 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2874 | } |
2875 | ||
2876 | /* | |
2877 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2878 | * Both runqueues must be locked. | |
2879 | */ | |
dd41f596 IM |
2880 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2881 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2882 | { |
2e1cb74a | 2883 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2884 | set_task_cpu(p, this_cpu); |
dd41f596 | 2885 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2886 | /* |
2887 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2888 | * to be always true for them. | |
2889 | */ | |
dd41f596 | 2890 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2891 | } |
2892 | ||
2893 | /* | |
2894 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2895 | */ | |
858119e1 | 2896 | static |
70b97a7f | 2897 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2898 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2899 | int *all_pinned) |
1da177e4 LT |
2900 | { |
2901 | /* | |
2902 | * We do not migrate tasks that are: | |
2903 | * 1) running (obviously), or | |
2904 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2905 | * 3) are cache-hot on their current CPU. | |
2906 | */ | |
cc367732 IM |
2907 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2908 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2909 | return 0; |
cc367732 | 2910 | } |
81026794 NP |
2911 | *all_pinned = 0; |
2912 | ||
cc367732 IM |
2913 | if (task_running(rq, p)) { |
2914 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2915 | return 0; |
cc367732 | 2916 | } |
1da177e4 | 2917 | |
da84d961 IM |
2918 | /* |
2919 | * Aggressive migration if: | |
2920 | * 1) task is cache cold, or | |
2921 | * 2) too many balance attempts have failed. | |
2922 | */ | |
2923 | ||
6bc1665b IM |
2924 | if (!task_hot(p, rq->clock, sd) || |
2925 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2926 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2927 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2928 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2929 | schedstat_inc(p, se.nr_forced_migrations); |
2930 | } | |
da84d961 IM |
2931 | #endif |
2932 | return 1; | |
2933 | } | |
2934 | ||
cc367732 IM |
2935 | if (task_hot(p, rq->clock, sd)) { |
2936 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2937 | return 0; |
cc367732 | 2938 | } |
1da177e4 LT |
2939 | return 1; |
2940 | } | |
2941 | ||
e1d1484f PW |
2942 | static unsigned long |
2943 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2944 | unsigned long max_load_move, struct sched_domain *sd, | |
2945 | enum cpu_idle_type idle, int *all_pinned, | |
2946 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2947 | { |
051c6764 | 2948 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2949 | struct task_struct *p; |
2950 | long rem_load_move = max_load_move; | |
1da177e4 | 2951 | |
e1d1484f | 2952 | if (max_load_move == 0) |
1da177e4 LT |
2953 | goto out; |
2954 | ||
81026794 NP |
2955 | pinned = 1; |
2956 | ||
1da177e4 | 2957 | /* |
dd41f596 | 2958 | * Start the load-balancing iterator: |
1da177e4 | 2959 | */ |
dd41f596 IM |
2960 | p = iterator->start(iterator->arg); |
2961 | next: | |
b82d9fdd | 2962 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2963 | goto out; |
051c6764 PZ |
2964 | |
2965 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2966 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2967 | p = iterator->next(iterator->arg); |
2968 | goto next; | |
1da177e4 LT |
2969 | } |
2970 | ||
dd41f596 | 2971 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2972 | pulled++; |
dd41f596 | 2973 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2974 | |
2dd73a4f | 2975 | /* |
b82d9fdd | 2976 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2977 | */ |
e1d1484f | 2978 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2979 | if (p->prio < *this_best_prio) |
2980 | *this_best_prio = p->prio; | |
dd41f596 IM |
2981 | p = iterator->next(iterator->arg); |
2982 | goto next; | |
1da177e4 LT |
2983 | } |
2984 | out: | |
2985 | /* | |
e1d1484f | 2986 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2987 | * so we can safely collect pull_task() stats here rather than |
2988 | * inside pull_task(). | |
2989 | */ | |
2990 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2991 | |
2992 | if (all_pinned) | |
2993 | *all_pinned = pinned; | |
e1d1484f PW |
2994 | |
2995 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2996 | } |
2997 | ||
dd41f596 | 2998 | /* |
43010659 PW |
2999 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3000 | * this_rq, as part of a balancing operation within domain "sd". | |
3001 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3002 | * |
3003 | * Called with both runqueues locked. | |
3004 | */ | |
3005 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3006 | unsigned long max_load_move, |
dd41f596 IM |
3007 | struct sched_domain *sd, enum cpu_idle_type idle, |
3008 | int *all_pinned) | |
3009 | { | |
5522d5d5 | 3010 | const struct sched_class *class = sched_class_highest; |
43010659 | 3011 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3012 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3013 | |
3014 | do { | |
43010659 PW |
3015 | total_load_moved += |
3016 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3017 | max_load_move - total_load_moved, |
a4ac01c3 | 3018 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3019 | class = class->next; |
c4acb2c0 GH |
3020 | |
3021 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3022 | break; | |
3023 | ||
43010659 | 3024 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3025 | |
43010659 PW |
3026 | return total_load_moved > 0; |
3027 | } | |
3028 | ||
e1d1484f PW |
3029 | static int |
3030 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3031 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3032 | struct rq_iterator *iterator) | |
3033 | { | |
3034 | struct task_struct *p = iterator->start(iterator->arg); | |
3035 | int pinned = 0; | |
3036 | ||
3037 | while (p) { | |
3038 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3039 | pull_task(busiest, p, this_rq, this_cpu); | |
3040 | /* | |
3041 | * Right now, this is only the second place pull_task() | |
3042 | * is called, so we can safely collect pull_task() | |
3043 | * stats here rather than inside pull_task(). | |
3044 | */ | |
3045 | schedstat_inc(sd, lb_gained[idle]); | |
3046 | ||
3047 | return 1; | |
3048 | } | |
3049 | p = iterator->next(iterator->arg); | |
3050 | } | |
3051 | ||
3052 | return 0; | |
3053 | } | |
3054 | ||
43010659 PW |
3055 | /* |
3056 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3057 | * part of active balancing operations within "domain". | |
3058 | * Returns 1 if successful and 0 otherwise. | |
3059 | * | |
3060 | * Called with both runqueues locked. | |
3061 | */ | |
3062 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3063 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3064 | { | |
5522d5d5 | 3065 | const struct sched_class *class; |
43010659 PW |
3066 | |
3067 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3068 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3069 | return 1; |
3070 | ||
3071 | return 0; | |
dd41f596 IM |
3072 | } |
3073 | ||
1da177e4 LT |
3074 | /* |
3075 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3076 | * domain. It calculates and returns the amount of weighted load which |
3077 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3078 | */ |
3079 | static struct sched_group * | |
3080 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3081 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3082 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3083 | { |
3084 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3085 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3086 | unsigned long max_pull; |
2dd73a4f PW |
3087 | unsigned long busiest_load_per_task, busiest_nr_running; |
3088 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3089 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3090 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3091 | int power_savings_balance = 1; | |
3092 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3093 | unsigned long min_nr_running = ULONG_MAX; | |
3094 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3095 | #endif | |
1da177e4 LT |
3096 | |
3097 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3098 | busiest_load_per_task = busiest_nr_running = 0; |
3099 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3100 | |
d15bcfdb | 3101 | if (idle == CPU_NOT_IDLE) |
7897986b | 3102 | load_idx = sd->busy_idx; |
d15bcfdb | 3103 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3104 | load_idx = sd->newidle_idx; |
3105 | else | |
3106 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3107 | |
3108 | do { | |
908a7c1b | 3109 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3110 | int local_group; |
3111 | int i; | |
908a7c1b | 3112 | int __group_imb = 0; |
783609c6 | 3113 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3114 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3115 | unsigned long sum_avg_load_per_task; |
3116 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3117 | |
3118 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3119 | ||
783609c6 SS |
3120 | if (local_group) |
3121 | balance_cpu = first_cpu(group->cpumask); | |
3122 | ||
1da177e4 | 3123 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3124 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3125 | sum_avg_load_per_task = avg_load_per_task = 0; |
3126 | ||
908a7c1b KC |
3127 | max_cpu_load = 0; |
3128 | min_cpu_load = ~0UL; | |
1da177e4 | 3129 | |
363ab6f1 | 3130 | for_each_cpu_mask_nr(i, group->cpumask) { |
0a2966b4 CL |
3131 | struct rq *rq; |
3132 | ||
3133 | if (!cpu_isset(i, *cpus)) | |
3134 | continue; | |
3135 | ||
3136 | rq = cpu_rq(i); | |
2dd73a4f | 3137 | |
9439aab8 | 3138 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3139 | *sd_idle = 0; |
3140 | ||
1da177e4 | 3141 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3142 | if (local_group) { |
3143 | if (idle_cpu(i) && !first_idle_cpu) { | |
3144 | first_idle_cpu = 1; | |
3145 | balance_cpu = i; | |
3146 | } | |
3147 | ||
a2000572 | 3148 | load = target_load(i, load_idx); |
908a7c1b | 3149 | } else { |
a2000572 | 3150 | load = source_load(i, load_idx); |
908a7c1b KC |
3151 | if (load > max_cpu_load) |
3152 | max_cpu_load = load; | |
3153 | if (min_cpu_load > load) | |
3154 | min_cpu_load = load; | |
3155 | } | |
1da177e4 LT |
3156 | |
3157 | avg_load += load; | |
2dd73a4f | 3158 | sum_nr_running += rq->nr_running; |
dd41f596 | 3159 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3160 | |
3161 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3162 | } |
3163 | ||
783609c6 SS |
3164 | /* |
3165 | * First idle cpu or the first cpu(busiest) in this sched group | |
3166 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3167 | * domains. In the newly idle case, we will allow all the cpu's |
3168 | * to do the newly idle load balance. | |
783609c6 | 3169 | */ |
9439aab8 SS |
3170 | if (idle != CPU_NEWLY_IDLE && local_group && |
3171 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3172 | *balance = 0; |
3173 | goto ret; | |
3174 | } | |
3175 | ||
1da177e4 | 3176 | total_load += avg_load; |
5517d86b | 3177 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3178 | |
3179 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3180 | avg_load = sg_div_cpu_power(group, |
3181 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3182 | |
408ed066 PZ |
3183 | |
3184 | /* | |
3185 | * Consider the group unbalanced when the imbalance is larger | |
3186 | * than the average weight of two tasks. | |
3187 | * | |
3188 | * APZ: with cgroup the avg task weight can vary wildly and | |
3189 | * might not be a suitable number - should we keep a | |
3190 | * normalized nr_running number somewhere that negates | |
3191 | * the hierarchy? | |
3192 | */ | |
3193 | avg_load_per_task = sg_div_cpu_power(group, | |
3194 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3195 | ||
3196 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3197 | __group_imb = 1; |
3198 | ||
5517d86b | 3199 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3200 | |
1da177e4 LT |
3201 | if (local_group) { |
3202 | this_load = avg_load; | |
3203 | this = group; | |
2dd73a4f PW |
3204 | this_nr_running = sum_nr_running; |
3205 | this_load_per_task = sum_weighted_load; | |
3206 | } else if (avg_load > max_load && | |
908a7c1b | 3207 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3208 | max_load = avg_load; |
3209 | busiest = group; | |
2dd73a4f PW |
3210 | busiest_nr_running = sum_nr_running; |
3211 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3212 | group_imb = __group_imb; |
1da177e4 | 3213 | } |
5c45bf27 SS |
3214 | |
3215 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3216 | /* | |
3217 | * Busy processors will not participate in power savings | |
3218 | * balance. | |
3219 | */ | |
dd41f596 IM |
3220 | if (idle == CPU_NOT_IDLE || |
3221 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3222 | goto group_next; | |
5c45bf27 SS |
3223 | |
3224 | /* | |
3225 | * If the local group is idle or completely loaded | |
3226 | * no need to do power savings balance at this domain | |
3227 | */ | |
3228 | if (local_group && (this_nr_running >= group_capacity || | |
3229 | !this_nr_running)) | |
3230 | power_savings_balance = 0; | |
3231 | ||
dd41f596 | 3232 | /* |
5c45bf27 SS |
3233 | * If a group is already running at full capacity or idle, |
3234 | * don't include that group in power savings calculations | |
dd41f596 IM |
3235 | */ |
3236 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3237 | || !sum_nr_running) |
dd41f596 | 3238 | goto group_next; |
5c45bf27 | 3239 | |
dd41f596 | 3240 | /* |
5c45bf27 | 3241 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3242 | * This is the group from where we need to pick up the load |
3243 | * for saving power | |
3244 | */ | |
3245 | if ((sum_nr_running < min_nr_running) || | |
3246 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3247 | first_cpu(group->cpumask) < |
3248 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3249 | group_min = group; |
3250 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3251 | min_load_per_task = sum_weighted_load / |
3252 | sum_nr_running; | |
dd41f596 | 3253 | } |
5c45bf27 | 3254 | |
dd41f596 | 3255 | /* |
5c45bf27 | 3256 | * Calculate the group which is almost near its |
dd41f596 IM |
3257 | * capacity but still has some space to pick up some load |
3258 | * from other group and save more power | |
3259 | */ | |
3260 | if (sum_nr_running <= group_capacity - 1) { | |
3261 | if (sum_nr_running > leader_nr_running || | |
3262 | (sum_nr_running == leader_nr_running && | |
3263 | first_cpu(group->cpumask) > | |
3264 | first_cpu(group_leader->cpumask))) { | |
3265 | group_leader = group; | |
3266 | leader_nr_running = sum_nr_running; | |
3267 | } | |
48f24c4d | 3268 | } |
5c45bf27 SS |
3269 | group_next: |
3270 | #endif | |
1da177e4 LT |
3271 | group = group->next; |
3272 | } while (group != sd->groups); | |
3273 | ||
2dd73a4f | 3274 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3275 | goto out_balanced; |
3276 | ||
3277 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3278 | ||
3279 | if (this_load >= avg_load || | |
3280 | 100*max_load <= sd->imbalance_pct*this_load) | |
3281 | goto out_balanced; | |
3282 | ||
2dd73a4f | 3283 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3284 | if (group_imb) |
3285 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3286 | ||
1da177e4 LT |
3287 | /* |
3288 | * We're trying to get all the cpus to the average_load, so we don't | |
3289 | * want to push ourselves above the average load, nor do we wish to | |
3290 | * reduce the max loaded cpu below the average load, as either of these | |
3291 | * actions would just result in more rebalancing later, and ping-pong | |
3292 | * tasks around. Thus we look for the minimum possible imbalance. | |
3293 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3294 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3295 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3296 | * appear as very large values with unsigned longs. |
3297 | */ | |
2dd73a4f PW |
3298 | if (max_load <= busiest_load_per_task) |
3299 | goto out_balanced; | |
3300 | ||
3301 | /* | |
3302 | * In the presence of smp nice balancing, certain scenarios can have | |
3303 | * max load less than avg load(as we skip the groups at or below | |
3304 | * its cpu_power, while calculating max_load..) | |
3305 | */ | |
3306 | if (max_load < avg_load) { | |
3307 | *imbalance = 0; | |
3308 | goto small_imbalance; | |
3309 | } | |
0c117f1b SS |
3310 | |
3311 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3312 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3313 | |
1da177e4 | 3314 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3315 | *imbalance = min(max_pull * busiest->__cpu_power, |
3316 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3317 | / SCHED_LOAD_SCALE; |
3318 | ||
2dd73a4f PW |
3319 | /* |
3320 | * if *imbalance is less than the average load per runnable task | |
3321 | * there is no gaurantee that any tasks will be moved so we'll have | |
3322 | * a think about bumping its value to force at least one task to be | |
3323 | * moved | |
3324 | */ | |
7fd0d2dd | 3325 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3326 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3327 | unsigned int imbn; |
3328 | ||
3329 | small_imbalance: | |
3330 | pwr_move = pwr_now = 0; | |
3331 | imbn = 2; | |
3332 | if (this_nr_running) { | |
3333 | this_load_per_task /= this_nr_running; | |
3334 | if (busiest_load_per_task > this_load_per_task) | |
3335 | imbn = 1; | |
3336 | } else | |
408ed066 | 3337 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3338 | |
408ed066 | 3339 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3340 | busiest_load_per_task * imbn) { |
2dd73a4f | 3341 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3342 | return busiest; |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * OK, we don't have enough imbalance to justify moving tasks, | |
3347 | * however we may be able to increase total CPU power used by | |
3348 | * moving them. | |
3349 | */ | |
3350 | ||
5517d86b ED |
3351 | pwr_now += busiest->__cpu_power * |
3352 | min(busiest_load_per_task, max_load); | |
3353 | pwr_now += this->__cpu_power * | |
3354 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3355 | pwr_now /= SCHED_LOAD_SCALE; |
3356 | ||
3357 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3358 | tmp = sg_div_cpu_power(busiest, |
3359 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3360 | if (max_load > tmp) |
5517d86b | 3361 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3362 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3363 | |
3364 | /* Amount of load we'd add */ | |
5517d86b | 3365 | if (max_load * busiest->__cpu_power < |
33859f7f | 3366 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3367 | tmp = sg_div_cpu_power(this, |
3368 | max_load * busiest->__cpu_power); | |
1da177e4 | 3369 | else |
5517d86b ED |
3370 | tmp = sg_div_cpu_power(this, |
3371 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3372 | pwr_move += this->__cpu_power * | |
3373 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3374 | pwr_move /= SCHED_LOAD_SCALE; |
3375 | ||
3376 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3377 | if (pwr_move > pwr_now) |
3378 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3379 | } |
3380 | ||
1da177e4 LT |
3381 | return busiest; |
3382 | ||
3383 | out_balanced: | |
5c45bf27 | 3384 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3385 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3386 | goto ret; |
1da177e4 | 3387 | |
5c45bf27 SS |
3388 | if (this == group_leader && group_leader != group_min) { |
3389 | *imbalance = min_load_per_task; | |
3390 | return group_min; | |
3391 | } | |
5c45bf27 | 3392 | #endif |
783609c6 | 3393 | ret: |
1da177e4 LT |
3394 | *imbalance = 0; |
3395 | return NULL; | |
3396 | } | |
3397 | ||
3398 | /* | |
3399 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3400 | */ | |
70b97a7f | 3401 | static struct rq * |
d15bcfdb | 3402 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3403 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3404 | { |
70b97a7f | 3405 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3406 | unsigned long max_load = 0; |
1da177e4 LT |
3407 | int i; |
3408 | ||
363ab6f1 | 3409 | for_each_cpu_mask_nr(i, group->cpumask) { |
dd41f596 | 3410 | unsigned long wl; |
0a2966b4 CL |
3411 | |
3412 | if (!cpu_isset(i, *cpus)) | |
3413 | continue; | |
3414 | ||
48f24c4d | 3415 | rq = cpu_rq(i); |
dd41f596 | 3416 | wl = weighted_cpuload(i); |
2dd73a4f | 3417 | |
dd41f596 | 3418 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3419 | continue; |
1da177e4 | 3420 | |
dd41f596 IM |
3421 | if (wl > max_load) { |
3422 | max_load = wl; | |
48f24c4d | 3423 | busiest = rq; |
1da177e4 LT |
3424 | } |
3425 | } | |
3426 | ||
3427 | return busiest; | |
3428 | } | |
3429 | ||
77391d71 NP |
3430 | /* |
3431 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3432 | * so long as it is large enough. | |
3433 | */ | |
3434 | #define MAX_PINNED_INTERVAL 512 | |
3435 | ||
1da177e4 LT |
3436 | /* |
3437 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3438 | * tasks if there is an imbalance. | |
1da177e4 | 3439 | */ |
70b97a7f | 3440 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3441 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3442 | int *balance, cpumask_t *cpus) |
1da177e4 | 3443 | { |
43010659 | 3444 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3445 | struct sched_group *group; |
1da177e4 | 3446 | unsigned long imbalance; |
70b97a7f | 3447 | struct rq *busiest; |
fe2eea3f | 3448 | unsigned long flags; |
5969fe06 | 3449 | |
7c16ec58 MT |
3450 | cpus_setall(*cpus); |
3451 | ||
89c4710e SS |
3452 | /* |
3453 | * When power savings policy is enabled for the parent domain, idle | |
3454 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3455 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3456 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3457 | */ |
d15bcfdb | 3458 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3459 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3460 | sd_idle = 1; |
1da177e4 | 3461 | |
2d72376b | 3462 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3463 | |
0a2966b4 | 3464 | redo: |
c8cba857 | 3465 | update_shares(sd); |
0a2966b4 | 3466 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3467 | cpus, balance); |
783609c6 | 3468 | |
06066714 | 3469 | if (*balance == 0) |
783609c6 | 3470 | goto out_balanced; |
783609c6 | 3471 | |
1da177e4 LT |
3472 | if (!group) { |
3473 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3474 | goto out_balanced; | |
3475 | } | |
3476 | ||
7c16ec58 | 3477 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3478 | if (!busiest) { |
3479 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3480 | goto out_balanced; | |
3481 | } | |
3482 | ||
db935dbd | 3483 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3484 | |
3485 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3486 | ||
43010659 | 3487 | ld_moved = 0; |
1da177e4 LT |
3488 | if (busiest->nr_running > 1) { |
3489 | /* | |
3490 | * Attempt to move tasks. If find_busiest_group has found | |
3491 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3492 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3493 | * correctly treated as an imbalance. |
3494 | */ | |
fe2eea3f | 3495 | local_irq_save(flags); |
e17224bf | 3496 | double_rq_lock(this_rq, busiest); |
43010659 | 3497 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3498 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3499 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3500 | local_irq_restore(flags); |
81026794 | 3501 | |
46cb4b7c SS |
3502 | /* |
3503 | * some other cpu did the load balance for us. | |
3504 | */ | |
43010659 | 3505 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3506 | resched_cpu(this_cpu); |
3507 | ||
81026794 | 3508 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3509 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3510 | cpu_clear(cpu_of(busiest), *cpus); |
3511 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3512 | goto redo; |
81026794 | 3513 | goto out_balanced; |
0a2966b4 | 3514 | } |
1da177e4 | 3515 | } |
81026794 | 3516 | |
43010659 | 3517 | if (!ld_moved) { |
1da177e4 LT |
3518 | schedstat_inc(sd, lb_failed[idle]); |
3519 | sd->nr_balance_failed++; | |
3520 | ||
3521 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3522 | |
fe2eea3f | 3523 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3524 | |
3525 | /* don't kick the migration_thread, if the curr | |
3526 | * task on busiest cpu can't be moved to this_cpu | |
3527 | */ | |
3528 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3529 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3530 | all_pinned = 1; |
3531 | goto out_one_pinned; | |
3532 | } | |
3533 | ||
1da177e4 LT |
3534 | if (!busiest->active_balance) { |
3535 | busiest->active_balance = 1; | |
3536 | busiest->push_cpu = this_cpu; | |
81026794 | 3537 | active_balance = 1; |
1da177e4 | 3538 | } |
fe2eea3f | 3539 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3540 | if (active_balance) |
1da177e4 LT |
3541 | wake_up_process(busiest->migration_thread); |
3542 | ||
3543 | /* | |
3544 | * We've kicked active balancing, reset the failure | |
3545 | * counter. | |
3546 | */ | |
39507451 | 3547 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3548 | } |
81026794 | 3549 | } else |
1da177e4 LT |
3550 | sd->nr_balance_failed = 0; |
3551 | ||
81026794 | 3552 | if (likely(!active_balance)) { |
1da177e4 LT |
3553 | /* We were unbalanced, so reset the balancing interval */ |
3554 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3555 | } else { |
3556 | /* | |
3557 | * If we've begun active balancing, start to back off. This | |
3558 | * case may not be covered by the all_pinned logic if there | |
3559 | * is only 1 task on the busy runqueue (because we don't call | |
3560 | * move_tasks). | |
3561 | */ | |
3562 | if (sd->balance_interval < sd->max_interval) | |
3563 | sd->balance_interval *= 2; | |
1da177e4 LT |
3564 | } |
3565 | ||
43010659 | 3566 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3567 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3568 | ld_moved = -1; |
3569 | ||
3570 | goto out; | |
1da177e4 LT |
3571 | |
3572 | out_balanced: | |
1da177e4 LT |
3573 | schedstat_inc(sd, lb_balanced[idle]); |
3574 | ||
16cfb1c0 | 3575 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3576 | |
3577 | out_one_pinned: | |
1da177e4 | 3578 | /* tune up the balancing interval */ |
77391d71 NP |
3579 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3580 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3581 | sd->balance_interval *= 2; |
3582 | ||
48f24c4d | 3583 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3584 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3585 | ld_moved = -1; |
3586 | else | |
3587 | ld_moved = 0; | |
3588 | out: | |
c8cba857 PZ |
3589 | if (ld_moved) |
3590 | update_shares(sd); | |
c09595f6 | 3591 | return ld_moved; |
1da177e4 LT |
3592 | } |
3593 | ||
3594 | /* | |
3595 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3596 | * tasks if there is an imbalance. | |
3597 | * | |
d15bcfdb | 3598 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3599 | * this_rq is locked. |
3600 | */ | |
48f24c4d | 3601 | static int |
7c16ec58 MT |
3602 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3603 | cpumask_t *cpus) | |
1da177e4 LT |
3604 | { |
3605 | struct sched_group *group; | |
70b97a7f | 3606 | struct rq *busiest = NULL; |
1da177e4 | 3607 | unsigned long imbalance; |
43010659 | 3608 | int ld_moved = 0; |
5969fe06 | 3609 | int sd_idle = 0; |
969bb4e4 | 3610 | int all_pinned = 0; |
7c16ec58 MT |
3611 | |
3612 | cpus_setall(*cpus); | |
5969fe06 | 3613 | |
89c4710e SS |
3614 | /* |
3615 | * When power savings policy is enabled for the parent domain, idle | |
3616 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3617 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3618 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3619 | */ |
3620 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3621 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3622 | sd_idle = 1; |
1da177e4 | 3623 | |
2d72376b | 3624 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3625 | redo: |
3e5459b4 | 3626 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3627 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3628 | &sd_idle, cpus, NULL); |
1da177e4 | 3629 | if (!group) { |
d15bcfdb | 3630 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3631 | goto out_balanced; |
1da177e4 LT |
3632 | } |
3633 | ||
7c16ec58 | 3634 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3635 | if (!busiest) { |
d15bcfdb | 3636 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3637 | goto out_balanced; |
1da177e4 LT |
3638 | } |
3639 | ||
db935dbd NP |
3640 | BUG_ON(busiest == this_rq); |
3641 | ||
d15bcfdb | 3642 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3643 | |
43010659 | 3644 | ld_moved = 0; |
d6d5cfaf NP |
3645 | if (busiest->nr_running > 1) { |
3646 | /* Attempt to move tasks */ | |
3647 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3648 | /* this_rq->clock is already updated */ |
3649 | update_rq_clock(busiest); | |
43010659 | 3650 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3651 | imbalance, sd, CPU_NEWLY_IDLE, |
3652 | &all_pinned); | |
1b12bbc7 | 3653 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3654 | |
969bb4e4 | 3655 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3656 | cpu_clear(cpu_of(busiest), *cpus); |
3657 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3658 | goto redo; |
3659 | } | |
d6d5cfaf NP |
3660 | } |
3661 | ||
43010659 | 3662 | if (!ld_moved) { |
d15bcfdb | 3663 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3664 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3665 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3666 | return -1; |
3667 | } else | |
16cfb1c0 | 3668 | sd->nr_balance_failed = 0; |
1da177e4 | 3669 | |
3e5459b4 | 3670 | update_shares_locked(this_rq, sd); |
43010659 | 3671 | return ld_moved; |
16cfb1c0 NP |
3672 | |
3673 | out_balanced: | |
d15bcfdb | 3674 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3675 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3676 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3677 | return -1; |
16cfb1c0 | 3678 | sd->nr_balance_failed = 0; |
48f24c4d | 3679 | |
16cfb1c0 | 3680 | return 0; |
1da177e4 LT |
3681 | } |
3682 | ||
3683 | /* | |
3684 | * idle_balance is called by schedule() if this_cpu is about to become | |
3685 | * idle. Attempts to pull tasks from other CPUs. | |
3686 | */ | |
70b97a7f | 3687 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3688 | { |
3689 | struct sched_domain *sd; | |
dd41f596 IM |
3690 | int pulled_task = -1; |
3691 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3692 | cpumask_t tmpmask; |
1da177e4 LT |
3693 | |
3694 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3695 | unsigned long interval; |
3696 | ||
3697 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3698 | continue; | |
3699 | ||
3700 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3701 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3702 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3703 | sd, &tmpmask); | |
92c4ca5c CL |
3704 | |
3705 | interval = msecs_to_jiffies(sd->balance_interval); | |
3706 | if (time_after(next_balance, sd->last_balance + interval)) | |
3707 | next_balance = sd->last_balance + interval; | |
3708 | if (pulled_task) | |
3709 | break; | |
1da177e4 | 3710 | } |
dd41f596 | 3711 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3712 | /* |
3713 | * We are going idle. next_balance may be set based on | |
3714 | * a busy processor. So reset next_balance. | |
3715 | */ | |
3716 | this_rq->next_balance = next_balance; | |
dd41f596 | 3717 | } |
1da177e4 LT |
3718 | } |
3719 | ||
3720 | /* | |
3721 | * active_load_balance is run by migration threads. It pushes running tasks | |
3722 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3723 | * running on each physical CPU where possible, and avoids physical / | |
3724 | * logical imbalances. | |
3725 | * | |
3726 | * Called with busiest_rq locked. | |
3727 | */ | |
70b97a7f | 3728 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3729 | { |
39507451 | 3730 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3731 | struct sched_domain *sd; |
3732 | struct rq *target_rq; | |
39507451 | 3733 | |
48f24c4d | 3734 | /* Is there any task to move? */ |
39507451 | 3735 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3736 | return; |
3737 | ||
3738 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3739 | |
3740 | /* | |
39507451 | 3741 | * This condition is "impossible", if it occurs |
41a2d6cf | 3742 | * we need to fix it. Originally reported by |
39507451 | 3743 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3744 | */ |
39507451 | 3745 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3746 | |
39507451 NP |
3747 | /* move a task from busiest_rq to target_rq */ |
3748 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3749 | update_rq_clock(busiest_rq); |
3750 | update_rq_clock(target_rq); | |
39507451 NP |
3751 | |
3752 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3753 | for_each_domain(target_cpu, sd) { |
39507451 | 3754 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3755 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3756 | break; |
c96d145e | 3757 | } |
39507451 | 3758 | |
48f24c4d | 3759 | if (likely(sd)) { |
2d72376b | 3760 | schedstat_inc(sd, alb_count); |
39507451 | 3761 | |
43010659 PW |
3762 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3763 | sd, CPU_IDLE)) | |
48f24c4d IM |
3764 | schedstat_inc(sd, alb_pushed); |
3765 | else | |
3766 | schedstat_inc(sd, alb_failed); | |
3767 | } | |
1b12bbc7 | 3768 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3769 | } |
3770 | ||
46cb4b7c SS |
3771 | #ifdef CONFIG_NO_HZ |
3772 | static struct { | |
3773 | atomic_t load_balancer; | |
41a2d6cf | 3774 | cpumask_t cpu_mask; |
46cb4b7c SS |
3775 | } nohz ____cacheline_aligned = { |
3776 | .load_balancer = ATOMIC_INIT(-1), | |
3777 | .cpu_mask = CPU_MASK_NONE, | |
3778 | }; | |
3779 | ||
7835b98b | 3780 | /* |
46cb4b7c SS |
3781 | * This routine will try to nominate the ilb (idle load balancing) |
3782 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3783 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3784 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3785 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3786 | * arrives... | |
3787 | * | |
3788 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3789 | * for idle load balancing. ilb owner will still be part of | |
3790 | * nohz.cpu_mask.. | |
7835b98b | 3791 | * |
46cb4b7c SS |
3792 | * While stopping the tick, this cpu will become the ilb owner if there |
3793 | * is no other owner. And will be the owner till that cpu becomes busy | |
3794 | * or if all cpus in the system stop their ticks at which point | |
3795 | * there is no need for ilb owner. | |
3796 | * | |
3797 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3798 | * next busy scheduler_tick() | |
3799 | */ | |
3800 | int select_nohz_load_balancer(int stop_tick) | |
3801 | { | |
3802 | int cpu = smp_processor_id(); | |
3803 | ||
3804 | if (stop_tick) { | |
3805 | cpu_set(cpu, nohz.cpu_mask); | |
3806 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3807 | ||
3808 | /* | |
3809 | * If we are going offline and still the leader, give up! | |
3810 | */ | |
e761b772 | 3811 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3812 | atomic_read(&nohz.load_balancer) == cpu) { |
3813 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3814 | BUG(); | |
3815 | return 0; | |
3816 | } | |
3817 | ||
3818 | /* time for ilb owner also to sleep */ | |
3819 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3820 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3821 | atomic_set(&nohz.load_balancer, -1); | |
3822 | return 0; | |
3823 | } | |
3824 | ||
3825 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3826 | /* make me the ilb owner */ | |
3827 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3828 | return 1; | |
3829 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3830 | return 1; | |
3831 | } else { | |
3832 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3833 | return 0; | |
3834 | ||
3835 | cpu_clear(cpu, nohz.cpu_mask); | |
3836 | ||
3837 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3838 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3839 | BUG(); | |
3840 | } | |
3841 | return 0; | |
3842 | } | |
3843 | #endif | |
3844 | ||
3845 | static DEFINE_SPINLOCK(balancing); | |
3846 | ||
3847 | /* | |
7835b98b CL |
3848 | * It checks each scheduling domain to see if it is due to be balanced, |
3849 | * and initiates a balancing operation if so. | |
3850 | * | |
3851 | * Balancing parameters are set up in arch_init_sched_domains. | |
3852 | */ | |
a9957449 | 3853 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3854 | { |
46cb4b7c SS |
3855 | int balance = 1; |
3856 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3857 | unsigned long interval; |
3858 | struct sched_domain *sd; | |
46cb4b7c | 3859 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3860 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3861 | int update_next_balance = 0; |
d07355f5 | 3862 | int need_serialize; |
7c16ec58 | 3863 | cpumask_t tmp; |
1da177e4 | 3864 | |
46cb4b7c | 3865 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3866 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3867 | continue; | |
3868 | ||
3869 | interval = sd->balance_interval; | |
d15bcfdb | 3870 | if (idle != CPU_IDLE) |
1da177e4 LT |
3871 | interval *= sd->busy_factor; |
3872 | ||
3873 | /* scale ms to jiffies */ | |
3874 | interval = msecs_to_jiffies(interval); | |
3875 | if (unlikely(!interval)) | |
3876 | interval = 1; | |
dd41f596 IM |
3877 | if (interval > HZ*NR_CPUS/10) |
3878 | interval = HZ*NR_CPUS/10; | |
3879 | ||
d07355f5 | 3880 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3881 | |
d07355f5 | 3882 | if (need_serialize) { |
08c183f3 CL |
3883 | if (!spin_trylock(&balancing)) |
3884 | goto out; | |
3885 | } | |
3886 | ||
c9819f45 | 3887 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3888 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3889 | /* |
3890 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3891 | * longer idle, or one of our SMT siblings is |
3892 | * not idle. | |
3893 | */ | |
d15bcfdb | 3894 | idle = CPU_NOT_IDLE; |
1da177e4 | 3895 | } |
1bd77f2d | 3896 | sd->last_balance = jiffies; |
1da177e4 | 3897 | } |
d07355f5 | 3898 | if (need_serialize) |
08c183f3 CL |
3899 | spin_unlock(&balancing); |
3900 | out: | |
f549da84 | 3901 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3902 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3903 | update_next_balance = 1; |
3904 | } | |
783609c6 SS |
3905 | |
3906 | /* | |
3907 | * Stop the load balance at this level. There is another | |
3908 | * CPU in our sched group which is doing load balancing more | |
3909 | * actively. | |
3910 | */ | |
3911 | if (!balance) | |
3912 | break; | |
1da177e4 | 3913 | } |
f549da84 SS |
3914 | |
3915 | /* | |
3916 | * next_balance will be updated only when there is a need. | |
3917 | * When the cpu is attached to null domain for ex, it will not be | |
3918 | * updated. | |
3919 | */ | |
3920 | if (likely(update_next_balance)) | |
3921 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3922 | } |
3923 | ||
3924 | /* | |
3925 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3926 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3927 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3928 | */ | |
3929 | static void run_rebalance_domains(struct softirq_action *h) | |
3930 | { | |
dd41f596 IM |
3931 | int this_cpu = smp_processor_id(); |
3932 | struct rq *this_rq = cpu_rq(this_cpu); | |
3933 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3934 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3935 | |
dd41f596 | 3936 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3937 | |
3938 | #ifdef CONFIG_NO_HZ | |
3939 | /* | |
3940 | * If this cpu is the owner for idle load balancing, then do the | |
3941 | * balancing on behalf of the other idle cpus whose ticks are | |
3942 | * stopped. | |
3943 | */ | |
dd41f596 IM |
3944 | if (this_rq->idle_at_tick && |
3945 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3946 | cpumask_t cpus = nohz.cpu_mask; |
3947 | struct rq *rq; | |
3948 | int balance_cpu; | |
3949 | ||
dd41f596 | 3950 | cpu_clear(this_cpu, cpus); |
363ab6f1 | 3951 | for_each_cpu_mask_nr(balance_cpu, cpus) { |
46cb4b7c SS |
3952 | /* |
3953 | * If this cpu gets work to do, stop the load balancing | |
3954 | * work being done for other cpus. Next load | |
3955 | * balancing owner will pick it up. | |
3956 | */ | |
3957 | if (need_resched()) | |
3958 | break; | |
3959 | ||
de0cf899 | 3960 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3961 | |
3962 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3963 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3964 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3965 | } |
3966 | } | |
3967 | #endif | |
3968 | } | |
3969 | ||
3970 | /* | |
3971 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3972 | * | |
3973 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3974 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3975 | * if the whole system is idle. | |
3976 | */ | |
dd41f596 | 3977 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3978 | { |
46cb4b7c SS |
3979 | #ifdef CONFIG_NO_HZ |
3980 | /* | |
3981 | * If we were in the nohz mode recently and busy at the current | |
3982 | * scheduler tick, then check if we need to nominate new idle | |
3983 | * load balancer. | |
3984 | */ | |
3985 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3986 | rq->in_nohz_recently = 0; | |
3987 | ||
3988 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3989 | cpu_clear(cpu, nohz.cpu_mask); | |
3990 | atomic_set(&nohz.load_balancer, -1); | |
3991 | } | |
3992 | ||
3993 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3994 | /* | |
3995 | * simple selection for now: Nominate the | |
3996 | * first cpu in the nohz list to be the next | |
3997 | * ilb owner. | |
3998 | * | |
3999 | * TBD: Traverse the sched domains and nominate | |
4000 | * the nearest cpu in the nohz.cpu_mask. | |
4001 | */ | |
4002 | int ilb = first_cpu(nohz.cpu_mask); | |
4003 | ||
434d53b0 | 4004 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4005 | resched_cpu(ilb); |
4006 | } | |
4007 | } | |
4008 | ||
4009 | /* | |
4010 | * If this cpu is idle and doing idle load balancing for all the | |
4011 | * cpus with ticks stopped, is it time for that to stop? | |
4012 | */ | |
4013 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
4014 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
4015 | resched_cpu(cpu); | |
4016 | return; | |
4017 | } | |
4018 | ||
4019 | /* | |
4020 | * If this cpu is idle and the idle load balancing is done by | |
4021 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4022 | */ | |
4023 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
4024 | cpu_isset(cpu, nohz.cpu_mask)) | |
4025 | return; | |
4026 | #endif | |
4027 | if (time_after_eq(jiffies, rq->next_balance)) | |
4028 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4029 | } |
dd41f596 IM |
4030 | |
4031 | #else /* CONFIG_SMP */ | |
4032 | ||
1da177e4 LT |
4033 | /* |
4034 | * on UP we do not need to balance between CPUs: | |
4035 | */ | |
70b97a7f | 4036 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4037 | { |
4038 | } | |
dd41f596 | 4039 | |
1da177e4 LT |
4040 | #endif |
4041 | ||
1da177e4 LT |
4042 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4043 | ||
4044 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4045 | ||
4046 | /* | |
41b86e9c IM |
4047 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
4048 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 4049 | */ |
41b86e9c | 4050 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 4051 | { |
1da177e4 | 4052 | unsigned long flags; |
41b86e9c IM |
4053 | u64 ns, delta_exec; |
4054 | struct rq *rq; | |
48f24c4d | 4055 | |
41b86e9c IM |
4056 | rq = task_rq_lock(p, &flags); |
4057 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4058 | if (task_current(rq, p)) { |
a8e504d2 IM |
4059 | update_rq_clock(rq); |
4060 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4061 | if ((s64)delta_exec > 0) |
4062 | ns += delta_exec; | |
4063 | } | |
4064 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4065 | |
1da177e4 LT |
4066 | return ns; |
4067 | } | |
4068 | ||
1da177e4 LT |
4069 | /* |
4070 | * Account user cpu time to a process. | |
4071 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4072 | * @cputime: the cpu time spent in user space since the last update |
4073 | */ | |
4074 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4075 | { | |
4076 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4077 | cputime64_t tmp; | |
4078 | ||
4079 | p->utime = cputime_add(p->utime, cputime); | |
4080 | ||
4081 | /* Add user time to cpustat. */ | |
4082 | tmp = cputime_to_cputime64(cputime); | |
4083 | if (TASK_NICE(p) > 0) | |
4084 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4085 | else | |
4086 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4087 | /* Account for user time used */ |
4088 | acct_update_integrals(p); | |
1da177e4 LT |
4089 | } |
4090 | ||
94886b84 LV |
4091 | /* |
4092 | * Account guest cpu time to a process. | |
4093 | * @p: the process that the cpu time gets accounted to | |
4094 | * @cputime: the cpu time spent in virtual machine since the last update | |
4095 | */ | |
f7402e03 | 4096 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4097 | { |
4098 | cputime64_t tmp; | |
4099 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4100 | ||
4101 | tmp = cputime_to_cputime64(cputime); | |
4102 | ||
4103 | p->utime = cputime_add(p->utime, cputime); | |
4104 | p->gtime = cputime_add(p->gtime, cputime); | |
4105 | ||
4106 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4107 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4108 | } | |
4109 | ||
c66f08be MN |
4110 | /* |
4111 | * Account scaled user cpu time to a process. | |
4112 | * @p: the process that the cpu time gets accounted to | |
4113 | * @cputime: the cpu time spent in user space since the last update | |
4114 | */ | |
4115 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4116 | { | |
4117 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4118 | } | |
4119 | ||
1da177e4 LT |
4120 | /* |
4121 | * Account system cpu time to a process. | |
4122 | * @p: the process that the cpu time gets accounted to | |
4123 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4124 | * @cputime: the cpu time spent in kernel space since the last update | |
4125 | */ | |
4126 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4127 | cputime_t cputime) | |
4128 | { | |
4129 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4130 | struct rq *rq = this_rq(); |
1da177e4 LT |
4131 | cputime64_t tmp; |
4132 | ||
983ed7a6 HH |
4133 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4134 | account_guest_time(p, cputime); | |
4135 | return; | |
4136 | } | |
94886b84 | 4137 | |
1da177e4 LT |
4138 | p->stime = cputime_add(p->stime, cputime); |
4139 | ||
4140 | /* Add system time to cpustat. */ | |
4141 | tmp = cputime_to_cputime64(cputime); | |
4142 | if (hardirq_count() - hardirq_offset) | |
4143 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4144 | else if (softirq_count()) | |
4145 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4146 | else if (p != rq->idle) |
1da177e4 | 4147 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4148 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4149 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4150 | else | |
4151 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4152 | /* Account for system time used */ | |
4153 | acct_update_integrals(p); | |
1da177e4 LT |
4154 | } |
4155 | ||
c66f08be MN |
4156 | /* |
4157 | * Account scaled system cpu time to a process. | |
4158 | * @p: the process that the cpu time gets accounted to | |
4159 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4160 | * @cputime: the cpu time spent in kernel space since the last update | |
4161 | */ | |
4162 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4163 | { | |
4164 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4165 | } | |
4166 | ||
1da177e4 LT |
4167 | /* |
4168 | * Account for involuntary wait time. | |
4169 | * @p: the process from which the cpu time has been stolen | |
4170 | * @steal: the cpu time spent in involuntary wait | |
4171 | */ | |
4172 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4173 | { | |
4174 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4175 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4176 | struct rq *rq = this_rq(); |
1da177e4 LT |
4177 | |
4178 | if (p == rq->idle) { | |
4179 | p->stime = cputime_add(p->stime, steal); | |
4180 | if (atomic_read(&rq->nr_iowait) > 0) | |
4181 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4182 | else | |
4183 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4184 | } else |
1da177e4 LT |
4185 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4186 | } | |
4187 | ||
7835b98b CL |
4188 | /* |
4189 | * This function gets called by the timer code, with HZ frequency. | |
4190 | * We call it with interrupts disabled. | |
4191 | * | |
4192 | * It also gets called by the fork code, when changing the parent's | |
4193 | * timeslices. | |
4194 | */ | |
4195 | void scheduler_tick(void) | |
4196 | { | |
7835b98b CL |
4197 | int cpu = smp_processor_id(); |
4198 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4199 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4200 | |
4201 | sched_clock_tick(); | |
dd41f596 IM |
4202 | |
4203 | spin_lock(&rq->lock); | |
3e51f33f | 4204 | update_rq_clock(rq); |
f1a438d8 | 4205 | update_cpu_load(rq); |
fa85ae24 | 4206 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4207 | spin_unlock(&rq->lock); |
7835b98b | 4208 | |
e418e1c2 | 4209 | #ifdef CONFIG_SMP |
dd41f596 IM |
4210 | rq->idle_at_tick = idle_cpu(cpu); |
4211 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4212 | #endif |
1da177e4 LT |
4213 | } |
4214 | ||
6cd8a4bb SR |
4215 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4216 | defined(CONFIG_PREEMPT_TRACER)) | |
4217 | ||
4218 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4219 | { | |
4220 | if (in_lock_functions(addr)) { | |
4221 | addr = CALLER_ADDR2; | |
4222 | if (in_lock_functions(addr)) | |
4223 | addr = CALLER_ADDR3; | |
4224 | } | |
4225 | return addr; | |
4226 | } | |
1da177e4 | 4227 | |
43627582 | 4228 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4229 | { |
6cd8a4bb | 4230 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4231 | /* |
4232 | * Underflow? | |
4233 | */ | |
9a11b49a IM |
4234 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4235 | return; | |
6cd8a4bb | 4236 | #endif |
1da177e4 | 4237 | preempt_count() += val; |
6cd8a4bb | 4238 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4239 | /* |
4240 | * Spinlock count overflowing soon? | |
4241 | */ | |
33859f7f MOS |
4242 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4243 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4244 | #endif |
4245 | if (preempt_count() == val) | |
4246 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4247 | } |
4248 | EXPORT_SYMBOL(add_preempt_count); | |
4249 | ||
43627582 | 4250 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4251 | { |
6cd8a4bb | 4252 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4253 | /* |
4254 | * Underflow? | |
4255 | */ | |
9a11b49a IM |
4256 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4257 | return; | |
1da177e4 LT |
4258 | /* |
4259 | * Is the spinlock portion underflowing? | |
4260 | */ | |
9a11b49a IM |
4261 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4262 | !(preempt_count() & PREEMPT_MASK))) | |
4263 | return; | |
6cd8a4bb | 4264 | #endif |
9a11b49a | 4265 | |
6cd8a4bb SR |
4266 | if (preempt_count() == val) |
4267 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4268 | preempt_count() -= val; |
4269 | } | |
4270 | EXPORT_SYMBOL(sub_preempt_count); | |
4271 | ||
4272 | #endif | |
4273 | ||
4274 | /* | |
dd41f596 | 4275 | * Print scheduling while atomic bug: |
1da177e4 | 4276 | */ |
dd41f596 | 4277 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4278 | { |
838225b4 SS |
4279 | struct pt_regs *regs = get_irq_regs(); |
4280 | ||
4281 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4282 | prev->comm, prev->pid, preempt_count()); | |
4283 | ||
dd41f596 | 4284 | debug_show_held_locks(prev); |
e21f5b15 | 4285 | print_modules(); |
dd41f596 IM |
4286 | if (irqs_disabled()) |
4287 | print_irqtrace_events(prev); | |
838225b4 SS |
4288 | |
4289 | if (regs) | |
4290 | show_regs(regs); | |
4291 | else | |
4292 | dump_stack(); | |
dd41f596 | 4293 | } |
1da177e4 | 4294 | |
dd41f596 IM |
4295 | /* |
4296 | * Various schedule()-time debugging checks and statistics: | |
4297 | */ | |
4298 | static inline void schedule_debug(struct task_struct *prev) | |
4299 | { | |
1da177e4 | 4300 | /* |
41a2d6cf | 4301 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4302 | * schedule() atomically, we ignore that path for now. |
4303 | * Otherwise, whine if we are scheduling when we should not be. | |
4304 | */ | |
3f33a7ce | 4305 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4306 | __schedule_bug(prev); |
4307 | ||
1da177e4 LT |
4308 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4309 | ||
2d72376b | 4310 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4311 | #ifdef CONFIG_SCHEDSTATS |
4312 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4313 | schedstat_inc(this_rq(), bkl_count); |
4314 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4315 | } |
4316 | #endif | |
dd41f596 IM |
4317 | } |
4318 | ||
4319 | /* | |
4320 | * Pick up the highest-prio task: | |
4321 | */ | |
4322 | static inline struct task_struct * | |
ff95f3df | 4323 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4324 | { |
5522d5d5 | 4325 | const struct sched_class *class; |
dd41f596 | 4326 | struct task_struct *p; |
1da177e4 LT |
4327 | |
4328 | /* | |
dd41f596 IM |
4329 | * Optimization: we know that if all tasks are in |
4330 | * the fair class we can call that function directly: | |
1da177e4 | 4331 | */ |
dd41f596 | 4332 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4333 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4334 | if (likely(p)) |
4335 | return p; | |
1da177e4 LT |
4336 | } |
4337 | ||
dd41f596 IM |
4338 | class = sched_class_highest; |
4339 | for ( ; ; ) { | |
fb8d4724 | 4340 | p = class->pick_next_task(rq); |
dd41f596 IM |
4341 | if (p) |
4342 | return p; | |
4343 | /* | |
4344 | * Will never be NULL as the idle class always | |
4345 | * returns a non-NULL p: | |
4346 | */ | |
4347 | class = class->next; | |
4348 | } | |
4349 | } | |
1da177e4 | 4350 | |
dd41f596 IM |
4351 | /* |
4352 | * schedule() is the main scheduler function. | |
4353 | */ | |
4354 | asmlinkage void __sched schedule(void) | |
4355 | { | |
4356 | struct task_struct *prev, *next; | |
67ca7bde | 4357 | unsigned long *switch_count; |
dd41f596 | 4358 | struct rq *rq; |
31656519 | 4359 | int cpu; |
dd41f596 IM |
4360 | |
4361 | need_resched: | |
4362 | preempt_disable(); | |
4363 | cpu = smp_processor_id(); | |
4364 | rq = cpu_rq(cpu); | |
4365 | rcu_qsctr_inc(cpu); | |
4366 | prev = rq->curr; | |
4367 | switch_count = &prev->nivcsw; | |
4368 | ||
4369 | release_kernel_lock(prev); | |
4370 | need_resched_nonpreemptible: | |
4371 | ||
4372 | schedule_debug(prev); | |
1da177e4 | 4373 | |
31656519 | 4374 | if (sched_feat(HRTICK)) |
f333fdc9 | 4375 | hrtick_clear(rq); |
8f4d37ec | 4376 | |
1e819950 IM |
4377 | /* |
4378 | * Do the rq-clock update outside the rq lock: | |
4379 | */ | |
4380 | local_irq_disable(); | |
3e51f33f | 4381 | update_rq_clock(rq); |
1e819950 IM |
4382 | spin_lock(&rq->lock); |
4383 | clear_tsk_need_resched(prev); | |
1da177e4 | 4384 | |
1da177e4 | 4385 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4386 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4387 | prev->state = TASK_RUNNING; |
16882c1e | 4388 | else |
2e1cb74a | 4389 | deactivate_task(rq, prev, 1); |
dd41f596 | 4390 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4391 | } |
4392 | ||
9a897c5a SR |
4393 | #ifdef CONFIG_SMP |
4394 | if (prev->sched_class->pre_schedule) | |
4395 | prev->sched_class->pre_schedule(rq, prev); | |
4396 | #endif | |
f65eda4f | 4397 | |
dd41f596 | 4398 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4399 | idle_balance(cpu, rq); |
1da177e4 | 4400 | |
31ee529c | 4401 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4402 | next = pick_next_task(rq, prev); |
1da177e4 | 4403 | |
1da177e4 | 4404 | if (likely(prev != next)) { |
673a90a1 DS |
4405 | sched_info_switch(prev, next); |
4406 | ||
1da177e4 LT |
4407 | rq->nr_switches++; |
4408 | rq->curr = next; | |
4409 | ++*switch_count; | |
4410 | ||
dd41f596 | 4411 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4412 | /* |
4413 | * the context switch might have flipped the stack from under | |
4414 | * us, hence refresh the local variables. | |
4415 | */ | |
4416 | cpu = smp_processor_id(); | |
4417 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4418 | } else |
4419 | spin_unlock_irq(&rq->lock); | |
4420 | ||
8f4d37ec | 4421 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4422 | goto need_resched_nonpreemptible; |
8f4d37ec | 4423 | |
1da177e4 LT |
4424 | preempt_enable_no_resched(); |
4425 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4426 | goto need_resched; | |
4427 | } | |
1da177e4 LT |
4428 | EXPORT_SYMBOL(schedule); |
4429 | ||
4430 | #ifdef CONFIG_PREEMPT | |
4431 | /* | |
2ed6e34f | 4432 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4433 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4434 | * occur there and call schedule directly. |
4435 | */ | |
4436 | asmlinkage void __sched preempt_schedule(void) | |
4437 | { | |
4438 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4439 | |
1da177e4 LT |
4440 | /* |
4441 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4442 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4443 | */ |
beed33a8 | 4444 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4445 | return; |
4446 | ||
3a5c359a AK |
4447 | do { |
4448 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4449 | schedule(); |
3a5c359a | 4450 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4451 | |
3a5c359a AK |
4452 | /* |
4453 | * Check again in case we missed a preemption opportunity | |
4454 | * between schedule and now. | |
4455 | */ | |
4456 | barrier(); | |
4457 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4458 | } |
1da177e4 LT |
4459 | EXPORT_SYMBOL(preempt_schedule); |
4460 | ||
4461 | /* | |
2ed6e34f | 4462 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4463 | * off of irq context. |
4464 | * Note, that this is called and return with irqs disabled. This will | |
4465 | * protect us against recursive calling from irq. | |
4466 | */ | |
4467 | asmlinkage void __sched preempt_schedule_irq(void) | |
4468 | { | |
4469 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4470 | |
2ed6e34f | 4471 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4472 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4473 | ||
3a5c359a AK |
4474 | do { |
4475 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4476 | local_irq_enable(); |
4477 | schedule(); | |
4478 | local_irq_disable(); | |
3a5c359a | 4479 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4480 | |
3a5c359a AK |
4481 | /* |
4482 | * Check again in case we missed a preemption opportunity | |
4483 | * between schedule and now. | |
4484 | */ | |
4485 | barrier(); | |
4486 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4487 | } |
4488 | ||
4489 | #endif /* CONFIG_PREEMPT */ | |
4490 | ||
95cdf3b7 IM |
4491 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4492 | void *key) | |
1da177e4 | 4493 | { |
48f24c4d | 4494 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4495 | } |
1da177e4 LT |
4496 | EXPORT_SYMBOL(default_wake_function); |
4497 | ||
4498 | /* | |
41a2d6cf IM |
4499 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4500 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4501 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4502 | * | |
4503 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4504 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4505 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4506 | */ | |
4507 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4508 | int nr_exclusive, int sync, void *key) | |
4509 | { | |
2e45874c | 4510 | wait_queue_t *curr, *next; |
1da177e4 | 4511 | |
2e45874c | 4512 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4513 | unsigned flags = curr->flags; |
4514 | ||
1da177e4 | 4515 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4516 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4517 | break; |
4518 | } | |
4519 | } | |
4520 | ||
4521 | /** | |
4522 | * __wake_up - wake up threads blocked on a waitqueue. | |
4523 | * @q: the waitqueue | |
4524 | * @mode: which threads | |
4525 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4526 | * @key: is directly passed to the wakeup function |
1da177e4 | 4527 | */ |
7ad5b3a5 | 4528 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4529 | int nr_exclusive, void *key) |
1da177e4 LT |
4530 | { |
4531 | unsigned long flags; | |
4532 | ||
4533 | spin_lock_irqsave(&q->lock, flags); | |
4534 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4535 | spin_unlock_irqrestore(&q->lock, flags); | |
4536 | } | |
1da177e4 LT |
4537 | EXPORT_SYMBOL(__wake_up); |
4538 | ||
4539 | /* | |
4540 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4541 | */ | |
7ad5b3a5 | 4542 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4543 | { |
4544 | __wake_up_common(q, mode, 1, 0, NULL); | |
4545 | } | |
4546 | ||
4547 | /** | |
67be2dd1 | 4548 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4549 | * @q: the waitqueue |
4550 | * @mode: which threads | |
4551 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4552 | * | |
4553 | * The sync wakeup differs that the waker knows that it will schedule | |
4554 | * away soon, so while the target thread will be woken up, it will not | |
4555 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4556 | * with each other. This can prevent needless bouncing between CPUs. | |
4557 | * | |
4558 | * On UP it can prevent extra preemption. | |
4559 | */ | |
7ad5b3a5 | 4560 | void |
95cdf3b7 | 4561 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4562 | { |
4563 | unsigned long flags; | |
4564 | int sync = 1; | |
4565 | ||
4566 | if (unlikely(!q)) | |
4567 | return; | |
4568 | ||
4569 | if (unlikely(!nr_exclusive)) | |
4570 | sync = 0; | |
4571 | ||
4572 | spin_lock_irqsave(&q->lock, flags); | |
4573 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4574 | spin_unlock_irqrestore(&q->lock, flags); | |
4575 | } | |
4576 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4577 | ||
b15136e9 | 4578 | void complete(struct completion *x) |
1da177e4 LT |
4579 | { |
4580 | unsigned long flags; | |
4581 | ||
4582 | spin_lock_irqsave(&x->wait.lock, flags); | |
4583 | x->done++; | |
d9514f6c | 4584 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4585 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4586 | } | |
4587 | EXPORT_SYMBOL(complete); | |
4588 | ||
b15136e9 | 4589 | void complete_all(struct completion *x) |
1da177e4 LT |
4590 | { |
4591 | unsigned long flags; | |
4592 | ||
4593 | spin_lock_irqsave(&x->wait.lock, flags); | |
4594 | x->done += UINT_MAX/2; | |
d9514f6c | 4595 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4596 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4597 | } | |
4598 | EXPORT_SYMBOL(complete_all); | |
4599 | ||
8cbbe86d AK |
4600 | static inline long __sched |
4601 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4602 | { |
1da177e4 LT |
4603 | if (!x->done) { |
4604 | DECLARE_WAITQUEUE(wait, current); | |
4605 | ||
4606 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4607 | __add_wait_queue_tail(&x->wait, &wait); | |
4608 | do { | |
009e577e MW |
4609 | if ((state == TASK_INTERRUPTIBLE && |
4610 | signal_pending(current)) || | |
4611 | (state == TASK_KILLABLE && | |
4612 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4613 | timeout = -ERESTARTSYS; |
4614 | break; | |
8cbbe86d AK |
4615 | } |
4616 | __set_current_state(state); | |
1da177e4 LT |
4617 | spin_unlock_irq(&x->wait.lock); |
4618 | timeout = schedule_timeout(timeout); | |
4619 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4620 | } while (!x->done && timeout); |
1da177e4 | 4621 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4622 | if (!x->done) |
4623 | return timeout; | |
1da177e4 LT |
4624 | } |
4625 | x->done--; | |
ea71a546 | 4626 | return timeout ?: 1; |
1da177e4 | 4627 | } |
1da177e4 | 4628 | |
8cbbe86d AK |
4629 | static long __sched |
4630 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4631 | { |
1da177e4 LT |
4632 | might_sleep(); |
4633 | ||
4634 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4635 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4636 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4637 | return timeout; |
4638 | } | |
1da177e4 | 4639 | |
b15136e9 | 4640 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4641 | { |
4642 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4643 | } |
8cbbe86d | 4644 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4645 | |
b15136e9 | 4646 | unsigned long __sched |
8cbbe86d | 4647 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4648 | { |
8cbbe86d | 4649 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4650 | } |
8cbbe86d | 4651 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4652 | |
8cbbe86d | 4653 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4654 | { |
51e97990 AK |
4655 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4656 | if (t == -ERESTARTSYS) | |
4657 | return t; | |
4658 | return 0; | |
0fec171c | 4659 | } |
8cbbe86d | 4660 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4661 | |
b15136e9 | 4662 | unsigned long __sched |
8cbbe86d AK |
4663 | wait_for_completion_interruptible_timeout(struct completion *x, |
4664 | unsigned long timeout) | |
0fec171c | 4665 | { |
8cbbe86d | 4666 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4667 | } |
8cbbe86d | 4668 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4669 | |
009e577e MW |
4670 | int __sched wait_for_completion_killable(struct completion *x) |
4671 | { | |
4672 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4673 | if (t == -ERESTARTSYS) | |
4674 | return t; | |
4675 | return 0; | |
4676 | } | |
4677 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4678 | ||
be4de352 DC |
4679 | /** |
4680 | * try_wait_for_completion - try to decrement a completion without blocking | |
4681 | * @x: completion structure | |
4682 | * | |
4683 | * Returns: 0 if a decrement cannot be done without blocking | |
4684 | * 1 if a decrement succeeded. | |
4685 | * | |
4686 | * If a completion is being used as a counting completion, | |
4687 | * attempt to decrement the counter without blocking. This | |
4688 | * enables us to avoid waiting if the resource the completion | |
4689 | * is protecting is not available. | |
4690 | */ | |
4691 | bool try_wait_for_completion(struct completion *x) | |
4692 | { | |
4693 | int ret = 1; | |
4694 | ||
4695 | spin_lock_irq(&x->wait.lock); | |
4696 | if (!x->done) | |
4697 | ret = 0; | |
4698 | else | |
4699 | x->done--; | |
4700 | spin_unlock_irq(&x->wait.lock); | |
4701 | return ret; | |
4702 | } | |
4703 | EXPORT_SYMBOL(try_wait_for_completion); | |
4704 | ||
4705 | /** | |
4706 | * completion_done - Test to see if a completion has any waiters | |
4707 | * @x: completion structure | |
4708 | * | |
4709 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4710 | * 1 if there are no waiters. | |
4711 | * | |
4712 | */ | |
4713 | bool completion_done(struct completion *x) | |
4714 | { | |
4715 | int ret = 1; | |
4716 | ||
4717 | spin_lock_irq(&x->wait.lock); | |
4718 | if (!x->done) | |
4719 | ret = 0; | |
4720 | spin_unlock_irq(&x->wait.lock); | |
4721 | return ret; | |
4722 | } | |
4723 | EXPORT_SYMBOL(completion_done); | |
4724 | ||
8cbbe86d AK |
4725 | static long __sched |
4726 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4727 | { |
0fec171c IM |
4728 | unsigned long flags; |
4729 | wait_queue_t wait; | |
4730 | ||
4731 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4732 | |
8cbbe86d | 4733 | __set_current_state(state); |
1da177e4 | 4734 | |
8cbbe86d AK |
4735 | spin_lock_irqsave(&q->lock, flags); |
4736 | __add_wait_queue(q, &wait); | |
4737 | spin_unlock(&q->lock); | |
4738 | timeout = schedule_timeout(timeout); | |
4739 | spin_lock_irq(&q->lock); | |
4740 | __remove_wait_queue(q, &wait); | |
4741 | spin_unlock_irqrestore(&q->lock, flags); | |
4742 | ||
4743 | return timeout; | |
4744 | } | |
4745 | ||
4746 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4747 | { | |
4748 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4749 | } |
1da177e4 LT |
4750 | EXPORT_SYMBOL(interruptible_sleep_on); |
4751 | ||
0fec171c | 4752 | long __sched |
95cdf3b7 | 4753 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4754 | { |
8cbbe86d | 4755 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4756 | } |
1da177e4 LT |
4757 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4758 | ||
0fec171c | 4759 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4760 | { |
8cbbe86d | 4761 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4762 | } |
1da177e4 LT |
4763 | EXPORT_SYMBOL(sleep_on); |
4764 | ||
0fec171c | 4765 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4766 | { |
8cbbe86d | 4767 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4768 | } |
1da177e4 LT |
4769 | EXPORT_SYMBOL(sleep_on_timeout); |
4770 | ||
b29739f9 IM |
4771 | #ifdef CONFIG_RT_MUTEXES |
4772 | ||
4773 | /* | |
4774 | * rt_mutex_setprio - set the current priority of a task | |
4775 | * @p: task | |
4776 | * @prio: prio value (kernel-internal form) | |
4777 | * | |
4778 | * This function changes the 'effective' priority of a task. It does | |
4779 | * not touch ->normal_prio like __setscheduler(). | |
4780 | * | |
4781 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4782 | */ | |
36c8b586 | 4783 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4784 | { |
4785 | unsigned long flags; | |
83b699ed | 4786 | int oldprio, on_rq, running; |
70b97a7f | 4787 | struct rq *rq; |
cb469845 | 4788 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4789 | |
4790 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4791 | ||
4792 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4793 | update_rq_clock(rq); |
b29739f9 | 4794 | |
d5f9f942 | 4795 | oldprio = p->prio; |
dd41f596 | 4796 | on_rq = p->se.on_rq; |
051a1d1a | 4797 | running = task_current(rq, p); |
0e1f3483 | 4798 | if (on_rq) |
69be72c1 | 4799 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4800 | if (running) |
4801 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4802 | |
4803 | if (rt_prio(prio)) | |
4804 | p->sched_class = &rt_sched_class; | |
4805 | else | |
4806 | p->sched_class = &fair_sched_class; | |
4807 | ||
b29739f9 IM |
4808 | p->prio = prio; |
4809 | ||
0e1f3483 HS |
4810 | if (running) |
4811 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4812 | if (on_rq) { |
8159f87e | 4813 | enqueue_task(rq, p, 0); |
cb469845 SR |
4814 | |
4815 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4816 | } |
4817 | task_rq_unlock(rq, &flags); | |
4818 | } | |
4819 | ||
4820 | #endif | |
4821 | ||
36c8b586 | 4822 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4823 | { |
dd41f596 | 4824 | int old_prio, delta, on_rq; |
1da177e4 | 4825 | unsigned long flags; |
70b97a7f | 4826 | struct rq *rq; |
1da177e4 LT |
4827 | |
4828 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4829 | return; | |
4830 | /* | |
4831 | * We have to be careful, if called from sys_setpriority(), | |
4832 | * the task might be in the middle of scheduling on another CPU. | |
4833 | */ | |
4834 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4835 | update_rq_clock(rq); |
1da177e4 LT |
4836 | /* |
4837 | * The RT priorities are set via sched_setscheduler(), but we still | |
4838 | * allow the 'normal' nice value to be set - but as expected | |
4839 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4840 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4841 | */ |
e05606d3 | 4842 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4843 | p->static_prio = NICE_TO_PRIO(nice); |
4844 | goto out_unlock; | |
4845 | } | |
dd41f596 | 4846 | on_rq = p->se.on_rq; |
c09595f6 | 4847 | if (on_rq) |
69be72c1 | 4848 | dequeue_task(rq, p, 0); |
1da177e4 | 4849 | |
1da177e4 | 4850 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4851 | set_load_weight(p); |
b29739f9 IM |
4852 | old_prio = p->prio; |
4853 | p->prio = effective_prio(p); | |
4854 | delta = p->prio - old_prio; | |
1da177e4 | 4855 | |
dd41f596 | 4856 | if (on_rq) { |
8159f87e | 4857 | enqueue_task(rq, p, 0); |
1da177e4 | 4858 | /* |
d5f9f942 AM |
4859 | * If the task increased its priority or is running and |
4860 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4861 | */ |
d5f9f942 | 4862 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4863 | resched_task(rq->curr); |
4864 | } | |
4865 | out_unlock: | |
4866 | task_rq_unlock(rq, &flags); | |
4867 | } | |
1da177e4 LT |
4868 | EXPORT_SYMBOL(set_user_nice); |
4869 | ||
e43379f1 MM |
4870 | /* |
4871 | * can_nice - check if a task can reduce its nice value | |
4872 | * @p: task | |
4873 | * @nice: nice value | |
4874 | */ | |
36c8b586 | 4875 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4876 | { |
024f4747 MM |
4877 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4878 | int nice_rlim = 20 - nice; | |
48f24c4d | 4879 | |
e43379f1 MM |
4880 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4881 | capable(CAP_SYS_NICE)); | |
4882 | } | |
4883 | ||
1da177e4 LT |
4884 | #ifdef __ARCH_WANT_SYS_NICE |
4885 | ||
4886 | /* | |
4887 | * sys_nice - change the priority of the current process. | |
4888 | * @increment: priority increment | |
4889 | * | |
4890 | * sys_setpriority is a more generic, but much slower function that | |
4891 | * does similar things. | |
4892 | */ | |
4893 | asmlinkage long sys_nice(int increment) | |
4894 | { | |
48f24c4d | 4895 | long nice, retval; |
1da177e4 LT |
4896 | |
4897 | /* | |
4898 | * Setpriority might change our priority at the same moment. | |
4899 | * We don't have to worry. Conceptually one call occurs first | |
4900 | * and we have a single winner. | |
4901 | */ | |
e43379f1 MM |
4902 | if (increment < -40) |
4903 | increment = -40; | |
1da177e4 LT |
4904 | if (increment > 40) |
4905 | increment = 40; | |
4906 | ||
4907 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4908 | if (nice < -20) | |
4909 | nice = -20; | |
4910 | if (nice > 19) | |
4911 | nice = 19; | |
4912 | ||
e43379f1 MM |
4913 | if (increment < 0 && !can_nice(current, nice)) |
4914 | return -EPERM; | |
4915 | ||
1da177e4 LT |
4916 | retval = security_task_setnice(current, nice); |
4917 | if (retval) | |
4918 | return retval; | |
4919 | ||
4920 | set_user_nice(current, nice); | |
4921 | return 0; | |
4922 | } | |
4923 | ||
4924 | #endif | |
4925 | ||
4926 | /** | |
4927 | * task_prio - return the priority value of a given task. | |
4928 | * @p: the task in question. | |
4929 | * | |
4930 | * This is the priority value as seen by users in /proc. | |
4931 | * RT tasks are offset by -200. Normal tasks are centered | |
4932 | * around 0, value goes from -16 to +15. | |
4933 | */ | |
36c8b586 | 4934 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4935 | { |
4936 | return p->prio - MAX_RT_PRIO; | |
4937 | } | |
4938 | ||
4939 | /** | |
4940 | * task_nice - return the nice value of a given task. | |
4941 | * @p: the task in question. | |
4942 | */ | |
36c8b586 | 4943 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4944 | { |
4945 | return TASK_NICE(p); | |
4946 | } | |
150d8bed | 4947 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4948 | |
4949 | /** | |
4950 | * idle_cpu - is a given cpu idle currently? | |
4951 | * @cpu: the processor in question. | |
4952 | */ | |
4953 | int idle_cpu(int cpu) | |
4954 | { | |
4955 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4956 | } | |
4957 | ||
1da177e4 LT |
4958 | /** |
4959 | * idle_task - return the idle task for a given cpu. | |
4960 | * @cpu: the processor in question. | |
4961 | */ | |
36c8b586 | 4962 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4963 | { |
4964 | return cpu_rq(cpu)->idle; | |
4965 | } | |
4966 | ||
4967 | /** | |
4968 | * find_process_by_pid - find a process with a matching PID value. | |
4969 | * @pid: the pid in question. | |
4970 | */ | |
a9957449 | 4971 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4972 | { |
228ebcbe | 4973 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4974 | } |
4975 | ||
4976 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4977 | static void |
4978 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4979 | { |
dd41f596 | 4980 | BUG_ON(p->se.on_rq); |
48f24c4d | 4981 | |
1da177e4 | 4982 | p->policy = policy; |
dd41f596 IM |
4983 | switch (p->policy) { |
4984 | case SCHED_NORMAL: | |
4985 | case SCHED_BATCH: | |
4986 | case SCHED_IDLE: | |
4987 | p->sched_class = &fair_sched_class; | |
4988 | break; | |
4989 | case SCHED_FIFO: | |
4990 | case SCHED_RR: | |
4991 | p->sched_class = &rt_sched_class; | |
4992 | break; | |
4993 | } | |
4994 | ||
1da177e4 | 4995 | p->rt_priority = prio; |
b29739f9 IM |
4996 | p->normal_prio = normal_prio(p); |
4997 | /* we are holding p->pi_lock already */ | |
4998 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4999 | set_load_weight(p); |
1da177e4 LT |
5000 | } |
5001 | ||
961ccddd RR |
5002 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5003 | struct sched_param *param, bool user) | |
1da177e4 | 5004 | { |
83b699ed | 5005 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5006 | unsigned long flags; |
cb469845 | 5007 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5008 | struct rq *rq; |
1da177e4 | 5009 | |
66e5393a SR |
5010 | /* may grab non-irq protected spin_locks */ |
5011 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5012 | recheck: |
5013 | /* double check policy once rq lock held */ | |
5014 | if (policy < 0) | |
5015 | policy = oldpolicy = p->policy; | |
5016 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5017 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5018 | policy != SCHED_IDLE) | |
b0a9499c | 5019 | return -EINVAL; |
1da177e4 LT |
5020 | /* |
5021 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5022 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5023 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5024 | */ |
5025 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5026 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5027 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5028 | return -EINVAL; |
e05606d3 | 5029 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5030 | return -EINVAL; |
5031 | ||
37e4ab3f OC |
5032 | /* |
5033 | * Allow unprivileged RT tasks to decrease priority: | |
5034 | */ | |
961ccddd | 5035 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5036 | if (rt_policy(policy)) { |
8dc3e909 | 5037 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5038 | |
5039 | if (!lock_task_sighand(p, &flags)) | |
5040 | return -ESRCH; | |
5041 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5042 | unlock_task_sighand(p, &flags); | |
5043 | ||
5044 | /* can't set/change the rt policy */ | |
5045 | if (policy != p->policy && !rlim_rtprio) | |
5046 | return -EPERM; | |
5047 | ||
5048 | /* can't increase priority */ | |
5049 | if (param->sched_priority > p->rt_priority && | |
5050 | param->sched_priority > rlim_rtprio) | |
5051 | return -EPERM; | |
5052 | } | |
dd41f596 IM |
5053 | /* |
5054 | * Like positive nice levels, dont allow tasks to | |
5055 | * move out of SCHED_IDLE either: | |
5056 | */ | |
5057 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5058 | return -EPERM; | |
5fe1d75f | 5059 | |
37e4ab3f OC |
5060 | /* can't change other user's priorities */ |
5061 | if ((current->euid != p->euid) && | |
5062 | (current->euid != p->uid)) | |
5063 | return -EPERM; | |
5064 | } | |
1da177e4 | 5065 | |
725aad24 | 5066 | if (user) { |
b68aa230 | 5067 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5068 | /* |
5069 | * Do not allow realtime tasks into groups that have no runtime | |
5070 | * assigned. | |
5071 | */ | |
5072 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) | |
5073 | return -EPERM; | |
b68aa230 PZ |
5074 | #endif |
5075 | ||
725aad24 JF |
5076 | retval = security_task_setscheduler(p, policy, param); |
5077 | if (retval) | |
5078 | return retval; | |
5079 | } | |
5080 | ||
b29739f9 IM |
5081 | /* |
5082 | * make sure no PI-waiters arrive (or leave) while we are | |
5083 | * changing the priority of the task: | |
5084 | */ | |
5085 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5086 | /* |
5087 | * To be able to change p->policy safely, the apropriate | |
5088 | * runqueue lock must be held. | |
5089 | */ | |
b29739f9 | 5090 | rq = __task_rq_lock(p); |
1da177e4 LT |
5091 | /* recheck policy now with rq lock held */ |
5092 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5093 | policy = oldpolicy = -1; | |
b29739f9 IM |
5094 | __task_rq_unlock(rq); |
5095 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5096 | goto recheck; |
5097 | } | |
2daa3577 | 5098 | update_rq_clock(rq); |
dd41f596 | 5099 | on_rq = p->se.on_rq; |
051a1d1a | 5100 | running = task_current(rq, p); |
0e1f3483 | 5101 | if (on_rq) |
2e1cb74a | 5102 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5103 | if (running) |
5104 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5105 | |
1da177e4 | 5106 | oldprio = p->prio; |
dd41f596 | 5107 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5108 | |
0e1f3483 HS |
5109 | if (running) |
5110 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5111 | if (on_rq) { |
5112 | activate_task(rq, p, 0); | |
cb469845 SR |
5113 | |
5114 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5115 | } |
b29739f9 IM |
5116 | __task_rq_unlock(rq); |
5117 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5118 | ||
95e02ca9 TG |
5119 | rt_mutex_adjust_pi(p); |
5120 | ||
1da177e4 LT |
5121 | return 0; |
5122 | } | |
961ccddd RR |
5123 | |
5124 | /** | |
5125 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5126 | * @p: the task in question. | |
5127 | * @policy: new policy. | |
5128 | * @param: structure containing the new RT priority. | |
5129 | * | |
5130 | * NOTE that the task may be already dead. | |
5131 | */ | |
5132 | int sched_setscheduler(struct task_struct *p, int policy, | |
5133 | struct sched_param *param) | |
5134 | { | |
5135 | return __sched_setscheduler(p, policy, param, true); | |
5136 | } | |
1da177e4 LT |
5137 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5138 | ||
961ccddd RR |
5139 | /** |
5140 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5141 | * @p: the task in question. | |
5142 | * @policy: new policy. | |
5143 | * @param: structure containing the new RT priority. | |
5144 | * | |
5145 | * Just like sched_setscheduler, only don't bother checking if the | |
5146 | * current context has permission. For example, this is needed in | |
5147 | * stop_machine(): we create temporary high priority worker threads, | |
5148 | * but our caller might not have that capability. | |
5149 | */ | |
5150 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5151 | struct sched_param *param) | |
5152 | { | |
5153 | return __sched_setscheduler(p, policy, param, false); | |
5154 | } | |
5155 | ||
95cdf3b7 IM |
5156 | static int |
5157 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5158 | { |
1da177e4 LT |
5159 | struct sched_param lparam; |
5160 | struct task_struct *p; | |
36c8b586 | 5161 | int retval; |
1da177e4 LT |
5162 | |
5163 | if (!param || pid < 0) | |
5164 | return -EINVAL; | |
5165 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5166 | return -EFAULT; | |
5fe1d75f ON |
5167 | |
5168 | rcu_read_lock(); | |
5169 | retval = -ESRCH; | |
1da177e4 | 5170 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5171 | if (p != NULL) |
5172 | retval = sched_setscheduler(p, policy, &lparam); | |
5173 | rcu_read_unlock(); | |
36c8b586 | 5174 | |
1da177e4 LT |
5175 | return retval; |
5176 | } | |
5177 | ||
5178 | /** | |
5179 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5180 | * @pid: the pid in question. | |
5181 | * @policy: new policy. | |
5182 | * @param: structure containing the new RT priority. | |
5183 | */ | |
41a2d6cf IM |
5184 | asmlinkage long |
5185 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5186 | { |
c21761f1 JB |
5187 | /* negative values for policy are not valid */ |
5188 | if (policy < 0) | |
5189 | return -EINVAL; | |
5190 | ||
1da177e4 LT |
5191 | return do_sched_setscheduler(pid, policy, param); |
5192 | } | |
5193 | ||
5194 | /** | |
5195 | * sys_sched_setparam - set/change the RT priority of a thread | |
5196 | * @pid: the pid in question. | |
5197 | * @param: structure containing the new RT priority. | |
5198 | */ | |
5199 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5200 | { | |
5201 | return do_sched_setscheduler(pid, -1, param); | |
5202 | } | |
5203 | ||
5204 | /** | |
5205 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5206 | * @pid: the pid in question. | |
5207 | */ | |
5208 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5209 | { | |
36c8b586 | 5210 | struct task_struct *p; |
3a5c359a | 5211 | int retval; |
1da177e4 LT |
5212 | |
5213 | if (pid < 0) | |
3a5c359a | 5214 | return -EINVAL; |
1da177e4 LT |
5215 | |
5216 | retval = -ESRCH; | |
5217 | read_lock(&tasklist_lock); | |
5218 | p = find_process_by_pid(pid); | |
5219 | if (p) { | |
5220 | retval = security_task_getscheduler(p); | |
5221 | if (!retval) | |
5222 | retval = p->policy; | |
5223 | } | |
5224 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5225 | return retval; |
5226 | } | |
5227 | ||
5228 | /** | |
5229 | * sys_sched_getscheduler - get the RT priority of a thread | |
5230 | * @pid: the pid in question. | |
5231 | * @param: structure containing the RT priority. | |
5232 | */ | |
5233 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5234 | { | |
5235 | struct sched_param lp; | |
36c8b586 | 5236 | struct task_struct *p; |
3a5c359a | 5237 | int retval; |
1da177e4 LT |
5238 | |
5239 | if (!param || pid < 0) | |
3a5c359a | 5240 | return -EINVAL; |
1da177e4 LT |
5241 | |
5242 | read_lock(&tasklist_lock); | |
5243 | p = find_process_by_pid(pid); | |
5244 | retval = -ESRCH; | |
5245 | if (!p) | |
5246 | goto out_unlock; | |
5247 | ||
5248 | retval = security_task_getscheduler(p); | |
5249 | if (retval) | |
5250 | goto out_unlock; | |
5251 | ||
5252 | lp.sched_priority = p->rt_priority; | |
5253 | read_unlock(&tasklist_lock); | |
5254 | ||
5255 | /* | |
5256 | * This one might sleep, we cannot do it with a spinlock held ... | |
5257 | */ | |
5258 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5259 | ||
1da177e4 LT |
5260 | return retval; |
5261 | ||
5262 | out_unlock: | |
5263 | read_unlock(&tasklist_lock); | |
5264 | return retval; | |
5265 | } | |
5266 | ||
b53e921b | 5267 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5268 | { |
1da177e4 | 5269 | cpumask_t cpus_allowed; |
b53e921b | 5270 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5271 | struct task_struct *p; |
5272 | int retval; | |
1da177e4 | 5273 | |
95402b38 | 5274 | get_online_cpus(); |
1da177e4 LT |
5275 | read_lock(&tasklist_lock); |
5276 | ||
5277 | p = find_process_by_pid(pid); | |
5278 | if (!p) { | |
5279 | read_unlock(&tasklist_lock); | |
95402b38 | 5280 | put_online_cpus(); |
1da177e4 LT |
5281 | return -ESRCH; |
5282 | } | |
5283 | ||
5284 | /* | |
5285 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5286 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5287 | * usage count and then drop tasklist_lock. |
5288 | */ | |
5289 | get_task_struct(p); | |
5290 | read_unlock(&tasklist_lock); | |
5291 | ||
5292 | retval = -EPERM; | |
5293 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5294 | !capable(CAP_SYS_NICE)) | |
5295 | goto out_unlock; | |
5296 | ||
e7834f8f DQ |
5297 | retval = security_task_setscheduler(p, 0, NULL); |
5298 | if (retval) | |
5299 | goto out_unlock; | |
5300 | ||
f9a86fcb | 5301 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5302 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5303 | again: |
7c16ec58 | 5304 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5305 | |
8707d8b8 | 5306 | if (!retval) { |
f9a86fcb | 5307 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5308 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5309 | /* | |
5310 | * We must have raced with a concurrent cpuset | |
5311 | * update. Just reset the cpus_allowed to the | |
5312 | * cpuset's cpus_allowed | |
5313 | */ | |
5314 | new_mask = cpus_allowed; | |
5315 | goto again; | |
5316 | } | |
5317 | } | |
1da177e4 LT |
5318 | out_unlock: |
5319 | put_task_struct(p); | |
95402b38 | 5320 | put_online_cpus(); |
1da177e4 LT |
5321 | return retval; |
5322 | } | |
5323 | ||
5324 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5325 | cpumask_t *new_mask) | |
5326 | { | |
5327 | if (len < sizeof(cpumask_t)) { | |
5328 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5329 | } else if (len > sizeof(cpumask_t)) { | |
5330 | len = sizeof(cpumask_t); | |
5331 | } | |
5332 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5333 | } | |
5334 | ||
5335 | /** | |
5336 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5337 | * @pid: pid of the process | |
5338 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5339 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5340 | */ | |
5341 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5342 | unsigned long __user *user_mask_ptr) | |
5343 | { | |
5344 | cpumask_t new_mask; | |
5345 | int retval; | |
5346 | ||
5347 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5348 | if (retval) | |
5349 | return retval; | |
5350 | ||
b53e921b | 5351 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5352 | } |
5353 | ||
1da177e4 LT |
5354 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5355 | { | |
36c8b586 | 5356 | struct task_struct *p; |
1da177e4 | 5357 | int retval; |
1da177e4 | 5358 | |
95402b38 | 5359 | get_online_cpus(); |
1da177e4 LT |
5360 | read_lock(&tasklist_lock); |
5361 | ||
5362 | retval = -ESRCH; | |
5363 | p = find_process_by_pid(pid); | |
5364 | if (!p) | |
5365 | goto out_unlock; | |
5366 | ||
e7834f8f DQ |
5367 | retval = security_task_getscheduler(p); |
5368 | if (retval) | |
5369 | goto out_unlock; | |
5370 | ||
2f7016d9 | 5371 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5372 | |
5373 | out_unlock: | |
5374 | read_unlock(&tasklist_lock); | |
95402b38 | 5375 | put_online_cpus(); |
1da177e4 | 5376 | |
9531b62f | 5377 | return retval; |
1da177e4 LT |
5378 | } |
5379 | ||
5380 | /** | |
5381 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5382 | * @pid: pid of the process | |
5383 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5384 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5385 | */ | |
5386 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5387 | unsigned long __user *user_mask_ptr) | |
5388 | { | |
5389 | int ret; | |
5390 | cpumask_t mask; | |
5391 | ||
5392 | if (len < sizeof(cpumask_t)) | |
5393 | return -EINVAL; | |
5394 | ||
5395 | ret = sched_getaffinity(pid, &mask); | |
5396 | if (ret < 0) | |
5397 | return ret; | |
5398 | ||
5399 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5400 | return -EFAULT; | |
5401 | ||
5402 | return sizeof(cpumask_t); | |
5403 | } | |
5404 | ||
5405 | /** | |
5406 | * sys_sched_yield - yield the current processor to other threads. | |
5407 | * | |
dd41f596 IM |
5408 | * This function yields the current CPU to other tasks. If there are no |
5409 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5410 | */ |
5411 | asmlinkage long sys_sched_yield(void) | |
5412 | { | |
70b97a7f | 5413 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5414 | |
2d72376b | 5415 | schedstat_inc(rq, yld_count); |
4530d7ab | 5416 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5417 | |
5418 | /* | |
5419 | * Since we are going to call schedule() anyway, there's | |
5420 | * no need to preempt or enable interrupts: | |
5421 | */ | |
5422 | __release(rq->lock); | |
8a25d5de | 5423 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5424 | _raw_spin_unlock(&rq->lock); |
5425 | preempt_enable_no_resched(); | |
5426 | ||
5427 | schedule(); | |
5428 | ||
5429 | return 0; | |
5430 | } | |
5431 | ||
e7b38404 | 5432 | static void __cond_resched(void) |
1da177e4 | 5433 | { |
8e0a43d8 IM |
5434 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5435 | __might_sleep(__FILE__, __LINE__); | |
5436 | #endif | |
5bbcfd90 IM |
5437 | /* |
5438 | * The BKS might be reacquired before we have dropped | |
5439 | * PREEMPT_ACTIVE, which could trigger a second | |
5440 | * cond_resched() call. | |
5441 | */ | |
1da177e4 LT |
5442 | do { |
5443 | add_preempt_count(PREEMPT_ACTIVE); | |
5444 | schedule(); | |
5445 | sub_preempt_count(PREEMPT_ACTIVE); | |
5446 | } while (need_resched()); | |
5447 | } | |
5448 | ||
02b67cc3 | 5449 | int __sched _cond_resched(void) |
1da177e4 | 5450 | { |
9414232f IM |
5451 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5452 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5453 | __cond_resched(); |
5454 | return 1; | |
5455 | } | |
5456 | return 0; | |
5457 | } | |
02b67cc3 | 5458 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5459 | |
5460 | /* | |
5461 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5462 | * call schedule, and on return reacquire the lock. | |
5463 | * | |
41a2d6cf | 5464 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5465 | * operations here to prevent schedule() from being called twice (once via |
5466 | * spin_unlock(), once by hand). | |
5467 | */ | |
95cdf3b7 | 5468 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5469 | { |
95c354fe | 5470 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5471 | int ret = 0; |
5472 | ||
95c354fe | 5473 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5474 | spin_unlock(lock); |
95c354fe NP |
5475 | if (resched && need_resched()) |
5476 | __cond_resched(); | |
5477 | else | |
5478 | cpu_relax(); | |
6df3cecb | 5479 | ret = 1; |
1da177e4 | 5480 | spin_lock(lock); |
1da177e4 | 5481 | } |
6df3cecb | 5482 | return ret; |
1da177e4 | 5483 | } |
1da177e4 LT |
5484 | EXPORT_SYMBOL(cond_resched_lock); |
5485 | ||
5486 | int __sched cond_resched_softirq(void) | |
5487 | { | |
5488 | BUG_ON(!in_softirq()); | |
5489 | ||
9414232f | 5490 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5491 | local_bh_enable(); |
1da177e4 LT |
5492 | __cond_resched(); |
5493 | local_bh_disable(); | |
5494 | return 1; | |
5495 | } | |
5496 | return 0; | |
5497 | } | |
1da177e4 LT |
5498 | EXPORT_SYMBOL(cond_resched_softirq); |
5499 | ||
1da177e4 LT |
5500 | /** |
5501 | * yield - yield the current processor to other threads. | |
5502 | * | |
72fd4a35 | 5503 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5504 | * thread runnable and calls sys_sched_yield(). |
5505 | */ | |
5506 | void __sched yield(void) | |
5507 | { | |
5508 | set_current_state(TASK_RUNNING); | |
5509 | sys_sched_yield(); | |
5510 | } | |
1da177e4 LT |
5511 | EXPORT_SYMBOL(yield); |
5512 | ||
5513 | /* | |
41a2d6cf | 5514 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5515 | * that process accounting knows that this is a task in IO wait state. |
5516 | * | |
5517 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5518 | * has set its backing_dev_info: the queue against which it should throttle) | |
5519 | */ | |
5520 | void __sched io_schedule(void) | |
5521 | { | |
70b97a7f | 5522 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5523 | |
0ff92245 | 5524 | delayacct_blkio_start(); |
1da177e4 LT |
5525 | atomic_inc(&rq->nr_iowait); |
5526 | schedule(); | |
5527 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5528 | delayacct_blkio_end(); |
1da177e4 | 5529 | } |
1da177e4 LT |
5530 | EXPORT_SYMBOL(io_schedule); |
5531 | ||
5532 | long __sched io_schedule_timeout(long timeout) | |
5533 | { | |
70b97a7f | 5534 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5535 | long ret; |
5536 | ||
0ff92245 | 5537 | delayacct_blkio_start(); |
1da177e4 LT |
5538 | atomic_inc(&rq->nr_iowait); |
5539 | ret = schedule_timeout(timeout); | |
5540 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5541 | delayacct_blkio_end(); |
1da177e4 LT |
5542 | return ret; |
5543 | } | |
5544 | ||
5545 | /** | |
5546 | * sys_sched_get_priority_max - return maximum RT priority. | |
5547 | * @policy: scheduling class. | |
5548 | * | |
5549 | * this syscall returns the maximum rt_priority that can be used | |
5550 | * by a given scheduling class. | |
5551 | */ | |
5552 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5553 | { | |
5554 | int ret = -EINVAL; | |
5555 | ||
5556 | switch (policy) { | |
5557 | case SCHED_FIFO: | |
5558 | case SCHED_RR: | |
5559 | ret = MAX_USER_RT_PRIO-1; | |
5560 | break; | |
5561 | case SCHED_NORMAL: | |
b0a9499c | 5562 | case SCHED_BATCH: |
dd41f596 | 5563 | case SCHED_IDLE: |
1da177e4 LT |
5564 | ret = 0; |
5565 | break; | |
5566 | } | |
5567 | return ret; | |
5568 | } | |
5569 | ||
5570 | /** | |
5571 | * sys_sched_get_priority_min - return minimum RT priority. | |
5572 | * @policy: scheduling class. | |
5573 | * | |
5574 | * this syscall returns the minimum rt_priority that can be used | |
5575 | * by a given scheduling class. | |
5576 | */ | |
5577 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5578 | { | |
5579 | int ret = -EINVAL; | |
5580 | ||
5581 | switch (policy) { | |
5582 | case SCHED_FIFO: | |
5583 | case SCHED_RR: | |
5584 | ret = 1; | |
5585 | break; | |
5586 | case SCHED_NORMAL: | |
b0a9499c | 5587 | case SCHED_BATCH: |
dd41f596 | 5588 | case SCHED_IDLE: |
1da177e4 LT |
5589 | ret = 0; |
5590 | } | |
5591 | return ret; | |
5592 | } | |
5593 | ||
5594 | /** | |
5595 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5596 | * @pid: pid of the process. | |
5597 | * @interval: userspace pointer to the timeslice value. | |
5598 | * | |
5599 | * this syscall writes the default timeslice value of a given process | |
5600 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5601 | */ | |
5602 | asmlinkage | |
5603 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5604 | { | |
36c8b586 | 5605 | struct task_struct *p; |
a4ec24b4 | 5606 | unsigned int time_slice; |
3a5c359a | 5607 | int retval; |
1da177e4 | 5608 | struct timespec t; |
1da177e4 LT |
5609 | |
5610 | if (pid < 0) | |
3a5c359a | 5611 | return -EINVAL; |
1da177e4 LT |
5612 | |
5613 | retval = -ESRCH; | |
5614 | read_lock(&tasklist_lock); | |
5615 | p = find_process_by_pid(pid); | |
5616 | if (!p) | |
5617 | goto out_unlock; | |
5618 | ||
5619 | retval = security_task_getscheduler(p); | |
5620 | if (retval) | |
5621 | goto out_unlock; | |
5622 | ||
77034937 IM |
5623 | /* |
5624 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5625 | * tasks that are on an otherwise idle runqueue: | |
5626 | */ | |
5627 | time_slice = 0; | |
5628 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5629 | time_slice = DEF_TIMESLICE; |
1868f958 | 5630 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5631 | struct sched_entity *se = &p->se; |
5632 | unsigned long flags; | |
5633 | struct rq *rq; | |
5634 | ||
5635 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5636 | if (rq->cfs.load.weight) |
5637 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5638 | task_rq_unlock(rq, &flags); |
5639 | } | |
1da177e4 | 5640 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5641 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5642 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5643 | return retval; |
3a5c359a | 5644 | |
1da177e4 LT |
5645 | out_unlock: |
5646 | read_unlock(&tasklist_lock); | |
5647 | return retval; | |
5648 | } | |
5649 | ||
7c731e0a | 5650 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5651 | |
82a1fcb9 | 5652 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5653 | { |
1da177e4 | 5654 | unsigned long free = 0; |
36c8b586 | 5655 | unsigned state; |
1da177e4 | 5656 | |
1da177e4 | 5657 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5658 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5659 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5660 | #if BITS_PER_LONG == 32 |
1da177e4 | 5661 | if (state == TASK_RUNNING) |
cc4ea795 | 5662 | printk(KERN_CONT " running "); |
1da177e4 | 5663 | else |
cc4ea795 | 5664 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5665 | #else |
5666 | if (state == TASK_RUNNING) | |
cc4ea795 | 5667 | printk(KERN_CONT " running task "); |
1da177e4 | 5668 | else |
cc4ea795 | 5669 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5670 | #endif |
5671 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5672 | { | |
10ebffde | 5673 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5674 | while (!*n) |
5675 | n++; | |
10ebffde | 5676 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5677 | } |
5678 | #endif | |
ba25f9dc | 5679 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5680 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5681 | |
5fb5e6de | 5682 | show_stack(p, NULL); |
1da177e4 LT |
5683 | } |
5684 | ||
e59e2ae2 | 5685 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5686 | { |
36c8b586 | 5687 | struct task_struct *g, *p; |
1da177e4 | 5688 | |
4bd77321 IM |
5689 | #if BITS_PER_LONG == 32 |
5690 | printk(KERN_INFO | |
5691 | " task PC stack pid father\n"); | |
1da177e4 | 5692 | #else |
4bd77321 IM |
5693 | printk(KERN_INFO |
5694 | " task PC stack pid father\n"); | |
1da177e4 LT |
5695 | #endif |
5696 | read_lock(&tasklist_lock); | |
5697 | do_each_thread(g, p) { | |
5698 | /* | |
5699 | * reset the NMI-timeout, listing all files on a slow | |
5700 | * console might take alot of time: | |
5701 | */ | |
5702 | touch_nmi_watchdog(); | |
39bc89fd | 5703 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5704 | sched_show_task(p); |
1da177e4 LT |
5705 | } while_each_thread(g, p); |
5706 | ||
04c9167f JF |
5707 | touch_all_softlockup_watchdogs(); |
5708 | ||
dd41f596 IM |
5709 | #ifdef CONFIG_SCHED_DEBUG |
5710 | sysrq_sched_debug_show(); | |
5711 | #endif | |
1da177e4 | 5712 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5713 | /* |
5714 | * Only show locks if all tasks are dumped: | |
5715 | */ | |
5716 | if (state_filter == -1) | |
5717 | debug_show_all_locks(); | |
1da177e4 LT |
5718 | } |
5719 | ||
1df21055 IM |
5720 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5721 | { | |
dd41f596 | 5722 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5723 | } |
5724 | ||
f340c0d1 IM |
5725 | /** |
5726 | * init_idle - set up an idle thread for a given CPU | |
5727 | * @idle: task in question | |
5728 | * @cpu: cpu the idle task belongs to | |
5729 | * | |
5730 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5731 | * flag, to make booting more robust. | |
5732 | */ | |
5c1e1767 | 5733 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5734 | { |
70b97a7f | 5735 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5736 | unsigned long flags; |
5737 | ||
dd41f596 IM |
5738 | __sched_fork(idle); |
5739 | idle->se.exec_start = sched_clock(); | |
5740 | ||
b29739f9 | 5741 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5742 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5743 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5744 | |
5745 | spin_lock_irqsave(&rq->lock, flags); | |
5746 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5747 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5748 | idle->oncpu = 1; | |
5749 | #endif | |
1da177e4 LT |
5750 | spin_unlock_irqrestore(&rq->lock, flags); |
5751 | ||
5752 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5753 | #if defined(CONFIG_PREEMPT) |
5754 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5755 | #else | |
a1261f54 | 5756 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5757 | #endif |
dd41f596 IM |
5758 | /* |
5759 | * The idle tasks have their own, simple scheduling class: | |
5760 | */ | |
5761 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5762 | } |
5763 | ||
5764 | /* | |
5765 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5766 | * indicates which cpus entered this state. This is used | |
5767 | * in the rcu update to wait only for active cpus. For system | |
5768 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5769 | * always be CPU_MASK_NONE. | |
5770 | */ | |
5771 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5772 | ||
19978ca6 IM |
5773 | /* |
5774 | * Increase the granularity value when there are more CPUs, | |
5775 | * because with more CPUs the 'effective latency' as visible | |
5776 | * to users decreases. But the relationship is not linear, | |
5777 | * so pick a second-best guess by going with the log2 of the | |
5778 | * number of CPUs. | |
5779 | * | |
5780 | * This idea comes from the SD scheduler of Con Kolivas: | |
5781 | */ | |
5782 | static inline void sched_init_granularity(void) | |
5783 | { | |
5784 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5785 | const unsigned long limit = 200000000; | |
5786 | ||
5787 | sysctl_sched_min_granularity *= factor; | |
5788 | if (sysctl_sched_min_granularity > limit) | |
5789 | sysctl_sched_min_granularity = limit; | |
5790 | ||
5791 | sysctl_sched_latency *= factor; | |
5792 | if (sysctl_sched_latency > limit) | |
5793 | sysctl_sched_latency = limit; | |
5794 | ||
5795 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
5796 | |
5797 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
5798 | } |
5799 | ||
1da177e4 LT |
5800 | #ifdef CONFIG_SMP |
5801 | /* | |
5802 | * This is how migration works: | |
5803 | * | |
70b97a7f | 5804 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5805 | * runqueue and wake up that CPU's migration thread. |
5806 | * 2) we down() the locked semaphore => thread blocks. | |
5807 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5808 | * thread off the CPU) | |
5809 | * 4) it gets the migration request and checks whether the migrated | |
5810 | * task is still in the wrong runqueue. | |
5811 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5812 | * it and puts it into the right queue. | |
5813 | * 6) migration thread up()s the semaphore. | |
5814 | * 7) we wake up and the migration is done. | |
5815 | */ | |
5816 | ||
5817 | /* | |
5818 | * Change a given task's CPU affinity. Migrate the thread to a | |
5819 | * proper CPU and schedule it away if the CPU it's executing on | |
5820 | * is removed from the allowed bitmask. | |
5821 | * | |
5822 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5823 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5824 | * call is not atomic; no spinlocks may be held. |
5825 | */ | |
cd8ba7cd | 5826 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5827 | { |
70b97a7f | 5828 | struct migration_req req; |
1da177e4 | 5829 | unsigned long flags; |
70b97a7f | 5830 | struct rq *rq; |
48f24c4d | 5831 | int ret = 0; |
1da177e4 LT |
5832 | |
5833 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5834 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5835 | ret = -EINVAL; |
5836 | goto out; | |
5837 | } | |
5838 | ||
9985b0ba DR |
5839 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5840 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5841 | ret = -EINVAL; | |
5842 | goto out; | |
5843 | } | |
5844 | ||
73fe6aae | 5845 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5846 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5847 | else { |
cd8ba7cd MT |
5848 | p->cpus_allowed = *new_mask; |
5849 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5850 | } |
5851 | ||
1da177e4 | 5852 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5853 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5854 | goto out; |
5855 | ||
cd8ba7cd | 5856 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5857 | /* Need help from migration thread: drop lock and wait. */ |
5858 | task_rq_unlock(rq, &flags); | |
5859 | wake_up_process(rq->migration_thread); | |
5860 | wait_for_completion(&req.done); | |
5861 | tlb_migrate_finish(p->mm); | |
5862 | return 0; | |
5863 | } | |
5864 | out: | |
5865 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5866 | |
1da177e4 LT |
5867 | return ret; |
5868 | } | |
cd8ba7cd | 5869 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5870 | |
5871 | /* | |
41a2d6cf | 5872 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5873 | * this because either it can't run here any more (set_cpus_allowed() |
5874 | * away from this CPU, or CPU going down), or because we're | |
5875 | * attempting to rebalance this task on exec (sched_exec). | |
5876 | * | |
5877 | * So we race with normal scheduler movements, but that's OK, as long | |
5878 | * as the task is no longer on this CPU. | |
efc30814 KK |
5879 | * |
5880 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5881 | */ |
efc30814 | 5882 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5883 | { |
70b97a7f | 5884 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5885 | int ret = 0, on_rq; |
1da177e4 | 5886 | |
e761b772 | 5887 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5888 | return ret; |
1da177e4 LT |
5889 | |
5890 | rq_src = cpu_rq(src_cpu); | |
5891 | rq_dest = cpu_rq(dest_cpu); | |
5892 | ||
5893 | double_rq_lock(rq_src, rq_dest); | |
5894 | /* Already moved. */ | |
5895 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5896 | goto done; |
1da177e4 LT |
5897 | /* Affinity changed (again). */ |
5898 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
b1e38734 | 5899 | goto fail; |
1da177e4 | 5900 | |
dd41f596 | 5901 | on_rq = p->se.on_rq; |
6e82a3be | 5902 | if (on_rq) |
2e1cb74a | 5903 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5904 | |
1da177e4 | 5905 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5906 | if (on_rq) { |
5907 | activate_task(rq_dest, p, 0); | |
5908 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5909 | } |
b1e38734 | 5910 | done: |
efc30814 | 5911 | ret = 1; |
b1e38734 | 5912 | fail: |
1da177e4 | 5913 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5914 | return ret; |
1da177e4 LT |
5915 | } |
5916 | ||
5917 | /* | |
5918 | * migration_thread - this is a highprio system thread that performs | |
5919 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5920 | * another runqueue. | |
5921 | */ | |
95cdf3b7 | 5922 | static int migration_thread(void *data) |
1da177e4 | 5923 | { |
1da177e4 | 5924 | int cpu = (long)data; |
70b97a7f | 5925 | struct rq *rq; |
1da177e4 LT |
5926 | |
5927 | rq = cpu_rq(cpu); | |
5928 | BUG_ON(rq->migration_thread != current); | |
5929 | ||
5930 | set_current_state(TASK_INTERRUPTIBLE); | |
5931 | while (!kthread_should_stop()) { | |
70b97a7f | 5932 | struct migration_req *req; |
1da177e4 | 5933 | struct list_head *head; |
1da177e4 | 5934 | |
1da177e4 LT |
5935 | spin_lock_irq(&rq->lock); |
5936 | ||
5937 | if (cpu_is_offline(cpu)) { | |
5938 | spin_unlock_irq(&rq->lock); | |
5939 | goto wait_to_die; | |
5940 | } | |
5941 | ||
5942 | if (rq->active_balance) { | |
5943 | active_load_balance(rq, cpu); | |
5944 | rq->active_balance = 0; | |
5945 | } | |
5946 | ||
5947 | head = &rq->migration_queue; | |
5948 | ||
5949 | if (list_empty(head)) { | |
5950 | spin_unlock_irq(&rq->lock); | |
5951 | schedule(); | |
5952 | set_current_state(TASK_INTERRUPTIBLE); | |
5953 | continue; | |
5954 | } | |
70b97a7f | 5955 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5956 | list_del_init(head->next); |
5957 | ||
674311d5 NP |
5958 | spin_unlock(&rq->lock); |
5959 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5960 | local_irq_enable(); | |
1da177e4 LT |
5961 | |
5962 | complete(&req->done); | |
5963 | } | |
5964 | __set_current_state(TASK_RUNNING); | |
5965 | return 0; | |
5966 | ||
5967 | wait_to_die: | |
5968 | /* Wait for kthread_stop */ | |
5969 | set_current_state(TASK_INTERRUPTIBLE); | |
5970 | while (!kthread_should_stop()) { | |
5971 | schedule(); | |
5972 | set_current_state(TASK_INTERRUPTIBLE); | |
5973 | } | |
5974 | __set_current_state(TASK_RUNNING); | |
5975 | return 0; | |
5976 | } | |
5977 | ||
5978 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5979 | |
5980 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5981 | { | |
5982 | int ret; | |
5983 | ||
5984 | local_irq_disable(); | |
5985 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5986 | local_irq_enable(); | |
5987 | return ret; | |
5988 | } | |
5989 | ||
054b9108 | 5990 | /* |
3a4fa0a2 | 5991 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5992 | * NOTE: interrupts should be disabled by the caller |
5993 | */ | |
48f24c4d | 5994 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5995 | { |
efc30814 | 5996 | unsigned long flags; |
1da177e4 | 5997 | cpumask_t mask; |
70b97a7f IM |
5998 | struct rq *rq; |
5999 | int dest_cpu; | |
1da177e4 | 6000 | |
3a5c359a AK |
6001 | do { |
6002 | /* On same node? */ | |
6003 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6004 | cpus_and(mask, mask, p->cpus_allowed); | |
6005 | dest_cpu = any_online_cpu(mask); | |
6006 | ||
6007 | /* On any allowed CPU? */ | |
434d53b0 | 6008 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
6009 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6010 | ||
6011 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 6012 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
6013 | cpumask_t cpus_allowed; |
6014 | ||
6015 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
6016 | /* |
6017 | * Try to stay on the same cpuset, where the | |
6018 | * current cpuset may be a subset of all cpus. | |
6019 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 6020 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
6021 | * called within calls to cpuset_lock/cpuset_unlock. |
6022 | */ | |
3a5c359a | 6023 | rq = task_rq_lock(p, &flags); |
470fd646 | 6024 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
6025 | dest_cpu = any_online_cpu(p->cpus_allowed); |
6026 | task_rq_unlock(rq, &flags); | |
1da177e4 | 6027 | |
3a5c359a AK |
6028 | /* |
6029 | * Don't tell them about moving exiting tasks or | |
6030 | * kernel threads (both mm NULL), since they never | |
6031 | * leave kernel. | |
6032 | */ | |
41a2d6cf | 6033 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
6034 | printk(KERN_INFO "process %d (%s) no " |
6035 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
6036 | task_pid_nr(p), p->comm, dead_cpu); |
6037 | } | |
3a5c359a | 6038 | } |
f7b4cddc | 6039 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
6040 | } |
6041 | ||
6042 | /* | |
6043 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6044 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6045 | * for performance reasons the counter is not stricly tracking tasks to | |
6046 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6047 | * to keep the global sum constant after CPU-down: | |
6048 | */ | |
70b97a7f | 6049 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6050 | { |
7c16ec58 | 6051 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
6052 | unsigned long flags; |
6053 | ||
6054 | local_irq_save(flags); | |
6055 | double_rq_lock(rq_src, rq_dest); | |
6056 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6057 | rq_src->nr_uninterruptible = 0; | |
6058 | double_rq_unlock(rq_src, rq_dest); | |
6059 | local_irq_restore(flags); | |
6060 | } | |
6061 | ||
6062 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6063 | static void migrate_live_tasks(int src_cpu) | |
6064 | { | |
48f24c4d | 6065 | struct task_struct *p, *t; |
1da177e4 | 6066 | |
f7b4cddc | 6067 | read_lock(&tasklist_lock); |
1da177e4 | 6068 | |
48f24c4d IM |
6069 | do_each_thread(t, p) { |
6070 | if (p == current) | |
1da177e4 LT |
6071 | continue; |
6072 | ||
48f24c4d IM |
6073 | if (task_cpu(p) == src_cpu) |
6074 | move_task_off_dead_cpu(src_cpu, p); | |
6075 | } while_each_thread(t, p); | |
1da177e4 | 6076 | |
f7b4cddc | 6077 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6078 | } |
6079 | ||
dd41f596 IM |
6080 | /* |
6081 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6082 | * It does so by boosting its priority to highest possible. |
6083 | * Used by CPU offline code. | |
1da177e4 LT |
6084 | */ |
6085 | void sched_idle_next(void) | |
6086 | { | |
48f24c4d | 6087 | int this_cpu = smp_processor_id(); |
70b97a7f | 6088 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6089 | struct task_struct *p = rq->idle; |
6090 | unsigned long flags; | |
6091 | ||
6092 | /* cpu has to be offline */ | |
48f24c4d | 6093 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6094 | |
48f24c4d IM |
6095 | /* |
6096 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6097 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6098 | */ |
6099 | spin_lock_irqsave(&rq->lock, flags); | |
6100 | ||
dd41f596 | 6101 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6102 | |
94bc9a7b DA |
6103 | update_rq_clock(rq); |
6104 | activate_task(rq, p, 0); | |
1da177e4 LT |
6105 | |
6106 | spin_unlock_irqrestore(&rq->lock, flags); | |
6107 | } | |
6108 | ||
48f24c4d IM |
6109 | /* |
6110 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6111 | * offline. |
6112 | */ | |
6113 | void idle_task_exit(void) | |
6114 | { | |
6115 | struct mm_struct *mm = current->active_mm; | |
6116 | ||
6117 | BUG_ON(cpu_online(smp_processor_id())); | |
6118 | ||
6119 | if (mm != &init_mm) | |
6120 | switch_mm(mm, &init_mm, current); | |
6121 | mmdrop(mm); | |
6122 | } | |
6123 | ||
054b9108 | 6124 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6125 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6126 | { |
70b97a7f | 6127 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6128 | |
6129 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6130 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6131 | |
6132 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6133 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6134 | |
48f24c4d | 6135 | get_task_struct(p); |
1da177e4 LT |
6136 | |
6137 | /* | |
6138 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6139 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6140 | * fine. |
6141 | */ | |
f7b4cddc | 6142 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6143 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6144 | spin_lock_irq(&rq->lock); |
1da177e4 | 6145 | |
48f24c4d | 6146 | put_task_struct(p); |
1da177e4 LT |
6147 | } |
6148 | ||
6149 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6150 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6151 | { | |
70b97a7f | 6152 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6153 | struct task_struct *next; |
48f24c4d | 6154 | |
dd41f596 IM |
6155 | for ( ; ; ) { |
6156 | if (!rq->nr_running) | |
6157 | break; | |
a8e504d2 | 6158 | update_rq_clock(rq); |
ff95f3df | 6159 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6160 | if (!next) |
6161 | break; | |
79c53799 | 6162 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6163 | migrate_dead(dead_cpu, next); |
e692ab53 | 6164 | |
1da177e4 LT |
6165 | } |
6166 | } | |
6167 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6168 | ||
e692ab53 NP |
6169 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6170 | ||
6171 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6172 | { |
6173 | .procname = "sched_domain", | |
c57baf1e | 6174 | .mode = 0555, |
e0361851 | 6175 | }, |
38605cae | 6176 | {0, }, |
e692ab53 NP |
6177 | }; |
6178 | ||
6179 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6180 | { |
c57baf1e | 6181 | .ctl_name = CTL_KERN, |
e0361851 | 6182 | .procname = "kernel", |
c57baf1e | 6183 | .mode = 0555, |
e0361851 AD |
6184 | .child = sd_ctl_dir, |
6185 | }, | |
38605cae | 6186 | {0, }, |
e692ab53 NP |
6187 | }; |
6188 | ||
6189 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6190 | { | |
6191 | struct ctl_table *entry = | |
5cf9f062 | 6192 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6193 | |
e692ab53 NP |
6194 | return entry; |
6195 | } | |
6196 | ||
6382bc90 MM |
6197 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6198 | { | |
cd790076 | 6199 | struct ctl_table *entry; |
6382bc90 | 6200 | |
cd790076 MM |
6201 | /* |
6202 | * In the intermediate directories, both the child directory and | |
6203 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6204 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6205 | * static strings and all have proc handlers. |
6206 | */ | |
6207 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6208 | if (entry->child) |
6209 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6210 | if (entry->proc_handler == NULL) |
6211 | kfree(entry->procname); | |
6212 | } | |
6382bc90 MM |
6213 | |
6214 | kfree(*tablep); | |
6215 | *tablep = NULL; | |
6216 | } | |
6217 | ||
e692ab53 | 6218 | static void |
e0361851 | 6219 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6220 | const char *procname, void *data, int maxlen, |
6221 | mode_t mode, proc_handler *proc_handler) | |
6222 | { | |
e692ab53 NP |
6223 | entry->procname = procname; |
6224 | entry->data = data; | |
6225 | entry->maxlen = maxlen; | |
6226 | entry->mode = mode; | |
6227 | entry->proc_handler = proc_handler; | |
6228 | } | |
6229 | ||
6230 | static struct ctl_table * | |
6231 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6232 | { | |
ace8b3d6 | 6233 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6234 | |
ad1cdc1d MM |
6235 | if (table == NULL) |
6236 | return NULL; | |
6237 | ||
e0361851 | 6238 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6239 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6240 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6241 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6242 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6243 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6244 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6245 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6246 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6247 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6248 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6249 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6250 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6251 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6252 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6253 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6254 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6255 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6256 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6257 | &sd->cache_nice_tries, |
6258 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6259 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6260 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6261 | /* &table[11] is terminator */ |
e692ab53 NP |
6262 | |
6263 | return table; | |
6264 | } | |
6265 | ||
9a4e7159 | 6266 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6267 | { |
6268 | struct ctl_table *entry, *table; | |
6269 | struct sched_domain *sd; | |
6270 | int domain_num = 0, i; | |
6271 | char buf[32]; | |
6272 | ||
6273 | for_each_domain(cpu, sd) | |
6274 | domain_num++; | |
6275 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6276 | if (table == NULL) |
6277 | return NULL; | |
e692ab53 NP |
6278 | |
6279 | i = 0; | |
6280 | for_each_domain(cpu, sd) { | |
6281 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6282 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6283 | entry->mode = 0555; |
e692ab53 NP |
6284 | entry->child = sd_alloc_ctl_domain_table(sd); |
6285 | entry++; | |
6286 | i++; | |
6287 | } | |
6288 | return table; | |
6289 | } | |
6290 | ||
6291 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6292 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6293 | { |
6294 | int i, cpu_num = num_online_cpus(); | |
6295 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6296 | char buf[32]; | |
6297 | ||
7378547f MM |
6298 | WARN_ON(sd_ctl_dir[0].child); |
6299 | sd_ctl_dir[0].child = entry; | |
6300 | ||
ad1cdc1d MM |
6301 | if (entry == NULL) |
6302 | return; | |
6303 | ||
97b6ea7b | 6304 | for_each_online_cpu(i) { |
e692ab53 | 6305 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6306 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6307 | entry->mode = 0555; |
e692ab53 | 6308 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6309 | entry++; |
e692ab53 | 6310 | } |
7378547f MM |
6311 | |
6312 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6313 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6314 | } | |
6382bc90 | 6315 | |
7378547f | 6316 | /* may be called multiple times per register */ |
6382bc90 MM |
6317 | static void unregister_sched_domain_sysctl(void) |
6318 | { | |
7378547f MM |
6319 | if (sd_sysctl_header) |
6320 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6321 | sd_sysctl_header = NULL; |
7378547f MM |
6322 | if (sd_ctl_dir[0].child) |
6323 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6324 | } |
e692ab53 | 6325 | #else |
6382bc90 MM |
6326 | static void register_sched_domain_sysctl(void) |
6327 | { | |
6328 | } | |
6329 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6330 | { |
6331 | } | |
6332 | #endif | |
6333 | ||
1f11eb6a GH |
6334 | static void set_rq_online(struct rq *rq) |
6335 | { | |
6336 | if (!rq->online) { | |
6337 | const struct sched_class *class; | |
6338 | ||
6339 | cpu_set(rq->cpu, rq->rd->online); | |
6340 | rq->online = 1; | |
6341 | ||
6342 | for_each_class(class) { | |
6343 | if (class->rq_online) | |
6344 | class->rq_online(rq); | |
6345 | } | |
6346 | } | |
6347 | } | |
6348 | ||
6349 | static void set_rq_offline(struct rq *rq) | |
6350 | { | |
6351 | if (rq->online) { | |
6352 | const struct sched_class *class; | |
6353 | ||
6354 | for_each_class(class) { | |
6355 | if (class->rq_offline) | |
6356 | class->rq_offline(rq); | |
6357 | } | |
6358 | ||
6359 | cpu_clear(rq->cpu, rq->rd->online); | |
6360 | rq->online = 0; | |
6361 | } | |
6362 | } | |
6363 | ||
1da177e4 LT |
6364 | /* |
6365 | * migration_call - callback that gets triggered when a CPU is added. | |
6366 | * Here we can start up the necessary migration thread for the new CPU. | |
6367 | */ | |
48f24c4d IM |
6368 | static int __cpuinit |
6369 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6370 | { |
1da177e4 | 6371 | struct task_struct *p; |
48f24c4d | 6372 | int cpu = (long)hcpu; |
1da177e4 | 6373 | unsigned long flags; |
70b97a7f | 6374 | struct rq *rq; |
1da177e4 LT |
6375 | |
6376 | switch (action) { | |
5be9361c | 6377 | |
1da177e4 | 6378 | case CPU_UP_PREPARE: |
8bb78442 | 6379 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6380 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6381 | if (IS_ERR(p)) |
6382 | return NOTIFY_BAD; | |
1da177e4 LT |
6383 | kthread_bind(p, cpu); |
6384 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6385 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6386 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6387 | task_rq_unlock(rq, &flags); |
6388 | cpu_rq(cpu)->migration_thread = p; | |
6389 | break; | |
48f24c4d | 6390 | |
1da177e4 | 6391 | case CPU_ONLINE: |
8bb78442 | 6392 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6393 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6394 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6395 | |
6396 | /* Update our root-domain */ | |
6397 | rq = cpu_rq(cpu); | |
6398 | spin_lock_irqsave(&rq->lock, flags); | |
6399 | if (rq->rd) { | |
6400 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6401 | |
6402 | set_rq_online(rq); | |
1f94ef59 GH |
6403 | } |
6404 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6405 | break; |
48f24c4d | 6406 | |
1da177e4 LT |
6407 | #ifdef CONFIG_HOTPLUG_CPU |
6408 | case CPU_UP_CANCELED: | |
8bb78442 | 6409 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6410 | if (!cpu_rq(cpu)->migration_thread) |
6411 | break; | |
41a2d6cf | 6412 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6413 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6414 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6415 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6416 | cpu_rq(cpu)->migration_thread = NULL; | |
6417 | break; | |
48f24c4d | 6418 | |
1da177e4 | 6419 | case CPU_DEAD: |
8bb78442 | 6420 | case CPU_DEAD_FROZEN: |
470fd646 | 6421 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6422 | migrate_live_tasks(cpu); |
6423 | rq = cpu_rq(cpu); | |
6424 | kthread_stop(rq->migration_thread); | |
6425 | rq->migration_thread = NULL; | |
6426 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6427 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6428 | update_rq_clock(rq); |
2e1cb74a | 6429 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6430 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6431 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6432 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6433 | migrate_dead_tasks(cpu); |
d2da272a | 6434 | spin_unlock_irq(&rq->lock); |
470fd646 | 6435 | cpuset_unlock(); |
1da177e4 LT |
6436 | migrate_nr_uninterruptible(rq); |
6437 | BUG_ON(rq->nr_running != 0); | |
6438 | ||
41a2d6cf IM |
6439 | /* |
6440 | * No need to migrate the tasks: it was best-effort if | |
6441 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6442 | * the requestors. | |
6443 | */ | |
1da177e4 LT |
6444 | spin_lock_irq(&rq->lock); |
6445 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6446 | struct migration_req *req; |
6447 | ||
1da177e4 | 6448 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6449 | struct migration_req, list); |
1da177e4 LT |
6450 | list_del_init(&req->list); |
6451 | complete(&req->done); | |
6452 | } | |
6453 | spin_unlock_irq(&rq->lock); | |
6454 | break; | |
57d885fe | 6455 | |
08f503b0 GH |
6456 | case CPU_DYING: |
6457 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6458 | /* Update our root-domain */ |
6459 | rq = cpu_rq(cpu); | |
6460 | spin_lock_irqsave(&rq->lock, flags); | |
6461 | if (rq->rd) { | |
6462 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6463 | set_rq_offline(rq); |
57d885fe GH |
6464 | } |
6465 | spin_unlock_irqrestore(&rq->lock, flags); | |
6466 | break; | |
1da177e4 LT |
6467 | #endif |
6468 | } | |
6469 | return NOTIFY_OK; | |
6470 | } | |
6471 | ||
6472 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6473 | * happens before everything else. | |
6474 | */ | |
26c2143b | 6475 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6476 | .notifier_call = migration_call, |
6477 | .priority = 10 | |
6478 | }; | |
6479 | ||
7babe8db | 6480 | static int __init migration_init(void) |
1da177e4 LT |
6481 | { |
6482 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6483 | int err; |
48f24c4d IM |
6484 | |
6485 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6486 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6487 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6488 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6489 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6490 | |
6491 | return err; | |
1da177e4 | 6492 | } |
7babe8db | 6493 | early_initcall(migration_init); |
1da177e4 LT |
6494 | #endif |
6495 | ||
6496 | #ifdef CONFIG_SMP | |
476f3534 | 6497 | |
3e9830dc | 6498 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6499 | |
099f98c8 GS |
6500 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6501 | { | |
6502 | switch (lvl) { | |
6503 | case SD_LV_NONE: | |
6504 | return "NONE"; | |
6505 | case SD_LV_SIBLING: | |
6506 | return "SIBLING"; | |
6507 | case SD_LV_MC: | |
6508 | return "MC"; | |
6509 | case SD_LV_CPU: | |
6510 | return "CPU"; | |
6511 | case SD_LV_NODE: | |
6512 | return "NODE"; | |
6513 | case SD_LV_ALLNODES: | |
6514 | return "ALLNODES"; | |
6515 | case SD_LV_MAX: | |
6516 | return "MAX"; | |
6517 | ||
6518 | } | |
6519 | return "MAX"; | |
6520 | } | |
6521 | ||
7c16ec58 MT |
6522 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6523 | cpumask_t *groupmask) | |
1da177e4 | 6524 | { |
4dcf6aff | 6525 | struct sched_group *group = sd->groups; |
434d53b0 | 6526 | char str[256]; |
1da177e4 | 6527 | |
434d53b0 | 6528 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6529 | cpus_clear(*groupmask); |
4dcf6aff IM |
6530 | |
6531 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6532 | ||
6533 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6534 | printk("does not load-balance\n"); | |
6535 | if (sd->parent) | |
6536 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6537 | " has parent"); | |
6538 | return -1; | |
41c7ce9a NP |
6539 | } |
6540 | ||
099f98c8 GS |
6541 | printk(KERN_CONT "span %s level %s\n", |
6542 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6543 | |
6544 | if (!cpu_isset(cpu, sd->span)) { | |
6545 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6546 | "CPU%d\n", cpu); | |
6547 | } | |
6548 | if (!cpu_isset(cpu, group->cpumask)) { | |
6549 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6550 | " CPU%d\n", cpu); | |
6551 | } | |
1da177e4 | 6552 | |
4dcf6aff | 6553 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6554 | do { |
4dcf6aff IM |
6555 | if (!group) { |
6556 | printk("\n"); | |
6557 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6558 | break; |
6559 | } | |
6560 | ||
4dcf6aff IM |
6561 | if (!group->__cpu_power) { |
6562 | printk(KERN_CONT "\n"); | |
6563 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6564 | "set\n"); | |
6565 | break; | |
6566 | } | |
1da177e4 | 6567 | |
4dcf6aff IM |
6568 | if (!cpus_weight(group->cpumask)) { |
6569 | printk(KERN_CONT "\n"); | |
6570 | printk(KERN_ERR "ERROR: empty group\n"); | |
6571 | break; | |
6572 | } | |
1da177e4 | 6573 | |
7c16ec58 | 6574 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6575 | printk(KERN_CONT "\n"); |
6576 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6577 | break; | |
6578 | } | |
1da177e4 | 6579 | |
7c16ec58 | 6580 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6581 | |
434d53b0 | 6582 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6583 | printk(KERN_CONT " %s", str); |
1da177e4 | 6584 | |
4dcf6aff IM |
6585 | group = group->next; |
6586 | } while (group != sd->groups); | |
6587 | printk(KERN_CONT "\n"); | |
1da177e4 | 6588 | |
7c16ec58 | 6589 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6590 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6591 | |
7c16ec58 | 6592 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6593 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6594 | "of domain->span\n"); | |
6595 | return 0; | |
6596 | } | |
1da177e4 | 6597 | |
4dcf6aff IM |
6598 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6599 | { | |
7c16ec58 | 6600 | cpumask_t *groupmask; |
4dcf6aff | 6601 | int level = 0; |
1da177e4 | 6602 | |
4dcf6aff IM |
6603 | if (!sd) { |
6604 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6605 | return; | |
6606 | } | |
1da177e4 | 6607 | |
4dcf6aff IM |
6608 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6609 | ||
7c16ec58 MT |
6610 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6611 | if (!groupmask) { | |
6612 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6613 | return; | |
6614 | } | |
6615 | ||
4dcf6aff | 6616 | for (;;) { |
7c16ec58 | 6617 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6618 | break; |
1da177e4 LT |
6619 | level++; |
6620 | sd = sd->parent; | |
33859f7f | 6621 | if (!sd) |
4dcf6aff IM |
6622 | break; |
6623 | } | |
7c16ec58 | 6624 | kfree(groupmask); |
1da177e4 | 6625 | } |
6d6bc0ad | 6626 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6627 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6628 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6629 | |
1a20ff27 | 6630 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6631 | { |
6632 | if (cpus_weight(sd->span) == 1) | |
6633 | return 1; | |
6634 | ||
6635 | /* Following flags need at least 2 groups */ | |
6636 | if (sd->flags & (SD_LOAD_BALANCE | | |
6637 | SD_BALANCE_NEWIDLE | | |
6638 | SD_BALANCE_FORK | | |
89c4710e SS |
6639 | SD_BALANCE_EXEC | |
6640 | SD_SHARE_CPUPOWER | | |
6641 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6642 | if (sd->groups != sd->groups->next) |
6643 | return 0; | |
6644 | } | |
6645 | ||
6646 | /* Following flags don't use groups */ | |
6647 | if (sd->flags & (SD_WAKE_IDLE | | |
6648 | SD_WAKE_AFFINE | | |
6649 | SD_WAKE_BALANCE)) | |
6650 | return 0; | |
6651 | ||
6652 | return 1; | |
6653 | } | |
6654 | ||
48f24c4d IM |
6655 | static int |
6656 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6657 | { |
6658 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6659 | ||
6660 | if (sd_degenerate(parent)) | |
6661 | return 1; | |
6662 | ||
6663 | if (!cpus_equal(sd->span, parent->span)) | |
6664 | return 0; | |
6665 | ||
6666 | /* Does parent contain flags not in child? */ | |
6667 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6668 | if (cflags & SD_WAKE_AFFINE) | |
6669 | pflags &= ~SD_WAKE_BALANCE; | |
6670 | /* Flags needing groups don't count if only 1 group in parent */ | |
6671 | if (parent->groups == parent->groups->next) { | |
6672 | pflags &= ~(SD_LOAD_BALANCE | | |
6673 | SD_BALANCE_NEWIDLE | | |
6674 | SD_BALANCE_FORK | | |
89c4710e SS |
6675 | SD_BALANCE_EXEC | |
6676 | SD_SHARE_CPUPOWER | | |
6677 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6678 | } |
6679 | if (~cflags & pflags) | |
6680 | return 0; | |
6681 | ||
6682 | return 1; | |
6683 | } | |
6684 | ||
57d885fe GH |
6685 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6686 | { | |
6687 | unsigned long flags; | |
57d885fe GH |
6688 | |
6689 | spin_lock_irqsave(&rq->lock, flags); | |
6690 | ||
6691 | if (rq->rd) { | |
6692 | struct root_domain *old_rd = rq->rd; | |
6693 | ||
1f11eb6a GH |
6694 | if (cpu_isset(rq->cpu, old_rd->online)) |
6695 | set_rq_offline(rq); | |
57d885fe | 6696 | |
dc938520 | 6697 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6698 | |
57d885fe GH |
6699 | if (atomic_dec_and_test(&old_rd->refcount)) |
6700 | kfree(old_rd); | |
6701 | } | |
6702 | ||
6703 | atomic_inc(&rd->refcount); | |
6704 | rq->rd = rd; | |
6705 | ||
dc938520 | 6706 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6707 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6708 | set_rq_online(rq); |
57d885fe GH |
6709 | |
6710 | spin_unlock_irqrestore(&rq->lock, flags); | |
6711 | } | |
6712 | ||
dc938520 | 6713 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6714 | { |
6715 | memset(rd, 0, sizeof(*rd)); | |
6716 | ||
dc938520 GH |
6717 | cpus_clear(rd->span); |
6718 | cpus_clear(rd->online); | |
6e0534f2 GH |
6719 | |
6720 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6721 | } |
6722 | ||
6723 | static void init_defrootdomain(void) | |
6724 | { | |
dc938520 | 6725 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6726 | atomic_set(&def_root_domain.refcount, 1); |
6727 | } | |
6728 | ||
dc938520 | 6729 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6730 | { |
6731 | struct root_domain *rd; | |
6732 | ||
6733 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6734 | if (!rd) | |
6735 | return NULL; | |
6736 | ||
dc938520 | 6737 | init_rootdomain(rd); |
57d885fe GH |
6738 | |
6739 | return rd; | |
6740 | } | |
6741 | ||
1da177e4 | 6742 | /* |
0eab9146 | 6743 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6744 | * hold the hotplug lock. |
6745 | */ | |
0eab9146 IM |
6746 | static void |
6747 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6748 | { |
70b97a7f | 6749 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6750 | struct sched_domain *tmp; |
6751 | ||
6752 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6753 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6754 | struct sched_domain *parent = tmp->parent; | |
6755 | if (!parent) | |
6756 | break; | |
1a848870 | 6757 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6758 | tmp->parent = parent->parent; |
1a848870 SS |
6759 | if (parent->parent) |
6760 | parent->parent->child = tmp; | |
6761 | } | |
245af2c7 SS |
6762 | } |
6763 | ||
1a848870 | 6764 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6765 | sd = sd->parent; |
1a848870 SS |
6766 | if (sd) |
6767 | sd->child = NULL; | |
6768 | } | |
1da177e4 LT |
6769 | |
6770 | sched_domain_debug(sd, cpu); | |
6771 | ||
57d885fe | 6772 | rq_attach_root(rq, rd); |
674311d5 | 6773 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6774 | } |
6775 | ||
6776 | /* cpus with isolated domains */ | |
67af63a6 | 6777 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6778 | |
6779 | /* Setup the mask of cpus configured for isolated domains */ | |
6780 | static int __init isolated_cpu_setup(char *str) | |
6781 | { | |
13b40c1e MT |
6782 | static int __initdata ints[NR_CPUS]; |
6783 | int i; | |
1da177e4 LT |
6784 | |
6785 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6786 | cpus_clear(cpu_isolated_map); | |
6787 | for (i = 1; i <= ints[0]; i++) | |
6788 | if (ints[i] < NR_CPUS) | |
6789 | cpu_set(ints[i], cpu_isolated_map); | |
6790 | return 1; | |
6791 | } | |
6792 | ||
8927f494 | 6793 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6794 | |
6795 | /* | |
6711cab4 SS |
6796 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6797 | * to a function which identifies what group(along with sched group) a CPU | |
6798 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6799 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6800 | * |
6801 | * init_sched_build_groups will build a circular linked list of the groups | |
6802 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6803 | * and ->cpu_power to 0. | |
6804 | */ | |
a616058b | 6805 | static void |
7c16ec58 | 6806 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6807 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6808 | struct sched_group **sg, |
6809 | cpumask_t *tmpmask), | |
6810 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6811 | { |
6812 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6813 | int i; |
6814 | ||
7c16ec58 MT |
6815 | cpus_clear(*covered); |
6816 | ||
363ab6f1 | 6817 | for_each_cpu_mask_nr(i, *span) { |
6711cab4 | 6818 | struct sched_group *sg; |
7c16ec58 | 6819 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6820 | int j; |
6821 | ||
7c16ec58 | 6822 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6823 | continue; |
6824 | ||
7c16ec58 | 6825 | cpus_clear(sg->cpumask); |
5517d86b | 6826 | sg->__cpu_power = 0; |
1da177e4 | 6827 | |
363ab6f1 | 6828 | for_each_cpu_mask_nr(j, *span) { |
7c16ec58 | 6829 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6830 | continue; |
6831 | ||
7c16ec58 | 6832 | cpu_set(j, *covered); |
1da177e4 LT |
6833 | cpu_set(j, sg->cpumask); |
6834 | } | |
6835 | if (!first) | |
6836 | first = sg; | |
6837 | if (last) | |
6838 | last->next = sg; | |
6839 | last = sg; | |
6840 | } | |
6841 | last->next = first; | |
6842 | } | |
6843 | ||
9c1cfda2 | 6844 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6845 | |
9c1cfda2 | 6846 | #ifdef CONFIG_NUMA |
198e2f18 | 6847 | |
9c1cfda2 JH |
6848 | /** |
6849 | * find_next_best_node - find the next node to include in a sched_domain | |
6850 | * @node: node whose sched_domain we're building | |
6851 | * @used_nodes: nodes already in the sched_domain | |
6852 | * | |
41a2d6cf | 6853 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6854 | * finds the closest node not already in the @used_nodes map. |
6855 | * | |
6856 | * Should use nodemask_t. | |
6857 | */ | |
c5f59f08 | 6858 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6859 | { |
6860 | int i, n, val, min_val, best_node = 0; | |
6861 | ||
6862 | min_val = INT_MAX; | |
6863 | ||
076ac2af | 6864 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6865 | /* Start at @node */ |
076ac2af | 6866 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6867 | |
6868 | if (!nr_cpus_node(n)) | |
6869 | continue; | |
6870 | ||
6871 | /* Skip already used nodes */ | |
c5f59f08 | 6872 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6873 | continue; |
6874 | ||
6875 | /* Simple min distance search */ | |
6876 | val = node_distance(node, n); | |
6877 | ||
6878 | if (val < min_val) { | |
6879 | min_val = val; | |
6880 | best_node = n; | |
6881 | } | |
6882 | } | |
6883 | ||
c5f59f08 | 6884 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6885 | return best_node; |
6886 | } | |
6887 | ||
6888 | /** | |
6889 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6890 | * @node: node whose cpumask we're constructing | |
73486722 | 6891 | * @span: resulting cpumask |
9c1cfda2 | 6892 | * |
41a2d6cf | 6893 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6894 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6895 | * out optimally. | |
6896 | */ | |
4bdbaad3 | 6897 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6898 | { |
c5f59f08 | 6899 | nodemask_t used_nodes; |
c5f59f08 | 6900 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6901 | int i; |
9c1cfda2 | 6902 | |
4bdbaad3 | 6903 | cpus_clear(*span); |
c5f59f08 | 6904 | nodes_clear(used_nodes); |
9c1cfda2 | 6905 | |
4bdbaad3 | 6906 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6907 | node_set(node, used_nodes); |
9c1cfda2 JH |
6908 | |
6909 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6910 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6911 | |
c5f59f08 | 6912 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6913 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6914 | } |
9c1cfda2 | 6915 | } |
6d6bc0ad | 6916 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6917 | |
5c45bf27 | 6918 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6919 | |
9c1cfda2 | 6920 | /* |
48f24c4d | 6921 | * SMT sched-domains: |
9c1cfda2 | 6922 | */ |
1da177e4 LT |
6923 | #ifdef CONFIG_SCHED_SMT |
6924 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6925 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6926 | |
41a2d6cf | 6927 | static int |
7c16ec58 MT |
6928 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6929 | cpumask_t *unused) | |
1da177e4 | 6930 | { |
6711cab4 SS |
6931 | if (sg) |
6932 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6933 | return cpu; |
6934 | } | |
6d6bc0ad | 6935 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6936 | |
48f24c4d IM |
6937 | /* |
6938 | * multi-core sched-domains: | |
6939 | */ | |
1e9f28fa SS |
6940 | #ifdef CONFIG_SCHED_MC |
6941 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6942 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6943 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6944 | |
6945 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6946 | static int |
7c16ec58 MT |
6947 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6948 | cpumask_t *mask) | |
1e9f28fa | 6949 | { |
6711cab4 | 6950 | int group; |
7c16ec58 MT |
6951 | |
6952 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6953 | cpus_and(*mask, *mask, *cpu_map); | |
6954 | group = first_cpu(*mask); | |
6711cab4 SS |
6955 | if (sg) |
6956 | *sg = &per_cpu(sched_group_core, group); | |
6957 | return group; | |
1e9f28fa SS |
6958 | } |
6959 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6960 | static int |
7c16ec58 MT |
6961 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6962 | cpumask_t *unused) | |
1e9f28fa | 6963 | { |
6711cab4 SS |
6964 | if (sg) |
6965 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6966 | return cpu; |
6967 | } | |
6968 | #endif | |
6969 | ||
1da177e4 | 6970 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6971 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6972 | |
41a2d6cf | 6973 | static int |
7c16ec58 MT |
6974 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6975 | cpumask_t *mask) | |
1da177e4 | 6976 | { |
6711cab4 | 6977 | int group; |
48f24c4d | 6978 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6979 | *mask = cpu_coregroup_map(cpu); |
6980 | cpus_and(*mask, *mask, *cpu_map); | |
6981 | group = first_cpu(*mask); | |
1e9f28fa | 6982 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6983 | *mask = per_cpu(cpu_sibling_map, cpu); |
6984 | cpus_and(*mask, *mask, *cpu_map); | |
6985 | group = first_cpu(*mask); | |
1da177e4 | 6986 | #else |
6711cab4 | 6987 | group = cpu; |
1da177e4 | 6988 | #endif |
6711cab4 SS |
6989 | if (sg) |
6990 | *sg = &per_cpu(sched_group_phys, group); | |
6991 | return group; | |
1da177e4 LT |
6992 | } |
6993 | ||
6994 | #ifdef CONFIG_NUMA | |
1da177e4 | 6995 | /* |
9c1cfda2 JH |
6996 | * The init_sched_build_groups can't handle what we want to do with node |
6997 | * groups, so roll our own. Now each node has its own list of groups which | |
6998 | * gets dynamically allocated. | |
1da177e4 | 6999 | */ |
9c1cfda2 | 7000 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7001 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7002 | |
9c1cfda2 | 7003 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 7004 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 7005 | |
6711cab4 | 7006 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 7007 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 7008 | { |
6711cab4 SS |
7009 | int group; |
7010 | ||
7c16ec58 MT |
7011 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
7012 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7013 | group = first_cpu(*nodemask); | |
6711cab4 SS |
7014 | |
7015 | if (sg) | |
7016 | *sg = &per_cpu(sched_group_allnodes, group); | |
7017 | return group; | |
1da177e4 | 7018 | } |
6711cab4 | 7019 | |
08069033 SS |
7020 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7021 | { | |
7022 | struct sched_group *sg = group_head; | |
7023 | int j; | |
7024 | ||
7025 | if (!sg) | |
7026 | return; | |
3a5c359a | 7027 | do { |
363ab6f1 | 7028 | for_each_cpu_mask_nr(j, sg->cpumask) { |
3a5c359a | 7029 | struct sched_domain *sd; |
08069033 | 7030 | |
3a5c359a AK |
7031 | sd = &per_cpu(phys_domains, j); |
7032 | if (j != first_cpu(sd->groups->cpumask)) { | |
7033 | /* | |
7034 | * Only add "power" once for each | |
7035 | * physical package. | |
7036 | */ | |
7037 | continue; | |
7038 | } | |
08069033 | 7039 | |
3a5c359a AK |
7040 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7041 | } | |
7042 | sg = sg->next; | |
7043 | } while (sg != group_head); | |
08069033 | 7044 | } |
6d6bc0ad | 7045 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7046 | |
a616058b | 7047 | #ifdef CONFIG_NUMA |
51888ca2 | 7048 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 7049 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 7050 | { |
a616058b | 7051 | int cpu, i; |
51888ca2 | 7052 | |
363ab6f1 | 7053 | for_each_cpu_mask_nr(cpu, *cpu_map) { |
51888ca2 SV |
7054 | struct sched_group **sched_group_nodes |
7055 | = sched_group_nodes_bycpu[cpu]; | |
7056 | ||
51888ca2 SV |
7057 | if (!sched_group_nodes) |
7058 | continue; | |
7059 | ||
076ac2af | 7060 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7061 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7062 | ||
7c16ec58 MT |
7063 | *nodemask = node_to_cpumask(i); |
7064 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7065 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
7066 | continue; |
7067 | ||
7068 | if (sg == NULL) | |
7069 | continue; | |
7070 | sg = sg->next; | |
7071 | next_sg: | |
7072 | oldsg = sg; | |
7073 | sg = sg->next; | |
7074 | kfree(oldsg); | |
7075 | if (oldsg != sched_group_nodes[i]) | |
7076 | goto next_sg; | |
7077 | } | |
7078 | kfree(sched_group_nodes); | |
7079 | sched_group_nodes_bycpu[cpu] = NULL; | |
7080 | } | |
51888ca2 | 7081 | } |
6d6bc0ad | 7082 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 7083 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
7084 | { |
7085 | } | |
6d6bc0ad | 7086 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7087 | |
89c4710e SS |
7088 | /* |
7089 | * Initialize sched groups cpu_power. | |
7090 | * | |
7091 | * cpu_power indicates the capacity of sched group, which is used while | |
7092 | * distributing the load between different sched groups in a sched domain. | |
7093 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7094 | * there are asymmetries in the topology. If there are asymmetries, group | |
7095 | * having more cpu_power will pickup more load compared to the group having | |
7096 | * less cpu_power. | |
7097 | * | |
7098 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7099 | * the maximum number of tasks a group can handle in the presence of other idle | |
7100 | * or lightly loaded groups in the same sched domain. | |
7101 | */ | |
7102 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7103 | { | |
7104 | struct sched_domain *child; | |
7105 | struct sched_group *group; | |
7106 | ||
7107 | WARN_ON(!sd || !sd->groups); | |
7108 | ||
7109 | if (cpu != first_cpu(sd->groups->cpumask)) | |
7110 | return; | |
7111 | ||
7112 | child = sd->child; | |
7113 | ||
5517d86b ED |
7114 | sd->groups->__cpu_power = 0; |
7115 | ||
89c4710e SS |
7116 | /* |
7117 | * For perf policy, if the groups in child domain share resources | |
7118 | * (for example cores sharing some portions of the cache hierarchy | |
7119 | * or SMT), then set this domain groups cpu_power such that each group | |
7120 | * can handle only one task, when there are other idle groups in the | |
7121 | * same sched domain. | |
7122 | */ | |
7123 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7124 | (child->flags & | |
7125 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7126 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7127 | return; |
7128 | } | |
7129 | ||
89c4710e SS |
7130 | /* |
7131 | * add cpu_power of each child group to this groups cpu_power | |
7132 | */ | |
7133 | group = child->groups; | |
7134 | do { | |
5517d86b | 7135 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7136 | group = group->next; |
7137 | } while (group != child->groups); | |
7138 | } | |
7139 | ||
7c16ec58 MT |
7140 | /* |
7141 | * Initializers for schedule domains | |
7142 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7143 | */ | |
7144 | ||
7145 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
7146 | #define SD_INIT_FUNC(type) \ | |
7147 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7148 | { \ | |
7149 | memset(sd, 0, sizeof(*sd)); \ | |
7150 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7151 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
7152 | } |
7153 | ||
7154 | SD_INIT_FUNC(CPU) | |
7155 | #ifdef CONFIG_NUMA | |
7156 | SD_INIT_FUNC(ALLNODES) | |
7157 | SD_INIT_FUNC(NODE) | |
7158 | #endif | |
7159 | #ifdef CONFIG_SCHED_SMT | |
7160 | SD_INIT_FUNC(SIBLING) | |
7161 | #endif | |
7162 | #ifdef CONFIG_SCHED_MC | |
7163 | SD_INIT_FUNC(MC) | |
7164 | #endif | |
7165 | ||
7166 | /* | |
7167 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7168 | * space as the usage in build_sched_domains() dictates. Used only | |
7169 | * if the amount of space is significant. | |
7170 | */ | |
7171 | struct allmasks { | |
7172 | cpumask_t tmpmask; /* make this one first */ | |
7173 | union { | |
7174 | cpumask_t nodemask; | |
7175 | cpumask_t this_sibling_map; | |
7176 | cpumask_t this_core_map; | |
7177 | }; | |
7178 | cpumask_t send_covered; | |
7179 | ||
7180 | #ifdef CONFIG_NUMA | |
7181 | cpumask_t domainspan; | |
7182 | cpumask_t covered; | |
7183 | cpumask_t notcovered; | |
7184 | #endif | |
7185 | }; | |
7186 | ||
7187 | #if NR_CPUS > 128 | |
7188 | #define SCHED_CPUMASK_ALLOC 1 | |
7189 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7190 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7191 | #else | |
7192 | #define SCHED_CPUMASK_ALLOC 0 | |
7193 | #define SCHED_CPUMASK_FREE(v) | |
7194 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7195 | #endif | |
7196 | ||
7197 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7198 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7199 | ||
1d3504fc HS |
7200 | static int default_relax_domain_level = -1; |
7201 | ||
7202 | static int __init setup_relax_domain_level(char *str) | |
7203 | { | |
30e0e178 LZ |
7204 | unsigned long val; |
7205 | ||
7206 | val = simple_strtoul(str, NULL, 0); | |
7207 | if (val < SD_LV_MAX) | |
7208 | default_relax_domain_level = val; | |
7209 | ||
1d3504fc HS |
7210 | return 1; |
7211 | } | |
7212 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7213 | ||
7214 | static void set_domain_attribute(struct sched_domain *sd, | |
7215 | struct sched_domain_attr *attr) | |
7216 | { | |
7217 | int request; | |
7218 | ||
7219 | if (!attr || attr->relax_domain_level < 0) { | |
7220 | if (default_relax_domain_level < 0) | |
7221 | return; | |
7222 | else | |
7223 | request = default_relax_domain_level; | |
7224 | } else | |
7225 | request = attr->relax_domain_level; | |
7226 | if (request < sd->level) { | |
7227 | /* turn off idle balance on this domain */ | |
7228 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7229 | } else { | |
7230 | /* turn on idle balance on this domain */ | |
7231 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7232 | } | |
7233 | } | |
7234 | ||
1da177e4 | 7235 | /* |
1a20ff27 DG |
7236 | * Build sched domains for a given set of cpus and attach the sched domains |
7237 | * to the individual cpus | |
1da177e4 | 7238 | */ |
1d3504fc HS |
7239 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7240 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7241 | { |
7242 | int i; | |
57d885fe | 7243 | struct root_domain *rd; |
7c16ec58 MT |
7244 | SCHED_CPUMASK_DECLARE(allmasks); |
7245 | cpumask_t *tmpmask; | |
d1b55138 JH |
7246 | #ifdef CONFIG_NUMA |
7247 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7248 | int sd_allnodes = 0; |
d1b55138 JH |
7249 | |
7250 | /* | |
7251 | * Allocate the per-node list of sched groups | |
7252 | */ | |
076ac2af | 7253 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7254 | GFP_KERNEL); |
d1b55138 JH |
7255 | if (!sched_group_nodes) { |
7256 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7257 | return -ENOMEM; |
d1b55138 | 7258 | } |
d1b55138 | 7259 | #endif |
1da177e4 | 7260 | |
dc938520 | 7261 | rd = alloc_rootdomain(); |
57d885fe GH |
7262 | if (!rd) { |
7263 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7264 | #ifdef CONFIG_NUMA |
7265 | kfree(sched_group_nodes); | |
7266 | #endif | |
57d885fe GH |
7267 | return -ENOMEM; |
7268 | } | |
7269 | ||
7c16ec58 MT |
7270 | #if SCHED_CPUMASK_ALLOC |
7271 | /* get space for all scratch cpumask variables */ | |
7272 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7273 | if (!allmasks) { | |
7274 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7275 | kfree(rd); | |
7276 | #ifdef CONFIG_NUMA | |
7277 | kfree(sched_group_nodes); | |
7278 | #endif | |
7279 | return -ENOMEM; | |
7280 | } | |
7281 | #endif | |
7282 | tmpmask = (cpumask_t *)allmasks; | |
7283 | ||
7284 | ||
7285 | #ifdef CONFIG_NUMA | |
7286 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7287 | #endif | |
7288 | ||
1da177e4 | 7289 | /* |
1a20ff27 | 7290 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7291 | */ |
363ab6f1 | 7292 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 | 7293 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7294 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7295 | |
7c16ec58 MT |
7296 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7297 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7298 | |
7299 | #ifdef CONFIG_NUMA | |
dd41f596 | 7300 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7301 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7302 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7303 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7304 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7305 | sd->span = *cpu_map; |
7c16ec58 | 7306 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7307 | p = sd; |
6711cab4 | 7308 | sd_allnodes = 1; |
9c1cfda2 JH |
7309 | } else |
7310 | p = NULL; | |
7311 | ||
1da177e4 | 7312 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7313 | SD_INIT(sd, NODE); |
1d3504fc | 7314 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7315 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7316 | sd->parent = p; |
1a848870 SS |
7317 | if (p) |
7318 | p->child = sd; | |
9c1cfda2 | 7319 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7320 | #endif |
7321 | ||
7322 | p = sd; | |
7323 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7324 | SD_INIT(sd, CPU); |
1d3504fc | 7325 | set_domain_attribute(sd, attr); |
7c16ec58 | 7326 | sd->span = *nodemask; |
1da177e4 | 7327 | sd->parent = p; |
1a848870 SS |
7328 | if (p) |
7329 | p->child = sd; | |
7c16ec58 | 7330 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7331 | |
1e9f28fa SS |
7332 | #ifdef CONFIG_SCHED_MC |
7333 | p = sd; | |
7334 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7335 | SD_INIT(sd, MC); |
1d3504fc | 7336 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7337 | sd->span = cpu_coregroup_map(i); |
7338 | cpus_and(sd->span, sd->span, *cpu_map); | |
7339 | sd->parent = p; | |
1a848870 | 7340 | p->child = sd; |
7c16ec58 | 7341 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7342 | #endif |
7343 | ||
1da177e4 LT |
7344 | #ifdef CONFIG_SCHED_SMT |
7345 | p = sd; | |
7346 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7347 | SD_INIT(sd, SIBLING); |
1d3504fc | 7348 | set_domain_attribute(sd, attr); |
d5a7430d | 7349 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7350 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7351 | sd->parent = p; |
1a848870 | 7352 | p->child = sd; |
7c16ec58 | 7353 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7354 | #endif |
7355 | } | |
7356 | ||
7357 | #ifdef CONFIG_SCHED_SMT | |
7358 | /* Set up CPU (sibling) groups */ | |
363ab6f1 | 7359 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7360 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7361 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7362 | ||
7363 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7364 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7365 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7366 | continue; |
7367 | ||
dd41f596 | 7368 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7369 | &cpu_to_cpu_group, |
7370 | send_covered, tmpmask); | |
1da177e4 LT |
7371 | } |
7372 | #endif | |
7373 | ||
1e9f28fa SS |
7374 | #ifdef CONFIG_SCHED_MC |
7375 | /* Set up multi-core groups */ | |
363ab6f1 | 7376 | for_each_cpu_mask_nr(i, *cpu_map) { |
7c16ec58 MT |
7377 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7378 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7379 | ||
7380 | *this_core_map = cpu_coregroup_map(i); | |
7381 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7382 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7383 | continue; |
7c16ec58 | 7384 | |
dd41f596 | 7385 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7386 | &cpu_to_core_group, |
7387 | send_covered, tmpmask); | |
1e9f28fa SS |
7388 | } |
7389 | #endif | |
7390 | ||
1da177e4 | 7391 | /* Set up physical groups */ |
076ac2af | 7392 | for (i = 0; i < nr_node_ids; i++) { |
7c16ec58 MT |
7393 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7394 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7395 | |
7c16ec58 MT |
7396 | *nodemask = node_to_cpumask(i); |
7397 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7398 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7399 | continue; |
7400 | ||
7c16ec58 MT |
7401 | init_sched_build_groups(nodemask, cpu_map, |
7402 | &cpu_to_phys_group, | |
7403 | send_covered, tmpmask); | |
1da177e4 LT |
7404 | } |
7405 | ||
7406 | #ifdef CONFIG_NUMA | |
7407 | /* Set up node groups */ | |
7c16ec58 MT |
7408 | if (sd_allnodes) { |
7409 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7410 | ||
7411 | init_sched_build_groups(cpu_map, cpu_map, | |
7412 | &cpu_to_allnodes_group, | |
7413 | send_covered, tmpmask); | |
7414 | } | |
9c1cfda2 | 7415 | |
076ac2af | 7416 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7417 | /* Set up node groups */ |
7418 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7419 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7420 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7421 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7422 | int j; |
7423 | ||
7c16ec58 MT |
7424 | *nodemask = node_to_cpumask(i); |
7425 | cpus_clear(*covered); | |
7426 | ||
7427 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7428 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7429 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7430 | continue; |
d1b55138 | 7431 | } |
9c1cfda2 | 7432 | |
4bdbaad3 | 7433 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7434 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7435 | |
15f0b676 | 7436 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7437 | if (!sg) { |
7438 | printk(KERN_WARNING "Can not alloc domain group for " | |
7439 | "node %d\n", i); | |
7440 | goto error; | |
7441 | } | |
9c1cfda2 | 7442 | sched_group_nodes[i] = sg; |
363ab6f1 | 7443 | for_each_cpu_mask_nr(j, *nodemask) { |
9c1cfda2 | 7444 | struct sched_domain *sd; |
9761eea8 | 7445 | |
9c1cfda2 JH |
7446 | sd = &per_cpu(node_domains, j); |
7447 | sd->groups = sg; | |
9c1cfda2 | 7448 | } |
5517d86b | 7449 | sg->__cpu_power = 0; |
7c16ec58 | 7450 | sg->cpumask = *nodemask; |
51888ca2 | 7451 | sg->next = sg; |
7c16ec58 | 7452 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7453 | prev = sg; |
7454 | ||
076ac2af | 7455 | for (j = 0; j < nr_node_ids; j++) { |
7c16ec58 | 7456 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
076ac2af | 7457 | int n = (i + j) % nr_node_ids; |
c5f59f08 | 7458 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7459 | |
7c16ec58 MT |
7460 | cpus_complement(*notcovered, *covered); |
7461 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7462 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7463 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7464 | break; |
7465 | ||
7c16ec58 MT |
7466 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7467 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7468 | continue; |
7469 | ||
15f0b676 SV |
7470 | sg = kmalloc_node(sizeof(struct sched_group), |
7471 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7472 | if (!sg) { |
7473 | printk(KERN_WARNING | |
7474 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7475 | goto error; |
9c1cfda2 | 7476 | } |
5517d86b | 7477 | sg->__cpu_power = 0; |
7c16ec58 | 7478 | sg->cpumask = *tmpmask; |
51888ca2 | 7479 | sg->next = prev->next; |
7c16ec58 | 7480 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7481 | prev->next = sg; |
7482 | prev = sg; | |
7483 | } | |
9c1cfda2 | 7484 | } |
1da177e4 LT |
7485 | #endif |
7486 | ||
7487 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7488 | #ifdef CONFIG_SCHED_SMT |
363ab6f1 | 7489 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7490 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7491 | ||
89c4710e | 7492 | init_sched_groups_power(i, sd); |
5c45bf27 | 7493 | } |
1da177e4 | 7494 | #endif |
1e9f28fa | 7495 | #ifdef CONFIG_SCHED_MC |
363ab6f1 | 7496 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7497 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7498 | ||
89c4710e | 7499 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7500 | } |
7501 | #endif | |
1e9f28fa | 7502 | |
363ab6f1 | 7503 | for_each_cpu_mask_nr(i, *cpu_map) { |
dd41f596 IM |
7504 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7505 | ||
89c4710e | 7506 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7507 | } |
7508 | ||
9c1cfda2 | 7509 | #ifdef CONFIG_NUMA |
076ac2af | 7510 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7511 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7512 | |
6711cab4 SS |
7513 | if (sd_allnodes) { |
7514 | struct sched_group *sg; | |
f712c0c7 | 7515 | |
7c16ec58 MT |
7516 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7517 | tmpmask); | |
f712c0c7 SS |
7518 | init_numa_sched_groups_power(sg); |
7519 | } | |
9c1cfda2 JH |
7520 | #endif |
7521 | ||
1da177e4 | 7522 | /* Attach the domains */ |
363ab6f1 | 7523 | for_each_cpu_mask_nr(i, *cpu_map) { |
1da177e4 LT |
7524 | struct sched_domain *sd; |
7525 | #ifdef CONFIG_SCHED_SMT | |
7526 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7527 | #elif defined(CONFIG_SCHED_MC) |
7528 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7529 | #else |
7530 | sd = &per_cpu(phys_domains, i); | |
7531 | #endif | |
57d885fe | 7532 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7533 | } |
51888ca2 | 7534 | |
7c16ec58 | 7535 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7536 | return 0; |
7537 | ||
a616058b | 7538 | #ifdef CONFIG_NUMA |
51888ca2 | 7539 | error: |
7c16ec58 MT |
7540 | free_sched_groups(cpu_map, tmpmask); |
7541 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7542 | return -ENOMEM; |
a616058b | 7543 | #endif |
1da177e4 | 7544 | } |
029190c5 | 7545 | |
1d3504fc HS |
7546 | static int build_sched_domains(const cpumask_t *cpu_map) |
7547 | { | |
7548 | return __build_sched_domains(cpu_map, NULL); | |
7549 | } | |
7550 | ||
029190c5 PJ |
7551 | static cpumask_t *doms_cur; /* current sched domains */ |
7552 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7553 | static struct sched_domain_attr *dattr_cur; |
7554 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7555 | |
7556 | /* | |
7557 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7558 | * cpumask_t) fails, then fallback to a single sched domain, | |
7559 | * as determined by the single cpumask_t fallback_doms. | |
7560 | */ | |
7561 | static cpumask_t fallback_doms; | |
7562 | ||
22e52b07 HC |
7563 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7564 | { | |
7565 | } | |
7566 | ||
1a20ff27 | 7567 | /* |
41a2d6cf | 7568 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7569 | * For now this just excludes isolated cpus, but could be used to |
7570 | * exclude other special cases in the future. | |
1a20ff27 | 7571 | */ |
51888ca2 | 7572 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7573 | { |
7378547f MM |
7574 | int err; |
7575 | ||
22e52b07 | 7576 | arch_update_cpu_topology(); |
029190c5 PJ |
7577 | ndoms_cur = 1; |
7578 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7579 | if (!doms_cur) | |
7580 | doms_cur = &fallback_doms; | |
7581 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7582 | dattr_cur = NULL; |
7378547f | 7583 | err = build_sched_domains(doms_cur); |
6382bc90 | 7584 | register_sched_domain_sysctl(); |
7378547f MM |
7585 | |
7586 | return err; | |
1a20ff27 DG |
7587 | } |
7588 | ||
7c16ec58 MT |
7589 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7590 | cpumask_t *tmpmask) | |
1da177e4 | 7591 | { |
7c16ec58 | 7592 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7593 | } |
1da177e4 | 7594 | |
1a20ff27 DG |
7595 | /* |
7596 | * Detach sched domains from a group of cpus specified in cpu_map | |
7597 | * These cpus will now be attached to the NULL domain | |
7598 | */ | |
858119e1 | 7599 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7600 | { |
7c16ec58 | 7601 | cpumask_t tmpmask; |
1a20ff27 DG |
7602 | int i; |
7603 | ||
6382bc90 MM |
7604 | unregister_sched_domain_sysctl(); |
7605 | ||
363ab6f1 | 7606 | for_each_cpu_mask_nr(i, *cpu_map) |
57d885fe | 7607 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7608 | synchronize_sched(); |
7c16ec58 | 7609 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7610 | } |
7611 | ||
1d3504fc HS |
7612 | /* handle null as "default" */ |
7613 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7614 | struct sched_domain_attr *new, int idx_new) | |
7615 | { | |
7616 | struct sched_domain_attr tmp; | |
7617 | ||
7618 | /* fast path */ | |
7619 | if (!new && !cur) | |
7620 | return 1; | |
7621 | ||
7622 | tmp = SD_ATTR_INIT; | |
7623 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7624 | new ? (new + idx_new) : &tmp, | |
7625 | sizeof(struct sched_domain_attr)); | |
7626 | } | |
7627 | ||
029190c5 PJ |
7628 | /* |
7629 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7630 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7631 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7632 | * It destroys each deleted domain and builds each new domain. | |
7633 | * | |
7634 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7635 | * The masks don't intersect (don't overlap.) We should setup one |
7636 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7637 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7638 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7639 | * it as it is. | |
7640 | * | |
41a2d6cf IM |
7641 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7642 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7643 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7644 | * and partition_sched_domains() will fallback to the single partition | |
e761b772 | 7645 | * 'fallback_doms', it also forces the domains to be rebuilt. |
029190c5 PJ |
7646 | * |
7647 | * Call with hotplug lock held | |
7648 | */ | |
1d3504fc HS |
7649 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7650 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7651 | { |
7652 | int i, j; | |
7653 | ||
712555ee | 7654 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7655 | |
7378547f MM |
7656 | /* always unregister in case we don't destroy any domains */ |
7657 | unregister_sched_domain_sysctl(); | |
7658 | ||
e761b772 MK |
7659 | if (doms_new == NULL) |
7660 | ndoms_new = 0; | |
029190c5 PJ |
7661 | |
7662 | /* Destroy deleted domains */ | |
7663 | for (i = 0; i < ndoms_cur; i++) { | |
7664 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7665 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7666 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7667 | goto match1; |
7668 | } | |
7669 | /* no match - a current sched domain not in new doms_new[] */ | |
7670 | detach_destroy_domains(doms_cur + i); | |
7671 | match1: | |
7672 | ; | |
7673 | } | |
7674 | ||
e761b772 MK |
7675 | if (doms_new == NULL) { |
7676 | ndoms_cur = 0; | |
7677 | ndoms_new = 1; | |
7678 | doms_new = &fallback_doms; | |
7679 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
7680 | dattr_new = NULL; | |
7681 | } | |
7682 | ||
029190c5 PJ |
7683 | /* Build new domains */ |
7684 | for (i = 0; i < ndoms_new; i++) { | |
7685 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7686 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7687 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7688 | goto match2; |
7689 | } | |
7690 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7691 | __build_sched_domains(doms_new + i, |
7692 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7693 | match2: |
7694 | ; | |
7695 | } | |
7696 | ||
7697 | /* Remember the new sched domains */ | |
7698 | if (doms_cur != &fallback_doms) | |
7699 | kfree(doms_cur); | |
1d3504fc | 7700 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7701 | doms_cur = doms_new; |
1d3504fc | 7702 | dattr_cur = dattr_new; |
029190c5 | 7703 | ndoms_cur = ndoms_new; |
7378547f MM |
7704 | |
7705 | register_sched_domain_sysctl(); | |
a1835615 | 7706 | |
712555ee | 7707 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7708 | } |
7709 | ||
5c45bf27 | 7710 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7711 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7712 | { |
95402b38 | 7713 | get_online_cpus(); |
e761b772 | 7714 | rebuild_sched_domains(); |
95402b38 | 7715 | put_online_cpus(); |
e761b772 | 7716 | return 0; |
5c45bf27 SS |
7717 | } |
7718 | ||
7719 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7720 | { | |
7721 | int ret; | |
7722 | ||
7723 | if (buf[0] != '0' && buf[0] != '1') | |
7724 | return -EINVAL; | |
7725 | ||
7726 | if (smt) | |
7727 | sched_smt_power_savings = (buf[0] == '1'); | |
7728 | else | |
7729 | sched_mc_power_savings = (buf[0] == '1'); | |
7730 | ||
7731 | ret = arch_reinit_sched_domains(); | |
7732 | ||
7733 | return ret ? ret : count; | |
7734 | } | |
7735 | ||
5c45bf27 | 7736 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7737 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7738 | char *page) | |
5c45bf27 SS |
7739 | { |
7740 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7741 | } | |
f718cd4a | 7742 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7743 | const char *buf, size_t count) |
5c45bf27 SS |
7744 | { |
7745 | return sched_power_savings_store(buf, count, 0); | |
7746 | } | |
f718cd4a AK |
7747 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7748 | sched_mc_power_savings_show, | |
7749 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7750 | #endif |
7751 | ||
7752 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7753 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7754 | char *page) | |
5c45bf27 SS |
7755 | { |
7756 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7757 | } | |
f718cd4a | 7758 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7759 | const char *buf, size_t count) |
5c45bf27 SS |
7760 | { |
7761 | return sched_power_savings_store(buf, count, 1); | |
7762 | } | |
f718cd4a AK |
7763 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7764 | sched_smt_power_savings_show, | |
6707de00 AB |
7765 | sched_smt_power_savings_store); |
7766 | #endif | |
7767 | ||
7768 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7769 | { | |
7770 | int err = 0; | |
7771 | ||
7772 | #ifdef CONFIG_SCHED_SMT | |
7773 | if (smt_capable()) | |
7774 | err = sysfs_create_file(&cls->kset.kobj, | |
7775 | &attr_sched_smt_power_savings.attr); | |
7776 | #endif | |
7777 | #ifdef CONFIG_SCHED_MC | |
7778 | if (!err && mc_capable()) | |
7779 | err = sysfs_create_file(&cls->kset.kobj, | |
7780 | &attr_sched_mc_power_savings.attr); | |
7781 | #endif | |
7782 | return err; | |
7783 | } | |
6d6bc0ad | 7784 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7785 | |
e761b772 | 7786 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7787 | /* |
e761b772 MK |
7788 | * Add online and remove offline CPUs from the scheduler domains. |
7789 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7790 | */ |
7791 | static int update_sched_domains(struct notifier_block *nfb, | |
7792 | unsigned long action, void *hcpu) | |
e761b772 MK |
7793 | { |
7794 | switch (action) { | |
7795 | case CPU_ONLINE: | |
7796 | case CPU_ONLINE_FROZEN: | |
7797 | case CPU_DEAD: | |
7798 | case CPU_DEAD_FROZEN: | |
7799 | partition_sched_domains(0, NULL, NULL); | |
7800 | return NOTIFY_OK; | |
7801 | ||
7802 | default: | |
7803 | return NOTIFY_DONE; | |
7804 | } | |
7805 | } | |
7806 | #endif | |
7807 | ||
7808 | static int update_runtime(struct notifier_block *nfb, | |
7809 | unsigned long action, void *hcpu) | |
1da177e4 | 7810 | { |
7def2be1 PZ |
7811 | int cpu = (int)(long)hcpu; |
7812 | ||
1da177e4 | 7813 | switch (action) { |
1da177e4 | 7814 | case CPU_DOWN_PREPARE: |
8bb78442 | 7815 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7816 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7817 | return NOTIFY_OK; |
7818 | ||
1da177e4 | 7819 | case CPU_DOWN_FAILED: |
8bb78442 | 7820 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7821 | case CPU_ONLINE: |
8bb78442 | 7822 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7823 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7824 | return NOTIFY_OK; |
7825 | ||
1da177e4 LT |
7826 | default: |
7827 | return NOTIFY_DONE; | |
7828 | } | |
1da177e4 | 7829 | } |
1da177e4 LT |
7830 | |
7831 | void __init sched_init_smp(void) | |
7832 | { | |
5c1e1767 NP |
7833 | cpumask_t non_isolated_cpus; |
7834 | ||
434d53b0 MT |
7835 | #if defined(CONFIG_NUMA) |
7836 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7837 | GFP_KERNEL); | |
7838 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7839 | #endif | |
95402b38 | 7840 | get_online_cpus(); |
712555ee | 7841 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7842 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7843 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7844 | if (cpus_empty(non_isolated_cpus)) |
7845 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7846 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7847 | put_online_cpus(); |
e761b772 MK |
7848 | |
7849 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7850 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7851 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7852 | #endif |
7853 | ||
7854 | /* RT runtime code needs to handle some hotplug events */ | |
7855 | hotcpu_notifier(update_runtime, 0); | |
7856 | ||
b328ca18 | 7857 | init_hrtick(); |
5c1e1767 NP |
7858 | |
7859 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7860 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7861 | BUG(); |
19978ca6 | 7862 | sched_init_granularity(); |
1da177e4 LT |
7863 | } |
7864 | #else | |
7865 | void __init sched_init_smp(void) | |
7866 | { | |
19978ca6 | 7867 | sched_init_granularity(); |
1da177e4 LT |
7868 | } |
7869 | #endif /* CONFIG_SMP */ | |
7870 | ||
7871 | int in_sched_functions(unsigned long addr) | |
7872 | { | |
1da177e4 LT |
7873 | return in_lock_functions(addr) || |
7874 | (addr >= (unsigned long)__sched_text_start | |
7875 | && addr < (unsigned long)__sched_text_end); | |
7876 | } | |
7877 | ||
a9957449 | 7878 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7879 | { |
7880 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7881 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7882 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7883 | cfs_rq->rq = rq; | |
7884 | #endif | |
67e9fb2a | 7885 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7886 | } |
7887 | ||
fa85ae24 PZ |
7888 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7889 | { | |
7890 | struct rt_prio_array *array; | |
7891 | int i; | |
7892 | ||
7893 | array = &rt_rq->active; | |
7894 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7895 | INIT_LIST_HEAD(array->queue + i); | |
7896 | __clear_bit(i, array->bitmap); | |
7897 | } | |
7898 | /* delimiter for bitsearch: */ | |
7899 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7900 | ||
052f1dc7 | 7901 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7902 | rt_rq->highest_prio = MAX_RT_PRIO; |
7903 | #endif | |
fa85ae24 PZ |
7904 | #ifdef CONFIG_SMP |
7905 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7906 | rt_rq->overloaded = 0; |
7907 | #endif | |
7908 | ||
7909 | rt_rq->rt_time = 0; | |
7910 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7911 | rt_rq->rt_runtime = 0; |
7912 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7913 | |
052f1dc7 | 7914 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7915 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7916 | rt_rq->rq = rq; |
7917 | #endif | |
fa85ae24 PZ |
7918 | } |
7919 | ||
6f505b16 | 7920 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7921 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7922 | struct sched_entity *se, int cpu, int add, | |
7923 | struct sched_entity *parent) | |
6f505b16 | 7924 | { |
ec7dc8ac | 7925 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7926 | tg->cfs_rq[cpu] = cfs_rq; |
7927 | init_cfs_rq(cfs_rq, rq); | |
7928 | cfs_rq->tg = tg; | |
7929 | if (add) | |
7930 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7931 | ||
7932 | tg->se[cpu] = se; | |
354d60c2 DG |
7933 | /* se could be NULL for init_task_group */ |
7934 | if (!se) | |
7935 | return; | |
7936 | ||
ec7dc8ac DG |
7937 | if (!parent) |
7938 | se->cfs_rq = &rq->cfs; | |
7939 | else | |
7940 | se->cfs_rq = parent->my_q; | |
7941 | ||
6f505b16 PZ |
7942 | se->my_q = cfs_rq; |
7943 | se->load.weight = tg->shares; | |
e05510d0 | 7944 | se->load.inv_weight = 0; |
ec7dc8ac | 7945 | se->parent = parent; |
6f505b16 | 7946 | } |
052f1dc7 | 7947 | #endif |
6f505b16 | 7948 | |
052f1dc7 | 7949 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7950 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7951 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7952 | struct sched_rt_entity *parent) | |
6f505b16 | 7953 | { |
ec7dc8ac DG |
7954 | struct rq *rq = cpu_rq(cpu); |
7955 | ||
6f505b16 PZ |
7956 | tg->rt_rq[cpu] = rt_rq; |
7957 | init_rt_rq(rt_rq, rq); | |
7958 | rt_rq->tg = tg; | |
7959 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7960 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7961 | if (add) |
7962 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7963 | ||
7964 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7965 | if (!rt_se) |
7966 | return; | |
7967 | ||
ec7dc8ac DG |
7968 | if (!parent) |
7969 | rt_se->rt_rq = &rq->rt; | |
7970 | else | |
7971 | rt_se->rt_rq = parent->my_q; | |
7972 | ||
6f505b16 | 7973 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7974 | rt_se->parent = parent; |
6f505b16 PZ |
7975 | INIT_LIST_HEAD(&rt_se->run_list); |
7976 | } | |
7977 | #endif | |
7978 | ||
1da177e4 LT |
7979 | void __init sched_init(void) |
7980 | { | |
dd41f596 | 7981 | int i, j; |
434d53b0 MT |
7982 | unsigned long alloc_size = 0, ptr; |
7983 | ||
7984 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7985 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7986 | #endif | |
7987 | #ifdef CONFIG_RT_GROUP_SCHED | |
7988 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7989 | #endif |
7990 | #ifdef CONFIG_USER_SCHED | |
7991 | alloc_size *= 2; | |
434d53b0 MT |
7992 | #endif |
7993 | /* | |
7994 | * As sched_init() is called before page_alloc is setup, | |
7995 | * we use alloc_bootmem(). | |
7996 | */ | |
7997 | if (alloc_size) { | |
5a9d3225 | 7998 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7999 | |
8000 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8001 | init_task_group.se = (struct sched_entity **)ptr; | |
8002 | ptr += nr_cpu_ids * sizeof(void **); | |
8003 | ||
8004 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8005 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8006 | |
8007 | #ifdef CONFIG_USER_SCHED | |
8008 | root_task_group.se = (struct sched_entity **)ptr; | |
8009 | ptr += nr_cpu_ids * sizeof(void **); | |
8010 | ||
8011 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8012 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8013 | #endif /* CONFIG_USER_SCHED */ |
8014 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8015 | #ifdef CONFIG_RT_GROUP_SCHED |
8016 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8017 | ptr += nr_cpu_ids * sizeof(void **); | |
8018 | ||
8019 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8020 | ptr += nr_cpu_ids * sizeof(void **); |
8021 | ||
8022 | #ifdef CONFIG_USER_SCHED | |
8023 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8024 | ptr += nr_cpu_ids * sizeof(void **); | |
8025 | ||
8026 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8027 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8028 | #endif /* CONFIG_USER_SCHED */ |
8029 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8030 | } |
dd41f596 | 8031 | |
57d885fe GH |
8032 | #ifdef CONFIG_SMP |
8033 | init_defrootdomain(); | |
8034 | #endif | |
8035 | ||
d0b27fa7 PZ |
8036 | init_rt_bandwidth(&def_rt_bandwidth, |
8037 | global_rt_period(), global_rt_runtime()); | |
8038 | ||
8039 | #ifdef CONFIG_RT_GROUP_SCHED | |
8040 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8041 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8042 | #ifdef CONFIG_USER_SCHED |
8043 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8044 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8045 | #endif /* CONFIG_USER_SCHED */ |
8046 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8047 | |
052f1dc7 | 8048 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8049 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8050 | INIT_LIST_HEAD(&init_task_group.children); |
8051 | ||
8052 | #ifdef CONFIG_USER_SCHED | |
8053 | INIT_LIST_HEAD(&root_task_group.children); | |
8054 | init_task_group.parent = &root_task_group; | |
8055 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8056 | #endif /* CONFIG_USER_SCHED */ |
8057 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8058 | |
0a945022 | 8059 | for_each_possible_cpu(i) { |
70b97a7f | 8060 | struct rq *rq; |
1da177e4 LT |
8061 | |
8062 | rq = cpu_rq(i); | |
8063 | spin_lock_init(&rq->lock); | |
7897986b | 8064 | rq->nr_running = 0; |
dd41f596 | 8065 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8066 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8067 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8068 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8069 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8070 | #ifdef CONFIG_CGROUP_SCHED |
8071 | /* | |
8072 | * How much cpu bandwidth does init_task_group get? | |
8073 | * | |
8074 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8075 | * gets 100% of the cpu resources in the system. This overall | |
8076 | * system cpu resource is divided among the tasks of | |
8077 | * init_task_group and its child task-groups in a fair manner, | |
8078 | * based on each entity's (task or task-group's) weight | |
8079 | * (se->load.weight). | |
8080 | * | |
8081 | * In other words, if init_task_group has 10 tasks of weight | |
8082 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8083 | * then A0's share of the cpu resource is: | |
8084 | * | |
8085 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8086 | * | |
8087 | * We achieve this by letting init_task_group's tasks sit | |
8088 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8089 | */ | |
ec7dc8ac | 8090 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8091 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8092 | root_task_group.shares = NICE_0_LOAD; |
8093 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8094 | /* |
8095 | * In case of task-groups formed thr' the user id of tasks, | |
8096 | * init_task_group represents tasks belonging to root user. | |
8097 | * Hence it forms a sibling of all subsequent groups formed. | |
8098 | * In this case, init_task_group gets only a fraction of overall | |
8099 | * system cpu resource, based on the weight assigned to root | |
8100 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8101 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8102 | * (init_cfs_rq) and having one entity represent this group of | |
8103 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8104 | */ | |
ec7dc8ac | 8105 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8106 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8107 | &per_cpu(init_sched_entity, i), i, 1, |
8108 | root_task_group.se[i]); | |
6f505b16 | 8109 | |
052f1dc7 | 8110 | #endif |
354d60c2 DG |
8111 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8112 | ||
8113 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8114 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8115 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8116 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8117 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8118 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8119 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8120 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8121 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8122 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8123 | root_task_group.rt_se[i]); | |
354d60c2 | 8124 | #endif |
dd41f596 | 8125 | #endif |
1da177e4 | 8126 | |
dd41f596 IM |
8127 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8128 | rq->cpu_load[j] = 0; | |
1da177e4 | 8129 | #ifdef CONFIG_SMP |
41c7ce9a | 8130 | rq->sd = NULL; |
57d885fe | 8131 | rq->rd = NULL; |
1da177e4 | 8132 | rq->active_balance = 0; |
dd41f596 | 8133 | rq->next_balance = jiffies; |
1da177e4 | 8134 | rq->push_cpu = 0; |
0a2966b4 | 8135 | rq->cpu = i; |
1f11eb6a | 8136 | rq->online = 0; |
1da177e4 LT |
8137 | rq->migration_thread = NULL; |
8138 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8139 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8140 | #endif |
8f4d37ec | 8141 | init_rq_hrtick(rq); |
1da177e4 | 8142 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8143 | } |
8144 | ||
2dd73a4f | 8145 | set_load_weight(&init_task); |
b50f60ce | 8146 | |
e107be36 AK |
8147 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8148 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8149 | #endif | |
8150 | ||
c9819f45 | 8151 | #ifdef CONFIG_SMP |
962cf36c | 8152 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8153 | #endif |
8154 | ||
b50f60ce HC |
8155 | #ifdef CONFIG_RT_MUTEXES |
8156 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8157 | #endif | |
8158 | ||
1da177e4 LT |
8159 | /* |
8160 | * The boot idle thread does lazy MMU switching as well: | |
8161 | */ | |
8162 | atomic_inc(&init_mm.mm_count); | |
8163 | enter_lazy_tlb(&init_mm, current); | |
8164 | ||
8165 | /* | |
8166 | * Make us the idle thread. Technically, schedule() should not be | |
8167 | * called from this thread, however somewhere below it might be, | |
8168 | * but because we are the idle thread, we just pick up running again | |
8169 | * when this runqueue becomes "idle". | |
8170 | */ | |
8171 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8172 | /* |
8173 | * During early bootup we pretend to be a normal task: | |
8174 | */ | |
8175 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8176 | |
8177 | scheduler_running = 1; | |
1da177e4 LT |
8178 | } |
8179 | ||
8180 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8181 | void __might_sleep(char *file, int line) | |
8182 | { | |
48f24c4d | 8183 | #ifdef in_atomic |
1da177e4 LT |
8184 | static unsigned long prev_jiffy; /* ratelimiting */ |
8185 | ||
8186 | if ((in_atomic() || irqs_disabled()) && | |
8187 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8188 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8189 | return; | |
8190 | prev_jiffy = jiffies; | |
91368d73 | 8191 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8192 | " context at %s:%d\n", file, line); |
8193 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8194 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8195 | debug_show_held_locks(current); |
3117df04 IM |
8196 | if (irqs_disabled()) |
8197 | print_irqtrace_events(current); | |
1da177e4 LT |
8198 | dump_stack(); |
8199 | } | |
8200 | #endif | |
8201 | } | |
8202 | EXPORT_SYMBOL(__might_sleep); | |
8203 | #endif | |
8204 | ||
8205 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8206 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8207 | { | |
8208 | int on_rq; | |
3e51f33f | 8209 | |
3a5e4dc1 AK |
8210 | update_rq_clock(rq); |
8211 | on_rq = p->se.on_rq; | |
8212 | if (on_rq) | |
8213 | deactivate_task(rq, p, 0); | |
8214 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8215 | if (on_rq) { | |
8216 | activate_task(rq, p, 0); | |
8217 | resched_task(rq->curr); | |
8218 | } | |
8219 | } | |
8220 | ||
1da177e4 LT |
8221 | void normalize_rt_tasks(void) |
8222 | { | |
a0f98a1c | 8223 | struct task_struct *g, *p; |
1da177e4 | 8224 | unsigned long flags; |
70b97a7f | 8225 | struct rq *rq; |
1da177e4 | 8226 | |
4cf5d77a | 8227 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8228 | do_each_thread(g, p) { |
178be793 IM |
8229 | /* |
8230 | * Only normalize user tasks: | |
8231 | */ | |
8232 | if (!p->mm) | |
8233 | continue; | |
8234 | ||
6cfb0d5d | 8235 | p->se.exec_start = 0; |
6cfb0d5d | 8236 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8237 | p->se.wait_start = 0; |
dd41f596 | 8238 | p->se.sleep_start = 0; |
dd41f596 | 8239 | p->se.block_start = 0; |
6cfb0d5d | 8240 | #endif |
dd41f596 IM |
8241 | |
8242 | if (!rt_task(p)) { | |
8243 | /* | |
8244 | * Renice negative nice level userspace | |
8245 | * tasks back to 0: | |
8246 | */ | |
8247 | if (TASK_NICE(p) < 0 && p->mm) | |
8248 | set_user_nice(p, 0); | |
1da177e4 | 8249 | continue; |
dd41f596 | 8250 | } |
1da177e4 | 8251 | |
4cf5d77a | 8252 | spin_lock(&p->pi_lock); |
b29739f9 | 8253 | rq = __task_rq_lock(p); |
1da177e4 | 8254 | |
178be793 | 8255 | normalize_task(rq, p); |
3a5e4dc1 | 8256 | |
b29739f9 | 8257 | __task_rq_unlock(rq); |
4cf5d77a | 8258 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8259 | } while_each_thread(g, p); |
8260 | ||
4cf5d77a | 8261 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8262 | } |
8263 | ||
8264 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8265 | |
8266 | #ifdef CONFIG_IA64 | |
8267 | /* | |
8268 | * These functions are only useful for the IA64 MCA handling. | |
8269 | * | |
8270 | * They can only be called when the whole system has been | |
8271 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8272 | * activity can take place. Using them for anything else would | |
8273 | * be a serious bug, and as a result, they aren't even visible | |
8274 | * under any other configuration. | |
8275 | */ | |
8276 | ||
8277 | /** | |
8278 | * curr_task - return the current task for a given cpu. | |
8279 | * @cpu: the processor in question. | |
8280 | * | |
8281 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8282 | */ | |
36c8b586 | 8283 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8284 | { |
8285 | return cpu_curr(cpu); | |
8286 | } | |
8287 | ||
8288 | /** | |
8289 | * set_curr_task - set the current task for a given cpu. | |
8290 | * @cpu: the processor in question. | |
8291 | * @p: the task pointer to set. | |
8292 | * | |
8293 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8294 | * are serviced on a separate stack. It allows the architecture to switch the |
8295 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8296 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8297 | * and caller must save the original value of the current task (see | |
8298 | * curr_task() above) and restore that value before reenabling interrupts and | |
8299 | * re-starting the system. | |
8300 | * | |
8301 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8302 | */ | |
36c8b586 | 8303 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8304 | { |
8305 | cpu_curr(cpu) = p; | |
8306 | } | |
8307 | ||
8308 | #endif | |
29f59db3 | 8309 | |
bccbe08a PZ |
8310 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8311 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8312 | { |
8313 | int i; | |
8314 | ||
8315 | for_each_possible_cpu(i) { | |
8316 | if (tg->cfs_rq) | |
8317 | kfree(tg->cfs_rq[i]); | |
8318 | if (tg->se) | |
8319 | kfree(tg->se[i]); | |
6f505b16 PZ |
8320 | } |
8321 | ||
8322 | kfree(tg->cfs_rq); | |
8323 | kfree(tg->se); | |
6f505b16 PZ |
8324 | } |
8325 | ||
ec7dc8ac DG |
8326 | static |
8327 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8328 | { |
29f59db3 | 8329 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8330 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8331 | struct rq *rq; |
29f59db3 SV |
8332 | int i; |
8333 | ||
434d53b0 | 8334 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8335 | if (!tg->cfs_rq) |
8336 | goto err; | |
434d53b0 | 8337 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8338 | if (!tg->se) |
8339 | goto err; | |
052f1dc7 PZ |
8340 | |
8341 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8342 | |
8343 | for_each_possible_cpu(i) { | |
9b5b7751 | 8344 | rq = cpu_rq(i); |
29f59db3 | 8345 | |
6f505b16 PZ |
8346 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8347 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8348 | if (!cfs_rq) |
8349 | goto err; | |
8350 | ||
6f505b16 PZ |
8351 | se = kmalloc_node(sizeof(struct sched_entity), |
8352 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8353 | if (!se) |
8354 | goto err; | |
8355 | ||
ec7dc8ac DG |
8356 | parent_se = parent ? parent->se[i] : NULL; |
8357 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8358 | } |
8359 | ||
8360 | return 1; | |
8361 | ||
8362 | err: | |
8363 | return 0; | |
8364 | } | |
8365 | ||
8366 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8367 | { | |
8368 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8369 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8370 | } | |
8371 | ||
8372 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8373 | { | |
8374 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8375 | } | |
6d6bc0ad | 8376 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8377 | static inline void free_fair_sched_group(struct task_group *tg) |
8378 | { | |
8379 | } | |
8380 | ||
ec7dc8ac DG |
8381 | static inline |
8382 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8383 | { |
8384 | return 1; | |
8385 | } | |
8386 | ||
8387 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8388 | { | |
8389 | } | |
8390 | ||
8391 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8392 | { | |
8393 | } | |
6d6bc0ad | 8394 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8395 | |
8396 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8397 | static void free_rt_sched_group(struct task_group *tg) |
8398 | { | |
8399 | int i; | |
8400 | ||
d0b27fa7 PZ |
8401 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8402 | ||
bccbe08a PZ |
8403 | for_each_possible_cpu(i) { |
8404 | if (tg->rt_rq) | |
8405 | kfree(tg->rt_rq[i]); | |
8406 | if (tg->rt_se) | |
8407 | kfree(tg->rt_se[i]); | |
8408 | } | |
8409 | ||
8410 | kfree(tg->rt_rq); | |
8411 | kfree(tg->rt_se); | |
8412 | } | |
8413 | ||
ec7dc8ac DG |
8414 | static |
8415 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8416 | { |
8417 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8418 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8419 | struct rq *rq; |
8420 | int i; | |
8421 | ||
434d53b0 | 8422 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8423 | if (!tg->rt_rq) |
8424 | goto err; | |
434d53b0 | 8425 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8426 | if (!tg->rt_se) |
8427 | goto err; | |
8428 | ||
d0b27fa7 PZ |
8429 | init_rt_bandwidth(&tg->rt_bandwidth, |
8430 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8431 | |
8432 | for_each_possible_cpu(i) { | |
8433 | rq = cpu_rq(i); | |
8434 | ||
6f505b16 PZ |
8435 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8436 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8437 | if (!rt_rq) | |
8438 | goto err; | |
29f59db3 | 8439 | |
6f505b16 PZ |
8440 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8441 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8442 | if (!rt_se) | |
8443 | goto err; | |
29f59db3 | 8444 | |
ec7dc8ac DG |
8445 | parent_se = parent ? parent->rt_se[i] : NULL; |
8446 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8447 | } |
8448 | ||
bccbe08a PZ |
8449 | return 1; |
8450 | ||
8451 | err: | |
8452 | return 0; | |
8453 | } | |
8454 | ||
8455 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8456 | { | |
8457 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8458 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8459 | } | |
8460 | ||
8461 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8462 | { | |
8463 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8464 | } | |
6d6bc0ad | 8465 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8466 | static inline void free_rt_sched_group(struct task_group *tg) |
8467 | { | |
8468 | } | |
8469 | ||
ec7dc8ac DG |
8470 | static inline |
8471 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8472 | { |
8473 | return 1; | |
8474 | } | |
8475 | ||
8476 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8477 | { | |
8478 | } | |
8479 | ||
8480 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8481 | { | |
8482 | } | |
6d6bc0ad | 8483 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8484 | |
d0b27fa7 | 8485 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8486 | static void free_sched_group(struct task_group *tg) |
8487 | { | |
8488 | free_fair_sched_group(tg); | |
8489 | free_rt_sched_group(tg); | |
8490 | kfree(tg); | |
8491 | } | |
8492 | ||
8493 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8494 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8495 | { |
8496 | struct task_group *tg; | |
8497 | unsigned long flags; | |
8498 | int i; | |
8499 | ||
8500 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8501 | if (!tg) | |
8502 | return ERR_PTR(-ENOMEM); | |
8503 | ||
ec7dc8ac | 8504 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8505 | goto err; |
8506 | ||
ec7dc8ac | 8507 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8508 | goto err; |
8509 | ||
8ed36996 | 8510 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8511 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8512 | register_fair_sched_group(tg, i); |
8513 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8514 | } |
6f505b16 | 8515 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8516 | |
8517 | WARN_ON(!parent); /* root should already exist */ | |
8518 | ||
8519 | tg->parent = parent; | |
f473aa5e | 8520 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8521 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8522 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8523 | |
9b5b7751 | 8524 | return tg; |
29f59db3 SV |
8525 | |
8526 | err: | |
6f505b16 | 8527 | free_sched_group(tg); |
29f59db3 SV |
8528 | return ERR_PTR(-ENOMEM); |
8529 | } | |
8530 | ||
9b5b7751 | 8531 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8532 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8533 | { |
29f59db3 | 8534 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8535 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8536 | } |
8537 | ||
9b5b7751 | 8538 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8539 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8540 | { |
8ed36996 | 8541 | unsigned long flags; |
9b5b7751 | 8542 | int i; |
29f59db3 | 8543 | |
8ed36996 | 8544 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8545 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8546 | unregister_fair_sched_group(tg, i); |
8547 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8548 | } |
6f505b16 | 8549 | list_del_rcu(&tg->list); |
f473aa5e | 8550 | list_del_rcu(&tg->siblings); |
8ed36996 | 8551 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8552 | |
9b5b7751 | 8553 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8554 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8555 | } |
8556 | ||
9b5b7751 | 8557 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8558 | * The caller of this function should have put the task in its new group |
8559 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8560 | * reflect its new group. | |
9b5b7751 SV |
8561 | */ |
8562 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8563 | { |
8564 | int on_rq, running; | |
8565 | unsigned long flags; | |
8566 | struct rq *rq; | |
8567 | ||
8568 | rq = task_rq_lock(tsk, &flags); | |
8569 | ||
29f59db3 SV |
8570 | update_rq_clock(rq); |
8571 | ||
051a1d1a | 8572 | running = task_current(rq, tsk); |
29f59db3 SV |
8573 | on_rq = tsk->se.on_rq; |
8574 | ||
0e1f3483 | 8575 | if (on_rq) |
29f59db3 | 8576 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8577 | if (unlikely(running)) |
8578 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8579 | |
6f505b16 | 8580 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8581 | |
810b3817 PZ |
8582 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8583 | if (tsk->sched_class->moved_group) | |
8584 | tsk->sched_class->moved_group(tsk); | |
8585 | #endif | |
8586 | ||
0e1f3483 HS |
8587 | if (unlikely(running)) |
8588 | tsk->sched_class->set_curr_task(rq); | |
8589 | if (on_rq) | |
7074badb | 8590 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8591 | |
29f59db3 SV |
8592 | task_rq_unlock(rq, &flags); |
8593 | } | |
6d6bc0ad | 8594 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8595 | |
052f1dc7 | 8596 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8597 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8598 | { |
8599 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8600 | int on_rq; |
8601 | ||
29f59db3 | 8602 | on_rq = se->on_rq; |
62fb1851 | 8603 | if (on_rq) |
29f59db3 SV |
8604 | dequeue_entity(cfs_rq, se, 0); |
8605 | ||
8606 | se->load.weight = shares; | |
e05510d0 | 8607 | se->load.inv_weight = 0; |
29f59db3 | 8608 | |
62fb1851 | 8609 | if (on_rq) |
29f59db3 | 8610 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8611 | } |
62fb1851 | 8612 | |
c09595f6 PZ |
8613 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8614 | { | |
8615 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8616 | struct rq *rq = cfs_rq->rq; | |
8617 | unsigned long flags; | |
8618 | ||
8619 | spin_lock_irqsave(&rq->lock, flags); | |
8620 | __set_se_shares(se, shares); | |
8621 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8622 | } |
8623 | ||
8ed36996 PZ |
8624 | static DEFINE_MUTEX(shares_mutex); |
8625 | ||
4cf86d77 | 8626 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8627 | { |
8628 | int i; | |
8ed36996 | 8629 | unsigned long flags; |
c61935fd | 8630 | |
ec7dc8ac DG |
8631 | /* |
8632 | * We can't change the weight of the root cgroup. | |
8633 | */ | |
8634 | if (!tg->se[0]) | |
8635 | return -EINVAL; | |
8636 | ||
18d95a28 PZ |
8637 | if (shares < MIN_SHARES) |
8638 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8639 | else if (shares > MAX_SHARES) |
8640 | shares = MAX_SHARES; | |
62fb1851 | 8641 | |
8ed36996 | 8642 | mutex_lock(&shares_mutex); |
9b5b7751 | 8643 | if (tg->shares == shares) |
5cb350ba | 8644 | goto done; |
29f59db3 | 8645 | |
8ed36996 | 8646 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8647 | for_each_possible_cpu(i) |
8648 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8649 | list_del_rcu(&tg->siblings); |
8ed36996 | 8650 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8651 | |
8652 | /* wait for any ongoing reference to this group to finish */ | |
8653 | synchronize_sched(); | |
8654 | ||
8655 | /* | |
8656 | * Now we are free to modify the group's share on each cpu | |
8657 | * w/o tripping rebalance_share or load_balance_fair. | |
8658 | */ | |
9b5b7751 | 8659 | tg->shares = shares; |
c09595f6 PZ |
8660 | for_each_possible_cpu(i) { |
8661 | /* | |
8662 | * force a rebalance | |
8663 | */ | |
8664 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8665 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8666 | } |
29f59db3 | 8667 | |
6b2d7700 SV |
8668 | /* |
8669 | * Enable load balance activity on this group, by inserting it back on | |
8670 | * each cpu's rq->leaf_cfs_rq_list. | |
8671 | */ | |
8ed36996 | 8672 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8673 | for_each_possible_cpu(i) |
8674 | register_fair_sched_group(tg, i); | |
f473aa5e | 8675 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8676 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8677 | done: |
8ed36996 | 8678 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8679 | return 0; |
29f59db3 SV |
8680 | } |
8681 | ||
5cb350ba DG |
8682 | unsigned long sched_group_shares(struct task_group *tg) |
8683 | { | |
8684 | return tg->shares; | |
8685 | } | |
052f1dc7 | 8686 | #endif |
5cb350ba | 8687 | |
052f1dc7 | 8688 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8689 | /* |
9f0c1e56 | 8690 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8691 | */ |
9f0c1e56 PZ |
8692 | static DEFINE_MUTEX(rt_constraints_mutex); |
8693 | ||
8694 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8695 | { | |
8696 | if (runtime == RUNTIME_INF) | |
8697 | return 1ULL << 16; | |
8698 | ||
6f6d6a1a | 8699 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8700 | } |
8701 | ||
b40b2e8e PZ |
8702 | #ifdef CONFIG_CGROUP_SCHED |
8703 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8704 | { | |
10b612f4 | 8705 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8706 | unsigned long total = 0; |
8707 | ||
8708 | if (!parent) { | |
8709 | if (global_rt_period() < period) | |
8710 | return 0; | |
8711 | ||
8712 | return to_ratio(period, runtime) < | |
8713 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8714 | } | |
8715 | ||
8716 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8717 | return 0; | |
8718 | ||
8719 | rcu_read_lock(); | |
8720 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8721 | if (tgi == tg) | |
8722 | continue; | |
8723 | ||
8724 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8725 | tgi->rt_bandwidth.rt_runtime); | |
8726 | } | |
8727 | rcu_read_unlock(); | |
8728 | ||
10b612f4 | 8729 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8730 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8731 | parent->rt_bandwidth.rt_runtime); | |
8732 | } | |
8733 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8734 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8735 | { |
8736 | struct task_group *tgi; | |
8737 | unsigned long total = 0; | |
9f0c1e56 | 8738 | unsigned long global_ratio = |
d0b27fa7 | 8739 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8740 | |
8741 | rcu_read_lock(); | |
9f0c1e56 PZ |
8742 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8743 | if (tgi == tg) | |
8744 | continue; | |
6f505b16 | 8745 | |
d0b27fa7 PZ |
8746 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8747 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8748 | } |
8749 | rcu_read_unlock(); | |
6f505b16 | 8750 | |
9f0c1e56 | 8751 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8752 | } |
b40b2e8e | 8753 | #endif |
6f505b16 | 8754 | |
521f1a24 DG |
8755 | /* Must be called with tasklist_lock held */ |
8756 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8757 | { | |
8758 | struct task_struct *g, *p; | |
8759 | do_each_thread(g, p) { | |
8760 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8761 | return 1; | |
8762 | } while_each_thread(g, p); | |
8763 | return 0; | |
8764 | } | |
8765 | ||
d0b27fa7 PZ |
8766 | static int tg_set_bandwidth(struct task_group *tg, |
8767 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8768 | { |
ac086bc2 | 8769 | int i, err = 0; |
9f0c1e56 | 8770 | |
9f0c1e56 | 8771 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8772 | read_lock(&tasklist_lock); |
ac086bc2 | 8773 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8774 | err = -EBUSY; |
8775 | goto unlock; | |
8776 | } | |
9f0c1e56 PZ |
8777 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8778 | err = -EINVAL; | |
8779 | goto unlock; | |
8780 | } | |
ac086bc2 PZ |
8781 | |
8782 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8783 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8784 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8785 | |
8786 | for_each_possible_cpu(i) { | |
8787 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8788 | ||
8789 | spin_lock(&rt_rq->rt_runtime_lock); | |
8790 | rt_rq->rt_runtime = rt_runtime; | |
8791 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8792 | } | |
8793 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8794 | unlock: |
521f1a24 | 8795 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8796 | mutex_unlock(&rt_constraints_mutex); |
8797 | ||
8798 | return err; | |
6f505b16 PZ |
8799 | } |
8800 | ||
d0b27fa7 PZ |
8801 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8802 | { | |
8803 | u64 rt_runtime, rt_period; | |
8804 | ||
8805 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8806 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8807 | if (rt_runtime_us < 0) | |
8808 | rt_runtime = RUNTIME_INF; | |
8809 | ||
8810 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8811 | } | |
8812 | ||
9f0c1e56 PZ |
8813 | long sched_group_rt_runtime(struct task_group *tg) |
8814 | { | |
8815 | u64 rt_runtime_us; | |
8816 | ||
d0b27fa7 | 8817 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8818 | return -1; |
8819 | ||
d0b27fa7 | 8820 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8821 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8822 | return rt_runtime_us; | |
8823 | } | |
d0b27fa7 PZ |
8824 | |
8825 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8826 | { | |
8827 | u64 rt_runtime, rt_period; | |
8828 | ||
8829 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8830 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8831 | ||
619b0488 R |
8832 | if (rt_period == 0) |
8833 | return -EINVAL; | |
8834 | ||
d0b27fa7 PZ |
8835 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8836 | } | |
8837 | ||
8838 | long sched_group_rt_period(struct task_group *tg) | |
8839 | { | |
8840 | u64 rt_period_us; | |
8841 | ||
8842 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8843 | do_div(rt_period_us, NSEC_PER_USEC); | |
8844 | return rt_period_us; | |
8845 | } | |
8846 | ||
8847 | static int sched_rt_global_constraints(void) | |
8848 | { | |
10b612f4 PZ |
8849 | struct task_group *tg = &root_task_group; |
8850 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8851 | int ret = 0; |
8852 | ||
10b612f4 PZ |
8853 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8854 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8855 | ||
d0b27fa7 | 8856 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8857 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8858 | ret = -EINVAL; |
8859 | mutex_unlock(&rt_constraints_mutex); | |
8860 | ||
8861 | return ret; | |
8862 | } | |
6d6bc0ad | 8863 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8864 | static int sched_rt_global_constraints(void) |
8865 | { | |
ac086bc2 PZ |
8866 | unsigned long flags; |
8867 | int i; | |
8868 | ||
8869 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8870 | for_each_possible_cpu(i) { | |
8871 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8872 | ||
8873 | spin_lock(&rt_rq->rt_runtime_lock); | |
8874 | rt_rq->rt_runtime = global_rt_runtime(); | |
8875 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8876 | } | |
8877 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8878 | ||
d0b27fa7 PZ |
8879 | return 0; |
8880 | } | |
6d6bc0ad | 8881 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8882 | |
8883 | int sched_rt_handler(struct ctl_table *table, int write, | |
8884 | struct file *filp, void __user *buffer, size_t *lenp, | |
8885 | loff_t *ppos) | |
8886 | { | |
8887 | int ret; | |
8888 | int old_period, old_runtime; | |
8889 | static DEFINE_MUTEX(mutex); | |
8890 | ||
8891 | mutex_lock(&mutex); | |
8892 | old_period = sysctl_sched_rt_period; | |
8893 | old_runtime = sysctl_sched_rt_runtime; | |
8894 | ||
8895 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8896 | ||
8897 | if (!ret && write) { | |
8898 | ret = sched_rt_global_constraints(); | |
8899 | if (ret) { | |
8900 | sysctl_sched_rt_period = old_period; | |
8901 | sysctl_sched_rt_runtime = old_runtime; | |
8902 | } else { | |
8903 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8904 | def_rt_bandwidth.rt_period = | |
8905 | ns_to_ktime(global_rt_period()); | |
8906 | } | |
8907 | } | |
8908 | mutex_unlock(&mutex); | |
8909 | ||
8910 | return ret; | |
8911 | } | |
68318b8e | 8912 | |
052f1dc7 | 8913 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8914 | |
8915 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8916 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8917 | { |
2b01dfe3 PM |
8918 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8919 | struct task_group, css); | |
68318b8e SV |
8920 | } |
8921 | ||
8922 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8923 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8924 | { |
ec7dc8ac | 8925 | struct task_group *tg, *parent; |
68318b8e | 8926 | |
2b01dfe3 | 8927 | if (!cgrp->parent) { |
68318b8e | 8928 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8929 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8930 | return &init_task_group.css; |
8931 | } | |
8932 | ||
ec7dc8ac DG |
8933 | parent = cgroup_tg(cgrp->parent); |
8934 | tg = sched_create_group(parent); | |
68318b8e SV |
8935 | if (IS_ERR(tg)) |
8936 | return ERR_PTR(-ENOMEM); | |
8937 | ||
8938 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8939 | tg->css.cgroup = cgrp; |
68318b8e SV |
8940 | |
8941 | return &tg->css; | |
8942 | } | |
8943 | ||
41a2d6cf IM |
8944 | static void |
8945 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8946 | { |
2b01dfe3 | 8947 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8948 | |
8949 | sched_destroy_group(tg); | |
8950 | } | |
8951 | ||
41a2d6cf IM |
8952 | static int |
8953 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8954 | struct task_struct *tsk) | |
68318b8e | 8955 | { |
b68aa230 PZ |
8956 | #ifdef CONFIG_RT_GROUP_SCHED |
8957 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8958 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8959 | return -EINVAL; |
8960 | #else | |
68318b8e SV |
8961 | /* We don't support RT-tasks being in separate groups */ |
8962 | if (tsk->sched_class != &fair_sched_class) | |
8963 | return -EINVAL; | |
b68aa230 | 8964 | #endif |
68318b8e SV |
8965 | |
8966 | return 0; | |
8967 | } | |
8968 | ||
8969 | static void | |
2b01dfe3 | 8970 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8971 | struct cgroup *old_cont, struct task_struct *tsk) |
8972 | { | |
8973 | sched_move_task(tsk); | |
8974 | } | |
8975 | ||
052f1dc7 | 8976 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8977 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8978 | u64 shareval) |
68318b8e | 8979 | { |
2b01dfe3 | 8980 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8981 | } |
8982 | ||
f4c753b7 | 8983 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8984 | { |
2b01dfe3 | 8985 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8986 | |
8987 | return (u64) tg->shares; | |
8988 | } | |
6d6bc0ad | 8989 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8990 | |
052f1dc7 | 8991 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8992 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8993 | s64 val) |
6f505b16 | 8994 | { |
06ecb27c | 8995 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8996 | } |
8997 | ||
06ecb27c | 8998 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8999 | { |
06ecb27c | 9000 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9001 | } |
d0b27fa7 PZ |
9002 | |
9003 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9004 | u64 rt_period_us) | |
9005 | { | |
9006 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9007 | } | |
9008 | ||
9009 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9010 | { | |
9011 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9012 | } | |
6d6bc0ad | 9013 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9014 | |
fe5c7cc2 | 9015 | static struct cftype cpu_files[] = { |
052f1dc7 | 9016 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9017 | { |
9018 | .name = "shares", | |
f4c753b7 PM |
9019 | .read_u64 = cpu_shares_read_u64, |
9020 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9021 | }, |
052f1dc7 PZ |
9022 | #endif |
9023 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9024 | { |
9f0c1e56 | 9025 | .name = "rt_runtime_us", |
06ecb27c PM |
9026 | .read_s64 = cpu_rt_runtime_read, |
9027 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9028 | }, |
d0b27fa7 PZ |
9029 | { |
9030 | .name = "rt_period_us", | |
f4c753b7 PM |
9031 | .read_u64 = cpu_rt_period_read_uint, |
9032 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9033 | }, |
052f1dc7 | 9034 | #endif |
68318b8e SV |
9035 | }; |
9036 | ||
9037 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9038 | { | |
fe5c7cc2 | 9039 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9040 | } |
9041 | ||
9042 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9043 | .name = "cpu", |
9044 | .create = cpu_cgroup_create, | |
9045 | .destroy = cpu_cgroup_destroy, | |
9046 | .can_attach = cpu_cgroup_can_attach, | |
9047 | .attach = cpu_cgroup_attach, | |
9048 | .populate = cpu_cgroup_populate, | |
9049 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9050 | .early_init = 1, |
9051 | }; | |
9052 | ||
052f1dc7 | 9053 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9054 | |
9055 | #ifdef CONFIG_CGROUP_CPUACCT | |
9056 | ||
9057 | /* | |
9058 | * CPU accounting code for task groups. | |
9059 | * | |
9060 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9061 | * (balbir@in.ibm.com). | |
9062 | */ | |
9063 | ||
9064 | /* track cpu usage of a group of tasks */ | |
9065 | struct cpuacct { | |
9066 | struct cgroup_subsys_state css; | |
9067 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9068 | u64 *cpuusage; | |
9069 | }; | |
9070 | ||
9071 | struct cgroup_subsys cpuacct_subsys; | |
9072 | ||
9073 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9074 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9075 | { |
32cd756a | 9076 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9077 | struct cpuacct, css); |
9078 | } | |
9079 | ||
9080 | /* return cpu accounting group to which this task belongs */ | |
9081 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9082 | { | |
9083 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9084 | struct cpuacct, css); | |
9085 | } | |
9086 | ||
9087 | /* create a new cpu accounting group */ | |
9088 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9089 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9090 | { |
9091 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9092 | ||
9093 | if (!ca) | |
9094 | return ERR_PTR(-ENOMEM); | |
9095 | ||
9096 | ca->cpuusage = alloc_percpu(u64); | |
9097 | if (!ca->cpuusage) { | |
9098 | kfree(ca); | |
9099 | return ERR_PTR(-ENOMEM); | |
9100 | } | |
9101 | ||
9102 | return &ca->css; | |
9103 | } | |
9104 | ||
9105 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9106 | static void |
32cd756a | 9107 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9108 | { |
32cd756a | 9109 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9110 | |
9111 | free_percpu(ca->cpuusage); | |
9112 | kfree(ca); | |
9113 | } | |
9114 | ||
9115 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9116 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9117 | { |
32cd756a | 9118 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9119 | u64 totalcpuusage = 0; |
9120 | int i; | |
9121 | ||
9122 | for_each_possible_cpu(i) { | |
9123 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9124 | ||
9125 | /* | |
9126 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9127 | * platforms. | |
9128 | */ | |
9129 | spin_lock_irq(&cpu_rq(i)->lock); | |
9130 | totalcpuusage += *cpuusage; | |
9131 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9132 | } | |
9133 | ||
9134 | return totalcpuusage; | |
9135 | } | |
9136 | ||
0297b803 DG |
9137 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9138 | u64 reset) | |
9139 | { | |
9140 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9141 | int err = 0; | |
9142 | int i; | |
9143 | ||
9144 | if (reset) { | |
9145 | err = -EINVAL; | |
9146 | goto out; | |
9147 | } | |
9148 | ||
9149 | for_each_possible_cpu(i) { | |
9150 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9151 | ||
9152 | spin_lock_irq(&cpu_rq(i)->lock); | |
9153 | *cpuusage = 0; | |
9154 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9155 | } | |
9156 | out: | |
9157 | return err; | |
9158 | } | |
9159 | ||
d842de87 SV |
9160 | static struct cftype files[] = { |
9161 | { | |
9162 | .name = "usage", | |
f4c753b7 PM |
9163 | .read_u64 = cpuusage_read, |
9164 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9165 | }, |
9166 | }; | |
9167 | ||
32cd756a | 9168 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9169 | { |
32cd756a | 9170 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9171 | } |
9172 | ||
9173 | /* | |
9174 | * charge this task's execution time to its accounting group. | |
9175 | * | |
9176 | * called with rq->lock held. | |
9177 | */ | |
9178 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9179 | { | |
9180 | struct cpuacct *ca; | |
9181 | ||
9182 | if (!cpuacct_subsys.active) | |
9183 | return; | |
9184 | ||
9185 | ca = task_ca(tsk); | |
9186 | if (ca) { | |
9187 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9188 | ||
9189 | *cpuusage += cputime; | |
9190 | } | |
9191 | } | |
9192 | ||
9193 | struct cgroup_subsys cpuacct_subsys = { | |
9194 | .name = "cpuacct", | |
9195 | .create = cpuacct_create, | |
9196 | .destroy = cpuacct_destroy, | |
9197 | .populate = cpuacct_populate, | |
9198 | .subsys_id = cpuacct_subsys_id, | |
9199 | }; | |
9200 | #endif /* CONFIG_CGROUP_CPUACCT */ |